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 #include <linux/err.h>
  37
  38 #ifdef HAVE_ZSTD_SUPPORT
  39 static int perf_session__process_compressed_event(struct perf_session *session,
  40                                                   union perf_event *event, u64 file_offset)
  41 {
  42         void *src;
  43         size_t decomp_size, src_size;
  44         u64 decomp_last_rem = 0;
  45         size_t mmap_len, decomp_len = session->header.env.comp_mmap_len;
  46         struct decomp *decomp, *decomp_last = session->decomp_last;
  47
  48         if (decomp_last) {
  49                 decomp_last_rem = decomp_last->size - decomp_last->head;
  50                 decomp_len += decomp_last_rem;
  51         }
  52
  53         mmap_len = sizeof(struct decomp) + decomp_len;
  54         decomp = mmap(NULL, mmap_len, PROT_READ|PROT_WRITE,
  55                       MAP_ANONYMOUS|MAP_PRIVATE, -1, 0);
  56         if (decomp == MAP_FAILED) {
  57                 pr_err("Couldn't allocate memory for decompression\n");
  58                 return -1;
  59         }
  60
  61         decomp->file_pos = file_offset;
  62         decomp->mmap_len = mmap_len;
  63         decomp->head = 0;
  64
  65         if (decomp_last_rem) {
  66                 memcpy(decomp->data, &(decomp_last->data[decomp_last->head]), decomp_last_rem);
  67                 decomp->size = decomp_last_rem;
  68         }
  69
  70         src = (void *)event + sizeof(struct perf_record_compressed);
  71         src_size = event->pack.header.size - sizeof(struct perf_record_compressed);
  72
  73         decomp_size = zstd_decompress_stream(&(session->zstd_data), src, src_size,
  74                                 &(decomp->data[decomp_last_rem]), decomp_len - decomp_last_rem);
  75         if (!decomp_size) {
  76                 munmap(decomp, mmap_len);
  77                 pr_err("Couldn't decompress data\n");
  78                 return -1;
  79         }
  80
  81         decomp->size += decomp_size;
  82
  83         if (session->decomp == NULL) {
  84                 session->decomp = decomp;
  85                 session->decomp_last = decomp;
  86         } else {
  87                 session->decomp_last->next = decomp;
  88                 session->decomp_last = decomp;
  89         }
  90
  91         pr_debug("decomp (B): %ld to %ld\n", src_size, decomp_size);
  92
  93         return 0;
  94 }
  95 #else /* !HAVE_ZSTD_SUPPORT */
  96 #define perf_session__process_compressed_event perf_session__process_compressed_event_stub
  97 #endif
  98
  99 static int perf_session__deliver_event(struct perf_session *session,
 100                                        union perf_event *event,
 101                                        struct perf_tool *tool,
 102                                        u64 file_offset);
 103
 104 static int perf_session__open(struct perf_session *session)
 105 {
 106         struct perf_data *data = session->data;
 107
 108         if (perf_session__read_header(session) < 0) {
 109                 pr_err("incompatible file format (rerun with -v to learn more)\n");
 110                 return -1;
 111         }
 112
 113         if (perf_data__is_pipe(data))
 114                 return 0;
 115
 116         if (perf_header__has_feat(&session->header, HEADER_STAT))
 117                 return 0;
 118
 119         if (!perf_evlist__valid_sample_type(session->evlist)) {
 120                 pr_err("non matching sample_type\n");
 121                 return -1;
 122         }
 123
 124         if (!perf_evlist__valid_sample_id_all(session->evlist)) {
 125                 pr_err("non matching sample_id_all\n");
 126                 return -1;
 127         }
 128
 129         if (!perf_evlist__valid_read_format(session->evlist)) {
 130                 pr_err("non matching read_format\n");
 131                 return -1;
 132         }
 133
 134         return 0;
 135 }
 136
 137 void perf_session__set_id_hdr_size(struct perf_session *session)
 138 {
 139         u16 id_hdr_size = perf_evlist__id_hdr_size(session->evlist);
 140
 141         machines__set_id_hdr_size(&session->machines, id_hdr_size);
 142 }
 143
 144 int perf_session__create_kernel_maps(struct perf_session *session)
 145 {
 146         int ret = machine__create_kernel_maps(&session->machines.host);
 147
 148         if (ret >= 0)
 149                 ret = machines__create_guest_kernel_maps(&session->machines);
 150         return ret;
 151 }
 152
 153 static void perf_session__destroy_kernel_maps(struct perf_session *session)
 154 {
 155         machines__destroy_kernel_maps(&session->machines);
 156 }
 157
 158 static bool perf_session__has_comm_exec(struct perf_session *session)
 159 {
 160         struct evsel *evsel;
 161
 162         evlist__for_each_entry(session->evlist, evsel) {
 163                 if (evsel->core.attr.comm_exec)
 164                         return true;
 165         }
 166
 167         return false;
 168 }
 169
 170 static void perf_session__set_comm_exec(struct perf_session *session)
 171 {
 172         bool comm_exec = perf_session__has_comm_exec(session);
 173
 174         machines__set_comm_exec(&session->machines, comm_exec);
 175 }
 176
 177 static int ordered_events__deliver_event(struct ordered_events *oe,
 178                                          struct ordered_event *event)
 179 {
 180         struct perf_session *session = container_of(oe, struct perf_session,
 181                                                     ordered_events);
 182
 183         return perf_session__deliver_event(session, event->event,
 184                                            session->tool, event->file_offset);
 185 }
 186
 187 struct perf_session *perf_session__new(struct perf_data *data,
 188                                        bool repipe, struct perf_tool *tool)
 189 {
 190         int ret = -ENOMEM;
 191         struct perf_session *session = zalloc(sizeof(*session));
 192
 193         if (!session)
 194                 goto out;
 195
 196         session->repipe = repipe;
 197         session->tool   = tool;
 198         INIT_LIST_HEAD(&session->auxtrace_index);
 199         machines__init(&session->machines);
 200         ordered_events__init(&session->ordered_events,
 201                              ordered_events__deliver_event, NULL);
 202
 203         perf_env__init(&session->header.env);
 204         if (data) {
 205                 ret = perf_data__open(data);
 206                 if (ret < 0)
 207                         goto out_delete;
 208
 209                 session->data = data;
 210
 211                 if (perf_data__is_read(data)) {
 212                         ret = perf_session__open(session);
 213                         if (ret < 0)
 214                                 goto out_delete;
 215
 216                         /*
 217                          * set session attributes that are present in perf.data
 218                          * but not in pipe-mode.
 219                          */
 220                         if (!data->is_pipe) {
 221                                 perf_session__set_id_hdr_size(session);
 222                                 perf_session__set_comm_exec(session);
 223                         }
 224
 225                         perf_evlist__init_trace_event_sample_raw(session->evlist);
 226
 227                         /* Open the directory data. */
 228                         if (data->is_dir) {
 229                                 ret = perf_data__open_dir(data);
 230                                 if (ret)
 231                                         goto out_delete;
 232                         }
 233
 234                         if (!symbol_conf.kallsyms_name &&
 235                             !symbol_conf.vmlinux_name)
 236                                 symbol_conf.kallsyms_name = perf_data__kallsyms_name(data);
 237                 }
 238         } else  {
 239                 session->machines.host.env = &perf_env;
 240         }
 241
 242         session->machines.host.single_address_space =
 243                 perf_env__single_address_space(session->machines.host.env);
 244
 245         if (!data || perf_data__is_write(data)) {
 246                 /*
 247                  * In O_RDONLY mode this will be performed when reading the
 248                  * kernel MMAP event, in perf_event__process_mmap().
 249                  */
 250                 if (perf_session__create_kernel_maps(session) < 0)
 251                         pr_warning("Cannot read kernel map\n");
 252         }
 253
 254         /*
 255          * In pipe-mode, evlist is empty until PERF_RECORD_HEADER_ATTR is
 256          * processed, so perf_evlist__sample_id_all is not meaningful here.
 257          */
 258         if ((!data || !data->is_pipe) && tool && tool->ordering_requires_timestamps &&
 259             tool->ordered_events && !perf_evlist__sample_id_all(session->evlist)) {
 260                 dump_printf("WARNING: No sample_id_all support, falling back to unordered processing\n");
 261                 tool->ordered_events = false;
 262         }
 263
 264         return session;
 265
 266  out_delete:
 267         perf_session__delete(session);
 268  out:
 269         return ERR_PTR(ret);
 270 }
 271
 272 static void perf_session__delete_threads(struct perf_session *session)
 273 {
 274         machine__delete_threads(&session->machines.host);
 275 }
 276
 277 static void perf_session__release_decomp_events(struct perf_session *session)
 278 {
 279         struct decomp *next, *decomp;
 280         size_t mmap_len;
 281         next = session->decomp;
 282         do {
 283                 decomp = next;
 284                 if (decomp == NULL)
 285                         break;
 286                 next = decomp->next;
 287                 mmap_len = decomp->mmap_len;
 288                 munmap(decomp, mmap_len);
 289         } while (1);
 290 }
 291
 292 void perf_session__delete(struct perf_session *session)
 293 {
 294         if (session == NULL)
 295                 return;
 296         auxtrace__free(session);
 297         auxtrace_index__free(&session->auxtrace_index);
 298         perf_session__destroy_kernel_maps(session);
 299         perf_session__delete_threads(session);
 300         perf_session__release_decomp_events(session);
 301         perf_env__exit(&session->header.env);
 302         machines__exit(&session->machines);
 303         if (session->data)
 304                 perf_data__close(session->data);
 305         free(session);
 306 }
 307
 308 static int process_event_synth_tracing_data_stub(struct perf_session *session
 309                                                  __maybe_unused,
 310                                                  union perf_event *event
 311                                                  __maybe_unused)
 312 {
 313         dump_printf(": unhandled!\n");
 314         return 0;
 315 }
 316
 317 static int process_event_synth_attr_stub(struct perf_tool *tool __maybe_unused,
 318                                          union perf_event *event __maybe_unused,
 319                                          struct evlist **pevlist
 320                                          __maybe_unused)
 321 {
 322         dump_printf(": unhandled!\n");
 323         return 0;
 324 }
 325
 326 static int process_event_synth_event_update_stub(struct perf_tool *tool __maybe_unused,
 327                                                  union perf_event *event __maybe_unused,
 328                                                  struct evlist **pevlist
 329                                                  __maybe_unused)
 330 {
 331         if (dump_trace)
 332                 perf_event__fprintf_event_update(event, stdout);
 333
 334         dump_printf(": unhandled!\n");
 335         return 0;
 336 }
 337
 338 static int process_event_sample_stub(struct perf_tool *tool __maybe_unused,
 339                                      union perf_event *event __maybe_unused,
 340                                      struct perf_sample *sample __maybe_unused,
 341                                      struct evsel *evsel __maybe_unused,
 342                                      struct machine *machine __maybe_unused)
 343 {
 344         dump_printf(": unhandled!\n");
 345         return 0;
 346 }
 347
 348 static int process_event_stub(struct perf_tool *tool __maybe_unused,
 349                               union perf_event *event __maybe_unused,
 350                               struct perf_sample *sample __maybe_unused,
 351                               struct machine *machine __maybe_unused)
 352 {
 353         dump_printf(": unhandled!\n");
 354         return 0;
 355 }
 356
 357 static int process_finished_round_stub(struct perf_tool *tool __maybe_unused,
 358                                        union perf_event *event __maybe_unused,
 359                                        struct ordered_events *oe __maybe_unused)
 360 {
 361         dump_printf(": unhandled!\n");
 362         return 0;
 363 }
 364
 365 static int process_finished_round(struct perf_tool *tool,
 366                                   union perf_event *event,
 367                                   struct ordered_events *oe);
 368
 369 static int skipn(int fd, off_t n)
 370 {
 371         char buf[4096];
 372         ssize_t ret;
 373
 374         while (n > 0) {
 375                 ret = read(fd, buf, min(n, (off_t)sizeof(buf)));
 376                 if (ret <= 0)
 377                         return ret;
 378                 n -= ret;
 379         }
 380
 381         return 0;
 382 }
 383
