xref: /openbmc/linux/tools/perf/builtin-report.c (revision 483eb062)
1 /*
2  * builtin-report.c
3  *
4  * Builtin report command: Analyze the perf.data input file,
5  * look up and read DSOs and symbol information and display
6  * a histogram of results, along various sorting keys.
7  */
8 #include "builtin.h"
9 
10 #include "util/util.h"
11 #include "util/cache.h"
12 
13 #include "util/annotate.h"
14 #include "util/color.h"
15 #include <linux/list.h>
16 #include <linux/rbtree.h>
17 #include "util/symbol.h"
18 #include "util/callchain.h"
19 #include "util/strlist.h"
20 #include "util/values.h"
21 
22 #include "perf.h"
23 #include "util/debug.h"
24 #include "util/evlist.h"
25 #include "util/evsel.h"
26 #include "util/header.h"
27 #include "util/session.h"
28 #include "util/tool.h"
29 
30 #include "util/parse-options.h"
31 #include "util/parse-events.h"
32 
33 #include "util/thread.h"
34 #include "util/sort.h"
35 #include "util/hist.h"
36 #include "util/data.h"
37 #include "arch/common.h"
38 
39 #include <dlfcn.h>
40 #include <linux/bitmap.h>
41 
42 struct report {
43 	struct perf_tool	tool;
44 	struct perf_session	*session;
45 	bool			force, use_tui, use_gtk, use_stdio;
46 	bool			hide_unresolved;
47 	bool			dont_use_callchains;
48 	bool			show_full_info;
49 	bool			show_threads;
50 	bool			inverted_callchain;
51 	bool			mem_mode;
52 	bool			header;
53 	bool			header_only;
54 	int			max_stack;
55 	struct perf_read_values	show_threads_values;
56 	const char		*pretty_printing_style;
57 	const char		*cpu_list;
58 	const char		*symbol_filter_str;
59 	float			min_percent;
60 	DECLARE_BITMAP(cpu_bitmap, MAX_NR_CPUS);
61 };
62 
63 static int report__config(const char *var, const char *value, void *cb)
64 {
65 	if (!strcmp(var, "report.group")) {
66 		symbol_conf.event_group = perf_config_bool(var, value);
67 		return 0;
68 	}
69 	if (!strcmp(var, "report.percent-limit")) {
70 		struct report *rep = cb;
71 		rep->min_percent = strtof(value, NULL);
72 		return 0;
73 	}
74 
75 	return perf_default_config(var, value, cb);
76 }
77 
78 static int report__add_mem_hist_entry(struct perf_tool *tool, struct addr_location *al,
79 				      struct perf_sample *sample, struct perf_evsel *evsel,
80 				      union perf_event *event)
81 {
82 	struct report *rep = container_of(tool, struct report, tool);
83 	struct symbol *parent = NULL;
84 	u8 cpumode = event->header.misc & PERF_RECORD_MISC_CPUMODE_MASK;
85 	struct hist_entry *he;
86 	struct mem_info *mi, *mx;
87 	uint64_t cost;
88 	int err = sample__resolve_callchain(sample, &parent, evsel, al, rep->max_stack);
89 
90 	if (err)
91 		return err;
92 
93 	mi = machine__resolve_mem(al->machine, al->thread, sample, cpumode);
94 	if (!mi)
95 		return -ENOMEM;
96 
97 	if (rep->hide_unresolved && !al->sym)
98 		return 0;
99 
100 	cost = sample->weight;
101 	if (!cost)
102 		cost = 1;
103 
104 	/*
105 	 * must pass period=weight in order to get the correct
106 	 * sorting from hists__collapse_resort() which is solely
107 	 * based on periods. We want sorting be done on nr_events * weight
108 	 * and this is indirectly achieved by passing period=weight here
109 	 * and the he_stat__add_period() function.
