xref: /openbmc/linux/tools/perf/builtin-report.c (revision 110e6f26)
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 <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 
40 #include "util/auxtrace.h"
41 
42 #include <dlfcn.h>
43 #include <linux/bitmap.h>
44 #include <linux/stringify.h>
45 
46 struct report {
47 	struct perf_tool	tool;
48 	struct perf_session	*session;
49 	bool			use_tui, use_gtk, use_stdio;
50 	bool			dont_use_callchains;
51 	bool			show_full_info;
52 	bool			show_threads;
53 	bool			inverted_callchain;
54 	bool			mem_mode;
55 	bool			header;
56 	bool			header_only;
57 	bool			nonany_branch_mode;
58 	int			max_stack;
59 	struct perf_read_values	show_threads_values;
60 	const char		*pretty_printing_style;
61 	const char		*cpu_list;
62 	const char		*symbol_filter_str;
63 	float			min_percent;
64 	u64			nr_entries;
65 	u64			queue_size;
66 	int			socket_filter;
67 	DECLARE_BITMAP(cpu_bitmap, MAX_NR_CPUS);
68 };
69 
70 static int report__config(const char *var, const char *value, void *cb)
71 {
72 	struct report *rep = cb;
73 
74 	if (!strcmp(var, "report.group")) {
75 		symbol_conf.event_group = perf_config_bool(var, value);
76 		return 0;
77 	}
78 	if (!strcmp(var, "report.percent-limit")) {
79 		double pcnt = strtof(value, NULL);
80 
81 		rep->min_percent = pcnt;
82 		callchain_param.min_percent = pcnt;
83 		return 0;
84 	}
85 	if (!strcmp(var, "report.children")) {
86 		symbol_conf.cumulate_callchain = perf_config_bool(var, value);
87 		return 0;
88 	}
89 	if (!strcmp(var, "report.queue-size")) {
90 		rep->queue_size = perf_config_u64(var, value);
91 		return 0;
92 	}
93 
94 	return 0;
95 }
96 
97 static int hist_iter__report_callback(struct hist_entry_iter *iter,
98 				      struct addr_location *al, bool single,
99 				      void *arg)
100 {
101 	int err = 0;
102 	struct report *rep = arg;
103 	struct hist_entry *he = iter->he;
104 	struct perf_evsel *evsel = iter->evsel;
105 	struct mem_info *mi;
106 	struct branch_info *bi;
107 
108 	if (!ui__has_annotation())
109 		return 0;
110 
111 	hist__account_cycles(iter->sample->branch_stack, al, iter->sample,
112 			     rep->nonany_branch_mode);
113 
114 	if (sort__mode == SORT_MODE__BRANCH) {
115 		bi = he->branch_info;
116 		err = addr_map_symbol__inc_samples(&bi->from, evsel->idx);
117 		if (err)
118 			goto out;
119 
120 		err = addr_map_symbol__inc_samples(&bi->to, evsel->idx);
121 
122 	} else if (rep->mem_mode) {
123 		mi = he->mem_info;
124 		err = addr_map_symbol__inc_samples(&mi->daddr, evsel->idx);
125 		if (err)
126 			goto out;
127 
128 		err = hist_entry__inc_addr_samples(he, evsel->idx, al->addr);
129 
130 	} else if (symbol_conf.cumulate_callchain) {
131 		if (single)
132 			err = hist_entry__inc_addr_samples(he, evsel->idx,
133 							   al->addr);
134 	} else {
135 		err = hist_entry__inc_addr_samples(he, evsel->idx, al->addr);
136 	}
137 
138 out:
139 	return err;
140 }
141 
142 static int process_sample_event(struct perf_tool *tool,
143 				union perf_event *event,
144 				struct perf_sample *sample,
145 				struct perf_evsel *evsel,
146 				struct machine *machine)
147 {
148 	struct report *rep = container_of(tool, struct report, tool);
149 	struct addr_location al;
150 	struct hist_entry_iter iter = {
151 		.evsel 			= evsel,
152 		.sample 		= sample,
153 		.hide_unresolved 	= symbol_conf.hide_unresolved,
154 		.add_entry_cb 		= hist_iter__report_callback,
155 	};
156 	int ret = 0;
157 
158 	if (machine__resolve(machine, &al, sample) < 0) {
159 		pr_debug("problem processing %d event, skipping it.\n",
160 			 event->header.type);
161 		return -1;
162 	}
163 
164 	if (symbol_conf.hide_unresolved && al.sym == NULL)
165 		goto out_put;
166 
167 	if (rep->cpu_list && !test_bit(sample->cpu, rep->cpu_bitmap))
168 		goto out_put;
169 
170 	if (sort__mode == SORT_MODE__BRANCH) {
171 		/*
172 		 * A non-synthesized event might not have a branch stack if
173 		 * branch stacks have been synthesized (using itrace options).
