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 "vdso.h"
  18
  19 static int perf_session__open(struct perf_session *self, bool force)
  20 {
  21         struct stat input_stat;
  22
  23         if (!strcmp(self->filename, "-")) {
  24                 self->fd_pipe = true;
  25                 self->fd = STDIN_FILENO;
  26
  27                 if (perf_session__read_header(self) < 0)
  28                         pr_err("incompatible file format (rerun with -v to learn more)");
  29
  30                 return 0;
  31         }
  32
  33         self->fd = open(self->filename, O_RDONLY);
  34         if (self->fd < 0) {
  35                 int err = errno;
  36
  37                 pr_err("failed to open %s: %s", self->filename, strerror(err));
  38                 if (err == ENOENT && !strcmp(self->filename, "perf.data"))
  39                         pr_err("  (try 'perf record' first)");
  40                 pr_err("\n");
  41                 return -errno;
  42         }
  43
  44         if (fstat(self->fd, &input_stat) < 0)
  45                 goto out_close;
  46
  47         if (!force && input_stat.st_uid && (input_stat.st_uid != geteuid())) {
  48                 pr_err("file %s not owned by current user or root\n",
  49                        self->filename);
  50                 goto out_close;
  51         }
  52
  53         if (!input_stat.st_size) {
  54                 pr_info("zero-sized file (%s), nothing to do!\n",
  55                         self->filename);
  56                 goto out_close;
  57         }
  58
  59         if (perf_session__read_header(self) < 0) {
  60                 pr_err("incompatible file format (rerun with -v to learn more)");
  61                 goto out_close;
  62         }
  63
  64         if (!perf_evlist__valid_sample_type(self->evlist)) {
  65                 pr_err("non matching sample_type");
  66                 goto out_close;
  67         }
  68
  69         if (!perf_evlist__valid_sample_id_all(self->evlist)) {
  70                 pr_err("non matching sample_id_all");
  71                 goto out_close;
  72         }
  73
  74         if (!perf_evlist__valid_read_format(self->evlist)) {
  75                 pr_err("non matching read_format");
  76                 goto out_close;
  77         }
  78
  79         self->size = input_stat.st_size;
  80         return 0;
  81
  82 out_close:
  83         close(self->fd);
  84         self->fd = -1;
  85         return -1;
  86 }
  87
  88 void perf_session__set_id_hdr_size(struct perf_session *session)
  89 {
  90         u16 id_hdr_size = perf_evlist__id_hdr_size(session->evlist);
  91
  92         machines__set_id_hdr_size(&session->machines, id_hdr_size);
  93 }
  94
  95 int perf_session__create_kernel_maps(struct perf_session *self)
  96 {
  97         int ret = machine__create_kernel_maps(&self->machines.host);
  98
  99         if (ret >= 0)
 100                 ret = machines__create_guest_kernel_maps(&self->machines);
 101         return ret;
 102 }
 103
 104 static void perf_session__destroy_kernel_maps(struct perf_session *self)
 105 {
 106         machines__destroy_kernel_maps(&self->machines);
 107 }
 108
 109 struct perf_session *perf_session__new(const char *filename, int mode,
 110                                        bool force, bool repipe,
 111                                        struct perf_tool *tool)
 112 {
 113         struct perf_session *self;
 114         struct stat st;
 115         size_t len;
 116
 117         if (!filename || !strlen(filename)) {
 118                 if (!fstat(STDIN_FILENO, &st) && S_ISFIFO(st.st_mode))
 119                         filename = "-";
 120                 else
 121                         filename = "perf.data";
 122         }
 123
 124         len = strlen(filename);
 125         self = zalloc(sizeof(*self) + len);
 126
 127         if (self == NULL)
 128                 goto out;
 129
 130         memcpy(self->filename, filename, len);
 131         self->repipe = repipe;
 132         INIT_LIST_HEAD(&self->ordered_samples.samples);
 133         INIT_LIST_HEAD(&self->ordered_samples.sample_cache);
 134         INIT_LIST_HEAD(&self->ordered_samples.to_free);
 135         machines__init(&self->machines);
 136
 137         if (mode == O_RDONLY) {
 138                 if (perf_session__open(self, force) < 0)
 139                         goto out_delete;
 140                 perf_session__set_id_hdr_size(self);
 141         } else if (mode == O_WRONLY) {
 142                 /*
 143                  * In O_RDONLY mode this will be performed when reading the
 144                  * kernel MMAP event, in perf_event__process_mmap().
 145                  */
 146                 if (perf_session__create_kernel_maps(self) < 0)
 147                         goto out_delete;
 148         }
 149
 150         if (tool && tool->ordering_requires_timestamps &&
 151             tool->ordered_samples && !perf_evlist__sample_id_all(self->evlist)) {
 152                 dump_printf("WARNING: No sample_id_all support, falling back to unordered processing\n");
 153                 tool->ordered_samples = false;
 154         }
 155
 156 out:
 157         return self;
 158 out_delete:
 159         perf_session__delete(self);
 160         return NULL;
 161 }
 162
 163 static void perf_session__delete_dead_threads(struct perf_session *session)
 164 {
 165         machine__delete_dead_threads(&session->machines.host);
 166 }
 167
 168 static void perf_session__delete_threads(struct perf_session *session)
 169 {
 170         machine__delete_threads(&session->machines.host);
 171 }
 172
 173 static void perf_session_env__delete(struct perf_session_env *env)
 174 {
 175         free(env->hostname);
 176         free(env->os_release);
 177         free(env->version);
 178         free(env->arch);
 179         free(env->cpu_desc);
 180         free(env->cpuid);
 181
 182         free(env->cmdline);
 183         free(env->sibling_cores);
 184         free(env->sibling_threads);
 185         free(env->numa_nodes);
 186         free(env->pmu_mappings);
 187 }
 188
 189 void perf_session__delete(struct perf_session *self)
 190 {
 191         perf_session__destroy_kernel_maps(self);
 192         perf_session__delete_dead_threads(self);
 193         perf_session__delete_threads(self);
 194         perf_session_env__delete(&self->header.env);
 195         machines__exit(&self->machines);
 196         close(self->fd);
 197         free(self);
 198         vdso__exit();
 199 }
 200
 201 static int process_event_synth_tracing_data_stub(struct perf_tool *tool
 202                                                  __maybe_unused,
 203                                                  union perf_event *event
 204                                                  __maybe_unused,
 205                                                  struct perf_session *session
 206                                                 __maybe_unused)
 207 {
 208         dump_printf(": unhandled!\n");
 209         return 0;
 210 }
 211
 212 static int process_event_synth_attr_stub(struct perf_tool *tool __maybe_unused,
 213                                          union perf_event *event __maybe_unused,
 214                                          struct perf_evlist **pevlist
 215                                          __maybe_unused)
 216 {
 217         dump_printf(": unhandled!\n");
 218         return 0;
 219 }
 220
 221 static int process_event_sample_stub(struct perf_tool *tool __maybe_unused,
 222                                      union perf_event *event __maybe_unused,
 223                                      struct perf_sample *sample __maybe_unused,
 224                                      struct perf_evsel *evsel __maybe_unused,
 225                                      struct machine *machine __maybe_unused)
 226 {
