tools/perf/util/session.c

   1 #include <linux/kernel.h>
   2 #include <traceevent/event-parse.h>
   3
   4 #include <byteswap.h>
   5 #include <unistd.h>
   6 #include <sys/types.h>
   7 #include <sys/mman.h>
   8
   9 #include "evlist.h"
  10 #include "evsel.h"
  11 #include "session.h"
  12 #include "tool.h"
  13 #include "sort.h"
  14 #include "util.h"
  15 #include "cpumap.h"
  16 #include "perf_regs.h"
  17 #include "asm/bug.h"
  18
  19 static int perf_session__open(struct perf_session *session)
  20 {
  21         struct perf_data_file *file = session->file;
  22
  23         if (perf_session__read_header(session) < 0) {
  24                 pr_err("incompatible file format (rerun with -v to learn more)");
  25                 return -1;
  26         }
  27
  28         if (perf_data_file__is_pipe(file))
  29                 return 0;
  30
  31         if (!perf_evlist__valid_sample_type(session->evlist)) {
  32                 pr_err("non matching sample_type");
  33                 return -1;
  34         }
  35
  36         if (!perf_evlist__valid_sample_id_all(session->evlist)) {
  37                 pr_err("non matching sample_id_all");
  38                 return -1;
  39         }
  40
  41         if (!perf_evlist__valid_read_format(session->evlist)) {
  42                 pr_err("non matching read_format");
  43                 return -1;
  44         }
  45
  46         return 0;
  47 }
  48
  49 void perf_session__set_id_hdr_size(struct perf_session *session)
  50 {
  51         u16 id_hdr_size = perf_evlist__id_hdr_size(session->evlist);
  52
  53         machines__set_id_hdr_size(&session->machines, id_hdr_size);
  54 }
  55
  56 int perf_session__create_kernel_maps(struct perf_session *session)
  57 {
  58         int ret = machine__create_kernel_maps(&session->machines.host);
  59
  60         if (ret >= 0)
  61                 ret = machines__create_guest_kernel_maps(&session->machines);
  62         return ret;
  63 }
  64
  65 static void perf_session__destroy_kernel_maps(struct perf_session *session)
  66 {
  67         machines__destroy_kernel_maps(&session->machines);
  68 }
  69
  70 static bool perf_session__has_comm_exec(struct perf_session *session)
  71 {
  72         struct perf_evsel *evsel;
  73
  74         evlist__for_each(session->evlist, evsel) {
  75                 if (evsel->attr.comm_exec)
  76                         return true;
  77         }
  78
  79         return false;
  80 }
  81
  82 static void perf_session__set_comm_exec(struct perf_session *session)
  83 {
  84         bool comm_exec = perf_session__has_comm_exec(session);
  85
  86         machines__set_comm_exec(&session->machines, comm_exec);
  87 }
  88
  89 struct perf_session *perf_session__new(struct perf_data_file *file,
  90                                        bool repipe, struct perf_tool *tool)
  91 {
  92         struct perf_session *session = zalloc(sizeof(*session));
  93
  94         if (!session)
  95                 goto out;
  96
  97         session->repipe = repipe;
  98         ordered_events__init(&session->ordered_events);
  99         machines__init(&session->machines);
 100
 101         if (file) {
 102                 if (perf_data_file__open(file))
 103                         goto out_delete;
 104
 105                 session->file = file;
 106
 107                 if (perf_data_file__is_read(file)) {
 108                         if (perf_session__open(session) < 0)
 109                                 goto out_close;
 110
 111                         perf_session__set_id_hdr_size(session);
 112                         perf_session__set_comm_exec(session);
 113                 }
 114         }
 115
 116         if (!file || perf_data_file__is_write(file)) {
 117                 /*
 118                  * In O_RDONLY mode this will be performed when reading the
 119                  * kernel MMAP event, in perf_event__process_mmap().
 120                  */
 121                 if (perf_session__create_kernel_maps(session) < 0)
 122                         pr_warning("Cannot read kernel map\n");
 123         }
 124
 125         if (tool && tool->ordering_requires_timestamps &&
 126             tool->ordered_events && !perf_evlist__sample_id_all(session->evlist)) {
 127                 dump_printf("WARNING: No sample_id_all support, falling back to unordered processing\n");
 128                 tool->ordered_events = false;
 129         }
 130
 131         return session;
 132
 133  out_close:
 134         perf_data_file__close(file);
 135  out_delete:
 136         perf_session__delete(session);
 137  out:
 138         return NULL;
 139 }
 140
 141 static void perf_session__delete_dead_threads(struct perf_session *session)
 142 {
 143         machine__delete_dead_threads(&session->machines.host);
 144 }
 145
 146 static void perf_session__delete_threads(struct perf_session *session)
 147 {
 148         machine__delete_threads(&session->machines.host);
 149 }
 150
 151 static void perf_session_env__delete(struct perf_session_env *env)
 152 {
 153         zfree(&env->hostname);
 154         zfree(&env->os_release);
 155         zfree(&env->version);
 156         zfree(&env->arch);
 157         zfree(&env->cpu_desc);
 158         zfree(&env->cpuid);
 159
 160         zfree(&env->cmdline);
 161         zfree(&env->sibling_cores);
 162         zfree(&env->sibling_threads);
 163         zfree(&env->numa_nodes);
 164         zfree(&env->pmu_mappings);
 165 }
 166
 167 void perf_session__delete(struct perf_session *session)
 168 {
 169         perf_session__destroy_kernel_maps(session);
 170         perf_session__delete_dead_threads(session);
 171         perf_session__delete_threads(session);
 172         perf_session_env__delete(&session->header.env);
 173         machines__exit(&session->machines);
 174         if (session->file)
 175                 perf_data_file__close(session->file);
 176         free(session);
 177 }
 178
 179 static int process_event_synth_tracing_data_stub(struct perf_tool *tool
 180                                                  __maybe_unused,
 181                                                  union perf_event *event
 182                                                  __maybe_unused,
 183                                                  struct perf_session *session
 184                                                 __maybe_unused)
 185 {
 186         dump_printf(": unhandled!\n");
 187         return 0;
 188 }
 189
 190 static int process_event_synth_attr_stub(struct perf_tool *tool __maybe_unused,
 191                                          union perf_event *event __maybe_unused,
 192                                          struct perf_evlist **pevlist
 193                                          __maybe_unused)
 194 {
 195         dump_printf(": unhandled!\n");
 196         return 0;
 197 }
 198
 199 static int process_event_sample_stub(struct perf_tool *tool __maybe_unused,
 200                                      union perf_event *event __maybe_unused,
 201                                      struct perf_sample *sample __maybe_unused,
 202                                      struct perf_evsel *evsel __maybe_unused,
 203                                      struct machine *machine __maybe_unused)
 204 {
 205         dump_printf(": unhandled!\n");
 206         return 0;
 207 }
 208
 209 static int process_event_stub(struct perf_tool *tool __maybe_unused,
 210                               union perf_event *event __maybe_unused,
 211                               struct perf_sample *sample __maybe_unused,
 212                               struct machine *machine __maybe_unused)
 213 {
 214         dump_printf(": unhandled!\n");
 215         return 0;
 216 }
 217
