1 // SPDX-License-Identifier: GPL-2.0 2 #include <errno.h> 3 #include <inttypes.h> 4 #include <linux/kernel.h> 5 #include <traceevent/event-parse.h> 6 #include <api/fs/fs.h> 7 8 #include <byteswap.h> 9 #include <unistd.h> 10 #include <sys/types.h> 11 #include <sys/mman.h> 12 13 #include "evlist.h" 14 #include "evsel.h" 15 #include "memswap.h" 16 #include "map.h" 17 #include "symbol.h" 18 #include "session.h" 19 #include "tool.h" 20 #include "sort.h" 21 #include "util.h" 22 #include "cpumap.h" 23 #include "perf_regs.h" 24 #include "asm/bug.h" 25 #include "auxtrace.h" 26 #include "thread.h" 27 #include "thread-stack.h" 28 #include "sample-raw.h" 29 #include "stat.h" 30 #include "arch/common.h" 31 32 static int perf_session__deliver_event(struct perf_session *session, 33 union perf_event *event, 34 struct perf_tool *tool, 35 u64 file_offset); 36 37 static int perf_session__open(struct perf_session *session) 38 { 39 struct perf_data *data = session->data; 40 41 if (perf_session__read_header(session) < 0) { 42 pr_err("incompatible file format (rerun with -v to learn more)\n"); 43 return -1; 44 } 45 46 if (perf_data__is_pipe(data)) 47 return 0; 48 49 if (perf_header__has_feat(&session->header, HEADER_STAT)) 50 return 0; 51 52 if (!perf_evlist__valid_sample_type(session->evlist)) { 53 pr_err("non matching sample_type\n"); 54 return -1; 55 } 56 57 if (!perf_evlist__valid_sample_id_all(session->evlist)) { 58 pr_err("non matching sample_id_all\n"); 59 return -1; 60 } 61 62 if (!perf_evlist__valid_read_format(session->evlist)) { 63 pr_err("non matching read_format\n"); 64 return -1; 65 } 66 67 return 0; 68 } 69 70 void perf_session__set_id_hdr_size(struct perf_session *session) 71 { 72 u16 id_hdr_size = perf_evlist__id_hdr_size(session->evlist); 73 74 machines__set_id_hdr_size(&session->machines, id_hdr_size); 75 } 76 77 int perf_session__create_kernel_maps(struct perf_session *session) 78 { 79 int ret = machine__create_kernel_maps(&session->machines.host); 80 81 if (ret >= 0) 82 ret = machines__create_guest_kernel_maps(&session->machines); 83 return ret; 84 } 85 86 static void perf_session__destroy_kernel_maps(struct perf_session *session) 87 { 88 machines__destroy_kernel_maps(&session->machines); 89 } 90 91 static bool perf_session__has_comm_exec(struct perf_session *session) 92 { 93 struct perf_evsel *evsel; 94 95 evlist__for_each_entry(session->evlist, evsel) { 96 if (evsel->attr.comm_exec) 97 return true; 98 } 99 100 return false; 101 } 102 103 static void perf_session__set_comm_exec(struct perf_session *session) 104 { 105 bool comm_exec = perf_session__has_comm_exec(session); 106 107 machines__set_comm_exec(&session->machines, comm_exec); 108 } 109 110 static int ordered_events__deliver_event(struct ordered_events *oe, 111 struct ordered_event *event) 112 { 113 struct perf_session *session = container_of(oe, struct perf_session, 114 ordered_events); 115 116 return perf_session__deliver_event(session, event->event, 117 session->tool, event->file_offset); 118 } 119 120 struct perf_session *perf_session__new(struct perf_data *data, 121 bool repipe, struct perf_tool *tool) 122 { 123 struct perf_session *session = zalloc(sizeof(*session)); 124 125 if (!session) 126 goto out; 127 128 session->repipe = repipe; 129 session->tool = tool; 130 INIT_LIST_HEAD(&session->auxtrace_index); 131 machines__init(&session->machines); 132 ordered_events__init(&session->ordered_events, 133 ordered_events__deliver_event, NULL); 134 135 if (data) { 136 if (perf_data__open(data)) 137 goto out_delete; 138 139 session->data = data; 140 141 if (perf_data__is_read(data)) { 142 if (perf_session__open(session) < 0) 143 goto out_close; 144 145 /* 146 * set session attributes that are present in perf.data 147 * but not in pipe-mode. 148 */ 149 if (!data->is_pipe) { 150 perf_session__set_id_hdr_size(session); 151 perf_session__set_comm_exec(session); 152 } 153 154 perf_evlist__init_trace_event_sample_raw(session->evlist); 155 } 156 } else { 157 session->machines.host.env = &perf_env; 158 } 159 160 session->machines.host.single_address_space = 161 perf_env__single_address_space(session->machines.host.env); 162 163 if (!data || perf_data__is_write(data)) { 164 /* 165 * In O_RDONLY mode this will be performed when reading the 166 * kernel MMAP event, in perf_event__process_mmap(). 167 */ 168 if (perf_session__create_kernel_maps(session) < 0) 169 pr_warning("Cannot read kernel map\n"); 170 } 171 172 /* 173 * In pipe-mode, evlist is empty until PERF_RECORD_HEADER_ATTR is 174 * processed, so perf_evlist__sample_id_all is not meaningful here. 175 */ 176 if ((!data || !data->is_pipe) && tool && tool->ordering_requires_timestamps && 177 tool->ordered_events && !perf_evlist__sample_id_all(session->evlist)) { 178 dump_printf("WARNING: No sample_id_all support, falling back to unordered processing\n"); 179 tool->ordered_events = false; 180 } 181 182 return session; 183 184 out_close: 185 perf_data__close(data); 186 out_delete: 187 perf_session__delete(session); 188 out: 189 return NULL; 190 } 191 192 static void perf_session__delete_threads(struct perf_session *session) 193 { 194 machine__delete_threads(&session->machines.host); 195 } 196 197 void perf_session__delete(struct perf_session *session) 198 { 199 if (session == NULL) 200 return; 201 auxtrace__free(session); 202 auxtrace_index__free(&session->auxtrace_index); 203 perf_session__destroy_kernel_maps(session); 204 perf_session__delete_threads(session); 205 perf_env__exit(&session->header.env); 206 machines__exit(&session->machines); 207 if (session->data) 208 perf_data__close(session->data); 209 free(session); 210 } 211 212 static int process_event_synth_tracing_data_stub(struct perf_session *session 213 __maybe_unused, 214 union perf_event *event 215 __maybe_unused) 216 { 217 dump_printf(": unhandled!\n"); 218 return 0; 219 } 220 221 static int process_event_synth_attr_stub(struct perf_tool *tool __maybe_unused, 222 union perf_event *event __maybe_unused, 223 struct perf_evlist **pevlist 224 __maybe_unused) 225 { 226 dump_printf(": unhandled!\n"); 227 return 0; 228 } 229 230 static int process_event_synth_event_update_stub(struct perf_tool *tool __maybe_unused, 231 union perf_event *event __maybe_unused, 232 struct perf_evlist **pevlist 233 __maybe_unused) 234 { 235 if (dump_trace) 236 perf_event__fprintf_event_update(event, stdout); 237 238 dump_printf(": unhandled!\n"); 239 return 0; 240 } 241 242 static int process_event_sample_stub(struct perf_tool *tool __maybe_unused, 243 union perf_event *event __maybe_unused, 244 struct perf_sample *sample __maybe_unused, 245 struct perf_evsel *evsel __maybe_unused, 246 struct machine *machine __maybe_unused) 247 { 248 dump_printf(": unhandled!\n"); 249 return 0; 250 } 251 252 static int process_event_stub(struct perf_tool *tool __maybe_unused, 253 union perf_event *event __maybe_unused, 254 struct perf_sample *sample __maybe_unused, 255 struct machine *machine __maybe_unused) 256 { 257 dump_printf(": unhandled!\n"); 258 return 0; 259 } 260 261 static int process_finished_round_stub(struct perf_tool *tool __maybe_unused, 262 union perf_event *event __maybe_unused, 263 struct ordered_events *oe __maybe_unused) 264 { 265 dump_printf(": unhandled!