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                 if (syms == NULL)
  96                         return -ENOMEM;
  97         }
  98
  99         attr = perf_header__find_attr(data->id, &self->header);
 100         if (attr)
 101                 stats = get_stats(self, data->id, attr->type, attr->config);
 102         else
 103                 stats = get_stats(self, data->id, 0, 0);
 104         if (stats == NULL)
 105                 return -ENOMEM;
 106         he = __perf_session__add_hist_entry(&stats->hists, al, parent,
 107                                             data->period, &hit);
 108         if (he == NULL)
 109                 return -ENOMEM;
 110
 111         if (hit)
 112                 he->count += data->period;
 113
 114         if (symbol_conf.use_callchain) {
 115                 if (!hit)
 116                         callchain_init(he->callchain);
 117                 err = append_chain(he->callchain, data->callchain, syms);
 118                 free(syms);
 119
 120                 if (err)
 121                         return err;
 122         }
 123
 124         return 0;
 125 }
 126
 127 static int validate_chain(struct ip_callchain *chain, event_t *event)
 128 {
 129         unsigned int chain_size;
 130
 131         chain_size = event->header.size;
 132         chain_size -= (unsigned long)&event->ip.__more_data - (unsigned long)event;
 133
 134         if (chain->nr*sizeof(u64) > chain_size)
 135                 return -1;
 136
 137         return 0;
 138 }
 139
 140 static int add_event_total(struct perf_session *session,
 141                            struct sample_data *data,
 142                            struct perf_event_attr *attr)
 143 {
 144         struct event_stat_id *stats;
 145
 146         if (attr)
 147                 stats = get_stats(session, data->id, attr->type, attr->config);
 148         else
 149                 stats = get_stats(session, data->id, 0, 0);
 150
 151         if (!stats)
 152                 return -ENOMEM;
 153
 154         stats->stats.total += data->period;
 155         session->events_stats.total += data->period;
 156         return 0;
 157 }
 158
 159 static int process_sample_event(event_t *event, struct perf_session *session)
 160 {
 161         struct sample_data data = { .period = 1, };
 162         struct addr_location al;
 163         struct perf_event_attr *attr;
 164
 165         event__parse_sample(event, session->sample_type, &data);
 166
 167         dump_printf("(IP, %d): %d/%d: %#Lx period: %Ld\n", event->header.misc,
 168                     data.pid, data.tid, data.ip, data.period);
 169
 170         if (session->sample_type & PERF_SAMPLE_CALLCHAIN) {
 171                 unsigned int i;
 172
 173                 dump_printf("... chain: nr:%Lu\n", data.callchain->nr);
 174
 175                 if (validate_chain(data.callchain, event) < 0) {
 176                         pr_debug("call-chain problem with event, "
 177                                  "skipping it.\n");
 178                         return 0;
 179                 }
 180
 181                 if (dump_trace) {
 182                         for (i = 0; i < data.callchain->nr; i++)
 183                                 dump_printf("..... %2d: %016Lx\n",
 184                                             i, data.callchain->ips[i]);
 185                 }
 186         }
 187
 188         if (event__preprocess_sample(event, session, &al, NULL) < 0) {
 189                 fprintf(stderr, "problem processing %d event, skipping it.\n",
 190                         event->header.type);
 191                 return -1;
 192         }
 193
 194         if (al.filtered || (hide_unresolved && al.sym == NULL))
 195                 return 0;
 196
 197         if (perf_session__add_hist_entry(session, &al, &data)) {
 198                 pr_debug("problem incrementing symbol count, skipping event\n");
 199                 return -1;
 200         }
 201
 202         attr = perf_header__find_attr(data.id, &session->header);
 203
 204         if (add_event_total(session, &data, attr)) {
 205                 pr_debug("problem adding event count\n");
 206                 return -1;
 207         }
 208
 209         return 0;
 210 }
 211
 212 static int process_read_event(event_t *event, struct perf_session *session __used)
 213 {
 214         struct perf_event_attr *attr;
 215
 216         attr = perf_header__find_attr(event->read.id, &session->header);
 217
 218         if (show_threads) {
 219                 const char *name = attr ? __event_name(attr->type, attr->config)
 220                                    : "unknown";
 221                 perf_read_values_add_value(&show_threads_values,
 222                                            event->read.pid, event->read.tid,
 223                                            event->read.id,
 224                                            name,
 225                                            event->read.value);
 226         }
 227
 228         dump_printf(": %d %d %s %Lu\n", event->read.pid, event->read.tid,
 229                     attr ? __event_name(attr->type, attr->config) : "FAIL",
 230                     event->read.value);
 231
 232         return 0;
 233 }
 234
 235 static int perf_session__setup_sample_type(struct perf_session *self)
 236 {
 237         if (!(self->sample_type & PERF_SAMPLE_CALLCHAIN)) {
 238                 if (sort__has_parent) {
 239                         fprintf(stderr, "selected --sort parent, but no"
 240                                         " callchain data. Did you call"
 241                                         " perf record without -g?\n");
 242                         return -EINVAL;
 243                 }
 244                 if (symbol_conf.use_callchain) {
 245                         fprintf(stderr, "selected -g but no callchain data."
