xref: /openbmc/linux/tools/perf/builtin-top.c (revision 5104d265)
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 "util/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 
44 #include "util/debug.h"
45 
46 #include <assert.h>
47 #include <elf.h>
48 #include <fcntl.h>
49 
50 #include <stdio.h>
51 #include <termios.h>
52 #include <unistd.h>
53 #include <inttypes.h>
54 
55 #include <errno.h>
56 #include <time.h>
57 #include <sched.h>
58 
59 #include <sys/syscall.h>
60 #include <sys/ioctl.h>
61 #include <sys/poll.h>
62 #include <sys/prctl.h>
63 #include <sys/wait.h>
64 #include <sys/uio.h>
65 #include <sys/utsname.h>
66 #include <sys/mman.h>
67 
68 #include <linux/unistd.h>
69 #include <linux/types.h>
70 
71 static volatile int done;
72 
73 #define HEADER_LINE_NR  5
74 
75 static void perf_top__update_print_entries(struct perf_top *top)
76 {
77 	top->print_entries = top->winsize.ws_row - HEADER_LINE_NR;
78 }
79 
80 static void perf_top__sig_winch(int sig __maybe_unused,
81 				siginfo_t *info __maybe_unused, void *arg)
82 {
83 	struct perf_top *top = arg;
84 
85 	get_term_dimensions(&top->winsize);
86 	perf_top__update_print_entries(top);
87 }
88 
89 static int perf_top__parse_source(struct perf_top *top, struct hist_entry *he)
90 {
91 	struct symbol *sym;
92 	struct annotation *notes;
93 	struct map *map;
94 	int err = -1;
95 
96 	if (!he || !he->ms.sym)
97 		return -1;
98 
99 	sym = he->ms.sym;
100 	map = he->ms.map;
101 
102 	/*
103 	 * We can't annotate with just /proc/kallsyms
104 	 */
105 	if (map->dso->symtab_type == DSO_BINARY_TYPE__KALLSYMS) {
106 		pr_err("Can't annotate %s: No vmlinux file was found in the "
107 		       "path\n", sym->name);
108 		sleep(1);
109 		return -1;
110 	}
111 
112 	notes = symbol__annotation(sym);
113 	if (notes->src != NULL) {
114 		pthread_mutex_lock(&notes->lock);
115 		goto out_assign;
116 	}
117 
118 	pthread_mutex_lock(&notes->lock);
119 
120 	if (symbol__alloc_hist(sym) < 0) {
121 		pthread_mutex_unlock(&notes->lock);
122 		pr_err("Not enough memory for annotating '%s' symbol!\n",
123 		       sym->name);
124 		sleep(1);
125 		return err;
126 	}
127 
128 	err = symbol__annotate(sym, map, 0);
129 	if (err == 0) {
130 out_assign:
131 		top->sym_filter_entry = he;
132 	}
133 
134 	pthread_mutex_unlock(&notes->lock);
135 	return err;
136 }
137 
138 static void __zero_source_counters(struct hist_entry *he)
139 {
140 	struct symbol *sym = he->ms.sym;
141 	symbol__annotate_zero_histograms(sym);
142 }
143 
144 static void ui__warn_map_erange(struct map *map, struct symbol *sym, u64 ip)
145 {
146 	struct utsname uts;
147 	int err = uname(&uts);
148 
149 	ui__warning("Out of bounds address found:\n\n"
150 		    "Addr:   %" PRIx64 "\n"
151 		    "DSO:    %s %c\n"
152 		    "Map:    %" PRIx64 "-%" PRIx64 "\n"
153 		    "Symbol: %" PRIx64 "-%" PRIx64 " %c %s\n"
154 		    "Arch:   %s\n"
155 		    "Kernel: %s\n"
156 		    "Tools:  %s\n\n"
157 		    "Not all samples will be on the annotation output.\n\n"
158 		    "Please report to linux-kernel@vger.kernel.org\n",
159 		    ip, map->dso->long_name, dso__symtab_origin(map->dso),
160 		    map->start, map->end, sym->start, sym->end,
161 		    sym->binding == STB_GLOBAL ? 'g' :
162 		    sym->binding == STB_LOCAL  ? 'l' : 'w', sym->name,
163 		    err ? "[unknown]" : uts.machine,
164 		    err ? "[unknown]" : uts.release, perf_version_string);
165 	if (use_browser <= 0)
166 		sleep(5);
167 
168 	map->erange_warned = true;
169 }
170 
171 static void perf_top__record_precise_ip(struct perf_top *top,
172 					struct hist_entry *he,
173 					int counter, u64 ip)
174 {
175 	struct annotation *notes;
176 	struct symbol *sym;
177 	int err;
178 
179 	if (he == NULL || he->ms.sym == NULL ||
180 	    ((top->sym_filter_entry == NULL ||
181 	      top->sym_filter_entry->ms.sym != he->ms.sym) && use_browser != 1))
182 		return;
183 
184 	sym = he->ms.sym;
185 	notes = symbol__annotation(sym);
186 
187 	if (pthread_mutex_trylock(&notes->lock))
188 		return;
189 
190 	if (notes->src == NULL && symbol__alloc_hist(sym) < 0) {
191 		pthread_mutex_unlock(&notes->lock);
192 		pr_err("Not enough memory for annotating '%s' symbol!\n",
193 		       sym->name);
194 		sleep(1);
195 		return;
196 	}
197 
198 	ip = he->ms.map->map_ip(he->ms.map, ip);
199 	err = symbol__inc_addr_samples(sym, he->ms.map, counter, ip);
200 
