1 #include <errno.h> 2 #include <inttypes.h> 3 #include <linux/kernel.h> 4 #include <traceevent/event-parse.h> 5 #include <api/fs/fs.h> 6 7 #include <byteswap.h> 8 #include <unistd.h> 9 #include <sys/types.h> 10 #include <sys/mman.h> 11 12 #include "evlist.h" 13 #include "evsel.h" 14 #include "memswap.h" 15 #include "session.h" 16 #include "tool.h" 17 #include "sort.h" 18 #include "util.h" 19 #include "cpumap.h" 20 #include "perf_regs.h" 21 #include "asm/bug.h" 22 #include "auxtrace.h" 23 #include "thread.h" 24 #include "thread-stack.h" 25 #include "stat.h" 26 27 static int perf_session__deliver_event(struct perf_session *session, 28 union perf_event *event, 29 struct perf_sample *sample, 30 struct perf_tool *tool, 31 u64 file_offset); 32 33 static int perf_session__open(struct perf_session *session) 34 { 35 struct perf_data_file *file = session->file; 36 37 if (perf_session__read_header(session) < 0) { 38 pr_err("incompatible file format (rerun with -v to learn more)\n"); 39 return -1; 40 } 41 42 if (perf_data_file__is_pipe(file)) 43 return 0; 44 45 if (perf_header__has_feat(&session->header, HEADER_STAT)) 46 return 0; 47 48 if (!perf_evlist__valid_sample_type(session->evlist)) { 49 pr_err("non matching sample_type\n"); 50 return -1; 51 } 52 53 if (!perf_evlist__valid_sample_id_all(session->evlist)) { 54 pr_err("non matching sample_id_all\n"); 55 return -1; 56 } 57 58 if (!perf_evlist__valid_read_format(session->evlist)) { 59 pr_err("non matching read_format\n"); 60 return -1; 61 } 62 63 return 0; 64 } 65 66 void perf_session__set_id_hdr_size(struct perf_session *session) 67 { 68 u16 id_hdr_size = perf_evlist__id_hdr_size(session->evlist); 69 70 machines__set_id_hdr_size(&session->machines, id_hdr_size); 71 } 72 73 int perf_session__create_kernel_maps(struct perf_session *session) 74 { 75 int ret = machine__create_kernel_maps(&session->machines.host); 76 77 if (ret >= 0) 78 ret = machines__create_guest_kernel_maps(&session->machines); 79 return ret; 80 } 81 82 static void perf_session__destroy_kernel_maps(struct perf_session *session) 83 { 84 machines__destroy_kernel_maps(&session->machines); 85 } 86 87 static bool perf_session__has_comm_exec(struct perf_session *session) 88 { 89 struct perf_evsel *evsel; 90 91 evlist__for_each_entry(session->evlist, evsel) { 92 if (evsel->attr.comm_exec) 93 return true; 94 } 95 96 return false; 97 } 98 99 static void perf_session__set_comm_exec(struct perf_session *session) 100 { 101 bool comm_exec = perf_session__has_comm_exec(session); 102 103 machines__set_comm_exec(&session->machines, comm_exec); 104 } 105 106 static int ordered_events__deliver_event(struct ordered_events *oe, 107 struct ordered_event *event) 108 { 109 struct perf_sample sample; 110 struct perf_session *session = container_of(oe, struct perf_session, 111 ordered_events); 112 int ret = perf_evlist__parse_sample(session->evlist, event->event, &sample); 113 114 if (ret) { 115 pr_err("Can't parse sample, err = %d\n", ret); 116 return ret; 117 } 118 119 return perf_session__deliver_event(session, event->event, &sample, 120 session->tool, event->file_offset); 121 } 122 123 struct perf_session *perf_session__new(struct perf_data_file *file, 124 bool repipe, struct perf_tool *tool) 125 { 126 struct perf_session *session = zalloc(sizeof(*session)); 127 128 if (!session) 129 goto out; 130 131 session->repipe = repipe; 132 session->tool = tool; 133 INIT_LIST_HEAD(&session->auxtrace_index); 134 machines__init(&session->machines); 135 ordered_events__init(&session->ordered_events, ordered_events__deliver_event); 136 137 if (file) { 138 if (perf_data_file__open(file)) 139 goto out_delete; 140 141 session->file = file; 142 143 if (perf_data_file__is_read(file)) { 144 if (perf_session__open(session) < 0) 145 goto out_close; 146 147 /* 148 * set session attributes that are present in perf.data 149 * but not in pipe-mode. 150 */ 151 if (!file->is_pipe) { 152 perf_session__set_id_hdr_size(session); 153 perf_session__set_comm_exec(session); 154 } 155 } 156 } else { 157 session->machines.host.env = &perf_env; 158 } 159 160 if (!file || perf_data_file__is_write(file)) { 161 /* 162 * In O_RDONLY mode this will be performed when reading the 163 * kernel MMAP event, in perf_event__process_mmap(). 164 */ 165 if (perf_session__create_kernel_maps(session) < 0) 166 pr_warning("Cannot read kernel map\n"); 167 } 168 169 /* 170 * In pipe-mode, evlist is empty until PERF_RECORD_HEADER_ATTR is 171 * processed, so perf_evlist__sample_id_all is not meaningful here. 172 */ 173 if ((!file || !file->is_pipe) && tool && tool->ordering_requires_timestamps && 174 tool->ordered_events && !perf_evlist__sample_id_all(session->evlist)) { 175 dump_printf("WARNING: No sample_id_all support, falling back to unordered processing\n"); 176 tool->ordered_events = false; 177 } 178 179 return session; 180 181 out_close: 182 perf_data_file__close(file); 183 out_delete: 184 perf_session__delete(session); 185 out: 186 return NULL; 187 } 188 189 static void perf_session__delete_threads(struct perf_session *session) 190 { 191 machine__delete_threads(&session->machines.host); 192 } 193 194 void perf_session__delete(struct perf_session *session) 195 { 196 if (session == NULL) 197 return; 198 auxtrace__free(session); 199 auxtrace_index__free(&session->auxtrace_index); 200 perf_session__destroy_kernel_maps(session); 201 perf_session__delete_threads(session); 202 perf_env__exit(&session->header.env); 203 machines__exit(&session->machines); 204 if (session->file) 205 perf_data_file__close(session->file); 206 free(session); 207 } 208 209 static int process_event_synth_tracing_data_stub(struct perf_tool *tool 210 __maybe_unused, 211 union perf_event *event 212 __maybe_unused, 213 struct perf_session *session 214 __maybe_unused) 215 { 216 dump_printf(": unhandled!\n"); 217 return 0; 218 } 219 220 static int process_event_synth_attr_stub(struct perf_tool *tool __maybe_unused, 221 union perf_event *event __maybe_unused, 222 struct perf_evlist **pevlist 223 __maybe_unused) 224 { 225 dump_printf(": unhandled!\n"); 226 return 0; 227 } 228 229 static int process_event_synth_event_update_stub(struct perf_tool *tool __maybe_unused, 230 union perf_event *event __maybe_unused, 231 struct perf_evlist **pevlist 232 __maybe_unused) 233 { 234 if (dump_trace) 235 perf_event__fprintf_event_update(event, stdout); 236 237 dump_printf(": unhandled!\n"); 238 return 0; 239 } 240 241 static int process_event_sample_stub(struct perf_tool *tool __maybe_unused, 242 union perf_event *event __maybe_unused, 243 struct perf_sample *sample __maybe_unused, 244 struct perf_evsel *evsel __maybe_unused, 245 struct machine *machine __maybe_unused) 246 { 247 dump_printf(": unhandled!\n"); 248 return 0; 249 } 250 251 static int process_event_stub(struct perf_tool *tool __maybe_unused, 252 union perf_event *event __maybe_unused, 253 struct perf_sample *sample __maybe_unused, 254 struct machine *machine __maybe_unused) 255 { 256 dump_printf(": unhandled!\n"); 257 return 0; 258 } 259 260 static int process_finished_round_stub(struct perf_tool *tool __maybe_unused, 261 union perf_event *event __maybe_unused, 262 struct ordered_events *oe __maybe_unused) 263 { 264 dump_printf(": unhandled!\n"); 265 return 0; 266 } 267 268 static int process_finished_round(struct perf_tool *tool, 269 union perf_event *event, 270 struct ordered_events *oe); 271 272 static int skipn(int fd, off_t n) 273 { 274 char buf[4096]; 275 ssize_t ret; 276 277 while (n > 0) { 278 ret = read(fd, buf, min(n, (off_t)sizeof(buf))); 279 if (ret <= 0) 280 return ret; 281 n -= ret; 282 } 283 284 return 0; 285 } 286 287 static s64 process_event_auxtrace_stub(struct perf_tool *tool __maybe_unused, 288 union perf_event *event, 289 struct perf_session *session 290 __maybe_unused) 291 { 292 dump_printf(": unhandled!\n"); 293 if (perf_data_file__is_pipe(session->file)) 294 skipn(perf_data_file__fd(session->file), event->auxtrace.size); 295 return event->auxtrace.size; 296 } 297 298 static int process_event_op2_stub(struct perf_tool *tool __maybe_unused, 299 union perf_event *event __maybe_unused, 300 struct perf_session *session __maybe_unused) 301 { 302 dump_printf(": unhandled!\n"); 303 return 0; 304 } 305 306 307 static 308 int process_event_thread_map_stub(struct perf_tool *tool __maybe_unused, 309 union perf_event *event __maybe_unused, 310 struct perf_session *session __maybe_unused) 311 { 312 if (dump_trace) 313 perf_event__fprintf_thread_map(event, stdout); 314 315 dump_printf(": unhandled!