tools/perf/util/session.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 #include <errno.h>
   3 #include <inttypes.h>
   4 #include <linux/err.h>
   5 #include <linux/kernel.h>
   6 #include <linux/zalloc.h>
   7 #include <api/fs/fs.h>
   8
   9 #include <byteswap.h>
  10 #include <unistd.h>
  11 #include <sys/types.h>
  12 #include <sys/mman.h>
  13 #include <perf/cpumap.h>
  14
  15 #include "map_symbol.h"
  16 #include "branch.h"
  17 #include "debug.h"
  18 #include "env.h"
  19 #include "evlist.h"
  20 #include "evsel.h"
  21 #include "memswap.h"
  22 #include "map.h"
  23 #include "symbol.h"
  24 #include "session.h"
  25 #include "tool.h"
  26 #include "perf_regs.h"
  27 #include "asm/bug.h"
  28 #include "auxtrace.h"
  29 #include "thread.h"
  30 #include "thread-stack.h"
  31 #include "sample-raw.h"
  32 #include "stat.h"
  33 #include "tsc.h"
  34 #include "ui/progress.h"
  35 #include "../perf.h"
  36 #include "arch/common.h"
  37 #include "units.h"
  38 #include <internal/lib.h>
  39
  40 #ifdef HAVE_ZSTD_SUPPORT
  41 static int perf_session__process_compressed_event(struct perf_session *session,
  42                                                   union perf_event *event, u64 file_offset,
  43                                                   const char *file_path)
  44 {
  45         void *src;
  46         size_t decomp_size, src_size;
  47         u64 decomp_last_rem = 0;
  48         size_t mmap_len, decomp_len = session->header.env.comp_mmap_len;
  49         struct decomp *decomp, *decomp_last = session->active_decomp->decomp_last;
  50
  51         if (decomp_last) {
  52                 decomp_last_rem = decomp_last->size - decomp_last->head;
  53                 decomp_len += decomp_last_rem;
  54         }
  55
  56         mmap_len = sizeof(struct decomp) + decomp_len;
  57         decomp = mmap(NULL, mmap_len, PROT_READ|PROT_WRITE,
  58                       MAP_ANONYMOUS|MAP_PRIVATE, -1, 0);
  59         if (decomp == MAP_FAILED) {
  60                 pr_err("Couldn't allocate memory for decompression\n");
  61                 return -1;
  62         }
  63
  64         decomp->file_pos = file_offset;
  65         decomp->file_path = file_path;
  66         decomp->mmap_len = mmap_len;
  67         decomp->head = 0;
  68
  69         if (decomp_last_rem) {
  70                 memcpy(decomp->data, &(decomp_last->data[decomp_last->head]), decomp_last_rem);
  71                 decomp->size = decomp_last_rem;
  72         }
  73
  74         src = (void *)event + sizeof(struct perf_record_compressed);
  75         src_size = event->pack.header.size - sizeof(struct perf_record_compressed);
  76
  77         decomp_size = zstd_decompress_stream(session->active_decomp->zstd_decomp, src, src_size,
  78                                 &(decomp->data[decomp_last_rem]), decomp_len - decomp_last_rem);
  79         if (!decomp_size) {
  80                 munmap(decomp, mmap_len);
  81                 pr_err("Couldn't decompress data\n");
  82                 return -1;
  83         }
  84
  85         decomp->size += decomp_size;
  86
  87         if (session->active_decomp->decomp == NULL)
  88                 session->active_decomp->decomp = decomp;
  89         else
  90                 session->active_decomp->decomp_last->next = decomp;
  91
  92         session->active_decomp->decomp_last = decomp;
  93
  94         pr_debug("decomp (B): %zd to %zd\n", src_size, decomp_size);
  95
  96         return 0;
  97 }
  98 #else /* !HAVE_ZSTD_SUPPORT */
  99 #define perf_session__process_compressed_event perf_session__process_compressed_event_stub
 100 #endif
 101
 102 static int perf_session__deliver_event(struct perf_session *session,
 103                                        union perf_event *event,
 104                                        struct perf_tool *tool,
 105                                        u64 file_offset,
 106                                        const char *file_path);
 107
 108 static int perf_session__open(struct perf_session *session, int repipe_fd)
 109 {
 110         struct perf_data *data = session->data;
 111
 112         if (perf_session__read_header(session, repipe_fd) < 0) {
 113                 pr_err("incompatible file format (rerun with -v to learn more)\n");
 114                 return -1;
 115         }
 116
 117         if (perf_data__is_pipe(data))
 118                 return 0;
 119
 120         if (perf_header__has_feat(&session->header, HEADER_STAT))
 121                 return 0;
 122
 123         if (!evlist__valid_sample_type(session->evlist)) {
 124                 pr_err("non matching sample_type\n");
 125                 return -1;
 126         }
 127
 128         if (!evlist__valid_sample_id_all(session->evlist)) {
 129                 pr_err("non matching sample_id_all\n");
 130                 return -1;
 131         }
 132
 133         if (!evlist__valid_read_format(session->evlist)) {
 134                 pr_err("non matching read_format\n");
 135                 return -1;
 136         }
 137
 138         return 0;
 139 }
 140
 141 void perf_session__set_id_hdr_size(struct perf_session *session)
 142 {
 143         u16 id_hdr_size = evlist__id_hdr_size(session->evlist);
 144
 145         machines__set_id_hdr_size(&session->machines, id_hdr_size);
 146 }
 147
 148 int perf_session__create_kernel_maps(struct perf_session *session)
 149 {
 150         int ret = machine__create_kernel_maps(&session->machines.host);
 151
 152         if (ret >= 0)
 153                 ret = machines__create_guest_kernel_maps(&session->machines);
 154         return ret;
 155 }
 156
 157 static void perf_session__destroy_kernel_maps(struct perf_session *session)
 158 {
 159         machines__destroy_kernel_maps(&session->machines);
 160 }
 161
 162 static bool perf_session__has_comm_exec(struct perf_session *session)
 163 {
 164         struct evsel *evsel;
 165
 166         evlist__for_each_entry(session->evlist, evsel) {
 167                 if (evsel->core.attr.comm_exec)
 168                         return true;
 169         }
 170
 171         return false;
 172 }
 173
 174 static void perf_session__set_comm_exec(struct perf_session *session)
 175 {
 176         bool comm_exec = perf_session__has_comm_exec(session);
 177
 178         machines__set_comm_exec(&session->machines, comm_exec);
 179 }
 180
 181 static int ordered_events__deliver_event(struct ordered_events *oe,
 182                                          struct ordered_event *event)
 183 {
 184         struct perf_session *session = container_of(oe, struct perf_session,
 185                                                     ordered_events);
 186
 187         return perf_session__deliver_event(session, event->event,
 188                                            session->tool, event->file_offset,
 189                                            event->file_path);
 190 }
 191
 192 struct perf_session *__perf_session__new(struct perf_data *data,
 193                                          bool repipe, int repipe_fd,
 194                                          struct perf_tool *tool)
 195 {
 196         int ret = -ENOMEM;
 197         struct perf_session *session = zalloc(sizeof(*session));
 198
 199         if (!session)
 200                 goto out;
 201
 202         session->repipe = repipe;
 203         session->tool   = tool;
 204         session->decomp_data.zstd_decomp = &session->zstd_data;
 205         session->active_decomp = &session->decomp_data;
 206         INIT_LIST_HEAD(&session->auxtrace_index);
 207         machines__init(&session->machines);
 208         ordered_events__init(&session->ordered_events,
 209                              ordered_events__deliver_event, NULL);
 210
 211         perf_env__init(&session->header.env);
 212         if (data) {
 213                 ret = perf_data__open(data);
 214                 if (ret < 0)
 215                         goto out_delete;
 216
 217                 session->data = data;
 218
 219                 if (perf_data__is_read(data)) {
 220                         ret = perf_session__open(session, repipe_fd);
 221                         if (ret < 0)
 222                                 goto out_delete;
 223
 224                         /*
 225                          * set session attributes that are present in perf.data
 226                          * but not in pipe-mode.
 227                          */
 228                         if (!data->is_pipe) {
 229                                 perf_session__set_id_hdr_size(session);
 230                                 perf_session__set_comm_exec(session);
 231                         }
 232
 233                         evlist__init_trace_event_sample_raw(session->evlist);
 234
 235                         /* Open the directory data. */
 236                         if (data->is_dir) {
 237                                 ret = perf_data__open_dir(data);
 238                                 if (ret)
 239                                         goto out_delete;
 240                         }
 241
 242                         if (!symbol_conf.kallsyms_name &&
 243                             !symbol_conf.vmlinux_name)
 244                                 symbol_conf.kallsyms_name = perf_data__kallsyms_name(data);
 245                 }
 246         } else  {
 247                 session->machines.host.env = &perf_env;
 248         }
 249
 250         session->machines.host.single_address_space =
 251                 perf_env__single_address_space(session->machines.host.env);
 252
 253         if (!data || perf_data__is_write(data)) {
 254                 /*
 255                  * In O_RDONLY mode this will be performed when reading the
 256                  * kernel MMAP event, in perf_event__process_mmap().
 257                  */
 258                 if (perf_session__create_kernel_maps(session) < 0)
 259                         pr_warning("Cannot read kernel map\n");
 260         }
 261
 262         /*
 263          * In pipe-mode, evlist is empty until PERF_RECORD_HEADER_ATTR is
 264          * processed, so evlist__sample_id_all is not meaningful here.
 265          */
 266         if ((!data || !data->is_pipe) && tool && tool->ordering_requires_timestamps &&
 267             tool->ordered_events && !evlist__sample_id_all(session->evlist)) {
 268                 dump_printf("WARNING: No sample_id_all support, falling back to unordered processing\n");
 269                 tool->ordered_events = false;
 270         }
 271
 272         return session;
 273
 274  out_delete:
 275         perf_session__delete(session);
 276  out:
 277         return ERR_PTR(ret);
 278 }
 279
 280 static void perf_session__delete_threads(struct perf_session *session)
 281 {
 282         machine__delete_threads(&session->machines.host);
 283 }
 284
 285 static void perf_decomp__release_events(struct decomp *next)
 286 {
 287         struct decomp *decomp;
 288         size_t mmap_len;
 289
 290         do {
 291                 decomp = next;
 292                 if (decomp == NULL)
 293                         break;
 294                 next = decomp->next;
 295                 mmap_len = decomp->mmap_len;
 296                 munmap(decomp, mmap_len);
 297         } while (1);
 298 }
 299
 300 void perf_session__delete(struct perf_session *session)
 301 {
 302         if (session == NULL)
 303                 return;
 304         auxtrace__free(session);
 305         auxtrace_index__free(&session->auxtrace_index);
 306         perf_session__destroy_kernel_maps(session);
 307         perf_session__delete_threads(session);
 308         perf_decomp__release_events(session->decomp_data.decomp);
 309         perf_env__exit(&session->header.env);
 310         machines__exit(&session->machines);
 311         if (session->data) {
 312                 if (perf_data__is_read(session->data))
 313                         evlist__delete(session->evlist);
 314                 perf_data__close(session->data);
 315         }
 316         trace_event__cleanup(&session->tevent);
 317         free(session);
 318 }
 319
 320 static int process_event_synth_tracing_data_stub(struct perf_session *session
 321                                                  __maybe_unused,
 322                                                  union perf_event *event
 323                                                  __maybe_unused)
 324 {
 325         dump_printf(": unhandled!\n");
 326         return 0;
 327 }
 328
 329 static int process_event_synth_attr_stub(struct perf_tool *tool __maybe_unused,
 330                                          union perf_event *event __maybe_unused,
 331                                          struct evlist **pevlist
 332                                          __maybe_unused)
 333 {
 334         dump_printf(": unhandled!\n");
 335         return 0;
 336 }
 337
 338 static int process_event_synth_event_update_stub(struct perf_tool *tool __maybe_unused,
 339                                                  union perf_event *event __maybe_unused,
 340                                                  struct evlist **pevlist
 341                                                  __maybe_unused)
 342 {
 343         if (dump_trace)
 344                 perf_event__fprintf_event_update(event, stdout);
 345
 346         dump_printf(": unhandled!\n");
 347         return 0;
 348 }
 349
 350 static int process_event_sample_stub(struct perf_tool *tool __maybe_unused,
 351                                      union perf_event *event __maybe_unused,
 352                                      struct perf_sample *sample __maybe_unused,
 353                                      struct evsel *evsel __maybe_unused,
 354                                      struct machine *machine __maybe_unused)
 355 {
 356         dump_printf(": unhandled!\n");
 357         return 0;
 358 }
 359
 360 static int process_event_stub(struct perf_tool *tool __maybe_unused,
 361                               union perf_event *event __maybe_unused,
 362                               struct perf_sample *sample __maybe_unused,
 363                               struct machine *machine __maybe_unused)
 364 {
 365         dump_printf(": unhandled!\n");
 366         return 0;
 367 }
 368
 369 static int process_finished_round_stub(struct perf_tool *tool __maybe_unused,
 370                                        union perf_event *event __maybe_unused,
 371                                        struct ordered_events *oe __maybe_unused)
 372 {
 373         dump_printf(": unhandled!\n");
 374         return 0;
 375 }
 376
 377 static int skipn(int fd, off_t n)
 378 {
 379         char buf[4096];
 380         ssize_t ret;
 381
 382         while (n > 0) {
 383                 ret = read(fd, buf, min(n, (off_t)sizeof(buf)));
 384                 if (ret <= 0)
 385                         return ret;
 386                 n -= ret;
 387         }
 388
 389         return 0;
 390 }
 391
 392 static s64 process_event_auxtrace_stub(struct perf_session *session __maybe_unused,
 393                                        union perf_event *event)
 394 {
 395         dump_printf(": unhandled!\n");
 396         if (perf_data__is_pipe(session->data))
 397                 skipn(perf_data__fd(session->data), event->auxtrace.size);
 398         return event->auxtrace.size;
 399 }
 400
 401 static int process_event_op2_stub(struct perf_session *session __maybe_unused,
