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