tools/perf/util/session.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 #include <errno.h>
   3 #include <inttypes.h>
   4 #include <linux/kernel.h>
   5 #include <linux/zalloc.h>
   6 #include <traceevent/event-parse.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
  14 #include "evlist.h"
  15 #include "evsel.h"
  16 #include "memswap.h"
  17 #include "map.h"
  18 #include "symbol.h"
  19 #include "session.h"
  20 #include "tool.h"
  21 #include "sort.h"
  22 #include "cpumap.h"
  23 #include "perf_regs.h"
  24 #include "asm/bug.h"
  25 #include "auxtrace.h"
  26 #include "thread.h"
  27 #include "thread-stack.h"
  28 #include "sample-raw.h"
  29 #include "stat.h"
  30 #include "arch/common.h"
  31
  32 #ifdef HAVE_ZSTD_SUPPORT
  33 static int perf_session__process_compressed_event(struct perf_session *session,
  34                                                   union perf_event *event, u64 file_offset)
  35 {
  36         void *src;
  37         size_t decomp_size, src_size;
  38         u64 decomp_last_rem = 0;
  39         size_t mmap_len, decomp_len = session->header.env.comp_mmap_len;
  40         struct decomp *decomp, *decomp_last = session->decomp_last;
  41
  42         if (decomp_last) {
  43                 decomp_last_rem = decomp_last->size - decomp_last->head;
  44                 decomp_len += decomp_last_rem;
  45         }
  46
  47         mmap_len = sizeof(struct decomp) + decomp_len;
  48         decomp = mmap(NULL, mmap_len, PROT_READ|PROT_WRITE,
  49                       MAP_ANONYMOUS|MAP_PRIVATE, -1, 0);
  50         if (decomp == MAP_FAILED) {
  51                 pr_err("Couldn't allocate memory for decompression\n");
  52                 return -1;
  53         }
  54
  55         decomp->file_pos = file_offset;
  56         decomp->mmap_len = mmap_len;
  57         decomp->head = 0;
  58
  59         if (decomp_last_rem) {
  60                 memcpy(decomp->data, &(decomp_last->data[decomp_last->head]), decomp_last_rem);
  61                 decomp->size = decomp_last_rem;
  62         }
  63
  64         src = (void *)event + sizeof(struct compressed_event);
  65         src_size = event->pack.header.size - sizeof(struct compressed_event);
  66
  67         decomp_size = zstd_decompress_stream(&(session->zstd_data), src, src_size,
  68                                 &(decomp->data[decomp_last_rem]), decomp_len - decomp_last_rem);
  69         if (!decomp_size) {
  70                 munmap(decomp, mmap_len);
  71                 pr_err("Couldn't decompress data\n");
  72                 return -1;
  73         }
  74
  75         decomp->size += decomp_size;
  76
  77         if (session->decomp == NULL) {
  78                 session->decomp = decomp;
  79                 session->decomp_last = decomp;
  80         } else {
  81                 session->decomp_last->next = decomp;
  82                 session->decomp_last = decomp;
  83         }
  84
  85         pr_debug("decomp (B): %ld to %ld\n", src_size, decomp_size);
  86
  87         return 0;
  88 }
  89 #else /* !HAVE_ZSTD_SUPPORT */
  90 #define perf_session__process_compressed_event perf_session__process_compressed_event_stub
  91 #endif
  92
  93 static int perf_session__deliver_event(struct perf_session *session,
  94                                        union perf_event *event,
  95                                        struct perf_tool *tool,
  96                                        u64 file_offset);
  97
  98 static int perf_session__open(struct perf_session *session)
  99 {
 100         struct perf_data *data = session->data;
 101
 102         if (perf_session__read_header(session) < 0) {
 103                 pr_err("incompatible file format (rerun with -v to learn more)\n");
 104                 return -1;
 105         }
 106
 107         if (perf_data__is_pipe(data))
 108                 return 0;
 109
 110         if (perf_header__has_feat(&session->header, HEADER_STAT))
 111                 return 0;
 112
 113         if (!perf_evlist__valid_sample_type(session->evlist)) {
 114                 pr_err("non matching sample_type\n");
 115                 return -1;
 116         }
 117
 118         if (!perf_evlist__valid_sample_id_all(session->evlist)) {
 119                 pr_err("non matching sample_id_all\n");
 120                 return -1;
 121         }
 122
 123         if (!perf_evlist__valid_read_format(session->evlist)) {
 124                 pr_err("non matching read_format\n");
 125                 return -1;
 126         }
 127
 128         return 0;
 129 }
 130
 131 void perf_session__set_id_hdr_size(struct perf_session *session)
 132 {
 133         u16 id_hdr_size = perf_evlist__id_hdr_size(session->evlist);
 134
 135         machines__set_id_hdr_size(&session->machines, id_hdr_size);
 136 }
 137
 138 int perf_session__create_kernel_maps(struct perf_session *session)
 139 {
 140         int ret = machine__create_kernel_maps(&session->machines.host);
 141
 142         if (ret >= 0)
 143                 ret = machines__create_guest_kernel_maps(&session->machines);
 144         return ret;
 145 }
 146
 147 static void perf_session__destroy_kernel_maps(struct perf_session *session)
 148 {
 149         machines__destroy_kernel_maps(&session->machines);
 150 }
 151
 152 static bool perf_session__has_comm_exec(struct perf_session *session)
 153 {
 154         struct perf_evsel *evsel;
 155
 156         evlist__for_each_entry(session->evlist, evsel) {
 157                 if (evsel->attr.comm_exec)
 158                         return true;
 159         }
 160
 161         return false;
 162 }
 163
 164 static void perf_session__set_comm_exec(struct perf_session *session)
 165 {
 166         bool comm_exec = perf_session__has_comm_exec(session);
 167
 168         machines__set_comm_exec(&session->machines, comm_exec);
 169 }
 170
 171 static int ordered_events__deliver_event(struct ordered_events *oe,
 172                                          struct ordered_event *event)
 173 {
 174         struct perf_session *session = container_of(oe, struct perf_session,
 175                                                     ordered_events);
 176
 177         return perf_session__deliver_event(session, event->event,
 178                                            session->tool, event->file_offset);
 179 }
 180
 181 struct perf_session *perf_session__new(struct perf_data *data,
 182                                        bool repipe, struct perf_tool *tool)
 183 {
 184         struct perf_session *session = zalloc(sizeof(*session));
 185
 186         if (!session)
 187                 goto out;
 188
 189         session->repipe = repipe;
 190         session->tool   = tool;
 191         INIT_LIST_HEAD(&session->auxtrace_index);
 192         machines__init(&session->machines);
 193         ordered_events__init(&session->ordered_events,
 194                              ordered_events__deliver_event, NULL);
 195
 196         perf_env__init(&session->header.env);
 197         if (data) {
 198                 if (perf_data__open(data))
 199                         goto out_delete;
 200
 201                 session->data = data;
 202
 203                 if (perf_data__is_read(data)) {
 204                         if (perf_session__open(session) < 0)
 205                                 goto out_delete;
 206
 207                         /*
 208                          * set session attributes that are present in perf.data
 209                          * but not in pipe-mode.
 210                          */
 211                         if (!data->is_pipe) {
 212                                 perf_session__set_id_hdr_size(session);
 213                                 perf_session__set_comm_exec(session);
 214                         }
 215
 216                         perf_evlist__init_trace_event_sample_raw(session->evlist);
 217
 218                         /* Open the directory data. */
 219                         if (data->is_dir && perf_data__open_dir(data))
 220                                 goto out_delete;
 221                 }
 222         } else  {
 223                 session->machines.host.env = &perf_env;
 224         }
 225
 226         session->machines.host.single_address_space =
 227                 perf_env__single_address_space(session->machines.host.env);
 228
 229         if (!data || perf_data__is_write(data)) {
 230                 /*
 231                  * In O_RDONLY mode this will be performed when reading the
 232                  * kernel MMAP event, in perf_event__process_mmap().
 233                  */
 234                 if (perf_session__create_kernel_maps(session) < 0)
 235                         pr_warning("Cannot read kernel map\n");
 236         }
 237
 238         /*
 239          * In pipe-mode, evlist is empty until PERF_RECORD_HEADER_ATTR is
 240          * processed, so perf_evlist__sample_id_all is not meaningful here.