\n"); 266 return 0; 267 } 268 269 static int process_finished_round(struct perf_tool *tool, 270 union perf_event *event, 271 struct ordered_events *oe); 272 273 static int skipn(int fd, off_t n) 274 { 275 char buf[4096]; 276 ssize_t ret; 277 278 while (n > 0) { 279 ret = read(fd, buf, min(n, (off_t)sizeof(buf))); 280 if (ret <= 0) 281 return ret; 282 n -= ret; 283 } 284 285 return 0; 286 } 287 288 static s64 process_event_auxtrace_stub(struct perf_session *session __maybe_unused, 289 union perf_event *event) 290 { 291 dump_printf(": unhandled!\n"); 292 if (perf_data__is_pipe(session->data)) 293 skipn(perf_data__fd(session->data), event->auxtrace.size); 294 return event->auxtrace.size; 295 } 296 297 static int process_event_op2_stub(struct perf_session *session __maybe_unused, 298 union perf_event *event __maybe_unused) 299 { 300 dump_printf(": unhandled!\n"); 301 return 0; 302 } 303 304 305 static 306 int process_event_thread_map_stub(struct perf_session *session __maybe_unused, 307 union perf_event *event __maybe_unused) 308 { 309 if (dump_trace) 310 perf_event__fprintf_thread_map(event, stdout); 311 312 dump_printf(": unhandled!\n"); 313 return 0; 314 } 315 316 static 317 int process_event_cpu_map_stub(struct perf_session *session __maybe_unused, 318 union perf_event *event __maybe_unused) 319 { 320 if (dump_trace) 321 perf_event__fprintf_cpu_map(event, stdout); 322 323 dump_printf(": unhandled!\n"); 324 return 0; 325 } 326 327 static 328 int process_event_stat_config_stub(struct perf_session *session __maybe_unused, 329 union perf_event *event __maybe_unused) 330 { 331 if (dump_trace) 332 perf_event__fprintf_stat_config(event, stdout); 333 334 dump_printf(": unhandled!\n"); 335 return 0; 336 } 337 338 static int process_stat_stub(struct perf_session *perf_session __maybe_unused, 339 union perf_event *event) 340 { 341 if (dump_trace) 342 perf_event__fprintf_stat(event, stdout); 343 344 dump_printf(": unhandled!\n"); 345 return 0; 346 } 347 348 static int process_stat_round_stub(struct perf_session *perf_session __maybe_unused, 349 union perf_event *event) 350 { 351 if (dump_trace) 352 perf_event__fprintf_stat_round(event, stdout); 353 354 dump_printf(": unhandled!\n"); 355 return 0; 356 } 357 358 void perf_tool__fill_defaults(struct perf_tool *tool) 359 { 360 if (tool->sample == NULL) 361 tool->sample = process_event_sample_stub; 362 if (tool->mmap == NULL) 363 tool->mmap = process_event_stub; 364 if (tool->mmap2 == NULL) 365 tool->mmap2 = process_event_stub; 366 if (tool->comm == NULL) 367 tool->comm = process_event_stub; 368 if (tool->namespaces == NULL) 369 tool->namespaces = process_event_stub; 370 if (tool->fork == NULL) 371 tool->fork = process_event_stub; 372 if (tool->exit == NULL) 373 tool->exit = process_event_stub; 374 if (tool->lost == NULL) 375 tool->lost = perf_event__process_lost; 376 if (tool->lost_samples == NULL) 377 tool->lost_samples = perf_event__process_lost_samples; 378 if (tool->aux == NULL) 379 tool->aux = perf_event__process_aux; 380 if (tool->itrace_start == NULL) 381 tool->itrace_start = perf_event__process_itrace_start; 382 if (tool->context_switch == NULL) 383 tool->context_switch = perf_event__process_switch; 384 if (tool->ksymbol == NULL) 385 tool->ksymbol = perf_event__process_ksymbol; 386 if (tool->bpf_event == NULL) 387 tool->bpf_event = perf_event__process_bpf_event; 388 if (tool->read == NULL) 389 tool->read = process_event_sample_stub; 390 if (tool->throttle == NULL) 391 tool->throttle = process_event_stub; 392 if (tool->unthrottle == NULL) 393 tool->unthrottle = process_event_stub; 394 if (tool->attr == NULL) 395 tool->attr = process_event_synth_attr_stub; 396 if (tool->event_update == NULL) 397 tool->event_update = process_event_synth_event_update_stub; 398 if (tool->tracing_data == NULL) 399 tool->tracing_data = process_event_synth_tracing_data_stub; 400 if (tool->build_id == NULL) 401 tool->build_id = process_event_op2_stub; 402 if (tool->finished_round == NULL) { 403 if (tool->ordered_events) 404 tool->finished_round = process_finished_round; 405 else 406 tool->finished_round = process_finished_round_stub; 407 } 408 if (tool->id_index == NULL) 409 tool->id_index = process_event_op2_stub; 410 if (tool->auxtrace_info == NULL) 411 tool->auxtrace_info = process_event_op2_stub; 412 if (tool->auxtrace == NULL) 413 tool->auxtrace = process_event_auxtrace_stub; 414 if (tool->auxtrace_error == NULL) 415 tool->auxtrace_error = process_event_op2_stub; 416 if (tool->thread_map == NULL) 417 tool->thread_map = process_event_thread_map_stub; 418 if (tool->cpu_map == NULL) 419 tool->cpu_map = process_event_cpu_map_stub; 420 if (tool->stat_config == NULL) 421 tool->stat_config = process_event_stat_config_stub; 422 if (tool->stat == NULL) 423 tool->stat = process_stat_stub; 424 if (tool->stat_round == NULL) 425 tool->stat_round = process_stat_round_stub; 426 if (tool->time_conv == NULL) 427 tool->time_conv = process_event_op2_stub; 428 if (tool->feature == NULL) 429 tool->feature = process_event_op2_stub; 430 } 431 432 static void swap_sample_id_all(union perf_event *event, void *data) 433 { 434 void *end = (void *) event + event->header.size; 435 int size = end - data; 436 437 BUG_ON(size % sizeof(u64)); 438 mem_bswap_64(data, size); 439 } 440 441 static void perf_event__all64_swap(union perf_event *event, 442 bool sample_id_all __maybe_unused) 443 { 444 struct perf_event_header *hdr = &event->header; 445 mem_bswap_64(hdr + 1, event->header.size - sizeof(*hdr)); 446 } 447 448 static void perf_event__comm_swap(union perf_event *event, bool sample_id_all) 449 { 450 event->comm.pid = bswap_32(event->comm.pid); 451 event->comm.tid = bswap_32(event->comm.tid); 452 453 if (sample_id_all) { 454 void *data = &event->comm.comm; 455 456 data += PERF_ALIGN(strlen(data) + 1, sizeof(u64)); 457 swap_sample_id_all(event, data); 458 } 459 } 460 461 static void perf_event__mmap_swap(union perf_event *event, 462 bool sample_id_all) 463 { 464 event->mmap.pid = bswap_32(event->mmap.pid); 465 event->mmap.tid = bswap_32(event->mmap.tid); 466 event->mmap.start = bswap_64(event->mmap.start); 467 event->mmap.len = bswap_64(event->mmap.len); 468 event->mmap.pgoff = bswap_64(event->mmap.pgoff); 469 470 if (sample_id_all) { 471 void *data = &event->mmap.filename; 472 473 data += PERF_ALIGN(strlen(data) + 1, sizeof(u64)); 474 swap_sample_id_all(event, data); 475 } 476 } 477 478 static void perf_event__mmap2_swap(union perf_event *event, 479 bool sample_id_all) 480 { 481 event->mmap2.pid = bswap_32(event->mmap2.pid); 482 event->mmap2.tid = bswap_32(event->mmap2.tid); 483 event->mmap2.start = bswap_64(event->mmap2.start); 484 event->mmap2.len = bswap_64(event->mmap2.len); 485 event->mmap2.pgoff = bswap_64(event->mmap2.pgoff); 486 event->mmap2.maj = bswap_32(event->mmap2.maj); 487 event->mmap2.min = bswap_32(event->mmap2.min); 488 event->mmap2.ino = bswap_64(event->mmap2.ino); 489 490 if (sample_id_all) { 491 void *data = &event->mmap2.filename; 492 493 data += PERF_ALIGN(strlen(data) + 