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.
10 #include "util/util.h"
12 #include "util/annotate.h"
13 #include "util/color.h"
14 #include <linux/list.h>
15 #include "util/cache.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"
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"
30 #include "util/parse-options.h"
31 #include "util/parse-events.h"
33 #include "util/thread.h"
34 #include "util/sort.h"
35 #include "util/hist.h"
37 #include <linux/bitmap.h>
40 struct perf_tool tool;
41 struct perf_session *session;
42 char const *input_name;
43 bool force, use_tui, use_stdio;
45 bool dont_use_callchains;
48 bool inverted_callchain;
49 struct perf_read_values show_threads_values;
50 const char *pretty_printing_style;
51 symbol_filter_t annotate_init;
53 DECLARE_BITMAP(cpu_bitmap, MAX_NR_CPUS);
56 static int perf_report__add_branch_hist_entry(struct perf_tool *tool,
57 struct addr_location *al,
58 struct perf_sample *sample,
59 struct perf_evsel *evsel,
60 struct machine *machine)
62 struct perf_report *rep = container_of(tool, struct perf_report, tool);
63 struct symbol *parent = NULL;
66 struct hist_entry *he;
67 struct branch_info *bi, *bx;
69 if ((sort__has_parent || symbol_conf.use_callchain)
70 && sample->callchain) {
71 err = machine__resolve_callchain(machine, evsel, al->thread,
72 sample->callchain, &parent);
77 bi = machine__resolve_bstack(machine, al->thread,
78 sample->branch_stack);
82 for (i = 0; i < sample->branch_stack->nr; i++) {
83 if (rep->hide_unresolved && !(bi[i].from.sym && bi[i].to.sym))
86 * The report shows the percentage of total branches captured
87 * and not events sampled. Thus we use a pseudo period of 1.
89 he = __hists__add_branch_entry(&evsel->hists, al, parent,
92 struct annotation *notes;
95 if (bx->from.sym && use_browser > 0) {
96 notes = symbol__annotation(bx->from.sym);
98 && symbol__alloc_hist(bx->from.sym) < 0)
101 err = symbol__inc_addr_samples(bx->from.sym,
109 if (bx->to.sym && use_browser > 0) {
110 notes = symbol__annotation(bx->to.sym);
112 && symbol__alloc_hist(bx->to.sym) < 0)
115 err = symbol__inc_addr_samples(bx->to.sym,
122 evsel->hists.stats.total_period += 1;
123 hists__inc_nr_events(&evsel->hists, PERF_RECORD_SAMPLE);
132 static int perf_evsel__add_hist_entry(struct perf_evsel *evsel,
133 struct addr_location *al,
134 struct perf_sample *sample,
135 struct machine *machine)
137 struct symbol *parent = NULL;
139 struct hist_entry *he;
141 if ((sort__has_parent || symbol_conf.use_callchain) && sample->callchain) {
142 err = machine__resolve_callchain(machine, evsel, al->thread,
143 sample->callchain, &parent);
148 he = __hists__add_entry(&evsel->hists, al, parent, sample->period);
152 if (symbol_conf.use_callchain) {
153 err = callchain_append(he->callchain,
154 &evsel->hists.callchain_cursor,
160 * Only in the newt browser we are doing integrated annotation,
161 * so we don't allocated the extra space needed because the stdio
162 * code will not use it.
