builtin-top.c 47.4 KB
Newer Older
1
// SPDX-License-Identifier: GPL-2.0-only
2
/*
3 4 5 6 7 8
 * 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>
9
 *		 2011, Red Hat Inc, Arnaldo Carvalho de Melo <acme@redhat.com>
10 11 12 13 14 15 16 17
 *
 * 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>
18
 */
19
#include "builtin.h"
20

21
#include "perf.h"
22

23
#include "util/annotate.h"
24
#include "util/bpf-event.h"
25
#include "util/cgroup.h"
26
#include "util/config.h"
27
#include "util/color.h"
28
#include "util/dso.h"
29
#include "util/evlist.h"
30
#include "util/evsel.h"
31
#include "util/evsel_config.h"
32
#include "util/event.h"
33
#include "util/machine.h"
34
#include "util/map.h"
35
#include "util/mmap.h"
36
#include "util/session.h"
37
#include "util/thread.h"
38
#include "util/symbol.h"
39
#include "util/synthetic-events.h"
40
#include "util/top.h"
41
#include "util/util.h"
42
#include <linux/rbtree.h>
43
#include <subcmd/parse-options.h>
44
#include "util/parse-events.h"
45
#include "util/callchain.h"
46
#include "util/cpumap.h"
47
#include "util/sort.h"
48
#include "util/string2.h"
49
#include "util/term.h"
50
#include "util/intlist.h"
51
#include "util/parse-branch-options.h"
52
#include "arch/common.h"
53
#include "ui/ui.h"
54

55
#include "util/debug.h"
J
Jiri Olsa 已提交
56
#include "util/ordered-events.h"
57
#include "util/pfm.h"
58

59
#include <assert.h>
60
#include <elf.h>
61
#include <fcntl.h>
62

63
#include <stdio.h>
64 65
#include <termios.h>
#include <unistd.h>
66
#include <inttypes.h>
67

68 69 70
#include <errno.h>
#include <time.h>
#include <sched.h>
71
#include <signal.h>
72 73 74

#include <sys/syscall.h>
#include <sys/ioctl.h>
75
#include <poll.h>
76 77 78
#include <sys/prctl.h>
#include <sys/wait.h>
#include <sys/uio.h>
79
#include <sys/utsname.h>
80 81
#include <sys/mman.h>

82
#include <linux/stringify.h>
83
#include <linux/time64.h>
84
#include <linux/types.h>
85
#include <linux/err.h>
86

87
#include <linux/ctype.h>
88
#include <perf/mmap.h>
89

90
static volatile int done;
91
static volatile int resize;
92

N
Namhyung Kim 已提交
93 94
#define HEADER_LINE_NR  5

95
static void perf_top__update_print_entries(struct perf_top *top)
96
{
N
Namhyung Kim 已提交
97
	top->print_entries = top->winsize.ws_row - HEADER_LINE_NR;
98 99
}

100
static void winch_sig(int sig __maybe_unused)
101
{
102 103
	resize = 1;
}
104

105 106
static void perf_top__resize(struct perf_top *top)
{
107 108
	get_term_dimensions(&top->winsize);
	perf_top__update_print_entries(top);
109 110
}

111
static int perf_top__parse_source(struct perf_top *top, struct hist_entry *he)
112
{
113
	struct evsel *evsel;
114
	struct symbol *sym;
115
	struct annotation *notes;
116
	struct map *map;
117
	int err = -1;
118

119
	if (!he || !he->ms.sym)
120 121
		return -1;

122 123
	evsel = hists_to_evsel(he->hists);

124 125
	sym = he->ms.sym;
	map = he->ms.map;
126 127 128 129

	/*
	 * We can't annotate with just /proc/kallsyms
	 */
130 131
	if (map->dso->symtab_type == DSO_BINARY_TYPE__KALLSYMS &&
	    !dso__is_kcore(map->dso)) {
132 133 134
		pr_err("Can't annotate %s: No vmlinux file was found in the "
		       "path\n", sym->name);
		sleep(1);
135
		return -1;
136 137
	}

138 139
	notes = symbol__annotation(sym);
	pthread_mutex_lock(&notes->lock);
140

141
	if (!symbol__hists(sym, top->evlist->core.nr_entries)) {
142
		pthread_mutex_unlock(&notes->lock);
143 144
		pr_err("Not enough memory for annotating '%s' symbol!\n",
		       sym->name);
145
		sleep(1);
146
		return err;
147
	}
148

149
	err = symbol__annotate(&he->ms, evsel, &top->annotation_opts, NULL);
150
	if (err == 0) {
151
		top->sym_filter_entry = he;
152 153
	} else {
		char msg[BUFSIZ];
154
		symbol__strerror_disassemble(&he->ms, err, msg, sizeof(msg));
155
		pr_err("Couldn't annotate %s: %s\n", sym->name, msg);
156
	}
157

158
	pthread_mutex_unlock(&notes->lock);
159
	return err;
160 161
}

162
static void __zero_source_counters(struct hist_entry *he)
163
{
164
	struct symbol *sym = he->ms.sym;
165
	symbol__annotate_zero_histograms(sym);
166 167
}

168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190
static void ui__warn_map_erange(struct map *map, struct symbol *sym, u64 ip)
{
	struct utsname uts;
	int err = uname(&uts);

	ui__warning("Out of bounds address found:\n\n"
		    "Addr:   %" PRIx64 "\n"
		    "DSO:    %s %c\n"
		    "Map:    %" PRIx64 "-%" PRIx64 "\n"
		    "Symbol: %" PRIx64 "-%" PRIx64 " %c %s\n"
		    "Arch:   %s\n"
		    "Kernel: %s\n"
		    "Tools:  %s\n\n"
		    "Not all samples will be on the annotation output.\n\n"
		    "Please report to linux-kernel@vger.kernel.org\n",
		    ip, map->dso->long_name, dso__symtab_origin(map->dso),
		    map->start, map->end, sym->start, sym->end,
		    sym->binding == STB_GLOBAL ? 'g' :
		    sym->binding == STB_LOCAL  ? 'l' : 'w', sym->name,
		    err ? "[unknown]" : uts.machine,
		    err ? "[unknown]" : uts.release, perf_version_string);
	if (use_browser <= 0)
		sleep(5);
191

192 193 194
	map->erange_warned = true;
}

195 196
static void perf_top__record_precise_ip(struct perf_top *top,
					struct hist_entry *he,
197
					struct perf_sample *sample,
198
					struct evsel *evsel, u64 ip)
199
{
200
	struct annotation *notes;
201
	struct symbol *sym = he->ms.sym;
202
	int err = 0;
203

204 205 206
	if (sym == NULL || (use_browser == 0 &&
			    (top->sym_filter_entry == NULL ||
			     top->sym_filter_entry->ms.sym != sym)))
207 208
		return;

209 210 211
	notes = symbol__annotation(sym);

	if (pthread_mutex_trylock(&notes->lock))
212 213
		return;

214
	err = hist_entry__inc_addr_samples(he, sample, evsel, ip);
215

216
	pthread_mutex_unlock(&notes->lock);
217

218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233
	if (unlikely(err)) {
		/*
		 * This function is now called with he->hists->lock held.
		 * Release it before going to sleep.
		 */
		pthread_mutex_unlock(&he->hists->lock);

		if (err == -ERANGE && !he->ms.map->erange_warned)
			ui__warn_map_erange(he->ms.map, sym, ip);
		else if (err == -ENOMEM) {
			pr_err("Not enough memory for annotating '%s' symbol!\n",
			       sym->name);
			sleep(1);
		}

		pthread_mutex_lock(&he->hists->lock);
234
	}
235 236
}

237
static void perf_top__show_details(struct perf_top *top)
238
{
239
	struct hist_entry *he = top->sym_filter_entry;
240
	struct evsel *evsel;
241
	struct annotation *notes;
242
	struct symbol *symbol;
243
	int more;
244

245
	if (!he)
246 247
		return;

248 249
	evsel = hists_to_evsel(he->hists);

250
	symbol = he->ms.sym;
251 252 253 254
	notes = symbol__annotation(symbol);

	pthread_mutex_lock(&notes->lock);

255 256
	symbol__calc_percent(symbol, evsel);

257 258
	if (notes->src == NULL)
		goto out_unlock;
259

260
	printf("Showing %s for %s\n", evsel__name(top->sym_evsel), symbol->name);
261
	printf("  Events  Pcnt (>=%d%%)\n", top->annotation_opts.min_pcnt);
262

