tools/perf/util/session.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 #include <errno.h>
   3 #include <inttypes.h>
   4 #include <linux/err.h>
   5 #include <linux/kernel.h>
   6 #include <linux/zalloc.h>
   7 #include <api/fs/fs.h>
   8
   9 #include <byteswap.h>
  10 #include <unistd.h>
  11 #include <sys/types.h>
  12 #include <sys/mman.h>
  13 #include <perf/cpumap.h>
  14
  15 #include "map_symbol.h"
  16 #include "branch.h"
  17 #include "debug.h"
  18 #include "evlist.h"
  19 #include "evsel.h"
  20 #include "memswap.h"
  21 #include "map.h"
  22 #include "symbol.h"
  23 #include "session.h"
  24 #include "tool.h"
  25 #include "perf_regs.h"
  26 #include "asm/bug.h"
  27 #include "auxtrace.h"
  28 #include "thread.h"
  29 #include "thread-stack.h"
  30 #include "sample-raw.h"
  31 #include "stat.h"
  32 #include "tsc.h"
  33 #include "ui/progress.h"
  34 #include "../perf.h"
  35 #include "arch/common.h"
  36 #include "units.h"
  37 #include <internal/lib.h>
  38
  39 #ifdef HAVE_ZSTD_SUPPORT
  40 static int perf_session__process_compressed_event(struct perf_session *session,
  41                                                   union perf_event *event, u64 file_offset)
  42 {
  43         void *src;
  44         size_t decomp_size, src_size;
  45         u64 decomp_last_rem = 0;
  46         size_t mmap_len, decomp_len = session->header.env.comp_mmap_len;
  47         struct decomp *decomp, *decomp_last = session->decomp_last;
  48
  49         if (decomp_last) {
  50                 decomp_last_rem = decomp_last->size - decomp_last->head;
  51                 decomp_len += decomp_last_rem;
  52         }
  53
  54         mmap_len = sizeof(struct decomp) + decomp_len;
  55         decomp = mmap(NULL, mmap_len, PROT_READ|PROT_WRITE,
  56                       MAP_ANONYMOUS|MAP_PRIVATE, -1, 0);
  57         if (decomp == MAP_FAILED) {
  58                 pr_err("Couldn't allocate memory for decompression\n");
  59                 return -1;
  60         }
  61
  62         decomp->file_pos = file_offset;
  63         decomp->mmap_len = mmap_len;
  64         decomp->head = 0;
  65
  66         if (decomp_last_rem) {
  67                 memcpy(decomp->data, &(decomp_last->data[decomp_last->head]), decomp_last_rem);
  68                 decomp->size = decomp_last_rem;
  69         }
  70
  71         src = (void *)event + sizeof(struct perf_record_compressed);
  72         src_size = event->pack.header.size - sizeof(struct perf_record_compressed);
  73
  74         decomp_size = zstd_decompress_stream(&(session->zstd_data), src, src_size,
  75                                 &(decomp->data[decomp_last_rem]), decomp_len - decomp_last_rem);
  76         if (!decomp_size) {
  77                 munmap(decomp, mmap_len);
  78                 pr_err("Couldn't decompress data\n");
  79                 return -1;
  80         }
  81
  82         decomp->size += decomp_size;
  83
  84         if (session->decomp == NULL) {
  85                 session->decomp = decomp;
  86                 session->decomp_last = decomp;
  87         } else {
  88                 session->decomp_last->next = decomp;
  89                 session->decomp_last = decomp;
  90         }
  91
  92         pr_debug("decomp (B): %zd to %zd\n", src_size, decomp_size);
  93
  94         return 0;
  95 }
  96 #else /* !HAVE_ZSTD_SUPPORT */
  97 #define perf_session__process_compressed_event perf_session__process_compressed_event_stub
  98 #endif
  99
 100 static int perf_session__deliver_event(struct perf_session *session,
 101                                        union perf_event *event,
 102                                        struct perf_tool *tool,
 103                                        u64 file_offset);
 104
 105 static int perf_session__open(struct perf_session *session, int repipe_fd)
 106 {
 107         struct perf_data *data = session->data;
 108
 109         if (perf_session__read_header(session, repipe_fd) < 0) {
 110                 pr_err("incompatible file format (rerun with -v to learn more)\n");
 111                 return -1;
 112         }
 113
 114         if (perf_data__is_pipe(data))
 115                 return 0;
 116
 117         if (perf_header__has_feat(&session->header, HEADER_STAT))
 118                 return 0;
 119
 120         if (!evlist__valid_sample_type(session->evlist)) {
 121                 pr_err("non matching sample_type\n");
 122                 return -1;
 123         }
 124
 125         if (!evlist__valid_sample_id_all(session->evlist)) {
 126                 pr_err("non matching sample_id_all\n");
 127                 return -1;
 128         }
 129
 130         if (!evlist__valid_read_format(session->evlist)) {
 131                 pr_err("non matching read_format\n");
 132                 return -1;
 133         }
 134
 135         return 0;
 136 }
 137
 138 void perf_session__set_id_hdr_size(struct perf_session *session)
 139 {
 140         u16 id_hdr_size = evlist__id_hdr_size(session->evlist);
 141
 142         machines__set_id_hdr_size(&session->machines, id_hdr_size);
 143 }
 144
 145 int perf_session__create_kernel_maps(struct perf_session *session)
 146 {
 147         int ret = machine__create_kernel_maps(&session->machines.host);
 148
 149         if (ret >= 0)
 150                 ret = machines__create_guest_kernel_maps(&session->machines);
 151         return ret;
 152 }
 153
 154 static void perf_session__destroy_kernel_maps(struct perf_session *session)
 155 {
 156         machines__destroy_kernel_maps(&session->machines);
 157 }
 158
 159 static bool perf_session__has_comm_exec(struct perf_session *session)
 160 {
 161         struct evsel *evsel;
 162
 163         evlist__for_each_entry(session->evlist, evsel) {
 164                 if (evsel->core.attr.comm_exec)
 165                         return true;
 166         }
 167
 168         return false;
 169 }
 170
 171 static void perf_session__set_comm_exec(struct perf_session *session)
 172 {
 173         bool comm_exec = perf_session__has_comm_exec(session);
 174
 175         machines__set_comm_exec(&session->machines, comm_exec);
 176 }
 177
 178 static int ordered_events__deliver_event(struct ordered_events *oe,
 179                                          struct ordered_event *event)
 180 {
 181         struct perf_session *session = container_of(oe, struct perf_session,
 182                                                     ordered_events);
 183
 184         return perf_session__deliver_event(session, event->event,
 185                                            session->tool, event->file_offset);
 186 }
 187
 188 struct perf_session *__perf_session__new(struct perf_data *data,
 189                                          bool repipe, int repipe_fd,
 190                                          struct perf_tool *tool)
 191 {
 192         int ret = -ENOMEM;
 193         struct perf_session *session = zalloc(sizeof(*session));
 194
 195         if (!session)
 196                 goto out;
 197
 198         session->repipe = repipe;
 199         session->tool   = tool;
 200         INIT_LIST_HEAD(&session->auxtrace_index);
 201         machines__init(&session->machines);
 202         ordered_events__init(&session->ordered_events,
 203                              ordered_events__deliver_event, NULL);
 204
 205         perf_env__init(&session->header.env);
 206         if (data) {
 207                 ret = perf_data__open(data);
 208                 if (ret < 0)
 209                         goto out_delete;
 210
 211                 session->data = data;
 212
 213                 if (perf_data__is_read(data)) {
 214                         ret = perf_session__open(session, repipe_fd);
 215                         if (ret < 0)
 216                                 goto out_delete;
 217
 218                         /*
 219                          * set session attributes that are present in perf.data
 220                          * but not in pipe-mode.
 221                          */
 222                         if (!data->is_pipe) {
 223                                 perf_session__set_id_hdr_size(session);
 224                                 perf_session__set_comm_exec(session);
 225                         }
 226
 227                         evlist__init_trace_event_sample_raw(session->evlist);
 228
 229                         /* Open the directory data. */
 230                         if (data->is_dir) {
 231                                 ret = perf_data__open_dir(data);
 232                                 if (ret)
 233                                         goto out_delete;
 234                         }
 235
 236                         if (!symbol_conf.kallsyms_name &&
 237                             !symbol_conf.vmlinux_name)
 238                                 symbol_conf.kallsyms_name = perf_data__kallsyms_name(data);
 239                 }
 240         } else  {
 241                 session->machines.host.env = &perf_env;
 242         }
 243
 244         session->machines.host.single_address_space =
 245                 perf_env__single_address_space(session->machines.host.env);
 246
 247         if (!data || perf_data__is_write(data)) {
 248                 /*
 249                  * In O_RDONLY mode this will be performed when reading the
 250                  * kernel MMAP event, in perf_event__process_mmap().
 251                  */
 252                 if (perf_session__create_kernel_maps(session) < 0)
 253                         pr_warning("Cannot read kernel map\n");
 254         }
 255
 256         /*
 257          * In pipe-mode, evlist is empty until PERF_RECORD_HEADER_ATTR is
 258          * processed, so evlist__sample_id_all is not meaningful here.
