Linux Audio

Check our new training course

Loading...
v4.6
   1/*
   2 * builtin-top.c
   3 *
   4 * Builtin top command: Display a continuously updated profile of
   5 * any workload, CPU or specific PID.
   6 *
   7 * Copyright (C) 2008, Red Hat Inc, Ingo Molnar <mingo@redhat.com>
   8 *		 2011, Red Hat Inc, Arnaldo Carvalho de Melo <acme@redhat.com>
   9 *
  10 * Improvements and fixes by:
  11 *
  12 *   Arjan van de Ven <arjan@linux.intel.com>
  13 *   Yanmin Zhang <yanmin.zhang@intel.com>
  14 *   Wu Fengguang <fengguang.wu@intel.com>
  15 *   Mike Galbraith <efault@gmx.de>
  16 *   Paul Mackerras <paulus@samba.org>
  17 *
  18 * Released under the GPL v2. (and only v2, not any later version)
  19 */
  20#include "builtin.h"
  21
  22#include "perf.h"
  23
  24#include "util/annotate.h"
  25#include "util/cache.h"
  26#include "util/color.h"
  27#include "util/evlist.h"
  28#include "util/evsel.h"
  29#include "util/machine.h"
  30#include "util/session.h"
  31#include "util/symbol.h"
  32#include "util/thread.h"
  33#include "util/thread_map.h"
  34#include "util/top.h"
  35#include "util/util.h"
  36#include <linux/rbtree.h>
  37#include <subcmd/parse-options.h>
  38#include "util/parse-events.h"
  39#include "util/cpumap.h"
  40#include "util/xyarray.h"
  41#include "util/sort.h"
  42#include "util/intlist.h"
  43#include "util/parse-branch-options.h"
  44#include "arch/common.h"
  45
  46#include "util/debug.h"
  47
  48#include <assert.h>
  49#include <elf.h>
  50#include <fcntl.h>
  51
  52#include <stdio.h>
  53#include <termios.h>
  54#include <unistd.h>
  55#include <inttypes.h>
  56
  57#include <errno.h>
  58#include <time.h>
  59#include <sched.h>
  60
  61#include <sys/syscall.h>
  62#include <sys/ioctl.h>
  63#include <poll.h>
  64#include <sys/prctl.h>
  65#include <sys/wait.h>
  66#include <sys/uio.h>
  67#include <sys/utsname.h>
  68#include <sys/mman.h>
  69
  70#include <linux/stringify.h>
  71#include <linux/types.h>
  72
  73static volatile int done;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
  74
  75#define HEADER_LINE_NR  5
 
 
  76
  77static void perf_top__update_print_entries(struct perf_top *top)
  78{
  79	top->print_entries = top->winsize.ws_row - HEADER_LINE_NR;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
  80}
  81
  82static void perf_top__sig_winch(int sig __maybe_unused,
  83				siginfo_t *info __maybe_unused, void *arg)
  84{
  85	struct perf_top *top = arg;
 
 
 
 
  86
  87	get_term_dimensions(&top->winsize);
  88	perf_top__update_print_entries(top);
 
 
  89}
  90
  91static int perf_top__parse_source(struct perf_top *top, struct hist_entry *he)
  92{
  93	struct symbol *sym;
  94	struct annotation *notes;
  95	struct map *map;
  96	int err = -1;
  97
  98	if (!he || !he->ms.sym)
  99		return -1;
 100
 101	sym = he->ms.sym;
 102	map = he->ms.map;
 103
 104	/*
 105	 * We can't annotate with just /proc/kallsyms
 106	 */
 107	if (map->dso->symtab_type == DSO_BINARY_TYPE__KALLSYMS &&
 108	    !dso__is_kcore(map->dso)) {
 109		pr_err("Can't annotate %s: No vmlinux file was found in the "
 110		       "path\n", sym->name);
 111		sleep(1);
 112		return -1;
 113	}
 114
 115	notes = symbol__annotation(sym);
 116	if (notes->src != NULL) {
 117		pthread_mutex_lock(&notes->lock);
 118		goto out_assign;
 119	}
 120
 121	pthread_mutex_lock(&notes->lock);
 122
 123	if (symbol__alloc_hist(sym) < 0) {
 124		pthread_mutex_unlock(&notes->lock);
 125		pr_err("Not enough memory for annotating '%s' symbol!\n",
 126		       sym->name);
 127		sleep(1);
 128		return err;
 129	}
 130
 131	err = symbol__annotate(sym, map, 0);
 132	if (err == 0) {
 133out_assign:
 134		top->sym_filter_entry = he;
 135	}
 136
 137	pthread_mutex_unlock(&notes->lock);
 138	return err;
 139}
 140
 141static void __zero_source_counters(struct hist_entry *he)
 142{
 143	struct symbol *sym = he->ms.sym;
 144	symbol__annotate_zero_histograms(sym);
 145}
 146
 147static void ui__warn_map_erange(struct map *map, struct symbol *sym, u64 ip)
 148{
 149	struct utsname uts;
 150	int err = uname(&uts);
 151
 152	ui__warning("Out of bounds address found:\n\n"
 153		    "Addr:   %" PRIx64 "\n"
 154		    "DSO:    %s %c\n"
 155		    "Map:    %" PRIx64 "-%" PRIx64 "\n"
 156		    "Symbol: %" PRIx64 "-%" PRIx64 " %c %s\n"
 157		    "Arch:   %s\n"
 158		    "Kernel: %s\n"
 159		    "Tools:  %s\n\n"
 160		    "Not all samples will be on the annotation output.\n\n"
 161		    "Please report to linux-kernel@vger.kernel.org\n",
 162		    ip, map->dso->long_name, dso__symtab_origin(map->dso),
 163		    map->start, map->end, sym->start, sym->end,
 164		    sym->binding == STB_GLOBAL ? 'g' :
 165		    sym->binding == STB_LOCAL  ? 'l' : 'w', sym->name,
 166		    err ? "[unknown]" : uts.machine,
 167		    err ? "[unknown]" : uts.release, perf_version_string);
 168	if (use_browser <= 0)
 169		sleep(5);
 170
 171	map->erange_warned = true;
 172}
 173
 174static void perf_top__record_precise_ip(struct perf_top *top,
 175					struct hist_entry *he,
 176					int counter, u64 ip)
 177{
 178	struct annotation *notes;
 179	struct symbol *sym = he->ms.sym;
 180	int err = 0;
 181
 182	if (sym == NULL || (use_browser == 0 &&
 183			    (top->sym_filter_entry == NULL ||
 184			     top->sym_filter_entry->ms.sym != sym)))
 185		return;
 186
 
 187	notes = symbol__annotation(sym);
 188
 189	if (pthread_mutex_trylock(&notes->lock))
 190		return;
 191
 192	err = hist_entry__inc_addr_samples(he, counter, ip);
 
 193
 194	pthread_mutex_unlock(&notes->lock);
 195
 196	if (unlikely(err)) {
 197		/*
 198		 * This function is now called with he->hists->lock held.
 199		 * Release it before going to sleep.
 200		 */
 201		pthread_mutex_unlock(&he->hists->lock);
 202
 203		if (err == -ERANGE && !he->ms.map->erange_warned)
 204			ui__warn_map_erange(he->ms.map, sym, ip);
 205		else if (err == -ENOMEM) {
 206			pr_err("Not enough memory for annotating '%s' symbol!\n",
 207			       sym->name);
 208			sleep(1);
 209		}
 210
 211		pthread_mutex_lock(&he->hists->lock);
 212	}
 213}
 214
 215static void perf_top__show_details(struct perf_top *top)
 216{
 217	struct hist_entry *he = top->sym_filter_entry;
 218	struct annotation *notes;
 219	struct symbol *symbol;
 220	int more;
 221
 222	if (!he)
 223		return;
 224
 225	symbol = he->ms.sym;
 226	notes = symbol__annotation(symbol);
 227
 228	pthread_mutex_lock(&notes->lock);
 229
 230	if (notes->src == NULL)
 231		goto out_unlock;
 232
 233	printf("Showing %s for %s\n", perf_evsel__name(top->sym_evsel), symbol->name);
 234	printf("  Events  Pcnt (>=%d%%)\n", top->sym_pcnt_filter);
 235
 236	more = symbol__annotate_printf(symbol, he->ms.map, top->sym_evsel,
 237				       0, top->sym_pcnt_filter, top->print_entries, 4);
 238
 239	if (top->evlist->enabled) {
 240		if (top->zero)
 241			symbol__annotate_zero_histogram(symbol, top->sym_evsel->idx);
 242		else
 243			symbol__annotate_decay_histogram(symbol, top->sym_evsel->idx);
 244	}
 245	if (more != 0)
 246		printf("%d lines not displayed, maybe increase display entries [e]\n", more);
 247out_unlock:
 248	pthread_mutex_unlock(&notes->lock);
 249}
 250
 251static void perf_top__print_sym_table(struct perf_top *top)
 
 
 
 
 
 
 
 252{
 253	char bf[160];
 254	int printed = 0;
 255	const int win_width = top->winsize.ws_col - 1;
 256	struct perf_evsel *evsel = top->sym_evsel;
 257	struct hists *hists = evsel__hists(evsel);
 
 
 
 258
 259	puts(CONSOLE_CLEAR);
 260
 261	perf_top__header_snprintf(top, bf, sizeof(bf));
 262	printf("%s\n", bf);
 263
 264	perf_top__reset_sample_counters(top);
 265
 266	printf("%-*.*s\n", win_width, win_width, graph_dotted_line);
 267
 268	if (hists->stats.nr_lost_warned !=
 269	    hists->stats.nr_events[PERF_RECORD_LOST]) {
 270		hists->stats.nr_lost_warned =
 271			      hists->stats.nr_events[PERF_RECORD_LOST];
 272		color_fprintf(stdout, PERF_COLOR_RED,
 273			      "WARNING: LOST %d chunks, Check IO/CPU overload",
 274			      hists->stats.nr_lost_warned);
 275		++printed;
 276	}
 277
 278	if (top->sym_filter_entry) {
 279		perf_top__show_details(top);
 280		return;
 281	}
 282
 283	if (top->evlist->enabled) {
 284		if (top->zero) {
 285			hists__delete_entries(hists);
 286		} else {
 287			hists__decay_entries(hists, top->hide_user_symbols,
 288					     top->hide_kernel_symbols);
 289		}
 290	}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 291
 292	hists__collapse_resort(hists, NULL);
 293	perf_evsel__output_resort(evsel, NULL);
 294
 295	hists__output_recalc_col_len(hists, top->print_entries - printed);
 296	putchar('\n');
 297	hists__fprintf(hists, false, top->print_entries - printed, win_width,
 298		       top->min_percent, stdout);
 
 
 
 
 
 
 
 
 