 218 static int process_finished_round_stub(struct perf_tool *tool __maybe_unused,
 219                                        union perf_event *event __maybe_unused,
 220                                        struct perf_session *perf_session
 221                                        __maybe_unused)
 222 {
 223         dump_printf(": unhandled!\n");
 224         return 0;
 225 }
 226
 227 static int process_finished_round(struct perf_tool *tool,
 228                                   union perf_event *event,
 229                                   struct perf_session *session);
 230
 231 static int process_id_index_stub(struct perf_tool *tool __maybe_unused,
 232                                  union perf_event *event __maybe_unused,
 233                                  struct perf_session *perf_session
 234                                  __maybe_unused)
 235 {
 236         dump_printf(": unhandled!\n");
 237         return 0;
 238 }
 239
 240 void perf_tool__fill_defaults(struct perf_tool *tool)
 241 {
 242         if (tool->sample == NULL)
 243                 tool->sample = process_event_sample_stub;
 244         if (tool->mmap == NULL)
 245                 tool->mmap = process_event_stub;
 246         if (tool->mmap2 == NULL)
 247                 tool->mmap2 = process_event_stub;
 248         if (tool->comm == NULL)
 249                 tool->comm = process_event_stub;
 250         if (tool->fork == NULL)
 251                 tool->fork = process_event_stub;
 252         if (tool->exit == NULL)
 253                 tool->exit = process_event_stub;
 254         if (tool->lost == NULL)
 255                 tool->lost = perf_event__process_lost;
 256         if (tool->read == NULL)
 257                 tool->read = process_event_sample_stub;
 258         if (tool->throttle == NULL)
 259                 tool->throttle = process_event_stub;
 260         if (tool->unthrottle == NULL)
 261                 tool->unthrottle = process_event_stub;
 262         if (tool->attr == NULL)
 263                 tool->attr = process_event_synth_attr_stub;
 264         if (tool->tracing_data == NULL)
 265                 tool->tracing_data = process_event_synth_tracing_data_stub;
 266         if (tool->build_id == NULL)
 267                 tool->build_id = process_finished_round_stub;
 268         if (tool->finished_round == NULL) {
 269                 if (tool->ordered_events)
 270                         tool->finished_round = process_finished_round;
 271                 else
 272                         tool->finished_round = process_finished_round_stub;
 273         }
 274         if (tool->id_index == NULL)
 275                 tool->id_index = process_id_index_stub;
 276 }
 277
 278 static void swap_sample_id_all(union perf_event *event, void *data)
 279 {
 280         void *end = (void *) event + event->header.size;
 281         int size = end - data;
 282
 283         BUG_ON(size % sizeof(u64));
 284         mem_bswap_64(data, size);
 285 }
 286
 287 static void perf_event__all64_swap(union perf_event *event,
 288                                    bool sample_id_all __maybe_unused)
 289 {
 290         struct perf_event_header *hdr = &event->header;
 291         mem_bswap_64(hdr + 1, event->header.size - sizeof(*hdr));
 292 }
 293
 294 static void perf_event__comm_swap(union perf_event *event, bool sample_id_all)
 295 {
 296         event->comm.pid = bswap_32(event->comm.pid);
 297         event->comm.tid = bswap_32(event->comm.tid);
 298
 299         if (sample_id_all) {
 300                 void *data = &event->comm.comm;
 301
 302                 data += PERF_ALIGN(strlen(data) + 1, sizeof(u64));
 303                 swap_sample_id_all(event, data);
 304         }
 305 }
 306
 307 static void perf_event__mmap_swap(union perf_event *event,
 308                                   bool sample_id_all)
 309 {
 310         event->mmap.pid   = bswap_32(event->mmap.pid);
 311         event->mmap.tid   = bswap_32(event->mmap.tid);
 312         event->mmap.start = bswap_64(event->mmap.start);
 313         event->mmap.len   = bswap_64(event->mmap.len);
 314         event->mmap.pgoff = bswap_64(event->mmap.pgoff);
 315
 316         if (sample_id_all) {
 317                 void *data = &event->mmap.filename;
 318
 319                 data += PERF_ALIGN(strlen(data) + 1, sizeof(u64));
 320                 swap_sample_id_all(event, data);
 321         }
 322 }
 323
 324 static void perf_event__mmap2_swap(union perf_event *event,
 325                                   bool sample_id_all)
 326 {
 327         event->mmap2.pid   = bswap_32(event->mmap2.pid);
 328         event->mmap2.tid   = bswap_32(event->mmap2.tid);
 329         event->mmap2.start = bswap_64(event->mmap2.start);
 330         event->mmap2.len   = bswap_64(event->mmap2.len);
 331         event->mmap2.pgoff = bswap_64(event->mmap2.pgoff);
 332         event->mmap2.maj   = bswap_32(event->mmap2.maj);
 333         event->mmap2.min   = bswap_32(event->mmap2.min);
 334         event->mmap2.ino   = bswap_64(event->mmap2.ino);
 335
 336         if (sample_id_all) {
 337                 void *data = &event->mmap2.filename;
 338
 339                 data += PERF_ALIGN(strlen(data) + 1, sizeof(u64));
 340                 swap_sample_id_all(event, data);
 341         }
 342 }
 343 static void perf_event__task_swap(union perf_event *event, bool sample_id_all)
 344 {
 345         event->fork.pid  = bswap_32(event->fork.pid);
 346         event->fork.tid  = bswap_32(event->fork.tid);
 347         event->fork.ppid = bswap_32(event->fork.ppid);
 348         event->fork.ptid = bswap_32(event->fork.ptid);
 349         event->fork.time = bswap_64(event->fork.time);
 350
 351         if (sample_id_all)
 352                 swap_sample_id_all(event, &event->fork + 1);
 353 }
 354
 355 static void perf_event__read_swap(union perf_event *event, bool sample_id_all)
 356 {
 357         event->read.pid          = bswap_32(event->read.pid);
 358         event->read.tid          = bswap_32(event->read.tid);
 359         event->read.value        = bswap_64(event->read.value);
 360         event->read.time_enabled = bswap_64(event->read.time_enabled);
 361         event->read.time_running = bswap_64(event->read.time_running);
 362         event->read.id           = bswap_64(event->read.id);
 363
 364         if (sample_id_all)
 365                 swap_sample_id_all(event, &event->read + 1);
 366 }
 367
 368 static void perf_event__throttle_swap(union perf_event *event,
 369                                       bool sample_id_all)
 370 {
 371         event->throttle.time      = bswap_64(event->throttle.time);
 372         event->throttle.id        = bswap_64(event->throttle.id);
 373         event->throttle.stream_id = bswap_64(event->throttle.stream_id);
 374
 375         if (sample_id_all)
 376                 swap_sample_id_all(event, &event->throttle + 1);
 377 }
 378
 379 static u8 revbyte(u8 b)
 380 {
 381         int rev = (b >> 4) | ((b & 0xf) << 4);
 382         rev = ((rev & 0xcc) >> 2) | ((rev & 0x33) << 2);
 383         rev = ((rev & 0xaa) >> 1) | ((rev & 0x55) << 1);
 384         return (u8) rev;
 385 }
 386
 387 /*
 388  * XXX this is hack in attempt to carry flags bitfield
 389  * throught endian village. ABI says:
 390  *
 391  * Bit-fields are allocated from right to left (least to most significant)
 392  * on little-endian implementations and from left to right (most to least
 393  * significant) on big-endian implementations.
 394  *
 395  * The above seems to be byte specific, so we need to reverse each
 396  * byte of the bitfield. 'Internet' also says this might be implementation
 397  * specific and we probably need proper fix and carry perf_event_attr
 398  * bitfield flags in separate data file FEAT_ section. Thought this seems
 399  * to work for now.
