xref: /openbmc/linux/tools/perf/builtin-report.c (revision 4a3fad70)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * builtin-report.c
4  *
5  * Builtin report command: Analyze the perf.data input file,
6  * look up and read DSOs and symbol information and display
7  * a histogram of results, along various sorting keys.
8  */
9 #include "builtin.h"
10 
11 #include "util/util.h"
12 #include "util/config.h"
13 
14 #include "util/annotate.h"
15 #include "util/color.h"
16 #include <linux/list.h>
17 #include <linux/rbtree.h>
18 #include "util/symbol.h"
19 #include "util/callchain.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 <subcmd/parse-options.h>
31 #include <subcmd/exec-cmd.h>
32 #include "util/parse-events.h"
33 
34 #include "util/thread.h"
35 #include "util/sort.h"
36 #include "util/hist.h"
37 #include "util/data.h"
38 #include "arch/common.h"
39 #include "util/time-utils.h"
40 #include "util/auxtrace.h"
41 #include "util/units.h"
42 #include "util/branch.h"
43 
44 #include <dlfcn.h>
45 #include <errno.h>
46 #include <inttypes.h>
47 #include <regex.h>
48 #include <signal.h>
49 #include <linux/bitmap.h>
50 #include <linux/stringify.h>
51 #include <sys/types.h>
52 #include <sys/stat.h>
53 #include <unistd.h>
54 
55 struct report {
56 	struct perf_tool	tool;
57 	struct perf_session	*session;
58 	bool			use_tui, use_gtk, use_stdio;
59 	bool			show_full_info;
60 	bool			show_threads;
61 	bool			inverted_callchain;
62 	bool			mem_mode;
63 	bool			header;
64 	bool			header_only;
65 	bool			nonany_branch_mode;
66 	int			max_stack;
67 	struct perf_read_values	show_threads_values;
68 	const char		*pretty_printing_style;
69 	const char		*cpu_list;
70 	const char		*symbol_filter_str;
71 	const char		*time_str;
72 	struct perf_time_interval ptime;
73 	float			min_percent;
74 	u64			nr_entries;
75 	u64			queue_size;
76 	int			socket_filter;
77 	DECLARE_BITMAP(cpu_bitmap, MAX_NR_CPUS);
78 	struct branch_type_stat	brtype_stat;
79 };
80 
81 static int report__config(const char *var, const char *value, void *cb)
82 {
83 	struct report *rep = cb;
84 
85 	if (!strcmp(var, "report.group")) {
86 		symbol_conf.event_group = perf_config_bool(var, value);
87 		return 0;
88 	}
89 	if (!strcmp(var, "report.percent-limit")) {
90 		double pcnt = strtof(value, NULL);
91 
92 		rep->min_percent = pcnt;
93 		callchain_param.min_percent = pcnt;
94 		return 0;
95 	}
96 	if (!strcmp(var, "report.children")) {
97 		symbol_conf.cumulate_callchain = perf_config_bool(var, value);
98 		return 0;
99 	}
100 	if (!strcmp(var, "report.queue-size"))
101 		return perf_config_u64(&rep->queue_size, var, value);
102 
103 	if (!strcmp(var, "report.sort_order")) {
104 		default_sort_order = strdup(value);
105 		return 0;
106 	}
107 
108 	return 0;
109 }
110 
111 static int hist_iter__report_callback(struct hist_entry_iter *iter,
112 				      struct addr_location *al, bool single,
113 				      void *arg)
114 {
115 	int err = 0;
116 	struct report *rep = arg;
117 	struct hist_entry *he = iter->he;
118 	struct perf_evsel *evsel = iter->evsel;
119 	struct perf_sample *sample = iter->sample;
120 	struct mem_info *mi;
121 	struct branch_info *bi;
122 
123 	if (!ui__has_annotation())
124 		return 0;
125 
126 	hist__account_cycles(sample->branch_stack, al, sample,
127 			     rep->nonany_branch_mode);
128 
129 	if (sort__mode == SORT_MODE__BRANCH) {
130 		bi = he->branch_info;
131 		err = addr_map_symbol__inc_samples(&bi->from, sample, evsel->idx);
132 		if (err)
133 			goto out;
134 
135 		err = addr_map_symbol__inc_samples(&bi->to, sample, evsel->idx);
136 
137 	} else if (rep->mem_mode) {
138 		mi = he->mem_info;
139 		err = addr_map_symbol__inc_samples(&mi->daddr, sample, evsel->idx);
140 		if (err)
141 			goto out;
142 
143 		err = hist_entry__inc_addr_samples(he, sample, evsel->idx, al->addr);
144 
145 	} else if (symbol_conf.cumulate_callchain) {
146 		if (single)
147 			err = hist_entry__inc_addr_samples(he, sample, evsel->idx,
148 							   al->addr);
149 	} else {
150 		err = hist_entry__inc_addr_samples(he, sample, evsel->idx, al->addr);
151 	}
152 
153 out:
154 	return err;
155 }
156 
157 static int hist_iter__branch_callback(struct hist_entry_iter *iter,
158 				      struct addr_location *al __maybe_unused,
159 				      bool single __maybe_unused,
160 				      void *arg)
161 {
162 	struct hist_entry *he = iter->he;
163 	struct report *rep = arg;
164 	struct branch_info *bi;
165 
166 	bi = he->branch_info;
167 	branch_type_count(&rep->brtype_stat, &bi->flags,
168 			  bi->from.addr, bi->to.addr);
169 
170 	return 0;
171 }
172 
173 static int process_sample_event(struct perf_tool *tool,
174 				union perf_event *event,
175 				struct perf_sample *sample,
176 				struct perf_evsel *evsel,
177 				struct machine *machine)
178 {
179 	struct report *rep = container_of(tool, struct report, tool);
180 	struct addr_location al;
181 	struct hist_entry_iter iter = {
182 		.evsel 			= evsel,
183 		.sample 		= sample,
184 		.hide_unresolved 	= symbol_conf.hide_unresolved,
185 		.add_entry_cb 		= hist_iter__report_callback,
186 	};
187 	int ret = 0;
188 
189 	if (perf_time__skip_sample(&rep->ptime, sample->time))
190 		return 0;
191 
192 	if (machine__resolve(machine, &al, sample) < 0) {
193 		pr_debug("problem processing %d event, skipping it.\n",
194 			 event->header.type);
195 		return -1;
196 	}
197 
198 	if (symbol_conf.hide_unresolved && al.sym == NULL)
199 		goto out_put;
200 
201 	if (rep->cpu_list && !test_bit(sample->cpu, rep->cpu_bitmap))
202 		goto out_put;
203 
204 	if (sort__mode == SORT_MODE__BRANCH) {
205 		/*
206 		 * A non-synthesized event might not have a branch stack if
207 		 * branch stacks have been synthesized (using itrace options).