 227         dump_printf(": unhandled!\n");
 228         return 0;
 229 }
 230
 231 static int process_event_stub(struct perf_tool *tool __maybe_unused,
 232                               union perf_event *event __maybe_unused,
 233                               struct perf_sample *sample __maybe_unused,
 234                               struct machine *machine __maybe_unused)
 235 {
 236         dump_printf(": unhandled!\n");
 237         return 0;
 238 }
 239
 240 static int process_finished_round_stub(struct perf_tool *tool __maybe_unused,
 241                                        union perf_event *event __maybe_unused,
 242                                        struct perf_session *perf_session
 243                                        __maybe_unused)
 244 {
 245         dump_printf(": unhandled!\n");
 246         return 0;
 247 }
 248
 249 static int process_finished_round(struct perf_tool *tool,
 250                                   union perf_event *event,
 251                                   struct perf_session *session);
 252
 253 void perf_tool__fill_defaults(struct perf_tool *tool)
 254 {
 255         if (tool->sample == NULL)
 256                 tool->sample = process_event_sample_stub;
 257         if (tool->mmap == NULL)
 258                 tool->mmap = process_event_stub;
 259         if (tool->mmap2 == NULL)
 260                 tool->mmap2 = process_event_stub;
 261         if (tool->comm == NULL)
 262                 tool->comm = process_event_stub;
 263         if (tool->fork == NULL)
 264                 tool->fork = process_event_stub;
 265         if (tool->exit == NULL)
 266                 tool->exit = process_event_stub;
 267         if (tool->lost == NULL)
 268                 tool->lost = perf_event__process_lost;
 269         if (tool->read == NULL)
 270                 tool->read = process_event_sample_stub;
 271         if (tool->throttle == NULL)
 272                 tool->throttle = process_event_stub;
 273         if (tool->unthrottle == NULL)
 274                 tool->unthrottle = process_event_stub;
 275         if (tool->attr == NULL)
 276                 tool->attr = process_event_synth_attr_stub;
 277         if (tool->tracing_data == NULL)
 278                 tool->tracing_data = process_event_synth_tracing_data_stub;
 279         if (tool->build_id == NULL)
 280                 tool->build_id = process_finished_round_stub;
 281         if (tool->finished_round == NULL) {
 282                 if (tool->ordered_samples)
 283                         tool->finished_round = process_finished_round;
 284                 else
 285                         tool->finished_round = process_finished_round_stub;
 286         }
 287 }
 288
 289 void mem_bswap_32(void *src, int byte_size)
 290 {
 291         u32 *m = src;
 292         while (byte_size > 0) {
 293                 *m = bswap_32(*m);
 294                 byte_size -= sizeof(u32);
 295                 ++m;
 296         }
 297 }
 298
 299 void mem_bswap_64(void *src, int byte_size)
 300 {
 301         u64 *m = src;
 302
 303         while (byte_size > 0) {
 304                 *m = bswap_64(*m);
 305                 byte_size -= sizeof(u64);
 306                 ++m;
 307         }
 308 }
 309
 310 static void swap_sample_id_all(union perf_event *event, void *data)
 311 {
 312         void *end = (void *) event + event->header.size;
 313         int size = end - data;
 314
 315         BUG_ON(size % sizeof(u64));
 316         mem_bswap_64(data, size);
 317 }
 318
 319 static void perf_event__all64_swap(union perf_event *event,
 320                                    bool sample_id_all __maybe_unused)
 321 {
 322         struct perf_event_header *hdr = &event->header;
 323         mem_bswap_64(hdr + 1, event->header.size - sizeof(*hdr));
 324 }
 325
 326 static void perf_event__comm_swap(union perf_event *event, bool sample_id_all)
 327 {
 328         event->comm.pid = bswap_32(event->comm.pid);
 329         event->comm.tid = bswap_32(event->comm.tid);
 330
 331         if (sample_id_all) {
 332                 void *data = &event->comm.comm;
 333
 334                 data += PERF_ALIGN(strlen(data) + 1, sizeof(u64));
 335                 swap_sample_id_all(event, data);
 336         }
 337 }
 338
 339 static void perf_event__mmap_swap(union perf_event *event,
 340                                   bool sample_id_all)
 341 {
 342         event->mmap.pid   = bswap_32(event->mmap.pid);
 343         event->mmap.tid   = bswap_32(event->mmap.tid);
 344         event->mmap.start = bswap_64(event->mmap.start);
 345         event->mmap.len   = bswap_64(event->mmap.len);
 346         event->mmap.pgoff = bswap_64(event->mmap.pgoff);
 347
 348         if (sample_id_all) {
 349                 void *data = &event->mmap.filename;
 350
 351                 data += PERF_ALIGN(strlen(data) + 1, sizeof(u64));
 352                 swap_sample_id_all(event, data);
 353         }
 354 }
 355
 356 static void perf_event__mmap2_swap(union perf_event *event,
 357                                   bool sample_id_all)
 358 {
 359         event->mmap2.pid   = bswap_32(event->mmap2.pid);
 360         event->mmap2.tid   = bswap_32(event->mmap2.tid);
 361         event->mmap2.start = bswap_64(event->mmap2.start);
 362         event->mmap2.len   = bswap_64(event->mmap2.len);
 363         event->mmap2.pgoff = bswap_64(event->mmap2.pgoff);
 364         event->mmap2.maj   = bswap_32(event->mmap2.maj);
 365         event->mmap2.min   = bswap_32(event->mmap2.min);
 366         event->mmap2.ino   = bswap_64(event->mmap2.ino);
 367
 368         if (sample_id_all) {
 369                 void *data = &event->mmap2.filename;
 370
 371                 data += PERF_ALIGN(strlen(data) + 1, sizeof(u64));
 372                 swap_sample_id_all(event, data);
 373         }
 374 }
 375 static void perf_event__task_swap(union perf_event *event, bool sample_id_all)
 376 {
 377         event->fork.pid  = bswap_32(event->fork.pid);
 378         event->fork.tid  = bswap_32(event->fork.tid);
 379         event->fork.ppid = bswap_32(event->fork.ppid);
 380         event->fork.ptid = bswap_32(event->fork.ptid);
 381         event->fork.time = bswap_64(event->fork.time);
 382
 383         if (sample_id_all)
 384                 swap_sample_id_all(event, &event->fork + 1);
 385 }
 386
 387 static void perf_event__read_swap(union perf_event *event, bool sample_id_all)
 388 {
 389         event->read.pid          = bswap_32(event->read.pid);
 390         event->read.tid          = bswap_32(event->read.tid);
 391         event->read.value        = bswap_64(event->read.value);
 392         event->read.time_enabled = bswap_64(event->read.time_enabled);
 393         event->read.time_running = bswap_64(event->read.time_running);
 394         event->read.id           = bswap_64(event->read.id);
 395
 396         if (sample_id_all)
 397                 swap_sample_id_all(event, &event->read + 1);
 398 }
 399
 400 static u8 revbyte(u8 b)
 401 {
 402         int rev = (b >> 4) | ((b & 0xf) << 4);
 403         rev = ((rev & 0xcc) >> 2) | ((rev & 0x33) << 2);
 404         rev = ((rev & 0xaa) >> 1) | ((rev & 0x55) << 1);
 405         return (u8) rev;
 406 }
 407
 408 /*
 409  * XXX this is hack in attempt to carry flags bitfield
 410  * throught endian village. ABI says:
 411  *
 412  * Bit-fields are allocated from right to left (least to most significant)
 413  * on little-endian implementations and from left to right (most to least
 414  * significant) on big-endian implementations.
 415  *
 416  * The above seems to be byte specific, so we need to reverse each
 417  * byte of the bitfield. 'Internet' also says this might be implementation
 418  * specific and we probably need proper fix and carry perf_event_attr
 419  * bitfield flags in separate data file FEAT_ section. Thought this seems
 420  * to work for now.
