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