xref: /openbmc/linux/tools/perf/builtin-report.c (revision f677b30b487ca3763c3de3f1b4d8c976c2961cd1)
1 /*
2  * builtin-report.c
3  *
4  * Builtin report command: Analyze the perf.data input file,
5  * look up and read DSOs and symbol information and display
6  * a histogram of results, along various sorting keys.
7  */
8 #include "builtin.h"
9 
10 #include "util/util.h"
11 #include "util/cache.h"
12 
13 #include "util/annotate.h"
14 #include "util/color.h"
15 #include <linux/list.h>
16 #include <linux/rbtree.h>
17 #include "util/symbol.h"
18 #include "util/callchain.h"
19 #include "util/strlist.h"
20 #include "util/values.h"
21 
22 #include "perf.h"
23 #include "util/debug.h"
24 #include "util/evlist.h"
25 #include "util/evsel.h"
26 #include "util/header.h"
27 #include "util/session.h"
28 #include "util/tool.h"
29 
30 #include "util/parse-options.h"
31 #include "util/parse-events.h"
32 
33 #include "util/thread.h"
34 #include "util/sort.h"
35 #include "util/hist.h"
36 #include "util/data.h"
37 #include "arch/common.h"
38 
39 #include <dlfcn.h>
40 #include <linux/bitmap.h>
41 
42 struct perf_report {
43 	struct perf_tool	tool;
44 	struct perf_session	*session;
45 	bool			force, use_tui, use_gtk, use_stdio;
46 	bool			hide_unresolved;
47 	bool			dont_use_callchains;
48 	bool			show_full_info;
49 	bool			show_threads;
50 	bool			inverted_callchain;
51 	bool			mem_mode;
52 	int			max_stack;
53 	struct perf_read_values	show_threads_values;
54 	const char		*pretty_printing_style;
55 	const char		*cpu_list;
56 	const char		*symbol_filter_str;
57 	float			min_percent;
58 	DECLARE_BITMAP(cpu_bitmap, MAX_NR_CPUS);
59 };
60 
61 static int perf_report_config(const char *var, const char *value, void *cb)
62 {
63 	if (!strcmp(var, "report.group")) {
64 		symbol_conf.event_group = perf_config_bool(var, value);
65 		return 0;
66 	}
67 	if (!strcmp(var, "report.percent-limit")) {
68 		struct perf_report *rep = cb;
69 		rep->min_percent = strtof(value, NULL);
70 		return 0;
71 	}
72 
73 	return perf_default_config(var, value, cb);
74 }
75 
76 static int perf_report__add_mem_hist_entry(struct perf_tool *tool,
77 					   struct addr_location *al,
78 					   struct perf_sample *sample,
79 					   struct perf_evsel *evsel,
80 					   struct machine *machine,
81 					   union perf_event *event)
82 {
83 	struct perf_report *rep = container_of(tool, struct perf_report, tool);
84 	struct symbol *parent = NULL;
85 	u8 cpumode = event->header.misc & PERF_RECORD_MISC_CPUMODE_MASK;
86 	int err = 0;
87 	struct hist_entry *he;
88 	struct mem_info *mi, *mx;
89 	uint64_t cost;
90 
91 	if ((sort__has_parent || symbol_conf.use_callchain) &&
92 	    sample->callchain) {
93 		err = machine__resolve_callchain(machine, evsel, al->thread,
94 						 sample, &parent, al,
95 						 rep->max_stack);
96 		if (err)
97 			return err;
98 	}
99 
100 	mi = machine__resolve_mem(machine, al->thread, sample, cpumode);
101 	if (!mi)
102 		return -ENOMEM;
103 
104 	if (rep->hide_unresolved && !al->sym)
105 		return 0;
106 
107 	cost = sample->weight;
108 	if (!cost)
109 		cost = 1;
110 
111 	/*
112 	 * must pass period=weight in order to get the correct
113 	 * sorting from hists__collapse_resort() which is solely
114 	 * based on periods. We want sorting be done on nr_events * weight
115 	 * and this is indirectly achieved by passing period=weight here
116 	 * and the he_stat__add_period() function.
117 	 */
118 	he = __hists__add_entry(&evsel->hists, al, parent, NULL, mi,
119 				cost, cost, 0);
120 	if (!he)
121 		return -ENOMEM;
122 
123 	/*
124 	 * In the TUI browser, we are doing integrated annotation,
125 	 * so we don't allocate the extra space needed because the stdio
126 	 * code will not use it.
