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 "evlist.h"
  19 #include "evsel.h"
  20 #include "memswap.h"
  21 #include "map.h"
  22 #include "symbol.h"
  23 #include "session.h"
  24 #include "tool.h"
  25 #include "perf_regs.h"
  26 #include "asm/bug.h"
  27 #include "auxtrace.h"
  28 #include "thread.h"
  29 #include "thread-stack.h"
  30 #include "sample-raw.h"
  31 #include "stat.h"
  32 #include "tsc.h"
  33 #include "ui/progress.h"
  34 #include "../perf.h"
  35 #include "arch/common.h"
  36 #include "units.h"
  37 #include <internal/lib.h>
  38
  39 #ifdef HAVE_ZSTD_SUPPORT
  40 static int perf_session__process_compressed_event(struct perf_session *session,
  41                                                   union perf_event *event, u64 file_offset)
  42 {
  43         void *src;
  44         size_t decomp_size, src_size;
  45         u64 decomp_last_rem = 0;
  46         size_t mmap_len, decomp_len = session->header.env.comp_mmap_len;
  47         struct decomp *decomp, *decomp_last = session->decomp_last;
  48
  49         if (decomp_last) {
  50                 decomp_last_rem = decomp_last->size - decomp_last->head;
  51                 decomp_len += decomp_last_rem;
  52         }
  53
  54         mmap_len = sizeof(struct decomp) + decomp_len;
  55         decomp = mmap(NULL, mmap_len, PROT_READ|PROT_WRITE,
  56                       MAP_ANONYMOUS|MAP_PRIVATE, -1, 0);
  57         if (decomp == MAP_FAILED) {
  58                 pr_err("Couldn't allocate memory for decompression\n");
  59                 return -1;
  60         }
  61
  62         decomp->file_pos = file_offset;
  63         decomp->mmap_len = mmap_len;
  64         decomp->head = 0;
  65
  66         if (decomp_last_rem) {
  67                 memcpy(decomp->data, &(decomp_last->data[decomp_last->head]), decomp_last_rem);
  68                 decomp->size = decomp_last_rem;
  69         }
  70
  71         src = (void *)event + sizeof(struct perf_record_compressed);
  72         src_size = event->pack.header.size - sizeof(struct perf_record_compressed);
  73
  74         decomp_size = zstd_decompress_stream(&(session->zstd_data), src, src_size,
  75                                 &(decomp->data[decomp_last_rem]), decomp_len - decomp_last_rem);
  76         if (!decomp_size) {
  77                 munmap(decomp, mmap_len);
  78                 pr_err("Couldn't decompress data\n");
  79                 return -1;
  80         }
  81
  82         decomp->size += decomp_size;
  83
  84         if (session->decomp == NULL) {
  85                 session->decomp = decomp;
  86                 session->decomp_last = decomp;
  87         } else {
  88                 session->decomp_last->next = decomp;
  89                 session->decomp_last = decomp;
  90         }
  91
  92         pr_debug("decomp (B): %zd to %zd\n", src_size, decomp_size);
  93
  94         return 0;
  95 }
  96 #else /* !HAVE_ZSTD_SUPPORT */
  97 #define perf_session__process_compressed_event perf_session__process_compressed_event_stub
  98 #endif
  99
 100 static int perf_session__deliver_event(struct perf_session *session,
 101                                        union perf_event *event,
 102                                        struct perf_tool *tool,
 103                                        u64 file_offset);
 104
 105 static int perf_session__open(struct perf_session *session)
 106 {
 107         struct perf_data *data = session->data;
 108
 109         if (perf_session__read_header(session) < 0) {
 110                 pr_err("incompatible file format (rerun with -v to learn more)\n");
 111                 return -1;
 112         }
 113
 114         if (perf_data__is_pipe(data))
 115                 return 0;
 116
 117         if (perf_header__has_feat(&session->header, HEADER_STAT))
 118                 return 0;
 119
 120         if (!evlist__valid_sample_type(session->evlist)) {
 121                 pr_err("non matching sample_type\n");
 122                 return -1;
 123         }
 124
 125         if (!evlist__valid_sample_id_all(session->evlist)) {
 126                 pr_err("non matching sample_id_all\n");
 127                 return -1;
 128         }
 129
 130         if (!evlist__valid_read_format(session->evlist)) {
 131                 pr_err("non matching read_format\n");
 132                 return -1;
 133         }
 134
 135         return 0;
 136 }
 137
 138 void perf_session__set_id_hdr_size(struct perf_session *session)
 139 {
 140         u16 id_hdr_size = evlist__id_hdr_size(session->evlist);
 141
 142         machines__set_id_hdr_size(&session->machines, id_hdr_size);
 143 }
 144
 145 int perf_session__create_kernel_maps(struct perf_session *session)
 146 {
 147         int ret = machine__create_kernel_maps(&session->machines.host);
 148
 149         if (ret >= 0)
 150                 ret = machines__create_guest_kernel_maps(&session->machines);
 151         return ret;
 152 }
 153
 154 static void perf_session__destroy_kernel_maps(struct perf_session *session)
 155 {
 156         machines__destroy_kernel_maps(&session->machines);
 157 }
 158
 159 static bool perf_session__has_comm_exec(struct perf_session *session)
 160 {
 161         struct evsel *evsel;
 162
 163         evlist__for_each_entry(session->evlist, evsel) {
 164                 if (evsel->core.attr.comm_exec)
 165                         return true;
 166         }
 167
 168         return false;
 169 }
 170
 171 static void perf_session__set_comm_exec(struct perf_session *session)
 172 {
 173         bool comm_exec = perf_session__has_comm_exec(session);
 174
 175         machines__set_comm_exec(&session->machines, comm_exec);
 176 }
 177
 178 static int ordered_events__deliver_event(struct ordered_events *oe,
 179                                          struct ordered_event *event)
 180 {
 181         struct perf_session *session = container_of(oe, struct perf_session,
 182                                                     ordered_events);
 183
 184         return perf_session__deliver_event(session, event->event,
 185                                            session->tool, event->file_offset);
 186 }
 187
 188 struct perf_session *perf_session__new(struct perf_data *data,
 189                                        bool repipe, struct perf_tool *tool)
 190 {
 191         int ret = -ENOMEM;
 192         struct perf_session *session = zalloc(sizeof(*session));
 193
 194         if (!session)
 195                 goto out;
 196
 197         session->repipe = repipe;
 198         session->tool   = tool;
 199         INIT_LIST_HEAD(&session->auxtrace_index);
 200         machines__init(&session->machines);
 201         ordered_events__init(&session->ordered_events,
 202                              ordered_events__deliver_event, NULL);
 203
 204         perf_env__init(&session->header.env);
 205         if (data) {
 206                 ret = perf_data__open(data);
 207                 if (ret < 0)
 208                         goto out_delete;
 209
 210                 session->data = data;
 211
 212                 if (perf_data__is_read(data)) {
 213                         ret = perf_session__open(session);
 214                         if (ret < 0)
 215                                 goto out_delete;
 216
 217                         /*
 218                          * set session attributes that are present in perf.data
 219                          * but not in pipe-mode.
 220                          */
 221                         if (!data->is_pipe) {
 222                                 perf_session__set_id_hdr_size(session);
 223                                 perf_session__set_comm_exec(session);
 224                         }
 225
 226                         evlist__init_trace_event_sample_raw(session->evlist);
 227
 228                         /* Open the directory data. */
 229                         if (data->is_dir) {
 230                                 ret = perf_data__open_dir(data);
 231                                 if (ret)
 232                                         goto out_delete;
 233                         }
 234
 235                         if (!symbol_conf.kallsyms_name &&
 236                             !symbol_conf.vmlinux_name)
 237                                 symbol_conf.kallsyms_name = perf_data__kallsyms_name(data);
 238                 }
 239         } else  {
 240                 session->machines.host.env = &perf_env;
 241         }
 242
 243         session->machines.host.single_address_space =
 244                 perf_env__single_address_space(session->machines.host.env);
 245
 246         if (!data || perf_data__is_write(data)) {
 247                 /*
 248                  * In O_RDONLY mode this will be performed when reading the
 249                  * kernel MMAP event, in perf_event__process_mmap().
 250                  */
 251                 if (perf_session__create_kernel_maps(session) < 0)
 252                         pr_warning("Cannot read kernel map\n");
 253         }
 254
 255         /*
 256          * In pipe-mode, evlist is empty until PERF_RECORD_HEADER_ATTR is
 257          * processed, so evlist__sample_id_all is not meaningful here.
 258          */
 259         if ((!data || !data->is_pipe) && tool && tool->ordering_requires_timestamps &&
 260             tool->ordered_events && !evlist__sample_id_all(session->evlist)) {
 261                 dump_printf("WARNING: No sample_id_all support, falling back to unordered processing\n");
 262                 tool->ordered_events = false;
 263         }
 264
 265         return session;
 266
 267  out_delete:
 268         perf_session__delete(session);
 269  out:
 270         return ERR_PTR(ret);
 271 }
 272
 273 static void perf_session__delete_threads(struct perf_session *session)
 274 {
 275         machine__delete_threads(&session->machines.host);
 276 }
 277
 278 static void perf_session__release_decomp_events(struct perf_session *session)
 279 {
 280         struct decomp *next, *decomp;
 281         size_t mmap_len;
 282         next = session->decomp;
 283         do {
 284                 decomp = next;
 285                 if (decomp == NULL)
 286                         break;
 287                 next = decomp->next;
 288                 mmap_len = decomp->mmap_len;
 289                 munmap(decomp, mmap_len);
 290         } while (1);
 291 }
 292
 293 void perf_session__delete(struct perf_session *session)
 294 {
 295         if (session == NULL)
 296                 return;
 297         auxtrace__free(session);
 298         auxtrace_index__free(&session->auxtrace_index);
 299         perf_session__destroy_kernel_maps(session);
 300         perf_session__delete_threads(session);
 301         perf_session__release_decomp_events(session);
 302         perf_env__exit(&session->header.env);
 303         machines__exit(&session->machines);
 304         if (session->data)
 305                 perf_data__close(session->data);
 306         free(session);
 307 }
 308
 309 static int process_event_synth_tracing_data_stub(struct perf_session *session
 310                                                  __maybe_unused,
 311                                                  union perf_event *event
 312                                                  __maybe_unused)
 313 {
 314         dump_printf(": unhandled!\n");
 315         return 0;
 316 }
 317
 318 static int process_event_synth_attr_stub(struct perf_tool *tool __maybe_unused,
 319                                          union perf_event *event __maybe_unused,
 320                                          struct evlist **pevlist