201 	pthread_mutex_unlock(&notes->lock);
202 
203 	if (err == -ERANGE && !he->ms.map->erange_warned)
204 		ui__warn_map_erange(he->ms.map, sym, ip);
205 }
206 
207 static void perf_top__show_details(struct perf_top *top)
208 {
209 	struct hist_entry *he = top->sym_filter_entry;
210 	struct annotation *notes;
211 	struct symbol *symbol;
212 	int more;
213 
214 	if (!he)
215 		return;
216 
217 	symbol = he->ms.sym;
218 	notes = symbol__annotation(symbol);
219 
220 	pthread_mutex_lock(&notes->lock);
221 
222 	if (notes->src == NULL)
223 		goto out_unlock;
224 
225 	printf("Showing %s for %s\n", perf_evsel__name(top->sym_evsel), symbol->name);
226 	printf("  Events  Pcnt (>=%d%%)\n", top->sym_pcnt_filter);
227 
228 	more = symbol__annotate_printf(symbol, he->ms.map, top->sym_evsel,
229 				       0, top->sym_pcnt_filter, top->print_entries, 4);
230 	if (top->zero)
231 		symbol__annotate_zero_histogram(symbol, top->sym_evsel->idx);
232 	else
233 		symbol__annotate_decay_histogram(symbol, top->sym_evsel->idx);
234 	if (more != 0)
235 		printf("%d lines not displayed, maybe increase display entries [e]\n", more);
236 out_unlock:
237 	pthread_mutex_unlock(&notes->lock);
238 }
239 
240 static const char		CONSOLE_CLEAR[] = "";
241 
242 static struct hist_entry *perf_evsel__add_hist_entry(struct perf_evsel *evsel,
243 						     struct addr_location *al,
244 						     struct perf_sample *sample)
245 {
246 	struct hist_entry *he;
247 
248 	pthread_mutex_lock(&evsel->hists.lock);
249 	he = __hists__add_entry(&evsel->hists, al, NULL, sample->period,
250 				sample->weight);
251 	pthread_mutex_unlock(&evsel->hists.lock);
252 
253 	if (he == NULL)
254 		return NULL;
255 
256 	hists__inc_nr_events(&evsel->hists, PERF_RECORD_SAMPLE);
257 	return he;
258 }
259 
260 static void perf_top__print_sym_table(struct perf_top *top)
261 {
262 	char bf[160];
263 	int printed = 0;
264 	const int win_width = top->winsize.ws_col - 1;
265 
266 	puts(CONSOLE_CLEAR);
267 
268 	perf_top__header_snprintf(top, bf, sizeof(bf));
269 	printf("%s\n", bf);
270 
271 	perf_top__reset_sample_counters(top);
272 
273 	printf("%-*.*s\n", win_width, win_width, graph_dotted_line);
274 
275 	if (top->sym_evsel->hists.stats.nr_lost_warned !=
276 	    top->sym_evsel->hists.stats.nr_events[PERF_RECORD_LOST]) {
277 		top->sym_evsel->hists.stats.nr_lost_warned =
278 			top->sym_evsel->hists.stats.nr_events[PERF_RECORD_LOST];
279 		color_fprintf(stdout, PERF_COLOR_RED,
280 			      "WARNING: LOST %d chunks, Check IO/CPU overload",
281 			      top->sym_evsel->hists.stats.nr_lost_warned);
282 		++printed;
283 	}
284 
285 	if (top->sym_filter_entry) {
286 		perf_top__show_details(top);
287 		return;
288 	}
289 
290 	hists__collapse_resort(&top->sym_evsel->hists);
291 	hists__output_resort(&top->sym_evsel->hists);
292 	hists__decay_entries(&top->sym_evsel->hists,
293 			     top->hide_user_symbols,
294 			     top->hide_kernel_symbols);
295 	hists__output_recalc_col_len(&top->sym_evsel->hists,
296 				     top->print_entries - printed);
297 	putchar('\n');
298 	hists__fprintf(&top->sym_evsel->hists, false,
299 		       top->print_entries - printed, win_width,
300 		       top->min_percent, stdout);
301 }
302 
303 static void prompt_integer(int *target, const char *msg)
304 {
305 	char *buf = malloc(0), *p;
306 	size_t dummy = 0;
307 	int tmp;
308 
309 	fprintf(stdout, "\n%s: ", msg);
310 	if (getline(&buf, &dummy, stdin) < 0)
311 		return;
312 
313 	p = strchr(buf, '\n');
314 	if (p)
315 		*p = 0;
316 
317 	p = buf;
318 	while(*p) {
319 		if (!isdigit(*p))
320 			goto out_free;
321 		p++;
322 	}
323 	tmp = strtoul(buf, NULL, 10);
324 	*target = tmp;
325 out_free:
326 	free(buf);
327 }
328 
329 static void prompt_percent(int *target, const char *msg)
330 {
331 	int tmp = 0;
332 
333 	prompt_integer(&tmp, msg);
334 	if (tmp >= 0 && tmp <= 100)
335 		*target = tmp;
336 }
337 
338 static void perf_top__prompt_symbol(struct perf_top *top, const char *msg)
339 {
340 	char *buf = malloc(0), *p;
341 	struct hist_entry *syme = top->sym_filter_entry, *n, *found = NULL;
342 	struct rb_node *next;
343 	size_t dummy = 0;
344 
345 	/* zero counters of active symbol */
346 	if (syme) {
347 		__zero_source_counters(syme);
348 		top->sym_filter_entry = NULL;
349 	}
350 
351 	fprintf(stdout, "\n%s: ", msg);
352 	if (getline(&buf, &dummy, stdin) < 0)