110 	 */
111 	he = __hists__add_entry(&evsel->hists, al, parent, NULL, mi,
112 				cost, cost, 0);
113 	if (!he)
114 		return -ENOMEM;
115 
116 	err = hist_entry__inc_addr_samples(he, evsel->idx, al->addr);
117 	if (err)
118 		goto out;
119 
120 	mx = he->mem_info;
121 	err = addr_map_symbol__inc_samples(&mx->daddr, evsel->idx);
122 	if (err)
123 		goto out;
124 
125 	evsel->hists.stats.total_period += cost;
126 	hists__inc_nr_events(&evsel->hists, PERF_RECORD_SAMPLE);
127 	err = hist_entry__append_callchain(he, sample);
128 out:
129 	return err;
130 }
131 
132 static int report__add_branch_hist_entry(struct perf_tool *tool, struct addr_location *al,
133 					 struct perf_sample *sample, struct perf_evsel *evsel)
134 {
135 	struct report *rep = container_of(tool, struct report, tool);
136 	struct symbol *parent = NULL;
137 	unsigned i;
138 	struct hist_entry *he;
139 	struct branch_info *bi, *bx;
140 	int err = sample__resolve_callchain(sample, &parent, evsel, al, rep->max_stack);
141 
142 	if (err)
143 		return err;
144 
145 	bi = machine__resolve_bstack(al->machine, al->thread,
146 				     sample->branch_stack);
147 	if (!bi)
148 		return -ENOMEM;
149 
150 	for (i = 0; i < sample->branch_stack->nr; i++) {
151 		if (rep->hide_unresolved && !(bi[i].from.sym && bi[i].to.sym))
152 			continue;
153 
154 		err = -ENOMEM;
155 
156 		/* overwrite the 'al' to branch-to info */
157 		al->map = bi[i].to.map;
158 		al->sym = bi[i].to.sym;
159 		al->addr = bi[i].to.addr;
160 		/*
161 		 * The report shows the percentage of total branches captured
162 		 * and not events sampled. Thus we use a pseudo period of 1.
163 		 */
164 		he = __hists__add_entry(&evsel->hists, al, parent, &bi[i], NULL,
165 					1, 1, 0);
166 		if (he) {
167 			bx = he->branch_info;
168 			err = addr_map_symbol__inc_samples(&bx->from, evsel->idx);
169 			if (err)
170 				goto out;
171 
172 			err = addr_map_symbol__inc_samples(&bx->to, evsel->idx);
173 			if (err)
174 				goto out;
175 
176 			evsel->hists.stats.total_period += 1;
177 			hists__inc_nr_events(&evsel->hists, PERF_RECORD_SAMPLE);
178 		} else
179 			goto out;
180 	}
181 	err = 0;
182 out:
183 	free(bi);
184 	return err;
185 }
186 
187 static int report__add_hist_entry(struct perf_tool *tool, struct perf_evsel *evsel,
188 				  struct addr_location *al, struct perf_sample *sample)
189 {
190 	struct report *rep = container_of(tool, struct report, tool);
191 	struct symbol *parent = NULL;
192 	struct hist_entry *he;
193 	int err = sample__resolve_callchain(sample, &parent, evsel, al, rep->max_stack);
194 
195 	if (err)
196 		return err;
197 
198 	he = __hists__add_entry(&evsel->hists, al, parent, NULL, NULL,
199 				sample->period, sample->weight,
200 				sample->transaction);
201 	if (he == NULL)
202 		return -ENOMEM;
203 
204 	err = hist_entry__append_callchain(he, sample);
205 	if (err)
206 		goto out;
207 
208 	err = hist_entry__inc_addr_samples(he, evsel->idx, al->addr);
209 	evsel->hists.stats.total_period += sample->period;
210 	hists__inc_nr_events(&evsel->hists, PERF_RECORD_SAMPLE);
211 out:
212 	return err;
213 }
214 
215 
216 static int process_sample_event(struct perf_tool *tool,
217 				union perf_event *event,
218 				struct perf_sample *sample,
219 				struct perf_evsel *evsel,
220 				struct machine *machine)
221 {
222 	struct report *rep = container_of(tool, struct report, tool);
223 	struct addr_location al;
224 	int ret;
225 
226 	if (perf_event__preprocess_sample(event, machine, &al, sample) < 0) {
227 		pr_debug("problem processing %d event, skipping it.\n",
228 			 event->header.type);
229 		return -1;
230 	}
231 
232 	if (al.filtered || (rep->hide_unresolved && al.sym == NULL))
233 		return 0;
234 
235 	if (rep->cpu_list && !test_bit(sample->cpu, rep->cpu_bitmap))
236 		return 0;
237 
238 	if (sort__mode == SORT_MODE__BRANCH) {
239 		ret = report__add_branch_hist_entry(tool, &al, sample, evsel);
240 		if (ret < 0)
241 			pr_debug("problem adding lbr entry, skipping event\n");