 402                                   union perf_event *event __maybe_unused)
 403 {
 404         dump_printf(": unhandled!\n");
 405         return 0;
 406 }
 407
 408
 409 static
 410 int process_event_thread_map_stub(struct perf_session *session __maybe_unused,
 411                                   union perf_event *event __maybe_unused)
 412 {
 413         if (dump_trace)
 414                 perf_event__fprintf_thread_map(event, stdout);
 415
 416         dump_printf(": unhandled!\n");
 417         return 0;
 418 }
 419
 420 static
 421 int process_event_cpu_map_stub(struct perf_session *session __maybe_unused,
 422                                union perf_event *event __maybe_unused)
 423 {
 424         if (dump_trace)
 425                 perf_event__fprintf_cpu_map(event, stdout);
 426
 427         dump_printf(": unhandled!\n");
 428         return 0;
 429 }
 430
 431 static
 432 int process_event_stat_config_stub(struct perf_session *session __maybe_unused,
 433                                    union perf_event *event __maybe_unused)
 434 {
 435         if (dump_trace)
 436                 perf_event__fprintf_stat_config(event, stdout);
 437
 438         dump_printf(": unhandled!\n");
 439         return 0;
 440 }
 441
 442 static int process_stat_stub(struct perf_session *perf_session __maybe_unused,
 443                              union perf_event *event)
 444 {
 445         if (dump_trace)
 446                 perf_event__fprintf_stat(event, stdout);
 447
 448         dump_printf(": unhandled!\n");
 449         return 0;
 450 }
 451
 452 static int process_stat_round_stub(struct perf_session *perf_session __maybe_unused,
 453                                    union perf_event *event)
 454 {
 455         if (dump_trace)
 456                 perf_event__fprintf_stat_round(event, stdout);
 457
 458         dump_printf(": unhandled!\n");
 459         return 0;
 460 }
 461
 462 static int process_event_time_conv_stub(struct perf_session *perf_session __maybe_unused,
 463                                         union perf_event *event)
 464 {
 465         if (dump_trace)
 466                 perf_event__fprintf_time_conv(event, stdout);
 467
 468         dump_printf(": unhandled!\n");
 469         return 0;
 470 }
 471
 472 static int perf_session__process_compressed_event_stub(struct perf_session *session __maybe_unused,
 473                                                        union perf_event *event __maybe_unused,
 474                                                        u64 file_offset __maybe_unused,
 475                                                        const char *file_path __maybe_unused)
 476 {
 477        dump_printf(": unhandled!\n");
 478        return 0;
 479 }
 480
 481 void perf_tool__fill_defaults(struct perf_tool *tool)
 482 {
 483         if (tool->sample == NULL)
 484                 tool->sample = process_event_sample_stub;
 485         if (tool->mmap == NULL)
 486                 tool->mmap = process_event_stub;
 487         if (tool->mmap2 == NULL)
 488                 tool->mmap2 = process_event_stub;
 489         if (tool->comm == NULL)
 490                 tool->comm = process_event_stub;
 491         if (tool->namespaces == NULL)
 492                 tool->namespaces = process_event_stub;
 493         if (tool->cgroup == NULL)
 494                 tool->cgroup = process_event_stub;
 495         if (tool->fork == NULL)
 496                 tool->fork = process_event_stub;
 497         if (tool->exit == NULL)
 498                 tool->exit = process_event_stub;
 499         if (tool->lost == NULL)
 500                 tool->lost = perf_event__process_lost;
 501         if (tool->lost_samples == NULL)
 502                 tool->lost_samples = perf_event__process_lost_samples;
 503         if (tool->aux == NULL)
 504                 tool->aux = perf_event__process_aux;
 505         if (tool->itrace_start == NULL)
 506                 tool->itrace_start = perf_event__process_itrace_start;
 507         if (tool->context_switch == NULL)
 508                 tool->context_switch = perf_event__process_switch;
 509         if (tool->ksymbol == NULL)
 510                 tool->ksymbol = perf_event__process_ksymbol;
 511         if (tool->bpf == NULL)
 512                 tool->bpf = perf_event__process_bpf;
 513         if (tool->text_poke == NULL)
 514                 tool->text_poke = perf_event__process_text_poke;
 515         if (tool->aux_output_hw_id == NULL)
 516                 tool->aux_output_hw_id = perf_event__process_aux_output_hw_id;
 517         if (tool->read == NULL)
 518                 tool->read = process_event_sample_stub;
 519         if (tool->throttle == NULL)
 520                 tool->throttle = process_event_stub;
 521         if (tool->unthrottle == NULL)
 522                 tool->unthrottle = process_event_stub;
 523         if (tool->attr == NULL)
 524                 tool->attr = process_event_synth_attr_stub;
 525         if (tool->event_update == NULL)
 526                 tool->event_update = process_event_synth_event_update_stub;
 527         if (tool->tracing_data == NULL)
 528                 tool->tracing_data = process_event_synth_tracing_data_stub;
 529         if (tool->build_id == NULL)
 530                 tool->build_id = process_event_op2_stub;
 531         if (tool->finished_round == NULL) {
 532                 if (tool->ordered_events)
 533                         tool->finished_round = perf_event__process_finished_round;
 534                 else
 535                         tool->finished_round = process_finished_round_stub;
 536         }
 537         if (tool->id_index == NULL)
 538                 tool->id_index = process_event_op2_stub;
 539         if (tool->auxtrace_info == NULL)
 540                 tool->auxtrace_info = process_event_op2_stub;
 541         if (tool->auxtrace == NULL)
 542                 tool->auxtrace = process_event_auxtrace_stub;
 543         if (tool->auxtrace_error == NULL)
 544                 tool->auxtrace_error = process_event_op2_stub;
 545         if (tool->thread_map == NULL)
 546                 tool->thread_map = process_event_thread_map_stub;
 547         if (tool->cpu_map == NULL)
 548                 tool->cpu_map = process_event_cpu_map_stub;
 549         if (tool->stat_config == NULL)
 550                 tool->stat_config = process_event_stat_config_stub;
 551         if (tool->stat == NULL)
 552                 tool->stat = process_stat_stub;
 553         if (tool->stat_round == NULL)
 554                 tool->stat_round = process_stat_round_stub;
 555         if (tool->time_conv == NULL)
 556                 tool->time_conv = process_event_time_conv_stub;
 557         if (tool->feature == NULL)
 558                 tool->feature = process_event_op2_stub;
 559         if (tool->compressed == NULL)
 560                 tool->compressed = perf_session__process_compressed_event;
 561         if (tool->finished_init == NULL)
 562                 tool->finished_init = process_event_op2_stub;
 563 }
 564
 565 static void swap_sample_id_all(union perf_event *event, void *data)
 566 {
 567         void *end = (void *) event + event->header.size;
 568         int size = end - data;
 569
 570         BUG_ON(size % sizeof(u64));
 571         mem_bswap_64(data, size);
 572 }
 573
 574 static void perf_event__all64_swap(union perf_event *event,
 575                                    bool sample_id_all __maybe_unused)
 576 {
 577         struct perf_event_header *hdr = &event->header;
 578         mem_bswap_64(hdr + 1, event->header.size - sizeof(*hdr));
 579 }
 580
 581 static void perf_event__comm_swap(union perf_event *event, bool sample_id_all)
 582 {
 583         event->comm.pid = bswap_32(event->comm.pid);
 584         event->comm.tid = bswap_32(event->comm.tid);
 585
 586         if (sample_id_all) {
 587                 void *data = &event->comm.comm;
 588
 589                 data += PERF_ALIGN(strlen(data) + 1, sizeof(u64));
 590                 swap_sample_id_all(event, data);
 591         }
 592 }
 593
 594 static void perf_event__mmap_swap(union perf_event *event,
 595                                   bool sample_id_all)
 596 {
 597         event->mmap.pid   = bswap_32(event->mmap.pid);
 598         event->mmap.tid   = bswap_32(event->mmap.tid);
 599         event->mmap.start = bswap_64(event->mmap.start);
 600         event->mmap.len   = bswap_64(event->mmap.len);
 601         event->mmap.pgoff = bswap_64(event->mmap.pgoff);
 602
 603         if (sample_id_all) {
 604                 void *data = &event->mmap.filename;
 605
 606                 data += PERF_ALIGN(strlen(data) + 1, sizeof(u64));
 607                 swap_sample_id_all(event, data);
 608         }
 609 }
 610
 611 static void perf_event__mmap2_swap(union perf_event *event,
 612                                   bool sample_id_all)
 613 {
 614         event->mmap2.pid   = bswap_32(event->mmap2.pid);
 615         event->mmap2.tid   = bswap_32(event->mmap2.tid);
 616         event->mmap2.start = bswap_64(event->mmap2.start);
 617         event->mmap2.len   = bswap_64(event->mmap2.len);
 618         event->mmap2.pgoff = bswap_64(event->mmap2.pgoff);
 619
 620         if (!(event->header.misc & PERF_RECORD_MISC_MMAP_BUILD_ID)) {
 621                 event->mmap2.maj   = bswap_32(event->mmap2.maj);
 622                 event->mmap2.min   = bswap_32(event->mmap2.min);
 623                 event->mmap2.ino   = bswap_64(event->mmap2.ino);
 624                 event->mmap2.ino_generation = bswap_64(event->mmap2.ino_generation);
 625         }
 626
 627         if (sample_id_all) {
 628                 void *data = &event->mmap2.filename;
 629
 630                 data += PERF_ALIGN(strlen(data) + 1, sizeof(u64));
 631                 swap_sample_id_all(event, data);
 632         }
 633 }
 634 static void perf_event__task_swap(union perf_event *event, bool sample_id_all)
 635 {
 636         event->fork.pid  = bswap_32(event->fork.pid);
 637         event->fork.tid  = bswap_32(event->fork.tid);
 638         event->fork.ppid = bswap_32(event->fork.ppid);
 639         event->fork.ptid = bswap_32(event->fork.ptid);
 640         event->fork.time = bswap_64(event->fork.time);
 641
 642         if (sample_id_all)
 643                 swap_sample_id_all(event, &event->fork + 1);
 644 }
 645
 646 static void perf_event__read_swap(union perf_event *event, bool sample_id_all)
 647 {
 648         event->read.pid          = bswap_32(event->read.pid);
 649         event->read.tid          = bswap_32(event->read.tid);
 650         event->read.value        = bswap_64(event->read.value);
 651         event->read.time_enabled = bswap_64(event->read.time_enabled);
 652         event->read.time_running = bswap_64(event->read.time_running);
 653         event->read.id           = bswap_64(event->read.id);
 654
 655         if (sample_id_all)
 656                 swap_sample_id_all(event, &event->read + 1);
 657 }
 658
 659 static void perf_event__aux_swap(union perf_event *event, bool sample_id_all)
 660 {
 661         event->aux.aux_offset = bswap_64(event->aux.aux_offset);
 662         event->aux.aux_size   = bswap_64(event->aux.aux_size);
 663         event->aux.flags      = bswap_64(event->aux.flags);
 664
 665         if (sample_id_all)
 666                 swap_sample_id_all(event, &event->aux + 1);
 667 }
 668
 669 static void perf_event__itrace_start_swap(union perf_event *event,
 670                                           bool sample_id_all)
 671 {
 672         event->itrace_start.pid  = bswap_32(event->itrace_start.pid);
 673         event->itrace_start.tid  = bswap_32(event->itrace_start.tid);
 674
 675         if (sample_id_all)
 676                 swap_sample_id_all(event, &event->itrace_start + 1);
 677 }
 678
 679 static void perf_event__switch_swap(union perf_event *event, bool sample_id_all)
 680 {
 681         if (event->header.type == PERF_RECORD_SWITCH_CPU_WIDE) {
 682                 event->context_switch.next_prev_pid =
 683                                 bswap_32(event->context_switch.next_prev_pid);
 684                 event->context_switch.next_prev_tid =
 685                                 bswap_32(event->context_switch.next_prev_tid);
 686         }
 687
 688         if (sample_id_all)
 689                 swap_sample_id_all(event, &event->context_switch + 1);
 690 }
 691
 692 static void perf_event__text_poke_swap(union perf_event *event, bool sample_id_all)
 693 {
 694         event->text_poke.addr    = bswap_64(event->text_poke.addr);
 695         event->text_poke.old_len = bswap_16(event->text_poke.old_len);
 696         event->text_poke.new_len = bswap_16(event->text_poke.new_len);
 697
 698         if (sample_id_all) {
 699                 size_t len = sizeof(event->text_poke.old_len) +
 700                              sizeof(event->text_poke.new_len) +
 701                              event->text_poke.old_len +
 702                              event->text_poke.new_len;
 703                 void *data = &event->text_poke.old_len;
 704
 705                 data += PERF_ALIGN(len, sizeof(u64));
 706                 swap_sample_id_all(event, data);
 707         }
 708 }
 709
 710 static void perf_event__throttle_swap(union perf_event *event,
 711                                       bool sample_id_all)
 712 {
 713         event->throttle.time      = bswap_64(event->throttle.time);
 714         event->throttle.id        = bswap_64(event->throttle.id);
 715         event->throttle.stream_id = bswap_64(event->throttle.stream_id);
 716
 717         if (sample_id_all)
 718                 swap_sample_id_all(event, &event->throttle + 1);
 719 }
 720
 721 static void perf_event__namespaces_swap(union perf_event *event,
 722                                         bool sample_id_all)
 723 {
 724         u64 i;
 725
 726         event->namespaces.pid           = bswap_32(event->namespaces.pid);
 727         event->namespaces.tid           = bswap_32(event->namespaces.tid);
 728         event->namespaces.nr_namespaces = bswap_64(event->namespaces.nr_namespaces);
 729
 730         for (i = 0; i < event->namespaces.nr_namespaces; i++) {
 731                 struct perf_ns_link_info *ns = &event->namespaces.link_info[i];
 732
 733                 ns->dev = bswap_64(ns->dev);
 734                 ns->ino = bswap_64(ns->ino);
 735         }
 736
 737         if (sample_id_all)
 738                 swap_sample_id_all(event, &event->namespaces.link_info[i]);
 739 }
 740
 741 static void perf_event__cgroup_swap(union perf_event *event, bool sample_id_all)
 742 {
 743         event->cgroup.id = bswap_64(event->cgroup.id);
 744
 745         if (sample_id_all) {
 746                 void *data = &event->cgroup.path;
 747
 748                 data += PERF_ALIGN(strlen(data) + 1, sizeof(u64));
 749                 swap_sample_id_all(event, data);
 750         }
 751 }
 752
 753 static u8 revbyte(u8 b)
 754 {
 755         int rev = (b >> 4) | ((b & 0xf) << 4);
 756         rev = ((rev & 0xcc) >> 2) | ((rev & 0x33) << 2);
 757         rev = ((rev & 0xaa) >> 1) | ((rev & 0x55) << 1);
 758         return (u8) rev;
 759 }
 760
 761 /*
 762  * XXX this is hack in attempt to carry flags bitfield
 763  * through endian village. ABI says:
 764  *
 765  * Bit-fields are allocated from right to left (least to most significant)
 766  * on little-endian implementations and from left to right (most to least
 767  * significant) on big-endian implementations.