164 if (al->sym != NULL && use_browser > 0) {
165 struct annotation *notes = symbol__annotation(he->ms.sym);
167 assert(evsel != NULL);
170 if (notes->src == NULL && symbol__alloc_hist(he->ms.sym) < 0)
173 err = hist_entry__inc_addr_samples(he, evsel->idx, al->addr);
176 evsel->hists.stats.total_period += sample->period;
177 hists__inc_nr_events(&evsel->hists, PERF_RECORD_SAMPLE);
183 static int process_sample_event(struct perf_tool *tool,
184 union perf_event *event,
185 struct perf_sample *sample,
186 struct perf_evsel *evsel,
187 struct machine *machine)
189 struct perf_report *rep = container_of(tool, struct perf_report, tool);
190 struct addr_location al;
192 if (perf_event__preprocess_sample(event, machine, &al, sample,
193 rep->annotate_init) < 0) {
194 fprintf(stderr, "problem processing %d event, skipping it.\n",
199 if (al.filtered || (rep->hide_unresolved && al.sym == NULL))
202 if (rep->cpu_list && !test_bit(sample->cpu, rep->cpu_bitmap))
205 if (sort__branch_mode == 1) {
206 if (perf_report__add_branch_hist_entry(tool, &al, sample,
208 pr_debug("problem adding lbr entry, skipping event\n");
213 al.map->dso->hit = 1;
215 if (perf_evsel__add_hist_entry(evsel, &al, sample, machine)) {
216 pr_debug("problem incrementing symbol period, skipping event\n");
223 static int process_read_event(struct perf_tool *tool,
224 union perf_event *event,
225 struct perf_sample *sample __used,
226 struct perf_evsel *evsel,
227 struct machine *machine __used)
229 struct perf_report *rep = container_of(tool, struct perf_report, tool);
231 if (rep->show_threads) {
232 const char *name = evsel ? event_name(evsel) : "unknown";
233 perf_read_values_add_value(&rep->show_threads_values,
234 event->read.pid, event->read.tid,
240 dump_printf(": %d %d %s %" PRIu64 "\n", event->read.pid, event->read.tid,
241 evsel ? event_name(evsel) : "FAIL",
247 static int perf_report__setup_sample_type(struct perf_report *rep)
249 struct perf_session *self = rep->session;
251 if (!(self->sample_type & PERF_SAMPLE_CALLCHAIN)) {
252 if (sort__has_parent) {
253 ui__warning("Selected --sort parent, but no "
254 "callchain data. Did you call "
255 "'perf record' without -g?\n");
258 if (symbol_conf.use_callchain) {
259 ui__warning("Selected -g but no callchain data. Did "
260 "you call 'perf record' without -g?\n");
263 } else if (!rep->dont_use_callchains &&
264 callchain_param.mode != CHAIN_NONE &&
265 !symbol_conf.use_callchain) {
266 symbol_conf.use_callchain = true;
267 if (callchain_register_param(&callchain_param) < 0) {
268 ui__warning("Can't register callchain "
274 if (sort__branch_mode == 1) {
275 if (!(self->sample_type & PERF_SAMPLE_BRANCH_STACK)) {
276 fprintf(stderr, "selected -b but no branch data."
277 " Did you call perf record without"
286 extern volatile int session_done;
288 static void sig_handler(int sig __used)
293 static size_t hists__fprintf_nr_sample_events(struct hists *self,
294 const char *evname, FILE *fp)
298 unsigned long nr_events = self->stats.nr_events[PERF_RECORD_SAMPLE];
300 nr_events = convert_unit(nr_events, &unit);
301 ret = fprintf(fp, "# Events: %lu%c", nr_events, unit);
303 ret += fprintf(fp, " %s", evname);
304 return ret + fprintf(fp, "\n#\n");
307 static int perf_evlist__tty_browse_hists(struct perf_evlist *evlist,
308 struct perf_report *rep,
311 struct perf_evsel *pos;
313 list_for_each_entry(pos, &evlist->entries, node) {
314 struct hists *hists = &pos->hists;
315 const char *evname = event_name(pos);
317 hists__fprintf_nr_sample_events(hists, evname, stdout);
318 hists__fprintf(hists, NULL, false, true, 0, 0, stdout);
319 fprintf(stdout, "\n\n");
322 if (sort_order == default_sort_order &&
323 parent_pattern == default_parent_pattern) {
324 fprintf(stdout, "#\n# (%s)\n#\n", help);
326 if (rep->show_threads) {
327 bool style = !strcmp(rep->pretty_printing_style, "raw");
328 perf_read_values_display(stdout, &rep->show_threads_values,
330 perf_read_values_destroy(&rep->show_threads_values);
337 static int __cmd_report(struct perf_report *rep)
341 struct perf_session *session = rep->session;
342 struct perf_evsel *pos;
343 struct map *kernel_map;
344 struct kmap *kernel_kmap;
345 const char *help = "For a higher level overview, try: perf report --sort comm,dso";
347 signal(SIGINT, sig_handler);
350 ret = perf_session__cpu_bitmap(session, rep->cpu_list,
356 if (use_browser <= 0)
357 perf_session__fprintf_info(session, stdout, rep->show_full_info);
359 if (rep->show_threads)
360 perf_read_values_init(&rep->show_threads_values);
362 ret = perf_report__setup_sample_type(rep);
366 ret = perf_session__process_events(session, &rep->tool);
370 kernel_map = session->host_machine.vmlinux_maps[MAP__FUNCTION];
371 kernel_kmap = map__kmap(kernel_map);
372 if (kernel_map == NULL ||
373 (kernel_map->dso->hit &&
374 (kernel_kmap->ref_reloc_sym == NULL ||
375 kernel_kmap->ref_reloc_sym->addr == 0))) {
376 const struct dso *kdso = kernel_map->dso;
379 "Kernel address maps (/proc/{kallsyms,modules}) were restricted.\n\n"
380 "Check /proc/sys/kernel/kptr_restrict before running 'perf record'.\n\n%s\n\n"
