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;
57 const char *sym_hist_filter;
59 DECLARE_BITMAP(cpu_bitmap, MAX_NR_CPUS);
63 * Given one basic block:
73 * where the horizontal are the branches and the vertical is the executed
74 * block of instructions.
76 * We count, for each 'instruction', the number of blocks that covered it as
77 * well as count the ratio each branch is taken.
79 * We can do this without knowing the actual instruction stream by keeping
80 * track of the address ranges. We break down ranges such that there is no
81 * overlap and iterate from the start until the end.
83 * @acme: once we parse the objdump output _before_ processing the samples,
84 * we can easily fold the branch.cycles IPC bits in.
86 static void process_basic_block(struct addr_map_symbol *start,
87 struct addr_map_symbol *end,
88 struct branch_flags *flags)
90 struct symbol *sym = start->ms.sym;
91 struct annotation *notes = sym ? symbol__annotation(sym) : NULL;
92 struct block_range_iter iter;
93 struct block_range *entry;
96 * Sanity; NULL isn't executable and the CPU cannot execute backwards
98 if (!start->addr || start->addr > end->addr)
101 iter = block_range__create(start->addr, end->addr);
102 if (!block_range_iter__valid(&iter))
106 * First block in range is a branch target.
108 entry = block_range_iter(&iter);
109 assert(entry->is_target);
113 entry = block_range_iter(&iter);
119 notes->max_coverage = max(notes->max_coverage, entry->coverage);
121 } while (block_range_iter__next(&iter));
124 * Last block in rage is a branch.
126 entry = block_range_iter(&iter);
127 assert(entry->is_branch);
129 if (flags->predicted)
133 static void process_branch_stack(struct branch_stack *bs, struct addr_location *al,
134 struct perf_sample *sample)
136 struct addr_map_symbol *prev = NULL;
137 struct branch_info *bi;
143 bi = sample__resolve_bstack(sample, al);
147 for (i = bs->nr - 1; i >= 0; i--) {
149 * XXX filter against symbol
152 process_basic_block(prev, &bi[i].from, &bi[i].flags);
159 static int hist_iter__branch_callback(struct hist_entry_iter *iter,
160 struct addr_location *al __maybe_unused,
161 bool single __maybe_unused,
162 void *arg __maybe_unused)
164 struct hist_entry *he = iter->he;
165 struct branch_info *bi;
166 struct perf_sample *sample = iter->sample;
167 struct evsel *evsel = iter->evsel;
170 bi = he->branch_info;
171 err = addr_map_symbol__inc_samples(&bi->from, sample, evsel);
176 err = addr_map_symbol__inc_samples(&bi->to, sample, evsel);
182 static int process_branch_callback(struct evsel *evsel,
183 struct perf_sample *sample,
184 struct addr_location *al __maybe_unused,
185 struct perf_annotate *ann,
186 struct machine *machine)
188 struct hist_entry_iter iter = {
191 .add_entry_cb = hist_iter__branch_callback,
192 .hide_unresolved = symbol_conf.hide_unresolved,
193 .ops = &hist_iter_branch,
196 struct addr_location a;
198 if (machine__resolve(machine, &a, sample) < 0)
207 hist__account_cycles(sample->branch_stack, al, sample, false, NULL);
209 return hist_entry_iter__add(&iter, &a, PERF_MAX_STACK_DEPTH, ann);
212 static bool has_annotation(struct perf_annotate *ann)
214 return ui__has_annotation() || ann->use_stdio2;
217 static int evsel__add_sample(struct evsel *evsel, struct perf_sample *sample,
218 struct addr_location *al, struct perf_annotate *ann,
219 struct machine *machine)
221 struct hists *hists = evsel__hists(evsel);
222 struct hist_entry *he;
225 if ((!ann->has_br_stack || !has_annotation(ann)) &&
226 ann->sym_hist_filter != NULL &&
228 strcmp(ann->sym_hist_filter, al->sym->name) != 0)) {
229 /* We're only interested in a symbol named sym_hist_filter */
231 * FIXME: why isn't this done in the symbol_filter when loading
234 if (al->sym != NULL) {
235 rb_erase_cached(&al->sym->rb_node,
236 &al->map->dso->symbols);
237 symbol__delete(al->sym);
238 dso__reset_find_symbol_cache(al->map->dso);
244 * XXX filtered samples can still have branch entries pointing into our
245 * symbol and are missed.