 241          */
 242         if ((!data || !data->is_pipe) && tool && tool->ordering_requires_timestamps &&
 243             tool->ordered_events && !perf_evlist__sample_id_all(session->evlist)) {
 244                 dump_printf("WARNING: No sample_id_all support, falling back to unordered processing\n");
 245                 tool->ordered_events = false;
 246         }
 247
 248         return session;
 249
 250  out_delete:
 251         perf_session__delete(session);
 252  out:
 253         return NULL;
 254 }
 255
 256 static void perf_session__delete_threads(struct perf_session *session)
 257 {
 258         machine__delete_threads(&session->machines.host);
 259 }
 260
 261 static void perf_session__release_decomp_events(struct perf_session *session)
 262 {
 263         struct decomp *next, *decomp;
 264         size_t mmap_len;
 265         next = session->decomp;
 266         do {
 267                 decomp = next;
 268                 if (decomp == NULL)
 269                         break;
 270                 next = decomp->next;
 271                 mmap_len = decomp->mmap_len;
 272                 munmap(decomp, mmap_len);
 273         } while (1);
 274 }
 275
 276 void perf_session__delete(struct perf_session *session)
 277 {
 278         if (session == NULL)
 279                 return;
 280         auxtrace__free(session);
 281         auxtrace_index__free(&session->auxtrace_index);
 282         perf_session__destroy_kernel_maps(session);
 283         perf_session__delete_threads(session);
 284         perf_session__release_decomp_events(session);
 285         perf_env__exit(&session->header.env);
 286         machines__exit(&session->machines);
 287         if (session->data)
 288                 perf_data__close(session->data);
 289         free(session);
 290 }
 291
 292 static int process_event_synth_tracing_data_stub(struct perf_session *session
 293                                                  __maybe_unused,
 294                                                  union perf_event *event
 295                                                  __maybe_unused)
 296 {
 297         dump_printf(": unhandled!\n");
 298         return 0;
 299 }
 300
 301 static int process_event_synth_attr_stub(struct perf_tool *tool __maybe_unused,
 302                                          union perf_event *event __maybe_unused,
 303                                          struct perf_evlist **pevlist
 304                                          __maybe_unused)
 305 {
 306         dump_printf(": unhandled!\n");
 307         return 0;
 308 }
 309
 310 static int process_event_synth_event_update_stub(struct perf_tool *tool __maybe_unused,
 311                                                  union perf_event *event __maybe_unused,
 312                                                  struct perf_evlist **pevlist
 313                                                  __maybe_unused)
 314 {
 315         if (dump_trace)
 316                 perf_event__fprintf_event_update(event, stdout);
 317
 318         dump_printf(": unhandled!\n");
 319         return 0;
 320 }
 321
 322 static int process_event_sample_stub(struct perf_tool *tool __maybe_unused,
 323                                      union perf_event *event __maybe_unused,
 324                                      struct perf_sample *sample __maybe_unused,
 325                                      struct perf_evsel *evsel __maybe_unused,
 326                                      struct machine *machine __maybe_unused)
 327 {
 328         dump_printf(": unhandled!\n");
 329         return 0;
 330 }
 331
 332 static int process_event_stub(struct perf_tool *tool __maybe_unused,
 333                               union perf_event *event __maybe_unused,
 334                               struct perf_sample *sample __maybe_unused,
 335                               struct machine *machine __maybe_unused)
 336 {
 337         dump_printf(": unhandled!\n");
 338         return 0;
 339 }
 340
 341 static int process_finished_round_stub(struct perf_tool *tool __maybe_unused,
 342                                        union perf_event *event __maybe_unused,
 343                                        struct ordered_events *oe __maybe_unused)
 344 {
 345         dump_printf(": unhandled!\n");
 346         return 0;
 347 }
 348
 349 static int process_finished_round(struct perf_tool *tool,
 350                                   union perf_event *event,
 351                                   struct ordered_events *oe);
 352
 353 static int skipn(int fd, off_t n)
 354 {
 355         char buf[4096];
 356         ssize_t ret;
 357
 358         while (n > 0) {
 359                 ret = read(fd, buf, min(n, (off_t)sizeof(buf)));
 360                 if (ret <= 0)
 361                         return ret;
 362                 n -= ret;
 363         }
 364
 365         return 0;
 366 }
 367
 368 static s64 process_event_auxtrace_stub(struct perf_session *session __maybe_unused,
 369                                        union perf_event *event)
 370 {
 371         dump_printf(": unhandled!\n");
 372         if (perf_data__is_pipe(session->data))
 373                 skipn(perf_data__fd(session->data), event->auxtrace.size);
 374         return event->auxtrace.size;
 375 }
 376
 377 static int process_event_op2_stub(struct perf_session *session __maybe_unused,
 378                                   union perf_event *event __maybe_unused)
 379 {
 380         dump_printf(": unhandled!\n");
 381         return 0;
 382 }
 383
 384
 385 static
 386 int process_event_thread_map_stub(struct perf_session *session __maybe_unused,
 387                                   union perf_event *event __maybe_unused)
 388 {
 389         if (dump_trace)
 390                 perf_event__fprintf_thread_map(event, stdout);
 391
 392         dump_printf(": unhandled!\n");
 393         return 0;
 394 }
 395
 396 static
 397 int process_event_cpu_map_stub(struct perf_session *session __maybe_unused,
 398                                union perf_event *event __maybe_unused)
 399 {
 400         if (dump_trace)
 401                 perf_event__fprintf_cpu_map(event, stdout);
 402
 403         dump_printf(": unhandled!\n");
 404         return 0;
 405 }
 406
 407 static
 408 int process_event_stat_config_stub(struct perf_session *session __maybe_unused,
 409                                    union perf_event *event __maybe_unused)
 410 {
 411         if (dump_trace)
 412                 perf_event__fprintf_stat_config(event, stdout);
 413
 414         dump_printf(": unhandled!\n");
 415         return 0;
 416 }
 417
 418 static int process_stat_stub(struct perf_session *perf_session __maybe_unused,
 419                              union perf_event *event)
 420 {
 421         if (dump_trace)
 422                 perf_event__fprintf_stat(event, stdout);
 423
 424         dump_printf(": unhandled!\n");
 425         return 0;
 426 }
 427
 428 static int process_stat_round_stub(struct perf_session *perf_session __maybe_unused,
 429                                    union perf_event *event)
 430 {
 431         if (dump_trace)
 432                 perf_event__fprintf_stat_round(event, stdout);
 433
 434         dump_printf(": unhandled!\n");
 435         return 0;
 436 }
 437
 438 static int perf_session__process_compressed_event_stub(struct perf_session *session __maybe_unused,
 439                                                        union perf_event *event __maybe_unused,
 440                                                        u64 file_offset __maybe_unused)
 441 {
 442        dump_printf(": unhandled!\n");
 443        return 0;
 444 }
 445
 446 void perf_tool__fill_defaults(struct perf_tool *tool)
 447 {
 448         if (tool->sample == NULL)
 449                 tool->sample = process_event_sample_stub;
 450         if (tool->mmap == NULL)
 451                 tool->mmap = process_event_stub;
 452         if (tool->mmap2 == NULL)
 453                 tool->mmap2 = process_event_stub;
 454         if (tool->comm == NULL)
 455                 tool->comm = process_event_stub;
 456         if (tool->namespaces == NULL)
 457                 tool->namespaces = process_event_stub;
 458         if (tool->fork == NULL)
 459                 tool->fork = process_event_stub;
 460         if (tool->exit == NULL)
 461                 tool->exit = process_event_stub;
 462         if (tool->lost == NULL)
 463                 tool->lost = perf_event__process_lost;
 464         if (tool->lost_samples == NULL)
 465                 tool->lost_samples = perf_event__process_lost_samples;
 466         if (tool->aux == NULL)
 467                 tool->aux = perf_event__process_aux;
 468         if (tool->itrace_start == NULL)
 469                 tool->itrace_start = perf_event__process_itrace_start;
 470         if (tool->context_switch == NULL)
 471                 tool->context_switch = perf_event__process_switch;
 472         if (tool->ksymbol == NULL)
 473                 tool->ksymbol = perf_event__process_ksymbol;
 474         if (tool->bpf_event == NULL)
 475                 tool->bpf_event = perf_event__process_bpf_event;
 476         if (tool->read == NULL)
 477                 tool->read = process_event_sample_stub;
 478         if (tool->throttle == NULL)
 479                 tool->throttle = process_event_stub;
 480         if (tool->unthrottle == NULL)
 481                 tool->unthrottle = process_event_stub;
 482         if (tool->attr == NULL)
 483                 tool->attr = process_event_synth_attr_stub;
 484         if (tool->event_update == NULL)
 485                 tool->event_update = process_event_synth_event_update_stub;
 486         if (tool->tracing_data == NULL)
 487                 tool->tracing_data = process_event_synth_tracing_data_stub;
 488         if (tool->build_id == NULL)
 489                 tool->build_id = process_event_op2_stub;
 490         if (tool->finished_round == NULL) {
 491                 if (tool->ordered_events)
 492                         tool->finished_round = process_finished_round;
 493                 else
 494                         tool->finished_round = process_finished_round_stub;
 495         }
 496         if (tool->id_index == NULL)
 497                 tool->id_index = process_event_op2_stub;
 498         if (tool->auxtrace_info == NULL)
 499                 tool->auxtrace_info = process_event_op2_stub;
 500         if (tool->auxtrace == NULL)
 501                 tool->auxtrace = process_event_auxtrace_stub;
 502         if (tool->auxtrace_error == NULL)
 503                 tool->auxtrace_error = process_event_op2_stub;
 504         if (tool->thread_map == NULL)
 505                 tool->thread_map = process_event_thread_map_stub;
 506         if (tool->cpu_map == NULL)
 507                 tool->cpu_map = process_event_cpu_map_stub;
 508         if (tool->stat_config == NULL)
 509                 tool->stat_config = process_event_stat_config_stub;
 510         if (tool->stat == NULL)
 511                 tool->stat = process_stat_stub;
 512         if (tool->stat_round == NULL)
 513                 tool->stat_round = process_stat_round_stub;
 514         if (tool->time_conv == NULL)
 515                 tool->time_conv = process_event_op2_stub;
 516         if (tool->feature == NULL)
 517                 tool->feature = process_event_op2_stub;
 518         if (tool->compressed == NULL)
 519                 tool->compressed = perf_session__process_compressed_event;
 520 }
 521
 522 static void swap_sample_id_all(union perf_event *event, void *data)
 523 {
 524         void *end = (void *) event + event->header.size;
 525         int size = end - data;
 526
 527         BUG_ON(size % sizeof(u64));
 528         mem_bswap_64(data, size);
 529 }
 530
 531 static void perf_event__all64_swap(union perf_event *event,
 532                                    bool sample_id_all __maybe_unused)
 533 {
 534         struct perf_event_header *hdr = &event->header;
 535         mem_bswap_64(hdr + 1, event->header.size - sizeof(*hdr));
 536 }
 537
 538 static void perf_event__comm_swap(union perf_event *event, bool sample_id_all)
 539 {
 540         event->comm.pid = bswap_32(event->comm.pid);
 541         event->comm.tid = bswap_32(event->comm.tid);
 542
 543         if (sample_id_all) {
 544                 void *data = &event->comm.comm;
 545
 546                 data += PERF_ALIGN(strlen(data) + 1, sizeof(u64));
 547                 swap_sample_id_all(event, data);
 548         }
 549 }
 550
 551 static void perf_event__mmap_swap(union perf_event *event,
 552                                   bool sample_id_all)
 553 {
 554         event->mmap.pid   = bswap_32(event->mmap.pid);
 555         event->mmap.tid   = bswap_32(event->mmap.tid);
 556         event->mmap.start = bswap_64(event->mmap.start);
 557         event->mmap.len   = bswap_64(event->mmap.len);
 558         event->mmap.pgoff = bswap_64(event->mmap.pgoff);
 559
 560         if (sample_id_all) {
 561                 void *data = &event->mmap.filename;
 562
 563                 data += PERF_ALIGN(strlen(data) + 1, sizeof(u64));
 564                 swap_sample_id_all(event, data);
 565         }
 566 }
 567
 568 static void perf_event__mmap2_swap(union perf_event *event,
 569                                   bool sample_id_all)
 570 {
 571         event->mmap2.pid   = bswap_32(event->mmap2.pid);
 572         event->mmap2.tid   = bswap_32(event->mmap2.tid);
 573         event->mmap2.start = bswap_64(event->mmap2.start);
 574         event->mmap2.len   = bswap_64(event->mmap2.len);
 575         event->mmap2.pgoff = bswap_64(event->mmap2.pgoff);
 576         event->mmap2.maj   = bswap_32(event->mmap2.maj);
 577         event->mmap2.min   = bswap_32(event->mmap2.min);
 578         event->mmap2.ino   = bswap_64(event->mmap2.ino);
 579
 580         if (sample_id_all) {
 581                 void *data = &event->mmap2.filename;
 582
 583                 data += PERF_ALIGN(strlen(data) + 1, sizeof(u64));
 584                 swap_sample_id_all(event, data);
 585         }
 586 }
 587 static void perf_event__task_swap(union perf_event *event, bool sample_id_all)
 588 {
 589         event->fork.pid  = bswap_32(event->fork.pid);
 590         event->fork.tid  = bswap_32(event->fork.tid);
 591         event->fork.ppid = bswap_32(event->fork.ppid);
 592         event->fork.ptid = bswap_32(event->fork.ptid);
 593         event->fork.time = bswap_64(event->fork.time);
 594
 595         if (sample_id_all)
 596                 swap_sample_id_all(event, &event->fork + 1);
 597 }
 598
 599 static void perf_event__read_swap(union perf_event *event, bool sample_id_all)
 600 {
 601         event->read.pid          = bswap_32(event->read.pid);
 602         event->read.tid          = bswap_32(event->read.tid);
 603         event->read.value        = bswap_64(event->read.value);
 604         event->read.time_enabled = bswap_64(event->read.time_enabled);
 605         event->read.time_running = bswap_64(event->read.time_running);
 606         event->read.id           = bswap_64(event->read.id);
 607
 608         if (sample_id_all)
 609                 swap_sample_id_all(event, &event->read + 1);
 610 }
 611
 612 static void perf_event__aux_swap(union perf_event *event, bool sample_id_all)
 613 {
 614         event->aux.aux_offset = bswap_64(event->aux.aux_offset);
 615         event->aux.aux_size   = bswap_64(event->aux.aux_size);
 616         event->aux.flags      = bswap_64(event->aux.flags);
 617
 618         if (sample_id_all)
 619                 swap_sample_id_all(event, &event->aux + 1);
 620 }
 621
 622 static void perf_event__itrace_start_swap(union perf_event *event,
 623                                           bool sample_id_all)
 624 {
 625         event->itrace_start.pid  = bswap_32(event->itrace_start.pid);
 626         event->itrace_start.tid  = bswap_32(event->itrace_start.tid);
 627
 628         if (sample_id_all)
 629                 swap_sample_id_all(event, &event->itrace_start + 1);
 630 }
 631
 632 static void perf_event__switch_swap(union perf_event *event, bool sample_id_all)
 633 {
 634         if (event->header.type == PERF_RECORD_SWITCH_CPU_WIDE) {
 635                 event->context_switch.next_prev_pid =
 636                                 bswap_32(event->context_switch.next_prev_pid);
 637                 event->context_switch.next_prev_tid =
 638                                 bswap_32(event->context_switch.next_prev_tid);
 639         }
 640
 641         if (sample_id_all)
 642                 swap_sample_id_all(event, &event->context_switch + 1);
 643 }
 644
 645 static void perf_event__throttle_swap(union perf_event *event,
 646                                       bool sample_id_all)
 647 {
 648         event->throttle.time      = bswap_64(event->throttle.time);
 649         event->throttle.id        = bswap_64(event->throttle.id);
 650         event->throttle.stream_id = bswap_64(event->throttle.stream_id);
 651
 652         if (sample_id_all)
 653                 swap_sample_id_all(event, &event->throttle + 1);
 654 }
 655
 656 static void perf_event__namespaces_swap(union perf_event *event,
 657                                         bool sample_id_all)
 658 {
 659         u64 i;
 660
 661         event->namespaces.pid           = bswap_32(event->namespaces.pid);
 662         event->namespaces.tid           = bswap_32(event->namespaces.tid);
 663         event->namespaces.nr_namespaces = bswap_64(event->namespaces.nr_namespaces);
 664
 665         for (i = 0; i < event->namespaces.nr_namespaces; i++) {
 666                 struct perf_ns_link_info *ns = &event->namespaces.link_info[i];
 667
 668                 ns->dev = bswap_64(ns->dev);
 669                 ns->ino = bswap_64(ns->ino);
 670         }
 671
 672         if (sample_id_all)
 673                 swap_sample_id_all(event, &event->namespaces.link_info[i]);
 674 }
 675
 676 static u8 revbyte(u8 b)
 677 {
 678         int rev = (b >> 4) | ((b & 0xf) << 4);
 679         rev = ((rev & 0xcc) >> 2) | ((rev & 0x33) << 2);
 680         rev = ((rev & 0xaa) >> 1) | ((rev & 0x55) << 1);
 681         return (u8) rev;
 682 }
 683
 684 /*
 685  * XXX this is hack in attempt to carry flags bitfield
 686  * through endian village. ABI says:
 687  *
 688  * Bit-fields are allocated from right to left (least to most significant)
 689  * on little-endian implementations and from left to right (most to least
 690  * significant) on big-endian implementations.