174 		 */
175 		if (!sample->branch_stack)
176 			goto out_put;
177 		iter.ops = &hist_iter_branch;
178 	} else if (rep->mem_mode) {
179 		iter.ops = &hist_iter_mem;
180 	} else if (symbol_conf.cumulate_callchain) {
181 		iter.ops = &hist_iter_cumulative;
182 	} else {
183 		iter.ops = &hist_iter_normal;
184 	}
185 
186 	if (al.map != NULL)
187 		al.map->dso->hit = 1;
188 
189 	ret = hist_entry_iter__add(&iter, &al, rep->max_stack, rep);
190 	if (ret < 0)
191 		pr_debug("problem adding hist entry, skipping event\n");
192 out_put:
193 	addr_location__put(&al);
194 	return ret;
195 }
196 
197 static int process_read_event(struct perf_tool *tool,
198 			      union perf_event *event,
199 			      struct perf_sample *sample __maybe_unused,
200 			      struct perf_evsel *evsel,
201 			      struct machine *machine __maybe_unused)
202 {
203 	struct report *rep = container_of(tool, struct report, tool);
204 
205 	if (rep->show_threads) {
206 		const char *name = evsel ? perf_evsel__name(evsel) : "unknown";
207 		perf_read_values_add_value(&rep->show_threads_values,
208 					   event->read.pid, event->read.tid,
209 					   event->read.id,
210 					   name,
211 					   event->read.value);
212 	}
213 
214 	dump_printf(": %d %d %s %" PRIu64 "\n", event->read.pid, event->read.tid,
215 		    evsel ? perf_evsel__name(evsel) : "FAIL",
216 		    event->read.value);
217 
218 	return 0;
219 }
220 
221 /* For pipe mode, sample_type is not currently set */
222 static int report__setup_sample_type(struct report *rep)
223 {
224 	struct perf_session *session = rep->session;
225 	u64 sample_type = perf_evlist__combined_sample_type(session->evlist);
226 	bool is_pipe = perf_data_file__is_pipe(session->file);
227 
228 	if (session->itrace_synth_opts->callchain ||
229 	    (!is_pipe &&
230 	     perf_header__has_feat(&session->header, HEADER_AUXTRACE) &&
231 	     !session->itrace_synth_opts->set))
232 		sample_type |= PERF_SAMPLE_CALLCHAIN;
233 
234 	if (session->itrace_synth_opts->last_branch)
235 		sample_type |= PERF_SAMPLE_BRANCH_STACK;
236 
237 	if (!is_pipe && !(sample_type & PERF_SAMPLE_CALLCHAIN)) {
238 		if (sort__has_parent) {
239 			ui__error("Selected --sort parent, but no "
240 				    "callchain data. Did you call "
241 				    "'perf record' without -g?\n");
242 			return -EINVAL;
243 		}
244 		if (symbol_conf.use_callchain) {
245 			ui__error("Selected -g or --branch-history but no "
246 				  "callchain data. Did\n"
247 				  "you call 'perf record' without -g?\n");
248 			return -1;
249 		}
250 	} else if (!rep->dont_use_callchains &&
251 		   callchain_param.mode != CHAIN_NONE &&
252 		   !symbol_conf.use_callchain) {
253 			symbol_conf.use_callchain = true;
254 			if (callchain_register_param(&callchain_param) < 0) {
255 				ui__error("Can't register callchain params.\n");
256 				return -EINVAL;
257 			}
258 	}
259 
260 	if (symbol_conf.cumulate_callchain) {
261 		/* Silently ignore if callchain is missing */
262 		if (!(sample_type & PERF_SAMPLE_CALLCHAIN)) {
263 			symbol_conf.cumulate_callchain = false;
264 			perf_hpp__cancel_cumulate();
265 		}
266 	}
267 
268 	if (sort__mode == SORT_MODE__BRANCH) {
269 		if (!is_pipe &&
270 		    !(sample_type & PERF_SAMPLE_BRANCH_STACK)) {
271 			ui__error("Selected -b but no branch data. "
272 				  "Did you call perf record without -b?\n");
273 			return -1;
274 		}
275 	}
276 
277 	if (symbol_conf.use_callchain || symbol_conf.cumulate_callchain) {
278 		if ((sample_type & PERF_SAMPLE_REGS_USER) &&
279 		    (sample_type & PERF_SAMPLE_STACK_USER))
280 			callchain_param.record_mode = CALLCHAIN_DWARF;
281 		else if (sample_type & PERF_SAMPLE_BRANCH_STACK)
282 			callchain_param.record_mode = CALLCHAIN_LBR;
283 		else
284 			callchain_param.record_mode = CALLCHAIN_FP;
285 	}
286 
287 	/* ??? handle more cases than just ANY? */
288 	if (!(perf_evlist__combined_branch_type(session->evlist) &
289 				PERF_SAMPLE_BRANCH_ANY))
290 		rep->nonany_branch_mode = true;
291 
292 	return 0;
293 }
294 
295 static void sig_handler(int sig __maybe_unused)
296 {
297 	session_done = 1;
298 }
299 
300 static size_t hists__fprintf_nr_sample_events(struct hists *hists, struct report *rep,
301 					      const char *evname, FILE *fp)
302 {
303 	size_t ret;