 259          */
 260         if ((!data || !data->is_pipe) && tool && tool->ordering_requires_timestamps &&
 261             tool->ordered_events && !evlist__sample_id_all(session->evlist)) {
 262                 dump_printf("WARNING: No sample_id_all support, falling back to unordered processing\n");
 263                 tool->ordered_events = false;
 264         }
 265
 266         return session;
 267
 268  out_delete:
 269         perf_session__delete(session);
 270  out:
 271         return ERR_PTR(ret);
 272 }
 273
 274 static void perf_session__delete_threads(struct perf_session *session)
 275 {
 276         machine__delete_threads(&session->machines.host);
 277 }
 278
 279 static void perf_session__release_decomp_events(struct perf_session *session)
 280 {
 281         struct decomp *next, *decomp;
 282         size_t mmap_len;
 283         next = session->decomp;
 284         do {
 285                 decomp = next;
 286                 if (decomp == NULL)
 287                         break;
 288                 next = decomp->next;
 289                 mmap_len = decomp->mmap_len;
 290                 munmap(decomp, mmap_len);
 291         } while (1);
 292 }
 293
 294 void perf_session__delete(struct perf_session *session)
 295 {
 296         if (session == NULL)
 297                 return;
 298         auxtrace__free(session);
 299         auxtrace_index__free(&session->auxtrace_index);
 300         perf_session__destroy_kernel_maps(session);
 301         perf_session__delete_threads(session);
 302         perf_session__release_decomp_events(session);
 303         perf_env__exit(&session->header.env);
 304         machines__exit(&session->machines);
 305         if (session->data) {
 306                 if (perf_data__is_read(session->data))
 307                         evlist__delete(session->evlist);
 308                 perf_data__close(session->data);
 309         }
 310         trace_event__cleanup(&session->tevent);
 311         free(session);
 312 }
 313
 314 static int process_event_synth_tracing_data_stub(struct perf_session *session
 315                                                  __maybe_unused,
 316                                                  union perf_event *event
 317                                                  __maybe_unused)
 318 {
 319         dump_printf(": unhandled!\n");
 320         return 0;
 321 }
 322
 323 static int process_event_synth_attr_stub(struct perf_tool *tool __maybe_unused,
 324                                          union perf_event *event __maybe_unused,
 325                                          struct evlist **pevlist
 326                                          __maybe_unused)
 327 {
 328         dump_printf(": unhandled!\n");
 329         return 0;
 330 }
 331
 332 static int process_event_synth_event_update_stub(struct perf_tool *tool __maybe_unused,
 333                                                  union perf_event *event __maybe_unused,
 334                                                  struct evlist **pevlist
 335                                                  __maybe_unused)
 336 {
 337         if (dump_trace)
 338                 perf_event__fprintf_event_update(event, stdout);
 339
 340         dump_printf(": unhandled!\n");
 341         return 0;
 342 }
 343
 344 static int process_event_sample_stub(struct perf_tool *tool __maybe_unused,
 345                                      union perf_event *event __maybe_unused,
 346                                      struct perf_sample *sample __maybe_unused,
 347                                      struct evsel *evsel __maybe_unused,
 348                                      struct machine *machine __maybe_unused)
 349 {
 350         dump_printf(": unhandled!\n");
 351         return 0;
 352 }
 353
 354 static int process_event_stub(struct perf_tool *tool __maybe_unused,
 355                               union perf_event *event __maybe_unused,
 356                               struct perf_sample *sample __maybe_unused,
 357                               struct machine *machine __maybe_unused)
 358 {
 359         dump_printf(": unhandled!\n");
 360         return 0;
 361 }
 362
 363 static int process_finished_round_stub(struct perf_tool *tool __maybe_unused,
 364                                        union perf_event *event __maybe_unused,
 365                                        struct ordered_events *oe __maybe_unused)
 366 {
 367         dump_printf(": unhandled!\n");
 368         return 0;
 369 }
 370
 371 static int process_finished_round(struct perf_tool *tool,
 372                                   union perf_event *event,
 373                                   struct ordered_events *oe);
 374
 375 static int skipn(int fd, off_t n)
 376 {
 377         char buf[4096];
 378         ssize_t ret;
 379
 380         while (n > 0) {
 381                 ret = read(fd, buf, min(n, (off_t)sizeof(buf)));
 382                 if (ret <= 0)
 383                         return ret;
 384                 n -= ret;
 385         }
 386
 387         return 0;
 388 }
 389
 390 static s64 process_event_auxtrace_stub(struct perf_session *session __maybe_unused,
 391                                        union perf_event *event)
 392 {
 393         dump_printf(": unhandled!\n");
 394         if (perf_data__is_pipe(session->data))
 395                 skipn(perf_data__fd(session->data), event->auxtrace.size);
 396         return event->auxtrace.size;
 397 }
 398
 399 static int process_event_op2_stub(struct perf_session *session __maybe_unused,
 400                                   union perf_event *event __maybe_unused)
 401 {
 402         dump_printf(": unhandled!\n");
 403         return 0;
 404 }
 405
 406
 407 static
 408 int process_event_thread_map_stub(struct perf_session *session __maybe_unused,
 409                                   union perf_event *event __maybe_unused)
 410 {
 411         if (dump_trace)
 412                 perf_event__fprintf_thread_map(event, stdout);
 413
 414         dump_printf(": unhandled!\n");
 415         return 0;
 416 }
 417
 418 static
 419 int process_event_cpu_map_stub(struct perf_session *session __maybe_unused,
 420                                union perf_event *event __maybe_unused)
 421 {
 422         if (dump_trace)
 423                 perf_event__fprintf_cpu_map(event, stdout);
 424
 425         dump_printf(": unhandled!\n");
 426         return 0;
 427 }
 428
 429 static
 430 int process_event_stat_config_stub(struct perf_session *session __maybe_unused,
 431                                    union perf_event *event __maybe_unused)
 432 {
 433         if (dump_trace)
 434                 perf_event__fprintf_stat_config(event, stdout);
 435
 436         dump_printf(": unhandled!\n");
 437         return 0;
 438 }
 439
 440 static int process_stat_stub(struct perf_session *perf_session __maybe_unused,
 441                              union perf_event *event)
 442 {
 443         if (dump_trace)
 444                 perf_event__fprintf_stat(event, stdout);
 445
 446         dump_printf(": unhandled!\n");
 447         return 0;
 448 }
 449
 450 static int process_stat_round_stub(struct perf_session *perf_session __maybe_unused,
 451                                    union perf_event *event)
 452 {
 453         if (dump_trace)
 454                 perf_event__fprintf_stat_round(event, stdout);
 455
 456         dump_printf(": unhandled!\n");
 457         return 0;
 458 }
 459
 460 static int process_event_time_conv_stub(struct perf_session *perf_session __maybe_unused,
 461                                         union perf_event *event)
 462 {
 463         if (dump_trace)
 464                 perf_event__fprintf_time_conv(event, stdout);
 465
 466         dump_printf(": unhandled!\n");
 467         return 0;
 468 }
 469
 470 static int perf_session__process_compressed_event_stub(struct perf_session *session __maybe_unused,
 471                                                        union perf_event *event __maybe_unused,
 472                                                        u64 file_offset __maybe_unused)
 473 {
 474        dump_printf(": unhandled!\n");
 475        return 0;
 476 }
 477
 478 void perf_tool__fill_defaults(struct perf_tool *tool)
 479 {
 480         if (tool->sample == NULL)
 481                 tool->sample = process_event_sample_stub;
 482         if (tool->mmap == NULL)
 483                 tool->mmap = process_event_stub;
 484         if (tool->mmap2 == NULL)
 485                 tool->mmap2 = process_event_stub;
 486         if (tool->comm == NULL)
 487                 tool->comm = process_event_stub;
 488         if (tool->namespaces == NULL)
 489                 tool->namespaces = process_event_stub;
 490         if (tool->cgroup == NULL)
 491                 tool->cgroup = process_event_stub;
 492         if (tool->fork == NULL)
 493                 tool->fork = process_event_stub;
 494         if (tool->exit == NULL)
 495                 tool->exit = process_event_stub;
 496         if (tool->lost == NULL)
 497                 tool->lost = perf_event__process_lost;
 498         if (tool->lost_samples == NULL)
 499                 tool->lost_samples = perf_event__process_lost_samples;
 500         if (tool->aux == NULL)
 501                 tool->aux = perf_event__process_aux;
 502         if (tool->itrace_start == NULL)
 503                 tool->itrace_start = perf_event__process_itrace_start;
 504         if (tool->context_switch == NULL)
 505                 tool->context_switch = perf_event__process_switch;
 506         if (tool->ksymbol == NULL)
 507                 tool->ksymbol = perf_event__process_ksymbol;
 508         if (tool->bpf == NULL)
 509                 tool->bpf = perf_event__process_bpf;
 510         if (tool->text_poke == NULL)
 511                 tool->text_poke = perf_event__process_text_poke;
 512         if (tool->read == NULL)
 513                 tool->read = process_event_sample_stub;
 514         if (tool->throttle == NULL)
 515                 tool->throttle = process_event_stub;
 516         if (tool->unthrottle == NULL)
 517                 tool->unthrottle = process_event_stub;
 518         if (tool->attr == NULL)
 519                 tool->attr = process_event_synth_attr_stub;
 520         if (tool->event_update == NULL)
 521                 tool->event_update = process_event_synth_event_update_stub;
 522         if (tool->tracing_data == NULL)
 523                 tool->tracing_data = process_event_synth_tracing_data_stub;
 524         if (tool->build_id == NULL)
 525                 tool->build_id = process_event_op2_stub;
 526         if (tool->finished_round == NULL) {
 527                 if (tool->ordered_events)
 528                         tool->finished_round = process_finished_round;
 529                 else
 530                         tool->finished_round = process_finished_round_stub;
 531         }
 532         if (tool->id_index == NULL)
 533                 tool->id_index = process_event_op2_stub;
 534         if (tool->auxtrace_info == NULL)
 535                 tool->auxtrace_info = process_event_op2_stub;
 536         if (tool->auxtrace == NULL)
 537                 tool->auxtrace = process_event_auxtrace_stub;
 538         if (tool->auxtrace_error == NULL)
 539                 tool->auxtrace_error = process_event_op2_stub;
 540         if (tool->thread_map == NULL)
 541                 tool->thread_map = process_event_thread_map_stub;
 542         if (tool->cpu_map == NULL)
 543                 tool->cpu_map = process_event_cpu_map_stub;
 544         if (tool->stat_config == NULL)
 545                 tool->stat_config = process_event_stat_config_stub;
 546         if (tool->stat == NULL)
 547                 tool->stat = process_stat_stub;
 548         if (tool->stat_round == NULL)
 549                 tool->stat_round = process_stat_round_stub;
 550         if (tool->time_conv == NULL)
 551                 tool->time_conv = process_event_time_conv_stub;
 552         if (tool->feature == NULL)
 553                 tool->feature = process_event_op2_stub;
 554         if (tool->compressed == NULL)
 555                 tool->compressed = perf_session__process_compressed_event;
 556 }
 557
 558 static void swap_sample_id_all(union perf_event *event, void *data)