 384 static s64 process_event_auxtrace_stub(struct perf_session *session __maybe_unused,
 385                                        union perf_event *event)
 386 {
 387         dump_printf(": unhandled!\n");
 388         if (perf_data__is_pipe(session->data))
 389                 skipn(perf_data__fd(session->data), event->auxtrace.size);
 390         return event->auxtrace.size;
 391 }
 392
 393 static int process_event_op2_stub(struct perf_session *session __maybe_unused,
 394                                   union perf_event *event __maybe_unused)
 395 {
 396         dump_printf(": unhandled!\n");
 397         return 0;
 398 }
 399
 400
 401 static
 402 int process_event_thread_map_stub(struct perf_session *session __maybe_unused,
 403                                   union perf_event *event __maybe_unused)
 404 {
 405         if (dump_trace)
 406                 perf_event__fprintf_thread_map(event, stdout);
 407
 408         dump_printf(": unhandled!\n");
 409         return 0;
 410 }
 411
 412 static
 413 int process_event_cpu_map_stub(struct perf_session *session __maybe_unused,
 414                                union perf_event *event __maybe_unused)
 415 {
 416         if (dump_trace)
 417                 perf_event__fprintf_cpu_map(event, stdout);
 418
 419         dump_printf(": unhandled!\n");
 420         return 0;
 421 }
 422
 423 static
 424 int process_event_stat_config_stub(struct perf_session *session __maybe_unused,
 425                                    union perf_event *event __maybe_unused)
 426 {
 427         if (dump_trace)
 428                 perf_event__fprintf_stat_config(event, stdout);
 429
 430         dump_printf(": unhandled!\n");
 431         return 0;
 432 }
 433
 434 static int process_stat_stub(struct perf_session *perf_session __maybe_unused,
 435                              union perf_event *event)
 436 {
 437         if (dump_trace)
 438                 perf_event__fprintf_stat(event, stdout);
 439
 440         dump_printf(": unhandled!\n");
 441         return 0;
 442 }
 443
 444 static int process_stat_round_stub(struct perf_session *perf_session __maybe_unused,
 445                                    union perf_event *event)
 446 {
 447         if (dump_trace)
 448                 perf_event__fprintf_stat_round(event, stdout);
 449
 450         dump_printf(": unhandled!\n");
 451         return 0;
 452 }
 453
 454 static int perf_session__process_compressed_event_stub(struct perf_session *session __maybe_unused,
 455                                                        union perf_event *event __maybe_unused,
 456                                                        u64 file_offset __maybe_unused)
 457 {
 458        dump_printf(": unhandled!\n");
 459        return 0;
 460 }
 461
 462 void perf_tool__fill_defaults(struct perf_tool *tool)
 463 {
 464         if (tool->sample == NULL)
 465                 tool->sample = process_event_sample_stub;
 466         if (tool->mmap == NULL)
 467                 tool->mmap = process_event_stub;
 468         if (tool->mmap2 == NULL)
 469                 tool->mmap2 = process_event_stub;
 470         if (tool->comm == NULL)
 471                 tool->comm = process_event_stub;
 472         if (tool->namespaces == NULL)
 473                 tool->namespaces = process_event_stub;
 474         if (tool->cgroup == NULL)
 475                 tool->cgroup = process_event_stub;
 476         if (tool->fork == NULL)
 477                 tool->fork = process_event_stub;
 478         if (tool->exit == NULL)
 479                 tool->exit = process_event_stub;
 480         if (tool->lost == NULL)
 481                 tool->lost = perf_event__process_lost;
 482         if (tool->lost_samples == NULL)
 483                 tool->lost_samples = perf_event__process_lost_samples;
 484         if (tool->aux == NULL)
 485                 tool->aux = perf_event__process_aux;
 486         if (tool->itrace_start == NULL)
 487                 tool->itrace_start = perf_event__process_itrace_start;
 488         if (tool->context_switch == NULL)
 489                 tool->context_switch = perf_event__process_switch;
 490         if (tool->ksymbol == NULL)
 491                 tool->ksymbol = perf_event__process_ksymbol;
 492         if (tool->bpf == NULL)
 493                 tool->bpf = perf_event__process_bpf;
 494         if (tool->read == NULL)
 495                 tool->read = process_event_sample_stub;
 496         if (tool->throttle == NULL)
 497                 tool->throttle = process_event_stub;
 498         if (tool->unthrottle == NULL)
 499                 tool->unthrottle = process_event_stub;
 500         if (tool->attr == NULL)
 501                 tool->attr = process_event_synth_attr_stub;
 502         if (tool->event_update == NULL)
 503                 tool->event_update = process_event_synth_event_update_stub;
 504         if (tool->tracing_data == NULL)
 505                 tool->tracing_data = process_event_synth_tracing_data_stub;
 506         if (tool->build_id == NULL)
 507                 tool->build_id = process_event_op2_stub;
 508         if (tool->finished_round == NULL) {
 509                 if (tool->ordered_events)
 510                         tool->finished_round = process_finished_round;
 511                 else
 512                         tool->finished_round = process_finished_round_stub;
 513         }
 514         if (tool->id_index == NULL)
 515                 tool->id_index = process_event_op2_stub;
 516         if (tool->auxtrace_info == NULL)
 517                 tool->auxtrace_info = process_event_op2_stub;
 518         if (tool->auxtrace == NULL)
 519                 tool->auxtrace = process_event_auxtrace_stub;
 520         if (tool->auxtrace_error == NULL)
 521                 tool->auxtrace_error = process_event_op2_stub;
 522         if (tool->thread_map == NULL)
 523                 tool->thread_map = process_event_thread_map_stub;
 524         if (tool->cpu_map == NULL)
 525                 tool->cpu_map = process_event_cpu_map_stub;
 526         if (tool->stat_config == NULL)
 527                 tool->stat_config = process_event_stat_config_stub;
 528         if (tool->stat == NULL)
 529                 tool->stat = process_stat_stub;
 530         if (tool->stat_round == NULL)
 531                 tool->stat_round = process_stat_round_stub;
 532         if (tool->time_conv == NULL)
 533                 tool->time_conv = process_event_op2_stub;
 534         if (tool->feature == NULL)
 535                 tool->feature = process_event_op2_stub;
 536         if (tool->compressed == NULL)
 537                 tool->compressed = perf_session__process_compressed_event;
 538 }
 539
 540 static void swap_sample_id_all(union perf_event *event, void *data)
 541 {
 542         void *end = (void *) event + event->header.size;
 543         int size = end - data;
 544
 545         BUG_ON(size % sizeof(u64));
 546         mem_bswap_64(data, size);
 547 }
 548
 549 static void perf_event__all64_swap(union perf_event *event,
 550                                    bool sample_id_all __maybe_unused)
 551 {
 552         struct perf_event_header *hdr = &event->header;
 553         mem_bswap_64(hdr + 1, event->header.size - sizeof(*hdr));
 554 }
 555
 556 static void perf_event__comm_swap(union perf_event *event, bool sample_id_all)
 557 {
 558         event->comm.pid = bswap_32(event->comm.pid);
 559         event->comm.tid = bswap_32(event->comm.tid);
 560
 561         if (sample_id_all) {
 562                 void *data = &event->comm.comm;
 563
 564                 data += PERF_ALIGN(strlen(data) + 1, sizeof(u64));
 565                 swap_sample_id_all(event, data);
 566         }
 567 }
 568
 569 static void perf_event__mmap_swap(union perf_event *event,
 570                                   bool sample_id_all)
 571 {
 572         event->mmap.pid   = bswap_32(event->mmap.pid);
 573         event->mmap.tid   = bswap_32(event->mmap.tid);
 574         event->mmap.start = bswap_64(event->mmap.start);
 575         event->mmap.len   = bswap_64(event->mmap.len);
 576         event->mmap.pgoff = bswap_64(event->mmap.pgoff);
 577
 578         if (sample_id_all) {
 579                 void *data = &event->mmap.filename;
 580
 581                 data += PERF_ALIGN(strlen(data) + 1, sizeof(u64));
 582                 swap_sample_id_all(event, data);
 583         }
 584 }
 585
 586 static void perf_event__mmap2_swap(union perf_event *event,
 587                                   bool sample_id_all)
 588 {
 589         event->mmap2.pid   = bswap_32(event->mmap2.pid);
 590         event->mmap2.tid   = bswap_32(event->mmap2.tid);
 591         event->mmap2.start = bswap_64(event->mmap2.start);
 592         event->mmap2.len   = bswap_64(event->mmap2.len);
 593         event->mmap2.pgoff = bswap_64(event->mmap2.pgoff);
 594         event->mmap2.maj   = bswap_32(event->mmap2.maj);
 595         event->mmap2.min   = bswap_32(event->mmap2.min);
 596         event->mmap2.ino   = bswap_64(event->mmap2.ino);
 597
 598         if (sample_id_all) {
 599                 void *data = &event->mmap2.filename;
 600
 601                 data += PERF_ALIGN(strlen(data) + 1, sizeof(u64));
 602                 swap_sample_id_all(event, data);
 603         }
 604 }
 605 static void perf_event__task_swap(union perf_event *event, bool sample_id_all)
 606 {
 607         event->fork.pid  = bswap_32(event->fork.pid);
 608         event->fork.tid  = bswap_32(event->fork.tid);
 609         event->fork.ppid = bswap_32(event->fork.ppid);
 610         event->fork.ptid = bswap_32(event->fork.ptid);
 611         event->fork.time = bswap_64(event->fork.time);
 612
 613         if (sample_id_all)
 614                 swap_sample_id_all(event, &event->fork + 1);
 615 }
 616
 617 static void perf_event__read_swap(union perf_event *event, bool sample_id_all)
 618 {
 619         event->read.pid          = bswap_32(event->read.pid);
 620         event->read.tid          = bswap_32(event->read.tid);
 621         event->read.value        = bswap_64(event->read.value);
 622         event->read.time_enabled = bswap_64(event->read.time_enabled);
 623         event->read.time_running = bswap_64(event->read.time_running);
 624         event->read.id           = bswap_64(event->read.id);
 625
 626         if (sample_id_all)
 627                 swap_sample_id_all(event, &event->read + 1);
 628 }
 629
 630 static void perf_event__aux_swap(union perf_event *event, bool sample_id_all)
 631 {
 632         event->aux.aux_offset = bswap_64(event->aux.aux_offset);
 633         event->aux.aux_size   = bswap_64(event->aux.aux_size);
 634         event->aux.flags      = bswap_64(event->aux.flags);
 635
 636         if (sample_id_all)
 637                 swap_sample_id_all(event, &event->aux + 1);
 638 }
 639
 640 static void perf_event__itrace_start_swap(union perf_event *event,
 641                                           bool sample_id_all)
 642 {
 643         event->itrace_start.pid  = bswap_32(event->itrace_start.pid);
 644         event->itrace_start.tid  = bswap_32(event->itrace_start.tid);
 645
 646         if (sample_id_all)
 647                 swap_sample_id_all(event, &event->itrace_start + 1);
 648 }
 649
 650 static void perf_event__switch_swap(union perf_event *event, bool sample_id_all)
 651 {
 652         if (event->header.type == PERF_RECORD_SWITCH_CPU_WIDE) {
 653                 event->context_switch.next_prev_pid =
 654                                 bswap_32(event->context_switch.next_prev_pid);
 655                 event->context_switch.next_prev_tid =
 656                                 bswap_32(event->context_switch.next_prev_tid);
 657         }
 658
 659         if (sample_id_all)
 660                 swap_sample_id_all(event, &event->context_switch + 1);
 661 }
 662
 663 static void perf_event__throttle_swap(union perf_event *event,
 664                                       bool sample_id_all)
 665 {
 666         event->throttle.time      = bswap_64(event->throttle.time);
 667         event->throttle.id        = bswap_64(event->throttle.id);
 668         event->throttle.stream_id = bswap_64(event->throttle.stream_id);
 669
 670         if (sample_id_all)
 671                 swap_sample_id_all(event, &event->throttle + 1);
 672 }
 673
 674 static void perf_event__namespaces_swap(union perf_event *event,
 675                                         bool sample_id_all)
 676 {
 677         u64 i;
 678
 679         event->namespaces.pid           = bswap_32(event->namespaces.pid);
 680         event->namespaces.tid           = bswap_32(event->namespaces.tid);
 681         event->namespaces.nr_namespaces = bswap_64(event->namespaces.nr_namespaces);
 682
 683         for (i = 0; i < event->namespaces.nr_namespaces; i++) {
 684                 struct perf_ns_link_info *ns = &event->namespaces.link_info[i];
 685
 686                 ns->dev = bswap_64(ns->dev);
 687                 ns->ino = bswap_64(ns->ino);
 688         }
 689
 690         if (sample_id_all)
 691                 swap_sample_id_all(event, &event->namespaces.link_info[i]);
 692 }
 693
 694 static u8 revbyte(u8 b)
 695 {
 696         int rev = (b >> 4) | ((b & 0xf) << 4);
 697         rev = ((rev & 0xcc) >> 2) | ((rev & 0x33) << 2);
 698         rev = ((rev & 0xaa) >> 1) | ((rev & 0x55) << 1);
 699         return (u8) rev;
 700 }
 701
 702 /*
 703  * XXX this is hack in attempt to carry flags bitfield
 704  * through endian village. ABI says:
 705  *
 706  * Bit-fields are allocated from right to left (least to most significant)
 707  * on little-endian implementations and from left to right (most to least
 708  * significant) on big-endian implementations.
