tools/perf/util/session.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 #include <errno.h>
   3 #include <inttypes.h>
   4 #include <linux/err.h>
   5 #include <linux/kernel.h>
   6 #include <linux/zalloc.h>
   7 #include <api/fs/fs.h>
   8
   9 #include <byteswap.h>
  10 #include <unistd.h>
  11 #include <sys/types.h>
  12 #include <sys/mman.h>
  13 #include <perf/cpumap.h>
  14
  15 #include "map_symbol.h"
  16 #include "branch.h"
  17 #include "debug.h"
  18 #include "evlist.h"
  19 #include "evsel.h"
  20 #include "memswap.h"
  21 #include "map.h"
  22 #include "symbol.h"
  23 #include "session.h"
  24 #include "tool.h"
  25 #include "perf_regs.h"
  26 #include "asm/bug.h"
  27 #include "auxtrace.h"
  28 #include "thread.h"
  29 #include "thread-stack.h"
  30 #include "sample-raw.h"
  31 #include "stat.h"
  32 #include "ui/progress.h"
  33 #include "../perf.h"
  34 #include "arch/common.h"
  35 #include <internal/lib.h>
  36
  37 #ifdef HAVE_ZSTD_SUPPORT
  38 static int perf_session__process_compressed_event(struct perf_session *session,
  39                                                   union perf_event *event, u64 file_offset)
  40 {
  41         void *src;
  42         size_t decomp_size, src_size;
  43         u64 decomp_last_rem = 0;
  44         size_t mmap_len, decomp_len = session->header.env.comp_mmap_len;
  45         struct decomp *decomp, *decomp_last = session->decomp_last;
  46
  47         if (decomp_last) {
  48                 decomp_last_rem = decomp_last->size - decomp_last->head;
  49                 decomp_len += decomp_last_rem;
  50         }
  51
  52         mmap_len = sizeof(struct decomp) + decomp_len;
  53         decomp = mmap(NULL, mmap_len, PROT_READ|PROT_WRITE,
  54                       MAP_ANONYMOUS|MAP_PRIVATE, -1, 0);
  55         if (decomp == MAP_FAILED) {
  56                 pr_err("Couldn't allocate memory for decompression\n");
  57                 return -1;
  58         }
  59
  60         decomp->file_pos = file_offset;
  61         decomp->mmap_len = mmap_len;
  62         decomp->head = 0;
  63
  64         if (decomp_last_rem) {
  65                 memcpy(decomp->data, &(decomp_last->data[decomp_last->head]), decomp_last_rem);
  66                 decomp->size = decomp_last_rem;
  67         }
  68
  69         src = (void *)event + sizeof(struct perf_record_compressed);
  70         src_size = event->pack.header.size - sizeof(struct perf_record_compressed);
  71
  72         decomp_size = zstd_decompress_stream(&(session->zstd_data), src, src_size,
  73                                 &(decomp->data[decomp_last_rem]), decomp_len - decomp_last_rem);
  74         if (!decomp_size) {
  75                 munmap(decomp, mmap_len);
  76                 pr_err("Couldn't decompress data\n");
  77                 return -1;
  78         }
  79
  80         decomp->size += decomp_size;
  81
  82         if (session->decomp == NULL) {
  83                 session->decomp = decomp;
  84                 session->decomp_last = decomp;
  85         } else {
  86                 session->decomp_last->next = decomp;
  87                 session->decomp_last = decomp;
  88         }
  89
  90         pr_debug("decomp (B): %ld to %ld\n", src_size, decomp_size);
  91
  92         return 0;
  93 }
  94 #else /* !HAVE_ZSTD_SUPPORT */
  95 #define perf_session__process_compressed_event perf_session__process_compressed_event_stub
  96 #endif
  97
  98 static int perf_session__deliver_event(struct perf_session *session,
  99                                        union perf_event *event,
 100                                        struct perf_tool *tool,
 101                                        u64 file_offset);
 102
 103 static int perf_session__open(struct perf_session *session)
 104 {
 105         struct perf_data *data = session->data;
 106
 107         if (perf_session__read_header(session) < 0) {
 108                 pr_err("incompatible file format (rerun with -v to learn more)\n");
 109                 return -1;
 110         }
 111
 112         if (perf_data__is_pipe(data))
 113                 return 0;
 114
 115         if (perf_header__has_feat(&session->header, HEADER_STAT))
 116                 return 0;
 117
 118         if (!perf_evlist__valid_sample_type(session->evlist)) {
 119                 pr_err("non matching sample_type\n");
 120                 return -1;
 121         }
 122
 123         if (!perf_evlist__valid_sample_id_all(session->evlist)) {
 124                 pr_err("non matching sample_id_all\n");
 125                 return -1;
 126         }
 127
 128         if (!perf_evlist__valid_read_format(session->evlist)) {
 129                 pr_err("non matching read_format\n");
 130                 return -1;
 131         }
 132
 133         return 0;
 134 }
 135
 136 void perf_session__set_id_hdr_size(struct perf_session *session)
 137 {
 138         u16 id_hdr_size = perf_evlist__id_hdr_size(session->evlist);
 139
 140         machines__set_id_hdr_size(&session->machines, id_hdr_size);
 141 }
 142
 143 int perf_session__create_kernel_maps(struct perf_session *session)
 144 {
 145         int ret = machine__create_kernel_maps(&session->machines.host);
 146
 147         if (ret >= 0)
 148                 ret = machines__create_guest_kernel_maps(&session->machines);
 149         return ret;
 150 }
 151
 152 static void perf_session__destroy_kernel_maps(struct perf_session *session)
 153 {
 154         machines__destroy_kernel_maps(&session->machines);
 155 }
 156
 157 static bool perf_session__has_comm_exec(struct perf_session *session)
 158 {
 159         struct evsel *evsel;
 160
 161         evlist__for_each_entry(session->evlist, evsel) {
 162                 if (evsel->core.attr.comm_exec)
 163                         return true;
 164         }
 165
 166         return false;
 167 }
 168
 169 static void perf_session__set_comm_exec(struct perf_session *session)
 170 {
 171         bool comm_exec = perf_session__has_comm_exec(session);
 172
 173         machines__set_comm_exec(&session->machines, comm_exec);
 174 }
 175
 176 static int ordered_events__deliver_event(struct ordered_events *oe,
 177                                          struct ordered_event *event)
 178 {
 179         struct perf_session *session = container_of(oe, struct perf_session,
 180                                                     ordered_events);
 181
 182         return perf_session__deliver_event(session, event->event,
 183                                            session->tool, event->file_offset);
 184 }
 185
 186 struct perf_session *perf_session__new(struct perf_data *data,
 187                                        bool repipe, struct perf_tool *tool)
 188 {
 189         int ret = -ENOMEM;
 190         struct perf_session *session = zalloc(sizeof(*session));
 191
 192         if (!session)
 193                 goto out;
 194
 195         session->repipe = repipe;
 196         session->tool   = tool;
 197         INIT_LIST_HEAD(&session->auxtrace_index);
 198         machines__init(&session->machines);
 199         ordered_events__init(&session->ordered_events,
 200                              ordered_events__deliver_event, NULL);
 201
 202         perf_env__init(&session->header.env);
 203         if (data) {
 204                 ret = perf_data__open(data);
 205                 if (ret < 0)
 206                         goto out_delete;
 207
 208                 session->data = data;
 209
 210                 if (perf_data__is_read(data)) {
 211                         ret = perf_session__open(session);
 212                         if (ret < 0)
 213                                 goto out_delete;
 214
 215                         /*
 216                          * set session attributes that are present in perf.data
 217                          * but not in pipe-mode.
 218                          */
 219                         if (!data->is_pipe) {
 220                                 perf_session__set_id_hdr_size(session);
 221                                 perf_session__set_comm_exec(session);
 222                         }
 223
 224                         perf_evlist__init_trace_event_sample_raw(session->evlist);
 225
 226                         /* Open the directory data. */
 227                         if (data->is_dir) {
 228                                 ret = perf_data__open_dir(data);
 229                                 if (ret)
 230                                         goto out_delete;
 231                         }
 232
 233                         if (!symbol_conf.kallsyms_name &&
 234                             !symbol_conf.vmlinux_name)
 235                                 symbol_conf.kallsyms_name = perf_data__kallsyms_name(data);
 236                 }
 237         } else  {
 238                 session->machines.host.env = &perf_env;
 239         }
 240
 241         session->machines.host.single_address_space =
 242                 perf_env__single_address_space(session->machines.host.env);
 243
 244         if (!data || perf_data__is_write(data)) {
 245                 /*
 246                  * In O_RDONLY mode this will be performed when reading the
 247                  * kernel MMAP event, in perf_event__process_mmap().
 248                  */
 249                 if (perf_session__create_kernel_maps(session) < 0)
 250                         pr_warning("Cannot read kernel map\n");
 251         }
 252
 253         /*
 254          * In pipe-mode, evlist is empty until PERF_RECORD_HEADER_ATTR is
 255          * processed, so perf_evlist__sample_id_all is not meaningful here.