 246                                         " Did you call perf record without"
 247                                         " -g?\n");
 248                         return -1;
 249                 }
 250         } else if (!dont_use_callchains && callchain_param.mode != CHAIN_NONE &&
 251                    !symbol_conf.use_callchain) {
 252                         symbol_conf.use_callchain = true;
 253                         if (register_callchain_param(&callchain_param) < 0) {
 254                                 fprintf(stderr, "Can't register callchain"
 255                                                 " params\n");
 256                                 return -EINVAL;
 257                         }
 258         }
 259
 260         return 0;
 261 }
 262
 263 static struct perf_event_ops event_ops = {
 264         .sample = process_sample_event,
 265         .mmap   = event__process_mmap,
 266         .comm   = event__process_comm,
 267         .exit   = event__process_task,
 268         .fork   = event__process_task,
 269         .lost   = event__process_lost,
 270         .read   = process_read_event,
 271 };
 272
 273 static int __cmd_report(void)
 274 {
 275         int ret = -EINVAL;
 276         struct perf_session *session;
 277         struct rb_node *next;
 278         const char *help = "For a higher level overview, try: perf report --sort comm,dso";
 279
 280         session = perf_session__new(input_name, O_RDONLY, force);
 281         if (session == NULL)
 282                 return -ENOMEM;
 283
 284         if (show_threads)
 285                 perf_read_values_init(&show_threads_values);
 286
 287         ret = perf_session__setup_sample_type(session);
 288         if (ret)
 289                 goto out_delete;
 290
 291         ret = perf_session__process_events(session, &event_ops);
 292         if (ret)
 293                 goto out_delete;
 294
 295         if (dump_trace) {
 296                 event__print_totals();
 297                 goto out_delete;
 298         }
 299
 300         if (verbose > 3)
 301                 perf_session__fprintf(session, stdout);
 302
 303         if (verbose > 2)
 304                 dsos__fprintf(stdout);
 305
 306         next = rb_first(&session->stats_by_id);
 307         while (next) {
 308                 struct event_stat_id *stats;
 309                 u64 nr_hists;
 310
 311                 stats = rb_entry(next, struct event_stat_id, rb_node);
 312                 perf_session__collapse_resort(&stats->hists);
 313                 nr_hists = perf_session__output_resort(&stats->hists,
 314                                                        stats->stats.total);
 315                 if (use_browser)
 316                         perf_session__browse_hists(&stats->hists, nr_hists,
 317                                                    stats->stats.total, help);
 318                 else {
 319                         if (rb_first(&session->stats_by_id) ==
 320                             rb_last(&session->stats_by_id))
 321                                 fprintf(stdout, "# Samples: %Ld\n#\n",
 322                                         stats->stats.total);
 323                         else
 324                                 fprintf(stdout, "# Samples: %Ld %s\n#\n",
 325                                         stats->stats.total,
 326                                         __event_name(stats->type, stats->config));
 327
 328                         perf_session__fprintf_hists(&stats->hists, NULL, false, stdout,
 329                                             stats->stats.total);
 330                         fprintf(stdout, "\n\n");
 331                 }
 332
 333                 next = rb_next(&stats->rb_node);
 334         }
 335
 336         if (!use_browser && sort_order == default_sort_order &&
 337             parent_pattern == default_parent_pattern) {
 338                 fprintf(stdout, "#\n# (%s)\n#\n", help);
 339
 340                 if (show_threads) {
 341                         bool style = !strcmp(pretty_printing_style, "raw");
 342                         perf_read_values_display(stdout, &show_threads_values,
 343                                                  style);
 344                         perf_read_values_destroy(&show_threads_values);
 345                 }
 346         }
 347 out_delete:
 348         perf_session__delete(session);
 349         return ret;
 350 }
 351
 352 static int
 353 parse_callchain_opt(const struct option *opt __used, const char *arg,
 354                     int unset)
 355 {
 356         char *tok;
 357         char *endptr;
 358
 359         /*
 360          * --no-call-graph
 361          */
 362         if (unset) {
 363                 dont_use_callchains = true;
 364                 return 0;
 365         }
 366
 367         symbol_conf.use_callchain = true;
 368
 369         if (!arg)
 370                 return 0;
 371
 372         tok = strtok((char *)arg, ",");
 373         if (!tok)
 374                 return -1;
 375
 376         /* get the output mode */
 377         if (!strncmp(tok, "graph", strlen(arg)))
 378                 callchain_param.mode = CHAIN_GRAPH_ABS;
 379
 380         else if (!strncmp(tok, "flat", strlen(arg)))