 709  *
 710  * The above seems to be byte specific, so we need to reverse each
 711  * byte of the bitfield. 'Internet' also says this might be implementation
 712  * specific and we probably need proper fix and carry perf_event_attr
 713  * bitfield flags in separate data file FEAT_ section. Thought this seems
 714  * to work for now.
 715  */
 716 static void swap_bitfield(u8 *p, unsigned len)
 717 {
 718         unsigned i;
 719
 720         for (i = 0; i < len; i++) {
 721                 *p = revbyte(*p);
 722                 p++;
 723         }
 724 }
 725
 726 /* exported for swapping attributes in file header */
 727 void perf_event__attr_swap(struct perf_event_attr *attr)
 728 {
 729         attr->type              = bswap_32(attr->type);
 730         attr->size              = bswap_32(attr->size);
 731
 732 #define bswap_safe(f, n)                                        \
 733         (attr->size > (offsetof(struct perf_event_attr, f) +    \
 734                        sizeof(attr->f) * (n)))
 735 #define bswap_field(f, sz)                      \
 736 do {                                            \
 737         if (bswap_safe(f, 0))                   \
 738                 attr->f = bswap_##sz(attr->f);  \
 739 } while(0)
 740 #define bswap_field_16(f) bswap_field(f, 16)
 741 #define bswap_field_32(f) bswap_field(f, 32)
 742 #define bswap_field_64(f) bswap_field(f, 64)
 743
 744         bswap_field_64(config);
 745         bswap_field_64(sample_period);
 746         bswap_field_64(sample_type);
 747         bswap_field_64(read_format);
 748         bswap_field_32(wakeup_events);
 749         bswap_field_32(bp_type);
 750         bswap_field_64(bp_addr);
 751         bswap_field_64(bp_len);
 752         bswap_field_64(branch_sample_type);
 753         bswap_field_64(sample_regs_user);
 754         bswap_field_32(sample_stack_user);
 755         bswap_field_32(aux_watermark);
 756         bswap_field_16(sample_max_stack);
 757         bswap_field_32(aux_sample_size);
 758
 759         /*
 760          * After read_format are bitfields. Check read_format because
 761          * we are unable to use offsetof on bitfield.
 762          */
 763         if (bswap_safe(read_format, 1))
 764                 swap_bitfield((u8 *) (&attr->read_format + 1),
 765                               sizeof(u64));
 766 #undef bswap_field_64
 767 #undef bswap_field_32
 768 #undef bswap_field
 769 #undef bswap_safe
 770 }
 771
 772 static void perf_event__hdr_attr_swap(union perf_event *event,
 773                                       bool sample_id_all __maybe_unused)
 774 {
 775         size_t size;
 776
 777         perf_event__attr_swap(&event->attr.attr);
 778
 779         size = event->header.size;
 780         size -= (void *)&event->attr.id - (void *)event;
 781         mem_bswap_64(event->attr.id, size);
 782 }
 783
 784 static void perf_event__event_update_swap(union perf_event *event,
 785                                           bool sample_id_all __maybe_unused)
 786 {
 787         event->event_update.type = bswap_64(event->event_update.type);
 788         event->event_update.id   = bswap_64(event->event_update.id);
 789 }
 790
 791 static void perf_event__event_type_swap(union perf_event *event,
 792                                         bool sample_id_all __maybe_unused)
 793 {
 794         event->event_type.event_type.event_id =
 795                 bswap_64(event->event_type.event_type.event_id);
 796 }
 797
 798 static void perf_event__tracing_data_swap(union perf_event *event,
 799                                           bool sample_id_all __maybe_unused)
 800 {
 801         event->tracing_data.size = bswap_32(event->tracing_data.size);
 802 }
 803
 804 static void perf_event__auxtrace_info_swap(union perf_event *event,
 805                                            bool sample_id_all __maybe_unused)
 806 {
 807         size_t size;
 808
 809         event->auxtrace_info.type = bswap_32(event->auxtrace_info.type);
 810
 811         size = event->header.size;
 812         size -= (void *)&event->auxtrace_info.priv - (void *)event;
 813         mem_bswap_64(event->auxtrace_info.priv, size);
 814 }
 815
 816 static void perf_event__auxtrace_swap(union perf_event *event,
 817                                       bool sample_id_all __maybe_unused)
 818 {
 819         event->auxtrace.size      = bswap_64(event->auxtrace.size);
 820         event->auxtrace.offset    = bswap_64(event->auxtrace.offset);
 821         event->auxtrace.reference = bswap_64(event->auxtrace.reference);
 822         event->auxtrace.idx       = bswap_32(event->auxtrace.idx);
 823         event->auxtrace.tid       = bswap_32(event->auxtrace.tid);
 824         event->auxtrace.cpu       = bswap_32(event->auxtrace.cpu);
 825 }
 826
 827 static void perf_event__auxtrace_error_swap(union perf_event *event,
 828                                             bool sample_id_all __maybe_unused)
 829 {
 830         event->auxtrace_error.type = bswap_32(event->auxtrace_error.type);
 831         event->auxtrace_error.code = bswap_32(event->auxtrace_error.code);
 832         event->auxtrace_error.cpu  = bswap_32(event->auxtrace_error.cpu);
 833         event->auxtrace_error.pid  = bswap_32(event->auxtrace_error.pid);
 834         event->auxtrace_error.tid  = bswap_32(event->auxtrace_error.tid);
 835         event->auxtrace_error.fmt  = bswap_32(event->auxtrace_error.fmt);
 836         event->auxtrace_error.ip   = bswap_64(event->auxtrace_error.ip);
 837         if (event->auxtrace_error.fmt)
 838                 event->auxtrace_error.time = bswap_64(event->auxtrace_error.time);
 839 }
 840
 841 static void perf_event__thread_map_swap(union perf_event *event,
 842                                         bool sample_id_all __maybe_unused)
 843 {
 844         unsigned i;
 845
 846         event->thread_map.nr = bswap_64(event->thread_map.nr);
 847
 848         for (i = 0; i < event->thread_map.nr; i++)
 849                 event->thread_map.entries[i].pid = bswap_64(event->thread_map.entries[i].pid);
 850 }
 851
 852 static void perf_event__cpu_map_swap(union perf_event *event,
 853                                      bool sample_id_all __maybe_unused)
 854 {
 855         struct perf_record_cpu_map_data *data = &event->cpu_map.data;
 856         struct cpu_map_entries *cpus;
 857         struct perf_record_record_cpu_map *mask;
 858         unsigned i;
 859
 860         data->type = bswap_64(data->type);
 861
 862         switch (data->type) {
 863         case PERF_CPU_MAP__CPUS:
 864                 cpus = (struct cpu_map_entries *)data->data;
 865
 866                 cpus->nr = bswap_16(cpus->nr);
 867
 868                 for (i = 0; i < cpus->nr; i++)
 869                         cpus->cpu[i] = bswap_16(cpus->cpu[i]);
 870                 break;
 871         case PERF_CPU_MAP__MASK:
 872                 mask = (struct perf_record_record_cpu_map *)data->data;
 873
 874                 mask->nr = bswap_16(mask->nr);
 875                 mask->long_size = bswap_16(mask->long_size);
 876
 877                 switch (mask->long_size) {
 878                 case 4: mem_bswap_32(&mask->mask, mask->nr); break;
 879                 case 8: mem_bswap_64(&mask->mask, mask->nr); break;
 880                 default:
 881                         pr_err("cpu_map swap: unsupported long size\n");
 882                 }
 883         default:
 884                 break;
 885         }
 886 }
 887
 888 static void perf_event__stat_config_swap(union perf_event *event,
 889                                          bool sample_id_all __maybe_unused)
 890 {
 891         u64 size;
 892
 893         size  = event->stat_config.nr * sizeof(event->stat_config.data[0]);
 894         size += 1; /* nr item itself */
 895         mem_bswap_64(&event->stat_config.nr, size);
 896 }
 897
 898 static void perf_event__stat_swap(union perf_event *event,
 899                                   bool sample_id_all __maybe_unused)
 900 {
 901         event->stat.id     = bswap_64(event->stat.id);
 902         event->stat.thread = bswap_32(event->stat.thread);
 903         event->stat.cpu    = bswap_32(event->stat.cpu);
 904         event->stat.val    = bswap_64(event->stat.val);
 905         event->stat.ena    = bswap_64(event->stat.ena);
 906         event->stat.run    = bswap_64(event->stat.run);
 907 }
 908
 909 static void perf_event__stat_round_swap(union perf_event *event,
 910                                         bool sample_id_all __maybe_unused)
 911 {
 912         event->stat_round.type = bswap_64(event->stat_round.type);
 913         event->stat_round.time = bswap_64(event->stat_round.time);
 914 }
 915
 916 typedef void (*perf_event__swap_op)(union perf_event *event,
 917                                     bool sample_id_all);
 918
 919 static perf_event__swap_op perf_event__swap_ops[] = {
 920         [PERF_RECORD_MMAP]                = perf_event__mmap_swap,
 921         [PERF_RECORD_MMAP2]               = perf_event__mmap2_swap,
 922         [PERF_RECORD_COMM]                = perf_event__comm_swap,
 923         [PERF_RECORD_FORK]                = perf_event__task_swap,
 924         [PERF_RECORD_EXIT]                = perf_event__task_swap,
 925         [PERF_RECORD_LOST]                = perf_event__all64_swap,
 926         [PERF_RECORD_READ]                = perf_event__read_swap,
 927         [PERF_RECORD_THROTTLE]            = perf_event__throttle_swap,
 928         [PERF_RECORD_UNTHROTTLE]          = perf_event__throttle_swap,
 929         [PERF_RECORD_SAMPLE]              = perf_event__all64_swap,
 930         [PERF_RECORD_AUX]                 = perf_event__aux_swap,
 931         [PERF_RECORD_ITRACE_START]        = perf_event__itrace_start_swap,
 932         [PERF_RECORD_LOST_SAMPLES]        = perf_event__all64_swap,
 933         [PERF_RECORD_SWITCH]              = perf_event__switch_swap,
 934         [PERF_RECORD_SWITCH_CPU_WIDE]     = perf_event__switch_swap,
 935         [PERF_RECORD_NAMESPACES]          = perf_event__namespaces_swap,
 936         [PERF_RECORD_HEADER_ATTR]         = perf_event__hdr_attr_swap,
 937         [PERF_RECORD_HEADER_EVENT_TYPE]   = perf_event__event_type_swap,
 938         [PERF_RECORD_HEADER_TRACING_DATA] = perf_event__tracing_data_swap,
 939         [PERF_RECORD_HEADER_BUILD_ID]     = NULL,
 940         [PERF_RECORD_ID_INDEX]            = perf_event__all64_swap,
 941         [PERF_RECORD_AUXTRACE_INFO]       = perf_event__auxtrace_info_swap,
 942         [PERF_RECORD_AUXTRACE]            = perf_event__auxtrace_swap,
 943         [PERF_RECORD_AUXTRACE_ERROR]      = perf_event__auxtrace_error_swap,
 944         [PERF_RECORD_THREAD_MAP]          = perf_event__thread_map_swap,
 945         [PERF_RECORD_CPU_MAP]             = perf_event__cpu_map_swap,
 946         [PERF_RECORD_STAT_CONFIG]         = perf_event__stat_config_swap,
 947         [PERF_RECORD_STAT]                = perf_event__stat_swap,
 948         [PERF_RECORD_STAT_ROUND]          = perf_event__stat_round_swap,
 949         [PERF_RECORD_EVENT_UPDATE]        = perf_event__event_update_swap,
 950         [PERF_RECORD_TIME_CONV]           = perf_event__all64_swap,
 951         [PERF_RECORD_HEADER_MAX]          = NULL,
 952 };
 953
 954 /*
 955  * When perf record finishes a pass on every buffers, it records this pseudo
 956  * event.