 256          */
 257         if ((!data || !data->is_pipe) && tool && tool->ordering_requires_timestamps &&
 258             tool->ordered_events && !perf_evlist__sample_id_all(session->evlist)) {
 259                 dump_printf("WARNING: No sample_id_all support, falling back to unordered processing\n");
 260                 tool->ordered_events = false;
 261         }
 262
 263         return session;
 264
 265  out_delete:
 266         perf_session__delete(session);
 267  out:
 268         return ERR_PTR(ret);
 269 }
 270
 271 static void perf_session__delete_threads(struct perf_session *session)
 272 {
 273         machine__delete_threads(&session->machines.host);
 274 }
 275
 276 static void perf_session__release_decomp_events(struct perf_session *session)
 277 {
 278         struct decomp *next, *decomp;
 279         size_t mmap_len;
 280         next = session->decomp;
 281         do {
 282                 decomp = next;
 283                 if (decomp == NULL)
 284                         break;
 285                 next = decomp->next;
 286                 mmap_len = decomp->mmap_len;
 287                 munmap(decomp, mmap_len);
 288         } while (1);
 289 }
 290
 291 void perf_session__delete(struct perf_session *session)
 292 {
 293         if (session == NULL)
 294                 return;
 295         auxtrace__free(session);
 296         auxtrace_index__free(&session->auxtrace_index);
 297         perf_session__destroy_kernel_maps(session);
 298         perf_session__delete_threads(session);
 299         perf_session__release_decomp_events(session);
 300         perf_env__exit(&session->header.env);
 301         machines__exit(&session->machines);
 302         if (session->data)
 303                 perf_data__close(session->data);
 304         free(session);
 305 }
 306
 307 static int process_event_synth_tracing_data_stub(struct perf_session *session
 308                                                  __maybe_unused,
 309                                                  union perf_event *event
 310                                                  __maybe_unused)
 311 {
 312         dump_printf(": unhandled!\n");
 313         return 0;
 314 }
 315
 316 static int process_event_synth_attr_stub(struct perf_tool *tool __maybe_unused,
 317                                          union perf_event *event __maybe_unused,
 318                                          struct evlist **pevlist
 319                                          __maybe_unused)
 320 {
 321         dump_printf(": unhandled!\n");
 322         return 0;
 323 }
 324
 325 static int process_event_synth_event_update_stub(struct perf_tool *tool __maybe_unused,
 326                                                  union perf_event *event __maybe_unused,
 327                                                  struct evlist **pevlist
 328                                                  __maybe_unused)
 329 {
 330         if (dump_trace)
 331                 perf_event__fprintf_event_update(event, stdout);
 332
 333         dump_printf(": unhandled!\n");
 334         return 0;
 335 }
 336
 337 static int process_event_sample_stub(struct perf_tool *tool __maybe_unused,
 338                                      union perf_event *event __maybe_unused,
 339                                      struct perf_sample *sample __maybe_unused,
 340                                      struct evsel *evsel __maybe_unused,
 341                                      struct machine *machine __maybe_unused)
 342 {
 343         dump_printf(": unhandled!\n");
 344         return 0;
 345 }
 346
 347 static int process_event_stub(struct perf_tool *tool __maybe_unused,
 348                               union perf_event *event __maybe_unused,
 349                               struct perf_sample *sample __maybe_unused,
 350                               struct machine *machine __maybe_unused)
 351 {
 352         dump_printf(": unhandled!\n");
 353         return 0;
 354 }
 355
 356 static int process_finished_round_stub(struct perf_tool *tool __maybe_unused,
 357                                        union perf_event *event __maybe_unused,
 358                                        struct ordered_events *oe __maybe_unused)
 359 {
 360         dump_printf(": unhandled!\n");
 361         return 0;
 362 }
 363
 364 static int process_finished_round(struct perf_tool *tool,
 365                                   union perf_event *event,
 366                                   struct ordered_events *oe);
 367
 368 static int skipn(int fd, off_t n)
 369 {
 370         char buf[4096];
 371         ssize_t ret;
 372
 373         while (n > 0) {
 374                 ret = read(fd, buf, min(n, (off_t)sizeof(buf)));
 375                 if (ret <= 0)
 376                         return ret;
 377                 n -= ret;
 378         }
 379
 380         return 0;
 381 }
 382
 383 static s64 process_event_auxtrace_stub(struct perf_session *session __maybe_unused,
 384                                        union perf_event *event)
 385 {
 386         dump_printf(": unhandled!\n");
 387         if (perf_data__is_pipe(session->data))
 388                 skipn(perf_data__fd(session->data), event->auxtrace.size);
 389         return event->auxtrace.size;
 390 }
 391
 392 static int process_event_op2_stub(struct perf_session *session __maybe_unused,
 393                                   union perf_event *event __maybe_unused)
 394 {
 395         dump_printf(": unhandled!\n");
 396         return 0;
 397 }
 398
 399
 400 static
 401 int process_event_thread_map_stub(struct perf_session *session __maybe_unused,
 402                                   union perf_event *event __maybe_unused)
 403 {
 404         if (dump_trace)
 405                 perf_event__fprintf_thread_map(event, stdout);
 406
 407         dump_printf(": unhandled!\n");
 408         return 0;
 409 }
 410
 411 static
 412 int process_event_cpu_map_stub(struct perf_session *session __maybe_unused,
 413                                union perf_event *event __maybe_unused)
 414 {
 415         if (dump_trace)
 416                 perf_event__fprintf_cpu_map(event, stdout);
 417
 418         dump_printf(": unhandled!\n");
 419         return 0;
 420 }
 421
 422 static
 423 int process_event_stat_config_stub(struct perf_session *session __maybe_unused,
 424                                    union perf_event *event __maybe_unused)
 425 {
 426         if (dump_trace)
 427                 perf_event__fprintf_stat_config(event, stdout);
 428
 429         dump_printf(": unhandled!\n");
 430         return 0;
 431 }
 432
 433 static int process_stat_stub(struct perf_session *perf_session __maybe_unused,
 434                              union perf_event *event)
 435 {
 436         if (dump_trace)
 437                 perf_event__fprintf_stat(event, stdout);
 438
 439         dump_printf(": unhandled!\n");
 440         return 0;
 441 }
 442
 443 static int process_stat_round_stub(struct perf_session *perf_session __maybe_unused,
 444                                    union perf_event *event)
 445 {
 446         if (dump_trace)
 447                 perf_event__fprintf_stat_round(event, stdout);
 448
 449         dump_printf(": unhandled!\n");
 450         return 0;
 451 }
 452
 453 static int perf_session__process_compressed_event_stub(struct perf_session *session __maybe_unused,
 454                                                        union perf_event *event __maybe_unused,
 455                                                        u64 file_offset __maybe_unused)
 456 {
 457        dump_printf(": unhandled!\n");
 458        return 0;
 459 }
 460
 461 void perf_tool__fill_defaults(struct perf_tool *tool)
 462 {
 463         if (tool->sample == NULL)
 464                 tool->sample = process_event_sample_stub;
 465         if (tool->mmap == NULL)
 466                 tool->mmap = process_event_stub;
 467         if (tool->mmap2 == NULL)
 468                 tool->mmap2 = process_event_stub;
 469         if (tool->comm == NULL)
 470                 tool->comm = process_event_stub;
 471         if (tool->namespaces == NULL)
 472                 tool->namespaces = process_event_stub;
 473         if (tool->cgroup == NULL)
 474                 tool->cgroup = process_event_stub;
 475         if (tool->fork == NULL)
 476                 tool->fork = process_event_stub;
 477         if (tool->exit == NULL)
 478                 tool->exit = process_event_stub;
 479         if (tool->lost == NULL)
 480                 tool->lost = perf_event__process_lost;
 481         if (tool->lost_samples == NULL)
 482                 tool->lost_samples = perf_event__process_lost_samples;
 483         if (tool->aux == NULL)
 484                 tool->aux = perf_event__process_aux;
 485         if (tool->itrace_start == NULL)
 486                 tool->itrace_start = perf_event__process_itrace_start;
 487         if (tool->context_switch == NULL)
 488                 tool->context_switch = perf_event__process_switch;
 489         if (tool->ksymbol == NULL)
 490                 tool->ksymbol = perf_event__process_ksymbol;
 491         if (tool->bpf == NULL)
 492                 tool->bpf = perf_event__process_bpf;
 493         if (tool->read == NULL)
 494                 tool->read = process_event_sample_stub;
 495         if (tool->throttle == NULL)
 496                 tool->throttle = process_event_stub;
 497         if (tool->unthrottle == NULL)
 498                 tool->unthrottle = process_event_stub;
 499         if (tool->attr == NULL)
 500                 tool->attr = process_event_synth_attr_stub;
 501         if (tool->event_update == NULL)
 502                 tool->event_update = process_event_synth_event_update_stub;
 503         if (tool->tracing_data == NULL)
 504                 tool->tracing_data = process_event_synth_tracing_data_stub;
 505         if (tool->build_id == NULL)
 506                 tool->build_id = process_event_op2_stub;
 507         if (tool->finished_round == NULL) {
 508                 if (tool->ordered_events)
 509                         tool->finished_round = process_finished_round;
 510                 else
 511                         tool->finished_round = process_finished_round_stub;
 512         }
 513         if (tool->id_index == NULL)
 514                 tool->id_index = process_event_op2_stub;
 515         if (tool->auxtrace_info == NULL)
 516                 tool->auxtrace_info = process_event_op2_stub;
 517         if (tool->auxtrace == NULL)
 518                 tool->auxtrace = process_event_auxtrace_stub;
 519         if (tool->auxtrace_error == NULL)
 520                 tool->auxtrace_error = process_event_op2_stub;
 521         if (tool->thread_map == NULL)
 522                 tool->thread_map = process_event_thread_map_stub;
 523         if (tool->cpu_map == NULL)
 524                 tool->cpu_map = process_event_cpu_map_stub;
 525         if (tool->stat_config == NULL)
 526                 tool->stat_config = process_event_stat_config_stub;
 527         if (tool->stat == NULL)
 528                 tool->stat = process_stat_stub;
 529         if (tool->stat_round == NULL)
 530                 tool->stat_round = process_stat_round_stub;
 531         if (tool->time_conv == NULL)
 532                 tool->time_conv = process_event_op2_stub;
 533         if (tool->feature == NULL)
 534                 tool->feature = process_event_op2_stub;
 535         if (tool->compressed == NULL)
 536                 tool->compressed = perf_session__process_compressed_event;
 537 }
 538
 539 static void swap_sample_id_all(union perf_event *event, void *data)
 540 {
 541         void *end = (void *) event + event->header.size;
 542         int size = end - data;
 543
 544         BUG_ON(size % sizeof(u64));
 545         mem_bswap_64(data, size);
 546 }
 547
 548 static void perf_event__all64_swap(union perf_event *event,
 549                                    bool sample_id_all __maybe_unused)
 550 {
 551         struct perf_event_header *hdr = &event->header;
 552         mem_bswap_64(hdr + 1, event->header.size - sizeof(*hdr));
 553 }
 554
 555 static void perf_event__comm_swap(union perf_event *event, bool sample_id_all)
 556 {
 557         event->comm.pid = bswap_32(event->comm.pid);
 558         event->comm.tid = bswap_32(event->comm.tid);
 559
 560         if (sample_id_all) {
 561                 void *data = &event->comm.comm;
 562
 563                 data += PERF_ALIGN(strlen(data) + 1, sizeof(u64));
 564                 swap_sample_id_all(event, data);
 565         }
 566 }
 567
 568 static void perf_event__mmap_swap(union perf_event *event,
 569                                   bool sample_id_all)
 570 {
 571         event->mmap.pid   = bswap_32(event->mmap.pid);
 572         event->mmap.tid   = bswap_32(event->mmap.tid);
 573         event->mmap.start = bswap_64(event->mmap.start);
 574         event->mmap.len   = bswap_64(event->mmap.len);
 575         event->mmap.pgoff = bswap_64(event->mmap.pgoff);
 576
 577         if (sample_id_all) {
 578                 void *data = &event->mmap.filename;
 579
 580                 data += PERF_ALIGN(strlen(data) + 1, sizeof(u64));
 581                 swap_sample_id_all(event, data);
 582         }
 583 }
 584
 585 static void perf_event__mmap2_swap(union perf_event *event,
 586                                   bool sample_id_all)
 587 {
 588         event->mmap2.pid   = bswap_32(event->mmap2.pid);
 589         event->mmap2.tid   = bswap_32(event->mmap2.tid);
 590         event->mmap2.start = bswap_64(event->mmap2.start);
 591         event->mmap2.len   = bswap_64(event->mmap2.len);
 592         event->mmap2.pgoff = bswap_64(event->mmap2.pgoff);
 593         event->mmap2.maj   = bswap_32(event->mmap2.maj);
 594         event->mmap2.min   = bswap_32(event->mmap2.min);
 595         event->mmap2.ino   = bswap_64(event->mmap2.ino);
 596
 597         if (sample_id_all) {
 598                 void *data = &event->mmap2.filename;
 599
 600                 data += PERF_ALIGN(strlen(data) + 1, sizeof(u64));
 601                 swap_sample_id_all(event, data);
 602         }
 603 }
 604 static void perf_event__task_swap(union perf_event *event, bool sample_id_all)
 605 {
 606         event->fork.pid  = bswap_32(event->fork.pid);
 607         event->fork.tid  = bswap_32(event->fork.tid);
 608         event->fork.ppid = bswap_32(event->fork.ppid);
 609         event->fork.ptid = bswap_32(event->fork.ptid);
 610         event->fork.time = bswap_64(event->fork.time);
 611
 612         if (sample_id_all)
 613                 swap_sample_id_all(event, &event->fork + 1);
 614 }
 615
 616 static void perf_event__read_swap(union perf_event *event, bool sample_id_all)
 617 {
 618         event->read.pid          = bswap_32(event->read.pid);
 619         event->read.tid          = bswap_32(event->read.tid);
 620         event->read.value        = bswap_64(event->read.value);
 621         event->read.time_enabled = bswap_64(event->read.time_enabled);
 622         event->read.time_running = bswap_64(event->read.time_running);
 623         event->read.id           = bswap_64(event->read.id);
 624
 625         if (sample_id_all)
 626                 swap_sample_id_all(event, &event->read + 1);
 627 }
 628
 629 static void perf_event__aux_swap(union perf_event *event, bool sample_id_all)
 630 {
 631         event->aux.aux_offset = bswap_64(event->aux.aux_offset);
 632         event->aux.aux_size   = bswap_64(event->aux.aux_size);
 633         event->aux.flags      = bswap_64(event->aux.flags);
 634
 635         if (sample_id_all)
 636                 swap_sample_id_all(event, &event->aux + 1);
 637 }
 638
 639 static void perf_event__itrace_start_swap(union perf_event *event,
 640                                           bool sample_id_all)
 641 {
 642         event->itrace_start.pid  = bswap_32(event->itrace_start.pid);
 643         event->itrace_start.tid  = bswap_32(event->itrace_start.tid);
 644
 645         if (sample_id_all)
 646                 swap_sample_id_all(event, &event->itrace_start + 1);
 647 }
 648
 649 static void perf_event__switch_swap(union perf_event *event, bool sample_id_all)
 650 {
 651         if (event->header.type == PERF_RECORD_SWITCH_CPU_WIDE) {
 652                 event->context_switch.next_prev_pid =
 653                                 bswap_32(event->context_switch.next_prev_pid);
 654                 event->context_switch.next_prev_tid =
 655                                 bswap_32(event->context_switch.next_prev_tid);
 656         }
 657
 658         if (sample_id_all)
 659                 swap_sample_id_all(event, &event->context_switch + 1);
 660 }
 661
 662 static void perf_event__throttle_swap(union perf_event *event,
 663                                       bool sample_id_all)
 664 {
 665         event->throttle.time      = bswap_64(event->throttle.time);
 666         event->throttle.id        = bswap_64(event->throttle.id);
 667         event->throttle.stream_id = bswap_64(event->throttle.stream_id);
 668
 669         if (sample_id_all)
 670                 swap_sample_id_all(event, &event->throttle + 1);
 671 }
 672
 673 static void perf_event__namespaces_swap(union perf_event *event,
 674                                         bool sample_id_all)
 675 {
 676         u64 i;
 677
 678         event->namespaces.pid           = bswap_32(event->namespaces.pid);
 679         event->namespaces.tid           = bswap_32(event->namespaces.tid);
 680         event->namespaces.nr_namespaces = bswap_64(event->namespaces.nr_namespaces);
 681
 682         for (i = 0; i < event->namespaces.nr_namespaces; i++) {
 683                 struct perf_ns_link_info *ns = &event->namespaces.link_info[i];
 684
 685                 ns->dev = bswap_64(ns->dev);
 686                 ns->ino = bswap_64(ns->ino);
 687         }
 688
 689         if (sample_id_all)
 690                 swap_sample_id_all(event, &event->namespaces.link_info[i]);
 691 }
 692
 693 static u8 revbyte(u8 b)
 694 {
 695         int rev = (b >> 4) | ((b & 0xf) << 4);
 696         rev = ((rev & 0xcc) >> 2) | ((rev & 0x33) << 2);
 697         rev = ((rev & 0xaa) >> 1) | ((rev & 0x55) << 1);
 698         return (u8) rev;
 699 }
 700
 701 /*
 702  * XXX this is hack in attempt to carry flags bitfield
 703  * through endian village. ABI says:
 704  *
 705  * Bit-fields are allocated from right to left (least to most significant)
 706  * on little-endian implementations and from left to right (most to least
 707  * significant) on big-endian implementations.