242 	} else if (rep->mem_mode == 1) {
243 		ret = report__add_mem_hist_entry(tool, &al, sample, evsel, event);
244 		if (ret < 0)
245 			pr_debug("problem adding mem entry, skipping event\n");
246 	} else {
247 		if (al.map != NULL)
248 			al.map->dso->hit = 1;
249 
250 		ret = report__add_hist_entry(tool, evsel, &al, sample);
251 		if (ret < 0)
252 			pr_debug("problem incrementing symbol period, skipping event\n");
253 	}
254 	return ret;
255 }
256 
257 static int process_read_event(struct perf_tool *tool,
258 			      union perf_event *event,
259 			      struct perf_sample *sample __maybe_unused,
260 			      struct perf_evsel *evsel,
261 			      struct machine *machine __maybe_unused)
262 {
263 	struct report *rep = container_of(tool, struct report, tool);
264 
265 	if (rep->show_threads) {
266 		const char *name = evsel ? perf_evsel__name(evsel) : "unknown";
267 		perf_read_values_add_value(&rep->show_threads_values,
268 					   event->read.pid, event->read.tid,
269 					   event->read.id,
270 					   name,
271 					   event->read.value);
272 	}
273 
274 	dump_printf(": %d %d %s %" PRIu64 "\n", event->read.pid, event->read.tid,
275 		    evsel ? perf_evsel__name(evsel) : "FAIL",
276 		    event->read.value);
277 
278 	return 0;
279 }
280 
281 /* For pipe mode, sample_type is not currently set */
282 static int report__setup_sample_type(struct report *rep)
283 {
284 	struct perf_session *session = rep->session;
285 	u64 sample_type = perf_evlist__combined_sample_type(session->evlist);
286 	bool is_pipe = perf_data_file__is_pipe(session->file);
287 
288 	if (!is_pipe && !(sample_type & PERF_SAMPLE_CALLCHAIN)) {
289 		if (sort__has_parent) {
290 			ui__error("Selected --sort parent, but no "
291 				    "callchain data. Did you call "
292 				    "'perf record' without -g?\n");
293 			return -EINVAL;
294 		}
295 		if (symbol_conf.use_callchain) {
296 			ui__error("Selected -g but no callchain data. Did "
297 				    "you call 'perf record' without -g?\n");
298 			return -1;
299 		}
300 	} else if (!rep->dont_use_callchains &&
301 		   callchain_param.mode != CHAIN_NONE &&
302 		   !symbol_conf.use_callchain) {
303 			symbol_conf.use_callchain = true;
304 			if (callchain_register_param(&callchain_param) < 0) {
305 				ui__error("Can't register callchain params.\n");
306 				return -EINVAL;
307 			}
308 	}
309 
310 	if (sort__mode == SORT_MODE__BRANCH) {
311 		if (!is_pipe &&
312 		    !(sample_type & PERF_SAMPLE_BRANCH_STACK)) {
313 			ui__error("Selected -b but no branch data. "
314 				  "Did you call perf record without -b?\n");
315 			return -1;
316 		}
317 	}
318 
319 	return 0;
320 }
321 
322 static void sig_handler(int sig __maybe_unused)
323 {
324 	session_done = 1;
325 }
326 
327 static size_t hists__fprintf_nr_sample_events(struct hists *hists, struct report *rep,
328 					      const char *evname, FILE *fp)
329 {
330 	size_t ret;
331 	char unit;
332 	unsigned long nr_samples = hists->stats.nr_events[PERF_RECORD_SAMPLE];
333 	u64 nr_events = hists->stats.total_period;
334 	struct perf_evsel *evsel = hists_to_evsel(hists);
335 	char buf[512];
336 	size_t size = sizeof(buf);
337 
338 	if (perf_evsel__is_group_event(evsel)) {
339 		struct perf_evsel *pos;
340 
341 		perf_evsel__group_desc(evsel, buf, size);
342 		evname = buf;
343 
344 		for_each_group_member(pos, evsel) {
345 			nr_samples += pos->hists.stats.nr_events[PERF_RECORD_SAMPLE];
346 			nr_events += pos->hists.stats.total_period;
347 		}
348 	}
349 
350 	nr_samples = convert_unit(nr_samples, &unit);
351 	ret = fprintf(fp, "# Samples: %lu%c", nr_samples, unit);
352 	if (evname != NULL)
353 		ret += fprintf(fp, " of event '%s'", evname);
354 
355 	if (rep->mem_mode) {
356 		ret += fprintf(fp, "\n# Total weight : %" PRIu64, nr_events);
357 		ret += fprintf(fp, "\n# Sort order   : %s", sort_order);
358 	} else
359 		ret += fprintf(fp, "\n# Event count (approx.): %" PRIu64, nr_events);
360 	return ret + fprintf(fp, "\n#\n");
361 }
362 