 559 {
 560         void *end = (void *) event + event->header.size;
 561         int size = end - data;
 562
 563         BUG_ON(size % sizeof(u64));
 564         mem_bswap_64(data, size);
 565 }
 566
 567 static void perf_event__all64_swap(union perf_event *event,
 568                                    bool sample_id_all __maybe_unused)
 569 {
 570         struct perf_event_header *hdr = &event->header;
 571         mem_bswap_64(hdr + 1, event->header.size - sizeof(*hdr));
 572 }
 573
 574 static void perf_event__comm_swap(union perf_event *event, bool sample_id_all)
 575 {
 576         event->comm.pid = bswap_32(event->comm.pid);
 577         event->comm.tid = bswap_32(event->comm.tid);
 578
 579         if (sample_id_all) {
 580                 void *data = &event->comm.comm;
 581
 582                 data += PERF_ALIGN(strlen(data) + 1, sizeof(u64));
 583                 swap_sample_id_all(event, data);
 584         }
 585 }
 586
 587 static void perf_event__mmap_swap(union perf_event *event,
 588                                   bool sample_id_all)
 589 {
 590         event->mmap.pid   = bswap_32(event->mmap.pid);
 591         event->mmap.tid   = bswap_32(event->mmap.tid);
 592         event->mmap.start = bswap_64(event->mmap.start);
 593         event->mmap.len   = bswap_64(event->mmap.len);
 594         event->mmap.pgoff = bswap_64(event->mmap.pgoff);
 595
 596         if (sample_id_all) {
 597                 void *data = &event->mmap.filename;
 598
 599                 data += PERF_ALIGN(strlen(data) + 1, sizeof(u64));
 600                 swap_sample_id_all(event, data);
 601         }
 602 }
 603
 604 static void perf_event__mmap2_swap(union perf_event *event,
 605                                   bool sample_id_all)
 606 {
 607         event->mmap2.pid   = bswap_32(event->mmap2.pid);
 608         event->mmap2.tid   = bswap_32(event->mmap2.tid);
 609         event->mmap2.start = bswap_64(event->mmap2.start);
 610         event->mmap2.len   = bswap_64(event->mmap2.len);
 611         event->mmap2.pgoff = bswap_64(event->mmap2.pgoff);
 612
 613         if (!(event->header.misc & PERF_RECORD_MISC_MMAP_BUILD_ID)) {
 614                 event->mmap2.maj   = bswap_32(event->mmap2.maj);
 615                 event->mmap2.min   = bswap_32(event->mmap2.min);
 616                 event->mmap2.ino   = bswap_64(event->mmap2.ino);
 617                 event->mmap2.ino_generation = bswap_64(event->mmap2.ino_generation);
 618         }
 619
 620         if (sample_id_all) {
 621                 void *data = &event->mmap2.filename;
 622
 623                 data += PERF_ALIGN(strlen(data) + 1, sizeof(u64));
 624                 swap_sample_id_all(event, data);
 625         }
 626 }
 627 static void perf_event__task_swap(union perf_event *event, bool sample_id_all)
 628 {
 629         event->fork.pid  = bswap_32(event->fork.pid);
 630         event->fork.tid  = bswap_32(event->fork.tid);
 631         event->fork.ppid = bswap_32(event->fork.ppid);
 632         event->fork.ptid = bswap_32(event->fork.ptid);
 633         event->fork.time = bswap_64(event->fork.time);
 634
 635         if (sample_id_all)
 636                 swap_sample_id_all(event, &event->fork + 1);
 637 }
 638
 639 static void perf_event__read_swap(union perf_event *event, bool sample_id_all)
 640 {
 641         event->read.pid          = bswap_32(event->read.pid);
 642         event->read.tid          = bswap_32(event->read.tid);
 643         event->read.value        = bswap_64(event->read.value);
 644         event->read.time_enabled = bswap_64(event->read.time_enabled);
 645         event->read.time_running = bswap_64(event->read.time_running);
 646         event->read.id           = bswap_64(event->read.id);
 647
 648         if (sample_id_all)
 649                 swap_sample_id_all(event, &event->read + 1);
 650 }
 651
 652 static void perf_event__aux_swap(union perf_event *event, bool sample_id_all)
 653 {
 654         event->aux.aux_offset = bswap_64(event->aux.aux_offset);
 655         event->aux.aux_size   = bswap_64(event->aux.aux_size);
 656         event->aux.flags      = bswap_64(event->aux.flags);
 657
 658         if (sample_id_all)
 659                 swap_sample_id_all(event, &event->aux + 1);
 660 }
 661
 662 static void perf_event__itrace_start_swap(union perf_event *event,
 663                                           bool sample_id_all)
 664 {
 665         event->itrace_start.pid  = bswap_32(event->itrace_start.pid);
 666         event->itrace_start.tid  = bswap_32(event->itrace_start.tid);
 667
 668         if (sample_id_all)
 669                 swap_sample_id_all(event, &event->itrace_start + 1);
 670 }
 671
 672 static void perf_event__switch_swap(union perf_event *event, bool sample_id_all)
 673 {
 674         if (event->header.type == PERF_RECORD_SWITCH_CPU_WIDE) {
 675                 event->context_switch.next_prev_pid =
 676                                 bswap_32(event->context_switch.next_prev_pid);
 677                 event->context_switch.next_prev_tid =
 678                                 bswap_32(event->context_switch.next_prev_tid);
 679         }
 680
 681         if (sample_id_all)
 682                 swap_sample_id_all(event, &event->context_switch + 1);
 683 }
 684
 685 static void perf_event__text_poke_swap(union perf_event *event, bool sample_id_all)
 686 {
 687         event->text_poke.addr    = bswap_64(event->text_poke.addr);
 688         event->text_poke.old_len = bswap_16(event->text_poke.old_len);
 689         event->text_poke.new_len = bswap_16(event->text_poke.new_len);
 690
 691         if (sample_id_all) {
 692                 size_t len = sizeof(event->text_poke.old_len) +
 693                              sizeof(event->text_poke.new_len) +
 694                              event->text_poke.old_len +
 695                              event->text_poke.new_len;
 696                 void *data = &event->text_poke.old_len;
 697
 698                 data += PERF_ALIGN(len, sizeof(u64));
 699                 swap_sample_id_all(event, data);
 700         }
 701 }
 702
 703 static void perf_event__throttle_swap(union perf_event *event,
 704                                       bool sample_id_all)
 705 {
 706         event->throttle.time      = bswap_64(event->throttle.time);
 707         event->throttle.id        = bswap_64(event->throttle.id);
 708         event->throttle.stream_id = bswap_64(event->throttle.stream_id);
 709
 710         if (sample_id_all)
 711                 swap_sample_id_all(event, &event->throttle + 1);
 712 }
 713
 714 static void perf_event__namespaces_swap(union perf_event *event,
 715                                         bool sample_id_all)
 716 {
 717         u64 i;
 718
 719         event->namespaces.pid           = bswap_32(event->namespaces.pid);
 720         event->namespaces.tid           = bswap_32(event->namespaces.tid);
 721         event->namespaces.nr_namespaces = bswap_64(event->namespaces.nr_namespaces);
 722
 723         for (i = 0; i < event->namespaces.nr_namespaces; i++) {
 724                 struct perf_ns_link_info *ns = &event->namespaces.link_info[i];
 725
 726                 ns->dev = bswap_64(ns->dev);
 727                 ns->ino = bswap_64(ns->ino);
 728         }
 729
 730         if (sample_id_all)
 731                 swap_sample_id_all(event, &event->namespaces.link_info[i]);
 732 }
 733
 734 static void perf_event__cgroup_swap(union perf_event *event, bool sample_id_all)
 735 {
 736         event->cgroup.id = bswap_64(event->cgroup.id);
 737
 738         if (sample_id_all) {
 739                 void *data = &event->cgroup.path;
 740
 741                 data += PERF_ALIGN(strlen(data) + 1, sizeof(u64));
 742                 swap_sample_id_all(event, data);
 743         }
 744 }
 745
 746 static u8 revbyte(u8 b)
 747 {
 748         int rev = (b >> 4) | ((b & 0xf) << 4);
 749         rev = ((rev & 0xcc) >> 2) | ((rev & 0x33) << 2);
 750         rev = ((rev & 0xaa) >> 1) | ((rev & 0x55) << 1);
 751         return (u8) rev;
 752 }
 753
 754 /*
 755  * XXX this is hack in attempt to carry flags bitfield
 756  * through endian village. ABI says:
 757  *
 758  * Bit-fields are allocated from right to left (least to most significant)
 759  * on little-endian implementations and from left to right (most to least
 760  * significant) on big-endian implementations.
 761  *
 762  * The above seems to be byte specific, so we need to reverse each
 763  * byte of the bitfield. 'Internet' also says this might be implementation
 764  * specific and we probably need proper fix and carry perf_event_attr
 765  * bitfield flags in separate data file FEAT_ section. Thought this seems
 766  * to work for now.
 767  */
 768 static void swap_bitfield(u8 *p, unsigned len)
 769 {
 770         unsigned i;
 771
 772         for (i = 0; i < len; i++) {
 773                 *p = revbyte(*p);
 774                 p++;
 775         }
 776 }
 777
 778 /* exported for swapping attributes in file header */
 779 void perf_event__attr_swap(struct perf_event_attr *attr)
 780 {
 781         attr->type              = bswap_32(attr->type);
 782         attr->size              = bswap_32(attr->size);
 783
 784 #define bswap_safe(f, n)                                        \
 785         (attr->size > (offsetof(struct perf_event_attr, f) +    \
 786                        sizeof(attr->f) * (n)))
 787 #define bswap_field(f, sz)                      \
 788 do {                                            \
 789         if (bswap_safe(f, 0))                   \
 790                 attr->f = bswap_##sz(attr->f);  \
 791 } while(0)
 792 #define bswap_field_16(f) bswap_field(f, 16)
 793 #define bswap_field_32(f) bswap_field(f, 32)
 794 #define bswap_field_64(f) bswap_field(f, 64)
 795
 796         bswap_field_64(config);
 797         bswap_field_64(sample_period);
 798         bswap_field_64(sample_type);
 799         bswap_field_64(read_format);
 800         bswap_field_32(wakeup_events);
 801         bswap_field_32(bp_type);
 802         bswap_field_64(bp_addr);
 803         bswap_field_64(bp_len);
 804         bswap_field_64(branch_sample_type);
 805         bswap_field_64(sample_regs_user);
 806         bswap_field_32(sample_stack_user);
 807         bswap_field_32(aux_watermark);
 808         bswap_field_16(sample_max_stack);
 809         bswap_field_32(aux_sample_size);
 810
 811         /*
 812          * After read_format are bitfields. Check read_format because
 813          * we are unable to use offsetof on bitfield.
 814          */
 815         if (bswap_safe(read_format, 1))
 816                 swap_bitfield((u8 *) (&attr->read_format + 1),
 817                               sizeof(u64));
 818 #undef bswap_field_64
 819 #undef bswap_field_32
 820 #undef bswap_field
 821 #undef bswap_safe
 822 }
 823
 824 static void perf_event__hdr_attr_swap(union perf_event *event,
 825                                       bool sample_id_all __maybe_unused)
 826 {
 827         size_t size;
 828
 829         perf_event__attr_swap(&event->attr.attr);
 830
 831         size = event->header.size;
 832         size -= (void *)&event->attr.id - (void *)event;
 833         mem_bswap_64(event->attr.id, size);
 834 }
 835
 836 static void perf_event__event_update_swap(union perf_event *event,
 837                                           bool sample_id_all __maybe_unused)
 838 {
 839         event->event_update.type = bswap_64(event->event_update.type);
 840         event->event_update.id   = bswap_64(event->event_update.id);
 841 }
 842
 843 static void perf_event__event_type_swap(union perf_event *event,
 844                                         bool sample_id_all __maybe_unused)
 845 {
 846         event->event_type.event_type.event_id =
 847                 bswap_64(event->event_type.event_type.event_id);
 848 }
 849
 850 static void perf_event__tracing_data_swap(union perf_event *event,
 851                                           bool sample_id_all __maybe_unused)
 852 {
 853         event->tracing_data.size = bswap_32(event->tracing_data.size);
 854 }
 855
 856 static void perf_event__auxtrace_info_swap(union perf_event *event,
 857                                            bool sample_id_all __maybe_unused)
 858 {
 859         size_t size;
 860
