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 "debug.h"
  16 #include "evlist.h"
  17 #include "evsel.h"
  18 #include "memswap.h"
  19 #include "map.h"
  20 #include "symbol.h"
  21 #include "session.h"
  22 #include "tool.h"
  23 #include "cpumap.h"
  24 #include "perf_regs.h"
  25 #include "asm/bug.h"
  26 #include "auxtrace.h"
  27 #include "thread.h"
  28 #include "thread-stack.h"
  29 #include "sample-raw.h"
  30 #include "stat.h"
  31 #include "util.h"
  32 #include "ui/progress.h"
  33 #include "../perf.h"
  34 #include "arch/common.h"
  35
  36 #ifdef HAVE_ZSTD_SUPPORT
  37 static int perf_session__process_compressed_event(struct perf_session *session,
  38                                                   union perf_event *event, u64 file_offset)
  39 {
  40         void *src;
  41         size_t decomp_size, src_size;
  42         u64 decomp_last_rem = 0;
  43         size_t mmap_len, decomp_len = session->header.env.comp_mmap_len;
  44         struct decomp *decomp, *decomp_last = session->decomp_last;
  45
  46         if (decomp_last) {
  47                 decomp_last_rem = decomp_last->size - decomp_last->head;
  48                 decomp_len += decomp_last_rem;
  49         }
  50
  51         mmap_len = sizeof(struct decomp) + decomp_len;
  52         decomp = mmap(NULL, mmap_len, PROT_READ|PROT_WRITE,
  53                       MAP_ANONYMOUS|MAP_PRIVATE, -1, 0);
  54         if (decomp == MAP_FAILED) {
  55                 pr_err("Couldn't allocate memory for decompression\n");
  56                 return -1;
  57         }
  58
  59         decomp->file_pos = file_offset;
  60         decomp->mmap_len = mmap_len;
  61         decomp->head = 0;
  62
  63         if (decomp_last_rem) {
  64                 memcpy(decomp->data, &(decomp_last->data[decomp_last->head]), decomp_last_rem);
  65                 decomp->size = decomp_last_rem;
  66         }
  67
  68         src = (void *)event + sizeof(struct perf_record_compressed);
  69         src_size = event->pack.header.size - sizeof(struct perf_record_compressed);
  70
  71         decomp_size = zstd_decompress_stream(&(session->zstd_data), src, src_size,
  72                                 &(decomp->data[decomp_last_rem]), decomp_len - decomp_last_rem);
  73         if (!decomp_size) {
  74                 munmap(decomp, mmap_len);
  75                 pr_err("Couldn't decompress data\n");
  76                 return -1;
  77         }
  78
  79         decomp->size += decomp_size;
  80
  81         if (session->decomp == NULL) {
  82                 session->decomp = decomp;
  83                 session->decomp_last = decomp;
  84         } else {
  85                 session->decomp_last->next = decomp;
  86                 session->decomp_last = decomp;
  87         }
  88
  89         pr_debug("decomp (B): %ld to %ld\n", src_size, decomp_size);
  90
  91         return 0;
  92 }
  93 #else /* !HAVE_ZSTD_SUPPORT */
  94 #define perf_session__process_compressed_event perf_session__process_compressed_event_stub
  95 #endif
  96
  97 static int perf_session__deliver_event(struct perf_session *session,
  98                                        union perf_event *event,
  99                                        struct perf_tool *tool,
 100                                        u64 file_offset);
 101
 102 static int perf_session__open(struct perf_session *session)
 103 {
 104         struct perf_data *data = session->data;
 105
 106         if (perf_session__read_header(session) < 0) {
 107                 pr_err("incompatible file format (rerun with -v to learn more)\n");
 108                 return -1;
 109         }
 110
 111         if (perf_data__is_pipe(data))
 112                 return 0;
 113
 114         if (perf_header__has_feat(&session->header, HEADER_STAT))
 115                 return 0;
 116
 117         if (!perf_evlist__valid_sample_type(session->evlist)) {
 118                 pr_err("non matching sample_type\n");
 119                 return -1;
 120         }
 121
 122         if (!perf_evlist__valid_sample_id_all(session->evlist)) {
 123                 pr_err("non matching sample_id_all\n");
 124                 return -1;
 125         }
 126
 127         if (!perf_evlist__valid_read_format(session->evlist)) {
 128                 pr_err("non matching read_format\n");
 129                 return -1;
 130         }
 131
 132         return 0;
 133 }
 134
 135 void perf_session__set_id_hdr_size(struct perf_session *session)
 136 {
 137         u16 id_hdr_size = perf_evlist__id_hdr_size(session->evlist);
 138
 139         machines__set_id_hdr_size(&session->machines, id_hdr_size);
 140 }
 141
 142 int perf_session__create_kernel_maps(struct perf_session *session)
 143 {
 144         int ret = machine__create_kernel_maps(&session->machines.host);
 145
 146         if (ret >= 0)
 147                 ret = machines__create_guest_kernel_maps(&session->machines);
 148         return ret;
 149 }
 150
 151 static void perf_session__destroy_kernel_maps(struct perf_session *session)
 152 {
 153         machines__destroy_kernel_maps(&session->machines);
 154 }
 155
 156 static bool perf_session__has_comm_exec(struct perf_session *session)
 157 {
 158         struct evsel *evsel;
 159
 160         evlist__for_each_entry(session->evlist, evsel) {
 161                 if (evsel->core.attr.comm_exec)
 162                         return true;
 163         }
 164
 165         return false;
 166 }
 167
 168 static void perf_session__set_comm_exec(struct perf_session *session)
 169 {
 170         bool comm_exec = perf_session__has_comm_exec(session);
 171
 172         machines__set_comm_exec(&session->machines, comm_exec);
 173 }
 174
 175 static int ordered_events__deliver_event(struct ordered_events *oe,
 176                                          struct ordered_event *event)
 177 {
 178         struct perf_session *session = container_of(oe, struct perf_session,
 179                                                     ordered_events);
 180
 181         return perf_session__deliver_event(session, event->event,
 182                                            session->tool, event->file_offset);
 183 }
 184
 185 struct perf_session *perf_session__new(struct perf_data *data,
 186                                        bool repipe, struct perf_tool *tool)
 187 {
 188         struct perf_session *session = zalloc(sizeof(*session));
 189
 190         if (!session)
 191                 goto out;
 192
 193         session->repipe = repipe;
 194         session->tool   = tool;
 195         INIT_LIST_HEAD(&session->auxtrace_index);
 196         machines__init(&session->machines);
 197         ordered_events__init(&session->ordered_events,
 198                              ordered_events__deliver_event, NULL);
 199
 200         perf_env__init(&session->header.env);
 201         if (data) {
 202                 if (perf_data__open(data))
 203                         goto out_delete;
 204
 205                 session->data = data;
 206
 207                 if (perf_data__is_read(data)) {
 208                         if (perf_session__open(session) < 0)
 209                                 goto out_delete;
 210
 211                         /*
 212                          * set session attributes that are present in perf.data
 213                          * but not in pipe-mode.
 214                          */
 215                         if (!data->is_pipe) {
 216                                 perf_session__set_id_hdr_size(session);
 217                                 perf_session__set_comm_exec(session);
 218                         }
 219
 220                         perf_evlist__init_trace_event_sample_raw(session->evlist);
 221
 222                         /* Open the directory data. */
 223                         if (data->is_dir && perf_data__open_dir(data))
 224                                 goto out_delete;
 225                 }
 226         } else  {
 227                 session->machines.host.env = &perf_env;
 228         }
 229
 230         session->machines.host.single_address_space =
 231                 perf_env__single_address_space(session->machines.host.env);
 232
 233         if (!data || perf_data__is_write(data)) {
 234                 /*
 235                  * In O_RDONLY mode this will be performed when reading the
 236                  * kernel MMAP event, in perf_event__process_mmap().
 237                  */
 238                 if (perf_session__create_kernel_maps(session) < 0)
 239                         pr_warning("Cannot read kernel map\n");
 240         }
 241
 242         /*
 243          * In pipe-mode, evlist is empty until PERF_RECORD_HEADER_ATTR is
 244          * processed, so perf_evlist__sample_id_all is not meaningful here.
 245          */
 246         if ((!data || !data->is_pipe) && tool && tool->ordering_requires_timestamps &&
 247             tool->ordered_events && !perf_evlist__sample_id_all(session->evlist)) {
 248                 dump_printf("WARNING: No sample_id_all support, falling back to unordered processing\n");
 249                 tool->ordered_events = false;
 250         }
 251
 252         return session;
 253
 254  out_delete:
 255         perf_session__delete(session);
 256  out:
 257         return NULL;
 258 }
 259
 260 static void perf_session__delete_threads(struct perf_session *session)
 261 {
 262         machine__delete_threads(&session->machines.host);
 263 }
 264
 265 static void perf_session__release_decomp_events(struct perf_session *session)
 266 {
 267         struct decomp *next, *decomp;
 268         size_t mmap_len;
 269         next = session->decomp;
 270         do {
 271                 decomp = next;
 272                 if (decomp == NULL)
 273                         break;
 274                 next = decomp->next;
 275                 mmap_len = decomp->mmap_len;
 276                 munmap(decomp, mmap_len);
 277         } while (1);
 278 }
 279
 280 void perf_session__delete(struct perf_session *session)
 281 {
 282         if (session == NULL)
 283                 return;
 284         auxtrace__free(session);
 285         auxtrace_index__free(&session->auxtrace_index);
 286         perf_session__destroy_kernel_maps(session);
 287         perf_session__delete_threads(session);
 288         perf_session__release_decomp_events(session);
 289         perf_env__exit(&session->header.env);
 290         machines__exit(&session->machines);
 291         if (session->data)
 292                 perf_data__close(session->data);
 293         free(session);
 294 }
 295
 296 static int process_event_synth_tracing_data_stub(struct perf_session *session
 297                                                  __maybe_unused,
 298                                                  union perf_event *event
 299                                                  __maybe_unused)
 300 {
 301         dump_printf(": unhandled!\n");
 302         return 0;
 303 }
 304
 305 static int process_event_synth_attr_stub(struct perf_tool *tool __maybe_unused,
 306                                          union perf_event *event __maybe_unused,
 307                                          struct evlist **pevlist
 308                                          __maybe_unused)
 309 {
 310         dump_printf(": unhandled!\n");
 311         return 0;
 312 }
 313
 314 static int process_event_synth_event_update_stub(struct perf_tool *tool __maybe_unused,
 315                                                  union perf_event *event __maybe_unused,
 316                                                  struct evlist **pevlist
 317                                                  __maybe_unused)
 318 {
 319         if (dump_trace)
 320                 perf_event__fprintf_event_update(event, stdout);
 321
 322         dump_printf(": unhandled!\n");
 323         return 0;
 324 }
 325
 326 static int process_event_sample_stub(struct perf_tool *tool __maybe_unused,
 327                                      union perf_event *event __maybe_unused,
 328                                      struct perf_sample *sample __maybe_unused,
 329                                      struct evsel *evsel __maybe_unused,
 330                                      struct machine *machine __maybe_unused)
 331 {
 332         dump_printf(": unhandled!\n");
 333         return 0;
 334 }
 335
 336 static int process_event_stub(struct perf_tool *tool __maybe_unused,
 337                               union perf_event *event __maybe_unused,
 338                               struct perf_sample *sample __maybe_unused,
 339                               struct machine *machine __maybe_unused)
 340 {
 341         dump_printf(": unhandled!\n");
 342         return 0;
 343 }
 344
 345 static int process_finished_round_stub(struct perf_tool *tool __maybe_unused,
 346                                        union perf_event *event __maybe_unused,
 347                                        struct ordered_events *oe __maybe_unused)
 348 {
 349         dump_printf(": unhandled!\n");
 350         return 0;
 351 }
 352
 353 static int process_finished_round(struct perf_tool *tool,
 354                                   union perf_event *event,
 355                                   struct ordered_events *oe);
 356
 357 static int skipn(int fd, off_t n)
 358 {
 359         char buf[4096];
 360         ssize_t ret;
 361
 362         while (n > 0) {
 363                 ret = read(fd, buf, min(n, (off_t)sizeof(buf)));
 364                 if (ret <= 0)
 365                         return ret;
 366                 n -= ret;
 367         }
 368