 
 299}
 300
 301static void prompt_integer(int *target, const char *msg)
 302{
 303	char *buf = malloc(0), *p;
 304	size_t dummy = 0;
 305	int tmp;
 306
 307	fprintf(stdout, "\n%s: ", msg);
 308	if (getline(&buf, &dummy, stdin) < 0)
 309		return;
 310
 311	p = strchr(buf, '\n');
 312	if (p)
 313		*p = 0;
 314
 315	p = buf;
 316	while(*p) {
 317		if (!isdigit(*p))
 318			goto out_free;
 319		p++;
 320	}
 321	tmp = strtoul(buf, NULL, 10);
 322	*target = tmp;
 323out_free:
 324	free(buf);
 325}
 326
 327static void prompt_percent(int *target, const char *msg)
 328{
 329	int tmp = 0;
 330
 331	prompt_integer(&tmp, msg);
 332	if (tmp >= 0 && tmp <= 100)
 333		*target = tmp;
 334}
 335
 336static void perf_top__prompt_symbol(struct perf_top *top, const char *msg)
 337{
 338	char *buf = malloc(0), *p;
 339	struct hist_entry *syme = top->sym_filter_entry, *n, *found = NULL;
 340	struct hists *hists = evsel__hists(top->sym_evsel);
 341	struct rb_node *next;
 342	size_t dummy = 0;
 343
 344	/* zero counters of active symbol */
 345	if (syme) {
 346		__zero_source_counters(syme);
 347		top->sym_filter_entry = NULL;
 348	}
 349
 350	fprintf(stdout, "\n%s: ", msg);
 351	if (getline(&buf, &dummy, stdin) < 0)
 352		goto out_free;
 353
 354	p = strchr(buf, '\n');
 355	if (p)
 356		*p = 0;
 357
 358	next = rb_first(&hists->entries);
 359	while (next) {
 360		n = rb_entry(next, struct hist_entry, rb_node);
 361		if (n->ms.sym && !strcmp(buf, n->ms.sym->name)) {
 362			found = n;
 
 
 
 
 363			break;
 364		}
 365		next = rb_next(&n->rb_node);
 366	}
 367
 368	if (!found) {
 369		fprintf(stderr, "Sorry, %s is not active.\n", buf);
 370		sleep(1);
 
 371	} else
 372		perf_top__parse_source(top, found);
 373
 374out_free:
 375	free(buf);
 376}
 377
 378static void perf_top__print_mapped_keys(struct perf_top *top)
 379{
 380	char *name = NULL;
 381
 382	if (top->sym_filter_entry) {
 383		struct symbol *sym = top->sym_filter_entry->ms.sym;
 384		name = sym->name;
 385	}
 386
 387	fprintf(stdout, "\nMapped keys:\n");
 388	fprintf(stdout, "\t[d]     display refresh delay.             \t(%d)\n", top->delay_secs);
 389	fprintf(stdout, "\t[e]     display entries (lines).           \t(%d)\n", top->print_entries);
 390
 391	if (top->evlist->nr_entries > 1)
 392		fprintf(stdout, "\t[E]     active event counter.              \t(%s)\n", perf_evsel__name(top->sym_evsel));
 393
 394	fprintf(stdout, "\t[f]     profile display filter (count).    \t(%d)\n", top->count_filter);
 395
 396	fprintf(stdout, "\t[F]     annotate display filter (percent). \t(%d%%)\n", top->sym_pcnt_filter);
 397	fprintf(stdout, "\t[s]     annotate symbol.                   \t(%s)\n", name?: "NULL");
 398	fprintf(stdout, "\t[S]     stop annotation.\n");
 399
 
 
 
 400	fprintf(stdout,
 401		"\t[K]     hide kernel_symbols symbols.     \t(%s)\n",
 402		top->hide_kernel_symbols ? "yes" : "no");
 403	fprintf(stdout,
 404		"\t[U]     hide user symbols.               \t(%s)\n",
 405		top->hide_user_symbols ? "yes" : "no");
 406	fprintf(stdout, "\t[z]     toggle sample zeroing.             \t(%d)\n", top->zero ? 1 : 0);
 407	fprintf(stdout, "\t[qQ]    quit.\n");
 408}
 409
 410static int perf_top__key_mapped(struct perf_top *top, int c)
 411{
 412	switch (c) {
 413		case 'd':
 414		case 'e':
 415		case 'f':
 416		case 'z':
 417		case 'q':
 418		case 'Q':
 419		case 'K':
 420		case 'U':
 421		case 'F':
 422		case 's':
 423		case 'S':
 424			return 1;
 425		case 'E':
 426			return top->evlist->nr_entries > 1 ? 1 : 0;
 
 427		default:
 428			break;
 429	}
 430
 431	return 0;
 432}
 433
 434static bool perf_top__handle_keypress(struct perf_top *top, int c)
 435{
 436	bool ret = true;
 437
 438	if (!perf_top__key_mapped(top, c)) {
 439		struct pollfd stdin_poll = { .fd = 0, .events = POLLIN };
 440		struct termios save;
 441
 442		perf_top__print_mapped_keys(top);
 443		fprintf(stdout, "\nEnter selection, or unmapped key to continue: ");
 444		fflush(stdout);
 445
 446		set_term_quiet_input(&save);
 
 
 
 
 
 447
 448		poll(&stdin_poll, 1, -1);
 449		c = getc(stdin);
 450
 451		tcsetattr(0, TCSAFLUSH, &save);
 452		if (!perf_top__key_mapped(top, c))
 453			return ret;
 454	}
 455
 456	switch (c) {
 457		case 'd':
 458			prompt_integer(&top->delay_secs, "Enter display delay");
 459			if (top->delay_secs < 1)
 460				top->delay_secs = 1;
 461			break;
 462		case 'e':
 463			prompt_integer(&top->print_entries, "Enter display entries (lines)");
 464			if (top->print_entries == 0) {
 465				struct sigaction act = {
 466					.sa_sigaction = perf_top__sig_winch,
 467					.sa_flags     = SA_SIGINFO,
 468				};
 469				perf_top__sig_winch(SIGWINCH, NULL, top);
 470				sigaction(SIGWINCH, &act, NULL);
 471			} else {
 472				signal(SIGWINCH, SIG_DFL);
 473			}
 474			break;
 475		case 'E':
 476			if (top->evlist->nr_entries > 1) {
 477				/* Select 0 as the default event: */
 478				int counter = 0;
 479
 480				fprintf(stderr, "\nAvailable events:");
 481
 482				evlist__for_each(top->evlist, top->sym_evsel)
 483					fprintf(stderr, "\n\t%d %s", top->sym_evsel->idx, perf_evsel__name(top->sym_evsel));
 484
 485				prompt_integer(&counter, "Enter details event counter");
 486
 487				if (counter >= top->evlist->nr_entries) {
 488					top->sym_evsel = perf_evlist__first(top->evlist);
 489					fprintf(stderr, "Sorry, no such event, using %s.\n", perf_evsel__name(top->sym_evsel));
 490					sleep(1);
 491					break;
 492				}
 493				evlist__for_each(top->evlist, top->sym_evsel)
 494					if (top->sym_evsel->idx == counter)
 495						break;
 496			} else
 497				top->sym_evsel = perf_evlist__first(top->evlist);
 498			break;
 499		case 'f':
 500			prompt_integer(&top->count_filter, "Enter display event count filter");
 501			break;
 502		case 'F':
 503			prompt_percent(&top->sym_pcnt_filter,
 504				       "Enter details display event filter (percent)");
 505			break;
 506		case 'K':
 507			top->hide_kernel_symbols = !top->hide_kernel_symbols;
 508			break;
 509		case 'q':
 510		case 'Q':
 511			printf("exiting.\n");
 512			if (top->dump_symtab)
 513				perf_session__fprintf_dsos(top->session, stderr);
 514			ret = false;
 515			break;
 516		case 's':
 517			perf_top__prompt_symbol(top, "Enter details symbol");
 518			break;
 519		case 'S':
 520			if (!top->sym_filter_entry)
 521				break;
 522			else {
 523				struct hist_entry *syme = top->sym_filter_entry;
 524
 525				top->sym_filter_entry = NULL;
 526				__zero_source_counters(syme);
 527			}
 528			break;
 529		case 'U':
 530			top->hide_user_symbols = !top->hide_user_symbols;
 
 
 
 531			break;
 532		case 'z':
 533			top->zero = !top->zero;
 534			break;
 535		default:
 536			break;
 537	}
 538
 539	return ret;
 540}
 541
 542static void perf_top__sort_new_samples(void *arg)
 543{
 544	struct perf_top *t = arg;
 545	struct perf_evsel *evsel = t->sym_evsel;
 546	struct hists *hists;
 547
 548	perf_top__reset_sample_counters(t);
 549
 550	if (t->evlist->selected != NULL)
 551		t->sym_evsel = t->evlist->selected;
 552
 553	hists = evsel__hists(evsel);
 554
 555	if (t->evlist->enabled) {
 556		if (t->zero) {
 557			hists__delete_entries(hists);
 558		} else {
 559			hists__decay_entries(hists, t->hide_user_symbols,
 560					     t->hide_kernel_symbols);
 561		}
 562	}
 563
 564	hists__collapse_resort(hists, NULL);
 565	perf_evsel__output_resort(evsel, NULL);
 566}
 567
 568static void *display_thread_tui(void *arg)
 569{
 570	struct perf_evsel *pos;
 571	struct perf_top *top = arg;
 572	const char *help = "For a higher level overview, try: perf top --sort comm,dso";
 573	struct hist_browser_timer hbt = {
 574		.timer		= perf_top__sort_new_samples,
 575		.arg		= top,
 576		.refresh	= top->delay_secs,
 577	};
 578
 579	perf_top__sort_new_samples(top);
 580
 581	/*
 582	 * Initialize the uid_filter_str, in the future the TUI will allow
 583	 * Zooming in/out UIDs. For now juse use whatever the user passed
 584	 * via --uid.
 585	 */
 586	evlist__for_each(top->evlist, pos) {
 587		struct hists *hists = evsel__hists(pos);
 588		hists->uid_filter_str = top->record_opts.target.uid_str;
 589	}
 590
 591	perf_evlist__tui_browse_hists(top->evlist, help, &hbt,
 592				      top->min_percent,
 593				      &top->session->header.env);
 594
 595	done = 1;
 596	return NULL;
 597}
 598
 599static void display_sig(int sig __maybe_unused)
 600{
 601	done = 1;
 602}
 603
 604static void display_setup_sig(void)
 605{
 606	signal(SIGSEGV, sighandler_dump_stack);
 607	signal(SIGFPE, sighandler_dump_stack);
 608	signal(SIGINT,  display_sig);
 609	signal(SIGQUIT, display_sig);
 610	signal(SIGTERM, display_sig);
 611}
 612
 613static void *display_thread(void *arg)
 614{
 615	struct pollfd stdin_poll = { .fd = 0, .events = POLLIN };
 616	struct termios save;
 617	struct perf_top *top = arg;
 618	int delay_msecs, c;
 
 
 
 
 
 
 
 619
 620	display_setup_sig();
 621	pthread__unblock_sigwinch();
 622repeat:
 623	delay_msecs = top->delay_secs * 1000;
 624	set_term_quiet_input(&save);
 625	/* trash return*/
 626	getc(stdin);
 627
 628	while (!done) {
 629		perf_top__print_sym_table(top);
 630		/*
 631		 * Either timeout expired or we got an EINTR due to SIGWINCH,
 632		 * refresh screen in both cases.
 633		 */
 634		switch (poll(&stdin_poll, 1, delay_msecs)) {
 635		case 0:
 636			continue;
 637		case -1:
 638			if (errno == EINTR)
 639				continue;
 640			/* Fall trhu */
 641		default:
 642			c = getc(stdin);
 643			tcsetattr(0, TCSAFLUSH, &save);
 644
 645			if (perf_top__handle_keypress(top, c))
 646				goto repeat;
 647			done = 1;
 648		}
 649	}
 650
 
 651	tcsetattr(0, TCSAFLUSH, &save);
 
 
 
 
 652	return NULL;
 653}
 654
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 655static int symbol_filter(struct map *map, struct symbol *sym)
 656{
 
 657	const char *name = sym->name;
 