 768  *
 769  * The above seems to be byte specific, so we need to reverse each
 770  * byte of the bitfield. 'Internet' also says this might be implementation
 771  * specific and we probably need proper fix and carry perf_event_attr
 772  * bitfield flags in separate data file FEAT_ section. Thought this seems
 773  * to work for now.
 774  */
 775 static void swap_bitfield(u8 *p, unsigned len)
 776 {
 777         unsigned i;
 778
 779         for (i = 0; i < len; i++) {
 780                 *p = revbyte(*p);
 781                 p++;
 782         }
 783 }
 784
 785 /* exported for swapping attributes in file header */
 786 void perf_event__attr_swap(struct perf_event_attr *attr)
 787 {
 788         attr->type              = bswap_32(attr->type);
 789         attr->size              = bswap_32(attr->size);
 790
 791 #define bswap_safe(f, n)                                        \
 792         (attr->size > (offsetof(struct perf_event_attr, f) +    \
 793                        sizeof(attr->f) * (n)))
 794 #define bswap_field(f, sz)                      \
 795 do {                                            \
 796         if (bswap_safe(f, 0))                   \
 797                 attr->f = bswap_##sz(attr->f);  \
 798 } while(0)
 799 #define bswap_field_16(f) bswap_field(f, 16)
 800 #define bswap_field_32(f) bswap_field(f, 32)
 801 #define bswap_field_64(f) bswap_field(f, 64)
 802
 803         bswap_field_64(config);
 804         bswap_field_64(sample_period);
 805         bswap_field_64(sample_type);
 806         bswap_field_64(read_format);
 807         bswap_field_32(wakeup_events);
 808         bswap_field_32(bp_type);
 809         bswap_field_64(bp_addr);
 810         bswap_field_64(bp_len);
 811         bswap_field_64(branch_sample_type);
 812         bswap_field_64(sample_regs_user);
 813         bswap_field_32(sample_stack_user);
 814         bswap_field_32(aux_watermark);
 815         bswap_field_16(sample_max_stack);
 816         bswap_field_32(aux_sample_size);
 817
 818         /*
 819          * After read_format are bitfields. Check read_format because
 820          * we are unable to use offsetof on bitfield.
 821          */
 822         if (bswap_safe(read_format, 1))
 823                 swap_bitfield((u8 *) (&attr->read_format + 1),
 824                               sizeof(u64));
 825 #undef bswap_field_64
 826 #undef bswap_field_32
 827 #undef bswap_field
 828 #undef bswap_safe
 829 }
 830
 831 static void perf_event__hdr_attr_swap(union perf_event *event,
 832                                       bool sample_id_all __maybe_unused)
 833 {
 834         size_t size;
 835
 836         perf_event__attr_swap(&event->attr.attr);
 837
 838         size = event->header.size;
 839         size -= (void *)&event->attr.id - (void *)event;
 840         mem_bswap_64(event->attr.id, size);
 841 }
 842
 843 static void perf_event__event_update_swap(union perf_event *event,
 844                                           bool sample_id_all __maybe_unused)
 845 {
 846         event->event_update.type = bswap_64(event->event_update.type);
 847         event->event_update.id   = bswap_64(event->event_update.id);
 848 }
 849
 850 static void perf_event__event_type_swap(union perf_event *event,
 851                                         bool sample_id_all __maybe_unused)
 852 {
 853         event->event_type.event_type.event_id =
 854                 bswap_64(event->event_type.event_type.event_id);
 855 }
 856
 857 static void perf_event__tracing_data_swap(union perf_event *event,
 858                                           bool sample_id_all __maybe_unused)
 859 {
 860         event->tracing_data.size = bswap_32(event->tracing_data.size);
 861 }
 862
 863 static void perf_event__auxtrace_info_swap(union perf_event *event,
 864                                            bool sample_id_all __maybe_unused)
 865 {
 866         size_t size;
 867
 868         event->auxtrace_info.type = bswap_32(event->auxtrace_info.type);
 869
 870         size = event->header.size;
 871         size -= (void *)&event->auxtrace_info.priv - (void *)event;
 872         mem_bswap_64(event->auxtrace_info.priv, size);
 873 }
 874
 875 static void perf_event__auxtrace_swap(union perf_event *event,
 876                                       bool sample_id_all __maybe_unused)
 877 {
 878         event->auxtrace.size      = bswap_64(event->auxtrace.size);
 879         event->auxtrace.offset    = bswap_64(event->auxtrace.offset);
 880         event->auxtrace.reference = bswap_64(event->auxtrace.reference);
 881         event->auxtrace.idx       = bswap_32(event->auxtrace.idx);
 882         event->auxtrace.tid       = bswap_32(event->auxtrace.tid);
 883         event->auxtrace.cpu       = bswap_32(event->auxtrace.cpu);
 884 }
 885
 886 static void perf_event__auxtrace_error_swap(union perf_event *event,
 887                                             bool sample_id_all __maybe_unused)
 888 {
 889         event->auxtrace_error.type = bswap_32(event->auxtrace_error.type);
 890         event->auxtrace_error.code = bswap_32(event->auxtrace_error.code);
 891         event->auxtrace_error.cpu  = bswap_32(event->auxtrace_error.cpu);
 892         event->auxtrace_error.pid  = bswap_32(event->auxtrace_error.pid);
 893         event->auxtrace_error.tid  = bswap_32(event->auxtrace_error.tid);
 894         event->auxtrace_error.fmt  = bswap_32(event->auxtrace_error.fmt);
 895         event->auxtrace_error.ip   = bswap_64(event->auxtrace_error.ip);
 896         if (event->auxtrace_error.fmt)
 897                 event->auxtrace_error.time = bswap_64(event->auxtrace_error.time);
 898         if (event->auxtrace_error.fmt >= 2) {
 899                 event->auxtrace_error.machine_pid = bswap_32(event->auxtrace_error.machine_pid);
 900                 event->auxtrace_error.vcpu = bswap_32(event->auxtrace_error.vcpu);
 901         }
 902 }
 903
 904 static void perf_event__thread_map_swap(union perf_event *event,
 905                                         bool sample_id_all __maybe_unused)
 906 {
 907         unsigned i;
 908
 909         event->thread_map.nr = bswap_64(event->thread_map.nr);
 910
 911         for (i = 0; i < event->thread_map.nr; i++)
 912                 event->thread_map.entries[i].pid = bswap_64(event->thread_map.entries[i].pid);
 913 }
 914
 915 static void perf_event__cpu_map_swap(union perf_event *event,
 916                                      bool sample_id_all __maybe_unused)
 917 {
 918         struct perf_record_cpu_map_data *data = &event->cpu_map.data;
 919
 920         data->type = bswap_16(data->type);
 921
 922         switch (data->type) {
 923         case PERF_CPU_MAP__CPUS:
 924                 data->cpus_data.nr = bswap_16(data->cpus_data.nr);
 925
 926                 for (unsigned i = 0; i < data->cpus_data.nr; i++)
 927                         data->cpus_data.cpu[i] = bswap_16(data->cpus_data.cpu[i]);
 928                 break;
 929         case PERF_CPU_MAP__MASK:
 930                 data->mask32_data.long_size = bswap_16(data->mask32_data.long_size);
 931
 932                 switch (data->mask32_data.long_size) {
 933                 case 4:
 934                         data->mask32_data.nr = bswap_16(data->mask32_data.nr);
 935                         for (unsigned i = 0; i < data->mask32_data.nr; i++)
 936                                 data->mask32_data.mask[i] = bswap_32(data->mask32_data.mask[i]);
 937                         break;
 938                 case 8:
 939                         data->mask64_data.nr = bswap_16(data->mask64_data.nr);
 940                         for (unsigned i = 0; i < data->mask64_data.nr; i++)
 941                                 data->mask64_data.mask[i] = bswap_64(data->mask64_data.mask[i]);
 942                         break;
 943                 default:
 944                         pr_err("cpu_map swap: unsupported long size\n");
 945                 }
 946         default:
 947                 break;
 948         }
 949 }
 950
 951 static void perf_event__stat_config_swap(union perf_event *event,
 952                                          bool sample_id_all __maybe_unused)
 953 {
 954         u64 size;
 955
 956         size  = bswap_64(event->stat_config.nr) * sizeof(event->stat_config.data[0]);
 957         size += 1; /* nr item itself */
 958         mem_bswap_64(&event->stat_config.nr, size);
 959 }
 960
 961 static void perf_event__stat_swap(union perf_event *event,
 962                                   bool sample_id_all __maybe_unused)
 963 {
 964         event->stat.id     = bswap_64(event->stat.id);
 965         event->stat.thread = bswap_32(event->stat.thread);
 966         event->stat.cpu    = bswap_32(event->stat.cpu);
 967         event->stat.val    = bswap_64(event->stat.val);
 968         event->stat.ena    = bswap_64(event->stat.ena);
 969         event->stat.run    = bswap_64(event->stat.run);
 970 }
 971
 972 static void perf_event__stat_round_swap(union perf_event *event,
 973                                         bool sample_id_all __maybe_unused)
 974 {
 975         event->stat_round.type = bswap_64(event->stat_round.type);
 976         event->stat_round.time = bswap_64(event->stat_round.time);
 977 }
 978
 979 static void perf_event__time_conv_swap(union perf_event *event,
 980                                        bool sample_id_all __maybe_unused)
 981 {
 982         event->time_conv.time_shift = bswap_64(event->time_conv.time_shift);
 983         event->time_conv.time_mult  = bswap_64(event->time_conv.time_mult);
 984         event->time_conv.time_zero  = bswap_64(event->time_conv.time_zero);
 985
 986         if (event_contains(event->time_conv, time_cycles)) {
 987                 event->time_conv.time_cycles = bswap_64(event->time_conv.time_cycles);
 988                 event->time_conv.time_mask = bswap_64(event->time_conv.time_mask);
 989         }
 990 }
 991
 992 typedef void (*perf_event__swap_op)(union perf_event *event,
 993                                     bool sample_id_all);
 994
 995 static perf_event__swap_op perf_event__swap_ops[] = {
 996         [PERF_RECORD_MMAP]                = perf_event__mmap_swap,
 997         [PERF_RECORD_MMAP2]               = perf_event__mmap2_swap,
 998         [PERF_RECORD_COMM]                = perf_event__comm_swap,
 999         [PERF_RECORD_FORK]                = perf_event__task_swap,
1000         [PERF_RECORD_EXIT]                = perf_event__task_swap,
1001         [PERF_RECORD_LOST]                = perf_event__all64_swap,
1002         [PERF_RECORD_READ]                = perf_event__read_swap,
1003         [PERF_RECORD_THROTTLE]            = perf_event__throttle_swap,
1004         [PERF_RECORD_UNTHROTTLE]          = perf_event__throttle_swap,
1005         [PERF_RECORD_SAMPLE]              = perf_event__all64_swap,
1006         [PERF_RECORD_AUX]                 = perf_event__aux_swap,
1007         [PERF_RECORD_ITRACE_START]        = perf_event__itrace_start_swap,
1008         [PERF_RECORD_LOST_SAMPLES]        = perf_event__all64_swap,
1009         [PERF_RECORD_SWITCH]              = perf_event__switch_swap,
1010         [PERF_RECORD_SWITCH_CPU_WIDE]     = perf_event__switch_swap,
1011         [PERF_RECORD_NAMESPACES]          = perf_event__namespaces_swap,
1012         [PERF_RECORD_CGROUP]              = perf_event__cgroup_swap,
1013         [PERF_RECORD_TEXT_POKE]           = perf_event__text_poke_swap,
1014         [PERF_RECORD_AUX_OUTPUT_HW_ID]    = perf_event__all64_swap,
1015         [PERF_RECORD_HEADER_ATTR]         = perf_event__hdr_attr_swap,
1016         [PERF_RECORD_HEADER_EVENT_TYPE]   = perf_event__event_type_swap,
1017         [PERF_RECORD_HEADER_TRACING_DATA] = perf_event__tracing_data_swap,
1018         [PERF_RECORD_HEADER_BUILD_ID]     = NULL,
1019         [PERF_RECORD_ID_INDEX]            = perf_event__all64_swap,
1020         [PERF_RECORD_AUXTRACE_INFO]       = perf_event__auxtrace_info_swap,
1021         [PERF_RECORD_AUXTRACE]            = perf_event__auxtrace_swap,
1022         [PERF_RECORD_AUXTRACE_ERROR]      = perf_event__auxtrace_error_swap,
1023         [PERF_RECORD_THREAD_MAP]          = perf_event__thread_map_swap,
1024         [PERF_RECORD_CPU_MAP]             = perf_event__cpu_map_swap,
1025         [PERF_RECORD_STAT_CONFIG]         = perf_event__stat_config_swap,
1026         [PERF_RECORD_STAT]                = perf_event__stat_swap,
1027         [PERF_RECORD_STAT_ROUND]          = perf_event__stat_round_swap,
1028         [PERF_RECORD_EVENT_UPDATE]        = perf_event__event_update_swap,
1029         [PERF_RECORD_TIME_CONV]           = perf_event__time_conv_swap,
1030         [PERF_RECORD_HEADER_MAX]          = NULL,
1031 };
1032
1033 /*
1034  * When perf record finishes a pass on every buffers, it records this pseudo
1035  * event.
