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/color.h"
  13 #include <linux/list.h>
  14 #include "util/cache.h"
  15 #include <linux/rbtree.h>
  16 #include "util/symbol.h"
  17 #include "util/string.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/header.h"
  25 #include "util/session.h"
  26
  27 #include "util/parse-options.h"
  28 #include "util/parse-events.h"
  29
  30 #include "util/thread.h"
  31 #include "util/sort.h"
  32 #include "util/hist.h"
  33
  34 static char             const *input_name = "perf.data";
  35
  36 static int              force;
  37 static bool             hide_unresolved;
  38 static bool             dont_use_callchains;
  39
  40 static int              show_threads;
  41 static struct perf_read_values  show_threads_values;
  42
  43 static char             default_pretty_printing_style[] = "normal";
  44 static char             *pretty_printing_style = default_pretty_printing_style;
  45
  46 static char             callchain_default_opt[] = "fractal,0.5";
  47
  48 static struct event_stat_id *get_stats(struct perf_session *self,
  49                                        u64 event_stream, u32 type, u64 config)
  50 {
  51         struct rb_node **p = &self->stats_by_id.rb_node;
  52         struct rb_node *parent = NULL;
  53         struct event_stat_id *iter, *new;
  54
  55         while (*p != NULL) {
  56                 parent = *p;
  57                 iter = rb_entry(parent, struct event_stat_id, rb_node);
  58                 if (iter->config == config)
  59                         return iter;
  60
  61
  62                 if (config > iter->config)
  63                         p = &(*p)->rb_right;
  64                 else
  65                         p = &(*p)->rb_left;
  66         }
  67
  68         new = malloc(sizeof(struct event_stat_id));
  69         if (new == NULL)
  70                 return NULL;
  71         memset(new, 0, sizeof(struct event_stat_id));
  72         new->event_stream = event_stream;
  73         new->config = config;
  74         new->type = type;
  75         rb_link_node(&new->rb_node, parent, p);
  76         rb_insert_color(&new->rb_node, &self->stats_by_id);
  77         return new;
  78 }
  79
  80 static int perf_session__add_hist_entry(struct perf_session *self,
  81                                         struct addr_location *al,
  82                                         struct sample_data *data)
  83 {
  84         struct symbol **syms = NULL, *parent = NULL;
  85         bool hit;
  86         struct hist_entry *he;
  87         struct event_stat_id *stats;
  88         struct perf_event_attr *attr;
  89
  90         if ((sort__has_parent || symbol_conf.use_callchain) && data->callchain)
  91                 syms = perf_session__resolve_callchain(self, al->thread,
  92                                                        data->callchain, &parent);
  93
  94         attr = perf_header__find_attr(data->id, &self->header);
  95         if (attr)
  96                 stats = get_stats(self, data->id, attr->type, attr->config);
  97         else
  98                 stats = get_stats(self, data->id, 0, 0);
  99         if (stats == NULL)
 100                 return -ENOMEM;
 101         he = __perf_session__add_hist_entry(&stats->hists, al, parent,
 102                                             data->period, &hit);
 103         if (he == NULL)
 104                 return -ENOMEM;
 105
 106         if (hit)
 107                 he->count += data->period;
 108
 109         if (symbol_conf.use_callchain) {
 110                 if (!hit)
 111                         callchain_init(&he->callchain);
 112                 append_chain(&he->callchain, data->callchain, syms);
 113                 free(syms);
 114         }
 115
 116         return 0;
 117 }
 118
 119 static int validate_chain(struct ip_callchain *chain, event_t *event)
 120 {
 121         unsigned int chain_size;
 122
 123         chain_size = event->header.size;
 124         chain_size -= (unsigned long)&event->ip.__more_data - (unsigned long)event;
 125
 126         if (chain->nr*sizeof(u64) > chain_size)
 127                 return -1;
 128
 129         return 0;
 130 }
 131
 132 static int add_event_total(struct perf_session *session,