208 		 */
209 		if (!sample->branch_stack)
210 			goto out_put;
211 
212 		iter.add_entry_cb = hist_iter__branch_callback;
213 		iter.ops = &hist_iter_branch;
214 	} else if (rep->mem_mode) {
215 		iter.ops = &hist_iter_mem;
216 	} else if (symbol_conf.cumulate_callchain) {
217 		iter.ops = &hist_iter_cumulative;
218 	} else {
219 		iter.ops = &hist_iter_normal;
220 	}
221 
222 	if (al.map != NULL)
223 		al.map->dso->hit = 1;
224 
225 	ret = hist_entry_iter__add(&iter, &al, rep->max_stack, rep);
226 	if (ret < 0)
227 		pr_debug("problem adding hist entry, skipping event\n");
228 out_put:
229 	addr_location__put(&al);
230 	return ret;
231 }
232 
233 static int process_read_event(struct perf_tool *tool,
234 			      union perf_event *event,
235 			      struct perf_sample *sample __maybe_unused,
236 			      struct perf_evsel *evsel,
237 			      struct machine *machine __maybe_unused)
238 {
239 	struct report *rep = container_of(tool, struct report, tool);
240 
241 	if (rep->show_threads) {
242 		const char *name = evsel ? perf_evsel__name(evsel) : "unknown";
243 		int err = perf_read_values_add_value(&rep->show_threads_values,
244 					   event->read.pid, event->read.tid,
245 					   evsel->idx,
246 					   name,
247 					   event->read.value);
248 
249 		if (err)
250 			return err;
251 	}
252 
253 	return 0;
254 }
255 
256 /* For pipe mode, sample_type is not currently set */
257 static int report__setup_sample_type(struct report *rep)
258 {
259 	struct perf_session *session = rep->session;
260 	u64 sample_type = perf_evlist__combined_sample_type(session->evlist);
261 	bool is_pipe = perf_data__is_pipe(session->data);
262 
263 	if (session->itrace_synth_opts->callchain ||
264 	    (!is_pipe &&
265 	     perf_header__has_feat(&session->header, HEADER_AUXTRACE) &&
266 	     !session->itrace_synth_opts->set))
267 		sample_type |= PERF_SAMPLE_CALLCHAIN;
268 
269 	if (session->itrace_synth_opts->last_branch)
270 		sample_type |= PERF_SAMPLE_BRANCH_STACK;
271 
272 	if (!is_pipe && !(sample_type & PERF_SAMPLE_CALLCHAIN)) {
273 		if (perf_hpp_list.parent) {
274 			ui__error("Selected --sort parent, but no "
275 				    "callchain data. Did you call "
276 				    "'perf record' without -g?\n");
277 			return -EINVAL;
278 		}
279 		if (symbol_conf.use_callchain &&
280 			!symbol_conf.show_branchflag_count) {
281 			ui__error("Selected -g or --branch-history.\n"
282 				  "But no callchain or branch data.\n"
283 				  "Did you call 'perf record' without -g or -b?\n");
284 			return -1;
285 		}
286 	} else if (!callchain_param.enabled &&
287 		   callchain_param.mode != CHAIN_NONE &&
288 		   !symbol_conf.use_callchain) {
289 			symbol_conf.use_callchain = true;
290 			if (callchain_register_param(&callchain_param) < 0) {
291 				ui__error("Can't register callchain params.\n");
292 				return -EINVAL;
293 			}
294 	}
295 
296 	if (symbol_conf.cumulate_callchain) {
297 		/* Silently ignore if callchain is missing */
298 		if (!(sample_type & PERF_SAMPLE_CALLCHAIN)) {
299 			symbol_conf.cumulate_callchain = false;
300 			perf_hpp__cancel_cumulate();
301 		}
302 	}
303 
304 	if (sort__mode == SORT_MODE__BRANCH) {
305 		if (!is_pipe &&
306 		    !(sample_type & PERF_SAMPLE_BRANCH_STACK)) {
307 			ui__error("Selected -b but no branch data. "
308 				  "Did you call perf record without -b?\n");
309 			return -1;
310 		}
311 	}
312 
313 	if (symbol_conf.use_callchain || symbol_conf.cumulate_callchain) {
314 		if ((sample_type & PERF_SAMPLE_REGS_USER) &&
315 		    (sample_type & PERF_SAMPLE_STACK_USER))
316 			callchain_param.record_mode = CALLCHAIN_DWARF;
317 		else if (sample_type & PERF_SAMPLE_BRANCH_STACK)
318 			callchain_param.record_mode = CALLCHAIN_LBR;
319 		else
320 			callchain_param.record_mode = CALLCHAIN_FP;
321 	}
322 
323 	/* ??? handle more cases than just ANY? */
324 	if (!(perf_evlist__combined_branch_type(session->evlist) &
325 				PERF_SAMPLE_BRANCH_ANY))
326 		rep->nonany_branch_mode = true;
327 
328 	return 0;
329 }
330 
331 static void sig_handler(int sig __maybe_unused)
332 {
333 	session_done = 1;
334 }
335 
336 static size_t hists__fprintf_nr_sample_events(struct hists *hists, struct report *rep,
337 					      const char *evname, FILE *fp)
338 {
339 	size_t ret;
340 	char unit;
341 	unsigned long nr_samples = hists->stats.nr_events[PERF_RECORD_SAMPLE];
342 	u64 nr_events = hists->stats.total_period;