263
	more = symbol__annotate_printf(&he->ms, top->sym_evsel, &top->annotation_opts);
264 265 266

	if (top->evlist->enabled) {
		if (top->zero)
267
			symbol__annotate_zero_histogram(symbol, top->sym_evsel->core.idx);
268
		else
269
			symbol__annotate_decay_histogram(symbol, top->sym_evsel->core.idx);
270
	}
271
	if (more != 0)
272
		printf("%d lines not displayed, maybe increase display entries [e]\n", more);
273 274
out_unlock:
	pthread_mutex_unlock(&notes->lock);
275
}
276

277
static void perf_top__resort_hists(struct perf_top *t)
278
{
279
	struct evlist *evlist = t->evlist;
280 281 282 283 284
	struct evsel *pos;

	evlist__for_each_entry(evlist, pos) {
		struct hists *hists = evsel__hists(pos);

285 286 287 288 289 290
		/*
		 * unlink existing entries so that they can be linked
		 * in a correct order in hists__match() below.
		 */
		hists__unlink(hists);

291 292 293 294 295 296 297 298 299
		if (evlist->enabled) {
			if (t->zero) {
				hists__delete_entries(hists);
			} else {
				hists__decay_entries(hists, t->hide_user_symbols,
						     t->hide_kernel_symbols);
			}
		}

300 301 302
		hists__collapse_resort(hists, NULL);

		/* Non-group events are considered as leader */
303
		if (symbol_conf.event_group && !evsel__is_group_leader(pos)) {
304
			struct hists *leader_hists = evsel__hists(evsel__leader(pos));
305 306 307 308 309 310 311

			hists__match(leader_hists, hists);
			hists__link(leader_hists, hists);
		}
	}

	evlist__for_each_entry(evlist, pos) {
312
		evsel__output_resort(pos, NULL);
313 314 315
	}
}

316
static void perf_top__print_sym_table(struct perf_top *top)
317
{
318 319
	char bf[160];
	int printed = 0;
320
	const int win_width = top->winsize.ws_col - 1;
321
	struct evsel *evsel = top->sym_evsel;
322
	struct hists *hists = evsel__hists(evsel);
323

324
	puts(CONSOLE_CLEAR);
325

326
	perf_top__header_snprintf(top, bf, sizeof(bf));
327
	printf("%s\n", bf);
328

329
	printf("%-*.*s\n", win_width, win_width, graph_dotted_line);
330

331
	if (!top->record_opts.overwrite &&
332 333 334 335
	    (top->evlist->stats.nr_lost_warned !=
	     top->evlist->stats.nr_events[PERF_RECORD_LOST])) {
		top->evlist->stats.nr_lost_warned =
			      top->evlist->stats.nr_events[PERF_RECORD_LOST];
336 337
		color_fprintf(stdout, PERF_COLOR_RED,
			      "WARNING: LOST %d chunks, Check IO/CPU overload",
338
			      top->evlist->stats.nr_lost_warned);
339
		++printed;
340 341
	}

342 343
	if (top->sym_filter_entry) {
		perf_top__show_details(top);
344 345 346
		return;
	}

347
	perf_top__resort_hists(top);
N
Namhyung Kim 已提交
348

349
	hists__output_recalc_col_len(hists, top->print_entries - printed);
350
	putchar('\n');
351
	hists__fprintf(hists, false, top->print_entries - printed, win_width,
352
		       top->min_percent, stdout, !symbol_conf.use_callchain);
353 354
}

355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389
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;
}

390
static void perf_top__prompt_symbol(struct perf_top *top, const char *msg)
391 392
{
	char *buf = malloc(0), *p;
393
	struct hist_entry *syme = top->sym_filter_entry, *n, *found = NULL;
394
	struct hists *hists = evsel__hists(top->sym_evsel);
395
	struct rb_node *next;
396 397 398 399 400
	size_t dummy = 0;

	/* zero counters of active symbol */
	if (syme) {
		__zero_source_counters(syme);
401
		top->sym_filter_entry = NULL;
402 403 404 405 406 407 408 409 410 411
	}

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

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

D
Davidlohr Bueso 已提交
412
	next = rb_first_cached(&hists->entries);
413 414 415 416
	while (next) {
		n = rb_entry(next, struct hist_entry, rb_node);
		if (n->ms.sym && !strcmp(buf, n->ms.sym->name)) {
			found = n;
417 418
			break;
		}
419
		next = rb_next(&n->rb_node);
420 421 422
	}

	if (!found) {
423
		fprintf(stderr, "Sorry, %s is not active.\n", buf);
424 425
		sleep(1);
	} else
426
		perf_top__parse_source(top, found);
427 428 429 430 431

out_free:
	free(buf);
}

432
static void perf_top__print_mapped_keys(struct perf_top *top)
433
{
434 435
	char *name = NULL;

436 437
	if (top->sym_filter_entry) {
		struct symbol *sym = top->sym_filter_entry->ms.sym;
438 439 440 441
		name = sym->name;
	}

	fprintf(stdout, "\nMapped keys:\n");
442 443
	fprintf(stdout, "\t[d]     display refresh delay.             \t(%d)\n", top->delay_secs);
	fprintf(stdout, "\t[e]     display entries (lines).           \t(%d)\n", top->print_entries);
444

445
	if (top->evlist->core.nr_entries > 1)
446
		fprintf(stdout, "\t[E]     active event counter.              \t(%s)\n", evsel__name(top->sym_evsel));
447

448
	fprintf(stdout, "\t[f]     profile display filter (count).    \t(%d)\n", top->count_filter);
449

450
	fprintf(stdout, "\t[F]     annotate display filter (percent). \t(%d%%)\n", top->annotation_opts.min_pcnt);
451 452
	fprintf(stdout, "\t[s]     annotate symbol.                   \t(%s)\n", name?: "NULL");
	fprintf(stdout, "\t[S]     stop annotation.\n");
453

454
	fprintf(stdout,
S
Sihyeon Jang 已提交
455
		"\t[K]     hide kernel symbols.             \t(%s)\n",
456
		top->hide_kernel_symbols ? "yes" : "no");
457 458
	fprintf(stdout,
		"\t[U]     hide user symbols.               \t(%s)\n",
459 460
		top->hide_user_symbols ? "yes" : "no");
	fprintf(stdout, "\t[z]     toggle sample zeroing.             \t(%d)\n", top->zero ? 1 : 0);
461 462 463
	fprintf(stdout, "\t[qQ]    quit.\n");
}

464
static int perf_top__key_mapped(struct perf_top *top, int c)
465 466 467 468 469 470 471 472
{
	switch (c) {
		case 'd':
		case 'e':
		case 'f':
		case 'z':
		case 'q':
		case 'Q':
473 474
		case 'K':
		case 'U':
475 476 477
		case 'F':
		case 's':
		case 'S':
478 479
			return 1;
		case 'E':
480
			return top->evlist->core.nr_entries > 1 ? 1 : 0;
481 482
		default:
			break;
483 484 485
	}

	return 0;
486 487
}

488
static bool perf_top__handle_keypress(struct perf_top *top, int c)
489
{
490 491
	bool ret = true;

492
	if (!perf_top__key_mapped(top, c)) {
493
		struct pollfd stdin_poll = { .fd = 0, .events = POLLIN };
494
		struct termios save;
495

496
		perf_top__print_mapped_keys(top);
497 498 499
		fprintf(stdout, "\nEnter selection, or unmapped key to continue: ");
		fflush(stdout);

500
		set_term_quiet_input(&save);
501 502 503 504 505

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

		tcsetattr(0, TCSAFLUSH, &save);
506
		if (!perf_top__key_mapped(top, c))
507
			return ret;
508 509
	}

510 511
	switch (c) {
		case 'd':
512 513 514
			prompt_integer(&top->delay_secs, "Enter display delay");
			if (top->delay_secs < 1)
				top->delay_secs = 1;
515 516
			break;
		case 'e':
517 518
			prompt_integer(&top->print_entries, "Enter display entries (lines)");
			if (top->print_entries == 0) {
519
				perf_top__resize(top);
520
				signal(SIGWINCH, winch_sig);
521
			} else {
522
				signal(SIGWINCH, SIG_DFL);
523
			}
524 525
			break;
		case 'E':
526
			if (top->evlist->core.nr_entries > 1) {
527 528 529
				/* Select 0 as the default event: */
				int counter = 0;

530
				fprintf(stderr, "\nAvailable events:");
531

532
				evlist__for_each_entry(top->evlist, top->sym_evsel)
533
					fprintf(stderr, "\n\t%d %s", top->sym_evsel->core.idx, evsel__name(top->sym_evsel));
534