247 process_branch_stack(sample->branch_stack, al, sample);
249 if (ann->has_br_stack && has_annotation(ann))
250 return process_branch_callback(evsel, sample, al, ann, machine);
252 he = hists__add_entry(hists, al, NULL, NULL, NULL, sample, true);
256 ret = hist_entry__inc_addr_samples(he, sample, evsel, al->addr);
257 hists__inc_nr_samples(hists, true);
261 static int process_sample_event(struct perf_tool *tool,
262 union perf_event *event,
263 struct perf_sample *sample,
265 struct machine *machine)
267 struct perf_annotate *ann = container_of(tool, struct perf_annotate, tool);
268 struct addr_location al;
271 if (machine__resolve(machine, &al, sample) < 0) {
272 pr_warning("problem processing %d event, skipping it.\n",
277 if (ann->cpu_list && !test_bit(sample->cpu, ann->cpu_bitmap))
281 evsel__add_sample(evsel, sample, &al, ann, machine)) {
282 pr_warning("problem incrementing symbol count, "
287 addr_location__put(&al);
291 static int process_feature_event(struct perf_session *session,
292 union perf_event *event)
294 if (event->feat.feat_id < HEADER_LAST_FEATURE)
295 return perf_event__process_feature(session, event);
299 static int hist_entry__tty_annotate(struct hist_entry *he,
301 struct perf_annotate *ann)
303 if (!ann->use_stdio2)
304 return symbol__tty_annotate(&he->ms, evsel, &ann->opts);
306 return symbol__tty_annotate2(&he->ms, evsel, &ann->opts);
309 static void hists__find_annotations(struct hists *hists,
311 struct perf_annotate *ann)
313 struct rb_node *nd = rb_first_cached(&hists->entries), *next;
317 struct hist_entry *he = rb_entry(nd, struct hist_entry, rb_node);
318 struct annotation *notes;
320 if (he->ms.sym == NULL || he->ms.map->dso->annotate_warned)
323 if (ann->sym_hist_filter &&
324 (strcmp(he->ms.sym->name, ann->sym_hist_filter) != 0))
327 notes = symbol__annotation(he->ms.sym);
328 if (notes->src == NULL) {
337 if (use_browser == 2) {
339 int (*annotate)(struct hist_entry *he,
341 struct hist_browser_timer *hbt);
343 annotate = dlsym(perf_gtk_handle,
344 "hist_entry__gtk_annotate");
345 if (annotate == NULL) {
346 ui__error("GTK browser not found!\n");
350 ret = annotate(he, evsel, NULL);
351 if (!ret || !ann->skip_missing)
354 /* skip missing symbols */
356 } else if (use_browser == 1) {
357 key = hist_entry__tui_annotate(he, evsel, NULL, &ann->opts);
361 if (!ann->skip_missing)
377 hist_entry__tty_annotate(he, evsel, ann);
383 static int __cmd_annotate(struct perf_annotate *ann)
386 struct perf_session *session = ann->session;
388 u64 total_nr_samples;
391 ret = perf_session__cpu_bitmap(session, ann->cpu_list,
397 if (!ann->opts.objdump_path) {
398 ret = perf_env__lookup_objdump(&session->header.env,
399 &ann->opts.objdump_path);
404 ret = perf_session__process_events(session);
409 perf_session__fprintf_nr_events(session, stdout, false);
410 evlist__fprintf_nr_events(session->evlist, stdout, false);
415 perf_session__fprintf(session, stdout);
418 perf_session__fprintf_dsos(session, stdout);
420 total_nr_samples = 0;
421 evlist__for_each_entry(session->evlist, pos) {
422 struct hists *hists = evsel__hists(pos);
423 u32 nr_samples = hists->stats.nr_samples;
425 if (nr_samples > 0) {
426 total_nr_samples += nr_samples;
427 hists__collapse_resort(hists, NULL);
428 /* Don't sort callchain */
429 evsel__reset_sample_bit(pos, CALLCHAIN);
430 evsel__output_resort(pos, NULL);
432 if (symbol_conf.event_group && !evsel__is_group_leader(pos))
435 hists__find_annotations(hists, pos, ann);
439 if (total_nr_samples == 0) {
440 ui__error("The %s data has no samples!\n", session->data->path);
