xref: /openbmc/linux/tools/perf/builtin-report.c (revision 75016ca3)
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/session.h"
35 #include "util/srcline.h"
36 #include "util/tool.h"
37 
38 #include <subcmd/parse-options.h>
39 #include <subcmd/exec-cmd.h>
40 #include "util/parse-events.h"
41 
42 #include "util/thread.h"
43 #include "util/sort.h"
44 #include "util/hist.h"
45 #include "util/data.h"
46 #include "arch/common.h"
47 #include "util/time-utils.h"
48 #include "util/auxtrace.h"
49 #include "util/units.h"
50 #include "util/util.h" // perf_tip()
51 #include "ui/ui.h"
52 #include "ui/progress.h"
53 #include "util/block-info.h"
54 
55 #include <dlfcn.h>
56 #include <errno.h>
57 #include <inttypes.h>
58 #include <regex.h>
59 #include <linux/ctype.h>
60 #include <signal.h>
61 #include <linux/bitmap.h>
62 #include <linux/string.h>
63 #include <linux/stringify.h>
64 #include <linux/time64.h>
65 #include <sys/types.h>
66 #include <sys/stat.h>
67 #include <unistd.h>
68 #include <linux/mman.h>
69 
70 struct report {
71 	struct perf_tool	tool;
72 	struct perf_session	*session;
73 	struct evswitch		evswitch;
74 	bool			use_tui, use_gtk, use_stdio;
75 	bool			show_full_info;
76 	bool			show_threads;
77 	bool			inverted_callchain;
78 	bool			mem_mode;
79 	bool			stats_mode;
80 	bool			tasks_mode;
81 	bool			mmaps_mode;
82 	bool			header;
83 	bool			header_only;
84 	bool			nonany_branch_mode;
85 	bool			group_set;
86 	bool			stitch_lbr;
87 	bool			disable_order;
88 	bool			skip_empty;
89 	int			max_stack;
90 	struct perf_read_values	show_threads_values;
91 	struct annotation_options annotation_opts;
92 	const char		*pretty_printing_style;
93 	const char		*cpu_list;
94 	const char		*symbol_filter_str;
95 	const char		*time_str;
96 	struct perf_time_interval *ptime_range;
97 	int			range_size;
98 	int			range_num;
99 	float			min_percent;
100 	u64			nr_entries;
101 	u64			queue_size;
102 	u64			total_cycles;
103 	int			socket_filter;
104 	DECLARE_BITMAP(cpu_bitmap, MAX_NR_CPUS);
105 	struct branch_type_stat	brtype_stat;
106 	bool			symbol_ipc;
107 	bool			total_cycles_mode;
108 	struct block_report	*block_reports;
109 	int			nr_block_reports;
110 };
111 
112 static int report__config(const char *var, const char *value, void *cb)
113 {
114 	struct report *rep = cb;
115 
116 	if (!strcmp(var, "report.group")) {
117 		symbol_conf.event_group = perf_config_bool(var, value);
118 		return 0;
119 	}
120 	if (!strcmp(var, "report.percent-limit")) {
121 		double pcnt = strtof(value, NULL);
122 
123 		rep->min_percent = pcnt;
124 		callchain_param.min_percent = pcnt;
125 		return 0;
126 	}
127 	if (!strcmp(var, "report.children")) {
128 		symbol_conf.cumulate_callchain = perf_config_bool(var, value);
129 		return 0;
130 	}
131 	if (!strcmp(var, "report.queue-size"))
132 		return perf_config_u64(&rep->queue_size, var, value);
133 
134 	if (!strcmp(var, "report.sort_order")) {
135 		default_sort_order = strdup(value);
136 		return 0;
137 	}
138 
139 	if (!strcmp(var, "report.skip-empty")) {
140 		rep->skip_empty = perf_config_bool(var, value);
141 		return 0;
142 	}
143 
144 	return 0;
145 }
146 
147 static int hist_iter__report_callback(struct hist_entry_iter *iter,
148 				      struct addr_location *al, bool single,
149 				      void *arg)
150 {
151 	int err = 0;
152 	struct report *rep = arg;
153 	struct hist_entry *he = iter->he;
154 	struct evsel *evsel = iter->evsel;
155 	struct perf_sample *sample = iter->sample;
156 	struct mem_info *mi;
157 	struct branch_info *bi;
158 
159 	if (!ui__has_annotation() && !rep->symbol_ipc)
160 		return 0;
161 
162 	if (sort__mode == SORT_MODE__BRANCH) {
163 		bi = he->branch_info;
164 		err = addr_map_symbol__inc_samples(&bi->from, sample, evsel);
165 		if (err)
166 			goto out;
167 
168 		err = addr_map_symbol__inc_samples(&bi->to, sample, evsel);
169 
170 	} else if (rep->mem_mode) {
171 		mi = he->mem_info;
172 		err = addr_map_symbol__inc_samples(&mi->daddr, sample, evsel);
173 		if (err)
174 			goto out;
175 
176 		err = hist_entry__inc_addr_samples(he, sample, evsel, al->addr);
177 
178 	} else if (symbol_conf.cumulate_callchain) {
179 		if (single)
180 			err = hist_entry__inc_addr_samples(he, sample, evsel, al->addr);
181 	} else {
182 		err = hist_entry__inc_addr_samples(he, sample, evsel, al->addr);
183 	}
184 
185 out:
186 	return err;
187 }
188 
189 static int hist_iter__branch_callback(struct hist_entry_iter *iter,
190 				      struct addr_location *al __maybe_unused,
191 				      bool single __maybe_unused,
192 				      void *arg)
193 {
194 	struct hist_entry *he = iter->he;
195 	struct report *rep = arg;
196 	struct branch_info *bi = he->branch_info;
197 	struct perf_sample *sample = iter->sample;
198 	struct evsel *evsel = iter->evsel;
199 	int err;
200 
201 	branch_type_count(&rep->brtype_stat, &bi->flags,
202 			  bi->from.addr, bi->to.addr);
203 
204 	if (!ui__has_annotation() && !rep->symbol_ipc)
205 		return 0;
206 
207 	err = addr_map_symbol__inc_samples(&bi->from, sample, evsel);
208 	if (err)
209 		goto out;
210 
211 	err = addr_map_symbol__inc_samples(&bi->to, sample, evsel);
212 
213 out:
214 	return err;
215 }
216 
217 static void setup_forced_leader(struct report *report,
218 				struct evlist *evlist)
219 {
220 	if (report->group_set)
221 		evlist__force_leader(evlist);
222 }
223 
224 static int process_feature_event(struct perf_session *session,
225 				 union perf_event *event)
226 {
227 	struct report *rep = container_of(session->tool, struct report, tool);
228 
229 	if (event->feat.feat_id < HEADER_LAST_FEATURE)
230 		return perf_event__process_feature(session, event);
231 
232 	if (event->feat.feat_id != HEADER_LAST_FEATURE) {
233 		pr_err("failed: wrong feature ID: %" PRI_lu64 "\n",
234 		       event->feat.feat_id);
235 		return -1;
236 	} else if (rep->header_only) {
237 		session_done = 1;
238 	}
239 
240 	/*
241 	 * (feat_id = HEADER_LAST_FEATURE) is the end marker which
242 	 * means all features are received, now we can force the
243 	 * group if needed.
244 	 */
245 	setup_forced_leader(rep, session->evlist);
246 	return 0;
247 }
248 
249 static int process_sample_event(struct perf_tool *tool,
250 				union perf_event *event,
251 				struct perf_sample *sample,
252 				struct evsel *evsel,
253 				struct machine *machine)
254 {
255 	struct report *rep = container_of(tool, struct report, tool);
256 	struct addr_location al;
257 	struct hist_entry_iter iter = {
258 		.evsel 			= evsel,
259 		.sample 		= sample,
260 		.hide_unresolved 	= symbol_conf.hide_unresolved,
261 		.add_entry_cb 		= hist_iter__report_callback,
262 	};
263 	int ret = 0;
264 
265 	if (perf_time__ranges_skip_sample(rep->ptime_range, rep->range_num,
266 					  sample->time)) {
267 		return 0;
268 	}
269 
270 	if (evswitch__discard(&rep->evswitch, evsel))
271 		return 0;
272 
273 	if (machine__resolve(machine, &al, sample) < 0) {
274 		pr_debug("problem processing %d event, skipping it.\n",
275 			 event->header.type);
276 		return -1;
277 	}
278 
279 	if (rep->stitch_lbr)
280 		al.thread->lbr_stitch_enable = true;
281 
282 	if (symbol_conf.hide_unresolved && al.sym == NULL)
283 		goto out_put;
284 
285 	if (rep->cpu_list && !test_bit(sample->cpu, rep->cpu_bitmap))
286 		goto out_put;
287 
288 	if (sort__mode == SORT_MODE__BRANCH) {
289 		/*
290 		 * A non-synthesized event might not have a branch stack if
291 		 * branch stacks have been synthesized (using itrace options).