 658
 659	if (!__map__is_kernel(map))
 660		return 0;
 661	/*
 662	 * ppc64 uses function descriptors and appends a '.' to the
 663	 * start of every instruction address. Remove it.
 664	 */
 665	if (name[0] == '.')
 666		name++;
 667
 668	if (!strcmp(name, "_text") ||
 669	    !strcmp(name, "_etext") ||
 670	    !strcmp(name, "_sinittext") ||
 671	    !strncmp("init_module", name, 11) ||
 672	    !strncmp("cleanup_module", name, 14) ||
 673	    strstr(name, "_text_start") ||
 674	    strstr(name, "_text_end"))
 675		return 1;
 676
 677	if (symbol__is_idle(sym))
 678		sym->ignore = true;
 679
 680	return 0;
 681}
 682
 683static int hist_iter__top_callback(struct hist_entry_iter *iter,
 684				   struct addr_location *al, bool single,
 685				   void *arg)
 686{
 687	struct perf_top *top = arg;
 688	struct hist_entry *he = iter->he;
 689	struct perf_evsel *evsel = iter->evsel;
 690
 691	if (sort__has_sym && single)
 692		perf_top__record_precise_ip(top, he, evsel->idx, al->addr);
 
 
 
 
 693
 694	hist__account_cycles(iter->sample->branch_stack, al, iter->sample,
 695		     !(top->record_opts.branch_stack & PERF_SAMPLE_BRANCH_ANY));
 696	return 0;
 697}
 698
 699static void perf_event__process_sample(struct perf_tool *tool,
 700				       const union perf_event *event,
 701				       struct perf_evsel *evsel,
 702				       struct perf_sample *sample,
 703				       struct machine *machine)
 704{
 705	struct perf_top *top = container_of(tool, struct perf_top, tool);
 
 706	struct addr_location al;
 707	int err;
 708
 709	if (!machine && perf_guest) {
 710		static struct intlist *seen;
 711
 712		if (!seen)
 713			seen = intlist__new(NULL);
 714
 715		if (!intlist__has_entry(seen, sample->pid)) {
 716			pr_err("Can't find guest [%d]'s kernel information\n",
 717				sample->pid);
 718			intlist__add(seen, sample->pid);
 719		}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 720		return;
 721	}
 722
 723	if (!machine) {
 724		pr_err("%u unprocessable samples recorded.\r",
 725		       top->session->evlist->stats.nr_unprocessable_samples++);
 726		return;
 727	}
 728
 729	if (event->header.misc & PERF_RECORD_MISC_EXACT_IP)
 730		top->exact_samples++;
 731
 732	if (machine__resolve(machine, &al, sample) < 0)
 
 
 733		return;
 734
 735	if (!top->kptr_restrict_warned &&
 736	    symbol_conf.kptr_restrict &&
 737	    al.cpumode == PERF_RECORD_MISC_KERNEL) {
 738		ui__warning(
 739"Kernel address maps (/proc/{kallsyms,modules}) are restricted.\n\n"
 740"Check /proc/sys/kernel/kptr_restrict.\n\n"
 741"Kernel%s samples will not be resolved.\n",
 742			  al.map && !RB_EMPTY_ROOT(&al.map->dso->symbols[MAP__FUNCTION]) ?
 743			  " modules" : "");
 744		if (use_browser <= 0)
 745			sleep(5);
 746		top->kptr_restrict_warned = true;
 747	}
 748
 749	if (al.sym == NULL) {
 750		const char *msg = "Kernel samples will not be resolved.\n";
 751		/*
 752		 * As we do lazy loading of symtabs we only will know if the
 753		 * specified vmlinux file is invalid when we actually have a
 754		 * hit in kernel space and then try to load it. So if we get
 755		 * here and there are _no_ symbols in the DSO backing the
 756		 * kernel map, bail out.
 757		 *
 758		 * We may never get here, for instance, if we use -K/
 759		 * --hide-kernel-symbols, even if the user specifies an
 760		 * invalid --vmlinux ;-)
 761		 */
 762		if (!top->kptr_restrict_warned && !top->vmlinux_warned &&
 763		    al.map == machine->vmlinux_maps[MAP__FUNCTION] &&
 764		    RB_EMPTY_ROOT(&al.map->dso->symbols[MAP__FUNCTION])) {
 765			if (symbol_conf.vmlinux_name) {
 766				char serr[256];
 767				dso__strerror_load(al.map->dso, serr, sizeof(serr));
 768				ui__warning("The %s file can't be used: %s\n%s",
 769					    symbol_conf.vmlinux_name, serr, msg);
 770			} else {
 771				ui__warning("A vmlinux file was not found.\n%s",
 772					    msg);
 773			}
 774
 775			if (use_browser <= 0)
 776				sleep(5);
 777			top->vmlinux_warned = true;
 778		}
 779	}
 780
 781	if (al.sym == NULL || !al.sym->ignore) {
 782		struct hists *hists = evsel__hists(evsel);
 783		struct hist_entry_iter iter = {
 784			.evsel		= evsel,
 785			.sample 	= sample,
 786			.add_entry_cb 	= hist_iter__top_callback,
 787		};
 788
 789		if (symbol_conf.cumulate_callchain)
 790			iter.ops = &hist_iter_cumulative;
 791		else
 792			iter.ops = &hist_iter_normal;
 793
 794		pthread_mutex_lock(&hists->lock);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 795
 796		err = hist_entry_iter__add(&iter, &al, top->max_stack, top);
 797		if (err < 0)
 798			pr_err("Problem incrementing symbol period, skipping event\n");
 799
 800		pthread_mutex_unlock(&hists->lock);
 
 
 
 
 
 
 
 
 
 
 
 
 
 801	}
 802
 803	addr_location__put(&al);
 804}
 805
 806static void perf_top__mmap_read_idx(struct perf_top *top, int idx)
 807{
 808	struct perf_sample sample;
 809	struct perf_evsel *evsel;
 810	struct perf_session *session = top->session;
 811	union perf_event *event;
 812	struct machine *machine;
 813	int ret;
 814
 815	while ((event = perf_evlist__mmap_read(top->evlist, idx)) != NULL) {
 816		ret = perf_evlist__parse_sample(top->evlist, event, &sample);
 817		if (ret) {
 818			pr_err("Can't parse sample, err = %d\n", ret);
 819			goto next_event;
 820		}
 821
 822		evsel = perf_evlist__id2evsel(session->evlist, sample.id);
 823		assert(evsel != NULL);
 824
 825		if (event->header.type == PERF_RECORD_SAMPLE)
 826			++top->samples;
 827
 828		switch (sample.cpumode) {
 829		case PERF_RECORD_MISC_USER:
 830			++top->us_samples;
 831			if (top->hide_user_symbols)
 832				goto next_event;
 833			machine = &session->machines.host;
 834			break;
 835		case PERF_RECORD_MISC_KERNEL:
 836			++top->kernel_samples;
 837			if (top->hide_kernel_symbols)
 838				goto next_event;
 839			machine = &session->machines.host;
 840			break;
 841		case PERF_RECORD_MISC_GUEST_KERNEL:
 842			++top->guest_kernel_samples;
 843			machine = perf_session__find_machine(session,
 844							     sample.pid);
 845			break;
 846		case PERF_RECORD_MISC_GUEST_USER:
 847			++top->guest_us_samples;
 848			/*
 849			 * TODO: we don't process guest user from host side
 850			 * except simple counting.
 851			 */
 852			goto next_event;
 853		default:
 854			if (event->header.type == PERF_RECORD_SAMPLE)
 855				goto next_event;
 856			machine = &session->machines.host;
 857			break;
 858		}
 859
 860
 861		if (event->header.type == PERF_RECORD_SAMPLE) {
 862			perf_event__process_sample(&top->tool, event, evsel,
 863						   &sample, machine);
 864		} else if (event->header.type < PERF_RECORD_MAX) {
 865			hists__inc_nr_events(evsel__hists(evsel), event->header.type);
 866			machine__process_event(machine, event, &sample);
 867		} else
 868			++session->evlist->stats.nr_unknown_events;
 869next_event:
 870		perf_evlist__mmap_consume(top->evlist, idx);
 871	}
 872}
 873
 874static void perf_top__mmap_read(struct perf_top *top)
 875{
 876	int i;
 877
 878	for (i = 0; i < top->evlist->nr_mmaps; i++)
 879		perf_top__mmap_read_idx(top, i);
 880}
 881
 882static int perf_top__start_counters(struct perf_top *top)
 883{
 884	char msg[512];
 885	struct perf_evsel *counter;
 886	struct perf_evlist *evlist = top->evlist;
 887	struct record_opts *opts = &top->record_opts;
 888
 889	perf_evlist__config(evlist, opts);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 890
 891	evlist__for_each(evlist, counter) {
 
 892try_again:
 893		if (perf_evsel__open(counter, top->evlist->cpus,
 894				     top->evlist->threads) < 0) {
 895			if (perf_evsel__fallback(counter, errno, msg, sizeof(msg))) {
 
 
 
 
 
 
 
 
 
 
 
 
 896				if (verbose)
 897					ui__warning("%s\n", msg);
 
 
 
 
 898				goto try_again;
 899			}
 900
 901			perf_evsel__open_strerror(counter, &opts->target,
 902						  errno, msg, sizeof(msg));
 903			ui__error("%s\n", msg);
 
 
 
 
 
 
 
 
 904			goto out_err;
 905		}
 906	}
 907
 908	if (perf_evlist__mmap(evlist, opts->mmap_pages, false) < 0) {
 909		ui__error("Failed to mmap with %d (%s)\n",
 910			    errno, strerror_r(errno, msg, sizeof(msg)));
 911		goto out_err;
 912	}
 913
 914	return 0;
 915
 916out_err:
 917	return -1;
 
 918}
 919
 920static int perf_top__setup_sample_type(struct perf_top *top __maybe_unused)
 921{
 922	if (!sort__has_sym) {
 923		if (symbol_conf.use_callchain) {
 924			ui__error("Selected -g but \"sym\" not present in --sort/-s.");
 925			return -EINVAL;
 926		}
 927	} else if (callchain_param.mode != CHAIN_NONE) {
 928		if (callchain_register_param(&callchain_param) < 0) {
 929			ui__error("Can't register callchain params.\n");
 930			return -EINVAL;
 931		}
 932	}
 933
 934	return 0;
 935}
 936
 937static int __cmd_top(struct perf_top *top)
 938{
 939	struct record_opts *opts = &top->record_opts;
 940	pthread_t thread;
 941	int ret;
 942
 943	top->session = perf_session__new(NULL, false, NULL);
 944	if (top->session == NULL)
 945		return -1;
 946
 947	machines__set_symbol_filter(&top->session->machines, symbol_filter);
 948
 949	if (!objdump_path) {
 950		ret = perf_env__lookup_objdump(&top->session->header.env);
 951		if (ret)
 952			goto out_delete;
 953	}
 954
 955	ret = perf_top__setup_sample_type(top);
 956	if (ret)
 957		goto out_delete;
 958
 959	if (perf_session__register_idle_thread(top->session) < 0)
 960		goto out_delete;
 961
 962	machine__synthesize_threads(&top->session->machines.host, &opts->target,
 963				    top->evlist->threads, false, opts->proc_map_timeout);
 964
 965	if (sort__has_socket) {
 966		ret = perf_env__read_cpu_topology_map(&perf_env);
 967		if (ret < 0)
 968			goto out_err_cpu_topo;
 969	}
 970
 971	ret = perf_top__start_counters(top);
 972	if (ret)
 973		goto out_delete;
 974
 975	top->session->evlist = top->evlist;
 976	perf_session__set_id_hdr_size(top->session);
 977
 978	/*
 979	 * When perf is starting the traced process, all the events (apart from
 980	 * group members) have enable_on_exec=1 set, so don't spoil it by
 981	 * prematurely enabling them.
 982	 *
 983	 * XXX 'top' still doesn't start workloads like record, trace, but should,
 984	 * so leave the check here.
 985	 */
 986        if (!target__none(&opts->target))
 987                perf_evlist__enable(top->evlist);
 