 400  */
 401 static void swap_bitfield(u8 *p, unsigned len)
 402 {
 403         unsigned i;
 404
 405         for (i = 0; i < len; i++) {
 406                 *p = revbyte(*p);
 407                 p++;
 408         }
 409 }
 410
 411 /* exported for swapping attributes in file header */
 412 void perf_event__attr_swap(struct perf_event_attr *attr)
 413 {
 414         attr->type              = bswap_32(attr->type);
 415         attr->size              = bswap_32(attr->size);
 416         attr->config            = bswap_64(attr->config);
 417         attr->sample_period     = bswap_64(attr->sample_period);
 418         attr->sample_type       = bswap_64(attr->sample_type);
 419         attr->read_format       = bswap_64(attr->read_format);
 420         attr->wakeup_events     = bswap_32(attr->wakeup_events);
 421         attr->bp_type           = bswap_32(attr->bp_type);
 422         attr->bp_addr           = bswap_64(attr->bp_addr);
 423         attr->bp_len            = bswap_64(attr->bp_len);
 424         attr->branch_sample_type = bswap_64(attr->branch_sample_type);
 425         attr->sample_regs_user   = bswap_64(attr->sample_regs_user);
 426         attr->sample_stack_user  = bswap_32(attr->sample_stack_user);
 427
 428         swap_bitfield((u8 *) (&attr->read_format + 1), sizeof(u64));
 429 }
 430
 431 static void perf_event__hdr_attr_swap(union perf_event *event,
 432                                       bool sample_id_all __maybe_unused)
 433 {
 434         size_t size;
 435
 436         perf_event__attr_swap(&event->attr.attr);
 437
 438         size = event->header.size;
 439         size -= (void *)&event->attr.id - (void *)event;
 440         mem_bswap_64(event->attr.id, size);
 441 }
 442
 443 static void perf_event__event_type_swap(union perf_event *event,
 444                                         bool sample_id_all __maybe_unused)
 445 {
 446         event->event_type.event_type.event_id =
 447                 bswap_64(event->event_type.event_type.event_id);
 448 }
 449
 450 static void perf_event__tracing_data_swap(union perf_event *event,
 451                                           bool sample_id_all __maybe_unused)
 452 {
 453         event->tracing_data.size = bswap_32(event->tracing_data.size);
 454 }
 455
 456 typedef void (*perf_event__swap_op)(union perf_event *event,
 457                                     bool sample_id_all);
 458
 459 static perf_event__swap_op perf_event__swap_ops[] = {
 460         [PERF_RECORD_MMAP]                = perf_event__mmap_swap,
 461         [PERF_RECORD_MMAP2]               = perf_event__mmap2_swap,
 462         [PERF_RECORD_COMM]                = perf_event__comm_swap,
 463         [PERF_RECORD_FORK]                = perf_event__task_swap,
 464         [PERF_RECORD_EXIT]                = perf_event__task_swap,
 465         [PERF_RECORD_LOST]                = perf_event__all64_swap,
 466         [PERF_RECORD_READ]                = perf_event__read_swap,
 467         [PERF_RECORD_THROTTLE]            = perf_event__throttle_swap,
 468         [PERF_RECORD_UNTHROTTLE]          = perf_event__throttle_swap,
 469         [PERF_RECORD_SAMPLE]              = perf_event__all64_swap,
 470         [PERF_RECORD_HEADER_ATTR]         = perf_event__hdr_attr_swap,
 471         [PERF_RECORD_HEADER_EVENT_TYPE]   = perf_event__event_type_swap,
 472         [PERF_RECORD_HEADER_TRACING_DATA] = perf_event__tracing_data_swap,
 473         [PERF_RECORD_HEADER_BUILD_ID]     = NULL,
 474         [PERF_RECORD_ID_INDEX]            = perf_event__all64_swap,
 475         [PERF_RECORD_HEADER_MAX]          = NULL,
 476 };
 477
 478 /*
 479  * When perf record finishes a pass on every buffers, it records this pseudo
 480  * event.
 481  * We record the max timestamp t found in the pass n.
 482  * Assuming these timestamps are monotonic across cpus, we know that if
 483  * a buffer still has events with timestamps below t, they will be all
 484  * available and then read in the pass n + 1.
 485  * Hence when we start to read the pass n + 2, we can safely flush every
 486  * events with timestamps below t.
 487  *
 488  *    ============ PASS n =================
 489  *       CPU 0         |   CPU 1
 490  *                     |
 491  *    cnt1 timestamps  |   cnt2 timestamps
 492  *          1          |         2
 493  *          2          |         3
 494  *          -          |         4  <--- max recorded
 495  *
 496  *    ============ PASS n + 1 ==============
 497  *       CPU 0         |   CPU 1
 498  *                     |
 499  *    cnt1 timestamps  |   cnt2 timestamps
 500  *          3          |         5
 501  *          4          |         6
 502  *          5          |         7 <---- max recorded
 503  *
 504  *      Flush every events below timestamp 4
 505  *
 506  *    ============ PASS n + 2 ==============
 507  *       CPU 0         |   CPU 1
 508  *                     |
 509  *    cnt1 timestamps  |   cnt2 timestamps
 510  *          6          |         8
 511  *          7          |         9
 512  *          -          |         10
 513  *
 514  *      Flush every events below timestamp 7
 515  *      etc...
 516  */
 517 static int process_finished_round(struct perf_tool *tool,
 518                                   union perf_event *event __maybe_unused,
 519                                   struct perf_session *session)
 520 {
 521         return ordered_events__flush(session, tool, OE_FLUSH__ROUND);
 522 }
 523
 524 int perf_session_queue_event(struct perf_session *s, union perf_event *event,
 525                              struct perf_tool *tool, struct perf_sample *sample,
 526                              u64 file_offset)
 527 {
 528         struct ordered_events *oe = &s->ordered_events;
 529         u64 timestamp = sample->time;
 530         struct ordered_event *new;
 531
 532         if (!timestamp || timestamp == ~0ULL)
 533                 return -ETIME;
 534
 535         if (timestamp < oe->last_flush) {
 536                 pr_oe_time(timestamp,      "out of order event\n");
 537                 pr_oe_time(oe->last_flush, "last flush, last_flush_type %d\n",
 538                            oe->last_flush_type);
 539
 540                 s->stats.nr_unordered_events++;
 541         }
 542
 543         new = ordered_events__new(oe, timestamp, event);
 544         if (!new) {
 545                 ordered_events__flush(s, tool, OE_FLUSH__HALF);
 546                 new = ordered_events__new(oe, timestamp, event);
 547         }
 548
 549         if (!new)
 550                 return -ENOMEM;
 551
 552         new->file_offset = file_offset;
 553         return 0;
 554 }
 555
 556 static void callchain__printf(struct perf_sample *sample)
 557 {
 558         unsigned int i;
 559
 560         printf("... chain: nr:%" PRIu64 "\n", sample->callchain->nr);
 561
 562         for (i = 0; i < sample->callchain->nr; i++)
 563                 printf("..... %2d: %016" PRIx64 "\n",
 564                        i, sample->callchain->ips[i]);
 565 }
 566
 567 static void branch_stack__printf(struct perf_sample *sample)
 568 {
 569         uint64_t i;
 570
 571         printf("... branch stack: nr:%" PRIu64 "\n", sample->branch_stack->nr);
 572
 573         for (i = 0; i < sample->branch_stack->nr; i++)
 574                 printf("..... %2"PRIu64": %016" PRIx64 " -> %016" PRIx64 "\n",
 575                         i, sample->branch_stack->entries[i].from,
 576                         sample->branch_stack->entries[i].to);
 577 }
 578
 579 static void regs_dump__printf(u64 mask, u64 *regs)
 580 {
 581         unsigned rid, i = 0;
 582
 583         for_each_set_bit(rid, (unsigned long *) &mask, sizeof(mask) * 8) {
 584                 u64 val = regs[i++];
 585
 586                 printf(".... %-5s 0x%" PRIx64 "\n",
 587                        perf_reg_name(rid), val);
 588         }
 589 }
 590
 591 static const char *regs_abi[] = {
 592         [PERF_SAMPLE_REGS_ABI_NONE] = "none",
 593         [PERF_SAMPLE_REGS_ABI_32] = "32-bit",
 594         [PERF_SAMPLE_REGS_ABI_64] = "64-bit",
 595 };
 596
 597 static inline const char *regs_dump_abi(struct regs_dump *d)
 598 {
 599         if (d->abi > PERF_SAMPLE_REGS_ABI_64)
 600                 return "unknown";
 601
 602         return regs_abi[d->abi];
 603 }
 604
 605 static void regs__printf(const char *type, struct regs_dump *regs)
 606 {
 607         u64 mask = regs->mask;
 608
 609         printf("... %s regs: mask 0x%" PRIx64 " ABI %s\n",