292 		 */
293 		if (!sample->branch_stack)
294 			goto out_put;
295 
296 		iter.add_entry_cb = hist_iter__branch_callback;
297 		iter.ops = &hist_iter_branch;
298 	} else if (rep->mem_mode) {
299 		iter.ops = &hist_iter_mem;
300 	} else if (symbol_conf.cumulate_callchain) {
301 		iter.ops = &hist_iter_cumulative;
302 	} else {
303 		iter.ops = &hist_iter_normal;
304 	}
305 
306 	if (al.map != NULL)
307 		al.map->dso->hit = 1;
308 
309 	if (ui__has_annotation() || rep->symbol_ipc || rep->total_cycles_mode) {
310 		hist__account_cycles(sample->branch_stack, &al, sample,
311 				     rep->nonany_branch_mode,
312 				     &rep->total_cycles);
313 	}
314 
315 	ret = hist_entry_iter__add(&iter, &al, rep->max_stack, rep);
316 	if (ret < 0)
317 		pr_debug("problem adding hist entry, skipping event\n");
318 out_put:
319 	addr_location__put(&al);
320 	return ret;
321 }
322 
323 static int process_read_event(struct perf_tool *tool,
324 			      union perf_event *event,
325 			      struct perf_sample *sample __maybe_unused,
326 			      struct evsel *evsel,
327 			      struct machine *machine __maybe_unused)
328 {
329 	struct report *rep = container_of(tool, struct report, tool);
330 
331 	if (rep->show_threads) {
332 		const char *name = evsel__name(evsel);
333 		int err = perf_read_values_add_value(&rep->show_threads_values,
334 					   event->read.pid, event->read.tid,
335 					   evsel->core.idx,
336 					   name,
337 					   event->read.value);
338 
339 		if (err)
340 			return err;
341 	}
342 
343 	return 0;
344 }
345 
346 /* For pipe mode, sample_type is not currently set */
347 static int report__setup_sample_type(struct report *rep)
348 {
349 	struct perf_session *session = rep->session;
350 	u64 sample_type = evlist__combined_sample_type(session->evlist);
351 	bool is_pipe = perf_data__is_pipe(session->data);
352 
353 	if (session->itrace_synth_opts->callchain ||
354 	    session->itrace_synth_opts->add_callchain ||
355 	    (!is_pipe &&
356 	     perf_header__has_feat(&session->header, HEADER_AUXTRACE) &&
357 	     !session->itrace_synth_opts->set))
358 		sample_type |= PERF_SAMPLE_CALLCHAIN;
359 
360 	if (session->itrace_synth_opts->last_branch ||
361 	    session->itrace_synth_opts->add_last_branch)
362 		sample_type |= PERF_SAMPLE_BRANCH_STACK;
363 
364 	if (!is_pipe && !(sample_type & PERF_SAMPLE_CALLCHAIN)) {
365 		if (perf_hpp_list.parent) {
366 			ui__error("Selected --sort parent, but no "
367 				    "callchain data. Did you call "
368 				    "'perf record' without -g?\n");
369 			return -EINVAL;
370 		}
371 		if (symbol_conf.use_callchain &&
372 			!symbol_conf.show_branchflag_count) {
373 			ui__error("Selected -g or --branch-history.\n"
374 				  "But no callchain or branch data.\n"
375 				  "Did you call 'perf record' without -g or -b?\n");
376 			return -1;
377 		}
378 	} else if (!callchain_param.enabled &&
379 		   callchain_param.mode != CHAIN_NONE &&
380 		   !symbol_conf.use_callchain) {
381 			symbol_conf.use_callchain = true;
382 			if (callchain_register_param(&callchain_param) < 0) {
383 				ui__error("Can't register callchain params.\n");
384 				return -EINVAL;
385 			}
386 	}
387 
388 	if (symbol_conf.cumulate_callchain) {
389 		/* Silently ignore if callchain is missing */
390 		if (!(sample_type & PERF_SAMPLE_CALLCHAIN)) {
391 			symbol_conf.cumulate_callchain = false;
392 			perf_hpp__cancel_cumulate();
393 		}
394 	}
395 
396 	if (sort__mode == SORT_MODE__BRANCH) {
397 		if (!is_pipe &&
398 		    !(sample_type & PERF_SAMPLE_BRANCH_STACK)) {
399 			ui__error("Selected -b but no branch data. "
400 				  "Did you call perf record without -b?\n");
401 			return -1;
402 		}
403 	}
404 
405 	if (sort__mode == SORT_MODE__MEMORY) {
406 		if (!is_pipe && !(sample_type & PERF_SAMPLE_DATA_SRC)) {
407 			ui__error("Selected --mem-mode but no mem data. "
408 				  "Did you call perf record without -d?\n");
409 			return -1;
410 		}
411 	}
412 
413 	callchain_param_setup(sample_type, perf_env__arch(&rep->session->header.env));
414 
415 	if (rep->stitch_lbr && (callchain_param.record_mode != CALLCHAIN_LBR)) {
416 		ui__warning("Can't find LBR callchain. Switch off --stitch-lbr.\n"
417 			    "Please apply --call-graph lbr when recording.\n");
418 		rep->stitch_lbr = false;
419 	}
420 
421 	/* ??? handle more cases than just ANY? */
422 	if (!(evlist__combined_branch_type(session->evlist) & PERF_SAMPLE_BRANCH_ANY))
423 		rep->nonany_branch_mode = true;
424 
425 #if !defined(HAVE_LIBUNWIND_SUPPORT) && !defined(HAVE_DWARF_SUPPORT)
426 	if (dwarf_callchain_users) {
427 		ui__warning("Please install libunwind or libdw "
428 			    "development packages during the perf build.\n");
429 	}
430 #endif
431 
432 	return 0;
433 }
434 
435 static void sig_handler(int sig __maybe_unused)
436 {
437 	session_done = 1;
438 }
439 
440 static size_t hists__fprintf_nr_sample_events(struct hists *hists, struct report *rep,
441 					      const char *evname, FILE *fp)
442 {
443 	size_t ret;
444 	char unit;
445 	unsigned long nr_samples = hists->stats.nr_samples;
446 	u64 nr_events = hists->stats.total_period;
447 	struct evsel *evsel = hists_to_evsel(hists);
448 	char buf[512];
449 	size_t size = sizeof(buf);
450 	int socked_id = hists->socket_filter;
451 
452 	if (quiet)
453 		return 0;
454 
455 	if (symbol_conf.filter_relative) {
456 		nr_samples = hists->stats.nr_non_filtered_samples;
457 		nr_events = hists->stats.total_non_filtered_period;
458 	}
459 
460 	if (evsel__is_group_event(evsel)) {
461 		struct evsel *pos;
462 
463 		evsel__group_desc(evsel, buf, size);
464 		evname = buf;
465 
466 		for_each_group_member(pos, evsel) {
467 			const struct hists *pos_hists = evsel__hists(pos);
468 
469 			if (symbol_conf.filter_relative) {