381 "Samples in kernel modules can't be resolved as well.\n\n",
382 RB_EMPTY_ROOT(&kdso->symbols[MAP__FUNCTION]) ?
383 "As no suitable kallsyms nor vmlinux was found, kernel samples\n"
384 "can't be resolved." :
385 "If some relocation was applied (e.g. kexec) symbols may be misresolved.");
389 perf_session__fprintf_nr_events(session, stdout);
394 perf_session__fprintf(session, stdout);
397 perf_session__fprintf_dsos(session, stdout);
400 list_for_each_entry(pos, &session->evlist->entries, node) {
401 struct hists *hists = &pos->hists;
403 hists__collapse_resort(hists);
404 hists__output_resort(hists);
405 nr_samples += hists->stats.nr_events[PERF_RECORD_SAMPLE];
408 if (nr_samples == 0) {
409 ui__warning("The %s file has no samples!\n", session->filename);
413 if (use_browser > 0) {
414 perf_evlist__tui_browse_hists(session->evlist, help,
417 perf_evlist__tty_browse_hists(session->evlist, rep, help);
421 * Speed up the exit process, for large files this can
422 * take quite a while.
424 * XXX Enable this when using valgrind or if we ever
425 * librarize this command.
427 * Also experiment with obstacks to see how much speed
430 * perf_session__delete(session);
436 parse_callchain_opt(const struct option *opt, const char *arg, int unset)
438 struct perf_report *rep = (struct perf_report *)opt->value;
446 rep->dont_use_callchains = true;
450 symbol_conf.use_callchain = true;
455 tok = strtok((char *)arg, ",");
459 /* get the output mode */
460 if (!strncmp(tok, "graph", strlen(arg)))
461 callchain_param.mode = CHAIN_GRAPH_ABS;
463 else if (!strncmp(tok, "flat", strlen(arg)))
464 callchain_param.mode = CHAIN_FLAT;
466 else if (!strncmp(tok, "fractal", strlen(arg)))
467 callchain_param.mode = CHAIN_GRAPH_REL;
469 else if (!strncmp(tok, "none", strlen(arg))) {
470 callchain_param.mode = CHAIN_NONE;
471 symbol_conf.use_callchain = false;
479 /* get the min percentage */
480 tok = strtok(NULL, ",");
484 callchain_param.min_percent = strtod(tok, &endptr);
488 /* get the print limit */
489 tok2 = strtok(NULL, ",");
493 if (tok2[0] != 'c') {
494 callchain_param.print_limit = strtoul(tok2, &endptr, 0);
495 tok2 = strtok(NULL, ",");
500 /* get the call chain order */
501 if (!strcmp(tok2, "caller"))
502 callchain_param.order = ORDER_CALLER;
503 else if (!strcmp(tok2, "callee"))
504 callchain_param.order = ORDER_CALLEE;
508 if (callchain_register_param(&callchain_param) < 0) {
509 fprintf(stderr, "Can't register callchain params\n");
516 parse_branch_mode(const struct option *opt __used, const char *str __used, int unset)
518 sort__branch_mode = !unset;
522 int cmd_report(int argc, const char **argv, const char *prefix __used)
524 struct perf_session *session;
526 bool has_br_stack = false;
528 char callchain_default_opt[] = "fractal,0.5,callee";
529 const char * const report_usage[] = {
530 "perf report [<options>]",
533 struct perf_report report = {
535 .sample = process_sample_event,
536 .mmap = perf_event__process_mmap,
537 .comm = perf_event__process_comm,
538 .exit = perf_event__process_task,
539 .fork = perf_event__process_task,
540 .lost = perf_event__process_lost,
541 .read = process_read_event,
542 .attr = perf_event__process_attr,
543 .event_type = perf_event__process_event_type,
544 .tracing_data = perf_event__process_tracing_data,
545 .build_id = perf_event__process_build_id,
546 .ordered_samples = true,