1, sizeof(u64)); 494 swap_sample_id_all(event, data); 495 } 496 } 497 static void perf_event__task_swap(union perf_event *event, bool sample_id_all) 498 { 499 event->fork.pid = bswap_32(event->fork.pid); 500 event->fork.tid = bswap_32(event->fork.tid); 501 event->fork.ppid = bswap_32(event->fork.ppid); 502 event->fork.ptid = bswap_32(event->fork.ptid); 503 event->fork.time = bswap_64(event->fork.time); 504 505 if (sample_id_all) 506 swap_sample_id_all(event, &event->fork + 1); 507 } 508 509 static void perf_event__read_swap(union perf_event *event, bool sample_id_all) 510 { 511 event->read.pid = bswap_32(event->read.pid); 512 event->read.tid = bswap_32(event->read.tid); 513 event->read.value = bswap_64(event->read.value); 514 event->read.time_enabled = bswap_64(event->read.time_enabled); 515 event->read.time_running = bswap_64(event->read.time_running); 516 event->read.id = bswap_64(event->read.id); 517 518 if (sample_id_all) 519 swap_sample_id_all(event, &event->read + 1); 520 } 521 522 static void perf_event__aux_swap(union perf_event *event, bool sample_id_all) 523 { 524 event->aux.aux_offset = bswap_64(event->aux.aux_offset); 525 event->aux.aux_size = bswap_64(event->aux.aux_size); 526 event->aux.flags = bswap_64(event->aux.flags); 527 528 if (sample_id_all) 529 swap_sample_id_all(event, &event->aux + 1); 530 } 531 532 static void perf_event__itrace_start_swap(union perf_event *event, 533 bool sample_id_all) 534 { 535 event->itrace_start.pid = bswap_32(event->itrace_start.pid); 536 event->itrace_start.tid = bswap_32(event->itrace_start.tid); 537 538 if (sample_id_all) 539 swap_sample_id_all(event, &event->itrace_start + 1); 540 } 541 542 static void perf_event__switch_swap(union perf_event *event, bool sample_id_all) 543 { 544 if (event->header.type == PERF_RECORD_SWITCH_CPU_WIDE) { 545 event->context_switch.next_prev_pid = 546 bswap_32(event->context_switch.next_prev_pid); 547 event->context_switch.next_prev_tid = 548 bswap_32(event->context_switch.next_prev_tid); 549 } 550 551 if (sample_id_all) 552 swap_sample_id_all(event, &event->context_switch + 1); 553 } 554 555 static void perf_event__throttle_swap(union perf_event *event, 556 bool sample_id_all) 557 { 558 event->throttle.time = bswap_64(event->throttle.time); 559 event->throttle.id = bswap_64(event->throttle.id); 560 event->throttle.stream_id = bswap_64(event->throttle.stream_id); 561 562 if (sample_id_all) 563 swap_sample_id_all(event, &event->throttle + 1); 564 } 565 566 static u8 revbyte(u8 b) 567 { 568 int rev = (b >> 4) | ((b & 0xf) << 4); 569 rev = ((rev & 0xcc) >> 2) | ((rev & 0x33) << 2); 570 rev = ((rev & 0xaa) >> 1) | ((rev & 0x55) << 1); 571 return (u8) rev; 572 } 573 574 /* 575 * XXX this is hack in attempt to carry flags bitfield 576 * through endian village. ABI says: 577 * 578 * Bit-fields are allocated from right to left (least to most significant) 579 * on little-endian implementations and from left to right (most to least 580 * significant) on big-endian implementations. 581 * 582 * The above seems to be byte specific, so we need to reverse each 583 * byte of the bitfield. 'Internet' also says this might be implementation 584 * specific and we probably need proper fix and carry perf_event_attr 585 * bitfield flags in separate data file FEAT_ section. Thought this seems 586 * to work for now. 587 */ 588 static void swap_bitfield(u8 *p, unsigned len) 589 { 590 unsigned i; 591 592 for (i = 0; i < len; i++) { 593 *p = revbyte(*p); 594 p++; 595 } 596 } 597 598 /* exported for swapping attributes in file header */ 599 void perf_event__attr_swap(struct perf_event_attr *attr) 600 { 601 attr->type = bswap_32(attr->type); 602 attr->size = bswap_32(attr->size); 603 604 #define bswap_safe(f, n) \ 605 (attr->size > (offsetof(struct perf_event_attr, f) + \ 606 sizeof(attr->f) * (n))) 607 #define bswap_field(f, sz) \ 608 do { \ 609 if (bswap_safe(f, 0)) \ 610 attr->f = bswap_##sz(attr->f); \ 611 } while(0) 612 #define bswap_field_16(f) bswap_field(f, 16) 613 #define bswap_field_32(f) bswap_field(f, 32) 614 #define bswap_field_64(f) bswap_field(f, 64) 615 616 bswap_field_64(config); 617 bswap_field_64(sample_period); 618 bswap_field_64(sample_type); 619 bswap_field_64(read_format); 620 bswap_field_32(wakeup_events); 621 bswap_field_32(bp_type); 622 bswap_field_64(bp_addr); 623 bswap_field_64(bp_len); 624 bswap_field_64(branch_sample_type); 625 bswap_field_64(sample_regs_user); 626 bswap_field_32(sample_stack_user); 627 bswap_field_32(aux_watermark); 628 bswap_field_16(sample_max_stack); 629 630 /* 631 * After read_format are bitfields. Check read_format because 632 * we are unable to use offsetof on bitfield. 633 */ 634 if (bswap_safe(read_format, 1)) 635 swap_bitfield((u8 *) (&attr->read_format + 1), 636 sizeof(u64)); 637 #undef bswap_field_64 638 #undef bswap_field_32 639 #undef bswap_field 640 #undef bswap_safe 641 } 642 643 static void perf_event__hdr_attr_swap(union perf_event *event, 644 bool sample_id_all __maybe_unused) 645 { 646 size_t size; 647 648 perf_event__attr_swap(&event->attr.attr); 649 650 size = event->header.size; 651 size -= (void *)&event->attr.id - (void *)event; 652 mem_bswap_64(event->attr.id, size); 653 } 654 655 static void perf_event__event_update_swap(union perf_event *event, 656 bool sample_id_all __maybe_unused) 657 { 658 event->event_update.type = bswap_64(event->event_update.type); 659 event->event_update.id = bswap_64(event->event_update.id); 660 } 661 662 static void perf_event__event_type_swap(union perf_event *event, 663 bool sample_id_all __maybe_unused) 664 { 665 event->event_type.event_type.event_id = 666 bswap_64(event->event_type.event_type.event_id); 667 } 668 669 static void perf_event__tracing_data_swap(union perf_event *event, 670 bool sample_id_all __maybe_unused) 671 { 672 event->tracing_data.size = bswap_32(event->tracing_data.size); 673 } 674 675 static void perf_event__auxtrace_info_swap(union perf_event *event, 676 bool sample_id_all __maybe_unused) 677 { 678 size_t size; 679 680 event->auxtrace_info.type = bswap_32(event->auxtrace_info.type); 681 682 size = event->header.size; 683 size -= (void *)&event->auxtrace_info.priv - (void *)event; 684 mem_bswap_64(event->auxtrace_info.priv, size); 685 } 686 687 static void perf_event__auxtrace_swap(union perf_event *event, 688 bool sample_id_all __maybe_unused) 689 { 690 event->auxtrace.size = bswap_64(event->auxtrace.size); 691 event->auxtrace.offset = bswap_64(event->auxtrace.offset); 692 event->auxtrace.reference = bswap_64(event->auxtrace.reference); 693 event->auxtrace.idx = bswap_32(event->auxtrace.idx); 694 event->auxtrace.tid = bswap_32(event->auxtrace.tid); 695 event->auxtrace.cpu = bswap_32(event->auxtrace.cpu); 696 } 697 698 static void perf_event__auxtrace_error_swap(union perf_event *event, 699 bool sample_id_all __maybe_unused) 700 { 701 event->auxtrace_error.type = bswap_32(event->auxtrace_error.type); 702 event->auxtrace_error.code = bswap_32(event->auxtrace_error.code); 703 event->auxtrace_error.cpu = bswap_32(event->auxtrace_error.cpu); 704 event->auxtrace_error.pid = bswap_32(event->auxtrace_error.pid); 