 421  */
 422 static void swap_bitfield(u8 *p, unsigned len)
 423 {
 424         unsigned i;
 425
 426         for (i = 0; i < len; i++) {
 427                 *p = revbyte(*p);
 428                 p++;
 429         }
 430 }
 431
 432 /* exported for swapping attributes in file header */
 433 void perf_event__attr_swap(struct perf_event_attr *attr)
 434 {
 435         attr->type              = bswap_32(attr->type);
 436         attr->size              = bswap_32(attr->size);
 437         attr->config            = bswap_64(attr->config);
 438         attr->sample_period     = bswap_64(attr->sample_period);
 439         attr->sample_type       = bswap_64(attr->sample_type);
 440         attr->read_format       = bswap_64(attr->read_format);
 441         attr->wakeup_events     = bswap_32(attr->wakeup_events);
 442         attr->bp_type           = bswap_32(attr->bp_type);
 443         attr->bp_addr           = bswap_64(attr->bp_addr);
 444         attr->bp_len            = bswap_64(attr->bp_len);
 445
 446         swap_bitfield((u8 *) (&attr->read_format + 1), sizeof(u64));
 447 }
 448
 449 static void perf_event__hdr_attr_swap(union perf_event *event,
 450                                       bool sample_id_all __maybe_unused)
 451 {
 452         size_t size;
 453
 454         perf_event__attr_swap(&event->attr.attr);
 455
 456         size = event->header.size;
 457         size -= (void *)&event->attr.id - (void *)event;
 458         mem_bswap_64(event->attr.id, size);
 459 }
 460
 461 static void perf_event__event_type_swap(union perf_event *event,
 462                                         bool sample_id_all __maybe_unused)
 463 {
 464         event->event_type.event_type.event_id =
 465                 bswap_64(event->event_type.event_type.event_id);
 466 }
 467
 468 static void perf_event__tracing_data_swap(union perf_event *event,
 469                                           bool sample_id_all __maybe_unused)
 470 {
 471         event->tracing_data.size = bswap_32(event->tracing_data.size);
 472 }
 473
 474 typedef void (*perf_event__swap_op)(union perf_event *event,
 475                                     bool sample_id_all);
 476
 477 static perf_event__swap_op perf_event__swap_ops[] = {
 478         [PERF_RECORD_MMAP]                = perf_event__mmap_swap,
 479         [PERF_RECORD_MMAP2]               = perf_event__mmap2_swap,
 480         [PERF_RECORD_COMM]                = perf_event__comm_swap,
 481         [PERF_RECORD_FORK]                = perf_event__task_swap,
 482         [PERF_RECORD_EXIT]                = perf_event__task_swap,
 483         [PERF_RECORD_LOST]                = perf_event__all64_swap,
 484         [PERF_RECORD_READ]                = perf_event__read_swap,
 485         [PERF_RECORD_SAMPLE]              = perf_event__all64_swap,
 486         [PERF_RECORD_HEADER_ATTR]         = perf_event__hdr_attr_swap,
 487         [PERF_RECORD_HEADER_EVENT_TYPE]   = perf_event__event_type_swap,
 488         [PERF_RECORD_HEADER_TRACING_DATA] = perf_event__tracing_data_swap,
 489         [PERF_RECORD_HEADER_BUILD_ID]     = NULL,
 490         [PERF_RECORD_HEADER_MAX]          = NULL,
 491 };
 492
 493 struct sample_queue {
 494         u64                     timestamp;
 495         u64                     file_offset;
 496         union perf_event        *event;
 497         struct list_head        list;
 498 };
 499
 500 static void perf_session_free_sample_buffers(struct perf_session *session)
 501 {
 502         struct ordered_samples *os = &session->ordered_samples;
 503
 504         while (!list_empty(&os->to_free)) {
 505                 struct sample_queue *sq;
 506
 507                 sq = list_entry(os->to_free.next, struct sample_queue, list);
 508                 list_del(&sq->list);
 509                 free(sq);
 510         }
 511 }
 512
 513 static int perf_session_deliver_event(struct perf_session *session,
 514                                       union perf_event *event,
 515                                       struct perf_sample *sample,
 516                                       struct perf_tool *tool,
 517                                       u64 file_offset);
 518
 519 static int flush_sample_queue(struct perf_session *s,
 520                        struct perf_tool *tool)
 521 {
 522         struct ordered_samples *os = &s->ordered_samples;
 523         struct list_head *head = &os->samples;
 524         struct sample_queue *tmp, *iter;
 525         struct perf_sample sample;
 526         u64 limit = os->next_flush;
 527         u64 last_ts = os->last_sample ? os->last_sample->timestamp : 0ULL;
 528         unsigned idx = 0, progress_next = os->nr_samples / 16;
 529         bool show_progress = limit == ULLONG_MAX;
 530         int ret;
 531
 532         if (!tool->ordered_samples || !limit)
 533                 return 0;
 534
 535         list_for_each_entry_safe(iter, tmp, head, list) {
 536                 if (session_done())
 537                         return 0;
 538
 539                 if (iter->timestamp > limit)
 540                         break;
 541
 542                 ret = perf_evlist__parse_sample(s->evlist, iter->event, &sample);
 543                 if (ret)
 544                         pr_err("Can't parse sample, err = %d\n", ret);
 545                 else {
 546                         ret = perf_session_deliver_event(s, iter->event, &sample, tool,
 547                                                          iter->file_offset);
 548                         if (ret)
 549                                 return ret;
 550                 }
 551
 552                 os->last_flush = iter->timestamp;
 553                 list_del(&iter->list);
 554                 list_add(&iter->list, &os->sample_cache);
 555                 if (show_progress && (++idx >= progress_next)) {
 556                         progress_next += os->nr_samples / 16;
 557                         ui_progress__update(idx, os->nr_samples,
 558                                             "Processing time ordered events...");
 559                 }
 560         }
 561
 562         if (list_empty(head)) {
 563                 os->last_sample = NULL;
 564         } else if (last_ts <= limit) {
 565                 os->last_sample =
 566                         list_entry(head->prev, struct sample_queue, list);
 567         }
 568
 569         os->nr_samples = 0;
 570
 571         return 0;
 572 }
 573
 574 /*
 575  * When perf record finishes a pass on every buffers, it records this pseudo
 576  * event.
 577  * We record the max timestamp t found in the pass n.
 578  * Assuming these timestamps are monotonic across cpus, we know that if
 579  * a buffer still has events with timestamps below t, they will be all
 580  * available and then read in the pass n + 1.
 581  * Hence when we start to read the pass n + 2, we can safely flush every
 582  * events with timestamps below t.
 583  *
 584  *    ============ PASS n =================
 585  *       CPU 0         |   CPU 1
 586  *                     |
 587  *    cnt1 timestamps  |   cnt2 timestamps
 588  *          1          |         2
 589  *          2          |         3
 590  *          -          |         4  <--- max recorded
 591  *
 592  *    ============ PASS n + 1 ==============
 593  *       CPU 0         |   CPU 1
 594  *                     |
 595  *    cnt1 timestamps  |   cnt2 timestamps
 596  *          3          |         5
 597  *          4          |         6
 598  *          5          |         7 <---- max recorded
 599  *
 600  *      Flush every events below timestamp 4
 601  *
 602  *    ============ PASS n + 2 ==============
 603  *       CPU 0         |   CPU 1
 604  *                     |
 605  *    cnt1 timestamps  |   cnt2 timestamps
 606  *          6          |         8
 607  *          7          |         9
 608  *          -          |         10
 609  *
 610  *      Flush every events below timestamp 7
 611  *      etc...