304 	char unit;
305 	unsigned long nr_samples = hists->stats.nr_events[PERF_RECORD_SAMPLE];
306 	u64 nr_events = hists->stats.total_period;
307 	struct perf_evsel *evsel = hists_to_evsel(hists);
308 	char buf[512];
309 	size_t size = sizeof(buf);
310 	int socked_id = hists->socket_filter;
311 
312 	if (symbol_conf.filter_relative) {
313 		nr_samples = hists->stats.nr_non_filtered_samples;
314 		nr_events = hists->stats.total_non_filtered_period;
315 	}
316 
317 	if (perf_evsel__is_group_event(evsel)) {
318 		struct perf_evsel *pos;
319 
320 		perf_evsel__group_desc(evsel, buf, size);
321 		evname = buf;
322 
323 		for_each_group_member(pos, evsel) {
324 			const struct hists *pos_hists = evsel__hists(pos);
325 
326 			if (symbol_conf.filter_relative) {
327 				nr_samples += pos_hists->stats.nr_non_filtered_samples;
328 				nr_events += pos_hists->stats.total_non_filtered_period;
329 			} else {
330 				nr_samples += pos_hists->stats.nr_events[PERF_RECORD_SAMPLE];
331 				nr_events += pos_hists->stats.total_period;
332 			}
333 		}
334 	}
335 
336 	nr_samples = convert_unit(nr_samples, &unit);
337 	ret = fprintf(fp, "# Samples: %lu%c", nr_samples, unit);
338 	if (evname != NULL)
339 		ret += fprintf(fp, " of event '%s'", evname);
340 
341 	if (symbol_conf.show_ref_callgraph &&
342 	    strstr(evname, "call-graph=no")) {
343 		ret += fprintf(fp, ", show reference callgraph");
344 	}
345 
346 	if (rep->mem_mode) {
347 		ret += fprintf(fp, "\n# Total weight : %" PRIu64, nr_events);
348 		ret += fprintf(fp, "\n# Sort order   : %s", sort_order ? : default_mem_sort_order);
349 	} else
350 		ret += fprintf(fp, "\n# Event count (approx.): %" PRIu64, nr_events);
351 
352 	if (socked_id > -1)
353 		ret += fprintf(fp, "\n# Processor Socket: %d", socked_id);
354 
355 	return ret + fprintf(fp, "\n#\n");
356 }
357 
358 static int perf_evlist__tty_browse_hists(struct perf_evlist *evlist,
359 					 struct report *rep,
360 					 const char *help)
361 {
362 	struct perf_evsel *pos;
363 
364 	fprintf(stdout, "#\n# Total Lost Samples: %" PRIu64 "\n#\n", evlist->stats.total_lost_samples);
365 	evlist__for_each(evlist, pos) {
366 		struct hists *hists = evsel__hists(pos);
367 		const char *evname = perf_evsel__name(pos);
368 
369 		if (symbol_conf.event_group &&
370 		    !perf_evsel__is_group_leader(pos))
371 			continue;
372 
373 		hists__fprintf_nr_sample_events(hists, rep, evname, stdout);
374 		hists__fprintf(hists, true, 0, 0, rep->min_percent, stdout);
375 		fprintf(stdout, "\n\n");
376 	}
377 
378 	if (sort_order == NULL &&
379 	    parent_pattern == default_parent_pattern)
380 		fprintf(stdout, "#\n# (%s)\n#\n", help);
381 
382 	if (rep->show_threads) {
383 		bool style = !strcmp(rep->pretty_printing_style, "raw");
384 		perf_read_values_display(stdout, &rep->show_threads_values,
385 					 style);
386 		perf_read_values_destroy(&rep->show_threads_values);
387 	}
388 
389 	return 0;
390 }
391 
392 static void report__warn_kptr_restrict(const struct report *rep)
393 {
394 	struct map *kernel_map = machine__kernel_map(&rep->session->machines.host);
395 	struct kmap *kernel_kmap = kernel_map ? map__kmap(kernel_map) : NULL;
396 
397 	if (kernel_map == NULL ||
398 	    (kernel_map->dso->hit &&
399 	     (kernel_kmap->ref_reloc_sym == NULL ||
400 	      kernel_kmap->ref_reloc_sym->addr == 0))) {
401 		const char *desc =
402 		    "As no suitable kallsyms nor vmlinux was found, kernel samples\n"
403 		    "can't be resolved.";
404 
405 		if (kernel_map) {
406 			const struct dso *kdso = kernel_map->dso;
407 			if (!RB_EMPTY_ROOT(&kdso->symbols[MAP__FUNCTION])) {
408 				desc = "If some relocation was applied (e.g. "
409 				       "kexec) symbols may be misresolved.";
410 			}
411 		}
412 
413 		ui__warning(
414 "Kernel address maps (/proc/{kallsyms,modules}) were restricted.\n\n"
415 "Check /proc/sys/kernel/kptr_restrict before running 'perf record'.\n\n%s\n\n"
416 "Samples in kernel modules can't be resolved as well.\n\n",
417 		desc);
418 	}
419 }
420 
421 static int report__gtk_browse_hists(struct report *rep, const char *help)
422 {
423 	int (*hist_browser)(struct perf_evlist *evlist, const char *help,
424 			    struct hist_browser_timer *timer, float min_pcnt);
425 