353 		goto out_free;
354 
355 	p = strchr(buf, '\n');
356 	if (p)
357 		*p = 0;
358 
359 	next = rb_first(&top->sym_evsel->hists.entries);
360 	while (next) {
361 		n = rb_entry(next, struct hist_entry, rb_node);
362 		if (n->ms.sym && !strcmp(buf, n->ms.sym->name)) {
363 			found = n;
364 			break;
365 		}
366 		next = rb_next(&n->rb_node);
367 	}
368 
369 	if (!found) {
370 		fprintf(stderr, "Sorry, %s is not active.\n", buf);
371 		sleep(1);
372 	} else
373 		perf_top__parse_source(top, found);
374 
375 out_free:
376 	free(buf);
377 }
378 
379 static void perf_top__print_mapped_keys(struct perf_top *top)
380 {
381 	char *name = NULL;
382 
383 	if (top->sym_filter_entry) {
384 		struct symbol *sym = top->sym_filter_entry->ms.sym;
385 		name = sym->name;
386 	}
387 
388 	fprintf(stdout, "\nMapped keys:\n");
389 	fprintf(stdout, "\t[d]     display refresh delay.             \t(%d)\n", top->delay_secs);
390 	fprintf(stdout, "\t[e]     display entries (lines).           \t(%d)\n", top->print_entries);
391 
392 	if (top->evlist->nr_entries > 1)
393 		fprintf(stdout, "\t[E]     active event counter.              \t(%s)\n", perf_evsel__name(top->sym_evsel));
394 
395 	fprintf(stdout, "\t[f]     profile display filter (count).    \t(%d)\n", top->count_filter);
396 
397 	fprintf(stdout, "\t[F]     annotate display filter (percent). \t(%d%%)\n", top->sym_pcnt_filter);
398 	fprintf(stdout, "\t[s]     annotate symbol.                   \t(%s)\n", name?: "NULL");
399 	fprintf(stdout, "\t[S]     stop annotation.\n");
400 
401 	fprintf(stdout,
402 		"\t[K]     hide kernel_symbols symbols.     \t(%s)\n",
403 		top->hide_kernel_symbols ? "yes" : "no");
404 	fprintf(stdout,
405 		"\t[U]     hide user symbols.               \t(%s)\n",
406 		top->hide_user_symbols ? "yes" : "no");
407 	fprintf(stdout, "\t[z]     toggle sample zeroing.             \t(%d)\n", top->zero ? 1 : 0);
408 	fprintf(stdout, "\t[qQ]    quit.\n");
409 }
410 
411 static int perf_top__key_mapped(struct perf_top *top, int c)
412 {
413 	switch (c) {
414 		case 'd':
415 		case 'e':
416 		case 'f':
417 		case 'z':
418 		case 'q':
419 		case 'Q':
420 		case 'K':
421 		case 'U':
422 		case 'F':
423 		case 's':
424 		case 'S':
425 			return 1;
426 		case 'E':
427 			return top->evlist->nr_entries > 1 ? 1 : 0;
428 		default:
429 			break;
430 	}
431 
432 	return 0;
433 }
434 
435 static bool perf_top__handle_keypress(struct perf_top *top, int c)
436 {
437 	bool ret = true;
438 
439 	if (!perf_top__key_mapped(top, c)) {
440 		struct pollfd stdin_poll = { .fd = 0, .events = POLLIN };
441 		struct termios tc, save;
442 
443 		perf_top__print_mapped_keys(top);
444 		fprintf(stdout, "\nEnter selection, or unmapped key to continue: ");
445 		fflush(stdout);
446 
447 		tcgetattr(0, &save);
448 		tc = save;
449 		tc.c_lflag &= ~(ICANON | ECHO);
450 		tc.c_cc[VMIN] = 0;
451 		tc.c_cc[VTIME] = 0;
452 		tcsetattr(0, TCSANOW, &tc);
453 
454 		poll(&stdin_poll, 1, -1);
455 		c = getc(stdin);
456 
457 		tcsetattr(0, TCSAFLUSH, &save);
458 		if (!perf_top__key_mapped(top, c))
459 			return ret;
460 	}
461 
462 	switch (c) {
463 		case 'd':
464 			prompt_integer(&top->delay_secs, "Enter display delay");
465 			if (top->delay_secs < 1)
466 				top->delay_secs = 1;
467 			break;
468 		case 'e':
469 			prompt_integer(&top->print_entries, "Enter display entries (lines)");
470 			if (top->print_entries == 0) {
471 				struct sigaction act = {
472 					.sa_sigaction = perf_top__sig_winch,
473 					.sa_flags     = SA_SIGINFO,
474 				};
475 				perf_top__sig_winch(SIGWINCH, NULL, top);
476 				sigaction(SIGWINCH, &act, NULL);
477 			} else {
478 				signal(SIGWINCH, SIG_DFL);
479 			}
480 			break;
481 		case 'E':
482 			if (top->evlist->nr_entries > 1) {
483 				/* Select 0 as the default event: */
484 				int counter = 0;
485 
486 				fprintf(stderr, "\nAvailable events:");
487 
488 				list_for_each_entry(top->sym_evsel, &top->evlist->entries, node)
489 					fprintf(stderr, "\n\t%d %s", top->sym_evsel->idx, perf_evsel__name(top->sym_evsel));
490 
491 				prompt_integer(&counter, "Enter details event counter");
492 
493 				if (counter >= top->evlist->nr_entries) {
494 					top->sym_evsel = perf_evlist__first(top->evlist);
495 					fprintf(stderr, "Sorry, no such event, using %s.\n", perf_evsel__name(top->sym_evsel));
496 					sleep(1);
497 					break;