 133                            struct sample_data *data,
 134                            struct perf_event_attr *attr)
 135 {
 136         struct event_stat_id *stats;
 137
 138         if (attr)
 139                 stats = get_stats(session, data->id, attr->type, attr->config);
 140         else
 141                 stats = get_stats(session, data->id, 0, 0);
 142
 143         if (!stats)
 144                 return -ENOMEM;
 145
 146         stats->stats.total += data->period;
 147         session->events_stats.total += data->period;
 148         return 0;
 149 }
 150
 151 static int process_sample_event(event_t *event, struct perf_session *session)
 152 {
 153         struct sample_data data = { .period = 1, };
 154         struct addr_location al;
 155         struct perf_event_attr *attr;
 156
 157         event__parse_sample(event, session->sample_type, &data);
 158
 159         dump_printf("(IP, %d): %d/%d: %#Lx period: %Ld\n", event->header.misc,
 160                     data.pid, data.tid, data.ip, data.period);
 161
 162         if (session->sample_type & PERF_SAMPLE_CALLCHAIN) {
 163                 unsigned int i;
 164
 165                 dump_printf("... chain: nr:%Lu\n", data.callchain->nr);
 166
 167                 if (validate_chain(data.callchain, event) < 0) {
 168                         pr_debug("call-chain problem with event, "
 169                                  "skipping it.\n");
 170                         return 0;
 171                 }
 172
 173                 if (dump_trace) {
 174                         for (i = 0; i < data.callchain->nr; i++)
 175                                 dump_printf("..... %2d: %016Lx\n",
 176                                             i, data.callchain->ips[i]);
 177                 }
 178         }
 179
 180         if (event__preprocess_sample(event, session, &al, NULL) < 0) {
 181                 fprintf(stderr, "problem processing %d event, skipping it.\n",
 182                         event->header.type);
 183                 return -1;
 184         }
 185
 186         if (al.filtered || (hide_unresolved && al.sym == NULL))
 187                 return 0;
 188
 189         if (perf_session__add_hist_entry(session, &al, &data)) {
 190                 pr_debug("problem incrementing symbol count, skipping event\n");
 191                 return -1;
 192         }
 193
 194         attr = perf_header__find_attr(data.id, &session->header);
 195
 196         if (add_event_total(session, &data, attr)) {
 197                 pr_debug("problem adding event count\n");
 198                 return -1;
 199         }
 200
 201         return 0;
 202 }
 203
 204 static int process_read_event(event_t *event, struct perf_session *session __used)
 205 {
 206         struct perf_event_attr *attr;
 207
 208         attr = perf_header__find_attr(event->read.id, &session->header);
 209
 210         if (show_threads) {
 211                 const char *name = attr ? __event_name(attr->type, attr->config)
 212                                    : "unknown";
 213                 perf_read_values_add_value(&show_threads_values,
 214                                            event->read.pid, event->read.tid,
 215                                            event->read.id,
 216                                            name,
 217                                            event->read.value);
 218         }
 219
 220         dump_printf(": %d %d %s %Lu\n", event->read.pid, event->read.tid,
 221                     attr ? __event_name(attr->type, attr->config) : "FAIL",
 222                     event->read.value);
 223
 224         return 0;
 225 }
 226
 227 static int perf_session__setup_sample_type(struct perf_session *self)
 228 {
 229         if (!(self->sample_type & PERF_SAMPLE_CALLCHAIN)) {
 230                 if (sort__has_parent) {
 231                         fprintf(stderr, "selected --sort parent, but no"
 232                                         " callchain data. Did you call"
 233                                         " perf record without -g?\n");
 234                         return -EINVAL;
 235                 }
 236                 if (symbol_conf.use_callchain) {
 237                         fprintf(stderr, "selected -g but no callchain data."