 708  *
 709  * The above seems to be byte specific, so we need to reverse each
 710  * byte of the bitfield. 'Internet' also says this might be implementation
 711  * specific and we probably need proper fix and carry perf_event_attr
 712  * bitfield flags in separate data file FEAT_ section. Thought this seems
 713  * to work for now.
 714  */
 715 static void swap_bitfield(u8 *p, unsigned len)
 716 {
 717         unsigned i;
 718
 719         for (i = 0; i < len; i++) {
 720                 *p = revbyte(*p);
 721                 p++;
 722         }
 723 }
 724
 725 /* exported for swapping attributes in file header */
 726 void perf_event__attr_swap(struct perf_event_attr *attr)
 727 {
 728         attr->type              = bswap_32(attr->type);
 729         attr->size              = bswap_32(attr->size);
 730
 731 #define bswap_safe(f, n)                                        \
 732         (attr->size > (offsetof(struct perf_event_attr, f) +    \
 733                        sizeof(attr->f) * (n)))
 734 #define bswap_field(f, sz)                      \
 735 do {                                            \
 736         if (bswap_safe(f, 0))                   \
 737                 attr->f = bswap_##sz(attr->f);  \
 738 } while(0)
 739 #define bswap_field_16(f) bswap_field(f, 16)
 740 #define bswap_field_32(f) bswap_field(f, 32)
 741 #define bswap_field_64(f) bswap_field(f, 64)
 742
 743         bswap_field_64(config);
 744         bswap_field_64(sample_period);
 745         bswap_field_64(sample_type);
 746         bswap_field_64(read_format);
 747         bswap_field_32(wakeup_events);
 748         bswap_field_32(bp_type);
 749         bswap_field_64(bp_addr);
 750         bswap_field_64(bp_len);
 751         bswap_field_64(branch_sample_type);
 752         bswap_field_64(sample_regs_user);
 753         bswap_field_32(sample_stack_user);
 754         bswap_field_32(aux_watermark);
 755         bswap_field_16(sample_max_stack);
 756         bswap_field_32(aux_sample_size);
 757
 758         /*
 759          * After read_format are bitfields. Check read_format because
 760          * we are unable to use offsetof on bitfield.
 761          */
 762         if (bswap_safe(read_format, 1))
 763                 swap_bitfield((u8 *) (&attr->read_format + 1),
 764                               sizeof(u64));
 765 #undef bswap_field_64
 766 #undef bswap_field_32
 767 #undef bswap_field
 768 #undef bswap_safe
 769 }
 770
 771 static void perf_event__hdr_attr_swap(union perf_event *event,
 772                                       bool sample_id_all __maybe_unused)
 773 {
 774         size_t size;
 775
 776         perf_event__attr_swap(&event->attr.attr);
 777
 778         size = event->header.size;
 779         size -= (void *)&event->attr.id - (void *)event;
 780         mem_bswap_64(event->attr.id, size);
 781 }
 782
 783 static void perf_event__event_update_swap(union perf_event *event,
 784                                           bool sample_id_all __maybe_unused)
 785 {
 786         event->event_update.type = bswap_64(event->event_update.type);
 787         event->event_update.id   = bswap_64(event->event_update.id);
 788 }
 789
 790 static void perf_event__event_type_swap(union perf_event *event,
 791                                         bool sample_id_all __maybe_unused)
 792 {
 793         event->event_type.event_type.event_id =
 794                 bswap_64(event->event_type.event_type.event_id);
 795 }
 796
 797 static void perf_event__tracing_data_swap(union perf_event *event,
 798                                           bool sample_id_all __maybe_unused)
 799 {
 800         event->tracing_data.size = bswap_32(event->tracing_data.size);
 801 }
 802
 803 static void perf_event__auxtrace_info_swap(union perf_event *event,
 804                                            bool sample_id_all __maybe_unused)
 805 {
 806         size_t size;
 807
 808         event->auxtrace_info.type = bswap_32(event->auxtrace_info.type);
 809
 810         size = event->header.size;
 811         size -= (void *)&event->auxtrace_info.priv - (void *)event;
 812         mem_bswap_64(event->auxtrace_info.priv, size);
 813 }
 814
 815 static void perf_event__auxtrace_swap(union perf_event *event,
 816                                       bool sample_id_all __maybe_unused)
 817 {
 818         event->auxtrace.size      = bswap_64(event->auxtrace.size);
 819         event->auxtrace.offset    = bswap_64(event->auxtrace.offset);
 820         event->auxtrace.reference = bswap_64(event->auxtrace.reference);
 821         event->auxtrace.idx       = bswap_32(event->auxtrace.idx);
 822         event->auxtrace.tid       = bswap_32(event->auxtrace.tid);
 823         event->auxtrace.cpu       = bswap_32(event->auxtrace.cpu);
 824 }
 825
 826 static void perf_event__auxtrace_error_swap(union perf_event *event,
 827                                             bool sample_id_all __maybe_unused)
 828 {
 829         event->auxtrace_error.type = bswap_32(event->auxtrace_error.type);
 830         event->auxtrace_error.code = bswap_32(event->auxtrace_error.code);
 831         event->auxtrace_error.cpu  = bswap_32(event->auxtrace_error.cpu);
 832         event->auxtrace_error.pid  = bswap_32(event->auxtrace_error.pid);
 833         event->auxtrace_error.tid  = bswap_32(event->auxtrace_error.tid);
 834         event->auxtrace_error.fmt  = bswap_32(event->auxtrace_error.fmt);
 835         event->auxtrace_error.ip   = bswap_64(event->auxtrace_error.ip);
 836         if (event->auxtrace_error.fmt)
 837                 event->auxtrace_error.time = bswap_64(event->auxtrace_error.time);
 838 }
 839
 840 static void perf_event__thread_map_swap(union perf_event *event,
 841                                         bool sample_id_all __maybe_unused)
 842 {
 843         unsigned i;
 844
 845         event->thread_map.nr = bswap_64(event->thread_map.nr);
 846
 847         for (i = 0; i < event->thread_map.nr; i++)
 848                 event->thread_map.entries[i].pid = bswap_64(event->thread_map.entries[i].pid);
 849 }
 850
 851 static void perf_event__cpu_map_swap(union perf_event *event,
 852                                      bool sample_id_all __maybe_unused)
 853 {
 854         struct perf_record_cpu_map_data *data = &event->cpu_map.data;
 855         struct cpu_map_entries *cpus;
 856         struct perf_record_record_cpu_map *mask;
 857         unsigned i;
 858
 859         data->type = bswap_64(data->type);
 860
 861         switch (data->type) {
 862         case PERF_CPU_MAP__CPUS:
 863                 cpus = (struct cpu_map_entries *)data->data;
 864
 865                 cpus->nr = bswap_16(cpus->nr);
 866
 867                 for (i = 0; i < cpus->nr; i++)
 868                         cpus->cpu[i] = bswap_16(cpus->cpu[i]);
 869                 break;
 870         case PERF_CPU_MAP__MASK:
 871                 mask = (struct perf_record_record_cpu_map *)data->data;
 872
 873                 mask->nr = bswap_16(mask->nr);
 874                 mask->long_size = bswap_16(mask->long_size);
 875
 876                 switch (mask->long_size) {
 877                 case 4: mem_bswap_32(&mask->mask, mask->nr); break;
 878                 case 8: mem_bswap_64(&mask->mask, mask->nr); break;
 879                 default:
 880                         pr_err("cpu_map swap: unsupported long size\n");
 881                 }
 882         default:
 883                 break;
 884         }
 885 }
 886
 887 static void perf_event__stat_config_swap(union perf_event *event,
 888                                          bool sample_id_all __maybe_unused)
 889 {
 890         u64 size;
 891
 892         size  = event->stat_config.nr * sizeof(event->stat_config.data[0]);
 893         size += 1; /* nr item itself */
 894         mem_bswap_64(&event->stat_config.nr, size);
 895 }
 896
 897 static void perf_event__stat_swap(union perf_event *event,
 898                                   bool sample_id_all __maybe_unused)
 899 {
 900         event->stat.id     = bswap_64(event->stat.id);
 901         event->stat.thread = bswap_32(event->stat.thread);
 902         event->stat.cpu    = bswap_32(event->stat.cpu);
 903         event->stat.val    = bswap_64(event->stat.val);
 904         event->stat.ena    = bswap_64(event->stat.ena);
 905         event->stat.run    = bswap_64(event->stat.run);
 906 }
 907
 908 static void perf_event__stat_round_swap(union perf_event *event,
 909                                         bool sample_id_all __maybe_unused)
 910 {
 911         event->stat_round.type = bswap_64(event->stat_round.type);
 912         event->stat_round.time = bswap_64(event->stat_round.time);
 913 }
 914
 915 typedef void (*perf_event__swap_op)(union perf_event *event,
 916                                     bool sample_id_all);
 917
 918 static perf_event__swap_op perf_event__swap_ops[] = {
 919         [PERF_RECORD_MMAP]                = perf_event__mmap_swap,
 920         [PERF_RECORD_MMAP2]               = perf_event__mmap2_swap,
 921         [PERF_RECORD_COMM]                = perf_event__comm_swap,
 922         [PERF_RECORD_FORK]                = perf_event__task_swap,
 923         [PERF_RECORD_EXIT]                = perf_event__task_swap,
 924         [PERF_RECORD_LOST]                = perf_event__all64_swap,
 925         [PERF_RECORD_READ]                = perf_event__read_swap,
 926         [PERF_RECORD_THROTTLE]            = perf_event__throttle_swap,
 927         [PERF_RECORD_UNTHROTTLE]          = perf_event__throttle_swap,
 928         [PERF_RECORD_SAMPLE]              = perf_event__all64_swap,
 929         [PERF_RECORD_AUX]                 = perf_event__aux_swap,
 930         [PERF_RECORD_ITRACE_START]        = perf_event__itrace_start_swap,
 931         [PERF_RECORD_LOST_SAMPLES]        = perf_event__all64_swap,
 932         [PERF_RECORD_SWITCH]              = perf_event__switch_swap,
 933         [PERF_RECORD_SWITCH_CPU_WIDE]     = perf_event__switch_swap,
 934         [PERF_RECORD_NAMESPACES]          = perf_event__namespaces_swap,
 935         [PERF_RECORD_HEADER_ATTR]         = perf_event__hdr_attr_swap,
 936         [PERF_RECORD_HEADER_EVENT_TYPE]   = perf_event__event_type_swap,
 937         [PERF_RECORD_HEADER_TRACING_DATA] = perf_event__tracing_data_swap,
 938         [PERF_RECORD_HEADER_BUILD_ID]     = NULL,
 939         [PERF_RECORD_ID_INDEX]            = perf_event__all64_swap,
 940         [PERF_RECORD_AUXTRACE_INFO]       = perf_event__auxtrace_info_swap,
 941         [PERF_RECORD_AUXTRACE]            = perf_event__auxtrace_swap,
 942         [PERF_RECORD_AUXTRACE_ERROR]      = perf_event__auxtrace_error_swap,
 943         [PERF_RECORD_THREAD_MAP]          = perf_event__thread_map_swap,
 944         [PERF_RECORD_CPU_MAP]             = perf_event__cpu_map_swap,
 945         [PERF_RECORD_STAT_CONFIG]         = perf_event__stat_config_swap,
 946         [PERF_RECORD_STAT]                = perf_event__stat_swap,
 947         [PERF_RECORD_STAT_ROUND]          = perf_event__stat_round_swap,
 948         [PERF_RECORD_EVENT_UPDATE]        = perf_event__event_update_swap,
 949         [PERF_RECORD_TIME_CONV]           = perf_event__all64_swap,
 950         [PERF_RECORD_HEADER_MAX]          = NULL,
 951 };
 952
 953 /*
 954  * When perf record finishes a pass on every buffers, it records this pseudo
 955  * event.