498 				}
499 				list_for_each_entry(top->sym_evsel, &top->evlist->entries, node)
500 					if (top->sym_evsel->idx == counter)
501 						break;
502 			} else
503 				top->sym_evsel = perf_evlist__first(top->evlist);
504 			break;
505 		case 'f':
506 			prompt_integer(&top->count_filter, "Enter display event count filter");
507 			break;
508 		case 'F':
509 			prompt_percent(&top->sym_pcnt_filter,
510 				       "Enter details display event filter (percent)");
511 			break;
512 		case 'K':
513 			top->hide_kernel_symbols = !top->hide_kernel_symbols;
514 			break;
515 		case 'q':
516 		case 'Q':
517 			printf("exiting.\n");
518 			if (top->dump_symtab)
519 				perf_session__fprintf_dsos(top->session, stderr);
520 			ret = false;
521 			break;
522 		case 's':
523 			perf_top__prompt_symbol(top, "Enter details symbol");
524 			break;
525 		case 'S':
526 			if (!top->sym_filter_entry)
527 				break;
528 			else {
529 				struct hist_entry *syme = top->sym_filter_entry;
530 
531 				top->sym_filter_entry = NULL;
532 				__zero_source_counters(syme);
533 			}
534 			break;
535 		case 'U':
536 			top->hide_user_symbols = !top->hide_user_symbols;
537 			break;
538 		case 'z':
539 			top->zero = !top->zero;
540 			break;
541 		default:
542 			break;
543 	}
544 
545 	return ret;
546 }
547 
548 static void perf_top__sort_new_samples(void *arg)
549 {
550 	struct perf_top *t = arg;
551 	perf_top__reset_sample_counters(t);
552 
553 	if (t->evlist->selected != NULL)
554 		t->sym_evsel = t->evlist->selected;
555 
556 	hists__collapse_resort(&t->sym_evsel->hists);
557 	hists__output_resort(&t->sym_evsel->hists);
558 	hists__decay_entries(&t->sym_evsel->hists,
559 			     t->hide_user_symbols,
560 			     t->hide_kernel_symbols);
561 }
562 
563 static void *display_thread_tui(void *arg)
564 {
565 	struct perf_evsel *pos;
566 	struct perf_top *top = arg;
567 	const char *help = "For a higher level overview, try: perf top --sort comm,dso";
568 	struct hist_browser_timer hbt = {
569 		.timer		= perf_top__sort_new_samples,
570 		.arg		= top,
571 		.refresh	= top->delay_secs,
572 	};
573 
574 	perf_top__sort_new_samples(top);
575 
576 	/*
577 	 * Initialize the uid_filter_str, in the future the TUI will allow
578 	 * Zooming in/out UIDs. For now juse use whatever the user passed
579 	 * via --uid.
580 	 */
581 	list_for_each_entry(pos, &top->evlist->entries, node)
582 		pos->hists.uid_filter_str = top->record_opts.target.uid_str;
583 
584 	perf_evlist__tui_browse_hists(top->evlist, help, &hbt, top->min_percent,
585 				      &top->session->header.env);
586 
587 	done = 1;
588 	return NULL;
589 }
590 
591 static void *display_thread(void *arg)
592 {
593 	struct pollfd stdin_poll = { .fd = 0, .events = POLLIN };
594 	struct termios tc, save;
595 	struct perf_top *top = arg;
596 	int delay_msecs, c;
597 
598 	tcgetattr(0, &save);
599 	tc = save;
600 	tc.c_lflag &= ~(ICANON | ECHO);
601 	tc.c_cc[VMIN] = 0;
602 	tc.c_cc[VTIME] = 0;
603 
604 	pthread__unblock_sigwinch();
605 repeat:
606 	delay_msecs = top->delay_secs * 1000;
607 	tcsetattr(0, TCSANOW, &tc);
608 	/* trash return*/
609 	getc(stdin);
610 
611 	while (!done) {
612 		perf_top__print_sym_table(top);
613 		/*
614 		 * Either timeout expired or we got an EINTR due to SIGWINCH,
615 		 * refresh screen in both cases.
616 		 */
617 		switch (poll(&stdin_poll, 1, delay_msecs)) {
618 		case 0:
619 			continue;
620 		case -1:
621 			if (errno == EINTR)
622 				continue;
623 			/* Fall trhu */
624 		default:
625 			c = getc(stdin);
626 			tcsetattr(0, TCSAFLUSH, &save);
627 
628 			if (perf_top__handle_keypress(top, c))
629 				goto repeat;
630 			done = 1;
631 		}
632 	}
633 
634 	return NULL;
635 }
636 
637 /* Tag samples to be skipped. */
638 static const char *skip_symbols[] = {
639 	"intel_idle",
640 	"default_idle",
641 	"native_safe_halt",
642 	"cpu_idle",
643 	"enter_idle",
644 	"exit_idle",
645 	"mwait_idle",
646 	"mwait_idle_with_hints",
647 	"poll_idle",
648 	"ppc64_runlatch_off",
649 	"pseries_dedicated_idle_sleep",
650 	NULL
651 };
652 
653 static int symbol_filter(struct map *map __maybe_unused, struct symbol *sym)
654 {
655 	const char *name = sym->name;
656 	int i;
657 
658 	/*
659 	 * ppc64 uses function descriptors and appends a '.' to the
660 	 * start of every instruction address. Remove it.
661 	 */
662 	if (name[0] == '.')