127 	 */
128 	if (sort__has_sym && he->ms.sym && use_browser > 0) {
129 		struct annotation *notes = symbol__annotation(he->ms.sym);
130 
131 		assert(evsel != NULL);
132 
133 		if (notes->src == NULL && symbol__alloc_hist(he->ms.sym) < 0)
134 			goto out;
135 
136 		err = hist_entry__inc_addr_samples(he, evsel->idx, al->addr);
137 		if (err)
138 			goto out;
139 	}
140 
141 	if (sort__has_sym && he->mem_info->daddr.sym && use_browser > 0) {
142 		struct annotation *notes;
143 
144 		mx = he->mem_info;
145 
146 		notes = symbol__annotation(mx->daddr.sym);
147 		if (notes->src == NULL && symbol__alloc_hist(mx->daddr.sym) < 0)
148 			goto out;
149 
150 		err = symbol__inc_addr_samples(mx->daddr.sym,
151 					       mx->daddr.map,
152 					       evsel->idx,
153 					       mx->daddr.al_addr);
154 		if (err)
155 			goto out;
156 	}
157 
158 	evsel->hists.stats.total_period += cost;
159 	hists__inc_nr_events(&evsel->hists, PERF_RECORD_SAMPLE);
160 	err = 0;
161 
162 	if (symbol_conf.use_callchain) {
163 		err = callchain_append(he->callchain,
164 				       &callchain_cursor,
165 				       sample->period);
166 	}
167 out:
168 	return err;
169 }
170 
171 static int perf_report__add_branch_hist_entry(struct perf_tool *tool,
172 					struct addr_location *al,
173 					struct perf_sample *sample,
174 					struct perf_evsel *evsel,
175 				      struct machine *machine)
176 {
177 	struct perf_report *rep = container_of(tool, struct perf_report, tool);
178 	struct symbol *parent = NULL;
179 	int err = 0;
180 	unsigned i;
181 	struct hist_entry *he;
182 	struct branch_info *bi, *bx;
183 
184 	if ((sort__has_parent || symbol_conf.use_callchain)
185 	    && sample->callchain) {
186 		err = machine__resolve_callchain(machine, evsel, al->thread,
187 						 sample, &parent, al,
188 						 rep->max_stack);
189 		if (err)
190 			return err;
191 	}
192 
193 	bi = machine__resolve_bstack(machine, al->thread,
194 				     sample->branch_stack);
195 	if (!bi)
196 		return -ENOMEM;
197 
198 	for (i = 0; i < sample->branch_stack->nr; i++) {
199 		if (rep->hide_unresolved && !(bi[i].from.sym && bi[i].to.sym))
200 			continue;
201 
202 		err = -ENOMEM;
203 
204 		/* overwrite the 'al' to branch-to info */
205 		al->map = bi[i].to.map;
206 		al->sym = bi[i].to.sym;
207 		al->addr = bi[i].to.addr;
208 		/*
209 		 * The report shows the percentage of total branches captured
210 		 * and not events sampled. Thus we use a pseudo period of 1.
211 		 */
212 		he = __hists__add_entry(&evsel->hists, al, parent, &bi[i], NULL,
213 					1, 1, 0);
214 		if (he) {
215 			struct annotation *notes;
216 			bx = he->branch_info;
217 			if (bx->from.sym && use_browser == 1 && sort__has_sym) {
218 				notes = symbol__annotation(bx->from.sym);
219 				if (!notes->src
220 				    && symbol__alloc_hist(bx->from.sym) < 0)
221 					goto out;
222 
223 				err = symbol__inc_addr_samples(bx->from.sym,
224 							       bx->from.map,
225 							       evsel->idx,
226 							       bx->from.al_addr);
227 				if (err)
228 					goto out;
229 			}
230 
231 			if (bx->to.sym && use_browser == 1 && sort__has_sym) {
232 				notes = symbol__annotation(bx->to.sym);
233 				if (!notes->src
234 				    && symbol__alloc_hist(bx->to.sym) < 0)
235 					goto out;
236 
237 				err = symbol__inc_addr_samples(bx->to.sym,
238 							       bx->to.map,
239 							       evsel->idx,
240 							       bx->to.al_addr);
241 				if (err)
242 					goto out;
243 			}
244 			evsel->hists.stats.total_period += 1;
245 			hists__inc_nr_events(&evsel->hists, PERF_RECORD_SAMPLE);
246 		} else
247 			goto out;
248 	}
249 	err = 0;
250 out:
251 	free(bi);
252 	return err;
253 }
254 
255 static int perf_evsel__add_hist_entry(struct perf_tool *tool,
256 				      struct perf_evsel *evsel,
257 				      struct addr_location *al,
258 				      struct perf_sample *sample,
259 				      struct machine *machine)
260 {
261 	struct perf_report *rep = container_of(tool, struct perf_report, tool);
262 	struct symbol *parent = NULL;
263 	int err = 0;
264 	struct hist_entry *he;
265 
266 	if ((sort__has_parent || symbol_conf.use_callchain) && sample->callchain) {
267 		err = machine__resolve_callchain(machine, evsel, al->thread,
268 						 sample, &parent, al,
269 						 rep->max_stack);
270 		if (err)
271 			return err;
272 	}
273 
274 	he = __hists__add_entry(&evsel->hists, al, parent, NULL, NULL,
275 				sample->period, sample->weight,
276 				sample->transaction);
277 	if (he == NULL)
278 		return -ENOMEM;
279 
280 	if (symbol_conf.use_callchain) {
281 		err = callchain_append(he->callchain,
282 				       &callchain_cursor,
283 				       sample->period);
284 		if (err)
285 			return err;
286 	}
287 	/*
288 	 * Only in the TUI browser we are doing integrated annotation,
289 	 * so we don't allocated the extra space needed because the stdio
290 	 * code will not use it.