 610                type,
 611                mask,
 612                regs_dump_abi(regs));
 613
 614         regs_dump__printf(mask, regs->regs);
 615 }
 616
 617 static void regs_user__printf(struct perf_sample *sample)
 618 {
 619         struct regs_dump *user_regs = &sample->user_regs;
 620
 621         if (user_regs->regs)
 622                 regs__printf("user", user_regs);
 623 }
 624
 625 static void regs_intr__printf(struct perf_sample *sample)
 626 {
 627         struct regs_dump *intr_regs = &sample->intr_regs;
 628
 629         if (intr_regs->regs)
 630                 regs__printf("intr", intr_regs);
 631 }
 632
 633 static void stack_user__printf(struct stack_dump *dump)
 634 {
 635         printf("... ustack: size %" PRIu64 ", offset 0x%x\n",
 636                dump->size, dump->offset);
 637 }
 638
 639 static void perf_session__print_tstamp(struct perf_session *session,
 640                                        union perf_event *event,
 641                                        struct perf_sample *sample)
 642 {
 643         u64 sample_type = __perf_evlist__combined_sample_type(session->evlist);
 644
 645         if (event->header.type != PERF_RECORD_SAMPLE &&
 646             !perf_evlist__sample_id_all(session->evlist)) {
 647                 fputs("-1 -1 ", stdout);
 648                 return;
 649         }
 650
 651         if ((sample_type & PERF_SAMPLE_CPU))
 652                 printf("%u ", sample->cpu);
 653
 654         if (sample_type & PERF_SAMPLE_TIME)
 655                 printf("%" PRIu64 " ", sample->time);
 656 }
 657
 658 static void sample_read__printf(struct perf_sample *sample, u64 read_format)
 659 {
 660         printf("... sample_read:\n");
 661
 662         if (read_format & PERF_FORMAT_TOTAL_TIME_ENABLED)
 663                 printf("...... time enabled %016" PRIx64 "\n",
 664                        sample->read.time_enabled);
 665
 666         if (read_format & PERF_FORMAT_TOTAL_TIME_RUNNING)
 667                 printf("...... time running %016" PRIx64 "\n",
 668                        sample->read.time_running);
 669
 670         if (read_format & PERF_FORMAT_GROUP) {
 671                 u64 i;
 672
 673                 printf(".... group nr %" PRIu64 "\n", sample->read.group.nr);
 674
 675                 for (i = 0; i < sample->read.group.nr; i++) {
 676                         struct sample_read_value *value;
 677
 678                         value = &sample->read.group.values[i];
 679                         printf("..... id %016" PRIx64
 680                                ", value %016" PRIx64 "\n",
 681                                value->id, value->value);
 682                 }
 683         } else
 684                 printf("..... id %016" PRIx64 ", value %016" PRIx64 "\n",
 685                         sample->read.one.id, sample->read.one.value);
 686 }
 687
 688 static void dump_event(struct perf_session *session, union perf_event *event,
 689                        u64 file_offset, struct perf_sample *sample)
 690 {
 691         if (!dump_trace)
 692                 return;
 693
 694         printf("\n%#" PRIx64 " [%#x]: event: %d\n",
 695                file_offset, event->header.size, event->header.type);
 696
 697         trace_event(event);
 698
 699         if (sample)
 700                 perf_session__print_tstamp(session, event, sample);
 701
 702         printf("%#" PRIx64 " [%#x]: PERF_RECORD_%s", file_offset,
 703                event->header.size, perf_event__name(event->header.type));
 704 }
 705
 706 static void dump_sample(struct perf_evsel *evsel, union perf_event *event,
 707                         struct perf_sample *sample)
 708 {
 709         u64 sample_type;
 710
 711         if (!dump_trace)
 712                 return;
 713
 714         printf("(IP, 0x%x): %d/%d: %#" PRIx64 " period: %" PRIu64 " addr: %#" PRIx64 "\n",
 715                event->header.misc, sample->pid, sample->tid, sample->ip,
 716                sample->period, sample->addr);
 717
 718         sample_type = evsel->attr.sample_type;
 719
 720         if (sample_type & PERF_SAMPLE_CALLCHAIN)
 721                 callchain__printf(sample);
 722
 723         if (sample_type & PERF_SAMPLE_BRANCH_STACK)
 724                 branch_stack__printf(sample);
 725
 726         if (sample_type & PERF_SAMPLE_REGS_USER)
 727                 regs_user__printf(sample);
 728
 729         if (sample_type & PERF_SAMPLE_REGS_INTR)
 730                 regs_intr__printf(sample);
 731
 732         if (sample_type & PERF_SAMPLE_STACK_USER)
 733                 stack_user__printf(&sample->user_stack);
 734
 735         if (sample_type & PERF_SAMPLE_WEIGHT)
 736                 printf("... weight: %" PRIu64 "\n", sample->weight);
 737
 738         if (sample_type & PERF_SAMPLE_DATA_SRC)
 739                 printf(" . data_src: 0x%"PRIx64"\n", sample->data_src);
 740
 741         if (sample_type & PERF_SAMPLE_TRANSACTION)
 742                 printf("... transaction: %" PRIx64 "\n", sample->transaction);
 743
 744         if (sample_type & PERF_SAMPLE_READ)
 745                 sample_read__printf(sample, evsel->attr.read_format);
 746 }
 747
 748 static struct machine *
 749         perf_session__find_machine_for_cpumode(struct perf_session *session,
 750                                                union perf_event *event,
 751                                                struct perf_sample *sample)
 752 {
 753         const u8 cpumode = event->header.misc & PERF_RECORD_MISC_CPUMODE_MASK;
 754         struct machine *machine;
 755
 756         if (perf_guest &&
 757             ((cpumode == PERF_RECORD_MISC_GUEST_KERNEL) ||
 758              (cpumode == PERF_RECORD_MISC_GUEST_USER))) {
 759                 u32 pid;
 760
 761                 if (event->header.type == PERF_RECORD_MMAP
 762                     || event->header.type == PERF_RECORD_MMAP2)
 763                         pid = event->mmap.pid;
 764                 else
 765                         pid = sample->pid;
 766
 767                 machine = perf_session__find_machine(session, pid);
 768                 if (!machine)
 769                         machine = perf_session__findnew_machine(session,
 770                                                 DEFAULT_GUEST_KERNEL_ID);
 771                 return machine;
 772         }
 773
 774         return &session->machines.host;
 775 }
 776
 777 static int deliver_sample_value(struct perf_session *session,
 778                                 struct perf_tool *tool,
 779                                 union perf_event *event,
 780                                 struct perf_sample *sample,
 781                                 struct sample_read_value *v,
 782                                 struct machine *machine)
 783 {
 784         struct perf_sample_id *sid;
 785
 786         sid = perf_evlist__id2sid(session->evlist, v->id);
 787         if (sid) {
 788                 sample->id     = v->id;
 789                 sample->period = v->value - sid->period;
 790                 sid->period    = v->value;
 791         }
 792
 793         if (!sid || sid->evsel == NULL) {
 794                 ++session->stats.nr_unknown_id;
 795                 return 0;
 796         }
 797
 798         return tool->sample(tool, event, sample, sid->evsel, machine);
 799 }
 800
 801 static int deliver_sample_group(struct perf_session *session,
 802                                 struct perf_tool *tool,
 803                                 union  perf_event *event,
 804                                 struct perf_sample *sample,
 805                                 struct machine *machine)
 806 {
 807         int ret = -EINVAL;
 808         u64 i;
 809
 810         for (i = 0; i < sample->read.group.nr; i++) {
 811                 ret = deliver_sample_value(session, tool, event, sample,
 812                                            &sample->read.group.values[i],
 813                                            machine);
 814                 if (ret)
 815                         break;
 816         }
 817
 818         return ret;
 819 }
 820
 821 static int
 822 perf_session__deliver_sample(struct perf_session *session,
 823                              struct perf_tool *tool,
 824                              union  perf_event *event,
 825                              struct perf_sample *sample,
 826                              struct perf_evsel *evsel,