 321                                          __maybe_unused)
 322 {
 323         dump_printf(": unhandled!\n");
 324         return 0;
 325 }
 326
 327 static int process_event_synth_event_update_stub(struct perf_tool *tool __maybe_unused,
 328                                                  union perf_event *event __maybe_unused,
 329                                                  struct evlist **pevlist
 330                                                  __maybe_unused)
 331 {
 332         if (dump_trace)
 333                 perf_event__fprintf_event_update(event, stdout);
 334
 335         dump_printf(": unhandled!\n");
 336         return 0;
 337 }
 338
 339 static int process_event_sample_stub(struct perf_tool *tool __maybe_unused,
 340                                      union perf_event *event __maybe_unused,
 341                                      struct perf_sample *sample __maybe_unused,
 342                                      struct evsel *evsel __maybe_unused,
 343                                      struct machine *machine __maybe_unused)
 344 {
 345         dump_printf(": unhandled!\n");
 346         return 0;
 347 }
 348
 349 static int process_event_stub(struct perf_tool *tool __maybe_unused,
 350                               union perf_event *event __maybe_unused,
 351                               struct perf_sample *sample __maybe_unused,
 352                               struct machine *machine __maybe_unused)
 353 {
 354         dump_printf(": unhandled!\n");
 355         return 0;
 356 }
 357
 358 static int process_finished_round_stub(struct perf_tool *tool __maybe_unused,
 359                                        union perf_event *event __maybe_unused,
 360                                        struct ordered_events *oe __maybe_unused)
 361 {
 362         dump_printf(": unhandled!\n");
 363         return 0;
 364 }
 365
 366 static int process_finished_round(struct perf_tool *tool,
 367                                   union perf_event *event,
 368                                   struct ordered_events *oe);
 369
 370 static int skipn(int fd, off_t n)
 371 {
 372         char buf[4096];
 373         ssize_t ret;
 374
 375         while (n > 0) {
 376                 ret = read(fd, buf, min(n, (off_t)sizeof(buf)));
 377                 if (ret <= 0)
 378                         return ret;
 379                 n -= ret;
 380         }
 381
 382         return 0;
 383 }
 384
 385 static s64 process_event_auxtrace_stub(struct perf_session *session __maybe_unused,
 386                                        union perf_event *event)
 387 {
 388         dump_printf(": unhandled!\n");
 389         if (perf_data__is_pipe(session->data))
 390                 skipn(perf_data__fd(session->data), event->auxtrace.size);
 391         return event->auxtrace.size;
 392 }
 393
 394 static int process_event_op2_stub(struct perf_session *session __maybe_unused,
 395                                   union perf_event *event __maybe_unused)
 396 {
 397         dump_printf(": unhandled!\n");
 398         return 0;
 399 }
 400
 401
 402 static
 403 int process_event_thread_map_stub(struct perf_session *session __maybe_unused,
 404                                   union perf_event *event __maybe_unused)
 405 {
 406         if (dump_trace)
 407                 perf_event__fprintf_thread_map(event, stdout);
 408
 409         dump_printf(": unhandled!\n");
 410         return 0;
 411 }
 412
 413 static
 414 int process_event_cpu_map_stub(struct perf_session *session __maybe_unused,
 415                                union perf_event *event __maybe_unused)
 416 {
 417         if (dump_trace)
 418                 perf_event__fprintf_cpu_map(event, stdout);
 419
 420         dump_printf(": unhandled!\n");
 421         return 0;
 422 }
 423
 424 static
 425 int process_event_stat_config_stub(struct perf_session *session __maybe_unused,
 426                                    union perf_event *event __maybe_unused)
 427 {
 428         if (dump_trace)
 429                 perf_event__fprintf_stat_config(event, stdout);
 430
 431         dump_printf(": unhandled!\n");
 432         return 0;
 433 }
 434
 435 static int process_stat_stub(struct perf_session *perf_session __maybe_unused,
 436                              union perf_event *event)
 437 {
 438         if (dump_trace)
 439                 perf_event__fprintf_stat(event, stdout);
 440
 441         dump_printf(": unhandled!\n");
 442         return 0;
 443 }
 444
 445 static int process_stat_round_stub(struct perf_session *perf_session __maybe_unused,
 446                                    union perf_event *event)
 447 {
 448         if (dump_trace)
 449                 perf_event__fprintf_stat_round(event, stdout);
 450
 451         dump_printf(": unhandled!\n");
 452         return 0;
 453 }
 454
 455 static int process_event_time_conv_stub(struct perf_session *perf_session __maybe_unused,
 456                                         union perf_event *event)
 457 {
 458         if (dump_trace)
 459                 perf_event__fprintf_time_conv(event, stdout);
 460
 461         dump_printf(": unhandled!\n");
 462         return 0;
 463 }
 464
 465 static int perf_session__process_compressed_event_stub(struct perf_session *session __maybe_unused,
 466                                                        union perf_event *event __maybe_unused,
 467                                                        u64 file_offset __maybe_unused)
 468 {
 469        dump_printf(": unhandled!\n");
 470        return 0;
 471 }
 472
 473 void perf_tool__fill_defaults(struct perf_tool *tool)
 474 {
 475         if (tool->sample == NULL)
 476                 tool->sample = process_event_sample_stub;
 477         if (tool->mmap == NULL)
 478                 tool->mmap = process_event_stub;
 479         if (tool->mmap2 == NULL)
 480                 tool->mmap2 = process_event_stub;
 481         if (tool->comm == NULL)
 482                 tool->comm = process_event_stub;
 483         if (tool->namespaces == NULL)
 484                 tool->namespaces = process_event_stub;
 485         if (tool->cgroup == NULL)
 486                 tool->cgroup = process_event_stub;
 487         if (tool->fork == NULL)
 488                 tool->fork = process_event_stub;
 489         if (tool->exit == NULL)
 490                 tool->exit = process_event_stub;
 491         if (tool->lost == NULL)
 492                 tool->lost = perf_event__process_lost;
 493         if (tool->lost_samples == NULL)
 494                 tool->lost_samples = perf_event__process_lost_samples;
 495         if (tool->aux == NULL)
 496                 tool->aux = perf_event__process_aux;
 497         if (tool->itrace_start == NULL)
 498                 tool->itrace_start = perf_event__process_itrace_start;
 499         if (tool->context_switch == NULL)
 500                 tool->context_switch = perf_event__process_switch;
 501         if (tool->ksymbol == NULL)
 502                 tool->ksymbol = perf_event__process_ksymbol;
 503         if (tool->bpf == NULL)
 504                 tool->bpf = perf_event__process_bpf;
 505         if (tool->text_poke == NULL)
 506                 tool->text_poke = perf_event__process_text_poke;
 507         if (tool->read == NULL)
 508                 tool->read = process_event_sample_stub;
 509         if (tool->throttle == NULL)
 510                 tool->throttle = process_event_stub;
 511         if (tool->unthrottle == NULL)
 512                 tool->unthrottle = process_event_stub;
 513         if (tool->attr == NULL)
 514                 tool->attr = process_event_synth_attr_stub;
 515         if (tool->event_update == NULL)
 516                 tool->event_update = process_event_synth_event_update_stub;
 517         if (tool->tracing_data == NULL)
 518                 tool->tracing_data = process_event_synth_tracing_data_stub;
 519         if (tool->build_id == NULL)
 520                 tool->build_id = process_event_op2_stub;
 521         if (tool->finished_round == NULL) {
 522                 if (tool->ordered_events)
 523                         tool->finished_round = process_finished_round;
 524                 else
 525                         tool->finished_round = process_finished_round_stub;
 526         }
 527         if (tool->id_index == NULL)
 528                 tool->id_index = process_event_op2_stub;
 529         if (tool->auxtrace_info == NULL)
 530                 tool->auxtrace_info = process_event_op2_stub;
 531         if (tool->auxtrace == NULL)
 532                 tool->auxtrace = process_event_auxtrace_stub;
 533         if (tool->auxtrace_error == NULL)
 534                 tool->auxtrace_error = process_event_op2_stub;
 535         if (tool->thread_map == NULL)
 536                 tool->thread_map = process_event_thread_map_stub;
 537         if (tool->cpu_map == NULL)
 538                 tool->cpu_map = process_event_cpu_map_stub;
 539         if (tool->stat_config == NULL)
 540                 tool->stat_config = process_event_stat_config_stub;
 541         if (tool->stat == NULL)
 542                 tool->stat = process_stat_stub;
 543         if (tool->stat_round == NULL)
 544                 tool->stat_round = process_stat_round_stub;
 545         if (tool->time_conv == NULL)
 546                 tool->time_conv = process_event_time_conv_stub;
 547         if (tool->feature == NULL)
 548                 tool->feature = process_event_op2_stub;
 549         if (tool->compressed == NULL)
 550                 tool->compressed = perf_session__process_compressed_event;
 551 }
 552
 553 static void swap_sample_id_all(union perf_event *event, void *data)
 554 {
 555         void *end = (void *) event + event->header.size;
 556         int size = end - data;
 557
 558         BUG_ON(size % sizeof(u64));
 559         mem_bswap_64(data, size);
 560 }
 561
 562 static void perf_event__all64_swap(union perf_event *event,
 563                                    bool sample_id_all __maybe_unused)
 564 {
 565         struct perf_event_header *hdr = &event->header;
 566         mem_bswap_64(hdr + 1, event->header.size - sizeof(*hdr));
 567 }
 568
 569 static void perf_event__comm_swap(union perf_event *event, bool sample_id_all)
 570 {
 571         event->comm.pid = bswap_32(event->comm.pid);
 572         event->comm.tid = bswap_32(event->comm.tid);
 573
 574         if (sample_id_all) {
 575                 void *data = &event->comm.comm;
 576
 577                 data += PERF_ALIGN(strlen(data) + 1, sizeof(u64));
 578                 swap_sample_id_all(event, data);
 579         }
 580 }
 581
 582 static void perf_event__mmap_swap(union perf_event *event,
 583                                   bool sample_id_all)
 584 {
 585         event->mmap.pid   = bswap_32(event->mmap.pid);
 586         event->mmap.tid   = bswap_32(event->mmap.tid);
 587         event->mmap.start = bswap_64(event->mmap.start);
 588         event->mmap.len   = bswap_64(event->mmap.len);
 589         event->mmap.pgoff = bswap_64(event->mmap.pgoff);
 590
 591         if (sample_id_all) {
 592                 void *data = &event->mmap.filename;
 593
 594                 data += PERF_ALIGN(strlen(data) + 1, sizeof(u64));
 595                 swap_sample_id_all(event, data);
 596         }
 597 }
 598
 599 static void perf_event__mmap2_swap(union perf_event *event,
 600                                   bool sample_id_all)
 601 {
 602         event->mmap2.pid   = bswap_32(event->mmap2.pid);
 603         event->mmap2.tid   = bswap_32(event->mmap2.tid);
 604         event->mmap2.start = bswap_64(event->mmap2.start);