 957  * We record the max timestamp t found in the pass n.
 958  * Assuming these timestamps are monotonic across cpus, we know that if
 959  * a buffer still has events with timestamps below t, they will be all
 960  * available and then read in the pass n + 1.
 961  * Hence when we start to read the pass n + 2, we can safely flush every
 962  * events with timestamps below t.
 963  *
 964  *    ============ PASS n =================
 965  *       CPU 0         |   CPU 1
 966  *                     |
 967  *    cnt1 timestamps  |   cnt2 timestamps
 968  *          1          |         2
 969  *          2          |         3
 970  *          -          |         4  <--- max recorded
 971  *
 972  *    ============ PASS n + 1 ==============
 973  *       CPU 0         |   CPU 1
 974  *                     |
 975  *    cnt1 timestamps  |   cnt2 timestamps
 976  *          3          |         5
 977  *          4          |         6
 978  *          5          |         7 <---- max recorded
 979  *
 980  *      Flush every events below timestamp 4
 981  *
 982  *    ============ PASS n + 2 ==============
 983  *       CPU 0         |   CPU 1
 984  *                     |
 985  *    cnt1 timestamps  |   cnt2 timestamps
 986  *          6          |         8
 987  *          7          |         9
 988  *          -          |         10
 989  *
 990  *      Flush every events below timestamp 7
 991  *      etc...
 992  */
 993 static int process_finished_round(struct perf_tool *tool __maybe_unused,
 994                                   union perf_event *event __maybe_unused,
 995                                   struct ordered_events *oe)
 996 {
 997         if (dump_trace)
 998                 fprintf(stdout, "\n");
 999         return ordered_events__flush(oe, OE_FLUSH__ROUND);
1000 }
1001
1002 int perf_session__queue_event(struct perf_session *s, union perf_event *event,
1003                               u64 timestamp, u64 file_offset)
1004 {
1005         return ordered_events__queue(&s->ordered_events, event, timestamp, file_offset);
1006 }
1007
1008 static void callchain__lbr_callstack_printf(struct perf_sample *sample)
1009 {
1010         struct ip_callchain *callchain = sample->callchain;
1011         struct branch_stack *lbr_stack = sample->branch_stack;
1012         struct branch_entry *entries = perf_sample__branch_entries(sample);
1013         u64 kernel_callchain_nr = callchain->nr;
1014         unsigned int i;
1015
1016         for (i = 0; i < kernel_callchain_nr; i++) {
1017                 if (callchain->ips[i] == PERF_CONTEXT_USER)
1018                         break;
1019         }
1020
1021         if ((i != kernel_callchain_nr) && lbr_stack->nr) {
1022                 u64 total_nr;
1023                 /*
1024                  * LBR callstack can only get user call chain,
1025                  * i is kernel call chain number,
1026                  * 1 is PERF_CONTEXT_USER.
1027                  *
1028                  * The user call chain is stored in LBR registers.
1029                  * LBR are pair registers. The caller is stored
1030                  * in "from" register, while the callee is stored
1031                  * in "to" register.
1032                  * For example, there is a call stack
1033                  * "A"->"B"->"C"->"D".
1034                  * The LBR registers will recorde like
1035                  * "C"->"D", "B"->"C", "A"->"B".
1036                  * So only the first "to" register and all "from"
1037                  * registers are needed to construct the whole stack.
1038                  */
1039                 total_nr = i + 1 + lbr_stack->nr + 1;
1040                 kernel_callchain_nr = i + 1;
1041
1042                 printf("... LBR call chain: nr:%" PRIu64 "\n", total_nr);
1043
1044                 for (i = 0; i < kernel_callchain_nr; i++)
1045                         printf("..... %2d: %016" PRIx64 "\n",
1046                                i, callchain->ips[i]);
1047
1048                 printf("..... %2d: %016" PRIx64 "\n",
1049                        (int)(kernel_callchain_nr), entries[0].to);
1050                 for (i = 0; i < lbr_stack->nr; i++)
1051                         printf("..... %2d: %016" PRIx64 "\n",
1052                                (int)(i + kernel_callchain_nr + 1), entries[i].from);
1053         }
1054 }
1055
1056 static void callchain__printf(struct evsel *evsel,
1057                               struct perf_sample *sample)
1058 {
1059         unsigned int i;
1060         struct ip_callchain *callchain = sample->callchain;
1061
1062         if (evsel__has_branch_callstack(evsel))
1063                 callchain__lbr_callstack_printf(sample);
1064
1065         printf("... FP chain: nr:%" PRIu64 "\n", callchain->nr);
1066
1067         for (i = 0; i < callchain->nr; i++)
1068                 printf("..... %2d: %016" PRIx64 "\n",
1069                        i, callchain->ips[i]);
1070 }
1071
1072 static void branch_stack__printf(struct perf_sample *sample, bool callstack)
1073 {
1074         struct branch_entry *entries = perf_sample__branch_entries(sample);
1075         uint64_t i;
1076
1077         printf("%s: nr:%" PRIu64 "\n",
1078                 !callstack ? "... branch stack" : "... branch callstack",
1079                 sample->branch_stack->nr);
1080
1081         for (i = 0; i < sample->branch_stack->nr; i++) {
1082                 struct branch_entry *e = &entries[i];
1083
1084                 if (!callstack) {
1085                         printf("..... %2"PRIu64": %016" PRIx64 " -> %016" PRIx64 " %hu cycles %s%s%s%s %x\n",
1086                                 i, e->from, e->to,
1087                                 (unsigned short)e->flags.cycles,
1088                                 e->flags.mispred ? "M" : " ",
1089                                 e->flags.predicted ? "P" : " ",
1090                                 e->flags.abort ? "A" : " ",
1091                                 e->flags.in_tx ? "T" : " ",
1092                                 (unsigned)e->flags.reserved);
1093                 } else {
1094                         printf("..... %2"PRIu64": %016" PRIx64 "\n",
1095                                 i, i > 0 ? e->from : e->to);
1096                 }
1097         }
1098 }
1099
1100 static void regs_dump__printf(u64 mask, u64 *regs)
1101 {
1102         unsigned rid, i = 0;
1103
1104         for_each_set_bit(rid, (unsigned long *) &mask, sizeof(mask) * 8) {
1105                 u64 val = regs[i++];
1106
1107                 printf(".... %-5s 0x%" PRIx64 "\n",
1108                        perf_reg_name(rid), val);
1109         }
1110 }
1111
1112 static const char *regs_abi[] = {
1113         [PERF_SAMPLE_REGS_ABI_NONE] = "none",
1114         [PERF_SAMPLE_REGS_ABI_32] = "32-bit",
1115         [PERF_SAMPLE_REGS_ABI_64] = "64-bit",
1116 };
1117
1118 static inline const char *regs_dump_abi(struct regs_dump *d)
1119 {
1120         if (d->abi > PERF_SAMPLE_REGS_ABI_64)
1121                 return "unknown";
1122
1123         return regs_abi[d->abi];
1124 }
1125
1126 static void regs__printf(const char *type, struct regs_dump *regs)
1127 {
1128         u64 mask = regs->mask;
1129
1130         printf("... %s regs: mask 0x%" PRIx64 " ABI %s\n",
1131                type,
1132                mask,
1133                regs_dump_abi(regs));
1134
1135         regs_dump__printf(mask, regs->regs);
1136 }
1137
1138 static void regs_user__printf(struct perf_sample *sample)
1139 {
1140         struct regs_dump *user_regs = &sample->user_regs;
1141
1142         if (user_regs->regs)
1143                 regs__printf("user", user_regs);
1144 }
1145
1146 static void regs_intr__printf(struct perf_sample *sample)
1147 {
1148         struct regs_dump *intr_regs = &sample->intr_regs;
1149
1150         if (intr_regs->regs)
1151                 regs__printf("intr", intr_regs);
1152 }
1153
1154 static void stack_user__printf(struct stack_dump *dump)
1155 {
1156         printf("... ustack: size %" PRIu64 ", offset 0x%x\n",
1157                dump->size, dump->offset);
1158 }
1159
1160 static void perf_evlist__print_tstamp(struct evlist *evlist,
1161                                        union perf_event *event,
1162                                        struct perf_sample *sample)
1163 {
1164         u64 sample_type = __perf_evlist__combined_sample_type(evlist);
1165
1166         if (event->header.type != PERF_RECORD_SAMPLE &&
1167             !perf_evlist__sample_id_all(evlist)) {
1168                 fputs("-1 -1 ", stdout);
1169                 return;
1170         }
1171
1172         if ((sample_type & PERF_SAMPLE_CPU))
1173                 printf("%u ", sample->cpu);
1174
1175         if (sample_type & PERF_SAMPLE_TIME)
1176                 printf("%" PRIu64 " ", sample->time);
1177 }
1178
1179 static void sample_read__printf(struct perf_sample *sample, u64 read_format)
1180 {
1181         printf("... sample_read:\n");
1182
1183         if (read_format & PERF_FORMAT_TOTAL_TIME_ENABLED)
1184                 printf("...... time enabled %016" PRIx64 "\n",
1185                        sample->read.time_enabled);
1186
1187         if (read_format & PERF_FORMAT_TOTAL_TIME_RUNNING)
1188                 printf("...... time running %016" PRIx64 "\n",
1189                        sample->read.time_running);
1190
1191         if (read_format & PERF_FORMAT_GROUP) {
1192                 u64 i;
1193
1194                 printf(".... group nr %" PRIu64 "\n", sample->read.group.nr);
1195
1196                 for (i = 0; i < sample->read.group.nr; i++) {
1197                         struct sample_read_value *value;
1198
1199                         value = &sample->read.group.values[i];
1200                         printf("..... id %016" PRIx64
1201                                ", value %016" PRIx64 "\n",
1202                                value->id, value->value);
1203                 }
1204         } else
1205                 printf("..... id %016" PRIx64 ", value %016" PRIx64 "\n",
1206                         sample->read.one.id, sample->read.one.value);
1207 }
1208
1209 static void dump_event(struct evlist *evlist, union perf_event *event,
1210                        u64 file_offset, struct perf_sample *sample)
1211 {
1212         if (!dump_trace)
1213                 return;
1214
1215         printf("\n%#" PRIx64 " [%#x]: event: %d\n",
1216                file_offset, event->header.size, event->header.type);
1217
1218         trace_event(event);
1219         if (event->header.type == PERF_RECORD_SAMPLE && evlist->trace_event_sample_raw)
1220                 evlist->trace_event_sample_raw(evlist, event, sample);
1221
1222         if (sample)
1223                 perf_evlist__print_tstamp(evlist, event, sample);
1224
1225         printf("%#" PRIx64 " [%#x]: PERF_RECORD_%s", file_offset,
1226                event->header.size, perf_event__name(event->header.type));
1227 }
1228
1229 static void dump_sample(struct evsel *evsel, union perf_event *event,
1230                         struct perf_sample *sample)
1231 {
1232         u64 sample_type;
1233
1234         if (!dump_trace)
1235                 return;
1236
1237         printf("(IP, 0x%x): %d/%d: %#" PRIx64 " period: %" PRIu64 " addr: %#" PRIx64 "\n",
1238                event->header.misc, sample->pid, sample->tid, sample->ip,
1239                sample->period, sample->addr);
1240
1241         sample_type = evsel->core.attr.sample_type;
1242
1243         if (evsel__has_callchain(evsel))
1244                 callchain__printf(evsel, sample);
1245
1246         if (evsel__has_br_stack(evsel))
1247                 branch_stack__printf(sample, evsel__has_branch_callstack(evsel));
1248
1249         if (sample_type & PERF_SAMPLE_REGS_USER)
1250                 regs_user__printf(sample);
1251
1252         if (sample_type & PERF_SAMPLE_REGS_INTR)
1253                 regs_intr__printf(sample);
1254
1255         if (sample_type & PERF_SAMPLE_STACK_USER)
1256                 stack_user__printf(&sample->user_stack);
1257
1258         if (sample_type & PERF_SAMPLE_WEIGHT)
1259                 printf("... weight: %" PRIu64 "\n", sample->weight);
1260
1261         if (sample_type & PERF_SAMPLE_DATA_SRC)
1262                 printf(" . data_src: 0x%"PRIx64"\n", sample->data_src);
1263
1264         if (sample_type & PERF_SAMPLE_PHYS_ADDR)