 691  *
 692  * The above seems to be byte specific, so we need to reverse each
 693  * byte of the bitfield. 'Internet' also says this might be implementation
 694  * specific and we probably need proper fix and carry perf_event_attr
 695  * bitfield flags in separate data file FEAT_ section. Thought this seems
 696  * to work for now.
 697  */
 698 static void swap_bitfield(u8 *p, unsigned len)
 699 {
 700         unsigned i;
 701
 702         for (i = 0; i < len; i++) {
 703                 *p = revbyte(*p);
 704                 p++;
 705         }
 706 }
 707
 708 /* exported for swapping attributes in file header */
 709 void perf_event__attr_swap(struct perf_event_attr *attr)
 710 {
 711         attr->type              = bswap_32(attr->type);
 712         attr->size              = bswap_32(attr->size);
 713
 714 #define bswap_safe(f, n)                                        \
 715         (attr->size > (offsetof(struct perf_event_attr, f) +    \
 716                        sizeof(attr->f) * (n)))
 717 #define bswap_field(f, sz)                      \
 718 do {                                            \
 719         if (bswap_safe(f, 0))                   \
 720                 attr->f = bswap_##sz(attr->f);  \
 721 } while(0)
 722 #define bswap_field_16(f) bswap_field(f, 16)
 723 #define bswap_field_32(f) bswap_field(f, 32)
 724 #define bswap_field_64(f) bswap_field(f, 64)
 725
 726         bswap_field_64(config);
 727         bswap_field_64(sample_period);
 728         bswap_field_64(sample_type);
 729         bswap_field_64(read_format);
 730         bswap_field_32(wakeup_events);
 731         bswap_field_32(bp_type);
 732         bswap_field_64(bp_addr);
 733         bswap_field_64(bp_len);
 734         bswap_field_64(branch_sample_type);
 735         bswap_field_64(sample_regs_user);
 736         bswap_field_32(sample_stack_user);
 737         bswap_field_32(aux_watermark);
 738         bswap_field_16(sample_max_stack);
 739
 740         /*
 741          * After read_format are bitfields. Check read_format because
 742          * we are unable to use offsetof on bitfield.
 743          */
 744         if (bswap_safe(read_format, 1))
 745                 swap_bitfield((u8 *) (&attr->read_format + 1),
 746                               sizeof(u64));
 747 #undef bswap_field_64
 748 #undef bswap_field_32
 749 #undef bswap_field
 750 #undef bswap_safe
 751 }
 752
 753 static void perf_event__hdr_attr_swap(union perf_event *event,
 754                                       bool sample_id_all __maybe_unused)
 755 {
 756         size_t size;
 757
 758         perf_event__attr_swap(&event->attr.attr);
 759
 760         size = event->header.size;
 761         size -= (void *)&event->attr.id - (void *)event;
 762         mem_bswap_64(event->attr.id, size);
 763 }
 764
 765 static void perf_event__event_update_swap(union perf_event *event,
 766                                           bool sample_id_all __maybe_unused)
 767 {
 768         event->event_update.type = bswap_64(event->event_update.type);
 769         event->event_update.id   = bswap_64(event->event_update.id);
 770 }
 771
 772 static void perf_event__event_type_swap(union perf_event *event,
 773                                         bool sample_id_all __maybe_unused)
 774 {
 775         event->event_type.event_type.event_id =
 776                 bswap_64(event->event_type.event_type.event_id);
 777 }
 778
 779 static void perf_event__tracing_data_swap(union perf_event *event,
 780                                           bool sample_id_all __maybe_unused)
 781 {
 782         event->tracing_data.size = bswap_32(event->tracing_data.size);
 783 }
 784
 785 static void perf_event__auxtrace_info_swap(union perf_event *event,
 786                                            bool sample_id_all __maybe_unused)
 787 {
 788         size_t size;
 789
 790         event->auxtrace_info.type = bswap_32(event->auxtrace_info.type);
 791
 792         size = event->header.size;
 793         size -= (void *)&event->auxtrace_info.priv - (void *)event;
 794         mem_bswap_64(event->auxtrace_info.priv, size);
 795 }
 796
 797 static void perf_event__auxtrace_swap(union perf_event *event,
 798                                       bool sample_id_all __maybe_unused)
 799 {
 800         event->auxtrace.size      = bswap_64(event->auxtrace.size);
 801         event->auxtrace.offset    = bswap_64(event->auxtrace.offset);
 802         event->auxtrace.reference = bswap_64(event->auxtrace.reference);
 803         event->auxtrace.idx       = bswap_32(event->auxtrace.idx);
 804         event->auxtrace.tid       = bswap_32(event->auxtrace.tid);
 805         event->auxtrace.cpu       = bswap_32(event->auxtrace.cpu);
 806 }
 807
 808 static void perf_event__auxtrace_error_swap(union perf_event *event,
 809                                             bool sample_id_all __maybe_unused)
 810 {
 811         event->auxtrace_error.type = bswap_32(event->auxtrace_error.type);
 812         event->auxtrace_error.code = bswap_32(event->auxtrace_error.code);
 813         event->auxtrace_error.cpu  = bswap_32(event->auxtrace_error.cpu);
 814         event->auxtrace_error.pid  = bswap_32(event->auxtrace_error.pid);
 815         event->auxtrace_error.tid  = bswap_32(event->auxtrace_error.tid);
 816         event->auxtrace_error.fmt  = bswap_32(event->auxtrace_error.fmt);
 817         event->auxtrace_error.ip   = bswap_64(event->auxtrace_error.ip);
 818         if (event->auxtrace_error.fmt)
 819                 event->auxtrace_error.time = bswap_64(event->auxtrace_error.time);
 820 }
 821
 822 static void perf_event__thread_map_swap(union perf_event *event,
 823                                         bool sample_id_all __maybe_unused)
 824 {
 825         unsigned i;
 826
 827         event->thread_map.nr = bswap_64(event->thread_map.nr);
 828
 829         for (i = 0; i < event->thread_map.nr; i++)
 830                 event->thread_map.entries[i].pid = bswap_64(event->thread_map.entries[i].pid);
 831 }
 832
 833 static void perf_event__cpu_map_swap(union perf_event *event,
 834                                      bool sample_id_all __maybe_unused)
 835 {
 836         struct cpu_map_data *data = &event->cpu_map.data;
 837         struct cpu_map_entries *cpus;
 838         struct cpu_map_mask *mask;
 839         unsigned i;
 840
 841         data->type = bswap_64(data->type);
 842
 843         switch (data->type) {
 844         case PERF_CPU_MAP__CPUS:
 845                 cpus = (struct cpu_map_entries *)data->data;
 846
 847                 cpus->nr = bswap_16(cpus->nr);
 848
 849                 for (i = 0; i < cpus->nr; i++)
 850                         cpus->cpu[i] = bswap_16(cpus->cpu[i]);
 851                 break;
 852         case PERF_CPU_MAP__MASK:
 853                 mask = (struct cpu_map_mask *) data->data;
 854
 855                 mask->nr = bswap_16(mask->nr);
 856                 mask->long_size = bswap_16(mask->long_size);
 857
 858                 switch (mask->long_size) {
 859                 case 4: mem_bswap_32(&mask->mask, mask->nr); break;
 860                 case 8: mem_bswap_64(&mask->mask, mask->nr); break;
 861                 default:
 862                         pr_err("cpu_map swap: unsupported long size\n");
 863                 }
 864         default:
 865                 break;
 866         }
 867 }
 868
 869 static void perf_event__stat_config_swap(union perf_event *event,
 870                                          bool sample_id_all __maybe_unused)
 871 {
 872         u64 size;
 873
 874         size  = event->stat_config.nr * sizeof(event->stat_config.data[0]);
 875         size += 1; /* nr item itself */
 876         mem_bswap_64(&event->stat_config.nr, size);
 877 }
 878
 879 static void perf_event__stat_swap(union perf_event *event,
 880                                   bool sample_id_all __maybe_unused)
 881 {
 882         event->stat.id     = bswap_64(event->stat.id);
 883         event->stat.thread = bswap_32(event->stat.thread);
 884         event->stat.cpu    = bswap_32(event->stat.cpu);
 885         event->stat.val    = bswap_64(event->stat.val);
 886         event->stat.ena    = bswap_64(event->stat.ena);
 887         event->stat.run    = bswap_64(event->stat.run);
 888 }
 889
 890 static void perf_event__stat_round_swap(union perf_event *event,
 891                                         bool sample_id_all __maybe_unused)
 892 {
 893         event->stat_round.type = bswap_64(event->stat_round.type);
 894         event->stat_round.time = bswap_64(event->stat_round.time);
 895 }
 896
 897 typedef void (*perf_event__swap_op)(union perf_event *event,
 898                                     bool sample_id_all);
 899
 900 static perf_event__swap_op perf_event__swap_ops[] = {
 901         [PERF_RECORD_MMAP]                = perf_event__mmap_swap,
 902         [PERF_RECORD_MMAP2]               = perf_event__mmap2_swap,
 903         [PERF_RECORD_COMM]                = perf_event__comm_swap,
 904         [PERF_RECORD_FORK]                = perf_event__task_swap,
 905         [PERF_RECORD_EXIT]                = perf_event__task_swap,
 906         [PERF_RECORD_LOST]                = perf_event__all64_swap,
 907         [PERF_RECORD_READ]                = perf_event__read_swap,
 908         [PERF_RECORD_THROTTLE]            = perf_event__throttle_swap,
 909         [PERF_RECORD_UNTHROTTLE]          = perf_event__throttle_swap,
 910         [PERF_RECORD_SAMPLE]              = perf_event__all64_swap,
 911         [PERF_RECORD_AUX]                 = perf_event__aux_swap,
 912         [PERF_RECORD_ITRACE_START]        = perf_event__itrace_start_swap,
 913         [PERF_RECORD_LOST_SAMPLES]        = perf_event__all64_swap,
 914         [PERF_RECORD_SWITCH]              = perf_event__switch_swap,
 915         [PERF_RECORD_SWITCH_CPU_WIDE]     = perf_event__switch_swap,
 916         [PERF_RECORD_NAMESPACES]          = perf_event__namespaces_swap,
 917         [PERF_RECORD_HEADER_ATTR]         = perf_event__hdr_attr_swap,
 918         [PERF_RECORD_HEADER_EVENT_TYPE]   = perf_event__event_type_swap,
 919         [PERF_RECORD_HEADER_TRACING_DATA] = perf_event__tracing_data_swap,
 920         [PERF_RECORD_HEADER_BUILD_ID]     = NULL,
 921         [PERF_RECORD_ID_INDEX]            = perf_event__all64_swap,
 922         [PERF_RECORD_AUXTRACE_INFO]       = perf_event__auxtrace_info_swap,
 923         [PERF_RECORD_AUXTRACE]            = perf_event__auxtrace_swap,
 924         [PERF_RECORD_AUXTRACE_ERROR]      = perf_event__auxtrace_error_swap,
 925         [PERF_RECORD_THREAD_MAP]          = perf_event__thread_map_swap,
 926         [PERF_RECORD_CPU_MAP]             = perf_event__cpu_map_swap,
 927         [PERF_RECORD_STAT_CONFIG]         = perf_event__stat_config_swap,
 928         [PERF_RECORD_STAT]                = perf_event__stat_swap,
 929         [PERF_RECORD_STAT_ROUND]          = perf_event__stat_round_swap,
 930         [PERF_RECORD_EVENT_UPDATE]        = perf_event__event_update_swap,
 931         [PERF_RECORD_TIME_CONV]           = perf_event__all64_swap,
 932         [PERF_RECORD_HEADER_MAX]          = NULL,
 933 };
 934
 935 /*
 936  * When perf record finishes a pass on every buffers, it records this pseudo
 937  * event.
