xref: /linux/tools/perf/builtin-report.c (revision 922ec313f061cf75157d8e93fe16bc6397f6ea25)
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/config.h"
12 
13 #include "util/annotate.h"
14 #include "util/color.h"
15 #include "util/dso.h"
16 #include <linux/list.h>
17 #include <linux/rbtree.h>
18 #include <linux/err.h>
19 #include <linux/zalloc.h>
20 #include "util/map.h"
21 #include "util/symbol.h"
22 #include "util/map_symbol.h"
23 #include "util/mem-events.h"
24 #include "util/branch.h"
25 #include "util/callchain.h"
26 #include "util/values.h"
27 
28 #include "perf.h"
29 #include "util/debug.h"
30 #include "util/evlist.h"
31 #include "util/evsel.h"
32 #include "util/evswitch.h"
33 #include "util/header.h"
34 #include "util/mem-info.h"
35 #include "util/session.h"
36 #include "util/srcline.h"
37 #include "util/tool.h"
38 
39 #include <subcmd/parse-options.h>
40 #include <subcmd/exec-cmd.h>
41 #include "util/parse-events.h"
42 
43 #include "util/thread.h"
44 #include "util/sort.h"
45 #include "util/hist.h"
46 #include "util/data.h"
47 #include "arch/common.h"
48 #include "util/time-utils.h"
49 #include "util/auxtrace.h"
50 #include "util/units.h"
51 #include "util/util.h" // perf_tip()
52 #include "ui/ui.h"
53 #include "ui/progress.h"
54 #include "util/block-info.h"
55 
56 #include <dlfcn.h>
57 #include <errno.h>
58 #include <inttypes.h>
59 #include <regex.h>
60 #include <linux/ctype.h>
61 #include <signal.h>
62 #include <linux/bitmap.h>
63 #include <linux/list_sort.h>
64 #include <linux/string.h>
65 #include <linux/stringify.h>
66 #include <linux/time64.h>
67 #include <sys/types.h>
68 #include <sys/stat.h>
69 #include <unistd.h>
70 #include <linux/mman.h>
71 
72 #ifdef HAVE_LIBTRACEEVENT
73 #include <traceevent/event-parse.h>
74 #endif
75 
76 struct report {
77 	struct perf_tool	tool;
78 	struct perf_session	*session;
79 	struct evswitch		evswitch;
80 #ifdef HAVE_SLANG_SUPPORT
81 	bool			use_tui;
82 #endif
83 #ifdef HAVE_GTK2_SUPPORT
84 	bool			use_gtk;
85 #endif
86 	bool			use_stdio;
87 	bool			show_full_info;
88 	bool			show_threads;
89 	bool			inverted_callchain;
90 	bool			mem_mode;
91 	bool			stats_mode;
92 	bool			tasks_mode;
93 	bool			mmaps_mode;
94 	bool			header;
95 	bool			header_only;
96 	bool			nonany_branch_mode;
97 	bool			group_set;
98 	bool			stitch_lbr;
99 	bool			disable_order;
100 	bool			skip_empty;
101 	bool			data_type;
102 	int			max_stack;
103 	struct perf_read_values	show_threads_values;
104 	const char		*pretty_printing_style;
105 	const char		*cpu_list;
106 	const char		*symbol_filter_str;
107 	const char		*time_str;
108 	struct perf_time_interval *ptime_range;
109 	int			range_size;
110 	int			range_num;
111 	float			min_percent;
112 	u64			nr_entries;
113 	u64			queue_size;
114 	u64			total_cycles;
115 	int			socket_filter;
116 	DECLARE_BITMAP(cpu_bitmap, MAX_NR_CPUS);
117 	struct branch_type_stat	brtype_stat;
118 	bool			symbol_ipc;
119 	bool			total_cycles_mode;
120 	struct block_report	*block_reports;
121 	int			nr_block_reports;
122 };
123 
124 static int report__config(const char *var, const char *value, void *cb)
125 {
126 	struct report *rep = cb;
127 
128 	if (!strcmp(var, "report.group")) {
129 		symbol_conf.event_group = perf_config_bool(var, value);
130 		return 0;
131 	}
132 	if (!strcmp(var, "report.percent-limit")) {
133 		double pcnt = strtof(value, NULL);
134 
135 		rep->min_percent = pcnt;
136 		callchain_param.min_percent = pcnt;
137 		return 0;
138 	}
139 	if (!strcmp(var, "report.children")) {
140 		symbol_conf.cumulate_callchain = perf_config_bool(var, value);
141 		return 0;
142 	}
143 	if (!strcmp(var, "report.queue-size"))
144 		return perf_config_u64(&rep->queue_size, var, value);
145 
146 	if (!strcmp(var, "report.sort_order")) {
147 		default_sort_order = strdup(value);
148 		if (!default_sort_order) {
149 			pr_err("Not enough memory for report.sort_order\n");
150 			return -1;
151 		}
152 		return 0;
153 	}
154 
155 	if (!strcmp(var, "report.skip-empty")) {
156 		rep->skip_empty = perf_config_bool(var, value);
157 		return 0;
158 	}
159 
160 	pr_debug("%s variable unknown, ignoring...", var);
161 	return 0;
162 }
163 
164 static int hist_iter__report_callback(struct hist_entry_iter *iter,
165 				      struct addr_location *al, bool single,
166 				      void *arg)
167 {
168 	int err = 0;
169 	struct report *rep = arg;
170 	struct hist_entry *he = iter->he;
171 	struct evsel *evsel = iter->evsel;
172 	struct perf_sample *sample = iter->sample;
173 	struct mem_info *mi;
174 	struct branch_info *bi;
175 
176 	if (!ui__has_annotation() && !rep->symbol_ipc)
177 		return 0;
178 
179 	if (sort__mode == SORT_MODE__BRANCH) {
180 		bi = he->branch_info;
181 		err = addr_map_symbol__inc_samples(&bi->from, sample, evsel);
182 		if (err)
183 			goto out;
184 
185 		err = addr_map_symbol__inc_samples(&bi->to, sample, evsel);
186 
187 	} else if (rep->mem_mode) {
188 		mi = he->mem_info;
189 		err = addr_map_symbol__inc_samples(mem_info__daddr(mi), sample, evsel);
190 		if (err)
191 			goto out;
192 
193 		err = hist_entry__inc_addr_samples(he, sample, evsel, al->addr);
194 
195 	} else if (symbol_conf.cumulate_callchain) {
196 		if (single)
197 			err = hist_entry__inc_addr_samples(he, sample, evsel, al->addr);
198 	} else {
199 		err = hist_entry__inc_addr_samples(he, sample, evsel, al->addr);
200 	}
201 
202 out:
203 	return err;
204 }
205 
206 static int hist_iter__branch_callback(struct hist_entry_iter *iter,
207 				      struct addr_location *al __maybe_unused,
208 				      bool single __maybe_unused,
209 				      void *arg)
210 {
211 	struct hist_entry *he = iter->he;
212 	struct report *rep = arg;
213 	struct branch_info *bi = he->branch_info;
214 	struct perf_sample *sample = iter->sample;
215 	struct evsel *evsel = iter->evsel;
216 	int err;
217 
218 	branch_type_count(&rep->brtype_stat, &bi->flags,
219 			  bi->from.addr, bi->to.addr);
220 
221 	if (!ui__has_annotation() && !rep->symbol_ipc)
222 		return 0;
223 
224 	err = addr_map_symbol__inc_samples(&bi->from, sample, evsel);
225 	if (err)
226 		goto out;
227 
228 	err = addr_map_symbol__inc_samples(&bi->to, sample, evsel);
229 
230 out:
231 	return err;
232 }
233 
234 static void setup_forced_leader(struct report *report,
235 				struct evlist *evlist)
236 {
237 	if (report->group_set)
238 		evlist__force_leader(evlist);
239 }
240 
241 static int process_feature_event(struct perf_session *session,
242 				 union perf_event *event)
243 {
244 	struct report *rep = container_of(session->tool, struct report, tool);
245 
246 	if (event->feat.feat_id < HEADER_LAST_FEATURE)
247 		return perf_event__process_feature(session, event);
248 
249 	if (event->feat.feat_id != HEADER_LAST_FEATURE) {
250 		pr_err("failed: wrong feature ID: %" PRI_lu64 "\n",
251 		       event->feat.feat_id);
252 		return -1;
253 	} else if (rep->header_only) {
254 		session_done = 1;
255 	}
256 
257 	/*
258 	 * (feat_id = HEADER_LAST_FEATURE) is the end marker which
259 	 * means all features are received, now we can force the
260 	 * group if needed.