535
				prompt_integer(&counter, "Enter details event counter");
536

537
				if (counter >= top->evlist->core.nr_entries) {
538
					top->sym_evsel = evlist__first(top->evlist);
539
					fprintf(stderr, "Sorry, no such event, using %s.\n", evsel__name(top->sym_evsel));
540
					sleep(1);
541
					break;
542
				}
543
				evlist__for_each_entry(top->evlist, top->sym_evsel)
544
					if (top->sym_evsel->core.idx == counter)
545
						break;
546
			} else
547
				top->sym_evsel = evlist__first(top->evlist);
548 549
			break;
		case 'f':
550
			prompt_integer(&top->count_filter, "Enter display event count filter");
551 552
			break;
		case 'F':
553
			prompt_percent(&top->annotation_opts.min_pcnt,
554
				       "Enter details display event filter (percent)");
555
			break;
556
		case 'K':
557
			top->hide_kernel_symbols = !top->hide_kernel_symbols;
558
			break;
559 560 561
		case 'q':
		case 'Q':
			printf("exiting.\n");
562 563
			if (top->dump_symtab)
				perf_session__fprintf_dsos(top->session, stderr);
564 565
			ret = false;
			break;
566
		case 's':
567
			perf_top__prompt_symbol(top, "Enter details symbol");
568 569
			break;
		case 'S':
570
			if (!top->sym_filter_entry)
571 572
				break;
			else {
573
				struct hist_entry *syme = top->sym_filter_entry;
574

575
				top->sym_filter_entry = NULL;
576 577 578
				__zero_source_counters(syme);
			}
			break;
579
		case 'U':
580
			top->hide_user_symbols = !top->hide_user_symbols;
581
			break;
582
		case 'z':
583
			top->zero = !top->zero;
584
			break;
585 586
		default:
			break;
587
	}
588 589

	return ret;
590 591
}

592 593 594
static void perf_top__sort_new_samples(void *arg)
{
	struct perf_top *t = arg;
595

596 597 598
	if (t->evlist->selected != NULL)
		t->sym_evsel = t->evlist->selected;

599
	perf_top__resort_hists(t);
600

601
	if (t->lost || t->drop)
602
		pr_warning("Too slow to read ring buffer (change period (-c/-F) or limit CPUs (-C)\n");
603 604
}

605 606 607 608 609 610
static void stop_top(void)
{
	session_done = 1;
	done = 1;
}

611
static void *display_thread_tui(void *arg)
612
{
613
	struct evsel *pos;
614
	struct perf_top *top = arg;
615
	const char *help = "For a higher level overview, try: perf top --sort comm,dso";
616 617 618 619 620
	struct hist_browser_timer hbt = {
		.timer		= perf_top__sort_new_samples,
		.arg		= top,
		.refresh	= top->delay_secs,
	};
621
	int ret;
622

623 624 625 626 627 628 629
	/* In order to read symbols from other namespaces perf to  needs to call
	 * setns(2).  This isn't permitted if the struct_fs has multiple users.
	 * unshare(2) the fs so that we may continue to setns into namespaces
	 * that we're observing.
	 */
	unshare(CLONE_FS);

630 631
	prctl(PR_SET_NAME, "perf-top-UI", 0, 0, 0);

632
repeat:
633
	perf_top__sort_new_samples(top);
634 635 636

	/*
	 * Initialize the uid_filter_str, in the future the TUI will allow
637
	 * Zooming in/out UIDs. For now just use whatever the user passed
638 639
	 * via --uid.
	 */
640
	evlist__for_each_entry(top->evlist, pos) {
641 642 643
		struct hists *hists = evsel__hists(pos);
		hists->uid_filter_str = top->record_opts.target.uid_str;
	}
644

645 646 647
	ret = evlist__tui_browse_hists(top->evlist, help, &hbt, top->min_percent,
				       &top->session->header.env, !top->record_opts.overwrite,
				       &top->annotation_opts);
648 649 650 651 652 653
	if (ret == K_RELOAD) {
		top->zero = true;
		goto repeat;
	} else
		stop_top();

654 655 656
	return NULL;
}

657 658
static void display_sig(int sig __maybe_unused)
{
659
	stop_top();
660 661 662 663
}

static void display_setup_sig(void)
{
664 665
	signal(SIGSEGV, sighandler_dump_stack);
	signal(SIGFPE, sighandler_dump_stack);
666 667 668 669 670
	signal(SIGINT,  display_sig);
	signal(SIGQUIT, display_sig);
	signal(SIGTERM, display_sig);
}

671
static void *display_thread(void *arg)
672
{
673
	struct pollfd stdin_poll = { .fd = 0, .events = POLLIN };
674
	struct termios save;
675
	struct perf_top *top = arg;
676 677
	int delay_msecs, c;

678 679 680 681 682 683 684
	/* In order to read symbols from other namespaces perf to  needs to call
	 * setns(2).  This isn't permitted if the struct_fs has multiple users.
	 * unshare(2) the fs so that we may continue to setns into namespaces
	 * that we're observing.
	 */
	unshare(CLONE_FS);

685 686
	prctl(PR_SET_NAME, "perf-top-UI", 0, 0, 0);

687
	display_setup_sig();
688
	pthread__unblock_sigwinch();
689
repeat:
690
	delay_msecs = top->delay_secs * MSEC_PER_SEC;
691
	set_term_quiet_input(&save);
692
	/* trash return*/
693 694 695
	clearerr(stdin);
	if (poll(&stdin_poll, 1, 0) > 0)
		getc(stdin);
696

697
	while (!done) {
698
		perf_top__print_sym_table(top);
699 700 701 702 703 704 705 706 707 708
		/*
		 * Either timeout expired or we got an EINTR due to SIGWINCH,
		 * refresh screen in both cases.
		 */
		switch (poll(&stdin_poll, 1, delay_msecs)) {
		case 0:
			continue;
		case -1:
			if (errno == EINTR)
				continue;
709
			__fallthrough;
710
		default:
711 712 713 714 715
			c = getc(stdin);
			tcsetattr(0, TCSAFLUSH, &save);

			if (perf_top__handle_keypress(top, c))
				goto repeat;
716
			stop_top();
717 718
		}
	}
719

720
	tcsetattr(0, TCSAFLUSH, &save);
721 722 723
	return NULL;
}

724 725 726 727 728 729
static int hist_iter__top_callback(struct hist_entry_iter *iter,
				   struct addr_location *al, bool single,
				   void *arg)
{
	struct perf_top *top = arg;
	struct hist_entry *he = iter->he;
730
	struct evsel *evsel = iter->evsel;
731

732
	if (perf_hpp_list.sym && single)
733
		perf_top__record_precise_ip(top, he, iter->sample, evsel, al->addr);
734

735
	hist__account_cycles(iter->sample->branch_stack, al, iter->sample,
736 737
		     !(top->record_opts.branch_stack & PERF_SAMPLE_BRANCH_ANY),
		     NULL);
738 739 740
	return 0;
}

741 742
static void perf_event__process_sample(struct perf_tool *tool,
				       const union perf_event *event,
743
				       struct evsel *evsel,
744
				       struct perf_sample *sample,
745
				       struct machine *machine)
746
{
747
	struct perf_top *top = container_of(tool, struct perf_top, tool);
748
	struct addr_location al;
749
	int err;
750

751
	if (!machine && perf_guest) {
752 753 754
		static struct intlist *seen;

		if (!seen)
755
			seen = intlist__new(NULL);
756

757
		if (!intlist__has_entry(seen, sample->pid)) {
758
			pr_err("Can't find guest [%d]'s kernel information\n",
759 760
				sample->pid);
			intlist__add(seen, sample->pid);
761
		}
762 763 764
		return;
	}

765
	if (!machine) {
766
		pr_err("%u unprocessable samples recorded.\r",
767
		       top->session->evlist->stats.nr_unprocessable_samples++);
768 769 770
		return;
	}

771
	if (event->header.misc & PERF_RECORD_MISC_EXACT_IP)
772
		top->exact_samples++;
773

774
	if (machine__resolve(machine, &al, sample) < 0)
775
		return;
776

777 778 779
	if (top->stitch_lbr)
		al.thread->lbr_stitch_enable = true;