1036  * We record the max timestamp t found in the pass n.
1037  * Assuming these timestamps are monotonic across cpus, we know that if
1038  * a buffer still has events with timestamps below t, they will be all
1039  * available and then read in the pass n + 1.
1040  * Hence when we start to read the pass n + 2, we can safely flush every
1041  * events with timestamps below t.
1042  *
1043  *    ============ PASS n =================
1044  *       CPU 0         |   CPU 1
1045  *                     |
1046  *    cnt1 timestamps  |   cnt2 timestamps
1047  *          1          |         2
1048  *          2          |         3
1049  *          -          |         4  <--- max recorded
1050  *
1051  *    ============ PASS n + 1 ==============
1052  *       CPU 0         |   CPU 1
1053  *                     |
1054  *    cnt1 timestamps  |   cnt2 timestamps
1055  *          3          |         5
1056  *          4          |         6
1057  *          5          |         7 <---- max recorded
1058  *
1059  *      Flush every events below timestamp 4
1060  *
1061  *    ============ PASS n + 2 ==============
1062  *       CPU 0         |   CPU 1
1063  *                     |
1064  *    cnt1 timestamps  |   cnt2 timestamps
1065  *          6          |         8
1066  *          7          |         9
1067  *          -          |         10
1068  *
1069  *      Flush every events below timestamp 7
1070  *      etc...
1071  */
1072 int perf_event__process_finished_round(struct perf_tool *tool __maybe_unused,
1073                                        union perf_event *event __maybe_unused,
1074                                        struct ordered_events *oe)
1075 {
1076         if (dump_trace)
1077                 fprintf(stdout, "\n");
1078         return ordered_events__flush(oe, OE_FLUSH__ROUND);
1079 }
1080
1081 int perf_session__queue_event(struct perf_session *s, union perf_event *event,
1082                               u64 timestamp, u64 file_offset, const char *file_path)
1083 {
1084         return ordered_events__queue(&s->ordered_events, event, timestamp, file_offset, file_path);
1085 }
1086
1087 static void callchain__lbr_callstack_printf(struct perf_sample *sample)
1088 {
1089         struct ip_callchain *callchain = sample->callchain;
1090         struct branch_stack *lbr_stack = sample->branch_stack;
1091         struct branch_entry *entries = perf_sample__branch_entries(sample);
1092         u64 kernel_callchain_nr = callchain->nr;
1093         unsigned int i;
1094
1095         for (i = 0; i < kernel_callchain_nr; i++) {
1096                 if (callchain->ips[i] == PERF_CONTEXT_USER)
1097                         break;
1098         }
1099
1100         if ((i != kernel_callchain_nr) && lbr_stack->nr) {
1101                 u64 total_nr;
1102                 /*
1103                  * LBR callstack can only get user call chain,
1104                  * i is kernel call chain number,
1105                  * 1 is PERF_CONTEXT_USER.
1106                  *
1107                  * The user call chain is stored in LBR registers.
1108                  * LBR are pair registers. The caller is stored
1109                  * in "from" register, while the callee is stored
1110                  * in "to" register.
1111                  * For example, there is a call stack
1112                  * "A"->"B"->"C"->"D".
1113                  * The LBR registers will be recorded like
1114                  * "C"->"D", "B"->"C", "A"->"B".
1115                  * So only the first "to" register and all "from"
1116                  * registers are needed to construct the whole stack.
1117                  */
1118                 total_nr = i + 1 + lbr_stack->nr + 1;
1119                 kernel_callchain_nr = i + 1;
1120
1121                 printf("... LBR call chain: nr:%" PRIu64 "\n", total_nr);
1122
1123                 for (i = 0; i < kernel_callchain_nr; i++)
1124                         printf("..... %2d: %016" PRIx64 "\n",
1125                                i, callchain->ips[i]);
1126
1127                 printf("..... %2d: %016" PRIx64 "\n",
1128                        (int)(kernel_callchain_nr), entries[0].to);
1129                 for (i = 0; i < lbr_stack->nr; i++)
1130                         printf("..... %2d: %016" PRIx64 "\n",
1131                                (int)(i + kernel_callchain_nr + 1), entries[i].from);
1132         }
1133 }
1134
1135 static void callchain__printf(struct evsel *evsel,
1136                               struct perf_sample *sample)
1137 {
1138         unsigned int i;
1139         struct ip_callchain *callchain = sample->callchain;
1140
1141         if (evsel__has_branch_callstack(evsel))
1142                 callchain__lbr_callstack_printf(sample);
1143
1144         printf("... FP chain: nr:%" PRIu64 "\n", callchain->nr);
1145
1146         for (i = 0; i < callchain->nr; i++)
1147                 printf("..... %2d: %016" PRIx64 "\n",
1148                        i, callchain->ips[i]);
1149 }
1150
1151 static void branch_stack__printf(struct perf_sample *sample, bool callstack)
1152 {
1153         struct branch_entry *entries = perf_sample__branch_entries(sample);
1154         uint64_t i;
1155
1156         if (!callstack) {
1157                 printf("%s: nr:%" PRIu64 "\n", "... branch stack", sample->branch_stack->nr);
1158         } else {
1159                 /* the reason of adding 1 to nr is because after expanding
1160                  * branch stack it generates nr + 1 callstack records. e.g.,
1161                  *         B()->C()
1162                  *         A()->B()
1163                  * the final callstack should be:
1164                  *         C()
1165                  *         B()
1166                  *         A()
1167                  */
1168                 printf("%s: nr:%" PRIu64 "\n", "... branch callstack", sample->branch_stack->nr+1);
1169         }
1170
1171         for (i = 0; i < sample->branch_stack->nr; i++) {
1172                 struct branch_entry *e = &entries[i];
1173
1174                 if (!callstack) {
1175                         printf("..... %2"PRIu64": %016" PRIx64 " -> %016" PRIx64 " %hu cycles %s%s%s%s %x %s\n",
1176                                 i, e->from, e->to,
1177                                 (unsigned short)e->flags.cycles,
1178                                 e->flags.mispred ? "M" : " ",
1179                                 e->flags.predicted ? "P" : " ",
1180                                 e->flags.abort ? "A" : " ",
1181                                 e->flags.in_tx ? "T" : " ",
1182                                 (unsigned)e->flags.reserved,
1183                                 e->flags.type ? branch_type_name(e->flags.type) : "");
1184                 } else {
1185                         if (i == 0) {
1186                                 printf("..... %2"PRIu64": %016" PRIx64 "\n"
1187                                        "..... %2"PRIu64": %016" PRIx64 "\n",
1188                                                 i, e->to, i+1, e->from);
1189                         } else {
1190                                 printf("..... %2"PRIu64": %016" PRIx64 "\n", i+1, e->from);
1191                         }
1192                 }
1193         }
1194 }
1195
1196 static void regs_dump__printf(u64 mask, u64 *regs, const char *arch)
1197 {
1198         unsigned rid, i = 0;
1199
1200         for_each_set_bit(rid, (unsigned long *) &mask, sizeof(mask) * 8) {
1201                 u64 val = regs[i++];
1202
1203                 printf(".... %-5s 0x%016" PRIx64 "\n",
1204                        perf_reg_name(rid, arch), val);
1205         }
1206 }
1207
1208 static const char *regs_abi[] = {
1209         [PERF_SAMPLE_REGS_ABI_NONE] = "none",
1210         [PERF_SAMPLE_REGS_ABI_32] = "32-bit",
1211         [PERF_SAMPLE_REGS_ABI_64] = "64-bit",
1212 };
1213
1214 static inline const char *regs_dump_abi(struct regs_dump *d)
1215 {
1216         if (d->abi > PERF_SAMPLE_REGS_ABI_64)
1217                 return "unknown";
1218
1219         return regs_abi[d->abi];
1220 }
1221
1222 static void regs__printf(const char *type, struct regs_dump *regs, const char *arch)
1223 {
1224         u64 mask = regs->mask;
1225
1226         printf("... %s regs: mask 0x%" PRIx64 " ABI %s\n",
1227                type,
1228                mask,
1229                regs_dump_abi(regs));
1230
1231         regs_dump__printf(mask, regs->regs, arch);
1232 }
1233
1234 static void regs_user__printf(struct perf_sample *sample, const char *arch)
1235 {
1236         struct regs_dump *user_regs = &sample->user_regs;
1237
1238         if (user_regs->regs)
1239                 regs__printf("user", user_regs, arch);
1240 }
1241
1242 static void regs_intr__printf(struct perf_sample *sample, const char *arch)
1243 {
1244         struct regs_dump *intr_regs = &sample->intr_regs;
1245
1246         if (intr_regs->regs)
1247                 regs__printf("intr", intr_regs, arch);
1248 }
1249
1250 static void stack_user__printf(struct stack_dump *dump)
1251 {
1252         printf("... ustack: size %" PRIu64 ", offset 0x%x\n",
1253                dump->size, dump->offset);
1254 }
1255
1256 static void evlist__print_tstamp(struct evlist *evlist, union perf_event *event, struct perf_sample *sample)
1257 {
1258         u64 sample_type = __evlist__combined_sample_type(evlist);
1259
1260         if (event->header.type != PERF_RECORD_SAMPLE &&
1261             !evlist__sample_id_all(evlist)) {
1262                 fputs("-1 -1 ", stdout);
1263                 return;
1264         }
1265
1266         if ((sample_type & PERF_SAMPLE_CPU))
1267                 printf("%u ", sample->cpu);
1268
1269         if (sample_type & PERF_SAMPLE_TIME)
1270                 printf("%" PRIu64 " ", sample->time);
1271 }
1272
1273 static void sample_read__printf(struct perf_sample *sample, u64 read_format)
1274 {
1275         printf("... sample_read:\n");
1276
1277         if (read_format & PERF_FORMAT_TOTAL_TIME_ENABLED)
1278                 printf("...... time enabled %016" PRIx64 "\n",
1279                        sample->read.time_enabled);
1280
1281         if (read_format & PERF_FORMAT_TOTAL_TIME_RUNNING)
1282                 printf("...... time running %016" PRIx64 "\n",
1283                        sample->read.time_running);
1284
1285         if (read_format & PERF_FORMAT_GROUP) {
1286                 struct sample_read_value *value = sample->read.group.values;
1287
1288                 printf(".... group nr %" PRIu64 "\n", sample->read.group.nr);
1289
1290                 sample_read_group__for_each(value, sample->read.group.nr, read_format) {
1291                         printf("..... id %016" PRIx64
1292                                ", value %016" PRIx64,
1293                                value->id, value->value);
1294                         if (read_format & PERF_FORMAT_LOST)
1295                                 printf(", lost %" PRIu64, value->lost);
1296                         printf("\n");
1297                 }
1298         } else {
1299                 printf("..... id %016" PRIx64 ", value %016" PRIx64,
1300                         sample->read.one.id, sample->read.one.value);
1301                 if (read_format & PERF_FORMAT_LOST)
1302                         printf(", lost %" PRIu64, sample->read.one.lost);
1303                 printf("\n");
1304         }
1305 }
1306
1307 static void dump_event(struct evlist *evlist, union perf_event *event,
1308                        u64 file_offset, struct perf_sample *sample,
1309                        const char *file_path)
1310 {
1311         if (!dump_trace)
1312                 return;
1313
1314         printf("\n%#" PRIx64 "@%s [%#x]: event: %d\n",
1315                file_offset, file_path, event->header.size, event->header.type);
1316
1317         trace_event(event);
1318         if (event->header.type == PERF_RECORD_SAMPLE && evlist->trace_event_sample_raw)
1319                 evlist->trace_event_sample_raw(evlist, event, sample);
1320
1321         if (sample)
1322                 evlist__print_tstamp(evlist, event, sample);
1323
1324         printf("%#" PRIx64 " [%#x]: PERF_RECORD_%s", file_offset,
1325                event->header.size, perf_event__name(event->header.type));
1326 }
1327
1328 char *get_page_size_name(u64 size, char *str)
1329 {
1330         if (!size || !unit_number__scnprintf(str, PAGE_SIZE_NAME_LEN, size))
1331                 snprintf(str, PAGE_SIZE_NAME_LEN, "%s", "N/A");
1332
1333         return str;
1334 }
1335
1336 static void dump_sample(struct evsel *evsel, union perf_event *event,
1337                         struct perf_sample *sample, const char *arch)
1338 {
1339         u64 sample_type;
1340         char str[PAGE_SIZE_NAME_LEN];
1341
1342         if (!dump_trace)
1343                 return;
1344
1345         printf("(IP, 0x%x): %d/%d: %#" PRIx64 " period: %" PRIu64 " addr: %#" PRIx64 "\n",
1346                event->header.misc, sample->pid, sample->tid, sample->ip,
1347                sample->period, sample->addr);
1348
1349         sample_type = evsel->core.attr.sample_type;
1350
1351         if (evsel__has_callchain(evsel))
1352                 callchain__printf(evsel, sample);
1353
1354         if (evsel__has_br_stack(evsel))
1355                 branch_stack__printf(sample, evsel__has_branch_callstack(evsel));
1356
1357         if (sample_type & PERF_SAMPLE_REGS_USER)
1358                 regs_user__printf(sample, arch);
1359
1360         if (sample_type & PERF_SAMPLE_REGS_INTR)
1361                 regs_intr__printf(sample, arch);
1362
1363         if (sample_type & PERF_SAMPLE_STACK_USER)
1364                 stack_user__printf(&sample->user_stack);
1365
1366         if (sample_type & PERF_SAMPLE_WEIGHT_TYPE) {
1367                 printf("... weight: %" PRIu64 "", sample->weight);
1368                         if (sample_type & PERF_SAMPLE_WEIGHT_STRUCT) {
1369                                 printf(",0x%"PRIx16"", sample->ins_lat);
1370                                 printf(",0x%"PRIx16"", sample->p_stage_cyc);
1371                         }
1372                 printf("\n");
1373         }
1374
1375         if (sample_type & PERF_SAMPLE_DATA_SRC)
1376                 printf(" . data_src: 0x%"PRIx64"\n", sample->data_src);
1377
1378         if (sample_type & PERF_SAMPLE_PHYS_ADDR)
1379                 printf(" .. phys_addr: 0x%"PRIx64"\n", sample->phys_addr);
1380
1381         if (sample_type & PERF_SAMPLE_DATA_PAGE_SIZE)
1382                 printf(" .. data page size: %s\n", get_page_size_name(sample->data_page_size, str));
1383
1384         if (sample_type & PERF_SAMPLE_CODE_PAGE_SIZE)
1385                 printf(" .. code page size: %s\n", get_page_size_name(sample->code_page_size, str));
1386
1387         if (sample_type & PERF_SAMPLE_TRANSACTION)
1388                 printf("... transaction: %" PRIx64 "\n", sample->transaction);
1389
1390         if (sample_type & PERF_SAMPLE_READ)
1391                 sample_read__printf(sample, evsel->core.attr.read_format);
1392 }
1393
1394 static void dump_read(struct evsel *evsel, union perf_event *event)
1395 {
1396         struct perf_record_read *read_event = &event->read;
1397         u64 read_format;
1398
1399         if (!dump_trace)
1400                 return;
1401
1402         printf(": %d %d %s %" PRI_lu64 "\n", event->read.pid, event->read.tid,
1403                evsel__name(evsel), event->read.value);
1404
1405         if (!evsel)
1406                 return;
1407
1408         read_format = evsel->core.attr.read_format;
1409
1410         if (read_format & PERF_FORMAT_TOTAL_TIME_ENABLED)
1411                 printf("... time enabled : %" PRI_lu64 "\n", read_event->time_enabled);
1412
1413         if (read_format & PERF_FORMAT_TOTAL_TIME_RUNNING)
1414                 printf("... time running : %" PRI_lu64 "\n", read_event->time_running);
1415
1416         if (read_format & PERF_FORMAT_ID)
1417                 printf("... id           : %" PRI_lu64 "\n", read_event->id);
1418
1419         if (read_format & PERF_FORMAT_LOST)
1420                 printf("... lost         : %" PRI_lu64 "\n", read_event->lost);
1421 }
1422
1423 static struct machine *machines__find_for_cpumode(struct machines *machines,
1424                                                union perf_event *event,
1425                                                struct perf_sample *sample)
1426 {
1427         if (perf_guest &&
1428             ((sample->cpumode == PERF_RECORD_MISC_GUEST_KERNEL) ||
1429              (sample->cpumode == PERF_RECORD_MISC_GUEST_USER))) {
1430                 u32 pid;
1431
1432                 if (sample->machine_pid)
1433                         pid = sample->machine_pid;
1434                 else if (event->header.type == PERF_RECORD_MMAP
1435                     || event->header.type == PERF_RECORD_MMAP2)
1436                         pid = event->mmap.pid;
1437                 else
1438                         pid = sample->pid;
1439
1440                 /*
1441                  * Guest code machine is created as needed and does not use
1442                  * DEFAULT_GUEST_KERNEL_ID.