291 	 */
292 	if (he->ms.sym != NULL && use_browser == 1 && sort__has_sym) {
293 		struct annotation *notes = symbol__annotation(he->ms.sym);
294 
295 		assert(evsel != NULL);
296 
297 		err = -ENOMEM;
298 		if (notes->src == NULL && symbol__alloc_hist(he->ms.sym) < 0)
299 			goto out;
300 
301 		err = hist_entry__inc_addr_samples(he, evsel->idx, al->addr);
302 	}
303 
304 	evsel->hists.stats.total_period += sample->period;
305 	hists__inc_nr_events(&evsel->hists, PERF_RECORD_SAMPLE);
306 out:
307 	return err;
308 }
309 
310 
311 static int process_sample_event(struct perf_tool *tool,
312 				union perf_event *event,
313 				struct perf_sample *sample,
314 				struct perf_evsel *evsel,
315 				struct machine *machine)
316 {
317 	struct perf_report *rep = container_of(tool, struct perf_report, tool);
318 	struct addr_location al;
319 	int ret;
320 
321 	if (perf_event__preprocess_sample(event, machine, &al, sample) < 0) {
322 		fprintf(stderr, "problem processing %d event, skipping it.\n",
323 			event->header.type);
324 		return -1;
325 	}
326 
327 	if (al.filtered || (rep->hide_unresolved && al.sym == NULL))
328 		return 0;
329 
330 	if (rep->cpu_list && !test_bit(sample->cpu, rep->cpu_bitmap))
331 		return 0;
332 
333 	if (sort__mode == SORT_MODE__BRANCH) {
334 		ret = perf_report__add_branch_hist_entry(tool, &al, sample,
335 							 evsel, machine);
336 		if (ret < 0)
337 			pr_debug("problem adding lbr entry, skipping event\n");
338 	} else if (rep->mem_mode == 1) {
339 		ret = perf_report__add_mem_hist_entry(tool, &al, sample,
340 						      evsel, machine, event);
341 		if (ret < 0)
342 			pr_debug("problem adding mem entry, skipping event\n");
343 	} else {
344 		if (al.map != NULL)
345 			al.map->dso->hit = 1;
346 
347 		ret = perf_evsel__add_hist_entry(tool, evsel, &al, sample,
348 						 machine);
349 		if (ret < 0)
350 			pr_debug("problem incrementing symbol period, skipping event\n");
351 	}
352 	return ret;
353 }
354 
355 static int process_read_event(struct perf_tool *tool,
356 			      union perf_event *event,
357 			      struct perf_sample *sample __maybe_unused,
358 			      struct perf_evsel *evsel,
359 			      struct machine *machine __maybe_unused)
360 {
361 	struct perf_report *rep = container_of(tool, struct perf_report, tool);
362 
363 	if (rep->show_threads) {
364 		const char *name = evsel ? perf_evsel__name(evsel) : "unknown";
365 		perf_read_values_add_value(&rep->show_threads_values,
366 					   event->read.pid, event->read.tid,
367 					   event->read.id,
368 					   name,
369 					   event->read.value);
370 	}
371 
372 	dump_printf(": %d %d %s %" PRIu64 "\n", event->read.pid, event->read.tid,
373 		    evsel ? perf_evsel__name(evsel) : "FAIL",
374 		    event->read.value);
375 
376 	return 0;
377 }
378 
379 /* For pipe mode, sample_type is not currently set */
380 static int perf_report__setup_sample_type(struct perf_report *rep)
381 {
382 	struct perf_session *session = rep->session;
383 	u64 sample_type = perf_evlist__combined_sample_type(session->evlist);
384 	bool is_pipe = perf_data_file__is_pipe(session->file);
385 
386 	if (!is_pipe && !(sample_type & PERF_SAMPLE_CALLCHAIN)) {
387 		if (sort__has_parent) {
388 			ui__error("Selected --sort parent, but no "
389 				    "callchain data. Did you call "
390 				    "'perf record' without -g?\n");
391 			return -EINVAL;
392 		}
393 		if (symbol_conf.use_callchain) {
394 			ui__error("Selected -g but no callchain data. Did "
395 				    "you call 'perf record' without -g?\n");
396 			return -1;
397 		}
398 	} else if (!rep->dont_use_callchains &&
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 (sort__mode == SORT_MODE__BRANCH) {
409 		if (!is_pipe &&
410 		    !(sample_type & PERF_SAMPLE_BRANCH_STACK)) {
411 			ui__error("Selected -b but no branch data. "
412 				  "Did you call perf record without -b?\n");
413 			return -1;
414 		}
415 	}
416 
417 	return 0;
418 }
419 
420 static void sig_handler(int sig __maybe_unused)
421 {
422 	session_done = 1;
423 }
424 