 938  * We record the max timestamp t found in the pass n.
 939  * Assuming these timestamps are monotonic across cpus, we know that if
 940  * a buffer still has events with timestamps below t, they will be all
 941  * available and then read in the pass n + 1.
 942  * Hence when we start to read the pass n + 2, we can safely flush every
 943  * events with timestamps below t.
 944  *
 945  *    ============ PASS n =================
 946  *       CPU 0         |   CPU 1
 947  *                     |
 948  *    cnt1 timestamps  |   cnt2 timestamps
 949  *          1          |         2
 950  *          2          |         3
 951  *          -          |         4  <--- max recorded
 952  *
 953  *    ============ PASS n + 1 ==============
 954  *       CPU 0         |   CPU 1
 955  *                     |
 956  *    cnt1 timestamps  |   cnt2 timestamps
 957  *          3          |         5
 958  *          4          |         6
 959  *          5          |         7 <---- max recorded
 960  *
 961  *      Flush every events below timestamp 4
 962  *
 963  *    ============ PASS n + 2 ==============
 964  *       CPU 0         |   CPU 1
 965  *                     |
 966  *    cnt1 timestamps  |   cnt2 timestamps
 967  *          6          |         8
 968  *          7          |         9
 969  *          -          |         10
 970  *
 971  *      Flush every events below timestamp 7
 972  *      etc...
 973  */
 974 static int process_finished_round(struct perf_tool *tool __maybe_unused,
 975                                   union perf_event *event __maybe_unused,
 976                                   struct ordered_events *oe)
 977 {
 978         if (dump_trace)
 979                 fprintf(stdout, "\n");
 980         return ordered_events__flush(oe, OE_FLUSH__ROUND);
 981 }
 982
 983 int perf_session__queue_event(struct perf_session *s, union perf_event *event,
 984                               u64 timestamp, u64 file_offset)
 985 {
 986         return ordered_events__queue(&s->ordered_events, event, timestamp, file_offset);
 987 }
 988
 989 static void callchain__lbr_callstack_printf(struct perf_sample *sample)
 990 {
 991         struct ip_callchain *callchain = sample->callchain;
 992         struct branch_stack *lbr_stack = sample->branch_stack;
 993         u64 kernel_callchain_nr = callchain->nr;
 994         unsigned int i;
 995
 996         for (i = 0; i < kernel_callchain_nr; i++) {
 997                 if (callchain->ips[i] == PERF_CONTEXT_USER)
 998                         break;
 999         }
1000
1001         if ((i != kernel_callchain_nr) && lbr_stack->nr) {
1002                 u64 total_nr;
1003                 /*
1004                  * LBR callstack can only get user call chain,
1005                  * i is kernel call chain number,
1006                  * 1 is PERF_CONTEXT_USER.
1007                  *
1008                  * The user call chain is stored in LBR registers.
1009                  * LBR are pair registers. The caller is stored
1010                  * in "from" register, while the callee is stored
1011                  * in "to" register.
1012                  * For example, there is a call stack
1013                  * "A"->"B"->"C"->"D".
1014                  * The LBR registers will recorde like
1015                  * "C"->"D", "B"->"C", "A"->"B".
1016                  * So only the first "to" register and all "from"
1017                  * registers are needed to construct the whole stack.
1018                  */
1019                 total_nr = i + 1 + lbr_stack->nr + 1;
1020                 kernel_callchain_nr = i + 1;
1021
1022                 printf("... LBR call chain: nr:%" PRIu64 "\n", total_nr);
1023
1024                 for (i = 0; i < kernel_callchain_nr; i++)
1025                         printf("..... %2d: %016" PRIx64 "\n",
1026                                i, callchain->ips[i]);
1027
1028                 printf("..... %2d: %016" PRIx64 "\n",
1029                        (int)(kernel_callchain_nr), lbr_stack->entries[0].to);
1030                 for (i = 0; i < lbr_stack->nr; i++)
1031                         printf("..... %2d: %016" PRIx64 "\n",
1032                                (int)(i + kernel_callchain_nr + 1), lbr_stack->entries[i].from);
1033         }
1034 }
1035
1036 static void callchain__printf(struct perf_evsel *evsel,
1037                               struct perf_sample *sample)
1038 {
1039         unsigned int i;
1040         struct ip_callchain *callchain = sample->callchain;
1041
1042         if (perf_evsel__has_branch_callstack(evsel))
1043                 callchain__lbr_callstack_printf(sample);
1044
1045         printf("... FP chain: nr:%" PRIu64 "\n", callchain->nr);
1046
1047         for (i = 0; i < callchain->nr; i++)
1048                 printf("..... %2d: %016" PRIx64 "\n",
1049                        i, callchain->ips[i]);
1050 }
1051
1052 static void branch_stack__printf(struct perf_sample *sample)
1053 {
1054         uint64_t i;
1055
1056         printf("... branch stack: nr:%" PRIu64 "\n", sample->branch_stack->nr);
1057
1058         for (i = 0; i < sample->branch_stack->nr; i++) {
1059                 struct branch_entry *e = &sample->branch_stack->entries[i];
1060
1061                 printf("..... %2"PRIu64": %016" PRIx64 " -> %016" PRIx64 " %hu cycles %s%s%s%s %x\n",
1062                         i, e->from, e->to,
1063                         (unsigned short)e->flags.cycles,
1064                         e->flags.mispred ? "M" : " ",
1065                         e->flags.predicted ? "P" : " ",
1066                         e->flags.abort ? "A" : " ",
1067                         e->flags.in_tx ? "T" : " ",
1068                         (unsigned)e->flags.reserved);
1069         }
1070 }
1071
1072 static void regs_dump__printf(u64 mask, u64 *regs)
1073 {
1074         unsigned rid, i = 0;
1075
1076         for_each_set_bit(rid, (unsigned long *) &mask, sizeof(mask) * 8) {
1077                 u64 val = regs[i++];
1078
1079                 printf(".... %-5s 0x%" PRIx64 "\n",
1080                        perf_reg_name(rid), val);
1081         }
1082 }
1083
1084 static const char *regs_abi[] = {
1085         [PERF_SAMPLE_REGS_ABI_NONE] = "none",
1086         [PERF_SAMPLE_REGS_ABI_32] = "32-bit",
1087         [PERF_SAMPLE_REGS_ABI_64] = "64-bit",
1088 };
1089
1090 static inline const char *regs_dump_abi(struct regs_dump *d)
1091 {
1092         if (d->abi > PERF_SAMPLE_REGS_ABI_64)
1093                 return "unknown";
1094
1095         return regs_abi[d->abi];
1096 }
1097
1098 static void regs__printf(const char *type, struct regs_dump *regs)
1099 {
1100         u64 mask = regs->mask;
1101
1102         printf("... %s regs: mask 0x%" PRIx64 " ABI %s\n",
1103                type,
1104                mask,
1105                regs_dump_abi(regs));
1106
1107         regs_dump__printf(mask, regs->regs);
1108 }
1109
1110 static void regs_user__printf(struct perf_sample *sample)
1111 {
1112         struct regs_dump *user_regs = &sample->user_regs;
1113
1114         if (user_regs->regs)
1115                 regs__printf("user", user_regs);
1116 }
1117
1118 static void regs_intr__printf(struct perf_sample *sample)
1119 {
1120         struct regs_dump *intr_regs = &sample->intr_regs;
1121
1122         if (intr_regs->regs)
1123                 regs__printf("intr", intr_regs);
1124 }
1125
1126 static void stack_user__printf(struct stack_dump *dump)
1127 {
1128         printf("... ustack: size %" PRIu64 ", offset 0x%x\n",
1129                dump->size, dump->offset);
1130 }
1131
1132 static void perf_evlist__print_tstamp(struct perf_evlist *evlist,
1133                                        union perf_event *event,
1134                                        struct perf_sample *sample)
1135 {
1136         u64 sample_type = __perf_evlist__combined_sample_type(evlist);
1137
1138         if (event->header.type != PERF_RECORD_SAMPLE &&
1139             !perf_evlist__sample_id_all(evlist)) {
1140                 fputs("-1 -1 ", stdout);
1141                 return;
1142         }
1143
1144         if ((sample_type & PERF_SAMPLE_CPU))
1145                 printf("%u ", sample->cpu);
1146
1147         if (sample_type & PERF_SAMPLE_TIME)
1148                 printf("%" PRIu64 " ", sample->time);
1149 }
1150
1151 static void sample_read__printf(struct perf_sample *sample, u64 read_format)
1152 {
1153         printf("... sample_read:\n");
1154
1155         if (read_format & PERF_FORMAT_TOTAL_TIME_ENABLED)
1156                 printf("...... time enabled %016" PRIx64 "\n",
1157                        sample->read.time_enabled);
1158
1159         if (read_format & PERF_FORMAT_TOTAL_TIME_RUNNING)
1160                 printf("...... time running %016" PRIx64 "\n",
1161                        sample->read.time_running);
1162
1163         if (read_format & PERF_FORMAT_GROUP) {
1164                 u64 i;
1165
1166                 printf(".... group nr %" PRIu64 "\n", sample->read.group.nr);
1167
1168                 for (i = 0; i < sample->read.group.nr; i++) {
1169                         struct sample_read_value *value;
1170
1171                         value = &sample->read.group.values[i];
1172                         printf("..... id %016" PRIx64
1173                                ", value %016" PRIx64 "\n",
1174                                value->id, value->value);
1175                 }
1176         } else
1177                 printf("..... id %016" PRIx64 ", value %016" PRIx64 "\n",
1178                         sample->read.one.id, sample->read.one.value);
1179 }
1180
1181 static void dump_event(struct perf_evlist *evlist, union perf_event *event,
1182                        u64 file_offset, struct perf_sample *sample)
1183 {
1184         if (!dump_trace)
1185                 return;
1186
1187         printf("\n%#" PRIx64 " [%#x]: event: %d\n",
1188                file_offset, event->header.size, event->header.type);
1189
1190         trace_event(event);
1191         if (event->header.type == PERF_RECORD_SAMPLE && evlist->trace_event_sample_raw)
1192                 evlist->trace_event_sample_raw(evlist, event, sample);
1193
1194         if (sample)
1195                 perf_evlist__print_tstamp(evlist, event, sample);
1196
1197         printf("%#" PRIx64 " [%#x]: PERF_RECORD_%s", file_offset,
1198                event->header.size, perf_event__name(event->header.type));
1199 }
1200
1201 static void dump_sample(struct perf_evsel *evsel, union perf_event *event,
1202                         struct perf_sample *sample)
1203 {
1204         u64 sample_type;
1205
1206         if (!dump_trace)
1207                 return;
1208
1209         printf("(IP, 0x%x): %d/%d: %#" PRIx64 " period: %" PRIu64 " addr: %#" PRIx64 "\n",
1210                event->header.misc, sample->pid, sample->tid, sample->ip,
1211                sample->period, sample->addr);
1212
1213         sample_type = evsel->attr.sample_type;
1214
1215         if (evsel__has_callchain(evsel))
1216                 callchain__printf(evsel, sample);
1217
1218         if ((sample_type & PERF_SAMPLE_BRANCH_STACK) && !perf_evsel__has_branch_callstack(evsel))
1219                 branch_stack__printf(sample);
1220
1221         if (sample_type & PERF_SAMPLE_REGS_USER)
1222                 regs_user__printf(sample);
1223
1224         if (sample_type & PERF_SAMPLE_REGS_INTR)
1225                 regs_intr__printf(sample);
1226
1227         if (sample_type & PERF_SAMPLE_STACK_USER)
1228                 stack_user__printf(&sample->user_stack);
1229
1230         if (sample_type & PERF_SAMPLE_WEIGHT)
1231                 printf("... weight: %" PRIu64 "\n", sample->weight);
1232
1233         if (sample_type & PERF_SAMPLE_DATA_SRC)
1234                 printf(" . data_src: 0x%"PRIx64"\n", sample->data_src);
1235
1236         if (sample_type & PERF_SAMPLE_PHYS_ADDR)
1237                 printf(" .. phys_addr: 0x%"PRIx64"\n", sample->phys_addr);