 
 
 
 
 
 
 
 
 
 
 988
 989	/* Wait for a minimal set of events before starting the snapshot */
 990	perf_evlist__poll(top->evlist, 100);
 991
 992	perf_top__mmap_read(top);
 993
 994	ret = -1;
 995	if (pthread_create(&thread, NULL, (use_browser > 0 ? display_thread_tui :
 996							    display_thread), top)) {
 997		ui__error("Could not create display thread.\n");
 998		goto out_delete;
 999	}
1000
1001	if (top->realtime_prio) {
1002		struct sched_param param;
1003
1004		param.sched_priority = top->realtime_prio;
1005		if (sched_setscheduler(0, SCHED_FIFO, &param)) {
1006			ui__error("Could not set realtime priority.\n");
1007			goto out_join;
1008		}
1009	}
1010
1011	while (!done) {
1012		u64 hits = top->samples;
1013
1014		perf_top__mmap_read(top);
1015
1016		if (hits == top->samples)
1017			ret = perf_evlist__poll(top->evlist, 100);
1018	}
1019
1020	ret = 0;
1021out_join:
1022	pthread_join(thread, NULL);
1023out_delete:
1024	perf_session__delete(top->session);
1025	top->session = NULL;
1026
1027	return ret;
1028
1029out_err_cpu_topo: {
1030	char errbuf[BUFSIZ];
1031	const char *err = strerror_r(-ret, errbuf, sizeof(errbuf));
1032
1033	ui__error("Could not read the CPU topology map: %s\n", err);
1034	goto out_delete;
1035}
1036}
1037
1038static int
1039callchain_opt(const struct option *opt, const char *arg, int unset)
1040{
1041	symbol_conf.use_callchain = true;
1042	return record_callchain_opt(opt, arg, unset);
1043}
1044
1045static int
1046parse_callchain_opt(const struct option *opt, const char *arg, int unset)
1047{
1048	struct record_opts *record = (struct record_opts *)opt->value;
1049
1050	record->callgraph_set = true;
1051	callchain_param.enabled = !unset;
1052	callchain_param.record_mode = CALLCHAIN_FP;
1053
1054	/*
1055	 * --no-call-graph
1056	 */
1057	if (unset) {
1058		symbol_conf.use_callchain = false;
1059		callchain_param.record_mode = CALLCHAIN_NONE;
1060		return 0;
1061	}
1062
1063	return parse_callchain_top_opt(arg);
1064}
1065
1066static int perf_top_config(const char *var, const char *value, void *cb __maybe_unused)
1067{
1068	if (!strcmp(var, "top.call-graph"))
1069		var = "call-graph.record-mode"; /* fall-through */
1070	if (!strcmp(var, "top.children")) {
1071		symbol_conf.cumulate_callchain = perf_config_bool(var, value);
1072		return 0;
1073	}
1074
1075	return 0;
1076}
1077
1078static int
1079parse_percent_limit(const struct option *opt, const char *arg,
1080		    int unset __maybe_unused)
1081{
1082	struct perf_top *top = opt->value;
1083
1084	top->min_percent = strtof(arg, NULL);
1085	return 0;
1086}
1087
1088const char top_callchain_help[] = CALLCHAIN_RECORD_HELP CALLCHAIN_REPORT_HELP
1089	"\n\t\t\t\tDefault: fp,graph,0.5,caller,function";
1090
1091int cmd_top(int argc, const char **argv, const char *prefix __maybe_unused)
1092{
1093	char errbuf[BUFSIZ];
1094	struct perf_top top = {
1095		.count_filter	     = 5,
1096		.delay_secs	     = 2,
1097		.record_opts = {
1098			.mmap_pages	= UINT_MAX,
1099			.user_freq	= UINT_MAX,
1100			.user_interval	= ULLONG_MAX,
1101			.freq		= 4000, /* 4 KHz */
1102			.target		= {
1103				.uses_mmap   = true,
1104			},
1105			.proc_map_timeout    = 500,
1106		},
1107		.max_stack	     = PERF_MAX_STACK_DEPTH,
1108		.sym_pcnt_filter     = 5,
1109	};
1110	struct record_opts *opts = &top.record_opts;
1111	struct target *target = &opts->target;
1112	const struct option options[] = {
1113	OPT_CALLBACK('e', "event", &top.evlist, "event",
1114		     "event selector. use 'perf list' to list available events",
1115		     parse_events_option),
1116	OPT_U64('c', "count", &opts->user_interval, "event period to sample"),
1117	OPT_STRING('p', "pid", &target->pid, "pid",
 
1118		    "profile events on existing process id"),
1119	OPT_STRING('t', "tid", &target->tid, "tid",
1120		    "profile events on existing thread id"),
1121	OPT_BOOLEAN('a', "all-cpus", &target->system_wide,
1122			    "system-wide collection from all CPUs"),
1123	OPT_STRING('C', "cpu", &target->cpu_list, "cpu",
1124		    "list of cpus to monitor"),
1125	OPT_STRING('k', "vmlinux", &symbol_conf.vmlinux_name,
1126		   "file", "vmlinux pathname"),
1127	OPT_BOOLEAN(0, "ignore-vmlinux", &symbol_conf.ignore_vmlinux,
1128		    "don't load vmlinux even if found"),
1129	OPT_BOOLEAN('K', "hide_kernel_symbols", &top.hide_kernel_symbols,
1130		    "hide kernel symbols"),
1131	OPT_CALLBACK('m', "mmap-pages", &opts->mmap_pages, "pages",
1132		     "number of mmap data pages",
1133		     perf_evlist__parse_mmap_pages),
1134	OPT_INTEGER('r', "realtime", &top.realtime_prio,
1135		    "collect data with this RT SCHED_FIFO priority"),
1136	OPT_INTEGER('d', "delay", &top.delay_secs,
1137		    "number of seconds to delay between refreshes"),
1138	OPT_BOOLEAN('D', "dump-symtab", &top.dump_symtab,
1139			    "dump the symbol table used for profiling"),
1140	OPT_INTEGER('f', "count-filter", &top.count_filter,
1141		    "only display functions with more events than this"),
1142	OPT_BOOLEAN(0, "group", &opts->group,
1143			    "put the counters into a counter group"),
1144	OPT_BOOLEAN('i', "no-inherit", &opts->no_inherit,
1145		    "child tasks do not inherit counters"),
1146	OPT_STRING(0, "sym-annotate", &top.sym_filter, "symbol name",
1147		    "symbol to annotate"),
1148	OPT_BOOLEAN('z', "zero", &top.zero, "zero history across updates"),
1149	OPT_UINTEGER('F', "freq", &opts->user_freq, "profile at this frequency"),
 
 
1150	OPT_INTEGER('E', "entries", &top.print_entries,
1151		    "display this many functions"),
1152	OPT_BOOLEAN('U', "hide_user_symbols", &top.hide_user_symbols,
1153		    "hide user symbols"),
1154	OPT_BOOLEAN(0, "tui", &top.use_tui, "Use the TUI interface"),
1155	OPT_BOOLEAN(0, "stdio", &top.use_stdio, "Use the stdio interface"),
1156	OPT_INCR('v', "verbose", &verbose,
1157		    "be more verbose (show counter open errors, etc)"),
1158	OPT_STRING('s', "sort", &sort_order, "key[,key2...]",
1159		   "sort by key(s): pid, comm, dso, symbol, parent, cpu, srcline, ..."
1160		   " Please refer the man page for the complete list."),
1161	OPT_STRING(0, "fields", &field_order, "key[,keys...]",
1162		   "output field(s): overhead, period, sample plus all of sort keys"),
1163	OPT_BOOLEAN('n', "show-nr-samples", &symbol_conf.show_nr_samples,
1164		    "Show a column with the number of samples"),
1165	OPT_CALLBACK_NOOPT('g', NULL, &top.record_opts,
1166			   NULL, "enables call-graph recording and display",
1167			   &callchain_opt),
1168	OPT_CALLBACK(0, "call-graph", &top.record_opts,
1169		     "record_mode[,record_size],print_type,threshold[,print_limit],order,sort_key[,branch]",
1170		     top_callchain_help, &parse_callchain_opt),
1171	OPT_BOOLEAN(0, "children", &symbol_conf.cumulate_callchain,
1172		    "Accumulate callchains of children and show total overhead as well"),
1173	OPT_INTEGER(0, "max-stack", &top.max_stack,
1174		    "Set the maximum stack depth when parsing the callchain. "
1175		    "Default: " __stringify(PERF_MAX_STACK_DEPTH)),
1176	OPT_CALLBACK(0, "ignore-callees", NULL, "regex",
1177		   "ignore callees of these functions in call graphs",
1178		   report_parse_ignore_callees_opt),
1179	OPT_BOOLEAN(0, "show-total-period", &symbol_conf.show_total_period,
1180		    "Show a column with the sum of periods"),
1181	OPT_STRING(0, "dsos", &symbol_conf.dso_list_str, "dso[,dso...]",
1182		   "only consider symbols in these dsos"),
1183	OPT_STRING(0, "comms", &symbol_conf.comm_list_str, "comm[,comm...]",
1184		   "only consider symbols in these comms"),
1185	OPT_STRING(0, "symbols", &symbol_conf.sym_list_str, "symbol[,symbol...]",
1186		   "only consider these symbols"),
1187	OPT_BOOLEAN(0, "source", &symbol_conf.annotate_src,
1188		    "Interleave source code with assembly code (default)"),
1189	OPT_BOOLEAN(0, "asm-raw", &symbol_conf.annotate_asm_raw,
1190		    "Display raw encoding of assembly instructions (default)"),
1191	OPT_BOOLEAN(0, "demangle-kernel", &symbol_conf.demangle_kernel,
1192		    "Enable kernel symbol demangling"),
1193	OPT_STRING(0, "objdump", &objdump_path, "path",
1194		    "objdump binary to use for disassembly and annotations"),
1195	OPT_STRING('M', "disassembler-style", &disassembler_style, "disassembler style",
1196		   "Specify disassembler style (e.g. -M intel for intel syntax)"),
1197	OPT_STRING('u', "uid", &target->uid_str, "user", "user to profile"),
1198	OPT_CALLBACK(0, "percent-limit", &top, "percent",
1199		     "Don't show entries under that percent", parse_percent_limit),
1200	OPT_CALLBACK(0, "percentage", NULL, "relative|absolute",
1201		     "How to display percentage of filtered entries", parse_filter_percentage),
1202	OPT_STRING('w', "column-widths", &symbol_conf.col_width_list_str,
1203		   "width[,width...]",
1204		   "don't try to adjust column width, use these fixed values"),
1205	OPT_UINTEGER(0, "proc-map-timeout", &opts->proc_map_timeout,
1206			"per thread proc mmap processing timeout in ms"),
1207	OPT_CALLBACK_NOOPT('b', "branch-any", &opts->branch_stack,
1208		     "branch any", "sample any taken branches",
1209		     parse_branch_stack),
1210	OPT_CALLBACK('j', "branch-filter", &opts->branch_stack,
1211		     "branch filter mask", "branch stack filter modes",
1212		     parse_branch_stack),
1213	OPT_BOOLEAN(0, "raw-trace", &symbol_conf.raw_trace,
1214		    "Show raw trace event output (do not use print fmt or plugins)"),
1215	OPT_BOOLEAN(0, "hierarchy", &symbol_conf.report_hierarchy,
1216		    "Show entries in a hierarchy"),
1217	OPT_END()
1218	};
1219	const char * const top_usage[] = {
1220		"perf top [<options>]",
1221		NULL
1222	};
1223	int status = hists__init();
1224
1225	if (status < 0)
1226		return status;
 