 369         return 0;
 370 }
 371
 372 static s64 process_event_auxtrace_stub(struct perf_session *session __maybe_unused,
 373                                        union perf_event *event)
 374 {
 375         dump_printf(": unhandled!\n");
 376         if (perf_data__is_pipe(session->data))
 377                 skipn(perf_data__fd(session->data), event->auxtrace.size);
 378         return event->auxtrace.size;
 379 }
 380
 381 static int process_event_op2_stub(struct perf_session *session __maybe_unused,
 382                                   union perf_event *event __maybe_unused)
 383 {
 384         dump_printf(": unhandled!\n");
 385         return 0;
 386 }
 387
 388
 389 static
 390 int process_event_thread_map_stub(struct perf_session *session __maybe_unused,
 391                                   union perf_event *event __maybe_unused)
 392 {
 393         if (dump_trace)
 394                 perf_event__fprintf_thread_map(event, stdout);
 395
 396         dump_printf(": unhandled!\n");
 397         return 0;
 398 }
 399
 400 static
 401 int process_event_cpu_map_stub(struct perf_session *session __maybe_unused,
 402                                union perf_event *event __maybe_unused)
 403 {
 404         if (dump_trace)
 405                 perf_event__fprintf_cpu_map(event, stdout);
 406
 407         dump_printf(": unhandled!\n");
 408         return 0;
 409 }
 410
 411 static
 412 int process_event_stat_config_stub(struct perf_session *session __maybe_unused,
 413                                    union perf_event *event __maybe_unused)
 414 {
 415         if (dump_trace)
 416                 perf_event__fprintf_stat_config(event, stdout);
 417
 418         dump_printf(": unhandled!\n");
 419         return 0;
 420 }
 421
 422 static int process_stat_stub(struct perf_session *perf_session __maybe_unused,
 423                              union perf_event *event)
 424 {
 425         if (dump_trace)
 426                 perf_event__fprintf_stat(event, stdout);
 427
 428         dump_printf(": unhandled!\n");
 429         return 0;
 430 }
 431
 432 static int process_stat_round_stub(struct perf_session *perf_session __maybe_unused,
 433                                    union perf_event *event)
 434 {
 435         if (dump_trace)
 436                 perf_event__fprintf_stat_round(event, stdout);
 437
 438         dump_printf(": unhandled!\n");
 439         return 0;
 440 }
 441
 442 static int perf_session__process_compressed_event_stub(struct perf_session *session __maybe_unused,
 443                                                        union perf_event *event __maybe_unused,
 444                                                        u64 file_offset __maybe_unused)
 445 {
 446        dump_printf(": unhandled!\n");
 447        return 0;
 448 }
 449
 450 void perf_tool__fill_defaults(struct perf_tool *tool)
 451 {
 452         if (tool->sample == NULL)
 453                 tool->sample = process_event_sample_stub;
 454         if (tool->mmap == NULL)
 455                 tool->mmap = process_event_stub;
 456         if (tool->mmap2 == NULL)
 457                 tool->mmap2 = process_event_stub;
 458         if (tool->comm == NULL)
 459                 tool->comm = process_event_stub;
 460         if (tool->namespaces == NULL)
 461                 tool->namespaces = process_event_stub;
 462         if (tool->fork == NULL)
 463                 tool->fork = process_event_stub;
 464         if (tool->exit == NULL)
 465                 tool->exit = process_event_stub;
 466         if (tool->lost == NULL)
 467                 tool->lost = perf_event__process_lost;
 468         if (tool->lost_samples == NULL)
 469                 tool->lost_samples = perf_event__process_lost_samples;
 470         if (tool->aux == NULL)
 471                 tool->aux = perf_event__process_aux;
 472         if (tool->itrace_start == NULL)
 473                 tool->itrace_start = perf_event__process_itrace_start;
 474         if (tool->context_switch == NULL)
 475                 tool->context_switch = perf_event__process_switch;
 476         if (tool->ksymbol == NULL)
 477                 tool->ksymbol = perf_event__process_ksymbol;
 478         if (tool->bpf == NULL)
 479                 tool->bpf = perf_event__process_bpf;
 480         if (tool->read == NULL)
 481                 tool->read = process_event_sample_stub;
 482         if (tool->throttle == NULL)
 483                 tool->throttle = process_event_stub;
 484         if (tool->unthrottle == NULL)
 485                 tool->unthrottle = process_event_stub;
 486         if (tool->attr == NULL)
 487                 tool->attr = process_event_synth_attr_stub;
 488         if (tool->event_update == NULL)
 489                 tool->event_update = process_event_synth_event_update_stub;
 490         if (tool->tracing_data == NULL)
 491                 tool->tracing_data = process_event_synth_tracing_data_stub;
 492         if (tool->build_id == NULL)
 493                 tool->build_id = process_event_op2_stub;
 494         if (tool->finished_round == NULL) {
 495                 if (tool->ordered_events)
 496                         tool->finished_round = process_finished_round;
 497                 else
 498                         tool->finished_round = process_finished_round_stub;
 499         }
 500         if (tool->id_index == NULL)
 501                 tool->id_index = process_event_op2_stub;
 502         if (tool->auxtrace_info == NULL)
 503                 tool->auxtrace_info = process_event_op2_stub;
 504         if (tool->auxtrace == NULL)
 505                 tool->auxtrace = process_event_auxtrace_stub;
 506         if (tool->auxtrace_error == NULL)
 507                 tool->auxtrace_error = process_event_op2_stub;
 508         if (tool->thread_map == NULL)
 509                 tool->thread_map = process_event_thread_map_stub;
 510         if (tool->cpu_map == NULL)
 511                 tool->cpu_map = process_event_cpu_map_stub;
 512         if (tool->stat_config == NULL)
 513                 tool->stat_config = process_event_stat_config_stub;
 514         if (tool->stat == NULL)
 515                 tool->stat = process_stat_stub;
 516         if (tool->stat_round == NULL)
 517                 tool->stat_round = process_stat_round_stub;
 518         if (tool->time_conv == NULL)
 519                 tool->time_conv = process_event_op2_stub;
 520         if (tool->feature == NULL)
 521                 tool->feature = process_event_op2_stub;
 522         if (tool->compressed == NULL)
 523                 tool->compressed = perf_session__process_compressed_event;
 524 }
 525
 526 static void swap_sample_id_all(union perf_event *event, void *data)
 527 {
 528         void *end = (void *) event + event->header.size;
 529         int size = end - data;
 530
 531         BUG_ON(size % sizeof(u64));
 532         mem_bswap_64(data, size);
 533 }
 534
 535 static void perf_event__all64_swap(union perf_event *event,
 536                                    bool sample_id_all __maybe_unused)
 537 {
 538         struct perf_event_header *hdr = &event->header;
 539         mem_bswap_64(hdr + 1, event->header.size - sizeof(*hdr));
 540 }
 541
 542 static void perf_event__comm_swap(union perf_event *event, bool sample_id_all)
 543 {
 544         event->comm.pid = bswap_32(event->comm.pid);
 545         event->comm.tid = bswap_32(event->comm.tid);
 546
 547         if (sample_id_all) {
 548                 void *data = &event->comm.comm;
 549
 550                 data += PERF_ALIGN(strlen(data) + 1, sizeof(u64));
 551                 swap_sample_id_all(event, data);
 552         }
 553 }
 554
 555 static void perf_event__mmap_swap(union perf_event *event,
 556                                   bool sample_id_all)
 557 {
 558         event->mmap.pid   = bswap_32(event->mmap.pid);
 559         event->mmap.tid   = bswap_32(event->mmap.tid);
 560         event->mmap.start = bswap_64(event->mmap.start);
 561         event->mmap.len   = bswap_64(event->mmap.len);
 562         event->mmap.pgoff = bswap_64(event->mmap.pgoff);
 563
 564         if (sample_id_all) {
 565                 void *data = &event->mmap.filename;
 566
 567                 data += PERF_ALIGN(strlen(data) + 1, sizeof(u64));
 568                 swap_sample_id_all(event, data);
 569         }
 570 }
 571
 572 static void perf_event__mmap2_swap(union perf_event *event,
 573                                   bool sample_id_all)
 574 {
 575         event->mmap2.pid   = bswap_32(event->mmap2.pid);
 576         event->mmap2.tid   = bswap_32(event->mmap2.tid);
 577         event->mmap2.start = bswap_64(event->mmap2.start);
 578         event->mmap2.len   = bswap_64(event->mmap2.len);
 579         event->mmap2.pgoff = bswap_64(event->mmap2.pgoff);
 580         event->mmap2.maj   = bswap_32(event->mmap2.maj);
 581         event->mmap2.min   = bswap_32(event->mmap2.min);
 582         event->mmap2.ino   = bswap_64(event->mmap2.ino);
 583
 584         if (sample_id_all) {
 585                 void *data = &event->mmap2.filename;
 586
 587                 data += PERF_ALIGN(strlen(data) + 1, sizeof(u64));
 588                 swap_sample_id_all(event, data);
 589         }
 590 }
 591 static void perf_event__task_swap(union perf_event *event, bool sample_id_all)
 592 {
 593         event->fork.pid  = bswap_32(event->fork.pid);
 594         event->fork.tid  = bswap_32(event->fork.tid);
 595         event->fork.ppid = bswap_32(event->fork.ppid);
 596         event->fork.ptid = bswap_32(event->fork.ptid);
 597         event->fork.time = bswap_64(event->fork.time);
 598
 599         if (sample_id_all)
 600                 swap_sample_id_all(event, &event->fork + 1);
 601 }
 602
 603 static void perf_event__read_swap(union perf_event *event, bool sample_id_all)
 604 {
 605         event->read.pid          = bswap_32(event->read.pid);
 606         event->read.tid          = bswap_32(event->read.tid);
 607         event->read.value        = bswap_64(event->read.value);
 608         event->read.time_enabled = bswap_64(event->read.time_enabled);
 609         event->read.time_running = bswap_64(event->read.time_running);
 610         event->read.id           = bswap_64(event->read.id);
 611
 612         if (sample_id_all)
 613                 swap_sample_id_all(event, &event->read + 1);
 614 }
 615
 616 static void perf_event__aux_swap(union perf_event *event, bool sample_id_all)
 617 {
 618         event->aux.aux_offset = bswap_64(event->aux.aux_offset);
 619         event->aux.aux_size   = bswap_64(event->aux.aux_size);
 620         event->aux.flags      = bswap_64(event->aux.flags);
 621
 622         if (sample_id_all)
 623                 swap_sample_id_all(event, &event->aux + 1);
 624 }
 625
 626 static void perf_event__itrace_start_swap(union perf_event *event,
 627                                           bool sample_id_all)
 628 {
 629         event->itrace_start.pid  = bswap_32(event->itrace_start.pid);
 630         event->itrace_start.tid  = bswap_32(event->itrace_start.tid);
 631
 632         if (sample_id_all)
 633                 swap_sample_id_all(event, &event->itrace_start + 1);
 634 }
 635
 636 static void perf_event__switch_swap(union perf_event *event, bool sample_id_all)
 637 {
 638         if (event->header.type == PERF_RECORD_SWITCH_CPU_WIDE) {
 639                 event->context_switch.next_prev_pid =
 640                                 bswap_32(event->context_switch.next_prev_pid);
 641                 event->context_switch.next_prev_tid =
 642                                 bswap_32(event->context_switch.next_prev_tid);
 643         }
 644
 645         if (sample_id_all)
 646                 swap_sample_id_all(event, &event->context_switch + 1);
 647 }
 648
 649 static void perf_event__throttle_swap(union perf_event *event,
 650                                       bool sample_id_all)
 651 {
 652         event->throttle.time      = bswap_64(event->throttle.time);
 653         event->throttle.id        = bswap_64(event->throttle.id);
 654         event->throttle.stream_id = bswap_64(event->throttle.stream_id);
 655
 656         if (sample_id_all)
 657                 swap_sample_id_all(event, &event->throttle + 1);
 658 }
 659
 660 static void perf_event__namespaces_swap(union perf_event *event,
 661                                         bool sample_id_all)
 662 {
 663         u64 i;
 664
 665         event->namespaces.pid           = bswap_32(event->namespaces.pid);
 666         event->namespaces.tid           = bswap_32(event->namespaces.tid);
 667         event->namespaces.nr_namespaces = bswap_64(event->namespaces.nr_namespaces);
 668
 669         for (i = 0; i < event->namespaces.nr_namespaces; i++) {
 670                 struct perf_ns_link_info *ns = &event->namespaces.link_info[i];
 671
 672                 ns->dev = bswap_64(ns->dev);
 673                 ns->ino = bswap_64(ns->ino);
 674         }
 675
 676         if (sample_id_all)
 677                 swap_sample_id_all(event, &event->namespaces.link_info[i]);
 678 }
 679
 680 static u8 revbyte(u8 b)
 681 {
 682         int rev = (b >> 4) | ((b & 0xf) << 4);
 683         rev = ((rev & 0xcc) >> 2) | ((rev & 0x33) << 2);
 684         rev = ((rev & 0xaa) >> 1) | ((rev & 0x55) << 1);
 685         return (u8) rev;
 686 }
 687
 688 /*
 689  * XXX this is hack in attempt to carry flags bitfield
 690  * through endian village. ABI says:
 691  *
 692  * Bit-fields are allocated from right to left (least to most significant)
 693  * on little-endian implementations and from left to right (most to least
 694  * significant) on big-endian implementations.