705 event->auxtrace_error.tid = bswap_32(event->auxtrace_error.tid); 706 event->auxtrace_error.fmt = bswap_32(event->auxtrace_error.fmt); 707 event->auxtrace_error.ip = bswap_64(event->auxtrace_error.ip); 708 if (event->auxtrace_error.fmt) 709 event->auxtrace_error.time = bswap_64(event->auxtrace_error.time); 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 u64 timestamp, u64 file_offset) 874 { 875 return ordered_events__queue(&s->ordered_events, event, timestamp, 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 if (event->header.type == PERF_RECORD_SAMPLE && evlist->trace_event_sample_raw) 1081 evlist->trace_event_sample_raw(evlist, event, sample); 1082 1083 if (sample) 1084 perf_evlist__print_tstamp(evlist, event, sample); 1085 1086 printf("%#" PRIx64 " [%#x]: PERF_RECORD_%s", file_offset, 1087 event->header.size, perf_event__name(event->header.type)); 1088 } 1089 1090 static void dump_sample(struct perf_evsel *evsel, union perf_event *event, 1091 struct perf_sample *sample) 1092 { 1093 u64 sample_type; 1094 1095 if (!dump_trace) 1096 return; 1097 1098 printf("(IP, 0x%x): %d/%d: %#" PRIx64 " period: %" PRIu64 " addr: %#" PRIx64 "\n", 1099 event->header.misc, sample->pid, sample->tid, sample->ip, 1100 sample->period, sample->addr); 1101 1102 sample_type = evsel->attr.sample_type; 1103 1104 if (evsel__has_callchain(evsel)) 1105 callchain__printf(evsel, sample); 1106 1107 if ((sample_type & PERF_SAMPLE_BRANCH_STACK) && !perf_evsel__has_branch_callstack(evsel)) 1108 branch_stack__printf(sample); 1109 1110 if (sample_type & PERF_SAMPLE_REGS_USER) 1111 regs_user__printf(sample); 1112 1113 if (sample_type & PERF_SAMPLE_REGS_INTR) 1114 regs_intr__printf(sample); 1115 1116 if (sample_type & PERF_SAMPLE_STACK_USER) 1117 stack_user__printf(&sample->user_stack); 1118 1119 if (sample_type & PERF_SAMPLE_WEIGHT) 1120 printf("... weight: %" PRIu64 "\n", sample->weight); 1121 1122 if (sample_type & PERF_SAMPLE_DATA_SRC) 1123 printf(" . data_src: 0x%"PRIx64"\n", sample->data_src); 1124 1125 if (sample_type & PERF_SAMPLE_PHYS_ADDR) 1126 printf(" .. phys_addr: 0x%"PRIx64"\n", sample->phys_addr); 1127 1128 if (sample_type & PERF_SAMPLE_TRANSACTION) 1129 printf("... transaction: %" PRIx64 "\n", sample->transaction); 1130 1131 if (sample_type & PERF_SAMPLE_READ) 1132 sample_read__printf(sample, evsel->attr.read_format); 1133 } 1134 1135 static void dump_read(struct perf_evsel *evsel, union perf_event *event) 1136 { 1137 struct read_event *read_event = &event->read; 1138 u64 read_format; 1139 1140 if (!dump_trace) 1141 return; 1142 1143 printf(": %d %d %s %" PRIu64 "\n", event->read.pid, event->read.tid, 1144 evsel ? perf_evsel__name(evsel) : "FAIL", 1145 event->read.value); 1146 1147 read_format = evsel->attr.read_format; 1148 1149 if (read_format & PERF_FORMAT_TOTAL_TIME_ENABLED) 1150 printf("... time enabled : %" PRIu64 "\n", read_event->time_enabled); 1151 1152 if (read_format & PERF_FORMAT_TOTAL_TIME_RUNNING) 1153 printf("... time running : %" PRIu64 "\n", read_event->time_running); 1154 1155 if (read_format & PERF_FORMAT_ID) 1156 printf("... id : %" PRIu64 "\n", read_event->id); 1157 } 1158 1159 static struct machine *machines__find_for_cpumode(struct machines *machines, 1160 union perf_event *event, 1161 struct perf_sample *sample) 1162 { 1163 struct machine *machine; 1164 1165 if (perf_guest && 1166 ((sample->cpumode == PERF_RECORD_MISC_GUEST_KERNEL) || 1167 (sample->cpumode == PERF_RECORD_MISC_GUEST_USER))) { 1168 u32 pid; 1169 1170 if (event->header.type == PERF_RECORD_MMAP 1171 || event->header.type == PERF_RECORD_MMAP2) 1172 pid = event->mmap.pid; 1173 else 1174 pid = sample->pid; 1175 1176 machine = machines__find(machines, pid); 1177 if (!machine) 1178 machine = machines__findnew(machines, DEFAULT_GUEST_KERNEL_ID); 1179 return machine; 1180 } 1181 1182 return &machines->host; 1183 } 1184 1185 static int deliver_sample_value(struct perf_evlist *evlist, 1186 struct perf_tool *tool, 1187 union perf_event *event, 1188 struct perf_sample *sample, 1189 struct sample_read_value *v, 1190 struct machine *machine) 1191 { 1192 struct perf_sample_id *sid = perf_evlist__id2sid(evlist, v->id); 1193 1194 if (sid) { 1195 sample->id = v->id; 1196 sample->period = v->value - sid->period; 1197 sid->period = v->value; 1198 } 1199 1200 if (!sid || sid->evsel == NULL) { 1201 ++evlist->stats.nr_unknown_id; 1202 return 0; 1203 } 1204 1205 return tool->sample(tool, event, sample, sid->evsel, machine); 1206 } 1207 1208 static int deliver_sample_group(struct perf_evlist *evlist, 1209 struct perf_tool *tool, 1210 union perf_event *event, 1211 struct perf_sample *sample, 1212 struct machine *machine) 1213 { 1214 int ret = -EINVAL; 1215 u64 i; 1216 1217 for (i = 0; i < sample->read.group.nr; i++) { 1218 ret = deliver_sample_value(evlist, tool, event, sample, 1219 &sample->read.group.values[i], 1220 machine); 1221 if (ret) 1222 break; 1223 } 1224 1225 return ret; 1226 } 1227 1228 static int 1229 perf_evlist__deliver_sample(struct perf_evlist *evlist, 1230 struct perf_tool *tool, 1231 union perf_event *event, 1232 struct perf_sample *sample, 1233 struct perf_evsel *evsel, 1234 struct machine *machine) 1235 { 1236 /* We know evsel != NULL. */ 1237 u64 sample_type = evsel->attr.sample_type; 1238 u64 read_format = evsel->attr.read_format; 1239 1240 /* Standard sample delivery. */ 1241 if (!(sample_type & PERF_SAMPLE_READ)) 1242 return tool->sample(tool, event, sample, evsel, machine); 1243 1244 /* For PERF_SAMPLE_READ we have either single or group mode. */ 1245 if (read_format & PERF_FORMAT_GROUP) 1246 return deliver_sample_group(evlist, tool, event, sample, 1247 machine); 1248 else 1249 return deliver_sample_value(evlist, tool, event, sample, 1250 &sample->read.one, machine); 1251 } 1252 1253 static int machines__deliver_event(struct machines *machines, 1254 struct perf_evlist *evlist, 1255 union perf_event *event, 1256 struct perf_sample *sample, 1257 struct perf_tool *tool, u64 file_offset) 1258 { 1259 struct perf_evsel *evsel; 1260 struct machine *machine; 1261 1262 dump_event(evlist, event, file_offset, sample); 1263 1264 evsel = perf_evlist__id2evsel(evlist, sample->id); 1265 1266 machine = machines__find_for_cpumode(machines, event, sample); 1267 1268 switch (event->header.type) { 1269 case PERF_RECORD_SAMPLE: 1270 if (evsel == NULL) { 1271 ++evlist->stats.nr_unknown_id; 1272 return 0; 1273 } 1274 dump_sample(evsel, event, sample); 1275 if (machine == NULL) { 1276 ++evlist->stats.nr_unprocessable_samples; 1277 return 0; 1278 } 1279 return perf_evlist__deliver_sample(evlist, tool, event, sample, evsel, machine); 1280 case PERF_RECORD_MMAP: 1281 return tool->mmap(tool, event, sample, machine); 1282 case PERF_RECORD_MMAP2: 1283 if (event->header.misc & PERF_RECORD_MISC_PROC_MAP_PARSE_TIMEOUT) 1284 ++evlist->stats.nr_proc_map_timeout; 1285 return tool->mmap2(tool, event, sample, machine); 1286 case PERF_RECORD_COMM: 1287 return tool->comm(tool, event, sample, machine); 1288 case PERF_RECORD_NAMESPACES: 1289 return tool->namespaces(tool, event, sample, machine); 1290 case PERF_RECORD_FORK: 1291 return