261 	 */
262 	setup_forced_leader(rep, session->evlist);
263 	return 0;
264 }
265 
266 static int process_sample_event(const struct perf_tool *tool,
267 				union perf_event *event,
268 				struct perf_sample *sample,
269 				struct evsel *evsel,
270 				struct machine *machine)
271 {
272 	struct report *rep = container_of(tool, struct report, tool);
273 	struct addr_location al;
274 	struct hist_entry_iter iter = {
275 		.evsel 			= evsel,
276 		.sample 		= sample,
277 		.hide_unresolved 	= symbol_conf.hide_unresolved,
278 		.add_entry_cb 		= hist_iter__report_callback,
279 	};
280 	int ret = 0;
281 
282 	if (perf_time__ranges_skip_sample(rep->ptime_range, rep->range_num,
283 					  sample->time)) {
284 		return 0;
285 	}
286 
287 	if (evswitch__discard(&rep->evswitch, evsel))
288 		return 0;
289 
290 	addr_location__init(&al);
291 	if (machine__resolve(machine, &al, sample) < 0) {
292 		pr_debug("problem processing %d event, skipping it.\n",
293 			 event->header.type);
294 		ret = -1;
295 		goto out_put;
296 	}
297 
298 	if (rep->stitch_lbr)
299 		thread__set_lbr_stitch_enable(al.thread, true);
300 
301 	if (symbol_conf.hide_unresolved && al.sym == NULL)
302 		goto out_put;
303 
304 	if (rep->cpu_list && !test_bit(sample->cpu, rep->cpu_bitmap))
305 		goto out_put;
306 
307 	if (sort__mode == SORT_MODE__BRANCH) {
308 		/*
309 		 * A non-synthesized event might not have a branch stack if
310 		 * branch stacks have been synthesized (using itrace options).
311 		 */
312 		if (!sample->branch_stack)
313 			goto out_put;
314 
315 		iter.add_entry_cb = hist_iter__branch_callback;
316 		iter.ops = &hist_iter_branch;
317 	} else if (rep->mem_mode) {
318 		iter.ops = &hist_iter_mem;
319 	} else if (symbol_conf.cumulate_callchain) {
320 		iter.ops = &hist_iter_cumulative;
321 	} else {
322 		iter.ops = &hist_iter_normal;
323 	}
324 
325 	if (al.map != NULL)
326 		dso__set_hit(map__dso(al.map));
327 
328 	if (ui__has_annotation() || rep->symbol_ipc || rep->total_cycles_mode) {
329 		hist__account_cycles(sample->branch_stack, &al, sample,
330 				     rep->nonany_branch_mode,
331 				     &rep->total_cycles, evsel);
332 	}
333 
334 	ret = hist_entry_iter__add(&iter, &al, rep->max_stack, rep);
335 	if (ret < 0)
336 		pr_debug("problem adding hist entry, skipping event\n");
337 out_put:
338 	addr_location__exit(&al);
339 	return ret;
340 }
341 
342 static int process_read_event(const struct perf_tool *tool,
343 			      union perf_event *event,
344 			      struct perf_sample *sample __maybe_unused,
345 			      struct evsel *evsel,
346 			      struct machine *machine __maybe_unused)
347 {
348 	struct report *rep = container_of(tool, struct report, tool);
349 
350 	if (rep->show_threads) {
351 		const char *name = evsel__name(evsel);
352 		int err = perf_read_values_add_value(&rep->show_threads_values,
353 					   event->read.pid, event->read.tid,
354 					   evsel->core.idx,
355 					   name,
356 					   event->read.value);
357 
358 		if (err)
359 			return err;
360 	}
361 
362 	return 0;
363 }
364 
365 /* For pipe mode, sample_type is not currently set */
366 static int report__setup_sample_type(struct report *rep)
367 {
368 	struct perf_session *session = rep->session;
369 	u64 sample_type = evlist__combined_sample_type(session->evlist);
370 	bool is_pipe = perf_data__is_pipe(session->data);
371 	struct evsel *evsel;
372 
373 	if (session->itrace_synth_opts->callchain ||
374 	    session->itrace_synth_opts->add_callchain ||
375 	    (!is_pipe &&
376 	     perf_header__has_feat(&session->header, HEADER_AUXTRACE) &&
377 	     !session->itrace_synth_opts->set))
378 		sample_type |= PERF_SAMPLE_CALLCHAIN;
379 
380 	if (session->itrace_synth_opts->last_branch ||
381 	    session->itrace_synth_opts->add_last_branch)
382 		sample_type |= PERF_SAMPLE_BRANCH_STACK;
383 
384 	if (!is_pipe && !(sample_type & PERF_SAMPLE_CALLCHAIN)) {
385 		if (perf_hpp_list.parent) {
386 			ui__error("Selected --sort parent, but no "
387 				    "callchain data. Did you call "
388 				    "'perf record' without -g?\n");
389 			return -EINVAL;
390 		}
391 		if (symbol_conf.use_callchain &&
392 			!symbol_conf.show_branchflag_count) {
393 			ui__error("Selected -g or --branch-history.\n"
394 				  "But no callchain or branch data.\n"
395 				  "Did you call 'perf record' without -g or -b?\n");
396 			return -1;
397 		}
398 	} else if (!callchain_param.enabled &&
399 		   callchain_param.mode != CHAIN_NONE &&
400 		   !symbol_conf.use_callchain) {
401 			symbol_conf.use_callchain = true;
402 			if (callchain_register_param(&callchain_param) < 0) {
403 				ui__error("Can't register callchain params.\n");
404 				return -EINVAL;
405 			}
406 	}
407 
408 	if (symbol_conf.cumulate_callchain) {
409 		/* Silently ignore if callchain is missing */
410 		if (!(sample_type & PERF_SAMPLE_CALLCHAIN)) {
411 			symbol_conf.cumulate_callchain = false;
412 			perf_hpp__cancel_cumulate();
413 		}
414 	}
415 
416 	if (sort__mode == SORT_MODE__BRANCH) {
417 		if (!is_pipe &&
418 		    !(sample_type & PERF_SAMPLE_BRANCH_STACK)) {
419 			ui__error("Selected -b but no branch data. "
420 				  "Did you call perf record without -b?\n");
421 			return -1;
422 		}
423 	}
424 
425 	if (sort__mode == SORT_MODE__MEMORY) {
426 		/*
427 		 * FIXUP: prior to kernel 5.18, Arm SPE missed to set
428 		 * PERF_SAMPLE_DATA_SRC bit in sample type.  For backward
429 		 * compatibility, set the bit if it's an old perf data file.
430 		 */
431 		evlist__for_each_entry(session->evlist, evsel) {
432 			if (strstr(evsel__name(evsel), "arm_spe") &&
433 				!(sample_type & PERF_SAMPLE_DATA_SRC)) {
434 				evsel->core.attr.sample_type |= PERF_SAMPLE_DATA_SRC;
435 				sample_type |= PERF_SAMPLE_DATA_SRC;
436 			}
437 		}
438 
439 		if (!is_pipe && !(sample_type & PERF_SAMPLE_DATA_SRC)) {
440 			ui__error("Selected --mem-mode but no mem data. "
441 				  "Did you call perf record without -d?\n");
442 			return -1;
443 		}
444 	}
445 
446 	callchain_param_setup(sample_type, perf_env__arch(&rep->session->header.env));
447 
448 	if (rep->stitch_lbr && (callchain_param.record_mode != CALLCHAIN_LBR)) {
449 		ui__warning("Can't find LBR callchain. Switch off --stitch-lbr.\n"
450 			    "Please apply --call-graph lbr when recording.\n");
451 		rep->stitch_lbr = false;
452 	}
453 
454 	/* ??? handle more cases than just ANY? */
455 	if (!(evlist__combined_branch_type(session->evlist) & PERF_SAMPLE_BRANCH_ANY))
456 		rep->nonany_branch_mode = true;
457 
458 #if !defined(HAVE_LIBUNWIND_SUPPORT) && !defined(HAVE_DWARF_SUPPORT)
459 	if (dwarf_callchain_users) {
460 		ui__warning("Please install libunwind or libdw "
461 			    "development packages during the perf build.\n");
462 	}
463 #endif
464 
465 	return 0;
466 }
467 
468 static void sig_handler(int sig __maybe_unused)
469 {
470 	session_done = 1;
471 }
472 
473 static size_t hists__fprintf_nr_sample_events(struct hists *hists, struct report *rep,
474 					      const char *evname, FILE *fp)
475 {
476 	size_t ret;
477 	char unit;
478 	unsigned long nr_samples = hists->stats.nr_samples;
479 	u64 nr_events = hists->stats.total_period;
480 	struct evsel *evsel = hists_to_evsel(hists);
481 	char buf[512];
482 	size_t size = sizeof(buf);
