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 map_symbol *syms = NULL;
  85         struct symbol *parent = NULL;
  86         bool hit;
  87         int err;
  88         struct hist_entry *he;
  89         struct event_stat_id *stats;
  90         struct perf_event_attr *attr;
  91
  92         if ((sort__has_parent || symbol_conf.use_callchain) && data->callchain)
  93                 syms = perf_session__resolve_callchain(self, al->thread,
  94                                                        data->callchain, &parent);
  95
  96         attr = perf_header__find_attr(data->id, &self->header);
  97         if (attr)
  98                 stats = get_stats(self, data->id, attr->type, attr->config);
  99         else
 100                 stats = get_stats(self, data->id, 0, 0);
 101         if (stats == NULL)
 102                 return -ENOMEM;
 103         he = __perf_session__add_hist_entry(&stats->hists, al, parent,
 104                                             data->period, &hit);
 105         if (he == NULL)
 106                 return -ENOMEM;
 107
 108         if (hit)
 109                 he->count += data->period;
 110
 111         if (symbol_conf.use_callchain) {
 112                 if (!hit)
 113                         callchain_init(&he->callchain);
 114                 err = append_chain(&he->callchain, data->callchain, syms);
 115                 free(syms);
 116
 117                 if (err)
 118                         return err;
 119         }
 120
 121         return 0;
 122 }
 123
 124 static int validate_chain(struct ip_callchain *chain, event_t *event)
 125 {
 126         unsigned int chain_size;
 127
 128         chain_size = event->header.size;
 129         chain_size -= (unsigned long)&event->ip.__more_data - (unsigned long)event;
 130
 131         if (chain->nr*sizeof(u64) > chain_size)
 132                 return -1;
 133
 134         return 0;
 135 }
 136
 137 static int add_event_total(struct perf_session *session,
 138                            struct sample_data *data,
 139                            struct perf_event_attr *attr)
 140 {
 141         struct event_stat_id *stats;
 142
 143         if (attr)
 144                 stats = get_stats(session, data->id, attr->type, attr->config);
 145         else
 146                 stats = get_stats(session, data->id, 0, 0);
 147
 148         if (!stats)
 149                 return -ENOMEM;
 150
 151         stats->stats.total += data->period;
 152         session->events_stats.total += data->period;
 153         return 0;
 154 }
 155
 156 static int process_sample_event(event_t *event, struct perf_session *session)
 157 {
 158         struct sample_data data = { .period = 1, };
 159         struct addr_location al;
 160         struct perf_event_attr *attr;
 161
 162         event__parse_sample(event, session->sample_type, &data);
 163
 164         dump_printf("(IP, %d): %d/%d: %#Lx period: %Ld\n", event->header.misc,
 165                     data.pid, data.tid, data.ip, data.period);
 166
 167         if (session->sample_type & PERF_SAMPLE_CALLCHAIN) {
 168                 unsigned int i;
 169
 170                 dump_printf("... chain: nr:%Lu\n", data.callchain->nr);
 171
 172                 if (validate_chain(data.callchain, event) < 0) {
 173                         pr_debug("call-chain problem with event, "
 174                                  "skipping it.\n");
 175                         return 0;
 176                 }
 177
 178                 if (dump_trace) {
 179                         for (i = 0; i < data.callchain->nr; i++)
 180                                 dump_printf("..... %2d: %016Lx\n",
 181                                             i, data.callchain->ips[i]);
 182                 }
 183         }
 184
 185         if (event__preprocess_sample(event, session, &al, NULL) < 0) {
 186                 fprintf(stderr, "problem processing %d event, skipping it.\n",
 187                         event->header.type);
 188                 return -1;
 189         }
 190
 191         if (al.filtered || (hide_unresolved && al.sym == NULL))
 192                 return 0;
 193
 194         if (perf_session__add_hist_entry(session, &al, &data)) {
 195                 pr_debug("problem incrementing symbol count, skipping event\n");
 196                 return -1;
 197         }
 198
 199         attr = perf_header__find_attr(data.id, &session->header);
 200
 201         if (add_event_total(session, &data, attr)) {
 202                 pr_debug("problem adding event count\n");
 203                 return -1;
 204         }
 205
 206         return 0;
 207 }
 208
 209 static int process_read_event(event_t *event, struct perf_session *session __used)
 210 {
 211         struct perf_event_attr *attr;
 212
 213         attr = perf_header__find_attr(event->read.id, &session->header);
 214
 215         if (show_threads) {
 216                 const char *name = attr ? __event_name(attr->type, attr->config)
 217                                    : "unknown";
 218                 perf_read_values_add_value(&show_threads_values,
 219                                            event->read.pid, event->read.tid,
 220                                            event->read.id,
 221                                            name,
 222                                            event->read.value);
 223         }
 224
 225         dump_printf(": %d %d %s %Lu\n", event->read.pid, event->read.tid,
 226                     attr ? __event_name(attr->type, attr->config) : "FAIL",
 227                     event->read.value);
 228
 229         return 0;
 230 }
 231
 232 static int perf_session__setup_sample_type(struct perf_session *self)
 233 {
 234         if (!(self->sample_type & PERF_SAMPLE_CALLCHAIN)) {
 235                 if (sort__has_parent) {
 236                         fprintf(stderr, "selected --sort parent, but no"
 237                                         " callchain data. Did you call"
 238                                         " perf record without -g?\n");
 239                         return -EINVAL;
 240                 }
 241                 if (symbol_conf.use_callchain) {
 242                         fprintf(stderr, "selected -g but no callchain data."