425 static size_t hists__fprintf_nr_sample_events(struct perf_report *rep,
426 					      struct hists *hists,
427 					      const char *evname, FILE *fp)
428 {
429 	size_t ret;
430 	char unit;
431 	unsigned long nr_samples = hists->stats.nr_events[PERF_RECORD_SAMPLE];
432 	u64 nr_events = hists->stats.total_period;
433 	struct perf_evsel *evsel = hists_to_evsel(hists);
434 	char buf[512];
435 	size_t size = sizeof(buf);
436 
437 	if (perf_evsel__is_group_event(evsel)) {
438 		struct perf_evsel *pos;
439 
440 		perf_evsel__group_desc(evsel, buf, size);
441 		evname = buf;
442 
443 		for_each_group_member(pos, evsel) {
444 			nr_samples += pos->hists.stats.nr_events[PERF_RECORD_SAMPLE];
445 			nr_events += pos->hists.stats.total_period;
446 		}
447 	}
448 
449 	nr_samples = convert_unit(nr_samples, &unit);
450 	ret = fprintf(fp, "# Samples: %lu%c", nr_samples, unit);
451 	if (evname != NULL)
452 		ret += fprintf(fp, " of event '%s'", evname);
453 
454 	if (rep->mem_mode) {
455 		ret += fprintf(fp, "\n# Total weight : %" PRIu64, nr_events);
456 		ret += fprintf(fp, "\n# Sort order   : %s", sort_order);
457 	} else
458 		ret += fprintf(fp, "\n# Event count (approx.): %" PRIu64, nr_events);
459 	return ret + fprintf(fp, "\n#\n");
460 }
461 
462 static int perf_evlist__tty_browse_hists(struct perf_evlist *evlist,
463 					 struct perf_report *rep,
464 					 const char *help)
465 {
466 	struct perf_evsel *pos;
467 
468 	list_for_each_entry(pos, &evlist->entries, node) {
469 		struct hists *hists = &pos->hists;
470 		const char *evname = perf_evsel__name(pos);
471 
472 		if (symbol_conf.event_group &&
473 		    !perf_evsel__is_group_leader(pos))
474 			continue;
475 
476 		hists__fprintf_nr_sample_events(rep, hists, evname, stdout);
477 		hists__fprintf(hists, true, 0, 0, rep->min_percent, stdout);
478 		fprintf(stdout, "\n\n");
479 	}
480 
481 	if (sort_order == default_sort_order &&
482 	    parent_pattern == default_parent_pattern) {
483 		fprintf(stdout, "#\n# (%s)\n#\n", help);
484 
485 		if (rep->show_threads) {
486 			bool style = !strcmp(rep->pretty_printing_style, "raw");
487 			perf_read_values_display(stdout, &rep->show_threads_values,
488 						 style);
489 			perf_read_values_destroy(&rep->show_threads_values);
490 		}
491 	}
492 
493 	return 0;
494 }
495 
496 static int __cmd_report(struct perf_report *rep)
497 {
498 	int ret = -EINVAL;
499 	u64 nr_samples;
500 	struct perf_session *session = rep->session;
501 	struct perf_evsel *pos;
502 	struct map *kernel_map;
503 	struct kmap *kernel_kmap;
504 	const char *help = "For a higher level overview, try: perf report --sort comm,dso";
505 	struct ui_progress prog;
506 	struct perf_data_file *file = session->file;
507 
508 	signal(SIGINT, sig_handler);
509 
510 	if (rep->cpu_list) {
511 		ret = perf_session__cpu_bitmap(session, rep->cpu_list,
512 					       rep->cpu_bitmap);
513 		if (ret)
514 			return ret;
515 	}
516 
517 	if (use_browser <= 0)
518 		perf_session__fprintf_info(session, stdout, rep->show_full_info);
519 
520 	if (rep->show_threads)
521 		perf_read_values_init(&rep->show_threads_values);
522 
523 	ret = perf_report__setup_sample_type(rep);
524 	if (ret)
525 		return ret;
526 
527 	ret = perf_session__process_events(session, &rep->tool);
528 	if (ret)
529 		return ret;
530 
531 	kernel_map = session->machines.host.vmlinux_maps[MAP__FUNCTION];
532 	kernel_kmap = map__kmap(kernel_map);
533 	if (kernel_map == NULL ||
534 	    (kernel_map->dso->hit &&
535 	     (kernel_kmap->ref_reloc_sym == NULL ||
536 	      kernel_kmap->ref_reloc_sym->addr == 0))) {
537 		const char *desc =
538 		    "As no suitable kallsyms nor vmlinux was found, kernel samples\n"
539 		    "can't be resolved.";
540 
541 		if (kernel_map) {
542 			const struct dso *kdso = kernel_map->dso;
543 			if (!RB_EMPTY_ROOT(&kdso->symbols[MAP__FUNCTION])) {
544 				desc = "If some relocation was applied (e.g. "
545 				       "kexec) symbols may be misresolved.";
546 			}
547 		}
548 
549 		ui__warning(