780
	if (!machine->kptr_restrict_warned &&
781 782
	    symbol_conf.kptr_restrict &&
	    al.cpumode == PERF_RECORD_MISC_KERNEL) {
783
		if (!evlist__exclude_kernel(top->session->evlist)) {
784
			ui__warning(
785
"Kernel address maps (/proc/{kallsyms,modules}) are restricted.\n\n"
786
"Check /proc/sys/kernel/kptr_restrict and /proc/sys/kernel/perf_event_paranoid.\n\n"
787
"Kernel%s samples will not be resolved.\n",
788
			  al.map && map__has_symbols(al.map) ?
789
			  " modules" : "");
790 791 792
			if (use_browser <= 0)
				sleep(5);
		}
793
		machine->kptr_restrict_warned = true;
794 795
	}

796
	if (al.sym == NULL && al.map != NULL) {
797
		const char *msg = "Kernel samples will not be resolved.\n";
798 799 800 801 802 803 804 805 806 807 808
		/*
		 * As we do lazy loading of symtabs we only will know if the
		 * specified vmlinux file is invalid when we actually have a
		 * hit in kernel space and then try to load it. So if we get
		 * here and there are _no_ symbols in the DSO backing the
		 * kernel map, bail out.
		 *
		 * We may never get here, for instance, if we use -K/
		 * --hide-kernel-symbols, even if the user specifies an
		 * invalid --vmlinux ;-)
		 */
809
		if (!machine->kptr_restrict_warned && !top->vmlinux_warned &&
810
		    __map__is_kernel(al.map) && map__has_symbols(al.map)) {
811
			if (symbol_conf.vmlinux_name) {
812 813 814 815
				char serr[256];
				dso__strerror_load(al.map->dso, serr, sizeof(serr));
				ui__warning("The %s file can't be used: %s\n%s",
					    symbol_conf.vmlinux_name, serr, msg);
816 817 818 819 820 821 822
			} else {
				ui__warning("A vmlinux file was not found.\n%s",
					    msg);
			}

			if (use_browser <= 0)
				sleep(5);
823
			top->vmlinux_warned = true;
824
		}
825 826
	}

827
	if (al.sym == NULL || !al.sym->idle) {
828
		struct hists *hists = evsel__hists(evsel);
829
		struct hist_entry_iter iter = {
830 831 832
			.evsel		= evsel,
			.sample 	= sample,
			.add_entry_cb 	= hist_iter__top_callback,
833
		};
834

835 836 837 838
		if (symbol_conf.cumulate_callchain)
			iter.ops = &hist_iter_cumulative;
		else
			iter.ops = &hist_iter_normal;
839

840
		pthread_mutex_lock(&hists->lock);
841

842
		err = hist_entry_iter__add(&iter, &al, top->max_stack, top);
843 844
		if (err < 0)
			pr_err("Problem incrementing symbol period, skipping event\n");
845

846
		pthread_mutex_unlock(&hists->lock);
847
	}
848

849
	addr_location__put(&al);
850 851
}

852 853
static void
perf_top__process_lost(struct perf_top *top, union perf_event *event,
854
		       struct evsel *evsel)
855 856 857
{
	top->lost += event->lost.lost;
	top->lost_total += event->lost.lost;
858
	evsel->evlist->stats.total_lost += event->lost.lost;
859 860 861 862 863
}

static void
perf_top__process_lost_samples(struct perf_top *top,
			       union perf_event *event,
864
			       struct evsel *evsel)
865 866 867
{
	top->lost += event->lost_samples.lost;
	top->lost_total += event->lost_samples.lost;
868
	evsel->evlist->stats.total_lost_samples += event->lost_samples.lost;
869 870
}

871 872
static u64 last_timestamp;

873
static void perf_top__mmap_read_idx(struct perf_top *top, int idx)
874
{
875
	struct record_opts *opts = &top->record_opts;
876
	struct evlist *evlist = top->evlist;
877
	struct mmap *md;
878
	union perf_event *event;
879

880
	md = opts->overwrite ? &evlist->overwrite_mmap[idx] : &evlist->mmap[idx];
881
	if (perf_mmap__read_init(&md->core) < 0)
882 883
		return;

884
	while ((event = perf_mmap__read_event(&md->core)) != NULL) {
J
Jiri Olsa 已提交
885
		int ret;
886

887
		ret = evlist__parse_sample_timestamp(evlist, event, &last_timestamp);
J
Jiri Olsa 已提交
888
		if (ret && ret != -1)
889 890
			break;

891
		ret = ordered_events__queue(top->qe.in, event, last_timestamp, 0);
J
Jiri Olsa 已提交
892 893
		if (ret)
			break;
894

895
		perf_mmap__consume(&md->core);
896 897 898 899 900 901 902

		if (top->qe.rotate) {
			pthread_mutex_lock(&top->qe.mutex);
			top->qe.rotate = false;
			pthread_cond_signal(&top->qe.cond);
			pthread_mutex_unlock(&top->qe.mutex);
		}
903
	}
904

905
	perf_mmap__read_done(&md->core);
906 907
}

908
static void perf_top__mmap_read(struct perf_top *top)
909
{
910
	bool overwrite = top->record_opts.overwrite;
911
	struct evlist *evlist = top->evlist;
912 913
	int i;

914
	if (overwrite)
915
		evlist__toggle_bkw_mmap(evlist, BKW_MMAP_DATA_PENDING);
916

917
	for (i = 0; i < top->evlist->core.nr_mmaps; i++)
918
		perf_top__mmap_read_idx(top, i);
919 920

	if (overwrite) {
921 922
		evlist__toggle_bkw_mmap(evlist, BKW_MMAP_EMPTY);
		evlist__toggle_bkw_mmap(evlist, BKW_MMAP_RUNNING);
923
	}
924 925
}

926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945
/*
 * Check per-event overwrite term.
 * perf top should support consistent term for all events.
 * - All events don't have per-event term
 *   E.g. "cpu/cpu-cycles/,cpu/instructions/"
 *   Nothing change, return 0.
 * - All events have same per-event term
 *   E.g. "cpu/cpu-cycles,no-overwrite/,cpu/instructions,no-overwrite/
 *   Using the per-event setting to replace the opts->overwrite if
 *   they are different, then return 0.
 * - Events have different per-event term
 *   E.g. "cpu/cpu-cycles,overwrite/,cpu/instructions,no-overwrite/"
 *   Return -1
 * - Some of the event set per-event term, but some not.
 *   E.g. "cpu/cpu-cycles/,cpu/instructions,no-overwrite/"
 *   Return -1
 */
static int perf_top__overwrite_check(struct perf_top *top)
{
	struct record_opts *opts = &top->record_opts;
946
	struct evlist *evlist = top->evlist;
947
	struct evsel_config_term *term;
948
	struct list_head *config_terms;
949
	struct evsel *evsel;
950 951 952 953 954 955
	int set, overwrite = -1;

	evlist__for_each_entry(evlist, evsel) {
		set = -1;
		config_terms = &evsel->config_terms;
		list_for_each_entry(term, config_terms, list) {
956
			if (term->type == EVSEL__CONFIG_TERM_OVERWRITE)
957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974
				set = term->val.overwrite ? 1 : 0;
		}

		/* no term for current and previous event (likely) */
		if ((overwrite < 0) && (set < 0))
			continue;

		/* has term for both current and previous event, compare */
		if ((overwrite >= 0) && (set >= 0) && (overwrite != set))
			return -1;

		/* no term for current event but has term for previous one */
		if ((overwrite >= 0) && (set < 0))
			return -1;

		/* has term for current event */
		if ((overwrite < 0) && (set >= 0)) {
			/* if it's first event, set overwrite */
975
			if (evsel == evlist__first(evlist))
976 977 978 979 980 981 982 983 984 985 986 987
				overwrite = set;
			else
				return -1;
		}
	}

	if ((overwrite >= 0) && (opts->overwrite != overwrite))
		opts->overwrite = overwrite;

	return 0;
}

K
Kan Liang 已提交
988
static int perf_top_overwrite_fallback(struct perf_top *top,
989
				       struct evsel *evsel)
K
Kan Liang 已提交
990 991
{
	struct record_opts *opts = &top->record_opts;
992
	struct evlist *evlist = top->evlist;
993
	struct evsel *counter;
K
Kan Liang 已提交
994 995 996 997 998

	if (!opts->overwrite)
		return 0;

	/* only fall back when first event fails */
999
	if (evsel != evlist__first(evlist))
K
Kan Liang 已提交
1000 1001 1002
		return 0;

	evlist__for_each_entry(evlist, counter)
1003
		counter->core.attr.write_backward = false;
K
Kan Liang 已提交
1004
	opts->overwrite = false;
1005
	pr_debug2("fall back to non-overwrite mode\n");
K
Kan Liang 已提交
1006 1007 1008
	return 1;
}

