tools/perf/builtin-report.c

   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
  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"
  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
  29 #include "util/parse-options.h"
  30 #include "util/parse-events.h"
  31
  32 #include "util/thread.h"
  33 #include "util/sort.h"
  34 #include "util/hist.h"
  35
  36 #include <linux/bitmap.h>
  37
  38 static char             const *input_name = "perf.data";
  39
  40 static bool             force, use_tui, use_stdio;
  41 static bool             hide_unresolved;
  42 static bool             dont_use_callchains;
  43 static bool             show_full_info;
  44
  45 static bool             show_threads;
  46 static struct perf_read_values  show_threads_values;
  47
  48 static const char       default_pretty_printing_style[] = "normal";
  49 static const char       *pretty_printing_style = default_pretty_printing_style;
  50
  51 static char             callchain_default_opt[] = "fractal,0.5,callee";
  52 static bool             inverted_callchain;
  53 static symbol_filter_t  annotate_init;
  54
  55 static const char       *cpu_list;
  56 static DECLARE_BITMAP(cpu_bitmap, MAX_NR_CPUS);
  57
  58 static int perf_session__add_hist_entry(struct perf_session *session,
  59                                         struct addr_location *al,
  60                                         struct perf_sample *sample,
  61                                         struct perf_evsel *evsel)
  62 {
  63         struct symbol *parent = NULL;
  64         int err = 0;
  65         struct hist_entry *he;
  66
  67         if ((sort__has_parent || symbol_conf.use_callchain) && sample->callchain) {
  68                 err = perf_session__resolve_callchain(session, al->thread,
  69                                                       sample->callchain, &parent);
  70                 if (err)
  71                         return err;
  72         }
  73
  74         he = __hists__add_entry(&evsel->hists, al, parent, sample->period);
  75         if (he == NULL)
  76                 return -ENOMEM;
  77
  78         if (symbol_conf.use_callchain) {
  79                 err = callchain_append(he->callchain, &session->callchain_cursor,
  80                                        sample->period);
  81                 if (err)
  82                         return err;
  83         }
  84         /*
  85          * Only in the newt browser we are doing integrated annotation,
  86          * so we don't allocated the extra space needed because the stdio
  87          * code will not use it.
  88          */
  89         if (al->sym != NULL && use_browser > 0) {
  90                 struct annotation *notes = symbol__annotation(he->ms.sym);
  91
  92                 assert(evsel != NULL);
  93
  94                 err = -ENOMEM;
  95                 if (notes->src == NULL &&
  96                     symbol__alloc_hist(he->ms.sym, session->evlist->nr_entries) < 0)
  97                         goto out;
  98
  99                 err = hist_entry__inc_addr_samples(he, evsel->idx, al->addr);
 100         }
 101
 102         evsel->hists.stats.total_period += sample->period;
 103         hists__inc_nr_events(&evsel->hists, PERF_RECORD_SAMPLE);
 104 out:
 105         return err;
 106 }
 107
 108
 109 static int process_sample_event(union perf_event *event,
 110                                 struct perf_sample *sample,
 111                                 struct perf_evsel *evsel,
 112                                 struct perf_session *session)
 113 {
 114         struct addr_location al;
 115
 116         if (perf_event__preprocess_sample(event, session, &al, sample,
 117                                           annotate_init) < 0) {
 118                 fprintf(stderr, "problem processing %d event, skipping it.\n",
 119                         event->header.type);
 120                 return -1;
 121         }
 122
 123         if (al.filtered || (hide_unresolved && al.sym == NULL))
 124                 return 0;
 125
 126         if (cpu_list && !test_bit(sample->cpu, cpu_bitmap))
 127                 return 0;
 128
 129         if (al.map != NULL)
 130                 al.map->dso->hit = 1;
 131
 132         if (perf_session__add_hist_entry(session, &al, sample, evsel)) {
 133                 pr_debug("problem incrementing symbol period, skipping event\n");
 134                 return -1;