\n"); 316 return 0; 317 } 318 319 static 320 int process_event_cpu_map_stub(struct perf_tool *tool __maybe_unused, 321 union perf_event *event __maybe_unused, 322 struct perf_session *session __maybe_unused) 323 { 324 if (dump_trace) 325 perf_event__fprintf_cpu_map(event, stdout); 326 327 dump_printf(": unhandled!\n"); 328 return 0; 329 } 330 331 static 332 int process_event_stat_config_stub(struct perf_tool *tool __maybe_unused, 333 union perf_event *event __maybe_unused, 334 struct perf_session *session __maybe_unused) 335 { 336 if (dump_trace) 337 perf_event__fprintf_stat_config(event, stdout); 338 339 dump_printf(": unhandled!\n"); 340 return 0; 341 } 342 343 static int process_stat_stub(struct perf_tool *tool __maybe_unused, 344 union perf_event *event __maybe_unused, 345 struct perf_session *perf_session 346 __maybe_unused) 347 { 348 if (dump_trace) 349 perf_event__fprintf_stat(event, stdout); 350 351 dump_printf(": unhandled!\n"); 352 return 0; 353 } 354 355 static int process_stat_round_stub(struct perf_tool *tool __maybe_unused, 356 union perf_event *event __maybe_unused, 357 struct perf_session *perf_session 358 __maybe_unused) 359 { 360 if (dump_trace) 361 perf_event__fprintf_stat_round(event, stdout); 362 363 dump_printf(": unhandled!\n"); 364 return 0; 365 } 366 367 void perf_tool__fill_defaults(struct perf_tool *tool) 368 { 369 if (tool->sample == NULL) 370 tool->sample = process_event_sample_stub; 371 if (tool->mmap == NULL) 372 tool->mmap = process_event_stub; 373 if (tool->mmap2 == NULL) 374 tool->mmap2 = process_event_stub; 375 if (tool->comm == NULL) 376 tool->comm = process_event_stub; 377 if (tool->fork == NULL) 378 tool->fork = process_event_stub; 379 if (tool->exit == NULL) 380 tool->exit = process_event_stub; 381 if (tool->lost == NULL) 382 tool->lost = perf_event__process_lost; 383 if (tool->lost_samples == NULL) 384 tool->lost_samples = perf_event__process_lost_samples; 385 if (tool->aux == NULL) 386 tool->aux = perf_event__process_aux; 387 if (tool->itrace_start == NULL) 388 tool->itrace_start = perf_event__process_itrace_start; 389 if (tool->context_switch == NULL) 390 tool->context_switch = perf_event__process_switch; 391 if (tool->read == NULL) 392 tool->read = process_event_sample_stub; 393 if (tool->throttle == NULL) 394 tool->throttle = process_event_stub; 395 if (tool->unthrottle == NULL) 396 tool->unthrottle = process_event_stub; 397 if (tool->attr == NULL) 398 tool->attr = process_event_synth_attr_stub; 399 if (tool->event_update == NULL) 400 tool->event_update = process_event_synth_event_update_stub; 401 if (tool->tracing_data == NULL) 402 tool->tracing_data = process_event_synth_tracing_data_stub; 403 if (tool->build_id == NULL) 404 tool->build_id = process_event_op2_stub; 405 if (tool->finished_round == NULL) { 406 if (tool->ordered_events) 407 tool->finished_round = process_finished_round; 408 else 409 tool->finished_round = process_finished_round_stub; 410 } 411 if (tool->id_index == NULL) 412 tool->id_index = process_event_op2_stub; 413 if (tool->auxtrace_info == NULL) 414 tool->auxtrace_info = process_event_op2_stub; 415 if (tool->auxtrace == NULL) 416 tool->auxtrace = process_event_auxtrace_stub; 417 if (tool->auxtrace_error == NULL) 418 tool->auxtrace_error = process_event_op2_stub; 419 if (tool->thread_map == NULL) 420 tool->thread_map = process_event_thread_map_stub; 421 if (tool->cpu_map == NULL) 422 tool->cpu_map = process_event_cpu_map_stub; 423 if (tool->stat_config == NULL) 424 tool->stat_config = process_event_stat_config_stub; 425 if (tool->stat == NULL) 426 tool->stat = process_stat_stub; 427 if (tool->stat_round == NULL) 428 tool->stat_round = process_stat_round_stub; 429 if (tool->time_conv == NULL) 430 tool->time_conv = process_event_op2_stub; 431 if (tool->feature == NULL) 432 tool->feature = process_event_op2_stub; 433 } 434 435 static void swap_sample_id_all(union perf_event *event, void *data) 436 { 437 void *end = (void *) event + event->header.size; 438 int size = end - data; 439 440 BUG_ON(size % sizeof(u64)); 441 mem_bswap_64(data, size); 442 } 443 444 static void perf_event__all64_swap(union perf_event *event, 445 bool sample_id_all __maybe_unused) 446 { 447 struct perf_event_header *hdr = &event->header; 448 mem_bswap_64(hdr + 1, event->header.size - sizeof(*hdr)); 449 } 450 451 static void perf_event__comm_swap(union perf_event *event, bool sample_id_all) 452 { 453 event->comm.pid = bswap_32(event->comm.pid); 454 event->comm.tid = bswap_32(event->comm.tid); 455 456 if (sample_id_all) { 457 void *data = &event->comm.comm; 458 459 data += PERF_ALIGN(strlen(data) + 1, sizeof(u64)); 460 swap_sample_id_all(event, data); 461 } 462 } 463 464 static void perf_event__mmap_swap(union perf_event *event, 465 bool sample_id_all) 466 { 467 event->mmap.pid = bswap_32(event->mmap.pid); 468 event->mmap.tid = bswap_32(event->mmap.tid); 469 event->mmap.start = bswap_64(event->mmap.start); 470 event->mmap.len = bswap_64(event->mmap.len); 471 event->mmap.pgoff = bswap_64(event->mmap.pgoff); 472 473 if (sample_id_all) { 474 void *data = &event->mmap.filename; 475 476 data += PERF_ALIGN(strlen(data) + 1, sizeof(u64)); 477 swap_sample_id_all(event, data); 478 } 479 } 480 481 static void perf_event__mmap2_swap(union perf_event *event, 482 bool sample_id_all) 483 { 484 event->mmap2.pid = bswap_32(event->mmap2.pid); 485 event->mmap2.tid = bswap_32(event->mmap2.tid); 486 event->mmap2.start = bswap_64(event->mmap2.start); 487 event->mmap2.len = bswap_64(event->mmap2.len); 488 event->mmap2.pgoff = bswap_64(event->mmap2.pgoff); 489 event->mmap2.maj = bswap_32(event->mmap2.maj); 490 event->mmap2.min = bswap_32(event->mmap2.min); 491 event->mmap2.ino = bswap_64(event->mmap2.ino); 492 493 if (sample_id_all) { 494 void *data = &event->mmap2.filename; 495 496 data += PERF_ALIGN(strlen(data) + 1, sizeof(u64)); 497 swap_sample_id_all(event, data); 498 } 499 } 500 static void perf_event__task_swap(union perf_event *event, bool sample_id_all) 501 { 502 event->fork.pid = bswap_32(event->fork.pid); 503 event->fork.tid = bswap_32(event->fork.tid); 504 event->fork.ppid = bswap_32(event->fork.ppid); 505 event->fork.ptid = bswap_32(event->fork.ptid); 506 event->fork.time = bswap_64(event->fork.time); 507 508 if (sample_id_all) 509 swap_sample_id_all(event, &event->fork + 1); 510 } 511 512 static void perf_event__read_swap(union perf_event *event, bool sample_id_all) 513 { 514 event->read.pid = bswap_32(event->read.pid); 515 event->read.tid = bswap_32(event->read.tid); 516 event->read.value = bswap_64(event->read.value); 517 event->read.time_enabled = bswap_64(event->read.time_enabled); 518 event->read.time_running = bswap_64(event->read.time_running); 519 event->read.id = bswap_64(event->read.id); 520 521 if (sample_id_all) 522 swap_sample_id_all(event, &event->read + 1); 523 } 524 525 static void perf_event__aux_swap(union perf_event *event, bool sample_id_all) 526 { 527 event->aux.aux_offset = bswap_64(event->aux.aux_offset); 528 event->aux.aux_size = bswap_64(event->aux.aux_size); 529 event->aux.flags = bswap_64(event->aux.flags); 530 531 if (sample_id_all) 532 swap_sample_id_all(event, &event->aux + 1); 533 } 534 535 static void perf_event__itrace_start_swap(union perf_event *event, 536 bool sample_id_all) 537 { 538 event->itrace_start.pid = bswap_32(event->itrace_start.pid); 539 event->itrace_start.tid = bswap_32(event->itrace_start.tid); 540 541 if (sample_id_all) 542 swap_sample_id_all(event, &event->itrace_start + 1); 543 } 544 545 static void perf_event__switch_swap(union perf_event *event, bool sample_id_all) 546 { 547 if (event->header.type == PERF_RECORD_SWITCH_CPU_WIDE) { 548 event->context_switch.next_prev_pid = 549 bswap_32(event->context_switch.next_prev_pid); 550 event->context_switch.next_prev_tid = 551 bswap_32(event->context_switch.next_prev_tid); 552 } 553 554 if (sample_id_all) 555 swap_sample_id_all(event, &event->context_switch + 1); 556 } 557 558 static void perf_event__throttle_swap(union perf_event *event, 559 bool sample_id_all) 560 { 561 event->throttle.time = bswap_64(event->throttle.time); 562 event->throttle.id = bswap_64(event->throttle.id); 563 event->throttle.stream_id = bswap_64(event->throttle.stream_id); 564 565 if (sample_id_all) 566 swap_sample_id_all(event, &event->throttle + 1); 567 } 568 569 static u8 revbyte(u8 b) 570 { 571 int rev = (b >> 4) | ((b & 0xf) << 4); 572 rev = ((rev & 0xcc) >> 2) | ((rev & 0x33) << 2); 573 rev = ((rev & 0xaa) >> 1) | ((rev & 0x55) << 1); 574 return (u8) rev; 575 } 576 577 /* 578 * XXX this is hack in attempt to carry flags bitfield 579 * through endian village. ABI says: 580 * 581 * Bit-fields are allocated from right to left (least to most significant) 582 * on little-endian implementations and from left to right (most to least 583 * significant) on big-endian implementations. 