1009
static int perf_top__start_counters(struct perf_top *top)
1010
{
1011
	char msg[BUFSIZ];
1012
	struct evsel *counter;
1013
	struct evlist *evlist = top->evlist;
1014
	struct record_opts *opts = &top->record_opts;
1015

1016 1017 1018 1019 1020 1021
	if (perf_top__overwrite_check(top)) {
		ui__error("perf top only support consistent per-event "
			  "overwrite setting for all events\n");
		goto out_err;
	}

1022
	evlist__config(evlist, opts, &callchain_param);
1023

1024
	evlist__for_each_entry(evlist, counter) {
1025
try_again:
1026
		if (evsel__open(counter, top->evlist->core.cpus,
1027
				     top->evlist->core.threads) < 0) {
K
Kan Liang 已提交
1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042

			/*
			 * Specially handle overwrite fall back.
			 * Because perf top is the only tool which has
			 * overwrite mode by default, support
			 * both overwrite and non-overwrite mode, and
			 * require consistent mode for all events.
			 *
			 * May move it to generic code with more tools
			 * have similar attribute.
			 */
			if (perf_missing_features.write_backward &&
			    perf_top_overwrite_fallback(top, counter))
				goto try_again;

1043
			if (evsel__fallback(counter, errno, msg, sizeof(msg))) {
1044
				if (verbose > 0)
1045
					ui__warning("%s\n", msg);
1046 1047
				goto try_again;
			}
1048

1049
			evsel__open_strerror(counter, &opts->target, errno, msg, sizeof(msg));
1050
			ui__error("%s\n", msg);
1051
			goto out_err;
1052
		}
1053
	}
1054

1055
	if (evlist__mmap(evlist, opts->mmap_pages) < 0) {
1056
		ui__error("Failed to mmap with %d (%s)\n",
1057
			    errno, str_error_r(errno, msg, sizeof(msg)));
1058 1059 1060
		goto out_err;
	}

1061
	return 0;
1062 1063

out_err:
1064
	return -1;
1065 1066
}

1067
static int callchain_param__setup_sample_type(struct callchain_param *callchain)
1068
{
1069
	if (callchain->mode != CHAIN_NONE) {
1070
		if (callchain_register_param(callchain) < 0) {
1071
			ui__error("Can't register callchain params.\n");
1072 1073 1074 1075 1076 1077 1078
			return -EINVAL;
		}
	}

	return 0;
}

J
Jiri Olsa 已提交
1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103
static struct ordered_events *rotate_queues(struct perf_top *top)
{
	struct ordered_events *in = top->qe.in;

	if (top->qe.in == &top->qe.data[1])
		top->qe.in = &top->qe.data[0];
	else
		top->qe.in = &top->qe.data[1];

	return in;
}

static void *process_thread(void *arg)
{
	struct perf_top *top = arg;

	while (!done) {
		struct ordered_events *out, *in = top->qe.in;

		if (!in->nr_events) {
			usleep(100);
			continue;
		}

		out = rotate_queues(top);
1104 1105 1106 1107 1108

		pthread_mutex_lock(&top->qe.mutex);
		top->qe.rotate = true;
		pthread_cond_wait(&top->qe.cond, &top->qe.mutex);
		pthread_mutex_unlock(&top->qe.mutex);
J
Jiri Olsa 已提交
1109 1110 1111 1112 1113 1114 1115 1116

		if (ordered_events__flush(out, OE_FLUSH__TOP))
			pr_err("failed to process events\n");
	}

	return NULL;
}

1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131
/*
 * Allow only 'top->delay_secs' seconds behind samples.
 */
static int should_drop(struct ordered_event *qevent, struct perf_top *top)
{
	union perf_event *event = qevent->event;
	u64 delay_timestamp;

	if (event->header.type != PERF_RECORD_SAMPLE)
		return false;

	delay_timestamp = qevent->timestamp + top->delay_secs * NSEC_PER_SEC;
	return delay_timestamp < last_timestamp;
}

J
Jiri Olsa 已提交
1132 1133 1134 1135
static int deliver_event(struct ordered_events *qe,
			 struct ordered_event *qevent)
{
	struct perf_top *top = qe->data;
1136
	struct evlist *evlist = top->evlist;
J
Jiri Olsa 已提交
1137 1138 1139
	struct perf_session *session = top->session;
	union perf_event *event = qevent->event;
	struct perf_sample sample;
1140
	struct evsel *evsel;
J
Jiri Olsa 已提交
1141 1142 1143
	struct machine *machine;
	int ret = -1;

1144 1145 1146
	if (should_drop(qevent, top)) {
		top->drop++;
		top->drop_total++;
1147
		return 0;
1148
	}
1149

1150
	ret = evlist__parse_sample(evlist, event, &sample);
J
Jiri Olsa 已提交
1151 1152 1153 1154 1155
	if (ret) {
		pr_err("Can't parse sample, err = %d\n", ret);
		goto next_event;
	}

1156
	evsel = evlist__id2evsel(session->evlist, sample.id);
J
Jiri Olsa 已提交
1157 1158
	assert(evsel != NULL);

1159 1160 1161
	if (event->header.type == PERF_RECORD_SAMPLE) {
		if (evswitch__discard(&top->evswitch, evsel))
			return 0;
J
Jiri Olsa 已提交
1162
		++top->samples;
1163
	}
J
Jiri Olsa 已提交
1164 1165 1166 1167 1168 1169 1170 1171 1172 1173 1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200 1201 1202 1203 1204

	switch (sample.cpumode) {
	case PERF_RECORD_MISC_USER:
		++top->us_samples;
		if (top->hide_user_symbols)
			goto next_event;
		machine = &session->machines.host;
		break;
	case PERF_RECORD_MISC_KERNEL:
		++top->kernel_samples;
		if (top->hide_kernel_symbols)
			goto next_event;
		machine = &session->machines.host;
		break;
	case PERF_RECORD_MISC_GUEST_KERNEL:
		++top->guest_kernel_samples;
		machine = perf_session__find_machine(session,
						     sample.pid);
		break;
	case PERF_RECORD_MISC_GUEST_USER:
		++top->guest_us_samples;
		/*
		 * TODO: we don't process guest user from host side
		 * except simple counting.
		 */
		goto next_event;
	default:
		if (event->header.type == PERF_RECORD_SAMPLE)
			goto next_event;
		machine = &session->machines.host;
		break;
	}

	if (event->header.type == PERF_RECORD_SAMPLE) {
		perf_event__process_sample(&top->tool, event, evsel,
					   &sample, machine);
	} else if (event->header.type == PERF_RECORD_LOST) {
		perf_top__process_lost(top, event, evsel);
	} else if (event->header.type == PERF_RECORD_LOST_SAMPLES) {
		perf_top__process_lost_samples(top, event, evsel);
	} else if (event->header.type < PERF_RECORD_MAX) {
1205
		events_stats__inc(&session->evlist->stats, event->header.type);
J
Jiri Olsa 已提交
1206 1207 1208 1209 1210 1211 1212 1213 1214 1215 1216 1217 1218 1219 1220 1221
		machine__process_event(machine, event, &sample);
	} else
		++session->evlist->stats.nr_unknown_events;

	ret = 0;
next_event:
	return ret;
}

static void init_process_thread(struct perf_top *top)
{
	ordered_events__init(&top->qe.data[0], deliver_event, top);
	ordered_events__init(&top->qe.data[1], deliver_event, top);
	ordered_events__set_copy_on_queue(&top->qe.data[0], true);
	ordered_events__set_copy_on_queue(&top->qe.data[1], true);
	top->qe.in = &top->qe.data[0];
1222 1223
	pthread_mutex_init(&top->qe.mutex, NULL);
	pthread_cond_init(&top->qe.cond, NULL);
J
Jiri Olsa 已提交
1224 1225
}

1226
static int __cmd_top(struct perf_top *top)
1227
{
1228
	struct record_opts *opts = &top->record_opts;
J
Jiri Olsa 已提交
1229
	pthread_t thread, thread_process;
1230
	int ret;
1231

1232 1233 1234
	if (!top->annotation_opts.objdump_path) {
		ret = perf_env__lookup_objdump(&top->session->header.env,
					       &top->annotation_opts.objdump_path);
1235
		if (ret)
1236
			return ret;
1237 1238
	}

1239
	ret = callchain_param__setup_sample_type(&callchain_param);
1240
	if (ret)
1241
		return ret;
1242

1243
	if (perf_session__register_idle_thread(top->session) < 0)
1244
		return ret;
N
Namhyung Kim 已提交
1245