 956  * We record the max timestamp t found in the pass n.
 957  * Assuming these timestamps are monotonic across cpus, we know that if
 958  * a buffer still has events with timestamps below t, they will be all
 959  * available and then read in the pass n + 1.
 960  * Hence when we start to read the pass n + 2, we can safely flush every
 961  * events with timestamps below t.
 962  *
 963  *    ============ PASS n =================
 964  *       CPU 0         |   CPU 1
 965  *                     |
 966  *    cnt1 timestamps  |   cnt2 timestamps
 967  *          1          |         2
 968  *          2          |         3
 969  *          -          |         4  <--- max recorded
 970  *
 971  *    ============ PASS n + 1 ==============
 972  *       CPU 0         |   CPU 1
 973  *                     |
 974  *    cnt1 timestamps  |   cnt2 timestamps
 975  *          3          |         5
 976  *          4          |         6
 977  *          5          |         7 <---- max recorded
 978  *
 979  *      Flush every events below timestamp 4
 980  *
 981  *    ============ PASS n + 2 ==============
 982  *       CPU 0         |   CPU 1
 983  *                     |
 984  *    cnt1 timestamps  |   cnt2 timestamps
 985  *          6          |         8
 986  *          7          |         9
 987  *          -          |         10
 988  *
 989  *      Flush every events below timestamp 7
 990  *      etc...
 991  */
 992 static int process_finished_round(struct perf_tool *tool __maybe_unused,
 993                                   union perf_event *event __maybe_unused,
 994                                   struct ordered_events *oe)
 995 {
 996         if (dump_trace)
 997                 fprintf(stdout, "\n");
 998         return ordered_events__flush(oe, OE_FLUSH__ROUND);
 999 }
1000
1001 int perf_session__queue_event(struct perf_session *s, union perf_event *event,
1002                               u64 timestamp, u64 file_offset)
1003 {
1004         return ordered_events__queue(&s->ordered_events, event, timestamp, file_offset);
1005 }
1006
1007 static void callchain__lbr_callstack_printf(struct perf_sample *sample)
1008 {
1009         struct ip_callchain *callchain = sample->callchain;
1010         struct branch_stack *lbr_stack = sample->branch_stack;
1011         struct branch_entry *entries = perf_sample__branch_entries(sample);
1012         u64 kernel_callchain_nr = callchain->nr;
1013         unsigned int i;
1014
1015         for (i = 0; i < kernel_callchain_nr; i++) {
1016                 if (callchain->ips[i] == PERF_CONTEXT_USER)
1017                         break;
1018         }
1019
1020         if ((i != kernel_callchain_nr) && lbr_stack->nr) {
1021                 u64 total_nr;
1022                 /*
1023                  * LBR callstack can only get user call chain,
1024                  * i is kernel call chain number,
1025                  * 1 is PERF_CONTEXT_USER.
1026                  *
1027                  * The user call chain is stored in LBR registers.
1028                  * LBR are pair registers. The caller is stored
1029                  * in "from" register, while the callee is stored
1030                  * in "to" register.
1031                  * For example, there is a call stack
1032                  * "A"->"B"->"C"->"D".
1033                  * The LBR registers will recorde like
1034                  * "C"->"D", "B"->"C", "A"->"B".
1035                  * So only the first "to" register and all "from"
1036                  * registers are needed to construct the whole stack.
1037                  */
1038                 total_nr = i + 1 + lbr_stack->nr + 1;
1039                 kernel_callchain_nr = i + 1;
1040
1041                 printf("... LBR call chain: nr:%" PRIu64 "\n", total_nr);
1042
1043                 for (i = 0; i < kernel_callchain_nr; i++)
1044                         printf("..... %2d: %016" PRIx64 "\n",
1045                                i, callchain->ips[i]);
1046
1047                 printf("..... %2d: %016" PRIx64 "\n",
1048                        (int)(kernel_callchain_nr), entries[0].to);
1049                 for (i = 0; i < lbr_stack->nr; i++)
1050                         printf("..... %2d: %016" PRIx64 "\n",
1051                                (int)(i + kernel_callchain_nr + 1), entries[i].from);
1052         }
1053 }
1054
1055 static void callchain__printf(struct evsel *evsel,
1056                               struct perf_sample *sample)
1057 {
1058         unsigned int i;
1059         struct ip_callchain *callchain = sample->callchain;
1060
1061         if (evsel__has_branch_callstack(evsel))
1062                 callchain__lbr_callstack_printf(sample);
1063
1064         printf("... FP chain: nr:%" PRIu64 "\n", callchain->nr);
1065
1066         for (i = 0; i < callchain->nr; i++)
1067                 printf("..... %2d: %016" PRIx64 "\n",
1068                        i, callchain->ips[i]);
1069 }
1070
1071 static void branch_stack__printf(struct perf_sample *sample, bool callstack)
1072 {
1073         struct branch_entry *entries = perf_sample__branch_entries(sample);
1074         uint64_t i;
1075
1076         printf("%s: nr:%" PRIu64 "\n",
1077                 !callstack ? "... branch stack" : "... branch callstack",
1078                 sample->branch_stack->nr);
1079
1080         for (i = 0; i < sample->branch_stack->nr; i++) {
1081                 struct branch_entry *e = &entries[i];
1082
1083                 if (!callstack) {
1084                         printf("..... %2"PRIu64": %016" PRIx64 " -> %016" PRIx64 " %hu cycles %s%s%s%s %x\n",
1085                                 i, e->from, e->to,
1086                                 (unsigned short)e->flags.cycles,
1087                                 e->flags.mispred ? "M" : " ",
1088                                 e->flags.predicted ? "P" : " ",
1089                                 e->flags.abort ? "A" : " ",
1090                                 e->flags.in_tx ? "T" : " ",
1091                                 (unsigned)e->flags.reserved);
1092                 } else {
1093                         printf("..... %2"PRIu64": %016" PRIx64 "\n",
1094                                 i, i > 0 ? e->from : e->to);
1095                 }
1096         }
1097 }
1098
1099 static void regs_dump__printf(u64 mask, u64 *regs)
1100 {
1101         unsigned rid, i = 0;
1102
1103         for_each_set_bit(rid, (unsigned long *) &mask, sizeof(mask) * 8) {
1104                 u64 val = regs[i++];
1105
1106                 printf(".... %-5s 0x%016" PRIx64 "\n",
1107                        perf_reg_name(rid), val);
1108         }
1109 }
1110
1111 static const char *regs_abi[] = {
1112         [PERF_SAMPLE_REGS_ABI_NONE] = "none",
1113         [PERF_SAMPLE_REGS_ABI_32] = "32-bit",
1114         [PERF_SAMPLE_REGS_ABI_64] = "64-bit",
1115 };
1116
1117 static inline const char *regs_dump_abi(struct regs_dump *d)
1118 {
1119         if (d->abi > PERF_SAMPLE_REGS_ABI_64)
1120                 return "unknown";
1121
1122         return regs_abi[d->abi];
1123 }
1124
1125 static void regs__printf(const char *type, struct regs_dump *regs)
1126 {
1127         u64 mask = regs->mask;
1128
1129         printf("... %s regs: mask 0x%" PRIx64 " ABI %s\n",
1130                type,
1131                mask,
1132                regs_dump_abi(regs));
1133
1134         regs_dump__printf(mask, regs->regs);
1135 }
1136
1137 static void regs_user__printf(struct perf_sample *sample)
1138 {
1139         struct regs_dump *user_regs = &sample->user_regs;
1140
1141         if (user_regs->regs)
1142                 regs__printf("user", user_regs);
1143 }
1144
1145 static void regs_intr__printf(struct perf_sample *sample)
1146 {
1147         struct regs_dump *intr_regs = &sample->intr_regs;
1148
1149         if (intr_regs->regs)
1150                 regs__printf("intr", intr_regs);
1151 }
1152
1153 static void stack_user__printf(struct stack_dump *dump)
1154 {
1155         printf("... ustack: size %" PRIu64 ", offset 0x%x\n",
1156                dump->size, dump->offset);
1157 }
1158
1159 static void perf_evlist__print_tstamp(struct evlist *evlist,
1160                                        union perf_event *event,
1161                                        struct perf_sample *sample)
1162 {
1163         u64 sample_type = __perf_evlist__combined_sample_type(evlist);
1164
1165         if (event->header.type != PERF_RECORD_SAMPLE &&
1166             !perf_evlist__sample_id_all(evlist)) {
1167                 fputs("-1 -1 ", stdout);
1168                 return;
1169         }
1170
1171         if ((sample_type & PERF_SAMPLE_CPU))
1172                 printf("%u ", sample->cpu);
1173
1174         if (sample_type & PERF_SAMPLE_TIME)
1175                 printf("%" PRIu64 " ", sample->time);
1176 }
1177
1178 static void sample_read__printf(struct perf_sample *sample, u64 read_format)
1179 {
1180         printf("... sample_read:\n");
1181
1182         if (read_format & PERF_FORMAT_TOTAL_TIME_ENABLED)
1183                 printf("...... time enabled %016" PRIx64 "\n",
1184                        sample->read.time_enabled);
1185
1186         if (read_format & PERF_FORMAT_TOTAL_TIME_RUNNING)
1187                 printf("...... time running %016" PRIx64 "\n",
1188                        sample->read.time_running);
1189
1190         if (read_format & PERF_FORMAT_GROUP) {
1191                 u64 i;
1192
1193                 printf(".... group nr %" PRIu64 "\n", sample->read.group.nr);
1194
1195                 for (i = 0; i < sample->read.group.nr; i++) {
1196                         struct sample_read_value *value;
1197
1198                         value = &sample->read.group.values[i];
1199                         printf("..... id %016" PRIx64
1200                                ", value %016" PRIx64 "\n",
1201                                value->id, value->value);
1202                 }
1203         } else
1204                 printf("..... id %016" PRIx64 ", value %016" PRIx64 "\n",
1205                         sample->read.one.id, sample->read.one.value);
1206 }
1207
1208 static void dump_event(struct evlist *evlist, union perf_event *event,
1209                        u64 file_offset, struct perf_sample *sample)
1210 {
1211         if (!dump_trace)
1212                 return;
1213
1214         printf("\n%#" PRIx64 " [%#x]: event: %d\n",
1215                file_offset, event->header.size, event->header.type);
1216
1217         trace_event(event);
1218         if (event->header.type == PERF_RECORD_SAMPLE && evlist->trace_event_sample_raw)
1219                 evlist->trace_event_sample_raw(evlist, event, sample);
1220
1221         if (sample)
1222                 perf_evlist__print_tstamp(evlist, event, sample);
1223
1224         printf("%#" PRIx64 " [%#x]: PERF_RECORD_%s", file_offset,
1225                event->header.size, perf_event__name(event->header.type));
1226 }
1227
1228 static void dump_sample(struct evsel *evsel, union perf_event *event,
1229                         struct perf_sample *sample)
1230 {
1231         u64 sample_type;
1232
1233         if (!dump_trace)
1234                 return;
1235
1236         printf("(IP, 0x%x): %d/%d: %#" PRIx64 " period: %" PRIu64 " addr: %#" PRIx64 "\n",
1237                event->header.misc, sample->pid, sample->tid, sample->ip,
1238                sample->period, sample->addr);
1239
1240         sample_type = evsel->core.attr.sample_type;
1241
1242         if (evsel__has_callchain(evsel))
1243                 callchain__printf(evsel, sample);
1244
1245         if (evsel__has_br_stack(evsel))
1246                 branch_stack__printf(sample, evsel__has_branch_callstack(evsel));
1247
1248         if (sample_type & PERF_SAMPLE_REGS_USER)
1249                 regs_user__printf(sample);
1250
1251         if (sample_type & PERF_SAMPLE_REGS_INTR)
1252                 regs_intr__printf(sample);
1253
1254         if (sample_type & PERF_SAMPLE_STACK_USER)
1255                 stack_user__printf(&sample->user_stack);
1256
1257         if (sample_type & PERF_SAMPLE_WEIGHT)
1258                 printf("... weight: %" PRIu64 "\n", sample->weight);
1259
1260         if (sample_type & PERF_SAMPLE_DATA_SRC)
1261                 printf(" . data_src: 0x%"PRIx64"\n", sample->data_src);
1262
1263         if (sample_type & PERF_SAMPLE_PHYS_ADDR)
