1 // SPDX-License-Identifier: GPL-2.0
5 * Builtin annotate 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.
11 #include "util/color.h"
12 #include <linux/list.h>
13 #include "util/cache.h"
14 #include <linux/rbtree.h>
15 #include <linux/zalloc.h>
16 #include "util/symbol.h"
19 #include "util/debug.h"
21 #include "util/evlist.h"
22 #include "util/evsel.h"
23 #include "util/annotate.h"
24 #include "util/event.h"
25 #include <subcmd/parse-options.h>
26 #include "util/parse-events.h"
27 #include "util/sort.h"
28 #include "util/hist.h"
30 #include "util/machine.h"
32 #include "util/session.h"
33 #include "util/tool.h"
34 #include "util/data.h"
35 #include "arch/common.h"
36 #include "util/block-range.h"
37 #include "util/map_symbol.h"
38 #include "util/branch.h"
42 #include <linux/bitmap.h>
43 #include <linux/err.h>
45 struct perf_annotate {
46 struct perf_tool tool;
47 struct perf_session *session;
48 struct annotation_options opts;
49 #ifdef HAVE_SLANG_SUPPORT
52 bool use_stdio, use_stdio2;
53 #ifdef HAVE_GTK2_SUPPORT
60 const char *sym_hist_filter;
62 DECLARE_BITMAP(cpu_bitmap, MAX_NR_CPUS);
66 * Given one basic block:
76 * where the horizontal are the branches and the vertical is the executed
77 * block of instructions.
79 * We count, for each 'instruction', the number of blocks that covered it as
80 * well as count the ratio each branch is taken.
82 * We can do this without knowing the actual instruction stream by keeping
83 * track of the address ranges. We break down ranges such that there is no
84 * overlap and iterate from the start until the end.
86 * @acme: once we parse the objdump output _before_ processing the samples,
87 * we can easily fold the branch.cycles IPC bits in.
89 static void process_basic_block(struct addr_map_symbol *start,
90 struct addr_map_symbol *end,
91 struct branch_flags *flags)
93 struct symbol *sym = start->ms.sym;
94 struct annotation *notes = sym ? symbol__annotation(sym) : NULL;
95 struct block_range_iter iter;
96 struct block_range *entry;
99 * Sanity; NULL isn't executable and the CPU cannot execute backwards
101 if (!start->addr || start->addr > end->addr)
104 iter = block_range__create(start->addr, end->addr);
105 if (!block_range_iter__valid(&iter))
109 * First block in range is a branch target.
111 entry = block_range_iter(&iter);
112 assert(entry->is_target);
116 entry = block_range_iter(&iter);
122 notes->max_coverage = max(notes->max_coverage, entry->coverage);
124 } while (block_range_iter__next(&iter));
127 * Last block in rage is a branch.
129 entry = block_range_iter(&iter);
130 assert(entry->is_branch);
132 if (flags->predicted)
136 static void process_branch_stack(struct branch_stack *bs, struct addr_location *al,
137 struct perf_sample *sample)
139 struct addr_map_symbol *prev = NULL;
140 struct branch_info *bi;
146 bi = sample__resolve_bstack(sample, al);
150 for (i = bs->nr - 1; i >= 0; i--) {
152 * XXX filter against symbol
155 process_basic_block(prev, &bi[i].from, &bi[i].flags);
162 static int hist_iter__branch_callback(struct hist_entry_iter *iter,
163 struct addr_location *al __maybe_unused,
164 bool single __maybe_unused,
165 void *arg __maybe_unused)
167 struct hist_entry *he = iter->he;
168 struct branch_info *bi;
169 struct perf_sample *sample = iter->sample;
170 struct evsel *evsel = iter->evsel;
173 bi = he->branch_info;
174 err = addr_map_symbol__inc_samples(&bi->from, sample, evsel);
179 err = addr_map_symbol__inc_samples(&bi->to, sample, evsel);
185 static int process_branch_callback(struct evsel *evsel,
186 struct perf_sample *sample,
187 struct addr_location *al __maybe_unused,
188 struct perf_annotate *ann,
189 struct machine *machine)
191 struct hist_entry_iter iter = {
194 .add_entry_cb = hist_iter__branch_callback,
195 .hide_unresolved = symbol_conf.hide_unresolved,
196 .ops = &hist_iter_branch,
199 struct addr_location a;
201 if (machine__resolve(machine, &a, sample) < 0)
210 hist__account_cycles(sample->branch_stack, al, sample, false, NULL);
212 return hist_entry_iter__add(&iter, &a, PERF_MAX_STACK_DEPTH, ann);
215 static bool has_annotation(struct perf_annotate *ann)
217 return ui__has_annotation() || ann->use_stdio2;
220 static int evsel__add_sample(struct evsel *evsel, struct perf_sample *sample,
221 struct addr_location *al, struct perf_annotate *ann,
222 struct machine *machine)
224 struct hists *hists = evsel__hists(evsel);
225 struct hist_entry *he;
228 if ((!ann->has_br_stack || !has_annotation(ann)) &&
229 ann->sym_hist_filter != NULL &&
231 strcmp(ann->sym_hist_filter, al->sym->name) != 0)) {
232 /* We're only interested in a symbol named sym_hist_filter */
234 * FIXME: why isn't this done in the symbol_filter when loading
237 if (al->sym != NULL) {
238 rb_erase_cached(&al->sym->rb_node,
239 &al->map->dso->symbols);
240 symbol__delete(al->sym);
241 dso__reset_find_symbol_cache(al->map->dso);
247 * XXX filtered samples can still have branch entries pointing into our
248 * symbol and are missed.