1265                 printf(" .. phys_addr: 0x%"PRIx64"\n", sample->phys_addr);
1266
1267         if (sample_type & PERF_SAMPLE_TRANSACTION)
1268                 printf("... transaction: %" PRIx64 "\n", sample->transaction);
1269
1270         if (sample_type & PERF_SAMPLE_READ)
1271                 sample_read__printf(sample, evsel->core.attr.read_format);
1272 }
1273
1274 static void dump_read(struct evsel *evsel, union perf_event *event)
1275 {
1276         struct perf_record_read *read_event = &event->read;
1277         u64 read_format;
1278
1279         if (!dump_trace)
1280                 return;
1281
1282         printf(": %d %d %s %" PRI_lu64 "\n", event->read.pid, event->read.tid,
1283                evsel__name(evsel), event->read.value);
1284
1285         if (!evsel)
1286                 return;
1287
1288         read_format = evsel->core.attr.read_format;
1289
1290         if (read_format & PERF_FORMAT_TOTAL_TIME_ENABLED)
1291                 printf("... time enabled : %" PRI_lu64 "\n", read_event->time_enabled);
1292
1293         if (read_format & PERF_FORMAT_TOTAL_TIME_RUNNING)
1294                 printf("... time running : %" PRI_lu64 "\n", read_event->time_running);
1295
1296         if (read_format & PERF_FORMAT_ID)
1297                 printf("... id           : %" PRI_lu64 "\n", read_event->id);
1298 }
1299
1300 static struct machine *machines__find_for_cpumode(struct machines *machines,
1301                                                union perf_event *event,
1302                                                struct perf_sample *sample)
1303 {
1304         struct machine *machine;
1305
1306         if (perf_guest &&
1307             ((sample->cpumode == PERF_RECORD_MISC_GUEST_KERNEL) ||
1308              (sample->cpumode == PERF_RECORD_MISC_GUEST_USER))) {
1309                 u32 pid;
1310
1311                 if (event->header.type == PERF_RECORD_MMAP
1312                     || event->header.type == PERF_RECORD_MMAP2)
1313                         pid = event->mmap.pid;
1314                 else
1315                         pid = sample->pid;
1316
1317                 machine = machines__find(machines, pid);
1318                 if (!machine)
1319                         machine = machines__findnew(machines, DEFAULT_GUEST_KERNEL_ID);
1320                 return machine;
1321         }
1322
1323         return &machines->host;
1324 }
1325
1326 static int deliver_sample_value(struct evlist *evlist,
1327                                 struct perf_tool *tool,
1328                                 union perf_event *event,
1329                                 struct perf_sample *sample,
1330                                 struct sample_read_value *v,
1331                                 struct machine *machine)
1332 {
1333         struct perf_sample_id *sid = perf_evlist__id2sid(evlist, v->id);
1334         struct evsel *evsel;
1335
1336         if (sid) {
1337                 sample->id     = v->id;
1338                 sample->period = v->value - sid->period;
1339                 sid->period    = v->value;
1340         }
1341
1342         if (!sid || sid->evsel == NULL) {
1343                 ++evlist->stats.nr_unknown_id;
1344                 return 0;
1345         }
1346
1347         /*
1348          * There's no reason to deliver sample
1349          * for zero period, bail out.
1350          */
1351         if (!sample->period)
1352                 return 0;
1353
1354         evsel = container_of(sid->evsel, struct evsel, core);
1355         return tool->sample(tool, event, sample, evsel, machine);
1356 }
1357
1358 static int deliver_sample_group(struct evlist *evlist,
1359                                 struct perf_tool *tool,
1360                                 union  perf_event *event,
1361                                 struct perf_sample *sample,
1362                                 struct machine *machine)
1363 {
1364         int ret = -EINVAL;
1365         u64 i;
1366
1367         for (i = 0; i < sample->read.group.nr; i++) {
1368                 ret = deliver_sample_value(evlist, tool, event, sample,
1369                                            &sample->read.group.values[i],
1370                                            machine);
1371                 if (ret)
1372                         break;
1373         }
1374
1375         return ret;
1376 }
1377
1378 static int
1379  perf_evlist__deliver_sample(struct evlist *evlist,
1380                              struct perf_tool *tool,
1381                              union  perf_event *event,
1382                              struct perf_sample *sample,
1383                              struct evsel *evsel,
1384                              struct machine *machine)
1385 {
1386         /* We know evsel != NULL. */
1387         u64 sample_type = evsel->core.attr.sample_type;
1388         u64 read_format = evsel->core.attr.read_format;
1389
1390         /* Standard sample delivery. */
1391         if (!(sample_type & PERF_SAMPLE_READ))
1392                 return tool->sample(tool, event, sample, evsel, machine);
1393
1394         /* For PERF_SAMPLE_READ we have either single or group mode. */
1395         if (read_format & PERF_FORMAT_GROUP)
1396                 return deliver_sample_group(evlist, tool, event, sample,
1397                                             machine);
1398         else
1399                 return deliver_sample_value(evlist, tool, event, sample,
1400                                             &sample->read.one, machine);
1401 }
1402
1403 static int machines__deliver_event(struct machines *machines,
1404                                    struct evlist *evlist,
1405                                    union perf_event *event,
1406                                    struct perf_sample *sample,
1407                                    struct perf_tool *tool, u64 file_offset)
1408 {
1409         struct evsel *evsel;
1410         struct machine *machine;
1411
1412         dump_event(evlist, event, file_offset, sample);
1413
1414         evsel = perf_evlist__id2evsel(evlist, sample->id);
1415
1416         machine = machines__find_for_cpumode(machines, event, sample);
1417
1418         switch (event->header.type) {
1419         case PERF_RECORD_SAMPLE:
1420                 if (evsel == NULL) {
1421                         ++evlist->stats.nr_unknown_id;
1422                         return 0;
1423                 }
1424                 dump_sample(evsel, event, sample);
1425                 if (machine == NULL) {
1426                         ++evlist->stats.nr_unprocessable_samples;
1427                         return 0;
1428                 }
1429                 return perf_evlist__deliver_sample(evlist, tool, event, sample, evsel, machine);
1430         case PERF_RECORD_MMAP:
1431                 return tool->mmap(tool, event, sample, machine);
1432         case PERF_RECORD_MMAP2:
1433                 if (event->header.misc & PERF_RECORD_MISC_PROC_MAP_PARSE_TIMEOUT)
1434                         ++evlist->stats.nr_proc_map_timeout;
1435                 return tool->mmap2(tool, event, sample, machine);
1436         case PERF_RECORD_COMM:
1437                 return tool->comm(tool, event, sample, machine);
1438         case PERF_RECORD_NAMESPACES:
1439                 return tool->namespaces(tool, event, sample, machine);
1440         case PERF_RECORD_CGROUP:
1441                 return tool->cgroup(tool, event, sample, machine);
1442         case PERF_RECORD_FORK:
1443                 return tool->fork(tool, event, sample, machine);
1444         case PERF_RECORD_EXIT:
1445                 return tool->exit(tool, event, sample, machine);
1446         case PERF_RECORD_LOST:
1447                 if (tool->lost == perf_event__process_lost)
1448                         evlist->stats.total_lost += event->lost.lost;
1449                 return tool->lost(tool, event, sample, machine);
1450         case PERF_RECORD_LOST_SAMPLES:
1451                 if (tool->lost_samples == perf_event__process_lost_samples)
1452                         evlist->stats.total_lost_samples += event->lost_samples.lost;
1453                 return tool->lost_samples(tool, event, sample, machine);
1454         case PERF_RECORD_READ:
1455                 dump_read(evsel, event);
1456                 return tool->read(tool, event, sample, evsel, machine);
1457         case PERF_RECORD_THROTTLE:
1458                 return tool->throttle(tool, event, sample, machine);
1459         case PERF_RECORD_UNTHROTTLE:
1460                 return tool->unthrottle(tool, event, sample, machine);
1461         case PERF_RECORD_AUX:
1462                 if (tool->aux == perf_event__process_aux) {
1463                         if (event->aux.flags & PERF_AUX_FLAG_TRUNCATED)
1464                                 evlist->stats.total_aux_lost += 1;
1465                         if (event->aux.flags & PERF_AUX_FLAG_PARTIAL)
1466                                 evlist->stats.total_aux_partial += 1;
1467                 }
1468                 return tool->aux(tool, event, sample, machine);
1469         case PERF_RECORD_ITRACE_START:
1470                 return tool->itrace_start(tool, event, sample, machine);
1471         case PERF_RECORD_SWITCH:
1472         case PERF_RECORD_SWITCH_CPU_WIDE:
1473                 return tool->context_switch(tool, event, sample, machine);
1474         case PERF_RECORD_KSYMBOL:
1475                 return tool->ksymbol(tool, event, sample, machine);
1476         case PERF_RECORD_BPF_EVENT:
1477                 return tool->bpf(tool, event, sample, machine);
1478         default:
1479                 ++evlist->stats.nr_unknown_events;
1480                 return -1;
1481         }
1482 }
1483
1484 static int perf_session__deliver_event(struct perf_session *session,
1485                                        union perf_event *event,
1486                                        struct perf_tool *tool,
1487                                        u64 file_offset)
1488 {
1489         struct perf_sample sample;
1490         int ret;
1491
1492         ret = perf_evlist__parse_sample(session->evlist, event, &sample);
1493         if (ret) {
1494                 pr_err("Can't parse sample, err = %d\n", ret);
1495                 return ret;
1496         }
1497
1498         ret = auxtrace__process_event(session, event, &sample, tool);
1499         if (ret < 0)
1500                 return ret;
1501         if (ret > 0)
1502                 return 0;
1503
1504         ret = machines__deliver_event(&session->machines, session->evlist,
1505                                       event, &sample, tool, file_offset);
1506
1507         if (dump_trace && sample.aux_sample.size)
1508                 auxtrace__dump_auxtrace_sample(session, &sample);
1509
1510         return ret;
1511 }
1512
1513 static s64 perf_session__process_user_event(struct perf_session *session,
1514                                             union perf_event *event,
1515                                             u64 file_offset)
1516 {
1517         struct ordered_events *oe = &session->ordered_events;
1518         struct perf_tool *tool = session->tool;
1519         struct perf_sample sample = { .time = 0, };
1520         int fd = perf_data__fd(session->data);
1521         int err;
1522
1523         if (event->header.type != PERF_RECORD_COMPRESSED ||
1524             tool->compressed == perf_session__process_compressed_event_stub)
1525                 dump_event(session->evlist, event, file_offset, &sample);
1526
1527         /* These events are processed right away */
1528         switch (event->header.type) {
1529         case PERF_RECORD_HEADER_ATTR:
1530                 err = tool->attr(tool, event, &session->evlist);
1531                 if (err == 0) {
1532                         perf_session__set_id_hdr_size(session);
1533                         perf_session__set_comm_exec(session);
1534                 }
1535                 return err;
1536         case PERF_RECORD_EVENT_UPDATE:
1537                 return tool->event_update(tool, event, &session->evlist);
1538         case PERF_RECORD_HEADER_EVENT_TYPE:
1539                 /*
1540                  * Depreceated, but we need to handle it for sake
1541                  * of old data files create in pipe mode.