 135         }
 136
 137         return 0;
 138 }
 139
 140 static int process_read_event(union perf_event *event,
 141                               struct perf_sample *sample __used,
 142                               struct perf_session *session)
 143 {
 144         struct perf_evsel *evsel = perf_evlist__id2evsel(session->evlist,
 145                                                          event->read.id);
 146         if (show_threads) {
 147                 const char *name = evsel ? event_name(evsel) : "unknown";
 148                 perf_read_values_add_value(&show_threads_values,
 149                                            event->read.pid, event->read.tid,
 150                                            event->read.id,
 151                                            name,
 152                                            event->read.value);
 153         }
 154
 155         dump_printf(": %d %d %s %" PRIu64 "\n", event->read.pid, event->read.tid,
 156                     evsel ? event_name(evsel) : "FAIL",
 157                     event->read.value);
 158
 159         return 0;
 160 }
 161
 162 static int perf_session__setup_sample_type(struct perf_session *self)
 163 {
 164         if (!(self->sample_type & PERF_SAMPLE_CALLCHAIN)) {
 165                 if (sort__has_parent) {
 166                         ui__warning("Selected --sort parent, but no "
 167                                     "callchain data. Did you call "
 168                                     "'perf record' without -g?\n");
 169                         return -EINVAL;
 170                 }
 171                 if (symbol_conf.use_callchain) {
 172                         ui__warning("Selected -g but no callchain data. Did "
 173                                     "you call 'perf record' without -g?\n");
 174                         return -1;
 175                 }
 176         } else if (!dont_use_callchains && callchain_param.mode != CHAIN_NONE &&
 177                    !symbol_conf.use_callchain) {
 178                         symbol_conf.use_callchain = true;
 179                         if (callchain_register_param(&callchain_param) < 0) {
 180                                 ui__warning("Can't register callchain "
 181                                             "params.\n");
 182                                 return -EINVAL;
 183                         }
 184         }
 185
 186         return 0;
 187 }
 188
 189 static struct perf_event_ops event_ops = {
 190         .sample          = process_sample_event,
 191         .mmap            = perf_event__process_mmap,
 192         .comm            = perf_event__process_comm,
 193         .exit            = perf_event__process_task,
 194         .fork            = perf_event__process_task,
 195         .lost            = perf_event__process_lost,
 196         .read            = process_read_event,
 197         .attr            = perf_event__process_attr,
 198         .event_type      = perf_event__process_event_type,
 199         .tracing_data    = perf_event__process_tracing_data,
 200         .build_id        = perf_event__process_build_id,
 201         .ordered_samples = true,
 202         .ordering_requires_timestamps = true,
 203 };
 204
 205 extern volatile int session_done;
 206
 207 static void sig_handler(int sig __used)
 208 {
 209         session_done = 1;
 210 }
 211
 212 static size_t hists__fprintf_nr_sample_events(struct hists *self,
 213                                               const char *evname, FILE *fp)
 214 {
 215         size_t ret;
 216         char unit;
 217         unsigned long nr_events = self->stats.nr_events[PERF_RECORD_SAMPLE];
 218
 219         nr_events = convert_unit(nr_events, &unit);
 220         ret = fprintf(fp, "# Events: %lu%c", nr_events, unit);
 221         if (evname != NULL)
 222                 ret += fprintf(fp, " %s", evname);
 223         return ret + fprintf(fp, "\n#\n");
 224 }
 225
 226 static int perf_evlist__tty_browse_hists(struct perf_evlist *evlist,
 227                                          const char *help)
 228 {
 229         struct perf_evsel *pos;
 230
 231         list_for_each_entry(pos, &evlist->entries, node) {
 232                 struct hists *hists = &pos->hists;
 233                 const char *evname = event_name(pos);
 234
 235                 hists__fprintf_nr_sample_events(hists, evname, stdout);
 236                 hists__fprintf(hists, NULL, false, true, 0, 0, stdout);