483 	int socked_id = hists->socket_filter;
484 
485 	if (quiet)
486 		return 0;
487 
488 	if (symbol_conf.filter_relative) {
489 		nr_samples = hists->stats.nr_non_filtered_samples;
490 		nr_events = hists->stats.total_non_filtered_period;
491 	}
492 
493 	if (evsel__is_group_event(evsel)) {
494 		struct evsel *pos;
495 
496 		evsel__group_desc(evsel, buf, size);
497 		evname = buf;
498 
499 		for_each_group_member(pos, evsel) {
500 			const struct hists *pos_hists = evsel__hists(pos);
501 
502 			if (symbol_conf.filter_relative) {
503 				nr_samples += pos_hists->stats.nr_non_filtered_samples;
504 				nr_events += pos_hists->stats.total_non_filtered_period;
505 			} else {
506 				nr_samples += pos_hists->stats.nr_samples;
507 				nr_events += pos_hists->stats.total_period;
508 			}
509 		}
510 	}
511 
512 	nr_samples = convert_unit(nr_samples, &unit);
513 	ret = fprintf(fp, "# Samples: %lu%c", nr_samples, unit);
514 	if (evname != NULL) {
515 		ret += fprintf(fp, " of event%s '%s'",
516 			       evsel->core.nr_members > 1 ? "s" : "", evname);
517 	}
518 
519 	if (rep->time_str)
520 		ret += fprintf(fp, " (time slices: %s)", rep->time_str);
521 
522 	if (symbol_conf.show_ref_callgraph && evname && strstr(evname, "call-graph=no")) {
523 		ret += fprintf(fp, ", show reference callgraph");
524 	}
525 
526 	if (rep->mem_mode) {
527 		ret += fprintf(fp, "\n# Total weight : %" PRIu64, nr_events);
528 		ret += fprintf(fp, "\n# Sort order   : %s", sort_order ? : default_mem_sort_order);
529 	} else
530 		ret += fprintf(fp, "\n# Event count (approx.): %" PRIu64, nr_events);
531 
532 	if (socked_id > -1)
533 		ret += fprintf(fp, "\n# Processor Socket: %d", socked_id);
534 
535 	return ret + fprintf(fp, "\n#\n");
536 }
537 
538 static int evlist__tui_block_hists_browse(struct evlist *evlist, struct report *rep)
539 {
540 	struct evsel *pos;
541 	int i = 0, ret;
542 
543 	evlist__for_each_entry(evlist, pos) {
544 		ret = report__browse_block_hists(&rep->block_reports[i++].hist,
545 						 rep->min_percent, pos,
546 						 &rep->session->header.env);
547 		if (ret != 0)
548 			return ret;
549 	}
550 
551 	return 0;
552 }
553 
554 static int evlist__tty_browse_hists(struct evlist *evlist, struct report *rep, const char *help)
555 {
556 	struct evsel *pos;
557 	int i = 0;
558 
559 	if (!quiet) {
560 		fprintf(stdout, "#\n# Total Lost Samples: %" PRIu64 "\n#\n",
561 			evlist->stats.total_lost_samples);
562 	}
563 
564 	evlist__for_each_entry(evlist, pos) {
565 		struct hists *hists = evsel__hists(pos);
566 		const char *evname = evsel__name(pos);
567 
568 		i++;
569 		if (symbol_conf.event_group && !evsel__is_group_leader(pos))
570 			continue;
571 
572 		if (rep->skip_empty && !hists->stats.nr_samples)
573 			continue;
574 
575 		hists__fprintf_nr_sample_events(hists, rep, evname, stdout);
576 
577 		if (rep->total_cycles_mode) {
578 			char *buf;
579 
580 			if (!annotation_br_cntr_abbr_list(&buf, pos, true)) {
581 				fprintf(stdout, "%s", buf);
582 				fprintf(stdout, "#\n");
583 				free(buf);
584 			}
585 			report__browse_block_hists(&rep->block_reports[i - 1].hist,
586 						   rep->min_percent, pos, NULL);
587 			continue;
588 		}
589 
590 		hists__fprintf(hists, !quiet, 0, 0, rep->min_percent, stdout,
591 			       !(symbol_conf.use_callchain ||
592 			         symbol_conf.show_branchflag_count));
593 		fprintf(stdout, "\n\n");
594 	}
595 
596 	if (!quiet)
597 		fprintf(stdout, "#\n# (%s)\n#\n", help);
598 
599 	if (rep->show_threads) {
600 		bool style = !strcmp(rep->pretty_printing_style, "raw");
601 		perf_read_values_display(stdout, &rep->show_threads_values,
602 					 style);
603 		perf_read_values_destroy(&rep->show_threads_values);
604 	}
605 
606 	if (sort__mode == SORT_MODE__BRANCH)
607 		branch_type_stat_display(stdout, &rep->brtype_stat);
608 
609 	return 0;
610 }
611 
612 static void report__warn_kptr_restrict(const struct report *rep)
613 {
614 	struct map *kernel_map = machine__kernel_map(&rep->session->machines.host);
615 	struct kmap *kernel_kmap = kernel_map ? map__kmap(kernel_map) : NULL;
616 
617 	if (evlist__exclude_kernel(rep->session->evlist))
618 		return;
619 
620 	if (kernel_map == NULL ||
621 	    (dso__hit(map__dso(kernel_map)) &&
622 	     (kernel_kmap->ref_reloc_sym == NULL ||
623 	      kernel_kmap->ref_reloc_sym->addr == 0))) {
624 		const char *desc =
625 		    "As no suitable kallsyms nor vmlinux was found, kernel samples\n"
626 		    "can't be resolved.";
627 
628 		if (kernel_map && map__has_symbols(kernel_map)) {
629 			desc = "If some relocation was applied (e.g. "
630 			       "kexec) symbols may be misresolved.";
631 		}
632 
633 		ui__warning(
634 "Kernel address maps (/proc/{kallsyms,modules}) were restricted.\n\n"
635 "Check /proc/sys/kernel/kptr_restrict before running 'perf record'.\n\n%s\n\n"
636 "Samples in kernel modules can't be resolved as well.\n\n",
637 		desc);
638 	}
639 }
640 
641 static int report__gtk_browse_hists(struct report *rep, const char *help)
642 {
643 	int (*hist_browser)(struct evlist *evlist, const char *help,
644 			    struct hist_browser_timer *timer, float min_pcnt);
645 
646 	hist_browser = dlsym(perf_gtk_handle, "evlist__gtk_browse_hists");
647 
648 	if (hist_browser == NULL) {
649 		ui__error("GTK browser not found!\n");
650 		return -1;
651 	}
652 
653 	return hist_browser(rep->session->evlist, help, NULL, rep->min_percent);
654 }
655 
656 static int report__browse_hists(struct report *rep)
657 {
658 	int ret;
659 	struct perf_session *session = rep->session;
660 	struct evlist *evlist = session->evlist;
661 	char *help = NULL, *path = NULL;
662 
663 	path = system_path(TIPDIR);
664 	if (perf_tip(&help, path) || help == NULL) {
665 		/* fallback for people who don't install perf ;-) */
666 		free(path);
667 		path = system_path(DOCDIR);
668 		if (perf_tip(&help, path) || help == NULL)
669 			help = strdup("Cannot load tips.txt file, please install perf!");
670 	}
671 	free(path);
672 
673 	switch (use_browser) {
674 	case 1:
675 		if (rep->total_cycles_mode) {
676 			ret = evlist__tui_block_hists_browse(evlist, rep);
677 			break;
678 		}
679 
680 		ret = evlist__tui_browse_hists(evlist, help, NULL, rep->min_percent,
681 					       &session->header.env, true);
682 		/*
683 		 * Usually "ret" is the last pressed key, and we only
684 		 * care if the key notifies us to switch data file.
685 		 */
686 		if (ret != K_SWITCH_INPUT_DATA && ret != K_RELOAD)
687 			ret = 0;
688 		break;
689 	case 2:
690 		ret = report__gtk_browse_hists(rep, help);
691 		break;
692 	default:
693 		ret = evlist__tty_browse_hists(evlist, rep, help);
694 		break;
695 	}
696 	free(help);
697 	return ret;
698 }
699 
700 static int report__collapse_hists(struct report *rep)
701 {
702 	struct perf_session *session = rep->session;
703 	struct evlist *evlist = session->evlist;
704 	struct ui_progress prog;
705 	struct evsel *pos;
706 	int ret = 0;
707 
708 	/*
709 	 * The pipe data needs to setup hierarchy hpp formats now, because it
710 	 * cannot know about evsels in the data before reading the data.  The
711 	 * normal file data saves the event (attribute) info in the header
712 	 * section, but pipe does not have the luxury.