tool->fork(tool, event, sample, machine); 1292 case PERF_RECORD_EXIT: 1293 return tool->exit(tool, event, sample, machine); 1294 case PERF_RECORD_LOST: 1295 if (tool->lost == perf_event__process_lost) 1296 evlist->stats.total_lost += event->lost.lost; 1297 return tool->lost(tool, event, sample, machine); 1298 case PERF_RECORD_LOST_SAMPLES: 1299 if (tool->lost_samples == perf_event__process_lost_samples) 1300 evlist->stats.total_lost_samples += event->lost_samples.lost; 1301 return tool->lost_samples(tool, event, sample, machine); 1302 case PERF_RECORD_READ: 1303 dump_read(evsel, event); 1304 return tool->read(tool, event, sample, evsel, machine); 1305 case PERF_RECORD_THROTTLE: 1306 return tool->throttle(tool, event, sample, machine); 1307 case PERF_RECORD_UNTHROTTLE: 1308 return tool->unthrottle(tool, event, sample, machine); 1309 case PERF_RECORD_AUX: 1310 if (tool->aux == perf_event__process_aux) { 1311 if (event->aux.flags & PERF_AUX_FLAG_TRUNCATED) 1312 evlist->stats.total_aux_lost += 1; 1313 if (event->aux.flags & PERF_AUX_FLAG_PARTIAL) 1314 evlist->stats.total_aux_partial += 1; 1315 } 1316 return tool->aux(tool, event, sample, machine); 1317 case PERF_RECORD_ITRACE_START: 1318 return tool->itrace_start(tool, event, sample, machine); 1319 case PERF_RECORD_SWITCH: 1320 case PERF_RECORD_SWITCH_CPU_WIDE: 1321 return tool->context_switch(tool, event, sample, machine); 1322 case PERF_RECORD_KSYMBOL: 1323 return tool->ksymbol(tool, event, sample, machine); 1324 case PERF_RECORD_BPF_EVENT: 1325 return tool->bpf_event(tool, event, sample, machine); 1326 default: 1327 ++evlist->stats.nr_unknown_events; 1328 return -1; 1329 } 1330 } 1331 1332 static int perf_session__deliver_event(struct perf_session *session, 1333 union perf_event *event, 1334 struct perf_tool *tool, 1335 u64 file_offset) 1336 { 1337 struct perf_sample sample; 1338 int ret; 1339 1340 ret = perf_evlist__parse_sample(session->evlist, event, &sample); 1341 if (ret) { 1342 pr_err("Can't parse sample, err = %d\n", ret); 1343 return ret; 1344 } 1345 1346 ret = auxtrace__process_event(session, event, &sample, tool); 1347 if (ret < 0) 1348 return ret; 1349 if (ret > 0) 1350 return 0; 1351 1352 return machines__deliver_event(&session->machines, session->evlist, 1353 event, &sample, tool, file_offset); 1354 } 1355 1356 static s64 perf_session__process_user_event(struct perf_session *session, 1357 union perf_event *event, 1358 u64 file_offset) 1359 { 1360 struct ordered_events *oe = &session->ordered_events; 1361 struct perf_tool *tool = session->tool; 1362 struct perf_sample sample = { .time = 0, }; 1363 int fd = perf_data__fd(session->data); 1364 int err; 1365 1366 dump_event(session->evlist, event, file_offset, &sample); 1367 1368 /* These events are processed right away */ 1369 switch (event->header.type) { 1370 case PERF_RECORD_HEADER_ATTR: 1371 err = tool->attr(tool, event, &session->evlist); 1372 if (err == 0) { 1373 perf_session__set_id_hdr_size(session); 1374 perf_session__set_comm_exec(session); 1375 } 1376 return err; 1377 case PERF_RECORD_EVENT_UPDATE: 1378 return tool->event_update(tool, event, &session->evlist); 1379 case PERF_RECORD_HEADER_EVENT_TYPE: 1380 /* 1381 * Depreceated, but we need to handle it for sake 1382 * of old data files create in pipe mode. 1383 */ 1384 return 0; 1385 case PERF_RECORD_HEADER_TRACING_DATA: 1386 /* setup for reading amidst mmap */ 1387 lseek(fd, file_offset, SEEK_SET); 1388 return tool->tracing_data(session, event); 1389 case PERF_RECORD_HEADER_BUILD_ID: 1390 return tool->build_id(session, event); 1391 case PERF_RECORD_FINISHED_ROUND: 1392 return tool->finished_round(tool, event, oe); 1393 case PERF_RECORD_ID_INDEX: 1394 return tool->id_index(session, event); 1395 case PERF_RECORD_AUXTRACE_INFO: 1396 return tool->auxtrace_info(session, event); 1397 case PERF_RECORD_AUXTRACE: 1398 /* setup for reading amidst mmap */ 1399 lseek(fd, file_offset + event->header.size, SEEK_SET); 1400 return tool->auxtrace(session, event); 1401 case PERF_RECORD_AUXTRACE_ERROR: 1402 perf_session__auxtrace_error_inc(session, event); 1403 return tool->auxtrace_error(session, event); 1404 case PERF_RECORD_THREAD_MAP: 1405 return tool->thread_map(session, event); 1406 case PERF_RECORD_CPU_MAP: 1407 return tool->cpu_map(session, event); 1408 case PERF_RECORD_STAT_CONFIG: 1409 return tool->stat_config(session, event); 1410 case PERF_RECORD_STAT: 1411 return tool->stat(session, event); 1412 case PERF_RECORD_STAT_ROUND: 1413 return tool->stat_round(session, event); 1414 case PERF_RECORD_TIME_CONV: 1415 session->time_conv = event->time_conv; 1416 return tool->time_conv(session, event); 1417 case PERF_RECORD_HEADER_FEATURE: 1418 return tool->feature(session, event); 1419 default: 1420 return -EINVAL; 1421 } 1422 } 1423 1424 int perf_session__deliver_synth_event(struct perf_session *session, 1425 union perf_event *event, 1426 struct perf_sample *sample) 1427 { 1428 struct perf_evlist *evlist = session->evlist; 1429 struct perf_tool *tool = session->tool; 1430 1431 events_stats__inc(&evlist->stats, event->header.type); 1432 1433 if (event->header.type >= PERF_RECORD_USER_TYPE_START) 1434 return perf_session__process_user_event(session, event, 0); 1435 1436 return machines__deliver_event(&session->machines, evlist, event, sample, tool, 0); 1437 } 1438 1439 static void event_swap(union perf_event *event, bool sample_id_all) 1440 { 1441 perf_event__swap_op swap; 1442 1443 swap = perf_event__swap_ops[event->header.type]; 1444 if (swap) 1445 swap(event, sample_id_all); 1446 } 1447 1448 int perf_session__peek_event(struct perf_session *session, off_t file_offset, 1449 void *buf, size_t buf_sz, 1450 union perf_event **event_ptr, 1451 struct perf_sample *sample) 1452 { 1453 union perf_event *event; 1454 size_t hdr_sz, rest; 1455 int fd; 1456 1457 if (session->one_mmap && !session->header.needs_swap) { 1458 event = file_offset - session->one_mmap_offset + 1459 session->one_mmap_addr; 1460 goto out_parse_sample; 1461 } 1462 1463 if (perf_data__is_pipe(session->data)) 1464 return -1; 1465 1466 fd = perf_data__fd(session->data); 1467 hdr_sz = sizeof(struct perf_event_header); 1468 1469 if (buf_sz < hdr_sz) 1470 return -1; 1471 1472 if (lseek(fd, file_offset, SEEK_SET) == (off_t)-1 || 1473 readn(fd, buf, hdr_sz) != (ssize_t)hdr_sz) 1474 return -1; 1475 1476 event = (union perf_event *)buf; 1477 1478 if (session->header.needs_swap) 1479 perf_event_header__bswap(&event->header); 1480 1481 if (event->header.size < hdr_sz || event->header.size > buf_sz) 1482 return -1; 1483 1484 rest = event->header.size - hdr_sz; 1485 1486 if (readn(fd, buf, rest) != (ssize_t)rest) 1487 return -1; 1488 1489 if (session->header.needs_swap) 1490 event_swap(event, perf_evlist__sample_id_all(session->evlist)); 1491 1492 out_parse_sample: 1493 1494 if (sample && event->header.type < PERF_RECORD_USER_TYPE_START && 1495 perf_evlist__parse_sample(session->evlist, event, sample)) 1496 return -1; 1497 1498 *event_ptr = event; 1499 1500 return 0; 1501 } 1502 1503 static s64 perf_session__process_event(struct perf_session *session, 1504 union perf_event *event, u64 