1238
1239         if (sample_type & PERF_SAMPLE_TRANSACTION)
1240                 printf("... transaction: %" PRIx64 "\n", sample->transaction);
1241
1242         if (sample_type & PERF_SAMPLE_READ)
1243                 sample_read__printf(sample, evsel->attr.read_format);
1244 }
1245
1246 static void dump_read(struct perf_evsel *evsel, union perf_event *event)
1247 {
1248         struct read_event *read_event = &event->read;
1249         u64 read_format;
1250
1251         if (!dump_trace)
1252                 return;
1253
1254         printf(": %d %d %s %" PRIu64 "\n", event->read.pid, event->read.tid,
1255                perf_evsel__name(evsel),
1256                event->read.value);
1257
1258         if (!evsel)
1259                 return;
1260
1261         read_format = evsel->attr.read_format;
1262
1263         if (read_format & PERF_FORMAT_TOTAL_TIME_ENABLED)
1264                 printf("... time enabled : %" PRIu64 "\n", read_event->time_enabled);
1265
1266         if (read_format & PERF_FORMAT_TOTAL_TIME_RUNNING)
1267                 printf("... time running : %" PRIu64 "\n", read_event->time_running);
1268
1269         if (read_format & PERF_FORMAT_ID)
1270                 printf("... id           : %" PRIu64 "\n", read_event->id);
1271 }
1272
1273 static struct machine *machines__find_for_cpumode(struct machines *machines,
1274                                                union perf_event *event,
1275                                                struct perf_sample *sample)
1276 {
1277         struct machine *machine;
1278
1279         if (perf_guest &&
1280             ((sample->cpumode == PERF_RECORD_MISC_GUEST_KERNEL) ||
1281              (sample->cpumode == PERF_RECORD_MISC_GUEST_USER))) {
1282                 u32 pid;
1283
1284                 if (event->header.type == PERF_RECORD_MMAP
1285                     || event->header.type == PERF_RECORD_MMAP2)
1286                         pid = event->mmap.pid;
1287                 else
1288                         pid = sample->pid;
1289
1290                 machine = machines__find(machines, pid);
1291                 if (!machine)
1292                         machine = machines__findnew(machines, DEFAULT_GUEST_KERNEL_ID);
1293                 return machine;
1294         }
1295
1296         return &machines->host;
1297 }
1298
1299 static int deliver_sample_value(struct perf_evlist *evlist,
1300                                 struct perf_tool *tool,
1301                                 union perf_event *event,
1302                                 struct perf_sample *sample,
1303                                 struct sample_read_value *v,
1304                                 struct machine *machine)
1305 {
1306         struct perf_sample_id *sid = perf_evlist__id2sid(evlist, v->id);
1307
1308         if (sid) {
1309                 sample->id     = v->id;
1310                 sample->period = v->value - sid->period;
1311                 sid->period    = v->value;
1312         }
1313
1314         if (!sid || sid->evsel == NULL) {
1315                 ++evlist->stats.nr_unknown_id;
1316                 return 0;
1317         }
1318
1319         /*
1320          * There's no reason to deliver sample
1321          * for zero period, bail out.
1322          */
1323         if (!sample->period)
1324                 return 0;
1325
1326         return tool->sample(tool, event, sample, sid->evsel, machine);
1327 }
1328
1329 static int deliver_sample_group(struct perf_evlist *evlist,
1330                                 struct perf_tool *tool,
1331                                 union  perf_event *event,
1332                                 struct perf_sample *sample,
1333                                 struct machine *machine)
1334 {
1335         int ret = -EINVAL;
1336         u64 i;
1337
1338         for (i = 0; i < sample->read.group.nr; i++) {
1339                 ret = deliver_sample_value(evlist, tool, event, sample,
1340                                            &sample->read.group.values[i],
1341                                            machine);
1342                 if (ret)
1343                         break;
1344         }
1345
1346         return ret;
1347 }
1348
1349 static int
1350  perf_evlist__deliver_sample(struct perf_evlist *evlist,
1351                              struct perf_tool *tool,
1352                              union  perf_event *event,
1353                              struct perf_sample *sample,
1354                              struct perf_evsel *evsel,
1355                              struct machine *machine)
1356 {
1357         /* We know evsel != NULL. */
1358         u64 sample_type = evsel->attr.sample_type;
1359         u64 read_format = evsel->attr.read_format;
1360
1361         /* Standard sample delivery. */
1362         if (!(sample_type & PERF_SAMPLE_READ))
1363                 return tool->sample(tool, event, sample, evsel, machine);
1364
1365         /* For PERF_SAMPLE_READ we have either single or group mode. */
1366         if (read_format & PERF_FORMAT_GROUP)
1367                 return deliver_sample_group(evlist, tool, event, sample,
1368                                             machine);
1369         else
1370                 return deliver_sample_value(evlist, tool, event, sample,
1371                                             &sample->read.one, machine);
1372 }
1373
1374 static int machines__deliver_event(struct machines *machines,
1375                                    struct perf_evlist *evlist,
1376                                    union perf_event *event,
1377                                    struct perf_sample *sample,
1378                                    struct perf_tool *tool, u64 file_offset)
1379 {
1380         struct perf_evsel *evsel;
1381         struct machine *machine;
1382
1383         dump_event(evlist, event, file_offset, sample);
1384
1385         evsel = perf_evlist__id2evsel(evlist, sample->id);
1386
1387         machine = machines__find_for_cpumode(machines, event, sample);
1388
1389         switch (event->header.type) {
1390         case PERF_RECORD_SAMPLE:
1391                 if (evsel == NULL) {
1392                         ++evlist->stats.nr_unknown_id;
1393                         return 0;
1394                 }
1395                 dump_sample(evsel, event, sample);
1396                 if (machine == NULL) {
1397                         ++evlist->stats.nr_unprocessable_samples;
1398                         return 0;
1399                 }
1400                 return perf_evlist__deliver_sample(evlist, tool, event, sample, evsel, machine);
1401         case PERF_RECORD_MMAP:
1402                 return tool->mmap(tool, event, sample, machine);
1403         case PERF_RECORD_MMAP2:
1404                 if (event->header.misc & PERF_RECORD_MISC_PROC_MAP_PARSE_TIMEOUT)
1405                         ++evlist->stats.nr_proc_map_timeout;
1406                 return tool->mmap2(tool, event, sample, machine);
1407         case PERF_RECORD_COMM:
1408                 return tool->comm(tool, event, sample, machine);
1409         case PERF_RECORD_NAMESPACES:
1410                 return tool->namespaces(tool, event, sample, machine);
1411         case PERF_RECORD_FORK:
1412                 return tool->fork(tool, event, sample, machine);
1413         case PERF_RECORD_EXIT:
1414                 return tool->exit(tool, event, sample, machine);
1415         case PERF_RECORD_LOST:
1416                 if (tool->lost == perf_event__process_lost)
1417                         evlist->stats.total_lost += event->lost.lost;
1418                 return tool->lost(tool, event, sample, machine);
1419         case PERF_RECORD_LOST_SAMPLES:
1420                 if (tool->lost_samples == perf_event__process_lost_samples)
1421                         evlist->stats.total_lost_samples += event->lost_samples.lost;
1422                 return tool->lost_samples(tool, event, sample, machine);
1423         case PERF_RECORD_READ:
1424                 dump_read(evsel, event);
1425                 return tool->read(tool, event, sample, evsel, machine);
1426         case PERF_RECORD_THROTTLE:
1427                 return tool->throttle(tool, event, sample, machine);
1428         case PERF_RECORD_UNTHROTTLE:
1429                 return tool->unthrottle(tool, event, sample, machine);
1430         case PERF_RECORD_AUX:
1431                 if (tool->aux == perf_event__process_aux) {
1432                         if (event->aux.flags & PERF_AUX_FLAG_TRUNCATED)
1433                                 evlist->stats.total_aux_lost += 1;
1434                         if (event->aux.flags & PERF_AUX_FLAG_PARTIAL)
1435                                 evlist->stats.total_aux_partial += 1;
1436                 }
1437                 return tool->aux(tool, event, sample, machine);
1438         case PERF_RECORD_ITRACE_START:
1439                 return tool->itrace_start(tool, event, sample, machine);
1440         case PERF_RECORD_SWITCH:
1441         case PERF_RECORD_SWITCH_CPU_WIDE:
1442                 return tool->context_switch(tool, event, sample, machine);
1443         case PERF_RECORD_KSYMBOL:
1444                 return tool->ksymbol(tool, event, sample, machine);
1445         case PERF_RECORD_BPF_EVENT:
1446                 return tool->bpf_event(tool, event, sample, machine);
1447         default:
1448                 ++evlist->stats.nr_unknown_events;
1449                 return -1;
1450         }
1451 }
1452
1453 static int perf_session__deliver_event(struct perf_session *session,
1454                                        union perf_event *event,
1455                                        struct perf_tool *tool,
1456                                        u64 file_offset)
1457 {
1458         struct perf_sample sample;
1459         int ret;
1460
1461         ret = perf_evlist__parse_sample(session->evlist, event, &sample);
1462         if (ret) {
1463                 pr_err("Can't parse sample, err = %d\n", ret);
1464                 return ret;
1465         }
1466
1467         ret = auxtrace__process_event(session, event, &sample, tool);
1468         if (ret < 0)
1469                 return ret;
1470         if (ret > 0)
1471                 return 0;
1472
1473         return machines__deliver_event(&session->machines, session->evlist,
1474                                        event, &sample, tool, file_offset);
1475 }
1476
1477 static s64 perf_session__process_user_event(struct perf_session *session,
1478                                             union perf_event *event,
1479                                             u64 file_offset)
1480 {
1481         struct ordered_events *oe = &session->ordered_events;
1482         struct perf_tool *tool = session->tool;
1483         struct perf_sample sample = { .time = 0, };
1484         int fd = perf_data__fd(session->data);
1485         int err;
1486
1487         if (event->header.type != PERF_RECORD_COMPRESSED ||
1488             tool->compressed == perf_session__process_compressed_event_stub)
1489                 dump_event(session->evlist, event, file_offset, &sample);
1490
1491         /* These events are processed right away */
1492         switch (event->header.type) {
1493         case PERF_RECORD_HEADER_ATTR:
1494                 err = tool->attr(tool, event, &session->evlist);
1495                 if (err == 0) {
1496                         perf_session__set_id_hdr_size(session);
1497                         perf_session__set_comm_exec(session);
1498                 }
1499                 return err;
1500         case PERF_RECORD_EVENT_UPDATE:
1501                 return tool->event_update(tool, event, &session->evlist);
1502         case PERF_RECORD_HEADER_EVENT_TYPE:
1503                 /*
1504                  * Depreceated, but we need to handle it for sake
1505                  * of old data files create in pipe mode.