 695  *
 696  * The above seems to be byte specific, so we need to reverse each
 697  * byte of the bitfield. 'Internet' also says this might be implementation
 698  * specific and we probably need proper fix and carry perf_event_attr
 699  * bitfield flags in separate data file FEAT_ section. Thought this seems
 700  * to work for now.
 701  */
 702 static void swap_bitfield(u8 *p, unsigned len)
 703 {
 704         unsigned i;
 705
 706         for (i = 0; i < len; i++) {
 707                 *p = revbyte(*p);
 708                 p++;
 709         }
 710 }
 711
 712 /* exported for swapping attributes in file header */
 713 void perf_event__attr_swap(struct perf_event_attr *attr)
 714 {
 715         attr->type              = bswap_32(attr->type);
 716         attr->size              = bswap_32(attr->size);
 717
 718 #define bswap_safe(f, n)                                        \
 719         (attr->size > (offsetof(struct perf_event_attr, f) +    \
 720                        sizeof(attr->f) * (n)))
 721 #define bswap_field(f, sz)                      \
 722 do {                                            \
 723         if (bswap_safe(f, 0))                   \
 724                 attr->f = bswap_##sz(attr->f);  \
 725 } while(0)
 726 #define bswap_field_16(f) bswap_field(f, 16)
 727 #define bswap_field_32(f) bswap_field(f, 32)
 728 #define bswap_field_64(f) bswap_field(f, 64)
 729
 730         bswap_field_64(config);
 731         bswap_field_64(sample_period);
 732         bswap_field_64(sample_type);
 733         bswap_field_64(read_format);
 734         bswap_field_32(wakeup_events);
 735         bswap_field_32(bp_type);
 736         bswap_field_64(bp_addr);
 737         bswap_field_64(bp_len);
 738         bswap_field_64(branch_sample_type);
 739         bswap_field_64(sample_regs_user);
 740         bswap_field_32(sample_stack_user);
 741         bswap_field_32(aux_watermark);
 742         bswap_field_16(sample_max_stack);
 743
 744         /*
 745          * After read_format are bitfields. Check read_format because
 746          * we are unable to use offsetof on bitfield.
 747          */
 748         if (bswap_safe(read_format, 1))
 749                 swap_bitfield((u8 *) (&attr->read_format + 1),
 750                               sizeof(u64));
 751 #undef bswap_field_64
 752 #undef bswap_field_32
 753 #undef bswap_field
 754 #undef bswap_safe
 755 }
 756
 757 static void perf_event__hdr_attr_swap(union perf_event *event,
 758                                       bool sample_id_all __maybe_unused)
 759 {
 760         size_t size;
 761
 762         perf_event__attr_swap(&event->attr.attr);
 763
 764         size = event->header.size;
 765         size -= (void *)&event->attr.id - (void *)event;
 766         mem_bswap_64(event->attr.id, size);
 767 }
 768
 769 static void perf_event__event_update_swap(union perf_event *event,
 770                                           bool sample_id_all __maybe_unused)
 771 {
 772         event->event_update.type = bswap_64(event->event_update.type);
 773         event->event_update.id   = bswap_64(event->event_update.id);
 774 }
 775
 776 static void perf_event__event_type_swap(union perf_event *event,
 777                                         bool sample_id_all __maybe_unused)
 778 {
 779         event->event_type.event_type.event_id =
 780                 bswap_64(event->event_type.event_type.event_id);
 781 }
 782
 783 static void perf_event__tracing_data_swap(union perf_event *event,
 784                                           bool sample_id_all __maybe_unused)
 785 {
 786         event->tracing_data.size = bswap_32(event->tracing_data.size);
 787 }
 788
 789 static void perf_event__auxtrace_info_swap(union perf_event *event,
 790                                            bool sample_id_all __maybe_unused)
 791 {
 792         size_t size;
 793
 794         event->auxtrace_info.type = bswap_32(event->auxtrace_info.type);
 795
 796         size = event->header.size;
 797         size -= (void *)&event->auxtrace_info.priv - (void *)event;
 798         mem_bswap_64(event->auxtrace_info.priv, size);
 799 }
 800
 801 static void perf_event__auxtrace_swap(union perf_event *event,
 802                                       bool sample_id_all __maybe_unused)
 803 {
 804         event->auxtrace.size      = bswap_64(event->auxtrace.size);
 805         event->auxtrace.offset    = bswap_64(event->auxtrace.offset);
 806         event->auxtrace.reference = bswap_64(event->auxtrace.reference);
 807         event->auxtrace.idx       = bswap_32(event->auxtrace.idx);
 808         event->auxtrace.tid       = bswap_32(event->auxtrace.tid);
 809         event->auxtrace.cpu       = bswap_32(event->auxtrace.cpu);
 810 }
 811
 812 static void perf_event__auxtrace_error_swap(union perf_event *event,
 813                                             bool sample_id_all __maybe_unused)
 814 {
 815         event->auxtrace_error.type = bswap_32(event->auxtrace_error.type);
 816         event->auxtrace_error.code = bswap_32(event->auxtrace_error.code);
 817         event->auxtrace_error.cpu  = bswap_32(event->auxtrace_error.cpu);
 818         event->auxtrace_error.pid  = bswap_32(event->auxtrace_error.pid);
 819         event->auxtrace_error.tid  = bswap_32(event->auxtrace_error.tid);
 820         event->auxtrace_error.fmt  = bswap_32(event->auxtrace_error.fmt);
 821         event->auxtrace_error.ip   = bswap_64(event->auxtrace_error.ip);
 822         if (event->auxtrace_error.fmt)
 823                 event->auxtrace_error.time = bswap_64(event->auxtrace_error.time);
 824 }
 825
 826 static void perf_event__thread_map_swap(union perf_event *event,
 827                                         bool sample_id_all __maybe_unused)
 828 {
 829         unsigned i;
 830
 831         event->thread_map.nr = bswap_64(event->thread_map.nr);
 832
 833         for (i = 0; i < event->thread_map.nr; i++)
 834                 event->thread_map.entries[i].pid = bswap_64(event->thread_map.entries[i].pid);
 835 }
 836
 837 static void perf_event__cpu_map_swap(union perf_event *event,
 838                                      bool sample_id_all __maybe_unused)
 839 {
 840         struct perf_record_cpu_map_data *data = &event->cpu_map.data;
 841         struct cpu_map_entries *cpus;
 842         struct perf_record_record_cpu_map *mask;
 843         unsigned i;
 844
 845         data->type = bswap_64(data->type);
 846
 847         switch (data->type) {
 848         case PERF_CPU_MAP__CPUS:
 849                 cpus = (struct cpu_map_entries *)data->data;
 850
 851                 cpus->nr = bswap_16(cpus->nr);
 852
 853                 for (i = 0; i < cpus->nr; i++)
 854                         cpus->cpu[i] = bswap_16(cpus->cpu[i]);
 855                 break;
 856         case PERF_CPU_MAP__MASK:
 857                 mask = (struct perf_record_record_cpu_map *)data->data;
 858
 859                 mask->nr = bswap_16(mask->nr);
 860                 mask->long_size = bswap_16(mask->long_size);
 861
 862                 switch (mask->long_size) {
 863                 case 4: mem_bswap_32(&mask->mask, mask->nr); break;
 864                 case 8: mem_bswap_64(&mask->mask, mask->nr); break;
 865                 default:
 866                         pr_err("cpu_map swap: unsupported long size\n");
 867                 }
 868         default:
 869                 break;
 870         }
 871 }
 872
 873 static void perf_event__stat_config_swap(union perf_event *event,
 874                                          bool sample_id_all __maybe_unused)
 875 {
 876         u64 size;
 877
 878         size  = event->stat_config.nr * sizeof(event->stat_config.data[0]);
 879         size += 1; /* nr item itself */
 880         mem_bswap_64(&event->stat_config.nr, size);
 881 }
 882
 883 static void perf_event__stat_swap(union perf_event *event,
 884                                   bool sample_id_all __maybe_unused)
 885 {
 886         event->stat.id     = bswap_64(event->stat.id);
 887         event->stat.thread = bswap_32(event->stat.thread);
 888         event->stat.cpu    = bswap_32(event->stat.cpu);
 889         event->stat.val    = bswap_64(event->stat.val);
 890         event->stat.ena    = bswap_64(event->stat.ena);
 891         event->stat.run    = bswap_64(event->stat.run);
 892 }
 893
 894 static void perf_event__stat_round_swap(union perf_event *event,
 895                                         bool sample_id_all __maybe_unused)
 896 {
 897         event->stat_round.type = bswap_64(event->stat_round.type);
 898         event->stat_round.time = bswap_64(event->stat_round.time);
 899 }
 900
 901 typedef void (*perf_event__swap_op)(union perf_event *event,
 902                                     bool sample_id_all);
 903
 904 static perf_event__swap_op perf_event__swap_ops[] = {
 905         [PERF_RECORD_MMAP]                = perf_event__mmap_swap,
 906         [PERF_RECORD_MMAP2]               = perf_event__mmap2_swap,
 907         [PERF_RECORD_COMM]                = perf_event__comm_swap,
 908         [PERF_RECORD_FORK]                = perf_event__task_swap,
 909         [PERF_RECORD_EXIT]                = perf_event__task_swap,
 910         [PERF_RECORD_LOST]                = perf_event__all64_swap,
 911         [PERF_RECORD_READ]                = perf_event__read_swap,
 912         [PERF_RECORD_THROTTLE]            = perf_event__throttle_swap,
 913         [PERF_RECORD_UNTHROTTLE]          = perf_event__throttle_swap,
 914         [PERF_RECORD_SAMPLE]              = perf_event__all64_swap,
 915         [PERF_RECORD_AUX]                 = perf_event__aux_swap,
 916         [PERF_RECORD_ITRACE_START]        = perf_event__itrace_start_swap,
 917         [PERF_RECORD_LOST_SAMPLES]        = perf_event__all64_swap,
 918         [PERF_RECORD_SWITCH]              = perf_event__switch_swap,
 919         [PERF_RECORD_SWITCH_CPU_WIDE]     = perf_event__switch_swap,
 920         [PERF_RECORD_NAMESPACES]          = perf_event__namespaces_swap,
 921         [PERF_RECORD_HEADER_ATTR]         = perf_event__hdr_attr_swap,
 922         [PERF_RECORD_HEADER_EVENT_TYPE]   = perf_event__event_type_swap,
 923         [PERF_RECORD_HEADER_TRACING_DATA] = perf_event__tracing_data_swap,
 924         [PERF_RECORD_HEADER_BUILD_ID]     = NULL,
 925         [PERF_RECORD_ID_INDEX]            = perf_event__all64_swap,
 926         [PERF_RECORD_AUXTRACE_INFO]       = perf_event__auxtrace_info_swap,
 927         [PERF_RECORD_AUXTRACE]            = perf_event__auxtrace_swap,
 928         [PERF_RECORD_AUXTRACE_ERROR]      = perf_event__auxtrace_error_swap,
 929         [PERF_RECORD_THREAD_MAP]          = perf_event__thread_map_swap,
 930         [PERF_RECORD_CPU_MAP]             = perf_event__cpu_map_swap,
 931         [PERF_RECORD_STAT_CONFIG]         = perf_event__stat_config_swap,
 932         [PERF_RECORD_STAT]                = perf_event__stat_swap,
 933         [PERF_RECORD_STAT_ROUND]          = perf_event__stat_round_swap,
 934         [PERF_RECORD_EVENT_UPDATE]        = perf_event__event_update_swap,
 935         [PERF_RECORD_TIME_CONV]           = perf_event__all64_swap,
 936         [PERF_RECORD_HEADER_MAX]          = NULL,
 937 };
 938
 939 /*
 940  * When perf record finishes a pass on every buffers, it records this pseudo
 941  * event.