 612  */
 613 static int process_finished_round(struct perf_tool *tool,
 614                                   union perf_event *event __maybe_unused,
 615                                   struct perf_session *session)
 616 {
 617         int ret = flush_sample_queue(session, tool);
 618         if (!ret)
 619                 session->ordered_samples.next_flush = session->ordered_samples.max_timestamp;
 620
 621         return ret;
 622 }
 623
 624 /* The queue is ordered by time */
 625 static void __queue_event(struct sample_queue *new, struct perf_session *s)
 626 {
 627         struct ordered_samples *os = &s->ordered_samples;
 628         struct sample_queue *sample = os->last_sample;
 629         u64 timestamp = new->timestamp;
 630         struct list_head *p;
 631
 632         ++os->nr_samples;
 633         os->last_sample = new;
 634
 635         if (!sample) {
 636                 list_add(&new->list, &os->samples);
 637                 os->max_timestamp = timestamp;
 638                 return;
 639         }
 640
 641         /*
 642          * last_sample might point to some random place in the list as it's
 643          * the last queued event. We expect that the new event is close to
 644          * this.
 645          */
 646         if (sample->timestamp <= timestamp) {
 647                 while (sample->timestamp <= timestamp) {
 648                         p = sample->list.next;
 649                         if (p == &os->samples) {
 650                                 list_add_tail(&new->list, &os->samples);
 651                                 os->max_timestamp = timestamp;
 652                                 return;
 653                         }
 654                         sample = list_entry(p, struct sample_queue, list);
 655                 }
 656                 list_add_tail(&new->list, &sample->list);
 657         } else {
 658                 while (sample->timestamp > timestamp) {
 659                         p = sample->list.prev;
 660                         if (p == &os->samples) {
 661                                 list_add(&new->list, &os->samples);
 662                                 return;
 663                         }
 664                         sample = list_entry(p, struct sample_queue, list);
 665                 }
 666                 list_add(&new->list, &sample->list);
 667         }
 668 }
 669
 670 #define MAX_SAMPLE_BUFFER       (64 * 1024 / sizeof(struct sample_queue))
 671
 672 int perf_session_queue_event(struct perf_session *s, union perf_event *event,
 673                                     struct perf_sample *sample, u64 file_offset)
 674 {
 675         struct ordered_samples *os = &s->ordered_samples;
 676         struct list_head *sc = &os->sample_cache;
 677         u64 timestamp = sample->time;
 678         struct sample_queue *new;
 679
 680         if (!timestamp || timestamp == ~0ULL)
 681                 return -ETIME;
 682
 683         if (timestamp < s->ordered_samples.last_flush) {
 684                 printf("Warning: Timestamp below last timeslice flush\n");
 685                 return -EINVAL;
 686         }
 687
 688         if (!list_empty(sc)) {
 689                 new = list_entry(sc->next, struct sample_queue, list);
 690                 list_del(&new->list);
 691         } else if (os->sample_buffer) {
 692                 new = os->sample_buffer + os->sample_buffer_idx;
 693                 if (++os->sample_buffer_idx == MAX_SAMPLE_BUFFER)
 694                         os->sample_buffer = NULL;
 695         } else {
 696                 os->sample_buffer = malloc(MAX_SAMPLE_BUFFER * sizeof(*new));
 697                 if (!os->sample_buffer)
 698                         return -ENOMEM;
 699                 list_add(&os->sample_buffer->list, &os->to_free);
 700                 os->sample_buffer_idx = 2;
 701                 new = os->sample_buffer + 1;
 702         }
 703
 704         new->timestamp = timestamp;
 705         new->file_offset = file_offset;
 706         new->event = event;
 707
 708         __queue_event(new, s);
 709
 710         return 0;
 711 }
 712
 713 static void callchain__printf(struct perf_sample *sample)
 714 {
 715         unsigned int i;
 716
 717         printf("... chain: nr:%" PRIu64 "\n", sample->callchain->nr);
 718
 719         for (i = 0; i < sample->callchain->nr; i++)
 720                 printf("..... %2d: %016" PRIx64 "\n",
 721                        i, sample->callchain->ips[i]);
 722 }
 723
 724 static void branch_stack__printf(struct perf_sample *sample)
 725 {
 726         uint64_t i;
 727
 728         printf("... branch stack: nr:%" PRIu64 "\n", sample->branch_stack->nr);
 729
 730         for (i = 0; i < sample->branch_stack->nr; i++)
 731                 printf("..... %2"PRIu64": %016" PRIx64 " -> %016" PRIx64 "\n",
 732                         i, sample->branch_stack->entries[i].from,
 733                         sample->branch_stack->entries[i].to);
 734 }
 735
 736 static void regs_dump__printf(u64 mask, u64 *regs)
 737 {
 738         unsigned rid, i = 0;
 739
 740         for_each_set_bit(rid, (unsigned long *) &mask, sizeof(mask) * 8) {
 741                 u64 val = regs[i++];
 742
 743                 printf(".... %-5s 0x%" PRIx64 "\n",
 744                        perf_reg_name(rid), val);
 745         }
 746 }
 747
 748 static void regs_user__printf(struct perf_sample *sample, u64 mask)
 749 {
 750         struct regs_dump *user_regs = &sample->user_regs;
 751
 752         if (user_regs->regs) {
 753                 printf("... user regs: mask 0x%" PRIx64 "\n", mask);
 754                 regs_dump__printf(mask, user_regs->regs);
 755         }
 756 }
 757
 758 static void stack_user__printf(struct stack_dump *dump)
 759 {
 760         printf("... ustack: size %" PRIu64 ", offset 0x%x\n",
 761                dump->size, dump->offset);
 762 }
 763
 764 static void perf_session__print_tstamp(struct perf_session *session,
 765                                        union perf_event *event,
 766                                        struct perf_sample *sample)
 767 {
 768         u64 sample_type = __perf_evlist__combined_sample_type(session->evlist);
 769
 770         if (event->header.type != PERF_RECORD_SAMPLE &&
 771             !perf_evlist__sample_id_all(session->evlist)) {
 772                 fputs("-1 -1 ", stdout);
 773                 return;
 774         }
 775
 776         if ((sample_type & PERF_SAMPLE_CPU))
 777                 printf("%u ", sample->cpu);
 778
 779         if (sample_type & PERF_SAMPLE_TIME)
 780                 printf("%" PRIu64 " ", sample->time);
 781 }
 782
 783 static void sample_read__printf(struct perf_sample *sample, u64 read_format)
 784 {
 785         printf("... sample_read:\n");
 786
 787         if (read_format & PERF_FORMAT_TOTAL_TIME_ENABLED)
 788                 printf("...... time enabled %016" PRIx64 "\n",
 789                        sample->read.time_enabled);
 790
 791         if (read_format & PERF_FORMAT_TOTAL_TIME_RUNNING)
 792                 printf("...... time running %016" PRIx64 "\n",
 793                        sample->read.time_running);
 794
 795         if (read_format & PERF_FORMAT_GROUP) {
 796                 u64 i;
 797
 798                 printf(".... group nr %" PRIu64 "\n", sample->read.group.nr);
 799
 800                 for (i = 0; i < sample->read.group.nr; i++) {
 801                         struct sample_read_value *value;
 802
 803                         value = &sample->read.group.values[i];
 804                         printf("..... id %016" PRIx64
 805                                ", value %016" PRIx64 "\n",
 806                                value->id, value->value);