1506                  */
1507                 return 0;
1508         case PERF_RECORD_HEADER_TRACING_DATA:
1509                 /* setup for reading amidst mmap */
1510                 lseek(fd, file_offset, SEEK_SET);
1511                 return tool->tracing_data(session, event);
1512         case PERF_RECORD_HEADER_BUILD_ID:
1513                 return tool->build_id(session, event);
1514         case PERF_RECORD_FINISHED_ROUND:
1515                 return tool->finished_round(tool, event, oe);
1516         case PERF_RECORD_ID_INDEX:
1517                 return tool->id_index(session, event);
1518         case PERF_RECORD_AUXTRACE_INFO:
1519                 return tool->auxtrace_info(session, event);
1520         case PERF_RECORD_AUXTRACE:
1521                 /* setup for reading amidst mmap */
1522                 lseek(fd, file_offset + event->header.size, SEEK_SET);
1523                 return tool->auxtrace(session, event);
1524         case PERF_RECORD_AUXTRACE_ERROR:
1525                 perf_session__auxtrace_error_inc(session, event);
1526                 return tool->auxtrace_error(session, event);
1527         case PERF_RECORD_THREAD_MAP:
1528                 return tool->thread_map(session, event);
1529         case PERF_RECORD_CPU_MAP:
1530                 return tool->cpu_map(session, event);
1531         case PERF_RECORD_STAT_CONFIG:
1532                 return tool->stat_config(session, event);
1533         case PERF_RECORD_STAT:
1534                 return tool->stat(session, event);
1535         case PERF_RECORD_STAT_ROUND:
1536                 return tool->stat_round(session, event);
1537         case PERF_RECORD_TIME_CONV:
1538                 session->time_conv = event->time_conv;
1539                 return tool->time_conv(session, event);
1540         case PERF_RECORD_HEADER_FEATURE:
1541                 return tool->feature(session, event);
1542         case PERF_RECORD_COMPRESSED:
1543                 err = tool->compressed(session, event, file_offset);
1544                 if (err)
1545                         dump_event(session->evlist, event, file_offset, &sample);
1546                 return err;
1547         default:
1548                 return -EINVAL;
1549         }
1550 }
1551
1552 int perf_session__deliver_synth_event(struct perf_session *session,
1553                                       union perf_event *event,
1554                                       struct perf_sample *sample)
1555 {
1556         struct perf_evlist *evlist = session->evlist;
1557         struct perf_tool *tool = session->tool;
1558
1559         events_stats__inc(&evlist->stats, event->header.type);
1560
1561         if (event->header.type >= PERF_RECORD_USER_TYPE_START)
1562                 return perf_session__process_user_event(session, event, 0);
1563
1564         return machines__deliver_event(&session->machines, evlist, event, sample, tool, 0);
1565 }
1566
1567 static void event_swap(union perf_event *event, bool sample_id_all)
1568 {
1569         perf_event__swap_op swap;
1570
1571         swap = perf_event__swap_ops[event->header.type];
1572         if (swap)
1573                 swap(event, sample_id_all);
1574 }
1575
1576 int perf_session__peek_event(struct perf_session *session, off_t file_offset,
1577                              void *buf, size_t buf_sz,
1578                              union perf_event **event_ptr,
1579                              struct perf_sample *sample)
1580 {
1581         union perf_event *event;
1582         size_t hdr_sz, rest;
1583         int fd;
1584
1585         if (session->one_mmap && !session->header.needs_swap) {
1586                 event = file_offset - session->one_mmap_offset +
1587                         session->one_mmap_addr;
1588                 goto out_parse_sample;
1589         }
1590
1591         if (perf_data__is_pipe(session->data))
1592                 return -1;
1593
1594         fd = perf_data__fd(session->data);
1595         hdr_sz = sizeof(struct perf_event_header);
1596
1597         if (buf_sz < hdr_sz)
1598                 return -1;
1599
1600         if (lseek(fd, file_offset, SEEK_SET) == (off_t)-1 ||
1601             readn(fd, buf, hdr_sz) != (ssize_t)hdr_sz)
1602                 return -1;
1603
1604         event = (union perf_event *)buf;
1605
1606         if (session->header.needs_swap)
1607                 perf_event_header__bswap(&event->header);
1608
1609         if (event->header.size < hdr_sz || event->header.size > buf_sz)
1610                 return -1;
1611
1612         rest = event->header.size - hdr_sz;
1613
1614         if (readn(fd, buf, rest) != (ssize_t)rest)
1615                 return -1;
1616
1617         if (session->header.needs_swap)
1618                 event_swap(event, perf_evlist__sample_id_all(session->evlist));
1619
1620 out_parse_sample:
1621
1622         if (sample && event->header.type < PERF_RECORD_USER_TYPE_START &&
1623             perf_evlist__parse_sample(session->evlist, event, sample))
1624                 return -1;
1625
1626         *event_ptr = event;
1627
1628         return 0;
1629 }
1630
1631 static s64 perf_session__process_event(struct perf_session *session,
1632                                        union perf_event *event, u64 file_offset)
1633 {
1634         struct perf_evlist *evlist = session->evlist;
1635         struct perf_tool *tool = session->tool;
1636         int ret;
1637
1638         if (session->header.needs_swap)
1639                 event_swap(event, perf_evlist__sample_id_all(evlist));
1640
1641         if (event->header.type >= PERF_RECORD_HEADER_MAX)
1642                 return -EINVAL;
1643
1644         events_stats__inc(&evlist->stats, event->header.type);
1645
1646         if (event->header.type >= PERF_RECORD_USER_TYPE_START)
1647                 return perf_session__process_user_event(session, event, file_offset);
1648
1649         if (tool->ordered_events) {
1650                 u64 timestamp = -1ULL;
1651
1652                 ret = perf_evlist__parse_sample_timestamp(evlist, event, &timestamp);
1653                 if (ret && ret != -1)
1654                         return ret;
1655
1656                 ret = perf_session__queue_event(session, event, timestamp, file_offset);
1657                 if (ret != -ETIME)
1658                         return ret;
1659         }
1660
1661         return perf_session__deliver_event(session, event, tool, file_offset);
1662 }
1663
1664 void perf_event_header__bswap(struct perf_event_header *hdr)
1665 {
1666         hdr->type = bswap_32(hdr->type);
1667         hdr->misc = bswap_16(hdr->misc);
1668         hdr->size = bswap_16(hdr->size);
1669 }
1670
1671 struct thread *perf_session__findnew(struct perf_session *session, pid_t pid)
1672 {
1673         return machine__findnew_thread(&session->machines.host, -1, pid);
1674 }
1675
1676 /*
1677  * Threads are identified by pid and tid, and the idle task has pid == tid == 0.
1678  * So here a single thread is created for that, but actually there is a separate
1679  * idle task per cpu, so there should be one 'struct thread' per cpu, but there
1680  * is only 1. That causes problems for some tools, requiring workarounds. For
1681  * example get_idle_thread() in builtin-sched.c, or thread_stack__per_cpu().
1682  */
1683 int perf_session__register_idle_thread(struct perf_session *session)
1684 {
1685         struct thread *thread;
1686         int err = 0;
1687
1688         thread = machine__findnew_thread(&session->machines.host, 0, 0);
1689         if (thread == NULL || thread__set_comm(thread, "swapper", 0)) {
1690                 pr_err("problem inserting idle task.\n");
1691                 err = -1;
1692         }
1693
1694         if (thread == NULL || thread__set_namespaces(thread, 0, NULL)) {
1695                 pr_err("problem inserting idle task.\n");
1696                 err = -1;
1697         }
1698
1699         /* machine__findnew_thread() got the thread, so put it */
1700         thread__put(thread);
1701         return err;
1702 }
1703
1704 static void
1705 perf_session__warn_order(const struct perf_session *session)
1706 {
1707         const struct ordered_events *oe = &session->ordered_events;
1708         struct perf_evsel *evsel;
1709         bool should_warn = true;
1710
1711         evlist__for_each_entry(session->evlist, evsel) {
1712                 if (evsel->attr.write_backward)
1713                         should_warn = false;
1714         }
1715
1716         if (!should_warn)
1717                 return;
1718         if (oe->nr_unordered_events != 0)
1719                 ui__warning("%u out of order events recorded.\n", oe->nr_unordered_events);
1720 }
1721
1722 static void perf_session__warn_about_errors(const struct perf_session *session)
1723 {
1724         const struct events_stats *stats = &session->evlist->stats;
1725
1726         if (session->tool->lost == perf_event__process_lost &&
1727             stats->nr_events[PERF_RECORD_LOST] != 0) {
1728                 ui__warning("Processed %d events and lost %d chunks!\n\n"
1729                             "Check IO/CPU overload!\n\n",
1730                             stats->nr_events[0],
1731                             stats->nr_events[PERF_RECORD_LOST]);
1732         }
1733
1734         if (session->tool->lost_samples == perf_event__process_lost_samples) {
1735                 double drop_rate;
1736
1737                 drop_rate = (double)stats->total_lost_samples /
1738                             (double) (stats->nr_events[PERF_RECORD_SAMPLE] + stats->total_lost_samples);
1739                 if (drop_rate > 0.05) {
1740                         ui__warning("Processed %" PRIu64 " samples and lost %3.2f%%!\n\n",
1741                                     stats->nr_events[PERF_RECORD_SAMPLE] + stats->total_lost_samples,
1742                                     drop_rate * 100.0);
1743                 }
1744         }
1745
1746         if (session->tool->aux == perf_event__process_aux &&
1747             stats->total_aux_lost != 0) {
1748                 ui__warning("AUX data lost %" PRIu64 " times out of %u!\n\n",
1749                             stats->total_aux_lost,
1750                             stats->nr_events[PERF_RECORD_AUX]);
1751         }
1752
1753         if (session->tool->aux == perf_event__process_aux &&
1754             stats->total_aux_partial != 0) {
1755                 bool vmm_exclusive = false;
1756
1757                 (void)sysfs__read_bool("module/kvm_intel/parameters/vmm_exclusive",
1758                                        &vmm_exclusive);
1759
1760                 ui__warning("AUX data had gaps in it %" PRIu64 " times out of %u!\n\n"
1761                             "Are you running a KVM guest in the background?%s\n\n",
1762                             stats->total_aux_partial,
1763                             stats->nr_events[PERF_RECORD_AUX],
1764                             vmm_exclusive ?
