tools/perf/util/session.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 #include <errno.h>
   3 #include <inttypes.h>
   4 #include <linux/kernel.h>
   5 #include <traceevent/event-parse.h>
   6 #include <api/fs/fs.h>
   7
   8 #include <byteswap.h>
   9 #include <unistd.h>
  10 #include <sys/types.h>
  11 #include <sys/mman.h>
  12
  13 #include "evlist.h"
  14 #include "evsel.h"
  15 #include "memswap.h"
  16 #include "session.h"
  17 #include "tool.h"
  18 #include "sort.h"
  19 #include "util.h"
  20 #include "cpumap.h"
  21 #include "perf_regs.h"
  22 #include "asm/bug.h"
  23 #include "auxtrace.h"
  24 #include "thread.h"
  25 #include "thread-stack.h"
  26 #include "stat.h"
  27
  28 static int perf_session__deliver_event(struct perf_session *session,
  29                                        union perf_event *event,
  30                                        struct perf_sample *sample,
  31                                        struct perf_tool *tool,
  32                                        u64 file_offset);
  33
  34 static int perf_session__open(struct perf_session *session)
  35 {
  36         struct perf_data_file *file = session->file;
  37
  38         if (perf_session__read_header(session) < 0) {
  39                 pr_err("incompatible file format (rerun with -v to learn more)\n");
  40                 return -1;
  41         }
  42
  43         if (perf_data_file__is_pipe(file))
  44                 return 0;
  45
  46         if (perf_header__has_feat(&session->header, HEADER_STAT))
  47                 return 0;
  48
  49         if (!perf_evlist__valid_sample_type(session->evlist)) {
  50                 pr_err("non matching sample_type\n");
  51                 return -1;
  52         }
  53
  54         if (!perf_evlist__valid_sample_id_all(session->evlist)) {
  55                 pr_err("non matching sample_id_all\n");
  56                 return -1;
  57         }
  58
  59         if (!perf_evlist__valid_read_format(session->evlist)) {
  60                 pr_err("non matching read_format\n");
  61                 return -1;
  62         }
  63
  64         return 0;
  65 }
  66
  67 void perf_session__set_id_hdr_size(struct perf_session *session)
  68 {
  69         u16 id_hdr_size = perf_evlist__id_hdr_size(session->evlist);
  70
  71         machines__set_id_hdr_size(&session->machines, id_hdr_size);
  72 }
  73
  74 int perf_session__create_kernel_maps(struct perf_session *session)
  75 {
  76         int ret = machine__create_kernel_maps(&session->machines.host);
  77
  78         if (ret >= 0)
  79                 ret = machines__create_guest_kernel_maps(&session->machines);
  80         return ret;
  81 }
  82
  83 static void perf_session__destroy_kernel_maps(struct perf_session *session)
  84 {
  85         machines__destroy_kernel_maps(&session->machines);
  86 }
  87
  88 static bool perf_session__has_comm_exec(struct perf_session *session)
  89 {
  90         struct perf_evsel *evsel;
  91
  92         evlist__for_each_entry(session->evlist, evsel) {
  93                 if (evsel->attr.comm_exec)
  94                         return true;
  95         }
  96
  97         return false;
  98 }
  99
 100 static void perf_session__set_comm_exec(struct perf_session *session)
 101 {
 102         bool comm_exec = perf_session__has_comm_exec(session);
 103
 104         machines__set_comm_exec(&session->machines, comm_exec);
 105 }
 106
 107 static int ordered_events__deliver_event(struct ordered_events *oe,
 108                                          struct ordered_event *event)
 109 {
 110         struct perf_sample sample;
 111         struct perf_session *session = container_of(oe, struct perf_session,
 112                                                     ordered_events);
 113         int ret = perf_evlist__parse_sample(session->evlist, event->event, &sample);
 114
 115         if (ret) {
 116                 pr_err("Can't parse sample, err = %d\n", ret);
 117                 return ret;
 118         }
 119
 120         return perf_session__deliver_event(session, event->event, &sample,
 121                                            session->tool, event->file_offset);
 122 }
 123
 124 struct perf_session *perf_session__new(struct perf_data_file *file,
 125                                        bool repipe, struct perf_tool *tool)
 126 {
 127         struct perf_session *session = zalloc(sizeof(*session));
 128
 129         if (!session)
 130                 goto out;
 131
 132         session->repipe = repipe;
 133         session->tool   = tool;
 134         INIT_LIST_HEAD(&session->auxtrace_index);
 135         machines__init(&session->machines);
 136         ordered_events__init(&session->ordered_events, ordered_events__deliver_event);
 137
 138         if (file) {
 139                 if (perf_data_file__open(file))
 140                         goto out_delete;
 141
 142                 session->file = file;
 143
 144                 if (perf_data_file__is_read(file)) {
 145                         if (perf_session__open(session) < 0)
 146                                 goto out_close;
 147
 148                         /*
 149                          * set session attributes that are present in perf.data
 150                          * but not in pipe-mode.
 151                          */
 152                         if (!file->is_pipe) {
 153                                 perf_session__set_id_hdr_size(session);
 154                                 perf_session__set_comm_exec(session);
 155                         }
 156                 }
 157         } else  {
 158                 session->machines.host.env = &perf_env;
 159         }
 160
 161         if (!file || perf_data_file__is_write(file)) {
 162                 /*
 163                  * In O_RDONLY mode this will be performed when reading the
 164                  * kernel MMAP event, in perf_event__process_mmap().
 165                  */
 166                 if (perf_session__create_kernel_maps(session) < 0)
 167                         pr_warning("Cannot read kernel map\n");
 168         }
 169
 170         /*
 171          * In pipe-mode, evlist is empty until PERF_RECORD_HEADER_ATTR is
 172          * processed, so perf_evlist__sample_id_all is not meaningful here.
 173          */
 174         if ((!file || !file->is_pipe) && tool && tool->ordering_requires_timestamps &&
 175             tool->ordered_events && !perf_evlist__sample_id_all(session->evlist)) {
 176                 dump_printf("WARNING: No sample_id_all support, falling back to unordered processing\n");
 177                 tool->ordered_events = false;
 178         }
 179
 180         return session;
 181
 182  out_close:
 183         perf_data_file__close(file);
 184  out_delete:
 185         perf_session__delete(session);
 186  out:
 187         return NULL;
 188 }
 189
 190 static void perf_session__delete_threads(struct perf_session *session)
 191 {
 192         machine__delete_threads(&session->machines.host);
 193 }
 194
 195 void perf_session__delete(struct perf_session *session)
 196 {
 197         if (session == NULL)
 198                 return;
 199         auxtrace__free(session);
 200         auxtrace_index__free(&session->auxtrace_index);
 201         perf_session__destroy_kernel_maps(session);
 202         perf_session__delete_threads(session);
 203         perf_env__exit(&session->header.env);
 204         machines__exit(&session->machines);
 205         if (session->file)
 206                 perf_data_file__close(session->file);
 207         free(session);
 208 }
 209
 210 static int process_event_synth_tracing_data_stub(struct perf_tool *tool
 211                                                  __maybe_unused,
 212                                                  union perf_event *event
 213                                                  __maybe_unused,
 214                                                  struct perf_session *session
 215                                                 __maybe_unused)
 216 {
 217         dump_printf(": unhandled!\n");
 218         return 0;
 219 }
 220
 221 static int process_event_synth_attr_stub(struct perf_tool *tool __maybe_unused,
 222                                          union perf_event *event __maybe_unused,
 223                                          struct perf_evlist **pevlist
 224                                          __maybe_unused)
 225 {
 226         dump_printf(": unhandled!\n");
 227         return 0;
 228 }
 229
 230 static int process_event_synth_event_update_stub(struct perf_tool *tool __maybe_unused,
 231                                                  union perf_event *event __maybe_unused,
 232                                                  struct perf_evlist **pevlist
 233                                                  __maybe_unused)
 234 {
 235         if (dump_trace)
 236                 perf_event__fprintf_event_update(event, stdout);
 237
 238         dump_printf(": unhandled!\n");
 239         return 0;
 240 }
 241
 242 static int process_event_sample_stub(struct perf_tool *tool __maybe_unused,
 243                                      union perf_event *event __maybe_unused,
 244                                      struct perf_sample *sample __maybe_unused,
 245                                      struct perf_evsel *evsel __maybe_unused,
 246                                      struct machine *machine __maybe_unused)
 247 {
 248         dump_printf(": unhandled!\n");
 249         return 0;
 250 }
 251
 252 static int process_event_stub(struct perf_tool *tool __maybe_unused,
 253                               union perf_event *event __maybe_unused,
 254                               struct perf_sample *sample __maybe_unused,
 255                               struct machine *machine __maybe_unused)
 256 {
 257         dump_printf(": unhandled!\n");
 258         return 0;
 259 }
 260
 261 static int process_finished_round_stub(struct perf_tool *tool __maybe_unused,
 262                                        union perf_event *event __maybe_unused,
 263                                        struct ordered_events *oe __maybe_unused)
 264 {
 265         dump_printf(": unhandled!\n");
 266         return 0;
 267 }
 268
 269 static int process_finished_round(struct perf_tool *tool,
 270                                   union perf_event *event,
 271                                   struct ordered_events *oe);
 272
 273 static int skipn(int fd, off_t n)
 274 {
 275         char buf[4096];
 276         ssize_t ret;
 277
 278         while (n > 0) {
 279                 ret = read(fd, buf, min(n, (off_t)sizeof(buf)));
 280                 if (ret <= 0)
 281                         return ret;
 282                 n -= ret;
 283         }
 284
 285         return 0;
 286 }
 287
 288 static s64 process_event_auxtrace_stub(struct perf_tool *tool __maybe_unused,
 289                                        union perf_event *event,
 290                                        struct perf_session *session
 291                                        __maybe_unused)
 292 {
 293         dump_printf(": unhandled!\n");
 294         if (perf_data_file__is_pipe(session->file))
 295                 skipn(perf_data_file__fd(session->file), event->auxtrace.size);
 296         return event->auxtrace.size;
 297 }
 298
 299 static int process_event_op2_stub(struct perf_tool *tool __maybe_unused,
 300                                   union perf_event *event __maybe_unused,
 301                                   struct perf_session *session __maybe_unused)
 302 {
 303         dump_printf(": unhandled!\n");
 304         return 0;
 305 }
 306
 307
 308 static
 309 int process_event_thread_map_stub(struct perf_tool *tool __maybe_unused,
 310                                   union perf_event *event __maybe_unused,