426 	hist_browser = dlsym(perf_gtk_handle, "perf_evlist__gtk_browse_hists");
427 
428 	if (hist_browser == NULL) {
429 		ui__error("GTK browser not found!\n");
430 		return -1;
431 	}
432 
433 	return hist_browser(rep->session->evlist, help, NULL, rep->min_percent);
434 }
435 
436 static int report__browse_hists(struct report *rep)
437 {
438 	int ret;
439 	struct perf_session *session = rep->session;
440 	struct perf_evlist *evlist = session->evlist;
441 	const char *help = perf_tip(system_path(TIPDIR));
442 
443 	if (help == NULL) {
444 		/* fallback for people who don't install perf ;-) */
445 		help = perf_tip(DOCDIR);
446 		if (help == NULL)
447 			help = "Cannot load tips.txt file, please install perf!";
448 	}
449 
450 	switch (use_browser) {
451 	case 1:
452 		ret = perf_evlist__tui_browse_hists(evlist, help, NULL,
453 						    rep->min_percent,
454 						    &session->header.env);
455 		/*
456 		 * Usually "ret" is the last pressed key, and we only
457 		 * care if the key notifies us to switch data file.
458 		 */
459 		if (ret != K_SWITCH_INPUT_DATA)
460 			ret = 0;
461 		break;
462 	case 2:
463 		ret = report__gtk_browse_hists(rep, help);
464 		break;
465 	default:
466 		ret = perf_evlist__tty_browse_hists(evlist, rep, help);
467 		break;
468 	}
469 
470 	return ret;
471 }
472 
473 static int report__collapse_hists(struct report *rep)
474 {
475 	struct ui_progress prog;
476 	struct perf_evsel *pos;
477 	int ret = 0;
478 
479 	ui_progress__init(&prog, rep->nr_entries, "Merging related events...");
480 
481 	evlist__for_each(rep->session->evlist, pos) {
482 		struct hists *hists = evsel__hists(pos);
483 
484 		if (pos->idx == 0)
485 			hists->symbol_filter_str = rep->symbol_filter_str;
486 
487 		hists->socket_filter = rep->socket_filter;
488 
489 		ret = hists__collapse_resort(hists, &prog);
490 		if (ret < 0)
491 			break;
492 
493 		/* Non-group events are considered as leader */
494 		if (symbol_conf.event_group &&
495 		    !perf_evsel__is_group_leader(pos)) {
496 			struct hists *leader_hists = evsel__hists(pos->leader);
497 
498 			hists__match(leader_hists, hists);
499 			hists__link(leader_hists, hists);
500 		}
501 	}
502 
503 	ui_progress__finish();
504 	return ret;
505 }
506 
507 static void report__output_resort(struct report *rep)
508 {
509 	struct ui_progress prog;
510 	struct perf_evsel *pos;
511 
512 	ui_progress__init(&prog, rep->nr_entries, "Sorting events for output...");
513 
514 	evlist__for_each(rep->session->evlist, pos)
515 		perf_evsel__output_resort(pos, &prog);
516 
517 	ui_progress__finish();
518 }
519 
520 static int __cmd_report(struct report *rep)
521 {
522 	int ret;
523 	struct perf_session *session = rep->session;
524 	struct perf_evsel *pos;
525 	struct perf_data_file *file = session->file;
526 
527 	signal(SIGINT, sig_handler);
528 
529 	if (rep->cpu_list) {
530 		ret = perf_session__cpu_bitmap(session, rep->cpu_list,
531 					       rep->cpu_bitmap);
532 		if (ret) {
533 			ui__error("failed to set cpu bitmap\n");
534 			return ret;
535 		}
536 	}
537 
538 	if (rep->show_threads)
539 		perf_read_values_init(&rep->show_threads_values);
540 
541 	ret = report__setup_sample_type(rep);
542 	if (ret) {
543 		/* report__setup_sample_type() already showed error message */
544 		return ret;
545 	}
546 
547 	ret = perf_session__process_events(session);
548 	if (ret) {
549 		ui__error("failed to process sample\n");
550 		return ret;
551 	}
552 
553 	report__warn_kptr_restrict(rep);
554 
555 	evlist__for_each(session->evlist, pos)
556 		rep->nr_entries += evsel__hists(pos)->nr_entries;
557 
558 	if (use_browser == 0) {
559 		if (verbose > 3)
560 			perf_session__fprintf(session, stdout);
561 
562 		if (verbose > 2)
563 			perf_session__fprintf_dsos(session, stdout);
564 
565 		if (dump_trace) {
566 			perf_session__fprintf_nr_events(session, stdout);
567 			perf_evlist__fprintf_nr_events(session->evlist, stdout);
568 			return 0;
569 		}
570 	}
571 
572 	ret = report__collapse_hists(rep);
573 	if (ret) {
574 		ui__error("failed to process hist entry\n");
575 		return ret;
576 	}
577 
578 	if (session_done())
579 		return 0;
580 
581 	/*
582 	 * recalculate number of entries after collapsing since it
583 	 * might be changed during the collapse phase.