 807                 }
 808         } else
 809                 printf("..... id %016" PRIx64 ", value %016" PRIx64 "\n",
 810                         sample->read.one.id, sample->read.one.value);
 811 }
 812
 813 static void dump_event(struct perf_session *session, union perf_event *event,
 814                        u64 file_offset, struct perf_sample *sample)
 815 {
 816         if (!dump_trace)
 817                 return;
 818
 819         printf("\n%#" PRIx64 " [%#x]: event: %d\n",
 820                file_offset, event->header.size, event->header.type);
 821
 822         trace_event(event);
 823
 824         if (sample)
 825                 perf_session__print_tstamp(session, event, sample);
 826
 827         printf("%#" PRIx64 " [%#x]: PERF_RECORD_%s", file_offset,
 828                event->header.size, perf_event__name(event->header.type));
 829 }
 830
 831 static void dump_sample(struct perf_evsel *evsel, union perf_event *event,
 832                         struct perf_sample *sample)
 833 {
 834         u64 sample_type;
 835
 836         if (!dump_trace)
 837                 return;
 838
 839         printf("(IP, %d): %d/%d: %#" PRIx64 " period: %" PRIu64 " addr: %#" PRIx64 "\n",
 840                event->header.misc, sample->pid, sample->tid, sample->ip,
 841                sample->period, sample->addr);
 842
 843         sample_type = evsel->attr.sample_type;
 844
 845         if (sample_type & PERF_SAMPLE_CALLCHAIN)
 846                 callchain__printf(sample);
 847
 848         if (sample_type & PERF_SAMPLE_BRANCH_STACK)
 849                 branch_stack__printf(sample);
 850
 851         if (sample_type & PERF_SAMPLE_REGS_USER)
 852                 regs_user__printf(sample, evsel->attr.sample_regs_user);
 853
 854         if (sample_type & PERF_SAMPLE_STACK_USER)
 855                 stack_user__printf(&sample->user_stack);
 856
 857         if (sample_type & PERF_SAMPLE_WEIGHT)
 858                 printf("... weight: %" PRIu64 "\n", sample->weight);
 859
 860         if (sample_type & PERF_SAMPLE_DATA_SRC)
 861                 printf(" . data_src: 0x%"PRIx64"\n", sample->data_src);
 862
 863         if (sample_type & PERF_SAMPLE_TRANSACTION)
 864                 printf("... transaction: %" PRIx64 "\n", sample->transaction);
 865
 866         if (sample_type & PERF_SAMPLE_READ)
 867                 sample_read__printf(sample, evsel->attr.read_format);
 868 }
 869
 870 static struct machine *
 871         perf_session__find_machine_for_cpumode(struct perf_session *session,
 872                                                union perf_event *event,
 873                                                struct perf_sample *sample)
 874 {
 875         const u8 cpumode = event->header.misc & PERF_RECORD_MISC_CPUMODE_MASK;
 876
 877         if (perf_guest &&
 878             ((cpumode == PERF_RECORD_MISC_GUEST_KERNEL) ||
 879              (cpumode == PERF_RECORD_MISC_GUEST_USER))) {
 880                 u32 pid;
 881
 882                 if (event->header.type == PERF_RECORD_MMAP
 883                     || event->header.type == PERF_RECORD_MMAP2)
 884                         pid = event->mmap.pid;
 885                 else
 886                         pid = sample->pid;
 887
 888                 return perf_session__findnew_machine(session, pid);
 889         }
 890
 891         return &session->machines.host;
 892 }
 893
 894 static int deliver_sample_value(struct perf_session *session,
 895                                 struct perf_tool *tool,
 896                                 union perf_event *event,
 897                                 struct perf_sample *sample,
 898                                 struct sample_read_value *v,
 899                                 struct machine *machine)
 900 {
 901         struct perf_sample_id *sid;
 902
 903         sid = perf_evlist__id2sid(session->evlist, v->id);
 904         if (sid) {
 905                 sample->id     = v->id;
 906                 sample->period = v->value - sid->period;
 907                 sid->period    = v->value;
 908         }
 909
 910         if (!sid || sid->evsel == NULL) {
 911                 ++session->stats.nr_unknown_id;
 912                 return 0;
 913         }
 914
 915         return tool->sample(tool, event, sample, sid->evsel, machine);
 916 }
 917
 918 static int deliver_sample_group(struct perf_session *session,
 919                                 struct perf_tool *tool,
 920                                 union  perf_event *event,
 921                                 struct perf_sample *sample,
 922                                 struct machine *machine)
 923 {
 924         int ret = -EINVAL;
 925         u64 i;
 926
 927         for (i = 0; i < sample->read.group.nr; i++) {
 928                 ret = deliver_sample_value(session, tool, event, sample,
 929                                            &sample->read.group.values[i],
 930                                            machine);
 931                 if (ret)
 932                         break;
 933         }
 934
 935         return ret;
 936 }
 937
 938 static int
 939 perf_session__deliver_sample(struct perf_session *session,
 940                              struct perf_tool *tool,
 941                              union  perf_event *event,
 942                              struct perf_sample *sample,
 943                              struct perf_evsel *evsel,
 944                              struct machine *machine)
 945 {
 946         /* We know evsel != NULL. */
 947         u64 sample_type = evsel->attr.sample_type;
 948         u64 read_format = evsel->attr.read_format;
 949
 950         /* Standard sample delievery. */
 951         if (!(sample_type & PERF_SAMPLE_READ))
 952                 return tool->sample(tool, event, sample, evsel, machine);
 953
 954         /* For PERF_SAMPLE_READ we have either single or group mode. */
 955         if (read_format & PERF_FORMAT_GROUP)
 956                 return deliver_sample_group(session, tool, event, sample,
 957                                             machine);
 958         else
 959                 return deliver_sample_value(session, tool, event, sample,
 960                                             &sample->read.one, machine);
 961 }
 962
 963 static int perf_session_deliver_event(struct perf_session *session,
 964                                       union perf_event *event,
 965                                       struct perf_sample *sample,
 966                                       struct perf_tool *tool,
 967                                       u64 file_offset)
 968 {
 969         struct perf_evsel *evsel;
 970         struct machine *machine;
 971
 972         dump_event(session, event, file_offset, sample);
 973
 974         evsel = perf_evlist__id2evsel(session->evlist, sample->id);
 975         if (evsel != NULL && event->header.type != PERF_RECORD_SAMPLE) {
 976                 /*
 977                  * XXX We're leaving PERF_RECORD_SAMPLE unnacounted here
 978                  * because the tools right now may apply filters, discarding
 979                  * some of the samples. For consistency, in the future we
 980                  * should have something like nr_filtered_samples and remove
 981                  * the sample->period from total_sample_period, etc, KISS for
 982                  * now tho.
 983                  *
 984                  * Also testing against NULL allows us to handle files without
 985                  * attr.sample_id_all and/or without PERF_SAMPLE_ID. In the
 986                  * future probably it'll be a good idea to restrict event
 987                  * processing via perf_session to files with both set.