 827                              struct machine *machine)
 828 {
 829         /* We know evsel != NULL. */
 830         u64 sample_type = evsel->attr.sample_type;
 831         u64 read_format = evsel->attr.read_format;
 832
 833         /* Standard sample delievery. */
 834         if (!(sample_type & PERF_SAMPLE_READ))
 835                 return tool->sample(tool, event, sample, evsel, machine);
 836
 837         /* For PERF_SAMPLE_READ we have either single or group mode. */
 838         if (read_format & PERF_FORMAT_GROUP)
 839                 return deliver_sample_group(session, tool, event, sample,
 840                                             machine);
 841         else
 842                 return deliver_sample_value(session, tool, event, sample,
 843                                             &sample->read.one, machine);
 844 }
 845
 846 int perf_session__deliver_event(struct perf_session *session,
 847                                 union perf_event *event,
 848                                 struct perf_sample *sample,
 849                                 struct perf_tool *tool, u64 file_offset)
 850 {
 851         struct perf_evsel *evsel;
 852         struct machine *machine;
 853
 854         dump_event(session, event, file_offset, sample);
 855
 856         evsel = perf_evlist__id2evsel(session->evlist, sample->id);
 857
 858         machine = perf_session__find_machine_for_cpumode(session, event,
 859                                                          sample);
 860
 861         switch (event->header.type) {
 862         case PERF_RECORD_SAMPLE:
 863                 dump_sample(evsel, event, sample);
 864                 if (evsel == NULL) {
 865                         ++session->stats.nr_unknown_id;
 866                         return 0;
 867                 }
 868                 if (machine == NULL) {
 869                         ++session->stats.nr_unprocessable_samples;
 870                         return 0;
 871                 }
 872                 return perf_session__deliver_sample(session, tool, event,
 873                                                     sample, evsel, machine);
 874         case PERF_RECORD_MMAP:
 875                 return tool->mmap(tool, event, sample, machine);
 876         case PERF_RECORD_MMAP2:
 877                 return tool->mmap2(tool, event, sample, machine);
 878         case PERF_RECORD_COMM:
 879                 return tool->comm(tool, event, sample, machine);
 880         case PERF_RECORD_FORK:
 881                 return tool->fork(tool, event, sample, machine);
 882         case PERF_RECORD_EXIT:
 883                 return tool->exit(tool, event, sample, machine);
 884         case PERF_RECORD_LOST:
 885                 if (tool->lost == perf_event__process_lost)
 886                         session->stats.total_lost += event->lost.lost;
 887                 return tool->lost(tool, event, sample, machine);
 888         case PERF_RECORD_READ:
 889                 return tool->read(tool, event, sample, evsel, machine);
 890         case PERF_RECORD_THROTTLE:
 891                 return tool->throttle(tool, event, sample, machine);
 892         case PERF_RECORD_UNTHROTTLE:
 893                 return tool->unthrottle(tool, event, sample, machine);
 894         default:
 895                 ++session->stats.nr_unknown_events;
 896                 return -1;
 897         }
 898 }
 899
 900 static s64 perf_session__process_user_event(struct perf_session *session,
 901                                             union perf_event *event,
 902                                             struct perf_tool *tool,
 903                                             u64 file_offset)
 904 {
 905         int fd = perf_data_file__fd(session->file);
 906         int err;
 907
 908         dump_event(session, event, file_offset, NULL);
 909
 910         /* These events are processed right away */
 911         switch (event->header.type) {
 912         case PERF_RECORD_HEADER_ATTR:
 913                 err = tool->attr(tool, event, &session->evlist);
 914                 if (err == 0) {
 915                         perf_session__set_id_hdr_size(session);
 916                         perf_session__set_comm_exec(session);
 917                 }
 918                 return err;
 919         case PERF_RECORD_HEADER_EVENT_TYPE:
 920                 /*
 921                  * Depreceated, but we need to handle it for sake
 922                  * of old data files create in pipe mode.
 923                  */
 924                 return 0;
 925         case PERF_RECORD_HEADER_TRACING_DATA:
 926                 /* setup for reading amidst mmap */
 927                 lseek(fd, file_offset, SEEK_SET);
 928                 return tool->tracing_data(tool, event, session);
 929         case PERF_RECORD_HEADER_BUILD_ID:
 930                 return tool->build_id(tool, event, session);
 931         case PERF_RECORD_FINISHED_ROUND:
 932                 return tool->finished_round(tool, event, session);
 933         case PERF_RECORD_ID_INDEX:
 934                 return tool->id_index(tool, event, session);
 935         default:
 936                 return -EINVAL;
 937         }
 938 }
 939
 940 int perf_session__deliver_synth_event(struct perf_session *session,
 941                                       union perf_event *event,
 942                                       struct perf_sample *sample,
 943                                       struct perf_tool *tool)
 944 {
 945         events_stats__inc(&session->stats, event->header.type);
 946
 947         if (event->header.type >= PERF_RECORD_USER_TYPE_START)
 948                 return perf_session__process_user_event(session, event, tool, 0);
 949
 950         return perf_session__deliver_event(session, event, sample, tool, 0);
 951 }
 952
 953 static void event_swap(union perf_event *event, bool sample_id_all)
 954 {
 955         perf_event__swap_op swap;
 956
 957         swap = perf_event__swap_ops[event->header.type];
 958         if (swap)
 959                 swap(event, sample_id_all);
 960 }
 961
 962 int perf_session__peek_event(struct perf_session *session, off_t file_offset,
 963                              void *buf, size_t buf_sz,
 964                              union perf_event **event_ptr,
 965                              struct perf_sample *sample)
 966 {
 967         union perf_event *event;
 968         size_t hdr_sz, rest;
 969         int fd;
 970
 971         if (session->one_mmap && !session->header.needs_swap) {
 972                 event = file_offset - session->one_mmap_offset +
 973                         session->one_mmap_addr;
 974                 goto out_parse_sample;
 975         }
 976
 977         if (perf_data_file__is_pipe(session->file))
 978                 return -1;
 979
 980         fd = perf_data_file__fd(session->file);
 981         hdr_sz = sizeof(struct perf_event_header);
 982
 983         if (buf_sz < hdr_sz)
 984                 return -1;
 985
 986         if (lseek(fd, file_offset, SEEK_SET) == (off_t)-1 ||
 987             readn(fd, &buf, hdr_sz) != (ssize_t)hdr_sz)
 988                 return -1;
 989
 990         event = (union perf_event *)buf;
 991
 992         if (session->header.needs_swap)
 993                 perf_event_header__bswap(&event->header);
 994
 995         if (event->header.size < hdr_sz)
 996                 return -1;
 997
 998         rest = event->header.size - hdr_sz;
 999
1000         if (readn(fd, &buf, rest) != (ssize_t)rest)
1001                 return -1;
1002
1003         if (session->header.needs_swap)
1004                 event_swap(event, perf_evlist__sample_id_all(session->evlist));
1005
1006 out_parse_sample:
1007
1008         if (sample && event->header.type < PERF_RECORD_USER_TYPE_START &&
1009             perf_evlist__parse_sample(session->evlist, event, sample))
1010                 return -1;
1011
1012         *event_ptr = event;
1013
1014         return 0;
1015 }
1016
1017 static s64 perf_session__process_event(struct perf_session *session,
1018                                        union perf_event *event,
1019                                        struct perf_tool *tool,
1020                                        u64 file_offset)
1021 {
1022         struct perf_sample sample;
1023         int ret;
1024
1025         if (session->header.needs_swap)
1026                 event_swap(event, perf_evlist__sample_id_all(session->evlist));
