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