470 				nr_samples += pos_hists->stats.nr_non_filtered_samples;
471 				nr_events += pos_hists->stats.total_non_filtered_period;
472 			} else {
473 				nr_samples += pos_hists->stats.nr_samples;
474 				nr_events += pos_hists->stats.total_period;
475 			}
476 		}
477 	}
478 
479 	nr_samples = convert_unit(nr_samples, &unit);
480 	ret = fprintf(fp, "# Samples: %lu%c", nr_samples, unit);
481 	if (evname != NULL) {
482 		ret += fprintf(fp, " of event%s '%s'",
483 			       evsel->core.nr_members > 1 ? "s" : "", evname);
484 	}
485 
486 	if (rep->time_str)
487 		ret += fprintf(fp, " (time slices: %s)", rep->time_str);
488 
489 	if (symbol_conf.show_ref_callgraph && evname && strstr(evname, "call-graph=no")) {
490 		ret += fprintf(fp, ", show reference callgraph");
491 	}
492 
493 	if (rep->mem_mode) {
494 		ret += fprintf(fp, "\n# Total weight : %" PRIu64, nr_events);
495 		ret += fprintf(fp, "\n# Sort order   : %s", sort_order ? : default_mem_sort_order);
496 	} else
497 		ret += fprintf(fp, "\n# Event count (approx.): %" PRIu64, nr_events);
498 
499 	if (socked_id > -1)
500 		ret += fprintf(fp, "\n# Processor Socket: %d", socked_id);
501 
502 	return ret + fprintf(fp, "\n#\n");
503 }
504 
505 static int evlist__tui_block_hists_browse(struct evlist *evlist, struct report *rep)
506 {
507 	struct evsel *pos;
508 	int i = 0, ret;
509 
510 	evlist__for_each_entry(evlist, pos) {
511 		ret = report__browse_block_hists(&rep->block_reports[i++].hist,
512 						 rep->min_percent, pos,
513 						 &rep->session->header.env,
514 						 &rep->annotation_opts);
515 		if (ret != 0)
516 			return ret;
517 	}
518 
519 	return 0;
520 }
521 
522 static int evlist__tty_browse_hists(struct evlist *evlist, struct report *rep, const char *help)
523 {
524 	struct evsel *pos;
525 	int i = 0;
526 
527 	if (!quiet) {
528 		fprintf(stdout, "#\n# Total Lost Samples: %" PRIu64 "\n#\n",
529 			evlist->stats.total_lost_samples);
530 	}
531 
532 	evlist__for_each_entry(evlist, pos) {
533 		struct hists *hists = evsel__hists(pos);
534 		const char *evname = evsel__name(pos);
535 
536 		if (symbol_conf.event_group && !evsel__is_group_leader(pos))
537 			continue;
538 
539 		if (rep->skip_empty && !hists->stats.nr_samples)
540 			continue;
541 
542 		hists__fprintf_nr_sample_events(hists, rep, evname, stdout);
543 
544 		if (rep->total_cycles_mode) {
545 			report__browse_block_hists(&rep->block_reports[i++].hist,
546 						   rep->min_percent, pos,
547 						   NULL, NULL);
548 			continue;
549 		}
550 
551 		hists__fprintf(hists, !quiet, 0, 0, rep->min_percent, stdout,
552 			       !(symbol_conf.use_callchain ||
553 			         symbol_conf.show_branchflag_count));
554 		fprintf(stdout, "\n\n");
555 	}
556 
557 	if (!quiet)
558 		fprintf(stdout, "#\n# (%s)\n#\n", help);
559 
560 	if (rep->show_threads) {
561 		bool style = !strcmp(rep->pretty_printing_style, "raw");
562 		perf_read_values_display(stdout, &rep->show_threads_values,
563 					 style);
564 		perf_read_values_destroy(&rep->show_threads_values);
565 	}
566 
567 	if (sort__mode == SORT_MODE__BRANCH)
568 		branch_type_stat_display(stdout, &rep->brtype_stat);
569 
570 	return 0;
571 }
572 
573 static void report__warn_kptr_restrict(const struct report *rep)
574 {
575 	struct map *kernel_map = machine__kernel_map(&rep->session->machines.host);
576 	struct kmap *kernel_kmap = kernel_map ? map__kmap(kernel_map) : NULL;
577 
578 	if (evlist__exclude_kernel(rep->session->evlist))
579 		return;
580 
581 	if (kernel_map == NULL ||
582 	    (kernel_map->dso->hit &&
583 	     (kernel_kmap->ref_reloc_sym == NULL ||
584 	      kernel_kmap->ref_reloc_sym->addr == 0))) {
585 		const char *desc =
586 		    "As no suitable kallsyms nor vmlinux was found, kernel samples\n"
587 		    "can't be resolved.";
588 
589 		if (kernel_map && map__has_symbols(kernel_map)) {
590 			desc = "If some relocation was applied (e.g. "
591 			       "kexec) symbols may be misresolved.";
592 		}
593 
594 		ui__warning(
595 "Kernel address maps (/proc/{kallsyms,modules}) were restricted.\n\n"
596 "Check /proc/sys/kernel/kptr_restrict before running 'perf record'.\n\n%s\n\n"
597 "Samples in kernel modules can't be resolved as well.\n\n",
598 		desc);
599 	}
600 }
601 
602 static int report__gtk_browse_hists(struct report *rep, const char *help)
603 {
604 	int (*hist_browser)(struct evlist *evlist, const char *help,
605 			    struct hist_browser_timer *timer, float min_pcnt);
606 
607 	hist_browser = dlsym(perf_gtk_handle, "evlist__gtk_browse_hists");
608 
609 	if (hist_browser == NULL) {
610 		ui__error("GTK browser not found!\n");
611 		return -1;
612 	}
613 
614 	return hist_browser(rep->session->evlist, help, NULL, rep->min_percent);
615 }
616 
617 static int report__browse_hists(struct report *rep)
618 {
619 	int ret;
620 	struct perf_session *session = rep->session;
621 	struct evlist *evlist = session->evlist;
622 	char *help = NULL, *path = NULL;
623 
624 	path = system_path(TIPDIR);
625 	if (perf_tip(&help, path) || help == NULL) {
626 		/* fallback for people who don't install perf ;-) */
627 		free(path);
628 		path = system_path(DOCDIR);
629 		if (perf_tip(&help, path) || help == NULL)
630 			help = strdup("Cannot load tips.txt file, please install perf!");
631 	}
632 	free(path);
633 
634 	switch (use_browser) {
635 	case 1:
636 		if (rep->total_cycles_mode) {
637 			ret = evlist__tui_block_hists_browse(evlist, rep);
638 			break;
639 		}
640 
641 		ret = evlist__tui_browse_hists(evlist, help, NULL, rep->min_percent,
642 					       &session->header.env, true, &rep->annotation_opts);
643 		/*
644 		 * Usually "ret" is the last pressed key, and we only
645 		 * care if the key notifies us to switch data file.