343 	struct perf_evsel *evsel = hists_to_evsel(hists);
344 	char buf[512];
345 	size_t size = sizeof(buf);
346 	int socked_id = hists->socket_filter;
347 
348 	if (quiet)
349 		return 0;
350 
351 	if (symbol_conf.filter_relative) {
352 		nr_samples = hists->stats.nr_non_filtered_samples;
353 		nr_events = hists->stats.total_non_filtered_period;
354 	}
355 
356 	if (perf_evsel__is_group_event(evsel)) {
357 		struct perf_evsel *pos;
358 
359 		perf_evsel__group_desc(evsel, buf, size);
360 		evname = buf;
361 
362 		for_each_group_member(pos, evsel) {
363 			const struct hists *pos_hists = evsel__hists(pos);
364 
365 			if (symbol_conf.filter_relative) {
366 				nr_samples += pos_hists->stats.nr_non_filtered_samples;
367 				nr_events += pos_hists->stats.total_non_filtered_period;
368 			} else {
369 				nr_samples += pos_hists->stats.nr_events[PERF_RECORD_SAMPLE];
370 				nr_events += pos_hists->stats.total_period;
371 			}
372 		}
373 	}
374 
375 	nr_samples = convert_unit(nr_samples, &unit);
376 	ret = fprintf(fp, "# Samples: %lu%c", nr_samples, unit);
377 	if (evname != NULL)
378 		ret += fprintf(fp, " of event '%s'", evname);
379 
380 	if (symbol_conf.show_ref_callgraph &&
381 	    strstr(evname, "call-graph=no")) {
382 		ret += fprintf(fp, ", show reference callgraph");
383 	}
384 
385 	if (rep->mem_mode) {
386 		ret += fprintf(fp, "\n# Total weight : %" PRIu64, nr_events);
387 		ret += fprintf(fp, "\n# Sort order   : %s", sort_order ? : default_mem_sort_order);
388 	} else
389 		ret += fprintf(fp, "\n# Event count (approx.): %" PRIu64, nr_events);
390 
391 	if (socked_id > -1)
392 		ret += fprintf(fp, "\n# Processor Socket: %d", socked_id);
393 
394 	return ret + fprintf(fp, "\n#\n");
395 }
396 
397 static int perf_evlist__tty_browse_hists(struct perf_evlist *evlist,
398 					 struct report *rep,
399 					 const char *help)
400 {
401 	struct perf_evsel *pos;
402 
403 	if (!quiet) {
404 		fprintf(stdout, "#\n# Total Lost Samples: %" PRIu64 "\n#\n",
405 			evlist->stats.total_lost_samples);
406 	}
407 
408 	evlist__for_each_entry(evlist, pos) {
409 		struct hists *hists = evsel__hists(pos);
410 		const char *evname = perf_evsel__name(pos);
411 
412 		if (symbol_conf.event_group &&
413 		    !perf_evsel__is_group_leader(pos))
414 			continue;
415 
416 		hists__fprintf_nr_sample_events(hists, rep, evname, stdout);
417 		hists__fprintf(hists, !quiet, 0, 0, rep->min_percent, stdout,
418 			       symbol_conf.use_callchain ||
419 			       symbol_conf.show_branchflag_count);
420 		fprintf(stdout, "\n\n");
421 	}
422 
423 	if (!quiet)
424 		fprintf(stdout, "#\n# (%s)\n#\n", help);
425 
426 	if (rep->show_threads) {
427 		bool style = !strcmp(rep->pretty_printing_style, "raw");
428 		perf_read_values_display(stdout, &rep->show_threads_values,
429 					 style);
430 		perf_read_values_destroy(&rep->show_threads_values);
431 	}
432 
433 	if (sort__mode == SORT_MODE__BRANCH)
434 		branch_type_stat_display(stdout, &rep->brtype_stat);
435 
436 	return 0;
437 }
438 
439 static void report__warn_kptr_restrict(const struct report *rep)
440 {
441 	struct map *kernel_map = machine__kernel_map(&rep->session->machines.host);
442 	struct kmap *kernel_kmap = kernel_map ? map__kmap(kernel_map) : NULL;
443 
444 	if (perf_evlist__exclude_kernel(rep->session->evlist))
445 		return;
446 
447 	if (kernel_map == NULL ||
448 	    (kernel_map->dso->hit &&
449 	     (kernel_kmap->ref_reloc_sym == NULL ||
450 	      kernel_kmap->ref_reloc_sym->addr == 0))) {
451 		const char *desc =
452 		    "As no suitable kallsyms nor vmlinux was found, kernel samples\n"
453 		    "can't be resolved.";
454 
455 		if (kernel_map) {
456 			const struct dso *kdso = kernel_map->dso;
457 			if (!RB_EMPTY_ROOT(&kdso->symbols[MAP__FUNCTION])) {
458 				desc = "If some relocation was applied (e.g. "
459 				       "kexec) symbols may be misresolved.";
460 			}
461 		}
462 
463 		ui__warning(
464 "Kernel address maps (/proc/{kallsyms,modules}) were restricted.\n\n"
465 "Check /proc/sys/kernel/kptr_restrict before running 'perf record'.\n\n%s\n\n"
466 "Samples in kernel modules can't be resolved as well.\n\n",
467 		desc);
468 	}
469 }
470 
471 static int report__gtk_browse_hists(struct report *rep, const char *help)
472 {
473 	int (*hist_browser)(struct perf_evlist *evlist, const char *help,
474 			    struct hist_browser_timer *timer, float min_pcnt);
475 