 605         event->mmap2.len   = bswap_64(event->mmap2.len);
 606         event->mmap2.pgoff = bswap_64(event->mmap2.pgoff);
 607
 608         if (!(event->header.misc & PERF_RECORD_MISC_MMAP_BUILD_ID)) {
 609                 event->mmap2.maj   = bswap_32(event->mmap2.maj);
 610                 event->mmap2.min   = bswap_32(event->mmap2.min);
 611                 event->mmap2.ino   = bswap_64(event->mmap2.ino);
 612                 event->mmap2.ino_generation = bswap_64(event->mmap2.ino_generation);
 613         }
 614
 615         if (sample_id_all) {
 616                 void *data = &event->mmap2.filename;
 617
 618                 data += PERF_ALIGN(strlen(data) + 1, sizeof(u64));
 619                 swap_sample_id_all(event, data);
 620         }
 621 }
 622 static void perf_event__task_swap(union perf_event *event, bool sample_id_all)
 623 {
 624         event->fork.pid  = bswap_32(event->fork.pid);
 625         event->fork.tid  = bswap_32(event->fork.tid);
 626         event->fork.ppid = bswap_32(event->fork.ppid);
 627         event->fork.ptid = bswap_32(event->fork.ptid);
 628         event->fork.time = bswap_64(event->fork.time);
 629
 630         if (sample_id_all)
 631                 swap_sample_id_all(event, &event->fork + 1);
 632 }
 633
 634 static void perf_event__read_swap(union perf_event *event, bool sample_id_all)
 635 {
 636         event->read.pid          = bswap_32(event->read.pid);
 637         event->read.tid          = bswap_32(event->read.tid);
 638         event->read.value        = bswap_64(event->read.value);
 639         event->read.time_enabled = bswap_64(event->read.time_enabled);
 640         event->read.time_running = bswap_64(event->read.time_running);
 641         event->read.id           = bswap_64(event->read.id);
 642
 643         if (sample_id_all)
 644                 swap_sample_id_all(event, &event->read + 1);
 645 }
 646
 647 static void perf_event__aux_swap(union perf_event *event, bool sample_id_all)
 648 {
 649         event->aux.aux_offset = bswap_64(event->aux.aux_offset);
 650         event->aux.aux_size   = bswap_64(event->aux.aux_size);
 651         event->aux.flags      = bswap_64(event->aux.flags);
 652
 653         if (sample_id_all)
 654                 swap_sample_id_all(event, &event->aux + 1);
 655 }
 656
 657 static void perf_event__itrace_start_swap(union perf_event *event,
 658                                           bool sample_id_all)
 659 {
 660         event->itrace_start.pid  = bswap_32(event->itrace_start.pid);
 661         event->itrace_start.tid  = bswap_32(event->itrace_start.tid);
 662
 663         if (sample_id_all)
 664                 swap_sample_id_all(event, &event->itrace_start + 1);
 665 }
 666
 667 static void perf_event__switch_swap(union perf_event *event, bool sample_id_all)
 668 {
 669         if (event->header.type == PERF_RECORD_SWITCH_CPU_WIDE) {
 670                 event->context_switch.next_prev_pid =
 671                                 bswap_32(event->context_switch.next_prev_pid);
 672                 event->context_switch.next_prev_tid =
 673                                 bswap_32(event->context_switch.next_prev_tid);
 674         }
 675
 676         if (sample_id_all)
 677                 swap_sample_id_all(event, &event->context_switch + 1);
 678 }
 679
 680 static void perf_event__text_poke_swap(union perf_event *event, bool sample_id_all)
 681 {
 682         event->text_poke.addr    = bswap_64(event->text_poke.addr);
 683         event->text_poke.old_len = bswap_16(event->text_poke.old_len);
 684         event->text_poke.new_len = bswap_16(event->text_poke.new_len);
 685
 686         if (sample_id_all) {
 687                 size_t len = sizeof(event->text_poke.old_len) +
 688                              sizeof(event->text_poke.new_len) +
 689                              event->text_poke.old_len +
 690                              event->text_poke.new_len;
 691                 void *data = &event->text_poke.old_len;
 692
 693                 data += PERF_ALIGN(len, sizeof(u64));
 694                 swap_sample_id_all(event, data);
 695         }
 696 }
 697
 698 static void perf_event__throttle_swap(union perf_event *event,
 699                                       bool sample_id_all)
 700 {
 701         event->throttle.time      = bswap_64(event->throttle.time);
 702         event->throttle.id        = bswap_64(event->throttle.id);
 703         event->throttle.stream_id = bswap_64(event->throttle.stream_id);
 704
 705         if (sample_id_all)
 706                 swap_sample_id_all(event, &event->throttle + 1);
 707 }
 708
 709 static void perf_event__namespaces_swap(union perf_event *event,
 710                                         bool sample_id_all)
 711 {
 712         u64 i;
 713
 714         event->namespaces.pid           = bswap_32(event->namespaces.pid);
 715         event->namespaces.tid           = bswap_32(event->namespaces.tid);
 716         event->namespaces.nr_namespaces = bswap_64(event->namespaces.nr_namespaces);
 717
 718         for (i = 0; i < event->namespaces.nr_namespaces; i++) {
 719                 struct perf_ns_link_info *ns = &event->namespaces.link_info[i];
 720
 721                 ns->dev = bswap_64(ns->dev);
 722                 ns->ino = bswap_64(ns->ino);
 723         }
 724
 725         if (sample_id_all)
 726                 swap_sample_id_all(event, &event->namespaces.link_info[i]);
 727 }
 728
 729 static void perf_event__cgroup_swap(union perf_event *event, bool sample_id_all)
 730 {
 731         event->cgroup.id = bswap_64(event->cgroup.id);
 732
 733         if (sample_id_all) {
 734                 void *data = &event->cgroup.path;
 735
 736                 data += PERF_ALIGN(strlen(data) + 1, sizeof(u64));
 737                 swap_sample_id_all(event, data);
 738         }
 739 }
 740
 741 static u8 revbyte(u8 b)
 742 {
 743         int rev = (b >> 4) | ((b & 0xf) << 4);
 744         rev = ((rev & 0xcc) >> 2) | ((rev & 0x33) << 2);
 745         rev = ((rev & 0xaa) >> 1) | ((rev & 0x55) << 1);
 746         return (u8) rev;
 747 }
 748
 749 /*
 750  * XXX this is hack in attempt to carry flags bitfield
 751  * through endian village. ABI says:
 752  *
 753  * Bit-fields are allocated from right to left (least to most significant)
 754  * on little-endian implementations and from left to right (most to least
 755  * significant) on big-endian implementations.
 756  *
 757  * The above seems to be byte specific, so we need to reverse each
 758  * byte of the bitfield. 'Internet' also says this might be implementation
 759  * specific and we probably need proper fix and carry perf_event_attr
 760  * bitfield flags in separate data file FEAT_ section. Thought this seems
 761  * to work for now.
 762  */
 763 static void swap_bitfield(u8 *p, unsigned len)
 764 {
 765         unsigned i;
 766
 767         for (i = 0; i < len; i++) {
 768                 *p = revbyte(*p);
 769                 p++;
 770         }
 771 }
 772
 773 /* exported for swapping attributes in file header */
 774 void perf_event__attr_swap(struct perf_event_attr *attr)
 775 {
 776         attr->type              = bswap_32(attr->type);
 777         attr->size              = bswap_32(attr->size);
 778
 779 #define bswap_safe(f, n)                                        \
 780         (attr->size > (offsetof(struct perf_event_attr, f) +    \
 781                        sizeof(attr->f) * (n)))
 782 #define bswap_field(f, sz)                      \
 783 do {                                            \
 784         if (bswap_safe(f, 0))                   \
 785                 attr->f = bswap_##sz(attr->f);  \
 786 } while(0)
 787 #define bswap_field_16(f) bswap_field(f, 16)
 788 #define bswap_field_32(f) bswap_field(f, 32)
 789 #define bswap_field_64(f) bswap_field(f, 64)
 790
 791         bswap_field_64(config);
 792         bswap_field_64(sample_period);
 793         bswap_field_64(sample_type);
 794         bswap_field_64(read_format);
 795         bswap_field_32(wakeup_events);
 796         bswap_field_32(bp_type);
 797         bswap_field_64(bp_addr);
 798         bswap_field_64(bp_len);
 799         bswap_field_64(branch_sample_type);
 800         bswap_field_64(sample_regs_user);
 801         bswap_field_32(sample_stack_user);
 802         bswap_field_32(aux_watermark);
 803         bswap_field_16(sample_max_stack);
 804         bswap_field_32(aux_sample_size);
 805
 806         /*
 807          * After read_format are bitfields. Check read_format because
 808          * we are unable to use offsetof on bitfield.
 809          */
 810         if (bswap_safe(read_format, 1))
 811                 swap_bitfield((u8 *) (&attr->read_format + 1),
 812                               sizeof(u64));
 813 #undef bswap_field_64
 814 #undef bswap_field_32
 815 #undef bswap_field
 816 #undef bswap_safe
 817 }
 818
 819 static void perf_event__hdr_attr_swap(union perf_event *event,
 820                                       bool sample_id_all __maybe_unused)
 821 {
 822         size_t size;
 823
 824         perf_event__attr_swap(&event->attr.attr);
 825
 826         size = event->header.size;
 827         size -= (void *)&event->attr.id - (void *)event;
 828         mem_bswap_64(event->attr.id, size);
 829 }
 830
 831 static void perf_event__event_update_swap(union perf_event *event,
 832                                           bool sample_id_all __maybe_unused)
 833 {
 834         event->event_update.type = bswap_64(event->event_update.type);
 835         event->event_update.id   = bswap_64(event->event_update.id);
 836 }
 837
 838 static void perf_event__event_type_swap(union perf_event *event,
 839                                         bool sample_id_all __maybe_unused)
 840 {
 841         event->event_type.event_type.event_id =
 842                 bswap_64(event->event_type.event_type.event_id);
 843 }
 844
 845 static void perf_event__tracing_data_swap(union perf_event *event,
 846                                           bool sample_id_all __maybe_unused)
 847 {
 848         event->tracing_data.size = bswap_32(event->tracing_data.size);
 849 }
 850
 851 static void perf_event__auxtrace_info_swap(union perf_event *event,
 852                                            bool sample_id_all __maybe_unused)
 853 {
 854         size_t size;
 855
 856         event->auxtrace_info.type = bswap_32(event->auxtrace_info.type);
 857
 858         size = event->header.size;
 859         size -= (void *)&event->auxtrace_info.priv - (void *)event;
 860         mem_bswap_64(event->auxtrace_info.priv, size);
 861 }
 862
 863 static void perf_event__auxtrace_swap(union perf_event *event,
 864                                       bool sample_id_all __maybe_unused)
 865 {
 866         event->auxtrace.size      = bswap_64(event->auxtrace.size);
 867         event->auxtrace.offset    = bswap_64(event->auxtrace.offset);
 868         event->auxtrace.reference = bswap_64(event->auxtrace.reference);
 869         event->auxtrace.idx       = bswap_32(event->auxtrace.idx);
 870         event->auxtrace.tid       = bswap_32(event->auxtrace.tid);
 871         event->auxtrace.cpu       = bswap_32(event->auxtrace.cpu);
 872 }
 873
 874 static void perf_event__auxtrace_error_swap(union perf_event *event,
 875                                             bool sample_id_all __maybe_unused)
 876 {
 877         event->auxtrace_error.type = bswap_32(event->auxtrace_error.type);
 878         event->auxtrace_error.code = bswap_32(event->auxtrace_error.code);
 879         event->auxtrace_error.cpu  = bswap_32(event->auxtrace_error.cpu);
 880         event->auxtrace_error.pid  = bswap_32(event->auxtrace_error.pid);