 237                 fprintf(stdout, "\n\n");
 238         }
 239
 240         if (sort_order == default_sort_order &&
 241             parent_pattern == default_parent_pattern) {
 242                 fprintf(stdout, "#\n# (%s)\n#\n", help);
 243
 244                 if (show_threads) {
 245                         bool style = !strcmp(pretty_printing_style, "raw");
 246                         perf_read_values_display(stdout, &show_threads_values,
 247                                                  style);
 248                         perf_read_values_destroy(&show_threads_values);
 249                 }
 250         }
 251
 252         return 0;
 253 }
 254
 255 static int __cmd_report(void)
 256 {
 257         int ret = -EINVAL;
 258         u64 nr_samples;
 259         struct perf_session *session;
 260         struct perf_evsel *pos;
 261         struct map *kernel_map;
 262         struct kmap *kernel_kmap;
 263         const char *help = "For a higher level overview, try: perf report --sort comm,dso";
 264
 265         signal(SIGINT, sig_handler);
 266
 267         session = perf_session__new(input_name, O_RDONLY, force, false, &event_ops);
 268         if (session == NULL)
 269                 return -ENOMEM;
 270
 271         if (cpu_list) {
 272                 ret = perf_session__cpu_bitmap(session, cpu_list, cpu_bitmap);
 273                 if (ret)
 274                         goto out_delete;
 275         }
 276
 277         if (use_browser <= 0)
 278                 perf_session__fprintf_info(session, stdout, show_full_info);
 279
 280         if (show_threads)
 281                 perf_read_values_init(&show_threads_values);
 282
 283         ret = perf_session__setup_sample_type(session);
 284         if (ret)
 285                 goto out_delete;
 286
 287         ret = perf_session__process_events(session, &event_ops);
 288         if (ret)
 289                 goto out_delete;
 290
 291         kernel_map = session->host_machine.vmlinux_maps[MAP__FUNCTION];
 292         kernel_kmap = map__kmap(kernel_map);
 293         if (kernel_map == NULL ||
 294             (kernel_map->dso->hit &&
 295              (kernel_kmap->ref_reloc_sym == NULL ||
 296               kernel_kmap->ref_reloc_sym->addr == 0))) {
 297                 const struct dso *kdso = kernel_map->dso;
 298
 299                 ui__warning(
 300 "Kernel address maps (/proc/{kallsyms,modules}) were restricted.\n\n"
 301 "Check /proc/sys/kernel/kptr_restrict before running 'perf record'.\n\n%s\n\n"
 302 "Samples in kernel modules can't be resolved as well.\n\n",
 303                             RB_EMPTY_ROOT(&kdso->symbols[MAP__FUNCTION]) ?
 304 "As no suitable kallsyms nor vmlinux was found, kernel samples\n"
 305 "can't be resolved." :
 306 "If some relocation was applied (e.g. kexec) symbols may be misresolved.");
 307         }
 308
 309         if (dump_trace) {
 310                 perf_session__fprintf_nr_events(session, stdout);
 311                 goto out_delete;
 312         }
 313
 314         if (verbose > 3)
 315                 perf_session__fprintf(session, stdout);
 316
 317         if (verbose > 2)
 318                 perf_session__fprintf_dsos(session, stdout);
 319
 320         nr_samples = 0;
 321         list_for_each_entry(pos, &session->evlist->entries, node) {
 322                 struct hists *hists = &pos->hists;
 323
 324                 hists__collapse_resort(hists);
 325                 hists__output_resort(hists);
 326                 nr_samples += hists->stats.nr_events[PERF_RECORD_SAMPLE];
 327         }
 328
 329         if (nr_samples == 0) {
 330                 ui__warning("The %s file has no samples!\n", input_name);
 331                 goto out_delete;
 332         }
 333
 334         if (use_browser > 0) {
 335                 perf_evlist__tui_browse_hists(session->evlist, help,
 336                                               NULL, NULL, 0);
 337         } else
 338                 perf_evlist__tty_browse_hists(session->evlist, help);
 339
 340 out_delete:
 341         /*
 342          * Speed up the exit process, for large files this can
 343          * take quite a while.
 344          *
 345          * XXX Enable this when using valgrind or if we ever
 346          * librarize this command.
 347          *
 348          * Also experiment with obstacks to see how much speed
 349          * up we'll get here.