 238                                         " Did you call perf record without"
 239                                         " -g?\n");
 240                         return -1;
 241                 }
 242         } else if (!dont_use_callchains && callchain_param.mode != CHAIN_NONE &&
 243                    !symbol_conf.use_callchain) {
 244                         symbol_conf.use_callchain = true;
 245                         if (register_callchain_param(&callchain_param) < 0) {
 246                                 fprintf(stderr, "Can't register callchain"
 247                                                 " params\n");
 248                                 return -EINVAL;
 249                         }
 250         }
 251
 252         return 0;
 253 }
 254
 255 static struct perf_event_ops event_ops = {
 256         .sample = process_sample_event,
 257         .mmap   = event__process_mmap,
 258         .comm   = event__process_comm,
 259         .exit   = event__process_task,
 260         .fork   = event__process_task,
 261         .lost   = event__process_lost,
 262         .read   = process_read_event,
 263 };
 264
 265 static int __cmd_report(void)
 266 {
 267         int ret = -EINVAL;
 268         struct perf_session *session;
 269         struct rb_node *next;
 270         const char *help = "For a higher level overview, try: perf report --sort comm,dso";
 271
 272         session = perf_session__new(input_name, O_RDONLY, force);
 273         if (session == NULL)
 274                 return -ENOMEM;
 275
 276         if (show_threads)
 277                 perf_read_values_init(&show_threads_values);
 278
 279         ret = perf_session__setup_sample_type(session);
 280         if (ret)
 281                 goto out_delete;
 282
 283         ret = perf_session__process_events(session, &event_ops);
 284         if (ret)
 285                 goto out_delete;
 286
 287         if (dump_trace) {
 288                 event__print_totals();
 289                 goto out_delete;
 290         }
 291
 292         if (verbose > 3)
 293                 perf_session__fprintf(session, stdout);
 294
 295         if (verbose > 2)
 296                 dsos__fprintf(stdout);
 297
 298         next = rb_first(&session->stats_by_id);
 299         while (next) {
 300                 struct event_stat_id *stats;
 301
 302                 stats = rb_entry(next, struct event_stat_id, rb_node);
 303                 perf_session__collapse_resort(&stats->hists);
 304                 perf_session__output_resort(&stats->hists, stats->stats.total);
 305
 306                 if (use_browser)
 307                         perf_session__browse_hists(&stats->hists,
 308                                                    stats->stats.total, help);
 309                 else {
 310                         if (rb_first(&session->stats_by_id) ==
 311                             rb_last(&session->stats_by_id))
 312                                 fprintf(stdout, "# Samples: %Ld\n#\n",
 313                                         stats->stats.total);
 314                         else
 315                                 fprintf(stdout, "# Samples: %Ld %s\n#\n",
 316                                         stats->stats.total,
 317                                         __event_name(stats->type, stats->config));
 318
 319                         perf_session__fprintf_hists(&stats->hists, NULL, false, stdout,
 320                                             stats->stats.total);
 321                         fprintf(stdout, "\n\n");
 322                 }
 323
 324                 next = rb_next(&stats->rb_node);
 325         }
 326
 327         if (!use_browser && sort_order == default_sort_order &&
 328             parent_pattern == default_parent_pattern) {
 329                 fprintf(stdout, "#\n# (%s)\n#\n", help);
 330
 331                 if (show_threads) {
 332                         bool style = !strcmp(pretty_printing_style, "raw");
 333                         perf_read_values_display(stdout, &show_threads_values,
 334                                                  style);
 335                         perf_read_values_destroy(&show_threads_values);
 336                 }
 337         }
 338 out_delete:
 339         perf_session__delete(session);
 340         return ret;
 341 }
 342
 343 static int
 344 parse_callchain_opt(const struct option *opt __used, const char *arg,
 345                     int unset)
 346 {
 347         char *tok;
 348         char *endptr;
 349
 350         /*
 351          * --no-call-graph
 352          */
 353         if (unset) {
 354                 dont_use_callchains = true;
 355                 return 0;
 356         }
 357
 358         symbol_conf.use_callchain = true;
 359
 360         if (!arg)
 361                 return 0;
 362
 363         tok = strtok((char *)arg, ",");
 364         if (!tok)
 365                 return -1;
 366
 367         /* get the output mode */
 368         if (!strncmp(tok, "graph", strlen(arg)))
 369                 callchain_param.mode = CHAIN_GRAPH_ABS;
 370
 371         else if (!strncmp(tok, "flat", strlen(arg)))