 881         event->auxtrace_error.tid  = bswap_32(event->auxtrace_error.tid);
 882         event->auxtrace_error.fmt  = bswap_32(event->auxtrace_error.fmt);
 883         event->auxtrace_error.ip   = bswap_64(event->auxtrace_error.ip);
 884         if (event->auxtrace_error.fmt)
 885                 event->auxtrace_error.time = bswap_64(event->auxtrace_error.time);
 886 }
 887
 888 static void perf_event__thread_map_swap(union perf_event *event,
 889                                         bool sample_id_all __maybe_unused)
 890 {
 891         unsigned i;
 892
 893         event->thread_map.nr = bswap_64(event->thread_map.nr);
 894
 895         for (i = 0; i < event->thread_map.nr; i++)
 896                 event->thread_map.entries[i].pid = bswap_64(event->thread_map.entries[i].pid);
 897 }
 898
 899 static void perf_event__cpu_map_swap(union perf_event *event,
 900                                      bool sample_id_all __maybe_unused)
 901 {
 902         struct perf_record_cpu_map_data *data = &event->cpu_map.data;
 903         struct cpu_map_entries *cpus;
 904         struct perf_record_record_cpu_map *mask;
 905         unsigned i;
 906
 907         data->type = bswap_16(data->type);
 908
 909         switch (data->type) {
 910         case PERF_CPU_MAP__CPUS:
 911                 cpus = (struct cpu_map_entries *)data->data;
 912
 913                 cpus->nr = bswap_16(cpus->nr);
 914
 915                 for (i = 0; i < cpus->nr; i++)
 916                         cpus->cpu[i] = bswap_16(cpus->cpu[i]);
 917                 break;
 918         case PERF_CPU_MAP__MASK:
 919                 mask = (struct perf_record_record_cpu_map *)data->data;
 920
 921                 mask->nr = bswap_16(mask->nr);
 922                 mask->long_size = bswap_16(mask->long_size);
 923
 924                 switch (mask->long_size) {
 925                 case 4: mem_bswap_32(&mask->mask, mask->nr); break;
 926                 case 8: mem_bswap_64(&mask->mask, mask->nr); break;
 927                 default:
 928                         pr_err("cpu_map swap: unsupported long size\n");
 929                 }
 930         default:
 931                 break;
 932         }
 933 }
 934
 935 static void perf_event__stat_config_swap(union perf_event *event,
 936                                          bool sample_id_all __maybe_unused)
 937 {
 938         u64 size;
 939
 940         size  = bswap_64(event->stat_config.nr) * sizeof(event->stat_config.data[0]);
 941         size += 1; /* nr item itself */
 942         mem_bswap_64(&event->stat_config.nr, size);
 943 }
 944
 945 static void perf_event__stat_swap(union perf_event *event,
 946                                   bool sample_id_all __maybe_unused)
 947 {
 948         event->stat.id     = bswap_64(event->stat.id);
 949         event->stat.thread = bswap_32(event->stat.thread);
 950         event->stat.cpu    = bswap_32(event->stat.cpu);
 951         event->stat.val    = bswap_64(event->stat.val);
 952         event->stat.ena    = bswap_64(event->stat.ena);
 953         event->stat.run    = bswap_64(event->stat.run);
 954 }
 955
 956 static void perf_event__stat_round_swap(union perf_event *event,
 957                                         bool sample_id_all __maybe_unused)
 958 {
 959         event->stat_round.type = bswap_64(event->stat_round.type);
 960         event->stat_round.time = bswap_64(event->stat_round.time);
 961 }
 962
 963 static void perf_event__time_conv_swap(union perf_event *event,
 964                                        bool sample_id_all __maybe_unused)
 965 {
 966         event->time_conv.time_shift = bswap_64(event->time_conv.time_shift);
 967         event->time_conv.time_mult  = bswap_64(event->time_conv.time_mult);
 968         event->time_conv.time_zero  = bswap_64(event->time_conv.time_zero);
 969
 970         if (event_contains(event->time_conv, time_cycles)) {
 971                 event->time_conv.time_cycles = bswap_64(event->time_conv.time_cycles);
 972                 event->time_conv.time_mask = bswap_64(event->time_conv.time_mask);
 973         }
 974 }
 975
 976 typedef void (*perf_event__swap_op)(union perf_event *event,
 977                                     bool sample_id_all);
 978
 979 static perf_event__swap_op perf_event__swap_ops[] = {
 980         [PERF_RECORD_MMAP]                = perf_event__mmap_swap,
 981         [PERF_RECORD_MMAP2]               = perf_event__mmap2_swap,
 982         [PERF_RECORD_COMM]                = perf_event__comm_swap,
 983         [PERF_RECORD_FORK]                = perf_event__task_swap,
 984         [PERF_RECORD_EXIT]                = perf_event__task_swap,
 985         [PERF_RECORD_LOST]                = perf_event__all64_swap,
 986         [PERF_RECORD_READ]                = perf_event__read_swap,
 987         [PERF_RECORD_THROTTLE]            = perf_event__throttle_swap,
 988         [PERF_RECORD_UNTHROTTLE]          = perf_event__throttle_swap,
 989         [PERF_RECORD_SAMPLE]              = perf_event__all64_swap,
 990         [PERF_RECORD_AUX]                 = perf_event__aux_swap,
 991         [PERF_RECORD_ITRACE_START]        = perf_event__itrace_start_swap,
 992         [PERF_RECORD_LOST_SAMPLES]        = perf_event__all64_swap,
 993         [PERF_RECORD_SWITCH]              = perf_event__switch_swap,
 994         [PERF_RECORD_SWITCH_CPU_WIDE]     = perf_event__switch_swap,
 995         [PERF_RECORD_NAMESPACES]          = perf_event__namespaces_swap,
 996         [PERF_RECORD_CGROUP]              = perf_event__cgroup_swap,
 997         [PERF_RECORD_TEXT_POKE]           = perf_event__text_poke_swap,
 998         [PERF_RECORD_HEADER_ATTR]         = perf_event__hdr_attr_swap,
 999         [PERF_RECORD_HEADER_EVENT_TYPE]   = perf_event__event_type_swap,
1000         [PERF_RECORD_HEADER_TRACING_DATA] = perf_event__tracing_data_swap,
1001         [PERF_RECORD_HEADER_BUILD_ID]     = NULL,
1002         [PERF_RECORD_ID_INDEX]            = perf_event__all64_swap,
1003         [PERF_RECORD_AUXTRACE_INFO]       = perf_event__auxtrace_info_swap,
1004         [PERF_RECORD_AUXTRACE]            = perf_event__auxtrace_swap,
1005         [PERF_RECORD_AUXTRACE_ERROR]      = perf_event__auxtrace_error_swap,
1006         [PERF_RECORD_THREAD_MAP]          = perf_event__thread_map_swap,
1007         [PERF_RECORD_CPU_MAP]             = perf_event__cpu_map_swap,
1008         [PERF_RECORD_STAT_CONFIG]         = perf_event__stat_config_swap,
1009         [PERF_RECORD_STAT]                = perf_event__stat_swap,
1010         [PERF_RECORD_STAT_ROUND]          = perf_event__stat_round_swap,
1011         [PERF_RECORD_EVENT_UPDATE]        = perf_event__event_update_swap,
1012         [PERF_RECORD_TIME_CONV]           = perf_event__time_conv_swap,
1013         [PERF_RECORD_HEADER_MAX]          = NULL,
1014 };
1015
1016 /*
1017  * When perf record finishes a pass on every buffers, it records this pseudo
1018  * event.
1019  * We record the max timestamp t found in the pass n.
1020  * Assuming these timestamps are monotonic across cpus, we know that if
1021  * a buffer still has events with timestamps below t, they will be all
1022  * available and then read in the pass n + 1.
1023  * Hence when we start to read the pass n + 2, we can safely flush every
1024  * events with timestamps below t.
1025  *
1026  *    ============ PASS n =================
1027  *       CPU 0         |   CPU 1
1028  *                     |
1029  *    cnt1 timestamps  |   cnt2 timestamps
1030  *          1          |         2
1031  *          2          |         3
1032  *          -          |         4  <--- max recorded
1033  *
1034  *    ============ PASS n + 1 ==============
1035  *       CPU 0         |   CPU 1
1036  *                     |
1037  *    cnt1 timestamps  |   cnt2 timestamps
1038  *          3          |         5
1039  *          4          |         6
1040  *          5          |         7 <---- max recorded
1041  *
1042  *      Flush every events below timestamp 4
1043  *
1044  *    ============ PASS n + 2 ==============
1045  *       CPU 0         |   CPU 1
1046  *                     |
1047  *    cnt1 timestamps  |   cnt2 timestamps
1048  *          6          |         8
1049  *          7          |         9
1050  *          -          |         10
1051  *
1052  *      Flush every events below timestamp 7
1053  *      etc...
1054  */
1055 static int process_finished_round(struct perf_tool *tool __maybe_unused,
1056                                   union perf_event *event __maybe_unused,
1057                                   struct ordered_events *oe)
1058 {
1059         if (dump_trace)
1060                 fprintf(stdout, "\n");
1061         return ordered_events__flush(oe, OE_FLUSH__ROUND);
1062 }
1063
1064 int perf_session__queue_event(struct perf_session *s, union perf_event *event,
1065                               u64 timestamp, u64 file_offset)
1066 {
1067         return ordered_events__queue(&s->ordered_events, event, timestamp, file_offset);
1068 }
1069
1070 static void callchain__lbr_callstack_printf(struct perf_sample *sample)
1071 {
1072         struct ip_callchain *callchain = sample->callchain;
1073         struct branch_stack *lbr_stack = sample->branch_stack;
1074         struct branch_entry *entries = perf_sample__branch_entries(sample);
1075         u64 kernel_callchain_nr = callchain->nr;
1076         unsigned int i;
1077
1078         for (i = 0; i < kernel_callchain_nr; i++) {
1079                 if (callchain->ips[i] == PERF_CONTEXT_USER)
1080                         break;
1081         }
1082
1083         if ((i != kernel_callchain_nr) && lbr_stack->nr) {
1084                 u64 total_nr;
1085                 /*
1086                  * LBR callstack can only get user call chain,
1087                  * i is kernel call chain number,
1088                  * 1 is PERF_CONTEXT_USER.
1089                  *
1090                  * The user call chain is stored in LBR registers.
1091                  * LBR are pair registers. The caller is stored
1092                  * in "from" register, while the callee is stored
1093                  * in "to" register.
1094                  * For example, there is a call stack
1095                  * "A"->"B"->"C"->"D".
1096                  * The LBR registers will be recorded like
1097                  * "C"->"D", "B"->"C", "A"->"B".
1098                  * So only the first "to" register and all "from"
1099                  * registers are needed to construct the whole stack.
1100                  */
1101                 total_nr = i + 1 + lbr_stack->nr + 1;
1102                 kernel_callchain_nr = i + 1;
1103
1104                 printf("... LBR call chain: nr:%" PRIu64 "\n", total_nr);
1105
1106                 for (i = 0; i < kernel_callchain_nr; i++)
1107                         printf("..... %2d: %016" PRIx64 "\n",
1108                                i, callchain->ips[i]);
1109
1110                 printf("..... %2d: %016" PRIx64 "\n",
1111                        (int)(kernel_callchain_nr), entries[0].to);
1112                 for (i = 0; i < lbr_stack->nr; i++)
1113                         printf("..... %2d: %016" PRIx64 "\n",
1114                                (int)(i + kernel_callchain_nr + 1), entries[i].from);
1115         }
1116 }
1117
1118 static void callchain__printf(struct evsel *evsel,
1119                               struct perf_sample *sample)
1120 {
1121         unsigned int i;
1122         struct ip_callchain *callchain = sample->callchain;
1123
1124         if (evsel__has_branch_callstack(evsel))
1125                 callchain__lbr_callstack_printf(sample);