 311                                   struct perf_session *session __maybe_unused)
 312 {
 313         if (dump_trace)
 314                 perf_event__fprintf_thread_map(event, stdout);
 315
 316         dump_printf(": unhandled!\n");
 317         return 0;
 318 }
 319
 320 static
 321 int process_event_cpu_map_stub(struct perf_tool *tool __maybe_unused,
 322                                union perf_event *event __maybe_unused,
 323                                struct perf_session *session __maybe_unused)
 324 {
 325         if (dump_trace)
 326                 perf_event__fprintf_cpu_map(event, stdout);
 327
 328         dump_printf(": unhandled!\n");
 329         return 0;
 330 }
 331
 332 static
 333 int process_event_stat_config_stub(struct perf_tool *tool __maybe_unused,
 334                                    union perf_event *event __maybe_unused,
 335                                    struct perf_session *session __maybe_unused)
 336 {
 337         if (dump_trace)
 338                 perf_event__fprintf_stat_config(event, stdout);
 339
 340         dump_printf(": unhandled!\n");
 341         return 0;
 342 }
 343
 344 static int process_stat_stub(struct perf_tool *tool __maybe_unused,
 345                              union perf_event *event __maybe_unused,
 346                              struct perf_session *perf_session
 347                              __maybe_unused)
 348 {
 349         if (dump_trace)
 350                 perf_event__fprintf_stat(event, stdout);
 351
 352         dump_printf(": unhandled!\n");
 353         return 0;
 354 }
 355
 356 static int process_stat_round_stub(struct perf_tool *tool __maybe_unused,
 357                                    union perf_event *event __maybe_unused,
 358                                    struct perf_session *perf_session
 359                                    __maybe_unused)
 360 {
 361         if (dump_trace)
 362                 perf_event__fprintf_stat_round(event, stdout);
 363
 364         dump_printf(": unhandled!\n");
 365         return 0;
 366 }
 367
 368 void perf_tool__fill_defaults(struct perf_tool *tool)
 369 {
 370         if (tool->sample == NULL)
 371                 tool->sample = process_event_sample_stub;
 372         if (tool->mmap == NULL)
 373                 tool->mmap = process_event_stub;
 374         if (tool->mmap2 == NULL)
 375                 tool->mmap2 = process_event_stub;
 376         if (tool->comm == NULL)
 377                 tool->comm = process_event_stub;
 378         if (tool->namespaces == NULL)
 379                 tool->namespaces = process_event_stub;
 380         if (tool->fork == NULL)
 381                 tool->fork = process_event_stub;
 382         if (tool->exit == NULL)
 383                 tool->exit = process_event_stub;
 384         if (tool->lost == NULL)
 385                 tool->lost = perf_event__process_lost;
 386         if (tool->lost_samples == NULL)
 387                 tool->lost_samples = perf_event__process_lost_samples;
 388         if (tool->aux == NULL)
 389                 tool->aux = perf_event__process_aux;
 390         if (tool->itrace_start == NULL)
 391                 tool->itrace_start = perf_event__process_itrace_start;
 392         if (tool->context_switch == NULL)
 393                 tool->context_switch = perf_event__process_switch;
 394         if (tool->read == NULL)
 395                 tool->read = process_event_sample_stub;
 396         if (tool->throttle == NULL)
 397                 tool->throttle = process_event_stub;
 398         if (tool->unthrottle == NULL)
 399                 tool->unthrottle = process_event_stub;
 400         if (tool->attr == NULL)
 401                 tool->attr = process_event_synth_attr_stub;
 402         if (tool->event_update == NULL)
 403                 tool->event_update = process_event_synth_event_update_stub;
 404         if (tool->tracing_data == NULL)
 405                 tool->tracing_data = process_event_synth_tracing_data_stub;
 406         if (tool->build_id == NULL)
 407                 tool->build_id = process_event_op2_stub;
 408         if (tool->finished_round == NULL) {
 409                 if (tool->ordered_events)
 410                         tool->finished_round = process_finished_round;
 411                 else
 412                         tool->finished_round = process_finished_round_stub;
 413         }
 414         if (tool->id_index == NULL)
 415                 tool->id_index = process_event_op2_stub;
 416         if (tool->auxtrace_info == NULL)
 417                 tool->auxtrace_info = process_event_op2_stub;
 418         if (tool->auxtrace == NULL)
 419                 tool->auxtrace = process_event_auxtrace_stub;
 420         if (tool->auxtrace_error == NULL)
 421                 tool->auxtrace_error = process_event_op2_stub;
 422         if (tool->thread_map == NULL)
 423                 tool->thread_map = process_event_thread_map_stub;
 424         if (tool->cpu_map == NULL)
 425                 tool->cpu_map = process_event_cpu_map_stub;
 426         if (tool->stat_config == NULL)
 427                 tool->stat_config = process_event_stat_config_stub;
 428         if (tool->stat == NULL)
 429                 tool->stat = process_stat_stub;
 430         if (tool->stat_round == NULL)
 431                 tool->stat_round = process_stat_round_stub;
 432         if (tool->time_conv == NULL)
 433                 tool->time_conv = process_event_op2_stub;
 434         if (tool->feature == NULL)
 435                 tool->feature = process_event_op2_stub;
 436 }
 437
 438 static void swap_sample_id_all(union perf_event *event, void *data)
 439 {
 440         void *end = (void *) event + event->header.size;
 441         int size = end - data;
 442
 443         BUG_ON(size % sizeof(u64));
 444         mem_bswap_64(data, size);
 445 }
 446
 447 static void perf_event__all64_swap(union perf_event *event,
 448                                    bool sample_id_all __maybe_unused)
 449 {
 450         struct perf_event_header *hdr = &event->header;
 451         mem_bswap_64(hdr + 1, event->header.size - sizeof(*hdr));
 452 }
 453
 454 static void perf_event__comm_swap(union perf_event *event, bool sample_id_all)
 455 {
 456         event->comm.pid = bswap_32(event->comm.pid);
 457         event->comm.tid = bswap_32(event->comm.tid);
 458
 459         if (sample_id_all) {
 460                 void *data = &event->comm.comm;
 461
 462                 data += PERF_ALIGN(strlen(data) + 1, sizeof(u64));
 463                 swap_sample_id_all(event, data);
 464         }
 465 }
 466
 467 static void perf_event__mmap_swap(union perf_event *event,
 468                                   bool sample_id_all)
 469 {
 470         event->mmap.pid   = bswap_32(event->mmap.pid);
 471         event->mmap.tid   = bswap_32(event->mmap.tid);
 472         event->mmap.start = bswap_64(event->mmap.start);
 473         event->mmap.len   = bswap_64(event->mmap.len);
 474         event->mmap.pgoff = bswap_64(event->mmap.pgoff);
 475
 476         if (sample_id_all) {
 477                 void *data = &event->mmap.filename;
 478
 479                 data += PERF_ALIGN(strlen(data) + 1, sizeof(u64));
 480                 swap_sample_id_all(event, data);
 481         }
 482 }
 483
 484 static void perf_event__mmap2_swap(union perf_event *event,
 485                                   bool sample_id_all)
 486 {
 487         event->mmap2.pid   = bswap_32(event->mmap2.pid);
 488         event->mmap2.tid   = bswap_32(event->mmap2.tid);
 489         event->mmap2.start = bswap_64(event->mmap2.start);
 490         event->mmap2.len   = bswap_64(event->mmap2.len);
 491         event->mmap2.pgoff = bswap_64(event->mmap2.pgoff);
 492         event->mmap2.maj   = bswap_32(event->mmap2.maj);
 493         event->mmap2.min   = bswap_32(event->mmap2.min);
 494         event->mmap2.ino   = bswap_64(event->mmap2.ino);
 495         event->mmap2.ino_generation = bswap_64(event->mmap2.ino_generation);
 496
 497         if (sample_id_all) {
 498                 void *data = &event->mmap2.filename;
 499
 500                 data += PERF_ALIGN(strlen(data) + 1, sizeof(u64));
 501                 swap_sample_id_all(event, data);
 502         }
 503 }
 504 static void perf_event__task_swap(union perf_event *event, bool sample_id_all)
 505 {
 506         event->fork.pid  = bswap_32(event->fork.pid);
 507         event->fork.tid  = bswap_32(event->fork.tid);
 508         event->fork.ppid = bswap_32(event->fork.ppid);
 509         event->fork.ptid = bswap_32(event->fork.ptid);
 510         event->fork.time = bswap_64(event->fork.time);
 511
 512         if (sample_id_all)
 513                 swap_sample_id_all(event, &event->fork + 1);
 514 }
 515
 516 static void perf_event__read_swap(union perf_event *event, bool sample_id_all)
 517 {
 518         event->read.pid          = bswap_32(event->read.pid);
 519         event->read.tid          = bswap_32(event->read.tid);
 520         event->read.value        = bswap_64(event->read.value);
 521         event->read.time_enabled = bswap_64(event->read.time_enabled);
 522         event->read.time_running = bswap_64(event->read.time_running);
 523         event->read.id           = bswap_64(event->read.id);
 524
 525         if (sample_id_all)
 526                 swap_sample_id_all(event, &event->read + 1);
 527 }
 528
 529 static void perf_event__aux_swap(union perf_event *event, bool sample_id_all)
 530 {
 531         event->aux.aux_offset = bswap_64(event->aux.aux_offset);
 532         event->aux.aux_size   = bswap_64(event->aux.aux_size);
 533         event->aux.flags      = bswap_64(event->aux.flags);
 534
 535         if (sample_id_all)
 536                 swap_sample_id_all(event, &event->aux + 1);
 537 }
 538
 539 static void perf_event__itrace_start_swap(union perf_event *event,
 540                                           bool sample_id_all)
 541 {
 542         event->itrace_start.pid  = bswap_32(event->itrace_start.pid);
 543         event->itrace_start.tid  = bswap_32(event->itrace_start.tid);
 544
 545         if (sample_id_all)
 546                 swap_sample_id_all(event, &event->itrace_start + 1);
 547 }
 548
 549 static void perf_event__switch_swap(union perf_event *event, bool sample_id_all)
 550 {
 551         if (event->header.type == PERF_RECORD_SWITCH_CPU_WIDE) {
 552                 event->context_switch.next_prev_pid =
 553                                 bswap_32(event->context_switch.next_prev_pid);
 554                 event->context_switch.next_prev_tid =
 555                                 bswap_32(event->context_switch.next_prev_tid);
 556         }
 557
 558         if (sample_id_all)
 559                 swap_sample_id_all(event, &event->context_switch + 1);
 560 }
 561
 562 static void perf_event__throttle_swap(union perf_event *event,
 563                                       bool sample_id_all)
 564 {
 565         event->throttle.time      = bswap_64(event->throttle.time);
 566         event->throttle.id        = bswap_64(event->throttle.id);
 567         event->throttle.stream_id = bswap_64(event->throttle.stream_id);
 568
 569         if (sample_id_all)
 570                 swap_sample_id_all(event, &event->throttle + 1);
 571 }
 572
 573 static u8 revbyte(u8 b)
 574 {
 575         int rev = (b >> 4) | ((b & 0xf) << 4);
 576         rev = ((rev & 0xcc) >> 2) | ((rev & 0x33) << 2);
 577         rev = ((rev & 0xaa) >> 1) | ((rev & 0x55) << 1);
 578         return (u8) rev;
 579 }
 580
 581 /*
 582  * XXX this is hack in attempt to carry flags bitfield
 583  * through endian village. ABI says:
 584  *
 585  * Bit-fields are allocated from right to left (least to most significant)
 586  * on little-endian implementations and from left to right (most to least
 587  * significant) on big-endian implementations.