713 	 */
714 	if (perf_data__is_pipe(session->data)) {
715 		if (perf_hpp__setup_hists_formats(&perf_hpp_list, evlist) < 0) {
716 			ui__error("Failed to setup hierarchy output formats\n");
717 			return -1;
718 		}
719 	}
720 
721 	ui_progress__init(&prog, rep->nr_entries, "Merging related events...");
722 
723 	evlist__for_each_entry(rep->session->evlist, pos) {
724 		struct hists *hists = evsel__hists(pos);
725 
726 		if (pos->core.idx == 0)
727 			hists->symbol_filter_str = rep->symbol_filter_str;
728 
729 		hists->socket_filter = rep->socket_filter;
730 
731 		ret = hists__collapse_resort(hists, &prog);
732 		if (ret < 0)
733 			break;
734 
735 		/* Non-group events are considered as leader */
736 		if (symbol_conf.event_group && !evsel__is_group_leader(pos)) {
737 			struct hists *leader_hists = evsel__hists(evsel__leader(pos));
738 
739 			hists__match(leader_hists, hists);
740 			hists__link(leader_hists, hists);
741 		}
742 	}
743 
744 	ui_progress__finish();
745 	return ret;
746 }
747 
748 static int hists__resort_cb(struct hist_entry *he, void *arg)
749 {
750 	struct report *rep = arg;
751 	struct symbol *sym = he->ms.sym;
752 
753 	if (rep->symbol_ipc && sym && !sym->annotate2) {
754 		struct evsel *evsel = hists_to_evsel(he->hists);
755 
756 		symbol__annotate2(&he->ms, evsel, NULL);
757 	}
758 
759 	return 0;
760 }
761 
762 static void report__output_resort(struct report *rep)
763 {
764 	struct ui_progress prog;
765 	struct evsel *pos;
766 
767 	ui_progress__init(&prog, rep->nr_entries, "Sorting events for output...");
768 
769 	evlist__for_each_entry(rep->session->evlist, pos) {
770 		evsel__output_resort_cb(pos, &prog, hists__resort_cb, rep);
771 	}
772 
773 	ui_progress__finish();
774 }
775 
776 static int count_sample_event(const struct perf_tool *tool __maybe_unused,
777 			      union perf_event *event __maybe_unused,
778 			      struct perf_sample *sample __maybe_unused,
779 			      struct evsel *evsel,
780 			      struct machine *machine __maybe_unused)
781 {
782 	struct hists *hists = evsel__hists(evsel);
783 
784 	hists__inc_nr_events(hists);
785 	return 0;
786 }
787 
788 static int count_lost_samples_event(const struct perf_tool *tool,
789 				    union perf_event *event,
790 				    struct perf_sample *sample,
791 				    struct machine *machine __maybe_unused)
792 {
793 	struct report *rep = container_of(tool, struct report, tool);
794 	struct evsel *evsel;
795 
796 	evsel = evlist__id2evsel(rep->session->evlist, sample->id);
797 	if (evsel) {
798 		hists__inc_nr_lost_samples(evsel__hists(evsel),
799 					   event->lost_samples.lost);
800 	}
801 	return 0;
802 }
803 
804 static int process_attr(const struct perf_tool *tool __maybe_unused,
805 			union perf_event *event,
806 			struct evlist **pevlist);
807 
808 static void stats_setup(struct report *rep)
809 {
810 	perf_tool__init(&rep->tool, /*ordered_events=*/false);
811 	rep->tool.attr = process_attr;
812 	rep->tool.sample = count_sample_event;
813 	rep->tool.lost_samples = count_lost_samples_event;
814 	rep->tool.no_warn = true;
815 }
816 
817 static int stats_print(struct report *rep)
818 {
819 	struct perf_session *session = rep->session;
820 
821 	perf_session__fprintf_nr_events(session, stdout);
822 	evlist__fprintf_nr_events(session->evlist, stdout);
823 	return 0;
824 }
825 
826 static void tasks_setup(struct report *rep)
827 {
828 	perf_tool__init(&rep->tool, /*ordered_events=*/true);
829 	if (rep->mmaps_mode) {
830 		rep->tool.mmap = perf_event__process_mmap;
831 		rep->tool.mmap2 = perf_event__process_mmap2;
832 	}
833 	rep->tool.attr = process_attr;
834 	rep->tool.comm = perf_event__process_comm;
835 	rep->tool.exit = perf_event__process_exit;
836 	rep->tool.fork = perf_event__process_fork;
837 	rep->tool.no_warn = true;
838 }
839 
840 struct maps__fprintf_task_args {
841 	int indent;
842 	FILE *fp;
843 	size_t printed;
844 };
845 
846 static int maps__fprintf_task_cb(struct map *map, void *data)
847 {
848 	struct maps__fprintf_task_args *args = data;
849 	const struct dso *dso = map__dso(map);
850 	u32 prot = map__prot(map);
851 	int ret;
852 
853 	ret = fprintf(args->fp,
854 		"%*s  %" PRIx64 "-%" PRIx64 " %c%c%c%c %08" PRIx64 " %" PRIu64 " %s\n",
855 		args->indent, "", map__start(map), map__end(map),
856 		prot & PROT_READ ? 'r' : '-',
857 		prot & PROT_WRITE ? 'w' : '-',
858 		prot & PROT_EXEC ? 'x' : '-',
859 		map__flags(map) ? 's' : 'p',
860 		map__pgoff(map),
861 		dso__id_const(dso)->ino, dso__name(dso));
862 
863 	if (ret < 0)
864 		return ret;
865 
866 	args->printed += ret;
867 	return 0;
868 }
869 
870 static size_t maps__fprintf_task(struct maps *maps, int indent, FILE *fp)
871 {
872 	struct maps__fprintf_task_args args = {
873 		.indent = indent,
874 		.fp = fp,
875 		.printed = 0,
876 	};
877 
878 	maps__for_each_map(maps, maps__fprintf_task_cb, &args);
879 
880 	return args.printed;
881 }
882 
883 static int thread_level(struct machine *machine, const struct thread *thread)
884 {
885 	struct thread *parent_thread;
886 	int res;
887 
888 	if (thread__tid(thread) <= 0)
889 		return 0;
890 
891 	if (thread__ppid(thread) <= 0)
892 		return 1;
893 
894 	parent_thread = machine__find_thread(machine, -1, thread__ppid(thread));
895 	if (!parent_thread) {
896 		pr_err("Missing parent thread of %d\n", thread__tid(thread));
897 		return 0;
898 	}
899 	res = 1 + thread_level(machine, parent_thread);
900 	thread__put(parent_thread);
901 	return res;
902 }
903 
904 static void task__print_level(struct machine *machine, struct thread *thread, FILE *fp)
905 {
906 	int level = thread_level(machine, thread);
907 	int comm_indent = fprintf(fp, "  %8d %8d %8d |%*s",
908 				  thread__pid(thread), thread__tid(thread),
909 				  thread__ppid(thread), level, "");
910 
911 	fprintf(fp, "%s\n", thread__comm_str(thread));
912 
913 	maps__fprintf_task(thread__maps(thread), comm_indent, fp);
914 }
915 
916 /*
917  * Sort two thread list nodes such that they form a tree. The first node is the
918  * root of the tree, its children are ordered numerically after it. If a child
919  * has children itself then they appear immediately after their parent. For
920  * example, the 4 threads in the order they'd appear in the list:
921  * - init with a TID 1 and a parent of 0
922  * - systemd with a TID 3000 and a parent of init/1
923  * - systemd child thread with TID 4000, the parent is 3000
924  * - NetworkManager is a child of init with a TID of 3500.
925  */
926 static int task_list_cmp(void *priv, const struct list_head *la, const struct list_head *lb)
927 {
928 	struct machine *machine = priv;
929 	struct thread_list *task_a = list_entry(la, struct thread_list, list);
930 	struct thread_list *task_b = list_entry(lb, struct thread_list, list);
931 	struct thread *a = task_a->thread;
932 	struct thread *b = task_b->thread;
933 	int level_a, level_b, res;
934 
935 	/* Same thread? */
936 	if (thread__tid(a) == thread__tid(b))
937 		return 0;
938 
939 	/* Compare a and b to root. */
940 	if (thread__tid(a) == 0)
941 		return -1;
942 
943 	if (thread__tid(b) == 0)
944 		return 1;
945 
946 	/* If parents match sort by tid. */
947 	if (thread__ppid(a) == thread__ppid(b))
948 		return thread__tid(a) < thread__tid(b) ? -1 : 1;
949 
950 	/*
951 	 * Find a and b such that if they are a child of each other a and b's
952 	 * tid's match, otherwise a and b have a common parent and distinct
953 	 * tid's to sort by. First make the depths of the threads match.