file_offset) 1505 { 1506 struct perf_evlist *evlist = session->evlist; 1507 struct perf_tool *tool = session->tool; 1508 int ret; 1509 1510 if (session->header.needs_swap) 1511 event_swap(event, perf_evlist__sample_id_all(evlist)); 1512 1513 if (event->header.type >= PERF_RECORD_HEADER_MAX) 1514 return -EINVAL; 1515 1516 events_stats__inc(&evlist->stats, event->header.type); 1517 1518 if (event->header.type >= PERF_RECORD_USER_TYPE_START) 1519 return perf_session__process_user_event(session, event, file_offset); 1520 1521 if (tool->ordered_events) { 1522 u64 timestamp = -1ULL; 1523 1524 ret = perf_evlist__parse_sample_timestamp(evlist, event, ×tamp); 1525 if (ret && ret != -1) 1526 return ret; 1527 1528 ret = perf_session__queue_event(session, event, timestamp, file_offset); 1529 if (ret != -ETIME) 1530 return ret; 1531 } 1532 1533 return perf_session__deliver_event(session, event, tool, file_offset); 1534 } 1535 1536 void perf_event_header__bswap(struct perf_event_header *hdr) 1537 { 1538 hdr->type = bswap_32(hdr->type); 1539 hdr->misc = bswap_16(hdr->misc); 1540 hdr->size = bswap_16(hdr->size); 1541 } 1542 1543 struct thread *perf_session__findnew(struct perf_session *session, pid_t pid) 1544 { 1545 return machine__findnew_thread(&session->machines.host, -1, pid); 1546 } 1547 1548 /* 1549 * Threads are identified by pid and tid, and the idle task has pid == tid == 0. 1550 * So here a single thread is created for that, but actually there is a separate 1551 * idle task per cpu, so there should be one 'struct thread' per cpu, but there 1552 * is only 1. That causes problems for some tools, requiring workarounds. For 1553 * example get_idle_thread() in builtin-sched.c, or thread_stack__per_cpu(). 1554 */ 1555 int perf_session__register_idle_thread(struct perf_session *session) 1556 { 1557 struct thread *thread; 1558 int err = 0; 1559 1560 thread = machine__findnew_thread(&session->machines.host, 0, 0); 1561 if (thread == NULL || thread__set_comm(thread, "swapper", 0)) { 1562 pr_err("problem inserting idle task.\n"); 1563 err = -1; 1564 } 1565 1566 if (thread == NULL || thread__set_namespaces(thread, 0, NULL)) { 1567 pr_err("problem inserting idle task.\n"); 1568 err = -1; 1569 } 1570 1571 /* machine__findnew_thread() got the thread, so put it */ 1572 thread__put(thread); 1573 return err; 1574 } 1575 1576 static void 1577 perf_session__warn_order(const struct perf_session *session) 1578 { 1579 const struct ordered_events *oe = &session->ordered_events; 1580 struct perf_evsel *evsel; 1581 bool should_warn = true; 1582 1583 evlist__for_each_entry(session->evlist, evsel) { 1584 if (evsel->attr.write_backward) 1585 should_warn = false; 1586 } 1587 1588 if (!should_warn) 1589 return; 1590 if (oe->nr_unordered_events != 0) 1591 ui__warning("%u out of order events recorded.\n", oe->nr_unordered_events); 1592 } 1593 1594 static void perf_session__warn_about_errors(const struct perf_session *session) 1595 { 1596 const struct events_stats *stats = &session->evlist->stats; 1597 1598 if (session->tool->lost == perf_event__process_lost && 1599 stats->nr_events[PERF_RECORD_LOST] != 0) { 1600 ui__warning("Processed %d events and lost %d chunks!\n\n" 1601 "Check IO/CPU overload!\n\n", 1602 stats->nr_events[0], 1603 stats->nr_events[PERF_RECORD_LOST]); 1604 } 1605 1606 if (session->tool->lost_samples == perf_event__process_lost_samples) { 1607 double drop_rate; 1608 1609 drop_rate = (double)stats->total_lost_samples / 1610 (double) (stats->nr_events[PERF_RECORD_SAMPLE] + stats->total_lost_samples); 1611 if (drop_rate > 0.05) { 1612 ui__warning("Processed %" PRIu64 " samples and lost %3.2f%%!\n\n", 1613 stats->nr_events[PERF_RECORD_SAMPLE] + stats->total_lost_samples, 1614 drop_rate * 100.0); 1615 } 1616 } 1617 1618 if (session->tool->aux == perf_event__process_aux && 1619 stats->total_aux_lost != 0) { 1620 ui__warning("AUX data lost %" PRIu64 " times out of %u!\n\n", 1621 stats->total_aux_lost, 1622 stats->nr_events[PERF_RECORD_AUX]); 1623 } 1624 1625 if (session->tool->aux == perf_event__process_aux && 1626 stats->total_aux_partial != 0) { 1627 bool vmm_exclusive = false; 1628 1629 (void)sysfs__read_bool("module/kvm_intel/parameters/vmm_exclusive", 1630 &vmm_exclusive); 1631 1632 ui__warning("AUX data had gaps in it %" PRIu64 " times out of %u!\n\n" 1633 "Are you running a KVM guest in the background?%s\n\n", 1634 stats->total_aux_partial, 1635 stats->nr_events[PERF_RECORD_AUX], 1636 vmm_exclusive ? 1637 "\nReloading kvm_intel module with vmm_exclusive=0\n" 1638 "will reduce the gaps to only guest's timeslices." : 1639 ""); 1640 } 1641 1642 if (stats->nr_unknown_events != 0) { 1643 ui__warning("Found %u unknown events!\n\n" 1644 "Is this an older tool processing a perf.data " 1645 "file generated by a more recent tool?\n\n" 1646 "If that is not the case, consider " 1647 "reporting to linux-kernel@vger.kernel.org.\n\n", 1648 stats->nr_unknown_events); 1649 } 1650 1651 if (stats->nr_unknown_id != 0) { 1652 ui__warning("%u samples with id not present in the header\n", 1653 stats->nr_unknown_id); 1654 } 1655 1656 if (stats->nr_invalid_chains != 0) { 1657 ui__warning("Found invalid callchains!\n\n" 1658 "%u out of %u events were discarded for this reason.\n\n" 1659 "Consider reporting to linux-kernel@vger.kernel.org.\n\n", 1660 stats->nr_invalid_chains, 1661 stats->nr_events[PERF_RECORD_SAMPLE]); 1662 } 1663 1664 if (stats->nr_unprocessable_samples != 0) { 1665 ui__warning("%u unprocessable samples recorded.\n" 1666 "Do you have a KVM guest running and not using 'perf kvm'?\n", 1667 stats->nr_unprocessable_samples); 1668 } 1669 1670 perf_session__warn_order(session); 1671 1672 events_stats__auxtrace_error_warn(stats); 1673 1674 if (stats->nr_proc_map_timeout != 0) { 1675 ui__warning("%d map information files for pre-existing threads were\n" 1676 "not processed, if there are samples for addresses they\n" 1677 "will not be resolved, you may find out which are these\n" 1678 "threads by running with -v and redirecting the output\n" 1679 "to a file.\n" 1680 "The time limit to process proc map is too short?