584 * 585 * The above seems to be byte specific, so we need to reverse each 586 * byte of the bitfield. 'Internet' also says this might be implementation 587 * specific and we probably need proper fix and carry perf_event_attr 588 * bitfield flags in separate data file FEAT_ section. Thought this seems 589 * to work for now. 590 */ 591 static void swap_bitfield(u8 *p, unsigned len) 592 { 593 unsigned i; 594 595 for (i = 0; i < len; i++) { 596 *p = revbyte(*p); 597 p++; 598 } 599 } 600 601 /* exported for swapping attributes in file header */ 602 void perf_event__attr_swap(struct perf_event_attr *attr) 603 { 604 attr->type = bswap_32(attr->type); 605 attr->size = bswap_32(attr->size); 606 607 #define bswap_safe(f, n) \ 608 (attr->size > (offsetof(struct perf_event_attr, f) + \ 609 sizeof(attr->f) * (n))) 610 #define bswap_field(f, sz) \ 611 do { \ 612 if (bswap_safe(f, 0)) \ 613 attr->f = bswap_##sz(attr->f); \ 614 } while(0) 615 #define bswap_field_16(f) bswap_field(f, 16) 616 #define bswap_field_32(f) bswap_field(f, 32) 617 #define bswap_field_64(f) bswap_field(f, 64) 618 619 bswap_field_64(config); 620 bswap_field_64(sample_period); 621 bswap_field_64(sample_type); 622 bswap_field_64(read_format); 623 bswap_field_32(wakeup_events); 624 bswap_field_32(bp_type); 625 bswap_field_64(bp_addr); 626 bswap_field_64(bp_len); 627 bswap_field_64(branch_sample_type); 628 bswap_field_64(sample_regs_user); 629 bswap_field_32(sample_stack_user); 630 bswap_field_32(aux_watermark); 631 bswap_field_16(sample_max_stack); 632 633 /* 634 * After read_format are bitfields. Check read_format because 635 * we are unable to use offsetof on bitfield. 636 */ 637 if (bswap_safe(read_format, 1)) 638 swap_bitfield((u8 *) (&attr->read_format + 1), 639 sizeof(u64)); 640 #undef bswap_field_64 641 #undef bswap_field_32 642 #undef bswap_field 643 #undef bswap_safe 644 } 645 646 static void perf_event__hdr_attr_swap(union perf_event *event, 647 bool sample_id_all __maybe_unused) 648 { 649 size_t size; 650 651 perf_event__attr_swap(&event->attr.attr); 652 653 size = event->header.size; 654 size -= (void *)&event->attr.id - (void *)event; 655 mem_bswap_64(event->attr.id, size); 656 } 657 658 static void perf_event__event_update_swap(union perf_event *event, 659 bool sample_id_all __maybe_unused) 660 { 661 event->event_update.type = bswap_64(event->event_update.type); 662 event->event_update.id = bswap_64(event->event_update.id); 663 } 664 665 static void perf_event__event_type_swap(union perf_event *event, 666 bool sample_id_all __maybe_unused) 667 { 668 event->event_type.event_type.event_id = 669 bswap_64(event->event_type.event_type.event_id); 670 } 671 672 static void perf_event__tracing_data_swap(union perf_event *event, 673 bool sample_id_all __maybe_unused) 674 { 675 event->tracing_data.size = bswap_32(event->tracing_data.size); 676 } 677 678 static void perf_event__auxtrace_info_swap(union perf_event *event, 679 bool sample_id_all __maybe_unused) 680 { 681 size_t size; 682 683 event->auxtrace_info.type = bswap_32(event->auxtrace_info.type); 684 685 size = event->header.size; 686 size -= (void *)&event->auxtrace_info.priv - (void *)event; 687 mem_bswap_64(event->auxtrace_info.priv, size); 688 } 689 690 static void perf_event__auxtrace_swap(union perf_event *event, 691 bool sample_id_all __maybe_unused) 692 { 693 event->auxtrace.size = bswap_64(event->auxtrace.size); 694 event->auxtrace.offset = bswap_64(event->auxtrace.offset); 695 event->auxtrace.reference = bswap_64(event->auxtrace.reference); 696 event->auxtrace.idx = bswap_32(event->auxtrace.idx); 697 event->auxtrace.tid = bswap_32(event->auxtrace.tid); 698 event->auxtrace.cpu = bswap_32(event->auxtrace.cpu); 699 } 700 701 static void perf_event__auxtrace_error_swap(union perf_event *event, 702 bool sample_id_all __maybe_unused) 703 { 704 event->auxtrace_error.type = bswap_32(event->auxtrace_error.type); 705 event->auxtrace_error.code = bswap_32(event->auxtrace_error.code); 706 event->auxtrace_error.cpu = bswap_32(event->auxtrace_error.cpu); 707 event->auxtrace_error.pid = bswap_32(event->auxtrace_error.pid); 708 event->auxtrace_error.tid = bswap_32(event->auxtrace_error.tid); 709 event->auxtrace_error.ip = bswap_64(event->auxtrace_error.ip); 710 } 711 712 static void perf_event__thread_map_swap(union perf_event *event, 713 bool sample_id_all __maybe_unused) 714 { 715 unsigned i; 716 717 event->thread_map.nr = bswap_64(event->thread_map.nr); 718 719 for (i = 0; i < event->thread_map.nr; i++) 720 event->thread_map.entries[i].pid = bswap_64(event->thread_map.entries[i].pid); 721 } 722 723 static void perf_event__cpu_map_swap(union perf_event *event, 724 bool sample_id_all __maybe_unused) 725 { 726 struct cpu_map_data *data = &event->cpu_map.data; 727 struct cpu_map_entries *cpus; 728 struct cpu_map_mask *mask; 729 unsigned i; 730 731 data->type = bswap_64(data->type); 732 733 switch (data->type) { 734 case PERF_CPU_MAP__CPUS: 735 cpus = (struct cpu_map_entries *)data->data; 736 737 cpus->nr = bswap_16(cpus->nr); 738 739 for (i = 0; i < cpus->nr; i++) 740 cpus->cpu[i] = bswap_16(cpus->cpu[i]); 741 break; 742 case PERF_CPU_MAP__MASK: 743 mask = (struct cpu_map_mask *) data->data; 744 745 mask->nr = bswap_16(mask->nr); 746 mask->long_size = bswap_16(mask->long_size); 747 748 switch (mask->long_size) { 749 case 4: mem_bswap_32(&mask->mask, mask->nr); break; 750 case 8: mem_bswap_64(&mask->mask, mask->nr); break; 751 default: 752 pr_err("cpu_map swap: unsupported long size\n"); 753 } 754 default: 755 break; 756 } 757 } 758 759 static void perf_event__stat_config_swap(union perf_event *event, 760 bool sample_id_all __maybe_unused) 761 { 762 u64 size; 763 764 size = event->stat_config.nr * sizeof(event->stat_config.data[0]); 765 size += 1; /* nr item itself */ 766 mem_bswap_64(&event->stat_config.nr, size); 767 } 768 769 static void perf_event__stat_swap(union perf_event *event, 770 bool sample_id_all __maybe_unused) 771 { 772 event->stat.id = bswap_64(event->stat.id); 773 event->stat.thread = bswap_32(event->stat.thread); 774 event->stat.cpu = bswap_32(event->stat.cpu); 775 event->stat.val = bswap_64(event->stat.val); 776 event->stat.ena = bswap_64(event->stat.ena); 777 event->stat.run = bswap_64(event->stat.run); 778 } 779 780 static void perf_event__stat_round_swap(union perf_event *event, 781 bool sample_id_all __maybe_unused) 782 { 783 event->stat_round.type = bswap_64(event->stat_round.type); 784 event->stat_round.time = bswap_64(event->stat_round.time); 785 } 786 787 typedef void (*perf_event__swap_op)(union perf_event *event, 788 bool sample_id_all); 789 790 static perf_event__swap_op perf_event__swap_ops[] = { 791 [PERF_RECORD_MMAP] = perf_event__mmap_swap, 792 [PERF_RECORD_MMAP2] = perf_event__mmap2_swap, 793 [PERF_RECORD_COMM] = perf_event__comm_swap, 794 [PERF_RECORD_FORK] = perf_event__task_swap, 795 [PERF_RECORD_EXIT] = perf_event__task_swap, 796 [PERF_RECORD_LOST] = perf_event__all64_swap, 797 [PERF_RECORD_READ] = perf_event__read_swap, 798 [PERF_RECORD_THROTTLE] = perf_event__throttle_swap, 799 [PERF_RECORD_UNTHROTTLE] = perf_event__throttle_swap, 800 [PERF_RECORD_SAMPLE] = perf_event__all64_swap, 801 [PERF_RECORD_AUX] = perf_event__aux_swap, 802 [PERF_RECORD_ITRACE_START] = perf_event__itrace_start_swap, 803 [PERF_RECORD_LOST_SAMPLES] = perf_event__all64_swap, 804 [PERF_RECORD_SWITCH] = perf_event__switch_swap, 805 [PERF_RECORD_SWITCH_CPU_WIDE] = perf_event__switch_swap, 806 [PERF_RECORD_HEADER_ATTR] = perf_event__hdr_attr_swap, 807 [PERF_RECORD_HEADER_EVENT_TYPE] = perf_event__event_type_swap, 808 [PERF_RECORD_HEADER_TRACING_DATA] = perf_event__tracing_data_swap, 809 [PERF_RECORD_HEADER_BUILD_ID] = NULL, 810 [PERF_RECORD_ID_INDEX] = perf_event__all64_swap, 811 [PERF_RECORD_AUXTRACE_INFO] = perf_event__auxtrace_info_swap, 812 [PERF_RECORD_AUXTRACE] = perf_event__auxtrace_swap, 813 [PERF_RECORD_AUXTRACE_ERROR] = perf_event__auxtrace_error_swap, 814 [PERF_RECORD_THREAD_MAP] = perf_event__thread_map_swap, 815 [PERF_RECORD_CPU_MAP] = perf_event__cpu_map_swap, 816 [PERF_RECORD_STAT_CONFIG] = perf_event__stat_config_swap, 817 [PERF_RECORD_STAT] = perf_event__stat_swap, 818 [PERF_RECORD_STAT_ROUND] = perf_event__stat_round_swap, 819 [PERF_RECORD_EVENT_UPDATE] = perf_event__event_update_swap, 820 [PERF_RECORD_TIME_CONV] = perf_event__all64_swap, 821 [PERF_RECORD_HEADER_MAX] = NULL, 822 }; 823 824 /* 825 * When perf record finishes a pass on every buffers, it records this pseudo 826 * event. 