444 if (use_browser == 2) {
445 void (*show_annotations)(void);
447 show_annotations = dlsym(perf_gtk_handle,
448 "perf_gtk__show_annotations");
449 if (show_annotations == NULL) {
450 ui__error("GTK browser not found!\n");
460 static const char * const annotate_usage[] = {
461 "perf annotate [<options>]",
465 int cmd_annotate(int argc, const char **argv)
467 struct perf_annotate annotate = {
469 .sample = process_sample_event,
470 .mmap = perf_event__process_mmap,
471 .mmap2 = perf_event__process_mmap2,
472 .comm = perf_event__process_comm,
473 .exit = perf_event__process_exit,
474 .fork = perf_event__process_fork,
475 .namespaces = perf_event__process_namespaces,
476 .attr = perf_event__process_attr,
477 .build_id = perf_event__process_build_id,
478 .tracing_data = perf_event__process_tracing_data,
479 .id_index = perf_event__process_id_index,
480 .auxtrace_info = perf_event__process_auxtrace_info,
481 .auxtrace = perf_event__process_auxtrace,
482 .feature = process_feature_event,
483 .ordered_events = true,
484 .ordering_requires_timestamps = true,
486 .opts = annotation__default_options,
488 struct perf_data data = {
489 .mode = PERF_DATA_MODE_READ,
491 struct itrace_synth_opts itrace_synth_opts = {
494 struct option options[] = {
495 OPT_STRING('i', "input", &input_name, "file",
497 OPT_STRING('d', "dsos", &symbol_conf.dso_list_str, "dso[,dso...]",
498 "only consider symbols in these dsos"),
499 OPT_STRING('s', "symbol", &annotate.sym_hist_filter, "symbol",
500 "symbol to annotate"),
501 OPT_BOOLEAN('f', "force", &data.force, "don't complain, do it"),
502 OPT_INCR('v', "verbose", &verbose,
503 "be more verbose (show symbol address, etc)"),
504 OPT_BOOLEAN('q', "quiet", &quiet, "do now show any message"),
505 OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace,
506 "dump raw trace in ASCII"),
507 OPT_BOOLEAN(0, "gtk", &annotate.use_gtk, "Use the GTK interface"),
508 #ifdef HAVE_SLANG_SUPPORT
509 OPT_BOOLEAN(0, "tui", &annotate.use_tui, "Use the TUI interface"),
511 OPT_BOOLEAN(0, "stdio", &annotate.use_stdio, "Use the stdio interface"),
512 OPT_BOOLEAN(0, "stdio2", &annotate.use_stdio2, "Use the stdio interface"),
513 OPT_BOOLEAN(0, "ignore-vmlinux", &symbol_conf.ignore_vmlinux,
514 "don't load vmlinux even if found"),
515 OPT_STRING('k', "vmlinux", &symbol_conf.vmlinux_name,
516 "file", "vmlinux pathname"),
517 OPT_BOOLEAN('m', "modules", &symbol_conf.use_modules,
518 "load module symbols - WARNING: use only with -k and LIVE kernel"),
519 OPT_BOOLEAN('l', "print-line", &annotate.opts.print_lines,
520 "print matching source lines (may be slow)"),
521 OPT_BOOLEAN('P', "full-paths", &annotate.opts.full_path,
522 "Don't shorten the displayed pathnames"),
523 OPT_BOOLEAN(0, "skip-missing", &annotate.skip_missing,
524 "Skip symbols that cannot be annotated"),
525 OPT_BOOLEAN_SET(0, "group", &symbol_conf.event_group,
527 "Show event group information together"),
528 OPT_STRING('C', "cpu", &annotate.cpu_list, "cpu", "list of cpus to profile"),
529 OPT_CALLBACK(0, "symfs", NULL, "directory",
530 "Look for files with symbols relative to this directory",
531 symbol__config_symfs),
532 OPT_BOOLEAN(0, "source", &annotate.opts.annotate_src,
533 "Interleave source code with assembly code (default)"),
534 OPT_BOOLEAN(0, "asm-raw", &annotate.opts.show_asm_raw,
535 "Display raw encoding of assembly instructions (default)"),
536 OPT_STRING('M', "disassembler-style", &annotate.opts.disassembler_style, "disassembler style",
537 "Specify disassembler style (e.g. -M intel for intel syntax)"),