1264                 printf(" .. phys_addr: 0x%"PRIx64"\n", sample->phys_addr);
1265
1266         if (sample_type & PERF_SAMPLE_TRANSACTION)
1267                 printf("... transaction: %" PRIx64 "\n", sample->transaction);
1268
1269         if (sample_type & PERF_SAMPLE_READ)
1270                 sample_read__printf(sample, evsel->core.attr.read_format);
1271 }
1272
1273 static void dump_read(struct evsel *evsel, union perf_event *event)
1274 {
1275         struct perf_record_read *read_event = &event->read;
1276         u64 read_format;
1277
1278         if (!dump_trace)
1279                 return;
1280
1281         printf(": %d %d %s %" PRI_lu64 "\n", event->read.pid, event->read.tid,
1282                evsel__name(evsel), event->read.value);
1283
1284         if (!evsel)
1285                 return;
1286
1287         read_format = evsel->core.attr.read_format;
1288
1289         if (read_format & PERF_FORMAT_TOTAL_TIME_ENABLED)
1290                 printf("... time enabled : %" PRI_lu64 "\n", read_event->time_enabled);
1291
1292         if (read_format & PERF_FORMAT_TOTAL_TIME_RUNNING)
1293                 printf("... time running : %" PRI_lu64 "\n", read_event->time_running);
1294
1295         if (read_format & PERF_FORMAT_ID)
1296                 printf("... id           : %" PRI_lu64 "\n", read_event->id);
1297 }
1298
1299 static struct machine *machines__find_for_cpumode(struct machines *machines,
1300                                                union perf_event *event,
1301                                                struct perf_sample *sample)
1302 {
1303         struct machine *machine;
1304
1305         if (perf_guest &&
1306             ((sample->cpumode == PERF_RECORD_MISC_GUEST_KERNEL) ||
1307              (sample->cpumode == PERF_RECORD_MISC_GUEST_USER))) {
1308                 u32 pid;
1309
1310                 if (event->header.type == PERF_RECORD_MMAP
1311                     || event->header.type == PERF_RECORD_MMAP2)
1312                         pid = event->mmap.pid;
1313                 else
1314                         pid = sample->pid;
1315
1316                 machine = machines__find(machines, pid);
1317                 if (!machine)
1318                         machine = machines__findnew(machines, DEFAULT_GUEST_KERNEL_ID);
1319                 return machine;
1320         }
1321
1322         return &machines->host;
1323 }
1324
1325 static int deliver_sample_value(struct evlist *evlist,
1326                                 struct perf_tool *tool,
1327                                 union perf_event *event,
1328                                 struct perf_sample *sample,
1329                                 struct sample_read_value *v,
1330                                 struct machine *machine)
1331 {
1332         struct perf_sample_id *sid = perf_evlist__id2sid(evlist, v->id);
1333         struct evsel *evsel;
1334
1335         if (sid) {
1336                 sample->id     = v->id;
1337                 sample->period = v->value - sid->period;
1338                 sid->period    = v->value;
1339         }
1340
1341         if (!sid || sid->evsel == NULL) {
1342                 ++evlist->stats.nr_unknown_id;
1343                 return 0;
1344         }
1345
1346         /*
1347          * There's no reason to deliver sample
1348          * for zero period, bail out.
1349          */
1350         if (!sample->period)
1351                 return 0;
1352
1353         evsel = container_of(sid->evsel, struct evsel, core);
1354         return tool->sample(tool, event, sample, evsel, machine);
1355 }
1356
1357 static int deliver_sample_group(struct evlist *evlist,
1358                                 struct perf_tool *tool,
1359                                 union  perf_event *event,
1360                                 struct perf_sample *sample,
1361                                 struct machine *machine)
1362 {
1363         int ret = -EINVAL;
1364         u64 i;
1365
1366         for (i = 0; i < sample->read.group.nr; i++) {
1367                 ret = deliver_sample_value(evlist, tool, event, sample,
1368                                            &sample->read.group.values[i],
1369                                            machine);
1370                 if (ret)
1371                         break;
1372         }
1373
1374         return ret;
1375 }
1376
1377 static int
1378  perf_evlist__deliver_sample(struct evlist *evlist,
1379                              struct perf_tool *tool,
1380                              union  perf_event *event,
1381                              struct perf_sample *sample,
1382                              struct evsel *evsel,
1383                              struct machine *machine)
1384 {
1385         /* We know evsel != NULL. */
1386         u64 sample_type = evsel->core.attr.sample_type;
1387         u64 read_format = evsel->core.attr.read_format;
1388
1389         /* Standard sample delivery. */
1390         if (!(sample_type & PERF_SAMPLE_READ))
1391                 return tool->sample(tool, event, sample, evsel, machine);
1392
1393         /* For PERF_SAMPLE_READ we have either single or group mode. */
1394         if (read_format & PERF_FORMAT_GROUP)
1395                 return deliver_sample_group(evlist, tool, event, sample,
1396                                             machine);
1397         else
1398                 return deliver_sample_value(evlist, tool, event, sample,
1399                                             &sample->read.one, machine);
1400 }
1401
1402 static int machines__deliver_event(struct machines *machines,
1403                                    struct evlist *evlist,
1404                                    union perf_event *event,
1405                                    struct perf_sample *sample,
1406                                    struct perf_tool *tool, u64 file_offset)
1407 {
1408         struct evsel *evsel;
1409         struct machine *machine;
1410
1411         dump_event(evlist, event, file_offset, sample);
1412
1413         evsel = perf_evlist__id2evsel(evlist, sample->id);
1414
1415         machine = machines__find_for_cpumode(machines, event, sample);
1416
1417         switch (event->header.type) {
1418         case PERF_RECORD_SAMPLE:
1419                 if (evsel == NULL) {
1420                         ++evlist->stats.nr_unknown_id;
1421                         return 0;
1422                 }
1423                 dump_sample(evsel, event, sample);
1424                 if (machine == NULL) {
1425                         ++evlist->stats.nr_unprocessable_samples;
1426                         return 0;
1427                 }
1428                 return perf_evlist__deliver_sample(evlist, tool, event, sample, evsel, machine);
1429         case PERF_RECORD_MMAP:
1430                 return tool->mmap(tool, event, sample, machine);
1431         case PERF_RECORD_MMAP2:
1432                 if (event->header.misc & PERF_RECORD_MISC_PROC_MAP_PARSE_TIMEOUT)
1433                         ++evlist->stats.nr_proc_map_timeout;
1434                 return tool->mmap2(tool, event, sample, machine);
1435         case PERF_RECORD_COMM:
1436                 return tool->comm(tool, event, sample, machine);
1437         case PERF_RECORD_NAMESPACES:
1438                 return tool->namespaces(tool, event, sample, machine);
1439         case PERF_RECORD_CGROUP:
1440                 return tool->cgroup(tool, event, sample, machine);
1441         case PERF_RECORD_FORK:
1442                 return tool->fork(tool, event, sample, machine);
1443         case PERF_RECORD_EXIT:
1444                 return tool->exit(tool, event, sample, machine);
1445         case PERF_RECORD_LOST:
1446                 if (tool->lost == perf_event__process_lost)
1447                         evlist->stats.total_lost += event->lost.lost;
1448                 return tool->lost(tool, event, sample, machine);
1449         case PERF_RECORD_LOST_SAMPLES:
1450                 if (tool->lost_samples == perf_event__process_lost_samples)
1451                         evlist->stats.total_lost_samples += event->lost_samples.lost;
1452                 return tool->lost_samples(tool, event, sample, machine);
1453         case PERF_RECORD_READ:
1454                 dump_read(evsel, event);
1455                 return tool->read(tool, event, sample, evsel, machine);
1456         case PERF_RECORD_THROTTLE:
1457                 return tool->throttle(tool, event, sample, machine);
1458         case PERF_RECORD_UNTHROTTLE:
1459                 return tool->unthrottle(tool, event, sample, machine);
1460         case PERF_RECORD_AUX:
1461                 if (tool->aux == perf_event__process_aux) {
1462                         if (event->aux.flags & PERF_AUX_FLAG_TRUNCATED)
1463                                 evlist->stats.total_aux_lost += 1;
1464                         if (event->aux.flags & PERF_AUX_FLAG_PARTIAL)
1465                                 evlist->stats.total_aux_partial += 1;
1466                 }
1467                 return tool->aux(tool, event, sample, machine);
1468         case PERF_RECORD_ITRACE_START:
1469                 return tool->itrace_start(tool, event, sample, machine);
1470         case PERF_RECORD_SWITCH:
1471         case PERF_RECORD_SWITCH_CPU_WIDE:
1472                 return tool->context_switch(tool, event, sample, machine);
1473         case PERF_RECORD_KSYMBOL:
1474                 return tool->ksymbol(tool, event, sample, machine);
1475         case PERF_RECORD_BPF_EVENT:
1476                 return tool->bpf(tool, event, sample, machine);
1477         default:
1478                 ++evlist->stats.nr_unknown_events;
1479                 return -1;
1480         }
1481 }
1482
1483 static int perf_session__deliver_event(struct perf_session *session,
1484                                        union perf_event *event,
1485                                        struct perf_tool *tool,
1486                                        u64 file_offset)
1487 {
1488         struct perf_sample sample;
1489         int ret;
1490
1491         ret = perf_evlist__parse_sample(session->evlist, event, &sample);
1492         if (ret) {
1493                 pr_err("Can't parse sample, err = %d\n", ret);
1494                 return ret;
1495         }
1496
1497         ret = auxtrace__process_event(session, event, &sample, tool);
1498         if (ret < 0)
1499                 return ret;
1500         if (ret > 0)
1501                 return 0;
1502
1503         ret = machines__deliver_event(&session->machines, session->evlist,
1504                                       event, &sample, tool, file_offset);
1505
1506         if (dump_trace && sample.aux_sample.size)
1507                 auxtrace__dump_auxtrace_sample(session, &sample);
1508
1509         return ret;
1510 }
1511
1512 static s64 perf_session__process_user_event(struct perf_session *session,
1513                                             union perf_event *event,
1514                                             u64 file_offset)
1515 {
1516         struct ordered_events *oe = &session->ordered_events;
1517         struct perf_tool *tool = session->tool;
1518         struct perf_sample sample = { .time = 0, };
1519         int fd = perf_data__fd(session->data);
1520         int err;
1521
1522         if (event->header.type != PERF_RECORD_COMPRESSED ||
1523             tool->compressed == perf_session__process_compressed_event_stub)
1524                 dump_event(session->evlist, event, file_offset, &sample);
1525
1526         /* These events are processed right away */
1527         switch (event->header.type) {
1528         case PERF_RECORD_HEADER_ATTR:
1529                 err = tool->attr(tool, event, &session->evlist);
1530                 if (err == 0) {
1531                         perf_session__set_id_hdr_size(session);
1532                         perf_session__set_comm_exec(session);
1533                 }
1534                 return err;
1535         case PERF_RECORD_EVENT_UPDATE:
1536                 return tool->event_update(tool, event, &session->evlist);
1537         case PERF_RECORD_HEADER_EVENT_TYPE:
1538                 /*
1539                  * Depreceated, but we need to handle it for sake
1540                  * of old data files create in pipe mode.