646 		 */
647 		if (ret != K_SWITCH_INPUT_DATA && ret != K_RELOAD)
648 			ret = 0;
649 		break;
650 	case 2:
651 		ret = report__gtk_browse_hists(rep, help);
652 		break;
653 	default:
654 		ret = evlist__tty_browse_hists(evlist, rep, help);
655 		break;
656 	}
657 	free(help);
658 	return ret;
659 }
660 
661 static int report__collapse_hists(struct report *rep)
662 {
663 	struct ui_progress prog;
664 	struct evsel *pos;
665 	int ret = 0;
666 
667 	ui_progress__init(&prog, rep->nr_entries, "Merging related events...");
668 
669 	evlist__for_each_entry(rep->session->evlist, pos) {
670 		struct hists *hists = evsel__hists(pos);
671 
672 		if (pos->core.idx == 0)
673 			hists->symbol_filter_str = rep->symbol_filter_str;
674 
675 		hists->socket_filter = rep->socket_filter;
676 
677 		ret = hists__collapse_resort(hists, &prog);
678 		if (ret < 0)
679 			break;
680 
681 		/* Non-group events are considered as leader */
682 		if (symbol_conf.event_group && !evsel__is_group_leader(pos)) {
683 			struct hists *leader_hists = evsel__hists(evsel__leader(pos));
684 
685 			hists__match(leader_hists, hists);
686 			hists__link(leader_hists, hists);
687 		}
688 	}
689 
690 	ui_progress__finish();
691 	return ret;
692 }
693 
694 static int hists__resort_cb(struct hist_entry *he, void *arg)
695 {
696 	struct report *rep = arg;
697 	struct symbol *sym = he->ms.sym;
698 
699 	if (rep->symbol_ipc && sym && !sym->annotate2) {
700 		struct evsel *evsel = hists_to_evsel(he->hists);
701 
702 		symbol__annotate2(&he->ms, evsel,
703 				  &annotation__default_options, NULL);
704 	}
705 
706 	return 0;
707 }
708 
709 static void report__output_resort(struct report *rep)
710 {
711 	struct ui_progress prog;
712 	struct evsel *pos;
713 
714 	ui_progress__init(&prog, rep->nr_entries, "Sorting events for output...");
715 
716 	evlist__for_each_entry(rep->session->evlist, pos) {
717 		evsel__output_resort_cb(pos, &prog, hists__resort_cb, rep);
718 	}
719 
720 	ui_progress__finish();
721 }
722 
723 static int count_sample_event(struct perf_tool *tool __maybe_unused,
724 			      union perf_event *event __maybe_unused,
725 			      struct perf_sample *sample __maybe_unused,
726 			      struct evsel *evsel,
727 			      struct machine *machine __maybe_unused)
728 {
729 	struct hists *hists = evsel__hists(evsel);
730 
731 	hists__inc_nr_events(hists);
732 	return 0;
733 }
734 
735 static int process_attr(struct perf_tool *tool __maybe_unused,
736 			union perf_event *event,
737 			struct evlist **pevlist);
738 
739 static void stats_setup(struct report *rep)
740 {
741 	memset(&rep->tool, 0, sizeof(rep->tool));
742 	rep->tool.attr = process_attr;
743 	rep->tool.sample = count_sample_event;
744 	rep->tool.no_warn = true;
745 }
746 
747 static int stats_print(struct report *rep)
748 {
749 	struct perf_session *session = rep->session;
750 
751 	perf_session__fprintf_nr_events(session, stdout, rep->skip_empty);
752 	evlist__fprintf_nr_events(session->evlist, stdout, rep->skip_empty);
753 	return 0;
754 }
755 
756 static void tasks_setup(struct report *rep)
757 {
758 	memset(&rep->tool, 0, sizeof(rep->tool));
759 	rep->tool.ordered_events = true;
760 	if (rep->mmaps_mode) {
761 		rep->tool.mmap = perf_event__process_mmap;
762 		rep->tool.mmap2 = perf_event__process_mmap2;
763 	}
764 	rep->tool.attr = process_attr;
765 	rep->tool.comm = perf_event__process_comm;
766 	rep->tool.exit = perf_event__process_exit;
767 	rep->tool.fork = perf_event__process_fork;
768 	rep->tool.no_warn = true;
769 }
770 
771 struct task {
772 	struct thread		*thread;
773 	struct list_head	 list;
774 	struct list_head	 children;
775 };
776 
777 static struct task *tasks_list(struct task *task, struct machine *machine)
778 {
779 	struct thread *parent_thread, *thread = task->thread;
780 	struct task   *parent_task;
781 
782 	/* Already listed. */
783 	if (!list_empty(&task->list))
784 		return NULL;
785 
786 	/* Last one in the chain. */
787 	if (thread->ppid == -1)
788 		return task;
789 
790 	parent_thread = machine__find_thread(machine, -1, thread->ppid);
791 	if (!parent_thread)
792 		return ERR_PTR(-ENOENT);
793 
794 	parent_task = thread__priv(parent_thread);
795 	list_add_tail(&task->list, &parent_task->children);
796 	return tasks_list(parent_task, machine);
797 }
798 
799 static size_t maps__fprintf_task(struct maps *maps, int indent, FILE *fp)
800 {
801 	size_t printed = 0;
802 	struct map *map;
803 
804 	maps__for_each_entry(maps, map) {
805 		printed += fprintf(fp, "%*s  %" PRIx64 "-%" PRIx64 " %c%c%c%c %08" PRIx64 " %" PRIu64 " %s\n",
806 				   indent, "", map->start, map->end,
807 				   map->prot & PROT_READ ? 'r' : '-',
808 				   map->prot & PROT_WRITE ? 'w' : '-',
809 				   map->prot & PROT_EXEC ? 'x' : '-',
810 				   map->flags & MAP_SHARED ? 's' : 'p',
811 				   map->pgoff,
812 				   map->dso->id.ino, map->dso->name);
813 	}
814 
815 	return printed;
816 }
817 
818 static void task__print_level(struct task *task, FILE *fp, int level)
819 {
820 	struct thread *thread = task->thread;
821 	struct task *child;
822 	int comm_indent = fprintf(fp, "  %8d %8d %8d |%*s",
823 				  thread->pid_, thread->tid, thread->ppid,
824 				  level, "");
825 
826 	fprintf(fp, "%s\n", thread__comm_str(thread));
827 
828 	maps__fprintf_task(thread->maps, comm_indent, fp);
829 
830 	if (!list_empty(&task->children)) {
831 		list_for_each_entry(child, &task->children, list)
832 			task__print_level(child, fp, level + 1);
833 	}
834 }
835 
836 static int tasks_print(struct report *rep, FILE *fp)
837 {
838 	struct perf_session *session = rep->session;
839 	struct machine      *machine = &session->machines.host;
840 	struct task *tasks, *task;
841 	unsigned int nr = 0, itask = 0, i;
842 	struct rb_node *nd;
843 	LIST_HEAD(list);
844 
845 	/*
846 	 * No locking needed while accessing machine->threads,
847 	 * because --tasks is single threaded command.
848 	 */
849 
850 	/* Count all the threads. */
851 	for (i = 0; i < THREADS__TABLE_SIZE; i++)
852 		nr += machine->threads[i].nr;
853 
854 	tasks = malloc(sizeof(*tasks) * nr);
855 	if (!tasks)
856 		return -ENOMEM;
857 
858 	for (i = 0; i < THREADS__TABLE_SIZE; i++) {
859 		struct threads *threads = &machine->threads[i];
860 
861 		for (nd = rb_first_cached(&threads->entries); nd;
862 		     nd = rb_next(nd)) {
863 			task = tasks + itask++;
864 
865 			task->thread = rb_entry(nd, struct thread, rb_node);
866 			INIT_LIST_HEAD(&task->children);
867 			INIT_LIST_HEAD(&task->list);
868 			thread__set_priv(task->thread, task);
869 		}
870 	}
871 
872 	/*
873 	 * Iterate every task down to the unprocessed parent
874 	 * and link all in task children list. Task with no
875 	 * parent is added into 'list'.
876 	 */
877 	for (itask = 0; itask < nr; itask++) {
878 		task = tasks + itask;
879 
880 		if (!list_empty(&task->list))
881 			continue;
882 
883 		task = tasks_list(task, machine);
884 		if (IS_ERR(task)) {
885 			pr_err("Error: failed to process tasks\n");
886 			free(tasks);
887 			return PTR_ERR(task);
888 		}
889 
890 		if (task)
891 			list_add_tail(&task->list, &list);
892 	}
893 
894 	fprintf(fp, "# %8s %8s %8s  %s\n", "pid", "tid", "ppid", "comm");
895 
896 	list_for_each_entry(task, &list, list)
897 		task__print_level(task, fp, 0);
898 
899 	free(tasks);
900 	return 0;
901 }
902 
903 static int __cmd_report(struct report *rep)
904 {
905 	int ret;
906 	struct perf_session *session = rep->session;
907 	struct evsel *pos;
908 	struct perf_data *data = session->data;
909 
910 	signal(SIGINT, sig_handler);
911 
912 	if (rep->cpu_list) {
913 		ret = perf_session__cpu_bitmap(session, rep->cpu_list,
914 					       rep->cpu_bitmap);
915 		if (ret) {
916 			ui__error("failed to set cpu bitmap\n");
917 			return ret;
918 		}
919 		session->itrace_synth_opts->cpu_bitmap = rep->cpu_bitmap;
920 	}
921 
922 	if (rep->show_threads) {
923 		ret = perf_read_values_init(&rep->show_threads_values);
924 		if (ret)
925 			return ret;
926 	}
927 
928 	ret = report__setup_sample_type(rep);
929 	if (ret) {
930 		/* report__setup_sample_type() already showed error message */
931 		return ret;
932 	}
933 
934 	if (rep->stats_mode)
935 		stats_setup(rep);
936 
937 	if (rep->tasks_mode)
938 		tasks_setup(rep);
939 
940 	ret = perf_session__process_events(session);
941 	if (ret) {
942 		ui__error("failed to process sample\n");
943 		return ret;
944 	}
945 
946 	evlist__check_mem_load_aux(session->evlist);
947 
948 	if (rep->stats_mode)
949 		return stats_print(rep);
950 
951 	if (rep->tasks_mode)
952 		return tasks_print(rep, stdout);
953 
954 	report__warn_kptr_restrict(rep);
955 
956 	evlist__for_each_entry(session->evlist, pos)
957 		rep->nr_entries += evsel__hists(pos)->nr_entries;
958 
959 	if (use_browser == 0) {
960 		if (verbose > 3)
961 			perf_session__fprintf(session, stdout);
962 
963 		if (verbose > 2)
964 			perf_session__fprintf_dsos(session, stdout);
965 
966 		if (dump_trace) {
967 			perf_session__fprintf_nr_events(session, stdout,
968 							rep->skip_empty);
969 			evlist__fprintf_nr_events(session->evlist, stdout,
970 						  rep->skip_empty);
971 			return 0;
972 		}
973 	}
974 
975 	ret = report__collapse_hists(rep);
976 	if (ret) {
977 		ui__error("failed to process hist entry\n");
978 		return ret;
979 	}
980 
981 	if (session_done())
982 		return 0;
983 
984 	/*
985 	 * recalculate number of entries after collapsing since it
986 	 * might be changed during the collapse phase.