550 "Kernel address maps (/proc/{kallsyms,modules}) were restricted.\n\n"
551 "Check /proc/sys/kernel/kptr_restrict before running 'perf record'.\n\n%s\n\n"
552 "Samples in kernel modules can't be resolved as well.\n\n",
553 		desc);
554 	}
555 
556 	if (verbose > 3)
557 		perf_session__fprintf(session, stdout);
558 
559 	if (verbose > 2)
560 		perf_session__fprintf_dsos(session, stdout);
561 
562 	if (dump_trace) {
563 		perf_session__fprintf_nr_events(session, stdout);
564 		return 0;
565 	}
566 
567 	nr_samples = 0;
568 	list_for_each_entry(pos, &session->evlist->entries, node)
569 		nr_samples += pos->hists.nr_entries;
570 
571 	ui_progress__init(&prog, nr_samples, "Merging related events...");
572 
573 	nr_samples = 0;
574 	list_for_each_entry(pos, &session->evlist->entries, node) {
575 		struct hists *hists = &pos->hists;
576 
577 		if (pos->idx == 0)
578 			hists->symbol_filter_str = rep->symbol_filter_str;
579 
580 		hists__collapse_resort(hists, &prog);
581 		nr_samples += hists->stats.nr_events[PERF_RECORD_SAMPLE];
582 
583 		/* Non-group events are considered as leader */
584 		if (symbol_conf.event_group &&
585 		    !perf_evsel__is_group_leader(pos)) {
586 			struct hists *leader_hists = &pos->leader->hists;
587 
588 			hists__match(leader_hists, hists);
589 			hists__link(leader_hists, hists);
590 		}
591 	}
592 	ui_progress__finish();
593 
594 	if (session_done())
595 		return 0;
596 
597 	if (nr_samples == 0) {
598 		ui__error("The %s file has no samples!\n", file->path);
599 		return 0;
600 	}
601 
602 	list_for_each_entry(pos, &session->evlist->entries, node)
603 		hists__output_resort(&pos->hists);
604 
605 	if (use_browser > 0) {
606 		if (use_browser == 1) {
607 			ret = perf_evlist__tui_browse_hists(session->evlist,
608 							help, NULL,
609 							rep->min_percent,
610 							&session->header.env);
611 			/*
612 			 * Usually "ret" is the last pressed key, and we only
613 			 * care if the key notifies us to switch data file.
614 			 */
615 			if (ret != K_SWITCH_INPUT_DATA)
616 				ret = 0;
617 
618 		} else if (use_browser == 2) {
619 			int (*hist_browser)(struct perf_evlist *,
620 					    const char *,
621 					    struct hist_browser_timer *,
622 					    float min_pcnt);
623 
624 			hist_browser = dlsym(perf_gtk_handle,
625 					     "perf_evlist__gtk_browse_hists");
626 			if (hist_browser == NULL) {
627 				ui__error("GTK browser not found!\n");
628 				return ret;
629 			}
630 			hist_browser(session->evlist, help, NULL,
631 				     rep->min_percent);
632 		}
633 	} else
634 		perf_evlist__tty_browse_hists(session->evlist, rep, help);
635 
636 	return ret;
637 }
638 
639 static int
640 parse_callchain_opt(const struct option *opt, const char *arg, int unset)
641 {
642 	struct perf_report *rep = (struct perf_report *)opt->value;
643 	char *tok, *tok2;
644 	char *endptr;
645 
646 	/*
647 	 * --no-call-graph
648 	 */
649 	if (unset) {
650 		rep->dont_use_callchains = true;
651 		return 0;
652 	}
653 
654 	symbol_conf.use_callchain = true;
655 
656 	if (!arg)
657 		return 0;
658 
659 	tok = strtok((char *)arg, ",");
660 	if (!tok)
661 		return -1;
662 
663 	/* get the output mode */
664 	if (!strncmp(tok, "graph", strlen(arg)))
665 		callchain_param.mode = CHAIN_GRAPH_ABS;
666 
667 	else if (!strncmp(tok, "flat", strlen(arg)))
668 		callchain_param.mode = CHAIN_FLAT;
669 
670 	else if (!strncmp(tok, "fractal", strlen(arg)))
671 		callchain_param.mode = CHAIN_GRAPH_REL;
672 
673 	else if (!strncmp(tok, "none", strlen(arg))) {
674 		callchain_param.mode = CHAIN_NONE;
675 		symbol_conf.use_callchain = false;
676 
677 		return 0;
678 	}
679 
680 	else
681 		return -1;
682 
683 	/* get the min percentage */
684 	tok = strtok(NULL, ",");
685 	if (!tok)
686 		goto setup;
687 
688 	callchain_param.min_percent = strtod(tok, &endptr);
689 	if (tok == endptr)
690 		return -1;
691 
692 	/* get the print limit */
693 	tok2 = strtok(NULL, ",");
694 	if (!tok2)
695 		goto setup;
696 
697 	if (tok2[0] != 'c') {
698 		callchain_param.print_limit = strtoul(tok2, &endptr, 0);