1443                  */
1444                 if (symbol_conf.guest_code)
1445                         return machines__findnew(machines, pid);
1446
1447                 return machines__find_guest(machines, pid);
1448         }
1449
1450         return &machines->host;
1451 }
1452
1453 static int deliver_sample_value(struct evlist *evlist,
1454                                 struct perf_tool *tool,
1455                                 union perf_event *event,
1456                                 struct perf_sample *sample,
1457                                 struct sample_read_value *v,
1458                                 struct machine *machine)
1459 {
1460         struct perf_sample_id *sid = evlist__id2sid(evlist, v->id);
1461         struct evsel *evsel;
1462
1463         if (sid) {
1464                 sample->id     = v->id;
1465                 sample->period = v->value - sid->period;
1466                 sid->period    = v->value;
1467         }
1468
1469         if (!sid || sid->evsel == NULL) {
1470                 ++evlist->stats.nr_unknown_id;
1471                 return 0;
1472         }
1473
1474         /*
1475          * There's no reason to deliver sample
1476          * for zero period, bail out.
1477          */
1478         if (!sample->period)
1479                 return 0;
1480
1481         evsel = container_of(sid->evsel, struct evsel, core);
1482         return tool->sample(tool, event, sample, evsel, machine);
1483 }
1484
1485 static int deliver_sample_group(struct evlist *evlist,
1486                                 struct perf_tool *tool,
1487                                 union  perf_event *event,
1488                                 struct perf_sample *sample,
1489                                 struct machine *machine,
1490                                 u64 read_format)
1491 {
1492         int ret = -EINVAL;
1493         struct sample_read_value *v = sample->read.group.values;
1494
1495         sample_read_group__for_each(v, sample->read.group.nr, read_format) {
1496                 ret = deliver_sample_value(evlist, tool, event, sample, v,
1497                                            machine);
1498                 if (ret)
1499                         break;
1500         }
1501
1502         return ret;
1503 }
1504
1505 static int evlist__deliver_sample(struct evlist *evlist, struct perf_tool *tool,
1506                                   union  perf_event *event, struct perf_sample *sample,
1507                                   struct evsel *evsel, struct machine *machine)
1508 {
1509         /* We know evsel != NULL. */
1510         u64 sample_type = evsel->core.attr.sample_type;
1511         u64 read_format = evsel->core.attr.read_format;
1512
1513         /* Standard sample delivery. */
1514         if (!(sample_type & PERF_SAMPLE_READ))
1515                 return tool->sample(tool, event, sample, evsel, machine);
1516
1517         /* For PERF_SAMPLE_READ we have either single or group mode. */
1518         if (read_format & PERF_FORMAT_GROUP)
1519                 return deliver_sample_group(evlist, tool, event, sample,
1520                                             machine, read_format);
1521         else
1522                 return deliver_sample_value(evlist, tool, event, sample,
1523                                             &sample->read.one, machine);
1524 }
1525
1526 static int machines__deliver_event(struct machines *machines,
1527                                    struct evlist *evlist,
1528                                    union perf_event *event,
1529                                    struct perf_sample *sample,
1530                                    struct perf_tool *tool, u64 file_offset,
1531                                    const char *file_path)
1532 {
1533         struct evsel *evsel;
1534         struct machine *machine;
1535
1536         dump_event(evlist, event, file_offset, sample, file_path);
1537
1538         evsel = evlist__id2evsel(evlist, sample->id);
1539
1540         machine = machines__find_for_cpumode(machines, event, sample);
1541
1542         switch (event->header.type) {
1543         case PERF_RECORD_SAMPLE:
1544                 if (evsel == NULL) {
1545                         ++evlist->stats.nr_unknown_id;
1546                         return 0;
1547                 }
1548                 if (machine == NULL) {
1549                         ++evlist->stats.nr_unprocessable_samples;
1550                         dump_sample(evsel, event, sample, perf_env__arch(NULL));
1551                         return 0;
1552                 }
1553                 dump_sample(evsel, event, sample, perf_env__arch(machine->env));
1554                 return evlist__deliver_sample(evlist, tool, event, sample, evsel, machine);
1555         case PERF_RECORD_MMAP:
1556                 return tool->mmap(tool, event, sample, machine);
1557         case PERF_RECORD_MMAP2:
1558                 if (event->header.misc & PERF_RECORD_MISC_PROC_MAP_PARSE_TIMEOUT)
1559                         ++evlist->stats.nr_proc_map_timeout;
1560                 return tool->mmap2(tool, event, sample, machine);
1561         case PERF_RECORD_COMM:
1562                 return tool->comm(tool, event, sample, machine);
1563         case PERF_RECORD_NAMESPACES:
1564                 return tool->namespaces(tool, event, sample, machine);
1565         case PERF_RECORD_CGROUP:
1566                 return tool->cgroup(tool, event, sample, machine);
1567         case PERF_RECORD_FORK:
1568                 return tool->fork(tool, event, sample, machine);
1569         case PERF_RECORD_EXIT:
1570                 return tool->exit(tool, event, sample, machine);
1571         case PERF_RECORD_LOST:
1572                 if (tool->lost == perf_event__process_lost)
1573                         evlist->stats.total_lost += event->lost.lost;
1574                 return tool->lost(tool, event, sample, machine);
1575         case PERF_RECORD_LOST_SAMPLES:
1576                 if (tool->lost_samples == perf_event__process_lost_samples)
1577                         evlist->stats.total_lost_samples += event->lost_samples.lost;
1578                 return tool->lost_samples(tool, event, sample, machine);
1579         case PERF_RECORD_READ:
1580                 dump_read(evsel, event);
1581                 return tool->read(tool, event, sample, evsel, machine);
1582         case PERF_RECORD_THROTTLE:
1583                 return tool->throttle(tool, event, sample, machine);
1584         case PERF_RECORD_UNTHROTTLE:
1585                 return tool->unthrottle(tool, event, sample, machine);
1586         case PERF_RECORD_AUX:
1587                 if (tool->aux == perf_event__process_aux) {
1588                         if (event->aux.flags & PERF_AUX_FLAG_TRUNCATED)
1589                                 evlist->stats.total_aux_lost += 1;
1590                         if (event->aux.flags & PERF_AUX_FLAG_PARTIAL)
1591                                 evlist->stats.total_aux_partial += 1;
1592                         if (event->aux.flags & PERF_AUX_FLAG_COLLISION)
1593                                 evlist->stats.total_aux_collision += 1;
1594                 }
1595                 return tool->aux(tool, event, sample, machine);
1596         case PERF_RECORD_ITRACE_START:
1597                 return tool->itrace_start(tool, event, sample, machine);
1598         case PERF_RECORD_SWITCH:
1599         case PERF_RECORD_SWITCH_CPU_WIDE:
1600                 return tool->context_switch(tool, event, sample, machine);
1601         case PERF_RECORD_KSYMBOL:
1602                 return tool->ksymbol(tool, event, sample, machine);
1603         case PERF_RECORD_BPF_EVENT:
1604                 return tool->bpf(tool, event, sample, machine);
1605         case PERF_RECORD_TEXT_POKE:
1606                 return tool->text_poke(tool, event, sample, machine);
1607         case PERF_RECORD_AUX_OUTPUT_HW_ID:
1608                 return tool->aux_output_hw_id(tool, event, sample, machine);
1609         default:
1610                 ++evlist->stats.nr_unknown_events;
1611                 return -1;
1612         }
1613 }
1614
1615 static int perf_session__deliver_event(struct perf_session *session,
1616                                        union perf_event *event,
1617                                        struct perf_tool *tool,
1618                                        u64 file_offset,
1619                                        const char *file_path)
1620 {
1621         struct perf_sample sample;
1622         int ret = evlist__parse_sample(session->evlist, event, &sample);
1623
1624         if (ret) {
1625                 pr_err("Can't parse sample, err = %d\n", ret);
1626                 return ret;
1627         }
1628
1629         ret = auxtrace__process_event(session, event, &sample, tool);
1630         if (ret < 0)
1631                 return ret;
1632         if (ret > 0)
1633                 return 0;
1634
1635         ret = machines__deliver_event(&session->machines, session->evlist,
1636                                       event, &sample, tool, file_offset, file_path);
1637
1638         if (dump_trace && sample.aux_sample.size)
1639                 auxtrace__dump_auxtrace_sample(session, &sample);
1640
1641         return ret;
1642 }
1643
1644 static s64 perf_session__process_user_event(struct perf_session *session,
1645                                             union perf_event *event,
1646                                             u64 file_offset,
1647                                             const char *file_path)
1648 {
1649         struct ordered_events *oe = &session->ordered_events;
1650         struct perf_tool *tool = session->tool;
1651         struct perf_sample sample = { .time = 0, };
1652         int fd = perf_data__fd(session->data);
1653         int err;
1654
1655         if (event->header.type != PERF_RECORD_COMPRESSED ||
1656             tool->compressed == perf_session__process_compressed_event_stub)
1657                 dump_event(session->evlist, event, file_offset, &sample, file_path);
1658
1659         /* These events are processed right away */
1660         switch (event->header.type) {
1661         case PERF_RECORD_HEADER_ATTR:
1662                 err = tool->attr(tool, event, &session->evlist);
1663                 if (err == 0) {
1664                         perf_session__set_id_hdr_size(session);
1665                         perf_session__set_comm_exec(session);
1666                 }
1667                 return err;
1668         case PERF_RECORD_EVENT_UPDATE:
1669                 return tool->event_update(tool, event, &session->evlist);
1670         case PERF_RECORD_HEADER_EVENT_TYPE:
1671                 /*
1672                  * Deprecated, but we need to handle it for sake
1673                  * of old data files create in pipe mode.
1674                  */
1675                 return 0;
1676         case PERF_RECORD_HEADER_TRACING_DATA:
1677                 /*
1678                  * Setup for reading amidst mmap, but only when we
1679                  * are in 'file' mode. The 'pipe' fd is in proper
1680                  * place already.