363 static int perf_evlist__tty_browse_hists(struct perf_evlist *evlist,
364 					 struct report *rep,
365 					 const char *help)
366 {
367 	struct perf_evsel *pos;
368 
369 	evlist__for_each(evlist, pos) {
370 		struct hists *hists = &pos->hists;
371 		const char *evname = perf_evsel__name(pos);
372 
373 		if (symbol_conf.event_group &&
374 		    !perf_evsel__is_group_leader(pos))
375 			continue;
376 
377 		hists__fprintf_nr_sample_events(hists, rep, evname, stdout);
378 		hists__fprintf(hists, true, 0, 0, rep->min_percent, stdout);
379 		fprintf(stdout, "\n\n");
380 	}
381 
382 	if (sort_order == default_sort_order &&
383 	    parent_pattern == default_parent_pattern) {
384 		fprintf(stdout, "#\n# (%s)\n#\n", help);
385 
386 		if (rep->show_threads) {
387 			bool style = !strcmp(rep->pretty_printing_style, "raw");
388 			perf_read_values_display(stdout, &rep->show_threads_values,
389 						 style);
390 			perf_read_values_destroy(&rep->show_threads_values);
391 		}
392 	}
393 
394 	return 0;
395 }
396 
397 static void report__warn_kptr_restrict(const struct report *rep)
398 {
399 	struct map *kernel_map = rep->session->machines.host.vmlinux_maps[MAP__FUNCTION];
400 	struct kmap *kernel_kmap = map__kmap(kernel_map);
401 
402 	if (kernel_map == NULL ||
403 	    (kernel_map->dso->hit &&
404 	     (kernel_kmap->ref_reloc_sym == NULL ||
405 	      kernel_kmap->ref_reloc_sym->addr == 0))) {
406 		const char *desc =
407 		    "As no suitable kallsyms nor vmlinux was found, kernel samples\n"
408 		    "can't be resolved.";
409 
410 		if (kernel_map) {
411 			const struct dso *kdso = kernel_map->dso;
412 			if (!RB_EMPTY_ROOT(&kdso->symbols[MAP__FUNCTION])) {
413 				desc = "If some relocation was applied (e.g. "
414 				       "kexec) symbols may be misresolved.";
415 			}
416 		}
417 
418 		ui__warning(
419 "Kernel address maps (/proc/{kallsyms,modules}) were restricted.\n\n"
420 "Check /proc/sys/kernel/kptr_restrict before running 'perf record'.\n\n%s\n\n"
421 "Samples in kernel modules can't be resolved as well.\n\n",
422 		desc);
423 	}
424 }
425 
426 static int report__gtk_browse_hists(struct report *rep, const char *help)
427 {
428 	int (*hist_browser)(struct perf_evlist *evlist, const char *help,
429 			    struct hist_browser_timer *timer, float min_pcnt);
430 
431 	hist_browser = dlsym(perf_gtk_handle, "perf_evlist__gtk_browse_hists");
432 
433 	if (hist_browser == NULL) {
434 		ui__error("GTK browser not found!\n");
435 		return -1;
436 	}
437 
438 	return hist_browser(rep->session->evlist, help, NULL, rep->min_percent);
439 }
440 
441 static int report__browse_hists(struct report *rep)
442 {
443 	int ret;
444 	struct perf_session *session = rep->session;
445 	struct perf_evlist *evlist = session->evlist;
446 	const char *help = "For a higher level overview, try: perf report --sort comm,dso";
447 
448 	switch (use_browser) {
449 	case 1:
450 		ret = perf_evlist__tui_browse_hists(evlist, help, NULL,
451 						    rep->min_percent,
452 						    &session->header.env);
453 		/*
454 		 * Usually "ret" is the last pressed key, and we only
455 		 * care if the key notifies us to switch data file.
456 		 */
457 		if (ret != K_SWITCH_INPUT_DATA)
458 			ret = 0;
459 		break;
460 	case 2:
461 		ret = report__gtk_browse_hists(rep, help);
462 		break;
463 	default:
464 		ret = perf_evlist__tty_browse_hists(evlist, rep, help);
465 		break;
466 	}
467 
468 	return ret;
469 }
470 
471 static u64 report__collapse_hists(struct report *rep)
472 {
473 	struct ui_progress prog;
474 	struct perf_evsel *pos;
475 	u64 nr_samples = 0;
476 	/*
477  	 * Count number of histogram entries to use when showing progress,
478  	 * reusing nr_samples variable.
479  	 */
480 	evlist__for_each(rep->session->evlist, pos)
481 		nr_samples += pos->hists.nr_entries;
482 
483 	ui_progress__init(&prog, nr_samples, "Merging related events...");
484 	/*
485 	 * Count total number of samples, will be used to check if this
486  	 * session had any.