827 * We record the max timestamp t found in the pass n. 828 * Assuming these timestamps are monotonic across cpus, we know that if 829 * a buffer still has events with timestamps below t, they will be all 830 * available and then read in the pass n + 1. 831 * Hence when we start to read the pass n + 2, we can safely flush every 832 * events with timestamps below t. 833 * 834 * ============ PASS n ================= 835 * CPU 0 | CPU 1 836 * | 837 * cnt1 timestamps | cnt2 timestamps 838 * 1 | 2 839 * 2 | 3 840 * - | 4 <--- max recorded 841 * 842 * ============ PASS n + 1 ============== 843 * CPU 0 | CPU 1 844 * | 845 * cnt1 timestamps | cnt2 timestamps 846 * 3 | 5 847 * 4 | 6 848 * 5 | 7 <---- max recorded 849 * 850 * Flush every events below timestamp 4 851 * 852 * ============ PASS n + 2 ============== 853 * CPU 0 | CPU 1 854 * | 855 * cnt1 timestamps | cnt2 timestamps 856 * 6 | 8 857 * 7 | 9 858 * - | 10 859 * 860 * Flush every events below timestamp 7 861 * etc... 862 */ 863 static int process_finished_round(struct perf_tool *tool __maybe_unused, 864 union perf_event *event __maybe_unused, 865 struct ordered_events *oe) 866 { 867 if (dump_trace) 868 fprintf(stdout, "\n"); 869 return ordered_events__flush(oe, OE_FLUSH__ROUND); 870 } 871 872 int perf_session__queue_event(struct perf_session *s, union perf_event *event, 873 struct perf_sample *sample, u64 file_offset) 874 { 875 return ordered_events__queue(&s->ordered_events, event, sample, file_offset); 876 } 877 878 static void callchain__lbr_callstack_printf(struct perf_sample *sample) 879 { 880 struct ip_callchain *callchain = sample->callchain; 881 struct branch_stack *lbr_stack = sample->branch_stack; 882 u64 kernel_callchain_nr = callchain->nr; 883 unsigned int i; 884 885 for (i = 0; i < kernel_callchain_nr; i++) { 886 if (callchain->ips[i] == PERF_CONTEXT_USER) 887 break; 888 } 889 890 if ((i != kernel_callchain_nr) && lbr_stack->nr) { 891 u64 total_nr; 892 /* 893 * LBR callstack can only get user call chain, 894 * i is kernel call chain number, 895 * 1 is PERF_CONTEXT_USER. 896 * 897 * The user call chain is stored in LBR registers. 898 * LBR are pair registers. The caller is stored 899 * in "from" register, while the callee is stored 900 * in "to" register. 901 * For example, there is a call stack 902 * "A"->"B"->"C"->"D". 903 * The LBR registers will recorde like 904 * "C"->"D", "B"->"C", "A"->"B". 905 * So only the first "to" register and all "from" 906 * registers are needed to construct the whole stack. 907 */ 908 total_nr = i + 1 + lbr_stack->nr + 1; 909 kernel_callchain_nr = i + 1; 910 911 printf("... LBR call chain: nr:%" PRIu64 "\n", total_nr); 912 913 for (i = 0; i < kernel_callchain_nr; i++) 914 printf("..... %2d: %016" PRIx64 "\n", 915 i, callchain->ips[i]); 916 917 printf("..... %2d: %016" PRIx64 "\n", 918 (int)(kernel_callchain_nr), lbr_stack->entries[0].to); 919 for (i = 0; i < lbr_stack->nr; i++) 920 printf("..... %2d: %016" PRIx64 "\n", 921 (int)(i + kernel_callchain_nr + 1), lbr_stack->entries[i].from); 922 } 923 } 924 925 static void callchain__printf(struct perf_evsel *evsel, 926 struct perf_sample *sample) 927 { 928 unsigned int i; 929 struct ip_callchain *callchain = sample->callchain; 930 931 if (perf_evsel__has_branch_callstack(evsel)) 932 callchain__lbr_callstack_printf(sample); 933 934 printf("... FP chain: nr:%" PRIu64 "\n", callchain->nr); 935 936 for (i = 0; i < callchain->nr; i++) 937 printf("..... %2d: %016" PRIx64 "\n", 938 i, callchain->ips[i]); 939 } 940 941 static void branch_stack__printf(struct perf_sample *sample) 942 { 943 uint64_t i; 944 945 printf("... branch stack: nr:%" PRIu64 "\n", sample->branch_stack->nr); 946 947 for (i = 0; i < sample->branch_stack->nr; i++) { 948 struct branch_entry *e = &sample->branch_stack->entries[i]; 949 950 printf("..... %2"PRIu64": %016" PRIx64 " -> %016" PRIx64 " %hu cycles %s%s%s%s %x\n", 951 i, e->from, e->to, 952 (unsigned short)e->flags.cycles, 953 e->flags.mispred ? "M" : " ", 954 e->flags.predicted ? "P" : " ", 955 e->flags.abort ? "A" : " ", 956 e->flags.in_tx ? "T" : " ", 957 (unsigned)e->flags.reserved); 958 } 959 } 960 961 static void regs_dump__printf(u64 mask, u64 *regs) 962 { 963 unsigned rid, i = 0; 964 965 for_each_set_bit(rid, (unsigned long *) &mask, sizeof(mask) * 8) { 966 u64 val = regs[i++]; 967 968 printf(".... %-5s 0x%" PRIx64 "\n", 969 perf_reg_name(rid), val); 970 } 971 } 972 973 static const char *regs_abi[] = { 974 [PERF_SAMPLE_REGS_ABI_NONE] = "none", 975 [PERF_SAMPLE_REGS_ABI_32] = "32-bit", 976 [PERF_SAMPLE_REGS_ABI_64] = "64-bit", 977 }; 978 979 static inline const char *regs_dump_abi(struct regs_dump *d) 980 { 981 if (d->abi > PERF_SAMPLE_REGS_ABI_64) 982 return "unknown"; 983 984 return regs_abi[d->abi]; 985 } 986 987 static void regs__printf(const char *type, struct regs_dump *regs) 988 { 989 u64 mask = regs->mask; 990 991 printf("... %s regs: mask 0x%" PRIx64 " ABI %s\n", 992 type, 993 mask, 994 regs_dump_abi(regs)); 995 996 regs_dump__printf(mask, regs->regs); 997 } 998 999 static void regs_user__printf(struct perf_sample *sample) 1000 { 1001 struct regs_dump *user_regs = &sample->user_regs; 1002 1003 if (user_regs->regs) 1004 regs__printf("user", user_regs); 1005 } 1006 1007 static void regs_intr__printf(struct perf_sample *sample) 1008 { 1009 struct regs_dump *intr_regs = &sample->intr_regs; 1010 1011 if (intr_regs->regs) 1012 regs__printf("intr", intr_regs); 1013 } 1014 1015 static void stack_user__printf(struct stack_dump *dump) 1016 { 1017 printf("... ustack: size %" PRIu64 ", offset 0x%x\n", 1018 dump->size, dump->offset); 1019 } 1020 1021 static void perf_evlist__print_tstamp(struct perf_evlist *evlist, 1022 union perf_event *event, 1023 struct perf_sample *sample) 1024 { 1025 u64 sample_type = __perf_evlist__combined_sample_type(evlist); 1026 1027 if (event->header.type != PERF_RECORD_SAMPLE && 1028 !perf_evlist__sample_id_all(evlist)) { 1029 fputs("-1 -1 ", stdout); 1030 return; 1031 } 1032 1033 if ((sample_type & PERF_SAMPLE_CPU)) 1034 printf("%u ", sample->cpu); 1035 1036 if (sample_type & PERF_SAMPLE_TIME) 1037 printf("%" PRIu64 " ", sample->time); 1038 } 1039 1040 static void sample_read__printf(struct perf_sample *sample, u64 read_format) 1041 { 1042 printf("... sample_read:\n"); 1043 1044 if (read_format & PERF_FORMAT_TOTAL_TIME_ENABLED) 1045 printf("...... time enabled %016" PRIx64 "\n", 1046 sample->read.time_enabled); 1047 1048 if (read_format & PERF_FORMAT_TOTAL_TIME_RUNNING) 1049 printf("...... time running %016" PRIx64 "\n", 1050 sample->read.time_running); 1051 1052 if (read_format & PERF_FORMAT_GROUP) { 1053 u64 i; 1054 1055 printf(".... group nr %" PRIu64 "\n", sample->read.group.nr); 1056 1057 for (i = 0; i < sample->read.group.nr; i++) { 1058 struct sample_read_value *value; 1059 1060 value = &sample->read.group.values[i]; 1061 printf("..... id %016" PRIx64 1062 ", value %016" PRIx64 "\n", 1063 value->id, value->value); 1064 } 1065 } else 1066 printf("..... id %016" PRIx64 ", value %016" PRIx64 "\n", 1067 sample->read.one.id, sample->read.one.value); 1068 } 1069 1070 static void dump_event(struct perf_evlist *evlist, union perf_event *event, 1071 u64 file_offset, struct perf_sample *sample) 1072 { 1073 if (!dump_trace) 1074 return; 1075 1076 printf("\n%#" PRIx64 " [%#x]: event: %d\n", 1077 file_offset, event->header.size, event->header.type); 1078 1079 trace_event(event); 1080 1081 if (sample) 1082 perf_evlist__print_tstamp(evlist, event, sample); 1083 1084 printf("%#" PRIx64 " [%#x]: PERF_RECORD_%s", file_offset, 1085 event->header.size, perf_event__name(event->header.type)); 1086 } 1087 1088 static void dump_sample(struct perf_evsel *evsel, union perf_event *event, 1089 struct perf_sample *sample) 1090 { 1091 u64 sample_type; 1092 1093 if (!dump_trace) 