1246 1247
	if (top->nr_threads_synthesize > 1)
		perf_set_multithreaded();
1248

J
Jiri Olsa 已提交
1249 1250
	init_process_thread(top);

N
Namhyung Kim 已提交
1251 1252
	if (opts->record_namespaces)
		top->tool.namespace_events = true;
1253 1254 1255 1256 1257 1258 1259 1260
	if (opts->record_cgroup) {
#ifdef HAVE_FILE_HANDLE
		top->tool.cgroup_events = true;
#else
		pr_err("cgroup tracking is not supported.\n");
		return -1;
#endif
	}
N
Namhyung Kim 已提交
1261

1262
	ret = perf_event__synthesize_bpf_events(top->session, perf_event__process,
1263 1264 1265
						&top->session->machines.host,
						&top->record_opts);
	if (ret < 0)
1266
		pr_debug("Couldn't synthesize BPF events: Pre-existing BPF programs won't have symbols resolved.\n");
1267

1268 1269 1270 1271 1272
	ret = perf_event__synthesize_cgroups(&top->tool, perf_event__process,
					     &top->session->machines.host);
	if (ret < 0)
		pr_debug("Couldn't synthesize cgroup events.\n");

1273
	machine__synthesize_threads(&top->session->machines.host, &opts->target,
1274
				    top->evlist->core.threads, false,
1275
				    top->nr_threads_synthesize);
1276

1277 1278
	if (top->nr_threads_synthesize > 1)
		perf_set_singlethreaded();
1279

1280
	if (perf_hpp_list.socket) {
1281
		ret = perf_env__read_cpu_topology_map(&perf_env);
1282 1283 1284 1285 1286 1287 1288
		if (ret < 0) {
			char errbuf[BUFSIZ];
			const char *err = str_error_r(-ret, errbuf, sizeof(errbuf));

			ui__error("Could not read the CPU topology map: %s\n", err);
			return ret;
		}
1289 1290
	}

1291 1292
	ret = perf_top__start_counters(top);
	if (ret)
1293
		return ret;
1294

1295
	top->session->evlist = top->evlist;
1296
	perf_session__set_id_hdr_size(top->session);
1297

1298 1299 1300 1301 1302 1303 1304 1305
	/*
	 * When perf is starting the traced process, all the events (apart from
	 * group members) have enable_on_exec=1 set, so don't spoil it by
	 * prematurely enabling them.
	 *
	 * XXX 'top' still doesn't start workloads like record, trace, but should,
	 * so leave the check here.
	 */
1306
        if (!target__none(&opts->target))
1307
		evlist__enable(top->evlist);
1308

1309
	ret = -1;
J
Jiri Olsa 已提交
1310 1311
	if (pthread_create(&thread_process, NULL, process_thread, top)) {
		ui__error("Could not create process thread.\n");
1312
		return ret;
J
Jiri Olsa 已提交
1313 1314
	}

1315
	if (pthread_create(&thread, NULL, (use_browser > 0 ? display_thread_tui :
1316
							    display_thread), top)) {
1317
		ui__error("Could not create display thread.\n");
J
Jiri Olsa 已提交
1318
		goto out_join_thread;
1319 1320
	}

1321
	if (top->realtime_prio) {
1322 1323
		struct sched_param param;

1324
		param.sched_priority = top->realtime_prio;
1325
		if (sched_setscheduler(0, SCHED_FIFO, &param)) {
1326
			ui__error("Could not set realtime priority.\n");
1327
			goto out_join;
1328 1329 1330
		}
	}

1331
	/* Wait for a minimal set of events before starting the snapshot */
1332
	evlist__poll(top->evlist, 100);
1333 1334 1335

	perf_top__mmap_read(top);

1336
	while (!done) {
1337
		u64 hits = top->samples;
1338

1339
		perf_top__mmap_read(top);
1340

1341
		if (opts->overwrite || (hits == top->samples))
1342
			ret = evlist__poll(top->evlist, 100);
1343 1344 1345 1346 1347

		if (resize) {
			perf_top__resize(top);
			resize = 0;
		}
1348 1349
	}

1350
	ret = 0;
1351 1352
out_join:
	pthread_join(thread, NULL);
J
Jiri Olsa 已提交
1353
out_join_thread:
1354
	pthread_cond_signal(&top->qe.cond);
J
Jiri Olsa 已提交
1355
	pthread_join(thread_process, NULL);
1356
	return ret;
1357 1358 1359
}

static int
J
Jiri Olsa 已提交
1360
callchain_opt(const struct option *opt, const char *arg, int unset)
1361 1362
{
	symbol_conf.use_callchain = true;
J
Jiri Olsa 已提交
1363 1364
	return record_callchain_opt(opt, arg, unset);
}
1365

J
Jiri Olsa 已提交
1366 1367 1368
static int
parse_callchain_opt(const struct option *opt, const char *arg, int unset)
{
1369
	struct callchain_param *callchain = opt->value;
1370

1371 1372
	callchain->enabled = !unset;
	callchain->record_mode = CALLCHAIN_FP;
1373 1374 1375 1376 1377 1378

	/*
	 * --no-call-graph
	 */
	if (unset) {
		symbol_conf.use_callchain = false;
1379
		callchain->record_mode = CALLCHAIN_NONE;
1380 1381 1382 1383
		return 0;
	}

	return parse_callchain_top_opt(arg);
1384
}
1385

1386
static int perf_top_config(const char *var, const char *value, void *cb __maybe_unused)
1387
{
1388 1389 1390 1391
	if (!strcmp(var, "top.call-graph")) {
		var = "call-graph.record-mode";
		return perf_default_config(var, value, cb);
	}
1392 1393 1394 1395
	if (!strcmp(var, "top.children")) {
		symbol_conf.cumulate_callchain = perf_config_bool(var, value);
		return 0;
	}
1396

1397
	return 0;
1398 1399
}

1400 1401 1402 1403 1404 1405 1406 1407 1408 1409
static int
parse_percent_limit(const struct option *opt, const char *arg,
		    int unset __maybe_unused)
{
	struct perf_top *top = opt->value;

	top->min_percent = strtof(arg, NULL);
	return 0;
}

1410 1411
const char top_callchain_help[] = CALLCHAIN_RECORD_HELP CALLCHAIN_REPORT_HELP
	"\n\t\t\t\tDefault: fp,graph,0.5,caller,function";
1412