 381                 callchain_param.mode = CHAIN_FLAT;
 382
 383         else if (!strncmp(tok, "fractal", strlen(arg)))
 384                 callchain_param.mode = CHAIN_GRAPH_REL;
 385
 386         else if (!strncmp(tok, "none", strlen(arg))) {
 387                 callchain_param.mode = CHAIN_NONE;
 388                 symbol_conf.use_callchain = false;
 389
 390                 return 0;
 391         }
 392
 393         else
 394                 return -1;
 395
 396         /* get the min percentage */
 397         tok = strtok(NULL, ",");
 398         if (!tok)
 399                 goto setup;
 400
 401         callchain_param.min_percent = strtod(tok, &endptr);
 402         if (tok == endptr)
 403                 return -1;
 404
 405 setup:
 406         if (register_callchain_param(&callchain_param) < 0) {
 407                 fprintf(stderr, "Can't register callchain params\n");
 408                 return -1;
 409         }
 410         return 0;
 411 }
 412
 413 static const char * const report_usage[] = {
 414         "perf report [<options>] <command>",
 415         NULL
 416 };
 417
 418 static const struct option options[] = {
 419         OPT_STRING('i', "input", &input_name, "file",
 420                     "input file name"),
 421         OPT_BOOLEAN('v', "verbose", &verbose,
 422                     "be more verbose (show symbol address, etc)"),
 423         OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace,
 424                     "dump raw trace in ASCII"),
 425         OPT_STRING('k', "vmlinux", &symbol_conf.vmlinux_name,
 426                    "file", "vmlinux pathname"),
 427         OPT_BOOLEAN('f', "force", &force, "don't complain, do it"),
 428         OPT_BOOLEAN('m', "modules", &symbol_conf.use_modules,
 429                     "load module symbols - WARNING: use only with -k and LIVE kernel"),
 430         OPT_BOOLEAN('n', "show-nr-samples", &symbol_conf.show_nr_samples,
 431                     "Show a column with the number of samples"),
 432         OPT_BOOLEAN('T', "threads", &show_threads,
 433                     "Show per-thread event counters"),
 434         OPT_STRING(0, "pretty", &pretty_printing_style, "key",
 435                    "pretty printing style key: normal raw"),
 436         OPT_STRING('s', "sort", &sort_order, "key[,key2...]",
 437                    "sort by key(s): pid, comm, dso, symbol, parent"),
 438         OPT_BOOLEAN('P', "full-paths", &symbol_conf.full_paths,
 439                     "Don't shorten the pathnames taking into account the cwd"),
 440         OPT_STRING('p', "parent", &parent_pattern, "regex",
 441                    "regex filter to identify parent, see: '--sort parent'"),
 442         OPT_BOOLEAN('x', "exclude-other", &symbol_conf.exclude_other,
 443                     "Only display entries with parent-match"),
 444         OPT_CALLBACK_DEFAULT('g', "call-graph", NULL, "output_type,min_percent",
 445                      "Display callchains using output_type and min percent threshold. "
 446                      "Default: fractal,0.5", &parse_callchain_opt, callchain_default_opt),
 447         OPT_STRING('d', "dsos", &symbol_conf.dso_list_str, "dso[,dso...]",
 448                    "only consider symbols in these dsos"),
 449         OPT_STRING('C', "comms", &symbol_conf.comm_list_str, "comm[,comm...]",
 450                    "only consider symbols in these comms"),
 451         OPT_STRING('S', "symbols", &symbol_conf.sym_list_str, "symbol[,symbol...]",
 452                    "only consider these symbols"),
 453         OPT_STRING('w', "column-widths", &symbol_conf.col_width_list_str,
 454                    "width[,width...]",
 455                    "don't try to adjust column width, use these fixed values"),
 456         OPT_STRING('t', "field-separator", &symbol_conf.field_sep, "separator",
 457                    "separator for columns, no spaces will be added between "
 458                    "columns '.' is reserved."),
 459         OPT_BOOLEAN('U', "hide-unresolved", &hide_unresolved,
 460                     "Only display entries resolved to a symbol"),
 461         OPT_END()
 462 };
 463
 464 int cmd_report(int argc, const char **argv, const char *prefix __used)
 465 {
 466         argc = parse_options(argc, argv, options, report_usage, 0);
 467
 468         setup_browser();
 469
 470         if (symbol__init() < 0)
 471                 return -1;
 472
 473         setup_sorting(report_usage, options);
 474
 475         if (parent_pattern != default_parent_pattern) {
 476                 if (sort_dimension__add("parent") < 0)
 477                         return -1;
 478                 sort_parent.elide = 1;
 479         } else
 480                 symbol_conf.exclude_other = false;
 481
 482         /*
 483          * Any (unrecognized) arguments left?
 484          */
 485         if (argc)
 486                 usage_with_options(report_usage, options);
 487
 488         sort_entry__setup_elide(&sort_dso, symbol_conf.dso_list, "dso", stdout);
 489         sort_entry__setup_elide(&sort_comm, symbol_conf.comm_list, "comm", stdout);
 490         sort_entry__setup_elide(&sort_sym, symbol_conf.sym_list, "symbol", stdout);
 491
 492         return __cmd_report();
 493 }