954 	 */
955 	level_a = thread_level(machine, a);
956 	level_b = thread_level(machine, b);
957 	a = thread__get(a);
958 	b = thread__get(b);
959 	for (int i = level_a; i > level_b; i--) {
960 		struct thread *parent = machine__find_thread(machine, -1, thread__ppid(a));
961 
962 		thread__put(a);
963 		if (!parent) {
964 			pr_err("Missing parent thread of %d\n", thread__tid(a));
965 			thread__put(b);
966 			return -1;
967 		}
968 		a = parent;
969 	}
970 	for (int i = level_b; i > level_a; i--) {
971 		struct thread *parent = machine__find_thread(machine, -1, thread__ppid(b));
972 
973 		thread__put(b);
974 		if (!parent) {
975 			pr_err("Missing parent thread of %d\n", thread__tid(b));
976 			thread__put(a);
977 			return 1;
978 		}
979 		b = parent;
980 	}
981 	/* Search up to a common parent. */
982 	while (thread__ppid(a) != thread__ppid(b)) {
983 		struct thread *parent;
984 
985 		parent = machine__find_thread(machine, -1, thread__ppid(a));
986 		thread__put(a);
987 		if (!parent)
988 			pr_err("Missing parent thread of %d\n", thread__tid(a));
989 		a = parent;
990 		parent = machine__find_thread(machine, -1, thread__ppid(b));
991 		thread__put(b);
992 		if (!parent)
993 			pr_err("Missing parent thread of %d\n", thread__tid(b));
994 		b = parent;
995 		if (!a || !b) {
996 			/* Handle missing parent (unexpected) with some sanity. */
997 			thread__put(a);
998 			thread__put(b);
999 			return !a && !b ? 0 : (!a ? -1 : 1);
1000 		}
1001 	}
1002 	if (thread__tid(a) == thread__tid(b)) {
1003 		/* a is a child of b or vice-versa, deeper levels appear later. */
1004 		res = level_a < level_b ? -1 : (level_a > level_b ? 1 : 0);
1005 	} else {
1006 		/* Sort by tid now the parent is the same. */
1007 		res = thread__tid(a) < thread__tid(b) ? -1 : 1;
1008 	}
1009 	thread__put(a);
1010 	thread__put(b);
1011 	return res;
1012 }
1013 
1014 static int tasks_print(struct report *rep, FILE *fp)
1015 {
1016 	struct machine *machine = &rep->session->machines.host;
1017 	LIST_HEAD(tasks);
1018 	int ret;
1019 
1020 	ret = machine__thread_list(machine, &tasks);
1021 	if (!ret) {
1022 		struct thread_list *task;
1023 
1024 		list_sort(machine, &tasks, task_list_cmp);
1025 
1026 		fprintf(fp, "# %8s %8s %8s  %s\n", "pid", "tid", "ppid", "comm");
1027 
1028 		list_for_each_entry(task, &tasks, list)
1029 			task__print_level(machine, task->thread, fp);
1030 	}
1031 	thread_list__delete(&tasks);
1032 	return ret;
1033 }
1034 
1035 static int __cmd_report(struct report *rep)
1036 {
1037 	int ret;
1038 	struct perf_session *session = rep->session;
1039 	struct evsel *pos;
1040 	struct perf_data *data = session->data;
1041 
1042 	signal(SIGINT, sig_handler);
1043 
1044 	if (rep->cpu_list) {
1045 		ret = perf_session__cpu_bitmap(session, rep->cpu_list,
1046 					       rep->cpu_bitmap);
1047 		if (ret) {
1048 			ui__error("failed to set cpu bitmap\n");
1049 			return ret;
1050 		}
1051 		session->itrace_synth_opts->cpu_bitmap = rep->cpu_bitmap;
1052 	}
1053 
1054 	if (rep->show_threads) {
1055 		ret = perf_read_values_init(&rep->show_threads_values);
1056 		if (ret)
1057 			return ret;
1058 	}
1059 
1060 	ret = report__setup_sample_type(rep);
1061 	if (ret) {
1062 		/* report__setup_sample_type() already showed error message */
1063 		return ret;
1064 	}
1065 
1066 	if (rep->stats_mode)
1067 		stats_setup(rep);
1068 
1069 	if (rep->tasks_mode)
1070 		tasks_setup(rep);
1071 
1072 	ret = perf_session__process_events(session);
1073 	if (ret) {
1074 		ui__error("failed to process sample\n");
1075 		return ret;
1076 	}
1077 
1078 	evlist__check_mem_load_aux(session->evlist);
1079 
1080 	if (rep->stats_mode)
1081 		return stats_print(rep);
1082 
1083 	if (rep->tasks_mode)
1084 		return tasks_print(rep, stdout);
1085 
1086 	report__warn_kptr_restrict(rep);
1087 
1088 	evlist__for_each_entry(session->evlist, pos)
1089 		rep->nr_entries += evsel__hists(pos)->nr_entries;
1090 
1091 	if (use_browser == 0) {
1092 		if (verbose > 3)
1093 			perf_session__fprintf(session, stdout);
1094 
1095 		if (verbose > 2)
1096 			perf_session__fprintf_dsos(session, stdout);
1097 
1098 		if (dump_trace) {
1099 			stats_print(rep);
1100 			return 0;
1101 		}
1102 	}
1103 
1104 	ret = report__collapse_hists(rep);
1105 	if (ret) {
1106 		ui__error("failed to process hist entry\n");
1107 		return ret;
1108 	}
1109 
1110 	if (session_done())
1111 		return 0;
1112 
1113 	/*
1114 	 * recalculate number of entries after collapsing since it
1115 	 * might be changed during the collapse phase.
1116 	 */
1117 	rep->nr_entries = 0;
1118 	evlist__for_each_entry(session->evlist, pos)
1119 		rep->nr_entries += evsel__hists(pos)->nr_entries;
1120 
1121 	if (rep->nr_entries == 0) {
1122 		ui__error("The %s data has no samples!\n", data->path);
1123 		return 0;
1124 	}
1125 
1126 	report__output_resort(rep);
1127 
1128 	if (rep->total_cycles_mode) {
1129 		int nr_hpps = 4;
1130 		int block_hpps[PERF_HPP_REPORT__BLOCK_MAX_INDEX] = {
1131 			PERF_HPP_REPORT__BLOCK_TOTAL_CYCLES_PCT,
1132 			PERF_HPP_REPORT__BLOCK_LBR_CYCLES,
1133 			PERF_HPP_REPORT__BLOCK_CYCLES_PCT,
1134 			PERF_HPP_REPORT__BLOCK_AVG_CYCLES,
1135 		};
1136 
1137 		if (session->evlist->nr_br_cntr > 0)
1138 			block_hpps[nr_hpps++] = PERF_HPP_REPORT__BLOCK_BRANCH_COUNTER;
1139 
1140 		block_hpps[nr_hpps++] = PERF_HPP_REPORT__BLOCK_RANGE;
1141 		block_hpps[nr_hpps++] = PERF_HPP_REPORT__BLOCK_DSO;
1142 
1143 		rep->block_reports = block_info__create_report(session->evlist,
1144 							       rep->total_cycles,
1145 							       block_hpps, nr_hpps,
1146 							       &rep->nr_block_reports);
1147 		if (!rep->block_reports)
1148 			return -1;
1149 	}
1150 
1151 	return report__browse_hists(rep);
1152 }
1153 
1154 static int
1155 report_parse_callchain_opt(const struct option *opt, const char *arg, int unset)
1156 {
1157 	struct callchain_param *callchain = opt->value;
1158 
1159 	callchain->enabled = !unset;
1160 	/*
1161 	 * --no-call-graph
1162 	 */
1163 	if (unset) {
1164 		symbol_conf.use_callchain = false;
1165 		callchain->mode = CHAIN_NONE;
1166 		return 0;
1167 	}
1168 
1169 	return parse_callchain_report_opt(arg);
1170 }
1171 
1172 static int
1173 parse_time_quantum(const struct option *opt, const char *arg,
1174 		   int unset __maybe_unused)
1175 {
1176 	unsigned long *time_q = opt->value;
1177 	char *end;
1178 
1179 	*time_q = strtoul(arg, &end, 0);
1180 	if (end == arg)
1181 		goto parse_err;
1182 	if (*time_q == 0) {
1183 		pr_err("time quantum cannot be 0");
1184 		return -1;
1185 	}
1186 	end = skip_spaces(end);
1187 	if (*end == 0)
1188 		return 0;
1189 	if (!strcmp(end, "s")) {
1190 		*time_q *= NSEC_PER_SEC;
1191 		return 0;
1192 	}
1193 	if (!strcmp(end, "ms")) {
1194 		*time_q *= NSEC_PER_MSEC;
1195 		return 0;
1196 	}
1197 	if (!strcmp(end, "us")) {
1198 		*time_q *= NSEC_PER_USEC;
1199 		return 0;
1200 	}
1201 	if (!strcmp(end, "ns"))
1202 		return 0;
1203 parse_err:
1204 	pr_err("Cannot parse time quantum `%s'\n", arg);
1205 	return -1;
1206 }
1207 
1208 int
1209 report_parse_ignore_callees_opt(const struct option *opt __maybe_unused,
1210 				const char *arg, int unset __maybe_unused)
1211 {
1212 	if (arg) {
1213 		int err = regcomp(&ignore_callees_regex, arg, REG_EXTENDED);
1214 		if (err) {
1215 			char buf[BUFSIZ];
1216 			regerror(err, &ignore_callees_regex, buf, sizeof(buf));
1217 			pr_err("Invalid --ignore-callees regex: %s\n%s", arg, buf);
1218 			return -1;
1219 		}
1220 		have_ignore_callees = 1;
1221 	}
1222 
1223 	return 0;
1224 }
1225 
1226 static int
1227 parse_branch_mode(const struct option *opt,
1228 		  const char *str __maybe_unused, int unset)
1229 {
1230 	int *branch_mode = opt->value;
1231 
1232 	*branch_mode = !unset;
1233 	return 0;
1234 }
1235 
1236 static int
1237 parse_percent_limit(const struct option *opt, const char *str,
1238 		    int unset __maybe_unused)
1239 {
1240 	struct report *rep = opt->value;
1241 	double pcnt = strtof(str, NULL);
1242 
1243 	rep->min_percent = pcnt;
1244 	callchain_param.min_percent = pcnt;
1245 	return 0;
1246 }
1247 
1248 static int process_attr(const struct perf_tool *tool __maybe_unused,
1249 			union perf_event *event,
1250 			struct evlist **pevlist)
1251 {
1252 	u64 sample_type;
1253 	int err;
1254 
1255 	err = perf_event__process_attr(tool, event, pevlist);
1256 	if (err)
1257 		return err;
1258 
1259 	/*
1260 	 * Check if we need to enable callchains based
1261 	 * on events sample_type.