1413
int cmd_top(int argc, const char **argv)
1414
{
1415
	char errbuf[BUFSIZ];
1416 1417 1418
	struct perf_top top = {
		.count_filter	     = 5,
		.delay_secs	     = 2,
1419 1420 1421 1422 1423
		.record_opts = {
			.mmap_pages	= UINT_MAX,
			.user_freq	= UINT_MAX,
			.user_interval	= ULLONG_MAX,
			.freq		= 4000, /* 4 KHz */
1424
			.target		= {
1425 1426
				.uses_mmap   = true,
			},
1427 1428 1429 1430 1431 1432 1433 1434
			/*
			 * FIXME: This will lose PERF_RECORD_MMAP and other metadata
			 * when we pause, fix that and reenable. Probably using a
			 * separate evlist with a dummy event, i.e. a non-overwrite
			 * ring buffer just for metadata events, while PERF_RECORD_SAMPLE
			 * stays in overwrite mode. -acme
			 * */
			.overwrite	= 0,
J
Jiri Olsa 已提交
1435
			.sample_time	= true,
1436
			.sample_time_set = true,
N
Namhyung Kim 已提交
1437
		},
1438
		.max_stack	     = sysctl__max_stack(),
1439
		.annotation_opts     = annotation__default_options,
1440
		.nr_threads_synthesize = UINT_MAX,
1441
	};
1442
	struct record_opts *opts = &top.record_opts;
1443
	struct target *target = &opts->target;
1444
	const struct option options[] = {
1445
	OPT_CALLBACK('e', "event", &top.evlist, "event",
1446
		     "event selector. use 'perf list' to list available events",
1447
		     parse_events_option),
1448 1449
	OPT_U64('c', "count", &opts->user_interval, "event period to sample"),
	OPT_STRING('p', "pid", &target->pid, "pid",
1450
		    "profile events on existing process id"),
1451
	OPT_STRING('t', "tid", &target->tid, "tid",
1452
		    "profile events on existing thread id"),
1453
	OPT_BOOLEAN('a', "all-cpus", &target->system_wide,
1454
			    "system-wide collection from all CPUs"),
1455
	OPT_STRING('C', "cpu", &target->cpu_list, "cpu",
1456
		    "list of cpus to monitor"),
1457 1458
	OPT_STRING('k', "vmlinux", &symbol_conf.vmlinux_name,
		   "file", "vmlinux pathname"),
1459 1460
	OPT_BOOLEAN(0, "ignore-vmlinux", &symbol_conf.ignore_vmlinux,
		    "don't load vmlinux even if found"),
1461 1462
	OPT_STRING(0, "kallsyms", &symbol_conf.kallsyms_name,
		   "file", "kallsyms pathname"),
1463
	OPT_BOOLEAN('K', "hide_kernel_symbols", &top.hide_kernel_symbols,
1464
		    "hide kernel symbols"),
1465
	OPT_CALLBACK('m', "mmap-pages", &opts->mmap_pages, "pages",
1466
		     "number of mmap data pages", evlist__parse_mmap_pages),
1467
	OPT_INTEGER('r', "realtime", &top.realtime_prio,
1468
		    "collect data with this RT SCHED_FIFO priority"),
1469
	OPT_INTEGER('d', "delay", &top.delay_secs,
1470
		    "number of seconds to delay between refreshes"),
1471
	OPT_BOOLEAN('D', "dump-symtab", &top.dump_symtab,
1472
			    "dump the symbol table used for profiling"),
1473
	OPT_INTEGER('f', "count-filter", &top.count_filter,
1474
		    "only display functions with more events than this"),
1475
	OPT_BOOLEAN(0, "group", &opts->group,
1476
			    "put the counters into a counter group"),
1477 1478
	OPT_BOOLEAN('i', "no-inherit", &opts->no_inherit,
		    "child tasks do not inherit counters"),
1479
	OPT_STRING(0, "sym-annotate", &top.sym_filter, "symbol name",
1480
		    "symbol to annotate"),
1481
	OPT_BOOLEAN('z', "zero", &top.zero, "zero history across updates"),
1482 1483 1484
	OPT_CALLBACK('F', "freq", &top.record_opts, "freq or 'max'",
		     "profile at this frequency",
		      record__parse_freq),
1485
	OPT_INTEGER('E', "entries", &top.print_entries,
1486
		    "display this many functions"),
1487
	OPT_BOOLEAN('U', "hide_user_symbols", &top.hide_user_symbols,
1488
		    "hide user symbols"),
1489 1490
	OPT_BOOLEAN(0, "tui", &top.use_tui, "Use the TUI interface"),
	OPT_BOOLEAN(0, "stdio", &top.use_stdio, "Use the stdio interface"),
1491
	OPT_INCR('v', "verbose", &verbose,
1492
		    "be more verbose (show counter open errors, etc)"),
1493
	OPT_STRING('s', "sort", &sort_order, "key[,key2...]",
1494 1495
		   "sort by key(s): pid, comm, dso, symbol, parent, cpu, srcline, ..."
		   " Please refer the man page for the complete list."),
1496 1497
	OPT_STRING(0, "fields", &field_order, "key[,keys...]",
		   "output field(s): overhead, period, sample plus all of sort keys"),
1498 1499
	OPT_BOOLEAN('n', "show-nr-samples", &symbol_conf.show_nr_samples,
		    "Show a column with the number of samples"),
1500
	OPT_CALLBACK_NOOPT('g', NULL, &callchain_param,
1501
			   NULL, "enables call-graph recording and display",
J
Jiri Olsa 已提交
1502
			   &callchain_opt),
1503
	OPT_CALLBACK(0, "call-graph", &callchain_param,
1504
		     "record_mode[,record_size],print_type,threshold[,print_limit],order,sort_key[,branch]",
1505
		     top_callchain_help, &parse_callchain_opt),
N
Namhyung Kim 已提交
1506 1507
	OPT_BOOLEAN(0, "children", &symbol_conf.cumulate_callchain,
		    "Accumulate callchains of children and show total overhead as well"),
1508 1509
	OPT_INTEGER(0, "max-stack", &top.max_stack,
		    "Set the maximum stack depth when parsing the callchain. "
1510
		    "Default: kernel.perf_event_max_stack or " __stringify(PERF_MAX_STACK_DEPTH)),
1511 1512 1513
	OPT_CALLBACK(0, "ignore-callees", NULL, "regex",
		   "ignore callees of these functions in call graphs",
		   report_parse_ignore_callees_opt),
1514 1515 1516 1517 1518 1519 1520 1521
	OPT_BOOLEAN(0, "show-total-period", &symbol_conf.show_total_period,
		    "Show a column with the sum of periods"),
	OPT_STRING(0, "dsos", &symbol_conf.dso_list_str, "dso[,dso...]",
		   "only consider symbols in these dsos"),
	OPT_STRING(0, "comms", &symbol_conf.comm_list_str, "comm[,comm...]",
		   "only consider symbols in these comms"),
	OPT_STRING(0, "symbols", &symbol_conf.sym_list_str, "symbol[,symbol...]",
		   "only consider these symbols"),
1522
	OPT_BOOLEAN(0, "source", &top.annotation_opts.annotate_src,
1523
		    "Interleave source code with assembly code (default)"),
1524
	OPT_BOOLEAN(0, "asm-raw", &top.annotation_opts.show_asm_raw,
1525
		    "Display raw encoding of assembly instructions (default)"),
1526 1527
	OPT_BOOLEAN(0, "demangle-kernel", &symbol_conf.demangle_kernel,
		    "Enable kernel symbol demangling"),
S
Song Liu 已提交
1528
	OPT_BOOLEAN(0, "no-bpf-event", &top.record_opts.no_bpf_event, "do not record bpf events"),
1529
	OPT_STRING(0, "objdump", &top.annotation_opts.objdump_path, "path",
1530
		    "objdump binary to use for disassembly and annotations"),
1531
	OPT_STRING('M', "disassembler-style", &top.annotation_opts.disassembler_style, "disassembler style",
1532
		   "Specify disassembler style (e.g. -M intel for intel syntax)"),
1533 1534 1535 1536
	OPT_STRING(0, "prefix", &top.annotation_opts.prefix, "prefix",
		    "Add prefix to source file path names in programs (with --prefix-strip)"),
	OPT_STRING(0, "prefix-strip", &top.annotation_opts.prefix_strip, "N",
		    "Strip first N entries of source file path name in programs (with --prefix)"),
1537
	OPT_STRING('u', "uid", &target->uid_str, "user", "user to profile"),
1538 1539
	OPT_CALLBACK(0, "percent-limit", &top, "percent",
		     "Don't show entries under that percent", parse_percent_limit),
N
Namhyung Kim 已提交
1540 1541
	OPT_CALLBACK(0, "percentage", NULL, "relative|absolute",
		     "How to display percentage of filtered entries", parse_filter_percentage),
1542 1543 1544
	OPT_STRING('w', "column-widths", &symbol_conf.col_width_list_str,
		   "width[,width...]",
		   "don't try to adjust column width, use these fixed values"),
1545
	OPT_UINTEGER(0, "proc-map-timeout", &proc_map_timeout,
1546
			"per thread proc mmap processing timeout in ms"),
1547 1548 1549 1550 1551 1552
	OPT_CALLBACK_NOOPT('b', "branch-any", &opts->branch_stack,
		     "branch any", "sample any taken branches",
		     parse_branch_stack),
	OPT_CALLBACK('j', "branch-filter", &opts->branch_stack,
		     "branch filter mask", "branch stack filter modes",
		     parse_branch_stack),
1553 1554
	OPT_BOOLEAN(0, "raw-trace", &symbol_conf.raw_trace,
		    "Show raw trace event output (do not use print fmt or plugins)"),
N
Namhyung Kim 已提交
1555 1556
	OPT_BOOLEAN(0, "hierarchy", &symbol_conf.report_hierarchy,
		    "Show entries in a hierarchy"),
1557
	OPT_BOOLEAN(0, "overwrite", &top.record_opts.overwrite,
1558
		    "Use a backward ring buffer, default: no"),
1559
	OPT_BOOLEAN(0, "force", &symbol_conf.force, "don't complain, do it"),
1560 1561
	OPT_UINTEGER(0, "num-thread-synthesize", &top.nr_threads_synthesize,
			"number of thread to run event synthesize"),
1562 1563
	OPT_CALLBACK('G', "cgroup", &top.evlist, "name",
		     "monitor event in cgroup name only", parse_cgroups),
N
Namhyung Kim 已提交
1564 1565
	OPT_BOOLEAN(0, "namespaces", &opts->record_namespaces,
		    "Record namespaces events"),
1566 1567
	OPT_BOOLEAN(0, "all-cgroups", &opts->record_cgroup,
		    "Record cgroup events"),
1568 1569 1570 1571
	OPT_INTEGER(0, "group-sort-idx", &symbol_conf.group_sort_idx,
		    "Sort the output by the event at the index n in group. "
		    "If n is invalid, sort by the first event. "
		    "WARNING: should be used on grouped events."),
1572 1573
	OPT_BOOLEAN(0, "stitch-lbr", &top.stitch_lbr,
		    "Enable LBR callgraph stitching approach"),
1574 1575 1576 1577 1578
#ifdef HAVE_LIBPFM
	OPT_CALLBACK(0, "pfm-events", &top.evlist, "event",
		"libpfm4 event selector. use 'perf list' to list available events",
		parse_libpfm_events_option),
#endif
1579
	OPTS_EVSWITCH(&top.evswitch),
1580
	OPT_END()
1581
	};
1582 1583 1584 1585
	const char * const top_usage[] = {
		"perf top [<options>]",
		NULL
	};
1586 1587 1588 1589
	int status = hists__init();

	if (status < 0)
		return status;
1590

1591 1592 1593
	top.annotation_opts.min_pcnt = 5;
	top.annotation_opts.context  = 4;