476 	hist_browser = dlsym(perf_gtk_handle, "perf_evlist__gtk_browse_hists");
477 
478 	if (hist_browser == NULL) {
479 		ui__error("GTK browser not found!\n");
480 		return -1;
481 	}
482 
483 	return hist_browser(rep->session->evlist, help, NULL, rep->min_percent);
484 }
485 
486 static int report__browse_hists(struct report *rep)
487 {
488 	int ret;
489 	struct perf_session *session = rep->session;
490 	struct perf_evlist *evlist = session->evlist;
491 	const char *help = perf_tip(system_path(TIPDIR));
492 
493 	if (help == NULL) {
494 		/* fallback for people who don't install perf ;-) */
495 		help = perf_tip(DOCDIR);
496 		if (help == NULL)
497 			help = "Cannot load tips.txt file, please install perf!";
498 	}
499 
500 	switch (use_browser) {
501 	case 1:
502 		ret = perf_evlist__tui_browse_hists(evlist, help, NULL,
503 						    rep->min_percent,
504 						    &session->header.env);
505 		/*
506 		 * Usually "ret" is the last pressed key, and we only
507 		 * care if the key notifies us to switch data file.
508 		 */
509 		if (ret != K_SWITCH_INPUT_DATA)
510 			ret = 0;
511 		break;
512 	case 2:
513 		ret = report__gtk_browse_hists(rep, help);
514 		break;
515 	default:
516 		ret = perf_evlist__tty_browse_hists(evlist, rep, help);
517 		break;
518 	}
519 
520 	return ret;
521 }
522 
523 static int report__collapse_hists(struct report *rep)
524 {
525 	struct ui_progress prog;
526 	struct perf_evsel *pos;
527 	int ret = 0;
528 
529 	ui_progress__init(&prog, rep->nr_entries, "Merging related events...");
530 
531 	evlist__for_each_entry(rep->session->evlist, pos) {
532 		struct hists *hists = evsel__hists(pos);
533 
534 		if (pos->idx == 0)
535 			hists->symbol_filter_str = rep->symbol_filter_str;
536 
537 		hists->socket_filter = rep->socket_filter;
538 
539 		ret = hists__collapse_resort(hists, &prog);
540 		if (ret < 0)
541 			break;
542 
543 		/* Non-group events are considered as leader */
544 		if (symbol_conf.event_group &&
545 		    !perf_evsel__is_group_leader(pos)) {
546 			struct hists *leader_hists = evsel__hists(pos->leader);
547 
548 			hists__match(leader_hists, hists);
549 			hists__link(leader_hists, hists);
550 		}
551 	}
552 
553 	ui_progress__finish();
554 	return ret;
555 }
556 
557 static void report__output_resort(struct report *rep)
558 {
559 	struct ui_progress prog;
560 	struct perf_evsel *pos;
561 
562 	ui_progress__init(&prog, rep->nr_entries, "Sorting events for output...");
563 
564 	evlist__for_each_entry(rep->session->evlist, pos)
565 		perf_evsel__output_resort(pos, &prog);
566 
567 	ui_progress__finish();
568 }
569 
570 static int __cmd_report(struct report *rep)
571 {
572 	int ret;
573 	struct perf_session *session = rep->session;
574 	struct perf_evsel *pos;
575 	struct perf_data *data = session->data;
576 
577 	signal(SIGINT, sig_handler);
578 
579 	if (rep->cpu_list) {
580 		ret = perf_session__cpu_bitmap(session, rep->cpu_list,
581 					       rep->cpu_bitmap);
582 		if (ret) {
583 			ui__error("failed to set cpu bitmap\n");
584 			return ret;
585 		}
586 		session->itrace_synth_opts->cpu_bitmap = rep->cpu_bitmap;
587 	}
588 
589 	if (rep->show_threads) {
590 		ret = perf_read_values_init(&rep->show_threads_values);
591 		if (ret)
592 			return ret;
593 	}
594 
595 	ret = report__setup_sample_type(rep);
596 	if (ret) {
597 		/* report__setup_sample_type() already showed error message */
598 		return ret;
599 	}
600 
601 	ret = perf_session__process_events(session);
602 	if (ret) {
603 		ui__error("failed to process sample\n");
604 		return ret;
605 	}
606 
607 	report__warn_kptr_restrict(rep);
608 
609 	evlist__for_each_entry(session->evlist, pos)
610 		rep->nr_entries += evsel__hists(pos)->nr_entries;
611 
612 	if (use_browser == 0) {
613 		if (verbose > 3)
614 			perf_session__fprintf(session, stdout);
615 
616 		if (verbose > 2)
617 			perf_session__fprintf_dsos(session, stdout);
618 
619 		if (dump_trace) {
620 			perf_session__fprintf_nr_events(session, stdout);
621 			perf_evlist__fprintf_nr_events(session->evlist, stdout);
622 			return 0;
623 		}
624 	}
625 
626 	ret = report__collapse_hists(rep);
627 	if (ret) {
628 		ui__error("failed to process hist entry\n");
629 		return ret;
630 	}
631 
632 	if (session_done())
633 		return 0;
634 
635 	/*
636 	 * recalculate number of entries after collapsing since it
637 	 * might be changed during the collapse phase.