 588  *
 589  * The above seems to be byte specific, so we need to reverse each
 590  * byte of the bitfield. 'Internet' also says this might be implementation
 591  * specific and we probably need proper fix and carry perf_event_attr
 592  * bitfield flags in separate data file FEAT_ section. Thought this seems
 593  * to work for now.
 594  */
 595 static void swap_bitfield(u8 *p, unsigned len)
 596 {
 597         unsigned i;
 598
 599         for (i = 0; i < len; i++) {
 600                 *p = revbyte(*p);
 601                 p++;
 602         }
 603 }
 604
 605 /* exported for swapping attributes in file header */
 606 void perf_event__attr_swap(struct perf_event_attr *attr)
 607 {
 608         attr->type              = bswap_32(attr->type);
 609         attr->size              = bswap_32(attr->size);
 610
 611 #define bswap_safe(f, n)                                        \
 612         (attr->size > (offsetof(struct perf_event_attr, f) +    \
 613                        sizeof(attr->f) * (n)))
 614 #define bswap_field(f, sz)                      \
 615 do {                                            \
 616         if (bswap_safe(f, 0))                   \
 617                 attr->f = bswap_##sz(attr->f);  \
 618 } while(0)
 619 #define bswap_field_16(f) bswap_field(f, 16)
 620 #define bswap_field_32(f) bswap_field(f, 32)
 621 #define bswap_field_64(f) bswap_field(f, 64)
 622
 623         bswap_field_64(config);
 624         bswap_field_64(sample_period);
 625         bswap_field_64(sample_type);
 626         bswap_field_64(read_format);
 627         bswap_field_32(wakeup_events);
 628         bswap_field_32(bp_type);
 629         bswap_field_64(bp_addr);
 630         bswap_field_64(bp_len);
 631         bswap_field_64(branch_sample_type);
 632         bswap_field_64(sample_regs_user);
 633         bswap_field_32(sample_stack_user);
 634         bswap_field_32(aux_watermark);
 635         bswap_field_16(sample_max_stack);
 636
 637         /*
 638          * After read_format are bitfields. Check read_format because
 639          * we are unable to use offsetof on bitfield.
 640          */
 641         if (bswap_safe(read_format, 1))
 642                 swap_bitfield((u8 *) (&attr->read_format + 1),
 643                               sizeof(u64));
 644 #undef bswap_field_64
 645 #undef bswap_field_32
 646 #undef bswap_field
 647 #undef bswap_safe
 648 }
 649
 650 static void perf_event__hdr_attr_swap(union perf_event *event,
 651                                       bool sample_id_all __maybe_unused)
 652 {
 653         size_t size;
 654
 655         perf_event__attr_swap(&event->attr.attr);
 656
 657         size = event->header.size;
 658         size -= (void *)&event->attr.id - (void *)event;
 659         mem_bswap_64(event->attr.id, size);
 660 }
 661
 662 static void perf_event__event_update_swap(union perf_event *event,
 663                                           bool sample_id_all __maybe_unused)
 664 {
 665         event->event_update.type = bswap_64(event->event_update.type);
 666         event->event_update.id   = bswap_64(event->event_update.id);
 667 }
 668
 669 static void perf_event__event_type_swap(union perf_event *event,
 670                                         bool sample_id_all __maybe_unused)
 671 {
 672         event->event_type.event_type.event_id =
 673                 bswap_64(event->event_type.event_type.event_id);
 674 }
 675
 676 static void perf_event__tracing_data_swap(union perf_event *event,
 677                                           bool sample_id_all __maybe_unused)
 678 {
 679         event->tracing_data.size = bswap_32(event->tracing_data.size);
 680 }
 681
 682 static void perf_event__auxtrace_info_swap(union perf_event *event,
 683                                            bool sample_id_all __maybe_unused)
 684 {
 685         size_t size;
 686
 687         event->auxtrace_info.type = bswap_32(event->auxtrace_info.type);
 688
 689         size = event->header.size;
 690         size -= (void *)&event->auxtrace_info.priv - (void *)event;
 691         mem_bswap_64(event->auxtrace_info.priv, size);
 692 }
 693
 694 static void perf_event__auxtrace_swap(union perf_event *event,
 695                                       bool sample_id_all __maybe_unused)
 696 {
 697         event->auxtrace.size      = bswap_64(event->auxtrace.size);
 698         event->auxtrace.offset    = bswap_64(event->auxtrace.offset);
 699         event->auxtrace.reference = bswap_64(event->auxtrace.reference);
 700         event->auxtrace.idx       = bswap_32(event->auxtrace.idx);
 701         event->auxtrace.tid       = bswap_32(event->auxtrace.tid);
 702         event->auxtrace.cpu       = bswap_32(event->auxtrace.cpu);
 703 }
 704
 705 static void perf_event__auxtrace_error_swap(union perf_event *event,
 706                                             bool sample_id_all __maybe_unused)
 707 {
 708         event->auxtrace_error.type = bswap_32(event->auxtrace_error.type);
 709         event->auxtrace_error.code = bswap_32(event->auxtrace_error.code);
 710         event->auxtrace_error.cpu  = bswap_32(event->auxtrace_error.cpu);
 711         event->auxtrace_error.pid  = bswap_32(event->auxtrace_error.pid);
 712         event->auxtrace_error.tid  = bswap_32(event->auxtrace_error.tid);
 713         event->auxtrace_error.ip   = bswap_64(event->auxtrace_error.ip);
 714 }
 715
 716 static void perf_event__thread_map_swap(union perf_event *event,
 717                                         bool sample_id_all __maybe_unused)
 718 {
 719         unsigned i;
 720
 721         event->thread_map.nr = bswap_64(event->thread_map.nr);
 722
 723         for (i = 0; i < event->thread_map.nr; i++)
 724                 event->thread_map.entries[i].pid = bswap_64(event->thread_map.entries[i].pid);
 725 }
 726
 727 static void perf_event__cpu_map_swap(union perf_event *event,
 728                                      bool sample_id_all __maybe_unused)
 729 {
 730         struct cpu_map_data *data = &event->cpu_map.data;
 731         struct cpu_map_entries *cpus;
 732         struct cpu_map_mask *mask;
 733         unsigned i;
 734
 735         data->type = bswap_64(data->type);
 736
 737         switch (data->type) {
 738         case PERF_CPU_MAP__CPUS:
 739                 cpus = (struct cpu_map_entries *)data->data;
 740
 741                 cpus->nr = bswap_16(cpus->nr);
 742
 743                 for (i = 0; i < cpus->nr; i++)
 744                         cpus->cpu[i] = bswap_16(cpus->cpu[i]);
 745                 break;
 746         case PERF_CPU_MAP__MASK:
 747                 mask = (struct cpu_map_mask *) data->data;
 748
 749                 mask->nr = bswap_16(mask->nr);
 750                 mask->long_size = bswap_16(mask->long_size);
 751
 752                 switch (mask->long_size) {
 753                 case 4: mem_bswap_32(&mask->mask, mask->nr); break;
 754                 case 8: mem_bswap_64(&mask->mask, mask->nr); break;
 755                 default:
 756                         pr_err("cpu_map swap: unsupported long size\n");
 757                 }
 758         default:
 759                 break;
 760         }
 761 }
 762
 763 static void perf_event__stat_config_swap(union perf_event *event,
 764                                          bool sample_id_all __maybe_unused)
 765 {
 766         u64 size;
 767
 768         size  = event->stat_config.nr * sizeof(event->stat_config.data[0]);
 769         size += 1; /* nr item itself */
 770         mem_bswap_64(&event->stat_config.nr, size);
 771 }
 772
 773 static void perf_event__stat_swap(union perf_event *event,
 774                                   bool sample_id_all __maybe_unused)
 775 {
 776         event->stat.id     = bswap_64(event->stat.id);
 777         event->stat.thread = bswap_32(event->stat.thread);
 778         event->stat.cpu    = bswap_32(event->stat.cpu);
 779         event->stat.val    = bswap_64(event->stat.val);
 780         event->stat.ena    = bswap_64(event->stat.ena);
 781         event->stat.run    = bswap_64(event->stat.run);
 782 }
 783
 784 static void perf_event__stat_round_swap(union perf_event *event,
 785                                         bool sample_id_all __maybe_unused)
 786 {
 787         event->stat_round.type = bswap_64(event->stat_round.type);
 788         event->stat_round.time = bswap_64(event->stat_round.time);
 789 }
 790
 791 typedef void (*perf_event__swap_op)(union perf_event *event,
 792                                     bool sample_id_all);
 793
 794 static perf_event__swap_op perf_event__swap_ops[] = {
 795         [PERF_RECORD_MMAP]                = perf_event__mmap_swap,
 796         [PERF_RECORD_MMAP2]               = perf_event__mmap2_swap,
 797         [PERF_RECORD_COMM]                = perf_event__comm_swap,
 798         [PERF_RECORD_FORK]                = perf_event__task_swap,
 799         [PERF_RECORD_EXIT]                = perf_event__task_swap,
 800         [PERF_RECORD_LOST]                = perf_event__all64_swap,
 801         [PERF_RECORD_READ]                = perf_event__read_swap,
 802         [PERF_RECORD_THROTTLE]            = perf_event__throttle_swap,
 803         [PERF_RECORD_UNTHROTTLE]          = perf_event__throttle_swap,
 804         [PERF_RECORD_SAMPLE]              = perf_event__all64_swap,
 805         [PERF_RECORD_AUX]                 = perf_event__aux_swap,
 806         [PERF_RECORD_ITRACE_START]        = perf_event__itrace_start_swap,
 807         [PERF_RECORD_LOST_SAMPLES]        = perf_event__all64_swap,
 808         [PERF_RECORD_SWITCH]              = perf_event__switch_swap,
 809         [PERF_RECORD_SWITCH_CPU_WIDE]     = perf_event__switch_swap,
 810         [PERF_RECORD_HEADER_ATTR]         = perf_event__hdr_attr_swap,
 811         [PERF_RECORD_HEADER_EVENT_TYPE]   = perf_event__event_type_swap,
 812         [PERF_RECORD_HEADER_TRACING_DATA] = perf_event__tracing_data_swap,
 813         [PERF_RECORD_HEADER_BUILD_ID]     = NULL,
 814         [PERF_RECORD_ID_INDEX]            = perf_event__all64_swap,
 815         [PERF_RECORD_AUXTRACE_INFO]       = perf_event__auxtrace_info_swap,
 816         [PERF_RECORD_AUXTRACE]            = perf_event__auxtrace_swap,
 817         [PERF_RECORD_AUXTRACE_ERROR]      = perf_event__auxtrace_error_swap,
 818         [PERF_RECORD_THREAD_MAP]          = perf_event__thread_map_swap,
 819         [PERF_RECORD_CPU_MAP]             = perf_event__cpu_map_swap,
 820         [PERF_RECORD_STAT_CONFIG]         = perf_event__stat_config_swap,
 821         [PERF_RECORD_STAT]                = perf_event__stat_swap,
 822         [PERF_RECORD_STAT_ROUND]          = perf_event__stat_round_swap,
 823         [PERF_RECORD_EVENT_UPDATE]        = perf_event__event_update_swap,
 824         [PERF_RECORD_TIME_CONV]           = perf_event__all64_swap,
 825         [PERF_RECORD_HEADER_MAX]          = NULL,
 826 };
 827
 828 /*
 829  * When perf record finishes a pass on every buffers, it records this pseudo
 830  * event.