\n" 1681 "Increase it by --proc-map-timeout\n", 1682 stats->nr_proc_map_timeout); 1683 } 1684 } 1685 1686 static int perf_session__flush_thread_stack(struct thread *thread, 1687 void *p __maybe_unused) 1688 { 1689 return thread_stack__flush(thread); 1690 } 1691 1692 static int perf_session__flush_thread_stacks(struct perf_session *session) 1693 { 1694 return machines__for_each_thread(&session->machines, 1695 perf_session__flush_thread_stack, 1696 NULL); 1697 } 1698 1699 volatile int session_done; 1700 1701 static int __perf_session__process_pipe_events(struct perf_session *session) 1702 { 1703 struct ordered_events *oe = &session->ordered_events; 1704 struct perf_tool *tool = session->tool; 1705 int fd = perf_data__fd(session->data); 1706 union perf_event *event; 1707 uint32_t size, cur_size = 0; 1708 void *buf = NULL; 1709 s64 skip = 0; 1710 u64 head; 1711 ssize_t err; 1712 void *p; 1713 1714 perf_tool__fill_defaults(tool); 1715 1716 head = 0; 1717 cur_size = sizeof(union perf_event); 1718 1719 buf = malloc(cur_size); 1720 if (!buf) 1721 return -errno; 1722 ordered_events__set_copy_on_queue(oe, true); 1723 more: 1724 event = buf; 1725 err = readn(fd, event, sizeof(struct perf_event_header)); 1726 if (err <= 0) { 1727 if (err == 0) 1728 goto done; 1729 1730 pr_err("failed to read event header\n"); 1731 goto out_err; 1732 } 1733 1734 if (session->header.needs_swap) 1735 perf_event_header__bswap(&event->header); 1736 1737 size = event->header.size; 1738 if (size < sizeof(struct perf_event_header)) { 1739 pr_err("bad event header size\n"); 1740 goto out_err; 1741 } 1742 1743 if (size > cur_size) { 1744 void *new = realloc(buf, size); 1745 if (!new) { 1746 pr_err("failed to allocate memory to read event\n"); 1747 goto out_err; 1748 } 1749 buf = new; 1750 cur_size = size; 1751 event = buf; 1752 } 1753 p = event; 1754 p += sizeof(struct perf_event_header); 1755 1756 if (size - sizeof(struct perf_event_header)) { 1757 err = readn(fd, p, size - sizeof(struct perf_event_header)); 1758 if (err <= 0) { 1759 if (err == 0) { 1760 pr_err("unexpected end of event stream\n"); 1761 goto done; 1762 } 1763 1764 pr_err("failed to read event data\n"); 1765 goto out_err; 1766 } 1767 } 1768 1769 if ((skip = perf_session__process_event(session, event, head)) < 0) { 1770 pr_err("%#" PRIx64 " [%#x]: failed to process type: %d\n", 1771 head, event->header.size, event->header.type); 1772 err = -EINVAL; 1773 goto out_err; 1774 } 1775 1776 head += size; 1777 1778 if (skip > 0) 1779 head += skip; 1780 1781 if (!session_done()) 1782 goto more; 1783 done: 1784 /* do the final flush for ordered samples */ 1785 err = ordered_events__flush(oe, OE_FLUSH__FINAL); 1786 if (err) 1787 goto out_err; 1788 err = auxtrace__flush_events(session, tool); 1789 if (err) 1790 goto out_err; 1791 err = perf_session__flush_thread_stacks(session); 1792 out_err: 1793 free(buf); 1794 if (!tool->no_warn) 1795 perf_session__warn_about_errors(session); 1796 ordered_events__free(&session->ordered_events); 1797 auxtrace__free_events(session); 1798 return err; 1799 } 1800 1801 static union perf_event * 1802 fetch_mmaped_event(struct perf_session *session, 1803 u64 head, size_t mmap_size, char *buf) 1804 { 1805 union perf_event *event; 1806 1807 /* 1808 * Ensure we have enough space remaining to read 1809 * the size of the event in the headers. 1810 */ 1811 if (head + sizeof(event->header) > mmap_size) 1812 return NULL; 1813 1814 event = (union perf_event *)(buf + head); 1815 1816 if (session->header.needs_swap) 1817 perf_event_header__bswap(&event->header); 1818 1819 if (head + event->header.size > mmap_size) { 1820 /* We're not fetching the event so swap back again */ 1821 if (session->header.needs_swap) 1822 perf_event_header__bswap(&event->header); 1823 return NULL; 1824 } 1825 1826 return event; 1827 } 1828 1829 /* 1830 * On 64bit we can mmap the data file in one go. No need for tiny mmap 1831 * slices. On 32bit we use 32MB. 1832 */ 1833 #if BITS_PER_LONG == 64 1834 #define MMAP_SIZE ULLONG_MAX 1835 #define NUM_MMAPS 1 1836 #else 1837 #define MMAP_SIZE (32 * 1024 * 1024ULL) 1838 #define NUM_MMAPS 128 1839 #endif 1840 1841 struct reader { 1842 int fd; 1843 u64 data_size; 1844 u64 data_offset; 1845 }; 1846 1847 static int 1848 reader__process_events(struct reader *rd, struct perf_session *session, 1849 struct ui_progress *prog) 1850 { 1851 u64 data_size = rd->data_size; 1852 u64 head, page_offset, file_offset, file_pos, size; 1853 int err = 0, mmap_prot, mmap_flags, map_idx = 0; 1854 size_t mmap_size; 1855 char *buf, *mmaps[NUM_MMAPS]; 1856 union perf_event *event; 1857 s64 skip; 1858 1859 page_offset = page_size * (rd->data_offset / page_size); 1860 file_offset = page_offset; 1861 head = rd->data_offset - page_offset; 1862 1863 ui_progress__init_size(prog, data_size, "Processing events..."); 1864 1865 data_size += rd->data_offset; 1866 1867 mmap_size = MMAP_SIZE; 1868 if (mmap_size > data_size) { 1869 mmap_size = data_size; 1870 session->one_mmap = true; 1871 } 1872 1873 memset(mmaps, 0, sizeof(mmaps)); 1874 1875 mmap_prot = PROT_READ; 1876 mmap_flags = MAP_SHARED; 1877 1878 if (session->header.needs_swap) { 1879 mmap_prot |= PROT_WRITE; 1880 mmap_flags = MAP_PRIVATE; 1881 } 1882 remap: 1883 buf = mmap(NULL, mmap_size, mmap_prot, mmap_flags, rd->fd, 1884 file_offset); 1885 if (buf == MAP_FAILED) { 1886 pr_err("failed to mmap file\n"); 1887 err = -errno; 1888 goto out; 1889 } 1890 mmaps[map_idx] = buf; 1891 map_idx = (map_idx + 1) & (ARRAY_SIZE(mmaps) - 1); 1892 file_pos = file_offset + head; 1893 if (session->one_mmap) { 1894 session->one_mmap_addr = buf; 1895 session->one_mmap_offset = file_offset; 1896 } 1897 1898 more: 1899 event = fetch_mmaped_event(session, head, mmap_size, buf); 1900 if (!event) { 1901 if (mmaps[map_idx]) { 1902 munmap(mmaps[map_idx], mmap_size); 1903 mmaps[map_idx] = NULL; 1904 } 1905 1906 page_offset = page_size * (head / page_size); 1907 file_offset += page_offset; 1908 head -= page_offset; 1909 goto remap; 1910 } 1911 1912 size = event->header.size; 1913 1914 if (size < sizeof(struct perf_event_header) || 1915 (skip = perf_session__process_event(session, event, file_pos)) < 0) { 1916 pr_err("%#" PRIx64 " [%#x]: failed to process type: %d\n", 1917 file_offset + head, event->header.size, 1918 event->header.type); 1919 err = -EINVAL; 1920 goto out; 1921 } 1922 1923 if (skip) 1924 size += skip; 1925 1926 head += size; 1927 file_pos += size; 1928 1929 ui_progress__update(prog, size); 1930 1931 if (session_done()) 1932 goto out; 1933 1934 if (file_pos < data_size) 1935 goto more; 1936 1937 out: 1938 return err; 1939 } 1940 1941 static int __perf_session__process_events(struct perf_session *session) 1942 { 1943 struct reader rd = { 1944 .fd = perf_data__fd(session->data), 1945 .data_size = session->header.data_size, 1946 .data_offset = session->header.data_offset, 1947 }; 1948 struct ordered_events *oe = &session->ordered_events; 1949 struct perf_tool *tool = session->tool; 1950 struct ui_progress prog; 1951 int err; 1952 1953 perf_tool__fill_defaults(tool); 1954 1955 if (rd.data_size == 0) 1956 return -1; 1957 1958 ui_progress__init_size(&prog, rd.data_size, "Processing events..."); 1959 1960 err = reader__process_events(&rd, session, &prog); 1961 if (err) 1962 goto out_err; 1963 /* do the final flush for ordered samples */ 1964 err = ordered_events__flush(oe, OE_FLUSH__FINAL); 1965 if (err) 1966 goto out_err; 1967 err = auxtrace__flush_events(session, tool); 1968 if (err) 1969 goto out_err; 1970 err = perf_session__flush_thread_stacks(session); 1971 out_err: 1972 ui_progress__finish(); 1973 if (!tool->no_warn) 1974 perf_session__warn_about_errors(session); 1975 /* 1976 * We may switching perf.data output, make ordered_events 1977 * reusable. 