1027
1028         if (event->header.type >= PERF_RECORD_HEADER_MAX)
1029                 return -EINVAL;
1030
1031         events_stats__inc(&session->stats, event->header.type);
1032
1033         if (event->header.type >= PERF_RECORD_USER_TYPE_START)
1034                 return perf_session__process_user_event(session, event, tool, file_offset);
1035
1036         /*
1037          * For all kernel events we get the sample data
1038          */
1039         ret = perf_evlist__parse_sample(session->evlist, event, &sample);
1040         if (ret)
1041                 return ret;
1042
1043         if (tool->ordered_events) {
1044                 ret = perf_session_queue_event(session, event, tool, &sample,
1045                                                file_offset);
1046                 if (ret != -ETIME)
1047                         return ret;
1048         }
1049
1050         return perf_session__deliver_event(session, event, &sample, tool,
1051                                            file_offset);
1052 }
1053
1054 void perf_event_header__bswap(struct perf_event_header *hdr)
1055 {
1056         hdr->type = bswap_32(hdr->type);
1057         hdr->misc = bswap_16(hdr->misc);
1058         hdr->size = bswap_16(hdr->size);
1059 }
1060
1061 struct thread *perf_session__findnew(struct perf_session *session, pid_t pid)
1062 {
1063         return machine__findnew_thread(&session->machines.host, -1, pid);
1064 }
1065
1066 static struct thread *perf_session__register_idle_thread(struct perf_session *session)
1067 {
1068         struct thread *thread;
1069
1070         thread = machine__findnew_thread(&session->machines.host, 0, 0);
1071         if (thread == NULL || thread__set_comm(thread, "swapper", 0)) {
1072                 pr_err("problem inserting idle task.\n");
1073                 thread = NULL;
1074         }
1075
1076         return thread;
1077 }
1078
1079 static void perf_session__warn_about_errors(const struct perf_session *session,
1080                                             const struct perf_tool *tool)
1081 {
1082         if (tool->lost == perf_event__process_lost &&
1083             session->stats.nr_events[PERF_RECORD_LOST] != 0) {
1084                 ui__warning("Processed %d events and lost %d chunks!\n\n"
1085                             "Check IO/CPU overload!\n\n",
1086                             session->stats.nr_events[0],
1087                             session->stats.nr_events[PERF_RECORD_LOST]);
1088         }
1089
1090         if (session->stats.nr_unknown_events != 0) {
1091                 ui__warning("Found %u unknown events!\n\n"
1092                             "Is this an older tool processing a perf.data "
1093                             "file generated by a more recent tool?\n\n"
1094                             "If that is not the case, consider "
1095                             "reporting to [email protected].\n\n",
1096                             session->stats.nr_unknown_events);
1097         }
1098
1099         if (session->stats.nr_unknown_id != 0) {
1100                 ui__warning("%u samples with id not present in the header\n",
1101                             session->stats.nr_unknown_id);
1102         }
1103
1104         if (session->stats.nr_invalid_chains != 0) {
1105                 ui__warning("Found invalid callchains!\n\n"
1106                             "%u out of %u events were discarded for this reason.\n\n"
1107                             "Consider reporting to [email protected].\n\n",
1108                             session->stats.nr_invalid_chains,
1109                             session->stats.nr_events[PERF_RECORD_SAMPLE]);
1110         }
1111
1112         if (session->stats.nr_unprocessable_samples != 0) {
1113                 ui__warning("%u unprocessable samples recorded.\n"
1114                             "Do you have a KVM guest running and not using 'perf kvm'?\n",
1115                             session->stats.nr_unprocessable_samples);
1116         }
1117
1118         if (session->stats.nr_unordered_events != 0)
1119                 ui__warning("%u out of order events recorded.\n", session->stats.nr_unordered_events);
1120 }
1121
1122 volatile int session_done;
1123
1124 static int __perf_session__process_pipe_events(struct perf_session *session,
1125                                                struct perf_tool *tool)
1126 {
1127         int fd = perf_data_file__fd(session->file);
1128         union perf_event *event;
1129         uint32_t size, cur_size = 0;
1130         void *buf = NULL;
1131         s64 skip = 0;
1132         u64 head;
1133         ssize_t err;
1134         void *p;
1135
1136         perf_tool__fill_defaults(tool);
1137
1138         head = 0;
1139         cur_size = sizeof(union perf_event);
1140
1141         buf = malloc(cur_size);
1142         if (!buf)
1143                 return -errno;
1144 more:
1145         event = buf;
1146         err = readn(fd, event, sizeof(struct perf_event_header));
1147         if (err <= 0) {
1148                 if (err == 0)
1149                         goto done;
1150
1151                 pr_err("failed to read event header\n");
1152                 goto out_err;
1153         }
1154
1155         if (session->header.needs_swap)
1156                 perf_event_header__bswap(&event->header);
1157
1158         size = event->header.size;
1159         if (size < sizeof(struct perf_event_header)) {
1160                 pr_err("bad event header size\n");
1161                 goto out_err;
1162         }
1163
1164         if (size > cur_size) {
1165                 void *new = realloc(buf, size);
1166                 if (!new) {
1167                         pr_err("failed to allocate memory to read event\n");
1168                         goto out_err;
1169                 }
1170                 buf = new;
1171                 cur_size = size;
1172                 event = buf;
1173         }
1174         p = event;
1175         p += sizeof(struct perf_event_header);
1176
1177         if (size - sizeof(struct perf_event_header)) {
1178                 err = readn(fd, p, size - sizeof(struct perf_event_header));
1179                 if (err <= 0) {
1180                         if (err == 0) {
1181                                 pr_err("unexpected end of event stream\n");
1182                                 goto done;
1183                         }
1184
1185                         pr_err("failed to read event data\n");
1186                         goto out_err;
1187                 }
1188         }
1189
1190         if ((skip = perf_session__process_event(session, event, tool, head)) < 0) {
1191                 pr_err("%#" PRIx64 " [%#x]: failed to process type: %d\n",
1192                        head, event->header.size, event->header.type);
1193                 err = -EINVAL;
1194                 goto out_err;
1195         }
1196
1197         head += size;
1198
1199         if (skip > 0)
1200                 head += skip;
1201
1202         if (!session_done())
1203                 goto more;
1204 done:
1205         /* do the final flush for ordered samples */
1206         err = ordered_events__flush(session, tool, OE_FLUSH__FINAL);
1207 out_err:
1208         free(buf);
1209         perf_session__warn_about_errors(session, tool);
1210         ordered_events__free(&session->ordered_events);
1211         return err;
1212 }
1213
1214 static union perf_event *
1215 fetch_mmaped_event(struct perf_session *session,
1216                    u64 head, size_t mmap_size, char *buf)
1217 {
1218         union perf_event *event;
1219
1220         /*
1221          * Ensure we have enough space remaining to read
1222          * the size of the event in the headers.
1223          */
1224         if (head + sizeof(event->header) > mmap_size)
1225                 return NULL;
1226
1227         event = (union perf_event *)(buf + head);
1228
1229         if (session->header.needs_swap)
1230                 perf_event_header__bswap(&event->header);
1231
1232         if (head + event->header.size > mmap_size) {
1233                 /* We're not fetching the event so swap back again */
1234                 if (session->header.needs_swap)
1235                         perf_event_header__bswap(&event->header);
1236                 return NULL;
1237         }
1238
1239         return event;
1240 }
1241
1242 /*
1243  * On 64bit we can mmap the data file in one go. No need for tiny mmap
1244  * slices. On 32bit we use 32MB.