1126
1127         printf("... FP chain: nr:%" PRIu64 "\n", callchain->nr);
1128
1129         for (i = 0; i < callchain->nr; i++)
1130                 printf("..... %2d: %016" PRIx64 "\n",
1131                        i, callchain->ips[i]);
1132 }
1133
1134 static void branch_stack__printf(struct perf_sample *sample, bool callstack)
1135 {
1136         struct branch_entry *entries = perf_sample__branch_entries(sample);
1137         uint64_t i;
1138
1139         printf("%s: nr:%" PRIu64 "\n",
1140                 !callstack ? "... branch stack" : "... branch callstack",
1141                 sample->branch_stack->nr);
1142
1143         for (i = 0; i < sample->branch_stack->nr; i++) {
1144                 struct branch_entry *e = &entries[i];
1145
1146                 if (!callstack) {
1147                         printf("..... %2"PRIu64": %016" PRIx64 " -> %016" PRIx64 " %hu cycles %s%s%s%s %x\n",
1148                                 i, e->from, e->to,
1149                                 (unsigned short)e->flags.cycles,
1150                                 e->flags.mispred ? "M" : " ",
1151                                 e->flags.predicted ? "P" : " ",
1152                                 e->flags.abort ? "A" : " ",
1153                                 e->flags.in_tx ? "T" : " ",
1154                                 (unsigned)e->flags.reserved);
1155                 } else {
1156                         printf("..... %2"PRIu64": %016" PRIx64 "\n",
1157                                 i, i > 0 ? e->from : e->to);
1158                 }
1159         }
1160 }
1161
1162 static void regs_dump__printf(u64 mask, u64 *regs)
1163 {
1164         unsigned rid, i = 0;
1165
1166         for_each_set_bit(rid, (unsigned long *) &mask, sizeof(mask) * 8) {
1167                 u64 val = regs[i++];
1168
1169                 printf(".... %-5s 0x%016" PRIx64 "\n",
1170                        perf_reg_name(rid), val);
1171         }
1172 }
1173
1174 static const char *regs_abi[] = {
1175         [PERF_SAMPLE_REGS_ABI_NONE] = "none",
1176         [PERF_SAMPLE_REGS_ABI_32] = "32-bit",
1177         [PERF_SAMPLE_REGS_ABI_64] = "64-bit",
1178 };
1179
1180 static inline const char *regs_dump_abi(struct regs_dump *d)
1181 {
1182         if (d->abi > PERF_SAMPLE_REGS_ABI_64)
1183                 return "unknown";
1184
1185         return regs_abi[d->abi];
1186 }
1187
1188 static void regs__printf(const char *type, struct regs_dump *regs)
1189 {
1190         u64 mask = regs->mask;
1191
1192         printf("... %s regs: mask 0x%" PRIx64 " ABI %s\n",
1193                type,
1194                mask,
1195                regs_dump_abi(regs));
1196
1197         regs_dump__printf(mask, regs->regs);
1198 }
1199
1200 static void regs_user__printf(struct perf_sample *sample)
1201 {
1202         struct regs_dump *user_regs = &sample->user_regs;
1203
1204         if (user_regs->regs)
1205                 regs__printf("user", user_regs);
1206 }
1207
1208 static void regs_intr__printf(struct perf_sample *sample)
1209 {
1210         struct regs_dump *intr_regs = &sample->intr_regs;
1211
1212         if (intr_regs->regs)
1213                 regs__printf("intr", intr_regs);
1214 }
1215
1216 static void stack_user__printf(struct stack_dump *dump)
1217 {
1218         printf("... ustack: size %" PRIu64 ", offset 0x%x\n",
1219                dump->size, dump->offset);
1220 }
1221
1222 static void evlist__print_tstamp(struct evlist *evlist, union perf_event *event, struct perf_sample *sample)
1223 {
1224         u64 sample_type = __evlist__combined_sample_type(evlist);
1225
1226         if (event->header.type != PERF_RECORD_SAMPLE &&
1227             !evlist__sample_id_all(evlist)) {
1228                 fputs("-1 -1 ", stdout);
1229                 return;
1230         }
1231
1232         if ((sample_type & PERF_SAMPLE_CPU))
1233                 printf("%u ", sample->cpu);
1234
1235         if (sample_type & PERF_SAMPLE_TIME)
1236                 printf("%" PRIu64 " ", sample->time);
1237 }
1238
1239 static void sample_read__printf(struct perf_sample *sample, u64 read_format)
1240 {
1241         printf("... sample_read:\n");
1242
1243         if (read_format & PERF_FORMAT_TOTAL_TIME_ENABLED)
1244                 printf("...... time enabled %016" PRIx64 "\n",
1245                        sample->read.time_enabled);
1246
1247         if (read_format & PERF_FORMAT_TOTAL_TIME_RUNNING)
1248                 printf("...... time running %016" PRIx64 "\n",
1249                        sample->read.time_running);
1250
1251         if (read_format & PERF_FORMAT_GROUP) {
1252                 u64 i;
1253
1254                 printf(".... group nr %" PRIu64 "\n", sample->read.group.nr);
1255
1256                 for (i = 0; i < sample->read.group.nr; i++) {
1257                         struct sample_read_value *value;
1258
1259                         value = &sample->read.group.values[i];
1260                         printf("..... id %016" PRIx64
1261                                ", value %016" PRIx64 "\n",
1262                                value->id, value->value);
1263                 }
1264         } else
1265                 printf("..... id %016" PRIx64 ", value %016" PRIx64 "\n",
1266                         sample->read.one.id, sample->read.one.value);
1267 }
1268
1269 static void dump_event(struct evlist *evlist, union perf_event *event,
1270                        u64 file_offset, struct perf_sample *sample)
1271 {
1272         if (!dump_trace)
1273                 return;
1274
1275         printf("\n%#" PRIx64 " [%#x]: event: %d\n",
1276                file_offset, event->header.size, event->header.type);
1277
1278         trace_event(event);
1279         if (event->header.type == PERF_RECORD_SAMPLE && evlist->trace_event_sample_raw)
1280                 evlist->trace_event_sample_raw(evlist, event, sample);
1281
1282         if (sample)
1283                 evlist__print_tstamp(evlist, event, sample);
1284
1285         printf("%#" PRIx64 " [%#x]: PERF_RECORD_%s", file_offset,
1286                event->header.size, perf_event__name(event->header.type));
1287 }
1288
1289 char *get_page_size_name(u64 size, char *str)
1290 {
1291         if (!size || !unit_number__scnprintf(str, PAGE_SIZE_NAME_LEN, size))
1292                 snprintf(str, PAGE_SIZE_NAME_LEN, "%s", "N/A");
1293
1294         return str;
1295 }
1296
1297 static void dump_sample(struct evsel *evsel, union perf_event *event,
1298                         struct perf_sample *sample)
1299 {
1300         u64 sample_type;
1301         char str[PAGE_SIZE_NAME_LEN];
1302
1303         if (!dump_trace)
1304                 return;
1305
1306         printf("(IP, 0x%x): %d/%d: %#" PRIx64 " period: %" PRIu64 " addr: %#" PRIx64 "\n",
1307                event->header.misc, sample->pid, sample->tid, sample->ip,
1308                sample->period, sample->addr);
1309
1310         sample_type = evsel->core.attr.sample_type;
1311
1312         if (evsel__has_callchain(evsel))
1313                 callchain__printf(evsel, sample);
1314
1315         if (evsel__has_br_stack(evsel))
1316                 branch_stack__printf(sample, evsel__has_branch_callstack(evsel));
1317
1318         if (sample_type & PERF_SAMPLE_REGS_USER)
1319                 regs_user__printf(sample);
1320
1321         if (sample_type & PERF_SAMPLE_REGS_INTR)
1322                 regs_intr__printf(sample);
1323
1324         if (sample_type & PERF_SAMPLE_STACK_USER)
1325                 stack_user__printf(&sample->user_stack);
1326
1327         if (sample_type & PERF_SAMPLE_WEIGHT_TYPE) {
1328                 printf("... weight: %" PRIu64 "", sample->weight);
1329                         if (sample_type & PERF_SAMPLE_WEIGHT_STRUCT) {
1330                                 printf(",0x%"PRIx16"", sample->ins_lat);
1331                                 printf(",0x%"PRIx16"", sample->p_stage_cyc);
1332                         }
1333                 printf("\n");
1334         }
1335
1336         if (sample_type & PERF_SAMPLE_DATA_SRC)
1337                 printf(" . data_src: 0x%"PRIx64"\n", sample->data_src);
1338
1339         if (sample_type & PERF_SAMPLE_PHYS_ADDR)
1340                 printf(" .. phys_addr: 0x%"PRIx64"\n", sample->phys_addr);
1341
1342         if (sample_type & PERF_SAMPLE_DATA_PAGE_SIZE)
1343                 printf(" .. data page size: %s\n", get_page_size_name(sample->data_page_size, str));
1344
1345         if (sample_type & PERF_SAMPLE_CODE_PAGE_SIZE)
1346                 printf(" .. code page size: %s\n", get_page_size_name(sample->code_page_size, str));
1347
1348         if (sample_type & PERF_SAMPLE_TRANSACTION)
1349                 printf("... transaction: %" PRIx64 "\n", sample->transaction);
1350
1351         if (sample_type & PERF_SAMPLE_READ)
1352                 sample_read__printf(sample, evsel->core.attr.read_format);
1353 }
1354
1355 static void dump_read(struct evsel *evsel, union perf_event *event)
1356 {
1357         struct perf_record_read *read_event = &event->read;
1358         u64 read_format;
1359
1360         if (!dump_trace)
1361                 return;
1362
1363         printf(": %d %d %s %" PRI_lu64 "\n", event->read.pid, event->read.tid,
1364                evsel__name(evsel), event->read.value);
1365
1366         if (!evsel)
1367                 return;
1368
1369         read_format = evsel->core.attr.read_format;
1370
1371         if (read_format & PERF_FORMAT_TOTAL_TIME_ENABLED)
1372                 printf("... time enabled : %" PRI_lu64 "\n", read_event->time_enabled);
1373
1374         if (read_format & PERF_FORMAT_TOTAL_TIME_RUNNING)
1375                 printf("... time running : %" PRI_lu64 "\n", read_event->time_running);
1376
1377         if (read_format & PERF_FORMAT_ID)
1378                 printf("... id           : %" PRI_lu64 "\n", read_event->id);
1379 }
1380
1381 static struct machine *machines__find_for_cpumode(struct machines *machines,
1382                                                union perf_event *event,
1383                                                struct perf_sample *sample)
1384 {
1385         if (perf_guest &&
1386             ((sample->cpumode == PERF_RECORD_MISC_GUEST_KERNEL) ||
1387              (sample->cpumode == PERF_RECORD_MISC_GUEST_USER))) {
1388                 u32 pid;
1389
1390                 if (event->header.type == PERF_RECORD_MMAP
1391                     || event->header.type == PERF_RECORD_MMAP2)
1392                         pid = event->mmap.pid;
1393                 else
1394                         pid = sample->pid;
1395
1396                 return machines__find_guest(machines, pid);
1397         }
1398
1399         return &machines->host;
1400 }
1401
1402 static int deliver_sample_value(struct evlist *evlist,
1403                                 struct perf_tool *tool,
1404                                 union perf_event *event,
1405                                 struct perf_sample *sample,
1406                                 struct sample_read_value *v,
1407                                 struct machine *machine)
1408 {
1409         struct perf_sample_id *sid = evlist__id2sid(evlist, v->id);
1410         struct evsel *evsel;
1411
1412         if (sid) {
1413                 sample->id     = v->id;
1414                 sample->period = v->value - sid->period;
1415                 sid->period    = v->value;
1416         }
1417
1418         if (!sid || sid->evsel == NULL) {
1419                 ++evlist->stats.nr_unknown_id;
1420                 return 0;
1421         }
1422
1423         /*
1424          * There's no reason to deliver sample
1425          * for zero period, bail out.