 831  * We record the max timestamp t found in the pass n.
 832  * Assuming these timestamps are monotonic across cpus, we know that if
 833  * a buffer still has events with timestamps below t, they will be all
 834  * available and then read in the pass n + 1.
 835  * Hence when we start to read the pass n + 2, we can safely flush every
 836  * events with timestamps below t.
 837  *
 838  *    ============ PASS n =================
 839  *       CPU 0         |   CPU 1
 840  *                     |
 841  *    cnt1 timestamps  |   cnt2 timestamps
 842  *          1          |         2
 843  *          2          |         3
 844  *          -          |         4  <--- max recorded
 845  *
 846  *    ============ PASS n + 1 ==============
 847  *       CPU 0         |   CPU 1
 848  *                     |
 849  *    cnt1 timestamps  |   cnt2 timestamps
 850  *          3          |         5
 851  *          4          |         6
 852  *          5          |         7 <---- max recorded
 853  *
 854  *      Flush every events below timestamp 4
 855  *
 856  *    ============ PASS n + 2 ==============
 857  *       CPU 0         |   CPU 1
 858  *                     |
 859  *    cnt1 timestamps  |   cnt2 timestamps
 860  *          6          |         8
 861  *          7          |         9
 862  *          -          |         10
 863  *
 864  *      Flush every events below timestamp 7
 865  *      etc...
 866  */
 867 static int process_finished_round(struct perf_tool *tool __maybe_unused,
 868                                   union perf_event *event __maybe_unused,
 869                                   struct ordered_events *oe)
 870 {
 871         if (dump_trace)
 872                 fprintf(stdout, "\n");
 873         return ordered_events__flush(oe, OE_FLUSH__ROUND);
 874 }
 875
 876 int perf_session__queue_event(struct perf_session *s, union perf_event *event,
 877                               struct perf_sample *sample, u64 file_offset)
 878 {
 879         return ordered_events__queue(&s->ordered_events, event, sample, file_offset);
 880 }
 881
 882 static void callchain__lbr_callstack_printf(struct perf_sample *sample)
 883 {
 884         struct ip_callchain *callchain = sample->callchain;
 885         struct branch_stack *lbr_stack = sample->branch_stack;
 886         u64 kernel_callchain_nr = callchain->nr;
 887         unsigned int i;
 888
 889         for (i = 0; i < kernel_callchain_nr; i++) {
 890                 if (callchain->ips[i] == PERF_CONTEXT_USER)
 891                         break;
 892         }
 893
 894         if ((i != kernel_callchain_nr) && lbr_stack->nr) {
 895                 u64 total_nr;
 896                 /*
 897                  * LBR callstack can only get user call chain,
 898                  * i is kernel call chain number,
 899                  * 1 is PERF_CONTEXT_USER.
 900                  *
 901                  * The user call chain is stored in LBR registers.
 902                  * LBR are pair registers. The caller is stored
 903                  * in "from" register, while the callee is stored
 904                  * in "to" register.
 905                  * For example, there is a call stack
 906                  * "A"->"B"->"C"->"D".
 907                  * The LBR registers will recorde like
 908                  * "C"->"D", "B"->"C", "A"->"B".
 909                  * So only the first "to" register and all "from"
 910                  * registers are needed to construct the whole stack.
 911                  */
 912                 total_nr = i + 1 + lbr_stack->nr + 1;
 913                 kernel_callchain_nr = i + 1;
 914
 915                 printf("... LBR call chain: nr:%" PRIu64 "\n", total_nr);
 916
 917                 for (i = 0; i < kernel_callchain_nr; i++)
 918                         printf("..... %2d: %016" PRIx64 "\n",
 919                                i, callchain->ips[i]);
 920
 921                 printf("..... %2d: %016" PRIx64 "\n",
 922                        (int)(kernel_callchain_nr), lbr_stack->entries[0].to);
 923                 for (i = 0; i < lbr_stack->nr; i++)
 924                         printf("..... %2d: %016" PRIx64 "\n",
 925                                (int)(i + kernel_callchain_nr + 1), lbr_stack->entries[i].from);
 926         }
 927 }
 928
 929 static void callchain__printf(struct perf_evsel *evsel,
 930                               struct perf_sample *sample)
 931 {
 932         unsigned int i;
 933         struct ip_callchain *callchain = sample->callchain;
 934
 935         if (perf_evsel__has_branch_callstack(evsel))
 936                 callchain__lbr_callstack_printf(sample);
 937
 938         printf("... FP chain: nr:%" PRIu64 "\n", callchain->nr);
 939
 940         for (i = 0; i < callchain->nr; i++)
 941                 printf("..... %2d: %016" PRIx64 "\n",
 942                        i, callchain->ips[i]);
 943 }
 944
 945 static void branch_stack__printf(struct perf_sample *sample)
 946 {
 947         uint64_t i;
 948
 949         printf("... branch stack: nr:%" PRIu64 "\n", sample->branch_stack->nr);
 950
 951         for (i = 0; i < sample->branch_stack->nr; i++) {
 952                 struct branch_entry *e = &sample->branch_stack->entries[i];
 953
 954                 printf("..... %2"PRIu64": %016" PRIx64 " -> %016" PRIx64 " %hu cycles %s%s%s%s %x\n",
 955                         i, e->from, e->to,
 956                         (unsigned short)e->flags.cycles,
 957                         e->flags.mispred ? "M" : " ",
 958                         e->flags.predicted ? "P" : " ",
 959                         e->flags.abort ? "A" : " ",
 960                         e->flags.in_tx ? "T" : " ",
 961                         (unsigned)e->flags.reserved);
 962         }
 963 }
 964
 965 static void regs_dump__printf(u64 mask, u64 *regs)
 966 {
 967         unsigned rid, i = 0;
 968
 969         for_each_set_bit(rid, (unsigned long *) &mask, sizeof(mask) * 8) {
 970                 u64 val = regs[i++];
 971
 972                 printf(".... %-5s 0x%" PRIx64 "\n",
 973                        perf_reg_name(rid), val);
 974         }
 975 }
 976
 977 static const char *regs_abi[] = {
 978         [PERF_SAMPLE_REGS_ABI_NONE] = "none",
 979         [PERF_SAMPLE_REGS_ABI_32] = "32-bit",
 980         [PERF_SAMPLE_REGS_ABI_64] = "64-bit",
 981 };
 982
 983 static inline const char *regs_dump_abi(struct regs_dump *d)
 984 {
 985         if (d->abi > PERF_SAMPLE_REGS_ABI_64)
 986                 return "unknown";
 987
 988         return regs_abi[d->abi];
 989 }
 990
 991 static void regs__printf(const char *type, struct regs_dump *regs)
 992 {
 993         u64 mask = regs->mask;
 994
 995         printf("... %s regs: mask 0x%" PRIx64 " ABI %s\n",
 996                type,
 997                mask,
 998                regs_dump_abi(regs));
 999
1000         regs_dump__printf(mask, regs->regs);
1001 }
1002
1003 static void regs_user__printf(struct perf_sample *sample)
1004 {
1005         struct regs_dump *user_regs = &sample->user_regs;
1006
1007         if (user_regs->regs)
1008                 regs__printf("user", user_regs);
1009 }
1010
1011 static void regs_intr__printf(struct perf_sample *sample)
1012 {
1013         struct regs_dump *intr_regs = &sample->intr_regs;
1014
1015         if (intr_regs->regs)
1016                 regs__printf("intr", intr_regs);
1017 }
1018
1019 static void stack_user__printf(struct stack_dump *dump)
1020 {
1021         printf("... ustack: size %" PRIu64 ", offset 0x%x\n",
1022                dump->size, dump->offset);
1023 }
1024
1025 static void perf_evlist__print_tstamp(struct perf_evlist *evlist,
1026                                        union perf_event *event,
1027                                        struct perf_sample *sample)
1028 {
1029         u64 sample_type = __perf_evlist__combined_sample_type(evlist);
1030
1031         if (event->header.type != PERF_RECORD_SAMPLE &&
1032             !perf_evlist__sample_id_all(evlist)) {
1033                 fputs("-1 -1 ", stdout);
1034                 return;
1035         }
1036
1037         if ((sample_type & PERF_SAMPLE_CPU))
1038                 printf("%u ", sample->cpu);
1039
1040         if (sample_type & PERF_SAMPLE_TIME)
1041                 printf("%" PRIu64 " ", sample->time);
1042 }
1043
1044 static void sample_read__printf(struct perf_sample *sample, u64 read_format)
1045 {
1046         printf("... sample_read:\n");
1047
1048         if (read_format & PERF_FORMAT_TOTAL_TIME_ENABLED)
1049                 printf("...... time enabled %016" PRIx64 "\n",
1050                        sample->read.time_enabled);
1051
1052         if (read_format & PERF_FORMAT_TOTAL_TIME_RUNNING)
1053                 printf("...... time running %016" PRIx64 "\n",
1054                        sample->read.time_running);
1055
1056         if (read_format & PERF_FORMAT_GROUP) {
1057                 u64 i;
1058
1059                 printf(".... group nr %" PRIu64 "\n", sample->read.group.nr);
1060
1061                 for (i = 0; i < sample->read.group.nr; i++) {
1062                         struct sample_read_value *value;
1063
1064                         value = &sample->read.group.values[i];
1065                         printf("..... id %016" PRIx64
1066                                ", value %016" PRIx64 "\n",
1067                                value->id, value->value);
1068                 }
1069         } else
1070                 printf("..... id %016" PRIx64 ", value %016" PRIx64 "\n",
1071                         sample->read.one.id, sample->read.one.value);
1072 }
1073
1074 static void dump_event(struct perf_evlist *evlist, union perf_event *event,
1075                        u64 file_offset, struct perf_sample *sample)
1076 {
1077         if (!dump_trace)
1078                 return;
1079
1080         printf("\n%#" PRIx64 " [%#x]: event: %d\n",
1081                file_offset, event->header.size, event->header.type);
1082
1083         trace_event(event);
1084
1085         if (sample)
1086                 perf_evlist__print_tstamp(evlist, event, sample);
1087
1088         printf("%#" PRIx64 " [%#x]: PERF_RECORD_%s", file_offset,
1089                event->header.size, perf_event__name(event->header.type));
1090 }
1091
1092 static void dump_sample(struct perf_evsel *evsel, union perf_event *event,
1093                         struct perf_sample *sample)
1094 {
1095         u64 sample_type;
1096
1097         if (!dump_trace)
1098                 return;
1099
1100         printf("(IP, 0x%x): %d/%d: %#" PRIx64 " period: %" PRIu64 " addr: %#" PRIx64 "\n",
1101                event->header.misc, sample->pid, sample->tid, sample->ip,
1102                sample->period, sample->addr);
1103
1104         sample_type = evsel->attr.sample_type;
1105
1106         if (sample_type & PERF_SAMPLE_CALLCHAIN)
1107                 callchain__printf(evsel, sample);
1108
1109         if ((sample_type & PERF_SAMPLE_BRANCH_STACK) && !perf_evsel__has_branch_callstack(evsel))
1110                 branch_stack__printf(sample);
1111
1112         if (sample_type & PERF_SAMPLE_REGS_USER)
1113                 regs_user__printf(sample);
1114
1115         if (sample_type & PERF_SAMPLE_REGS_INTR)
1116                 regs_intr__printf(sample);
1117
1118         if (sample_type & PERF_SAMPLE_STACK_USER)
1119                 stack_user__printf(&sample->user_stack);
1120
1121         if (sample_type & PERF_SAMPLE_WEIGHT)