 243                                         " Did you call perf record without"
 244                                         " -g?\n");
 245                         return -1;
 246                 }
 247         } else if (!dont_use_callchains && callchain_param.mode != CHAIN_NONE &&
 248                    !symbol_conf.use_callchain) {
 249                         symbol_conf.use_callchain = true;
 250                         if (register_callchain_param(&callchain_param) < 0) {
 251                                 fprintf(stderr, "Can't register callchain"
 252                                                 " params\n");
 253                                 return -EINVAL;
 254                         }
 255         }
 256
 257         return 0;
 258 }
 259
 260 static struct perf_event_ops event_ops = {
 261         .sample = process_sample_event,
 262         .mmap   = event__process_mmap,
 263         .comm   = event__process_comm,
 264         .exit   = event__process_task,
 265         .fork   = event__process_task,
 266         .lost   = event__process_lost,
 267         .read   = process_read_event,
 268 };
 269
 270 static int __cmd_report(void)
 271 {
 272         int ret = -EINVAL;
 273         struct perf_session *session;
 274         struct rb_node *next;
 275         const char *help = "For a higher level overview, try: perf report --sort comm,dso";
 276
 277         session = perf_session__new(input_name, O_RDONLY, force);
 278         if (session == NULL)
 279                 return -ENOMEM;
 280
 281         if (show_threads)
 282                 perf_read_values_init(&show_threads_values);
 283
 284         ret = perf_session__setup_sample_type(session);
 285         if (ret)
 286                 goto out_delete;
 287
 288         ret = perf_session__process_events(session, &event_ops);
 289         if (ret)
 290                 goto out_delete;
 291
 292         if (dump_trace) {
 293                 event__print_totals();
 294                 goto out_delete;
 295         }
 296
 297         if (verbose > 3)
 298                 perf_session__fprintf(session, stdout);
 299
 300         if (verbose > 2)
 301                 dsos__fprintf(stdout);
 302
 303         next = rb_first(&session->stats_by_id);
 304         while (next) {
 305                 struct event_stat_id *stats;
 306
 307                 stats = rb_entry(next, struct event_stat_id, rb_node);
 308                 perf_session__collapse_resort(&stats->hists);
 309                 perf_session__output_resort(&stats->hists, stats->stats.total);
 310
 311                 if (use_browser)
 312                         perf_session__browse_hists(&stats->hists,
 313                                                    stats->stats.total, help);
 314                 else {
 315                         if (rb_first(&session->stats_by_id) ==
 316                             rb_last(&session->stats_by_id))
 317                                 fprintf(stdout, "# Samples: %Ld\n#\n",
 318                                         stats->stats.total);
 319                         else
 320                                 fprintf(stdout, "# Samples: %Ld %s\n#\n",
 321                                         stats->stats.total,
 322                                         __event_name(stats->type, stats->config));
 323
 324                         perf_session__fprintf_hists(&stats->hists, NULL, false, stdout,
 325                                             stats->stats.total);
 326                         fprintf(stdout, "\n\n");
 327                 }
 328
 329                 next = rb_next(&stats->rb_node);
 330         }
 331
 332         if (!use_browser && sort_order == default_sort_order &&
 333             parent_pattern == default_parent_pattern) {
 334                 fprintf(stdout, "#\n# (%s)\n#\n", help);
 335
 336                 if (show_threads) {
 337                         bool style = !strcmp(pretty_printing_style, "raw");
 338                         perf_read_values_display(stdout, &show_threads_values,
 339                                                  style);
 340                         perf_read_values_destroy(&show_threads_values);
 341                 }
 342         }
 343 out_delete:
 344         perf_session__delete(session);
 345         return ret;
 346 }
 347
 348 static int
 349 parse_callchain_opt(const struct option *opt __used, const char *arg,
 350                     int unset)
 351 {
 352         char *tok;
 353         char *endptr;
 354
 355         /*
 356          * --no-call-graph
 357          */
 358         if (unset) {
 359                 dont_use_callchains = true;
 360                 return 0;
 361         }
 362
 363         symbol_conf.use_callchain = true;
 364
 365         if (!arg)
 366                 return 0;
 367
 368         tok = strtok((char *)arg, ",");
 369         if (!tok)
 370                 return -1;
 371
 372         /* get the output mode */
 373         if (!strncmp(tok, "graph", strlen(arg)))
 374                 callchain_param.mode = CHAIN_GRAPH_ABS;
 375
 376         else if (!strncmp(tok, "flat", strlen(arg)))
