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