1122                 printf("... weight: %" PRIu64 "\n", sample->weight);
1123
1124         if (sample_type & PERF_SAMPLE_DATA_SRC)
1125                 printf(" . data_src: 0x%"PRIx64"\n", sample->data_src);
1126
1127         if (sample_type & PERF_SAMPLE_PHYS_ADDR)
1128                 printf(" .. phys_addr: 0x%"PRIx64"\n", sample->phys_addr);
1129
1130         if (sample_type & PERF_SAMPLE_TRANSACTION)
1131                 printf("... transaction: %" PRIx64 "\n", sample->transaction);
1132
1133         if (sample_type & PERF_SAMPLE_READ)
1134                 sample_read__printf(sample, evsel->attr.read_format);
1135 }
1136
1137 static void dump_read(struct perf_evsel *evsel, union perf_event *event)
1138 {
1139         struct read_event *read_event = &event->read;
1140         u64 read_format;
1141
1142         if (!dump_trace)
1143                 return;
1144
1145         printf(": %d %d %s %" PRIu64 "\n", event->read.pid, event->read.tid,
1146                evsel ? perf_evsel__name(evsel) : "FAIL",
1147                event->read.value);
1148
1149         if (!evsel)
1150                 return;
1151
1152         read_format = evsel->attr.read_format;
1153
1154         if (read_format & PERF_FORMAT_TOTAL_TIME_ENABLED)
1155                 printf("... time enabled : %" PRIu64 "\n", read_event->time_enabled);
1156
1157         if (read_format & PERF_FORMAT_TOTAL_TIME_RUNNING)
1158                 printf("... time running : %" PRIu64 "\n", read_event->time_running);
1159
1160         if (read_format & PERF_FORMAT_ID)
1161                 printf("... id           : %" PRIu64 "\n", read_event->id);
1162 }
1163
1164 static struct machine *machines__find_for_cpumode(struct machines *machines,
1165                                                union perf_event *event,
1166                                                struct perf_sample *sample)
1167 {
1168         struct machine *machine;
1169
1170         if (perf_guest &&
1171             ((sample->cpumode == PERF_RECORD_MISC_GUEST_KERNEL) ||
1172              (sample->cpumode == PERF_RECORD_MISC_GUEST_USER))) {
1173                 u32 pid;
1174
1175                 if (event->header.type == PERF_RECORD_MMAP
1176                     || event->header.type == PERF_RECORD_MMAP2)
1177                         pid = event->mmap.pid;
1178                 else
1179                         pid = sample->pid;
1180
1181                 machine = machines__find(machines, pid);
1182                 if (!machine)
1183                         machine = machines__findnew(machines, DEFAULT_GUEST_KERNEL_ID);
1184                 return machine;
1185         }
1186
1187         return &machines->host;
1188 }
1189
1190 static int deliver_sample_value(struct perf_evlist *evlist,
1191                                 struct perf_tool *tool,
1192                                 union perf_event *event,
1193                                 struct perf_sample *sample,
1194                                 struct sample_read_value *v,
1195                                 struct machine *machine)
1196 {
1197         struct perf_sample_id *sid = perf_evlist__id2sid(evlist, v->id);
1198
1199         if (sid) {
1200                 sample->id     = v->id;
1201                 sample->period = v->value - sid->period;
1202                 sid->period    = v->value;
1203         }
1204
1205         if (!sid || sid->evsel == NULL) {
1206                 ++evlist->stats.nr_unknown_id;
1207                 return 0;
1208         }
1209
1210         return tool->sample(tool, event, sample, sid->evsel, machine);
1211 }
1212
1213 static int deliver_sample_group(struct perf_evlist *evlist,
1214                                 struct perf_tool *tool,
1215                                 union  perf_event *event,
1216                                 struct perf_sample *sample,
1217                                 struct machine *machine)
1218 {
1219         int ret = -EINVAL;
1220         u64 i;
1221
1222         for (i = 0; i < sample->read.group.nr; i++) {
1223                 ret = deliver_sample_value(evlist, tool, event, sample,
1224                                            &sample->read.group.values[i],
1225                                            machine);
1226                 if (ret)
1227                         break;
1228         }
1229
1230         return ret;
1231 }
1232
1233 static int
1234  perf_evlist__deliver_sample(struct perf_evlist *evlist,
1235                              struct perf_tool *tool,
1236                              union  perf_event *event,
1237                              struct perf_sample *sample,
1238                              struct perf_evsel *evsel,
1239                              struct machine *machine)
1240 {
1241         /* We know evsel != NULL. */
1242         u64 sample_type = evsel->attr.sample_type;
1243         u64 read_format = evsel->attr.read_format;
1244
1245         /* Standard sample delivery. */
1246         if (!(sample_type & PERF_SAMPLE_READ))
1247                 return tool->sample(tool, event, sample, evsel, machine);
1248
1249         /* For PERF_SAMPLE_READ we have either single or group mode. */
1250         if (read_format & PERF_FORMAT_GROUP)
1251                 return deliver_sample_group(evlist, tool, event, sample,
1252                                             machine);
1253         else
1254                 return deliver_sample_value(evlist, tool, event, sample,
1255                                             &sample->read.one, machine);
1256 }
1257
1258 static int machines__deliver_event(struct machines *machines,
1259                                    struct perf_evlist *evlist,
1260                                    union perf_event *event,
1261                                    struct perf_sample *sample,
1262                                    struct perf_tool *tool, u64 file_offset)
1263 {
1264         struct perf_evsel *evsel;
1265         struct machine *machine;
1266
1267         dump_event(evlist, event, file_offset, sample);
1268
1269         evsel = perf_evlist__id2evsel(evlist, sample->id);
1270
1271         machine = machines__find_for_cpumode(machines, event, sample);
1272
1273         switch (event->header.type) {
1274         case PERF_RECORD_SAMPLE:
1275                 if (evsel == NULL) {
1276                         ++evlist->stats.nr_unknown_id;
1277                         return 0;
1278                 }
1279                 dump_sample(evsel, event, sample);
1280                 if (machine == NULL) {
1281                         ++evlist->stats.nr_unprocessable_samples;
1282                         return 0;
1283                 }
1284                 return perf_evlist__deliver_sample(evlist, tool, event, sample, evsel, machine);
1285         case PERF_RECORD_MMAP:
1286                 return tool->mmap(tool, event, sample, machine);
1287         case PERF_RECORD_MMAP2:
1288                 if (event->header.misc & PERF_RECORD_MISC_PROC_MAP_PARSE_TIMEOUT)
1289                         ++evlist->stats.nr_proc_map_timeout;
1290                 return tool->mmap2(tool, event, sample, machine);
1291         case PERF_RECORD_COMM:
1292                 return tool->comm(tool, event, sample, machine);
1293         case PERF_RECORD_NAMESPACES:
1294                 return tool->namespaces(tool, event, sample, machine);
1295         case PERF_RECORD_FORK:
1296                 return tool->fork(tool, event, sample, machine);
1297         case PERF_RECORD_EXIT:
1298                 return tool->exit(tool, event, sample, machine);
1299         case PERF_RECORD_LOST:
1300                 if (tool->lost == perf_event__process_lost)
1301                         evlist->stats.total_lost += event->lost.lost;
1302                 return tool->lost(tool, event, sample, machine);
1303         case PERF_RECORD_LOST_SAMPLES:
1304                 if (tool->lost_samples == perf_event__process_lost_samples)
1305                         evlist->stats.total_lost_samples += event->lost_samples.lost;
1306                 return tool->lost_samples(tool, event, sample, machine);
1307         case PERF_RECORD_READ:
1308                 dump_read(evsel, event);
1309                 return tool->read(tool, event, sample, evsel, machine);
1310         case PERF_RECORD_THROTTLE:
1311                 return tool->throttle(tool, event, sample, machine);
1312         case PERF_RECORD_UNTHROTTLE:
1313                 return tool->unthrottle(tool, event, sample, machine);
1314         case PERF_RECORD_AUX:
1315                 if (tool->aux == perf_event__process_aux) {
1316                         if (event->aux.flags & PERF_AUX_FLAG_TRUNCATED)
1317                                 evlist->stats.total_aux_lost += 1;
1318                         if (event->aux.flags & PERF_AUX_FLAG_PARTIAL)
1319                                 evlist->stats.total_aux_partial += 1;
1320                 }
1321                 return tool->aux(tool, event, sample, machine);
1322         case PERF_RECORD_ITRACE_START:
1323                 return tool->itrace_start(tool, event, sample, machine);
1324         case PERF_RECORD_SWITCH:
1325         case PERF_RECORD_SWITCH_CPU_WIDE:
1326                 return tool->context_switch(tool, event, sample, machine);
1327         default:
1328                 ++evlist->stats.nr_unknown_events;
1329                 return -1;
1330         }
1331 }
1332
1333 static int perf_session__deliver_event(struct perf_session *session,
1334                                        union perf_event *event,
1335                                        struct perf_sample *sample,
1336                                        struct perf_tool *tool,
1337                                        u64 file_offset)
1338 {
1339         int ret;
1340
1341         ret = auxtrace__process_event(session, event, sample, tool);
1342         if (ret < 0)
1343                 return ret;
1344         if (ret > 0)
1345                 return 0;
1346
1347         return machines__deliver_event(&session->machines, session->evlist,
1348                                        event, sample, tool, file_offset);
1349 }
1350
1351 static s64 perf_session__process_user_event(struct perf_session *session,
1352                                             union perf_event *event,
1353                                             u64 file_offset)
1354 {
1355         struct ordered_events *oe = &session->ordered_events;
1356         struct perf_tool *tool = session->tool;
1357         int fd = perf_data_file__fd(session->file);
1358         int err;
1359
1360         dump_event(session->evlist, event, file_offset, NULL);
1361
1362         /* These events are processed right away */
1363         switch (event->header.type) {
1364         case PERF_RECORD_HEADER_ATTR:
1365                 err = tool->attr(tool, event, &session->evlist);
1366                 if (err == 0) {
1367                         perf_session__set_id_hdr_size(session);
1368                         perf_session__set_comm_exec(session);
1369                 }
1370                 return err;
1371         case PERF_RECORD_EVENT_UPDATE:
1372                 return tool->event_update(tool, event, &session->evlist);
1373         case PERF_RECORD_HEADER_EVENT_TYPE:
1374                 /*
1375                  * Depreceated, but we need to handle it for sake
1376                  * of old data files create in pipe mode.