 
1227
1228	top.evlist = perf_evlist__new();
1229	if (top.evlist == NULL)
1230		return -ENOMEM;
1231
1232	perf_config(perf_top_config, &top);
1233
1234	argc = parse_options(argc, argv, options, top_usage, 0);
1235	if (argc)
1236		usage_with_options(top_usage, options);
1237
1238	if (!top.evlist->nr_entries &&
1239	    perf_evlist__add_default(top.evlist) < 0) {
1240		pr_err("Not enough memory for event selector list\n");
1241		goto out_delete_evlist;
1242	}
1243
1244	if (symbol_conf.report_hierarchy) {
1245		/* disable incompatible options */
1246		symbol_conf.event_group = false;
1247		symbol_conf.cumulate_callchain = false;
1248
1249		if (field_order) {
1250			pr_err("Error: --hierarchy and --fields options cannot be used together\n");
1251			parse_options_usage(top_usage, options, "fields", 0);
1252			parse_options_usage(NULL, options, "hierarchy", 0);
1253			goto out_delete_evlist;
1254		}
1255	}
1256
1257	sort__mode = SORT_MODE__TOP;
1258	/* display thread wants entries to be collapsed in a different tree */
1259	sort__need_collapse = 1;
1260
1261	if (top.use_stdio)
1262		use_browser = 0;
1263	else if (top.use_tui)
1264		use_browser = 1;
1265
1266	setup_browser(false);
1267
1268	if (setup_sorting(top.evlist) < 0) {
1269		if (sort_order)
1270			parse_options_usage(top_usage, options, "s", 1);
1271		if (field_order)
1272			parse_options_usage(sort_order ? NULL : top_usage,
1273					    options, "fields", 0);
1274		goto out_delete_evlist;
1275	}
1276
1277	status = target__validate(target);
1278	if (status) {
1279		target__strerror(target, status, errbuf, BUFSIZ);
1280		ui__warning("%s\n", errbuf);
1281	}
1282
1283	status = target__parse_uid(target);
1284	if (status) {
1285		int saved_errno = errno;
1286
1287		target__strerror(target, status, errbuf, BUFSIZ);
1288		ui__error("%s\n", errbuf);
1289
1290		status = -saved_errno;
1291		goto out_delete_evlist;
1292	}
1293
1294	if (target__none(target))
1295		target->system_wide = true;
1296
1297	if (perf_evlist__create_maps(top.evlist, target) < 0) {
1298		ui__error("Couldn't create thread/CPU maps: %s\n",
1299			  errno == ENOENT ? "No such process" : strerror_r(errno, errbuf, sizeof(errbuf)));
1300		goto out_delete_evlist;
1301	}
1302
1303	symbol_conf.nr_events = top.evlist->nr_entries;
1304
1305	if (top.delay_secs < 1)
1306		top.delay_secs = 1;
1307
1308	if (record_opts__config(opts)) {
1309		status = -EINVAL;
1310		goto out_delete_evlist;
 
 
 
 
 
 
 
1311	}
1312
1313	top.sym_evsel = perf_evlist__first(top.evlist);
 
 
 
 
 
 
 
 
1314
1315	if (!symbol_conf.use_callchain) {
1316		symbol_conf.cumulate_callchain = false;
1317		perf_hpp__cancel_cumulate();
1318	}
1319
1320	if (symbol_conf.cumulate_callchain && !callchain_param.order_set)
1321		callchain_param.order = ORDER_CALLER;
 
1322
1323	symbol_conf.priv_size = sizeof(struct annotation);
 
 
 
1324
1325	symbol_conf.try_vmlinux_path = (symbol_conf.vmlinux_name == NULL);
1326	if (symbol__init(NULL) < 0)
1327		return -1;
1328
1329	sort__setup_elide(stdout);
1330
1331	get_term_dimensions(&top.winsize);
1332	if (top.print_entries == 0) {
1333		struct sigaction act = {
1334			.sa_sigaction = perf_top__sig_winch,
1335			.sa_flags     = SA_SIGINFO,
1336		};
1337		perf_top__update_print_entries(&top);
1338		sigaction(SIGWINCH, &act, NULL);
1339	}
1340
1341	status = __cmd_top(&top);
1342
1343out_delete_evlist:
1344	perf_evlist__delete(top.evlist);
1345
1346	return status;
1347}
v3.1
   1/*
   2 * builtin-top.c
   3 *
   4 * Builtin top command: Display a continuously updated profile of
   5 * any workload, CPU or specific PID.
   6 *
   7 * Copyright (C) 2008, Red Hat Inc, Ingo Molnar <mingo@redhat.com>
 
   8 *
   9 * Improvements and fixes by:
  10 *
  11 *   Arjan van de Ven <arjan@linux.intel.com>
  12 *   Yanmin Zhang <yanmin.zhang@intel.com>
  13 *   Wu Fengguang <fengguang.wu@intel.com>
  14 *   Mike Galbraith <efault@gmx.de>
  15 *   Paul Mackerras <paulus@samba.org>
  16 *
  17 * 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/annotate.h"
  24#include "util/cache.h"
  25#include "util/color.h"
  26#include "util/evlist.h"
  27#include "util/evsel.h"
 
  28#include "util/session.h"
  29#include "util/symbol.h"
  30#include "util/thread.h"
  31#include "util/thread_map.h"
  32#include "util/top.h"
  33#include "util/util.h"
  34#include <linux/rbtree.h>
  35#include "util/parse-options.h"
  36#include "util/parse-events.h"
  37#include "util/cpumap.h"
  38#include "util/xyarray.h"
 
 
 
 
  39
  40#include "util/debug.h"
  41
  42#include <assert.h>
 
  43#include <fcntl.h>
  44
  45#include <stdio.h>
  46#include <termios.h>
  47#include <unistd.h>
  48#include <inttypes.h>
  49
  50#include <errno.h>
  51#include <time.h>
  52#include <sched.h>
  53
  54#include <sys/syscall.h>
  55#include <sys/ioctl.h>
  56#include <sys/poll.h>
  57#include <sys/prctl.h>
  58#include <sys/wait.h>
  59#include <sys/uio.h>
 
  60#include <sys/mman.h>
  61
  62#include <linux/unistd.h>
  63#include <linux/types.h>
  64
  65static struct perf_top top = {
  66	.count_filter		= 5,
  67	.delay_secs		= 2,
  68	.display_weighted	= -1,
  69	.target_pid		= -1,
  70	.target_tid		= -1,
  71	.active_symbols		= LIST_HEAD_INIT(top.active_symbols),
  72	.active_symbols_lock	= PTHREAD_MUTEX_INITIALIZER,
  73	.active_symbols_cond	= PTHREAD_COND_INITIALIZER,
  74	.freq			= 1000, /* 1 KHz */
  75};
  76
  77static bool			system_wide			=  false;
  78
  79static bool			use_tui, use_stdio;
  80
  81static int			default_interval		=      0;
  82
  83static bool			kptr_restrict_warned;
  84static bool			vmlinux_warned;
  85static bool			inherit				=  false;
  86static int			realtime_prio			=      0;
  87static bool			group				=  false;
  88static unsigned int		page_size;
  89static unsigned int		mmap_pages			=    128;
  90
  91static bool			dump_symtab                     =  false;
  92
  93static struct winsize		winsize;
  94
  95static const char		*sym_filter			=   NULL;
  96struct sym_entry		*sym_filter_entry_sched		=   NULL;
  97static int			sym_pcnt_filter			=      5;
  98
  99/*
 100 * Source functions
 101 */
 102
 103void get_term_dimensions(struct winsize *ws)
 104{
 105	char *s = getenv("LINES");
 106
 107	if (s != NULL) {
 108		ws->ws_row = atoi(s);
 109		s = getenv("COLUMNS");
 110		if (s != NULL) {
 111			ws->ws_col = atoi(s);
 112			if (ws->ws_row && ws->ws_col)
 113				return;
 114		}
 115	}
 116#ifdef TIOCGWINSZ
 117	if (ioctl(1, TIOCGWINSZ, ws) == 0 &&
 118	    ws->ws_row && ws->ws_col)
 119		return;
 120#endif
 121	ws->ws_row = 25;
 122	ws->ws_col = 80;
 123}
 124
 125static void update_print_entries(struct winsize *ws)
 
 126{
 127	top.print_entries = ws->ws_row;
 128
 129	if (top.print_entries > 9)
 130		top.print_entries -= 9;
 131}
 132
 133static void sig_winch_handler(int sig __used)
 134{
 135	get_term_dimensions(&winsize);
 136	update_print_entries(&winsize);
 137}
 138
 139static int parse_source(struct sym_entry *syme)
 140{
 141	struct symbol *sym;
 142	struct annotation *notes;
 143	struct map *map;
 144	int err = -1;
 145
 146	if (!syme)
 147		return -1;
 148
 149	sym = sym_entry__symbol(syme);
 150	map = syme->map;
 151
 152	/*
 153	 * We can't annotate with just /proc/kallsyms
 154	 */
 155	if (map->dso->symtab_type == SYMTAB__KALLSYMS) {
 
 156		pr_err("Can't annotate %s: No vmlinux file was found in the "
 157		       "path\n", sym->name);
 158		sleep(1);
 159		return -1;
 160	}
 161
 162	notes = symbol__annotation(sym);
 163	if (notes->src != NULL) {
 164		pthread_mutex_lock(&notes->lock);
 165		goto out_assign;
 166	}
 167
 168	pthread_mutex_lock(&notes->lock);
 169
 170	if (symbol__alloc_hist(sym, top.evlist->nr_entries) < 0) {
 171		pthread_mutex_unlock(&notes->lock);
 172		pr_err("Not enough memory for annotating '%s' symbol!\n",
 173		       sym->name);
 174		sleep(1);
 175		return err;
 176	}
 177
 178	err = symbol__annotate(sym, syme->map, 0);
 179	if (err == 0) {
 180out_assign:
 181		top.sym_filter_entry = syme;
 182	}
 183
 184	pthread_mutex_unlock(&notes->lock);
 185	return err;
 186}
 187
 188static void __zero_source_counters(struct sym_entry *syme)
 189{
 190	struct symbol *sym = sym_entry__symbol(syme);
 191	symbol__annotate_zero_histograms(sym);
 192}
 193
 194static void record_precise_ip(struct sym_entry *syme, struct map *map,
 195			      int counter, u64 ip)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 196{
 197	struct annotation *notes;
 198	struct symbol *sym;
 
 199
 200	if (syme != top.sym_filter_entry)
 
 
 201		return;
 202
 203	sym = sym_entry__symbol(syme);
 204	notes = symbol__annotation(sym);
 205
 206	if (pthread_mutex_trylock(&notes->lock))
 207		return;
 208
 209	ip = map->map_ip(map, ip);
 210	symbol__inc_addr_samples(sym, map, counter, ip);
 211
 212	pthread_mutex_unlock(&notes->lock);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 213}
 214
 215static void show_details(struct sym_entry *syme)
 216{
 