1542                  */
1543                 return 0;
1544         case PERF_RECORD_HEADER_TRACING_DATA:
1545                 /*
1546                  * Setup for reading amidst mmap, but only when we
1547                  * are in 'file' mode. The 'pipe' fd is in proper
1548                  * place already.
1549                  */
1550                 if (!perf_data__is_pipe(session->data))
1551                         lseek(fd, file_offset, SEEK_SET);
1552                 return tool->tracing_data(session, event);
1553         case PERF_RECORD_HEADER_BUILD_ID:
1554                 return tool->build_id(session, event);
1555         case PERF_RECORD_FINISHED_ROUND:
1556                 return tool->finished_round(tool, event, oe);
1557         case PERF_RECORD_ID_INDEX:
1558                 return tool->id_index(session, event);
1559         case PERF_RECORD_AUXTRACE_INFO:
1560                 return tool->auxtrace_info(session, event);
1561         case PERF_RECORD_AUXTRACE:
1562                 /* setup for reading amidst mmap */
1563                 lseek(fd, file_offset + event->header.size, SEEK_SET);
1564                 return tool->auxtrace(session, event);
1565         case PERF_RECORD_AUXTRACE_ERROR:
1566                 perf_session__auxtrace_error_inc(session, event);
1567                 return tool->auxtrace_error(session, event);
1568         case PERF_RECORD_THREAD_MAP:
1569                 return tool->thread_map(session, event);
1570         case PERF_RECORD_CPU_MAP:
1571                 return tool->cpu_map(session, event);
1572         case PERF_RECORD_STAT_CONFIG:
1573                 return tool->stat_config(session, event);
1574         case PERF_RECORD_STAT:
1575                 return tool->stat(session, event);
1576         case PERF_RECORD_STAT_ROUND:
1577                 return tool->stat_round(session, event);
1578         case PERF_RECORD_TIME_CONV:
1579                 session->time_conv = event->time_conv;
1580                 return tool->time_conv(session, event);
1581         case PERF_RECORD_HEADER_FEATURE:
1582                 return tool->feature(session, event);
1583         case PERF_RECORD_COMPRESSED:
1584                 err = tool->compressed(session, event, file_offset);
1585                 if (err)
1586                         dump_event(session->evlist, event, file_offset, &sample);
1587                 return err;
1588         default:
1589                 return -EINVAL;
1590         }
1591 }
1592
1593 int perf_session__deliver_synth_event(struct perf_session *session,
1594                                       union perf_event *event,
1595                                       struct perf_sample *sample)
1596 {
1597         struct evlist *evlist = session->evlist;
1598         struct perf_tool *tool = session->tool;
1599
1600         events_stats__inc(&evlist->stats, event->header.type);
1601
1602         if (event->header.type >= PERF_RECORD_USER_TYPE_START)
1603                 return perf_session__process_user_event(session, event, 0);
1604
1605         return machines__deliver_event(&session->machines, evlist, event, sample, tool, 0);
1606 }
1607
1608 static void event_swap(union perf_event *event, bool sample_id_all)
1609 {
1610         perf_event__swap_op swap;
1611
1612         swap = perf_event__swap_ops[event->header.type];
1613         if (swap)
1614                 swap(event, sample_id_all);
1615 }
1616
1617 int perf_session__peek_event(struct perf_session *session, off_t file_offset,
1618                              void *buf, size_t buf_sz,
1619                              union perf_event **event_ptr,
1620                              struct perf_sample *sample)
1621 {
1622         union perf_event *event;
1623         size_t hdr_sz, rest;
1624         int fd;
1625
1626         if (session->one_mmap && !session->header.needs_swap) {
1627                 event = file_offset - session->one_mmap_offset +
1628                         session->one_mmap_addr;
1629                 goto out_parse_sample;
1630         }
1631
1632         if (perf_data__is_pipe(session->data))
1633                 return -1;
1634
1635         fd = perf_data__fd(session->data);
1636         hdr_sz = sizeof(struct perf_event_header);
1637
1638         if (buf_sz < hdr_sz)
1639                 return -1;
1640
1641         if (lseek(fd, file_offset, SEEK_SET) == (off_t)-1 ||
1642             readn(fd, buf, hdr_sz) != (ssize_t)hdr_sz)
1643                 return -1;
1644
1645         event = (union perf_event *)buf;
1646
1647         if (session->header.needs_swap)
1648                 perf_event_header__bswap(&event->header);
1649
1650         if (event->header.size < hdr_sz || event->header.size > buf_sz)
1651                 return -1;
1652
1653         rest = event->header.size - hdr_sz;
1654
1655         if (readn(fd, buf, rest) != (ssize_t)rest)
1656                 return -1;
1657
1658         if (session->header.needs_swap)
1659                 event_swap(event, perf_evlist__sample_id_all(session->evlist));
1660
1661 out_parse_sample:
1662
1663         if (sample && event->header.type < PERF_RECORD_USER_TYPE_START &&
1664             perf_evlist__parse_sample(session->evlist, event, sample))
1665                 return -1;
1666
1667         *event_ptr = event;
1668
1669         return 0;
1670 }
1671
1672 int perf_session__peek_events(struct perf_session *session, u64 offset,
1673                               u64 size, peek_events_cb_t cb, void *data)
1674 {
1675         u64 max_offset = offset + size;
1676         char buf[PERF_SAMPLE_MAX_SIZE];
1677         union perf_event *event;
1678         int err;
1679
1680         do {
1681                 err = perf_session__peek_event(session, offset, buf,
1682                                                PERF_SAMPLE_MAX_SIZE, &event,
1683                                                NULL);
1684                 if (err)
1685                         return err;
1686
1687                 err = cb(session, event, offset, data);
1688                 if (err)
1689                         return err;
1690
1691                 offset += event->header.size;
1692                 if (event->header.type == PERF_RECORD_AUXTRACE)
1693                         offset += event->auxtrace.size;
1694
1695         } while (offset < max_offset);
1696
1697         return err;
1698 }
1699
1700 static s64 perf_session__process_event(struct perf_session *session,
1701                                        union perf_event *event, u64 file_offset)
1702 {
1703         struct evlist *evlist = session->evlist;
1704         struct perf_tool *tool = session->tool;
1705         int ret;
1706
1707         if (session->header.needs_swap)
1708                 event_swap(event, perf_evlist__sample_id_all(evlist));
1709
1710         if (event->header.type >= PERF_RECORD_HEADER_MAX)
1711                 return -EINVAL;
1712
1713         events_stats__inc(&evlist->stats, event->header.type);
1714
1715         if (event->header.type >= PERF_RECORD_USER_TYPE_START)
1716                 return perf_session__process_user_event(session, event, file_offset);
1717
1718         if (tool->ordered_events) {
1719                 u64 timestamp = -1ULL;
1720
1721                 ret = perf_evlist__parse_sample_timestamp(evlist, event, &timestamp);
1722                 if (ret && ret != -1)
1723                         return ret;
1724
1725                 ret = perf_session__queue_event(session, event, timestamp, file_offset);
1726                 if (ret != -ETIME)
1727                         return ret;
1728         }
1729
1730         return perf_session__deliver_event(session, event, tool, file_offset);
1731 }
1732
1733 void perf_event_header__bswap(struct perf_event_header *hdr)
1734 {
1735         hdr->type = bswap_32(hdr->type);
1736         hdr->misc = bswap_16(hdr->misc);
1737         hdr->size = bswap_16(hdr->size);
1738 }
1739
1740 struct thread *perf_session__findnew(struct perf_session *session, pid_t pid)
1741 {
1742         return machine__findnew_thread(&session->machines.host, -1, pid);
1743 }
1744
1745 /*
1746  * Threads are identified by pid and tid, and the idle task has pid == tid == 0.
1747  * So here a single thread is created for that, but actually there is a separate
1748  * idle task per cpu, so there should be one 'struct thread' per cpu, but there
1749  * is only 1. That causes problems for some tools, requiring workarounds. For
1750  * example get_idle_thread() in builtin-sched.c, or thread_stack__per_cpu().