638 	 */
639 	rep->nr_entries = 0;
640 	evlist__for_each_entry(session->evlist, pos)
641 		rep->nr_entries += evsel__hists(pos)->nr_entries;
642 
643 	if (rep->nr_entries == 0) {
644 		ui__error("The %s file has no samples!\n", data->file.path);
645 		return 0;
646 	}
647 
648 	report__output_resort(rep);
649 
650 	return report__browse_hists(rep);
651 }
652 
653 static int
654 report_parse_callchain_opt(const struct option *opt, const char *arg, int unset)
655 {
656 	struct callchain_param *callchain = opt->value;
657 
658 	callchain->enabled = !unset;
659 	/*
660 	 * --no-call-graph
661 	 */
662 	if (unset) {
663 		symbol_conf.use_callchain = false;
664 		callchain->mode = CHAIN_NONE;
665 		return 0;
666 	}
667 
668 	return parse_callchain_report_opt(arg);
669 }
670 
671 int
672 report_parse_ignore_callees_opt(const struct option *opt __maybe_unused,
673 				const char *arg, int unset __maybe_unused)
674 {
675 	if (arg) {
676 		int err = regcomp(&ignore_callees_regex, arg, REG_EXTENDED);
677 		if (err) {
678 			char buf[BUFSIZ];
679 			regerror(err, &ignore_callees_regex, buf, sizeof(buf));
680 			pr_err("Invalid --ignore-callees regex: %s\n%s", arg, buf);
681 			return -1;
682 		}
683 		have_ignore_callees = 1;
684 	}
685 
686 	return 0;
687 }
688 
689 static int
690 parse_branch_mode(const struct option *opt,
691 		  const char *str __maybe_unused, int unset)
692 {
693 	int *branch_mode = opt->value;
694 
695 	*branch_mode = !unset;
696 	return 0;
697 }
698 
699 static int
700 parse_percent_limit(const struct option *opt, const char *str,
701 		    int unset __maybe_unused)
702 {
703 	struct report *rep = opt->value;
704 	double pcnt = strtof(str, NULL);
705 
706 	rep->min_percent = pcnt;
707 	callchain_param.min_percent = pcnt;
708 	return 0;
709 }
710 
711 #define CALLCHAIN_DEFAULT_OPT  "graph,0.5,caller,function,percent"
712 
713 const char report_callchain_help[] = "Display call graph (stack chain/backtrace):\n\n"
714 				     CALLCHAIN_REPORT_HELP
715 				     "\n\t\t\t\tDefault: " CALLCHAIN_DEFAULT_OPT;
716 
717 int cmd_report(int argc, const char **argv)
718 {
719 	struct perf_session *session;
720 	struct itrace_synth_opts itrace_synth_opts = { .set = 0, };
721 	struct stat st;
722 	bool has_br_stack = false;
723 	int branch_mode = -1;
724 	bool branch_call_mode = false;
725 	char callchain_default_opt[] = CALLCHAIN_DEFAULT_OPT;
726 	const char * const report_usage[] = {
727 		"perf report [<options>]",
728 		NULL
729 	};
730 	struct report report = {
731 		.tool = {
732 			.sample		 = process_sample_event,
733 			.mmap		 = perf_event__process_mmap,
734 			.mmap2		 = perf_event__process_mmap2,
735 			.comm		 = perf_event__process_comm,
736 			.namespaces	 = perf_event__process_namespaces,
737 			.exit		 = perf_event__process_exit,
738 			.fork		 = perf_event__process_fork,
739 			.lost		 = perf_event__process_lost,
740 			.read		 = process_read_event,
741 			.attr		 = perf_event__process_attr,
742 			.tracing_data	 = perf_event__process_tracing_data,
743 			.build_id	 = perf_event__process_build_id,
744 			.id_index	 = perf_event__process_id_index,
745 			.auxtrace_info	 = perf_event__process_auxtrace_info,
746 			.auxtrace	 = perf_event__process_auxtrace,
747 			.feature	 = perf_event__process_feature,
748 			.ordered_events	 = true,
749 			.ordering_requires_timestamps = true,
750 		},
751 		.max_stack		 = PERF_MAX_STACK_DEPTH,
752 		.pretty_printing_style	 = "normal",
753 		.socket_filter		 = -1,
754 	};
755 	const struct option options[] = {
756 	OPT_STRING('i', "input", &input_name, "file",
757 		    "input file name"),
758 	OPT_INCR('v', "verbose", &verbose,
759 		    "be more verbose (show symbol address, etc)"),
760 	OPT_BOOLEAN('q', "quiet", &quiet, "Do not show any message"),
761 	OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace,
762 		    "dump raw trace in ASCII"),
763 	OPT_STRING('k', "vmlinux", &symbol_conf.vmlinux_name,
764 		   "file", "vmlinux pathname"),
765 	OPT_STRING(0, "kallsyms", &symbol_conf.kallsyms_name,
766 		   "file", "kallsyms pathname"),
767 	OPT_BOOLEAN('f', "force", &symbol_conf.force, "don't complain, do it"),
768 	OPT_BOOLEAN('m', "modules", &symbol_conf.use_modules,
769 		    "load module symbols - WARNING: use only with -k and LIVE kernel"),
770 	OPT_BOOLEAN('n', "show-nr-samples", &symbol_conf.show_nr_samples,
771 		    "Show a column with the number of samples"),
772 	OPT_BOOLEAN('T', "threads", &report.show_threads,
773 		    "Show per-thread event counters"),
774 	OPT_STRING(0, "pretty", &report.pretty_printing_style, "key",