1541                  */
1542                 return 0;
1543         case PERF_RECORD_HEADER_TRACING_DATA:
1544                 /*
1545                  * Setup for reading amidst mmap, but only when we
1546                  * are in 'file' mode. The 'pipe' fd is in proper
1547                  * place already.
1548                  */
1549                 if (!perf_data__is_pipe(session->data))
1550                         lseek(fd, file_offset, SEEK_SET);
1551                 return tool->tracing_data(session, event);
1552         case PERF_RECORD_HEADER_BUILD_ID:
1553                 return tool->build_id(session, event);
1554         case PERF_RECORD_FINISHED_ROUND:
1555                 return tool->finished_round(tool, event, oe);
1556         case PERF_RECORD_ID_INDEX:
1557                 return tool->id_index(session, event);
1558         case PERF_RECORD_AUXTRACE_INFO:
1559                 return tool->auxtrace_info(session, event);
1560         case PERF_RECORD_AUXTRACE:
1561                 /* setup for reading amidst mmap */
1562                 lseek(fd, file_offset + event->header.size, SEEK_SET);
1563                 return tool->auxtrace(session, event);
1564         case PERF_RECORD_AUXTRACE_ERROR:
1565                 perf_session__auxtrace_error_inc(session, event);
1566                 return tool->auxtrace_error(session, event);
1567         case PERF_RECORD_THREAD_MAP:
1568                 return tool->thread_map(session, event);
1569         case PERF_RECORD_CPU_MAP:
1570                 return tool->cpu_map(session, event);
1571         case PERF_RECORD_STAT_CONFIG:
1572                 return tool->stat_config(session, event);
1573         case PERF_RECORD_STAT:
1574                 return tool->stat(session, event);
1575         case PERF_RECORD_STAT_ROUND:
1576                 return tool->stat_round(session, event);
1577         case PERF_RECORD_TIME_CONV:
1578                 session->time_conv = event->time_conv;
1579                 return tool->time_conv(session, event);
1580         case PERF_RECORD_HEADER_FEATURE:
1581                 return tool->feature(session, event);
1582         case PERF_RECORD_COMPRESSED:
1583                 err = tool->compressed(session, event, file_offset);
1584                 if (err)
1585                         dump_event(session->evlist, event, file_offset, &sample);
1586                 return err;
1587         default:
1588                 return -EINVAL;
1589         }
1590 }
1591
1592 int perf_session__deliver_synth_event(struct perf_session *session,
1593                                       union perf_event *event,
1594                                       struct perf_sample *sample)
1595 {
1596         struct evlist *evlist = session->evlist;
1597         struct perf_tool *tool = session->tool;
1598
1599         events_stats__inc(&evlist->stats, event->header.type);
1600
1601         if (event->header.type >= PERF_RECORD_USER_TYPE_START)
1602                 return perf_session__process_user_event(session, event, 0);
1603
1604         return machines__deliver_event(&session->machines, evlist, event, sample, tool, 0);
1605 }
1606
1607 static void event_swap(union perf_event *event, bool sample_id_all)
1608 {
1609         perf_event__swap_op swap;
1610
1611         swap = perf_event__swap_ops[event->header.type];
1612         if (swap)
1613                 swap(event, sample_id_all);
1614 }
1615
1616 int perf_session__peek_event(struct perf_session *session, off_t file_offset,
1617                              void *buf, size_t buf_sz,
1618                              union perf_event **event_ptr,
1619                              struct perf_sample *sample)
1620 {
1621         union perf_event *event;
1622         size_t hdr_sz, rest;
1623         int fd;
1624
1625         if (session->one_mmap && !session->header.needs_swap) {
1626                 event = file_offset - session->one_mmap_offset +
1627                         session->one_mmap_addr;
1628                 goto out_parse_sample;
1629         }
1630
1631         if (perf_data__is_pipe(session->data))
1632                 return -1;
1633
1634         fd = perf_data__fd(session->data);
1635         hdr_sz = sizeof(struct perf_event_header);
1636
1637         if (buf_sz < hdr_sz)
1638                 return -1;
1639
1640         if (lseek(fd, file_offset, SEEK_SET) == (off_t)-1 ||
1641             readn(fd, buf, hdr_sz) != (ssize_t)hdr_sz)
1642                 return -1;
1643
1644         event = (union perf_event *)buf;
1645
1646         if (session->header.needs_swap)
1647                 perf_event_header__bswap(&event->header);
1648
1649         if (event->header.size < hdr_sz || event->header.size > buf_sz)
1650                 return -1;
1651
1652         rest = event->header.size - hdr_sz;
1653
1654         if (readn(fd, buf, rest) != (ssize_t)rest)
1655                 return -1;
1656
1657         if (session->header.needs_swap)
1658                 event_swap(event, perf_evlist__sample_id_all(session->evlist));
1659
1660 out_parse_sample:
1661
1662         if (sample && event->header.type < PERF_RECORD_USER_TYPE_START &&
1663             perf_evlist__parse_sample(session->evlist, event, sample))
1664                 return -1;
1665
1666         *event_ptr = event;
1667
1668         return 0;
1669 }
1670
1671 int perf_session__peek_events(struct perf_session *session, u64 offset,
1672                               u64 size, peek_events_cb_t cb, void *data)
1673 {
1674         u64 max_offset = offset + size;
1675         char buf[PERF_SAMPLE_MAX_SIZE];
1676         union perf_event *event;
1677         int err;
1678
1679         do {
1680                 err = perf_session__peek_event(session, offset, buf,
1681                                                PERF_SAMPLE_MAX_SIZE, &event,
1682                                                NULL);
1683                 if (err)
1684                         return err;
1685
1686                 err = cb(session, event, offset, data);
1687                 if (err)
1688                         return err;
1689
1690                 offset += event->header.size;
1691                 if (event->header.type == PERF_RECORD_AUXTRACE)
1692                         offset += event->auxtrace.size;
1693
1694         } while (offset < max_offset);
1695
1696         return err;
1697 }
1698
1699 static s64 perf_session__process_event(struct perf_session *session,
1700                                        union perf_event *event, u64 file_offset)
1701 {
1702         struct evlist *evlist = session->evlist;
1703         struct perf_tool *tool = session->tool;
1704         int ret;
1705
1706         if (session->header.needs_swap)
1707                 event_swap(event, perf_evlist__sample_id_all(evlist));
1708
1709         if (event->header.type >= PERF_RECORD_HEADER_MAX)
1710                 return -EINVAL;
1711
1712         events_stats__inc(&evlist->stats, event->header.type);
1713
1714         if (event->header.type >= PERF_RECORD_USER_TYPE_START)
1715                 return perf_session__process_user_event(session, event, file_offset);
1716
1717         if (tool->ordered_events) {
1718                 u64 timestamp = -1ULL;
1719
1720                 ret = perf_evlist__parse_sample_timestamp(evlist, event, &timestamp);
1721                 if (ret && ret != -1)
1722                         return ret;
1723
1724                 ret = perf_session__queue_event(session, event, timestamp, file_offset);
1725                 if (ret != -ETIME)
1726                         return ret;
1727         }
1728
1729         return perf_session__deliver_event(session, event, tool, file_offset);
1730 }
1731
1732 void perf_event_header__bswap(struct perf_event_header *hdr)
1733 {
1734         hdr->type = bswap_32(hdr->type);
1735         hdr->misc = bswap_16(hdr->misc);
1736         hdr->size = bswap_16(hdr->size);
1737 }
1738
1739 struct thread *perf_session__findnew(struct perf_session *session, pid_t pid)
1740 {
1741         return machine__findnew_thread(&session->machines.host, -1, pid);
1742 }
1743
1744 /*
1745  * Threads are identified by pid and tid, and the idle task has pid == tid == 0.
1746  * So here a single thread is created for that, but actually there is a separate
1747  * idle task per cpu, so there should be one 'struct thread' per cpu, but there
1748  * is only 1. That causes problems for some tools, requiring workarounds. For
1749  * example get_idle_thread() in builtin-sched.c, or thread_stack__per_cpu().