1094 return; 1095 1096 printf("(IP, 0x%x): %d/%d: %#" PRIx64 " period: %" PRIu64 " addr: %#" PRIx64 "\n", 1097 event->header.misc, sample->pid, sample->tid, sample->ip, 1098 sample->period, sample->addr); 1099 1100 sample_type = evsel->attr.sample_type; 1101 1102 if (sample_type & PERF_SAMPLE_CALLCHAIN) 1103 callchain__printf(evsel, sample); 1104 1105 if ((sample_type & PERF_SAMPLE_BRANCH_STACK) && !perf_evsel__has_branch_callstack(evsel)) 1106 branch_stack__printf(sample); 1107 1108 if (sample_type & PERF_SAMPLE_REGS_USER) 1109 regs_user__printf(sample); 1110 1111 if (sample_type & PERF_SAMPLE_REGS_INTR) 1112 regs_intr__printf(sample); 1113 1114 if (sample_type & PERF_SAMPLE_STACK_USER) 1115 stack_user__printf(&sample->user_stack); 1116 1117 if (sample_type & PERF_SAMPLE_WEIGHT) 1118 printf("... weight: %" PRIu64 "\n", sample->weight); 1119 1120 if (sample_type & PERF_SAMPLE_DATA_SRC) 1121 printf(" . data_src: 0x%"PRIx64"\n", sample->data_src); 1122 1123 if (sample_type & PERF_SAMPLE_PHYS_ADDR) 1124 printf(" .. phys_addr: 0x%"PRIx64"\n", sample->phys_addr); 1125 1126 if (sample_type & PERF_SAMPLE_TRANSACTION) 1127 printf("... transaction: %" PRIx64 "\n", sample->transaction); 1128 1129 if (sample_type & PERF_SAMPLE_READ) 1130 sample_read__printf(sample, evsel->attr.read_format); 1131 } 1132 1133 static void dump_read(struct perf_evsel *evsel, union perf_event *event) 1134 { 1135 struct read_event *read_event = &event->read; 1136 u64 read_format; 1137 1138 if (!dump_trace) 1139 return; 1140 1141 printf(": %d %d %s %" PRIu64 "\n", event->read.pid, event->read.tid, 1142 evsel ? perf_evsel__name(evsel) : "FAIL", 1143 event->read.value); 1144 1145 read_format = evsel->attr.read_format; 1146 1147 if (read_format & PERF_FORMAT_TOTAL_TIME_ENABLED) 1148 printf("... time enabled : %" PRIu64 "\n", read_event->time_enabled); 1149 1150 if (read_format & PERF_FORMAT_TOTAL_TIME_RUNNING) 1151 printf("... time running : %" PRIu64 "\n", read_event->time_running); 1152 1153 if (read_format & PERF_FORMAT_ID) 1154 printf("... id : %" PRIu64 "\n", read_event->id); 1155 } 1156 1157 static struct machine *machines__find_for_cpumode(struct machines *machines, 1158 union perf_event *event, 1159 struct perf_sample *sample) 1160 { 1161 struct machine *machine; 1162 1163 if (perf_guest && 1164 ((sample->cpumode == PERF_RECORD_MISC_GUEST_KERNEL) || 1165 (sample->cpumode == PERF_RECORD_MISC_GUEST_USER))) { 1166 u32 pid; 1167 1168 if (event->header.type == PERF_RECORD_MMAP 1169 || event->header.type == PERF_RECORD_MMAP2) 1170 pid = event->mmap.pid; 1171 else 1172 pid = sample->pid; 1173 1174 machine = machines__find(machines, pid); 1175 if (!machine) 1176 machine = machines__findnew(machines, DEFAULT_GUEST_KERNEL_ID); 1177 return machine; 1178 } 1179 1180 return &machines->host; 1181 } 1182 1183 static int deliver_sample_value(struct perf_evlist *evlist, 1184 struct perf_tool *tool, 1185 union perf_event *event, 1186 struct perf_sample *sample, 1187 struct sample_read_value *v, 1188 struct machine *machine) 1189 { 1190 struct perf_sample_id *sid = perf_evlist__id2sid(evlist, v->id); 1191 1192 if (sid) { 1193 sample->id = v->id; 1194 sample->period = v->value - sid->period; 1195 sid->period = v->value; 1196 } 1197 1198 if (!sid || sid->evsel == NULL) { 1199 ++evlist->stats.nr_unknown_id; 1200 return 0; 1201 } 1202 1203 return tool->sample(tool, event, sample, sid->evsel, machine); 1204 } 1205 1206 static int deliver_sample_group(struct perf_evlist *evlist, 1207 struct perf_tool *tool, 1208 union perf_event *event, 1209 struct perf_sample *sample, 1210 struct machine *machine) 1211 { 1212 int ret = -EINVAL; 1213 u64 i; 1214 1215 for (i = 0; i < sample->read.group.nr; i++) { 1216 ret = deliver_sample_value(evlist, tool, event, sample, 1217 &sample->read.group.values[i], 1218 machine); 1219 if (ret) 1220 break; 1221 } 1222 1223 return ret; 1224 } 1225 1226 static int 1227 perf_evlist__deliver_sample(struct perf_evlist *evlist, 1228 struct perf_tool *tool, 1229 union perf_event *event, 1230 struct perf_sample *sample, 1231 struct perf_evsel *evsel, 1232 struct machine *machine) 1233 { 1234 /* We know evsel != NULL. */ 1235 u64 sample_type = evsel->attr.sample_type; 1236 u64 read_format = evsel->attr.read_format; 1237 1238 /* Standard sample delivery. */ 1239 if (!(sample_type & PERF_SAMPLE_READ)) 1240 return tool->sample(tool, event, sample, evsel, machine); 1241 1242 /* For PERF_SAMPLE_READ we have either single or group mode. */ 1243 if (read_format & PERF_FORMAT_GROUP) 1244 return deliver_sample_group(evlist, tool, event, sample, 1245 machine); 1246 else 1247 return deliver_sample_value(evlist, tool, event, sample, 1248 &sample->read.one, machine); 1249 } 1250 1251 static int machines__deliver_event(struct machines *machines, 1252 struct perf_evlist *evlist, 1253 union perf_event *event, 1254 struct perf_sample *sample, 1255 struct perf_tool *tool, u64 file_offset) 1256 { 1257 struct perf_evsel *evsel; 1258 struct machine *machine; 1259 1260 dump_event(evlist, event, file_offset, sample); 1261 1262 evsel = perf_evlist__id2evsel(evlist, sample->id); 1263 1264 machine = machines__find_for_cpumode(machines, event, sample); 1265 1266 switch (event->header.type) { 1267 case PERF_RECORD_SAMPLE: 1268 if (evsel == NULL) { 1269 ++evlist->stats.nr_unknown_id; 1270 return 0; 1271 } 1272 dump_sample(evsel, event, sample); 1273 if (machine == NULL) { 1274 ++evlist->stats.nr_unprocessable_samples; 1275 return 0; 1276 } 1277 return perf_evlist__deliver_sample(evlist, tool, event, sample, evsel, machine); 1278 case PERF_RECORD_MMAP: 1279 return tool->mmap(tool, event, sample, machine); 1280 case PERF_RECORD_MMAP2: 1281 if (event->header.misc & PERF_RECORD_MISC_PROC_MAP_PARSE_TIMEOUT) 1282 ++evlist->stats.nr_proc_map_timeout; 1283 return tool->mmap2(tool, event, sample, machine); 1284 case PERF_RECORD_COMM: 1285 return tool->comm(tool, event, sample, machine); 1286 case PERF_RECORD_NAMESPACES: 1287 return tool->namespaces(tool, event, sample, machine); 1288 case PERF_RECORD_FORK: 1289 return tool->fork(tool, event, sample, machine); 1290 case PERF_RECORD_EXIT: 1291 return tool->exit(tool, event, sample, machine); 1292 case PERF_RECORD_LOST: 1293 if (tool->lost == perf_event__process_lost) 1294 evlist->stats.total_lost += event->lost.lost; 1295 return tool->lost(tool, event, sample, machine); 1296 case PERF_RECORD_LOST_SAMPLES: 1297 if (tool->lost_samples == perf_event__process_lost_samples) 1298 evlist->stats.total_lost_samples += event->lost_samples.lost; 1299 return tool->lost_samples(tool, event, sample, machine); 1300 case PERF_RECORD_READ: 1301 dump_read(evsel, event); 1302 return tool->read(tool, event, sample, evsel, machine); 1303 case PERF_RECORD_THROTTLE: 1304 return tool->throttle(tool, event, sample, machine); 1305 case PERF_RECORD_UNTHROTTLE: 1306 return tool->unthrottle(tool, event, sample, machine); 1307 case PERF_RECORD_AUX: 1308 if (tool->aux == perf_event__process_aux) { 1309 if (event->aux.flags & PERF_AUX_FLAG_TRUNCATED) 1310 evlist->stats.total_aux_lost += 1; 1311 if (event->aux.flags & PERF_AUX_FLAG_PARTIAL) 1312 evlist->stats.total_aux_partial += 1; 1313 } 1314 return tool->aux(tool, event, sample, machine); 1315 case PERF_RECORD_ITRACE_START: 1316 return tool->itrace_start(tool, event, sample, machine); 1317 case PERF_RECORD_SWITCH: 1318 case PERF_RECORD_SWITCH_CPU_WIDE: 1319 return tool->context_switch(tool, event, sample, machine); 1320 default: 1321 ++evlist->stats.nr_unknown_events; 1322 return -1; 1323 } 1324 } 1325 1326 static int perf_session__deliver_event(struct perf_session *session, 1327 union perf_event *event, 1328 struct perf_sample *sample, 1329 struct perf_tool *tool, 1330 u64 file_offset) 1331 { 1332 int ret; 1333 1334 ret = auxtrace__process_event(session, event, sample, tool); 1335 if (ret < 0) 1336 return ret; 1337 if (ret > 0) 1338 return 0; 1339 1340 return machines__deliver_event(&session->machines, session->evlist, 1341 event, sample, tool, file_offset); 1342 } 1343 1344 static s64 perf_session__process_user_event(struct perf_session *session, 1345 union perf_event *event, 1346 u64 file_offset) 1347 { 1348 struct ordered_events *oe = &session->ordered_events; 1349 struct perf_tool *tool = session->tool; 1350 int fd = perf_data_file__fd(session->file); 1351 int err; 1352 1353 dump_event(session->evlist, event, file_offset, NULL); 1354 1355 /* These