250 process_branch_stack(sample->branch_stack, al, sample);
252 if (ann->has_br_stack && has_annotation(ann))
253 return process_branch_callback(evsel, sample, al, ann, machine);
255 he = hists__add_entry(hists, al, NULL, NULL, NULL, sample, true);
259 ret = hist_entry__inc_addr_samples(he, sample, evsel, al->addr);
260 hists__inc_nr_samples(hists, true);
264 static int process_sample_event(struct perf_tool *tool,
265 union perf_event *event,
266 struct perf_sample *sample,
268 struct machine *machine)
270 struct perf_annotate *ann = container_of(tool, struct perf_annotate, tool);
271 struct addr_location al;
274 if (machine__resolve(machine, &al, sample) < 0) {
275 pr_warning("problem processing %d event, skipping it.\n",
280 if (ann->cpu_list && !test_bit(sample->cpu, ann->cpu_bitmap))
284 evsel__add_sample(evsel, sample, &al, ann, machine)) {
285 pr_warning("problem incrementing symbol count, "
290 addr_location__put(&al);
294 static int process_feature_event(struct perf_session *session,
295 union perf_event *event)
297 if (event->feat.feat_id < HEADER_LAST_FEATURE)
298 return perf_event__process_feature(session, event);
302 static int hist_entry__tty_annotate(struct hist_entry *he,
304 struct perf_annotate *ann)
306 if (!ann->use_stdio2)
307 return symbol__tty_annotate(&he->ms, evsel, &ann->opts);
309 return symbol__tty_annotate2(&he->ms, evsel, &ann->opts);
312 static void hists__find_annotations(struct hists *hists,
314 struct perf_annotate *ann)
316 struct rb_node *nd = rb_first_cached(&hists->entries), *next;
320 struct hist_entry *he = rb_entry(nd, struct hist_entry, rb_node);
321 struct annotation *notes;
323 if (he->ms.sym == NULL || he->ms.map->dso->annotate_warned)
326 if (ann->sym_hist_filter &&
327 (strcmp(he->ms.sym->name, ann->sym_hist_filter) != 0))
330 if (ann->min_percent) {
332 u64 total = hists__total_period(hists);
335 percent = 100.0 * he->stat.period / total;
337 if (percent < ann->min_percent)
341 notes = symbol__annotation(he->ms.sym);
342 if (notes->src == NULL) {
351 if (use_browser == 2) {
353 int (*annotate)(struct hist_entry *he,
355 struct hist_browser_timer *hbt);
357 annotate = dlsym(perf_gtk_handle,
358 "hist_entry__gtk_annotate");
359 if (annotate == NULL) {
360 ui__error("GTK browser not found!\n");
364 ret = annotate(he, evsel, NULL);
365 if (!ret || !ann->skip_missing)
368 /* skip missing symbols */
370 } else if (use_browser == 1) {
371 key = hist_entry__tui_annotate(he, evsel, NULL, &ann->opts);
375 if (!ann->skip_missing)
391 hist_entry__tty_annotate(he, evsel, ann);
397 static int __cmd_annotate(struct perf_annotate *ann)
400 struct perf_session *session = ann->session;
402 u64 total_nr_samples;
405 ret = perf_session__cpu_bitmap(session, ann->cpu_list,
411 if (!ann->opts.objdump_path) {
412 ret = perf_env__lookup_objdump(&session->header.env,
413 &ann->opts.objdump_path);
418 ret = perf_session__process_events(session);
423 perf_session__fprintf_nr_events(session, stdout, false);
424 evlist__fprintf_nr_events(session->evlist, stdout, false);
429 perf_session__fprintf(session, stdout);
432 perf_session__fprintf_dsos(session, stdout);
434 total_nr_samples = 0;
435 evlist__for_each_entry(session->evlist, pos) {
436 struct hists *hists = evsel__hists(pos);
437 u32 nr_samples = hists->stats.nr_samples;
439 if (nr_samples > 0) {
440 total_nr_samples += nr_samples;
441 hists__collapse_resort(hists, NULL);
442 /* Don't sort callchain */
443 evsel__reset_sample_bit(pos, CALLCHAIN);
444 evsel__output_resort(pos, NULL);
446 if (symbol_conf.event_group && !evsel__is_group_leader(pos))
449 hists__find_annotations(hists, pos, ann);
453 if (total_nr_samples == 0) {
454 ui__error("The %s data has no samples!\n", session->data->path);
458 if (use_browser == 2) {
459 void (*show_annotations)(void);