1750  */
1751 int perf_session__register_idle_thread(struct perf_session *session)
1752 {
1753         struct thread *thread;
1754         int err = 0;
1755
1756         thread = machine__findnew_thread(&session->machines.host, 0, 0);
1757         if (thread == NULL || thread__set_comm(thread, "swapper", 0)) {
1758                 pr_err("problem inserting idle task.\n");
1759                 err = -1;
1760         }
1761
1762         if (thread == NULL || thread__set_namespaces(thread, 0, NULL)) {
1763                 pr_err("problem inserting idle task.\n");
1764                 err = -1;
1765         }
1766
1767         /* machine__findnew_thread() got the thread, so put it */
1768         thread__put(thread);
1769         return err;
1770 }
1771
1772 static void
1773 perf_session__warn_order(const struct perf_session *session)
1774 {
1775         const struct ordered_events *oe = &session->ordered_events;
1776         struct evsel *evsel;
1777         bool should_warn = true;
1778
1779         evlist__for_each_entry(session->evlist, evsel) {
1780                 if (evsel->core.attr.write_backward)
1781                         should_warn = false;
1782         }
1783
1784         if (!should_warn)
1785                 return;
1786         if (oe->nr_unordered_events != 0)
1787                 ui__warning("%u out of order events recorded.\n", oe->nr_unordered_events);
1788 }
1789
1790 static void perf_session__warn_about_errors(const struct perf_session *session)
1791 {
1792         const struct events_stats *stats = &session->evlist->stats;
1793
1794         if (session->tool->lost == perf_event__process_lost &&
1795             stats->nr_events[PERF_RECORD_LOST] != 0) {
1796                 ui__warning("Processed %d events and lost %d chunks!\n\n"
1797                             "Check IO/CPU overload!\n\n",
1798                             stats->nr_events[0],
1799                             stats->nr_events[PERF_RECORD_LOST]);
1800         }
1801
1802         if (session->tool->lost_samples == perf_event__process_lost_samples) {
1803                 double drop_rate;
1804
1805                 drop_rate = (double)stats->total_lost_samples /
1806                             (double) (stats->nr_events[PERF_RECORD_SAMPLE] + stats->total_lost_samples);
1807                 if (drop_rate > 0.05) {
1808                         ui__warning("Processed %" PRIu64 " samples and lost %3.2f%%!\n\n",
1809                                     stats->nr_events[PERF_RECORD_SAMPLE] + stats->total_lost_samples,
1810                                     drop_rate * 100.0);
1811                 }
1812         }
1813
1814         if (session->tool->aux == perf_event__process_aux &&
1815             stats->total_aux_lost != 0) {
1816                 ui__warning("AUX data lost %" PRIu64 " times out of %u!\n\n",
1817                             stats->total_aux_lost,
1818                             stats->nr_events[PERF_RECORD_AUX]);
1819         }
1820
1821         if (session->tool->aux == perf_event__process_aux &&
1822             stats->total_aux_partial != 0) {
1823                 bool vmm_exclusive = false;
1824
1825                 (void)sysfs__read_bool("module/kvm_intel/parameters/vmm_exclusive",
1826                                        &vmm_exclusive);
1827
1828                 ui__warning("AUX data had gaps in it %" PRIu64 " times out of %u!\n\n"
1829                             "Are you running a KVM guest in the background?%s\n\n",
1830                             stats->total_aux_partial,
1831                             stats->nr_events[PERF_RECORD_AUX],
1832                             vmm_exclusive ?
1833                             "\nReloading kvm_intel module with vmm_exclusive=0\n"
1834                             "will reduce the gaps to only guest's timeslices." :
1835                             "");
1836         }
1837
1838         if (stats->nr_unknown_events != 0) {
1839                 ui__warning("Found %u unknown events!\n\n"
1840                             "Is this an older tool processing a perf.data "
1841                             "file generated by a more recent tool?\n\n"
1842                             "If that is not the case, consider "
1843                             "reporting to linux-kernel@vger.kernel.org.\n\n",
1844                             stats->nr_unknown_events);
1845         }
1846
1847         if (stats->nr_unknown_id != 0) {
1848                 ui__warning("%u samples with id not present in the header\n",
1849                             stats->nr_unknown_id);
1850         }
1851
1852         if (stats->nr_invalid_chains != 0) {
1853                 ui__warning("Found invalid callchains!\n\n"
1854                             "%u out of %u events were discarded for this reason.\n\n"
1855                             "Consider reporting to linux-kernel@vger.kernel.org.\n\n",
1856                             stats->nr_invalid_chains,
1857                             stats->nr_events[PERF_RECORD_SAMPLE]);
1858         }
1859
1860         if (stats->nr_unprocessable_samples != 0) {
1861                 ui__warning("%u unprocessable samples recorded.\n"
1862                             "Do you have a KVM guest running and not using 'perf kvm'?\n",
1863                             stats->nr_unprocessable_samples);
1864         }
1865
1866         perf_session__warn_order(session);
1867
1868         events_stats__auxtrace_error_warn(stats);
1869
1870         if (stats->nr_proc_map_timeout != 0) {
1871                 ui__warning("%d map information files for pre-existing threads were\n"
1872                             "not processed, if there are samples for addresses they\n"
1873                             "will not be resolved, you may find out which are these\n"
1874                             "threads by running with -v and redirecting the output\n"
1875                             "to a file.\n"
1876                             "The time limit to process proc map is too short?\n"
1877                             "Increase it by --proc-map-timeout\n",
1878                             stats->nr_proc_map_timeout);
1879         }
1880 }
1881
1882 static int perf_session__flush_thread_stack(struct thread *thread,
1883                                             void *p __maybe_unused)
1884 {
1885         return thread_stack__flush(thread);
1886 }
1887
1888 static int perf_session__flush_thread_stacks(struct perf_session *session)
1889 {
1890         return machines__for_each_thread(&session->machines,
1891                                          perf_session__flush_thread_stack,
1892                                          NULL);
1893 }
1894
1895 volatile int session_done;
1896
1897 static int __perf_session__process_decomp_events(struct perf_session *session);
1898
1899 static int __perf_session__process_pipe_events(struct perf_session *session)
1900 {
1901         struct ordered_events *oe = &session->ordered_events;
1902         struct perf_tool *tool = session->tool;
1903         int fd = perf_data__fd(session->data);
1904         union perf_event *event;
1905         uint32_t size, cur_size = 0;
1906         void *buf = NULL;
1907         s64 skip = 0;
1908         u64 head;
1909         ssize_t err;
1910         void *p;
1911
1912         perf_tool__fill_defaults(tool);
1913
1914         head = 0;
1915         cur_size = sizeof(union perf_event);
1916
1917         buf = malloc(cur_size);
1918         if (!buf)
1919                 return -errno;
1920         ordered_events__set_copy_on_queue(oe, true);
1921 more:
1922         event = buf;
1923         err = readn(fd, event, sizeof(struct perf_event_header));
1924         if (err <= 0) {
1925                 if (err == 0)
1926                         goto done;
1927
1928                 pr_err("failed to read event header\n");
1929                 goto out_err;
1930         }
1931
1932         if (session->header.needs_swap)
1933                 perf_event_header__bswap(&event->header);
1934
1935         size = event->header.size;
1936         if (size < sizeof(struct perf_event_header)) {
1937                 pr_err("bad event header size\n");
1938                 goto out_err;
1939         }
1940
1941         if (size > cur_size) {
1942                 void *new = realloc(buf, size);
1943                 if (!new) {
1944                         pr_err("failed to allocate memory to read event\n");
1945                         goto out_err;
1946                 }
1947                 buf = new;
1948                 cur_size = size;
1949                 event = buf;
1950         }
1951         p = event;
1952         p += sizeof(struct perf_event_header);
1953
1954         if (size - sizeof(struct perf_event_header)) {
1955                 err = readn(fd, p, size - sizeof(struct perf_event_header));
1956                 if (err <= 0) {
1957                         if (err == 0) {
1958                                 pr_err("unexpected end of event stream\n");
1959                                 goto done;
1960                         }
1961
1962                         pr_err("failed to read event data\n");
1963                         goto out_err;
1964                 }
1965         }
1966
1967         if ((skip = perf_session__process_event(session, event, head)) < 0) {
1968                 pr_err("%#" PRIx64 " [%#x]: failed to process type: %d\n",
1969                        head, event->header.size, event->header.type);
1970                 err = -EINVAL;
1971                 goto out_err;
1972         }
1973
1974         head += size;
1975
1976         if (skip > 0)
1977                 head += skip;
1978
1979         err = __perf_session__process_decomp_events(session);
1980         if (err)
1981                 goto out_err;
1982
1983         if (!session_done())
1984                 goto more;
1985 done:
1986         /* do the final flush for ordered samples */
1987         err = ordered_events__flush(oe, OE_FLUSH__FINAL);
1988         if (err)
1989                 goto out_err;
1990         err = auxtrace__flush_events(session, tool);
1991         if (err)
1992                 goto out_err;
1993         err = perf_session__flush_thread_stacks(session);
1994 out_err:
1995         free(buf);
1996         if (!tool->no_warn)
1997                 perf_session__warn_about_errors(session);
1998         ordered_events__free(&session->ordered_events);
1999         auxtrace__free_events(session);
2000         return err;
2001 }
2002
2003 static union perf_event *
2004 prefetch_event(char *buf, u64 head, size_t mmap_size,
2005                bool needs_swap, union perf_event *error)
2006 {
2007         union perf_event *event;
2008
2009         /*
2010          * Ensure we have enough space remaining to read
2011          * the size of the event in the headers.
2012          */
2013         if (head + sizeof(event->header) > mmap_size)
2014                 return NULL;
2015
2016         event = (union perf_event *)(buf + head);
2017         if (needs_swap)
2018                 perf_event_header__bswap(&event->header);
2019
2020         if (head + event->header.size <= mmap_size)
2021                 return event;
2022
2023         /* We're not fetching the event so swap back again */
2024         if (needs_swap)
2025                 perf_event_header__bswap(&event->header);
2026
2027         pr_debug("%s: head=%#" PRIx64 " event->header_size=%#x, mmap_size=%#zx:"
2028                  " fuzzed or compressed perf.data?\n",__func__, head, event->header.size, mmap_size);
2029
2030         return error;
2031 }
2032
2033 static union perf_event *
2034 fetch_mmaped_event(u64 head, size_t mmap_size, char *buf, bool needs_swap)
2035 {
2036         return prefetch_event(buf, head, mmap_size, needs_swap, ERR_PTR(-EINVAL));
2037 }
2038
2039 static union perf_event *
2040 fetch_decomp_event(u64 head, size_t mmap_size, char *buf, bool needs_swap)
2041 {
2042         return prefetch_event(buf, head, mmap_size, needs_swap, NULL);
2043 }
2044
2045 static int __perf_session__process_decomp_events(struct perf_session *session)
2046 {
2047         s64 skip;
2048         u64 size, file_pos = 0;
2049         struct decomp *decomp = session->decomp_last;
2050
2051         if (!decomp)
2052                 return 0;
2053
2054         while (decomp->head < decomp->size && !session_done()) {
2055                 union perf_event *event = fetch_decomp_event(decomp->head, decomp->size, decomp->data,
2056                                                              session->header.needs_swap);
2057
2058                 if (!event)
2059                         break;
2060
2061                 size = event->header.size;
2062
2063                 if (size < sizeof(struct perf_event_header) ||
2064                     (skip = perf_session__process_event(session, event, file_pos)) < 0) {
2065                         pr_err("%#" PRIx64 " [%#x]: failed to process type: %d\n",
2066                                 decomp->file_pos + decomp->head, event->header.size, event->header.type);
2067                         return -EINVAL;
2068                 }
2069
2070                 if (skip)
2071                         size += skip;
2072
2073                 decomp->head += size;
2074         }
2075
2076         return 0;
2077 }
2078
2079 /*
2080  * On 64bit we can mmap the data file in one go. No need for tiny mmap
2081  * slices. On 32bit we use 32MB.