 942  * We record the max timestamp t found in the pass n.
 943  * Assuming these timestamps are monotonic across cpus, we know that if
 944  * a buffer still has events with timestamps below t, they will be all
 945  * available and then read in the pass n + 1.
 946  * Hence when we start to read the pass n + 2, we can safely flush every
 947  * events with timestamps below t.
 948  *
 949  *    ============ PASS n =================
 950  *       CPU 0         |   CPU 1
 951  *                     |
 952  *    cnt1 timestamps  |   cnt2 timestamps
 953  *          1          |         2
 954  *          2          |         3
 955  *          -          |         4  <--- max recorded
 956  *
 957  *    ============ PASS n + 1 ==============
 958  *       CPU 0         |   CPU 1
 959  *                     |
 960  *    cnt1 timestamps  |   cnt2 timestamps
 961  *          3          |         5
 962  *          4          |         6
 963  *          5          |         7 <---- max recorded
 964  *
 965  *      Flush every events below timestamp 4
 966  *
 967  *    ============ PASS n + 2 ==============
 968  *       CPU 0         |   CPU 1
 969  *                     |
 970  *    cnt1 timestamps  |   cnt2 timestamps
 971  *          6          |         8
 972  *          7          |         9
 973  *          -          |         10
 974  *
 975  *      Flush every events below timestamp 7
 976  *      etc...
 977  */
 978 static int process_finished_round(struct perf_tool *tool __maybe_unused,
 979                                   union perf_event *event __maybe_unused,
 980                                   struct ordered_events *oe)
 981 {
 982         if (dump_trace)
 983                 fprintf(stdout, "\n");
 984         return ordered_events__flush(oe, OE_FLUSH__ROUND);
 985 }
 986
 987 int perf_session__queue_event(struct perf_session *s, union perf_event *event,
 988                               u64 timestamp, u64 file_offset)
 989 {
 990         return ordered_events__queue(&s->ordered_events, event, timestamp, file_offset);
 991 }
 992
 993 static void callchain__lbr_callstack_printf(struct perf_sample *sample)
 994 {
 995         struct ip_callchain *callchain = sample->callchain;
 996         struct branch_stack *lbr_stack = sample->branch_stack;
 997         u64 kernel_callchain_nr = callchain->nr;
 998         unsigned int i;
 999
1000         for (i = 0; i < kernel_callchain_nr; i++) {
1001                 if (callchain->ips[i] == PERF_CONTEXT_USER)
1002                         break;
1003         }
1004
1005         if ((i != kernel_callchain_nr) && lbr_stack->nr) {
1006                 u64 total_nr;
1007                 /*
1008                  * LBR callstack can only get user call chain,
1009                  * i is kernel call chain number,
1010                  * 1 is PERF_CONTEXT_USER.
1011                  *
1012                  * The user call chain is stored in LBR registers.
1013                  * LBR are pair registers. The caller is stored
1014                  * in "from" register, while the callee is stored
1015                  * in "to" register.
1016                  * For example, there is a call stack
1017                  * "A"->"B"->"C"->"D".
1018                  * The LBR registers will recorde like
1019                  * "C"->"D", "B"->"C", "A"->"B".
1020                  * So only the first "to" register and all "from"
1021                  * registers are needed to construct the whole stack.
1022                  */
1023                 total_nr = i + 1 + lbr_stack->nr + 1;
1024                 kernel_callchain_nr = i + 1;
1025
1026                 printf("... LBR call chain: nr:%" PRIu64 "\n", total_nr);
1027
1028                 for (i = 0; i < kernel_callchain_nr; i++)
1029                         printf("..... %2d: %016" PRIx64 "\n",
1030                                i, callchain->ips[i]);
1031
1032                 printf("..... %2d: %016" PRIx64 "\n",
1033                        (int)(kernel_callchain_nr), lbr_stack->entries[0].to);
1034                 for (i = 0; i < lbr_stack->nr; i++)
1035                         printf("..... %2d: %016" PRIx64 "\n",
1036                                (int)(i + kernel_callchain_nr + 1), lbr_stack->entries[i].from);
1037         }
1038 }
1039
1040 static void callchain__printf(struct evsel *evsel,
1041                               struct perf_sample *sample)
1042 {
1043         unsigned int i;
1044         struct ip_callchain *callchain = sample->callchain;
1045
1046         if (perf_evsel__has_branch_callstack(evsel))
1047                 callchain__lbr_callstack_printf(sample);
1048
1049         printf("... FP chain: nr:%" PRIu64 "\n", callchain->nr);
1050
1051         for (i = 0; i < callchain->nr; i++)
1052                 printf("..... %2d: %016" PRIx64 "\n",
1053                        i, callchain->ips[i]);
1054 }
1055
1056 static void branch_stack__printf(struct perf_sample *sample, bool callstack)
1057 {
1058         uint64_t i;
1059
1060         printf("%s: nr:%" PRIu64 "\n",
1061                 !callstack ? "... branch stack" : "... branch callstack",
1062                 sample->branch_stack->nr);
1063
1064         for (i = 0; i < sample->branch_stack->nr; i++) {
1065                 struct branch_entry *e = &sample->branch_stack->entries[i];
1066
1067                 if (!callstack) {
1068                         printf("..... %2"PRIu64": %016" PRIx64 " -> %016" PRIx64 " %hu cycles %s%s%s%s %x\n",
1069                                 i, e->from, e->to,
1070                                 (unsigned short)e->flags.cycles,
1071                                 e->flags.mispred ? "M" : " ",
1072                                 e->flags.predicted ? "P" : " ",
1073                                 e->flags.abort ? "A" : " ",
1074                                 e->flags.in_tx ? "T" : " ",
1075                                 (unsigned)e->flags.reserved);
1076                 } else {
1077                         printf("..... %2"PRIu64": %016" PRIx64 "\n",
1078                                 i, i > 0 ? e->from : e->to);
1079                 }
1080         }
1081 }
1082
1083 static void regs_dump__printf(u64 mask, u64 *regs)
1084 {
1085         unsigned rid, i = 0;
1086
1087         for_each_set_bit(rid, (unsigned long *) &mask, sizeof(mask) * 8) {
1088                 u64 val = regs[i++];
1089
1090                 printf(".... %-5s 0x%" PRIx64 "\n",
1091                        perf_reg_name(rid), val);
1092         }
1093 }
1094
1095 static const char *regs_abi[] = {
1096         [PERF_SAMPLE_REGS_ABI_NONE] = "none",
1097         [PERF_SAMPLE_REGS_ABI_32] = "32-bit",
1098         [PERF_SAMPLE_REGS_ABI_64] = "64-bit",
1099 };
1100
1101 static inline const char *regs_dump_abi(struct regs_dump *d)
1102 {
1103         if (d->abi > PERF_SAMPLE_REGS_ABI_64)
1104                 return "unknown";
1105
1106         return regs_abi[d->abi];
1107 }
1108
1109 static void regs__printf(const char *type, struct regs_dump *regs)
1110 {
1111         u64 mask = regs->mask;
1112
1113         printf("... %s regs: mask 0x%" PRIx64 " ABI %s\n",
1114                type,
1115                mask,
1116                regs_dump_abi(regs));
1117
1118         regs_dump__printf(mask, regs->regs);
1119 }
1120
1121 static void regs_user__printf(struct perf_sample *sample)
1122 {
1123         struct regs_dump *user_regs = &sample->user_regs;
1124
1125         if (user_regs->regs)
1126                 regs__printf("user", user_regs);
1127 }
1128
1129 static void regs_intr__printf(struct perf_sample *sample)
1130 {
1131         struct regs_dump *intr_regs = &sample->intr_regs;
1132
1133         if (intr_regs->regs)
1134                 regs__printf("intr", intr_regs);
1135 }
1136
1137 static void stack_user__printf(struct stack_dump *dump)
1138 {
1139         printf("... ustack: size %" PRIu64 ", offset 0x%x\n",
1140                dump->size, dump->offset);
1141 }
1142
1143 static void perf_evlist__print_tstamp(struct evlist *evlist,
1144                                        union perf_event *event,
1145                                        struct perf_sample *sample)
1146 {
1147         u64 sample_type = __perf_evlist__combined_sample_type(evlist);
1148
1149         if (event->header.type != PERF_RECORD_SAMPLE &&
1150             !perf_evlist__sample_id_all(evlist)) {
1151                 fputs("-1 -1 ", stdout);
1152                 return;
1153         }
1154
1155         if ((sample_type & PERF_SAMPLE_CPU))
1156                 printf("%u ", sample->cpu);
1157
1158         if (sample_type & PERF_SAMPLE_TIME)
1159                 printf("%" PRIu64 " ", sample->time);
1160 }
1161
1162 static void sample_read__printf(struct perf_sample *sample, u64 read_format)
1163 {
1164         printf("... sample_read:\n");
1165
1166         if (read_format & PERF_FORMAT_TOTAL_TIME_ENABLED)
1167                 printf("...... time enabled %016" PRIx64 "\n",
1168                        sample->read.time_enabled);
1169
1170         if (read_format & PERF_FORMAT_TOTAL_TIME_RUNNING)
1171                 printf("...... time running %016" PRIx64 "\n",
1172                        sample->read.time_running);
1173
1174         if (read_format & PERF_FORMAT_GROUP) {
1175                 u64 i;
1176
1177                 printf(".... group nr %" PRIu64 "\n", sample->read.group.nr);
1178
1179                 for (i = 0; i < sample->read.group.nr; i++) {
1180                         struct sample_read_value *value;
1181
1182                         value = &sample->read.group.values[i];
1183                         printf("..... id %016" PRIx64
1184                                ", value %016" PRIx64 "\n",
1185                                value->id, value->value);
1186                 }
1187         } else
1188                 printf("..... id %016" PRIx64 ", value %016" PRIx64 "\n",
1189                         sample->read.one.id, sample->read.one.value);
1190 }
1191
1192 static void dump_event(struct evlist *evlist, union perf_event *event,
1193                        u64 file_offset, struct perf_sample *sample)
1194 {
1195         if (!dump_trace)
1196                 return;
1197
1198         printf("\n%#" PRIx64 " [%#x]: event: %d\n",
1199                file_offset, event->header.size, event->header.type);
1200
1201         trace_event(event);
1202         if (event->header.type == PERF_RECORD_SAMPLE && evlist->trace_event_sample_raw)
1203                 evlist->trace_event_sample_raw(evlist, event, sample);
1204
1205         if (sample)
1206                 perf_evlist__print_tstamp(evlist, event, sample);
1207
1208         printf("%#" PRIx64 " [%#x]: PERF_RECORD_%s", file_offset,
1209                event->header.size, perf_event__name(event->header.type));
1210 }
1211
1212 static void dump_sample(struct evsel *evsel, union perf_event *event,
1213                         struct perf_sample *sample)
1214 {
1215         u64 sample_type;
1216
1217         if (!dump_trace)
1218                 return;
1219
1220         printf("(IP, 0x%x): %d/%d: %#" PRIx64 " period: %" PRIu64 " addr: %#" PRIx64 "\n",
1221                event->header.misc, sample->pid, sample->tid, sample->ip,
1222                sample->period, sample->addr);
1223
1224         sample_type = evsel->core.attr.sample_type;
1225
1226         if (evsel__has_callchain(evsel))
1227                 callchain__printf(evsel, sample);
1228
1229         if (sample_type & PERF_SAMPLE_BRANCH_STACK)
1230                 branch_stack__printf(sample, perf_evsel__has_branch_callstack(evsel));
1231
1232         if (sample_type & PERF_SAMPLE_REGS_USER)
1233                 regs_user__printf(sample);
1234
1235         if (sample_type & PERF_SAMPLE_REGS_INTR)
1236                 regs_intr__printf(sample);
1237
1238         if (sample_type & PERF_SAMPLE_STACK_USER)
1239                 stack_user__printf(&sample->user_stack);
1240
1241         if (sample_type & PERF_SAMPLE_WEIGHT)
1242                 printf("... weight: %" PRIu64 "\n", sample->weight);
1243
1244         if (sample_type & PERF_SAMPLE_DATA_SRC)