461 show_annotations = dlsym(perf_gtk_handle,
462 "perf_gtk__show_annotations");
463 if (show_annotations == NULL) {
464 ui__error("GTK browser not found!\n");
474 static int parse_percent_limit(const struct option *opt, const char *str,
475 int unset __maybe_unused)
477 struct perf_annotate *ann = opt->value;
478 double pcnt = strtof(str, NULL);
480 ann->min_percent = pcnt;
484 static const char * const annotate_usage[] = {
485 "perf annotate [<options>]",
489 int cmd_annotate(int argc, const char **argv)
491 struct perf_annotate annotate = {
493 .sample = process_sample_event,
494 .mmap = perf_event__process_mmap,
495 .mmap2 = perf_event__process_mmap2,
496 .comm = perf_event__process_comm,
497 .exit = perf_event__process_exit,
498 .fork = perf_event__process_fork,
499 .namespaces = perf_event__process_namespaces,
500 .attr = perf_event__process_attr,
501 .build_id = perf_event__process_build_id,
502 #ifdef HAVE_LIBTRACEEVENT
503 .tracing_data = perf_event__process_tracing_data,
505 .id_index = perf_event__process_id_index,
506 .auxtrace_info = perf_event__process_auxtrace_info,
507 .auxtrace = perf_event__process_auxtrace,
508 .feature = process_feature_event,
509 .ordered_events = true,
510 .ordering_requires_timestamps = true,
512 .opts = annotation__default_options,
514 struct perf_data data = {
515 .mode = PERF_DATA_MODE_READ,
517 struct itrace_synth_opts itrace_synth_opts = {
520 struct option options[] = {
521 OPT_STRING('i', "input", &input_name, "file",
523 OPT_STRING('d', "dsos", &symbol_conf.dso_list_str, "dso[,dso...]",
524 "only consider symbols in these dsos"),
525 OPT_STRING('s', "symbol", &annotate.sym_hist_filter, "symbol",
526 "symbol to annotate"),
527 OPT_BOOLEAN('f', "force", &data.force, "don't complain, do it"),
528 OPT_INCR('v', "verbose", &verbose,
529 "be more verbose (show symbol address, etc)"),
530 OPT_BOOLEAN('q', "quiet", &quiet, "do now show any warnings or messages"),
531 OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace,
532 "dump raw trace in ASCII"),
533 #ifdef HAVE_GTK2_SUPPORT
534 OPT_BOOLEAN(0, "gtk", &annotate.use_gtk, "Use the GTK interface"),
536 #ifdef HAVE_SLANG_SUPPORT
537 OPT_BOOLEAN(0, "tui", &annotate.use_tui, "Use the TUI interface"),
539 OPT_BOOLEAN(0, "stdio", &annotate.use_stdio, "Use the stdio interface"),
540 OPT_BOOLEAN(0, "stdio2", &annotate.use_stdio2, "Use the stdio interface"),
541 OPT_BOOLEAN(0, "ignore-vmlinux", &symbol_conf.ignore_vmlinux,
542 "don't load vmlinux even if found"),
543 OPT_STRING('k', "vmlinux", &symbol_conf.vmlinux_name,
544 "file", "vmlinux pathname"),
545 OPT_BOOLEAN('m', "modules", &symbol_conf.use_modules,
546 "load module symbols - WARNING: use only with -k and LIVE kernel"),
547 OPT_BOOLEAN('l', "print-line", &annotate.opts.print_lines,
548 "print matching source lines (may be slow)"),
549 OPT_BOOLEAN('P', "full-paths", &annotate.opts.full_path,
550 "Don't shorten the displayed pathnames"),
551 OPT_BOOLEAN(0, "skip-missing", &annotate.skip_missing,
552 "Skip symbols that cannot be annotated"),
553 OPT_BOOLEAN_SET(0, "group", &symbol_conf.event_group,
555 "Show event group information together"),
556 OPT_STRING('C', "cpu", &annotate.cpu_list, "cpu", "list of cpus to profile"),
557 OPT_CALLBACK(0, "symfs", NULL, "directory",
558 "Look for files with symbols relative to this directory",
559 symbol__config_symfs),
560 OPT_BOOLEAN(0, "source", &annotate.opts.annotate_src,
561 "Interleave source code with assembly code (default)"),
562 OPT_BOOLEAN(0, "asm-raw", &annotate.opts.show_asm_raw,
563 "Display raw encoding of assembly instructions (default)"),
564 OPT_STRING('M', "disassembler-style", &annotate.opts.disassembler_style, "disassembler style",
565 "Specify disassembler style (e.g. -M intel for intel syntax)"),