 988                  */
 989                 hists__inc_nr_events(&evsel->hists, event->header.type);
 990         }
 991
 992         machine = perf_session__find_machine_for_cpumode(session, event,
 993                                                          sample);
 994
 995         switch (event->header.type) {
 996         case PERF_RECORD_SAMPLE:
 997                 dump_sample(evsel, event, sample);
 998                 if (evsel == NULL) {
 999                         ++session->stats.nr_unknown_id;
1000                         return 0;
1001                 }
1002                 if (machine == NULL) {
1003                         ++session->stats.nr_unprocessable_samples;
1004                         return 0;
1005                 }
1006                 return perf_session__deliver_sample(session, tool, event,
1007                                                     sample, evsel, machine);
1008         case PERF_RECORD_MMAP:
1009                 return tool->mmap(tool, event, sample, machine);
1010         case PERF_RECORD_MMAP2:
1011                 return tool->mmap2(tool, event, sample, machine);
1012         case PERF_RECORD_COMM:
1013                 return tool->comm(tool, event, sample, machine);
1014         case PERF_RECORD_FORK:
1015                 return tool->fork(tool, event, sample, machine);
1016         case PERF_RECORD_EXIT:
1017                 return tool->exit(tool, event, sample, machine);
1018         case PERF_RECORD_LOST:
1019                 if (tool->lost == perf_event__process_lost)
1020                         session->stats.total_lost += event->lost.lost;
1021                 return tool->lost(tool, event, sample, machine);
1022         case PERF_RECORD_READ:
1023                 return tool->read(tool, event, sample, evsel, machine);
1024         case PERF_RECORD_THROTTLE:
1025                 return tool->throttle(tool, event, sample, machine);
1026         case PERF_RECORD_UNTHROTTLE:
1027                 return tool->unthrottle(tool, event, sample, machine);
1028         default:
1029                 ++session->stats.nr_unknown_events;
1030                 return -1;
1031         }
1032 }
1033
1034 static int perf_session__process_user_event(struct perf_session *session, union perf_event *event,
1035                                             struct perf_tool *tool, u64 file_offset)
1036 {
1037         int err;
1038
1039         dump_event(session, event, file_offset, NULL);
1040
1041         /* These events are processed right away */
1042         switch (event->header.type) {
1043         case PERF_RECORD_HEADER_ATTR:
1044                 err = tool->attr(tool, event, &session->evlist);
1045                 if (err == 0)
1046                         perf_session__set_id_hdr_size(session);
1047                 return err;
1048         case PERF_RECORD_HEADER_TRACING_DATA:
1049                 /* setup for reading amidst mmap */
1050                 lseek(session->fd, file_offset, SEEK_SET);
1051                 return tool->tracing_data(tool, event, session);
1052         case PERF_RECORD_HEADER_BUILD_ID:
1053                 return tool->build_id(tool, event, session);
1054         case PERF_RECORD_FINISHED_ROUND:
1055                 return tool->finished_round(tool, event, session);
1056         default:
1057                 return -EINVAL;
1058         }
1059 }
1060
1061 static void event_swap(union perf_event *event, bool sample_id_all)
1062 {
1063         perf_event__swap_op swap;
1064
1065         swap = perf_event__swap_ops[event->header.type];
1066         if (swap)
1067                 swap(event, sample_id_all);
1068 }
1069
1070 static int perf_session__process_event(struct perf_session *session,
1071                                        union perf_event *event,
1072                                        struct perf_tool *tool,
1073                                        u64 file_offset)
1074 {
1075         struct perf_sample sample;
1076         int ret;
1077
1078         if (session->header.needs_swap)
1079                 event_swap(event, perf_evlist__sample_id_all(session->evlist));
1080
1081         if (event->header.type >= PERF_RECORD_HEADER_MAX)
1082                 return -EINVAL;
1083
1084         events_stats__inc(&session->stats, event->header.type);
1085
1086         if (event->header.type >= PERF_RECORD_USER_TYPE_START)
1087                 return perf_session__process_user_event(session, event, tool, file_offset);
1088
1089         /*
1090          * For all kernel events we get the sample data
1091          */
1092         ret = perf_evlist__parse_sample(session->evlist, event, &sample);
1093         if (ret)
1094                 return ret;
1095
1096         if (tool->ordered_samples) {
1097                 ret = perf_session_queue_event(session, event, &sample,
1098                                                file_offset);
1099                 if (ret != -ETIME)
1100                         return ret;
1101         }
1102
1103         return perf_session_deliver_event(session, event, &sample, tool,
1104                                           file_offset);
1105 }
1106
1107 void perf_event_header__bswap(struct perf_event_header *self)
1108 {
1109         self->type = bswap_32(self->type);
1110         self->misc = bswap_16(self->misc);
1111         self->size = bswap_16(self->size);
1112 }
1113
1114 struct thread *perf_session__findnew(struct perf_session *session, pid_t pid)
1115 {
1116         return machine__findnew_thread(&session->machines.host, 0, pid);
1117 }
1118
1119 static struct thread *perf_session__register_idle_thread(struct perf_session *self)
1120 {
1121         struct thread *thread = perf_session__findnew(self, 0);
1122
1123         if (thread == NULL || thread__set_comm(thread, "swapper")) {
1124                 pr_err("problem inserting idle task.\n");
1125                 thread = NULL;
1126         }
1127
1128         return thread;
1129 }
1130
1131 static void perf_session__warn_about_errors(const struct perf_session *session,
1132                                             const struct perf_tool *tool)
1133 {
1134         if (tool->lost == perf_event__process_lost &&
1135             session->stats.nr_events[PERF_RECORD_LOST] != 0) {
1136                 ui__warning("Processed %d events and lost %d chunks!\n\n"
1137                             "Check IO/CPU overload!\n\n",
1138                             session->stats.nr_events[0],
1139                             session->stats.nr_events[PERF_RECORD_LOST]);
1140         }
1141
1142         if (session->stats.nr_unknown_events != 0) {
1143                 ui__warning("Found %u unknown events!\n\n"
1144                             "Is this an older tool processing a perf.data "
1145                             "file generated by a more recent tool?\n\n"
1146                             "If that is not the case, consider "
1147                             "reporting to linux-kernel@vger.kernel.org.\n\n",
1148                             session->stats.nr_unknown_events);
1149         }
1150
1151         if (session->stats.nr_unknown_id != 0) {
1152                 ui__warning("%u samples with id not present in the header\n",
1153                             session->stats.nr_unknown_id);
1154         }
1155
1156         if (session->stats.nr_invalid_chains != 0) {
1157                 ui__warning("Found invalid callchains!\n\n"
1158                             "%u out of %u events were discarded for this reason.\n\n"
1159                             "Consider reporting to linux-kernel@vger.kernel.org.\n\n",
1160                             session->stats.nr_invalid_chains,
1161                             session->stats.nr_events[PERF_RECORD_SAMPLE]);
1162         }
1163
1164         if (session->stats.nr_unprocessable_samples != 0) {
1165                 ui__warning("%u unprocessable samples recorded.\n"
1166                             "Do you have a KVM guest running and not using 'perf kvm'?\n",
1167                             session->stats.nr_unprocessable_samples);
1168         }
1169 }
1170
1171 volatile int session_done;
1172
1173 static int __perf_session__process_pipe_events(struct perf_session *self,
1174                                                struct perf_tool *tool)
1175 {
1176         union perf_event *event;
1177         uint32_t size, cur_size = 0;
1178         void *buf = NULL;
1179         int skip = 0;
1180         u64 head;
1181         int err;
1182         void *p;
1183
1184         perf_tool__fill_defaults(tool);
1185
1186         head = 0;
1187         cur_size = sizeof(union perf_event);
1188
1189         buf = malloc(cur_size);
1190         if (!buf)
1191                 return -errno;
1192 more:
1193         event = buf;
1194         err = readn(self->fd, event, sizeof(struct perf_event_header));
1195         if (err <= 0) {
1196                 if (err == 0)
1197                         goto done;
1198
1199                 pr_err("failed to read event header\n");
1200                 goto out_err;
1201         }
1202
1203         if (self->header.needs_swap)
1204                 perf_event_header__bswap(&event->header);
1205
1206         size = event->header.size;
1207         if (size < sizeof(struct perf_event_header)) {
1208                 pr_err("bad event header size\n");
1209                 goto out_err;
1210         }
1211
1212         if (size > cur_size) {
1213                 void *new = realloc(buf, size);
1214                 if (!new) {
1215                         pr_err("failed to allocate memory to read event\n");
1216                         goto out_err;
1217                 }
1218                 buf = new;
1219                 cur_size = size;
1220                 event = buf;
1221         }
1222         p = event;
1223         p += sizeof(struct perf_event_header);
1224
1225         if (size - sizeof(struct perf_event_header)) {
1226                 err = readn(self->fd, p, size - sizeof(struct perf_event_header));
1227                 if (err <= 0) {
1228                         if (err == 0) {
1229                                 pr_err("unexpected end of event stream\n");
1230                                 goto done;
1231                         }
1232
1233                         pr_err("failed to read event data\n");
1234                         goto out_err;
1235                 }
1236         }
1237
1238         if ((skip = perf_session__process_event(self, event, tool, head)) < 0) {
1239                 pr_err("%#" PRIx64 " [%#x]: failed to process type: %d\n",
1240                        head, event->header.size, event->header.type);
1241                 err = -EINVAL;
1242                 goto out_err;
1243         }
1244
1245         head += size;
1246
1247         if (skip > 0)
1248                 head += skip;
1249
1250         if (!session_done())
1251                 goto more;
1252 done:
1253         err = 0;
1254 out_err:
1255         free(buf);
1256         perf_session__warn_about_errors(self, tool);
1257         perf_session_free_sample_buffers(self);
1258         return err;
1259 }
1260
1261 static union perf_event *
1262 fetch_mmaped_event(struct perf_session *session,
1263                    u64 head, size_t mmap_size, char *buf)
1264 {
1265         union perf_event *event;
1266
1267         /*
1268          * Ensure we have enough space remaining to read
1269          * the size of the event in the headers.