584 	 */
585 	rep->nr_entries = 0;
586 	evlist__for_each(session->evlist, pos)
587 		rep->nr_entries += evsel__hists(pos)->nr_entries;
588 
589 	if (rep->nr_entries == 0) {
590 		ui__error("The %s file has no samples!\n", file->path);
591 		return 0;
592 	}
593 
594 	report__output_resort(rep);
595 
596 	return report__browse_hists(rep);
597 }
598 
599 static int
600 report_parse_callchain_opt(const struct option *opt, const char *arg, int unset)
601 {
602 	struct report *rep = (struct report *)opt->value;
603 
604 	/*
605 	 * --no-call-graph
606 	 */
607 	if (unset) {
608 		rep->dont_use_callchains = true;
609 		return 0;
610 	}
611 
612 	return parse_callchain_report_opt(arg);
613 }
614 
615 int
616 report_parse_ignore_callees_opt(const struct option *opt __maybe_unused,
617 				const char *arg, int unset __maybe_unused)
618 {
619 	if (arg) {
620 		int err = regcomp(&ignore_callees_regex, arg, REG_EXTENDED);
621 		if (err) {
622 			char buf[BUFSIZ];
623 			regerror(err, &ignore_callees_regex, buf, sizeof(buf));
624 			pr_err("Invalid --ignore-callees regex: %s\n%s", arg, buf);
625 			return -1;
626 		}
627 		have_ignore_callees = 1;
628 	}
629 
630 	return 0;
631 }
632 
633 static int
634 parse_branch_mode(const struct option *opt __maybe_unused,
635 		  const char *str __maybe_unused, int unset)
636 {
637 	int *branch_mode = opt->value;
638 
639 	*branch_mode = !unset;
640 	return 0;
641 }
642 
643 static int
644 parse_percent_limit(const struct option *opt, const char *str,
645 		    int unset __maybe_unused)
646 {
647 	struct report *rep = opt->value;
648 	double pcnt = strtof(str, NULL);
649 
650 	rep->min_percent = pcnt;
651 	callchain_param.min_percent = pcnt;
652 	return 0;
653 }
654 
655 #define CALLCHAIN_DEFAULT_OPT  "graph,0.5,caller,function,percent"
656 
657 const char report_callchain_help[] = "Display call graph (stack chain/backtrace):\n\n"
658 				     CALLCHAIN_REPORT_HELP
659 				     "\n\t\t\t\tDefault: " CALLCHAIN_DEFAULT_OPT;
660 
661 int cmd_report(int argc, const char **argv, const char *prefix __maybe_unused)
662 {
663 	struct perf_session *session;
664 	struct itrace_synth_opts itrace_synth_opts = { .set = 0, };
665 	struct stat st;
666 	bool has_br_stack = false;
667 	int branch_mode = -1;
668 	bool branch_call_mode = false;
669 	char callchain_default_opt[] = CALLCHAIN_DEFAULT_OPT;
670 	const char * const report_usage[] = {
671 		"perf report [<options>]",
672 		NULL
673 	};
674 	struct report report = {
675 		.tool = {
676 			.sample		 = process_sample_event,
677 			.mmap		 = perf_event__process_mmap,
678 			.mmap2		 = perf_event__process_mmap2,
679 			.comm		 = perf_event__process_comm,
680 			.exit		 = perf_event__process_exit,
681 			.fork		 = perf_event__process_fork,
682 			.lost		 = perf_event__process_lost,
683 			.read		 = process_read_event,
684 			.attr		 = perf_event__process_attr,
685 			.tracing_data	 = perf_event__process_tracing_data,
686 			.build_id	 = perf_event__process_build_id,
687 			.id_index	 = perf_event__process_id_index,
688 			.auxtrace_info	 = perf_event__process_auxtrace_info,
689 			.auxtrace	 = perf_event__process_auxtrace,
690 			.ordered_events	 = true,
691 			.ordering_requires_timestamps = true,
692 		},
693 		.max_stack		 = PERF_MAX_STACK_DEPTH,
694 		.pretty_printing_style	 = "normal",
695 		.socket_filter		 = -1,
696 	};
697 	const struct option options[] = {
698 	OPT_STRING('i', "input", &input_name, "file",
699 		    "input file name"),
700 	OPT_INCR('v', "verbose", &verbose,
701 		    "be more verbose (show symbol address, etc)"),
702 	OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace,
703 		    "dump raw trace in ASCII"),
704 	OPT_STRING('k', "vmlinux", &symbol_conf.vmlinux_name,
705 		   "file", "vmlinux pathname"),
706 	OPT_STRING(0, "kallsyms", &symbol_conf.kallsyms_name,
707 		   "file", "kallsyms pathname"),
708 	OPT_BOOLEAN('f', "force", &symbol_conf.force, "don't complain, do it"),
709 	OPT_BOOLEAN('m', "modules", &symbol_conf.use_modules,
710 		    "load module symbols - WARNING: use only with -k and LIVE kernel"),
711 	OPT_BOOLEAN('n', "show-nr-samples", &symbol_conf.show_nr_samples,
712 		    "Show a column with the number of samples"),
713 	OPT_BOOLEAN('T', "threads", &report.show_threads,
714 		    "Show per-thread event counters"),