566 OPT_STRING(0, "prefix", &annotate.opts.prefix, "prefix",
567 "Add prefix to source file path names in programs (with --prefix-strip)"),
568 OPT_STRING(0, "prefix-strip", &annotate.opts.prefix_strip, "N",
569 "Strip first N entries of source file path name in programs (with --prefix)"),
570 OPT_STRING(0, "objdump", &annotate.opts.objdump_path, "path",
571 "objdump binary to use for disassembly and annotations"),
572 OPT_BOOLEAN(0, "demangle", &symbol_conf.demangle,
573 "Enable symbol demangling"),
574 OPT_BOOLEAN(0, "demangle-kernel", &symbol_conf.demangle_kernel,
575 "Enable kernel symbol demangling"),
576 OPT_BOOLEAN(0, "group", &symbol_conf.event_group,
577 "Show event group information together"),
578 OPT_BOOLEAN(0, "show-total-period", &symbol_conf.show_total_period,
579 "Show a column with the sum of periods"),
580 OPT_BOOLEAN('n', "show-nr-samples", &symbol_conf.show_nr_samples,
581 "Show a column with the number of samples"),
582 OPT_CALLBACK_DEFAULT(0, "stdio-color", NULL, "mode",
583 "'always' (default), 'never' or 'auto' only applicable to --stdio mode",
584 stdio__config_color, "always"),
585 OPT_CALLBACK(0, "percent-type", &annotate.opts, "local-period",
586 "Set percent type local/global-period/hits",
587 annotate_parse_percent_type),
588 OPT_CALLBACK(0, "percent-limit", &annotate, "percent",
589 "Don't show entries under that percent", parse_percent_limit),
590 OPT_CALLBACK_OPTARG(0, "itrace", &itrace_synth_opts, NULL, "opts",
591 "Instruction Tracing options\n" ITRACE_HELP,
592 itrace_parse_synth_opts),
598 set_option_flag(options, 0, "show-total-period", PARSE_OPT_EXCLUSIVE);
599 set_option_flag(options, 0, "show-nr-samples", PARSE_OPT_EXCLUSIVE);
606 annotation_config__init(&annotate.opts);
608 argc = parse_options(argc, argv, options, annotate_usage, 0);
611 * Special case: if there's an argument left then assume that
612 * it's a symbol filter:
615 usage_with_options(annotate_usage, options);
617 annotate.sym_hist_filter = argv[0];
620 if (annotate_check_args(&annotate.opts) < 0)
623 #ifdef HAVE_GTK2_SUPPORT
624 if (symbol_conf.show_nr_samples && annotate.use_gtk) {
625 pr_err("--show-nr-samples is not available in --gtk mode at this time\n");
630 ret = symbol__validate_sym_arguments();
637 data.path = input_name;
639 annotate.session = perf_session__new(&data, &annotate.tool);
640 if (IS_ERR(annotate.session))
641 return PTR_ERR(annotate.session);
643 annotate.session->itrace_synth_opts = &itrace_synth_opts;
645 annotate.has_br_stack = perf_header__has_feat(&annotate.session->header,
646 HEADER_BRANCH_STACK);
648 if (annotate.group_set)
649 evlist__force_leader(annotate.session->evlist);
651 ret = symbol__annotation_init();
655 symbol_conf.try_vmlinux_path = true;
657 ret = symbol__init(&annotate.session->header.env);
661 if (annotate.use_stdio || annotate.use_stdio2)
663 #ifdef HAVE_SLANG_SUPPORT
664 else if (annotate.use_tui)
667 #ifdef HAVE_GTK2_SUPPORT
668 else if (annotate.use_gtk)
675 * Events of different processes may correspond to the same
676 * symbol, we do not care about the processes in annotate,
677 * set sort order to avoid repeated output.
679 sort_order = "dso,symbol";
682 * Set SORT_MODE__BRANCH so that annotate display IPC/Cycle
683 * if branch info is in perf data in TUI mode.
685 if ((use_browser == 1 || annotate.use_stdio2) && annotate.has_br_stack)
686 sort__mode = SORT_MODE__BRANCH;
688 if (setup_sorting(NULL) < 0)
689 usage_with_options(annotate_usage, options);
691 ret = __cmd_annotate(&annotate);
695 * Speed up the exit process, for large files this can
696 * take quite a while.
698 * XXX Enable this when using valgrind or if we ever
699 * librarize this command.
701 * Also experiment with obstacks to see how much speed
704 * perf_session__delete(session);