987 	 */
988 	rep->nr_entries = 0;
989 	evlist__for_each_entry(session->evlist, pos)
990 		rep->nr_entries += evsel__hists(pos)->nr_entries;
991 
992 	if (rep->nr_entries == 0) {
993 		ui__error("The %s data has no samples!\n", data->path);
994 		return 0;
995 	}
996 
997 	report__output_resort(rep);
998 
999 	if (rep->total_cycles_mode) {
1000 		int block_hpps[6] = {
1001 			PERF_HPP_REPORT__BLOCK_TOTAL_CYCLES_PCT,
1002 			PERF_HPP_REPORT__BLOCK_LBR_CYCLES,
1003 			PERF_HPP_REPORT__BLOCK_CYCLES_PCT,
1004 			PERF_HPP_REPORT__BLOCK_AVG_CYCLES,
1005 			PERF_HPP_REPORT__BLOCK_RANGE,
1006 			PERF_HPP_REPORT__BLOCK_DSO,
1007 		};
1008 
1009 		rep->block_reports = block_info__create_report(session->evlist,
1010 							       rep->total_cycles,
1011 							       block_hpps, 6,
1012 							       &rep->nr_block_reports);
1013 		if (!rep->block_reports)
1014 			return -1;
1015 	}
1016 
1017 	return report__browse_hists(rep);
1018 }
1019 
1020 static int
1021 report_parse_callchain_opt(const struct option *opt, const char *arg, int unset)
1022 {
1023 	struct callchain_param *callchain = opt->value;
1024 
1025 	callchain->enabled = !unset;
1026 	/*
1027 	 * --no-call-graph
1028 	 */
1029 	if (unset) {
1030 		symbol_conf.use_callchain = false;
1031 		callchain->mode = CHAIN_NONE;
1032 		return 0;
1033 	}
1034 
1035 	return parse_callchain_report_opt(arg);
1036 }
1037 
1038 static int
1039 parse_time_quantum(const struct option *opt, const char *arg,
1040 		   int unset __maybe_unused)
1041 {
1042 	unsigned long *time_q = opt->value;
1043 	char *end;
1044 
1045 	*time_q = strtoul(arg, &end, 0);
1046 	if (end == arg)
1047 		goto parse_err;
1048 	if (*time_q == 0) {
1049 		pr_err("time quantum cannot be 0");
1050 		return -1;
1051 	}
1052 	end = skip_spaces(end);
1053 	if (*end == 0)
1054 		return 0;
1055 	if (!strcmp(end, "s")) {
1056 		*time_q *= NSEC_PER_SEC;
1057 		return 0;
1058 	}
1059 	if (!strcmp(end, "ms")) {
1060 		*time_q *= NSEC_PER_MSEC;
1061 		return 0;
1062 	}
1063 	if (!strcmp(end, "us")) {
1064 		*time_q *= NSEC_PER_USEC;
1065 		return 0;
1066 	}
1067 	if (!strcmp(end, "ns"))
1068 		return 0;
1069 parse_err:
1070 	pr_err("Cannot parse time quantum `%s'\n", arg);
1071 	return -1;
1072 }
1073 
1074 int
1075 report_parse_ignore_callees_opt(const struct option *opt __maybe_unused,
1076 				const char *arg, int unset __maybe_unused)
1077 {
1078 	if (arg) {
1079 		int err = regcomp(&ignore_callees_regex, arg, REG_EXTENDED);
1080 		if (err) {
1081 			char buf[BUFSIZ];
1082 			regerror(err, &ignore_callees_regex, buf, sizeof(buf));
1083 			pr_err("Invalid --ignore-callees regex: %s\n%s", arg, buf);
1084 			return -1;
1085 		}
1086 		have_ignore_callees = 1;
1087 	}
1088 
1089 	return 0;
1090 }
1091 
1092 static int
1093 parse_branch_mode(const struct option *opt,
1094 		  const char *str __maybe_unused, int unset)
1095 {
1096 	int *branch_mode = opt->value;
1097 
1098 	*branch_mode = !unset;
1099 	return 0;
1100 }
1101 
1102 static int
1103 parse_percent_limit(const struct option *opt, const char *str,
1104 		    int unset __maybe_unused)
1105 {
1106 	struct report *rep = opt->value;
1107 	double pcnt = strtof(str, NULL);
1108 
1109 	rep->min_percent = pcnt;
1110 	callchain_param.min_percent = pcnt;
1111 	return 0;
1112 }
1113 
1114 static int process_attr(struct perf_tool *tool __maybe_unused,
1115 			union perf_event *event,
1116 			struct evlist **pevlist)
1117 {
1118 	u64 sample_type;
1119 	int err;
1120 
1121 	err = perf_event__process_attr(tool, event, pevlist);
1122 	if (err)
1123 		return err;
1124 
1125 	/*
1126 	 * Check if we need to enable callchains based
1127 	 * on events sample_type.