1262 	 */
1263 	sample_type = evlist__combined_sample_type(*pevlist);
1264 	callchain_param_setup(sample_type, perf_env__arch((*pevlist)->env));
1265 	return 0;
1266 }
1267 
1268 int cmd_report(int argc, const char **argv)
1269 {
1270 	struct perf_session *session;
1271 	struct itrace_synth_opts itrace_synth_opts = { .set = 0, };
1272 	struct stat st;
1273 	bool has_br_stack = false;
1274 	int branch_mode = -1;
1275 	int last_key = 0;
1276 	bool branch_call_mode = false;
1277 #define CALLCHAIN_DEFAULT_OPT  "graph,0.5,caller,function,percent"
1278 	static const char report_callchain_help[] = "Display call graph (stack chain/backtrace):\n\n"
1279 						    CALLCHAIN_REPORT_HELP
1280 						    "\n\t\t\t\tDefault: " CALLCHAIN_DEFAULT_OPT;
1281 	char callchain_default_opt[] = CALLCHAIN_DEFAULT_OPT;
1282 	const char * const report_usage[] = {
1283 		"perf report [<options>]",
1284 		NULL
1285 	};
1286 	struct report report = {
1287 		.max_stack		 = PERF_MAX_STACK_DEPTH,
1288 		.pretty_printing_style	 = "normal",
1289 		.socket_filter		 = -1,
1290 		.skip_empty		 = true,
1291 	};
1292 	char *sort_order_help = sort_help("sort by key(s):", SORT_MODE__NORMAL);
1293 	char *field_order_help = sort_help("output field(s):", SORT_MODE__NORMAL);
1294 	const char *disassembler_style = NULL, *objdump_path = NULL, *addr2line_path = NULL;
1295 	const struct option options[] = {
1296 	OPT_STRING('i', "input", &input_name, "file",
1297 		    "input file name"),
1298 	OPT_INCR('v', "verbose", &verbose,
1299 		    "be more verbose (show symbol address, etc)"),
1300 	OPT_BOOLEAN('q', "quiet", &quiet, "Do not show any warnings or messages"),
1301 	OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace,
1302 		    "dump raw trace in ASCII"),
1303 	OPT_BOOLEAN(0, "stats", &report.stats_mode, "Display event stats"),
1304 	OPT_BOOLEAN(0, "tasks", &report.tasks_mode, "Display recorded tasks"),
1305 	OPT_BOOLEAN(0, "mmaps", &report.mmaps_mode, "Display recorded tasks memory maps"),
1306 	OPT_STRING('k', "vmlinux", &symbol_conf.vmlinux_name,
1307 		   "file", "vmlinux pathname"),
1308 	OPT_BOOLEAN(0, "ignore-vmlinux", &symbol_conf.ignore_vmlinux,
1309                     "don't load vmlinux even if found"),
1310 	OPT_STRING(0, "kallsyms", &symbol_conf.kallsyms_name,
1311 		   "file", "kallsyms pathname"),
1312 	OPT_BOOLEAN('f', "force", &symbol_conf.force, "don't complain, do it"),
1313 	OPT_BOOLEAN('m', "modules", &symbol_conf.use_modules,
1314 		    "load module symbols - WARNING: use only with -k and LIVE kernel"),
1315 	OPT_BOOLEAN('n', "show-nr-samples", &symbol_conf.show_nr_samples,
1316 		    "Show a column with the number of samples"),
1317 	OPT_BOOLEAN('T', "threads", &report.show_threads,
1318 		    "Show per-thread event counters"),
1319 	OPT_STRING(0, "pretty", &report.pretty_printing_style, "key",
1320 		   "pretty printing style key: normal raw"),
1321 #ifdef HAVE_SLANG_SUPPORT
1322 	OPT_BOOLEAN(0, "tui", &report.use_tui, "Use the TUI interface"),
1323 #endif
1324 #ifdef HAVE_GTK2_SUPPORT
1325 	OPT_BOOLEAN(0, "gtk", &report.use_gtk, "Use the GTK2 interface"),
1326 #endif
1327 	OPT_BOOLEAN(0, "stdio", &report.use_stdio,
1328 		    "Use the stdio interface"),
1329 	OPT_BOOLEAN(0, "header", &report.header, "Show data header."),
1330 	OPT_BOOLEAN(0, "header-only", &report.header_only,
1331 		    "Show only data header."),
1332 	OPT_STRING('s', "sort", &sort_order, "key[,key2...]",
1333 		   sort_order_help),
1334 	OPT_STRING('F', "fields", &field_order, "key[,keys...]",
1335 		   field_order_help),
1336 	OPT_BOOLEAN(0, "show-cpu-utilization", &symbol_conf.show_cpu_utilization,
1337 		    "Show sample percentage for different cpu modes"),
1338 	OPT_BOOLEAN_FLAG(0, "showcpuutilization", &symbol_conf.show_cpu_utilization,
1339 		    "Show sample percentage for different cpu modes", PARSE_OPT_HIDDEN),
1340 	OPT_STRING('p', "parent", &parent_pattern, "regex",
1341 		   "regex filter to identify parent, see: '--sort parent'"),
1342 	OPT_BOOLEAN('x', "exclude-other", &symbol_conf.exclude_other,
1343 		    "Only display entries with parent-match"),
1344 	OPT_CALLBACK_DEFAULT('g', "call-graph", &callchain_param,
1345 			     "print_type,threshold[,print_limit],order,sort_key[,branch],value",
1346 			     report_callchain_help, &report_parse_callchain_opt,
1347 			     callchain_default_opt),
1348 	OPT_BOOLEAN(0, "children", &symbol_conf.cumulate_callchain,
1349 		    "Accumulate callchains of children and show total overhead as well. "
1350 		    "Enabled by default, use --no-children to disable."),
1351 	OPT_INTEGER(0, "max-stack", &report.max_stack,
1352 		    "Set the maximum stack depth when parsing the callchain, "
1353 		    "anything beyond the specified depth will be ignored. "
1354 		    "Default: kernel.perf_event_max_stack or " __stringify(PERF_MAX_STACK_DEPTH)),
1355 	OPT_BOOLEAN('G', "inverted", &report.inverted_callchain,
1356 		    "alias for inverted call graph"),
1357 	OPT_CALLBACK(0, "ignore-callees", NULL, "regex",
1358 		   "ignore callees of these functions in call graphs",
1359 		   report_parse_ignore_callees_opt),
1360 	OPT_STRING('d', "dsos", &symbol_conf.dso_list_str, "dso[,dso...]",
1361 		   "only consider symbols in these dsos"),
1362 	OPT_STRING('c', "comms", &symbol_conf.comm_list_str, "comm[,comm...]",
1363 		   "only consider symbols in these comms"),
1364 	OPT_STRING(0, "pid", &symbol_conf.pid_list_str, "pid[,pid...]",
1365 		   "only consider symbols in these pids"),
1366 	OPT_STRING(0, "tid", &symbol_conf.tid_list_str, "tid[,tid...]",
1367 		   "only consider symbols in these tids"),
1368 	OPT_STRING('S', "symbols", &symbol_conf.sym_list_str, "symbol[,symbol...]",
1369 		   "only consider these symbols"),
1370 	OPT_STRING(0, "symbol-filter", &report.symbol_filter_str, "filter",
1371 		   "only show symbols that (partially) match with this filter"),
1372 	OPT_STRING('w', "column-widths", &symbol_conf.col_width_list_str,
1373 		   "width[,width...]",
1374 		   "don't try to adjust column width, use these fixed values"),
1375 	OPT_STRING_NOEMPTY('t', "field-separator", &symbol_conf.field_sep, "separator",
1376 		   "separator for columns, no spaces will be added between "
1377 		   "columns '.' is reserved."),
1378 	OPT_BOOLEAN('U', "hide-unresolved", &symbol_conf.hide_unresolved,
1379 		    "Only display entries resolved to a symbol"),
1380 	OPT_CALLBACK(0, "symfs", NULL, "directory",
1381 		     "Look for files with symbols relative to this directory",
1382 		     symbol__config_symfs),
1383 	OPT_STRING('C', "cpu", &report.cpu_list, "cpu",
1384 		   "list of cpus to profile"),
1385 	OPT_BOOLEAN('I', "show-info", &report.show_full_info,
1386 		    "Display extended information about perf.data file"),
1387 	OPT_BOOLEAN(0, "source", &annotate_opts.annotate_src,
1388 		    "Interleave source code with assembly code (default)"),
1389 	OPT_BOOLEAN(0, "asm-raw", &annotate_opts.show_asm_raw,
1390 		    "Display raw encoding of assembly instructions (default)"),
1391 	OPT_STRING('M', "disassembler-style", &disassembler_style, "disassembler style",
1392 		   "Specify disassembler style (e.g. -M intel for intel syntax)"),
1393 	OPT_STRING(0, "prefix", &annotate_opts.prefix, "prefix",
1394 		    "Add prefix to source file path names in programs (with --prefix-strip)"),
1395 	OPT_STRING(0, "prefix-strip", &annotate_opts.prefix_strip, "N",
1396 		    "Strip first N entries of source file path name in programs (with --prefix)"),
1397 	OPT_BOOLEAN(0, "show-total-period", &symbol_conf.show_total_period,
1398 		    "Show a column with the sum of periods"),
1399 	OPT_BOOLEAN_SET(0, "group", &symbol_conf.event_group, &report.group_set,
1400 		    "Show event group information together"),
1401 	OPT_INTEGER(0, "group-sort-idx", &symbol_conf.group_sort_idx,
1402 		    "Sort the output by the event at the index n in group. "
1403 		    "If n is invalid, sort by the first event. "
1404 		    "WARNING: should be used on grouped events."),
1405 	OPT_CALLBACK_NOOPT('b', "branch-stack", &branch_mode, "",
1406 		    "use branch records for per branch histogram filling",
1407 		    parse_branch_mode),
1408 	OPT_BOOLEAN(0, "branch-history", &branch_call_mode,
1409 		    "add last branch records to call history"),
1410 	OPT_STRING(0, "objdump", &objdump_path, "path",