1978 */ 1979 ordered_events__reinit(&session->ordered_events); 1980 auxtrace__free_events(session); 1981 session->one_mmap = false; 1982 return err; 1983 } 1984 1985 int perf_session__process_events(struct perf_session *session) 1986 { 1987 if (perf_session__register_idle_thread(session) < 0) 1988 return -ENOMEM; 1989 1990 if (perf_data__is_pipe(session->data)) 1991 return __perf_session__process_pipe_events(session); 1992 1993 return __perf_session__process_events(session); 1994 } 1995 1996 bool perf_session__has_traces(struct perf_session *session, const char *msg) 1997 { 1998 struct perf_evsel *evsel; 1999 2000 evlist__for_each_entry(session->evlist, evsel) { 2001 if (evsel->attr.type == PERF_TYPE_TRACEPOINT) 2002 return true; 2003 } 2004 2005 pr_err("No trace sample to read. Did you call 'perf %s'?\n", msg); 2006 return false; 2007 } 2008 2009 int map__set_kallsyms_ref_reloc_sym(struct map *map, const char *symbol_name, u64 addr) 2010 { 2011 char *bracket; 2012 struct ref_reloc_sym *ref; 2013 struct kmap *kmap; 2014 2015 ref = zalloc(sizeof(struct ref_reloc_sym)); 2016 if (ref == NULL) 2017 return -ENOMEM; 2018 2019 ref->name = strdup(symbol_name); 2020 if (ref->name == NULL) { 2021 free(ref); 2022 return -ENOMEM; 2023 } 2024 2025 bracket = strchr(ref->name, ']'); 2026 if (bracket) 2027 *bracket = '\0'; 2028 2029 ref->addr = addr; 2030 2031 kmap = map__kmap(map); 2032 if (kmap) 2033 kmap->ref_reloc_sym = ref; 2034 2035 return 0; 2036 } 2037 2038 size_t perf_session__fprintf_dsos(struct perf_session *session, FILE *fp) 2039 { 2040 return machines__fprintf_dsos(&session->machines, fp); 2041 } 2042 2043 size_t perf_session__fprintf_dsos_buildid(struct perf_session *session, FILE *fp, 2044 bool (skip)(struct dso *dso, int parm), int parm) 2045 { 2046 return machines__fprintf_dsos_buildid(&session->machines, fp, skip, parm); 2047 } 2048 2049 size_t perf_session__fprintf_nr_events(struct perf_session *session, FILE *fp) 2050 { 2051 size_t ret; 2052 const char *msg = ""; 2053 2054 if (perf_header__has_feat(&session->header, HEADER_AUXTRACE)) 2055 msg = " (excludes AUX area (e.g. instruction trace) decoded / synthesized events)"; 2056 2057 ret = fprintf(fp, "\nAggregated stats:%s\n", msg); 2058 2059 ret += events_stats__fprintf(&session->evlist->stats, fp); 2060 return ret; 2061 } 2062 2063 size_t perf_session__fprintf(struct perf_session *session, FILE *fp) 2064 { 2065 /* 2066 * FIXME: Here we have to actually print all the machines in this 2067 * session, not just the host... 2068 */ 2069 return machine__fprintf(&session->machines.host, fp); 2070 } 2071 2072 struct perf_evsel *perf_session__find_first_evtype(struct perf_session *session, 2073 unsigned int type) 2074 { 2075 struct perf_evsel *pos; 2076 2077 evlist__for_each_entry(session->evlist, pos) { 2078 if (pos->attr.type == type) 2079 return pos; 2080 } 2081 return NULL; 2082 } 2083 2084 int perf_session__cpu_bitmap(struct perf_session *session, 2085 const char *cpu_list, unsigned long *cpu_bitmap) 2086 { 2087 int i, err = -1; 2088 struct cpu_map *map; 2089 2090 for (i = 0; i < PERF_TYPE_MAX; ++i) { 2091 struct perf_evsel *evsel; 2092 2093 evsel = perf_session__find_first_evtype(session, i); 2094 if (!evsel) 2095 continue; 2096 2097 if (!(evsel->attr.sample_type & PERF_SAMPLE_CPU)) { 2098 pr_err("File does not contain CPU events. " 2099 "Remove -C option to proceed.\n"); 2100 return -1; 2101 } 2102 } 2103 2104 map = cpu_map__new(cpu_list); 2105 if (map == NULL) { 2106 pr_err("Invalid cpu_list\n"); 2107 return -1; 2108 } 2109 2110 for (i = 0; i < map->nr; i++) { 2111 int cpu = map->map[i]; 2112 2113 if (cpu >= MAX_NR_CPUS) { 2114 pr_err("Requested CPU %d too large. " 2115 "Consider raising MAX_NR_CPUS\n", cpu); 2116 goto out_delete_map; 2117 } 2118 2119 set_bit(cpu, cpu_bitmap); 2120 } 2121 2122 err = 0; 2123 2124 out_delete_map: 2125 cpu_map__put(map); 2126 return err; 2127 } 2128 2129 void perf_session__fprintf_info(struct perf_session *session, FILE *fp, 2130 bool full) 2131 { 2132 if (session == NULL || fp == NULL) 2133 return; 2134 2135 fprintf(fp, "# ========\n"); 2136 perf_header__fprintf_info(session, fp, full); 2137 fprintf(fp, "# ========\n#\n"); 2138 } 2139 2140 2141 int __perf_session__set_tracepoints_handlers(struct perf_session *session, 2142 const struct perf_evsel_str_handler *assocs, 2143 size_t nr_assocs) 2144 { 2145 struct perf_evsel *evsel; 2146 size_t i; 2147 int err; 2148 2149 for (i = 0; i < nr_assocs; i++) { 2150 /* 2151 * Adding a handler for an event not in the session, 2152 * just ignore it. 2153 */ 2154 evsel = perf_evlist__find_tracepoint_by_name(session->evlist, assocs[i].name); 2155 if (evsel == NULL) 2156 continue; 2157 2158 err = -EEXIST; 2159 if (evsel->handler != NULL) 2160 goto out; 2161 evsel->handler = assocs[i].handler; 2162 } 2163 2164 err = 0; 2165 out: 2166 return err; 2167 } 2168 2169 int perf_event__process_id_index(struct perf_session *session, 2170 union perf_event *event) 2171 { 2172 struct perf_evlist *evlist = session->evlist; 2173 struct id_index_event *ie = &event->id_index; 2174 size_t i, nr, max_nr; 2175 2176 max_nr = (ie->header.size - sizeof(struct id_index_event)) / 2177 sizeof(struct id_index_entry); 2178 nr = ie->nr; 2179 if (nr > max_nr) 2180 return -EINVAL; 2181 2182 if (dump_trace) 2183 fprintf(stdout, " nr: %zu\n", nr); 2184 2185 for (i = 0; i < nr; i++) { 2186 struct id_index_entry *e = &ie->entries[i]; 2187 struct perf_sample_id *sid; 2188 2189 if (dump_trace) { 2190 fprintf(stdout, " ... id: %"PRIu64, e->id); 2191 fprintf(stdout, " idx: %"PRIu64, e->idx); 2192 fprintf(stdout, " cpu: %"PRId64, e->cpu); 2193 fprintf(stdout, " tid: %"PRId64"\n", e->tid); 2194 } 2195 2196 sid = perf_evlist__id2sid(evlist, e->id); 2197 if (!sid) 2198 return -ENOENT; 2199 sid->idx = e->idx; 2200 sid->cpu = e->cpu; 2201 sid->tid = e->tid; 2202 } 2203 return 0; 2204 } 2205 2206 int perf_event__synthesize_id_index(struct perf_tool *tool, 2207 perf_event__handler_t process, 2208 struct perf_evlist *evlist, 2209 struct machine *machine) 2210 { 2211 union perf_event *ev; 2212 struct perf_evsel *evsel; 2213 size_t nr = 0, i = 0, sz, max_nr, n; 2214 int err; 2215 2216 pr_debug2("Synthesizing id index\n"); 2217 2218 max_nr = (UINT16_MAX - sizeof(struct id_index_event)) / 2219 sizeof(struct id_index_entry); 2220 2221 evlist__for_each_entry(evlist, evsel) 2222 nr += evsel->ids; 2223 2224 n = nr > max_nr ? max_nr : nr; 2225 sz = sizeof(struct id_index_event) + n * sizeof(struct id_index_entry); 2226 ev = zalloc(sz); 2227 if (!ev) 2228 return -ENOMEM; 2229 2230 ev->id_index.header.type = PERF_RECORD_ID_INDEX; 2231 ev->id_index.header.size = sz; 2232 ev->id_index.nr = n; 2233 2234 evlist__for_each_entry(evlist, evsel) { 2235 u32 j; 2236 2237 for (j = 0; j < evsel->ids; j++) { 2238 struct id_index_entry *e; 2239 struct perf_sample_id *sid; 2240 2241 if (i >= n) { 2242 err = process(tool, ev, NULL, machine); 2243 if (err) 2244 goto out_err; 2245 nr -= n; 2246 i = 0; 2247 } 2248 2249 e = &ev->id_index.entries[i++]; 2250 2251 e->id = evsel->id[j]; 2252 2253 sid = perf_evlist__id2sid(evlist, e->id); 2254 if (!sid) { 2255 free(ev); 2256 return -ENOENT; 2257 } 2258 2259 e->idx = sid->idx; 2260 e->cpu = sid->cpu; 2261 e->tid = sid->tid; 2262 } 2263 } 2264 2265 sz = sizeof(struct id_index_event) + nr * sizeof(struct id_index_entry); 2266 ev->id_index.header.size = sz; 2267 ev->id_index.nr = nr; 2268 2269 err = process(tool, ev, NULL, machine); 2270 out_err: 2271 free(ev); 2272 2273 return err; 2274 } 2275