1128 	 */
1129 	sample_type = evlist__combined_sample_type(*pevlist);
1130 	callchain_param_setup(sample_type, perf_env__arch((*pevlist)->env));
1131 	return 0;
1132 }
1133 
1134 int cmd_report(int argc, const char **argv)
1135 {
1136 	struct perf_session *session;
1137 	struct itrace_synth_opts itrace_synth_opts = { .set = 0, };
1138 	struct stat st;
1139 	bool has_br_stack = false;
1140 	int branch_mode = -1;
1141 	int last_key = 0;
1142 	bool branch_call_mode = false;
1143 #define CALLCHAIN_DEFAULT_OPT  "graph,0.5,caller,function,percent"
1144 	static const char report_callchain_help[] = "Display call graph (stack chain/backtrace):\n\n"
1145 						    CALLCHAIN_REPORT_HELP
1146 						    "\n\t\t\t\tDefault: " CALLCHAIN_DEFAULT_OPT;
1147 	char callchain_default_opt[] = CALLCHAIN_DEFAULT_OPT;
1148 	const char * const report_usage[] = {
1149 		"perf report [<options>]",
1150 		NULL
1151 	};
1152 	struct report report = {
1153 		.tool = {
1154 			.sample		 = process_sample_event,
1155 			.mmap		 = perf_event__process_mmap,
1156 			.mmap2		 = perf_event__process_mmap2,
1157 			.comm		 = perf_event__process_comm,
1158 			.namespaces	 = perf_event__process_namespaces,
1159 			.cgroup		 = perf_event__process_cgroup,
1160 			.exit		 = perf_event__process_exit,
1161 			.fork		 = perf_event__process_fork,
1162 			.lost		 = perf_event__process_lost,
1163 			.read		 = process_read_event,
1164 			.attr		 = process_attr,
1165 			.tracing_data	 = perf_event__process_tracing_data,
1166 			.build_id	 = perf_event__process_build_id,
1167 			.id_index	 = perf_event__process_id_index,
1168 			.auxtrace_info	 = perf_event__process_auxtrace_info,
1169 			.auxtrace	 = perf_event__process_auxtrace,
1170 			.event_update	 = perf_event__process_event_update,
1171 			.feature	 = process_feature_event,
1172 			.ordered_events	 = true,
1173 			.ordering_requires_timestamps = true,
1174 		},
1175 		.max_stack		 = PERF_MAX_STACK_DEPTH,
1176 		.pretty_printing_style	 = "normal",
1177 		.socket_filter		 = -1,
1178 		.annotation_opts	 = annotation__default_options,
1179 		.skip_empty		 = true,
1180 	};
1181 	char *sort_order_help = sort_help("sort by key(s):");
1182 	char *field_order_help = sort_help("output field(s): overhead period sample ");
1183 	const struct option options[] = {
1184 	OPT_STRING('i', "input", &input_name, "file",
1185 		    "input file name"),
1186 	OPT_INCR('v', "verbose", &verbose,
1187 		    "be more verbose (show symbol address, etc)"),
1188 	OPT_BOOLEAN('q', "quiet", &quiet, "Do not show any message"),
1189 	OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace,
1190 		    "dump raw trace in ASCII"),
1191 	OPT_BOOLEAN(0, "stats", &report.stats_mode, "Display event stats"),
1192 	OPT_BOOLEAN(0, "tasks", &report.tasks_mode, "Display recorded tasks"),
1193 	OPT_BOOLEAN(0, "mmaps", &report.mmaps_mode, "Display recorded tasks memory maps"),
1194 	OPT_STRING('k', "vmlinux", &symbol_conf.vmlinux_name,
1195 		   "file", "vmlinux pathname"),
1196 	OPT_BOOLEAN(0, "ignore-vmlinux", &symbol_conf.ignore_vmlinux,
1197                     "don't load vmlinux even if found"),
1198 	OPT_STRING(0, "kallsyms", &symbol_conf.kallsyms_name,
1199 		   "file", "kallsyms pathname"),
1200 	OPT_BOOLEAN('f', "force", &symbol_conf.force, "don't complain, do it"),
1201 	OPT_BOOLEAN('m', "modules", &symbol_conf.use_modules,
1202 		    "load module symbols - WARNING: use only with -k and LIVE kernel"),
1203 	OPT_BOOLEAN('n', "show-nr-samples", &symbol_conf.show_nr_samples,
1204 		    "Show a column with the number of samples"),
1205 	OPT_BOOLEAN('T', "threads", &report.show_threads,
1206 		    "Show per-thread event counters"),
1207 	OPT_STRING(0, "pretty", &report.pretty_printing_style, "key",
1208 		   "pretty printing style key: normal raw"),
1209 	OPT_BOOLEAN(0, "tui", &report.use_tui, "Use the TUI interface"),
1210 	OPT_BOOLEAN(0, "gtk", &report.use_gtk, "Use the GTK2 interface"),
1211 	OPT_BOOLEAN(0, "stdio", &report.use_stdio,
1212 		    "Use the stdio interface"),
1213 	OPT_BOOLEAN(0, "header", &report.header, "Show data header."),
1214 	OPT_BOOLEAN(0, "header-only", &report.header_only,
1215 		    "Show only data header."),
1216 	OPT_STRING('s', "sort", &sort_order, "key[,key2...]",
1217 		   sort_order_help),
1218 	OPT_STRING('F', "fields", &field_order, "key[,keys...]",
1219 		   field_order_help),
1220 	OPT_BOOLEAN(0, "show-cpu-utilization", &symbol_conf.show_cpu_utilization,
1221 		    "Show sample percentage for different cpu modes"),
1222 	OPT_BOOLEAN_FLAG(0, "showcpuutilization", &symbol_conf.show_cpu_utilization,
1223 		    "Show sample percentage for different cpu modes", PARSE_OPT_HIDDEN),
1224 	OPT_STRING('p', "parent", &parent_pattern, "regex",
1225 		   "regex filter to identify parent, see: '--sort parent'"),
1226 	OPT_BOOLEAN('x', "exclude-other", &symbol_conf.exclude_other,
1227 		    "Only display entries with parent-match"),
1228 	OPT_CALLBACK_DEFAULT('g', "call-graph", &callchain_param,
1229 			     "print_type,threshold[,print_limit],order,sort_key[,branch],value",
1230 			     report_callchain_help, &report_parse_callchain_opt,
1231 			     callchain_default_opt),
1232 	OPT_BOOLEAN(0, "children", &symbol_conf.cumulate_callchain,
1233 		    "Accumulate callchains of children and show total overhead as well. "
1234 		    "Enabled by default, use --no-children to disable."),
1235 	OPT_INTEGER(0, "max-stack", &report.max_stack,
1236 		    "Set the maximum stack depth when parsing the callchain, "
1237 		    "anything beyond the specified depth will be ignored. "
1238 		    "Default: kernel.perf_event_max_stack or " __stringify(PERF_MAX_STACK_DEPTH)),
1239 	OPT_BOOLEAN('G', "inverted", &report.inverted_callchain,
1240 		    "alias for inverted call graph"),
1241 	OPT_CALLBACK(0, "ignore-callees", NULL, "regex",
1242 		   "ignore callees of these functions in call graphs",
1243 		   report_parse_ignore_callees_opt),
1244 	OPT_STRING('d', "dsos", &symbol_conf.dso_list_str, "dso[,dso...]",
1245 		   "only consider symbols in these dsos"),
1246 	OPT_STRING('c', "comms", &symbol_conf.comm_list_str, "comm[,comm...]",
1247 		   "only consider symbols in these comms"),
1248 	OPT_STRING(0, "pid", &symbol_conf.pid_list_str, "pid[,pid...]",
1249 		   "only consider symbols in these pids"),
1250 	OPT_STRING(0, "tid", &symbol_conf.tid_list_str, "tid[,tid...]",
1251 		   "only consider symbols in these tids"),
1252 	OPT_STRING('S', "symbols", &symbol_conf.sym_list_str, "symbol[,symbol...]",
1253 		   "only consider these symbols"),
1254 	OPT_STRING(0, "symbol-filter", &report.symbol_filter_str, "filter",
1255 		   "only show symbols that (partially) match with this filter"),
1256 	OPT_STRING('w', "column-widths", &symbol_conf.col_width_list_str,
1257 		   "width[,width...]",
1258 		   "don't try to adjust column width, use these fixed values"),
1259 	OPT_STRING_NOEMPTY('t', "field-separator", &symbol_conf.field_sep, "separator",
1260 		   "separator for columns, no spaces will be added between "
1261 		   "columns '.' is reserved."),
1262 	OPT_BOOLEAN('U', "hide-unresolved", &symbol_conf.hide_unresolved,
1263 		    "Only display entries resolved to a symbol"),
1264 	OPT_CALLBACK(0, "symfs", NULL, "directory",
1265 		     "Look for files with symbols relative to this directory",
1266 		     symbol__config_symfs),
1267 	OPT_STRING('C', "cpu", &report.cpu_list, "cpu",
1268 		   "list of cpus to profile"),
1269 	OPT_BOOLEAN('I', "show-info", &report.show_full_info,
1270 		    "Display extended information about perf.data file"),
1271 	OPT_BOOLEAN(0, "source", &report.annotation_opts.annotate_src,
1272 		    "Interleave source code with assembly code (default)"),