1765                             "\nReloading kvm_intel module with vmm_exclusive=0\n"
1766                             "will reduce the gaps to only guest's timeslices." :
1767                             "");
1768         }
1769
1770         if (stats->nr_unknown_events != 0) {
1771                 ui__warning("Found %u unknown events!\n\n"
1772                             "Is this an older tool processing a perf.data "
1773                             "file generated by a more recent tool?\n\n"
1774                             "If that is not the case, consider "
1775                             "reporting to linux-kernel@vger.kernel.org.\n\n",
1776                             stats->nr_unknown_events);
1777         }
1778
1779         if (stats->nr_unknown_id != 0) {
1780                 ui__warning("%u samples with id not present in the header\n",
1781                             stats->nr_unknown_id);
1782         }
1783
1784         if (stats->nr_invalid_chains != 0) {
1785                 ui__warning("Found invalid callchains!\n\n"
1786                             "%u out of %u events were discarded for this reason.\n\n"
1787                             "Consider reporting to linux-kernel@vger.kernel.org.\n\n",
1788                             stats->nr_invalid_chains,
1789                             stats->nr_events[PERF_RECORD_SAMPLE]);
1790         }
1791
1792         if (stats->nr_unprocessable_samples != 0) {
1793                 ui__warning("%u unprocessable samples recorded.\n"
1794                             "Do you have a KVM guest running and not using 'perf kvm'?\n",
1795                             stats->nr_unprocessable_samples);
1796         }
1797
1798         perf_session__warn_order(session);
1799
1800         events_stats__auxtrace_error_warn(stats);
1801
1802         if (stats->nr_proc_map_timeout != 0) {
1803                 ui__warning("%d map information files for pre-existing threads were\n"
1804                             "not processed, if there are samples for addresses they\n"
1805                             "will not be resolved, you may find out which are these\n"
1806                             "threads by running with -v and redirecting the output\n"
1807                             "to a file.\n"
1808                             "The time limit to process proc map is too short?\n"
1809                             "Increase it by --proc-map-timeout\n",
1810                             stats->nr_proc_map_timeout);
1811         }
1812 }
1813
1814 static int perf_session__flush_thread_stack(struct thread *thread,
1815                                             void *p __maybe_unused)
1816 {
1817         return thread_stack__flush(thread);
1818 }
1819
1820 static int perf_session__flush_thread_stacks(struct perf_session *session)
1821 {
1822         return machines__for_each_thread(&session->machines,
1823                                          perf_session__flush_thread_stack,
1824                                          NULL);
1825 }
1826
1827 volatile int session_done;
1828
1829 static int __perf_session__process_decomp_events(struct perf_session *session);
1830
1831 static int __perf_session__process_pipe_events(struct perf_session *session)
1832 {
1833         struct ordered_events *oe = &session->ordered_events;
1834         struct perf_tool *tool = session->tool;
1835         int fd = perf_data__fd(session->data);
1836         union perf_event *event;
1837         uint32_t size, cur_size = 0;
1838         void *buf = NULL;
1839         s64 skip = 0;
1840         u64 head;
1841         ssize_t err;
1842         void *p;
1843
1844         perf_tool__fill_defaults(tool);
1845
1846         head = 0;
1847         cur_size = sizeof(union perf_event);
1848
1849         buf = malloc(cur_size);
1850         if (!buf)
1851                 return -errno;
1852         ordered_events__set_copy_on_queue(oe, true);
1853 more:
1854         event = buf;
1855         err = readn(fd, event, sizeof(struct perf_event_header));
1856         if (err <= 0) {
1857                 if (err == 0)
1858                         goto done;
1859
1860                 pr_err("failed to read event header\n");
1861                 goto out_err;
1862         }
1863
1864         if (session->header.needs_swap)
1865                 perf_event_header__bswap(&event->header);
1866
1867         size = event->header.size;
1868         if (size < sizeof(struct perf_event_header)) {
1869                 pr_err("bad event header size\n");
1870                 goto out_err;
1871         }
1872
1873         if (size > cur_size) {
1874                 void *new = realloc(buf, size);
1875                 if (!new) {
1876                         pr_err("failed to allocate memory to read event\n");
1877                         goto out_err;
1878                 }
1879                 buf = new;
1880                 cur_size = size;
1881                 event = buf;
1882         }
1883         p = event;
1884         p += sizeof(struct perf_event_header);
1885
1886         if (size - sizeof(struct perf_event_header)) {
1887                 err = readn(fd, p, size - sizeof(struct perf_event_header));
1888                 if (err <= 0) {
1889                         if (err == 0) {
1890                                 pr_err("unexpected end of event stream\n");
1891                                 goto done;
1892                         }
1893
1894                         pr_err("failed to read event data\n");
1895                         goto out_err;
1896                 }
1897         }
1898
1899         if ((skip = perf_session__process_event(session, event, head)) < 0) {
1900                 pr_err("%#" PRIx64 " [%#x]: failed to process type: %d\n",
1901                        head, event->header.size, event->header.type);
1902                 err = -EINVAL;
1903                 goto out_err;
1904         }
1905
1906         head += size;
1907
1908         if (skip > 0)
1909                 head += skip;
1910
1911         err = __perf_session__process_decomp_events(session);
1912         if (err)
1913                 goto out_err;
1914
1915         if (!session_done())
1916                 goto more;
1917 done:
1918         /* do the final flush for ordered samples */
1919         err = ordered_events__flush(oe, OE_FLUSH__FINAL);
1920         if (err)
1921                 goto out_err;
1922         err = auxtrace__flush_events(session, tool);
1923         if (err)
1924                 goto out_err;
1925         err = perf_session__flush_thread_stacks(session);
1926 out_err:
1927         free(buf);
1928         if (!tool->no_warn)
1929                 perf_session__warn_about_errors(session);
1930         ordered_events__free(&session->ordered_events);
1931         auxtrace__free_events(session);
1932         return err;
1933 }
1934
1935 static union perf_event *
1936 fetch_mmaped_event(struct perf_session *session,
1937                    u64 head, size_t mmap_size, char *buf)
1938 {
1939         union perf_event *event;
1940
1941         /*
1942          * Ensure we have enough space remaining to read
1943          * the size of the event in the headers.
1944          */
1945         if (head + sizeof(event->header) > mmap_size)
1946                 return NULL;
1947
1948         event = (union perf_event *)(buf + head);
1949
1950         if (session->header.needs_swap)
1951                 perf_event_header__bswap(&event->header);
1952
1953         if (head + event->header.size > mmap_size) {
1954                 /* We're not fetching the event so swap back again */
1955                 if (session->header.needs_swap)
1956                         perf_event_header__bswap(&event->header);
1957                 return NULL;
1958         }
1959
1960         return event;
1961 }
1962
1963 static int __perf_session__process_decomp_events(struct perf_session *session)
1964 {
1965         s64 skip;
1966         u64 size, file_pos = 0;
1967         struct decomp *decomp = session->decomp_last;
1968
1969         if (!decomp)
1970                 return 0;
1971
1972         while (decomp->head < decomp->size && !session_done()) {
1973                 union perf_event *event = fetch_mmaped_event(session, decomp->head, decomp->size, decomp->data);
1974
1975                 if (!event)
1976                         break;
1977
1978                 size = event->header.size;
1979
1980                 if (size < sizeof(struct perf_event_header) ||
1981                     (skip = perf_session__process_event(session, event, file_pos)) < 0) {
1982                         pr_err("%#" PRIx64 " [%#x]: failed to process type: %d\n",
1983                                 decomp->file_pos + decomp->head, event->header.size, event->header.type);
1984                         return -EINVAL;
1985                 }
1986
1987                 if (skip)
1988                         size += skip;
1989
1990                 decomp->head += size;
1991         }
1992
1993         return 0;
1994 }
1995
1996 /*
1997  * On 64bit we can mmap the data file in one go. No need for tiny mmap
1998  * slices. On 32bit we use 32MB.
1999  */
2000 #if BITS_PER_LONG == 64
2001 #define MMAP_SIZE ULLONG_MAX
2002 #define NUM_MMAPS 1
2003 #else
2004 #define MMAP_SIZE (32 * 1024 * 1024ULL)
2005 #define NUM_MMAPS 128
2006 #endif
2007
2008 struct reader;
2009
2010 typedef s64 (*reader_cb_t)(struct perf_session *session,
2011                            union perf_event *event,
2012                            u64 file_offset);
2013
2014 struct reader {
2015         int              fd;
2016         u64              data_size;
2017         u64              data_offset;
2018         reader_cb_t      process;
2019 };
2020
2021 static int
2022 reader__process_events(struct reader *rd, struct perf_session *session,
2023                        struct ui_progress *prog)
2024 {
2025         u64 data_size = rd->data_size;
2026         u64 head, page_offset, file_offset, file_pos, size;
2027         int err = 0, mmap_prot, mmap_flags, map_idx = 0;
2028         size_t  mmap_size;
2029         char *buf, *mmaps[NUM_MMAPS];
2030         union perf_event *event;
2031         s64 skip;
2032
2033         page_offset = page_size * (rd->data_offset / page_size);
2034         file_offset = page_offset;
2035         head = rd->data_offset - page_offset;
2036
2037         ui_progress__init_size(prog, data_size, "Processing events...");
2038
2039         data_size += rd->data_offset;
2040
2041         mmap_size = MMAP_SIZE;
2042         if (mmap_size > data_size) {
2043                 mmap_size = data_size;
2044                 session->one_mmap = true;
2045         }
2046
2047         memset(mmaps, 0, sizeof(mmaps));
2048
2049         mmap_prot  = PROT_READ;
2050         mmap_flags = MAP_SHARED;
2051
2052         if (session->header.needs_swap) {
2053                 mmap_prot  |= PROT_WRITE;
2054                 mmap_flags = MAP_PRIVATE;
2055         }
2056 remap:
2057         buf = mmap(NULL, mmap_size, mmap_prot, mmap_flags, rd->fd,
2058                    file_offset);
2059         if (buf == MAP_FAILED) {
2060                 pr_err("failed to mmap file\n");
2061                 err = -errno;
2062                 goto out;
2063         }
2064         mmaps[map_idx] = buf;
2065         map_idx = (map_idx + 1) & (ARRAY_SIZE(mmaps) - 1);
2066         file_pos = file_offset + head;
2067         if (session->one_mmap) {
2068                 session->one_mmap_addr = buf;
2069                 session->one_mmap_offset = file_offset;
2070         }
2071
2072 more:
2073         event = fetch_mmaped_event(session, head, mmap_size, buf);
2074         if (!event) {
2075                 if (mmaps[map_idx]) {
2076                         munmap(mmaps[map_idx], mmap_size);
2077                         mmaps[map_idx] = NULL;
2078                 }
2079
2080                 page_offset = page_size * (head / page_size);
2081                 file_offset += page_offset;
2082                 head -= page_offset;
2083                 goto remap;
2084         }
2085
2086         size = event->header.size;
2087
2088         skip = -EINVAL;
2089
2090         if (size < sizeof(struct perf_event_header) ||
2091             (skip = rd->process(session, event, file_pos)) < 0) {
2092                 pr_err("%#" PRIx64 " [%#x]: failed to process type: %d [%s]\n",
2093                        file_offset + head, event->header.size,
2094                        event->header.type, strerror(-skip));
2095                 err = skip;
2096                 goto out;
2097         }
2098
2099         if (skip)
2100                 size += skip;
2101
2102         head += size;
2103         file_pos += size;
2104
2105         err = __perf_session__process_decomp_events(session);
2106         if (err)
2107                 goto out;
2108
2109         ui_progress__update(prog, size);
2110
2111         if (session_done())
2112                 goto out;
2113
2114         if (file_pos < data_size)
2115                 goto more;
2116
2117 out:
2118         return err;
2119 }
2120
2121 static s64 process_simple(struct perf_session *session,
2122                           union perf_event *event,
2123                           u64 file_offset)
2124 {
2125         return perf_session__process_event(session, event, file_offset);
2126 }
2127
2128 static int __perf_session__process_events(struct perf_session *session)
2129 {
2130         struct reader rd = {
2131                 .fd             = perf_data__fd(session->data),
2132                 .data_size      = session->header.data_size,
2133                 .data_offset    = session->header.data_offset,
2134                 .process        = process_simple,
2135         };
2136         struct ordered_events *oe = &session->ordered_events;
2137         struct perf_tool *tool = session->tool;
2138         struct ui_progress prog;
2139         int err;
2140
2141         perf_tool__fill_defaults(tool);
2142
2143         if (rd.data_size == 0)
2144                 return -1;
2145
2146         ui_progress__init_size(&prog, rd.data_size, "Processing events...");
2147
2148         err = reader__process_events(&rd, session, &prog);
2149         if (err)
2150                 goto out_err;
2151         /* do the final flush for ordered samples */
2152         err = ordered_events__flush(oe, OE_FLUSH__FINAL);
2153         if (err)
2154                 goto out_err;
2155         err = auxtrace__flush_events(session, tool);
2156         if (err)
2157                 goto out_err;
2158         err = perf_session__flush_thread_stacks(session);
2159 out_err:
2160         ui_progress__finish();
2161         if (!tool->no_warn)
2162                 perf_session__warn_about_errors(session);
2163         /*
2164          * We may switching perf.data output, make ordered_events
2165          * reusable.