775 		   "pretty printing style key: normal raw"),
776 	OPT_BOOLEAN(0, "tui", &report.use_tui, "Use the TUI interface"),
777 	OPT_BOOLEAN(0, "gtk", &report.use_gtk, "Use the GTK2 interface"),
778 	OPT_BOOLEAN(0, "stdio", &report.use_stdio,
779 		    "Use the stdio interface"),
780 	OPT_BOOLEAN(0, "header", &report.header, "Show data header."),
781 	OPT_BOOLEAN(0, "header-only", &report.header_only,
782 		    "Show only data header."),
783 	OPT_STRING('s', "sort", &sort_order, "key[,key2...]",
784 		   "sort by key(s): pid, comm, dso, symbol, parent, cpu, srcline, ..."
785 		   " Please refer the man page for the complete list."),
786 	OPT_STRING('F', "fields", &field_order, "key[,keys...]",
787 		   "output field(s): overhead, period, sample plus all of sort keys"),
788 	OPT_BOOLEAN(0, "show-cpu-utilization", &symbol_conf.show_cpu_utilization,
789 		    "Show sample percentage for different cpu modes"),
790 	OPT_BOOLEAN_FLAG(0, "showcpuutilization", &symbol_conf.show_cpu_utilization,
791 		    "Show sample percentage for different cpu modes", PARSE_OPT_HIDDEN),
792 	OPT_STRING('p', "parent", &parent_pattern, "regex",
793 		   "regex filter to identify parent, see: '--sort parent'"),
794 	OPT_BOOLEAN('x', "exclude-other", &symbol_conf.exclude_other,
795 		    "Only display entries with parent-match"),
796 	OPT_CALLBACK_DEFAULT('g', "call-graph", &callchain_param,
797 			     "print_type,threshold[,print_limit],order,sort_key[,branch],value",
798 			     report_callchain_help, &report_parse_callchain_opt,
799 			     callchain_default_opt),
800 	OPT_BOOLEAN(0, "children", &symbol_conf.cumulate_callchain,
801 		    "Accumulate callchains of children and show total overhead as well"),
802 	OPT_INTEGER(0, "max-stack", &report.max_stack,
803 		    "Set the maximum stack depth when parsing the callchain, "
804 		    "anything beyond the specified depth will be ignored. "
805 		    "Default: kernel.perf_event_max_stack or " __stringify(PERF_MAX_STACK_DEPTH)),
806 	OPT_BOOLEAN('G', "inverted", &report.inverted_callchain,
807 		    "alias for inverted call graph"),
808 	OPT_CALLBACK(0, "ignore-callees", NULL, "regex",
809 		   "ignore callees of these functions in call graphs",
810 		   report_parse_ignore_callees_opt),
811 	OPT_STRING('d', "dsos", &symbol_conf.dso_list_str, "dso[,dso...]",
812 		   "only consider symbols in these dsos"),
813 	OPT_STRING('c', "comms", &symbol_conf.comm_list_str, "comm[,comm...]",
814 		   "only consider symbols in these comms"),
815 	OPT_STRING(0, "pid", &symbol_conf.pid_list_str, "pid[,pid...]",
816 		   "only consider symbols in these pids"),
817 	OPT_STRING(0, "tid", &symbol_conf.tid_list_str, "tid[,tid...]",
818 		   "only consider symbols in these tids"),
819 	OPT_STRING('S', "symbols", &symbol_conf.sym_list_str, "symbol[,symbol...]",
820 		   "only consider these symbols"),
821 	OPT_STRING(0, "symbol-filter", &report.symbol_filter_str, "filter",
822 		   "only show symbols that (partially) match with this filter"),
823 	OPT_STRING('w', "column-widths", &symbol_conf.col_width_list_str,
824 		   "width[,width...]",
825 		   "don't try to adjust column width, use these fixed values"),
826 	OPT_STRING_NOEMPTY('t', "field-separator", &symbol_conf.field_sep, "separator",
827 		   "separator for columns, no spaces will be added between "
828 		   "columns '.' is reserved."),
829 	OPT_BOOLEAN('U', "hide-unresolved", &symbol_conf.hide_unresolved,
830 		    "Only display entries resolved to a symbol"),
831 	OPT_CALLBACK(0, "symfs", NULL, "directory",
832 		     "Look for files with symbols relative to this directory",
833 		     symbol__config_symfs),
834 	OPT_STRING('C', "cpu", &report.cpu_list, "cpu",
835 		   "list of cpus to profile"),
836 	OPT_BOOLEAN('I', "show-info", &report.show_full_info,
837 		    "Display extended information about perf.data file"),
838 	OPT_BOOLEAN(0, "source", &symbol_conf.annotate_src,
839 		    "Interleave source code with assembly code (default)"),
840 	OPT_BOOLEAN(0, "asm-raw", &symbol_conf.annotate_asm_raw,
841 		    "Display raw encoding of assembly instructions (default)"),
842 	OPT_STRING('M', "disassembler-style", &disassembler_style, "disassembler style",
843 		   "Specify disassembler style (e.g. -M intel for intel syntax)"),
844 	OPT_BOOLEAN(0, "show-total-period", &symbol_conf.show_total_period,
845 		    "Show a column with the sum of periods"),
846 	OPT_BOOLEAN(0, "group", &symbol_conf.event_group,
847 		    "Show event group information together"),
848 	OPT_CALLBACK_NOOPT('b', "branch-stack", &branch_mode, "",
849 		    "use branch records for per branch histogram filling",