 377                 callchain_param.mode = CHAIN_FLAT;
 378
 379         else if (!strncmp(tok, "fractal", strlen(arg)))
 380                 callchain_param.mode = CHAIN_GRAPH_REL;
 381
 382         else if (!strncmp(tok, "none", strlen(arg))) {
 383                 callchain_param.mode = CHAIN_NONE;
 384                 symbol_conf.use_callchain = false;
 385
 386                 return 0;
 387         }
 388
 389         else
 390                 return -1;
 391
 392         /* get the min percentage */
 393         tok = strtok(NULL, ",");
 394         if (!tok)
 395                 goto setup;
 396
 397         callchain_param.min_percent = strtod(tok, &endptr);
 398         if (tok == endptr)
 399                 return -1;
 400
 401 setup:
 402         if (register_callchain_param(&callchain_param) < 0) {
 403                 fprintf(stderr, "Can't register callchain params\n");
 404                 return -1;
 405         }
 406         return 0;
 407 }
 408
 409 static const char * const report_usage[] = {
 410         "perf report [<options>] <command>",
 411         NULL
 412 };
 413
 414 static const struct option options[] = {
 415         OPT_STRING('i', "input", &input_name, "file",
 416                     "input file name"),
 417         OPT_BOOLEAN('v', "verbose", &verbose,
 418                     "be more verbose (show symbol address, etc)"),
 419         OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace,
 420                     "dump raw trace in ASCII"),
 421         OPT_STRING('k', "vmlinux", &symbol_conf.vmlinux_name,
 422                    "file", "vmlinux pathname"),
 423         OPT_BOOLEAN('f', "force", &force, "don't complain, do it"),
 424         OPT_BOOLEAN('m', "modules", &symbol_conf.use_modules,
 425                     "load module symbols - WARNING: use only with -k and LIVE kernel"),
 426         OPT_BOOLEAN('n', "show-nr-samples", &symbol_conf.show_nr_samples,
 427                     "Show a column with the number of samples"),
 428         OPT_BOOLEAN('T', "threads", &show_threads,
 429                     "Show per-thread event counters"),
 430         OPT_STRING(0, "pretty", &pretty_printing_style, "key",
 431                    "pretty printing style key: normal raw"),
 432         OPT_STRING('s', "sort", &sort_order, "key[,key2...]",
 433                    "sort by key(s): pid, comm, dso, symbol, parent"),
 434         OPT_BOOLEAN('P', "full-paths", &symbol_conf.full_paths,
 435                     "Don't shorten the pathnames taking into account the cwd"),
 436         OPT_STRING('p', "parent", &parent_pattern, "regex",
 437                    "regex filter to identify parent, see: '--sort parent'"),
 438         OPT_BOOLEAN('x', "exclude-other", &symbol_conf.exclude_other,
 439                     "Only display entries with parent-match"),
 440         OPT_CALLBACK_DEFAULT('g', "call-graph", NULL, "output_type,min_percent",
 441                      "Display callchains using output_type and min percent threshold. "
 442                      "Default: fractal,0.5", &parse_callchain_opt, callchain_default_opt),
 443         OPT_STRING('d', "dsos", &symbol_conf.dso_list_str, "dso[,dso...]",
 444                    "only consider symbols in these dsos"),
 445         OPT_STRING('C', "comms", &symbol_conf.comm_list_str, "comm[,comm...]",
 446                    "only consider symbols in these comms"),
 447         OPT_STRING('S', "symbols", &symbol_conf.sym_list_str, "symbol[,symbol...]",
 448                    "only consider these symbols"),
 449         OPT_STRING('w', "column-widths", &symbol_conf.col_width_list_str,
 450                    "width[,width...]",
 451                    "don't try to adjust column width, use these fixed values"),
 452         OPT_STRING('t', "field-separator", &symbol_conf.field_sep, "separator",
 453                    "separator for columns, no spaces will be added between "
 454                    "columns '.' is reserved."),
 455         OPT_BOOLEAN('U', "hide-unresolved", &hide_unresolved,
 456                     "Only display entries resolved to a symbol"),
 457         OPT_END()
 458 };
 459
 460 int cmd_report(int argc, const char **argv, const char *prefix __used)
 461 {
 462         argc = parse_options(argc, argv, options, report_usage, 0);
 463
 464         setup_browser();
 465
 466         if (symbol__init() < 0)
 467                 return -1;
 468
 469         setup_sorting(report_usage, options);
 470
 471         if (parent_pattern != default_parent_pattern) {
 472                 sort_dimension__add("parent");
 473                 sort_parent.elide = 1;
 474         } else
 475                 symbol_conf.exclude_other = false;
 476
 477         /*
 478          * Any (unrecognized) arguments left?
 479          */
 480         if (argc)
 481                 usage_with_options(report_usage, options);
 482
 483         sort_entry__setup_elide(&sort_dso, symbol_conf.dso_list, "dso", stdout);
 484         sort_entry__setup_elide(&sort_comm, symbol_conf.comm_list, "comm", stdout);
 485         sort_entry__setup_elide(&sort_sym, symbol_conf.sym_list, "symbol", stdout);
 486
 487         return __cmd_report();
 488 }