1377                  */
1378                 return 0;
1379         case PERF_RECORD_HEADER_TRACING_DATA:
1380                 /* setup for reading amidst mmap */
1381                 lseek(fd, file_offset, SEEK_SET);
1382                 return tool->tracing_data(tool, event, session);
1383         case PERF_RECORD_HEADER_BUILD_ID:
1384                 return tool->build_id(tool, event, session);
1385         case PERF_RECORD_FINISHED_ROUND:
1386                 return tool->finished_round(tool, event, oe);
1387         case PERF_RECORD_ID_INDEX:
1388                 return tool->id_index(tool, event, session);
1389         case PERF_RECORD_AUXTRACE_INFO:
1390                 return tool->auxtrace_info(tool, event, session);
1391         case PERF_RECORD_AUXTRACE:
1392                 /* setup for reading amidst mmap */
1393                 lseek(fd, file_offset + event->header.size, SEEK_SET);
1394                 return tool->auxtrace(tool, event, session);
1395         case PERF_RECORD_AUXTRACE_ERROR:
1396                 perf_session__auxtrace_error_inc(session, event);
1397                 return tool->auxtrace_error(tool, event, session);
1398         case PERF_RECORD_THREAD_MAP:
1399                 return tool->thread_map(tool, event, session);
1400         case PERF_RECORD_CPU_MAP:
1401                 return tool->cpu_map(tool, event, session);
1402         case PERF_RECORD_STAT_CONFIG:
1403                 return tool->stat_config(tool, event, session);
1404         case PERF_RECORD_STAT:
1405                 return tool->stat(tool, event, session);
1406         case PERF_RECORD_STAT_ROUND:
1407                 return tool->stat_round(tool, event, session);
1408         case PERF_RECORD_TIME_CONV:
1409                 session->time_conv = event->time_conv;
1410                 return tool->time_conv(tool, event, session);
1411         case PERF_RECORD_HEADER_FEATURE:
1412                 return tool->feature(tool, event, session);
1413         default:
1414                 return -EINVAL;
1415         }
1416 }
1417
1418 int perf_session__deliver_synth_event(struct perf_session *session,
1419                                       union perf_event *event,
1420                                       struct perf_sample *sample)
1421 {
1422         struct perf_evlist *evlist = session->evlist;
1423         struct perf_tool *tool = session->tool;
1424
1425         events_stats__inc(&evlist->stats, event->header.type);
1426
1427         if (event->header.type >= PERF_RECORD_USER_TYPE_START)
1428                 return perf_session__process_user_event(session, event, 0);
1429
1430         return machines__deliver_event(&session->machines, evlist, event, sample, tool, 0);
1431 }
1432
1433 static void event_swap(union perf_event *event, bool sample_id_all)
1434 {
1435         perf_event__swap_op swap;
1436
1437         swap = perf_event__swap_ops[event->header.type];
1438         if (swap)
1439                 swap(event, sample_id_all);
1440 }
1441
1442 int perf_session__peek_event(struct perf_session *session, off_t file_offset,
1443                              void *buf, size_t buf_sz,
1444                              union perf_event **event_ptr,
1445                              struct perf_sample *sample)
1446 {
1447         union perf_event *event;
1448         size_t hdr_sz, rest;
1449         int fd;
1450
1451         if (session->one_mmap && !session->header.needs_swap) {
1452                 event = file_offset - session->one_mmap_offset +
1453                         session->one_mmap_addr;
1454                 goto out_parse_sample;
1455         }
1456
1457         if (perf_data_file__is_pipe(session->file))
1458                 return -1;
1459
1460         fd = perf_data_file__fd(session->file);
1461         hdr_sz = sizeof(struct perf_event_header);
1462
1463         if (buf_sz < hdr_sz)
1464                 return -1;
1465
1466         if (lseek(fd, file_offset, SEEK_SET) == (off_t)-1 ||
1467             readn(fd, buf, hdr_sz) != (ssize_t)hdr_sz)
1468                 return -1;
1469
1470         event = (union perf_event *)buf;
1471
1472         if (session->header.needs_swap)
1473                 perf_event_header__bswap(&event->header);
1474
1475         if (event->header.size < hdr_sz || event->header.size > buf_sz)
1476                 return -1;
1477
1478         buf += hdr_sz;
1479         rest = event->header.size - hdr_sz;
1480
1481         if (readn(fd, buf, rest) != (ssize_t)rest)
1482                 return -1;
1483
1484         if (session->header.needs_swap)
1485                 event_swap(event, perf_evlist__sample_id_all(session->evlist));
1486
1487 out_parse_sample:
1488
1489         if (sample && event->header.type < PERF_RECORD_USER_TYPE_START &&
1490             perf_evlist__parse_sample(session->evlist, event, sample))
1491                 return -1;
1492
1493         *event_ptr = event;
1494
1495         return 0;
1496 }
1497
1498 static s64 perf_session__process_event(struct perf_session *session,
1499                                        union perf_event *event, u64 file_offset)
1500 {
1501         struct perf_evlist *evlist = session->evlist;
1502         struct perf_tool *tool = session->tool;
1503         struct perf_sample sample;
1504         int ret;
1505
1506         if (session->header.needs_swap)
1507                 event_swap(event, perf_evlist__sample_id_all(evlist));
1508
1509         if (event->header.type >= PERF_RECORD_HEADER_MAX)
1510                 return -EINVAL;
1511
1512         events_stats__inc(&evlist->stats, event->header.type);
1513
1514         if (event->header.type >= PERF_RECORD_USER_TYPE_START)
1515                 return perf_session__process_user_event(session, event, file_offset);
1516
1517         /*
1518          * For all kernel events we get the sample data
1519          */
1520         ret = perf_evlist__parse_sample(evlist, event, &sample);
1521         if (ret)
1522                 return ret;
1523
1524         if (tool->ordered_events) {
1525                 ret = perf_session__queue_event(session, event, &sample, file_offset);
1526                 if (ret != -ETIME)
1527                         return ret;
1528         }
1529
1530         return perf_session__deliver_event(session, event, &sample, tool,
1531                                            file_offset);
1532 }
1533
1534 void perf_event_header__bswap(struct perf_event_header *hdr)
1535 {
1536         hdr->type = bswap_32(hdr->type);
1537         hdr->misc = bswap_16(hdr->misc);
1538         hdr->size = bswap_16(hdr->size);
1539 }
1540
1541 struct thread *perf_session__findnew(struct perf_session *session, pid_t pid)
1542 {
1543         return machine__findnew_thread(&session->machines.host, -1, pid);
1544 }
1545
1546 int perf_session__register_idle_thread(struct perf_session *session)
1547 {
1548         struct thread *thread;
1549         int err = 0;
1550
1551         thread = machine__findnew_thread(&session->machines.host, 0, 0);
1552         if (thread == NULL || thread__set_comm(thread, "swapper", 0)) {
1553                 pr_err("problem inserting idle task.\n");
1554                 err = -1;
1555         }
1556
1557         if (thread == NULL || thread__set_namespaces(thread, 0, NULL)) {
1558                 pr_err("problem inserting idle task.\n");
1559                 err = -1;
1560         }
1561
1562         /* machine__findnew_thread() got the thread, so put it */
1563         thread__put(thread);
1564         return err;
1565 }
1566
1567 static void
1568 perf_session__warn_order(const struct perf_session *session)
1569 {
1570         const struct ordered_events *oe = &session->ordered_events;
1571         struct perf_evsel *evsel;
1572         bool should_warn = true;
1573
1574         evlist__for_each_entry(session->evlist, evsel) {
1575                 if (evsel->attr.write_backward)
1576                         should_warn = false;
1577         }
1578
1579         if (!should_warn)
1580                 return;
1581         if (oe->nr_unordered_events != 0)
1582                 ui__warning("%u out of order events recorded.\n", oe->nr_unordered_events);
1583 }
1584
1585 static void perf_session__warn_about_errors(const struct perf_session *session)
1586 {
1587         const struct events_stats *stats = &session->evlist->stats;
1588
1589         if (session->tool->lost == perf_event__process_lost &&
1590             stats->nr_events[PERF_RECORD_LOST] != 0) {
1591                 ui__warning("Processed %d events and lost %d chunks!\n\n"
1592                             "Check IO/CPU overload!\n\n",
1593                             stats->nr_events[0],
1594                             stats->nr_events[PERF_RECORD_LOST]);
1595         }
1596
1597         if (session->tool->lost_samples == perf_event__process_lost_samples) {
1598                 double drop_rate;
1599
1600                 drop_rate = (double)stats->total_lost_samples /
1601                             (double) (stats->nr_events[PERF_RECORD_SAMPLE] + stats->total_lost_samples);
1602                 if (drop_rate > 0.05) {
1603                         ui__warning("Processed %" PRIu64 " samples and lost %3.2f%% samples!\n\n",
1604                                     stats->nr_events[PERF_RECORD_SAMPLE] + stats->total_lost_samples,
1605                                     drop_rate * 100.0);
1606                 }
1607         }
1608
1609         if (session->tool->aux == perf_event__process_aux &&
1610             stats->total_aux_lost != 0) {
1611                 ui__warning("AUX data lost %" PRIu64 " times out of %u!\n\n",
1612                             stats->total_aux_lost,
1613                             stats->nr_events[PERF_RECORD_AUX]);
1614         }
1615
1616         if (session->tool->aux == perf_event__process_aux &&
1617             stats->total_aux_partial != 0) {
1618                 bool vmm_exclusive = false;
1619
1620                 (void)sysfs__read_bool("module/kvm_intel/parameters/vmm_exclusive",
1621                                        &vmm_exclusive);
1622
1623                 ui__warning("AUX data had gaps in it %" PRIu64 " times out of %u!\n\n"
1624                             "Are you running a KVM guest in the background?%s\n\n",
1625                             stats->total_aux_partial,
1626                             stats->nr_events[PERF_RECORD_AUX],
1627                             vmm_exclusive ?