850 		    parse_branch_mode),
851 	OPT_BOOLEAN(0, "branch-history", &branch_call_mode,
852 		    "add last branch records to call history"),
853 	OPT_STRING(0, "objdump", &objdump_path, "path",
854 		   "objdump binary to use for disassembly and annotations"),
855 	OPT_BOOLEAN(0, "demangle", &symbol_conf.demangle,
856 		    "Disable symbol demangling"),
857 	OPT_BOOLEAN(0, "demangle-kernel", &symbol_conf.demangle_kernel,
858 		    "Enable kernel symbol demangling"),
859 	OPT_BOOLEAN(0, "mem-mode", &report.mem_mode, "mem access profile"),
860 	OPT_CALLBACK(0, "percent-limit", &report, "percent",
861 		     "Don't show entries under that percent", parse_percent_limit),
862 	OPT_CALLBACK(0, "percentage", NULL, "relative|absolute",
863 		     "how to display percentage of filtered entries", parse_filter_percentage),
864 	OPT_CALLBACK_OPTARG(0, "itrace", &itrace_synth_opts, NULL, "opts",
865 			    "Instruction Tracing options",
866 			    itrace_parse_synth_opts),
867 	OPT_BOOLEAN(0, "full-source-path", &srcline_full_filename,
868 			"Show full source file name path for source lines"),
869 	OPT_BOOLEAN(0, "show-ref-call-graph", &symbol_conf.show_ref_callgraph,
870 		    "Show callgraph from reference event"),
871 	OPT_INTEGER(0, "socket-filter", &report.socket_filter,
872 		    "only show processor socket that match with this filter"),
873 	OPT_BOOLEAN(0, "raw-trace", &symbol_conf.raw_trace,
874 		    "Show raw trace event output (do not use print fmt or plugins)"),
875 	OPT_BOOLEAN(0, "hierarchy", &symbol_conf.report_hierarchy,
876 		    "Show entries in a hierarchy"),
877 	OPT_CALLBACK_DEFAULT(0, "stdio-color", NULL, "mode",
878 			     "'always' (default), 'never' or 'auto' only applicable to --stdio mode",
879 			     stdio__config_color, "always"),
880 	OPT_STRING(0, "time", &report.time_str, "str",
881 		   "Time span of interest (start,stop)"),
882 	OPT_BOOLEAN(0, "inline", &symbol_conf.inline_name,
883 		    "Show inline function"),
884 	OPT_END()
885 	};
886 	struct perf_data data = {
887 		.mode  = PERF_DATA_MODE_READ,
888 	};
889 	int ret = hists__init();
890 
891 	if (ret < 0)
892 		return ret;
893 
894 	ret = perf_config(report__config, &report);
895 	if (ret)
896 		return ret;
897 
898 	argc = parse_options(argc, argv, options, report_usage, 0);
899 	if (argc) {
900 		/*
901 		 * Special case: if there's an argument left then assume that
902 		 * it's a symbol filter:
903 		 */
904 		if (argc > 1)
905 			usage_with_options(report_usage, options);
906 
907 		report.symbol_filter_str = argv[0];
908 	}
909 
910 	if (quiet)
911 		perf_quiet_option();
912 
913 	if (symbol_conf.vmlinux_name &&
914 	    access(symbol_conf.vmlinux_name, R_OK)) {
915 		pr_err("Invalid file: %s\n", symbol_conf.vmlinux_name);
916 		return -EINVAL;
917 	}
918 	if (symbol_conf.kallsyms_name &&
919 	    access(symbol_conf.kallsyms_name, R_OK)) {
920 		pr_err("Invalid file: %s\n", symbol_conf.kallsyms_name);
921 		return -EINVAL;
922 	}
923 
924 	if (report.use_stdio)
925 		use_browser = 0;
926 	else if (report.use_tui)
927 		use_browser = 1;
928 	else if (report.use_gtk)
929 		use_browser = 2;
930 
931 	if (report.inverted_callchain)
932 		callchain_param.order = ORDER_CALLER;
933 	if (symbol_conf.cumulate_callchain && !callchain_param.order_set)
934 		callchain_param.order = ORDER_CALLER;
935 
936 	if (itrace_synth_opts.callchain &&
937 	    (int)itrace_synth_opts.callchain_sz > report.max_stack)
938 		report.max_stack = itrace_synth_opts.callchain_sz;
939 
940 	if (!input_name || !strlen(input_name)) {
941 		if (!fstat(STDIN_FILENO, &st) && S_ISFIFO(st.st_mode))
942 			input_name = "-";
943 		else
944 			input_name = "perf.data";
945 	}
946 
947 	data.file.path = input_name;
948 	data.force     = symbol_conf.force;
949 
950 repeat:
951 	session = perf_session__new(&data, false, &report.tool);
952 	if (session == NULL)
953 		return -1;
954 
955 	if (report.queue_size) {
956 		ordered_events__set_alloc_size(&session->ordered_events,
957 					       report.queue_size);
958 	}
959 
960 	session->itrace_synth_opts = &itrace_synth_opts;
961 
962 	report.session = session;
963 
964 	has_br_stack = perf_header__has_feat(&session->header,
965 					     HEADER_BRANCH_STACK);
966 
967 	if (itrace_synth_opts.last_branch)
968 		has_br_stack = true;
969 
970 	if (has_br_stack && branch_call_mode)
971 		symbol_conf.show_branchflag_count = true;
972 
973 	memset(&report.brtype_stat, 0, sizeof(struct branch_type_stat));
974 
975 	/*
976 	 * Branch mode is a tristate:
977 	 * -1 means default, so decide based on the file having branch data.