 350          *
 351          * perf_session__delete(session);
 352          */
 353         return ret;
 354 }
 355
 356 static int
 357 parse_callchain_opt(const struct option *opt __used, const char *arg,
 358                     int unset)
 359 {
 360         char *tok, *tok2;
 361         char *endptr;
 362
 363         /*
 364          * --no-call-graph
 365          */
 366         if (unset) {
 367                 dont_use_callchains = true;
 368                 return 0;
 369         }
 370
 371         symbol_conf.use_callchain = true;
 372
 373         if (!arg)
 374                 return 0;
 375
 376         tok = strtok((char *)arg, ",");
 377         if (!tok)
 378                 return -1;
 379
 380         /* get the output mode */
 381         if (!strncmp(tok, "graph", strlen(arg)))
 382                 callchain_param.mode = CHAIN_GRAPH_ABS;
 383
 384         else if (!strncmp(tok, "flat", strlen(arg)))
 385                 callchain_param.mode = CHAIN_FLAT;
 386
 387         else if (!strncmp(tok, "fractal", strlen(arg)))
 388                 callchain_param.mode = CHAIN_GRAPH_REL;
 389
 390         else if (!strncmp(tok, "none", strlen(arg))) {
 391                 callchain_param.mode = CHAIN_NONE;
 392                 symbol_conf.use_callchain = false;
 393
 394                 return 0;
 395         }
 396
 397         else
 398                 return -1;
 399
 400         /* get the min percentage */
 401         tok = strtok(NULL, ",");
 402         if (!tok)
 403                 goto setup;
 404
 405         callchain_param.min_percent = strtod(tok, &endptr);
 406         if (tok == endptr)
 407                 return -1;
 408
 409         /* get the print limit */
 410         tok2 = strtok(NULL, ",");
 411         if (!tok2)
 412                 goto setup;
 413
 414         if (tok2[0] != 'c') {
 415                 callchain_param.print_limit = strtod(tok2, &endptr);
 416                 tok2 = strtok(NULL, ",");
 417                 if (!tok2)
 418                         goto setup;
 419         }
 420
 421         /* get the call chain order */
 422         if (!strcmp(tok2, "caller"))
 423                 callchain_param.order = ORDER_CALLER;
 424         else if (!strcmp(tok2, "callee"))
 425                 callchain_param.order = ORDER_CALLEE;
 426         else
 427                 return -1;
 428 setup:
 429         if (callchain_register_param(&callchain_param) < 0) {
 430                 fprintf(stderr, "Can't register callchain params\n");
 431                 return -1;
 432         }
 433         return 0;
 434 }
 435
 436 static const char * const report_usage[] = {
 437         "perf report [<options>] <command>",
 438         NULL
 439 };
 440
 441 static const struct option options[] = {
 442         OPT_STRING('i', "input", &input_name, "file",
 443                     "input file name"),
 444         OPT_INCR('v', "verbose", &verbose,
 445                     "be more verbose (show symbol address, etc)"),
 446         OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace,
 447                     "dump raw trace in ASCII"),
 448         OPT_STRING('k', "vmlinux", &symbol_conf.vmlinux_name,
 449                    "file", "vmlinux pathname"),
 450         OPT_STRING(0, "kallsyms", &symbol_conf.kallsyms_name,
 451                    "file", "kallsyms pathname"),
 452         OPT_BOOLEAN('f', "force", &force, "don't complain, do it"),
 453         OPT_BOOLEAN('m', "modules", &symbol_conf.use_modules,
 454                     "load module symbols - WARNING: use only with -k and LIVE kernel"),
 455         OPT_BOOLEAN('n', "show-nr-samples", &symbol_conf.show_nr_samples,
 456                     "Show a column with the number of samples"),
 457         OPT_BOOLEAN('T', "threads", &show_threads,
 458                     "Show per-thread event counters"),
 459         OPT_STRING(0, "pretty", &pretty_printing_style, "key",
 460                    "pretty printing style key: normal raw"),
 461         OPT_BOOLEAN(0, "tui", &use_tui, "Use the TUI interface"),
 462         OPT_BOOLEAN(0, "stdio", &use_stdio, "Use the stdio interface"),
 463         OPT_STRING('s', "sort", &sort_order, "key[,key2...]",
 464                    "sort by key(s): pid, comm, dso, symbol, parent"),
 465         OPT_BOOLEAN(0, "showcpuutilization", &symbol_conf.show_cpu_utilization,
 466                     "Show sample percentage for different cpu modes"),
 467         OPT_STRING('p', "parent", &parent_pattern, "regex",
 468                    "regex filter to identify parent, see: '--sort parent'"),
 469         OPT_BOOLEAN('x', "exclude-other", &symbol_conf.exclude_other,
 470                     "Only display entries with parent-match"),
 471         OPT_CALLBACK_DEFAULT('g', "call-graph", NULL, "output_type,min_percent, call_order",
 472                      "Display callchains using output_type (graph, flat, fractal, or none) , min percent threshold and callchain order. "
 473                      "Default: fractal,0.5,callee", &parse_callchain_opt, callchain_default_opt),