1426          */
1427         if (!sample->period)
1428                 return 0;
1429
1430         evsel = container_of(sid->evsel, struct evsel, core);
1431         return tool->sample(tool, event, sample, evsel, machine);
1432 }
1433
1434 static int deliver_sample_group(struct evlist *evlist,
1435                                 struct perf_tool *tool,
1436                                 union  perf_event *event,
1437                                 struct perf_sample *sample,
1438                                 struct machine *machine)
1439 {
1440         int ret = -EINVAL;
1441         u64 i;
1442
1443         for (i = 0; i < sample->read.group.nr; i++) {
1444                 ret = deliver_sample_value(evlist, tool, event, sample,
1445                                            &sample->read.group.values[i],
1446                                            machine);
1447                 if (ret)
1448                         break;
1449         }
1450
1451         return ret;
1452 }
1453
1454 static int evlist__deliver_sample(struct evlist *evlist, struct perf_tool *tool,
1455                                   union  perf_event *event, struct perf_sample *sample,
1456                                   struct evsel *evsel, struct machine *machine)
1457 {
1458         /* We know evsel != NULL. */
1459         u64 sample_type = evsel->core.attr.sample_type;
1460         u64 read_format = evsel->core.attr.read_format;
1461
1462         /* Standard sample delivery. */
1463         if (!(sample_type & PERF_SAMPLE_READ))
1464                 return tool->sample(tool, event, sample, evsel, machine);
1465
1466         /* For PERF_SAMPLE_READ we have either single or group mode. */
1467         if (read_format & PERF_FORMAT_GROUP)
1468                 return deliver_sample_group(evlist, tool, event, sample,
1469                                             machine);
1470         else
1471                 return deliver_sample_value(evlist, tool, event, sample,
1472                                             &sample->read.one, machine);
1473 }
1474
1475 static int machines__deliver_event(struct machines *machines,
1476                                    struct evlist *evlist,
1477                                    union perf_event *event,
1478                                    struct perf_sample *sample,
1479                                    struct perf_tool *tool, u64 file_offset)
1480 {
1481         struct evsel *evsel;
1482         struct machine *machine;
1483
1484         dump_event(evlist, event, file_offset, sample);
1485
1486         evsel = evlist__id2evsel(evlist, sample->id);
1487
1488         machine = machines__find_for_cpumode(machines, event, sample);
1489
1490         switch (event->header.type) {
1491         case PERF_RECORD_SAMPLE:
1492                 if (evsel == NULL) {
1493                         ++evlist->stats.nr_unknown_id;
1494                         return 0;
1495                 }
1496                 dump_sample(evsel, event, sample);
1497                 if (machine == NULL) {
1498                         ++evlist->stats.nr_unprocessable_samples;
1499                         return 0;
1500                 }
1501                 return evlist__deliver_sample(evlist, tool, event, sample, evsel, machine);
1502         case PERF_RECORD_MMAP:
1503                 return tool->mmap(tool, event, sample, machine);
1504         case PERF_RECORD_MMAP2:
1505                 if (event->header.misc & PERF_RECORD_MISC_PROC_MAP_PARSE_TIMEOUT)
1506                         ++evlist->stats.nr_proc_map_timeout;
1507                 return tool->mmap2(tool, event, sample, machine);
1508         case PERF_RECORD_COMM:
1509                 return tool->comm(tool, event, sample, machine);
1510         case PERF_RECORD_NAMESPACES:
1511                 return tool->namespaces(tool, event, sample, machine);
1512         case PERF_RECORD_CGROUP:
1513                 return tool->cgroup(tool, event, sample, machine);
1514         case PERF_RECORD_FORK:
1515                 return tool->fork(tool, event, sample, machine);
1516         case PERF_RECORD_EXIT:
1517                 return tool->exit(tool, event, sample, machine);
1518         case PERF_RECORD_LOST:
1519                 if (tool->lost == perf_event__process_lost)
1520                         evlist->stats.total_lost += event->lost.lost;
1521                 return tool->lost(tool, event, sample, machine);
1522         case PERF_RECORD_LOST_SAMPLES:
1523                 if (tool->lost_samples == perf_event__process_lost_samples)
1524                         evlist->stats.total_lost_samples += event->lost_samples.lost;
1525                 return tool->lost_samples(tool, event, sample, machine);
1526         case PERF_RECORD_READ:
1527                 dump_read(evsel, event);
1528                 return tool->read(tool, event, sample, evsel, machine);
1529         case PERF_RECORD_THROTTLE:
1530                 return tool->throttle(tool, event, sample, machine);
1531         case PERF_RECORD_UNTHROTTLE:
1532                 return tool->unthrottle(tool, event, sample, machine);
1533         case PERF_RECORD_AUX:
1534                 if (tool->aux == perf_event__process_aux) {
1535                         if (event->aux.flags & PERF_AUX_FLAG_TRUNCATED)
1536                                 evlist->stats.total_aux_lost += 1;
1537                         if (event->aux.flags & PERF_AUX_FLAG_PARTIAL)
1538                                 evlist->stats.total_aux_partial += 1;
1539                 }
1540                 return tool->aux(tool, event, sample, machine);
1541         case PERF_RECORD_ITRACE_START:
1542                 return tool->itrace_start(tool, event, sample, machine);
1543         case PERF_RECORD_SWITCH:
1544         case PERF_RECORD_SWITCH_CPU_WIDE:
1545                 return tool->context_switch(tool, event, sample, machine);
1546         case PERF_RECORD_KSYMBOL:
1547                 return tool->ksymbol(tool, event, sample, machine);
1548         case PERF_RECORD_BPF_EVENT:
1549                 return tool->bpf(tool, event, sample, machine);
1550         case PERF_RECORD_TEXT_POKE:
1551                 return tool->text_poke(tool, event, sample, machine);
1552         default:
1553                 ++evlist->stats.nr_unknown_events;
1554                 return -1;
1555         }
1556 }
1557
1558 static int perf_session__deliver_event(struct perf_session *session,
1559                                        union perf_event *event,
1560                                        struct perf_tool *tool,
1561                                        u64 file_offset)
1562 {
1563         struct perf_sample sample;
1564         int ret = evlist__parse_sample(session->evlist, event, &sample);
1565
1566         if (ret) {
1567                 pr_err("Can't parse sample, err = %d\n", ret);
1568                 return ret;
1569         }
1570
1571         ret = auxtrace__process_event(session, event, &sample, tool);
1572         if (ret < 0)
1573                 return ret;
1574         if (ret > 0)
1575                 return 0;
1576
1577         ret = machines__deliver_event(&session->machines, session->evlist,
1578                                       event, &sample, tool, file_offset);
1579
1580         if (dump_trace && sample.aux_sample.size)
1581                 auxtrace__dump_auxtrace_sample(session, &sample);
1582
1583         return ret;
1584 }
1585
1586 static s64 perf_session__process_user_event(struct perf_session *session,
1587                                             union perf_event *event,
1588                                             u64 file_offset)
1589 {
1590         struct ordered_events *oe = &session->ordered_events;
1591         struct perf_tool *tool = session->tool;
1592         struct perf_sample sample = { .time = 0, };
1593         int fd = perf_data__fd(session->data);
1594         int err;
1595
1596         if (event->header.type != PERF_RECORD_COMPRESSED ||
1597             tool->compressed == perf_session__process_compressed_event_stub)
1598                 dump_event(session->evlist, event, file_offset, &sample);
1599
1600         /* These events are processed right away */
1601         switch (event->header.type) {
1602         case PERF_RECORD_HEADER_ATTR:
1603                 err = tool->attr(tool, event, &session->evlist);
1604                 if (err == 0) {
1605                         perf_session__set_id_hdr_size(session);
1606                         perf_session__set_comm_exec(session);
1607                 }
1608                 return err;
1609         case PERF_RECORD_EVENT_UPDATE:
1610                 return tool->event_update(tool, event, &session->evlist);
1611         case PERF_RECORD_HEADER_EVENT_TYPE:
1612                 /*
1613                  * Deprecated, but we need to handle it for sake
1614                  * of old data files create in pipe mode.
1615                  */
1616                 return 0;
1617         case PERF_RECORD_HEADER_TRACING_DATA:
1618                 /*
1619                  * Setup for reading amidst mmap, but only when we
1620                  * are in 'file' mode. The 'pipe' fd is in proper
1621                  * place already.
1622                  */
1623                 if (!perf_data__is_pipe(session->data))
1624                         lseek(fd, file_offset, SEEK_SET);
1625                 return tool->tracing_data(session, event);
1626         case PERF_RECORD_HEADER_BUILD_ID:
1627                 return tool->build_id(session, event);
1628         case PERF_RECORD_FINISHED_ROUND:
1629                 return tool->finished_round(tool, event, oe);
1630         case PERF_RECORD_ID_INDEX:
1631                 return tool->id_index(session, event);
1632         case PERF_RECORD_AUXTRACE_INFO:
1633                 return tool->auxtrace_info(session, event);
1634         case PERF_RECORD_AUXTRACE:
1635                 /* setup for reading amidst mmap */
1636                 lseek(fd, file_offset + event->header.size, SEEK_SET);
1637                 return tool->auxtrace(session, event);
1638         case PERF_RECORD_AUXTRACE_ERROR:
1639                 perf_session__auxtrace_error_inc(session, event);
1640                 return tool->auxtrace_error(session, event);
1641         case PERF_RECORD_THREAD_MAP:
1642                 return tool->thread_map(session, event);
1643         case PERF_RECORD_CPU_MAP:
1644                 return tool->cpu_map(session, event);
1645         case PERF_RECORD_STAT_CONFIG:
1646                 return tool->stat_config(session, event);
1647         case PERF_RECORD_STAT:
1648                 return tool->stat(session, event);
1649         case PERF_RECORD_STAT_ROUND:
1650                 return tool->stat_round(session, event);
1651         case PERF_RECORD_TIME_CONV:
1652                 session->time_conv = event->time_conv;
1653                 return tool->time_conv(session, event);
1654         case PERF_RECORD_HEADER_FEATURE:
1655                 return tool->feature(session, event);
1656         case PERF_RECORD_COMPRESSED:
1657                 err = tool->compressed(session, event, file_offset);
1658                 if (err)
1659                         dump_event(session->evlist, event, file_offset, &sample);
1660                 return err;
1661         default:
1662                 return -EINVAL;
1663         }
1664 }
1665
1666 int perf_session__deliver_synth_event(struct perf_session *session,
1667                                       union perf_event *event,
1668                                       struct perf_sample *sample)
1669 {
1670         struct evlist *evlist = session->evlist;
1671         struct perf_tool *tool = session->tool;
1672
1673         events_stats__inc(&evlist->stats, event->header.type);
1674
1675         if (event->header.type >= PERF_RECORD_USER_TYPE_START)
1676                 return perf_session__process_user_event(session, event, 0);
1677
1678         return machines__deliver_event(&session->machines, evlist, event, sample, tool, 0);
1679 }
1680
1681 static void event_swap(union perf_event *event, bool sample_id_all)
1682 {
1683         perf_event__swap_op swap;
1684
1685         swap = perf_event__swap_ops[event->header.type];
1686         if (swap)
1687                 swap(event, sample_id_all);
1688 }
1689
1690 int perf_session__peek_event(struct perf_session *session, off_t file_offset,
1691                              void *buf, size_t buf_sz,
1692                              union perf_event **event_ptr,
1693                              struct perf_sample *sample)
1694 {
1695         union perf_event *event;
1696         size_t hdr_sz, rest;
1697         int fd;
1698
1699         if (session->one_mmap && !session->header.needs_swap) {
1700                 event = file_offset - session->one_mmap_offset +
1701                         session->one_mmap_addr;
1702                 goto out_parse_sample;
1703         }
1704
1705         if (perf_data__is_pipe(session->data))
1706                 return -1;
1707
1708         fd = perf_data__fd(session->data);
1709         hdr_sz = sizeof(struct perf_event_header);
1710
1711         if (buf_sz < hdr_sz)
1712                 return -1;
1713
1714         if (lseek(fd, file_offset, SEEK_SET) == (off_t)-1 ||
1715             readn(fd, buf, hdr_sz) != (ssize_t)hdr_sz)
1716                 return -1;
1717
1718         event = (union perf_event *)buf;
1719
1720         if (session->header.needs_swap)
1721                 perf_event_header__bswap(&event->header);
1722
1723         if (event->header.size < hdr_sz || event->header.size > buf_sz)
1724                 return -1;
1725
1726         buf += hdr_sz;
1727         rest = event->header.size - hdr_sz;
1728
1729         if (readn(fd, buf, rest) != (ssize_t)rest)
1730                 return -1;
1731
1732         if (session->header.needs_swap)
1733                 event_swap(event, evlist__sample_id_all(session->evlist));
1734
1735 out_parse_sample:
1736
1737         if (sample && event->header.type < PERF_RECORD_USER_TYPE_START &&
1738             evlist__parse_sample(session->evlist, event, sample))
1739                 return -1;
1740
1741         *event_ptr = event;
1742
1743         return 0;
1744 }
1745
1746 int perf_session__peek_events(struct perf_session *session, u64 offset,
1747                               u64 size, peek_events_cb_t cb, void *data)
1748 {
1749         u64 max_offset = offset + size;
1750         char buf[PERF_SAMPLE_MAX_SIZE];
1751         union perf_event *event;
1752         int err;
1753
1754         do {
1755                 err = perf_session__peek_event(session, offset, buf,
1756                                                PERF_SAMPLE_MAX_SIZE, &event,
1757                                                NULL);
1758                 if (err)
1759                         return err;
1760
1761                 err = cb(session, event, offset, data);
1762                 if (err)
1763                         return err;
1764
1765                 offset += event->header.size;
1766                 if (event->header.type == PERF_RECORD_AUXTRACE)
1767                         offset += event->auxtrace.size;
1768
1769         } while (offset < max_offset);
1770
1771         return err;
1772 }
1773
1774 static s64 perf_session__process_event(struct perf_session *session,
1775                                        union perf_event *event, u64 file_offset)
1776 {
1777         struct evlist *evlist = session->evlist;
1778         struct perf_tool *tool = session->tool;
1779         int ret;
1780
1781         if (session->header.needs_swap)
1782                 event_swap(event, evlist__sample_id_all(evlist));
1783
1784         if (event->header.type >= PERF_RECORD_HEADER_MAX)
1785                 return -EINVAL;
1786
1787         events_stats__inc(&evlist->stats, event->header.type);
1788
1789         if (event->header.type >= PERF_RECORD_USER_TYPE_START)
1790                 return perf_session__process_user_event(session, event, file_offset);
1791
1792         if (tool->ordered_events) {
1793                 u64 timestamp = -1ULL;
1794
1795                 ret = evlist__parse_sample_timestamp(evlist, event, &timestamp);
1796                 if (ret && ret != -1)
1797                         return ret;
1798
1799                 ret = perf_session__queue_event(session, event, timestamp, file_offset);
1800                 if (ret != -ETIME)
1801                         return ret;
1802         }
1803
1804         return perf_session__deliver_event(session, event, tool, file_offset);
1805 }
1806
1807 void perf_event_header__bswap(struct perf_event_header *hdr)
1808 {
1809         hdr->type = bswap_32(hdr->type);
1810         hdr->misc = bswap_16(hdr->misc);
1811         hdr->size = bswap_16(hdr->size);
1812 }
1813
1814 struct thread *perf_session__findnew(struct perf_session *session, pid_t pid)
1815 {
1816         return machine__findnew_thread(&session->machines.host, -1, pid);
1817 }
1818
1819 int perf_session__register_idle_thread(struct perf_session *session)
1820 {
1821         struct thread *thread = machine__idle_thread(&session->machines.host);
1822
1823         /* machine__idle_thread() got the thread, so put it */
1824         thread__put(thread);
1825         return thread ? 0 : -1;
1826 }
1827
1828 static void
1829 perf_session__warn_order(const struct perf_session *session)
1830 {
1831         const struct ordered_events *oe = &session->ordered_events;
1832         struct evsel *evsel;
1833         bool should_warn = true;
1834
1835         evlist__for_each_entry(session->evlist, evsel) {
1836                 if (evsel->core.attr.write_backward)
1837                         should_warn = false;
1838         }
1839
1840         if (!should_warn)
1841                 return;
1842         if (oe->nr_unordered_events != 0)
1843                 ui__warning("%u out of order events recorded.\n", oe->nr_unordered_events);
1844 }
1845
1846 static void perf_session__warn_about_errors(const struct perf_session *session)
1847 {
1848         const struct events_stats *stats = &session->evlist->stats;
1849
1850         if (session->tool->lost == perf_event__process_lost &&
1851             stats->nr_events[PERF_RECORD_LOST] != 0) {
1852                 ui__warning("Processed %d events and lost %d chunks!\n\n"
1853                             "Check IO/CPU overload!\n\n",
1854                             stats->nr_events[0],
1855                             stats->nr_events[PERF_RECORD_LOST]);
1856         }
1857
1858         if (session->tool->lost_samples == perf_event__process_lost_samples) {
1859                 double drop_rate;
1860
1861                 drop_rate = (double)stats->total_lost_samples /
1862                             (double) (stats->nr_events[PERF_RECORD_SAMPLE] + stats->total_lost_samples);
1863                 if (drop_rate > 0.05) {
1864                         ui__warning("Processed %" PRIu64 " samples and lost %3.2f%%!\n\n",
1865                                     stats->nr_events[PERF_RECORD_SAMPLE] + stats->total_lost_samples,
1866                                     drop_rate * 100.0);
1867                 }
1868         }
1869
1870         if (session->tool->aux == perf_event__process_aux &&
1871             stats->total_aux_lost != 0) {
1872                 ui__warning("AUX data lost %" PRIu64 " times out of %u!\n\n",
1873                             stats->total_aux_lost,
1874                             stats->nr_events[PERF_RECORD_AUX]);
1875         }
1876
1877         if (session->tool->aux == perf_event__process_aux &&
1878             stats->total_aux_partial != 0) {
1879                 bool vmm_exclusive = false;
1880
1881                 (void)sysfs__read_bool("module/kvm_intel/parameters/vmm_exclusive",
1882                                        &vmm_exclusive);
1883
1884                 ui__warning("AUX data had gaps in it %" PRIu64 " times out of %u!\n\n"
1885                             "Are you running a KVM guest in the background?%s\n\n",
1886                             stats->total_aux_partial,
1887                             stats->nr_events[PERF_RECORD_AUX],
1888                             vmm_exclusive ?