events are processed right away */ 1356 switch (event->header.type) { 1357 case PERF_RECORD_HEADER_ATTR: 1358 err = tool->attr(tool, event, &session->evlist); 1359 if (err == 0) { 1360 perf_session__set_id_hdr_size(session); 1361 perf_session__set_comm_exec(session); 1362 } 1363 return err; 1364 case PERF_RECORD_EVENT_UPDATE: 1365 return tool->event_update(tool, event, &session->evlist); 1366 case PERF_RECORD_HEADER_EVENT_TYPE: 1367 /* 1368 * Depreceated, but we need to handle it for sake 1369 * of old data files create in pipe mode. 1370 */ 1371 return 0; 1372 case PERF_RECORD_HEADER_TRACING_DATA: 1373 /* setup for reading amidst mmap */ 1374 lseek(fd, file_offset, SEEK_SET); 1375 return tool->tracing_data(tool, event, session); 1376 case PERF_RECORD_HEADER_BUILD_ID: 1377 return tool->build_id(tool, event, session); 1378 case PERF_RECORD_FINISHED_ROUND: 1379 return tool->finished_round(tool, event, oe); 1380 case PERF_RECORD_ID_INDEX: 1381 return tool->id_index(tool, event, session); 1382 case PERF_RECORD_AUXTRACE_INFO: 1383 return tool->auxtrace_info(tool, event, session); 1384 case PERF_RECORD_AUXTRACE: 1385 /* setup for reading amidst mmap */ 1386 lseek(fd, file_offset + event->header.size, SEEK_SET); 1387 return tool->auxtrace(tool, event, session); 1388 case PERF_RECORD_AUXTRACE_ERROR: 1389 perf_session__auxtrace_error_inc(session, event); 1390 return tool->auxtrace_error(tool, event, session); 1391 case PERF_RECORD_THREAD_MAP: 1392 return tool->thread_map(tool, event, session); 1393 case PERF_RECORD_CPU_MAP: 1394 return tool->cpu_map(tool, event, session); 1395 case PERF_RECORD_STAT_CONFIG: 1396 return tool->stat_config(tool, event, session); 1397 case PERF_RECORD_STAT: 1398 return tool->stat(tool, event, session); 1399 case PERF_RECORD_STAT_ROUND: 1400 return tool->stat_round(tool, event, session); 1401 case PERF_RECORD_TIME_CONV: 1402 session->time_conv = event->time_conv; 1403 return tool->time_conv(tool, event, session); 1404 case PERF_RECORD_HEADER_FEATURE: 1405 return tool->feature(tool, event, session); 1406 default: 1407 return -EINVAL; 1408 } 1409 } 1410 1411 int perf_session__deliver_synth_event(struct perf_session *session, 1412 union perf_event *event, 1413 struct perf_sample *sample) 1414 { 1415 struct perf_evlist *evlist = session->evlist; 1416 struct perf_tool *tool = session->tool; 1417 1418 events_stats__inc(&evlist->stats, event->header.type); 1419 1420 if (event->header.type >= PERF_RECORD_USER_TYPE_START) 1421 return perf_session__process_user_event(session, event, 0); 1422 1423 return machines__deliver_event(&session->machines, evlist, event, sample, tool, 0); 1424 } 1425 1426 static void event_swap(union perf_event *event, bool sample_id_all) 1427 { 1428 perf_event__swap_op swap; 1429 1430 swap = perf_event__swap_ops[event->header.type]; 1431 if (swap) 1432 swap(event, sample_id_all); 1433 } 1434 1435 int perf_session__peek_event(struct perf_session *session, off_t file_offset, 1436 void *buf, size_t buf_sz, 1437 union perf_event **event_ptr, 1438 struct perf_sample *sample) 1439 { 1440 union perf_event *event; 1441 size_t hdr_sz, rest; 1442 int fd; 1443 1444 if (session->one_mmap && !session->header.needs_swap) { 1445 event = file_offset - session->one_mmap_offset + 1446 session->one_mmap_addr; 1447 goto out_parse_sample; 1448 } 1449 1450 if (perf_data_file__is_pipe(session->file)) 1451 return -1; 1452 1453 fd = perf_data_file__fd(session->file); 1454 hdr_sz = sizeof(struct perf_event_header); 1455 1456 if (buf_sz < hdr_sz) 1457 return -1; 1458 1459 if (lseek(fd, file_offset, SEEK_SET) == (off_t)-1 || 1460 readn(fd, buf, hdr_sz) != (ssize_t)hdr_sz) 1461 return -1; 1462 1463 event = (union perf_event *)buf; 1464 1465 if (session->header.needs_swap) 1466 perf_event_header__bswap(&event->header); 1467 1468 if (event->header.size < hdr_sz || event->header.size > buf_sz) 1469 return -1; 1470 1471 rest = event->header.size - hdr_sz; 1472 1473 if (readn(fd, buf, rest) != (ssize_t)rest) 1474 return -1; 1475 1476 if (session->header.needs_swap) 1477 event_swap(event, perf_evlist__sample_id_all(session->evlist)); 1478 1479 out_parse_sample: 1480 1481 if (sample && event->header.type < PERF_RECORD_USER_TYPE_START && 1482 perf_evlist__parse_sample(session->evlist, event, sample)) 1483 return -1; 1484 1485 *event_ptr = event; 1486 1487 return 0; 1488 } 1489 1490 static s64 perf_session__process_event(struct perf_session *session, 1491 union perf_event *event, u64 file_offset) 1492 { 1493 struct perf_evlist *evlist = session->evlist; 1494 struct perf_tool *tool = session->tool; 1495 struct perf_sample sample; 1496 int ret; 1497 1498 if (session->header.needs_swap) 1499 event_swap(event, perf_evlist__sample_id_all(evlist)); 1500 1501 if (event->header.type >= PERF_RECORD_HEADER_MAX) 1502 return -EINVAL; 1503 1504 events_stats__inc(&evlist->stats, event->header.type); 1505 1506 if (event->header.type >= PERF_RECORD_USER_TYPE_START) 1507 return perf_session__process_user_event(session, event, file_offset); 1508 1509 /* 1510 * For all kernel events we get the sample data 1511 */ 1512 ret = perf_evlist__parse_sample(evlist, event, &sample); 1513 if (ret) 1514 return ret; 1515 1516 if (tool->ordered_events) { 1517 ret = perf_session__queue_event(session, event, &sample, file_offset); 1518 if (ret != -ETIME) 1519 return ret; 1520 } 1521 1522 return perf_session__deliver_event(session, event, &sample, tool, 1523 file_offset); 1524 } 1525 1526 void perf_event_header__bswap(struct perf_event_header *hdr) 1527 { 1528 hdr->type = bswap_32(hdr->type); 1529 hdr->misc = bswap_16(hdr->misc); 1530 hdr->size = bswap_16(hdr->size); 1531 } 1532 1533 struct thread *perf_session__findnew(struct perf_session *session, pid_t pid) 1534 { 1535 return machine__findnew_thread(&session->machines.host, -1, pid); 1536 } 1537 1538 int perf_session__register_idle_thread(struct perf_session *session) 1539 { 1540 struct thread *thread; 1541 int err = 0; 1542 1543 thread = machine__findnew_thread(&session->machines.host, 0, 0); 1544 if (thread == NULL || thread__set_comm(thread, "swapper", 0)) { 1545 pr_err("problem inserting idle task.\n"); 1546 err = -1; 1547 } 1548 1549 if (thread == NULL || thread__set_namespaces(thread, 0, NULL)) { 1550 pr_err("problem inserting idle task.\n"); 1551 err = -1; 1552 } 1553 1554 /* machine__findnew_thread() got the thread, so put it */ 1555 thread__put(thread); 1556 return err; 1557 } 1558 1559 static void 1560 perf_session__warn_order(const struct perf_session *session) 1561 { 1562 const struct ordered_events *oe = &session->ordered_events; 1563 struct perf_evsel *evsel; 1564 bool should_warn = true; 1565 1566 evlist__for_each_entry(session->evlist, evsel) { 1567 if (evsel->attr.write_backward) 1568 should_warn = false; 1569 } 1570 1571 if (!should_warn) 1572 return; 1573 if (oe->nr_unordered_events != 0) 1574 ui__warning("%u out of order events recorded.\n", oe->nr_unordered_events); 1575 } 1576 1577 static void perf_session__warn_about_errors(const struct perf_session *session) 1578 { 1579 const struct events_stats *stats = &session->evlist->stats; 1580 1581 if (session->tool->lost == perf_event__process_lost && 1582 stats->nr_events[PERF_RECORD_LOST] != 0) { 1583 ui__warning("Processed %d events and lost %d chunks!\n\n" 1584 "Check IO/CPU overload!\n\n", 1585 stats->nr_events[0], 1586 stats->nr_events[PERF_RECORD_LOST]); 1587 } 1588 1589 if (session->tool->lost_samples == perf_event__process_lost_samples) { 1590 double drop_rate; 1591 1592 drop_rate = (double)stats->total_lost_samples / 1593 (double) (stats->nr_events[PERF_RECORD_SAMPLE] + stats->total_lost_samples); 1594 if (drop_rate > 0.05) { 1595 ui__warning("Processed %" PRIu64 " samples and lost %3.2f%% samples!\n\n", 1596 stats->nr_events[PERF_RECORD_SAMPLE] + stats->total_lost_samples, 1597 drop_rate * 100.0); 1598 } 1599 } 1600 1601 if (session->tool->aux == perf_event__process_aux && 1602 stats->total_aux_lost != 0) { 1603 ui__warning("AUX data lost %" PRIu64 " times out of %u!\n\n", 1604 stats->total_aux_lost, 1605 stats->nr_events[PERF_RECORD_AUX]); 1606 } 1607 1608 if (session->tool->aux == perf_event__process_aux && 1609 stats->total_aux_partial != 0) { 1610 bool vmm_exclusive = false; 1611 1612 (void)sysfs__read_bool("module/kvm_intel/parameters/vmm_exclusive", 1613 &vmm_exclusive); 1614 1615 ui__warning("AUX data had gaps in it %" PRIu64 " times out of %u!