699 		tok2 = strtok(NULL, ",");
700 		if (!tok2)
701 			goto setup;
702 	}
703 
704 	/* get the call chain order */
705 	if (!strncmp(tok2, "caller", strlen("caller")))
706 		callchain_param.order = ORDER_CALLER;
707 	else if (!strncmp(tok2, "callee", strlen("callee")))
708 		callchain_param.order = ORDER_CALLEE;
709 	else
710 		return -1;
711 
712 	/* Get the sort key */
713 	tok2 = strtok(NULL, ",");
714 	if (!tok2)
715 		goto setup;
716 	if (!strncmp(tok2, "function", strlen("function")))
717 		callchain_param.key = CCKEY_FUNCTION;
718 	else if (!strncmp(tok2, "address", strlen("address")))
719 		callchain_param.key = CCKEY_ADDRESS;
720 	else
721 		return -1;
722 setup:
723 	if (callchain_register_param(&callchain_param) < 0) {
724 		fprintf(stderr, "Can't register callchain params\n");
725 		return -1;
726 	}
727 	return 0;
728 }
729 
730 int
731 report_parse_ignore_callees_opt(const struct option *opt __maybe_unused,
732 				const char *arg, int unset __maybe_unused)
733 {
734 	if (arg) {
735 		int err = regcomp(&ignore_callees_regex, arg, REG_EXTENDED);
736 		if (err) {
737 			char buf[BUFSIZ];
738 			regerror(err, &ignore_callees_regex, buf, sizeof(buf));
739 			pr_err("Invalid --ignore-callees regex: %s\n%s", arg, buf);
740 			return -1;
741 		}
742 		have_ignore_callees = 1;
743 	}
744 
745 	return 0;
746 }
747 
748 static int
749 parse_branch_mode(const struct option *opt __maybe_unused,
750 		  const char *str __maybe_unused, int unset)
751 {
752 	int *branch_mode = opt->value;
753 
754 	*branch_mode = !unset;
755 	return 0;
756 }
757 
758 static int
759 parse_percent_limit(const struct option *opt, const char *str,
760 		    int unset __maybe_unused)
761 {
762 	struct perf_report *rep = opt->value;
763 
764 	rep->min_percent = strtof(str, NULL);
765 	return 0;
766 }
767 
768 int cmd_report(int argc, const char **argv, const char *prefix __maybe_unused)
769 {
770 	struct perf_session *session;
771 	struct stat st;
772 	bool has_br_stack = false;
773 	int branch_mode = -1;
774 	int ret = -1;
775 	char callchain_default_opt[] = "fractal,0.5,callee";
776 	const char * const report_usage[] = {
777 		"perf report [<options>]",
778 		NULL
779 	};
780 	struct perf_report report = {
781 		.tool = {
782 			.sample		 = process_sample_event,
783 			.mmap		 = perf_event__process_mmap,
784 			.mmap2		 = perf_event__process_mmap2,
785 			.comm		 = perf_event__process_comm,
786 			.exit		 = perf_event__process_exit,
787 			.fork		 = perf_event__process_fork,
788 			.lost		 = perf_event__process_lost,
789 			.read		 = process_read_event,
790 			.attr		 = perf_event__process_attr,
791 			.tracing_data	 = perf_event__process_tracing_data,
792 			.build_id	 = perf_event__process_build_id,
793 			.ordered_samples = true,
794 			.ordering_requires_timestamps = true,
795 		},
796 		.max_stack		 = PERF_MAX_STACK_DEPTH,
797 		.pretty_printing_style	 = "normal",
798 	};
799 	const struct option options[] = {
800 	OPT_STRING('i', "input", &input_name, "file",
801 		    "input file name"),
802 	OPT_INCR('v', "verbose", &verbose,
803 		    "be more verbose (show symbol address, etc)"),
804 	OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace,
805 		    "dump raw trace in ASCII"),
806 	OPT_STRING('k', "vmlinux", &symbol_conf.vmlinux_name,
807 		   "file", "vmlinux pathname"),
808 	OPT_STRING(0, "kallsyms", &symbol_conf.kallsyms_name,
809 		   "file", "kallsyms pathname"),
810 	OPT_BOOLEAN('f', "force", &report.force, "don't complain, do it"),
811 	OPT_BOOLEAN('m', "modules", &symbol_conf.use_modules,
812 		    "load module symbols - WARNING: use only with -k and LIVE kernel"),
813 	OPT_BOOLEAN('n', "show-nr-samples", &symbol_conf.show_nr_samples,
814 		    "Show a column with the number of samples"),
815 	OPT_BOOLEAN('T', "threads", &report.show_threads,
816 		    "Show per-thread event counters"),
817 	OPT_STRING(0, "pretty", &report.pretty_printing_style, "key",
818 		   "pretty printing style key: normal raw"),
819 	OPT_BOOLEAN(0, "tui", &report.use_tui, "Use the TUI interface"),