487  	 */
488 	nr_samples = 0;
489 
490 	evlist__for_each(rep->session->evlist, pos) {
491 		struct hists *hists = &pos->hists;
492 
493 		if (pos->idx == 0)
494 			hists->symbol_filter_str = rep->symbol_filter_str;
495 
496 		hists__collapse_resort(hists, &prog);
497 		nr_samples += hists->stats.nr_events[PERF_RECORD_SAMPLE];
498 
499 		/* Non-group events are considered as leader */
500 		if (symbol_conf.event_group &&
501 		    !perf_evsel__is_group_leader(pos)) {
502 			struct hists *leader_hists = &pos->leader->hists;
503 
504 			hists__match(leader_hists, hists);
505 			hists__link(leader_hists, hists);
506 		}
507 	}
508 
509 	ui_progress__finish();
510 
511 	return nr_samples;
512 }
513 
514 static int __cmd_report(struct report *rep)
515 {
516 	int ret;
517 	u64 nr_samples;
518 	struct perf_session *session = rep->session;
519 	struct perf_evsel *pos;
520 	struct perf_data_file *file = session->file;
521 
522 	signal(SIGINT, sig_handler);
523 
524 	if (rep->cpu_list) {
525 		ret = perf_session__cpu_bitmap(session, rep->cpu_list,
526 					       rep->cpu_bitmap);
527 		if (ret)
528 			return ret;
529 	}
530 
531 	if (rep->show_threads)
532 		perf_read_values_init(&rep->show_threads_values);
533 
534 	ret = report__setup_sample_type(rep);
535 	if (ret)
536 		return ret;
537 
538 	ret = perf_session__process_events(session, &rep->tool);
539 	if (ret)
540 		return ret;
541 
542 	report__warn_kptr_restrict(rep);
543 
544 	if (use_browser == 0) {
545 		if (verbose > 3)
546 			perf_session__fprintf(session, stdout);
547 
548 		if (verbose > 2)
549 			perf_session__fprintf_dsos(session, stdout);
550 
551 		if (dump_trace) {
552 			perf_session__fprintf_nr_events(session, stdout);
553 			return 0;
554 		}
555 	}
556 
557 	nr_samples = report__collapse_hists(rep);
558 
559 	if (session_done())
560 		return 0;
561 
562 	if (nr_samples == 0) {
563 		ui__error("The %s file has no samples!\n", file->path);
564 		return 0;
565 	}
566 
567 	evlist__for_each(session->evlist, pos)
568 		hists__output_resort(&pos->hists);
569 
570 	return report__browse_hists(rep);
571 }
572 
573 static int
574 parse_callchain_opt(const struct option *opt, const char *arg, int unset)
575 {
576 	struct report *rep = (struct report *)opt->value;
577 	char *tok, *tok2;
578 	char *endptr;
579 
580 	/*
581 	 * --no-call-graph
582 	 */
583 	if (unset) {
584 		rep->dont_use_callchains = true;
585 		return 0;
586 	}
587 
588 	symbol_conf.use_callchain = true;
589 
590 	if (!arg)
591 		return 0;
592 
593 	tok = strtok((char *)arg, ",");
594 	if (!tok)
595 		return -1;
596 
597 	/* get the output mode */
598 	if (!strncmp(tok, "graph", strlen(arg)))
599 		callchain_param.mode = CHAIN_GRAPH_ABS;
600 
601 	else if (!strncmp(tok, "flat", strlen(arg)))
602 		callchain_param.mode = CHAIN_FLAT;
603 
604 	else if (!strncmp(tok, "fractal", strlen(arg)))
605 		callchain_param.mode = CHAIN_GRAPH_REL;
606 
607 	else if (!strncmp(tok, "none", strlen(arg))) {
608 		callchain_param.mode = CHAIN_NONE;
609 		symbol_conf.use_callchain = false;
610 
611 		return 0;
612 	}
613 
614 	else
615 		return -1;
616 
617 	/* get the min percentage */
618 	tok = strtok(NULL, ",");
619 	if (!tok)
620 		goto setup;
621 
622 	callchain_param.min_percent = strtod(tok, &endptr);
623 	if (tok == endptr)
624 		return -1;
625 
626 	/* get the print limit */
627 	tok2 = strtok(NULL, ",");
628 	if (!tok2)
629 		goto setup;
630 
631 	if (tok2[0] != 'c') {
632 		callchain_param.print_limit = strtoul(tok2, &endptr, 0);
633 		tok2 = strtok(NULL, ",");
634 		if (!tok2)
635 			goto setup;
636 	}
637 
638 	/* get the call chain order */
639 	if (!strncmp(tok2, "caller", strlen("caller")))
640 		callchain_param.order = ORDER_CALLER;
641 	else if (!strncmp(tok2, "callee", strlen("callee")))
642 		callchain_param.order = ORDER_CALLEE;
643 	else
644 		return -1;
645 
646 	/* Get the sort key */
647 	tok2 = strtok(NULL, ",");
648 	if (!tok2)
649 		goto setup;
650 	if (!strncmp(tok2, "function", strlen("function")))
651 		callchain_param.key = CCKEY_FUNCTION;