715 	OPT_STRING(0, "pretty", &report.pretty_printing_style, "key",
716 		   "pretty printing style key: normal raw"),
717 	OPT_BOOLEAN(0, "tui", &report.use_tui, "Use the TUI interface"),
718 	OPT_BOOLEAN(0, "gtk", &report.use_gtk, "Use the GTK2 interface"),
719 	OPT_BOOLEAN(0, "stdio", &report.use_stdio,
720 		    "Use the stdio interface"),
721 	OPT_BOOLEAN(0, "header", &report.header, "Show data header."),
722 	OPT_BOOLEAN(0, "header-only", &report.header_only,
723 		    "Show only data header."),
724 	OPT_STRING('s', "sort", &sort_order, "key[,key2...]",
725 		   "sort by key(s): pid, comm, dso, symbol, parent, cpu, srcline, ..."
726 		   " Please refer the man page for the complete list."),
727 	OPT_STRING('F', "fields", &field_order, "key[,keys...]",
728 		   "output field(s): overhead, period, sample plus all of sort keys"),
729 	OPT_BOOLEAN(0, "show-cpu-utilization", &symbol_conf.show_cpu_utilization,
730 		    "Show sample percentage for different cpu modes"),
731 	OPT_BOOLEAN_FLAG(0, "showcpuutilization", &symbol_conf.show_cpu_utilization,
732 		    "Show sample percentage for different cpu modes", PARSE_OPT_HIDDEN),
733 	OPT_STRING('p', "parent", &parent_pattern, "regex",
734 		   "regex filter to identify parent, see: '--sort parent'"),
735 	OPT_BOOLEAN('x', "exclude-other", &symbol_conf.exclude_other,
736 		    "Only display entries with parent-match"),
737 	OPT_CALLBACK_DEFAULT('g', "call-graph", &report,
738 			     "print_type,threshold[,print_limit],order,sort_key[,branch],value",
739 			     report_callchain_help, &report_parse_callchain_opt,
740 			     callchain_default_opt),
741 	OPT_BOOLEAN(0, "children", &symbol_conf.cumulate_callchain,
742 		    "Accumulate callchains of children and show total overhead as well"),
743 	OPT_INTEGER(0, "max-stack", &report.max_stack,
744 		    "Set the maximum stack depth when parsing the callchain, "
745 		    "anything beyond the specified depth will be ignored. "
746 		    "Default: " __stringify(PERF_MAX_STACK_DEPTH)),
747 	OPT_BOOLEAN('G', "inverted", &report.inverted_callchain,
748 		    "alias for inverted call graph"),
749 	OPT_CALLBACK(0, "ignore-callees", NULL, "regex",
750 		   "ignore callees of these functions in call graphs",
751 		   report_parse_ignore_callees_opt),
752 	OPT_STRING('d', "dsos", &symbol_conf.dso_list_str, "dso[,dso...]",
753 		   "only consider symbols in these dsos"),
754 	OPT_STRING('c', "comms", &symbol_conf.comm_list_str, "comm[,comm...]",
755 		   "only consider symbols in these comms"),
756 	OPT_STRING(0, "pid", &symbol_conf.pid_list_str, "pid[,pid...]",
757 		   "only consider symbols in these pids"),
758 	OPT_STRING(0, "tid", &symbol_conf.tid_list_str, "tid[,tid...]",
759 		   "only consider symbols in these tids"),
760 	OPT_STRING('S', "symbols", &symbol_conf.sym_list_str, "symbol[,symbol...]",
761 		   "only consider these symbols"),
762 	OPT_STRING(0, "symbol-filter", &report.symbol_filter_str, "filter",
763 		   "only show symbols that (partially) match with this filter"),
764 	OPT_STRING('w', "column-widths", &symbol_conf.col_width_list_str,
765 		   "width[,width...]",
766 		   "don't try to adjust column width, use these fixed values"),
767 	OPT_STRING_NOEMPTY('t', "field-separator", &symbol_conf.field_sep, "separator",
768 		   "separator for columns, no spaces will be added between "
769 		   "columns '.' is reserved."),
770 	OPT_BOOLEAN('U', "hide-unresolved", &symbol_conf.hide_unresolved,
771 		    "Only display entries resolved to a symbol"),
772 	OPT_STRING(0, "symfs", &symbol_conf.symfs, "directory",
773 		    "Look for files with symbols relative to this directory"),
774 	OPT_STRING('C', "cpu", &report.cpu_list, "cpu",
775 		   "list of cpus to profile"),
776 	OPT_BOOLEAN('I', "show-info", &report.show_full_info,
777 		    "Display extended information about perf.data file"),
778 	OPT_BOOLEAN(0, "source", &symbol_conf.annotate_src,
779 		    "Interleave source code with assembly code (default)"),
780 	OPT_BOOLEAN(0, "asm-raw", &symbol_conf.annotate_asm_raw,
781 		    "Display raw encoding of assembly instructions (default)"),
782 	OPT_STRING('M', "disassembler-style", &disassembler_style, "disassembler style",
783 		   "Specify disassembler style (e.g. -M intel for intel syntax)"),