\n\n" 1616 "Are you running a KVM guest in the background?%s\n\n", 1617 stats->total_aux_partial, 1618 stats->nr_events[PERF_RECORD_AUX], 1619 vmm_exclusive ? 1620 "\nReloading kvm_intel module with vmm_exclusive=0\n" 1621 "will reduce the gaps to only guest's timeslices." : 1622 ""); 1623 } 1624 1625 if (stats->nr_unknown_events != 0) { 1626 ui__warning("Found %u unknown events!\n\n" 1627 "Is this an older tool processing a perf.data " 1628 "file generated by a more recent tool?\n\n" 1629 "If that is not the case, consider " 1630 "reporting to linux-kernel@vger.kernel.org.\n\n", 1631 stats->nr_unknown_events); 1632 } 1633 1634 if (stats->nr_unknown_id != 0) { 1635 ui__warning("%u samples with id not present in the header\n", 1636 stats->nr_unknown_id); 1637 } 1638 1639 if (stats->nr_invalid_chains != 0) { 1640 ui__warning("Found invalid callchains!\n\n" 1641 "%u out of %u events were discarded for this reason.\n\n" 1642 "Consider reporting to linux-kernel@vger.kernel.org.\n\n", 1643 stats->nr_invalid_chains, 1644 stats->nr_events[PERF_RECORD_SAMPLE]); 1645 } 1646 1647 if (stats->nr_unprocessable_samples != 0) { 1648 ui__warning("%u unprocessable samples recorded.\n" 1649 "Do you have a KVM guest running and not using 'perf kvm'?\n", 1650 stats->nr_unprocessable_samples); 1651 } 1652 1653 perf_session__warn_order(session); 1654 1655 events_stats__auxtrace_error_warn(stats); 1656 1657 if (stats->nr_proc_map_timeout != 0) { 1658 ui__warning("%d map information files for pre-existing threads were\n" 1659 "not processed, if there are samples for addresses they\n" 1660 "will not be resolved, you may find out which are these\n" 1661 "threads by running with -v and redirecting the output\n" 1662 "to a file.\n" 1663 "The time limit to process proc map is too short?\n" 1664 "Increase it by --proc-map-timeout\n", 1665 stats->nr_proc_map_timeout); 1666 } 1667 } 1668 1669 static int perf_session__flush_thread_stack(struct thread *thread, 1670 void *p __maybe_unused) 1671 { 1672 return thread_stack__flush(thread); 1673 } 1674 1675 static int perf_session__flush_thread_stacks(struct perf_session *session) 1676 { 1677 return machines__for_each_thread(&session->machines, 1678 perf_session__flush_thread_stack, 1679 NULL); 1680 } 1681 1682 volatile int session_done; 1683 1684 static int __perf_session__process_pipe_events(struct perf_session *session) 1685 { 1686 struct ordered_events *oe = &session->ordered_events; 1687 struct perf_tool *tool = session->tool; 1688 int fd = perf_data_file__fd(session->file); 1689 union perf_event *event; 1690 uint32_t size, cur_size = 0; 1691 void *buf = NULL; 1692 s64 skip = 0; 1693 u64 head; 1694 ssize_t err; 1695 void *p; 1696 1697 perf_tool__fill_defaults(tool); 1698 1699 head = 0; 1700 cur_size = sizeof(union perf_event); 1701 1702 buf = malloc(cur_size); 1703 if (!buf) 1704 return -errno; 1705 ordered_events__set_copy_on_queue(oe, true); 1706 more: 1707 event = buf; 1708 err = readn(fd, event, sizeof(struct perf_event_header)); 1709 if (err <= 0) { 1710 if (err == 0) 1711 goto done; 1712 1713 pr_err("failed to read event header\n"); 1714 goto out_err; 1715 } 1716 1717 if (session->header.needs_swap) 1718 perf_event_header__bswap(&event->header); 1719 1720 size = event->header.size; 1721 if (size < sizeof(struct perf_event_header)) { 1722 pr_err("bad event header size\n"); 1723 goto out_err; 1724 } 1725 1726 if (size > cur_size) { 1727 void *new = realloc(buf, size); 1728 if (!new) { 1729 pr_err("failed to allocate memory to read event\n"); 1730 goto out_err; 1731 } 1732 buf = new; 1733 cur_size = size; 1734 event = buf; 1735 } 1736 p = event; 1737 p += sizeof(struct perf_event_header); 1738 1739 if (size - sizeof(struct perf_event_header)) { 1740 err = readn(fd, p, size - sizeof(struct perf_event_header)); 1741 if (err <= 0) { 1742 if (err == 0) { 1743 pr_err("unexpected end of event stream\n"); 1744 goto done; 1745 } 1746 1747 pr_err("failed to read event data\n"); 1748 goto out_err; 1749 } 1750 } 1751 1752 if ((skip = perf_session__process_event(session, event, head)) < 0) { 1753 pr_err("%#" PRIx64 " [%#x]: failed to process type: %d\n", 1754 head, event->header.size, event->header.type); 1755 err = -EINVAL; 1756 goto out_err; 1757 } 1758 1759 head += size; 1760 1761 if (skip > 0) 1762 head += skip; 1763 1764 if (!session_done()) 1765 goto more; 1766 done: 1767 /* do the final flush for ordered samples */ 1768 err = ordered_events__flush(oe, OE_FLUSH__FINAL); 1769 if (err) 1770 goto out_err; 1771 err = auxtrace__flush_events(session, tool); 1772 if (err) 1773 goto out_err; 1774 err = perf_session__flush_thread_stacks(session); 1775 out_err: 1776 free(buf); 1777 perf_session__warn_about_errors(session); 1778 ordered_events__free(&session->ordered_events); 1779 auxtrace__free_events(session); 1780 return err; 1781 } 1782 1783 static union perf_event * 1784 fetch_mmaped_event(struct perf_session *session, 1785 u64 head, size_t mmap_size, char *buf) 1786 { 1787 union perf_event *event; 1788 1789 /* 1790 * Ensure we have enough space remaining to read 1791 * the size of the event in the headers. 1792 */ 1793 if (head + sizeof(event->header) > mmap_size) 1794 return NULL; 1795 1796 event = (union perf_event *)(buf + head); 1797 1798 if (session->header.needs_swap) 1799 perf_event_header__bswap(&event->header); 1800 1801 if (head + event->header.size > mmap_size) { 1802 /* We're not fetching the event so swap back again */ 1803 if (session->header.needs_swap) 1804 perf_event_header__bswap(&event->header); 1805 return NULL; 1806 } 1807 1808 return event; 1809 } 1810 1811 /* 1812 * On 64bit we can mmap the data file in one go. No need for tiny mmap 1813 * slices. On 32bit we use 32MB. 1814 */ 1815 #if BITS_PER_LONG == 64 1816 #define MMAP_SIZE ULLONG_MAX 1817 #define NUM_MMAPS 1 1818 #else 1819 #define MMAP_SIZE (32 * 1024 * 1024ULL) 1820 #define NUM_MMAPS 128 1821 #endif 1822 1823 static int __perf_session__process_events(struct perf_session *session, 1824 u64 data_offset, u64 data_size, 1825 u64 file_size) 1826 { 1827 struct ordered_events *oe = &session->ordered_events; 1828 struct perf_tool *tool = session->tool; 1829 int fd = perf_data_file__fd(session->file); 1830 u64 head, page_offset, file_offset, file_pos, size; 1831 int err, mmap_prot, mmap_flags, map_idx = 0; 1832 size_t mmap_size; 1833 char *buf, *mmaps[NUM_MMAPS]; 1834 union perf_event *event; 1835 struct ui_progress prog; 1836 s64 skip; 1837 1838 perf_tool__fill_defaults(tool); 1839 1840 page_offset = page_size * (data_offset / page_size); 1841 file_offset = page_offset; 1842 head = data_offset - page_offset; 1843 1844 if (data_size == 0) 1845 goto out; 1846 1847 if (data_offset + data_size < file_size) 1848 file_size = data_offset + data_size; 1849 1850 ui_progress__init(&prog, file_size, "Processing events..."); 1851 1852 mmap_size = MMAP_SIZE; 1853 if (mmap_size > file_size) { 1854 mmap_size = file_size; 1855 session->one_mmap = true; 1856 } 1857 1858 memset(mmaps, 0, sizeof(mmaps)); 1859 1860 mmap_prot = PROT_READ; 1861 mmap_flags = MAP_SHARED; 1862 1863 if (session->header.needs_swap) { 1864 mmap_prot |= PROT_WRITE; 1865 mmap_flags = MAP_PRIVATE; 1866 } 1867 remap: 1868 buf = mmap(NULL, mmap_size, mmap_prot, mmap_flags, fd, 1869 file_offset); 1870 if (buf == MAP_FAILED) { 1871 pr_err("failed to mmap file\n"); 1872 err = -errno; 1873 goto out_err; 1874 } 1875 mmaps[map_idx] = buf; 1876 map_idx = (map_idx + 1) & (ARRAY_SIZE(mmaps) - 1); 1877 file_pos = file_offset + head; 1878 if (session->one_mmap) { 1879 session->one_mmap_addr = buf; 1880 session->one_mmap_offset = file_offset; 1881 } 1882 1883 more: 1884 event = fetch_mmaped_event(session, head, mmap_size, buf); 1885 if (!event) { 1886 if (mmaps[map_idx]) { 1887 munmap(mmaps[map_idx], mmap_size); 1888 mmaps[map_idx] = NULL; 1889 } 1890 1891 page_offset = page_size * (head / page_size); 1892 file_offset += page_offset; 1893 head -= page_offset; 1894 goto remap; 1895 } 1896 1897 size = event->header.size; 1898 1899 if (size < sizeof(struct perf_event_header) || 1900 (skip = perf_session__process_event(session, event, file_pos)) < 0) { 1901 pr_err("%#" PRIx64 " [%#x]: failed to process type: %d\n", 1902 file_offset + head, event->header.size, 1903 event->header.type); 1904 err = -EINVAL; 1905 goto out_err; 1906 } 1907 1908 if (skip) 1909 size += skip; 1910 1911 head += size; 1912 file_pos += size; 1913 1914 ui_progress__update(&prog, size); 1915 1916 if (session_done()) 1917 goto out; 1918 1919 if (file_pos < file_size) 1920 goto more; 1921 1922 out: 1923 /* do the final flush for ordered samples */ 1924 err = ordered_events__flush(oe, OE_FLUSH__FINAL); 1925 if (err) 1926 goto out_err; 1927 err = auxtrace__flush_events(session, tool); 1928 if (err) 1929 goto out_err; 1930 err = perf_session__flush_thread_stacks(session); 1931 out_err: 1932 ui_progress__finish(); 1933 perf_session__warn_about_errors(session); 1934 /* 1935 * We may switching perf.data output, make ordered_events 1936 * reusable. 