1411 		   "objdump binary to use for disassembly and annotations"),
1412 	OPT_STRING(0, "addr2line", &addr2line_path, "path",
1413 		   "addr2line binary to use for line numbers"),
1414 	OPT_BOOLEAN(0, "demangle", &symbol_conf.demangle,
1415 		    "Disable symbol demangling"),
1416 	OPT_BOOLEAN(0, "demangle-kernel", &symbol_conf.demangle_kernel,
1417 		    "Enable kernel symbol demangling"),
1418 	OPT_BOOLEAN(0, "mem-mode", &report.mem_mode, "mem access profile"),
1419 	OPT_INTEGER(0, "samples", &symbol_conf.res_sample,
1420 		    "Number of samples to save per histogram entry for individual browsing"),
1421 	OPT_CALLBACK(0, "percent-limit", &report, "percent",
1422 		     "Don't show entries under that percent", parse_percent_limit),
1423 	OPT_CALLBACK(0, "percentage", NULL, "relative|absolute",
1424 		     "how to display percentage of filtered entries", parse_filter_percentage),
1425 	OPT_CALLBACK_OPTARG(0, "itrace", &itrace_synth_opts, NULL, "opts",
1426 			    "Instruction Tracing options\n" ITRACE_HELP,
1427 			    itrace_parse_synth_opts),
1428 	OPT_BOOLEAN(0, "full-source-path", &srcline_full_filename,
1429 			"Show full source file name path for source lines"),
1430 	OPT_BOOLEAN(0, "show-ref-call-graph", &symbol_conf.show_ref_callgraph,
1431 		    "Show callgraph from reference event"),
1432 	OPT_BOOLEAN(0, "stitch-lbr", &report.stitch_lbr,
1433 		    "Enable LBR callgraph stitching approach"),
1434 	OPT_INTEGER(0, "socket-filter", &report.socket_filter,
1435 		    "only show processor socket that match with this filter"),
1436 	OPT_BOOLEAN(0, "raw-trace", &symbol_conf.raw_trace,
1437 		    "Show raw trace event output (do not use print fmt or plugins)"),
1438 	OPT_BOOLEAN('H', "hierarchy", &symbol_conf.report_hierarchy,
1439 		    "Show entries in a hierarchy"),
1440 	OPT_CALLBACK_DEFAULT(0, "stdio-color", NULL, "mode",
1441 			     "'always' (default), 'never' or 'auto' only applicable to --stdio mode",
1442 			     stdio__config_color, "always"),
1443 	OPT_STRING(0, "time", &report.time_str, "str",
1444 		   "Time span of interest (start,stop)"),
1445 	OPT_BOOLEAN(0, "inline", &symbol_conf.inline_name,
1446 		    "Show inline function"),
1447 	OPT_CALLBACK(0, "percent-type", &annotate_opts, "local-period",
1448 		     "Set percent type local/global-period/hits",
1449 		     annotate_parse_percent_type),
1450 	OPT_BOOLEAN(0, "ns", &symbol_conf.nanosecs, "Show times in nanosecs"),
1451 	OPT_CALLBACK(0, "time-quantum", &symbol_conf.time_quantum, "time (ms|us|ns|s)",
1452 		     "Set time quantum for time sort key (default 100ms)",
1453 		     parse_time_quantum),
1454 	OPTS_EVSWITCH(&report.evswitch),
1455 	OPT_BOOLEAN(0, "total-cycles", &report.total_cycles_mode,
1456 		    "Sort all blocks by 'Sampled Cycles%'"),
1457 	OPT_BOOLEAN(0, "disable-order", &report.disable_order,
1458 		    "Disable raw trace ordering"),
1459 	OPT_BOOLEAN(0, "skip-empty", &report.skip_empty,
1460 		    "Do not display empty (or dummy) events in the output"),
1461 	OPT_END()
1462 	};
1463 	struct perf_data data = {
1464 		.mode  = PERF_DATA_MODE_READ,
1465 	};
1466 	int ret = hists__init();
1467 	char sort_tmp[128];
1468 	bool ordered_events = true;
1469 
1470 	if (ret < 0)
1471 		goto exit;
1472 
1473 	/*
1474 	 * tasks_mode require access to exited threads to list those that are in
1475 	 * the data file. Off-cpu events are synthesized after other events and
1476 	 * reference exited threads.
1477 	 */
1478 	symbol_conf.keep_exited_threads = true;
1479 
1480 	annotation_options__init();
1481 
1482 	ret = perf_config(report__config, &report);
1483 	if (ret)
1484 		goto exit;
1485 
1486 	argc = parse_options(argc, argv, options, report_usage, 0);
1487 	if (argc) {
1488 		/*
1489 		 * Special case: if there's an argument left then assume that
1490 		 * it's a symbol filter:
1491 		 */
1492 		if (argc > 1)
1493 			usage_with_options(report_usage, options);
1494 
1495 		report.symbol_filter_str = argv[0];
1496 	}
1497 
1498 	if (disassembler_style) {
1499 		annotate_opts.disassembler_style = strdup(disassembler_style);
1500 		if (!annotate_opts.disassembler_style)
1501 			return -ENOMEM;
1502 	}
1503 	if (objdump_path) {
1504 		annotate_opts.objdump_path = strdup(objdump_path);
1505 		if (!annotate_opts.objdump_path)
1506 			return -ENOMEM;
1507 	}
1508 	if (addr2line_path) {
1509 		symbol_conf.addr2line_path = strdup(addr2line_path);
1510 		if (!symbol_conf.addr2line_path)
1511 			return -ENOMEM;
1512 	}
1513 
1514 	if (annotate_check_args() < 0) {
1515 		ret = -EINVAL;
1516 		goto exit;
1517 	}
1518 
1519 	if (report.mmaps_mode)
1520 		report.tasks_mode = true;
1521 
1522 	if (dump_trace && report.disable_order)
1523 		ordered_events = false;
1524 
1525 	if (quiet)
1526 		perf_quiet_option();
1527 
1528 	ret = symbol__validate_sym_arguments();
1529 	if (ret)
1530 		goto exit;
1531 
1532 	if (report.inverted_callchain)
1533 		callchain_param.order = ORDER_CALLER;
1534 	if (symbol_conf.cumulate_callchain && !callchain_param.order_set)
1535 		callchain_param.order = ORDER_CALLER;
1536 
1537 	if ((itrace_synth_opts.callchain || itrace_synth_opts.add_callchain) &&
1538 	    (int)itrace_synth_opts.callchain_sz > report.max_stack)
1539 		report.max_stack = itrace_synth_opts.callchain_sz;
1540 
1541 	if (!input_name || !strlen(input_name)) {
1542 		if (!fstat(STDIN_FILENO, &st) && S_ISFIFO(st.st_mode))
1543 			input_name = "-";
1544 		else
1545 			input_name = "perf.data";
1546 	}
1547 
1548 	data.path  = input_name;
1549 	data.force = symbol_conf.force;
1550 
1551 	symbol_conf.skip_empty = report.skip_empty;
1552 
1553 repeat:
1554 	perf_tool__init(&report.tool, ordered_events);
1555 	report.tool.sample		 = process_sample_event;
1556 	report.tool.mmap		 = perf_event__process_mmap;
1557 	report.tool.mmap2		 = perf_event__process_mmap2;
1558 	report.tool.comm		 = perf_event__process_comm;
1559 	report.tool.namespaces		 = perf_event__process_namespaces;
1560 	report.tool.cgroup		 = perf_event__process_cgroup;
1561 	report.tool.exit		 = perf_event__process_exit;
1562 	report.tool.fork		 = perf_event__process_fork;
1563 	report.tool.lost		 = perf_event__process_lost;
1564 	report.tool.read		 = process_read_event;
1565 	report.tool.attr		 = process_attr;
1566 #ifdef HAVE_LIBTRACEEVENT
1567 	report.tool.tracing_data	 = perf_event__process_tracing_data;
1568 #endif
1569 	report.tool.build_id		 = perf_event__process_build_id;
1570 	report.tool.id_index		 = perf_event__process_id_index;
1571 	report.tool.auxtrace_info	 = perf_event__process_auxtrace_info;
1572 	report.tool.auxtrace		 = perf_event__process_auxtrace;
1573 	report.tool.event_update	 = perf_event__process_event_update;
1574 	report.tool.feature		 = process_feature_event;
1575 	report.tool.ordering_requires_timestamps = true;
1576 
1577 	session = perf_session__new(&data, &report.tool);
1578 	if (IS_ERR(session)) {
1579 		ret = PTR_ERR(session);
1580 		goto exit;
1581 	}
1582 
1583 	ret = evswitch__init(&report.evswitch, session->evlist, stderr);
1584 	if (ret)
1585 		goto exit;
1586 
1587 	if (zstd_init(&(session->zstd_data), 0) < 0)
1588 		pr_warning("Decompression initialization failed. Reported data may be incomplete.\n");
1589 
1590 	if (report.queue_size) {
1591 		ordered_events__set_alloc_size(&session->ordered_events,
1592 					       report.queue_size);
1593 	}
1594 
1595 	session->itrace_synth_opts = &itrace_synth_opts;
1596 
1597 	report.session = session;
1598 
1599 	has_br_stack = perf_header__has_feat(&session->header,
1600 					     HEADER_BRANCH_STACK);
1601 	if (evlist__combined_sample_type(session->evlist) & PERF_SAMPLE_STACK_USER)
1602 		has_br_stack = false;
1603 
1604 	setup_forced_leader(&report, session->evlist);
1605 
1606 	if (symbol_conf.group_sort_idx && evlist__nr_groups(session->evlist) == 0) {
1607 		parse_options_usage(NULL, options, "group-sort-idx", 0);
1608 		ret = -EINVAL;
1609 		goto error;
1610 	}
1611 
1612 	if (itrace_synth_opts.last_branch || itrace_synth_opts.add_last_branch)
1613 		has_br_stack = true;
1614 
1615 	if (has_br_stack && branch_call_mode)
1616 		symbol_conf.show_branchflag_count = true;
1617 
1618 	memset(&report.brtype_stat, 0, sizeof(struct branch_type_stat));
1619 
1620 	/*
1621 	 * Branch mode is a tristate:
1622 	 * -1 means default, so decide based on the file having branch data.