1273 	OPT_BOOLEAN(0, "asm-raw", &report.annotation_opts.show_asm_raw,
1274 		    "Display raw encoding of assembly instructions (default)"),
1275 	OPT_STRING('M', "disassembler-style", &report.annotation_opts.disassembler_style, "disassembler style",
1276 		   "Specify disassembler style (e.g. -M intel for intel syntax)"),
1277 	OPT_STRING(0, "prefix", &report.annotation_opts.prefix, "prefix",
1278 		    "Add prefix to source file path names in programs (with --prefix-strip)"),
1279 	OPT_STRING(0, "prefix-strip", &report.annotation_opts.prefix_strip, "N",
1280 		    "Strip first N entries of source file path name in programs (with --prefix)"),
1281 	OPT_BOOLEAN(0, "show-total-period", &symbol_conf.show_total_period,
1282 		    "Show a column with the sum of periods"),
1283 	OPT_BOOLEAN_SET(0, "group", &symbol_conf.event_group, &report.group_set,
1284 		    "Show event group information together"),
1285 	OPT_INTEGER(0, "group-sort-idx", &symbol_conf.group_sort_idx,
1286 		    "Sort the output by the event at the index n in group. "
1287 		    "If n is invalid, sort by the first event. "
1288 		    "WARNING: should be used on grouped events."),
1289 	OPT_CALLBACK_NOOPT('b', "branch-stack", &branch_mode, "",
1290 		    "use branch records for per branch histogram filling",
1291 		    parse_branch_mode),
1292 	OPT_BOOLEAN(0, "branch-history", &branch_call_mode,
1293 		    "add last branch records to call history"),
1294 	OPT_STRING(0, "objdump", &report.annotation_opts.objdump_path, "path",
1295 		   "objdump binary to use for disassembly and annotations"),
1296 	OPT_BOOLEAN(0, "demangle", &symbol_conf.demangle,
1297 		    "Disable symbol demangling"),
1298 	OPT_BOOLEAN(0, "demangle-kernel", &symbol_conf.demangle_kernel,
1299 		    "Enable kernel symbol demangling"),
1300 	OPT_BOOLEAN(0, "mem-mode", &report.mem_mode, "mem access profile"),
1301 	OPT_INTEGER(0, "samples", &symbol_conf.res_sample,
1302 		    "Number of samples to save per histogram entry for individual browsing"),
1303 	OPT_CALLBACK(0, "percent-limit", &report, "percent",
1304 		     "Don't show entries under that percent", parse_percent_limit),
1305 	OPT_CALLBACK(0, "percentage", NULL, "relative|absolute",
1306 		     "how to display percentage of filtered entries", parse_filter_percentage),
1307 	OPT_CALLBACK_OPTARG(0, "itrace", &itrace_synth_opts, NULL, "opts",
1308 			    "Instruction Tracing options\n" ITRACE_HELP,
1309 			    itrace_parse_synth_opts),
1310 	OPT_BOOLEAN(0, "full-source-path", &srcline_full_filename,
1311 			"Show full source file name path for source lines"),
1312 	OPT_BOOLEAN(0, "show-ref-call-graph", &symbol_conf.show_ref_callgraph,
1313 		    "Show callgraph from reference event"),
1314 	OPT_BOOLEAN(0, "stitch-lbr", &report.stitch_lbr,
1315 		    "Enable LBR callgraph stitching approach"),
1316 	OPT_INTEGER(0, "socket-filter", &report.socket_filter,
1317 		    "only show processor socket that match with this filter"),
1318 	OPT_BOOLEAN(0, "raw-trace", &symbol_conf.raw_trace,
1319 		    "Show raw trace event output (do not use print fmt or plugins)"),
1320 	OPT_BOOLEAN(0, "hierarchy", &symbol_conf.report_hierarchy,
1321 		    "Show entries in a hierarchy"),
1322 	OPT_CALLBACK_DEFAULT(0, "stdio-color", NULL, "mode",
1323 			     "'always' (default), 'never' or 'auto' only applicable to --stdio mode",
1324 			     stdio__config_color, "always"),
1325 	OPT_STRING(0, "time", &report.time_str, "str",
1326 		   "Time span of interest (start,stop)"),
1327 	OPT_BOOLEAN(0, "inline", &symbol_conf.inline_name,
1328 		    "Show inline function"),
1329 	OPT_CALLBACK(0, "percent-type", &report.annotation_opts, "local-period",
1330 		     "Set percent type local/global-period/hits",
1331 		     annotate_parse_percent_type),
1332 	OPT_BOOLEAN(0, "ns", &symbol_conf.nanosecs, "Show times in nanosecs"),
1333 	OPT_CALLBACK(0, "time-quantum", &symbol_conf.time_quantum, "time (ms|us|ns|s)",
1334 		     "Set time quantum for time sort key (default 100ms)",
1335 		     parse_time_quantum),
1336 	OPTS_EVSWITCH(&report.evswitch),
1337 	OPT_BOOLEAN(0, "total-cycles", &report.total_cycles_mode,
1338 		    "Sort all blocks by 'Sampled Cycles%'"),
1339 	OPT_BOOLEAN(0, "disable-order", &report.disable_order,
1340 		    "Disable raw trace ordering"),
1341 	OPT_BOOLEAN(0, "skip-empty", &report.skip_empty,
1342 		    "Do not display empty (or dummy) events in the output"),
1343 	OPT_END()
1344 	};
1345 	struct perf_data data = {
1346 		.mode  = PERF_DATA_MODE_READ,
1347 	};
1348 	int ret = hists__init();
1349 	char sort_tmp[128];
1350 
1351 	if (ret < 0)
1352 		goto exit;
1353 
1354 	ret = perf_config(report__config, &report);
1355 	if (ret)
1356 		goto exit;
1357 
1358 	argc = parse_options(argc, argv, options, report_usage, 0);
1359 	if (argc) {
1360 		/*
1361 		 * Special case: if there's an argument left then assume that
1362 		 * it's a symbol filter:
1363 		 */
1364 		if (argc > 1)
1365 			usage_with_options(report_usage, options);
1366 
1367 		report.symbol_filter_str = argv[0];
1368 	}
1369 
1370 	if (annotate_check_args(&report.annotation_opts) < 0) {
1371 		ret = -EINVAL;
1372 		goto exit;
1373 	}
1374 
1375 	if (report.mmaps_mode)
1376 		report.tasks_mode = true;
1377 
1378 	if (dump_trace && report.disable_order)
1379 		report.tool.ordered_events = false;
1380 
1381 	if (quiet)
1382 		perf_quiet_option();
1383 
1384 	ret = symbol__validate_sym_arguments();
1385 	if (ret)
1386 		goto exit;
1387 
1388 	if (report.inverted_callchain)
1389 		callchain_param.order = ORDER_CALLER;
1390 	if (symbol_conf.cumulate_callchain && !callchain_param.order_set)
1391 		callchain_param.order = ORDER_CALLER;
1392 
1393 	if ((itrace_synth_opts.callchain || itrace_synth_opts.add_callchain) &&
1394 	    (int)itrace_synth_opts.callchain_sz > report.max_stack)
1395 		report.max_stack = itrace_synth_opts.callchain_sz;
1396 
1397 	if (!input_name || !strlen(input_name)) {
1398 		if (!fstat(STDIN_FILENO, &st) && S_ISFIFO(st.st_mode))
1399 			input_name = "-";
1400 		else
1401 			input_name = "perf.data";
1402 	}
1403 
1404 	data.path  = input_name;
1405 	data.force = symbol_conf.force;
1406 
1407 repeat:
1408 	session = perf_session__new(&data, &report.tool);
1409 	if (IS_ERR(session)) {
1410 		ret = PTR_ERR(session);
1411 		goto exit;
1412 	}
1413 
1414 	ret = evswitch__init(&report.evswitch, session->evlist, stderr);
1415 	if (ret)
1416 		goto exit;
1417 
1418 	if (zstd_init(&(session->zstd_data), 0) < 0)
1419 		pr_warning("Decompression initialization failed. Reported data may be incomplete.\n");
1420 
1421 	if (report.queue_size) {
1422 		ordered_events__set_alloc_size(&session->ordered_events,
1423 					       report.queue_size);
1424 	}
1425 
1426 	session->itrace_synth_opts = &itrace_synth_opts;
1427 
1428 	report.session = session;
1429 
1430 	has_br_stack = perf_header__has_feat(&session->header,
1431 					     HEADER_BRANCH_STACK);
1432 	if (evlist__combined_sample_type(session->evlist) & PERF_SAMPLE_STACK_USER)
1433 		has_br_stack = false;
1434 
1435 	setup_forced_leader(&report, session->evlist);
1436 
1437 	if (symbol_conf.group_sort_idx && !session->evlist->core.nr_groups) {
1438 		parse_options_usage(NULL, options, "group-sort-idx", 0);
1439 		ret = -EINVAL;
1440 		goto error;
1441 	}
1442 
1443 	if (itrace_synth_opts.last_branch || itrace_synth_opts.add_last_branch)
1444 		has_br_stack = true;
1445 
1446 	if (has_br_stack && branch_call_mode)
1447 		symbol_conf.show_branchflag_count = true;
1448 
1449 	memset(&report.brtype_stat, 0, sizeof(struct branch_type_stat));
1450 
1451 	/*
1452 	 * Branch mode is a tristate:
1453 	 * -1 means default, so decide based on the file having branch data.