1751  */
1752 int perf_session__register_idle_thread(struct perf_session *session)
1753 {
1754         struct thread *thread;
1755         int err = 0;
1756
1757         thread = machine__findnew_thread(&session->machines.host, 0, 0);
1758         if (thread == NULL || thread__set_comm(thread, "swapper", 0)) {
1759                 pr_err("problem inserting idle task.\n");
1760                 err = -1;
1761         }
1762
1763         if (thread == NULL || thread__set_namespaces(thread, 0, NULL)) {
1764                 pr_err("problem inserting idle task.\n");
1765                 err = -1;
1766         }
1767
1768         /* machine__findnew_thread() got the thread, so put it */
1769         thread__put(thread);
1770         return err;
1771 }
1772
1773 static void
1774 perf_session__warn_order(const struct perf_session *session)
1775 {
1776         const struct ordered_events *oe = &session->ordered_events;
1777         struct evsel *evsel;
1778         bool should_warn = true;
1779
1780         evlist__for_each_entry(session->evlist, evsel) {
1781                 if (evsel->core.attr.write_backward)
1782                         should_warn = false;
1783         }
1784
1785         if (!should_warn)
1786                 return;
1787         if (oe->nr_unordered_events != 0)
1788                 ui__warning("%u out of order events recorded.\n", oe->nr_unordered_events);
1789 }
1790
1791 static void perf_session__warn_about_errors(const struct perf_session *session)
1792 {
1793         const struct events_stats *stats = &session->evlist->stats;
1794
1795         if (session->tool->lost == perf_event__process_lost &&
1796             stats->nr_events[PERF_RECORD_LOST] != 0) {
1797                 ui__warning("Processed %d events and lost %d chunks!\n\n"
1798                             "Check IO/CPU overload!\n\n",
1799                             stats->nr_events[0],
1800                             stats->nr_events[PERF_RECORD_LOST]);
1801         }
1802
1803         if (session->tool->lost_samples == perf_event__process_lost_samples) {
1804                 double drop_rate;
1805
1806                 drop_rate = (double)stats->total_lost_samples /
1807                             (double) (stats->nr_events[PERF_RECORD_SAMPLE] + stats->total_lost_samples);
1808                 if (drop_rate > 0.05) {
1809                         ui__warning("Processed %" PRIu64 " samples and lost %3.2f%%!\n\n",
1810                                     stats->nr_events[PERF_RECORD_SAMPLE] + stats->total_lost_samples,
1811                                     drop_rate * 100.0);
1812                 }
1813         }
1814
1815         if (session->tool->aux == perf_event__process_aux &&
1816             stats->total_aux_lost != 0) {
1817                 ui__warning("AUX data lost %" PRIu64 " times out of %u!\n\n",
1818                             stats->total_aux_lost,
1819                             stats->nr_events[PERF_RECORD_AUX]);
1820         }
1821
1822         if (session->tool->aux == perf_event__process_aux &&
1823             stats->total_aux_partial != 0) {
1824                 bool vmm_exclusive = false;
1825
1826                 (void)sysfs__read_bool("module/kvm_intel/parameters/vmm_exclusive",
1827                                        &vmm_exclusive);
1828
1829                 ui__warning("AUX data had gaps in it %" PRIu64 " times out of %u!\n\n"
1830                             "Are you running a KVM guest in the background?%s\n\n",
1831                             stats->total_aux_partial,
1832                             stats->nr_events[PERF_RECORD_AUX],
1833                             vmm_exclusive ?
1834                             "\nReloading kvm_intel module with vmm_exclusive=0\n"
1835                             "will reduce the gaps to only guest's timeslices." :
1836                             "");
1837         }
1838
1839         if (stats->nr_unknown_events != 0) {
1840                 ui__warning("Found %u unknown events!\n\n"
1841                             "Is this an older tool processing a perf.data "
1842                             "file generated by a more recent tool?\n\n"
1843                             "If that is not the case, consider "
1844                             "reporting to linux-kernel@vger.kernel.org.\n\n",
1845                             stats->nr_unknown_events);
1846         }
1847
1848         if (stats->nr_unknown_id != 0) {
1849                 ui__warning("%u samples with id not present in the header\n",
1850                             stats->nr_unknown_id);
1851         }
1852
1853         if (stats->nr_invalid_chains != 0) {
1854                 ui__warning("Found invalid callchains!\n\n"
1855                             "%u out of %u events were discarded for this reason.\n\n"
1856                             "Consider reporting to linux-kernel@vger.kernel.org.\n\n",
1857                             stats->nr_invalid_chains,
1858                             stats->nr_events[PERF_RECORD_SAMPLE]);
1859         }
1860
1861         if (stats->nr_unprocessable_samples != 0) {
1862                 ui__warning("%u unprocessable samples recorded.\n"
1863                             "Do you have a KVM guest running and not using 'perf kvm'?\n",
1864                             stats->nr_unprocessable_samples);
1865         }
1866
1867         perf_session__warn_order(session);
1868
1869         events_stats__auxtrace_error_warn(stats);
1870
1871         if (stats->nr_proc_map_timeout != 0) {
1872                 ui__warning("%d map information files for pre-existing threads were\n"
1873                             "not processed, if there are samples for addresses they\n"
1874                             "will not be resolved, you may find out which are these\n"
1875                             "threads by running with -v and redirecting the output\n"
1876                             "to a file.\n"
1877                             "The time limit to process proc map is too short?\n"
1878                             "Increase it by --proc-map-timeout\n",
1879                             stats->nr_proc_map_timeout);
1880         }
1881 }
1882
1883 static int perf_session__flush_thread_stack(struct thread *thread,
1884                                             void *p __maybe_unused)
1885 {
1886         return thread_stack__flush(thread);
1887 }
1888
1889 static int perf_session__flush_thread_stacks(struct perf_session *session)
1890 {
1891         return machines__for_each_thread(&session->machines,
1892                                          perf_session__flush_thread_stack,
1893                                          NULL);
1894 }
1895
1896 volatile int session_done;
1897
1898 static int __perf_session__process_decomp_events(struct perf_session *session);
1899
1900 static int __perf_session__process_pipe_events(struct perf_session *session)
1901 {
1902         struct ordered_events *oe = &session->ordered_events;
1903         struct perf_tool *tool = session->tool;
1904         int fd = perf_data__fd(session->data);
1905         union perf_event *event;
1906         uint32_t size, cur_size = 0;
1907         void *buf = NULL;
1908         s64 skip = 0;
1909         u64 head;
1910         ssize_t err;
1911         void *p;
1912
1913         perf_tool__fill_defaults(tool);
1914
1915         head = 0;
1916         cur_size = sizeof(union perf_event);
1917
1918         buf = malloc(cur_size);
1919         if (!buf)
1920                 return -errno;
1921         ordered_events__set_copy_on_queue(oe, true);
1922 more:
1923         event = buf;
1924         err = readn(fd, event, sizeof(struct perf_event_header));
1925         if (err <= 0) {
1926                 if (err == 0)
1927                         goto done;
1928
1929                 pr_err("failed to read event header\n");
1930                 goto out_err;
1931         }
1932
1933         if (session->header.needs_swap)
1934                 perf_event_header__bswap(&event->header);
1935
1936         size = event->header.size;
1937         if (size < sizeof(struct perf_event_header)) {
1938                 pr_err("bad event header size\n");
1939                 goto out_err;
1940         }
1941
1942         if (size > cur_size) {
1943                 void *new = realloc(buf, size);
1944                 if (!new) {
1945                         pr_err("failed to allocate memory to read event\n");
1946                         goto out_err;
1947                 }
1948                 buf = new;
1949                 cur_size = size;
1950                 event = buf;
1951         }
1952         p = event;
1953         p += sizeof(struct perf_event_header);
1954
1955         if (size - sizeof(struct perf_event_header)) {
1956                 err = readn(fd, p, size - sizeof(struct perf_event_header));
1957                 if (err <= 0) {
1958                         if (err == 0) {
1959                                 pr_err("unexpected end of event stream\n");
1960                                 goto done;
1961                         }
1962
1963                         pr_err("failed to read event data\n");
1964                         goto out_err;
1965                 }
1966         }
1967
1968         if ((skip = perf_session__process_event(session, event, head)) < 0) {
1969                 pr_err("%#" PRIx64 " [%#x]: failed to process type: %d\n",
1970                        head, event->header.size, event->header.type);
1971                 err = -EINVAL;
1972                 goto out_err;
1973         }
1974
1975         head += size;
1976
1977         if (skip > 0)
1978                 head += skip;
1979
1980         err = __perf_session__process_decomp_events(session);
1981         if (err)
1982                 goto out_err;
1983
1984         if (!session_done())
1985                 goto more;
1986 done:
1987         /* do the final flush for ordered samples */
1988         err = ordered_events__flush(oe, OE_FLUSH__FINAL);
1989         if (err)
1990                 goto out_err;
1991         err = auxtrace__flush_events(session, tool);
1992         if (err)
1993                 goto out_err;
1994         err = perf_session__flush_thread_stacks(session);
1995 out_err:
1996         free(buf);
1997         if (!tool->no_warn)
1998                 perf_session__warn_about_errors(session);
1999         ordered_events__free(&session->ordered_events);
2000         auxtrace__free_events(session);
2001         return err;
2002 }
2003
2004 static union perf_event *
2005 prefetch_event(char *buf, u64 head, size_t mmap_size,
2006                bool needs_swap, union perf_event *error)
2007 {
2008         union perf_event *event;
2009
2010         /*
2011          * Ensure we have enough space remaining to read
2012          * the size of the event in the headers.
2013          */
2014         if (head + sizeof(event->header) > mmap_size)
2015                 return NULL;
2016
2017         event = (union perf_event *)(buf + head);
2018         if (needs_swap)
2019                 perf_event_header__bswap(&event->header);
2020
2021         if (head + event->header.size <= mmap_size)
2022                 return event;
2023
2024         /* We're not fetching the event so swap back again */
2025         if (needs_swap)
2026                 perf_event_header__bswap(&event->header);
2027
2028         pr_debug("%s: head=%#" PRIx64 " event->header_size=%#x, mmap_size=%#zx:"
2029                  " fuzzed or compressed perf.data?\n",__func__, head, event->header.size, mmap_size);
2030
2031         return error;
2032 }
2033
2034 static union perf_event *
2035 fetch_mmaped_event(u64 head, size_t mmap_size, char *buf, bool needs_swap)
2036 {
2037         return prefetch_event(buf, head, mmap_size, needs_swap, ERR_PTR(-EINVAL));
2038 }
2039
2040 static union perf_event *
2041 fetch_decomp_event(u64 head, size_t mmap_size, char *buf, bool needs_swap)
2042 {
2043         return prefetch_event(buf, head, mmap_size, needs_swap, NULL);
2044 }
2045
2046 static int __perf_session__process_decomp_events(struct perf_session *session)
2047 {
2048         s64 skip;
2049         u64 size, file_pos = 0;
2050         struct decomp *decomp = session->decomp_last;
2051
2052         if (!decomp)
2053                 return 0;
2054
2055         while (decomp->head < decomp->size && !session_done()) {
2056                 union perf_event *event = fetch_decomp_event(decomp->head, decomp->size, decomp->data,
2057                                                              session->header.needs_swap);
2058
2059                 if (!event)
2060                         break;
2061
2062                 size = event->header.size;
2063
2064                 if (size < sizeof(struct perf_event_header) ||
2065                     (skip = perf_session__process_event(session, event, file_pos)) < 0) {
2066                         pr_err("%#" PRIx64 " [%#x]: failed to process type: %d\n",
2067                                 decomp->file_pos + decomp->head, event->header.size, event->header.type);
2068                         return -EINVAL;
2069                 }
2070
2071                 if (skip)
2072                         size += skip;
2073
2074                 decomp->head += size;
2075         }
2076
2077         return 0;
2078 }
2079
2080 /*
2081  * On 64bit we can mmap the data file in one go. No need for tiny mmap
2082  * slices. On 32bit we use 32MB.