1889                             "\nReloading kvm_intel module with vmm_exclusive=0\n"
1890                             "will reduce the gaps to only guest's timeslices." :
1891                             "");
1892         }
1893
1894         if (stats->nr_unknown_events != 0) {
1895                 ui__warning("Found %u unknown events!\n\n"
1896                             "Is this an older tool processing a perf.data "
1897                             "file generated by a more recent tool?\n\n"
1898                             "If that is not the case, consider "
1899                             "reporting to linux-kernel@vger.kernel.org.\n\n",
1900                             stats->nr_unknown_events);
1901         }
1902
1903         if (stats->nr_unknown_id != 0) {
1904                 ui__warning("%u samples with id not present in the header\n",
1905                             stats->nr_unknown_id);
1906         }
1907
1908         if (stats->nr_invalid_chains != 0) {
1909                 ui__warning("Found invalid callchains!\n\n"
1910                             "%u out of %u events were discarded for this reason.\n\n"
1911                             "Consider reporting to linux-kernel@vger.kernel.org.\n\n",
1912                             stats->nr_invalid_chains,
1913                             stats->nr_events[PERF_RECORD_SAMPLE]);
1914         }
1915
1916         if (stats->nr_unprocessable_samples != 0) {
1917                 ui__warning("%u unprocessable samples recorded.\n"
1918                             "Do you have a KVM guest running and not using 'perf kvm'?\n",
1919                             stats->nr_unprocessable_samples);
1920         }
1921
1922         perf_session__warn_order(session);
1923
1924         events_stats__auxtrace_error_warn(stats);
1925
1926         if (stats->nr_proc_map_timeout != 0) {
1927                 ui__warning("%d map information files for pre-existing threads were\n"
1928                             "not processed, if there are samples for addresses they\n"
1929                             "will not be resolved, you may find out which are these\n"
1930                             "threads by running with -v and redirecting the output\n"
1931                             "to a file.\n"
1932                             "The time limit to process proc map is too short?\n"
1933                             "Increase it by --proc-map-timeout\n",
1934                             stats->nr_proc_map_timeout);
1935         }
1936 }
1937
1938 static int perf_session__flush_thread_stack(struct thread *thread,
1939                                             void *p __maybe_unused)
1940 {
1941         return thread_stack__flush(thread);
1942 }
1943
1944 static int perf_session__flush_thread_stacks(struct perf_session *session)
1945 {
1946         return machines__for_each_thread(&session->machines,
1947                                          perf_session__flush_thread_stack,
1948                                          NULL);
1949 }
1950
1951 volatile int session_done;
1952
1953 static int __perf_session__process_decomp_events(struct perf_session *session);
1954
1955 static int __perf_session__process_pipe_events(struct perf_session *session)
1956 {
1957         struct ordered_events *oe = &session->ordered_events;
1958         struct perf_tool *tool = session->tool;
1959         union perf_event *event;
1960         uint32_t size, cur_size = 0;
1961         void *buf = NULL;
1962         s64 skip = 0;
1963         u64 head;
1964         ssize_t err;
1965         void *p;
1966
1967         perf_tool__fill_defaults(tool);
1968
1969         head = 0;
1970         cur_size = sizeof(union perf_event);
1971
1972         buf = malloc(cur_size);
1973         if (!buf)
1974                 return -errno;
1975         ordered_events__set_copy_on_queue(oe, true);
1976 more:
1977         event = buf;
1978         err = perf_data__read(session->data, event,
1979                               sizeof(struct perf_event_header));
1980         if (err <= 0) {
1981                 if (err == 0)
1982                         goto done;
1983
1984                 pr_err("failed to read event header\n");
1985                 goto out_err;
1986         }
1987
1988         if (session->header.needs_swap)
1989                 perf_event_header__bswap(&event->header);
1990
1991         size = event->header.size;
1992         if (size < sizeof(struct perf_event_header)) {
1993                 pr_err("bad event header size\n");
1994                 goto out_err;
1995         }
1996
1997         if (size > cur_size) {
1998                 void *new = realloc(buf, size);
1999                 if (!new) {
2000                         pr_err("failed to allocate memory to read event\n");
2001                         goto out_err;
2002                 }
2003                 buf = new;
2004                 cur_size = size;
2005                 event = buf;
2006         }
2007         p = event;
2008         p += sizeof(struct perf_event_header);
2009
2010         if (size - sizeof(struct perf_event_header)) {
2011                 err = perf_data__read(session->data, p,
2012                                       size - sizeof(struct perf_event_header));
2013                 if (err <= 0) {
2014                         if (err == 0) {
2015                                 pr_err("unexpected end of event stream\n");
2016                                 goto done;
2017                         }
2018
2019                         pr_err("failed to read event data\n");
2020                         goto out_err;
2021                 }
2022         }
2023
2024         if ((skip = perf_session__process_event(session, event, head)) < 0) {
2025                 pr_err("%#" PRIx64 " [%#x]: failed to process type: %d\n",
2026                        head, event->header.size, event->header.type);
2027                 err = -EINVAL;
2028                 goto out_err;
2029         }
2030
2031         head += size;
2032
2033         if (skip > 0)
2034                 head += skip;
2035
2036         err = __perf_session__process_decomp_events(session);
2037         if (err)
2038                 goto out_err;
2039
2040         if (!session_done())
2041                 goto more;
2042 done:
2043         /* do the final flush for ordered samples */
2044         err = ordered_events__flush(oe, OE_FLUSH__FINAL);
2045         if (err)
2046                 goto out_err;
2047         err = auxtrace__flush_events(session, tool);
2048         if (err)
2049                 goto out_err;
2050         err = perf_session__flush_thread_stacks(session);
2051 out_err:
2052         free(buf);
2053         if (!tool->no_warn)
2054                 perf_session__warn_about_errors(session);
2055         ordered_events__free(&session->ordered_events);
2056         auxtrace__free_events(session);
2057         return err;
2058 }
2059
2060 static union perf_event *
2061 prefetch_event(char *buf, u64 head, size_t mmap_size,
2062                bool needs_swap, union perf_event *error)
2063 {
2064         union perf_event *event;
2065
2066         /*
2067          * Ensure we have enough space remaining to read
2068          * the size of the event in the headers.
2069          */
2070         if (head + sizeof(event->header) > mmap_size)
2071                 return NULL;
2072
2073         event = (union perf_event *)(buf + head);
2074         if (needs_swap)
2075                 perf_event_header__bswap(&event->header);
2076
2077         if (head + event->header.size <= mmap_size)
2078                 return event;
2079
2080         /* We're not fetching the event so swap back again */
2081         if (needs_swap)
2082                 perf_event_header__bswap(&event->header);
2083
2084         pr_debug("%s: head=%#" PRIx64 " event->header_size=%#x, mmap_size=%#zx:"
2085                  " fuzzed or compressed perf.data?\n",__func__, head, event->header.size, mmap_size);
2086
2087         return error;
2088 }
2089
2090 static union perf_event *
2091 fetch_mmaped_event(u64 head, size_t mmap_size, char *buf, bool needs_swap)
2092 {
2093         return prefetch_event(buf, head, mmap_size, needs_swap, ERR_PTR(-EINVAL));
2094 }
2095
2096 static union perf_event *
2097 fetch_decomp_event(u64 head, size_t mmap_size, char *buf, bool needs_swap)
2098 {
2099         return prefetch_event(buf, head, mmap_size, needs_swap, NULL);
2100 }
2101
2102 static int __perf_session__process_decomp_events(struct perf_session *session)
2103 {
2104         s64 skip;
2105         u64 size, file_pos = 0;
2106         struct decomp *decomp = session->decomp_last;
2107
2108         if (!decomp)
2109                 return 0;
2110
2111         while (decomp->head < decomp->size && !session_done()) {
2112                 union perf_event *event = fetch_decomp_event(decomp->head, decomp->size, decomp->data,
2113                                                              session->header.needs_swap);
2114
2115                 if (!event)
2116                         break;
2117
2118                 size = event->header.size;
2119
2120                 if (size < sizeof(struct perf_event_header) ||
2121                     (skip = perf_session__process_event(session, event, file_pos)) < 0) {
2122                         pr_err("%#" PRIx64 " [%#x]: failed to process type: %d\n",
2123                                 decomp->file_pos + decomp->head, event->header.size, event->header.type);
2124                         return -EINVAL;
2125                 }
2126
2127                 if (skip)
2128                         size += skip;
2129
2130                 decomp->head += size;
2131         }
2132
2133         return 0;
2134 }
2135
2136 /*
2137  * On 64bit we can mmap the data file in one go. No need for tiny mmap
2138  * slices. On 32bit we use 32MB.