1623 	 * 0/1 means the user chose a mode.
1624 	 */
1625 	if (((branch_mode == -1 && has_br_stack) || branch_mode == 1) &&
1626 	    !branch_call_mode) {
1627 		sort__mode = SORT_MODE__BRANCH;
1628 		symbol_conf.cumulate_callchain = false;
1629 	}
1630 	if (branch_call_mode) {
1631 		callchain_param.key = CCKEY_ADDRESS;
1632 		callchain_param.branch_callstack = true;
1633 		symbol_conf.use_callchain = true;
1634 		callchain_register_param(&callchain_param);
1635 		if (sort_order == NULL)
1636 			sort_order = "srcline,symbol,dso";
1637 	}
1638 
1639 	if (report.mem_mode) {
1640 		if (sort__mode == SORT_MODE__BRANCH) {
1641 			pr_err("branch and mem mode incompatible\n");
1642 			goto error;
1643 		}
1644 		sort__mode = SORT_MODE__MEMORY;
1645 		symbol_conf.cumulate_callchain = false;
1646 	}
1647 
1648 	if (symbol_conf.report_hierarchy) {
1649 		/* disable incompatible options */
1650 		symbol_conf.cumulate_callchain = false;
1651 
1652 		if (field_order) {
1653 			pr_err("Error: --hierarchy and --fields options cannot be used together\n");
1654 			parse_options_usage(report_usage, options, "F", 1);
1655 			parse_options_usage(NULL, options, "hierarchy", 0);
1656 			goto error;
1657 		}
1658 
1659 		perf_hpp_list.need_collapse = true;
1660 	}
1661 
1662 	if (report.use_stdio)
1663 		use_browser = 0;
1664 #ifdef HAVE_SLANG_SUPPORT
1665 	else if (report.use_tui)
1666 		use_browser = 1;
1667 #endif
1668 #ifdef HAVE_GTK2_SUPPORT
1669 	else if (report.use_gtk)
1670 		use_browser = 2;
1671 #endif
1672 
1673 	/* Force tty output for header output and per-thread stat. */
1674 	if (report.header || report.header_only || report.show_threads)
1675 		use_browser = 0;
1676 	if (report.header || report.header_only)
1677 		report.tool.show_feat_hdr = SHOW_FEAT_HEADER;
1678 	if (report.show_full_info)
1679 		report.tool.show_feat_hdr = SHOW_FEAT_HEADER_FULL_INFO;
1680 	if (report.stats_mode || report.tasks_mode)
1681 		use_browser = 0;
1682 	if (report.stats_mode && report.tasks_mode) {
1683 		pr_err("Error: --tasks and --mmaps can't be used together with --stats\n");
1684 		goto error;
1685 	}
1686 
1687 	if (report.total_cycles_mode) {
1688 		if (sort__mode != SORT_MODE__BRANCH)
1689 			report.total_cycles_mode = false;
1690 		else
1691 			sort_order = NULL;
1692 	}
1693 
1694 	if (sort_order && strstr(sort_order, "type")) {
1695 		report.data_type = true;
1696 		annotate_opts.annotate_src = false;
1697 
1698 #ifndef HAVE_DWARF_GETLOCATIONS_SUPPORT
1699 		pr_err("Error: Data type profiling is disabled due to missing DWARF support\n");
1700 		goto error;
1701 #endif
1702 	}
1703 
1704 	if (strcmp(input_name, "-") != 0)
1705 		setup_browser(true);
1706 	else
1707 		use_browser = 0;
1708 
1709 	if (report.data_type && use_browser == 1) {
1710 		symbol_conf.annotate_data_member = true;
1711 		symbol_conf.annotate_data_sample = true;
1712 	}
1713 
1714 	if (sort_order && strstr(sort_order, "ipc")) {
1715 		parse_options_usage(report_usage, options, "s", 1);
1716 		goto error;
1717 	}
1718 
1719 	if (sort_order && strstr(sort_order, "symbol")) {
1720 		if (sort__mode == SORT_MODE__BRANCH) {
1721 			snprintf(sort_tmp, sizeof(sort_tmp), "%s,%s",
1722 				 sort_order, "ipc_lbr");
1723 			report.symbol_ipc = true;
1724 		} else {
1725 			snprintf(sort_tmp, sizeof(sort_tmp), "%s,%s",
1726 				 sort_order, "ipc_null");
1727 		}
1728 
1729 		sort_order = sort_tmp;
1730 	}
1731 
1732 	if ((last_key != K_SWITCH_INPUT_DATA && last_key != K_RELOAD) &&
1733 	    (setup_sorting(session->evlist) < 0)) {
1734 		if (sort_order)
1735 			parse_options_usage(report_usage, options, "s", 1);
1736 		if (field_order)
1737 			parse_options_usage(sort_order ? NULL : report_usage,
1738 					    options, "F", 1);
1739 		goto error;
1740 	}
1741 
1742 	if ((report.header || report.header_only) && !quiet) {
1743 		perf_session__fprintf_info(session, stdout,
1744 					   report.show_full_info);
1745 		if (report.header_only) {
1746 			if (data.is_pipe) {
1747 				/*
1748 				 * we need to process first few records
1749 				 * which contains PERF_RECORD_HEADER_FEATURE.
1750 				 */
1751 				perf_session__process_events(session);
1752 			}
1753 			ret = 0;
1754 			goto error;
1755 		}
1756 	} else if (use_browser == 0 && !quiet &&
1757 		   !report.stats_mode && !report.tasks_mode) {
1758 		fputs("# To display the perf.data header info, please use --header/--header-only options.\n#\n",
1759 		      stdout);
1760 	}
1761 
1762 	/*
1763 	 * Only in the TUI browser we are doing integrated annotation,
1764 	 * so don't allocate extra space that won't be used in the stdio
1765 	 * implementation.
1766 	 */
1767 	if (ui__has_annotation() || report.symbol_ipc || report.data_type ||
1768 	    report.total_cycles_mode) {
1769 		ret = symbol__annotation_init();
1770 		if (ret < 0)
1771 			goto error;
1772 		/*
1773  		 * For searching by name on the "Browse map details".
1774  		 * providing it only in verbose mode not to bloat too
1775  		 * much struct symbol.
1776  		 */
1777 		if (verbose > 0) {
1778 			/*
1779 			 * XXX: Need to provide a less kludgy way to ask for
1780 			 * more space per symbol, the u32 is for the index on
1781 			 * the ui browser.
1782 			 * See symbol__browser_index.
1783 			 */
1784 			symbol_conf.priv_size += sizeof(u32);
1785 		}
1786 		annotation_config__init();
1787 	}
1788 
1789 	if (symbol__init(&session->header.env) < 0)
1790 		goto error;
1791 
1792 	if (report.time_str) {
1793 		ret = perf_time__parse_for_ranges(report.time_str, session,
1794 						  &report.ptime_range,
1795 						  &report.range_size,
1796 						  &report.range_num);
1797 		if (ret < 0)
1798 			goto error;
1799 
1800 		itrace_synth_opts__set_time_range(&itrace_synth_opts,
1801 						  report.ptime_range,
1802 						  report.range_num);
1803 	}
1804 
1805 #ifdef HAVE_LIBTRACEEVENT
1806 	if (session->tevent.pevent &&
1807 	    tep_set_function_resolver(session->tevent.pevent,
1808 				      machine__resolve_kernel_addr,
1809 				      &session->machines.host) < 0) {
1810 		pr_err("%s: failed to set libtraceevent function resolver\n",
1811 		       __func__);
1812 		return -1;
1813 	}
1814 #endif
1815 	sort__setup_elide(stdout);
1816 
1817 	ret = __cmd_report(&report);
1818 	if (ret == K_SWITCH_INPUT_DATA || ret == K_RELOAD) {
1819 		perf_session__delete(session);
1820 		last_key = K_SWITCH_INPUT_DATA;
1821 		goto repeat;
1822 	} else
1823 		ret = 0;
1824 
1825 	if (!use_browser && (verbose > 2 || debug_kmaps))
1826 		perf_session__dump_kmaps(session);
1827 error:
1828 	if (report.ptime_range) {
1829 		itrace_synth_opts__clear_time_range(&itrace_synth_opts);
1830 		zfree(&report.ptime_range);
1831 	}
1832 
1833 	if (report.block_reports) {
1834 		block_info__free_report(report.block_reports,
1835 					report.nr_block_reports);
1836 		report.block_reports = NULL;
1837 	}
1838 
1839 	zstd_fini(&(session->zstd_data));
1840 	perf_session__delete(session);
1841 exit:
1842 	annotation_options__exit();
1843 	free(sort_order_help);
1844 	free(field_order_help);
1845 	return ret;
1846 }
1847