1454 	 * 0/1 means the user chose a mode.
1455 	 */
1456 	if (((branch_mode == -1 && has_br_stack) || branch_mode == 1) &&
1457 	    !branch_call_mode) {
1458 		sort__mode = SORT_MODE__BRANCH;
1459 		symbol_conf.cumulate_callchain = false;
1460 	}
1461 	if (branch_call_mode) {
1462 		callchain_param.key = CCKEY_ADDRESS;
1463 		callchain_param.branch_callstack = true;
1464 		symbol_conf.use_callchain = true;
1465 		callchain_register_param(&callchain_param);
1466 		if (sort_order == NULL)
1467 			sort_order = "srcline,symbol,dso";
1468 	}
1469 
1470 	if (report.mem_mode) {
1471 		if (sort__mode == SORT_MODE__BRANCH) {
1472 			pr_err("branch and mem mode incompatible\n");
1473 			goto error;
1474 		}
1475 		sort__mode = SORT_MODE__MEMORY;
1476 		symbol_conf.cumulate_callchain = false;
1477 	}
1478 
1479 	if (symbol_conf.report_hierarchy) {
1480 		/* disable incompatible options */
1481 		symbol_conf.cumulate_callchain = false;
1482 
1483 		if (field_order) {
1484 			pr_err("Error: --hierarchy and --fields options cannot be used together\n");
1485 			parse_options_usage(report_usage, options, "F", 1);
1486 			parse_options_usage(NULL, options, "hierarchy", 0);
1487 			goto error;
1488 		}
1489 
1490 		perf_hpp_list.need_collapse = true;
1491 	}
1492 
1493 	if (report.use_stdio)
1494 		use_browser = 0;
1495 	else if (report.use_tui)
1496 		use_browser = 1;
1497 	else if (report.use_gtk)
1498 		use_browser = 2;
1499 
1500 	/* Force tty output for header output and per-thread stat. */
1501 	if (report.header || report.header_only || report.show_threads)
1502 		use_browser = 0;
1503 	if (report.header || report.header_only)
1504 		report.tool.show_feat_hdr = SHOW_FEAT_HEADER;
1505 	if (report.show_full_info)
1506 		report.tool.show_feat_hdr = SHOW_FEAT_HEADER_FULL_INFO;
1507 	if (report.stats_mode || report.tasks_mode)
1508 		use_browser = 0;
1509 	if (report.stats_mode && report.tasks_mode) {
1510 		pr_err("Error: --tasks and --mmaps can't be used together with --stats\n");
1511 		goto error;
1512 	}
1513 
1514 	if (report.total_cycles_mode) {
1515 		if (sort__mode != SORT_MODE__BRANCH)
1516 			report.total_cycles_mode = false;
1517 		else
1518 			sort_order = NULL;
1519 	}
1520 
1521 	if (strcmp(input_name, "-") != 0)
1522 		setup_browser(true);
1523 	else
1524 		use_browser = 0;
1525 
1526 	if (sort_order && strstr(sort_order, "ipc")) {
1527 		parse_options_usage(report_usage, options, "s", 1);
1528 		goto error;
1529 	}
1530 
1531 	if (sort_order && strstr(sort_order, "symbol")) {
1532 		if (sort__mode == SORT_MODE__BRANCH) {
1533 			snprintf(sort_tmp, sizeof(sort_tmp), "%s,%s",
1534 				 sort_order, "ipc_lbr");
1535 			report.symbol_ipc = true;
1536 		} else {
1537 			snprintf(sort_tmp, sizeof(sort_tmp), "%s,%s",
1538 				 sort_order, "ipc_null");
1539 		}
1540 
1541 		sort_order = sort_tmp;
1542 	}
1543 
1544 	if ((last_key != K_SWITCH_INPUT_DATA && last_key != K_RELOAD) &&
1545 	    (setup_sorting(session->evlist) < 0)) {
1546 		if (sort_order)
1547 			parse_options_usage(report_usage, options, "s", 1);
1548 		if (field_order)
1549 			parse_options_usage(sort_order ? NULL : report_usage,
1550 					    options, "F", 1);
1551 		goto error;
1552 	}
1553 
1554 	if ((report.header || report.header_only) && !quiet) {
1555 		perf_session__fprintf_info(session, stdout,
1556 					   report.show_full_info);
1557 		if (report.header_only) {
1558 			if (data.is_pipe) {
1559 				/*
1560 				 * we need to process first few records
1561 				 * which contains PERF_RECORD_HEADER_FEATURE.
1562 				 */
1563 				perf_session__process_events(session);
1564 			}
1565 			ret = 0;
1566 			goto error;
1567 		}
1568 	} else if (use_browser == 0 && !quiet &&
1569 		   !report.stats_mode && !report.tasks_mode) {
1570 		fputs("# To display the perf.data header info, please use --header/--header-only options.\n#\n",
1571 		      stdout);
1572 	}
1573 
1574 	/*
1575 	 * Only in the TUI browser we are doing integrated annotation,
1576 	 * so don't allocate extra space that won't be used in the stdio
1577 	 * implementation.
1578 	 */
1579 	if (ui__has_annotation() || report.symbol_ipc ||
1580 	    report.total_cycles_mode) {
1581 		ret = symbol__annotation_init();
1582 		if (ret < 0)
1583 			goto error;
1584 		/*
1585  		 * For searching by name on the "Browse map details".
1586  		 * providing it only in verbose mode not to bloat too
1587  		 * much struct symbol.
1588  		 */
1589 		if (verbose > 0) {
1590 			/*
1591 			 * XXX: Need to provide a less kludgy way to ask for
1592 			 * more space per symbol, the u32 is for the index on
1593 			 * the ui browser.
1594 			 * See symbol__browser_index.
1595 			 */
1596 			symbol_conf.priv_size += sizeof(u32);
1597 			symbol_conf.sort_by_name = true;
1598 		}
1599 		annotation_config__init(&report.annotation_opts);
1600 	}
1601 
1602 	if (symbol__init(&session->header.env) < 0)
1603 		goto error;
1604 
1605 	if (report.time_str) {
1606 		ret = perf_time__parse_for_ranges(report.time_str, session,
1607 						  &report.ptime_range,
1608 						  &report.range_size,
1609 						  &report.range_num);
1610 		if (ret < 0)
1611 			goto error;
1612 
1613 		itrace_synth_opts__set_time_range(&itrace_synth_opts,
1614 						  report.ptime_range,
1615 						  report.range_num);
1616 	}
1617 
1618 	if (session->tevent.pevent &&
1619 	    tep_set_function_resolver(session->tevent.pevent,
1620 				      machine__resolve_kernel_addr,
1621 				      &session->machines.host) < 0) {
1622 		pr_err("%s: failed to set libtraceevent function resolver\n",
1623 		       __func__);
1624 		return -1;
1625 	}
1626 
1627 	sort__setup_elide(stdout);
1628 
1629 	ret = __cmd_report(&report);
1630 	if (ret == K_SWITCH_INPUT_DATA || ret == K_RELOAD) {
1631 		perf_session__delete(session);
1632 		last_key = K_SWITCH_INPUT_DATA;
1633 		goto repeat;
1634 	} else
1635 		ret = 0;
1636 
1637 error:
1638 	if (report.ptime_range) {
1639 		itrace_synth_opts__clear_time_range(&itrace_synth_opts);
1640 		zfree(&report.ptime_range);
1641 	}
1642 
1643 	if (report.block_reports) {
1644 		block_info__free_report(report.block_reports,
1645 					report.nr_block_reports);
1646 		report.block_reports = NULL;
1647 	}
1648 
1649 	zstd_fini(&(session->zstd_data));
1650 	perf_session__delete(session);
1651 exit:
1652 	free(sort_order_help);
1653 	free(field_order_help);
1654 	return ret;
1655 }
1656