652 	else if (!strncmp(tok2, "address", strlen("address")))
653 		callchain_param.key = CCKEY_ADDRESS;
654 	else
655 		return -1;
656 setup:
657 	if (callchain_register_param(&callchain_param) < 0) {
658 		pr_err("Can't register callchain params\n");
659 		return -1;
660 	}
661 	return 0;
662 }
663 
664 int
665 report_parse_ignore_callees_opt(const struct option *opt __maybe_unused,
666 				const char *arg, int unset __maybe_unused)
667 {
668 	if (arg) {
669 		int err = regcomp(&ignore_callees_regex, arg, REG_EXTENDED);
670 		if (err) {
671 			char buf[BUFSIZ];
672 			regerror(err, &ignore_callees_regex, buf, sizeof(buf));
673 			pr_err("Invalid --ignore-callees regex: %s\n%s", arg, buf);
674 			return -1;
675 		}
676 		have_ignore_callees = 1;
677 	}
678 
679 	return 0;
680 }
681 
682 static int
683 parse_branch_mode(const struct option *opt __maybe_unused,
684 		  const char *str __maybe_unused, int unset)
685 {
686 	int *branch_mode = opt->value;
687 
688 	*branch_mode = !unset;
689 	return 0;
690 }
691 
692 static int
693 parse_percent_limit(const struct option *opt, const char *str,
694 		    int unset __maybe_unused)
695 {
696 	struct report *rep = opt->value;
697 
698 	rep->min_percent = strtof(str, NULL);
699 	return 0;
700 }
701 
702 int cmd_report(int argc, const char **argv, const char *prefix __maybe_unused)
703 {
704 	struct perf_session *session;
705 	struct stat st;
706 	bool has_br_stack = false;
707 	int branch_mode = -1;
708 	int ret = -1;
709 	char callchain_default_opt[] = "fractal,0.5,callee";
710 	const char * const report_usage[] = {
711 		"perf report [<options>]",
712 		NULL
713 	};
714 	struct report report = {
715 		.tool = {
716 			.sample		 = process_sample_event,
717 			.mmap		 = perf_event__process_mmap,
718 			.mmap2		 = perf_event__process_mmap2,
719 			.comm		 = perf_event__process_comm,
720 			.exit		 = perf_event__process_exit,
721 			.fork		 = perf_event__process_fork,
722 			.lost		 = perf_event__process_lost,
723 			.read		 = process_read_event,
724 			.attr		 = perf_event__process_attr,
725 			.tracing_data	 = perf_event__process_tracing_data,
726 			.build_id	 = perf_event__process_build_id,
727 			.ordered_samples = true,
728 			.ordering_requires_timestamps = true,
729 		},
730 		.max_stack		 = PERF_MAX_STACK_DEPTH,
731 		.pretty_printing_style	 = "normal",
732 	};
733 	const struct option options[] = {
734 	OPT_STRING('i', "input", &input_name, "file",
735 		    "input file name"),
736 	OPT_INCR('v', "verbose", &verbose,
737 		    "be more verbose (show symbol address, etc)"),
738 	OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace,
739 		    "dump raw trace in ASCII"),
740 	OPT_STRING('k', "vmlinux", &symbol_conf.vmlinux_name,
741 		   "file", "vmlinux pathname"),
742 	OPT_STRING(0, "kallsyms", &symbol_conf.kallsyms_name,
743 		   "file", "kallsyms pathname"),
744 	OPT_BOOLEAN('f', "force", &report.force, "don't complain, do it"),
745 	OPT_BOOLEAN('m', "modules", &symbol_conf.use_modules,
746 		    "load module symbols - WARNING: use only with -k and LIVE kernel"),
747 	OPT_BOOLEAN('n', "show-nr-samples", &symbol_conf.show_nr_samples,
748 		    "Show a column with the number of samples"),
749 	OPT_BOOLEAN('T', "threads", &report.show_threads,
750 		    "Show per-thread event counters"),
751 	OPT_STRING(0, "pretty", &report.pretty_printing_style, "key",
752 		   "pretty printing style key: normal raw"),
753 	OPT_BOOLEAN(0, "tui", &report.use_tui, "Use the TUI interface"),
754 	OPT_BOOLEAN(0, "gtk", &report.use_gtk, "Use the GTK2 interface"),
755 	OPT_BOOLEAN(0, "stdio", &report.use_stdio,
756 		    "Use the stdio interface"),
757 	OPT_BOOLEAN(0, "header", &report.header, "Show data header."),
758 	OPT_BOOLEAN(0, "header-only", &report.header_only,
759 		    "Show only data header."),
760 	OPT_STRING('s', "sort", &sort_order, "key[,key2...]",
761 		   "sort by key(s): pid, comm, dso, symbol, parent, cpu, srcline,"
762 		   " dso_to, dso_from, symbol_to, symbol_from, mispredict,"
763 		   " weight, local_weight, mem, symbol_daddr, dso_daddr, tlb, "