 217	struct annotation *notes;
 218	struct symbol *symbol;
 219	int more;
 220
 221	if (!syme)
 222		return;
 223
 224	symbol = sym_entry__symbol(syme);
 225	notes = symbol__annotation(symbol);
 226
 227	pthread_mutex_lock(&notes->lock);
 228
 229	if (notes->src == NULL)
 230		goto out_unlock;
 231
 232	printf("Showing %s for %s\n", event_name(top.sym_evsel), symbol->name);
 233	printf("  Events  Pcnt (>=%d%%)\n", sym_pcnt_filter);
 
 
 
 234
 235	more = symbol__annotate_printf(symbol, syme->map, top.sym_evsel->idx,
 236				       0, sym_pcnt_filter, top.print_entries, 4);
 237	if (top.zero)
 238		symbol__annotate_zero_histogram(symbol, top.sym_evsel->idx);
 239	else
 240		symbol__annotate_decay_histogram(symbol, top.sym_evsel->idx);
 241	if (more != 0)
 242		printf("%d lines not displayed, maybe increase display entries [e]\n", more);
 243out_unlock:
 244	pthread_mutex_unlock(&notes->lock);
 245}
 246
 247static const char		CONSOLE_CLEAR[] = "";
 248
 249static void __list_insert_active_sym(struct sym_entry *syme)
 250{
 251	list_add(&syme->node, &top.active_symbols);
 252}
 253
 254static void print_sym_table(struct perf_session *session)
 255{
 256	char bf[160];
 257	int printed = 0;
 258	struct rb_node *nd;
 259	struct sym_entry *syme;
 260	struct rb_root tmp = RB_ROOT;
 261	const int win_width = winsize.ws_col - 1;
 262	int sym_width, dso_width, dso_short_width;
 263	float sum_ksamples = perf_top__decay_samples(&top, &tmp);
 264
 265	puts(CONSOLE_CLEAR);
 266
 267	perf_top__header_snprintf(&top, bf, sizeof(bf));
 268	printf("%s\n", bf);
 269
 270	perf_top__reset_sample_counters(&top);
 271
 272	printf("%-*.*s\n", win_width, win_width, graph_dotted_line);
 273
 274	if (session->hists.stats.total_lost != 0) {
 275		color_fprintf(stdout, PERF_COLOR_RED, "WARNING:");
 276		printf(" LOST %" PRIu64 " events, Check IO/CPU overload\n",
 277		       session->hists.stats.total_lost);
 
 
 
 
 278	}
 279
 280	if (top.sym_filter_entry) {
 281		show_details(top.sym_filter_entry);
 282		return;
 283	}
 284
 285	perf_top__find_widths(&top, &tmp, &dso_width, &dso_short_width,
 286			      &sym_width);
 287
 288	if (sym_width + dso_width > winsize.ws_col - 29) {
 289		dso_width = dso_short_width;
 290		if (sym_width + dso_width > winsize.ws_col - 29)
 291			sym_width = winsize.ws_col - dso_width - 29;
 292	}
 293	putchar('\n');
 294	if (top.evlist->nr_entries == 1)
 295		printf("             samples  pcnt");
 296	else
 297		printf("   weight    samples  pcnt");
 298
 299	if (verbose)
 300		printf("         RIP       ");
 301	printf(" %-*.*s DSO\n", sym_width, sym_width, "function");
 302	printf("   %s    _______ _____",
 303	       top.evlist->nr_entries == 1 ? "      " : "______");
 304	if (verbose)
 305		printf(" ________________");
 306	printf(" %-*.*s", sym_width, sym_width, graph_line);
 307	printf(" %-*.*s", dso_width, dso_width, graph_line);
 308	puts("\n");
 309
 310	for (nd = rb_first(&tmp); nd; nd = rb_next(nd)) {
 311		struct symbol *sym;
 312		double pcnt;
 313
 314		syme = rb_entry(nd, struct sym_entry, rb_node);
 315		sym = sym_entry__symbol(syme);
 316		if (++printed > top.print_entries ||
 317		    (int)syme->snap_count < top.count_filter)
 318			continue;
 319
 320		pcnt = 100.0 - (100.0 * ((sum_ksamples - syme->snap_count) /
 321					 sum_ksamples));
 322
 323		if (top.evlist->nr_entries == 1 || !top.display_weighted)
 324			printf("%20.2f ", syme->weight);
 325		else
 326			printf("%9.1f %10ld ", syme->weight, syme->snap_count);
 327
 328		percent_color_fprintf(stdout, "%4.1f%%", pcnt);
 329		if (verbose)
 330			printf(" %016" PRIx64, sym->start);
 331		printf(" %-*.*s", sym_width, sym_width, sym->name);
 332		printf(" %-*.*s\n", dso_width, dso_width,
 333		       dso_width >= syme->map->dso->long_name_len ?
 334					syme->map->dso->long_name :
 335					syme->map->dso->short_name);
 336	}
 337}
 338
 339static void prompt_integer(int *target, const char *msg)
 340{
 341	char *buf = malloc(0), *p;
 342	size_t dummy = 0;
 343	int tmp;
 344
 345	fprintf(stdout, "\n%s: ", msg);
 346	if (getline(&buf, &dummy, stdin) < 0)
 347		return;
 348
 349	p = strchr(buf, '\n');
 350	if (p)
 351		*p = 0;
 352
 353	p = buf;
 354	while(*p) {
 355		if (!isdigit(*p))
 356			goto out_free;
 357		p++;
 358	}
 359	tmp = strtoul(buf, NULL, 10);
 360	*target = tmp;
 361out_free:
 362	free(buf);
 363}
 364
 365static void prompt_percent(int *target, const char *msg)
 366{
 367	int tmp = 0;
 368
 369	prompt_integer(&tmp, msg);
 370	if (tmp >= 0 && tmp <= 100)
 371		*target = tmp;
 372}
 373
 374static void prompt_symbol(struct sym_entry **target, const char *msg)
 375{
 376	char *buf = malloc(0), *p;
 377	struct sym_entry *syme = *target, *n, *found = NULL;
 
 
 378	size_t dummy = 0;
 379
 380	/* zero counters of active symbol */
 381	if (syme) {
 382		__zero_source_counters(syme);
 383		*target = NULL;
 384	}
 385
 386	fprintf(stdout, "\n%s: ", msg);
 387	if (getline(&buf, &dummy, stdin) < 0)
 388		goto out_free;
 389
 390	p = strchr(buf, '\n');
 391	if (p)
 392		*p = 0;
 393
 394	pthread_mutex_lock(&top.active_symbols_lock);
 395	syme = list_entry(top.active_symbols.next, struct sym_entry, node);
 396	pthread_mutex_unlock(&top.active_symbols_lock);
 397
 398	list_for_each_entry_safe_from(syme, n, &top.active_symbols, node) {
 399		struct symbol *sym = sym_entry__symbol(syme);
 400
 401		if (!strcmp(buf, sym->name)) {
 402			found = syme;
 403			break;
 404		}
 
 405	}
 406
 407	if (!found) {
 408		fprintf(stderr, "Sorry, %s is not active.\n", buf);
 409		sleep(1);
 410		return;
 411	} else
 412		parse_source(found);
 413
 414out_free:
 415	free(buf);
 416}
 417
 418static void print_mapped_keys(void)
 419{
 420	char *name = NULL;
 421
 422	if (top.sym_filter_entry) {
 423		struct symbol *sym = sym_entry__symbol(top.sym_filter_entry);
 424		name = sym->name;
 425	}
 426
 427	fprintf(stdout, "\nMapped keys:\n");
 428	fprintf(stdout, "\t[d]     display refresh delay.             \t(%d)\n", top.delay_secs);
 429	fprintf(stdout, "\t[e]     display entries (lines).           \t(%d)\n", top.print_entries);
 430
 431	if (top.evlist->nr_entries > 1)
 432		fprintf(stdout, "\t[E]     active event counter.              \t(%s)\n", event_name(top.sym_evsel));
 433
 434	fprintf(stdout, "\t[f]     profile display filter (count).    \t(%d)\n", top.count_filter);
 435
 436	fprintf(stdout, "\t[F]     annotate display filter (percent). \t(%d%%)\n", sym_pcnt_filter);
 437	fprintf(stdout, "\t[s]     annotate symbol.                   \t(%s)\n", name?: "NULL");
 438	fprintf(stdout, "\t[S]     stop annotation.\n");
 439
 440	if (top.evlist->nr_entries > 1)
 441		fprintf(stdout, "\t[w]     toggle display weighted/count[E]r. \t(%d)\n", top.display_weighted ? 1 : 0);
 442
 443	fprintf(stdout,
 444		"\t[K]     hide kernel_symbols symbols.     \t(%s)\n",
 445		top.hide_kernel_symbols ? "yes" : "no");
 446	fprintf(stdout,
 447		"\t[U]     hide user symbols.               \t(%s)\n",
 448		top.hide_user_symbols ? "yes" : "no");
 449	fprintf(stdout, "\t[z]     toggle sample zeroing.             \t(%d)\n", top.zero ? 1 : 0);
 450	fprintf(stdout, "\t[qQ]    quit.\n");
 451}
 452
 453static int key_mapped(int c)
 454{
 455	switch (c) {
 456		case 'd':
 457		case 'e':
 458		case 'f':
 459		case 'z':
 460		case 'q':
 461		case 'Q':
 462		case 'K':
 463		case 'U':
 464		case 'F':
 465		case 's':
 466		case 'S':
 467			return 1;
 468		case 'E':
 469		case 'w':
 470			return top.evlist->nr_entries > 1 ? 1 : 0;
 471		default:
 472			break;
 473	}
 474
 475	return 0;
 476}
 477
 478static void handle_keypress(struct perf_session *session, int c)
 479{
 480	if (!key_mapped(c)) {
 
 
 481		struct pollfd stdin_poll = { .fd = 0, .events = POLLIN };
 482		struct termios tc, save;
 483
 484		print_mapped_keys();
 485		fprintf(stdout, "\nEnter selection, or unmapped key to continue: ");
 486		fflush(stdout);
 487
 488		tcgetattr(0, &save);
 489		tc = save;
 490		tc.c_lflag &= ~(ICANON | ECHO);
 491		tc.c_cc[VMIN] = 0;
 492		tc.c_cc[VTIME] = 0;
 493		tcsetattr(0, TCSANOW, &tc);
 494
 495		poll(&stdin_poll, 1, -1);
 496		c = getc(stdin);
 497
 498		tcsetattr(0, TCSAFLUSH, &save);
 499		if (!key_mapped(c))
 500			return;
 501	}
 502
 503	switch (c) {
 504		case 'd':
 505			prompt_integer(&top.delay_secs, "Enter display delay");
 506			if (top.delay_secs < 1)
 507				top.delay_secs = 1;
 508			break;
 509		case 'e':
 510			prompt_integer(&top.print_entries, "Enter display entries (lines)");
 511			if (top.print_entries == 0) {
 512				sig_winch_handler(SIGWINCH);
 513				signal(SIGWINCH, sig_winch_handler);
 514			} else
 
 
 
 
 515				signal(SIGWINCH, SIG_DFL);
 