820 	OPT_BOOLEAN(0, "gtk", &report.use_gtk, "Use the GTK2 interface"),
821 	OPT_BOOLEAN(0, "stdio", &report.use_stdio,
822 		    "Use the stdio interface"),
823 	OPT_STRING('s', "sort", &sort_order, "key[,key2...]",
824 		   "sort by key(s): pid, comm, dso, symbol, parent, cpu, srcline,"
825 		   " dso_to, dso_from, symbol_to, symbol_from, mispredict,"
826 		   " weight, local_weight, mem, symbol_daddr, dso_daddr, tlb, "
827 		   "snoop, locked, abort, in_tx, transaction"),
828 	OPT_BOOLEAN(0, "showcpuutilization", &symbol_conf.show_cpu_utilization,
829 		    "Show sample percentage for different cpu modes"),
830 	OPT_STRING('p', "parent", &parent_pattern, "regex",
831 		   "regex filter to identify parent, see: '--sort parent'"),
832 	OPT_BOOLEAN('x', "exclude-other", &symbol_conf.exclude_other,
833 		    "Only display entries with parent-match"),
834 	OPT_CALLBACK_DEFAULT('g', "call-graph", &report, "output_type,min_percent[,print_limit],call_order",
835 		     "Display callchains using output_type (graph, flat, fractal, or none) , min percent threshold, optional print limit, callchain order, key (function or address). "
836 		     "Default: fractal,0.5,callee,function", &parse_callchain_opt, callchain_default_opt),
837 	OPT_INTEGER(0, "max-stack", &report.max_stack,
838 		    "Set the maximum stack depth when parsing the callchain, "
839 		    "anything beyond the specified depth will be ignored. "
840 		    "Default: " __stringify(PERF_MAX_STACK_DEPTH)),
841 	OPT_BOOLEAN('G', "inverted", &report.inverted_callchain,
842 		    "alias for inverted call graph"),
843 	OPT_CALLBACK(0, "ignore-callees", NULL, "regex",
844 		   "ignore callees of these functions in call graphs",
845 		   report_parse_ignore_callees_opt),
846 	OPT_STRING('d', "dsos", &symbol_conf.dso_list_str, "dso[,dso...]",
847 		   "only consider symbols in these dsos"),
848 	OPT_STRING('c', "comms", &symbol_conf.comm_list_str, "comm[,comm...]",
849 		   "only consider symbols in these comms"),
850 	OPT_STRING('S', "symbols", &symbol_conf.sym_list_str, "symbol[,symbol...]",
851 		   "only consider these symbols"),
852 	OPT_STRING(0, "symbol-filter", &report.symbol_filter_str, "filter",
853 		   "only show symbols that (partially) match with this filter"),
854 	OPT_STRING('w', "column-widths", &symbol_conf.col_width_list_str,
855 		   "width[,width...]",
856 		   "don't try to adjust column width, use these fixed values"),
857 	OPT_STRING('t', "field-separator", &symbol_conf.field_sep, "separator",
858 		   "separator for columns, no spaces will be added between "
859 		   "columns '.' is reserved."),
860 	OPT_BOOLEAN('U', "hide-unresolved", &report.hide_unresolved,
861 		    "Only display entries resolved to a symbol"),
862 	OPT_STRING(0, "symfs", &symbol_conf.symfs, "directory",
863 		    "Look for files with symbols relative to this directory"),
864 	OPT_STRING('C', "cpu", &report.cpu_list, "cpu",
865 		   "list of cpus to profile"),
866 	OPT_BOOLEAN('I', "show-info", &report.show_full_info,
867 		    "Display extended information about perf.data file"),
868 	OPT_BOOLEAN(0, "source", &symbol_conf.annotate_src,
869 		    "Interleave source code with assembly code (default)"),
870 	OPT_BOOLEAN(0, "asm-raw", &symbol_conf.annotate_asm_raw,
871 		    "Display raw encoding of assembly instructions (default)"),
872 	OPT_STRING('M', "disassembler-style", &disassembler_style, "disassembler style",
873 		   "Specify disassembler style (e.g. -M intel for intel syntax)"),
874 	OPT_BOOLEAN(0, "show-total-period", &symbol_conf.show_total_period,
875 		    "Show a column with the sum of periods"),
876 	OPT_BOOLEAN(0, "group", &symbol_conf.event_group,
877 		    "Show event group information together"),
878 	OPT_CALLBACK_NOOPT('b', "branch-stack", &branch_mode, "",
879 		    "use branch records for histogram filling", parse_branch_mode),
880 	OPT_STRING(0, "objdump", &objdump_path, "path",
881 		   "objdump binary to use for disassembly and annotations"),
882 	OPT_BOOLEAN(0, "demangle", &symbol_conf.demangle,