978 	 * 0/1 means the user chose a mode.
979 	 */
980 	if (((branch_mode == -1 && has_br_stack) || branch_mode == 1) &&
981 	    !branch_call_mode) {
982 		sort__mode = SORT_MODE__BRANCH;
983 		symbol_conf.cumulate_callchain = false;
984 	}
985 	if (branch_call_mode) {
986 		callchain_param.key = CCKEY_ADDRESS;
987 		callchain_param.branch_callstack = 1;
988 		symbol_conf.use_callchain = true;
989 		callchain_register_param(&callchain_param);
990 		if (sort_order == NULL)
991 			sort_order = "srcline,symbol,dso";
992 	}
993 
994 	if (report.mem_mode) {
995 		if (sort__mode == SORT_MODE__BRANCH) {
996 			pr_err("branch and mem mode incompatible\n");
997 			goto error;
998 		}
999 		sort__mode = SORT_MODE__MEMORY;
1000 		symbol_conf.cumulate_callchain = false;
1001 	}
1002 
1003 	if (symbol_conf.report_hierarchy) {
1004 		/* disable incompatible options */
1005 		symbol_conf.cumulate_callchain = false;
1006 
1007 		if (field_order) {
1008 			pr_err("Error: --hierarchy and --fields options cannot be used together\n");
1009 			parse_options_usage(report_usage, options, "F", 1);
1010 			parse_options_usage(NULL, options, "hierarchy", 0);
1011 			goto error;
1012 		}
1013 
1014 		perf_hpp_list.need_collapse = true;
1015 	}
1016 
1017 	/* Force tty output for header output and per-thread stat. */
1018 	if (report.header || report.header_only || report.show_threads)
1019 		use_browser = 0;
1020 	if (report.header || report.header_only)
1021 		report.tool.show_feat_hdr = SHOW_FEAT_HEADER;
1022 	if (report.show_full_info)
1023 		report.tool.show_feat_hdr = SHOW_FEAT_HEADER_FULL_INFO;
1024 
1025 	if (strcmp(input_name, "-") != 0)
1026 		setup_browser(true);
1027 	else
1028 		use_browser = 0;
1029 
1030 	if (setup_sorting(session->evlist) < 0) {
1031 		if (sort_order)
1032 			parse_options_usage(report_usage, options, "s", 1);
1033 		if (field_order)
1034 			parse_options_usage(sort_order ? NULL : report_usage,
1035 					    options, "F", 1);
1036 		goto error;
1037 	}
1038 
1039 	if ((report.header || report.header_only) && !quiet) {
1040 		perf_session__fprintf_info(session, stdout,
1041 					   report.show_full_info);
1042 		if (report.header_only) {
1043 			ret = 0;
1044 			goto error;
1045 		}
1046 	} else if (use_browser == 0 && !quiet) {
1047 		fputs("# To display the perf.data header info, please use --header/--header-only options.\n#\n",
1048 		      stdout);
1049 	}
1050 
1051 	/*
1052 	 * Only in the TUI browser we are doing integrated annotation,
1053 	 * so don't allocate extra space that won't be used in the stdio
1054 	 * implementation.
1055 	 */
1056 	if (ui__has_annotation()) {
1057 		ret = symbol__annotation_init();
1058 		if (ret < 0)
1059 			goto error;
1060 		/*
1061  		 * For searching by name on the "Browse map details".
1062  		 * providing it only in verbose mode not to bloat too
1063  		 * much struct symbol.
1064  		 */
1065 		if (verbose > 0) {
1066 			/*
1067 			 * XXX: Need to provide a less kludgy way to ask for
1068 			 * more space per symbol, the u32 is for the index on
1069 			 * the ui browser.
1070 			 * See symbol__browser_index.
1071 			 */
1072 			symbol_conf.priv_size += sizeof(u32);
1073 			symbol_conf.sort_by_name = true;
1074 		}
1075 	}
1076 
1077 	if (symbol__init(&session->header.env) < 0)
1078 		goto error;
1079 
1080 	if (perf_time__parse_str(&report.ptime, report.time_str) != 0) {
1081 		pr_err("Invalid time string\n");
1082 		return -EINVAL;
1083 	}
1084 
1085 	sort__setup_elide(stdout);
1086 
1087 	ret = __cmd_report(&report);
1088 	if (ret == K_SWITCH_INPUT_DATA) {
1089 		perf_session__delete(session);
1090 		goto repeat;
1091 	} else
1092 		ret = 0;
1093 
1094 error:
1095 	perf_session__delete(session);
1096 	return ret;
1097 }
1098