784 	OPT_BOOLEAN(0, "show-total-period", &symbol_conf.show_total_period,
785 		    "Show a column with the sum of periods"),
786 	OPT_BOOLEAN(0, "group", &symbol_conf.event_group,
787 		    "Show event group information together"),
788 	OPT_CALLBACK_NOOPT('b', "branch-stack", &branch_mode, "",
789 		    "use branch records for per branch histogram filling",
790 		    parse_branch_mode),
791 	OPT_BOOLEAN(0, "branch-history", &branch_call_mode,
792 		    "add last branch records to call history"),
793 	OPT_STRING(0, "objdump", &objdump_path, "path",
794 		   "objdump binary to use for disassembly and annotations"),
795 	OPT_BOOLEAN(0, "demangle", &symbol_conf.demangle,
796 		    "Disable symbol demangling"),
797 	OPT_BOOLEAN(0, "demangle-kernel", &symbol_conf.demangle_kernel,
798 		    "Enable kernel symbol demangling"),
799 	OPT_BOOLEAN(0, "mem-mode", &report.mem_mode, "mem access profile"),
800 	OPT_CALLBACK(0, "percent-limit", &report, "percent",
801 		     "Don't show entries under that percent", parse_percent_limit),
802 	OPT_CALLBACK(0, "percentage", NULL, "relative|absolute",
803 		     "how to display percentage of filtered entries", parse_filter_percentage),
804 	OPT_CALLBACK_OPTARG(0, "itrace", &itrace_synth_opts, NULL, "opts",
805 			    "Instruction Tracing options",
806 			    itrace_parse_synth_opts),
807 	OPT_BOOLEAN(0, "full-source-path", &srcline_full_filename,
808 			"Show full source file name path for source lines"),
809 	OPT_BOOLEAN(0, "show-ref-call-graph", &symbol_conf.show_ref_callgraph,
810 		    "Show callgraph from reference event"),
811 	OPT_INTEGER(0, "socket-filter", &report.socket_filter,
812 		    "only show processor socket that match with this filter"),
813 	OPT_BOOLEAN(0, "raw-trace", &symbol_conf.raw_trace,
814 		    "Show raw trace event output (do not use print fmt or plugins)"),
815 	OPT_BOOLEAN(0, "hierarchy", &symbol_conf.report_hierarchy,
816 		    "Show entries in a hierarchy"),
817 	OPT_END()
818 	};
819 	struct perf_data_file file = {
820 		.mode  = PERF_DATA_MODE_READ,
821 	};
822 	int ret = hists__init();
823 
824 	if (ret < 0)
825 		return ret;
826 
827 	perf_config(report__config, &report);
828 
829 	argc = parse_options(argc, argv, options, report_usage, 0);
830 	if (argc) {
831 		/*
832 		 * Special case: if there's an argument left then assume that
833 		 * it's a symbol filter:
834 		 */
835 		if (argc > 1)
836 			usage_with_options(report_usage, options);
837 
838 		report.symbol_filter_str = argv[0];
839 	}
840 
841 	if (symbol_conf.vmlinux_name &&
842 	    access(symbol_conf.vmlinux_name, R_OK)) {
843 		pr_err("Invalid file: %s\n", symbol_conf.vmlinux_name);
844 		return -EINVAL;
845 	}
846 	if (symbol_conf.kallsyms_name &&
847 	    access(symbol_conf.kallsyms_name, R_OK)) {
848 		pr_err("Invalid file: %s\n", symbol_conf.kallsyms_name);
849 		return -EINVAL;
850 	}
851 
852 	if (report.use_stdio)
853 		use_browser = 0;
854 	else if (report.use_tui)
855 		use_browser = 1;
856 	else if (report.use_gtk)
857 		use_browser = 2;
858 
859 	if (report.inverted_callchain)
860 		callchain_param.order = ORDER_CALLER;
861 	if (symbol_conf.cumulate_callchain && !callchain_param.order_set)
862 		callchain_param.order = ORDER_CALLER;
863 
864 	if (itrace_synth_opts.callchain &&
865 	    (int)itrace_synth_opts.callchain_sz > report.max_stack)
866 		report.max_stack = itrace_synth_opts.callchain_sz;
867 
868 	if (!input_name || !strlen(input_name)) {
869 		if (!fstat(STDIN_FILENO, &st) && S_ISFIFO(st.st_mode))
870 			input_name = "-";
871 		else
872 			input_name = "perf.data";
873 	}
874 
875 	file.path  = input_name;
876 	file.force = symbol_conf.force;
877 
878 repeat:
879 	session = perf_session__new(&file, false, &report.tool);
880 	if (session == NULL)
881 		return -1;
882 
883 	if (report.queue_size) {
884 		ordered_events__set_alloc_size(&session->ordered_events,
885 					       report.queue_size);
886 	}
887 
888 	session->itrace_synth_opts = &itrace_synth_opts;
889 
890 	report.session = session;
891 
892 	has_br_stack = perf_header__has_feat(&session->header,
893 					     HEADER_BRANCH_STACK);
894 
895 	if (itrace_synth_opts.last_branch)
896 		has_br_stack = true;
897 
898 	/*
899 	 * Branch mode is a tristate:
900 	 * -1 means default, so decide based on the file having branch data.