1628                             "\nReloading kvm_intel module with vmm_exclusive=0\n"
1629                             "will reduce the gaps to only guest's timeslices." :
1630                             "");
1631         }
1632
1633         if (stats->nr_unknown_events != 0) {
1634                 ui__warning("Found %u unknown events!\n\n"
1635                             "Is this an older tool processing a perf.data "
1636                             "file generated by a more recent tool?\n\n"
1637                             "If that is not the case, consider "
1638                             "reporting to linux-kernel@vger.kernel.org.\n\n",
1639                             stats->nr_unknown_events);
1640         }
1641
1642         if (stats->nr_unknown_id != 0) {
1643                 ui__warning("%u samples with id not present in the header\n",
1644                             stats->nr_unknown_id);
1645         }
1646
1647         if (stats->nr_invalid_chains != 0) {
1648                 ui__warning("Found invalid callchains!\n\n"
1649                             "%u out of %u events were discarded for this reason.\n\n"
1650                             "Consider reporting to linux-kernel@vger.kernel.org.\n\n",
1651                             stats->nr_invalid_chains,
1652                             stats->nr_events[PERF_RECORD_SAMPLE]);
1653         }
1654
1655         if (stats->nr_unprocessable_samples != 0) {
1656                 ui__warning("%u unprocessable samples recorded.\n"
1657                             "Do you have a KVM guest running and not using 'perf kvm'?\n",
1658                             stats->nr_unprocessable_samples);
1659         }
1660
1661         perf_session__warn_order(session);
1662
1663         events_stats__auxtrace_error_warn(stats);
1664
1665         if (stats->nr_proc_map_timeout != 0) {
1666                 ui__warning("%d map information files for pre-existing threads were\n"
1667                             "not processed, if there are samples for addresses they\n"
1668                             "will not be resolved, you may find out which are these\n"
1669                             "threads by running with -v and redirecting the output\n"
1670                             "to a file.\n"
1671                             "The time limit to process proc map is too short?\n"
1672                             "Increase it by --proc-map-timeout\n",
1673                             stats->nr_proc_map_timeout);
1674         }
1675 }
1676
1677 static int perf_session__flush_thread_stack(struct thread *thread,
1678                                             void *p __maybe_unused)
1679 {
1680         return thread_stack__flush(thread);
1681 }
1682
1683 static int perf_session__flush_thread_stacks(struct perf_session *session)
1684 {
1685         return machines__for_each_thread(&session->machines,
1686                                          perf_session__flush_thread_stack,
1687                                          NULL);
1688 }
1689
1690 volatile int session_done;
1691
1692 static int __perf_session__process_pipe_events(struct perf_session *session)
1693 {
1694         struct ordered_events *oe = &session->ordered_events;
1695         struct perf_tool *tool = session->tool;
1696         int fd = perf_data_file__fd(session->file);
1697         union perf_event *event;
1698         uint32_t size, cur_size = 0;
1699         void *buf = NULL;
1700         s64 skip = 0;
1701         u64 head;
1702         ssize_t err;
1703         void *p;
1704
1705         perf_tool__fill_defaults(tool);
1706
1707         head = 0;
1708         cur_size = sizeof(union perf_event);
1709
1710         buf = malloc(cur_size);
1711         if (!buf)
1712                 return -errno;
1713         ordered_events__set_copy_on_queue(oe, true);
1714 more:
1715         event = buf;
1716         err = readn(fd, event, sizeof(struct perf_event_header));
1717         if (err <= 0) {
1718                 if (err == 0)
1719                         goto done;
1720
1721                 pr_err("failed to read event header\n");
1722                 goto out_err;
1723         }
1724
1725         if (session->header.needs_swap)
1726                 perf_event_header__bswap(&event->header);
1727
1728         size = event->header.size;
1729         if (size < sizeof(struct perf_event_header)) {
1730                 pr_err("bad event header size\n");
1731                 goto out_err;
1732         }
1733
1734         if (size > cur_size) {
1735                 void *new = realloc(buf, size);
1736                 if (!new) {
1737                         pr_err("failed to allocate memory to read event\n");
1738                         goto out_err;
1739                 }
1740                 buf = new;
1741                 cur_size = size;
1742                 event = buf;
1743         }
1744         p = event;
1745         p += sizeof(struct perf_event_header);
1746
1747         if (size - sizeof(struct perf_event_header)) {
1748                 err = readn(fd, p, size - sizeof(struct perf_event_header));
1749                 if (err <= 0) {
1750                         if (err == 0) {
1751                                 pr_err("unexpected end of event stream\n");
1752                                 goto done;
1753                         }
1754
1755                         pr_err("failed to read event data\n");
1756                         goto out_err;
1757                 }
1758         }
1759
1760         if ((skip = perf_session__process_event(session, event, head)) < 0) {
1761                 pr_err("%#" PRIx64 " [%#x]: failed to process type: %d\n",
1762                        head, event->header.size, event->header.type);
1763                 err = -EINVAL;
1764                 goto out_err;
1765         }
1766
1767         head += size;
1768
1769         if (skip > 0)
1770                 head += skip;
1771
1772         if (!session_done())
1773                 goto more;
1774 done:
1775         /* do the final flush for ordered samples */
1776         err = ordered_events__flush(oe, OE_FLUSH__FINAL);
1777         if (err)
1778                 goto out_err;
1779         err = auxtrace__flush_events(session, tool);
1780         if (err)
1781                 goto out_err;
1782         err = perf_session__flush_thread_stacks(session);
1783 out_err:
1784         free(buf);
1785         perf_session__warn_about_errors(session);
1786         ordered_events__free(&session->ordered_events);
1787         auxtrace__free_events(session);
1788         return err;
1789 }
1790
1791 static union perf_event *
1792 fetch_mmaped_event(struct perf_session *session,
1793                    u64 head, size_t mmap_size, char *buf)
1794 {
1795         union perf_event *event;
1796
1797         /*
1798          * Ensure we have enough space remaining to read
1799          * the size of the event in the headers.
1800          */
1801         if (head + sizeof(event->header) > mmap_size)
1802                 return NULL;
1803
1804         event = (union perf_event *)(buf + head);
1805
1806         if (session->header.needs_swap)
1807                 perf_event_header__bswap(&event->header);
1808
1809         if (head + event->header.size > mmap_size) {
1810                 /* We're not fetching the event so swap back again */
1811                 if (session->header.needs_swap)
1812                         perf_event_header__bswap(&event->header);
1813                 return NULL;
1814         }
1815
1816         return event;
1817 }
1818
1819 /*
1820  * On 64bit we can mmap the data file in one go. No need for tiny mmap
1821  * slices. On 32bit we use 32MB.
1822  */
1823 #if BITS_PER_LONG == 64
1824 #define MMAP_SIZE ULLONG_MAX
1825 #define NUM_MMAPS 1
1826 #else
1827 #define MMAP_SIZE (32 * 1024 * 1024ULL)
1828 #define NUM_MMAPS 128
1829 #endif
1830
1831 static int __perf_session__process_events(struct perf_session *session,
1832                                           u64 data_offset, u64 data_size,
1833                                           u64 file_size)
1834 {
1835         struct ordered_events *oe = &session->ordered_events;
1836         struct perf_tool *tool = session->tool;
1837         int fd = perf_data_file__fd(session->file);
1838         u64 head, page_offset, file_offset, file_pos, size;
1839         int err, mmap_prot, mmap_flags, map_idx = 0;
1840         size_t  mmap_size;
1841         char *buf, *mmaps[NUM_MMAPS];
1842         union perf_event *event;
1843         struct ui_progress prog;
1844         s64 skip;
1845
1846         perf_tool__fill_defaults(tool);
1847
1848         page_offset = page_size * (data_offset / page_size);
1849         file_offset = page_offset;
1850         head = data_offset - page_offset;
1851
1852         if (data_size == 0)
1853                 goto out;
1854
1855         if (data_offset + data_size < file_size)
1856                 file_size = data_offset + data_size;
1857
1858         ui_progress__init(&prog, file_size, "Processing events...");
1859
1860         mmap_size = MMAP_SIZE;
1861         if (mmap_size > file_size) {
1862                 mmap_size = file_size;
1863                 session->one_mmap = true;
1864         }
1865
1866         memset(mmaps, 0, sizeof(mmaps));
1867
1868         mmap_prot  = PROT_READ;
1869         mmap_flags = MAP_SHARED;
1870
1871         if (session->header.needs_swap) {
1872                 mmap_prot  |= PROT_WRITE;
1873                 mmap_flags = MAP_PRIVATE;
1874         }
1875 remap:
1876         buf = mmap(NULL, mmap_size, mmap_prot, mmap_flags, fd,
1877                    file_offset);
1878         if (buf == MAP_FAILED) {
1879                 pr_err("failed to mmap file\n");
1880                 err = -errno;
1881                 goto out_err;
1882         }
1883         mmaps[map_idx] = buf;
1884         map_idx = (map_idx + 1) & (ARRAY_SIZE(mmaps) - 1);
1885         file_pos = file_offset + head;
1886         if (session->one_mmap) {
1887                 session->one_mmap_addr = buf;
1888                 session->one_mmap_offset = file_offset;
1889         }
1890
1891 more:
1892         event = fetch_mmaped_event(session, head, mmap_size, buf);
1893         if (!event) {
1894                 if (mmaps[map_idx]) {
1895                         munmap(mmaps[map_idx], mmap_size);
1896                         mmaps[map_idx] = NULL;
1897                 }
1898
1899                 page_offset = page_size * (head / page_size);
1900                 file_offset += page_offset;
1901                 head -= page_offset;
1902                 goto remap;
1903         }
1904
1905         size = event->header.size;
1906
1907         if (size < sizeof(struct perf_event_header) ||
1908             (skip = perf_session__process_event(session, event, file_pos)) < 0) {
1909                 pr_err("%#" PRIx64 " [%#x]: failed to process type: %d\n",
1910                        file_offset + head, event->header.size,
1911                        event->header.type);
1912                 err = -EINVAL;
1913                 goto out_err;
1914         }
1915
1916         if (skip)
1917                 size += skip;
1918
1919         head += size;
1920         file_pos += size;
1921
1922         ui_progress__update(&prog, size);
1923
1924         if (session_done())
1925                 goto out;
1926
1927         if (file_pos < file_size)
1928                 goto more;
1929
1930 out:
1931         /* do the final flush for ordered samples */
1932         err = ordered_events__flush(oe, OE_FLUSH__FINAL);
1933         if (err)
1934                 goto out_err;
1935         err = auxtrace__flush_events(session, tool);
1936         if (err)
1937                 goto out_err;
1938         err = perf_session__flush_thread_stacks(session);
1939 out_err:
1940         ui_progress__finish();
1941         perf_session__warn_about_errors(session);
1942         /*
1943          * We may switching perf.data output, make ordered_events
1944          * reusable.
