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