 861         event->auxtrace_info.type = bswap_32(event->auxtrace_info.type);
 862
 863         size = event->header.size;
 864         size -= (void *)&event->auxtrace_info.priv - (void *)event;
 865         mem_bswap_64(event->auxtrace_info.priv, size);
 866 }
 867
 868 static void perf_event__auxtrace_swap(union perf_event *event,
 869                                       bool sample_id_all __maybe_unused)
 870 {
 871         event->auxtrace.size      = bswap_64(event->auxtrace.size);
 872         event->auxtrace.offset    = bswap_64(event->auxtrace.offset);
 873         event->auxtrace.reference = bswap_64(event->auxtrace.reference);
 874         event->auxtrace.idx       = bswap_32(event->auxtrace.idx);
 875         event->auxtrace.tid       = bswap_32(event->auxtrace.tid);
 876         event->auxtrace.cpu       = bswap_32(event->auxtrace.cpu);
 877 }
 878
 879 static void perf_event__auxtrace_error_swap(union perf_event *event,
 880                                             bool sample_id_all __maybe_unused)
 881 {
 882         event->auxtrace_error.type = bswap_32(event->auxtrace_error.type);
 883         event->auxtrace_error.code = bswap_32(event->auxtrace_error.code);
 884         event->auxtrace_error.cpu  = bswap_32(event->auxtrace_error.cpu);
 885         event->auxtrace_error.pid  = bswap_32(event->auxtrace_error.pid);
 886         event->auxtrace_error.tid  = bswap_32(event->auxtrace_error.tid);
 887         event->auxtrace_error.fmt  = bswap_32(event->auxtrace_error.fmt);
 888         event->auxtrace_error.ip   = bswap_64(event->auxtrace_error.ip);
 889         if (event->auxtrace_error.fmt)
 890                 event->auxtrace_error.time = bswap_64(event->auxtrace_error.time);
 891 }
 892
 893 static void perf_event__thread_map_swap(union perf_event *event,
 894                                         bool sample_id_all __maybe_unused)
 895 {
 896         unsigned i;
 897
 898         event->thread_map.nr = bswap_64(event->thread_map.nr);
 899
 900         for (i = 0; i < event->thread_map.nr; i++)
 901                 event->thread_map.entries[i].pid = bswap_64(event->thread_map.entries[i].pid);
 902 }
 903
 904 static void perf_event__cpu_map_swap(union perf_event *event,
 905                                      bool sample_id_all __maybe_unused)
 906 {
 907         struct perf_record_cpu_map_data *data = &event->cpu_map.data;
 908         struct cpu_map_entries *cpus;
 909         struct perf_record_record_cpu_map *mask;
 910         unsigned i;
 911
 912         data->type = bswap_16(data->type);
 913
 914         switch (data->type) {
 915         case PERF_CPU_MAP__CPUS:
 916                 cpus = (struct cpu_map_entries *)data->data;
 917
 918                 cpus->nr = bswap_16(cpus->nr);
 919
 920                 for (i = 0; i < cpus->nr; i++)
 921                         cpus->cpu[i] = bswap_16(cpus->cpu[i]);
 922                 break;
 923         case PERF_CPU_MAP__MASK:
 924                 mask = (struct perf_record_record_cpu_map *)data->data;
 925
 926                 mask->nr = bswap_16(mask->nr);
 927                 mask->long_size = bswap_16(mask->long_size);
 928
 929                 switch (mask->long_size) {
 930                 case 4: mem_bswap_32(&mask->mask, mask->nr); break;
 931                 case 8: mem_bswap_64(&mask->mask, mask->nr); break;
 932                 default:
 933                         pr_err("cpu_map swap: unsupported long size\n");
 934                 }
 935         default:
 936                 break;
 937         }
 938 }
 939
 940 static void perf_event__stat_config_swap(union perf_event *event,
 941                                          bool sample_id_all __maybe_unused)
 942 {
 943         u64 size;
 944
 945         size  = bswap_64(event->stat_config.nr) * sizeof(event->stat_config.data[0]);
 946         size += 1; /* nr item itself */
 947         mem_bswap_64(&event->stat_config.nr, size);
 948 }
 949
 950 static void perf_event__stat_swap(union perf_event *event,
 951                                   bool sample_id_all __maybe_unused)
 952 {
 953         event->stat.id     = bswap_64(event->stat.id);
 954         event->stat.thread = bswap_32(event->stat.thread);
 955         event->stat.cpu    = bswap_32(event->stat.cpu);
 956         event->stat.val    = bswap_64(event->stat.val);
 957         event->stat.ena    = bswap_64(event->stat.ena);
 958         event->stat.run    = bswap_64(event->stat.run);
 959 }
 960
 961 static void perf_event__stat_round_swap(union perf_event *event,
 962                                         bool sample_id_all __maybe_unused)
 963 {
 964         event->stat_round.type = bswap_64(event->stat_round.type);
 965         event->stat_round.time = bswap_64(event->stat_round.time);
 966 }
 967
 968 static void perf_event__time_conv_swap(union perf_event *event,
 969                                        bool sample_id_all __maybe_unused)
 970 {
 971         event->time_conv.time_shift = bswap_64(event->time_conv.time_shift);
 972         event->time_conv.time_mult  = bswap_64(event->time_conv.time_mult);
 973         event->time_conv.time_zero  = bswap_64(event->time_conv.time_zero);
 974
 975         if (event_contains(event->time_conv, time_cycles)) {
 976                 event->time_conv.time_cycles = bswap_64(event->time_conv.time_cycles);
 977                 event->time_conv.time_mask = bswap_64(event->time_conv.time_mask);
 978         }
 979 }
 980
 981 typedef void (*perf_event__swap_op)(union perf_event *event,
 982                                     bool sample_id_all);
 983
 984 static perf_event__swap_op perf_event__swap_ops[] = {
 985         [PERF_RECORD_MMAP]                = perf_event__mmap_swap,
 986         [PERF_RECORD_MMAP2]               = perf_event__mmap2_swap,
 987         [PERF_RECORD_COMM]                = perf_event__comm_swap,
 988         [PERF_RECORD_FORK]                = perf_event__task_swap,
 989         [PERF_RECORD_EXIT]                = perf_event__task_swap,
 990         [PERF_RECORD_LOST]                = perf_event__all64_swap,
 991         [PERF_RECORD_READ]                = perf_event__read_swap,
 992         [PERF_RECORD_THROTTLE]            = perf_event__throttle_swap,
 993         [PERF_RECORD_UNTHROTTLE]          = perf_event__throttle_swap,
 994         [PERF_RECORD_SAMPLE]              = perf_event__all64_swap,
 995         [PERF_RECORD_AUX]                 = perf_event__aux_swap,
 996         [PERF_RECORD_ITRACE_START]        = perf_event__itrace_start_swap,
 997         [PERF_RECORD_LOST_SAMPLES]        = perf_event__all64_swap,
 998         [PERF_RECORD_SWITCH]              = perf_event__switch_swap,
 999         [PERF_RECORD_SWITCH_CPU_WIDE]     = perf_event__switch_swap,
1000         [PERF_RECORD_NAMESPACES]          = perf_event__namespaces_swap,
1001         [PERF_RECORD_CGROUP]              = perf_event__cgroup_swap,
1002         [PERF_RECORD_TEXT_POKE]           = perf_event__text_poke_swap,
1003         [PERF_RECORD_HEADER_ATTR]         = perf_event__hdr_attr_swap,
1004         [PERF_RECORD_HEADER_EVENT_TYPE]   = perf_event__event_type_swap,
1005         [PERF_RECORD_HEADER_TRACING_DATA] = perf_event__tracing_data_swap,
1006         [PERF_RECORD_HEADER_BUILD_ID]     = NULL,
1007         [PERF_RECORD_ID_INDEX]            = perf_event__all64_swap,
1008         [PERF_RECORD_AUXTRACE_INFO]       = perf_event__auxtrace_info_swap,
1009         [PERF_RECORD_AUXTRACE]            = perf_event__auxtrace_swap,
1010         [PERF_RECORD_AUXTRACE_ERROR]      = perf_event__auxtrace_error_swap,
1011         [PERF_RECORD_THREAD_MAP]          = perf_event__thread_map_swap,
1012         [PERF_RECORD_CPU_MAP]             = perf_event__cpu_map_swap,
1013         [PERF_RECORD_STAT_CONFIG]         = perf_event__stat_config_swap,
1014         [PERF_RECORD_STAT]                = perf_event__stat_swap,
1015         [PERF_RECORD_STAT_ROUND]          = perf_event__stat_round_swap,
1016         [PERF_RECORD_EVENT_UPDATE]        = perf_event__event_update_swap,
1017         [PERF_RECORD_TIME_CONV]           = perf_event__time_conv_swap,
1018         [PERF_RECORD_HEADER_MAX]          = NULL,
1019 };
1020
1021 /*
1022  * When perf record finishes a pass on every buffers, it records this pseudo
1023  * event.
1024  * We record the max timestamp t found in the pass n.
1025  * Assuming these timestamps are monotonic across cpus, we know that if
1026  * a buffer still has events with timestamps below t, they will be all
1027  * available and then read in the pass n + 1.
1028  * Hence when we start to read the pass n + 2, we can safely flush every
1029  * events with timestamps below t.
1030  *
1031  *    ============ PASS n =================
1032  *       CPU 0         |   CPU 1
1033  *                     |
1034  *    cnt1 timestamps  |   cnt2 timestamps
1035  *          1          |         2
1036  *          2          |         3
1037  *          -          |         4  <--- max recorded
1038  *
1039  *    ============ PASS n + 1 ==============
1040  *       CPU 0         |   CPU 1
1041  *                     |
1042  *    cnt1 timestamps  |   cnt2 timestamps
1043  *          3          |         5
1044  *          4          |         6
1045  *          5          |         7 <---- max recorded
1046  *
1047  *      Flush every events below timestamp 4
1048  *
1049  *    ============ PASS n + 2 ==============
1050  *       CPU 0         |   CPU 1
1051  *                     |
1052  *    cnt1 timestamps  |   cnt2 timestamps
1053  *          6          |         8
1054  *          7          |         9
1055  *          -          |         10
1056  *
1057  *      Flush every events below timestamp 7
1058  *      etc...
1059  */
1060 static int process_finished_round(struct perf_tool *tool __maybe_unused,
1061                                   union perf_event *event __maybe_unused,
1062                                   struct ordered_events *oe)
1063 {
1064         if (dump_trace)
1065                 fprintf(stdout, "\n");
1066         return ordered_events__flush(oe, OE_FLUSH__ROUND);
1067 }
1068
1069 int perf_session__queue_event(struct perf_session *s, union perf_event *event,
1070                               u64 timestamp, u64 file_offset)
1071 {
1072         return ordered_events__queue(&s->ordered_events, event, timestamp, file_offset);
1073 }
1074
1075 static void callchain__lbr_callstack_printf(struct perf_sample *sample)
1076 {
1077         struct ip_callchain *callchain = sample->callchain;
1078         struct branch_stack *lbr_stack = sample->branch_stack;
1079         struct branch_entry *entries = perf_sample__branch_entries(sample);
1080         u64 kernel_callchain_nr = callchain->nr;
1081         unsigned int i;
1082
1083         for (i = 0; i < kernel_callchain_nr; i++) {
1084                 if (callchain->ips[i] == PERF_CONTEXT_USER)
1085                         break;
1086         }
1087
1088         if ((i != kernel_callchain_nr) && lbr_stack->nr) {
1089                 u64 total_nr;
1090                 /*
1091                  * LBR callstack can only get user call chain,
1092                  * i is kernel call chain number,
1093                  * 1 is PERF_CONTEXT_USER.
1094                  *
1095                  * The user call chain is stored in LBR registers.
1096                  * LBR are pair registers. The caller is stored
1097                  * in "from" register, while the callee is stored
1098                  * in "to" register.
1099                  * For example, there is a call stack
1100                  * "A"->"B"->"C"->"D".
1101                  * The LBR registers will be recorded like
1102                  * "C"->"D", "B"->"C", "A"->"B".
1103                  * So only the first "to" register and all "from"
1104                  * registers are needed to construct the whole stack.
1105                  */
1106                 total_nr = i + 1 + lbr_stack->nr + 1;
1107                 kernel_callchain_nr = i + 1;
1108
1109                 printf("... LBR call chain: nr:%" PRIu64 "\n", total_nr);
1110
1111                 for (i = 0; i < kernel_callchain_nr; i++)
1112                         printf("..... %2d: %016" PRIx64 "\n",
1113                                i, callchain->ips[i]);
1114
1115                 printf("..... %2d: %016" PRIx64 "\n",