2139  */
2140 #if BITS_PER_LONG == 64
2141 #define MMAP_SIZE ULLONG_MAX
2142 #define NUM_MMAPS 1
2143 #else
2144 #define MMAP_SIZE (32 * 1024 * 1024ULL)
2145 #define NUM_MMAPS 128
2146 #endif
2147
2148 struct reader;
2149
2150 typedef s64 (*reader_cb_t)(struct perf_session *session,
2151                            union perf_event *event,
2152                            u64 file_offset);
2153
2154 struct reader {
2155         int              fd;
2156         u64              data_size;
2157         u64              data_offset;
2158         reader_cb_t      process;
2159 };
2160
2161 static int
2162 reader__process_events(struct reader *rd, struct perf_session *session,
2163                        struct ui_progress *prog)
2164 {
2165         u64 data_size = rd->data_size;
2166         u64 head, page_offset, file_offset, file_pos, size;
2167         int err = 0, mmap_prot, mmap_flags, map_idx = 0;
2168         size_t  mmap_size;
2169         char *buf, *mmaps[NUM_MMAPS];
2170         union perf_event *event;
2171         s64 skip;
2172
2173         page_offset = page_size * (rd->data_offset / page_size);
2174         file_offset = page_offset;
2175         head = rd->data_offset - page_offset;
2176
2177         ui_progress__init_size(prog, data_size, "Processing events...");
2178
2179         data_size += rd->data_offset;
2180
2181         mmap_size = MMAP_SIZE;
2182         if (mmap_size > data_size) {
2183                 mmap_size = data_size;
2184                 session->one_mmap = true;
2185         }
2186
2187         memset(mmaps, 0, sizeof(mmaps));
2188
2189         mmap_prot  = PROT_READ;
2190         mmap_flags = MAP_SHARED;
2191
2192         if (session->header.needs_swap) {
2193                 mmap_prot  |= PROT_WRITE;
2194                 mmap_flags = MAP_PRIVATE;
2195         }
2196 remap:
2197         buf = mmap(NULL, mmap_size, mmap_prot, mmap_flags, rd->fd,
2198                    file_offset);
2199         if (buf == MAP_FAILED) {
2200                 pr_err("failed to mmap file\n");
2201                 err = -errno;
2202                 goto out;
2203         }
2204         mmaps[map_idx] = buf;
2205         map_idx = (map_idx + 1) & (ARRAY_SIZE(mmaps) - 1);
2206         file_pos = file_offset + head;
2207         if (session->one_mmap) {
2208                 session->one_mmap_addr = buf;
2209                 session->one_mmap_offset = file_offset;
2210         }
2211
2212 more:
2213         event = fetch_mmaped_event(head, mmap_size, buf, session->header.needs_swap);
2214         if (IS_ERR(event))
2215                 return PTR_ERR(event);
2216
2217         if (!event) {
2218                 if (mmaps[map_idx]) {
2219                         munmap(mmaps[map_idx], mmap_size);
2220                         mmaps[map_idx] = NULL;
2221                 }
2222
2223                 page_offset = page_size * (head / page_size);
2224                 file_offset += page_offset;
2225                 head -= page_offset;
2226                 goto remap;
2227         }
2228
2229         size = event->header.size;
2230
2231         skip = -EINVAL;
2232
2233         if (size < sizeof(struct perf_event_header) ||
2234             (skip = rd->process(session, event, file_pos)) < 0) {
2235                 pr_err("%#" PRIx64 " [%#x]: failed to process type: %d [%s]\n",
2236                        file_offset + head, event->header.size,
2237                        event->header.type, strerror(-skip));
2238                 err = skip;
2239                 goto out;
2240         }
2241
2242         if (skip)
2243                 size += skip;
2244
2245         head += size;
2246         file_pos += size;
2247
2248         err = __perf_session__process_decomp_events(session);
2249         if (err)
2250                 goto out;
2251
2252         ui_progress__update(prog, size);
2253
2254         if (session_done())
2255                 goto out;
2256
2257         if (file_pos < data_size)
2258                 goto more;
2259
2260 out:
2261         return err;
2262 }
2263
2264 static s64 process_simple(struct perf_session *session,
2265                           union perf_event *event,
2266                           u64 file_offset)
2267 {
2268         return perf_session__process_event(session, event, file_offset);
2269 }
2270
2271 static int __perf_session__process_events(struct perf_session *session)
2272 {
2273         struct reader rd = {
2274                 .fd             = perf_data__fd(session->data),
2275                 .data_size      = session->header.data_size,
2276                 .data_offset    = session->header.data_offset,
2277                 .process        = process_simple,
2278         };
2279         struct ordered_events *oe = &session->ordered_events;
2280         struct perf_tool *tool = session->tool;
2281         struct ui_progress prog;
2282         int err;
2283
2284         perf_tool__fill_defaults(tool);
2285
2286         if (rd.data_size == 0)
2287                 return -1;
2288
2289         ui_progress__init_size(&prog, rd.data_size, "Processing events...");
2290
2291         err = reader__process_events(&rd, session, &prog);
2292         if (err)
2293                 goto out_err;
2294         /* do the final flush for ordered samples */
2295         err = ordered_events__flush(oe, OE_FLUSH__FINAL);
2296         if (err)
2297                 goto out_err;
2298         err = auxtrace__flush_events(session, tool);
2299         if (err)
2300                 goto out_err;
2301         err = perf_session__flush_thread_stacks(session);
2302 out_err:
2303         ui_progress__finish();
2304         if (!tool->no_warn)
2305                 perf_session__warn_about_errors(session);
2306         /*
2307          * We may switching perf.data output, make ordered_events
2308          * reusable.
2309          */
2310         ordered_events__reinit(&session->ordered_events);
2311         auxtrace__free_events(session);
2312         session->one_mmap = false;
2313         return err;
2314 }
2315
2316 int perf_session__process_events(struct perf_session *session)
2317 {
2318         if (perf_session__register_idle_thread(session) < 0)
2319                 return -ENOMEM;
2320
2321         if (perf_data__is_pipe(session->data))
2322                 return __perf_session__process_pipe_events(session);
2323
2324         return __perf_session__process_events(session);
2325 }
2326
2327 bool perf_session__has_traces(struct perf_session *session, const char *msg)
2328 {
2329         struct evsel *evsel;
2330
2331         evlist__for_each_entry(session->evlist, evsel) {
2332                 if (evsel->core.attr.type == PERF_TYPE_TRACEPOINT)
2333                         return true;
2334         }
2335
2336         pr_err("No trace sample to read. Did you call 'perf %s'?\n", msg);
2337         return false;
2338 }
2339
2340 int map__set_kallsyms_ref_reloc_sym(struct map *map, const char *symbol_name, u64 addr)
2341 {
2342         char *bracket;
2343         struct ref_reloc_sym *ref;
2344         struct kmap *kmap;
2345
2346         ref = zalloc(sizeof(struct ref_reloc_sym));
2347         if (ref == NULL)
2348                 return -ENOMEM;
2349
2350         ref->name = strdup(symbol_name);
2351         if (ref->name == NULL) {
2352                 free(ref);
2353                 return -ENOMEM;
2354         }
2355
2356         bracket = strchr(ref->name, ']');
2357         if (bracket)
2358                 *bracket = '\0';
2359
2360         ref->addr = addr;
2361
2362         kmap = map__kmap(map);
2363         if (kmap)
2364                 kmap->ref_reloc_sym = ref;
2365
2366         return 0;
2367 }
2368
2369 size_t perf_session__fprintf_dsos(struct perf_session *session, FILE *fp)
2370 {
2371         return machines__fprintf_dsos(&session->machines, fp);
2372 }
2373
2374 size_t perf_session__fprintf_dsos_buildid(struct perf_session *session, FILE *fp,
2375                                           bool (skip)(struct dso *dso, int parm), int parm)
2376 {
2377         return machines__fprintf_dsos_buildid(&session->machines, fp, skip, parm);
2378 }
2379
2380 size_t perf_session__fprintf_nr_events(struct perf_session *session, FILE *fp,
2381                                        bool skip_empty)
2382 {
2383         size_t ret;
2384         const char *msg = "";
2385
2386         if (perf_header__has_feat(&session->header, HEADER_AUXTRACE))
2387                 msg = " (excludes AUX area (e.g. instruction trace) decoded / synthesized events)";
2388
2389         ret = fprintf(fp, "\nAggregated stats:%s\n", msg);
2390
2391         ret += events_stats__fprintf(&session->evlist->stats, fp, skip_empty);
2392         return ret;
2393 }
2394
2395 size_t perf_session__fprintf(struct perf_session *session, FILE *fp)
2396 {
2397         /*
2398          * FIXME: Here we have to actually print all the machines in this
2399          * session, not just the host...
2400          */
2401         return machine__fprintf(&session->machines.host, fp);
2402 }
2403
2404 struct evsel *perf_session__find_first_evtype(struct perf_session *session,
2405                                               unsigned int type)
2406 {
2407         struct evsel *pos;
2408
2409         evlist__for_each_entry(session->evlist, pos) {
2410                 if (pos->core.attr.type == type)
2411                         return pos;
2412         }
2413         return NULL;
2414 }
2415
2416 int perf_session__cpu_bitmap(struct perf_session *session,
2417                              const char *cpu_list, unsigned long *cpu_bitmap)
2418 {
2419         int i, err = -1;
2420         struct perf_cpu_map *map;
2421         int nr_cpus = min(session->header.env.nr_cpus_avail, MAX_NR_CPUS);
2422
2423         for (i = 0; i < PERF_TYPE_MAX; ++i) {
2424                 struct evsel *evsel;
2425
2426                 evsel = perf_session__find_first_evtype(session, i);
2427                 if (!evsel)
2428                         continue;
2429
2430                 if (!(evsel->core.attr.sample_type & PERF_SAMPLE_CPU)) {
2431                         pr_err("File does not contain CPU events. "
2432                                "Remove -C option to proceed.\n");
2433                         return -1;
2434                 }
2435         }
2436
2437         map = perf_cpu_map__new(cpu_list);
2438         if (map == NULL) {
2439                 pr_err("Invalid cpu_list\n");
2440                 return -1;
2441         }
2442
2443         for (i = 0; i < map->nr; i++) {
2444                 int cpu = map->map[i];
2445
2446                 if (cpu >= nr_cpus) {
2447                         pr_err("Requested CPU %d too large. "
2448                                "Consider raising MAX_NR_CPUS\n", cpu);
2449                         goto out_delete_map;
2450                 }
2451
2452                 set_bit(cpu, cpu_bitmap);
2453         }
2454
2455         err = 0;
2456
2457 out_delete_map:
2458         perf_cpu_map__put(map);
2459         return err;
2460 }
2461
2462 void perf_session__fprintf_info(struct perf_session *session, FILE *fp,
2463                                 bool full)
2464 {
2465         if (session == NULL || fp == NULL)
2466                 return;
2467
2468         fprintf(fp, "# ========\n");
2469         perf_header__fprintf_info(session, fp, full);
2470         fprintf(fp, "# ========\n#\n");
2471 }
2472
2473 int perf_event__process_id_index(struct perf_session *session,
2474                                  union perf_event *event)
2475 {
2476         struct evlist *evlist = session->evlist;
2477         struct perf_record_id_index *ie = &event->id_index;
2478         size_t i, nr, max_nr;
2479
2480         max_nr = (ie->header.size - sizeof(struct perf_record_id_index)) /
2481                  sizeof(struct id_index_entry);
2482         nr = ie->nr;
2483         if (nr > max_nr)
2484                 return -EINVAL;
2485
2486         if (dump_trace)
2487                 fprintf(stdout, " nr: %zu\n", nr);
2488
2489         for (i = 0; i < nr; i++) {
2490                 struct id_index_entry *e = &ie->entries[i];
2491                 struct perf_sample_id *sid;
2492
2493                 if (dump_trace) {
2494                         fprintf(stdout, " ... id: %"PRI_lu64, e->id);
2495                         fprintf(stdout, "  idx: %"PRI_lu64, e->idx);
2496                         fprintf(stdout, "  cpu: %"PRI_ld64, e->cpu);
2497                         fprintf(stdout, "  tid: %"PRI_ld64"\n", e->tid);
2498                 }
2499
2500                 sid = evlist__id2sid(evlist, e->id);
2501                 if (!sid)
2502                         return -ENOENT;
2503                 sid->idx = e->idx;
2504                 sid->cpu = e->cpu;
2505                 sid->tid = e->tid;
2506         }
2507         return 0;
2508 }