1270          */
1271         if (head + sizeof(event->header) > mmap_size)
1272                 return NULL;
1273
1274         event = (union perf_event *)(buf + head);
1275
1276         if (session->header.needs_swap)
1277                 perf_event_header__bswap(&event->header);
1278
1279         if (head + event->header.size > mmap_size) {
1280                 /* We're not fetching the event so swap back again */
1281                 if (session->header.needs_swap)
1282                         perf_event_header__bswap(&event->header);
1283                 return NULL;
1284         }
1285
1286         return event;
1287 }
1288
1289 /*
1290  * On 64bit we can mmap the data file in one go. No need for tiny mmap
1291  * slices. On 32bit we use 32MB.
1292  */
1293 #if BITS_PER_LONG == 64
1294 #define MMAP_SIZE ULLONG_MAX
1295 #define NUM_MMAPS 1
1296 #else
1297 #define MMAP_SIZE (32 * 1024 * 1024ULL)
1298 #define NUM_MMAPS 128
1299 #endif
1300
1301 int __perf_session__process_events(struct perf_session *session,
1302                                    u64 data_offset, u64 data_size,
1303                                    u64 file_size, struct perf_tool *tool)
1304 {
1305         u64 head, page_offset, file_offset, file_pos, progress_next;
1306         int err, mmap_prot, mmap_flags, map_idx = 0;
1307         size_t  mmap_size;
1308         char *buf, *mmaps[NUM_MMAPS];
1309         union perf_event *event;
1310         uint32_t size;
1311
1312         perf_tool__fill_defaults(tool);
1313
1314         page_offset = page_size * (data_offset / page_size);
1315         file_offset = page_offset;
1316         head = data_offset - page_offset;
1317
1318         if (data_size && (data_offset + data_size < file_size))
1319                 file_size = data_offset + data_size;
1320
1321         progress_next = file_size / 16;
1322
1323         mmap_size = MMAP_SIZE;
1324         if (mmap_size > file_size)
1325                 mmap_size = file_size;
1326
1327         memset(mmaps, 0, sizeof(mmaps));
1328
1329         mmap_prot  = PROT_READ;
1330         mmap_flags = MAP_SHARED;
1331
1332         if (session->header.needs_swap) {
1333                 mmap_prot  |= PROT_WRITE;
1334                 mmap_flags = MAP_PRIVATE;
1335         }
1336 remap:
1337         buf = mmap(NULL, mmap_size, mmap_prot, mmap_flags, session->fd,
1338                    file_offset);
1339         if (buf == MAP_FAILED) {
1340                 pr_err("failed to mmap file\n");
1341                 err = -errno;
1342                 goto out_err;
1343         }
1344         mmaps[map_idx] = buf;
1345         map_idx = (map_idx + 1) & (ARRAY_SIZE(mmaps) - 1);
1346         file_pos = file_offset + head;
1347
1348 more:
1349         event = fetch_mmaped_event(session, head, mmap_size, buf);
1350         if (!event) {
1351                 if (mmaps[map_idx]) {
1352                         munmap(mmaps[map_idx], mmap_size);
1353                         mmaps[map_idx] = NULL;
1354                 }
1355
1356                 page_offset = page_size * (head / page_size);
1357                 file_offset += page_offset;
1358                 head -= page_offset;
1359                 goto remap;
1360         }
1361
1362         size = event->header.size;
1363
1364         if (size < sizeof(struct perf_event_header) ||
1365             perf_session__process_event(session, event, tool, file_pos) < 0) {
1366                 pr_err("%#" PRIx64 " [%#x]: failed to process type: %d\n",
1367                        file_offset + head, event->header.size,
1368                        event->header.type);
1369                 err = -EINVAL;
1370                 goto out_err;
1371         }
1372
1373         head += size;
1374         file_pos += size;
1375
1376         if (file_pos >= progress_next) {
1377                 progress_next += file_size / 16;
1378                 ui_progress__update(file_pos, file_size,
1379                                     "Processing events...");
1380         }
1381
1382         err = 0;
1383         if (session_done())
1384                 goto out_err;
1385
1386         if (file_pos < file_size)
1387                 goto more;
1388
1389         /* do the final flush for ordered samples */
1390         session->ordered_samples.next_flush = ULLONG_MAX;
1391         err = flush_sample_queue(session, tool);
1392 out_err:
1393         ui_progress__finish();
1394         perf_session__warn_about_errors(session, tool);
1395         perf_session_free_sample_buffers(session);
1396         return err;
1397 }
1398
1399 int perf_session__process_events(struct perf_session *self,
1400                                  struct perf_tool *tool)
1401 {
1402         int err;
1403
1404         if (perf_session__register_idle_thread(self) == NULL)
1405                 return -ENOMEM;
1406
1407         if (!self->fd_pipe)
1408                 err = __perf_session__process_events(self,
1409                                                      self->header.data_offset,
1410                                                      self->header.data_size,
1411                                                      self->size, tool);
1412         else
1413                 err = __perf_session__process_pipe_events(self, tool);
1414
1415         return err;
1416 }
1417
1418 bool perf_session__has_traces(struct perf_session *session, const char *msg)
1419 {
1420         struct perf_evsel *evsel;
1421
1422         list_for_each_entry(evsel, &session->evlist->entries, node) {
1423                 if (evsel->attr.type == PERF_TYPE_TRACEPOINT)
1424                         return true;
1425         }
1426
1427         pr_err("No trace sample to read. Did you call 'perf %s'?\n", msg);
1428         return false;
1429 }
1430
1431 int maps__set_kallsyms_ref_reloc_sym(struct map **maps,
1432                                      const char *symbol_name, u64 addr)
1433 {
1434         char *bracket;
1435         enum map_type i;
1436         struct ref_reloc_sym *ref;
1437
1438         ref = zalloc(sizeof(struct ref_reloc_sym));
1439         if (ref == NULL)
1440                 return -ENOMEM;
1441
1442         ref->name = strdup(symbol_name);
1443         if (ref->name == NULL) {
1444                 free(ref);
1445                 return -ENOMEM;
1446         }
1447
1448         bracket = strchr(ref->name, ']');
1449         if (bracket)
1450                 *bracket = '\0';
1451
1452         ref->addr = addr;
1453
1454         for (i = 0; i < MAP__NR_TYPES; ++i) {
1455                 struct kmap *kmap = map__kmap(maps[i]);
1456                 kmap->ref_reloc_sym = ref;
1457         }
1458
1459         return 0;
1460 }
1461
1462 size_t perf_session__fprintf_dsos(struct perf_session *self, FILE *fp)
1463 {
1464         return machines__fprintf_dsos(&self->machines, fp);
1465 }
1466
1467 size_t perf_session__fprintf_dsos_buildid(struct perf_session *self, FILE *fp,
1468                                           bool (skip)(struct dso *dso, int parm), int parm)
1469 {
1470         return machines__fprintf_dsos_buildid(&self->machines, fp, skip, parm);
1471 }
1472
1473 size_t perf_session__fprintf_nr_events(struct perf_session *session, FILE *fp)
1474 {
1475         struct perf_evsel *pos;
1476         size_t ret = fprintf(fp, "Aggregated stats:\n");
1477
1478         ret += events_stats__fprintf(&session->stats, fp);
1479
1480         list_for_each_entry(pos, &session->evlist->entries, node) {
1481                 ret += fprintf(fp, "%s stats:\n", perf_evsel__name(pos));
1482                 ret += events_stats__fprintf(&pos->hists.stats, fp);
1483         }
1484
1485         return ret;
1486 }
1487
1488 size_t perf_session__fprintf(struct perf_session *session, FILE *fp)
1489 {
1490         /*
1491          * FIXME: Here we have to actually print all the machines in this
1492          * session, not just the host...