1245                 printf(" . data_src: 0x%"PRIx64"\n", sample->data_src);
1246
1247         if (sample_type & PERF_SAMPLE_PHYS_ADDR)
1248                 printf(" .. phys_addr: 0x%"PRIx64"\n", sample->phys_addr);
1249
1250         if (sample_type & PERF_SAMPLE_TRANSACTION)
1251                 printf("... transaction: %" PRIx64 "\n", sample->transaction);
1252
1253         if (sample_type & PERF_SAMPLE_READ)
1254                 sample_read__printf(sample, evsel->core.attr.read_format);
1255 }
1256
1257 static void dump_read(struct evsel *evsel, union perf_event *event)
1258 {
1259         struct perf_record_read *read_event = &event->read;
1260         u64 read_format;
1261
1262         if (!dump_trace)
1263                 return;
1264
1265         printf(": %d %d %s %" PRI_lu64 "\n", event->read.pid, event->read.tid,
1266                perf_evsel__name(evsel),
1267                event->read.value);
1268
1269         if (!evsel)
1270                 return;
1271
1272         read_format = evsel->core.attr.read_format;
1273
1274         if (read_format & PERF_FORMAT_TOTAL_TIME_ENABLED)
1275                 printf("... time enabled : %" PRI_lu64 "\n", read_event->time_enabled);
1276
1277         if (read_format & PERF_FORMAT_TOTAL_TIME_RUNNING)
1278                 printf("... time running : %" PRI_lu64 "\n", read_event->time_running);
1279
1280         if (read_format & PERF_FORMAT_ID)
1281                 printf("... id           : %" PRI_lu64 "\n", read_event->id);
1282 }
1283
1284 static struct machine *machines__find_for_cpumode(struct machines *machines,
1285                                                union perf_event *event,
1286                                                struct perf_sample *sample)
1287 {
1288         struct machine *machine;
1289
1290         if (perf_guest &&
1291             ((sample->cpumode == PERF_RECORD_MISC_GUEST_KERNEL) ||
1292              (sample->cpumode == PERF_RECORD_MISC_GUEST_USER))) {
1293                 u32 pid;
1294
1295                 if (event->header.type == PERF_RECORD_MMAP
1296                     || event->header.type == PERF_RECORD_MMAP2)
1297                         pid = event->mmap.pid;
1298                 else
1299                         pid = sample->pid;
1300
1301                 machine = machines__find(machines, pid);
1302                 if (!machine)
1303                         machine = machines__findnew(machines, DEFAULT_GUEST_KERNEL_ID);
1304                 return machine;
1305         }
1306
1307         return &machines->host;
1308 }
1309
1310 static int deliver_sample_value(struct evlist *evlist,
1311                                 struct perf_tool *tool,
1312                                 union perf_event *event,
1313                                 struct perf_sample *sample,
1314                                 struct sample_read_value *v,
1315                                 struct machine *machine)
1316 {
1317         struct perf_sample_id *sid = perf_evlist__id2sid(evlist, v->id);
1318
1319         if (sid) {
1320                 sample->id     = v->id;
1321                 sample->period = v->value - sid->period;
1322                 sid->period    = v->value;
1323         }
1324
1325         if (!sid || sid->evsel == NULL) {
1326                 ++evlist->stats.nr_unknown_id;
1327                 return 0;
1328         }
1329
1330         /*
1331          * There's no reason to deliver sample
1332          * for zero period, bail out.
1333          */
1334         if (!sample->period)
1335                 return 0;
1336
1337         return tool->sample(tool, event, sample, sid->evsel, machine);
1338 }
1339
1340 static int deliver_sample_group(struct evlist *evlist,
1341                                 struct perf_tool *tool,
1342                                 union  perf_event *event,
1343                                 struct perf_sample *sample,
1344                                 struct machine *machine)
1345 {
1346         int ret = -EINVAL;
1347         u64 i;
1348
1349         for (i = 0; i < sample->read.group.nr; i++) {
1350                 ret = deliver_sample_value(evlist, tool, event, sample,
1351                                            &sample->read.group.values[i],
1352                                            machine);
1353                 if (ret)
1354                         break;
1355         }
1356
1357         return ret;
1358 }
1359
1360 static int
1361  perf_evlist__deliver_sample(struct evlist *evlist,
1362                              struct perf_tool *tool,
1363                              union  perf_event *event,
1364                              struct perf_sample *sample,
1365                              struct evsel *evsel,
1366                              struct machine *machine)
1367 {
1368         /* We know evsel != NULL. */
1369         u64 sample_type = evsel->core.attr.sample_type;
1370         u64 read_format = evsel->core.attr.read_format;
1371
1372         /* Standard sample delivery. */
1373         if (!(sample_type & PERF_SAMPLE_READ))
1374                 return tool->sample(tool, event, sample, evsel, machine);
1375
1376         /* For PERF_SAMPLE_READ we have either single or group mode. */
1377         if (read_format & PERF_FORMAT_GROUP)
1378                 return deliver_sample_group(evlist, tool, event, sample,
1379                                             machine);
1380         else
1381                 return deliver_sample_value(evlist, tool, event, sample,
1382                                             &sample->read.one, machine);
1383 }
1384
1385 static int machines__deliver_event(struct machines *machines,
1386                                    struct evlist *evlist,
1387                                    union perf_event *event,
1388                                    struct perf_sample *sample,
1389                                    struct perf_tool *tool, u64 file_offset)
1390 {
1391         struct evsel *evsel;
1392         struct machine *machine;
1393
1394         dump_event(evlist, event, file_offset, sample);
1395
1396         evsel = perf_evlist__id2evsel(evlist, sample->id);
1397
1398         machine = machines__find_for_cpumode(machines, event, sample);
1399
1400         switch (event->header.type) {
1401         case PERF_RECORD_SAMPLE:
1402                 if (evsel == NULL) {
1403                         ++evlist->stats.nr_unknown_id;
1404                         return 0;
1405                 }
1406                 dump_sample(evsel, event, sample);
1407                 if (machine == NULL) {
1408                         ++evlist->stats.nr_unprocessable_samples;
1409                         return 0;
1410                 }
1411                 return perf_evlist__deliver_sample(evlist, tool, event, sample, evsel, machine);
1412         case PERF_RECORD_MMAP:
1413                 return tool->mmap(tool, event, sample, machine);
1414         case PERF_RECORD_MMAP2:
1415                 if (event->header.misc & PERF_RECORD_MISC_PROC_MAP_PARSE_TIMEOUT)
1416                         ++evlist->stats.nr_proc_map_timeout;
1417                 return tool->mmap2(tool, event, sample, machine);
1418         case PERF_RECORD_COMM:
1419                 return tool->comm(tool, event, sample, machine);
1420         case PERF_RECORD_NAMESPACES:
1421                 return tool->namespaces(tool, event, sample, machine);
1422         case PERF_RECORD_FORK:
1423                 return tool->fork(tool, event, sample, machine);
1424         case PERF_RECORD_EXIT:
1425                 return tool->exit(tool, event, sample, machine);
1426         case PERF_RECORD_LOST:
1427                 if (tool->lost == perf_event__process_lost)
1428                         evlist->stats.total_lost += event->lost.lost;
1429                 return tool->lost(tool, event, sample, machine);
1430         case PERF_RECORD_LOST_SAMPLES:
1431                 if (tool->lost_samples == perf_event__process_lost_samples)
1432                         evlist->stats.total_lost_samples += event->lost_samples.lost;
1433                 return tool->lost_samples(tool, event, sample, machine);
1434         case PERF_RECORD_READ:
1435                 dump_read(evsel, event);
1436                 return tool->read(tool, event, sample, evsel, machine);
1437         case PERF_RECORD_THROTTLE:
1438                 return tool->throttle(tool, event, sample, machine);
1439         case PERF_RECORD_UNTHROTTLE:
1440                 return tool->unthrottle(tool, event, sample, machine);
1441         case PERF_RECORD_AUX:
1442                 if (tool->aux == perf_event__process_aux) {
1443                         if (event->aux.flags & PERF_AUX_FLAG_TRUNCATED)
1444                                 evlist->stats.total_aux_lost += 1;
1445                         if (event->aux.flags & PERF_AUX_FLAG_PARTIAL)
1446                                 evlist->stats.total_aux_partial += 1;
1447                 }
1448                 return tool->aux(tool, event, sample, machine);
1449         case PERF_RECORD_ITRACE_START:
1450                 return tool->itrace_start(tool, event, sample, machine);
1451         case PERF_RECORD_SWITCH:
1452         case PERF_RECORD_SWITCH_CPU_WIDE:
1453                 return tool->context_switch(tool, event, sample, machine);
1454         case PERF_RECORD_KSYMBOL:
1455                 return tool->ksymbol(tool, event, sample, machine);
1456         case PERF_RECORD_BPF_EVENT:
1457                 return tool->bpf(tool, event, sample, machine);
1458         default:
1459                 ++evlist->stats.nr_unknown_events;
1460                 return -1;
1461         }
1462 }
1463
1464 static int perf_session__deliver_event(struct perf_session *session,
1465                                        union perf_event *event,
1466                                        struct perf_tool *tool,
1467                                        u64 file_offset)
1468 {
1469         struct perf_sample sample;
1470         int ret;
1471
1472         ret = perf_evlist__parse_sample(session->evlist, event, &sample);
1473         if (ret) {
1474                 pr_err("Can't parse sample, err = %d\n", ret);
1475                 return ret;
1476         }
1477
1478         ret = auxtrace__process_event(session, event, &sample, tool);
1479         if (ret < 0)
1480                 return ret;
1481         if (ret > 0)
1482                 return 0;
1483
1484         return machines__deliver_event(&session->machines, session->evlist,
1485                                        event, &sample, tool, file_offset);
1486 }
1487
1488 static s64 perf_session__process_user_event(struct perf_session *session,
1489                                             union perf_event *event,
1490                                             u64 file_offset)
1491 {
1492         struct ordered_events *oe = &session->ordered_events;
1493         struct perf_tool *tool = session->tool;
1494         struct perf_sample sample = { .time = 0, };
1495         int fd = perf_data__fd(session->data);
1496         int err;
1497
1498         if (event->header.type != PERF_RECORD_COMPRESSED ||
1499             tool->compressed == perf_session__process_compressed_event_stub)
1500                 dump_event(session->evlist, event, file_offset, &sample);
1501
1502         /* These events are processed right away */
1503         switch (event->header.type) {
1504         case PERF_RECORD_HEADER_ATTR:
1505                 err = tool->attr(tool, event, &session->evlist);
1506                 if (err == 0) {
1507                         perf_session__set_id_hdr_size(session);
1508                         perf_session__set_comm_exec(session);
1509                 }
1510                 return err;
1511         case PERF_RECORD_EVENT_UPDATE:
1512                 return tool->event_update(tool, event, &session->evlist);
1513         case PERF_RECORD_HEADER_EVENT_TYPE:
1514                 /*
1515                  * Depreceated, but we need to handle it for sake
1516                  * of old data files create in pipe mode.