1594
	top.evlist = evlist__new();
1595
	if (top.evlist == NULL)
1596 1597
		return -ENOMEM;

1598 1599 1600
	status = perf_config(perf_top_config, &top);
	if (status)
		return status;
1601 1602 1603 1604 1605 1606
	/*
	 * Since the per arch annotation init routine may need the cpuid, read
	 * it here, since we are not getting this from the perf.data header.
	 */
	status = perf_env__read_cpuid(&perf_env);
	if (status) {
1607 1608
		/*
		 * Some arches do not provide a get_cpuid(), so just use pr_debug, otherwise
1609
		 * warn the user explicitly.
1610 1611 1612 1613
		 */
		eprintf(status == ENOSYS ? 1 : 0, verbose,
			"Couldn't read the cpuid for this machine: %s\n",
			str_error_r(errno, errbuf, sizeof(errbuf)));
1614 1615
	}
	top.evlist->env = &perf_env;
1616

1617 1618 1619 1620
	argc = parse_options(argc, argv, options, top_usage, 0);
	if (argc)
		usage_with_options(top_usage, options);

1621 1622 1623
	if (annotate_check_args(&top.annotation_opts) < 0)
		goto out_delete_evlist;

1624
	if (!top.evlist->core.nr_entries &&
1625
	    evlist__add_default(top.evlist) < 0) {
1626 1627 1628 1629
		pr_err("Not enough memory for event selector list\n");
		goto out_delete_evlist;
	}

1630 1631 1632 1633
	status = evswitch__init(&top.evswitch, top.evlist, stderr);
	if (status)
		goto out_delete_evlist;

N
Namhyung Kim 已提交
1634 1635 1636 1637 1638 1639 1640 1641 1642 1643 1644 1645 1646
	if (symbol_conf.report_hierarchy) {
		/* disable incompatible options */
		symbol_conf.event_group = false;
		symbol_conf.cumulate_callchain = false;

		if (field_order) {
			pr_err("Error: --hierarchy and --fields options cannot be used together\n");
			parse_options_usage(top_usage, options, "fields", 0);
			parse_options_usage(NULL, options, "hierarchy", 0);
			goto out_delete_evlist;
		}
	}

1647 1648 1649 1650 1651
	if (top.stitch_lbr && !(callchain_param.record_mode == CALLCHAIN_LBR)) {
		pr_err("Error: --stitch-lbr must be used with --call-graph lbr\n");
		goto out_delete_evlist;
	}

1652 1653 1654 1655 1656
	if (nr_cgroups > 0 && opts->record_cgroup) {
		pr_err("--cgroup and --all-cgroups cannot be used together\n");
		goto out_delete_evlist;
	}

1657 1658 1659
	if (opts->branch_stack && callchain_param.enabled)
		symbol_conf.show_branchflag_count = true;

1660
	sort__mode = SORT_MODE__TOP;
1661
	/* display thread wants entries to be collapsed in a different tree */
1662
	perf_hpp_list.need_collapse = 1;
1663

1664 1665 1666 1667 1668 1669 1670
	if (top.use_stdio)
		use_browser = 0;
	else if (top.use_tui)
		use_browser = 1;

	setup_browser(false);

1671
	if (setup_sorting(top.evlist) < 0) {
1672 1673 1674 1675 1676
		if (sort_order)
			parse_options_usage(top_usage, options, "s", 1);
		if (field_order)
			parse_options_usage(sort_order ? NULL : top_usage,
					    options, "fields", 0);
1677 1678
		goto out_delete_evlist;
	}
1679

1680
	status = target__validate(target);
1681
	if (status) {
1682
		target__strerror(target, status, errbuf, BUFSIZ);
1683
		ui__warning("%s\n", errbuf);
1684 1685
	}

1686
	status = target__parse_uid(target);
1687 1688
	if (status) {
		int saved_errno = errno;
1689

1690
		target__strerror(target, status, errbuf, BUFSIZ);
1691
		ui__error("%s\n", errbuf);
1692 1693

		status = -saved_errno;
1694
		goto out_delete_evlist;
1695
	}
1696

1697
	if (target__none(target))
1698
		target->system_wide = true;
1699

1700
	if (evlist__create_maps(top.evlist, target) < 0) {
1701
		ui__error("Couldn't create thread/CPU maps: %s\n",
1702
			  errno == ENOENT ? "No such process" : str_error_r(errno, errbuf, sizeof(errbuf)));
1703
		goto out_delete_evlist;
1704
	}
1705

1706 1707
	if (top.delay_secs < 1)
		top.delay_secs = 1;
1708

1709
	if (record_opts__config(opts)) {
1710
		status = -EINVAL;
1711
		goto out_delete_evlist;
1712 1713
	}

1714
	top.sym_evsel = evlist__first(top.evlist);
1715

1716
	if (!callchain_param.enabled) {
N
Namhyung Kim 已提交
1717 1718 1719 1720
		symbol_conf.cumulate_callchain = false;
		perf_hpp__cancel_cumulate();
	}

1721 1722 1723
	if (symbol_conf.cumulate_callchain && !callchain_param.order_set)
		callchain_param.order = ORDER_CALLER;

1724 1725 1726
	status = symbol__annotation_init();
	if (status < 0)
		goto out_delete_evlist;
1727

1728
	annotation_config__init(&top.annotation_opts);
1729

1730
	symbol_conf.try_vmlinux_path = (symbol_conf.vmlinux_name == NULL);
1731 1732 1733
	status = symbol__init(NULL);
	if (status < 0)
		goto out_delete_evlist;
1734

1735
	sort__setup_elide(stdout);
1736

1737
	get_term_dimensions(&top.winsize);
1738
	if (top.print_entries == 0) {
1739
		perf_top__update_print_entries(&top);
1740
		signal(SIGWINCH, winch_sig);
1741 1742
	}

1743
	top.session = perf_session__new(NULL, NULL);
1744 1745
	if (IS_ERR(top.session)) {
		status = PTR_ERR(top.session);
1746 1747 1748
		goto out_delete_evlist;
	}

1749
#ifdef HAVE_LIBBPF_SUPPORT
1750 1751 1752 1753 1754 1755 1756 1757 1758 1759 1760 1761 1762
	if (!top.record_opts.no_bpf_event) {
		top.sb_evlist = evlist__new();

		if (top.sb_evlist == NULL) {
			pr_err("Couldn't create side band evlist.\n.");
			goto out_delete_evlist;
		}

		if (evlist__add_bpf_sb_event(top.sb_evlist, &perf_env)) {
			pr_err("Couldn't ask for PERF_RECORD_BPF_EVENT side band events.\n.");
			goto out_delete_evlist;
		}
	}
1763
#endif
1764

1765
	if (evlist__start_sb_thread(top.sb_evlist, target)) {
1766 1767 1768 1769
		pr_debug("Couldn't start the BPF side band thread:\nBPF programs starting from now on won't be annotatable\n");
		opts->no_bpf_event = true;
	}

1770
	status = __cmd_top(&top);
1771

1772
	if (!opts->no_bpf_event)
1773
		evlist__stop_sb_thread(top.sb_evlist);
1774

1775
out_delete_evlist:
1776
	evlist__delete(top.evlist);
1777
	perf_session__delete(top.session);
1778 1779

	return status;
1780
}