1493          */
1494         return machine__fprintf(&session->machines.host, fp);
1495 }
1496
1497 struct perf_evsel *perf_session__find_first_evtype(struct perf_session *session,
1498                                               unsigned int type)
1499 {
1500         struct perf_evsel *pos;
1501
1502         list_for_each_entry(pos, &session->evlist->entries, node) {
1503                 if (pos->attr.type == type)
1504                         return pos;
1505         }
1506         return NULL;
1507 }
1508
1509 void perf_evsel__print_ip(struct perf_evsel *evsel, union perf_event *event,
1510                           struct perf_sample *sample, struct machine *machine,
1511                           unsigned int print_opts, unsigned int stack_depth)
1512 {
1513         struct addr_location al;
1514         struct callchain_cursor_node *node;
1515         int print_ip = print_opts & PRINT_IP_OPT_IP;
1516         int print_sym = print_opts & PRINT_IP_OPT_SYM;
1517         int print_dso = print_opts & PRINT_IP_OPT_DSO;
1518         int print_symoffset = print_opts & PRINT_IP_OPT_SYMOFFSET;
1519         int print_oneline = print_opts & PRINT_IP_OPT_ONELINE;
1520         char s = print_oneline ? ' ' : '\t';
1521
1522         if (perf_event__preprocess_sample(event, machine, &al, sample) < 0) {
1523                 error("problem processing %d event, skipping it.\n",
1524                         event->header.type);
1525                 return;
1526         }
1527
1528         if (symbol_conf.use_callchain && sample->callchain) {
1529
1530                 if (machine__resolve_callchain(machine, evsel, al.thread,
1531                                                sample, NULL, NULL) != 0) {
1532                         if (verbose)
1533                                 error("Failed to resolve callchain. Skipping\n");
1534                         return;
1535                 }
1536                 callchain_cursor_commit(&callchain_cursor);
1537
1538                 while (stack_depth) {
1539                         node = callchain_cursor_current(&callchain_cursor);
1540                         if (!node)
1541                                 break;
1542
1543                         if (print_ip)
1544                                 printf("%c%16" PRIx64, s, node->ip);
1545
1546                         if (print_sym) {
1547                                 printf(" ");
1548                                 if (print_symoffset) {
1549                                         al.addr = node->ip;
1550                                         al.map  = node->map;
1551                                         symbol__fprintf_symname_offs(node->sym, &al, stdout);
1552                                 } else
1553                                         symbol__fprintf_symname(node->sym, stdout);
1554                         }
1555
1556                         if (print_dso) {
1557                                 printf(" (");
1558                                 map__fprintf_dsoname(node->map, stdout);
1559                                 printf(")");
1560                         }
1561
1562                         if (!print_oneline)
1563                                 printf("\n");
1564
1565                         callchain_cursor_advance(&callchain_cursor);
1566
1567                         stack_depth--;
1568                 }
1569
1570         } else {
1571                 if (print_ip)
1572                         printf("%16" PRIx64, sample->ip);
1573
1574                 if (print_sym) {
1575                         printf(" ");
1576                         if (print_symoffset)
1577                                 symbol__fprintf_symname_offs(al.sym, &al,
1578                                                              stdout);
1579                         else
1580                                 symbol__fprintf_symname(al.sym, stdout);
1581                 }
1582
1583                 if (print_dso) {
1584                         printf(" (");
1585                         map__fprintf_dsoname(al.map, stdout);
1586                         printf(")");
1587                 }
1588         }
1589 }
1590
1591 int perf_session__cpu_bitmap(struct perf_session *session,
1592                              const char *cpu_list, unsigned long *cpu_bitmap)
1593 {
1594         int i;
1595         struct cpu_map *map;
1596
1597         for (i = 0; i < PERF_TYPE_MAX; ++i) {
1598                 struct perf_evsel *evsel;
1599
1600                 evsel = perf_session__find_first_evtype(session, i);
1601                 if (!evsel)
1602                         continue;
1603
1604                 if (!(evsel->attr.sample_type & PERF_SAMPLE_CPU)) {
1605                         pr_err("File does not contain CPU events. "
1606                                "Remove -c option to proceed.\n");
1607                         return -1;
1608                 }
1609         }
1610
1611         map = cpu_map__new(cpu_list);
1612         if (map == NULL) {
1613                 pr_err("Invalid cpu_list\n");
1614                 return -1;
1615         }
1616
1617         for (i = 0; i < map->nr; i++) {
1618                 int cpu = map->map[i];
1619
1620                 if (cpu >= MAX_NR_CPUS) {
1621                         pr_err("Requested CPU %d too large. "
1622                                "Consider raising MAX_NR_CPUS\n", cpu);
1623                         return -1;
1624                 }
1625
1626                 set_bit(cpu, cpu_bitmap);
1627         }
1628
1629         return 0;
1630 }
1631
1632 void perf_session__fprintf_info(struct perf_session *session, FILE *fp,
1633                                 bool full)
1634 {
1635         struct stat st;
1636         int ret;
1637
1638         if (session == NULL || fp == NULL)
1639                 return;
1640
1641         ret = fstat(session->fd, &st);
1642         if (ret == -1)
1643                 return;
1644
1645         fprintf(fp, "# ========\n");
1646         fprintf(fp, "# captured on: %s", ctime(&st.st_ctime));
1647         perf_header__fprintf_info(session, fp, full);
1648         fprintf(fp, "# ========\n#\n");
1649 }
1650
1651
1652 int __perf_session__set_tracepoints_handlers(struct perf_session *session,
1653                                              const struct perf_evsel_str_handler *assocs,
1654                                              size_t nr_assocs)
1655 {
1656         struct perf_evsel *evsel;
1657         size_t i;
1658         int err;
1659
1660         for (i = 0; i < nr_assocs; i++) {
1661                 /*
1662                  * Adding a handler for an event not in the session,
1663                  * just ignore it.
1664                  */
1665                 evsel = perf_evlist__find_tracepoint_by_name(session->evlist, assocs[i].name);
1666                 if (evsel == NULL)
1667                         continue;
1668
1669                 err = -EEXIST;
1670                 if (evsel->handler.func != NULL)
1671                         goto out;
1672                 evsel->handler.func = assocs[i].handler;
1673         }
1674
1675         err = 0;
1676 out:
1677         return err;
1678 }