1517                  */
1518                 return 0;
1519         case PERF_RECORD_HEADER_TRACING_DATA:
1520                 /* setup for reading amidst mmap */
1521                 lseek(fd, file_offset, SEEK_SET);
1522                 return tool->tracing_data(session, event);
1523         case PERF_RECORD_HEADER_BUILD_ID:
1524                 return tool->build_id(session, event);
1525         case PERF_RECORD_FINISHED_ROUND:
1526                 return tool->finished_round(tool, event, oe);
1527         case PERF_RECORD_ID_INDEX:
1528                 return tool->id_index(session, event);
1529         case PERF_RECORD_AUXTRACE_INFO:
1530                 return tool->auxtrace_info(session, event);
1531         case PERF_RECORD_AUXTRACE:
1532                 /* setup for reading amidst mmap */
1533                 lseek(fd, file_offset + event->header.size, SEEK_SET);
1534                 return tool->auxtrace(session, event);
1535         case PERF_RECORD_AUXTRACE_ERROR:
1536                 perf_session__auxtrace_error_inc(session, event);
1537                 return tool->auxtrace_error(session, event);
1538         case PERF_RECORD_THREAD_MAP:
1539                 return tool->thread_map(session, event);
1540         case PERF_RECORD_CPU_MAP:
1541                 return tool->cpu_map(session, event);
1542         case PERF_RECORD_STAT_CONFIG:
1543                 return tool->stat_config(session, event);
1544         case PERF_RECORD_STAT:
1545                 return tool->stat(session, event);
1546         case PERF_RECORD_STAT_ROUND:
1547                 return tool->stat_round(session, event);
1548         case PERF_RECORD_TIME_CONV:
1549                 session->time_conv = event->time_conv;
1550                 return tool->time_conv(session, event);
1551         case PERF_RECORD_HEADER_FEATURE:
1552                 return tool->feature(session, event);
1553         case PERF_RECORD_COMPRESSED:
1554                 err = tool->compressed(session, event, file_offset);
1555                 if (err)
1556                         dump_event(session->evlist, event, file_offset, &sample);
1557                 return err;
1558         default:
1559                 return -EINVAL;
1560         }
1561 }
1562
1563 int perf_session__deliver_synth_event(struct perf_session *session,
1564                                       union perf_event *event,
1565                                       struct perf_sample *sample)
1566 {
1567         struct evlist *evlist = session->evlist;
1568         struct perf_tool *tool = session->tool;
1569
1570         events_stats__inc(&evlist->stats, event->header.type);
1571
1572         if (event->header.type >= PERF_RECORD_USER_TYPE_START)
1573                 return perf_session__process_user_event(session, event, 0);
1574
1575         return machines__deliver_event(&session->machines, evlist, event, sample, tool, 0);
1576 }
1577
1578 static void event_swap(union perf_event *event, bool sample_id_all)
1579 {
1580         perf_event__swap_op swap;
1581
1582         swap = perf_event__swap_ops[event->header.type];
1583         if (swap)
1584                 swap(event, sample_id_all);
1585 }
1586
1587 int perf_session__peek_event(struct perf_session *session, off_t file_offset,
1588                              void *buf, size_t buf_sz,
1589                              union perf_event **event_ptr,
1590                              struct perf_sample *sample)
1591 {
1592         union perf_event *event;
1593         size_t hdr_sz, rest;
1594         int fd;
1595
1596         if (session->one_mmap && !session->header.needs_swap) {
1597                 event = file_offset - session->one_mmap_offset +
1598                         session->one_mmap_addr;
1599                 goto out_parse_sample;
1600         }
1601
1602         if (perf_data__is_pipe(session->data))
1603                 return -1;
1604
1605         fd = perf_data__fd(session->data);
1606         hdr_sz = sizeof(struct perf_event_header);
1607
1608         if (buf_sz < hdr_sz)
1609                 return -1;
1610
1611         if (lseek(fd, file_offset, SEEK_SET) == (off_t)-1 ||
1612             readn(fd, buf, hdr_sz) != (ssize_t)hdr_sz)
1613                 return -1;
1614
1615         event = (union perf_event *)buf;
1616
1617         if (session->header.needs_swap)
1618                 perf_event_header__bswap(&event->header);
1619
1620         if (event->header.size < hdr_sz || event->header.size > buf_sz)
1621                 return -1;
1622
1623         rest = event->header.size - hdr_sz;
1624
1625         if (readn(fd, buf, rest) != (ssize_t)rest)
1626                 return -1;
1627
1628         if (session->header.needs_swap)
1629                 event_swap(event, perf_evlist__sample_id_all(session->evlist));
1630
1631 out_parse_sample:
1632
1633         if (sample && event->header.type < PERF_RECORD_USER_TYPE_START &&
1634             perf_evlist__parse_sample(session->evlist, event, sample))
1635                 return -1;
1636
1637         *event_ptr = event;
1638
1639         return 0;
1640 }
1641
1642 static s64 perf_session__process_event(struct perf_session *session,
1643                                        union perf_event *event, u64 file_offset)
1644 {
1645         struct evlist *evlist = session->evlist;
1646         struct perf_tool *tool = session->tool;
1647         int ret;
1648
1649         if (session->header.needs_swap)
1650                 event_swap(event, perf_evlist__sample_id_all(evlist));
1651
1652         if (event->header.type >= PERF_RECORD_HEADER_MAX)
1653                 return -EINVAL;
1654
1655         events_stats__inc(&evlist->stats, event->header.type);
1656
1657         if (event->header.type >= PERF_RECORD_USER_TYPE_START)
1658                 return perf_session__process_user_event(session, event, file_offset);
1659
1660         if (tool->ordered_events) {
1661                 u64 timestamp = -1ULL;
1662
1663                 ret = perf_evlist__parse_sample_timestamp(evlist, event, &timestamp);
1664                 if (ret && ret != -1)
1665                         return ret;
1666
1667                 ret = perf_session__queue_event(session, event, timestamp, file_offset);
1668                 if (ret != -ETIME)
1669                         return ret;
1670         }
1671
1672         return perf_session__deliver_event(session, event, tool, file_offset);
1673 }
1674
1675 void perf_event_header__bswap(struct perf_event_header *hdr)
1676 {
1677         hdr->type = bswap_32(hdr->type);
1678         hdr->misc = bswap_16(hdr->misc);
1679         hdr->size = bswap_16(hdr->size);
1680 }
1681
1682 struct thread *perf_session__findnew(struct perf_session *session, pid_t pid)
1683 {
1684         return machine__findnew_thread(&session->machines.host, -1, pid);
1685 }
1686
1687 /*
1688  * Threads are identified by pid and tid, and the idle task has pid == tid == 0.
1689  * So here a single thread is created for that, but actually there is a separate
1690  * idle task per cpu, so there should be one 'struct thread' per cpu, but there
1691  * is only 1. That causes problems for some tools, requiring workarounds. For
1692  * example get_idle_thread() in builtin-sched.c, or thread_stack__per_cpu().
1693  */
1694 int perf_session__register_idle_thread(struct perf_session *session)
1695 {
1696         struct thread *thread;
1697         int err = 0;
1698
1699         thread = machine__findnew_thread(&session->machines.host, 0, 0);
1700         if (thread == NULL || thread__set_comm(thread, "swapper", 0)) {
1701                 pr_err("problem inserting idle task.\n");
1702                 err = -1;
1703         }
1704
1705         if (thread == NULL || thread__set_namespaces(thread, 0, NULL)) {
1706                 pr_err("problem inserting idle task.\n");
1707                 err = -1;
1708         }
1709
1710         /* machine__findnew_thread() got the thread, so put it */
1711         thread__put(thread);
1712         return err;
1713 }
1714
1715 static void
1716 perf_session__warn_order(const struct perf_session *session)
1717 {
1718         const struct ordered_events *oe = &session->ordered_events;
1719         struct evsel *evsel;
1720         bool should_warn = true;
1721
1722         evlist__for_each_entry(session->evlist, evsel) {
1723                 if (evsel->core.attr.write_backward)
1724                         should_warn = false;
1725         }
1726
1727         if (!should_warn)
1728                 return;
1729         if (oe->nr_unordered_events != 0)
1730                 ui__warning("%u out of order events recorded.\n", oe->nr_unordered_events);
1731 }
1732
1733 static void perf_session__warn_about_errors(const struct perf_session *session)
1734 {
1735         const struct events_stats *stats = &session->evlist->stats;
1736
1737         if (session->tool->lost == perf_event__process_lost &&
1738             stats->nr_events[PERF_RECORD_LOST] != 0) {
1739                 ui__warning("Processed %d events and lost %d chunks!\n\n"
1740                             "Check IO/CPU overload!\n\n",
1741                             stats->nr_events[0],
1742                             stats->nr_events[PERF_RECORD_LOST]);
1743         }
1744
1745         if (session->tool->lost_samples == perf_event__process_lost_samples) {
1746                 double drop_rate;
1747
1748                 drop_rate = (double)stats->total_lost_samples /
1749                             (double) (stats->nr_events[PERF_RECORD_SAMPLE] + stats->total_lost_samples);
1750                 if (drop_rate > 0.05) {
1751                         ui__warning("Processed %" PRIu64 " samples and lost %3.2f%%!\n\n",
1752                                     stats->nr_events[PERF_RECORD_SAMPLE] + stats->total_lost_samples,
1753                                     drop_rate * 100.0);
1754                 }
1755         }
1756
1757         if (session->tool->aux == perf_event__process_aux &&
1758             stats->total_aux_lost != 0) {
1759                 ui__warning("AUX data lost %" PRIu64 " times out of %u!\n\n",
1760                             stats->total_aux_lost,
1761                             stats->nr_events[PERF_RECORD_AUX]);
1762         }
1763
1764         if (session->tool->aux == perf_event__process_aux &&
1765             stats->total_aux_partial != 0) {
1766                 bool vmm_exclusive = false;
1767
1768                 (void)sysfs__read_bool("module/kvm_intel/parameters/vmm_exclusive",
1769                                        &vmm_exclusive);
1770
1771                 ui__warning("AUX data had gaps in it %" PRIu64 " times out of %u!\n\n"
1772                             "Are you running a KVM guest in the background?%s\n\n",
1773                             stats->total_aux_partial,
1774                             stats->nr_events[PERF_RECORD_AUX],
1775                             vmm_exclusive ?