663 		name++;
664 
665 	if (!strcmp(name, "_text") ||
666 	    !strcmp(name, "_etext") ||
667 	    !strcmp(name, "_sinittext") ||
668 	    !strncmp("init_module", name, 11) ||
669 	    !strncmp("cleanup_module", name, 14) ||
670 	    strstr(name, "_text_start") ||
671 	    strstr(name, "_text_end"))
672 		return 1;
673 
674 	for (i = 0; skip_symbols[i]; i++) {
675 		if (!strcmp(skip_symbols[i], name)) {
676 			sym->ignore = true;
677 			break;
678 		}
679 	}
680 
681 	return 0;
682 }
683 
684 static void perf_event__process_sample(struct perf_tool *tool,
685 				       const union perf_event *event,
686 				       struct perf_evsel *evsel,
687 				       struct perf_sample *sample,
688 				       struct machine *machine)
689 {
690 	struct perf_top *top = container_of(tool, struct perf_top, tool);
691 	struct symbol *parent = NULL;
692 	u64 ip = event->ip.ip;
693 	struct addr_location al;
694 	int err;
695 
696 	if (!machine && perf_guest) {
697 		static struct intlist *seen;
698 
699 		if (!seen)
700 			seen = intlist__new(NULL);
701 
702 		if (!intlist__has_entry(seen, event->ip.pid)) {
703 			pr_err("Can't find guest [%d]'s kernel information\n",
704 				event->ip.pid);
705 			intlist__add(seen, event->ip.pid);
706 		}
707 		return;
708 	}
709 
710 	if (!machine) {
711 		pr_err("%u unprocessable samples recorded.\r",
712 		       top->session->stats.nr_unprocessable_samples++);
713 		return;
714 	}
715 
716 	if (event->header.misc & PERF_RECORD_MISC_EXACT_IP)
717 		top->exact_samples++;
718 
719 	if (perf_event__preprocess_sample(event, machine, &al, sample,
720 					  symbol_filter) < 0 ||
721 	    al.filtered)
722 		return;
723 
724 	if (!top->kptr_restrict_warned &&
725 	    symbol_conf.kptr_restrict &&
726 	    al.cpumode == PERF_RECORD_MISC_KERNEL) {
727 		ui__warning(
728 "Kernel address maps (/proc/{kallsyms,modules}) are restricted.\n\n"
729 "Check /proc/sys/kernel/kptr_restrict.\n\n"
730 "Kernel%s samples will not be resolved.\n",
731 			  !RB_EMPTY_ROOT(&al.map->dso->symbols[MAP__FUNCTION]) ?
732 			  " modules" : "");
733 		if (use_browser <= 0)
734 			sleep(5);
735 		top->kptr_restrict_warned = true;
736 	}
737 
738 	if (al.sym == NULL) {
739 		const char *msg = "Kernel samples will not be resolved.\n";
740 		/*
741 		 * As we do lazy loading of symtabs we only will know if the
742 		 * specified vmlinux file is invalid when we actually have a
743 		 * hit in kernel space and then try to load it. So if we get
744 		 * here and there are _no_ symbols in the DSO backing the
745 		 * kernel map, bail out.
746 		 *
747 		 * We may never get here, for instance, if we use -K/
748 		 * --hide-kernel-symbols, even if the user specifies an
749 		 * invalid --vmlinux ;-)
750 		 */
751 		if (!top->kptr_restrict_warned && !top->vmlinux_warned &&
752 		    al.map == machine->vmlinux_maps[MAP__FUNCTION] &&
753 		    RB_EMPTY_ROOT(&al.map->dso->symbols[MAP__FUNCTION])) {
754 			if (symbol_conf.vmlinux_name) {
755 				ui__warning("The %s file can't be used.\n%s",
756 					    symbol_conf.vmlinux_name, msg);
757 			} else {
758 				ui__warning("A vmlinux file was not found.\n%s",
759 					    msg);
760 			}
761 
762 			if (use_browser <= 0)
763 				sleep(5);
764 			top->vmlinux_warned = true;
765 		}
766 	}
767 
768 	if (al.sym == NULL || !al.sym->ignore) {
769 		struct hist_entry *he;
770 
771 		if ((sort__has_parent || symbol_conf.use_callchain) &&
772 		    sample->callchain) {
773 			err = machine__resolve_callchain(machine, evsel,
774 							 al.thread, sample,
775 							 &parent);
776 
777 			if (err)
778 				return;
779 		}
780 
781 		he = perf_evsel__add_hist_entry(evsel, &al, sample);
782 		if (he == NULL) {
783 			pr_err("Problem incrementing symbol period, skipping event\n");
784 			return;
785 		}
786 
787 		if (symbol_conf.use_callchain) {
788 			err = callchain_append(he->callchain, &callchain_cursor,
789 					       sample->period);
790 			if (err)
791 				return;
792 		}
793 
794 		if (sort__has_sym)
795 			perf_top__record_precise_ip(top, he, evsel->idx, ip);
796 	}
797 
798 	return;
799 }
800 
801 static void perf_top__mmap_read_idx(struct perf_top *top, int idx)
802 {
803 	struct perf_sample sample;
804 	struct perf_evsel *evsel;
805 	struct perf_session *session = top->session;
806 	union perf_event *event;
807 	struct machine *machine;
808 	u8 origin;
809 	int ret;
810 
811 	while ((event = perf_evlist__mmap_read(top->evlist, idx)) != NULL) {
812 		ret = perf_evlist__parse_sample(top->evlist, event, &sample);
813 		if (ret) {
814 			pr_err("Can't parse sample, err = %d\n", ret);
815 			continue;
816 		}
817 
818 		evsel = perf_evlist__id2evsel(session->evlist, sample.id);
819 		assert(evsel != NULL);
820 
821 		origin = event->header.misc & PERF_RECORD_MISC_CPUMODE_MASK;
822 
823 		if (event->header.type == PERF_RECORD_SAMPLE)
824 			++top->samples;
825 
826 		switch (origin) {
827 		case PERF_RECORD_MISC_USER:
828 			++top->us_samples;
829 			if (top->hide_user_symbols)
830 				continue;
831 			machine = &session->machines.host;
832 			break;
833 		case PERF_RECORD_MISC_KERNEL:
834 			++top->kernel_samples;
835 			if (top->hide_kernel_symbols)
836 				continue;
837 			machine = &session->machines.host;
838 			break;
839 		case PERF_RECORD_MISC_GUEST_KERNEL:
840 			++top->guest_kernel_samples;
841 			machine = perf_session__find_machine(session, event->ip.pid);
842 			break;
843 		case PERF_RECORD_MISC_GUEST_USER:
844 			++top->guest_us_samples;
845 			/*
846 			 * TODO: we don't process guest user from host side
847 			 * except simple counting.