 516			break;
 517		case 'E':
 518			if (top.evlist->nr_entries > 1) {
 519				/* Select 0 as the default event: */
 520				int counter = 0;
 521
 522				fprintf(stderr, "\nAvailable events:");
 523
 524				list_for_each_entry(top.sym_evsel, &top.evlist->entries, node)
 525					fprintf(stderr, "\n\t%d %s", top.sym_evsel->idx, event_name(top.sym_evsel));
 526
 527				prompt_integer(&counter, "Enter details event counter");
 528
 529				if (counter >= top.evlist->nr_entries) {
 530					top.sym_evsel = list_entry(top.evlist->entries.next, struct perf_evsel, node);
 531					fprintf(stderr, "Sorry, no such event, using %s.\n", event_name(top.sym_evsel));
 532					sleep(1);
 533					break;
 534				}
 535				list_for_each_entry(top.sym_evsel, &top.evlist->entries, node)
 536					if (top.sym_evsel->idx == counter)
 537						break;
 538			} else
 539				top.sym_evsel = list_entry(top.evlist->entries.next, struct perf_evsel, node);
 540			break;
 541		case 'f':
 542			prompt_integer(&top.count_filter, "Enter display event count filter");
 543			break;
 544		case 'F':
 545			prompt_percent(&sym_pcnt_filter, "Enter details display event filter (percent)");
 
 546			break;
 547		case 'K':
 548			top.hide_kernel_symbols = !top.hide_kernel_symbols;
 549			break;
 550		case 'q':
 551		case 'Q':
 552			printf("exiting.\n");
 553			if (dump_symtab)
 554				perf_session__fprintf_dsos(session, stderr);
 555			exit(0);
 
 556		case 's':
 557			prompt_symbol(&top.sym_filter_entry, "Enter details symbol");
 558			break;
 559		case 'S':
 560			if (!top.sym_filter_entry)
 561				break;
 562			else {
 563				struct sym_entry *syme = top.sym_filter_entry;
 564
 565				top.sym_filter_entry = NULL;
 566				__zero_source_counters(syme);
 567			}
 568			break;
 569		case 'U':
 570			top.hide_user_symbols = !top.hide_user_symbols;
 571			break;
 572		case 'w':
 573			top.display_weighted = ~top.display_weighted;
 574			break;
 575		case 'z':
 576			top.zero = !top.zero;
 577			break;
 578		default:
 579			break;
 580	}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 581}
 582
 583static void *display_thread_tui(void *arg __used)
 584{
 585	int err = 0;
 586	pthread_mutex_lock(&top.active_symbols_lock);
 587	while (list_empty(&top.active_symbols)) {
 588		err = pthread_cond_wait(&top.active_symbols_cond,
 589					&top.active_symbols_lock);
 590		if (err)
 591			break;
 
 
 
 
 
 
 
 
 
 
 
 
 592	}
 593	pthread_mutex_unlock(&top.active_symbols_lock);
 594	if (!err)
 595		perf_top__tui_browser(&top);
 596	exit_browser(0);
 597	exit(0);
 
 598	return NULL;
 599}
 600
 601static void *display_thread(void *arg __used)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 602{
 603	struct pollfd stdin_poll = { .fd = 0, .events = POLLIN };
 604	struct termios tc, save;
 
 605	int delay_msecs, c;
 606	struct perf_session *session = (struct perf_session *) arg;
 607
 608	tcgetattr(0, &save);
 609	tc = save;
 610	tc.c_lflag &= ~(ICANON | ECHO);
 611	tc.c_cc[VMIN] = 0;
 612	tc.c_cc[VTIME] = 0;
 613
 
 
 614repeat:
 615	delay_msecs = top.delay_secs * 1000;
 616	tcsetattr(0, TCSANOW, &tc);
 617	/* trash return*/
 618	getc(stdin);
 619
 620	do {
 621		print_sym_table(session);
 622	} while (!poll(&stdin_poll, 1, delay_msecs) == 1);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 623
 624	c = getc(stdin);
 625	tcsetattr(0, TCSAFLUSH, &save);
 626
 627	handle_keypress(session, c);
 628	goto repeat;
 629
 630	return NULL;
 631}
 632
 633/* Tag samples to be skipped. */
 634static const char *skip_symbols[] = {
 635	"default_idle",
 636	"native_safe_halt",
 637	"cpu_idle",
 638	"enter_idle",
 639	"exit_idle",
 640	"mwait_idle",
 641	"mwait_idle_with_hints",
 642	"poll_idle",
 643	"ppc64_runlatch_off",
 644	"pseries_dedicated_idle_sleep",
 645	NULL
 646};
 647
 648static int symbol_filter(struct map *map, struct symbol *sym)
 649{
 650	struct sym_entry *syme;
 651	const char *name = sym->name;
 652	int i;
 653
 
 
 654	/*
 655	 * ppc64 uses function descriptors and appends a '.' to the
 656	 * start of every instruction address. Remove it.
 657	 */
 658	if (name[0] == '.')
 659		name++;
 660
 661	if (!strcmp(name, "_text") ||
 662	    !strcmp(name, "_etext") ||
 663	    !strcmp(name, "_sinittext") ||
 664	    !strncmp("init_module", name, 11) ||
 665	    !strncmp("cleanup_module", name, 14) ||
 666	    strstr(name, "_text_start") ||
 667	    strstr(name, "_text_end"))
 668		return 1;
 669
 670	syme = symbol__priv(sym);
 671	syme->map = map;
 672	symbol__annotate_init(map, sym);
 673
 674	if (!top.sym_filter_entry && sym_filter && !strcmp(name, sym_filter)) {
 675		/* schedule initial sym_filter_entry setup */
 676		sym_filter_entry_sched = syme;
 677		sym_filter = NULL;
 678	}
 
 
 
 
 679
 680	for (i = 0; skip_symbols[i]; i++) {
 681		if (!strcmp(skip_symbols[i], name)) {
 682			sym->ignore = true;
 683			break;
 684		}
 685	}
 686
 
 
 687	return 0;
 688}
 689
 690static void perf_event__process_sample(const union perf_event *event,
 
 
 691				       struct perf_sample *sample,
 692				       struct perf_session *session)
 693{
 694	u64 ip = event->ip.ip;
 695	struct sym_entry *syme;
 696	struct addr_location al;
 697	struct machine *machine;
 698	u8 origin = event->header.misc & PERF_RECORD_MISC_CPUMODE_MASK;
 
 
 699
 700	++top.samples;
 
 701
 702	switch (origin) {
 703	case PERF_RECORD_MISC_USER:
 704		++top.us_samples;
 705		if (top.hide_user_symbols)
 706			return;
 707		machine = perf_session__find_host_machine(session);
 708		break;
 709	case PERF_RECORD_MISC_KERNEL:
 710		++top.kernel_samples;
 711		if (top.hide_kernel_symbols)
 712			return;
 713		machine = perf_session__find_host_machine(session);
 714		break;
 715	case PERF_RECORD_MISC_GUEST_KERNEL:
 716		++top.guest_kernel_samples;
 717		machine = perf_session__find_machine(session, event->ip.pid);
 718		break;
 719	case PERF_RECORD_MISC_GUEST_USER:
 720		++top.guest_us_samples;
 721		/*
 722		 * TODO: we don't process guest user from host side
 723		 * except simple counting.
 724		 */
 725		return;
 726	default:
 727		return;
 728	}
 729
 730	if (!machine && perf_guest) {
 731		pr_err("Can't find guest [%d]'s kernel information\n",
 732			event->ip.pid);
 733		return;
 734	}
 735
 736	if (event->header.misc & PERF_RECORD_MISC_EXACT_IP)
 737		top.exact_samples++;
 738
 739	if (perf_event__preprocess_sample(event, session, &al, sample,
 740					  symbol_filter) < 0 ||
 741	    al.filtered)
 742		return;
 743
 744	if (!kptr_restrict_warned &&
 745	    symbol_conf.kptr_restrict &&
 746	    al.cpumode == PERF_RECORD_MISC_KERNEL) {
 747		ui__warning(
 748"Kernel address maps (/proc/{kallsyms,modules}) are restricted.\n\n"
 749"Check /proc/sys/kernel/kptr_restrict.\n\n"
 750"Kernel%s samples will not be resolved.\n",
 751			  !RB_EMPTY_ROOT(&al.map->dso->symbols[MAP__FUNCTION]) ?
 752			  " modules" : "");
 753		if (use_browser <= 0)
 754			sleep(5);
 755		kptr_restrict_warned = true;
 756	}
 757
 758	if (al.sym == NULL) {
 759		const char *msg = "Kernel samples will not be resolved.\n";
 760		/*
 761		 * As we do lazy loading of symtabs we only will know if the
 762		 * specified vmlinux file is invalid when we actually have a
 763		 * hit in kernel space and then try to load it. So if we get
 764		 * here and there are _no_ symbols in the DSO backing the
 765		 * kernel map, bail out.
 766		 *
 767		 * We may never get here, for instance, if we use -K/
 768		 * --hide-kernel-symbols, even if the user specifies an
 769		 * invalid --vmlinux ;-)
 770		 */
 771		if (!kptr_restrict_warned && !vmlinux_warned &&
 772		    al.map == machine->vmlinux_maps[MAP__FUNCTION] &&
 773		    RB_EMPTY_ROOT(&al.map->dso->symbols[MAP__FUNCTION])) {
 774			if (symbol_conf.vmlinux_name) {
 775				ui__warning("The %s file can't be used.\n%s",
 776					    symbol_conf.vmlinux_name, msg);
 
 
 777			} else {
 778				ui__warning("A vmlinux file was not found.\n%s",
 779					    msg);
 780			}
 781
 782			if (use_browser <= 0)
 783				sleep(5);
 784			vmlinux_warned = true;
 785		}
 
 786
 787		return;
 788	}
 
 
 
 
 
 
 
 
 
 
 789
 790	/* let's see, whether we need to install initial sym_filter_entry */
 791	if (sym_filter_entry_sched) {
 792		top.sym_filter_entry = sym_filter_entry_sched;
 793		sym_filter_entry_sched = NULL;
 794		if (parse_source(top.sym_filter_entry) < 0) {
 795			struct symbol *sym = sym_entry__symbol(top.sym_filter_entry);
 796
 797			pr_err("Can't annotate %s", sym->name);
 798			if (top.sym_filter_entry->map->dso->symtab_type == SYMTAB__KALLSYMS) {
 799				pr_err(": No vmlinux file was found in the path:\n");
 800				machine__fprintf_vmlinux_path(machine, stderr);
 801			} else
 802				pr_err(".\n");
 803			exit(1);
 804		}
 805	}
 806
 807	syme = symbol__priv(al.sym);
 808	if (!al.sym->ignore) {
 809		struct perf_evsel *evsel;
 810
 811		evsel = perf_evlist__id2evsel(top.evlist, sample->id);
 812		assert(evsel != NULL);
 813		syme->count[evsel->idx]++;
 814		record_precise_ip(syme, al.map, evsel->idx, ip);
 815		pthread_mutex_lock(&top.active_symbols_lock);
 816		if (list_empty(&syme->node) || !syme->node.next) {
 817			static bool first = true;
 818			__list_insert_active_sym(syme);
 819			if (first) {
 820				pthread_cond_broadcast(&top.active_symbols_cond);
 821				first = false;
 822			}
 823		}
 824		pthread_mutex_unlock(&top.active_symbols_lock);
 825	}
 
 
 826}
 827
 828static void perf_session__mmap_read_idx(struct perf_session *self, int idx)
 829{
 830	struct perf_sample sample;
 
 
 831	union perf_event *event;
 
 832	int ret;
 833
 834	while ((event = perf_evlist__mmap_read(top.evlist, idx)) != NULL) {
 835		ret = perf_session__parse_sample(self, event, &sample);
 836		if (ret) {
 837			pr_err("Can't parse sample, err = %d\n", ret);
 838			continue;
 839		}
 840
 
 
 