901 	 * 0/1 means the user chose a mode.
902 	 */
903 	if (((branch_mode == -1 && has_br_stack) || branch_mode == 1) &&
904 	    !branch_call_mode) {
905 		sort__mode = SORT_MODE__BRANCH;
906 		symbol_conf.cumulate_callchain = false;
907 	}
908 	if (branch_call_mode) {
909 		callchain_param.key = CCKEY_ADDRESS;
910 		callchain_param.branch_callstack = 1;
911 		symbol_conf.use_callchain = true;
912 		callchain_register_param(&callchain_param);
913 		if (sort_order == NULL)
914 			sort_order = "srcline,symbol,dso";
915 	}
916 
917 	if (report.mem_mode) {
918 		if (sort__mode == SORT_MODE__BRANCH) {
919 			pr_err("branch and mem mode incompatible\n");
920 			goto error;
921 		}
922 		sort__mode = SORT_MODE__MEMORY;
923 		symbol_conf.cumulate_callchain = false;
924 	}
925 
926 	if (symbol_conf.report_hierarchy) {
927 		/* disable incompatible options */
928 		symbol_conf.event_group = false;
929 		symbol_conf.cumulate_callchain = false;
930 
931 		if (field_order) {
932 			pr_err("Error: --hierarchy and --fields options cannot be used together\n");
933 			parse_options_usage(report_usage, options, "F", 1);
934 			parse_options_usage(NULL, options, "hierarchy", 0);
935 			goto error;
936 		}
937 
938 		sort__need_collapse = true;
939 	}
940 
941 	/* Force tty output for header output and per-thread stat. */
942 	if (report.header || report.header_only || report.show_threads)
943 		use_browser = 0;
944 
945 	if (strcmp(input_name, "-") != 0)
946 		setup_browser(true);
947 	else
948 		use_browser = 0;
949 
950 	if (setup_sorting(session->evlist) < 0) {
951 		if (sort_order)
952 			parse_options_usage(report_usage, options, "s", 1);
953 		if (field_order)
954 			parse_options_usage(sort_order ? NULL : report_usage,
955 					    options, "F", 1);
956 		goto error;
957 	}
958 
959 	if (report.header || report.header_only) {
960 		perf_session__fprintf_info(session, stdout,
961 					   report.show_full_info);
962 		if (report.header_only) {
963 			ret = 0;
964 			goto error;
965 		}
966 	} else if (use_browser == 0) {
967 		fputs("# To display the perf.data header info, please use --header/--header-only options.\n#\n",
968 		      stdout);
969 	}
970 
971 	/*
972 	 * Only in the TUI browser we are doing integrated annotation,
973 	 * so don't allocate extra space that won't be used in the stdio
974 	 * implementation.
975 	 */
976 	if (ui__has_annotation()) {
977 		symbol_conf.priv_size = sizeof(struct annotation);
978 		machines__set_symbol_filter(&session->machines,
979 					    symbol__annotate_init);
980 		/*
981  		 * For searching by name on the "Browse map details".
982  		 * providing it only in verbose mode not to bloat too
983  		 * much struct symbol.
984  		 */
985 		if (verbose) {
986 			/*
987 			 * XXX: Need to provide a less kludgy way to ask for
988 			 * more space per symbol, the u32 is for the index on
989 			 * the ui browser.
990 			 * See symbol__browser_index.
991 			 */
992 			symbol_conf.priv_size += sizeof(u32);
993 			symbol_conf.sort_by_name = true;
994 		}
995 	}
996 
997 	if (symbol__init(&session->header.env) < 0)
998 		goto error;
999 
1000 	sort__setup_elide(stdout);
1001 
1002 	ret = __cmd_report(&report);
1003 	if (ret == K_SWITCH_INPUT_DATA) {
1004 		perf_session__delete(session);
1005 		goto repeat;
1006 	} else
1007 		ret = 0;
1008 
1009 error:
1010 	perf_session__delete(session);
1011 	return ret;
1012 }
1013