764 		   "snoop, locked, abort, in_tx, transaction"),
765 	OPT_BOOLEAN(0, "showcpuutilization", &symbol_conf.show_cpu_utilization,
766 		    "Show sample percentage for different cpu modes"),
767 	OPT_STRING('p', "parent", &parent_pattern, "regex",
768 		   "regex filter to identify parent, see: '--sort parent'"),
769 	OPT_BOOLEAN('x', "exclude-other", &symbol_conf.exclude_other,
770 		    "Only display entries with parent-match"),
771 	OPT_CALLBACK_DEFAULT('g', "call-graph", &report, "output_type,min_percent[,print_limit],call_order",
772 		     "Display callchains using output_type (graph, flat, fractal, or none) , min percent threshold, optional print limit, callchain order, key (function or address). "
773 		     "Default: fractal,0.5,callee,function", &parse_callchain_opt, callchain_default_opt),
774 	OPT_INTEGER(0, "max-stack", &report.max_stack,
775 		    "Set the maximum stack depth when parsing the callchain, "
776 		    "anything beyond the specified depth will be ignored. "
777 		    "Default: " __stringify(PERF_MAX_STACK_DEPTH)),
778 	OPT_BOOLEAN('G', "inverted", &report.inverted_callchain,
779 		    "alias for inverted call graph"),
780 	OPT_CALLBACK(0, "ignore-callees", NULL, "regex",
781 		   "ignore callees of these functions in call graphs",
782 		   report_parse_ignore_callees_opt),
783 	OPT_STRING('d', "dsos", &symbol_conf.dso_list_str, "dso[,dso...]",
784 		   "only consider symbols in these dsos"),
785 	OPT_STRING('c', "comms", &symbol_conf.comm_list_str, "comm[,comm...]",
786 		   "only consider symbols in these comms"),
787 	OPT_STRING('S', "symbols", &symbol_conf.sym_list_str, "symbol[,symbol...]",
788 		   "only consider these symbols"),
789 	OPT_STRING(0, "symbol-filter", &report.symbol_filter_str, "filter",
790 		   "only show symbols that (partially) match with this filter"),
791 	OPT_STRING('w', "column-widths", &symbol_conf.col_width_list_str,
792 		   "width[,width...]",
793 		   "don't try to adjust column width, use these fixed values"),
794 	OPT_STRING('t', "field-separator", &symbol_conf.field_sep, "separator",
795 		   "separator for columns, no spaces will be added between "
796 		   "columns '.' is reserved."),
797 	OPT_BOOLEAN('U', "hide-unresolved", &report.hide_unresolved,
798 		    "Only display entries resolved to a symbol"),
799 	OPT_STRING(0, "symfs", &symbol_conf.symfs, "directory",
800 		    "Look for files with symbols relative to this directory"),
801 	OPT_STRING('C', "cpu", &report.cpu_list, "cpu",
802 		   "list of cpus to profile"),
803 	OPT_BOOLEAN('I', "show-info", &report.show_full_info,
804 		    "Display extended information about perf.data file"),
805 	OPT_BOOLEAN(0, "source", &symbol_conf.annotate_src,
806 		    "Interleave source code with assembly code (default)"),
807 	OPT_BOOLEAN(0, "asm-raw", &symbol_conf.annotate_asm_raw,
808 		    "Display raw encoding of assembly instructions (default)"),
809 	OPT_STRING('M', "disassembler-style", &disassembler_style, "disassembler style",
810 		   "Specify disassembler style (e.g. -M intel for intel syntax)"),
811 	OPT_BOOLEAN(0, "show-total-period", &symbol_conf.show_total_period,
812 		    "Show a column with the sum of periods"),
813 	OPT_BOOLEAN(0, "group", &symbol_conf.event_group,
814 		    "Show event group information together"),
815 	OPT_CALLBACK_NOOPT('b', "branch-stack", &branch_mode, "",
816 		    "use branch records for histogram filling", parse_branch_mode),
817 	OPT_STRING(0, "objdump", &objdump_path, "path",
818 		   "objdump binary to use for disassembly and annotations"),
819 	OPT_BOOLEAN(0, "demangle", &symbol_conf.demangle,
820 		    "Disable symbol demangling"),
821 	OPT_BOOLEAN(0, "mem-mode", &report.mem_mode, "mem access profile"),
822 	OPT_CALLBACK(0, "percent-limit", &report, "percent",
823 		     "Don't show entries under that percent", parse_percent_limit),
824 	OPT_END()
825 	};
826 	struct perf_data_file file = {
827 		.mode  = PERF_DATA_MODE_READ,
828 	};
829 
830 	perf_config(report__config, &report);
831 