 841		if (event->header.type == PERF_RECORD_SAMPLE)
 842			perf_event__process_sample(event, &sample, self);
 843		else
 844			perf_event__process(event, &sample, self);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 845	}
 846}
 847
 848static void perf_session__mmap_read(struct perf_session *self)
 849{
 850	int i;
 851
 852	for (i = 0; i < top.evlist->nr_mmaps; i++)
 853		perf_session__mmap_read_idx(self, i);
 854}
 855
 856static void start_counters(struct perf_evlist *evlist)
 857{
 
 858	struct perf_evsel *counter;
 
 
 859
 860	list_for_each_entry(counter, &evlist->entries, node) {
 861		struct perf_event_attr *attr = &counter->attr;
 862
 863		attr->sample_type = PERF_SAMPLE_IP | PERF_SAMPLE_TID;
 864
 865		if (top.freq) {
 866			attr->sample_type |= PERF_SAMPLE_PERIOD;
 867			attr->freq	  = 1;
 868			attr->sample_freq = top.freq;
 869		}
 870
 871		if (evlist->nr_entries > 1) {
 872			attr->sample_type |= PERF_SAMPLE_ID;
 873			attr->read_format |= PERF_FORMAT_ID;
 874		}
 875
 876		attr->mmap = 1;
 877		attr->inherit = inherit;
 878try_again:
 879		if (perf_evsel__open(counter, top.evlist->cpus,
 880				     top.evlist->threads, group) < 0) {
 881			int err = errno;
 882
 883			if (err == EPERM || err == EACCES) {
 884				ui__warning_paranoid();
 885				goto out_err;
 886			}
 887			/*
 888			 * If it's cycles then fall back to hrtimer
 889			 * based cpu-clock-tick sw counter, which
 890			 * is always available even if no PMU support:
 891			 */
 892			if (attr->type == PERF_TYPE_HARDWARE &&
 893			    attr->config == PERF_COUNT_HW_CPU_CYCLES) {
 894				if (verbose)
 895					ui__warning("Cycles event not supported,\n"
 896						    "trying to fall back to cpu-clock-ticks\n");
 897
 898				attr->type = PERF_TYPE_SOFTWARE;
 899				attr->config = PERF_COUNT_SW_CPU_CLOCK;
 900				goto try_again;
 901			}
 902
 903			if (err == ENOENT) {
 904				ui__warning("The %s event is not supported.\n",
 905					    event_name(counter));
 906				goto out_err;
 907			}
 908
 909			ui__warning("The sys_perf_event_open() syscall "
 910				    "returned with %d (%s).  /bin/dmesg "
 911				    "may provide additional information.\n"
 912				    "No CONFIG_PERF_EVENTS=y kernel support "
 913				    "configured?\n", err, strerror(err));
 914			goto out_err;
 915		}
 916	}
 917
 918	if (perf_evlist__mmap(evlist, mmap_pages, false) < 0) {
 919		ui__warning("Failed to mmap with %d (%s)\n",
 920			    errno, strerror(errno));
 921		goto out_err;
 922	}
 923
 924	return;
 925
 926out_err:
 927	exit_browser(0);
 928	exit(0);
 929}
 930
 931static int __cmd_top(void)
 932{
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 933	pthread_t thread;
 934	int ret __used;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 935	/*
 936	 * FIXME: perf_session__new should allow passing a O_MMAP, so that all this
 937	 * mmap reading, etc is encapsulated in it. Use O_WRONLY for now.
 
 
 
 
 938	 */
 939	struct perf_session *session = perf_session__new(NULL, O_WRONLY, false, false, NULL);
 940	if (session == NULL)
 941		return -ENOMEM;
 942
 943	if (top.target_tid != -1)
 944		perf_event__synthesize_thread_map(top.evlist->threads,
 945						  perf_event__process, session);
 946	else
 947		perf_event__synthesize_threads(perf_event__process, session);
 948
 949	start_counters(top.evlist);
 950	session->evlist = top.evlist;
 951	perf_session__update_sample_type(session);
 952
 953	/* Wait for a minimal set of events before starting the snapshot */
 954	poll(top.evlist->pollfd, top.evlist->nr_fds, 100);
 955
 956	perf_session__mmap_read(session);
 957
 
 958	if (pthread_create(&thread, NULL, (use_browser > 0 ? display_thread_tui :
 959							     display_thread), session)) {
 960		printf("Could not create display thread.\n");
 961		exit(-1);
 962	}
 963
 964	if (realtime_prio) {
 965		struct sched_param param;
 966
 967		param.sched_priority = realtime_prio;
 968		if (sched_setscheduler(0, SCHED_FIFO, &param)) {
 969			printf("Could not set realtime priority.\n");
 970			exit(-1);
 971		}
 972	}
 973
 974	while (1) {
 975		u64 hits = top.samples;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 976
 977		perf_session__mmap_read(session);
 
 978
 979		if (hits == top.samples)
 980			ret = poll(top.evlist->pollfd, top.evlist->nr_fds, 100);
 
 
 
 
 
 981	}
 982
 983	return 0;
 984}
 985
 986static const char * const top_usage[] = {
 987	"perf top [<options>]",
 988	NULL
 989};
 
 
 
 
 
 
 
 
 990
 991static const struct option options[] = {
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 992	OPT_CALLBACK('e', "event", &top.evlist, "event",
 993		     "event selector. use 'perf list' to list available events",
 994		     parse_events_option),
 995	OPT_INTEGER('c', "count", &default_interval,
 996		    "event period to sample"),
 997	OPT_INTEGER('p', "pid", &top.target_pid,
 998		    "profile events on existing process id"),
 999	OPT_INTEGER('t', "tid", &top.target_tid,
1000		    "profile events on existing thread id"),
1001	OPT_BOOLEAN('a', "all-cpus", &system_wide,
1002			    "system-wide collection from all CPUs"),
1003	OPT_STRING('C', "cpu", &top.cpu_list, "cpu",
1004		    "list of cpus to monitor"),
1005	OPT_STRING('k', "vmlinux", &symbol_conf.vmlinux_name,
1006		   "file", "vmlinux pathname"),
 
 
1007	OPT_BOOLEAN('K', "hide_kernel_symbols", &top.hide_kernel_symbols,
1008		    "hide kernel symbols"),
1009	OPT_UINTEGER('m', "mmap-pages", &mmap_pages, "number of mmap data pages"),
1010	OPT_INTEGER('r', "realtime", &realtime_prio,
 
 
1011		    "collect data with this RT SCHED_FIFO priority"),
1012	OPT_INTEGER('d', "delay", &top.delay_secs,
1013		    "number of seconds to delay between refreshes"),
1014	OPT_BOOLEAN('D', "dump-symtab", &dump_symtab,
1015			    "dump the symbol table used for profiling"),
1016	OPT_INTEGER('f', "count-filter", &top.count_filter,
1017		    "only display functions with more events than this"),
1018	OPT_BOOLEAN('g', "group", &group,
1019			    "put the counters into a counter group"),
1020	OPT_BOOLEAN('i', "inherit", &inherit,
1021		    "child tasks inherit counters"),
1022	OPT_STRING('s', "sym-annotate", &sym_filter, "symbol name",
1023		    "symbol to annotate"),
1024	OPT_BOOLEAN('z', "zero", &top.zero,
1025		    "zero history across updates"),
1026	OPT_INTEGER('F', "freq", &top.freq,
1027		    "profile at this frequency"),
1028	OPT_INTEGER('E', "entries", &top.print_entries,
1029		    "display this many functions"),
1030	OPT_BOOLEAN('U', "hide_user_symbols", &top.hide_user_symbols,
1031		    "hide user symbols"),
1032	OPT_BOOLEAN(0, "tui", &use_tui, "Use the TUI interface"),
1033	OPT_BOOLEAN(0, "stdio", &use_stdio, "Use the stdio interface"),
1034	OPT_INCR('v', "verbose", &verbose,
1035		    "be more verbose (show counter open errors, etc)"),
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1036	OPT_END()
1037};
 
 
 
 
 
1038
1039int cmd_top(int argc, const char **argv, const char *prefix __used)
1040{
1041	struct perf_evsel *pos;
1042	int status = -ENOMEM;
1043
1044	top.evlist = perf_evlist__new(NULL, NULL);
1045	if (top.evlist == NULL)
1046		return -ENOMEM;
1047
1048	page_size = sysconf(_SC_PAGE_SIZE);
1049
1050	argc = parse_options(argc, argv, options, top_usage, 0);
1051	if (argc)
1052		usage_with_options(top_usage, options);
1053
1054	/*
1055 	 * XXX For now start disabled, only using TUI if explicitely asked for.
1056 	 * Change that when handle_keys equivalent gets written, live annotation
1057 	 * done, etc.
1058 	 */
1059	use_browser = 0;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1060
1061	if (use_stdio)
1062		use_browser = 0;
1063	else if (use_tui)
1064		use_browser = 1;
1065
1066	setup_browser(false);
1067
1068	/* CPU and PID are mutually exclusive */
1069	if (top.target_tid > 0 && top.cpu_list) {
1070		printf("WARNING: PID switch overriding CPU\n");
1071		sleep(1);
1072		top.cpu_list = NULL;
 
 
 
 
 
 
 
 
1073	}
1074
1075	if (top.target_pid != -1)
1076		top.target_tid = top.target_pid;
 
 
 
 
1077
1078	if (perf_evlist__create_maps(top.evlist, top.target_pid,
1079				     top.target_tid, top.cpu_list) < 0)
1080		usage_with_options(top_usage, options);
 
 
 
1081
1082	if (!top.evlist->nr_entries &&
1083	    perf_evlist__add_default(top.evlist) < 0) {
1084		pr_err("Not enough memory for event selector list\n");
1085		return -ENOMEM;
1086	}
1087
 
 
1088	if (top.delay_secs < 1)
1089		top.delay_secs = 1;
1090
1091	/*
1092	 * User specified count overrides default frequency.
1093	 */
1094	if (default_interval)
1095		top.freq = 0;
1096	else if (top.freq) {
1097		default_interval = top.freq;
1098	} else {
1099		fprintf(stderr, "frequency and count are zero, aborting\n");
1100		exit(EXIT_FAILURE);
1101	}
1102
1103	list_for_each_entry(pos, &top.evlist->entries, node) {
1104		if (perf_evsel__alloc_fd(pos, top.evlist->cpus->nr,
1105					 top.evlist->threads->nr) < 0)
1106			goto out_free_fd;
1107		/*
1108		 * Fill in the ones not specifically initialized via -c:
1109		 */
1110		if (pos->attr.sample_period)
1111			continue;
1112
1113		pos->attr.sample_period = default_interval;
 
 
1114	}
1115
1116	if (perf_evlist__alloc_pollfd(top.evlist) < 0 ||
1117	    perf_evlist__alloc_mmap(top.evlist) < 0)
1118		goto out_free_fd;
1119
1120	top.sym_evsel = list_entry(top.evlist->entries.next, struct perf_evsel, node);
1121
1122	symbol_conf.priv_size = (sizeof(struct sym_entry) + sizeof(struct annotation) +
1123				 (top.evlist->nr_entries + 1) * sizeof(unsigned long));
1124
1125	symbol_conf.try_vmlinux_path = (symbol_conf.vmlinux_name == NULL);
1126	if (symbol__init() < 0)
1127		return -1;
1128
1129	get_term_dimensions(&winsize);
 
 
1130	if (top.print_entries == 0) {
1131		update_print_entries(&winsize);
1132		signal(SIGWINCH, sig_winch_handler);
 
 
 
 
1133	}
1134
1135	status = __cmd_top();
1136out_free_fd:
 
1137	perf_evlist__delete(top.evlist);
1138
1139	return status;
1140}