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