1681                  */
1682                 if (!perf_data__is_pipe(session->data))
1683                         lseek(fd, file_offset, SEEK_SET);
1684                 return tool->tracing_data(session, event);
1685         case PERF_RECORD_HEADER_BUILD_ID:
1686                 return tool->build_id(session, event);
1687         case PERF_RECORD_FINISHED_ROUND:
1688                 return tool->finished_round(tool, event, oe);
1689         case PERF_RECORD_ID_INDEX:
1690                 return tool->id_index(session, event);
1691         case PERF_RECORD_AUXTRACE_INFO:
1692                 return tool->auxtrace_info(session, event);
1693         case PERF_RECORD_AUXTRACE:
1694                 /* setup for reading amidst mmap */
1695                 lseek(fd, file_offset + event->header.size, SEEK_SET);
1696                 return tool->auxtrace(session, event);
1697         case PERF_RECORD_AUXTRACE_ERROR:
1698                 perf_session__auxtrace_error_inc(session, event);
1699                 return tool->auxtrace_error(session, event);
1700         case PERF_RECORD_THREAD_MAP:
1701                 return tool->thread_map(session, event);
1702         case PERF_RECORD_CPU_MAP:
1703                 return tool->cpu_map(session, event);
1704         case PERF_RECORD_STAT_CONFIG:
1705                 return tool->stat_config(session, event);
1706         case PERF_RECORD_STAT:
1707                 return tool->stat(session, event);
1708         case PERF_RECORD_STAT_ROUND:
1709                 return tool->stat_round(session, event);
1710         case PERF_RECORD_TIME_CONV:
1711                 session->time_conv = event->time_conv;
1712                 return tool->time_conv(session, event);
1713         case PERF_RECORD_HEADER_FEATURE:
1714                 return tool->feature(session, event);
1715         case PERF_RECORD_COMPRESSED:
1716                 err = tool->compressed(session, event, file_offset, file_path);
1717                 if (err)
1718                         dump_event(session->evlist, event, file_offset, &sample, file_path);
1719                 return err;
1720         case PERF_RECORD_FINISHED_INIT:
1721                 return tool->finished_init(session, event);
1722         default:
1723                 return -EINVAL;
1724         }
1725 }
1726
1727 int perf_session__deliver_synth_event(struct perf_session *session,
1728                                       union perf_event *event,
1729                                       struct perf_sample *sample)
1730 {
1731         struct evlist *evlist = session->evlist;
1732         struct perf_tool *tool = session->tool;
1733
1734         events_stats__inc(&evlist->stats, event->header.type);
1735
1736         if (event->header.type >= PERF_RECORD_USER_TYPE_START)
1737                 return perf_session__process_user_event(session, event, 0, NULL);
1738
1739         return machines__deliver_event(&session->machines, evlist, event, sample, tool, 0, NULL);
1740 }
1741
1742 static void event_swap(union perf_event *event, bool sample_id_all)
1743 {
1744         perf_event__swap_op swap;
1745
1746         swap = perf_event__swap_ops[event->header.type];
1747         if (swap)
1748                 swap(event, sample_id_all);
1749 }
1750
1751 int perf_session__peek_event(struct perf_session *session, off_t file_offset,
1752                              void *buf, size_t buf_sz,
1753                              union perf_event **event_ptr,
1754                              struct perf_sample *sample)
1755 {
1756         union perf_event *event;
1757         size_t hdr_sz, rest;
1758         int fd;
1759
1760         if (session->one_mmap && !session->header.needs_swap) {
1761                 event = file_offset - session->one_mmap_offset +
1762                         session->one_mmap_addr;
1763                 goto out_parse_sample;
1764         }
1765
1766         if (perf_data__is_pipe(session->data))
1767                 return -1;
1768
1769         fd = perf_data__fd(session->data);
1770         hdr_sz = sizeof(struct perf_event_header);
1771
1772         if (buf_sz < hdr_sz)
1773                 return -1;
1774
1775         if (lseek(fd, file_offset, SEEK_SET) == (off_t)-1 ||
1776             readn(fd, buf, hdr_sz) != (ssize_t)hdr_sz)
1777                 return -1;
1778
1779         event = (union perf_event *)buf;
1780
1781         if (session->header.needs_swap)
1782                 perf_event_header__bswap(&event->header);
1783
1784         if (event->header.size < hdr_sz || event->header.size > buf_sz)
1785                 return -1;
1786
1787         buf += hdr_sz;
1788         rest = event->header.size - hdr_sz;
1789
1790         if (readn(fd, buf, rest) != (ssize_t)rest)
1791                 return -1;
1792
1793         if (session->header.needs_swap)
1794                 event_swap(event, evlist__sample_id_all(session->evlist));
1795
1796 out_parse_sample:
1797
1798         if (sample && event->header.type < PERF_RECORD_USER_TYPE_START &&
1799             evlist__parse_sample(session->evlist, event, sample))
1800                 return -1;
1801
1802         *event_ptr = event;
1803
1804         return 0;
1805 }
1806
1807 int perf_session__peek_events(struct perf_session *session, u64 offset,
1808                               u64 size, peek_events_cb_t cb, void *data)
1809 {
1810         u64 max_offset = offset + size;
1811         char buf[PERF_SAMPLE_MAX_SIZE];
1812         union perf_event *event;
1813         int err;
1814
1815         do {
1816                 err = perf_session__peek_event(session, offset, buf,
1817                                                PERF_SAMPLE_MAX_SIZE, &event,
1818                                                NULL);
1819                 if (err)
1820                         return err;
1821
1822                 err = cb(session, event, offset, data);
1823                 if (err)
1824                         return err;
1825
1826                 offset += event->header.size;
1827                 if (event->header.type == PERF_RECORD_AUXTRACE)
1828                         offset += event->auxtrace.size;
1829
1830         } while (offset < max_offset);
1831
1832         return err;
1833 }
1834
1835 static s64 perf_session__process_event(struct perf_session *session,
1836                                        union perf_event *event, u64 file_offset,
1837                                        const char *file_path)
1838 {
1839         struct evlist *evlist = session->evlist;
1840         struct perf_tool *tool = session->tool;
1841         int ret;
1842
1843         if (session->header.needs_swap)
1844                 event_swap(event, evlist__sample_id_all(evlist));
1845
1846         if (event->header.type >= PERF_RECORD_HEADER_MAX)
1847                 return -EINVAL;
1848
1849         events_stats__inc(&evlist->stats, event->header.type);
1850
1851         if (event->header.type >= PERF_RECORD_USER_TYPE_START)
1852                 return perf_session__process_user_event(session, event, file_offset, file_path);
1853
1854         if (tool->ordered_events) {
1855                 u64 timestamp = -1ULL;
1856
1857                 ret = evlist__parse_sample_timestamp(evlist, event, &timestamp);
1858                 if (ret && ret != -1)
1859                         return ret;
1860
1861                 ret = perf_session__queue_event(session, event, timestamp, file_offset, file_path);
1862                 if (ret != -ETIME)
1863                         return ret;
1864         }
1865
1866         return perf_session__deliver_event(session, event, tool, file_offset, file_path);
1867 }
1868
1869 void perf_event_header__bswap(struct perf_event_header *hdr)
1870 {
1871         hdr->type = bswap_32(hdr->type);
1872         hdr->misc = bswap_16(hdr->misc);
1873         hdr->size = bswap_16(hdr->size);
1874 }
1875
1876 struct thread *perf_session__findnew(struct perf_session *session, pid_t pid)
1877 {
1878         return machine__findnew_thread(&session->machines.host, -1, pid);
1879 }
1880
1881 int perf_session__register_idle_thread(struct perf_session *session)
1882 {
1883         struct thread *thread = machine__idle_thread(&session->machines.host);
1884
1885         /* machine__idle_thread() got the thread, so put it */
1886         thread__put(thread);
1887         return thread ? 0 : -1;
1888 }
1889
1890 static void
1891 perf_session__warn_order(const struct perf_session *session)
1892 {
1893         const struct ordered_events *oe = &session->ordered_events;
1894         struct evsel *evsel;
1895         bool should_warn = true;
1896
1897         evlist__for_each_entry(session->evlist, evsel) {
1898                 if (evsel->core.attr.write_backward)
1899                         should_warn = false;
1900         }
1901
1902         if (!should_warn)
1903                 return;
1904         if (oe->nr_unordered_events != 0)
1905                 ui__warning("%u out of order events recorded.\n", oe->nr_unordered_events);
1906 }
1907
1908 static void perf_session__warn_about_errors(const struct perf_session *session)
1909 {
1910         const struct events_stats *stats = &session->evlist->stats;
1911
1912         if (session->tool->lost == perf_event__process_lost &&
1913             stats->nr_events[PERF_RECORD_LOST] != 0) {
1914                 ui__warning("Processed %d events and lost %d chunks!\n\n"
1915                             "Check IO/CPU overload!\n\n",
1916                             stats->nr_events[0],
1917                             stats->nr_events[PERF_RECORD_LOST]);
1918         }
1919
1920         if (session->tool->lost_samples == perf_event__process_lost_samples) {
1921                 double drop_rate;
1922
1923                 drop_rate = (double)stats->total_lost_samples /
1924                             (double) (stats->nr_events[PERF_RECORD_SAMPLE] + stats->total_lost_samples);
1925                 if (drop_rate > 0.05) {
1926                         ui__warning("Processed %" PRIu64 " samples and lost %3.2f%%!\n\n",
1927                                     stats->nr_events[PERF_RECORD_SAMPLE] + stats->total_lost_samples,
1928                                     drop_rate * 100.0);
1929                 }
1930         }
1931
1932         if (session->tool->aux == perf_event__process_aux &&
1933             stats->total_aux_lost != 0) {
1934                 ui__warning("AUX data lost %" PRIu64 " times out of %u!\n\n",
1935                             stats->total_aux_lost,
1936                             stats->nr_events[PERF_RECORD_AUX]);
1937         }
1938
1939         if (session->tool->aux == perf_event__process_aux &&
1940             stats->total_aux_partial != 0) {
1941                 bool vmm_exclusive = false;
1942
1943                 (void)sysfs__read_bool("module/kvm_intel/parameters/vmm_exclusive",
1944                                        &vmm_exclusive);
1945
1946                 ui__warning("AUX data had gaps in it %" PRIu64 " times out of %u!\n\n"
1947                             "Are you running a KVM guest in the background?%s\n\n",
1948                             stats->total_aux_partial,
1949                             stats->nr_events[PERF_RECORD_AUX],
1950                             vmm_exclusive ?
1951                             "\nReloading kvm_intel module with vmm_exclusive=0\n"
1952                             "will reduce the gaps to only guest's timeslices." :
1953                             "");
1954         }
1955
1956         if (session->tool->aux == perf_event__process_aux &&
1957             stats->total_aux_collision != 0) {
1958                 ui__warning("AUX data detected collision  %" PRIu64 " times out of %u!\n\n",
1959                             stats->total_aux_collision,
1960                             stats->nr_events[PERF_RECORD_AUX]);
1961         }
1962
1963         if (stats->nr_unknown_events != 0) {
1964                 ui__warning("Found %u unknown events!\n\n"
1965                             "Is this an older tool processing a perf.data "
1966                             "file generated by a more recent tool?\n\n"
1967                             "If that is not the case, consider "
1968                             "reporting to linux-kernel@vger.kernel.org.\n\n",
1969                             stats->nr_unknown_events);
1970         }
1971
1972         if (stats->nr_unknown_id != 0) {
1973                 ui__warning("%u samples with id not present in the header\n",
1974                             stats->nr_unknown_id);
1975         }
1976
1977         if (stats->nr_invalid_chains != 0) {
1978                 ui__warning("Found invalid callchains!\n\n"
1979                             "%u out of %u events were discarded for this reason.\n\n"
1980                             "Consider reporting to linux-kernel@vger.kernel.org.\n\n",
1981                             stats->nr_invalid_chains,
1982                             stats->nr_events[PERF_RECORD_SAMPLE]);
1983         }
1984
1985         if (stats->nr_unprocessable_samples != 0) {
1986                 ui__warning("%u unprocessable samples recorded.\n"
1987                             "Do you have a KVM guest running and not using 'perf kvm'?\n",
1988                             stats->nr_unprocessable_samples);
1989         }
1990
1991         perf_session__warn_order(session);
1992
1993         events_stats__auxtrace_error_warn(stats);
1994
1995         if (stats->nr_proc_map_timeout != 0) {
1996                 ui__warning("%d map information files for pre-existing threads were\n"
1997                             "not processed, if there are samples for addresses they\n"
1998                             "will not be resolved, you may find out which are these\n"
1999                             "threads by running with -v and redirecting the output\n"
2000                             "to a file.\n"
2001                             "The time limit to process proc map is too short?\n"
2002                             "Increase it by --proc-map-timeout\n",
2003                             stats->nr_proc_map_timeout);
2004         }
2005 }
2006
2007 static int perf_session__flush_thread_stack(struct thread *thread,
2008                                             void *p __maybe_unused)
2009 {
2010         return thread_stack__flush(thread);
2011 }
2012
2013 static int perf_session__flush_thread_stacks(struct perf_session *session)
2014 {
2015         return machines__for_each_thread(&session->machines,
2016                                          perf_session__flush_thread_stack,
2017                                          NULL);
2018 }
2019
2020 volatile int session_done;
2021
2022 static int __perf_session__process_decomp_events(struct perf_session *session);
2023
2024 static int __perf_session__process_pipe_events(struct perf_session *session)
2025 {
2026         struct ordered_events *oe = &session->ordered_events;
2027         struct perf_tool *tool = session->tool;
2028         union perf_event *event;
2029         uint32_t size, cur_size = 0;
2030         void *buf = NULL;
2031         s64 skip = 0;
2032         u64 head;
2033         ssize_t err;
2034         void *p;
2035
2036         perf_tool__fill_defaults(tool);
2037
2038         head = 0;
2039         cur_size = sizeof(union perf_event);
2040
2041         buf = malloc(cur_size);
2042         if (!buf)
2043                 return -errno;
2044         ordered_events__set_copy_on_queue(oe, true);
2045 more:
2046         event = buf;
2047         err = perf_data__read(session->data, event,
2048                               sizeof(struct perf_event_header));
2049         if (err <= 0) {
2050                 if (err == 0)
2051                         goto done;
2052
2053                 pr_err("failed to read event header\n");
2054                 goto out_err;
2055         }
2056
2057         if (session->header.needs_swap)
2058                 perf_event_header__bswap(&event->header);
2059
2060         size = event->header.size;
2061         if (size < sizeof(struct perf_event_header)) {
2062                 pr_err("bad event header size\n");
2063                 goto out_err;
2064         }
2065
2066         if (size > cur_size) {
2067                 void *new = realloc(buf, size);
2068                 if (!new) {
2069                         pr_err("failed to allocate memory to read event\n");
2070                         goto out_err;
2071                 }
2072                 buf = new;
2073                 cur_size = size;
2074                 event = buf;
2075         }
2076         p = event;
2077         p += sizeof(struct perf_event_header);
2078
2079         if (size - sizeof(struct perf_event_header)) {
2080                 err = perf_data__read(session->data, p,
2081                                       size - sizeof(struct perf_event_header));
2082                 if (err <= 0) {
2083                         if (err == 0) {
2084                                 pr_err("unexpected end of event stream\n");
2085                                 goto done;
2086                         }
2087
2088                         pr_err("failed to read event data\n");
2089                         goto out_err;
2090                 }
2091         }
2092
2093         if ((skip = perf_session__process_event(session, event, head, "pipe")) < 0) {
2094                 pr_err("%#" PRIx64 " [%#x]: failed to process type: %d\n",
2095                        head, event->header.size, event->header.type);
2096                 err = -EINVAL;
2097                 goto out_err;
2098         }
2099
2100         head += size;
2101
2102         if (skip > 0)
2103                 head += skip;
2104
2105         err = __perf_session__process_decomp_events(session);
2106         if (err)
2107                 goto out_err;
2108
2109         if (!session_done())
2110                 goto more;
2111 done:
2112         /* do the final flush for ordered samples */
2113         err = ordered_events__flush(oe, OE_FLUSH__FINAL);
2114         if (err)
2115                 goto out_err;
2116         err = auxtrace__flush_events(session, tool);
2117         if (err)
2118                 goto out_err;
2119         err = perf_session__flush_thread_stacks(session);
2120 out_err:
2121         free(buf);
2122         if (!tool->no_warn)
2123                 perf_session__warn_about_errors(session);
2124         ordered_events__free(&session->ordered_events);
2125         auxtrace__free_events(session);
2126         return err;
2127 }
2128
2129 static union perf_event *
2130 prefetch_event(char *buf, u64 head, size_t mmap_size,
2131                bool needs_swap, union perf_event *error)
2132 {
2133         union perf_event *event;
2134         u16 event_size;
2135
2136         /*
2137          * Ensure we have enough space remaining to read
2138          * the size of the event in the headers.