2082  */
2083 #if BITS_PER_LONG == 64
2084 #define MMAP_SIZE ULLONG_MAX
2085 #define NUM_MMAPS 1
2086 #else
2087 #define MMAP_SIZE (32 * 1024 * 1024ULL)
2088 #define NUM_MMAPS 128
2089 #endif
2090
2091 struct reader;
2092
2093 typedef s64 (*reader_cb_t)(struct perf_session *session,
2094                            union perf_event *event,
2095                            u64 file_offset);
2096
2097 struct reader {
2098         int              fd;
2099         u64              data_size;
2100         u64              data_offset;
2101         reader_cb_t      process;
2102 };
2103
2104 static int
2105 reader__process_events(struct reader *rd, struct perf_session *session,
2106                        struct ui_progress *prog)
2107 {
2108         u64 data_size = rd->data_size;
2109         u64 head, page_offset, file_offset, file_pos, size;
2110         int err = 0, mmap_prot, mmap_flags, map_idx = 0;
2111         size_t  mmap_size;
2112         char *buf, *mmaps[NUM_MMAPS];
2113         union perf_event *event;
2114         s64 skip;
2115
2116         page_offset = page_size * (rd->data_offset / page_size);
2117         file_offset = page_offset;
2118         head = rd->data_offset - page_offset;
2119
2120         ui_progress__init_size(prog, data_size, "Processing events...");
2121
2122         data_size += rd->data_offset;
2123
2124         mmap_size = MMAP_SIZE;
2125         if (mmap_size > data_size) {
2126                 mmap_size = data_size;
2127                 session->one_mmap = true;
2128         }
2129
2130         memset(mmaps, 0, sizeof(mmaps));
2131
2132         mmap_prot  = PROT_READ;
2133         mmap_flags = MAP_SHARED;
2134
2135         if (session->header.needs_swap) {
2136                 mmap_prot  |= PROT_WRITE;
2137                 mmap_flags = MAP_PRIVATE;
2138         }
2139 remap:
2140         buf = mmap(NULL, mmap_size, mmap_prot, mmap_flags, rd->fd,
2141                    file_offset);
2142         if (buf == MAP_FAILED) {
2143                 pr_err("failed to mmap file\n");
2144                 err = -errno;
2145                 goto out;
2146         }
2147         mmaps[map_idx] = buf;
2148         map_idx = (map_idx + 1) & (ARRAY_SIZE(mmaps) - 1);
2149         file_pos = file_offset + head;
2150         if (session->one_mmap) {
2151                 session->one_mmap_addr = buf;
2152                 session->one_mmap_offset = file_offset;
2153         }
2154
2155 more:
2156         event = fetch_mmaped_event(head, mmap_size, buf, session->header.needs_swap);
2157         if (IS_ERR(event))
2158                 return PTR_ERR(event);
2159
2160         if (!event) {
2161                 if (mmaps[map_idx]) {
2162                         munmap(mmaps[map_idx], mmap_size);
2163                         mmaps[map_idx] = NULL;
2164                 }
2165
2166                 page_offset = page_size * (head / page_size);
2167                 file_offset += page_offset;
2168                 head -= page_offset;
2169                 goto remap;
2170         }
2171
2172         size = event->header.size;
2173
2174         skip = -EINVAL;
2175
2176         if (size < sizeof(struct perf_event_header) ||
2177             (skip = rd->process(session, event, file_pos)) < 0) {
2178                 pr_err("%#" PRIx64 " [%#x]: failed to process type: %d [%s]\n",
2179                        file_offset + head, event->header.size,
2180                        event->header.type, strerror(-skip));
2181                 err = skip;
2182                 goto out;
2183         }
2184
2185         if (skip)
2186                 size += skip;
2187
2188         head += size;
2189         file_pos += size;
2190
2191         err = __perf_session__process_decomp_events(session);
2192         if (err)
2193                 goto out;
2194
2195         ui_progress__update(prog, size);
2196
2197         if (session_done())
2198                 goto out;
2199
2200         if (file_pos < data_size)
2201                 goto more;
2202
2203 out:
2204         return err;
2205 }
2206
2207 static s64 process_simple(struct perf_session *session,
2208                           union perf_event *event,
2209                           u64 file_offset)
2210 {
2211         return perf_session__process_event(session, event, file_offset);
2212 }
2213
2214 static int __perf_session__process_events(struct perf_session *session)
2215 {
2216         struct reader rd = {
2217                 .fd             = perf_data__fd(session->data),
2218                 .data_size      = session->header.data_size,
2219                 .data_offset    = session->header.data_offset,
2220                 .process        = process_simple,
2221         };
2222         struct ordered_events *oe = &session->ordered_events;
2223         struct perf_tool *tool = session->tool;
2224         struct ui_progress prog;
2225         int err;
2226
2227         perf_tool__fill_defaults(tool);
2228
2229         if (rd.data_size == 0)
2230                 return -1;
2231
2232         ui_progress__init_size(&prog, rd.data_size, "Processing events...");
2233
2234         err = reader__process_events(&rd, session, &prog);
2235         if (err)
2236                 goto out_err;
2237         /* do the final flush for ordered samples */
2238         err = ordered_events__flush(oe, OE_FLUSH__FINAL);
2239         if (err)
2240                 goto out_err;
2241         err = auxtrace__flush_events(session, tool);
2242         if (err)
2243                 goto out_err;
2244         err = perf_session__flush_thread_stacks(session);
2245 out_err:
2246         ui_progress__finish();
2247         if (!tool->no_warn)
2248                 perf_session__warn_about_errors(session);
2249         /*
2250          * We may switching perf.data output, make ordered_events
2251          * reusable.
2252          */
2253         ordered_events__reinit(&session->ordered_events);
2254         auxtrace__free_events(session);
2255         session->one_mmap = false;
2256         return err;
2257 }
2258
2259 int perf_session__process_events(struct perf_session *session)
2260 {
2261         if (perf_session__register_idle_thread(session) < 0)
2262                 return -ENOMEM;
2263
2264         if (perf_data__is_pipe(session->data))
2265                 return __perf_session__process_pipe_events(session);
2266
2267         return __perf_session__process_events(session);
2268 }
2269
2270 bool perf_session__has_traces(struct perf_session *session, const char *msg)
2271 {
2272         struct evsel *evsel;
2273
2274         evlist__for_each_entry(session->evlist, evsel) {
2275                 if (evsel->core.attr.type == PERF_TYPE_TRACEPOINT)
2276                         return true;
2277         }
2278
2279         pr_err("No trace sample to read. Did you call 'perf %s'?\n", msg);
2280         return false;
2281 }
2282
2283 int map__set_kallsyms_ref_reloc_sym(struct map *map, const char *symbol_name, u64 addr)
2284 {
2285         char *bracket;
2286         struct ref_reloc_sym *ref;
2287         struct kmap *kmap;
2288
2289         ref = zalloc(sizeof(struct ref_reloc_sym));
2290         if (ref == NULL)
2291                 return -ENOMEM;
2292
2293         ref->name = strdup(symbol_name);
2294         if (ref->name == NULL) {
2295                 free(ref);
2296                 return -ENOMEM;
2297         }
2298
2299         bracket = strchr(ref->name, ']');
2300         if (bracket)
2301                 *bracket = '\0';
2302
2303         ref->addr = addr;
2304
2305         kmap = map__kmap(map);
2306         if (kmap)
2307                 kmap->ref_reloc_sym = ref;
2308
2309         return 0;
2310 }
2311
2312 size_t perf_session__fprintf_dsos(struct perf_session *session, FILE *fp)
2313 {
2314         return machines__fprintf_dsos(&session->machines, fp);
2315 }
2316
2317 size_t perf_session__fprintf_dsos_buildid(struct perf_session *session, FILE *fp,
2318                                           bool (skip)(struct dso *dso, int parm), int parm)
2319 {
2320         return machines__fprintf_dsos_buildid(&session->machines, fp, skip, parm);
2321 }
2322
2323 size_t perf_session__fprintf_nr_events(struct perf_session *session, FILE *fp)
2324 {
2325         size_t ret;
2326         const char *msg = "";
2327
2328         if (perf_header__has_feat(&session->header, HEADER_AUXTRACE))
2329                 msg = " (excludes AUX area (e.g. instruction trace) decoded / synthesized events)";
2330
2331         ret = fprintf(fp, "\nAggregated stats:%s\n", msg);
2332
2333         ret += events_stats__fprintf(&session->evlist->stats, fp);
2334         return ret;
2335 }
2336
2337 size_t perf_session__fprintf(struct perf_session *session, FILE *fp)
2338 {
2339         /*
2340          * FIXME: Here we have to actually print all the machines in this
2341          * session, not just the host...
2342          */
2343         return machine__fprintf(&session->machines.host, fp);
2344 }
2345
2346 struct evsel *perf_session__find_first_evtype(struct perf_session *session,
2347                                               unsigned int type)
2348 {
2349         struct evsel *pos;
2350
2351         evlist__for_each_entry(session->evlist, pos) {
2352                 if (pos->core.attr.type == type)
2353                         return pos;
2354         }
2355         return NULL;
2356 }
2357
2358 int perf_session__cpu_bitmap(struct perf_session *session,
2359                              const char *cpu_list, unsigned long *cpu_bitmap)
2360 {
2361         int i, err = -1;
2362         struct perf_cpu_map *map;
2363         int nr_cpus = min(session->header.env.nr_cpus_online, MAX_NR_CPUS);
2364
2365         for (i = 0; i < PERF_TYPE_MAX; ++i) {
2366                 struct evsel *evsel;
2367
2368                 evsel = perf_session__find_first_evtype(session, i);
2369                 if (!evsel)
2370                         continue;
2371
2372                 if (!(evsel->core.attr.sample_type & PERF_SAMPLE_CPU)) {
2373                         pr_err("File does not contain CPU events. "
2374                                "Remove -C option to proceed.\n");
2375                         return -1;
2376                 }
2377         }
2378
2379         map = perf_cpu_map__new(cpu_list);
2380         if (map == NULL) {
2381                 pr_err("Invalid cpu_list\n");
2382                 return -1;
2383         }
2384
2385         for (i = 0; i < map->nr; i++) {
2386                 int cpu = map->map[i];
2387
2388                 if (cpu >= nr_cpus) {
2389                         pr_err("Requested CPU %d too large. "
2390                                "Consider raising MAX_NR_CPUS\n", cpu);
2391                         goto out_delete_map;
2392                 }
2393
2394                 set_bit(cpu, cpu_bitmap);
2395         }
2396
2397         err = 0;
2398
2399 out_delete_map:
2400         perf_cpu_map__put(map);
2401         return err;
2402 }
2403
2404 void perf_session__fprintf_info(struct perf_session *session, FILE *fp,
2405                                 bool full)
2406 {
2407         if (session == NULL || fp == NULL)
2408                 return;
2409
2410         fprintf(fp, "# ========\n");
2411         perf_header__fprintf_info(session, fp, full);
2412         fprintf(fp, "# ========\n#\n");
2413 }
2414
2415 int perf_event__process_id_index(struct perf_session *session,
2416                                  union perf_event *event)
2417 {
2418         struct evlist *evlist = session->evlist;
2419         struct perf_record_id_index *ie = &event->id_index;
2420         size_t i, nr, max_nr;
2421
2422         max_nr = (ie->header.size - sizeof(struct perf_record_id_index)) /
2423                  sizeof(struct id_index_entry);
2424         nr = ie->nr;
2425         if (nr > max_nr)
2426                 return -EINVAL;
2427
2428         if (dump_trace)
2429                 fprintf(stdout, " nr: %zu\n", nr);
2430
2431         for (i = 0; i < nr; i++) {
2432                 struct id_index_entry *e = &ie->entries[i];
2433                 struct perf_sample_id *sid;
2434
2435                 if (dump_trace) {
2436                         fprintf(stdout, " ... id: %"PRI_lu64, e->id);
2437                         fprintf(stdout, "  idx: %"PRI_lu64, e->idx);
2438                         fprintf(stdout, "  cpu: %"PRI_ld64, e->cpu);
2439                         fprintf(stdout, "  tid: %"PRI_ld64"\n", e->tid);
2440                 }
2441
2442                 sid = perf_evlist__id2sid(evlist, e->id);
2443                 if (!sid)
2444                         return -ENOENT;
2445                 sid->idx = e->idx;
2446                 sid->cpu = e->cpu;
2447                 sid->tid = e->tid;
2448         }
2449         return 0;
2450 }