2083  */
2084 #if BITS_PER_LONG == 64
2085 #define MMAP_SIZE ULLONG_MAX
2086 #define NUM_MMAPS 1
2087 #else
2088 #define MMAP_SIZE (32 * 1024 * 1024ULL)
2089 #define NUM_MMAPS 128
2090 #endif
2091
2092 struct reader;
2093
2094 typedef s64 (*reader_cb_t)(struct perf_session *session,
2095                            union perf_event *event,
2096                            u64 file_offset);
2097
2098 struct reader {
2099         int              fd;
2100         u64              data_size;
2101         u64              data_offset;
2102         reader_cb_t      process;
2103 };
2104
2105 static int
2106 reader__process_events(struct reader *rd, struct perf_session *session,
2107                        struct ui_progress *prog)
2108 {
2109         u64 data_size = rd->data_size;
2110         u64 head, page_offset, file_offset, file_pos, size;
2111         int err = 0, mmap_prot, mmap_flags, map_idx = 0;
2112         size_t  mmap_size;
2113         char *buf, *mmaps[NUM_MMAPS];
2114         union perf_event *event;
2115         s64 skip;
2116
2117         page_offset = page_size * (rd->data_offset / page_size);
2118         file_offset = page_offset;
2119         head = rd->data_offset - page_offset;
2120
2121         ui_progress__init_size(prog, data_size, "Processing events...");
2122
2123         data_size += rd->data_offset;
2124
2125         mmap_size = MMAP_SIZE;
2126         if (mmap_size > data_size) {
2127                 mmap_size = data_size;
2128                 session->one_mmap = true;
2129         }
2130
2131         memset(mmaps, 0, sizeof(mmaps));
2132
2133         mmap_prot  = PROT_READ;
2134         mmap_flags = MAP_SHARED;
2135
2136         if (session->header.needs_swap) {
2137                 mmap_prot  |= PROT_WRITE;
2138                 mmap_flags = MAP_PRIVATE;
2139         }
2140 remap:
2141         buf = mmap(NULL, mmap_size, mmap_prot, mmap_flags, rd->fd,
2142                    file_offset);
2143         if (buf == MAP_FAILED) {
2144                 pr_err("failed to mmap file\n");
2145                 err = -errno;
2146                 goto out;
2147         }
2148         mmaps[map_idx] = buf;
2149         map_idx = (map_idx + 1) & (ARRAY_SIZE(mmaps) - 1);
2150         file_pos = file_offset + head;
2151         if (session->one_mmap) {
2152                 session->one_mmap_addr = buf;
2153                 session->one_mmap_offset = file_offset;
2154         }
2155
2156 more:
2157         event = fetch_mmaped_event(head, mmap_size, buf, session->header.needs_swap);
2158         if (IS_ERR(event))
2159                 return PTR_ERR(event);
2160
2161         if (!event) {
2162                 if (mmaps[map_idx]) {
2163                         munmap(mmaps[map_idx], mmap_size);
2164                         mmaps[map_idx] = NULL;
2165                 }
2166
2167                 page_offset = page_size * (head / page_size);
2168                 file_offset += page_offset;
2169                 head -= page_offset;
2170                 goto remap;
2171         }
2172
2173         size = event->header.size;
2174
2175         skip = -EINVAL;
2176
2177         if (size < sizeof(struct perf_event_header) ||
2178             (skip = rd->process(session, event, file_pos)) < 0) {
2179                 pr_err("%#" PRIx64 " [%#x]: failed to process type: %d [%s]\n",
2180                        file_offset + head, event->header.size,
2181                        event->header.type, strerror(-skip));
2182                 err = skip;
2183                 goto out;
2184         }
2185
2186         if (skip)
2187                 size += skip;
2188
2189         head += size;
2190         file_pos += size;
2191
2192         err = __perf_session__process_decomp_events(session);
2193         if (err)
2194                 goto out;
2195
2196         ui_progress__update(prog, size);
2197
2198         if (session_done())
2199                 goto out;
2200
2201         if (file_pos < data_size)
2202                 goto more;
2203
2204 out:
2205         return err;
2206 }
2207
2208 static s64 process_simple(struct perf_session *session,
2209                           union perf_event *event,
2210                           u64 file_offset)
2211 {
2212         return perf_session__process_event(session, event, file_offset);
2213 }
2214
2215 static int __perf_session__process_events(struct perf_session *session)
2216 {
2217         struct reader rd = {
2218                 .fd             = perf_data__fd(session->data),
2219                 .data_size      = session->header.data_size,
2220                 .data_offset    = session->header.data_offset,
2221                 .process        = process_simple,
2222         };
2223         struct ordered_events *oe = &session->ordered_events;
2224         struct perf_tool *tool = session->tool;
2225         struct ui_progress prog;
2226         int err;
2227
2228         perf_tool__fill_defaults(tool);
2229
2230         if (rd.data_size == 0)
2231                 return -1;
2232
2233         ui_progress__init_size(&prog, rd.data_size, "Processing events...");
2234
2235         err = reader__process_events(&rd, session, &prog);
2236         if (err)
2237                 goto out_err;
2238         /* do the final flush for ordered samples */
2239         err = ordered_events__flush(oe, OE_FLUSH__FINAL);
2240         if (err)
2241                 goto out_err;
2242         err = auxtrace__flush_events(session, tool);
2243         if (err)
2244                 goto out_err;
2245         err = perf_session__flush_thread_stacks(session);
2246 out_err:
2247         ui_progress__finish();
2248         if (!tool->no_warn)
2249                 perf_session__warn_about_errors(session);
2250         /*
2251          * We may switching perf.data output, make ordered_events
2252          * reusable.
2253          */
2254         ordered_events__reinit(&session->ordered_events);
2255         auxtrace__free_events(session);
2256         session->one_mmap = false;
2257         return err;
2258 }
2259
2260 int perf_session__process_events(struct perf_session *session)
2261 {
2262         if (perf_session__register_idle_thread(session) < 0)
2263                 return -ENOMEM;
2264
2265         if (perf_data__is_pipe(session->data))
2266                 return __perf_session__process_pipe_events(session);
2267
2268         return __perf_session__process_events(session);
2269 }
2270
2271 bool perf_session__has_traces(struct perf_session *session, const char *msg)
2272 {
2273         struct evsel *evsel;
2274
2275         evlist__for_each_entry(session->evlist, evsel) {
2276                 if (evsel->core.attr.type == PERF_TYPE_TRACEPOINT)
2277                         return true;
2278         }
2279
2280         pr_err("No trace sample to read. Did you call 'perf %s'?\n", msg);
2281         return false;
2282 }
2283
2284 int map__set_kallsyms_ref_reloc_sym(struct map *map, const char *symbol_name, u64 addr)
2285 {
2286         char *bracket;
2287         struct ref_reloc_sym *ref;
2288         struct kmap *kmap;
2289
2290         ref = zalloc(sizeof(struct ref_reloc_sym));
2291         if (ref == NULL)
2292                 return -ENOMEM;
2293
2294         ref->name = strdup(symbol_name);
2295         if (ref->name == NULL) {
2296                 free(ref);
2297                 return -ENOMEM;
2298         }
2299
2300         bracket = strchr(ref->name, ']');
2301         if (bracket)
2302                 *bracket = '\0';
2303
2304         ref->addr = addr;
2305
2306         kmap = map__kmap(map);
2307         if (kmap)
2308                 kmap->ref_reloc_sym = ref;
2309
2310         return 0;
2311 }
2312
2313 size_t perf_session__fprintf_dsos(struct perf_session *session, FILE *fp)
2314 {
2315         return machines__fprintf_dsos(&session->machines, fp);
2316 }
2317
2318 size_t perf_session__fprintf_dsos_buildid(struct perf_session *session, FILE *fp,
2319                                           bool (skip)(struct dso *dso, int parm), int parm)
2320 {
2321         return machines__fprintf_dsos_buildid(&session->machines, fp, skip, parm);
2322 }
2323
2324 size_t perf_session__fprintf_nr_events(struct perf_session *session, FILE *fp)
2325 {
2326         size_t ret;
2327         const char *msg = "";
2328
2329         if (perf_header__has_feat(&session->header, HEADER_AUXTRACE))
2330                 msg = " (excludes AUX area (e.g. instruction trace) decoded / synthesized events)";
2331
2332         ret = fprintf(fp, "\nAggregated stats:%s\n", msg);
2333
2334         ret += events_stats__fprintf(&session->evlist->stats, fp);
2335         return ret;
2336 }
2337
2338 size_t perf_session__fprintf(struct perf_session *session, FILE *fp)
2339 {
2340         /*
2341          * FIXME: Here we have to actually print all the machines in this
2342          * session, not just the host...
2343          */
2344         return machine__fprintf(&session->machines.host, fp);
2345 }
2346
2347 struct evsel *perf_session__find_first_evtype(struct perf_session *session,
2348                                               unsigned int type)
2349 {
2350         struct evsel *pos;
2351
2352         evlist__for_each_entry(session->evlist, pos) {
2353                 if (pos->core.attr.type == type)
2354                         return pos;
2355         }
2356         return NULL;
2357 }
2358
2359 int perf_session__cpu_bitmap(struct perf_session *session,
2360                              const char *cpu_list, unsigned long *cpu_bitmap)
2361 {
2362         int i, err = -1;
2363         struct perf_cpu_map *map;
2364         int nr_cpus = min(session->header.env.nr_cpus_online, MAX_NR_CPUS);
2365
2366         for (i = 0; i < PERF_TYPE_MAX; ++i) {
2367                 struct evsel *evsel;
2368
2369                 evsel = perf_session__find_first_evtype(session, i);
2370                 if (!evsel)
2371                         continue;
2372
2373                 if (!(evsel->core.attr.sample_type & PERF_SAMPLE_CPU)) {
2374                         pr_err("File does not contain CPU events. "
2375                                "Remove -C option to proceed.\n");
2376                         return -1;
2377                 }
2378         }
2379
2380         map = perf_cpu_map__new(cpu_list);
2381         if (map == NULL) {
2382                 pr_err("Invalid cpu_list\n");
2383                 return -1;
2384         }
2385
2386         for (i = 0; i < map->nr; i++) {
2387                 int cpu = map->map[i];
2388
2389                 if (cpu >= nr_cpus) {
2390                         pr_err("Requested CPU %d too large. "
2391                                "Consider raising MAX_NR_CPUS\n", cpu);
2392                         goto out_delete_map;
2393                 }
2394
2395                 set_bit(cpu, cpu_bitmap);
2396         }
2397
2398         err = 0;
2399
2400 out_delete_map:
2401         perf_cpu_map__put(map);
2402         return err;
2403 }
2404
2405 void perf_session__fprintf_info(struct perf_session *session, FILE *fp,
2406                                 bool full)
2407 {
2408         if (session == NULL || fp == NULL)
2409                 return;
2410
2411         fprintf(fp, "# ========\n");
2412         perf_header__fprintf_info(session, fp, full);
2413         fprintf(fp, "# ========\n#\n");
2414 }
2415
2416 int perf_event__process_id_index(struct perf_session *session,
2417                                  union perf_event *event)
2418 {
2419         struct evlist *evlist = session->evlist;
2420         struct perf_record_id_index *ie = &event->id_index;
2421         size_t i, nr, max_nr;
2422
2423         max_nr = (ie->header.size - sizeof(struct perf_record_id_index)) /
2424                  sizeof(struct id_index_entry);
2425         nr = ie->nr;
2426         if (nr > max_nr)
2427                 return -EINVAL;
2428
2429         if (dump_trace)
2430                 fprintf(stdout, " nr: %zu\n", nr);
2431
2432         for (i = 0; i < nr; i++) {
2433                 struct id_index_entry *e = &ie->entries[i];
2434                 struct perf_sample_id *sid;
2435
2436                 if (dump_trace) {
2437                         fprintf(stdout, " ... id: %"PRI_lu64, e->id);
2438                         fprintf(stdout, "  idx: %"PRI_lu64, e->idx);
2439                         fprintf(stdout, "  cpu: %"PRI_ld64, e->cpu);
2440                         fprintf(stdout, "  tid: %"PRI_ld64"\n", e->tid);
2441                 }
2442
2443                 sid = perf_evlist__id2sid(evlist, e->id);
2444                 if (!sid)
2445                         return -ENOENT;
2446                 sid->idx = e->idx;
2447                 sid->cpu = e->cpu;
2448                 sid->tid = e->tid;
2449         }
2450         return 0;
2451 }