848 			 */
849 			/* Fall thru */
850 		default:
851 			continue;
852 		}
853 
854 
855 		if (event->header.type == PERF_RECORD_SAMPLE) {
856 			perf_event__process_sample(&top->tool, event, evsel,
857 						   &sample, machine);
858 		} else if (event->header.type < PERF_RECORD_MAX) {
859 			hists__inc_nr_events(&evsel->hists, event->header.type);
860 			machine__process_event(machine, event);
861 		} else
862 			++session->stats.nr_unknown_events;
863 	}
864 }
865 
866 static void perf_top__mmap_read(struct perf_top *top)
867 {
868 	int i;
869 
870 	for (i = 0; i < top->evlist->nr_mmaps; i++)
871 		perf_top__mmap_read_idx(top, i);
872 }
873 
874 static int perf_top__start_counters(struct perf_top *top)
875 {
876 	char msg[512];
877 	struct perf_evsel *counter;
878 	struct perf_evlist *evlist = top->evlist;
879 	struct perf_record_opts *opts = &top->record_opts;
880 
881 	perf_evlist__config(evlist, opts);
882 
883 	list_for_each_entry(counter, &evlist->entries, node) {
884 try_again:
885 		if (perf_evsel__open(counter, top->evlist->cpus,
886 				     top->evlist->threads) < 0) {
887 			if (perf_evsel__fallback(counter, errno, msg, sizeof(msg))) {
888 				if (verbose)
889 					ui__warning("%s\n", msg);
890 				goto try_again;
891 			}
892 
893 			perf_evsel__open_strerror(counter, &opts->target,
894 						  errno, msg, sizeof(msg));
895 			ui__error("%s\n", msg);
896 			goto out_err;
897 		}
898 	}
899 
900 	if (perf_evlist__mmap(evlist, opts->mmap_pages, false) < 0) {
901 		ui__error("Failed to mmap with %d (%s)\n",
902 			    errno, strerror(errno));
903 		goto out_err;
904 	}
905 
906 	return 0;
907 
908 out_err:
909 	return -1;
910 }
911 
912 static int perf_top__setup_sample_type(struct perf_top *top __maybe_unused)
913 {
914 	if (!sort__has_sym) {
915 		if (symbol_conf.use_callchain) {
916 			ui__error("Selected -g but \"sym\" not present in --sort/-s.");
917 			return -EINVAL;
918 		}
919 	} else if (callchain_param.mode != CHAIN_NONE) {
920 		if (callchain_register_param(&callchain_param) < 0) {
921 			ui__error("Can't register callchain params.\n");
922 			return -EINVAL;
923 		}
924 	}
925 
926 	return 0;
927 }
928 
929 static int __cmd_top(struct perf_top *top)
930 {
931 	struct perf_record_opts *opts = &top->record_opts;
932 	pthread_t thread;
933 	int ret;
934 	/*
935 	 * FIXME: perf_session__new should allow passing a O_MMAP, so that all this
936 	 * mmap reading, etc is encapsulated in it. Use O_WRONLY for now.
937 	 */
938 	top->session = perf_session__new(NULL, O_WRONLY, false, false, NULL);
939 	if (top->session == NULL)
940 		return -ENOMEM;
941 
942 	ret = perf_top__setup_sample_type(top);
943 	if (ret)
944 		goto out_delete;
945 
946 	if (perf_target__has_task(&opts->target))
947 		perf_event__synthesize_thread_map(&top->tool, top->evlist->threads,
948 						  perf_event__process,
949 						  &top->session->machines.host);
950 	else
951 		perf_event__synthesize_threads(&top->tool, perf_event__process,
952 					       &top->session->machines.host);
953 
954 	ret = perf_top__start_counters(top);
955 	if (ret)
956 		goto out_delete;
957 
958 	top->session->evlist = top->evlist;
959 	perf_session__set_id_hdr_size(top->session);
960 
961 	/*
962 	 * When perf is starting the traced process, all the events (apart from
963 	 * group members) have enable_on_exec=1 set, so don't spoil it by
964 	 * prematurely enabling them.
965 	 *
966 	 * XXX 'top' still doesn't start workloads like record, trace, but should,
967 	 * so leave the check here.