 474         OPT_BOOLEAN('G', "inverted", &inverted_callchain, "alias for inverted call graph"),
 475         OPT_STRING('d', "dsos", &symbol_conf.dso_list_str, "dso[,dso...]",
 476                    "only consider symbols in these dsos"),
 477         OPT_STRING('C', "comms", &symbol_conf.comm_list_str, "comm[,comm...]",
 478                    "only consider symbols in these comms"),
 479         OPT_STRING('S', "symbols", &symbol_conf.sym_list_str, "symbol[,symbol...]",
 480                    "only consider these symbols"),
 481         OPT_STRING('w', "column-widths", &symbol_conf.col_width_list_str,
 482                    "width[,width...]",
 483                    "don't try to adjust column width, use these fixed values"),
 484         OPT_STRING('t', "field-separator", &symbol_conf.field_sep, "separator",
 485                    "separator for columns, no spaces will be added between "
 486                    "columns '.' is reserved."),
 487         OPT_BOOLEAN('U', "hide-unresolved", &hide_unresolved,
 488                     "Only display entries resolved to a symbol"),
 489         OPT_STRING(0, "symfs", &symbol_conf.symfs, "directory",
 490                     "Look for files with symbols relative to this directory"),
 491         OPT_STRING('c', "cpu", &cpu_list, "cpu", "list of cpus to profile"),
 492         OPT_BOOLEAN('I', "show-info", &show_full_info,
 493                     "Display extended information about perf.data file"),
 494         OPT_BOOLEAN(0, "source", &symbol_conf.annotate_src,
 495                     "Interleave source code with assembly code (default)"),
 496         OPT_BOOLEAN(0, "asm-raw", &symbol_conf.annotate_asm_raw,
 497                     "Display raw encoding of assembly instructions (default)"),
 498         OPT_STRING('M', "disassembler-style", &disassembler_style, "disassembler style",
 499                    "Specify disassembler style (e.g. -M intel for intel syntax)"),
 500         OPT_BOOLEAN(0, "show-total-period", &symbol_conf.show_total_period,
 501                     "Show a column with the sum of periods"),
 502         OPT_END()
 503 };
 504
 505 int cmd_report(int argc, const char **argv, const char *prefix __used)
 506 {
 507         argc = parse_options(argc, argv, options, report_usage, 0);
 508
 509         if (use_stdio)
 510                 use_browser = 0;
 511         else if (use_tui)
 512                 use_browser = 1;
 513
 514         if (inverted_callchain)
 515                 callchain_param.order = ORDER_CALLER;
 516
 517         if (strcmp(input_name, "-") != 0)
 518                 setup_browser(true);
 519         else
 520                 use_browser = 0;
 521         /*
 522          * Only in the newt browser we are doing integrated annotation,
 523          * so don't allocate extra space that won't be used in the stdio
 524          * implementation.
 525          */
 526         if (use_browser > 0) {
 527                 symbol_conf.priv_size = sizeof(struct annotation);
 528                 annotate_init         = symbol__annotate_init;
 529                 /*
 530                  * For searching by name on the "Browse map details".
 531                  * providing it only in verbose mode not to bloat too
 532                  * much struct symbol.
 533                  */
 534                 if (verbose) {
 535                         /*
 536                          * XXX: Need to provide a less kludgy way to ask for
 537                          * more space per symbol, the u32 is for the index on
 538                          * the ui browser.
 539                          * See symbol__browser_index.
 540                          */
 541                         symbol_conf.priv_size += sizeof(u32);
 542                         symbol_conf.sort_by_name = true;
 543                 }
 544         }
 545
 546         if (symbol__init() < 0)
 547                 return -1;
 548
 549         setup_sorting(report_usage, options);
 550
 551         if (parent_pattern != default_parent_pattern) {
 552                 if (sort_dimension__add("parent") < 0)
 553                         return -1;
 554
 555                 /*
 556                  * Only show the parent fields if we explicitly
 557                  * sort that way. If we only use parent machinery
 558                  * for filtering, we don't want it.
 559                  */
 560                 if (!strstr(sort_order, "parent"))
 561                         sort_parent.elide = 1;
 562         } else
 563                 symbol_conf.exclude_other = false;
 564
 565         /*
 566          * Any (unrecognized) arguments left?
 567          */
 568         if (argc)
 569                 usage_with_options(report_usage, options);
 570
 571         sort_entry__setup_elide(&sort_dso, symbol_conf.dso_list, "dso", stdout);
 572         sort_entry__setup_elide(&sort_comm, symbol_conf.comm_list, "comm", stdout);
 573         sort_entry__setup_elide(&sort_sym, symbol_conf.sym_list, "symbol", stdout);
 574
 575         return __cmd_report();
 576 }