1937 */ 1938 ordered_events__reinit(&session->ordered_events); 1939 auxtrace__free_events(session); 1940 session->one_mmap = false; 1941 return err; 1942 } 1943 1944 int perf_session__process_events(struct perf_session *session) 1945 { 1946 u64 size = perf_data_file__size(session->file); 1947 int err; 1948 1949 if (perf_session__register_idle_thread(session) < 0) 1950 return -ENOMEM; 1951 1952 if (!perf_data_file__is_pipe(session->file)) 1953 err = __perf_session__process_events(session, 1954 session->header.data_offset, 1955 session->header.data_size, size); 1956 else 1957 err = __perf_session__process_pipe_events(session); 1958 1959 return err; 1960 } 1961 1962 bool perf_session__has_traces(struct perf_session *session, const char *msg) 1963 { 1964 struct perf_evsel *evsel; 1965 1966 evlist__for_each_entry(session->evlist, evsel) { 1967 if (evsel->attr.type == PERF_TYPE_TRACEPOINT) 1968 return true; 1969 } 1970 1971 pr_err("No trace sample to read. Did you call 'perf %s'?\n", msg); 1972 return false; 1973 } 1974 1975 int maps__set_kallsyms_ref_reloc_sym(struct map **maps, 1976 const char *symbol_name, u64 addr) 1977 { 1978 char *bracket; 1979 int i; 1980 struct ref_reloc_sym *ref; 1981 1982 ref = zalloc(sizeof(struct ref_reloc_sym)); 1983 if (ref == NULL) 1984 return -ENOMEM; 1985 1986 ref->name = strdup(symbol_name); 1987 if (ref->name == NULL) { 1988 free(ref); 1989 return -ENOMEM; 1990 } 1991 1992 bracket = strchr(ref->name, ']'); 1993 if (bracket) 1994 *bracket = '\0'; 1995 1996 ref->addr = addr; 1997 1998 for (i = 0; i < MAP__NR_TYPES; ++i) { 1999 struct kmap *kmap = map__kmap(maps[i]); 2000 2001 if (!kmap) 2002 continue; 2003 kmap->ref_reloc_sym = ref; 2004 } 2005 2006 return 0; 2007 } 2008 2009 size_t perf_session__fprintf_dsos(struct perf_session *session, FILE *fp) 2010 { 2011 return machines__fprintf_dsos(&session->machines, fp); 2012 } 2013 2014 size_t perf_session__fprintf_dsos_buildid(struct perf_session *session, FILE *fp, 2015 bool (skip)(struct dso *dso, int parm), int parm) 2016 { 2017 return machines__fprintf_dsos_buildid(&session->machines, fp, skip, parm); 2018 } 2019 2020 size_t perf_session__fprintf_nr_events(struct perf_session *session, FILE *fp) 2021 { 2022 size_t ret; 2023 const char *msg = ""; 2024 2025 if (perf_header__has_feat(&session->header, HEADER_AUXTRACE)) 2026 msg = " (excludes AUX area (e.g. instruction trace) decoded / synthesized events)"; 2027 2028 ret = fprintf(fp, "\nAggregated stats:%s\n", msg); 2029 2030 ret += events_stats__fprintf(&session->evlist->stats, fp); 2031 return ret; 2032 } 2033 2034 size_t perf_session__fprintf(struct perf_session *session, FILE *fp) 2035 { 2036 /* 2037 * FIXME: Here we have to actually print all the machines in this 2038 * session, not just the host... 2039 */ 2040 return machine__fprintf(&session->machines.host, fp); 2041 } 2042 2043 struct perf_evsel *perf_session__find_first_evtype(struct perf_session *session, 2044 unsigned int type) 2045 { 2046 struct perf_evsel *pos; 2047 2048 evlist__for_each_entry(session->evlist, pos) { 2049 if (pos->attr.type == type) 2050 return pos; 2051 } 2052 return NULL; 2053 } 2054 2055 int perf_session__cpu_bitmap(struct perf_session *session, 2056 const char *cpu_list, unsigned long *cpu_bitmap) 2057 { 2058 int i, err = -1; 2059 struct cpu_map *map; 2060 2061 for (i = 0; i < PERF_TYPE_MAX; ++i) { 2062 struct perf_evsel *evsel; 2063 2064 evsel = perf_session__find_first_evtype(session, i); 2065 if (!evsel) 2066 continue; 2067 2068 if (!(evsel->attr.sample_type & PERF_SAMPLE_CPU)) { 2069 pr_err("File does not contain CPU events. " 2070 "Remove -C option to proceed.\n"); 2071 return -1; 2072 } 2073 } 2074 2075 map = cpu_map__new(cpu_list); 2076 if (map == NULL) { 2077 pr_err("Invalid cpu_list\n"); 2078 return -1; 2079 } 2080 2081 for (i = 0; i < map->nr; i++) { 2082 int cpu = map->map[i]; 2083 2084 if (cpu >= MAX_NR_CPUS) { 2085 pr_err("Requested CPU %d too large. " 2086 "Consider raising MAX_NR_CPUS\n", cpu); 2087 goto out_delete_map; 2088 } 2089 2090 set_bit(cpu, cpu_bitmap); 2091 } 2092 2093 err = 0; 2094 2095 out_delete_map: 2096 cpu_map__put(map); 2097 return err; 2098 } 2099 2100 void perf_session__fprintf_info(struct perf_session *session, FILE *fp, 2101 bool full) 2102 { 2103 if (session == NULL || fp == NULL) 2104 return; 2105 2106 fprintf(fp, "# ========\n"); 2107 perf_header__fprintf_info(session, fp, full); 2108 fprintf(fp, "# ========\n#\n"); 2109 } 2110 2111 2112 int __perf_session__set_tracepoints_handlers(struct perf_session *session, 2113 const struct perf_evsel_str_handler *assocs, 2114 size_t nr_assocs) 2115 { 2116 struct perf_evsel *evsel; 2117 size_t i; 2118 int err; 2119 2120 for (i = 0; i < nr_assocs; i++) { 2121 /* 2122 * Adding a handler for an event not in the session, 2123 * just ignore it. 2124 */ 2125 evsel = perf_evlist__find_tracepoint_by_name(session->evlist, assocs[i].name); 2126 if (evsel == NULL) 2127 continue; 2128 2129 err = -EEXIST; 2130 if (evsel->handler != NULL) 2131 goto out; 2132 evsel->handler = assocs[i].handler; 2133 } 2134 2135 err = 0; 2136 out: 2137 return err; 2138 } 2139 2140 int perf_event__process_id_index(struct perf_tool *tool __maybe_unused, 2141 union perf_event *event, 2142 struct perf_session *session) 2143 { 2144 struct perf_evlist *evlist = session->evlist; 2145 struct id_index_event *ie = &event->id_index; 2146 size_t i, nr, max_nr; 2147 2148 max_nr = (ie->header.size - sizeof(struct id_index_event)) / 2149 sizeof(struct id_index_entry); 2150 nr = ie->nr; 2151 if (nr > max_nr) 2152 return -EINVAL; 2153 2154 if (dump_trace) 2155 fprintf(stdout, " nr: %zu\n", nr); 2156 2157 for (i = 0; i < nr; i++) { 2158 struct id_index_entry *e = &ie->entries[i]; 2159 struct perf_sample_id *sid; 2160 2161 if (dump_trace) { 2162 fprintf(stdout, " ... id: %"PRIu64, e->id); 2163 fprintf(stdout, " idx: %"PRIu64, e->idx); 2164 fprintf(stdout, " cpu: %"PRId64, e->cpu); 2165 fprintf(stdout, " tid: %"PRId64"\n", e->tid); 2166 } 2167 2168 sid = perf_evlist__id2sid(evlist, e->id); 2169 if (!sid) 2170 return -ENOENT; 2171 sid->idx = e->idx; 2172 sid->cpu = e->cpu; 2173 sid->tid = e->tid; 2174 } 2175 return 0; 2176 } 2177 2178 int perf_event__synthesize_id_index(struct perf_tool *tool, 2179 perf_event__handler_t process, 2180 struct perf_evlist *evlist, 2181 struct machine *machine) 2182 { 2183 union perf_event *ev; 2184 struct perf_evsel *evsel; 2185 size_t nr = 0, i = 0, sz, max_nr, n; 2186 int err; 2187 2188 pr_debug2("Synthesizing id index\n"); 2189 2190 max_nr = (UINT16_MAX - sizeof(struct id_index_event)) / 2191 sizeof(struct id_index_entry); 2192 2193 evlist__for_each_entry(evlist, evsel) 2194 nr += evsel->ids; 2195 2196 n = nr > max_nr ? max_nr : nr; 2197 sz = sizeof(struct id_index_event) + n * sizeof(struct id_index_entry); 2198 ev = zalloc(sz); 2199 if (!ev) 2200 return -ENOMEM; 2201 2202 ev->id_index.header.type = PERF_RECORD_ID_INDEX; 2203 ev->id_index.header.size = sz; 2204 ev->id_index.nr = n; 2205 2206 evlist__for_each_entry(evlist, evsel) { 2207 u32 j; 2208 2209 for (j = 0; j < evsel->ids; j++) { 2210 struct id_index_entry *e; 2211 struct perf_sample_id *sid; 2212 2213 if (i >= n) { 2214 err = process(tool, ev, NULL, machine); 2215 if (err) 2216 goto out_err; 2217 nr -= n; 2218 i = 0; 2219 } 2220 2221 e = &ev->id_index.entries[i++]; 2222 2223 e->id = evsel->id[j]; 2224 2225 sid = perf_evlist__id2sid(evlist, e->id); 2226 if (!sid) { 2227 free(ev); 2228 return -ENOENT; 2229 } 2230 2231 e->idx = sid->idx; 2232 e->cpu = sid->cpu; 2233 e->tid = sid->tid; 2234 } 2235 } 2236 2237 sz = sizeof(struct id_index_event) + nr * sizeof(struct id_index_entry); 2238 ev->id_index.header.size = sz; 2239 ev->id_index.nr = nr; 2240 2241 err = process(tool, ev, NULL, machine); 2242 out_err: 2243 free(ev); 2244 2245 return err; 2246 } 2247