2139          */
2140         if (head + sizeof(event->header) > mmap_size)
2141                 return NULL;
2142
2143         event = (union perf_event *)(buf + head);
2144         if (needs_swap)
2145                 perf_event_header__bswap(&event->header);
2146
2147         event_size = event->header.size;
2148         if (head + event_size <= mmap_size)
2149                 return event;
2150
2151         /* We're not fetching the event so swap back again */
2152         if (needs_swap)
2153                 perf_event_header__bswap(&event->header);
2154
2155         /* Check if the event fits into the next mmapped buf. */
2156         if (event_size <= mmap_size - head % page_size) {
2157                 /* Remap buf and fetch again. */
2158                 return NULL;
2159         }
2160
2161         /* Invalid input. Event size should never exceed mmap_size. */
2162         pr_debug("%s: head=%#" PRIx64 " event->header.size=%#x, mmap_size=%#zx:"
2163                  " fuzzed or compressed perf.data?\n", __func__, head, event_size, mmap_size);
2164
2165         return error;
2166 }
2167
2168 static union perf_event *
2169 fetch_mmaped_event(u64 head, size_t mmap_size, char *buf, bool needs_swap)
2170 {
2171         return prefetch_event(buf, head, mmap_size, needs_swap, ERR_PTR(-EINVAL));
2172 }
2173
2174 static union perf_event *
2175 fetch_decomp_event(u64 head, size_t mmap_size, char *buf, bool needs_swap)
2176 {
2177         return prefetch_event(buf, head, mmap_size, needs_swap, NULL);
2178 }
2179
2180 static int __perf_session__process_decomp_events(struct perf_session *session)
2181 {
2182         s64 skip;
2183         u64 size;
2184         struct decomp *decomp = session->active_decomp->decomp_last;
2185
2186         if (!decomp)
2187                 return 0;
2188
2189         while (decomp->head < decomp->size && !session_done()) {
2190                 union perf_event *event = fetch_decomp_event(decomp->head, decomp->size, decomp->data,
2191                                                              session->header.needs_swap);
2192
2193                 if (!event)
2194                         break;
2195
2196                 size = event->header.size;
2197
2198                 if (size < sizeof(struct perf_event_header) ||
2199                     (skip = perf_session__process_event(session, event, decomp->file_pos,
2200                                                         decomp->file_path)) < 0) {
2201                         pr_err("%#" PRIx64 " [%#x]: failed to process type: %d\n",
2202                                 decomp->file_pos + decomp->head, event->header.size, event->header.type);
2203                         return -EINVAL;
2204                 }
2205
2206                 if (skip)
2207                         size += skip;
2208
2209                 decomp->head += size;
2210         }
2211
2212         return 0;
2213 }
2214
2215 /*
2216  * On 64bit we can mmap the data file in one go. No need for tiny mmap
2217  * slices. On 32bit we use 32MB.
2218  */
2219 #if BITS_PER_LONG == 64
2220 #define MMAP_SIZE ULLONG_MAX
2221 #define NUM_MMAPS 1
2222 #else
2223 #define MMAP_SIZE (32 * 1024 * 1024ULL)
2224 #define NUM_MMAPS 128
2225 #endif
2226
2227 struct reader;
2228
2229 typedef s64 (*reader_cb_t)(struct perf_session *session,
2230                            union perf_event *event,
2231                            u64 file_offset,
2232                            const char *file_path);
2233
2234 struct reader {
2235         int              fd;
2236         const char       *path;
2237         u64              data_size;
2238         u64              data_offset;
2239         reader_cb_t      process;
2240         bool             in_place_update;
2241         char             *mmaps[NUM_MMAPS];
2242         size_t           mmap_size;
2243         int              mmap_idx;
2244         char             *mmap_cur;
2245         u64              file_pos;
2246         u64              file_offset;
2247         u64              head;
2248         u64              size;
2249         bool             done;
2250         struct zstd_data   zstd_data;
2251         struct decomp_data decomp_data;
2252 };
2253
2254 static int
2255 reader__init(struct reader *rd, bool *one_mmap)
2256 {
2257         u64 data_size = rd->data_size;
2258         char **mmaps = rd->mmaps;
2259
2260         rd->head = rd->data_offset;
2261         data_size += rd->data_offset;
2262
2263         rd->mmap_size = MMAP_SIZE;
2264         if (rd->mmap_size > data_size) {
2265                 rd->mmap_size = data_size;
2266                 if (one_mmap)
2267                         *one_mmap = true;
2268         }
2269
2270         memset(mmaps, 0, sizeof(rd->mmaps));
2271
2272         if (zstd_init(&rd->zstd_data, 0))
2273                 return -1;
2274         rd->decomp_data.zstd_decomp = &rd->zstd_data;
2275
2276         return 0;
2277 }
2278
2279 static void
2280 reader__release_decomp(struct reader *rd)
2281 {
2282         perf_decomp__release_events(rd->decomp_data.decomp);
2283         zstd_fini(&rd->zstd_data);
2284 }
2285
2286 static int
2287 reader__mmap(struct reader *rd, struct perf_session *session)
2288 {
2289         int mmap_prot, mmap_flags;
2290         char *buf, **mmaps = rd->mmaps;
2291         u64 page_offset;
2292
2293         mmap_prot  = PROT_READ;
2294         mmap_flags = MAP_SHARED;
2295
2296         if (rd->in_place_update) {
2297                 mmap_prot  |= PROT_WRITE;
2298         } else if (session->header.needs_swap) {
2299                 mmap_prot  |= PROT_WRITE;
2300                 mmap_flags = MAP_PRIVATE;
2301         }
2302
2303         if (mmaps[rd->mmap_idx]) {
2304                 munmap(mmaps[rd->mmap_idx], rd->mmap_size);
2305                 mmaps[rd->mmap_idx] = NULL;
2306         }
2307
2308         page_offset = page_size * (rd->head / page_size);
2309         rd->file_offset += page_offset;
2310         rd->head -= page_offset;
2311
2312         buf = mmap(NULL, rd->mmap_size, mmap_prot, mmap_flags, rd->fd,
2313                    rd->file_offset);
2314         if (buf == MAP_FAILED) {
2315                 pr_err("failed to mmap file\n");
2316                 return -errno;
2317         }
2318         mmaps[rd->mmap_idx] = rd->mmap_cur = buf;
2319         rd->mmap_idx = (rd->mmap_idx + 1) & (ARRAY_SIZE(rd->mmaps) - 1);
2320         rd->file_pos = rd->file_offset + rd->head;
2321         if (session->one_mmap) {
2322                 session->one_mmap_addr = buf;
2323                 session->one_mmap_offset = rd->file_offset;
2324         }
2325
2326         return 0;
2327 }
2328
2329 enum {
2330         READER_OK,
2331         READER_NODATA,
2332 };
2333
2334 static int
2335 reader__read_event(struct reader *rd, struct perf_session *session,
2336                    struct ui_progress *prog)
2337 {
2338         u64 size;
2339         int err = READER_OK;
2340         union perf_event *event;
2341         s64 skip;
2342
2343         event = fetch_mmaped_event(rd->head, rd->mmap_size, rd->mmap_cur,
2344                                    session->header.needs_swap);
2345         if (IS_ERR(event))
2346                 return PTR_ERR(event);
2347
2348         if (!event)
2349                 return READER_NODATA;
2350
2351         size = event->header.size;
2352
2353         skip = -EINVAL;
2354
2355         if (size < sizeof(struct perf_event_header) ||
2356             (skip = rd->process(session, event, rd->file_pos, rd->path)) < 0) {
2357                 pr_err("%#" PRIx64 " [%#x]: failed to process type: %d [%s]\n",
2358                        rd->file_offset + rd->head, event->header.size,
2359                        event->header.type, strerror(-skip));
2360                 err = skip;
2361                 goto out;
2362         }
2363
2364         if (skip)
2365                 size += skip;
2366
2367         rd->size += size;
2368         rd->head += size;
2369         rd->file_pos += size;
2370
2371         err = __perf_session__process_decomp_events(session);
2372         if (err)
2373                 goto out;
2374
2375         ui_progress__update(prog, size);
2376
2377 out:
2378         return err;
2379 }
2380
2381 static inline bool
2382 reader__eof(struct reader *rd)
2383 {
2384         return (rd->file_pos >= rd->data_size + rd->data_offset);
2385 }
2386
2387 static int
2388 reader__process_events(struct reader *rd, struct perf_session *session,
2389                        struct ui_progress *prog)
2390 {
2391         int err;
2392
2393         err = reader__init(rd, &session->one_mmap);
2394         if (err)
2395                 goto out;
2396
2397         session->active_decomp = &rd->decomp_data;
2398
2399 remap:
2400         err = reader__mmap(rd, session);
2401         if (err)
2402                 goto out;
2403
2404 more:
2405         err = reader__read_event(rd, session, prog);
2406         if (err < 0)
2407                 goto out;
2408         else if (err == READER_NODATA)
2409                 goto remap;
2410
2411         if (session_done())
2412                 goto out;
2413
2414         if (!reader__eof(rd))
2415                 goto more;
2416
2417 out:
2418         session->active_decomp = &session->decomp_data;
2419         return err;
2420 }
2421
2422 static s64 process_simple(struct perf_session *session,
2423                           union perf_event *event,
2424                           u64 file_offset,
2425                           const char *file_path)
2426 {
2427         return perf_session__process_event(session, event, file_offset, file_path);
2428 }
2429
2430 static int __perf_session__process_events(struct perf_session *session)
2431 {
2432         struct reader rd = {
2433                 .fd             = perf_data__fd(session->data),
2434                 .path           = session->data->file.path,
2435                 .data_size      = session->header.data_size,
2436                 .data_offset    = session->header.data_offset,
2437                 .process        = process_simple,
2438                 .in_place_update = session->data->in_place_update,
2439         };
2440         struct ordered_events *oe = &session->ordered_events;
2441         struct perf_tool *tool = session->tool;
2442         struct ui_progress prog;
2443         int err;
2444
2445         perf_tool__fill_defaults(tool);
2446
2447         if (rd.data_size == 0)
2448                 return -1;
2449
2450         ui_progress__init_size(&prog, rd.data_size, "Processing events...");
2451
2452         err = reader__process_events(&rd, session, &prog);
2453         if (err)
2454                 goto out_err;
2455         /* do the final flush for ordered samples */
2456         err = ordered_events__flush(oe, OE_FLUSH__FINAL);
2457         if (err)
2458                 goto out_err;
2459         err = auxtrace__flush_events(session, tool);
2460         if (err)
2461                 goto out_err;
2462         err = perf_session__flush_thread_stacks(session);
2463 out_err:
2464         ui_progress__finish();
2465         if (!tool->no_warn)
2466                 perf_session__warn_about_errors(session);
2467         /*
2468          * We may switching perf.data output, make ordered_events
2469          * reusable.
2470          */
2471         ordered_events__reinit(&session->ordered_events);
2472         auxtrace__free_events(session);
2473         reader__release_decomp(&rd);
2474         session->one_mmap = false;
2475         return err;
2476 }
2477
2478 /*
2479  * Processing 2 MB of data from each reader in sequence,
2480  * because that's the way the ordered events sorting works
2481  * most efficiently.
2482  */
2483 #define READER_MAX_SIZE (2 * 1024 * 1024)
2484
2485 /*
2486  * This function reads, merge and process directory data.