1245  */
1246 #if BITS_PER_LONG == 64
1247 #define MMAP_SIZE ULLONG_MAX
1248 #define NUM_MMAPS 1
1249 #else
1250 #define MMAP_SIZE (32 * 1024 * 1024ULL)
1251 #define NUM_MMAPS 128
1252 #endif
1253
1254 static int __perf_session__process_events(struct perf_session *session,
1255                                           u64 data_offset, u64 data_size,
1256                                           u64 file_size, struct perf_tool *tool)
1257 {
1258         int fd = perf_data_file__fd(session->file);
1259         u64 head, page_offset, file_offset, file_pos, size;
1260         int err, mmap_prot, mmap_flags, map_idx = 0;
1261         size_t  mmap_size;
1262         char *buf, *mmaps[NUM_MMAPS];
1263         union perf_event *event;
1264         struct ui_progress prog;
1265         s64 skip;
1266
1267         perf_tool__fill_defaults(tool);
1268
1269         page_offset = page_size * (data_offset / page_size);
1270         file_offset = page_offset;
1271         head = data_offset - page_offset;
1272
1273         if (data_size && (data_offset + data_size < file_size))
1274                 file_size = data_offset + data_size;
1275
1276         ui_progress__init(&prog, file_size, "Processing events...");
1277
1278         mmap_size = MMAP_SIZE;
1279         if (mmap_size > file_size) {
1280                 mmap_size = file_size;
1281                 session->one_mmap = true;
1282         }
1283
1284         memset(mmaps, 0, sizeof(mmaps));
1285
1286         mmap_prot  = PROT_READ;
1287         mmap_flags = MAP_SHARED;
1288
1289         if (session->header.needs_swap) {
1290                 mmap_prot  |= PROT_WRITE;
1291                 mmap_flags = MAP_PRIVATE;
1292         }
1293 remap:
1294         buf = mmap(NULL, mmap_size, mmap_prot, mmap_flags, fd,
1295                    file_offset);
1296         if (buf == MAP_FAILED) {
1297                 pr_err("failed to mmap file\n");
1298                 err = -errno;
1299                 goto out_err;
1300         }
1301         mmaps[map_idx] = buf;
1302         map_idx = (map_idx + 1) & (ARRAY_SIZE(mmaps) - 1);
1303         file_pos = file_offset + head;
1304         if (session->one_mmap) {
1305                 session->one_mmap_addr = buf;
1306                 session->one_mmap_offset = file_offset;
1307         }
1308
1309 more:
1310         event = fetch_mmaped_event(session, head, mmap_size, buf);
1311         if (!event) {
1312                 if (mmaps[map_idx]) {
1313                         munmap(mmaps[map_idx], mmap_size);
1314                         mmaps[map_idx] = NULL;
1315                 }
1316
1317                 page_offset = page_size * (head / page_size);
1318                 file_offset += page_offset;
1319                 head -= page_offset;
1320                 goto remap;
1321         }
1322
1323         size = event->header.size;
1324
1325         if (size < sizeof(struct perf_event_header) ||
1326             (skip = perf_session__process_event(session, event, tool, file_pos))
1327                                                                         < 0) {
1328                 pr_err("%#" PRIx64 " [%#x]: failed to process type: %d\n",
1329                        file_offset + head, event->header.size,
1330                        event->header.type);
1331                 err = -EINVAL;
1332                 goto out_err;
1333         }
1334
1335         if (skip)
1336                 size += skip;
1337
1338         head += size;
1339         file_pos += size;
1340
1341         ui_progress__update(&prog, size);
1342
1343         if (session_done())
1344                 goto out;
1345
1346         if (file_pos < file_size)
1347                 goto more;
1348
1349 out:
1350         /* do the final flush for ordered samples */
1351         err = ordered_events__flush(session, tool, OE_FLUSH__FINAL);
1352 out_err:
1353         ui_progress__finish();
1354         perf_session__warn_about_errors(session, tool);
1355         ordered_events__free(&session->ordered_events);
1356         session->one_mmap = false;
1357         return err;
1358 }
1359
1360 int perf_session__process_events(struct perf_session *session,
1361                                  struct perf_tool *tool)
1362 {
1363         u64 size = perf_data_file__size(session->file);
1364         int err;
1365
1366         if (perf_session__register_idle_thread(session) == NULL)
1367                 return -ENOMEM;
1368
1369         if (!perf_data_file__is_pipe(session->file))
1370                 err = __perf_session__process_events(session,
1371                                                      session->header.data_offset,
1372                                                      session->header.data_size,
1373                                                      size, tool);
1374         else
1375                 err = __perf_session__process_pipe_events(session, tool);
1376
1377         return err;
1378 }
1379
1380 bool perf_session__has_traces(struct perf_session *session, const char *msg)
1381 {
1382         struct perf_evsel *evsel;
1383
1384         evlist__for_each(session->evlist, evsel) {
1385                 if (evsel->attr.type == PERF_TYPE_TRACEPOINT)
1386                         return true;
1387         }
1388
1389         pr_err("No trace sample to read. Did you call 'perf %s'?\n", msg);
1390         return false;
1391 }
1392
1393 int maps__set_kallsyms_ref_reloc_sym(struct map **maps,
1394                                      const char *symbol_name, u64 addr)
1395 {
1396         char *bracket;
1397         enum map_type i;
1398         struct ref_reloc_sym *ref;
1399
1400         ref = zalloc(sizeof(struct ref_reloc_sym));
1401         if (ref == NULL)
1402                 return -ENOMEM;
1403
1404         ref->name = strdup(symbol_name);
1405         if (ref->name == NULL) {
1406                 free(ref);
1407                 return -ENOMEM;
1408         }
1409
1410         bracket = strchr(ref->name, ']');
1411         if (bracket)
1412                 *bracket = '\0';
1413
1414         ref->addr = addr;
1415
1416         for (i = 0; i < MAP__NR_TYPES; ++i) {
1417                 struct kmap *kmap = map__kmap(maps[i]);
1418                 kmap->ref_reloc_sym = ref;
1419         }
1420
1421         return 0;
1422 }
1423
1424 size_t perf_session__fprintf_dsos(struct perf_session *session, FILE *fp)
1425 {
1426         return machines__fprintf_dsos(&session->machines, fp);
1427 }
1428
1429 size_t perf_session__fprintf_dsos_buildid(struct perf_session *session, FILE *fp,
1430                                           bool (skip)(struct dso *dso, int parm), int parm)
1431 {
1432         return machines__fprintf_dsos_buildid(&session->machines, fp, skip, parm);
1433 }
1434
1435 size_t perf_session__fprintf_nr_events(struct perf_session *session, FILE *fp)
1436 {
1437         size_t ret = fprintf(fp, "Aggregated stats:\n");
1438
1439         ret += events_stats__fprintf(&session->stats, fp);
1440         return ret;
1441 }
1442
1443 size_t perf_session__fprintf(struct perf_session *session, FILE *fp)
1444 {
1445         /*
1446          * FIXME: Here we have to actually print all the machines in this
1447          * session, not just the host...
1448          */
1449         return machine__fprintf(&session->machines.host, fp);
1450 }
1451
1452 struct perf_evsel *perf_session__find_first_evtype(struct perf_session *session,
1453                                               unsigned int type)
1454 {
1455         struct perf_evsel *pos;
1456
1457         evlist__for_each(session->evlist, pos) {
1458                 if (pos->attr.type == type)
1459                         return pos;
1460         }
1461         return NULL;
1462 }
1463
1464 void perf_evsel__print_ip(struct perf_evsel *evsel, struct perf_sample *sample,
1465                           struct addr_location *al,
1466                           unsigned int print_opts, unsigned int stack_depth)
1467 {
1468         struct callchain_cursor_node *node;
1469         int print_ip = print_opts & PRINT_IP_OPT_IP;
1470         int print_sym = print_opts & PRINT_IP_OPT_SYM;
1471         int print_dso = print_opts & PRINT_IP_OPT_DSO;
1472         int print_symoffset = print_opts & PRINT_IP_OPT_SYMOFFSET;
1473         int print_oneline = print_opts & PRINT_IP_OPT_ONELINE;
1474         int print_srcline = print_opts & PRINT_IP_OPT_SRCLINE;
1475         char s = print_oneline ? ' ' : '\t';
1476
1477         if (symbol_conf.use_callchain && sample->callchain) {
1478                 struct addr_location node_al;
1479
1480                 if (thread__resolve_callchain(al->thread, evsel,
1481                                               sample, NULL, NULL,
1482                                               PERF_MAX_STACK_DEPTH) != 0) {
1483                         if (verbose)
1484                                 error("Failed to resolve callchain. Skipping\n");
1485                         return;
1486                 }
1487                 callchain_cursor_commit(&callchain_cursor);
1488
1489                 if (print_symoffset)
1490                         node_al = *al;
1491
1492                 while (stack_depth) {
1493                         u64 addr = 0;
1494
1495                         node = callchain_cursor_current(&callchain_cursor);
1496                         if (!node)
1497                                 break;
1498
1499                         if (node->sym && node->sym->ignore)
1500                                 goto next;
1501
1502                         if (print_ip)
1503                                 printf("%c%16" PRIx64, s, node->ip);
1504