538 OPT_STRING(0, "prefix", &annotate.opts.prefix, "prefix",
539 "Add prefix to source file path names in programs (with --prefix-strip)"),
540 OPT_STRING(0, "prefix-strip", &annotate.opts.prefix_strip, "N",
541 "Strip first N entries of source file path name in programs (with --prefix)"),
542 OPT_STRING(0, "objdump", &annotate.opts.objdump_path, "path",
543 "objdump binary to use for disassembly and annotations"),
544 OPT_BOOLEAN(0, "demangle", &symbol_conf.demangle,
545 "Enable symbol demangling"),
546 OPT_BOOLEAN(0, "demangle-kernel", &symbol_conf.demangle_kernel,
547 "Enable kernel symbol demangling"),
548 OPT_BOOLEAN(0, "group", &symbol_conf.event_group,
549 "Show event group information together"),
550 OPT_BOOLEAN(0, "show-total-period", &symbol_conf.show_total_period,
551 "Show a column with the sum of periods"),
552 OPT_BOOLEAN('n', "show-nr-samples", &symbol_conf.show_nr_samples,
553 "Show a column with the number of samples"),
554 OPT_CALLBACK_DEFAULT(0, "stdio-color", NULL, "mode",
555 "'always' (default), 'never' or 'auto' only applicable to --stdio mode",
556 stdio__config_color, "always"),
557 OPT_CALLBACK(0, "percent-type", &annotate.opts, "local-period",
558 "Set percent type local/global-period/hits",
559 annotate_parse_percent_type),
560 OPT_CALLBACK_OPTARG(0, "itrace", &itrace_synth_opts, NULL, "opts",
561 "Instruction Tracing options\n" ITRACE_HELP,
562 itrace_parse_synth_opts),
568 set_option_flag(options, 0, "show-total-period", PARSE_OPT_EXCLUSIVE);
569 set_option_flag(options, 0, "show-nr-samples", PARSE_OPT_EXCLUSIVE);
576 annotation_config__init(&annotate.opts);
578 argc = parse_options(argc, argv, options, annotate_usage, 0);
581 * Special case: if there's an argument left then assume that
582 * it's a symbol filter:
585 usage_with_options(annotate_usage, options);
587 annotate.sym_hist_filter = argv[0];
590 if (annotate_check_args(&annotate.opts) < 0)
593 if (symbol_conf.show_nr_samples && annotate.use_gtk) {
594 pr_err("--show-nr-samples is not available in --gtk mode at this time\n");
598 ret = symbol__validate_sym_arguments();
605 data.path = input_name;
607 annotate.session = perf_session__new(&data, &annotate.tool);
608 if (IS_ERR(annotate.session))
609 return PTR_ERR(annotate.session);
611 annotate.session->itrace_synth_opts = &itrace_synth_opts;
613 annotate.has_br_stack = perf_header__has_feat(&annotate.session->header,
614 HEADER_BRANCH_STACK);
616 if (annotate.group_set)
617 evlist__force_leader(annotate.session->evlist);
619 ret = symbol__annotation_init();
623 symbol_conf.try_vmlinux_path = true;
625 ret = symbol__init(&annotate.session->header.env);
629 if (annotate.use_stdio || annotate.use_stdio2)
631 #ifdef HAVE_SLANG_SUPPORT
632 else if (annotate.use_tui)
635 else if (annotate.use_gtk)
641 * Events of different processes may correspond to the same
642 * symbol, we do not care about the processes in annotate,
643 * set sort order to avoid repeated output.
645 sort_order = "dso,symbol";
648 * Set SORT_MODE__BRANCH so that annotate display IPC/Cycle
649 * if branch info is in perf data in TUI mode.
651 if ((use_browser == 1 || annotate.use_stdio2) && annotate.has_br_stack)
652 sort__mode = SORT_MODE__BRANCH;
654 if (setup_sorting(NULL) < 0)
655 usage_with_options(annotate_usage, options);
657 ret = __cmd_annotate(&annotate);
661 * Speed up the exit process, for large files this can
662 * take quite a while.
664 * XXX Enable this when using valgrind or if we ever
665 * librarize this command.
667 * Also experiment with obstacks to see how much speed
670 * perf_session__delete(session);