832 	argc = parse_options(argc, argv, options, report_usage, 0);
833 
834 	if (report.use_stdio)
835 		use_browser = 0;
836 	else if (report.use_tui)
837 		use_browser = 1;
838 	else if (report.use_gtk)
839 		use_browser = 2;
840 
841 	if (report.inverted_callchain)
842 		callchain_param.order = ORDER_CALLER;
843 
844 	if (!input_name || !strlen(input_name)) {
845 		if (!fstat(STDIN_FILENO, &st) && S_ISFIFO(st.st_mode))
846 			input_name = "-";
847 		else
848 			input_name = "perf.data";
849 	}
850 
851 	file.path  = input_name;
852 	file.force = report.force;
853 
854 repeat:
855 	session = perf_session__new(&file, false, &report.tool);
856 	if (session == NULL)
857 		return -ENOMEM;
858 
859 	report.session = session;
860 
861 	has_br_stack = perf_header__has_feat(&session->header,
862 					     HEADER_BRANCH_STACK);
863 
864 	if (branch_mode == -1 && has_br_stack)
865 		sort__mode = SORT_MODE__BRANCH;
866 
867 	/* sort__mode could be NORMAL if --no-branch-stack */
868 	if (sort__mode == SORT_MODE__BRANCH) {
869 		/*
870 		 * if no sort_order is provided, then specify
871 		 * branch-mode specific order
872 		 */
873 		if (sort_order == default_sort_order)
874 			sort_order = "comm,dso_from,symbol_from,"
875 				     "dso_to,symbol_to";
876 
877 	}
878 	if (report.mem_mode) {
879 		if (sort__mode == SORT_MODE__BRANCH) {
880 			pr_err("branch and mem mode incompatible\n");
881 			goto error;
882 		}
883 		sort__mode = SORT_MODE__MEMORY;
884 
885 		/*
886 		 * if no sort_order is provided, then specify
887 		 * branch-mode specific order
888 		 */
889 		if (sort_order == default_sort_order)
890 			sort_order = "local_weight,mem,sym,dso,symbol_daddr,dso_daddr,snoop,tlb,locked";
891 	}
892 
893 	if (setup_sorting() < 0) {
894 		parse_options_usage(report_usage, options, "s", 1);
895 		goto error;
896 	}
897 
898 	if (parent_pattern != default_parent_pattern) {
899 		if (sort_dimension__add("parent") < 0)
900 			goto error;
901 	}
902 
903 	/* Force tty output for header output. */
904 	if (report.header || report.header_only)
905 		use_browser = 0;
906 
907 	if (strcmp(input_name, "-") != 0)
908 		setup_browser(true);
909 	else {
910 		use_browser = 0;
911 		perf_hpp__init();
912 	}
913 
914 	if (report.header || report.header_only) {
915 		perf_session__fprintf_info(session, stdout,
916 					   report.show_full_info);
917 		if (report.header_only)
918 			return 0;
919 	} else if (use_browser == 0) {
920 		fputs("# To display the perf.data header info, please use --header/--header-only options.\n#\n",
921 		      stdout);
922 	}
923 
924 	/*
925 	 * Only in the TUI browser we are doing integrated annotation,
926 	 * so don't allocate extra space that won't be used in the stdio
927 	 * implementation.
928 	 */
929 	if (use_browser == 1 && sort__has_sym) {
930 		symbol_conf.priv_size = sizeof(struct annotation);
931 		machines__set_symbol_filter(&session->machines,
932 					    symbol__annotate_init);
933 		/*
934  		 * For searching by name on the "Browse map details".
935  		 * providing it only in verbose mode not to bloat too
936  		 * much struct symbol.
937  		 */
938 		if (verbose) {
939 			/*
940 			 * XXX: Need to provide a less kludgy way to ask for
941 			 * more space per symbol, the u32 is for the index on
942 			 * the ui browser.
943 			 * See symbol__browser_index.
944 			 */
945 			symbol_conf.priv_size += sizeof(u32);
946 			symbol_conf.sort_by_name = true;
947 		}
948 	}
949 
950 	if (symbol__init() < 0)
951 		goto error;
952 
953 	if (argc) {
954 		/*
955 		 * Special case: if there's an argument left then assume that
956 		 * it's a symbol filter:
957 		 */
958 		if (argc > 1)
959 			usage_with_options(report_usage, options);
960 
961 		report.symbol_filter_str = argv[0];
962 	}
963 
964 	sort__setup_elide(stdout);
965 
966 	ret = __cmd_report(&report);
967 	if (ret == K_SWITCH_INPUT_DATA) {
968 		perf_session__delete(session);
969 		goto repeat;
970 	} else
971 		ret = 0;
972 
973 error:
974 	perf_session__delete(session);
975 	return ret;
976 }
977