1505                         if (node->map)
1506                                 addr = node->map->map_ip(node->map, node->ip);
1507
1508                         if (print_sym) {
1509                                 printf(" ");
1510                                 if (print_symoffset) {
1511                                         node_al.addr = addr;
1512                                         node_al.map  = node->map;
1513                                         symbol__fprintf_symname_offs(node->sym, &node_al, stdout);
1514                                 } else
1515                                         symbol__fprintf_symname(node->sym, stdout);
1516                         }
1517
1518                         if (print_dso) {
1519                                 printf(" (");
1520                                 map__fprintf_dsoname(node->map, stdout);
1521                                 printf(")");
1522                         }
1523
1524                         if (print_srcline)
1525                                 map__fprintf_srcline(node->map, addr, "\n  ",
1526                                                      stdout);
1527
1528                         if (!print_oneline)
1529                                 printf("\n");
1530
1531                         stack_depth--;
1532 next:
1533                         callchain_cursor_advance(&callchain_cursor);
1534                 }
1535
1536         } else {
1537                 if (al->sym && al->sym->ignore)
1538                         return;
1539
1540                 if (print_ip)
1541                         printf("%16" PRIx64, sample->ip);
1542
1543                 if (print_sym) {
1544                         printf(" ");
1545                         if (print_symoffset)
1546                                 symbol__fprintf_symname_offs(al->sym, al,
1547                                                              stdout);
1548                         else
1549                                 symbol__fprintf_symname(al->sym, stdout);
1550                 }
1551
1552                 if (print_dso) {
1553                         printf(" (");
1554                         map__fprintf_dsoname(al->map, stdout);
1555                         printf(")");
1556                 }
1557
1558                 if (print_srcline)
1559                         map__fprintf_srcline(al->map, al->addr, "\n  ", stdout);
1560         }
1561 }
1562
1563 int perf_session__cpu_bitmap(struct perf_session *session,
1564                              const char *cpu_list, unsigned long *cpu_bitmap)
1565 {
1566         int i, err = -1;
1567         struct cpu_map *map;
1568
1569         for (i = 0; i < PERF_TYPE_MAX; ++i) {
1570                 struct perf_evsel *evsel;
1571
1572                 evsel = perf_session__find_first_evtype(session, i);
1573                 if (!evsel)
1574                         continue;
1575
1576                 if (!(evsel->attr.sample_type & PERF_SAMPLE_CPU)) {
1577                         pr_err("File does not contain CPU events. "
1578                                "Remove -c option to proceed.\n");
1579                         return -1;
1580                 }
1581         }
1582
1583         map = cpu_map__new(cpu_list);
1584         if (map == NULL) {
1585                 pr_err("Invalid cpu_list\n");
1586                 return -1;
1587         }
1588
1589         for (i = 0; i < map->nr; i++) {
1590                 int cpu = map->map[i];
1591
1592                 if (cpu >= MAX_NR_CPUS) {
1593                         pr_err("Requested CPU %d too large. "
1594                                "Consider raising MAX_NR_CPUS\n", cpu);
1595                         goto out_delete_map;
1596                 }
1597
1598                 set_bit(cpu, cpu_bitmap);
1599         }
1600
1601         err = 0;
1602
1603 out_delete_map:
1604         cpu_map__delete(map);
1605         return err;
1606 }
1607
1608 void perf_session__fprintf_info(struct perf_session *session, FILE *fp,
1609                                 bool full)
1610 {
1611         struct stat st;
1612         int fd, ret;
1613
1614         if (session == NULL || fp == NULL)
1615                 return;
1616
1617         fd = perf_data_file__fd(session->file);
1618
1619         ret = fstat(fd, &st);
1620         if (ret == -1)
1621                 return;
1622
1623         fprintf(fp, "# ========\n");
1624         fprintf(fp, "# captured on: %s", ctime(&st.st_ctime));
1625         perf_header__fprintf_info(session, fp, full);
1626         fprintf(fp, "# ========\n#\n");
1627 }
1628
1629
1630 int __perf_session__set_tracepoints_handlers(struct perf_session *session,
1631                                              const struct perf_evsel_str_handler *assocs,
1632                                              size_t nr_assocs)
1633 {
1634         struct perf_evsel *evsel;
1635         size_t i;
1636         int err;
1637
1638         for (i = 0; i < nr_assocs; i++) {
1639                 /*
1640                  * Adding a handler for an event not in the session,
1641                  * just ignore it.
1642                  */
1643                 evsel = perf_evlist__find_tracepoint_by_name(session->evlist, assocs[i].name);
1644                 if (evsel == NULL)
1645                         continue;
1646
1647                 err = -EEXIST;
1648                 if (evsel->handler != NULL)
1649                         goto out;
1650                 evsel->handler = assocs[i].handler;
1651         }
1652
1653         err = 0;
1654 out:
1655         return err;
1656 }
1657
1658 int perf_event__process_id_index(struct perf_tool *tool __maybe_unused,
1659                                  union perf_event *event,
1660                                  struct perf_session *session)
1661 {
1662         struct perf_evlist *evlist = session->evlist;
1663         struct id_index_event *ie = &event->id_index;
1664         size_t i, nr, max_nr;
1665
1666         max_nr = (ie->header.size - sizeof(struct id_index_event)) /
1667                  sizeof(struct id_index_entry);
1668         nr = ie->nr;
1669         if (nr > max_nr)
1670                 return -EINVAL;
1671
1672         if (dump_trace)
1673                 fprintf(stdout, " nr: %zu\n", nr);
1674
1675         for (i = 0; i < nr; i++) {
1676                 struct id_index_entry *e = &ie->entries[i];
1677                 struct perf_sample_id *sid;
1678
1679                 if (dump_trace) {
1680                         fprintf(stdout, " ... id: %"PRIu64, e->id);
1681                         fprintf(stdout, "  idx: %"PRIu64, e->idx);
1682                         fprintf(stdout, "  cpu: %"PRId64, e->cpu);
1683                         fprintf(stdout, "  tid: %"PRId64"\n", e->tid);
1684                 }
1685
1686                 sid = perf_evlist__id2sid(evlist, e->id);
1687                 if (!sid)
1688                         return -ENOENT;
1689                 sid->idx = e->idx;
1690                 sid->cpu = e->cpu;
1691                 sid->tid = e->tid;
1692         }
1693         return 0;
1694 }
1695
1696 int perf_event__synthesize_id_index(struct perf_tool *tool,
1697                                     perf_event__handler_t process,
1698                                     struct perf_evlist *evlist,
1699                                     struct machine *machine)
1700 {
1701         union perf_event *ev;
1702         struct perf_evsel *evsel;
1703         size_t nr = 0, i = 0, sz, max_nr, n;
1704         int err;
1705
1706         pr_debug2("Synthesizing id index\n");
1707
1708         max_nr = (UINT16_MAX - sizeof(struct id_index_event)) /
1709                  sizeof(struct id_index_entry);
1710
1711         evlist__for_each(evlist, evsel)
1712                 nr += evsel->ids;
1713
1714         n = nr > max_nr ? max_nr : nr;
1715         sz = sizeof(struct id_index_event) + n * sizeof(struct id_index_entry);
1716         ev = zalloc(sz);
1717         if (!ev)
1718                 return -ENOMEM;
1719
1720         ev->id_index.header.type = PERF_RECORD_ID_INDEX;
1721         ev->id_index.header.size = sz;
1722         ev->id_index.nr = n;
1723
1724         evlist__for_each(evlist, evsel) {
1725                 u32 j;
1726
1727                 for (j = 0; j < evsel->ids; j++) {
1728                         struct id_index_entry *e;
1729                         struct perf_sample_id *sid;
1730
1731                         if (i >= n) {
1732                                 err = process(tool, ev, NULL, machine);
1733                                 if (err)
1734                                         goto out_err;
1735                                 nr -= n;
1736                                 i = 0;
1737                         }
1738
1739                         e = &ev->id_index.entries[i++];
1740
1741                         e->id = evsel->id[j];
1742
1743                         sid = perf_evlist__id2sid(evlist, e->id);
1744                         if (!sid) {
1745                                 free(ev);
1746                                 return -ENOENT;
1747                         }
1748
1749                         e->idx = sid->idx;
1750                         e->cpu = sid->cpu;
1751                         e->tid = sid->tid;
1752                 }
1753         }
1754
1755         sz = sizeof(struct id_index_event) + nr * sizeof(struct id_index_entry);
1756         ev->id_index.header.size = sz;
1757         ev->id_index.nr = nr;
1758
1759         err = process(tool, ev, NULL, machine);
1760 out_err:
1761         free(ev);
1762
1763         return err;
1764 }