547 .ordering_requires_timestamps = true,
549 .pretty_printing_style = "normal",
551 const struct option options[] = {
552 OPT_STRING('i', "input", &report.input_name, "file",
554 OPT_INCR('v', "verbose", &verbose,
555 "be more verbose (show symbol address, etc)"),
556 OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace,
557 "dump raw trace in ASCII"),
558 OPT_STRING('k', "vmlinux", &symbol_conf.vmlinux_name,
559 "file", "vmlinux pathname"),
560 OPT_STRING(0, "kallsyms", &symbol_conf.kallsyms_name,
561 "file", "kallsyms pathname"),
562 OPT_BOOLEAN('f', "force", &report.force, "don't complain, do it"),
563 OPT_BOOLEAN('m', "modules", &symbol_conf.use_modules,
564 "load module symbols - WARNING: use only with -k and LIVE kernel"),
565 OPT_BOOLEAN('n', "show-nr-samples", &symbol_conf.show_nr_samples,
566 "Show a column with the number of samples"),
567 OPT_BOOLEAN('T', "threads", &report.show_threads,
568 "Show per-thread event counters"),
569 OPT_STRING(0, "pretty", &report.pretty_printing_style, "key",
570 "pretty printing style key: normal raw"),
571 OPT_BOOLEAN(0, "tui", &report.use_tui, "Use the TUI interface"),
572 OPT_BOOLEAN(0, "stdio", &report.use_stdio,
573 "Use the stdio interface"),
574 OPT_STRING('s', "sort", &sort_order, "key[,key2...]",
575 "sort by key(s): pid, comm, dso, symbol, parent, dso_to,"
576 " dso_from, symbol_to, symbol_from, mispredict"),
577 OPT_BOOLEAN(0, "showcpuutilization", &symbol_conf.show_cpu_utilization,
578 "Show sample percentage for different cpu modes"),
579 OPT_STRING('p', "parent", &parent_pattern, "regex",
580 "regex filter to identify parent, see: '--sort parent'"),
581 OPT_BOOLEAN('x', "exclude-other", &symbol_conf.exclude_other,
582 "Only display entries with parent-match"),
583 OPT_CALLBACK_DEFAULT('g', "call-graph", &report, "output_type,min_percent[,print_limit],call_order",
584 "Display callchains using output_type (graph, flat, fractal, or none) , min percent threshold, optional print limit and callchain order. "
585 "Default: fractal,0.5,callee", &parse_callchain_opt, callchain_default_opt),
586 OPT_BOOLEAN('G', "inverted", &report.inverted_callchain,
587 "alias for inverted call graph"),
588 OPT_STRING('d', "dsos", &symbol_conf.dso_list_str, "dso[,dso...]",
589 "only consider symbols in these dsos"),
590 OPT_STRING('c', "comms", &symbol_conf.comm_list_str, "comm[,comm...]",
591 "only consider symbols in these comms"),
592 OPT_STRING('S', "symbols", &symbol_conf.sym_list_str, "symbol[,symbol...]",
593 "only consider these symbols"),
594 OPT_STRING('w', "column-widths", &symbol_conf.col_width_list_str,
596 "don't try to adjust column width, use these fixed values"),
597 OPT_STRING('t', "field-separator", &symbol_conf.field_sep, "separator",
598 "separator for columns, no spaces will be added between "
599 "columns '.' is reserved."),
600 OPT_BOOLEAN('U', "hide-unresolved", &report.hide_unresolved,
601 "Only display entries resolved to a symbol"),
602 OPT_STRING(0, "symfs", &symbol_conf.symfs, "directory",
603 "Look for files with symbols relative to this directory"),
604 OPT_STRING('C', "cpu", &report.cpu_list, "cpu",
605 "list of cpus to profile"),
606 OPT_BOOLEAN('I', "show-info", &report.show_full_info,
607 "Display extended information about perf.data file"),
608 OPT_BOOLEAN(0, "source", &symbol_conf.annotate_src,
609 "Interleave source code with assembly code (default)"),