1116                        (int)(kernel_callchain_nr), entries[0].to);
1117                 for (i = 0; i < lbr_stack->nr; i++)
1118                         printf("..... %2d: %016" PRIx64 "\n",
1119                                (int)(i + kernel_callchain_nr + 1), entries[i].from);
1120         }
1121 }
1122
1123 static void callchain__printf(struct evsel *evsel,
1124                               struct perf_sample *sample)
1125 {
1126         unsigned int i;
1127         struct ip_callchain *callchain = sample->callchain;
1128
1129         if (evsel__has_branch_callstack(evsel))
1130                 callchain__lbr_callstack_printf(sample);
1131
1132         printf("... FP chain: nr:%" PRIu64 "\n", callchain->nr);
1133
1134         for (i = 0; i < callchain->nr; i++)
1135                 printf("..... %2d: %016" PRIx64 "\n",
1136                        i, callchain->ips[i]);
1137 }
1138
1139 static void branch_stack__printf(struct perf_sample *sample, bool callstack)
1140 {
1141         struct branch_entry *entries = perf_sample__branch_entries(sample);
1142         uint64_t i;
1143
1144         printf("%s: nr:%" PRIu64 "\n",
1145                 !callstack ? "... branch stack" : "... branch callstack",
1146                 sample->branch_stack->nr);
1147
1148         for (i = 0; i < sample->branch_stack->nr; i++) {
1149                 struct branch_entry *e = &entries[i];
1150
1151                 if (!callstack) {
1152                         printf("..... %2"PRIu64": %016" PRIx64 " -> %016" PRIx64 " %hu cycles %s%s%s%s %x\n",
1153                                 i, e->from, e->to,
1154                                 (unsigned short)e->flags.cycles,
1155                                 e->flags.mispred ? "M" : " ",
1156                                 e->flags.predicted ? "P" : " ",
1157                                 e->flags.abort ? "A" : " ",
1158                                 e->flags.in_tx ? "T" : " ",
1159                                 (unsigned)e->flags.reserved);
1160                 } else {
1161                         printf("..... %2"PRIu64": %016" PRIx64 "\n",
1162                                 i, i > 0 ? e->from : e->to);
1163                 }
1164         }
1165 }
1166
1167 static void regs_dump__printf(u64 mask, u64 *regs)
1168 {
1169         unsigned rid, i = 0;
1170
1171         for_each_set_bit(rid, (unsigned long *) &mask, sizeof(mask) * 8) {
1172                 u64 val = regs[i++];
1173
1174                 printf(".... %-5s 0x%016" PRIx64 "\n",
1175                        perf_reg_name(rid), val);
1176         }
1177 }
1178
1179 static const char *regs_abi[] = {
1180         [PERF_SAMPLE_REGS_ABI_NONE] = "none",
1181         [PERF_SAMPLE_REGS_ABI_32] = "32-bit",
1182         [PERF_SAMPLE_REGS_ABI_64] = "64-bit",
1183 };
1184
1185 static inline const char *regs_dump_abi(struct regs_dump *d)
1186 {
1187         if (d->abi > PERF_SAMPLE_REGS_ABI_64)
1188                 return "unknown";
1189
1190         return regs_abi[d->abi];
1191 }
1192
1193 static void regs__printf(const char *type, struct regs_dump *regs)
1194 {
1195         u64 mask = regs->mask;
1196
1197         printf("... %s regs: mask 0x%" PRIx64 " ABI %s\n",
1198                type,
1199                mask,
1200                regs_dump_abi(regs));
1201
1202         regs_dump__printf(mask, regs->regs);
1203 }
1204
1205 static void regs_user__printf(struct perf_sample *sample)
1206 {
1207         struct regs_dump *user_regs = &sample->user_regs;
1208
1209         if (user_regs->regs)
1210                 regs__printf("user", user_regs);
1211 }
1212
1213 static void regs_intr__printf(struct perf_sample *sample)
1214 {
1215         struct regs_dump *intr_regs = &sample->intr_regs;
1216
1217         if (intr_regs->regs)
1218                 regs__printf("intr", intr_regs);
1219 }
1220
1221 static void stack_user__printf(struct stack_dump *dump)
1222 {
1223         printf("... ustack: size %" PRIu64 ", offset 0x%x\n",
1224                dump->size, dump->offset);
1225 }
1226
1227 static void evlist__print_tstamp(struct evlist *evlist, union perf_event *event, struct perf_sample *sample)
1228 {
1229         u64 sample_type = __evlist__combined_sample_type(evlist);
1230
1231         if (event->header.type != PERF_RECORD_SAMPLE &&
1232             !evlist__sample_id_all(evlist)) {
1233                 fputs("-1 -1 ", stdout);
1234                 return;
1235         }
1236
1237         if ((sample_type & PERF_SAMPLE_CPU))
1238                 printf("%u ", sample->cpu);
1239
1240         if (sample_type & PERF_SAMPLE_TIME)
1241                 printf("%" PRIu64 " ", sample->time);
1242 }
1243
1244 static void sample_read__printf(struct perf_sample *sample, u64 read_format)
1245 {
1246         printf("... sample_read:\n");
1247
1248         if (read_format & PERF_FORMAT_TOTAL_TIME_ENABLED)
1249                 printf("...... time enabled %016" PRIx64 "\n",
1250                        sample->read.time_enabled);
1251
1252         if (read_format & PERF_FORMAT_TOTAL_TIME_RUNNING)
1253                 printf("...... time running %016" PRIx64 "\n",
1254                        sample->read.time_running);
1255
1256         if (read_format & PERF_FORMAT_GROUP) {
1257                 u64 i;
1258
1259                 printf(".... group nr %" PRIu64 "\n", sample->read.group.nr);
1260
1261                 for (i = 0; i < sample->read.group.nr; i++) {
1262                         struct sample_read_value *value;
1263
1264                         value = &sample->read.group.values[i];
1265                         printf("..... id %016" PRIx64
1266                                ", value %016" PRIx64 "\n",
1267                                value->id, value->value);
1268                 }
1269         } else
1270                 printf("..... id %016" PRIx64 ", value %016" PRIx64 "\n",
1271                         sample->read.one.id, sample->read.one.value);
1272 }
1273
1274 static void dump_event(struct evlist *evlist, union perf_event *event,
1275                        u64 file_offset, struct perf_sample *sample)
1276 {
1277         if (!dump_trace)
1278                 return;
1279
1280         printf("\n%#" PRIx64 " [%#x]: event: %d\n",
1281                file_offset, event->header.size, event->header.type);
1282
1283         trace_event(event);
1284         if (event->header.type == PERF_RECORD_SAMPLE && evlist->trace_event_sample_raw)
1285                 evlist->trace_event_sample_raw(evlist, event, sample);
1286
1287         if (sample)
1288                 evlist__print_tstamp(evlist, event, sample);
1289
1290         printf("%#" PRIx64 " [%#x]: PERF_RECORD_%s", file_offset,
1291                event->header.size, perf_event__name(event->header.type));
1292 }
1293
1294 char *get_page_size_name(u64 size, char *str)
1295 {
1296         if (!size || !unit_number__scnprintf(str, PAGE_SIZE_NAME_LEN, size))
1297                 snprintf(str, PAGE_SIZE_NAME_LEN, "%s", "N/A");
1298
1299         return str;
1300 }
1301
1302 static void dump_sample(struct evsel *evsel, union perf_event *event,
1303                         struct perf_sample *sample)
1304 {
1305         u64 sample_type;
1306         char str[PAGE_SIZE_NAME_LEN];
1307
1308         if (!dump_trace)
1309                 return;
1310
1311         printf("(IP, 0x%x): %d/%d: %#" PRIx64 " period: %" PRIu64 " addr: %#" PRIx64 "\n",
1312                event->header.misc, sample->pid, sample->tid, sample->ip,
1313                sample->period, sample->addr);
1314
1315         sample_type = evsel->core.attr.sample_type;
1316
1317         if (evsel__has_callchain(evsel))
1318                 callchain__printf(evsel, sample);
1319
1320         if (evsel__has_br_stack(evsel))
1321                 branch_stack__printf(sample, evsel__has_branch_callstack(evsel));
1322
1323         if (sample_type & PERF_SAMPLE_REGS_USER)
1324                 regs_user__printf(sample);
1325
1326         if (sample_type & PERF_SAMPLE_REGS_INTR)
1327                 regs_intr__printf(sample);
1328
1329         if (sample_type & PERF_SAMPLE_STACK_USER)
1330                 stack_user__printf(&sample->user_stack);
1331
1332         if (sample_type & PERF_SAMPLE_WEIGHT_TYPE) {
1333                 printf("... weight: %" PRIu64 "", sample->weight);
1334                         if (sample_type & PERF_SAMPLE_WEIGHT_STRUCT) {
1335                                 printf(",0x%"PRIx16"", sample->ins_lat);
1336                                 printf(",0x%"PRIx16"", sample->p_stage_cyc);
1337                         }
1338                 printf("\n");
1339         }
1340
1341         if (sample_type & PERF_SAMPLE_DATA_SRC)
1342                 printf(" . data_src: 0x%"PRIx64"\n", sample->data_src);
1343
1344         if (sample_type & PERF_SAMPLE_PHYS_ADDR)
1345                 printf(" .. phys_addr: 0x%"PRIx64"\n", sample->phys_addr);
1346
1347         if (sample_type & PERF_SAMPLE_DATA_PAGE_SIZE)
1348                 printf(" .. data page size: %s\n", get_page_size_name(sample->data_page_size, str));
1349
1350         if (sample_type & PERF_SAMPLE_CODE_PAGE_SIZE)
1351                 printf(" .. code page size: %s\n", get_page_size_name(sample->code_page_size, str));
1352
1353         if (sample_type & PERF_SAMPLE_TRANSACTION)
1354                 printf("... transaction: %" PRIx64 "\n", sample->transaction);
1355
1356         if (sample_type & PERF_SAMPLE_READ)
1357                 sample_read__printf(sample, evsel->core.attr.read_format);
1358 }
1359
1360 static void dump_read(struct evsel *evsel, union perf_event *event)
1361 {
1362         struct perf_record_read *read_event = &event->read;
1363         u64 read_format;
1364
1365         if (!dump_trace)
1366                 return;
1367
1368         printf(": %d %d %s %" PRI_lu64 "\n", event->read.pid, event->read.tid,
1369                evsel__name(evsel), event->read.value);
1370
1371         if (!evsel)
1372                 return;
1373
1374         read_format = evsel->core.attr.read_format;
1375
1376         if (read_format & PERF_FORMAT_TOTAL_TIME_ENABLED)
1377                 printf("... time enabled : %" PRI_lu64 "\n", read_event->time_enabled);
1378
1379         if (read_format & PERF_FORMAT_TOTAL_TIME_RUNNING)
1380                 printf("... time running : %" PRI_lu64 "\n", read_event->time_running);
1381
1382         if (read_format & PERF_FORMAT_ID)
1383                 printf("... id           : %" PRI_lu64 "\n", read_event->id);
1384 }
1385
1386 static struct machine *machines__find_for_cpumode(struct machines *machines,
1387                                                union perf_event *event,
1388                                                struct perf_sample *sample)
1389 {
1390         if (perf_guest &&
1391             ((sample->cpumode == PERF_RECORD_MISC_GUEST_KERNEL) ||
1392              (sample->cpumode == PERF_RECORD_MISC_GUEST_USER))) {
1393                 u32 pid;
1394
1395                 if (event->header.type == PERF_RECORD_MMAP
1396                     || event->header.type == PERF_RECORD_MMAP2)
1397                         pid = event->mmap.pid;
1398                 else
1399                         pid = sample->pid;
1400
1401                 return machines__find_guest(machines, pid);
1402         }
1403
1404         return &machines->host;
1405 }
1406
1407 static int deliver_sample_value(struct evlist *evlist,
1408                                 struct perf_tool *tool,
1409                                 union perf_event *event,
1410                                 struct perf_sample *sample,
1411                                 struct sample_read_value *v,
1412                                 struct machine *machine)
1413 {
1414         struct perf_sample_id *sid = evlist__id2sid(evlist, v->id);
1415         struct evsel *evsel;
1416
1417         if (sid) {
1418                 sample->id     = v->id;
1419                 sample->period = v->value - sid->period;
1420                 sid->period    = v->value;
1421         }
1422
1423         if (!sid || sid->evsel == NULL) {
1424                 ++evlist->stats.nr_unknown_id;
1425                 return 0;
1426         }
1427
1428         /*
1429          * There's no reason to deliver sample
1430          * for zero period, bail out.