1776                             "\nReloading kvm_intel module with vmm_exclusive=0\n"
1777                             "will reduce the gaps to only guest's timeslices." :
1778                             "");
1779         }
1780
1781         if (stats->nr_unknown_events != 0) {
1782                 ui__warning("Found %u unknown events!\n\n"
1783                             "Is this an older tool processing a perf.data "
1784                             "file generated by a more recent tool?\n\n"
1785                             "If that is not the case, consider "
1786                             "reporting to linux-kernel@vger.kernel.org.\n\n",
1787                             stats->nr_unknown_events);
1788         }
1789
1790         if (stats->nr_unknown_id != 0) {
1791                 ui__warning("%u samples with id not present in the header\n",
1792                             stats->nr_unknown_id);
1793         }
1794
1795         if (stats->nr_invalid_chains != 0) {
1796                 ui__warning("Found invalid callchains!\n\n"
1797                             "%u out of %u events were discarded for this reason.\n\n"
1798                             "Consider reporting to linux-kernel@vger.kernel.org.\n\n",
1799                             stats->nr_invalid_chains,
1800                             stats->nr_events[PERF_RECORD_SAMPLE]);
1801         }
1802
1803         if (stats->nr_unprocessable_samples != 0) {
1804                 ui__warning("%u unprocessable samples recorded.\n"
1805                             "Do you have a KVM guest running and not using 'perf kvm'?\n",
1806                             stats->nr_unprocessable_samples);
1807         }
1808
1809         perf_session__warn_order(session);
1810
1811         events_stats__auxtrace_error_warn(stats);
1812
1813         if (stats->nr_proc_map_timeout != 0) {
1814                 ui__warning("%d map information files for pre-existing threads were\n"
1815                             "not processed, if there are samples for addresses they\n"
1816                             "will not be resolved, you may find out which are these\n"
1817                             "threads by running with -v and redirecting the output\n"
1818                             "to a file.\n"
1819                             "The time limit to process proc map is too short?\n"
1820                             "Increase it by --proc-map-timeout\n",
1821                             stats->nr_proc_map_timeout);
1822         }
1823 }
1824
1825 static int perf_session__flush_thread_stack(struct thread *thread,
1826                                             void *p __maybe_unused)
1827 {
1828         return thread_stack__flush(thread);
1829 }
1830
1831 static int perf_session__flush_thread_stacks(struct perf_session *session)
1832 {
1833         return machines__for_each_thread(&session->machines,
1834                                          perf_session__flush_thread_stack,
1835                                          NULL);
1836 }
1837
1838 volatile int session_done;
1839
1840 static int __perf_session__process_decomp_events(struct perf_session *session);
1841
1842 static int __perf_session__process_pipe_events(struct perf_session *session)
1843 {
1844         struct ordered_events *oe = &session->ordered_events;
1845         struct perf_tool *tool = session->tool;
1846         int fd = perf_data__fd(session->data);
1847         union perf_event *event;
1848         uint32_t size, cur_size = 0;
1849         void *buf = NULL;
1850         s64 skip = 0;
1851         u64 head;
1852         ssize_t err;
1853         void *p;
1854
1855         perf_tool__fill_defaults(tool);
1856
1857         head = 0;
1858         cur_size = sizeof(union perf_event);
1859
1860         buf = malloc(cur_size);
1861         if (!buf)
1862                 return -errno;
1863         ordered_events__set_copy_on_queue(oe, true);
1864 more:
1865         event = buf;
1866         err = readn(fd, event, sizeof(struct perf_event_header));
1867         if (err <= 0) {
1868                 if (err == 0)
1869                         goto done;
1870
1871                 pr_err("failed to read event header\n");
1872                 goto out_err;
1873         }
1874
1875         if (session->header.needs_swap)
1876                 perf_event_header__bswap(&event->header);
1877
1878         size = event->header.size;
1879         if (size < sizeof(struct perf_event_header)) {
1880                 pr_err("bad event header size\n");
1881                 goto out_err;
1882         }
1883
1884         if (size > cur_size) {
1885                 void *new = realloc(buf, size);
1886                 if (!new) {
1887                         pr_err("failed to allocate memory to read event\n");
1888                         goto out_err;
1889                 }
1890                 buf = new;
1891                 cur_size = size;
1892                 event = buf;
1893         }
1894         p = event;
1895         p += sizeof(struct perf_event_header);
1896
1897         if (size - sizeof(struct perf_event_header)) {
1898                 err = readn(fd, p, size - sizeof(struct perf_event_header));
1899                 if (err <= 0) {
1900                         if (err == 0) {
1901                                 pr_err("unexpected end of event stream\n");
1902                                 goto done;
1903                         }
1904
1905                         pr_err("failed to read event data\n");
1906                         goto out_err;
1907                 }
1908         }
1909
1910         if ((skip = perf_session__process_event(session, event, head)) < 0) {
1911                 pr_err("%#" PRIx64 " [%#x]: failed to process type: %d\n",
1912                        head, event->header.size, event->header.type);
1913                 err = -EINVAL;
1914                 goto out_err;
1915         }
1916
1917         head += size;
1918
1919         if (skip > 0)
1920                 head += skip;
1921
1922         err = __perf_session__process_decomp_events(session);
1923         if (err)
1924                 goto out_err;
1925
1926         if (!session_done())
1927                 goto more;
1928 done:
1929         /* do the final flush for ordered samples */
1930         err = ordered_events__flush(oe, OE_FLUSH__FINAL);
1931         if (err)
1932                 goto out_err;
1933         err = auxtrace__flush_events(session, tool);
1934         if (err)
1935                 goto out_err;
1936         err = perf_session__flush_thread_stacks(session);
1937 out_err:
1938         free(buf);
1939         if (!tool->no_warn)
1940                 perf_session__warn_about_errors(session);
1941         ordered_events__free(&session->ordered_events);
1942         auxtrace__free_events(session);
1943         return err;
1944 }
1945
1946 static union perf_event *
1947 fetch_mmaped_event(struct perf_session *session,
1948                    u64 head, size_t mmap_size, char *buf)
1949 {
1950         union perf_event *event;
1951
1952         /*
1953          * Ensure we have enough space remaining to read
1954          * the size of the event in the headers.
1955          */
1956         if (head + sizeof(event->header) > mmap_size)
1957                 return NULL;
1958
1959         event = (union perf_event *)(buf + head);
1960
1961         if (session->header.needs_swap)
1962                 perf_event_header__bswap(&event->header);
1963
1964         if (head + event->header.size > mmap_size) {
1965                 /* We're not fetching the event so swap back again */
1966                 if (session->header.needs_swap)
1967                         perf_event_header__bswap(&event->header);
1968                 pr_debug("%s: head=%#" PRIx64 " event->header_size=%#x, mmap_size=%#zx: fuzzed perf.data?\n",
1969                          __func__, head, event->header.size, mmap_size);
1970                 return ERR_PTR(-EINVAL);
1971         }
1972
1973         return event;
1974 }
1975
1976 static int __perf_session__process_decomp_events(struct perf_session *session)
1977 {
1978         s64 skip;
1979         u64 size, file_pos = 0;
1980         struct decomp *decomp = session->decomp_last;
1981
1982         if (!decomp)
1983                 return 0;
1984
1985         while (decomp->head < decomp->size && !session_done()) {
1986                 union perf_event *event = fetch_mmaped_event(session, decomp->head, decomp->size, decomp->data);
1987
1988                 if (IS_ERR(event))
1989                         return PTR_ERR(event);
1990
1991                 if (!event)
1992                         break;
1993
1994                 size = event->header.size;
1995
1996                 if (size < sizeof(struct perf_event_header) ||
1997                     (skip = perf_session__process_event(session, event, file_pos)) < 0) {
1998                         pr_err("%#" PRIx64 " [%#x]: failed to process type: %d\n",
1999                                 decomp->file_pos + decomp->head, event->header.size, event->header.type);
2000                         return -EINVAL;
2001                 }
2002
2003                 if (skip)
2004                         size += skip;
2005
2006                 decomp->head += size;
2007         }
2008
2009         return 0;
2010 }
2011
2012 /*
2013  * On 64bit we can mmap the data file in one go. No need for tiny mmap
2014  * slices. On 32bit we use 32MB.
2015  */
2016 #if BITS_PER_LONG == 64
2017 #define MMAP_SIZE ULLONG_MAX
2018 #define NUM_MMAPS 1
2019 #else
2020 #define MMAP_SIZE (32 * 1024 * 1024ULL)
2021 #define NUM_MMAPS 128
2022 #endif
2023
2024 struct reader;
2025
2026 typedef s64 (*reader_cb_t)(struct perf_session *session,
2027                            union perf_event *event,
2028                            u64 file_offset);
2029
2030 struct reader {
2031         int              fd;
2032         u64              data_size;
2033         u64              data_offset;
2034         reader_cb_t      process;
2035 };
2036
2037 static int
2038 reader__process_events(struct reader *rd, struct perf_session *session,
2039                        struct ui_progress *prog)
2040 {
2041         u64 data_size = rd->data_size;
2042         u64 head, page_offset, file_offset, file_pos, size;
2043         int err = 0, mmap_prot, mmap_flags, map_idx = 0;
2044         size_t  mmap_size;
2045         char *buf, *mmaps[NUM_MMAPS];
2046         union perf_event *event;
2047         s64 skip;
2048
2049         page_offset = page_size * (rd->data_offset / page_size);
2050         file_offset = page_offset;
2051         head = rd->data_offset - page_offset;
2052
2053         ui_progress__init_size(prog, data_size, "Processing events...");
2054
2055         data_size += rd->data_offset;
2056
2057         mmap_size = MMAP_SIZE;
2058         if (mmap_size > data_size) {
2059                 mmap_size = data_size;
2060                 session->one_mmap = true;
2061         }
2062
2063         memset(mmaps, 0, sizeof(mmaps));
2064
2065         mmap_prot  = PROT_READ;
2066         mmap_flags = MAP_SHARED;
2067
2068         if (session->header.needs_swap) {
2069                 mmap_prot  |= PROT_WRITE;
2070                 mmap_flags = MAP_PRIVATE;
2071         }
2072 remap:
2073         buf = mmap(NULL, mmap_size, mmap_prot, mmap_flags, rd->fd,
2074                    file_offset);
2075         if (buf == MAP_FAILED) {
2076                 pr_err("failed to mmap file\n");
2077                 err = -errno;
2078                 goto out;
2079         }
2080         mmaps[map_idx] = buf;
2081         map_idx = (map_idx + 1) & (ARRAY_SIZE(mmaps) - 1);
2082         file_pos = file_offset + head;
2083         if (session->one_mmap) {
2084                 session->one_mmap_addr = buf;
2085                 session->one_mmap_offset = file_offset;
2086         }
2087
2088 more:
2089         event = fetch_mmaped_event(session, head, mmap_size, buf);
2090         if (IS_ERR(event))
2091                 return PTR_ERR(event);
2092
2093         if (!event) {
2094                 if (mmaps[map_idx]) {
2095                         munmap(mmaps[map_idx], mmap_size);
2096                         mmaps[map_idx] = NULL;
2097                 }
2098
2099                 page_offset = page_size * (head / page_size);
2100                 file_offset += page_offset;
2101                 head -= page_offset;
2102                 goto remap;
2103         }
2104
2105         size = event->header.size;
2106
2107         skip = -EINVAL;
2108
2109         if (size < sizeof(struct perf_event_header) ||
2110             (skip = rd->process(session, event, file_pos)) < 0) {
2111                 pr_err("%#" PRIx64 " [%#x]: failed to process type: %d [%s]\n",
2112                        file_offset + head, event->header.size,
2113                        event->header.type, strerror(-skip));
2114                 err = skip;
2115                 goto out;
2116         }
2117
2118         if (skip)
2119                 size += skip;
2120
2121         head += size;
2122         file_pos += size;
2123
2124         err = __perf_session__process_decomp_events(session);
2125         if (err)
2126                 goto out;
2127
2128         ui_progress__update(prog, size);
2129
2130         if (session_done())
2131                 goto out;
2132
2133         if (file_pos < data_size)
2134                 goto more;
2135
2136 out:
2137         return err;
2138 }
2139
2140 static s64 process_simple(struct perf_session *session,
2141                           union perf_event *event,
2142                           u64 file_offset)
2143 {
2144         return perf_session__process_event(session, event, file_offset);
2145 }
2146
2147 static int __perf_session__process_events(struct perf_session *session)
2148 {
2149         struct reader rd = {
2150                 .fd             = perf_data__fd(session->data),
2151                 .data_size      = session->header.data_size,
2152                 .data_offset    = session->header.data_offset,
2153                 .process        = process_simple,
2154         };
2155         struct ordered_events *oe = &session->ordered_events;
2156         struct perf_tool *tool = session->tool;
2157         struct ui_progress prog;
2158         int err;
2159
2160         perf_tool__fill_defaults(tool);
2161
2162         if (rd.data_size == 0)
2163                 return -1;
2164
2165         ui_progress__init_size(&prog, rd.data_size, "Processing events...");
2166
2167         err = reader__process_events(&rd, session, &prog);
2168         if (err)
2169                 goto out_err;
2170         /* do the final flush for ordered samples */
2171         err = ordered_events__flush(oe, OE_FLUSH__FINAL);
2172         if (err)
2173                 goto out_err;
2174         err = auxtrace__flush_events(session, tool);
2175         if (err)
2176                 goto out_err;
2177         err = perf_session__flush_thread_stacks(session);
2178 out_err:
2179         ui_progress__finish();
2180         if (!tool->no_warn)
2181                 perf_session__warn_about_errors(session);
2182         /*
2183          * We may switching perf.data output, make ordered_events
2184          * reusable.