968 	 */
969         if (!perf_target__none(&opts->target))
970                 perf_evlist__enable(top->evlist);
971 
972 	/* Wait for a minimal set of events before starting the snapshot */
973 	poll(top->evlist->pollfd, top->evlist->nr_fds, 100);
974 
975 	perf_top__mmap_read(top);
976 
977 	ret = -1;
978 	if (pthread_create(&thread, NULL, (use_browser > 0 ? display_thread_tui :
979 							    display_thread), top)) {
980 		ui__error("Could not create display thread.\n");
981 		goto out_delete;
982 	}
983 
984 	if (top->realtime_prio) {
985 		struct sched_param param;
986 
987 		param.sched_priority = top->realtime_prio;
988 		if (sched_setscheduler(0, SCHED_FIFO, &param)) {
989 			ui__error("Could not set realtime priority.\n");
990 			goto out_delete;
991 		}
992 	}
993 
994 	while (!done) {
995 		u64 hits = top->samples;
996 
997 		perf_top__mmap_read(top);
998 
999 		if (hits == top->samples)
1000 			ret = poll(top->evlist->pollfd, top->evlist->nr_fds, 100);
1001 	}
1002 
1003 	ret = 0;
1004 out_delete:
1005 	perf_session__delete(top->session);
1006 	top->session = NULL;
1007 
1008 	return ret;
1009 }
1010 
1011 static int
1012 parse_callchain_opt(const struct option *opt, const char *arg, int unset)
1013 {
1014 	/*
1015 	 * --no-call-graph
1016 	 */
1017 	if (unset)
1018 		return 0;
1019 
1020 	symbol_conf.use_callchain = true;
1021 
1022 	return record_parse_callchain_opt(opt, arg, unset);
1023 }
1024 
1025 static int
1026 parse_percent_limit(const struct option *opt, const char *arg,
1027 		    int unset __maybe_unused)
1028 {
1029 	struct perf_top *top = opt->value;
1030 
1031 	top->min_percent = strtof(arg, NULL);
1032 	return 0;
1033 }
1034 
1035 int cmd_top(int argc, const char **argv, const char *prefix __maybe_unused)
1036 {
1037 	int status;
1038 	char errbuf[BUFSIZ];
1039 	struct perf_top top = {
1040 		.count_filter	     = 5,
1041 		.delay_secs	     = 2,
1042 		.record_opts = {
1043 			.mmap_pages	= UINT_MAX,
1044 			.user_freq	= UINT_MAX,
1045 			.user_interval	= ULLONG_MAX,
1046 			.freq		= 4000, /* 4 KHz */
1047 			.target		     = {
1048 				.uses_mmap   = true,
1049 			},
1050 		},
1051 		.sym_pcnt_filter     = 5,
1052 	};
1053 	struct perf_record_opts *opts = &top.record_opts;
1054 	struct perf_target *target = &opts->target;
1055 	const struct option options[] = {
1056 	OPT_CALLBACK('e', "event", &top.evlist, "event",
1057 		     "event selector. use 'perf list' to list available events",
1058 		     parse_events_option),
1059 	OPT_U64('c', "count", &opts->user_interval, "event period to sample"),
1060 	OPT_STRING('p', "pid", &target->pid, "pid",
1061 		    "profile events on existing process id"),
1062 	OPT_STRING('t', "tid", &target->tid, "tid",
1063 		    "profile events on existing thread id"),
1064 	OPT_BOOLEAN('a', "all-cpus", &target->system_wide,
1065 			    "system-wide collection from all CPUs"),
1066 	OPT_STRING('C', "cpu", &target->cpu_list, "cpu",
1067 		    "list of cpus to monitor"),
1068 	OPT_STRING('k', "vmlinux", &symbol_conf.vmlinux_name,
1069 		   "file", "vmlinux pathname"),
1070 	OPT_BOOLEAN('K', "hide_kernel_symbols", &top.hide_kernel_symbols,
1071 		    "hide kernel symbols"),
1072 	OPT_UINTEGER('m', "mmap-pages", &opts->mmap_pages,
1073 		     "number of mmap data pages"),
1074 	OPT_INTEGER('r', "realtime", &top.realtime_prio,
1075 		    "collect data with this RT SCHED_FIFO priority"),
1076 	OPT_INTEGER('d', "delay", &top.delay_secs,
1077 		    "number of seconds to delay between refreshes"),
1078 	OPT_BOOLEAN('D', "dump-symtab", &top.dump_symtab,
1079 			    "dump the symbol table used for profiling"),
1080 	OPT_INTEGER('f', "count-filter", &top.count_filter,
1081 		    "only display functions with more events than this"),
1082 	OPT_BOOLEAN('g', "group", &opts->group,
1083 			    "put the counters into a counter group"),
1084 	OPT_BOOLEAN('i', "no-inherit", &opts->no_inherit,
1085 		    "child tasks do not inherit counters"),
1086 	OPT_STRING(0, "sym-annotate", &top.sym_filter, "symbol name",
1087 		    "symbol to annotate"),
1088 	OPT_BOOLEAN('z', "zero", &top.zero, "zero history across updates"),
1089 	OPT_UINTEGER('F', "freq", &opts->user_freq, "profile at this frequency"),
1090 	OPT_INTEGER('E', "entries", &top.print_entries,
1091 		    "display this many functions"),
1092 	OPT_BOOLEAN('U', "hide_user_symbols", &top.hide_user_symbols,
1093 		    "hide user symbols"),
1094 	OPT_BOOLEAN(0, "tui", &top.use_tui, "Use the TUI interface"),
1095 	OPT_BOOLEAN(0, "stdio", &top.use_stdio, "Use the stdio interface"),
1096 	OPT_INCR('v', "verbose", &verbose,
1097 		    "be more verbose (show counter open errors, etc)"),