 372                 callchain_param.mode = CHAIN_FLAT;
 373
 374         else if (!strncmp(tok, "fractal", strlen(arg)))
 375                 callchain_param.mode = CHAIN_GRAPH_REL;
 376
 377         else if (!strncmp(tok, "none", strlen(arg))) {
 378                 callchain_param.mode = CHAIN_NONE;
 379                 symbol_conf.use_callchain = false;
 380
 381                 return 0;
 382         }
 383
 384         else
 385                 return -1;
 386
 387         /* get the min percentage */
 388         tok = strtok(NULL, ",");
 389         if (!tok)
 390                 goto setup;
 391
 392         callchain_param.min_percent = strtod(tok, &endptr);
 393         if (tok == endptr)
 394                 return -1;
 395
 396 setup:
 397         if (register_callchain_param(&callchain_param) < 0) {
 398                 fprintf(stderr, "Can't register callchain params\n");
 399                 return -1;
 400         }
 401         return 0;
 402 }
 403
 404 static const char * const report_usage[] = {
 405         "perf report [<options>] <command>",
 406         NULL
 407 };
 408
 409 static const struct option options[] = {
 410         OPT_STRING('i', "input", &input_name, "file",
 411                     "input file name"),
 412         OPT_BOOLEAN('v', "verbose", &verbose,
 413                     "be more verbose (show symbol address, etc)"),
 414         OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace,
 415                     "dump raw trace in ASCII"),
 416         OPT_STRING('k', "vmlinux", &symbol_conf.vmlinux_name,
 417                    "file", "vmlinux pathname"),
 418         OPT_BOOLEAN('f', "force", &force, "don't complain, do it"),
 419         OPT_BOOLEAN('m', "modules", &symbol_conf.use_modules,
 420                     "load module symbols - WARNING: use only with -k and LIVE kernel"),
 421         OPT_BOOLEAN('n', "show-nr-samples", &symbol_conf.show_nr_samples,
 422                     "Show a column with the number of samples"),
 423         OPT_BOOLEAN('T', "threads", &show_threads,
 424                     "Show per-thread event counters"),
 425         OPT_STRING(0, "pretty", &pretty_printing_style, "key",
 426                    "pretty printing style key: normal raw"),
 427         OPT_STRING('s', "sort", &sort_order, "key[,key2...]",
 428                    "sort by key(s): pid, comm, dso, symbol, parent"),
 429         OPT_BOOLEAN('P', "full-paths", &symbol_conf.full_paths,
 430                     "Don't shorten the pathnames taking into account the cwd"),
 431         OPT_STRING('p', "parent", &parent_pattern, "regex",
 432                    "regex filter to identify parent, see: '--sort parent'"),
 433         OPT_BOOLEAN('x', "exclude-other", &symbol_conf.exclude_other,
 434                     "Only display entries with parent-match"),
 435         OPT_CALLBACK_DEFAULT('g', "call-graph", NULL, "output_type,min_percent",
 436                      "Display callchains using output_type and min percent threshold. "
 437                      "Default: fractal,0.5", &parse_callchain_opt, callchain_default_opt),
 438         OPT_STRING('d', "dsos", &symbol_conf.dso_list_str, "dso[,dso...]",
 439                    "only consider symbols in these dsos"),
 440         OPT_STRING('C', "comms", &symbol_conf.comm_list_str, "comm[,comm...]",
 441                    "only consider symbols in these comms"),
 442         OPT_STRING('S', "symbols", &symbol_conf.sym_list_str, "symbol[,symbol...]",
 443                    "only consider these symbols"),
 444         OPT_STRING('w', "column-widths", &symbol_conf.col_width_list_str,
 445                    "width[,width...]",
 446                    "don't try to adjust column width, use these fixed values"),
 447         OPT_STRING('t', "field-separator", &symbol_conf.field_sep, "separator",
 448                    "separator for columns, no spaces will be added between "
 449                    "columns '.' is reserved."),
 450         OPT_BOOLEAN('U', "hide-unresolved", &hide_unresolved,
 451                     "Only display entries resolved to a symbol"),
 452         OPT_END()
 453 };
 454
 455 int cmd_report(int argc, const char **argv, const char *prefix __used)
 456 {
 457         argc = parse_options(argc, argv, options, report_usage, 0);
 458
 459         setup_browser();
 460
 461         if (symbol__init() < 0)
 462                 return -1;
 463
 464         setup_sorting(report_usage, options);
 465
 466         if (parent_pattern != default_parent_pattern) {
 467                 sort_dimension__add("parent");
 468                 sort_parent.elide = 1;
 469         } else
 470                 symbol_conf.exclude_other = false;
 471
 472         /*
 473          * Any (unrecognized) arguments left?
 474          */
 475         if (argc)
 476                 usage_with_options(report_usage, options);
 477
 478         sort_entry__setup_elide(&sort_dso, symbol_conf.dso_list, "dso", stdout);
 479         sort_entry__setup_elide(&sort_comm, symbol_conf.comm_list, "comm", stdout);
 480         sort_entry__setup_elide(&sort_sym, symbol_conf.sym_list, "symbol", stdout);
 481
 482         return __cmd_report();
 483 }