2185          */
2186         ordered_events__reinit(&session->ordered_events);
2187         auxtrace__free_events(session);
2188         session->one_mmap = false;
2189         return err;
2190 }
2191
2192 int perf_session__process_events(struct perf_session *session)
2193 {
2194         if (perf_session__register_idle_thread(session) < 0)
2195                 return -ENOMEM;
2196
2197         if (perf_data__is_pipe(session->data))
2198                 return __perf_session__process_pipe_events(session);
2199
2200         return __perf_session__process_events(session);
2201 }
2202
2203 bool perf_session__has_traces(struct perf_session *session, const char *msg)
2204 {
2205         struct evsel *evsel;
2206
2207         evlist__for_each_entry(session->evlist, evsel) {
2208                 if (evsel->core.attr.type == PERF_TYPE_TRACEPOINT)
2209                         return true;
2210         }
2211
2212         pr_err("No trace sample to read. Did you call 'perf %s'?\n", msg);
2213         return false;
2214 }
2215
2216 int map__set_kallsyms_ref_reloc_sym(struct map *map, const char *symbol_name, u64 addr)
2217 {
2218         char *bracket;
2219         struct ref_reloc_sym *ref;
2220         struct kmap *kmap;
2221
2222         ref = zalloc(sizeof(struct ref_reloc_sym));
2223         if (ref == NULL)
2224                 return -ENOMEM;
2225
2226         ref->name = strdup(symbol_name);
2227         if (ref->name == NULL) {
2228                 free(ref);
2229                 return -ENOMEM;
2230         }
2231
2232         bracket = strchr(ref->name, ']');
2233         if (bracket)
2234                 *bracket = '\0';
2235
2236         ref->addr = addr;
2237
2238         kmap = map__kmap(map);
2239         if (kmap)
2240                 kmap->ref_reloc_sym = ref;
2241
2242         return 0;
2243 }
2244
2245 size_t perf_session__fprintf_dsos(struct perf_session *session, FILE *fp)
2246 {
2247         return machines__fprintf_dsos(&session->machines, fp);
2248 }
2249
2250 size_t perf_session__fprintf_dsos_buildid(struct perf_session *session, FILE *fp,
2251                                           bool (skip)(struct dso *dso, int parm), int parm)
2252 {
2253         return machines__fprintf_dsos_buildid(&session->machines, fp, skip, parm);
2254 }
2255
2256 size_t perf_session__fprintf_nr_events(struct perf_session *session, FILE *fp)
2257 {
2258         size_t ret;
2259         const char *msg = "";
2260
2261         if (perf_header__has_feat(&session->header, HEADER_AUXTRACE))
2262                 msg = " (excludes AUX area (e.g. instruction trace) decoded / synthesized events)";
2263
2264         ret = fprintf(fp, "\nAggregated stats:%s\n", msg);
2265
2266         ret += events_stats__fprintf(&session->evlist->stats, fp);
2267         return ret;
2268 }
2269
2270 size_t perf_session__fprintf(struct perf_session *session, FILE *fp)
2271 {
2272         /*
2273          * FIXME: Here we have to actually print all the machines in this
2274          * session, not just the host...
2275          */
2276         return machine__fprintf(&session->machines.host, fp);
2277 }
2278
2279 struct evsel *perf_session__find_first_evtype(struct perf_session *session,
2280                                               unsigned int type)
2281 {
2282         struct evsel *pos;
2283
2284         evlist__for_each_entry(session->evlist, pos) {
2285                 if (pos->core.attr.type == type)
2286                         return pos;
2287         }
2288         return NULL;
2289 }
2290
2291 int perf_session__cpu_bitmap(struct perf_session *session,
2292                              const char *cpu_list, unsigned long *cpu_bitmap)
2293 {
2294         int i, err = -1;
2295         struct perf_cpu_map *map;
2296         int nr_cpus = min(session->header.env.nr_cpus_online, MAX_NR_CPUS);
2297
2298         for (i = 0; i < PERF_TYPE_MAX; ++i) {
2299                 struct evsel *evsel;
2300
2301                 evsel = perf_session__find_first_evtype(session, i);
2302                 if (!evsel)
2303                         continue;
2304
2305                 if (!(evsel->core.attr.sample_type & PERF_SAMPLE_CPU)) {
2306                         pr_err("File does not contain CPU events. "
2307                                "Remove -C option to proceed.\n");
2308                         return -1;
2309                 }
2310         }
2311
2312         map = perf_cpu_map__new(cpu_list);
2313         if (map == NULL) {
2314                 pr_err("Invalid cpu_list\n");
2315                 return -1;
2316         }
2317
2318         for (i = 0; i < map->nr; i++) {
2319                 int cpu = map->map[i];
2320
2321                 if (cpu >= nr_cpus) {
2322                         pr_err("Requested CPU %d too large. "
2323                                "Consider raising MAX_NR_CPUS\n", cpu);
2324                         goto out_delete_map;
2325                 }
2326
2327                 set_bit(cpu, cpu_bitmap);
2328         }
2329
2330         err = 0;
2331
2332 out_delete_map:
2333         perf_cpu_map__put(map);
2334         return err;
2335 }
2336
2337 void perf_session__fprintf_info(struct perf_session *session, FILE *fp,
2338                                 bool full)
2339 {
2340         if (session == NULL || fp == NULL)
2341                 return;
2342
2343         fprintf(fp, "# ========\n");
2344         perf_header__fprintf_info(session, fp, full);
2345         fprintf(fp, "# ========\n#\n");
2346 }
2347
2348
2349 int __perf_session__set_tracepoints_handlers(struct perf_session *session,
2350                                              const struct evsel_str_handler *assocs,
2351                                              size_t nr_assocs)
2352 {
2353         struct evsel *evsel;
2354         size_t i;
2355         int err;
2356
2357         for (i = 0; i < nr_assocs; i++) {
2358                 /*
2359                  * Adding a handler for an event not in the session,
2360                  * just ignore it.
2361                  */
2362                 evsel = perf_evlist__find_tracepoint_by_name(session->evlist, assocs[i].name);
2363                 if (evsel == NULL)
2364                         continue;
2365
2366                 err = -EEXIST;
2367                 if (evsel->handler != NULL)
2368                         goto out;
2369                 evsel->handler = assocs[i].handler;
2370         }
2371
2372         err = 0;
2373 out:
2374         return err;
2375 }
2376
2377 int perf_event__process_id_index(struct perf_session *session,
2378                                  union perf_event *event)
2379 {
2380         struct evlist *evlist = session->evlist;
2381         struct perf_record_id_index *ie = &event->id_index;
2382         size_t i, nr, max_nr;
2383
2384         max_nr = (ie->header.size - sizeof(struct perf_record_id_index)) /
2385                  sizeof(struct id_index_entry);
2386         nr = ie->nr;
2387         if (nr > max_nr)
2388                 return -EINVAL;
2389
2390         if (dump_trace)
2391                 fprintf(stdout, " nr: %zu\n", nr);
2392
2393         for (i = 0; i < nr; i++) {
2394                 struct id_index_entry *e = &ie->entries[i];
2395                 struct perf_sample_id *sid;
2396
2397                 if (dump_trace) {
2398                         fprintf(stdout, " ... id: %"PRI_lu64, e->id);
2399                         fprintf(stdout, "  idx: %"PRI_lu64, e->idx);
2400                         fprintf(stdout, "  cpu: %"PRI_ld64, e->cpu);
2401                         fprintf(stdout, "  tid: %"PRI_ld64"\n", e->tid);
2402                 }
2403
2404                 sid = perf_evlist__id2sid(evlist, e->id);
2405                 if (!sid)
2406                         return -ENOENT;
2407                 sid->idx = e->idx;
2408                 sid->cpu = e->cpu;
2409                 sid->tid = e->tid;
2410         }
2411         return 0;
2412 }
2413
2414 int perf_event__synthesize_id_index(struct perf_tool *tool,
2415                                     perf_event__handler_t process,
2416                                     struct evlist *evlist,
2417                                     struct machine *machine)
2418 {
2419         union perf_event *ev;
2420         struct evsel *evsel;
2421         size_t nr = 0, i = 0, sz, max_nr, n;
2422         int err;
2423
2424         pr_debug2("Synthesizing id index\n");
2425
2426         max_nr = (UINT16_MAX - sizeof(struct perf_record_id_index)) /
2427                  sizeof(struct id_index_entry);
2428
2429         evlist__for_each_entry(evlist, evsel)
2430                 nr += evsel->ids;
2431
2432         n = nr > max_nr ? max_nr : nr;
2433         sz = sizeof(struct perf_record_id_index) + n * sizeof(struct id_index_entry);
2434         ev = zalloc(sz);
2435         if (!ev)
2436                 return -ENOMEM;
2437
2438         ev->id_index.header.type = PERF_RECORD_ID_INDEX;
2439         ev->id_index.header.size = sz;
2440         ev->id_index.nr = n;
2441
2442         evlist__for_each_entry(evlist, evsel) {
2443                 u32 j;
2444
2445                 for (j = 0; j < evsel->ids; j++) {
2446                         struct id_index_entry *e;
2447                         struct perf_sample_id *sid;
2448
2449                         if (i >= n) {
2450                                 err = process(tool, ev, NULL, machine);
2451                                 if (err)
2452                                         goto out_err;
2453                                 nr -= n;
2454                                 i = 0;
2455                         }
2456
2457                         e = &ev->id_index.entries[i++];
2458
2459                         e->id = evsel->id[j];
2460
2461                         sid = perf_evlist__id2sid(evlist, e->id);
2462                         if (!sid) {
2463                                 free(ev);
2464                                 return -ENOENT;
2465                         }
2466
2467                         e->idx = sid->idx;
2468                         e->cpu = sid->cpu;
2469                         e->tid = sid->tid;
2470                 }
2471         }
2472
2473         sz = sizeof(struct perf_record_id_index) + nr * sizeof(struct id_index_entry);
2474         ev->id_index.header.size = sz;
2475         ev->id_index.nr = nr;
2476
2477         err = process(tool, ev, NULL, machine);
2478 out_err:
2479         free(ev);
2480
2481         return err;
2482 }