610 OPT_BOOLEAN(0, "asm-raw", &symbol_conf.annotate_asm_raw,
611 "Display raw encoding of assembly instructions (default)"),
612 OPT_STRING('M', "disassembler-style", &disassembler_style, "disassembler style",
613 "Specify disassembler style (e.g. -M intel for intel syntax)"),
614 OPT_BOOLEAN(0, "show-total-period", &symbol_conf.show_total_period,
615 "Show a column with the sum of periods"),
616 OPT_CALLBACK_NOOPT('b', "branch-stack", &sort__branch_mode, "",
617 "use branch records for histogram filling", parse_branch_mode),
621 argc = parse_options(argc, argv, options, report_usage, 0);
623 if (report.use_stdio)
625 else if (report.use_tui)
628 if (report.inverted_callchain)
629 callchain_param.order = ORDER_CALLER;
631 if (!report.input_name || !strlen(report.input_name)) {
632 if (!fstat(STDIN_FILENO, &st) && S_ISFIFO(st.st_mode))
633 report.input_name = "-";
635 report.input_name = "perf.data";
637 session = perf_session__new(report.input_name, O_RDONLY,
638 report.force, false, &report.tool);
642 report.session = session;
644 has_br_stack = perf_header__has_feat(&session->header,
645 HEADER_BRANCH_STACK);
647 if (sort__branch_mode == -1 && has_br_stack)
648 sort__branch_mode = 1;
650 /* sort__branch_mode could be 0 if --no-branch-stack */
651 if (sort__branch_mode == 1) {
653 * if no sort_order is provided, then specify
654 * branch-mode specific order
656 if (sort_order == default_sort_order)
657 sort_order = "comm,dso_from,symbol_from,"
662 if (strcmp(report.input_name, "-") != 0) {
669 * Only in the newt browser we are doing integrated annotation,
670 * so don't allocate extra space that won't be used in the stdio
673 if (use_browser > 0) {
674 symbol_conf.priv_size = sizeof(struct annotation);
675 report.annotate_init = symbol__annotate_init;
677 * For searching by name on the "Browse map details".
678 * providing it only in verbose mode not to bloat too
679 * much struct symbol.
683 * XXX: Need to provide a less kludgy way to ask for
684 * more space per symbol, the u32 is for the index on
686 * See symbol__browser_index.
688 symbol_conf.priv_size += sizeof(u32);
689 symbol_conf.sort_by_name = true;
693 if (symbol__init() < 0)
696 setup_sorting(report_usage, options);
698 if (parent_pattern != default_parent_pattern) {
699 if (sort_dimension__add("parent") < 0)
703 * Only show the parent fields if we explicitly
704 * sort that way. If we only use parent machinery
705 * for filtering, we don't want it.
707 if (!strstr(sort_order, "parent"))
708 sort_parent.elide = 1;
710 symbol_conf.exclude_other = false;
713 * Any (unrecognized) arguments left?
716 usage_with_options(report_usage, options);
718 sort_entry__setup_elide(&sort_comm, symbol_conf.comm_list, "comm", stdout);
720 if (sort__branch_mode == 1) {
721 sort_entry__setup_elide(&sort_dso_from, symbol_conf.dso_from_list, "dso_from", stdout);
722 sort_entry__setup_elide(&sort_dso_to, symbol_conf.dso_to_list, "dso_to", stdout);
723 sort_entry__setup_elide(&sort_sym_from, symbol_conf.sym_from_list, "sym_from", stdout);
724 sort_entry__setup_elide(&sort_sym_to, symbol_conf.sym_to_list, "sym_to", stdout);
726 sort_entry__setup_elide(&sort_dso, symbol_conf.dso_list, "dso", stdout);
727 sort_entry__setup_elide(&sort_sym, symbol_conf.sym_list, "symbol", stdout);
730 ret = __cmd_report(&report);
732 perf_session__delete(session);