2487  * It assumens the version 1 of directory data, where each
2488  * data file holds per-cpu data, already sorted by kernel.
2489  */
2490 static int __perf_session__process_dir_events(struct perf_session *session)
2491 {
2492         struct perf_data *data = session->data;
2493         struct perf_tool *tool = session->tool;
2494         int i, ret, readers, nr_readers;
2495         struct ui_progress prog;
2496         u64 total_size = perf_data__size(session->data);
2497         struct reader *rd;
2498
2499         perf_tool__fill_defaults(tool);
2500
2501         ui_progress__init_size(&prog, total_size, "Sorting events...");
2502
2503         nr_readers = 1;
2504         for (i = 0; i < data->dir.nr; i++) {
2505                 if (data->dir.files[i].size)
2506                         nr_readers++;
2507         }
2508
2509         rd = zalloc(nr_readers * sizeof(struct reader));
2510         if (!rd)
2511                 return -ENOMEM;
2512
2513         rd[0] = (struct reader) {
2514                 .fd              = perf_data__fd(session->data),
2515                 .path            = session->data->file.path,
2516                 .data_size       = session->header.data_size,
2517                 .data_offset     = session->header.data_offset,
2518                 .process         = process_simple,
2519                 .in_place_update = session->data->in_place_update,
2520         };
2521         ret = reader__init(&rd[0], NULL);
2522         if (ret)
2523                 goto out_err;
2524         ret = reader__mmap(&rd[0], session);
2525         if (ret)
2526                 goto out_err;
2527         readers = 1;
2528
2529         for (i = 0; i < data->dir.nr; i++) {
2530                 if (!data->dir.files[i].size)
2531                         continue;
2532                 rd[readers] = (struct reader) {
2533                         .fd              = data->dir.files[i].fd,
2534                         .path            = data->dir.files[i].path,
2535                         .data_size       = data->dir.files[i].size,
2536                         .data_offset     = 0,
2537                         .process         = process_simple,
2538                         .in_place_update = session->data->in_place_update,
2539                 };
2540                 ret = reader__init(&rd[readers], NULL);
2541                 if (ret)
2542                         goto out_err;
2543                 ret = reader__mmap(&rd[readers], session);
2544                 if (ret)
2545                         goto out_err;
2546                 readers++;
2547         }
2548
2549         i = 0;
2550         while (readers) {
2551                 if (session_done())
2552                         break;
2553
2554                 if (rd[i].done) {
2555                         i = (i + 1) % nr_readers;
2556                         continue;
2557                 }
2558                 if (reader__eof(&rd[i])) {
2559                         rd[i].done = true;
2560                         readers--;
2561                         continue;
2562                 }
2563
2564                 session->active_decomp = &rd[i].decomp_data;
2565                 ret = reader__read_event(&rd[i], session, &prog);
2566                 if (ret < 0) {
2567                         goto out_err;
2568                 } else if (ret == READER_NODATA) {
2569                         ret = reader__mmap(&rd[i], session);
2570                         if (ret)
2571                                 goto out_err;
2572                 }
2573
2574                 if (rd[i].size >= READER_MAX_SIZE) {
2575                         rd[i].size = 0;
2576                         i = (i + 1) % nr_readers;
2577                 }
2578         }
2579
2580         ret = ordered_events__flush(&session->ordered_events, OE_FLUSH__FINAL);
2581         if (ret)
2582                 goto out_err;
2583
2584         ret = perf_session__flush_thread_stacks(session);
2585 out_err:
2586         ui_progress__finish();
2587
2588         if (!tool->no_warn)
2589                 perf_session__warn_about_errors(session);
2590
2591         /*
2592          * We may switching perf.data output, make ordered_events
2593          * reusable.
2594          */
2595         ordered_events__reinit(&session->ordered_events);
2596
2597         session->one_mmap = false;
2598
2599         session->active_decomp = &session->decomp_data;
2600         for (i = 0; i < nr_readers; i++)
2601                 reader__release_decomp(&rd[i]);
2602         zfree(&rd);
2603
2604         return ret;
2605 }
2606
2607 int perf_session__process_events(struct perf_session *session)
2608 {
2609         if (perf_session__register_idle_thread(session) < 0)
2610                 return -ENOMEM;
2611
2612         if (perf_data__is_pipe(session->data))
2613                 return __perf_session__process_pipe_events(session);
2614
2615         if (perf_data__is_dir(session->data) && session->data->dir.nr)
2616                 return __perf_session__process_dir_events(session);
2617
2618         return __perf_session__process_events(session);
2619 }
2620
2621 bool perf_session__has_traces(struct perf_session *session, const char *msg)
2622 {
2623         struct evsel *evsel;
2624
2625         evlist__for_each_entry(session->evlist, evsel) {
2626                 if (evsel->core.attr.type == PERF_TYPE_TRACEPOINT)
2627                         return true;
2628         }
2629
2630         pr_err("No trace sample to read. Did you call 'perf %s'?\n", msg);
2631         return false;
2632 }
2633
2634 int map__set_kallsyms_ref_reloc_sym(struct map *map, const char *symbol_name, u64 addr)
2635 {
2636         char *bracket;
2637         struct ref_reloc_sym *ref;
2638         struct kmap *kmap;
2639
2640         ref = zalloc(sizeof(struct ref_reloc_sym));
2641         if (ref == NULL)
2642                 return -ENOMEM;
2643
2644         ref->name = strdup(symbol_name);
2645         if (ref->name == NULL) {
2646                 free(ref);
2647                 return -ENOMEM;
2648         }
2649
2650         bracket = strchr(ref->name, ']');
2651         if (bracket)
2652                 *bracket = '\0';
2653
2654         ref->addr = addr;
2655
2656         kmap = map__kmap(map);
2657         if (kmap)
2658                 kmap->ref_reloc_sym = ref;
2659
2660         return 0;
2661 }
2662
2663 size_t perf_session__fprintf_dsos(struct perf_session *session, FILE *fp)
2664 {
2665         return machines__fprintf_dsos(&session->machines, fp);
2666 }
2667
2668 size_t perf_session__fprintf_dsos_buildid(struct perf_session *session, FILE *fp,
2669                                           bool (skip)(struct dso *dso, int parm), int parm)
2670 {
2671         return machines__fprintf_dsos_buildid(&session->machines, fp, skip, parm);
2672 }
2673
2674 size_t perf_session__fprintf_nr_events(struct perf_session *session, FILE *fp,
2675                                        bool skip_empty)
2676 {
2677         size_t ret;
2678         const char *msg = "";
2679
2680         if (perf_header__has_feat(&session->header, HEADER_AUXTRACE))
2681                 msg = " (excludes AUX area (e.g. instruction trace) decoded / synthesized events)";
2682
2683         ret = fprintf(fp, "\nAggregated stats:%s\n", msg);
2684
2685         ret += events_stats__fprintf(&session->evlist->stats, fp, skip_empty);
2686         return ret;
2687 }
2688
2689 size_t perf_session__fprintf(struct perf_session *session, FILE *fp)
2690 {
2691         /*
2692          * FIXME: Here we have to actually print all the machines in this
2693          * session, not just the host...
2694          */
2695         return machine__fprintf(&session->machines.host, fp);
2696 }
2697
2698 struct evsel *perf_session__find_first_evtype(struct perf_session *session,
2699                                               unsigned int type)
2700 {
2701         struct evsel *pos;
2702
2703         evlist__for_each_entry(session->evlist, pos) {
2704                 if (pos->core.attr.type == type)
2705                         return pos;
2706         }
2707         return NULL;
2708 }
2709
2710 int perf_session__cpu_bitmap(struct perf_session *session,
2711                              const char *cpu_list, unsigned long *cpu_bitmap)
2712 {
2713         int i, err = -1;
2714         struct perf_cpu_map *map;
2715         int nr_cpus = min(session->header.env.nr_cpus_avail, MAX_NR_CPUS);
2716
2717         for (i = 0; i < PERF_TYPE_MAX; ++i) {
2718                 struct evsel *evsel;
2719
2720                 evsel = perf_session__find_first_evtype(session, i);
2721                 if (!evsel)
2722                         continue;
2723
2724                 if (!(evsel->core.attr.sample_type & PERF_SAMPLE_CPU)) {
2725                         pr_err("File does not contain CPU events. "
2726                                "Remove -C option to proceed.\n");
2727                         return -1;
2728                 }
2729         }
2730
2731         map = perf_cpu_map__new(cpu_list);
2732         if (map == NULL) {
2733                 pr_err("Invalid cpu_list\n");
2734                 return -1;
2735         }
2736
2737         for (i = 0; i < perf_cpu_map__nr(map); i++) {
2738                 struct perf_cpu cpu = perf_cpu_map__cpu(map, i);
2739
2740                 if (cpu.cpu >= nr_cpus) {
2741                         pr_err("Requested CPU %d too large. "
2742                                "Consider raising MAX_NR_CPUS\n", cpu.cpu);
2743                         goto out_delete_map;
2744                 }
2745
2746                 set_bit(cpu.cpu, cpu_bitmap);
2747         }
2748
2749         err = 0;
2750
2751 out_delete_map:
2752         perf_cpu_map__put(map);
2753         return err;
2754 }
2755
2756 void perf_session__fprintf_info(struct perf_session *session, FILE *fp,
2757                                 bool full)
2758 {
2759         if (session == NULL || fp == NULL)
2760                 return;
2761
2762         fprintf(fp, "# ========\n");
2763         perf_header__fprintf_info(session, fp, full);
2764         fprintf(fp, "# ========\n#\n");
2765 }
2766
2767 static int perf_session__register_guest(struct perf_session *session, pid_t machine_pid)
2768 {
2769         struct machine *machine = machines__findnew(&session->machines, machine_pid);
2770         struct thread *thread;
2771
2772         if (!machine)
2773                 return -ENOMEM;
2774
2775         machine->single_address_space = session->machines.host.single_address_space;
2776
2777         thread = machine__idle_thread(machine);
2778         if (!thread)
2779                 return -ENOMEM;
2780         thread__put(thread);
2781
2782         machine->kallsyms_filename = perf_data__guest_kallsyms_name(session->data, machine_pid);
2783
2784         return 0;
2785 }
2786
2787 static int perf_session__set_guest_cpu(struct perf_session *session, pid_t pid,
2788                                        pid_t tid, int guest_cpu)
2789 {
2790         struct machine *machine = &session->machines.host;
2791         struct thread *thread = machine__findnew_thread(machine, pid, tid);
2792
2793         if (!thread)
2794                 return -ENOMEM;
2795         thread->guest_cpu = guest_cpu;
2796         thread__put(thread);
2797
2798         return 0;
2799 }
2800
2801 int perf_event__process_id_index(struct perf_session *session,
2802                                  union perf_event *event)
2803 {
2804         struct evlist *evlist = session->evlist;
2805         struct perf_record_id_index *ie = &event->id_index;
2806         size_t sz = ie->header.size - sizeof(*ie);
2807         size_t i, nr, max_nr;
2808         size_t e1_sz = sizeof(struct id_index_entry);
2809         size_t e2_sz = sizeof(struct id_index_entry_2);
2810         size_t etot_sz = e1_sz + e2_sz;
2811         struct id_index_entry_2 *e2;
2812         pid_t last_pid = 0;
2813
2814         max_nr = sz / e1_sz;
2815         nr = ie->nr;
2816         if (nr > max_nr) {
2817                 printf("Too big: nr %zu max_nr %zu\n", nr, max_nr);
2818                 return -EINVAL;
2819         }
2820
2821         if (sz >= nr * etot_sz) {
2822                 max_nr = sz / etot_sz;
2823                 if (nr > max_nr) {
2824                         printf("Too big2: nr %zu max_nr %zu\n", nr, max_nr);
2825                         return -EINVAL;
2826                 }
2827                 e2 = (void *)ie + sizeof(*ie) + nr * e1_sz;
2828         } else {
2829                 e2 = NULL;
2830         }
2831
2832         if (dump_trace)
2833                 fprintf(stdout, " nr: %zu\n", nr);
2834
2835         for (i = 0; i < nr; i++, (e2 ? e2++ : 0)) {
2836                 struct id_index_entry *e = &ie->entries[i];
2837                 struct perf_sample_id *sid;
2838                 int ret;
2839
2840                 if (dump_trace) {
2841                         fprintf(stdout, " ... id: %"PRI_lu64, e->id);
2842                         fprintf(stdout, "  idx: %"PRI_lu64, e->idx);
2843                         fprintf(stdout, "  cpu: %"PRI_ld64, e->cpu);
2844                         fprintf(stdout, "  tid: %"PRI_ld64, e->tid);
2845                         if (e2) {
2846                                 fprintf(stdout, "  machine_pid: %"PRI_ld64, e2->machine_pid);
2847                                 fprintf(stdout, "  vcpu: %"PRI_lu64"\n", e2->vcpu);
2848                         } else {
2849                                 fprintf(stdout, "\n");
2850                         }
2851                 }
2852
2853                 sid = evlist__id2sid(evlist, e->id);
2854                 if (!sid)
2855                         return -ENOENT;
2856
2857                 sid->idx = e->idx;
2858                 sid->cpu.cpu = e->cpu;
2859                 sid->tid = e->tid;
2860
2861                 if (!e2)
2862                         continue;
2863
2864                 sid->machine_pid = e2->machine_pid;
2865                 sid->vcpu.cpu = e2->vcpu;
2866
2867                 if (!sid->machine_pid)
2868                         continue;
2869
2870                 if (sid->machine_pid != last_pid) {
2871                         ret = perf_session__register_guest(session, sid->machine_pid);
2872                         if (ret)
2873                                 return ret;
2874                         last_pid = sid->machine_pid;
2875                         perf_guest = true;
2876                 }
2877
2878                 ret = perf_session__set_guest_cpu(session, sid->machine_pid, e->tid, e2->vcpu);
2879                 if (ret)
2880                         return ret;
2881         }
2882         return 0;
2883 }