2166          */
2167         ordered_events__reinit(&session->ordered_events);
2168         auxtrace__free_events(session);
2169         session->one_mmap = false;
2170         return err;
2171 }
2172
2173 int perf_session__process_events(struct perf_session *session)
2174 {
2175         if (perf_session__register_idle_thread(session) < 0)
2176                 return -ENOMEM;
2177
2178         if (perf_data__is_pipe(session->data))
2179                 return __perf_session__process_pipe_events(session);
2180
2181         return __perf_session__process_events(session);
2182 }
2183
2184 bool perf_session__has_traces(struct perf_session *session, const char *msg)
2185 {
2186         struct perf_evsel *evsel;
2187
2188         evlist__for_each_entry(session->evlist, evsel) {
2189                 if (evsel->attr.type == PERF_TYPE_TRACEPOINT)
2190                         return true;
2191         }
2192
2193         pr_err("No trace sample to read. Did you call 'perf %s'?\n", msg);
2194         return false;
2195 }
2196
2197 int map__set_kallsyms_ref_reloc_sym(struct map *map, const char *symbol_name, u64 addr)
2198 {
2199         char *bracket;
2200         struct ref_reloc_sym *ref;
2201         struct kmap *kmap;
2202
2203         ref = zalloc(sizeof(struct ref_reloc_sym));
2204         if (ref == NULL)
2205                 return -ENOMEM;
2206
2207         ref->name = strdup(symbol_name);
2208         if (ref->name == NULL) {
2209                 free(ref);
2210                 return -ENOMEM;
2211         }
2212
2213         bracket = strchr(ref->name, ']');
2214         if (bracket)
2215                 *bracket = '\0';
2216
2217         ref->addr = addr;
2218
2219         kmap = map__kmap(map);
2220         if (kmap)
2221                 kmap->ref_reloc_sym = ref;
2222
2223         return 0;
2224 }
2225
2226 size_t perf_session__fprintf_dsos(struct perf_session *session, FILE *fp)
2227 {
2228         return machines__fprintf_dsos(&session->machines, fp);
2229 }
2230
2231 size_t perf_session__fprintf_dsos_buildid(struct perf_session *session, FILE *fp,
2232                                           bool (skip)(struct dso *dso, int parm), int parm)
2233 {
2234         return machines__fprintf_dsos_buildid(&session->machines, fp, skip, parm);
2235 }
2236
2237 size_t perf_session__fprintf_nr_events(struct perf_session *session, FILE *fp)
2238 {
2239         size_t ret;
2240         const char *msg = "";
2241
2242         if (perf_header__has_feat(&session->header, HEADER_AUXTRACE))
2243                 msg = " (excludes AUX area (e.g. instruction trace) decoded / synthesized events)";
2244
2245         ret = fprintf(fp, "\nAggregated stats:%s\n", msg);
2246
2247         ret += events_stats__fprintf(&session->evlist->stats, fp);
2248         return ret;
2249 }
2250
2251 size_t perf_session__fprintf(struct perf_session *session, FILE *fp)
2252 {
2253         /*
2254          * FIXME: Here we have to actually print all the machines in this
2255          * session, not just the host...
2256          */
2257         return machine__fprintf(&session->machines.host, fp);
2258 }
2259
2260 struct perf_evsel *perf_session__find_first_evtype(struct perf_session *session,
2261                                               unsigned int type)
2262 {
2263         struct perf_evsel *pos;
2264
2265         evlist__for_each_entry(session->evlist, pos) {
2266                 if (pos->attr.type == type)
2267                         return pos;
2268         }
2269         return NULL;
2270 }
2271
2272 int perf_session__cpu_bitmap(struct perf_session *session,
2273                              const char *cpu_list, unsigned long *cpu_bitmap)
2274 {
2275         int i, err = -1;
2276         struct cpu_map *map;
2277
2278         for (i = 0; i < PERF_TYPE_MAX; ++i) {
2279                 struct perf_evsel *evsel;
2280
2281                 evsel = perf_session__find_first_evtype(session, i);
2282                 if (!evsel)
2283                         continue;
2284
2285                 if (!(evsel->attr.sample_type & PERF_SAMPLE_CPU)) {
2286                         pr_err("File does not contain CPU events. "
2287                                "Remove -C option to proceed.\n");
2288                         return -1;
2289                 }
2290         }
2291
2292         map = cpu_map__new(cpu_list);
2293         if (map == NULL) {
2294                 pr_err("Invalid cpu_list\n");
2295                 return -1;
2296         }
2297
2298         for (i = 0; i < map->nr; i++) {
2299                 int cpu = map->map[i];
2300
2301                 if (cpu >= MAX_NR_CPUS) {
2302                         pr_err("Requested CPU %d too large. "
2303                                "Consider raising MAX_NR_CPUS\n", cpu);
2304                         goto out_delete_map;
2305                 }
2306
2307                 set_bit(cpu, cpu_bitmap);
2308         }
2309
2310         err = 0;
2311
2312 out_delete_map:
2313         cpu_map__put(map);
2314         return err;
2315 }
2316
2317 void perf_session__fprintf_info(struct perf_session *session, FILE *fp,
2318                                 bool full)
2319 {
2320         if (session == NULL || fp == NULL)
2321                 return;
2322
2323         fprintf(fp, "# ========\n");
2324         perf_header__fprintf_info(session, fp, full);
2325         fprintf(fp, "# ========\n#\n");
2326 }
2327
2328
2329 int __perf_session__set_tracepoints_handlers(struct perf_session *session,
2330                                              const struct perf_evsel_str_handler *assocs,
2331                                              size_t nr_assocs)
2332 {
2333         struct perf_evsel *evsel;
2334         size_t i;
2335         int err;
2336
2337         for (i = 0; i < nr_assocs; i++) {
2338                 /*
2339                  * Adding a handler for an event not in the session,
2340                  * just ignore it.
2341                  */
2342                 evsel = perf_evlist__find_tracepoint_by_name(session->evlist, assocs[i].name);
2343                 if (evsel == NULL)
2344                         continue;
2345
2346                 err = -EEXIST;
2347                 if (evsel->handler != NULL)
2348                         goto out;
2349                 evsel->handler = assocs[i].handler;
2350         }
2351
2352         err = 0;
2353 out:
2354         return err;
2355 }
2356
2357 int perf_event__process_id_index(struct perf_session *session,
2358                                  union perf_event *event)
2359 {
2360         struct perf_evlist *evlist = session->evlist;
2361         struct id_index_event *ie = &event->id_index;
2362         size_t i, nr, max_nr;
2363
2364         max_nr = (ie->header.size - sizeof(struct id_index_event)) /
2365                  sizeof(struct id_index_entry);
2366         nr = ie->nr;
2367         if (nr > max_nr)
2368                 return -EINVAL;
2369
2370         if (dump_trace)
2371                 fprintf(stdout, " nr: %zu\n", nr);
2372
2373         for (i = 0; i < nr; i++) {
2374                 struct id_index_entry *e = &ie->entries[i];
2375                 struct perf_sample_id *sid;
2376
2377                 if (dump_trace) {
2378                         fprintf(stdout, " ... id: %"PRIu64, e->id);
2379                         fprintf(stdout, "  idx: %"PRIu64, e->idx);
2380                         fprintf(stdout, "  cpu: %"PRId64, e->cpu);
2381                         fprintf(stdout, "  tid: %"PRId64"\n", e->tid);
2382                 }
2383
2384                 sid = perf_evlist__id2sid(evlist, e->id);
2385                 if (!sid)
2386                         return -ENOENT;
2387                 sid->idx = e->idx;
2388                 sid->cpu = e->cpu;
2389                 sid->tid = e->tid;
2390         }
2391         return 0;
2392 }
2393
2394 int perf_event__synthesize_id_index(struct perf_tool *tool,
2395                                     perf_event__handler_t process,
2396                                     struct perf_evlist *evlist,
2397                                     struct machine *machine)
2398 {
2399         union perf_event *ev;
2400         struct perf_evsel *evsel;
2401         size_t nr = 0, i = 0, sz, max_nr, n;
2402         int err;
2403
2404         pr_debug2("Synthesizing id index\n");
2405
2406         max_nr = (UINT16_MAX - sizeof(struct id_index_event)) /
2407                  sizeof(struct id_index_entry);
2408
2409         evlist__for_each_entry(evlist, evsel)
2410                 nr += evsel->ids;
2411
2412         n = nr > max_nr ? max_nr : nr;
2413         sz = sizeof(struct id_index_event) + n * sizeof(struct id_index_entry);
2414         ev = zalloc(sz);
2415         if (!ev)
2416                 return -ENOMEM;
2417
2418         ev->id_index.header.type = PERF_RECORD_ID_INDEX;
2419         ev->id_index.header.size = sz;
2420         ev->id_index.nr = n;
2421
2422         evlist__for_each_entry(evlist, evsel) {
2423                 u32 j;
2424
2425                 for (j = 0; j < evsel->ids; j++) {
2426                         struct id_index_entry *e;
2427                         struct perf_sample_id *sid;
2428
2429                         if (i >= n) {
2430                                 err = process(tool, ev, NULL, machine);
2431                                 if (err)
2432                                         goto out_err;
2433                                 nr -= n;
2434                                 i = 0;
2435                         }
2436
2437                         e = &ev->id_index.entries[i++];
2438
2439                         e->id = evsel->id[j];
2440
2441                         sid = perf_evlist__id2sid(evlist, e->id);
2442                         if (!sid) {
2443                                 free(ev);
2444                                 return -ENOENT;
2445                         }
2446
2447                         e->idx = sid->idx;
2448                         e->cpu = sid->cpu;
2449                         e->tid = sid->tid;
2450                 }
2451         }
2452
2453         sz = sizeof(struct id_index_event) + nr * sizeof(struct id_index_entry);
2454         ev->id_index.header.size = sz;
2455         ev->id_index.nr = nr;
2456
2457         err = process(tool, ev, NULL, machine);
2458 out_err:
2459         free(ev);
2460
2461         return err;
2462 }