1098 	OPT_STRING('s', "sort", &sort_order, "key[,key2...]",
1099 		   "sort by key(s): pid, comm, dso, symbol, parent, weight, local_weight"),
1100 	OPT_BOOLEAN('n', "show-nr-samples", &symbol_conf.show_nr_samples,
1101 		    "Show a column with the number of samples"),
1102 	OPT_CALLBACK_DEFAULT('G', "call-graph", &top.record_opts,
1103 			     "mode[,dump_size]", record_callchain_help,
1104 			     &parse_callchain_opt, "fp"),
1105 	OPT_BOOLEAN(0, "show-total-period", &symbol_conf.show_total_period,
1106 		    "Show a column with the sum of periods"),
1107 	OPT_STRING(0, "dsos", &symbol_conf.dso_list_str, "dso[,dso...]",
1108 		   "only consider symbols in these dsos"),
1109 	OPT_STRING(0, "comms", &symbol_conf.comm_list_str, "comm[,comm...]",
1110 		   "only consider symbols in these comms"),
1111 	OPT_STRING(0, "symbols", &symbol_conf.sym_list_str, "symbol[,symbol...]",
1112 		   "only consider these symbols"),
1113 	OPT_BOOLEAN(0, "source", &symbol_conf.annotate_src,
1114 		    "Interleave source code with assembly code (default)"),
1115 	OPT_BOOLEAN(0, "asm-raw", &symbol_conf.annotate_asm_raw,
1116 		    "Display raw encoding of assembly instructions (default)"),
1117 	OPT_STRING('M', "disassembler-style", &disassembler_style, "disassembler style",
1118 		   "Specify disassembler style (e.g. -M intel for intel syntax)"),
1119 	OPT_STRING('u', "uid", &target->uid_str, "user", "user to profile"),
1120 	OPT_CALLBACK(0, "percent-limit", &top, "percent",
1121 		     "Don't show entries under that percent", parse_percent_limit),
1122 	OPT_END()
1123 	};
1124 	const char * const top_usage[] = {
1125 		"perf top [<options>]",
1126 		NULL
1127 	};
1128 
1129 	top.evlist = perf_evlist__new();
1130 	if (top.evlist == NULL)
1131 		return -ENOMEM;
1132 
1133 	argc = parse_options(argc, argv, options, top_usage, 0);
1134 	if (argc)
1135 		usage_with_options(top_usage, options);
1136 
1137 	if (sort_order == default_sort_order)
1138 		sort_order = "dso,symbol";
1139 
1140 	if (setup_sorting() < 0)
1141 		usage_with_options(top_usage, options);
1142 
1143 	/* display thread wants entries to be collapsed in a different tree */
1144 	sort__need_collapse = 1;
1145 
1146 	if (top.use_stdio)
1147 		use_browser = 0;
1148 	else if (top.use_tui)
1149 		use_browser = 1;
1150 
1151 	setup_browser(false);
1152 
1153 	status = perf_target__validate(target);
1154 	if (status) {
1155 		perf_target__strerror(target, status, errbuf, BUFSIZ);
1156 		ui__warning("%s", errbuf);
1157 	}
1158 
1159 	status = perf_target__parse_uid(target);
1160 	if (status) {
1161 		int saved_errno = errno;
1162 
1163 		perf_target__strerror(target, status, errbuf, BUFSIZ);
1164 		ui__error("%s", errbuf);
1165 
1166 		status = -saved_errno;
1167 		goto out_delete_evlist;
1168 	}
1169 
1170 	if (perf_target__none(target))
1171 		target->system_wide = true;
1172 
1173 	if (perf_evlist__create_maps(top.evlist, target) < 0)
1174 		usage_with_options(top_usage, options);
1175 
1176 	if (!top.evlist->nr_entries &&
1177 	    perf_evlist__add_default(top.evlist) < 0) {
1178 		ui__error("Not enough memory for event selector list\n");
1179 		goto out_delete_maps;
1180 	}
1181 
1182 	symbol_conf.nr_events = top.evlist->nr_entries;
1183 
1184 	if (top.delay_secs < 1)
1185 		top.delay_secs = 1;
1186 
1187 	if (opts->user_interval != ULLONG_MAX)
1188 		opts->default_interval = opts->user_interval;
1189 	if (opts->user_freq != UINT_MAX)
1190 		opts->freq = opts->user_freq;
1191 
1192 	/*
1193 	 * User specified count overrides default frequency.
1194 	 */
1195 	if (opts->default_interval)
1196 		opts->freq = 0;
1197 	else if (opts->freq) {
1198 		opts->default_interval = opts->freq;
1199 	} else {
1200 		ui__error("frequency and count are zero, aborting\n");
1201 		status = -EINVAL;
1202 		goto out_delete_maps;
1203 	}
1204 
1205 	top.sym_evsel = perf_evlist__first(top.evlist);
1206 
1207 	symbol_conf.priv_size = sizeof(struct annotation);
1208 
1209 	symbol_conf.try_vmlinux_path = (symbol_conf.vmlinux_name == NULL);
1210 	if (symbol__init() < 0)
1211 		return -1;
1212 
1213 	sort__setup_elide(stdout);
1214 
1215 	get_term_dimensions(&top.winsize);
1216 	if (top.print_entries == 0) {
1217 		struct sigaction act = {
1218 			.sa_sigaction = perf_top__sig_winch,
1219 			.sa_flags     = SA_SIGINFO,
1220 		};
1221 		perf_top__update_print_entries(&top);
1222 		sigaction(SIGWINCH, &act, NULL);
1223 	}
1224 
1225 	status = __cmd_top(&top);
1226 
1227 out_delete_maps:
1228 	perf_evlist__delete_maps(top.evlist);
1229 out_delete_evlist:
1230 	perf_evlist__delete(top.evlist);
1231 
1232 	return status;
1233 }
1234