1431          */
1432         if (!sample->period)
1433                 return 0;
1434
1435         evsel = container_of(sid->evsel, struct evsel, core);
1436         return tool->sample(tool, event, sample, evsel, machine);
1437 }
1438
1439 static int deliver_sample_group(struct evlist *evlist,
1440                                 struct perf_tool *tool,
1441                                 union  perf_event *event,
1442                                 struct perf_sample *sample,
1443                                 struct machine *machine)
1444 {
1445         int ret = -EINVAL;
1446         u64 i;
1447
1448         for (i = 0; i < sample->read.group.nr; i++) {
1449                 ret = deliver_sample_value(evlist, tool, event, sample,
1450                                            &sample->read.group.values[i],
1451                                            machine);
1452                 if (ret)
1453                         break;
1454         }
1455
1456         return ret;
1457 }
1458
1459 static int evlist__deliver_sample(struct evlist *evlist, struct perf_tool *tool,
1460                                   union  perf_event *event, struct perf_sample *sample,
1461                                   struct evsel *evsel, struct machine *machine)
1462 {
1463         /* We know evsel != NULL. */
1464         u64 sample_type = evsel->core.attr.sample_type;
1465         u64 read_format = evsel->core.attr.read_format;
1466
1467         /* Standard sample delivery. */
1468         if (!(sample_type & PERF_SAMPLE_READ))
1469                 return tool->sample(tool, event, sample, evsel, machine);
1470
1471         /* For PERF_SAMPLE_READ we have either single or group mode. */
1472         if (read_format & PERF_FORMAT_GROUP)
1473                 return deliver_sample_group(evlist, tool, event, sample,
1474                                             machine);
1475         else
1476                 return deliver_sample_value(evlist, tool, event, sample,
1477                                             &sample->read.one, machine);
1478 }
1479
1480 static int machines__deliver_event(struct machines *machines,
1481                                    struct evlist *evlist,
1482                                    union perf_event *event,
1483                                    struct perf_sample *sample,
1484                                    struct perf_tool *tool, u64 file_offset)
1485 {
1486         struct evsel *evsel;
1487         struct machine *machine;
1488
1489         dump_event(evlist, event, file_offset, sample);
1490
1491         evsel = evlist__id2evsel(evlist, sample->id);
1492
1493         machine = machines__find_for_cpumode(machines, event, sample);
1494
1495         switch (event->header.type) {
1496         case PERF_RECORD_SAMPLE:
1497                 if (evsel == NULL) {
1498                         ++evlist->stats.nr_unknown_id;
1499                         return 0;
1500                 }
1501                 dump_sample(evsel, event, sample);
1502                 if (machine == NULL) {
1503                         ++evlist->stats.nr_unprocessable_samples;
1504                         return 0;
1505                 }
1506                 return evlist__deliver_sample(evlist, tool, event, sample, evsel, machine);
1507         case PERF_RECORD_MMAP:
1508                 return tool->mmap(tool, event, sample, machine);
1509         case PERF_RECORD_MMAP2:
1510                 if (event->header.misc & PERF_RECORD_MISC_PROC_MAP_PARSE_TIMEOUT)
1511                         ++evlist->stats.nr_proc_map_timeout;
1512                 return tool->mmap2(tool, event, sample, machine);
1513         case PERF_RECORD_COMM:
1514                 return tool->comm(tool, event, sample, machine);
1515         case PERF_RECORD_NAMESPACES:
1516                 return tool->namespaces(tool, event, sample, machine);
1517         case PERF_RECORD_CGROUP:
1518                 return tool->cgroup(tool, event, sample, machine);
1519         case PERF_RECORD_FORK:
1520                 return tool->fork(tool, event, sample, machine);
1521         case PERF_RECORD_EXIT:
1522                 return tool->exit(tool, event, sample, machine);
1523         case PERF_RECORD_LOST:
1524                 if (tool->lost == perf_event__process_lost)
1525                         evlist->stats.total_lost += event->lost.lost;
1526                 return tool->lost(tool, event, sample, machine);
1527         case PERF_RECORD_LOST_SAMPLES:
1528                 if (tool->lost_samples == perf_event__process_lost_samples)
1529                         evlist->stats.total_lost_samples += event->lost_samples.lost;
1530                 return tool->lost_samples(tool, event, sample, machine);
1531         case PERF_RECORD_READ:
1532                 dump_read(evsel, event);
1533                 return tool->read(tool, event, sample, evsel, machine);
1534         case PERF_RECORD_THROTTLE:
1535                 return tool->throttle(tool, event, sample, machine);
1536         case PERF_RECORD_UNTHROTTLE:
1537                 return tool->unthrottle(tool, event, sample, machine);
1538         case PERF_RECORD_AUX:
1539                 if (tool->aux == perf_event__process_aux) {
1540                         if (event->aux.flags & PERF_AUX_FLAG_TRUNCATED)
1541                                 evlist->stats.total_aux_lost += 1;
1542                         if (event->aux.flags & PERF_AUX_FLAG_PARTIAL)
1543                                 evlist->stats.total_aux_partial += 1;
1544                         if (event->aux.flags & PERF_AUX_FLAG_COLLISION)
1545                                 evlist->stats.total_aux_collision += 1;
1546                 }
1547                 return tool->aux(tool, event, sample, machine);
1548         case PERF_RECORD_ITRACE_START:
1549                 return tool->itrace_start(tool, event, sample, machine);
1550         case PERF_RECORD_SWITCH:
1551         case PERF_RECORD_SWITCH_CPU_WIDE:
1552                 return tool->context_switch(tool, event, sample, machine);
1553         case PERF_RECORD_KSYMBOL:
1554                 return tool->ksymbol(tool, event, sample, machine);
1555         case PERF_RECORD_BPF_EVENT:
1556                 return tool->bpf(tool, event, sample, machine);
1557         case PERF_RECORD_TEXT_POKE:
1558                 return tool->text_poke(tool, event, sample, machine);
1559         default:
1560                 ++evlist->stats.nr_unknown_events;
1561                 return -1;
1562         }
1563 }
1564
1565 static int perf_session__deliver_event(struct perf_session *session,
1566                                        union perf_event *event,
1567                                        struct perf_tool *tool,
1568                                        u64 file_offset)
1569 {
1570         struct perf_sample sample;
1571         int ret = evlist__parse_sample(session->evlist, event, &sample);
1572
1573         if (ret) {
1574                 pr_err("Can't parse sample, err = %d\n", ret);
1575                 return ret;
1576         }
1577
1578         ret = auxtrace__process_event(session, event, &sample, tool);
1579         if (ret < 0)
1580                 return ret;
1581         if (ret > 0)
1582                 return 0;
1583
1584         ret = machines__deliver_event(&session->machines, session->evlist,
1585                                       event, &sample, tool, file_offset);
1586
1587         if (dump_trace && sample.aux_sample.size)
1588                 auxtrace__dump_auxtrace_sample(session, &sample);
1589
1590         return ret;
1591 }
1592
1593 static s64 perf_session__process_user_event(struct perf_session *session,
1594                                             union perf_event *event,
1595                                             u64 file_offset)
1596 {
1597         struct ordered_events *oe = &session->ordered_events;
1598         struct perf_tool *tool = session->tool;
1599         struct perf_sample sample = { .time = 0, };
1600         int fd = perf_data__fd(session->data);
1601         int err;
1602
1603         if (event->header.type != PERF_RECORD_COMPRESSED ||
1604             tool->compressed == perf_session__process_compressed_event_stub)
1605                 dump_event(session->evlist, event, file_offset, &sample);
1606
1607         /* These events are processed right away */
1608         switch (event->header.type) {
1609         case PERF_RECORD_HEADER_ATTR:
1610                 err = tool->attr(tool, event, &session->evlist);
1611                 if (err == 0) {
1612                         perf_session__set_id_hdr_size(session);
1613                         perf_session__set_comm_exec(session);
1614                 }
1615                 return err;
1616         case PERF_RECORD_EVENT_UPDATE:
1617                 return tool->event_update(tool, event, &session->evlist);
1618         case PERF_RECORD_HEADER_EVENT_TYPE:
1619                 /*
1620                  * Deprecated, but we need to handle it for sake
1621                  * of old data files create in pipe mode.
1622                  */
1623                 return 0;
1624         case PERF_RECORD_HEADER_TRACING_DATA:
1625                 /*
1626                  * Setup for reading amidst mmap, but only when we
1627                  * are in 'file' mode. The 'pipe' fd is in proper
1628                  * place already.
1629                  */
1630                 if (!perf_data__is_pipe(session->data))
1631                         lseek(fd, file_offset, SEEK_SET);
1632                 return tool->tracing_data(session, event);
1633         case PERF_RECORD_HEADER_BUILD_ID:
1634                 return tool->build_id(session, event);
1635         case PERF_RECORD_FINISHED_ROUND:
1636                 return tool->finished_round(tool, event, oe);
1637         case PERF_RECORD_ID_INDEX:
1638                 return tool->id_index(session, event);
1639         case PERF_RECORD_AUXTRACE_INFO:
1640                 return tool->auxtrace_info(session, event);
1641         case PERF_RECORD_AUXTRACE:
1642                 /* setup for reading amidst mmap */
1643                 lseek(fd, file_offset + event->header.size, SEEK_SET);
1644                 return tool->auxtrace(session, event);
1645         case PERF_RECORD_AUXTRACE_ERROR:
1646                 perf_session__auxtrace_error_inc(session, event);
1647                 return tool->auxtrace_error(session, event);
1648         case PERF_RECORD_THREAD_MAP:
1649                 return tool->thread_map(session, event);
1650         case PERF_RECORD_CPU_MAP:
1651                 return tool->cpu_map(session, event);
1652         case PERF_RECORD_STAT_CONFIG:
1653                 return tool->stat_config(session, event);
1654         case PERF_RECORD_STAT:
1655                 return tool->stat(session, event);
1656         case PERF_RECORD_STAT_ROUND:
1657                 return tool->stat_round(session, event);
1658         case PERF_RECORD_TIME_CONV:
1659                 session->time_conv = event->time_conv;
1660                 return tool->time_conv(session, event);
1661         case PERF_RECORD_HEADER_FEATURE:
1662                 return tool->feature(session, event);
1663         case PERF_RECORD_COMPRESSED:
1664                 err = tool->compressed(session, event, file_offset);
1665                 if (err)
1666                         dump_event(session->evlist, event, file_offset, &sample);
1667                 return err;
1668         default:
1669                 return -EINVAL;
1670         }
1671 }
1672
1673 int perf_session__deliver_synth_event(struct perf_session *session,
1674                                       union perf_event *event,
1675                                       struct perf_sample *sample)
1676 {
1677         struct evlist *evlist = session->evlist;
1678         struct perf_tool *tool = session->tool;
1679
1680         events_stats__inc(&evlist->stats, event->header.type);
1681
1682         if (event->header.type >= PERF_RECORD_USER_TYPE_START)
1683                 return perf_session__process_user_event(session, event, 0);
1684
1685         return machines__deliver_event(&session->machines, evlist, event, sample, tool, 0);
1686 }
1687
1688 static void event_swap(union perf_event *event, bool sample_id_all)
1689 {
1690         perf_event__swap_op swap;
1691
1692         swap = perf_event__swap_ops[event->header.type];
1693         if (swap)
1694                 swap(event, sample_id_all);
1695 }
1696
1697 int perf_session__peek_event(struct perf_session *session, off_t file_offset,
1698                              void *buf, size_t buf_sz,
1699                              union perf_event **event_ptr,
1700                              struct perf_sample *sample)
1701 {
1702         union perf_event *event;
1703         size_t hdr_sz, rest;
1704         int fd;
1705
1706         if (session->one_mmap && !session->header.needs_swap) {
1707                 event = file_offset - session->one_mmap_offset +
1708                         session->one_mmap_addr;
1709                 goto out_parse_sample;
1710         }
1711
1712         if (perf_data__is_pipe(session->data))
1713                 return -1;
1714
1715         fd = perf_data__fd(session->data);
1716         hdr_sz = sizeof(struct perf_event_header);
1717
1718         if (buf_sz < hdr_sz)
1719                 return -1;
1720
1721         if (lseek(fd, file_offset, SEEK_SET) == (off_t)-1 ||
1722             readn(fd, buf, hdr_sz) != (ssize_t)hdr_sz)
1723                 return -1;
1724
1725         event = (union perf_event *)buf;
1726
1727         if (session->header.needs_swap)
1728                 perf_event_header__bswap(&event->header);
1729
1730         if (event->header.size < hdr_sz || event->header.size > buf_sz)
1731                 return -1;
1732
1733         buf += hdr_sz;
1734         rest = event->header.size - hdr_sz;
1735
1736         if (readn(fd, buf, rest) != (ssize_t)rest)
1737                 return -1;
1738
1739         if (session->header.needs_swap)
1740                 event_swap(event, evlist__sample_id_all(session->evlist));
1741
1742 out_parse_sample:
1743
1744         if (sample && event->header.type < PERF_RECORD_USER_TYPE_START &&
1745             evlist__parse_sample(session->evlist, event, sample))
1746                 return -1;
1747
1748         *event_ptr = event;
1749
1750         return 0;
1751 }
1752
1753 int perf_session__peek_events(struct perf_session *session, u64 offset,
1754                               u64 size, peek_events_cb_t cb, void *data)
1755 {
1756         u64 max_offset = offset + size;
1757         char buf[PERF_SAMPLE_MAX_SIZE];
1758         union perf_event *event;
1759         int err;
1760
1761         do {
1762                 err = perf_session__peek_event(session, offset, buf,
1763                                                PERF_SAMPLE_MAX_SIZE, &event,
1764                                                NULL);
1765                 if (err)
1766                         return err;
1767
1768                 err = cb(session, event, offset, data);
1769                 if (err)
1770                         return err;
1771
1772                 offset += event->header.size;
1773                 if (event->header.type == PERF_RECORD_AUXTRACE)
1774                         offset += event->auxtrace.size;
1775
1776         } while (offset < max_offset);
1777
1778         return err;
1779 }
1780
1781 static s64 perf_session__process_event(struct perf_session *session,
1782                                        union perf_event *event, u64 file_offset)
1783 {
1784         struct evlist *evlist = session->evlist;
1785         struct perf_tool *tool = session->tool;
1786         int ret;
1787
1788         if (session->header.needs_swap)
1789                 event_swap(event, evlist__sample_id_all(evlist));
1790
1791         if (event->header.type >= PERF_RECORD_HEADER_MAX)
1792                 return -EINVAL;
1793
1794         events_stats__inc(&evlist->stats, event->header.type);
1795
1796         if (event->header.type >= PERF_RECORD_USER_TYPE_START)
1797                 return perf_session__process_user_event(session, event, file_offset);
1798
1799         if (tool->ordered_events) {
1800                 u64 timestamp = -1ULL;
1801
1802                 ret = evlist__parse_sample_timestamp(evlist, event, &timestamp);
1803                 if (ret && ret != -1)
1804                         return ret;
1805
1806                 ret = perf_session__queue_event(session, event, timestamp, file_offset);
1807                 if (ret != -ETIME)
1808                         return ret;
1809         }
1810
1811         return perf_session__deliver_event(session, event, tool, file_offset);
1812 }
1813
1814 void perf_event_header__bswap(struct perf_event_header *hdr)
1815 {
1816         hdr->type = bswap_32(hdr->type);
1817         hdr->misc = bswap_16(hdr->misc);
1818         hdr->size = bswap_16(hdr->size);
1819 }
1820
1821 struct thread *perf_session__findnew(struct perf_session *session, pid_t pid)
1822 {
1823         return machine__findnew_thread(&session->machines.host, -1, pid);
1824 }
1825
1826 int perf_session__register_idle_thread(struct perf_session *session)
1827 {
1828         struct thread *thread = machine__idle_thread(&session->machines.host);
1829
1830         /* machine__idle_thread() got the thread, so put it */
1831         thread__put(thread);
1832         return thread ? 0 : -1;
1833 }
1834
1835 static void
1836 perf_session__warn_order(const struct perf_session *session)
1837 {
1838         const struct ordered_events *oe = &session->ordered_events;
1839         struct evsel *evsel;
1840         bool should_warn = true;
1841
1842         evlist__for_each_entry(session->evlist, evsel) {
1843                 if (evsel->core.attr.write_backward)
1844                         should_warn = false;
1845         }
1846
1847         if (!should_warn)
1848                 return;
1849         if (oe->nr_unordered_events != 0)
1850                 ui__warning("%u out of order events recorded.\n", oe->nr_unordered_events);
1851 }
1852
1853 static void perf_session__warn_about_errors(const struct perf_session *session)
1854 {
1855         const struct events_stats *stats = &session->evlist->stats;
1856
1857         if (session->tool->lost == perf_event__process_lost &&
1858             stats->nr_events[PERF_RECORD_LOST] != 0) {
1859                 ui__warning("Processed %d events and lost %d chunks!\n\n"
1860                             "Check IO/CPU overload!\n\n",
1861                             stats->nr_events[0],
1862                             stats->nr_events[PERF_RECORD_LOST]);
1863         }
1864
1865         if (session->tool->lost_samples == perf_event__process_lost_samples) {
1866                 double drop_rate;
1867
1868                 drop_rate = (double)stats->total_lost_samples /
1869                             (double) (stats->nr_events[PERF_RECORD_SAMPLE] + stats->total_lost_samples);
1870                 if (drop_rate > 0.05) {
1871                         ui__warning("Processed %" PRIu64 " samples and lost %3.2f%%!\n\n",
1872                                     stats->nr_events[PERF_RECORD_SAMPLE] + stats->total_lost_samples,
1873                                     drop_rate * 100.0);
1874                 }
1875         }
1876
1877         if (session->tool->aux == perf_event__process_aux &&
1878             stats->total_aux_lost != 0) {
1879                 ui__warning("AUX data lost %" PRIu64 " times out of %u!\n\n",
1880                             stats->total_aux_lost,
1881                             stats->nr_events[PERF_RECORD_AUX]);
1882         }
1883
1884         if (session->tool->aux == perf_event__process_aux &&
1885             stats->total_aux_partial != 0) {
1886                 bool vmm_exclusive = false;
1887
1888                 (void)sysfs__read_bool("module/kvm_intel/parameters/vmm_exclusive",
1889                                        &vmm_exclusive);
1890
1891                 ui__warning("AUX data had gaps in it %" PRIu64 " times out of %u!\n\n"
1892                             "Are you running a KVM guest in the background?%s\n\n",
1893                             stats->total_aux_partial,
1894                             stats->nr_events[PERF_RECORD_AUX],
1895                             vmm_exclusive ?