883 		    "Disable symbol demangling"),
884 	OPT_BOOLEAN(0, "mem-mode", &report.mem_mode, "mem access profile"),
885 	OPT_CALLBACK(0, "percent-limit", &report, "percent",
886 		     "Don't show entries under that percent", parse_percent_limit),
887 	OPT_END()
888 	};
889 	struct perf_data_file file = {
890 		.mode  = PERF_DATA_MODE_READ,
891 	};
892 
893 	perf_config(perf_report_config, &report);
894 
895 	argc = parse_options(argc, argv, options, report_usage, 0);
896 
897 	if (report.use_stdio)
898 		use_browser = 0;
899 	else if (report.use_tui)
900 		use_browser = 1;
901 	else if (report.use_gtk)
902 		use_browser = 2;
903 
904 	if (report.inverted_callchain)
905 		callchain_param.order = ORDER_CALLER;
906 
907 	if (!input_name || !strlen(input_name)) {
908 		if (!fstat(STDIN_FILENO, &st) && S_ISFIFO(st.st_mode))
909 			input_name = "-";
910 		else
911 			input_name = "perf.data";
912 	}
913 
914 	file.path  = input_name;
915 	file.force = report.force;
916 
917 repeat:
918 	session = perf_session__new(&file, false, &report.tool);
919 	if (session == NULL)
920 		return -ENOMEM;
921 
922 	report.session = session;
923 
924 	has_br_stack = perf_header__has_feat(&session->header,
925 					     HEADER_BRANCH_STACK);
926 
927 	if (branch_mode == -1 && has_br_stack)
928 		sort__mode = SORT_MODE__BRANCH;
929 
930 	/* sort__mode could be NORMAL if --no-branch-stack */
931 	if (sort__mode == SORT_MODE__BRANCH) {
932 		/*
933 		 * if no sort_order is provided, then specify
934 		 * branch-mode specific order
935 		 */
936 		if (sort_order == default_sort_order)
937 			sort_order = "comm,dso_from,symbol_from,"
938 				     "dso_to,symbol_to";
939 
940 	}
941 	if (report.mem_mode) {
942 		if (sort__mode == SORT_MODE__BRANCH) {
943 			fprintf(stderr, "branch and mem mode incompatible\n");
944 			goto error;
945 		}
946 		sort__mode = SORT_MODE__MEMORY;
947 
948 		/*
949 		 * if no sort_order is provided, then specify
950 		 * branch-mode specific order
951 		 */
952 		if (sort_order == default_sort_order)
953 			sort_order = "local_weight,mem,sym,dso,symbol_daddr,dso_daddr,snoop,tlb,locked";
954 	}
955 
956 	if (setup_sorting() < 0) {
957 		parse_options_usage(report_usage, options, "s", 1);
958 		goto error;
959 	}
960 
961 	if (parent_pattern != default_parent_pattern) {
962 		if (sort_dimension__add("parent") < 0)
963 			goto error;
964 	}
965 
966 	if (strcmp(input_name, "-") != 0)
967 		setup_browser(true);
968 	else {
969 		use_browser = 0;
970 		perf_hpp__init();
971 	}
972 
973 	/*
974 	 * Only in the TUI browser we are doing integrated annotation,
975 	 * so don't allocate extra space that won't be used in the stdio
976 	 * implementation.
977 	 */
978 	if (use_browser == 1 && sort__has_sym) {
979 		symbol_conf.priv_size = sizeof(struct annotation);
980 		machines__set_symbol_filter(&session->machines,
981 					    symbol__annotate_init);
982 		/*
983  		 * For searching by name on the "Browse map details".
984  		 * providing it only in verbose mode not to bloat too
985  		 * much struct symbol.
986  		 */
987 		if (verbose) {
988 			/*
989 			 * XXX: Need to provide a less kludgy way to ask for
990 			 * more space per symbol, the u32 is for the index on
991 			 * the ui browser.
992 			 * See symbol__browser_index.
993 			 */
994 			symbol_conf.priv_size += sizeof(u32);
995 			symbol_conf.sort_by_name = true;
996 		}
997 	}
998 
999 	if (symbol__init() < 0)
1000 		goto error;
1001 
1002 	if (argc) {
1003 		/*
1004 		 * Special case: if there's an argument left then assume that
1005 		 * it's a symbol filter:
1006 		 */
1007 		if (argc > 1)
1008 			usage_with_options(report_usage, options);
1009 
1010 		report.symbol_filter_str = argv[0];
1011 	}
1012 
1013 	sort__setup_elide(stdout);
1014 
1015 	ret = __cmd_report(&report);
1016 	if (ret == K_SWITCH_INPUT_DATA) {
1017 		perf_session__delete(session);
1018 		goto repeat;
1019 	} else
1020 		ret = 0;
1021 
1022 error:
1023 	perf_session__delete(session);
1024 	return ret;
1025 }
1026