1945          */
1946         ordered_events__reinit(&session->ordered_events);
1947         auxtrace__free_events(session);
1948         session->one_mmap = false;
1949         return err;
1950 }
1951
1952 int perf_session__process_events(struct perf_session *session)
1953 {
1954         u64 size = perf_data_file__size(session->file);
1955         int err;
1956
1957         if (perf_session__register_idle_thread(session) < 0)
1958                 return -ENOMEM;
1959
1960         if (!perf_data_file__is_pipe(session->file))
1961                 err = __perf_session__process_events(session,
1962                                                      session->header.data_offset,
1963                                                      session->header.data_size, size);
1964         else
1965                 err = __perf_session__process_pipe_events(session);
1966
1967         return err;
1968 }
1969
1970 bool perf_session__has_traces(struct perf_session *session, const char *msg)
1971 {
1972         struct perf_evsel *evsel;
1973
1974         evlist__for_each_entry(session->evlist, evsel) {
1975                 if (evsel->attr.type == PERF_TYPE_TRACEPOINT)
1976                         return true;
1977         }
1978
1979         pr_err("No trace sample to read. Did you call 'perf %s'?\n", msg);
1980         return false;
1981 }
1982
1983 int maps__set_kallsyms_ref_reloc_sym(struct map **maps,
1984                                      const char *symbol_name, u64 addr)
1985 {
1986         char *bracket;
1987         int i;
1988         struct ref_reloc_sym *ref;
1989
1990         ref = zalloc(sizeof(struct ref_reloc_sym));
1991         if (ref == NULL)
1992                 return -ENOMEM;
1993
1994         ref->name = strdup(symbol_name);
1995         if (ref->name == NULL) {
1996                 free(ref);
1997                 return -ENOMEM;
1998         }
1999
2000         bracket = strchr(ref->name, ']');
2001         if (bracket)
2002                 *bracket = '\0';
2003
2004         ref->addr = addr;
2005
2006         for (i = 0; i < MAP__NR_TYPES; ++i) {
2007                 struct kmap *kmap = map__kmap(maps[i]);
2008
2009                 if (!kmap)
2010                         continue;
2011                 kmap->ref_reloc_sym = ref;
2012         }
2013
2014         return 0;
2015 }
2016
2017 size_t perf_session__fprintf_dsos(struct perf_session *session, FILE *fp)
2018 {
2019         return machines__fprintf_dsos(&session->machines, fp);
2020 }
2021
2022 size_t perf_session__fprintf_dsos_buildid(struct perf_session *session, FILE *fp,
2023                                           bool (skip)(struct dso *dso, int parm), int parm)
2024 {
2025         return machines__fprintf_dsos_buildid(&session->machines, fp, skip, parm);
2026 }
2027
2028 size_t perf_session__fprintf_nr_events(struct perf_session *session, FILE *fp)
2029 {
2030         size_t ret;
2031         const char *msg = "";
2032
2033         if (perf_header__has_feat(&session->header, HEADER_AUXTRACE))
2034                 msg = " (excludes AUX area (e.g. instruction trace) decoded / synthesized events)";
2035
2036         ret = fprintf(fp, "\nAggregated stats:%s\n", msg);
2037
2038         ret += events_stats__fprintf(&session->evlist->stats, fp);
2039         return ret;
2040 }
2041
2042 size_t perf_session__fprintf(struct perf_session *session, FILE *fp)
2043 {
2044         /*
2045          * FIXME: Here we have to actually print all the machines in this
2046          * session, not just the host...
2047          */
2048         return machine__fprintf(&session->machines.host, fp);
2049 }
2050
2051 struct perf_evsel *perf_session__find_first_evtype(struct perf_session *session,
2052                                               unsigned int type)
2053 {
2054         struct perf_evsel *pos;
2055
2056         evlist__for_each_entry(session->evlist, pos) {
2057                 if (pos->attr.type == type)
2058                         return pos;
2059         }
2060         return NULL;
2061 }
2062
2063 int perf_session__cpu_bitmap(struct perf_session *session,
2064                              const char *cpu_list, unsigned long *cpu_bitmap)
2065 {
2066         int i, err = -1;
2067         struct cpu_map *map;
2068
2069         for (i = 0; i < PERF_TYPE_MAX; ++i) {
2070                 struct perf_evsel *evsel;
2071
2072                 evsel = perf_session__find_first_evtype(session, i);
2073                 if (!evsel)
2074                         continue;
2075
2076                 if (!(evsel->attr.sample_type & PERF_SAMPLE_CPU)) {
2077                         pr_err("File does not contain CPU events. "
2078                                "Remove -C option to proceed.\n");
2079                         return -1;
2080                 }
2081         }
2082
2083         map = cpu_map__new(cpu_list);
2084         if (map == NULL) {
2085                 pr_err("Invalid cpu_list\n");
2086                 return -1;
2087         }
2088
2089         for (i = 0; i < map->nr; i++) {
2090                 int cpu = map->map[i];
2091
2092                 if (cpu >= MAX_NR_CPUS) {
2093                         pr_err("Requested CPU %d too large. "
2094                                "Consider raising MAX_NR_CPUS\n", cpu);
2095                         goto out_delete_map;
2096                 }
2097
2098                 set_bit(cpu, cpu_bitmap);
2099         }
2100
2101         err = 0;
2102
2103 out_delete_map:
2104         cpu_map__put(map);
2105         return err;
2106 }
2107
2108 void perf_session__fprintf_info(struct perf_session *session, FILE *fp,
2109                                 bool full)
2110 {
2111         if (session == NULL || fp == NULL)
2112                 return;
2113
2114         fprintf(fp, "# ========\n");
2115         perf_header__fprintf_info(session, fp, full);
2116         fprintf(fp, "# ========\n#\n");
2117 }
2118
2119
2120 int __perf_session__set_tracepoints_handlers(struct perf_session *session,
2121                                              const struct perf_evsel_str_handler *assocs,
2122                                              size_t nr_assocs)
2123 {
2124         struct perf_evsel *evsel;
2125         size_t i;
2126         int err;
2127
2128         for (i = 0; i < nr_assocs; i++) {
2129                 /*
2130                  * Adding a handler for an event not in the session,
2131                  * just ignore it.
2132                  */
2133                 evsel = perf_evlist__find_tracepoint_by_name(session->evlist, assocs[i].name);
2134                 if (evsel == NULL)
2135                         continue;
2136
2137                 err = -EEXIST;
2138                 if (evsel->handler != NULL)
2139                         goto out;
2140                 evsel->handler = assocs[i].handler;
2141         }
2142
2143         err = 0;
2144 out:
2145         return err;
2146 }
2147
2148 int perf_event__process_id_index(struct perf_tool *tool __maybe_unused,
2149                                  union perf_event *event,
2150                                  struct perf_session *session)
2151 {
2152         struct perf_evlist *evlist = session->evlist;
2153         struct id_index_event *ie = &event->id_index;
2154         size_t i, nr, max_nr;
2155
2156         max_nr = (ie->header.size - sizeof(struct id_index_event)) /
2157                  sizeof(struct id_index_entry);
2158         nr = ie->nr;
2159         if (nr > max_nr)
2160                 return -EINVAL;
2161
2162         if (dump_trace)
2163                 fprintf(stdout, " nr: %zu\n", nr);
2164
2165         for (i = 0; i < nr; i++) {
2166                 struct id_index_entry *e = &ie->entries[i];
2167                 struct perf_sample_id *sid;
2168
2169                 if (dump_trace) {
2170                         fprintf(stdout, " ... id: %"PRIu64, e->id);
2171                         fprintf(stdout, "  idx: %"PRIu64, e->idx);
2172                         fprintf(stdout, "  cpu: %"PRId64, e->cpu);
2173                         fprintf(stdout, "  tid: %"PRId64"\n", e->tid);
2174                 }
2175
2176                 sid = perf_evlist__id2sid(evlist, e->id);
2177                 if (!sid)
2178                         return -ENOENT;
2179                 sid->idx = e->idx;
2180                 sid->cpu = e->cpu;
2181                 sid->tid = e->tid;
2182         }
2183         return 0;
2184 }
2185
2186 int perf_event__synthesize_id_index(struct perf_tool *tool,
2187                                     perf_event__handler_t process,
2188                                     struct perf_evlist *evlist,
2189                                     struct machine *machine)
2190 {
2191         union perf_event *ev;
2192         struct perf_evsel *evsel;
2193         size_t nr = 0, i = 0, sz, max_nr, n;
2194         int err;
2195
2196         pr_debug2("Synthesizing id index\n");
2197
2198         max_nr = (UINT16_MAX - sizeof(struct id_index_event)) /
2199                  sizeof(struct id_index_entry);
2200
2201         evlist__for_each_entry(evlist, evsel)
2202                 nr += evsel->ids;
2203
2204         n = nr > max_nr ? max_nr : nr;
2205         sz = sizeof(struct id_index_event) + n * sizeof(struct id_index_entry);
2206         ev = zalloc(sz);
2207         if (!ev)
2208                 return -ENOMEM;
2209
2210         ev->id_index.header.type = PERF_RECORD_ID_INDEX;
2211         ev->id_index.header.size = sz;
2212         ev->id_index.nr = n;
2213
2214         evlist__for_each_entry(evlist, evsel) {
2215                 u32 j;
2216
2217                 for (j = 0; j < evsel->ids; j++) {
2218                         struct id_index_entry *e;
2219                         struct perf_sample_id *sid;
2220
2221                         if (i >= n) {
2222                                 err = process(tool, ev, NULL, machine);
2223                                 if (err)
2224                                         goto out_err;
2225                                 nr -= n;
2226                                 i = 0;
2227                         }
2228
2229                         e = &ev->id_index.entries[i++];
2230
2231                         e->id = evsel->id[j];
2232
2233                         sid = perf_evlist__id2sid(evlist, e->id);
2234                         if (!sid) {
2235                                 free(ev);
2236                                 return -ENOENT;
2237                         }
2238
2239                         e->idx = sid->idx;
2240                         e->cpu = sid->cpu;
2241                         e->tid = sid->tid;
2242                 }
2243         }
2244
2245         sz = sizeof(struct id_index_event) + nr * sizeof(struct id_index_entry);
2246         ev->id_index.header.size = sz;
2247         ev->id_index.nr = nr;
2248
2249         err = process(tool, ev, NULL, machine);
2250 out_err:
2251         free(ev);
2252
2253         return err;
2254 }