1896                             "\nReloading kvm_intel module with vmm_exclusive=0\n"
1897                             "will reduce the gaps to only guest's timeslices." :
1898                             "");
1899         }
1900
1901         if (session->tool->aux == perf_event__process_aux &&
1902             stats->total_aux_collision != 0) {
1903                 ui__warning("AUX data detected collision  %" PRIu64 " times out of %u!\n\n",
1904                             stats->total_aux_collision,
1905                             stats->nr_events[PERF_RECORD_AUX]);
1906         }
1907
1908         if (stats->nr_unknown_events != 0) {
1909                 ui__warning("Found %u unknown events!\n\n"
1910                             "Is this an older tool processing a perf.data "
1911                             "file generated by a more recent tool?\n\n"
1912                             "If that is not the case, consider "
1913                             "reporting to linux-kernel@vger.kernel.org.\n\n",
1914                             stats->nr_unknown_events);
1915         }
1916
1917         if (stats->nr_unknown_id != 0) {
1918                 ui__warning("%u samples with id not present in the header\n",
1919                             stats->nr_unknown_id);
1920         }
1921
1922         if (stats->nr_invalid_chains != 0) {
1923                 ui__warning("Found invalid callchains!\n\n"
1924                             "%u out of %u events were discarded for this reason.\n\n"
1925                             "Consider reporting to linux-kernel@vger.kernel.org.\n\n",
1926                             stats->nr_invalid_chains,
1927                             stats->nr_events[PERF_RECORD_SAMPLE]);
1928         }
1929
1930         if (stats->nr_unprocessable_samples != 0) {
1931                 ui__warning("%u unprocessable samples recorded.\n"
1932                             "Do you have a KVM guest running and not using 'perf kvm'?\n",
1933                             stats->nr_unprocessable_samples);
1934         }
1935
1936         perf_session__warn_order(session);
1937
1938         events_stats__auxtrace_error_warn(stats);
1939
1940         if (stats->nr_proc_map_timeout != 0) {
1941                 ui__warning("%d map information files for pre-existing threads were\n"
1942                             "not processed, if there are samples for addresses they\n"
1943                             "will not be resolved, you may find out which are these\n"
1944                             "threads by running with -v and redirecting the output\n"
1945                             "to a file.\n"
1946                             "The time limit to process proc map is too short?\n"
1947                             "Increase it by --proc-map-timeout\n",
1948                             stats->nr_proc_map_timeout);
1949         }
1950 }
1951
1952 static int perf_session__flush_thread_stack(struct thread *thread,
1953                                             void *p __maybe_unused)
1954 {
1955         return thread_stack__flush(thread);
1956 }
1957
1958 static int perf_session__flush_thread_stacks(struct perf_session *session)
1959 {
1960         return machines__for_each_thread(&session->machines,
1961                                          perf_session__flush_thread_stack,
1962                                          NULL);
1963 }
1964
1965 volatile int session_done;
1966
1967 static int __perf_session__process_decomp_events(struct perf_session *session);
1968
1969 static int __perf_session__process_pipe_events(struct perf_session *session)
1970 {
1971         struct ordered_events *oe = &session->ordered_events;
1972         struct perf_tool *tool = session->tool;
1973         union perf_event *event;
1974         uint32_t size, cur_size = 0;
1975         void *buf = NULL;
1976         s64 skip = 0;
1977         u64 head;
1978         ssize_t err;
1979         void *p;
1980
1981         perf_tool__fill_defaults(tool);
1982
1983         head = 0;
1984         cur_size = sizeof(union perf_event);
1985
1986         buf = malloc(cur_size);
1987         if (!buf)
1988                 return -errno;
1989         ordered_events__set_copy_on_queue(oe, true);
1990 more:
1991         event = buf;
1992         err = perf_data__read(session->data, event,
1993                               sizeof(struct perf_event_header));
1994         if (err <= 0) {
1995                 if (err == 0)
1996                         goto done;
1997
1998                 pr_err("failed to read event header\n");
1999                 goto out_err;
2000         }
2001
2002         if (session->header.needs_swap)
2003                 perf_event_header__bswap(&event->header);
2004
2005         size = event->header.size;
2006         if (size < sizeof(struct perf_event_header)) {
2007                 pr_err("bad event header size\n");
2008                 goto out_err;
2009         }
2010
2011         if (size > cur_size) {
2012                 void *new = realloc(buf, size);
2013                 if (!new) {
2014                         pr_err("failed to allocate memory to read event\n");
2015                         goto out_err;
2016                 }
2017                 buf = new;
2018                 cur_size = size;
2019                 event = buf;
2020         }
2021         p = event;
2022         p += sizeof(struct perf_event_header);
2023
2024         if (size - sizeof(struct perf_event_header)) {
2025                 err = perf_data__read(session->data, p,
2026                                       size - sizeof(struct perf_event_header));
2027                 if (err <= 0) {
2028                         if (err == 0) {
2029                                 pr_err("unexpected end of event stream\n");
2030                                 goto done;
2031                         }
2032
2033                         pr_err("failed to read event data\n");
2034                         goto out_err;
2035                 }
2036         }
2037
2038         if ((skip = perf_session__process_event(session, event, head)) < 0) {
2039                 pr_err("%#" PRIx64 " [%#x]: failed to process type: %d\n",
2040                        head, event->header.size, event->header.type);
2041                 err = -EINVAL;
2042                 goto out_err;
2043         }
2044
2045         head += size;
2046
2047         if (skip > 0)
2048                 head += skip;
2049
2050         err = __perf_session__process_decomp_events(session);
2051         if (err)
2052                 goto out_err;
2053
2054         if (!session_done())
2055                 goto more;
2056 done:
2057         /* do the final flush for ordered samples */
2058         err = ordered_events__flush(oe, OE_FLUSH__FINAL);
2059         if (err)
2060                 goto out_err;
2061         err = auxtrace__flush_events(session, tool);
2062         if (err)
2063                 goto out_err;
2064         err = perf_session__flush_thread_stacks(session);
2065 out_err:
2066         free(buf);
2067         if (!tool->no_warn)
2068                 perf_session__warn_about_errors(session);
2069         ordered_events__free(&session->ordered_events);
2070         auxtrace__free_events(session);
2071         return err;
2072 }
2073
2074 static union perf_event *
2075 prefetch_event(char *buf, u64 head, size_t mmap_size,
2076                bool needs_swap, union perf_event *error)
2077 {
2078         union perf_event *event;
2079
2080         /*
2081          * Ensure we have enough space remaining to read
2082          * the size of the event in the headers.
2083          */
2084         if (head + sizeof(event->header) > mmap_size)
2085                 return NULL;
2086
2087         event = (union perf_event *)(buf + head);
2088         if (needs_swap)
2089                 perf_event_header__bswap(&event->header);
2090
2091         if (head + event->header.size <= mmap_size)
2092                 return event;
2093
2094         /* We're not fetching the event so swap back again */
2095         if (needs_swap)
2096                 perf_event_header__bswap(&event->header);
2097
2098         pr_debug("%s: head=%#" PRIx64 " event->header_size=%#x, mmap_size=%#zx:"
2099                  " fuzzed or compressed perf.data?\n",__func__, head, event->header.size, mmap_size);
2100
2101         return error;
2102 }
2103
2104 static union perf_event *
2105 fetch_mmaped_event(u64 head, size_t mmap_size, char *buf, bool needs_swap)
2106 {
2107         return prefetch_event(buf, head, mmap_size, needs_swap, ERR_PTR(-EINVAL));
2108 }
2109
2110 static union perf_event *
2111 fetch_decomp_event(u64 head, size_t mmap_size, char *buf, bool needs_swap)
2112 {
2113         return prefetch_event(buf, head, mmap_size, needs_swap, NULL);
2114 }
2115
2116 static int __perf_session__process_decomp_events(struct perf_session *session)
2117 {
2118         s64 skip;
2119         u64 size, file_pos = 0;
2120         struct decomp *decomp = session->decomp_last;
2121
2122         if (!decomp)
2123                 return 0;
2124
2125         while (decomp->head < decomp->size && !session_done()) {
2126                 union perf_event *event = fetch_decomp_event(decomp->head, decomp->size, decomp->data,
2127                                                              session->header.needs_swap);
2128
2129                 if (!event)
2130                         break;
2131
2132                 size = event->header.size;
2133
2134                 if (size < sizeof(struct perf_event_header) ||
2135                     (skip = perf_session__process_event(session, event, file_pos)) < 0) {
2136                         pr_err("%#" PRIx64 " [%#x]: failed to process type: %d\n",
2137                                 decomp->file_pos + decomp->head, event->header.size, event->header.type);
2138                         return -EINVAL;
2139                 }
2140
2141                 if (skip)
2142                         size += skip;
2143
2144                 decomp->head += size;
2145         }
2146
2147         return 0;
2148 }
2149
2150 /*
2151  * On 64bit we can mmap the data file in one go. No need for tiny mmap
2152  * slices. On 32bit we use 32MB.
2153  */
2154 #if BITS_PER_LONG == 64
2155 #define MMAP_SIZE ULLONG_MAX
2156 #define NUM_MMAPS 1
2157 #else
2158 #define MMAP_SIZE (32 * 1024 * 1024ULL)
2159 #define NUM_MMAPS 128
2160 #endif
2161
2162 struct reader;
2163
2164 typedef s64 (*reader_cb_t)(struct perf_session *session,
2165                            union perf_event *event,
2166                            u64 file_offset);
2167
2168 struct reader {
2169         int              fd;
2170         u64              data_size;
2171         u64              data_offset;
2172         reader_cb_t      process;
2173         bool             in_place_update;
2174 };
2175
2176 static int
2177 reader__process_events(struct reader *rd, struct perf_session *session,
2178                        struct ui_progress *prog)
2179 {
2180         u64 data_size = rd->data_size;
2181         u64 head, page_offset, file_offset, file_pos, size;
2182         int err = 0, mmap_prot, mmap_flags, map_idx = 0;
2183         size_t  mmap_size;
2184         char *buf, *mmaps[NUM_MMAPS];
2185         union perf_event *event;
2186         s64 skip;
2187
2188         page_offset = page_size * (rd->data_offset / page_size);
2189         file_offset = page_offset;
2190         head = rd->data_offset - page_offset;
2191
2192         ui_progress__init_size(prog, data_size, "Processing events...");
2193
2194         data_size += rd->data_offset;
2195
2196         mmap_size = MMAP_SIZE;
2197         if (mmap_size > data_size) {
2198                 mmap_size = data_size;
2199                 session->one_mmap = true;
2200         }
2201
2202         memset(mmaps, 0, sizeof(mmaps));
2203
2204         mmap_prot  = PROT_READ;
2205         mmap_flags = MAP_SHARED;
2206
2207         if (rd->in_place_update) {
2208                 mmap_prot  |= PROT_WRITE;
2209         } else if (session->header.needs_swap) {
2210                 mmap_prot  |= PROT_WRITE;
2211                 mmap_flags = MAP_PRIVATE;
2212         }
2213 remap:
2214         buf = mmap(NULL, mmap_size, mmap_prot, mmap_flags, rd->fd,
2215                    file_offset);
2216         if (buf == MAP_FAILED) {
2217                 pr_err("failed to mmap file\n");
2218                 err = -errno;
2219                 goto out;
2220         }
2221         mmaps[map_idx] = buf;
2222         map_idx = (map_idx + 1) & (ARRAY_SIZE(mmaps) - 1);
2223         file_pos = file_offset + head;
2224         if (session->one_mmap) {
2225                 session->one_mmap_addr = buf;
2226                 session->one_mmap_offset = file_offset;
2227         }
2228
2229 more:
2230         event = fetch_mmaped_event(head, mmap_size, buf, session->header.needs_swap);
2231         if (IS_ERR(event))
2232                 return PTR_ERR(event);
2233
2234         if (!event) {
2235                 if (mmaps[map_idx]) {
2236                         munmap(mmaps[map_idx], mmap_size);
2237                         mmaps[map_idx] = NULL;
2238                 }
2239
2240                 page_offset = page_size * (head / page_size);
2241                 file_offset += page_offset;
2242                 head -= page_offset;
2243                 goto remap;
2244         }
2245
2246         size = event->header.size;
2247
2248         skip = -EINVAL;
2249
2250         if (size < sizeof(struct perf_event_header) ||
2251             (skip = rd->process(session, event, file_pos)) < 0) {
2252                 pr_err("%#" PRIx64 " [%#x]: failed to process type: %d [%s]\n",
2253                        file_offset + head, event->header.size,
2254                        event->header.type, strerror(-skip));
2255                 err = skip;
2256                 goto out;
2257         }
2258
2259         if (skip)
2260                 size += skip;
2261
2262         head += size;
2263         file_pos += size;
2264
2265         err = __perf_session__process_decomp_events(session);
2266         if (err)
2267                 goto out;
2268
2269         ui_progress__update(prog, size);
2270
2271         if (session_done())
2272                 goto out;
2273
2274         if (file_pos < data_size)
2275                 goto more;
2276
2277 out:
2278         return err;
2279 }
2280
2281 static s64 process_simple(struct perf_session *session,
2282                           union perf_event *event,
2283                           u64 file_offset)
2284 {
2285         return perf_session__process_event(session, event, file_offset);
2286 }
2287
2288 static int __perf_session__process_events(struct perf_session *session)
2289 {
2290         struct reader rd = {
2291                 .fd             = perf_data__fd(session->data),
2292                 .data_size      = session->header.data_size,
2293                 .data_offset    = session->header.data_offset,
2294                 .process        = process_simple,
2295                 .in_place_update = session->data->in_place_update,
2296         };
2297         struct ordered_events *oe = &session->ordered_events;
2298         struct perf_tool *tool = session->tool;
2299         struct ui_progress prog;
2300         int err;
2301
2302         perf_tool__fill_defaults(tool);
2303
2304         if (rd.data_size == 0)
2305                 return -1;
2306
2307         ui_progress__init_size(&prog, rd.data_size, "Processing events...");
2308
2309         err = reader__process_events(&rd, session, &prog);
2310         if (err)
2311                 goto out_err;
2312         /* do the final flush for ordered samples */
2313         err = ordered_events__flush(oe, OE_FLUSH__FINAL);
2314         if (err)
2315                 goto out_err;
2316         err = auxtrace__flush_events(session, tool);
2317         if (err)
2318                 goto out_err;
2319         err = perf_session__flush_thread_stacks(session);
2320 out_err:
2321         ui_progress__finish();
2322         if (!tool->no_warn)
2323                 perf_session__warn_about_errors(session);
2324         /*
2325          * We may switching perf.data output, make ordered_events
2326          * reusable.
2327          */
2328         ordered_events__reinit(&session->ordered_events);
2329         auxtrace__free_events(session);
2330         session->one_mmap = false;
2331         return err;
2332 }
2333
2334 int perf_session__process_events(struct perf_session *session)
2335 {
2336         if (perf_session__register_idle_thread(session) < 0)
2337                 return -ENOMEM;
2338
2339         if (perf_data__is_pipe(session->data))
2340                 return __perf_session__process_pipe_events(session);
2341
2342         return __perf_session__process_events(session);
2343 }
2344
2345 bool perf_session__has_traces(struct perf_session *session, const char *msg)
2346 {
2347         struct evsel *evsel;
2348
2349         evlist__for_each_entry(session->evlist, evsel) {
2350                 if (evsel->core.attr.type == PERF_TYPE_TRACEPOINT)
2351                         return true;
2352         }
2353
2354         pr_err("No trace sample to read. Did you call 'perf %s'?\n", msg);
2355         return false;
2356 }
2357
2358 int map__set_kallsyms_ref_reloc_sym(struct map *map, const char *symbol_name, u64 addr)
2359 {
2360         char *bracket;
2361         struct ref_reloc_sym *ref;
2362         struct kmap *kmap;
2363
2364         ref = zalloc(sizeof(struct ref_reloc_sym));
2365         if (ref == NULL)
2366                 return -ENOMEM;
2367
2368         ref->name = strdup(symbol_name);
2369         if (ref->name == NULL) {
2370                 free(ref);
2371                 return -ENOMEM;
2372         }
2373
2374         bracket = strchr(ref->name, ']');
2375         if (bracket)
2376                 *bracket = '\0';
2377
2378         ref->addr = addr;
2379
2380         kmap = map__kmap(map);
2381         if (kmap)
2382                 kmap->ref_reloc_sym = ref;
2383
2384         return 0;
2385 }
2386
2387 size_t perf_session__fprintf_dsos(struct perf_session *session, FILE *fp)
2388 {
2389         return machines__fprintf_dsos(&session->machines, fp);
2390 }
2391
2392 size_t perf_session__fprintf_dsos_buildid(struct perf_session *session, FILE *fp,
2393                                           bool (skip)(struct dso *dso, int parm), int parm)
2394 {
2395         return machines__fprintf_dsos_buildid(&session->machines, fp, skip, parm);
2396 }
2397
2398 size_t perf_session__fprintf_nr_events(struct perf_session *session, FILE *fp,
2399                                        bool skip_empty)
2400 {
2401         size_t ret;
2402         const char *msg = "";
2403
2404         if (perf_header__has_feat(&session->header, HEADER_AUXTRACE))
2405                 msg = " (excludes AUX area (e.g. instruction trace) decoded / synthesized events)";
2406
2407         ret = fprintf(fp, "\nAggregated stats:%s\n", msg);
2408
2409         ret += events_stats__fprintf(&session->evlist->stats, fp, skip_empty);
2410         return ret;
2411 }
2412
2413 size_t perf_session__fprintf(struct perf_session *session, FILE *fp)
2414 {
2415         /*
2416          * FIXME: Here we have to actually print all the machines in this
2417          * session, not just the host...
2418          */
2419         return machine__fprintf(&session->machines.host, fp);
2420 }
2421
2422 struct evsel *perf_session__find_first_evtype(struct perf_session *session,
2423                                               unsigned int type)
2424 {
2425         struct evsel *pos;
2426
2427         evlist__for_each_entry(session->evlist, pos) {
2428                 if (pos->core.attr.type == type)
2429                         return pos;
2430         }
2431         return NULL;
2432 }
2433
2434 int perf_session__cpu_bitmap(struct perf_session *session,
2435                              const char *cpu_list, unsigned long *cpu_bitmap)
2436 {
2437         int i, err = -1;
2438         struct perf_cpu_map *map;
2439         int nr_cpus = min(session->header.env.nr_cpus_avail, MAX_NR_CPUS);
2440
2441         for (i = 0; i < PERF_TYPE_MAX; ++i) {
2442                 struct evsel *evsel;
2443
2444                 evsel = perf_session__find_first_evtype(session, i);
2445                 if (!evsel)
2446                         continue;
2447
2448                 if (!(evsel->core.attr.sample_type & PERF_SAMPLE_CPU)) {
2449                         pr_err("File does not contain CPU events. "
2450                                "Remove -C option to proceed.\n");
2451                         return -1;
2452                 }
2453         }
2454
2455         map = perf_cpu_map__new(cpu_list);
2456         if (map == NULL) {
2457                 pr_err("Invalid cpu_list\n");
2458                 return -1;
2459         }
2460
2461         for (i = 0; i < map->nr; i++) {
2462                 int cpu = map->map[i];
2463
2464                 if (cpu >= nr_cpus) {
2465                         pr_err("Requested CPU %d too large. "
2466                                "Consider raising MAX_NR_CPUS\n", cpu);
2467                         goto out_delete_map;
2468                 }
2469
2470                 set_bit(cpu, cpu_bitmap);
2471         }
2472
2473         err = 0;
2474
2475 out_delete_map:
2476         perf_cpu_map__put(map);
2477         return err;
2478 }
2479
2480 void perf_session__fprintf_info(struct perf_session *session, FILE *fp,
2481                                 bool full)
2482 {
2483         if (session == NULL || fp == NULL)
2484                 return;
2485
2486         fprintf(fp, "# ========\n");
2487         perf_header__fprintf_info(session, fp, full);
2488         fprintf(fp, "# ========\n#\n");
2489 }
2490
2491 int perf_event__process_id_index(struct perf_session *session,
2492                                  union perf_event *event)
2493 {
2494         struct evlist *evlist = session->evlist;
2495         struct perf_record_id_index *ie = &event->id_index;
2496         size_t i, nr, max_nr;
2497
2498         max_nr = (ie->header.size - sizeof(struct perf_record_id_index)) /
2499                  sizeof(struct id_index_entry);
2500         nr = ie->nr;
2501         if (nr > max_nr)
2502                 return -EINVAL;
2503
2504         if (dump_trace)
2505                 fprintf(stdout, " nr: %zu\n", nr);
2506
2507         for (i = 0; i < nr; i++) {
2508                 struct id_index_entry *e = &ie->entries[i];
2509                 struct perf_sample_id *sid;
2510
2511                 if (dump_trace) {
2512                         fprintf(stdout, " ... id: %"PRI_lu64, e->id);
2513                         fprintf(stdout, "  idx: %"PRI_lu64, e->idx);
2514                         fprintf(stdout, "  cpu: %"PRI_ld64, e->cpu);
2515                         fprintf(stdout, "  tid: %"PRI_ld64"\n", e->tid);
2516                 }
2517
2518                 sid = evlist__id2sid(evlist, e->id);
2519                 if (!sid)
2520                         return -ENOENT;
2521                 sid->idx = e->idx;
2522                 sid->cpu = e->cpu;
2523                 sid->tid = e->tid;
2524         }
2525         return 0;
2526 }