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.ip = bswap_64(event->auxtrace_error.ip); 707 } 708 709 static void perf_event__thread_map_swap(union perf_event *event, 710 bool sample_id_all __maybe_unused) 711 { 712 unsigned i; 713 714 event->thread_map.nr = bswap_64(event->thread_map.nr); 715 716 for (i = 0; i < event->thread_map.nr; i++) 717 event->thread_map.entries[i].pid = bswap_64(event->thread_map.entries[i].pid); 718 } 719 720 static void perf_event__cpu_map_swap(union perf_event *event, 721 bool sample_id_all __maybe_unused) 722 { 723 struct cpu_map_data *data = &event->cpu_map.data; 724 struct cpu_map_entries *cpus; 725 struct cpu_map_mask *mask; 726 unsigned i; 727 728 data->type = bswap_64(data->type); 729 730 switch (data->type) { 731 case PERF_CPU_MAP__CPUS: 732 cpus = (struct cpu_map_entries *)data->data; 733 734 cpus->nr = bswap_16(cpus->nr); 735 736 for (i = 0; i < cpus->nr; i++) 737 cpus->cpu[i] = bswap_16(cpus->cpu[i]); 738 break; 739 case PERF_CPU_MAP__MASK: 740 mask = (struct cpu_map_mask *) data->data; 741 742 mask->nr = bswap_16(mask->nr); 743 mask->long_size = bswap_16(mask->long_size); 744 745 switch (mask->long_size) { 746 case 4: mem_bswap_32(&mask->mask, mask->nr); break; 747 case 8: mem_bswap_64(&mask->mask, mask->nr); break; 748 default: 749 pr_err("cpu_map swap: unsupported long size\n"); 750 } 751 default: 752 break; 753 } 754 } 755 756 static void perf_event__stat_config_swap(union perf_event *event, 757 bool sample_id_all __maybe_unused) 758 { 759 u64 size; 760 761 size = event->stat_config.nr * sizeof(event->stat_config.data[0]); 762 size += 1; /* nr item itself */ 763 mem_bswap_64(&event->stat_config.nr, size); 764 } 765 766 static void perf_event__stat_swap(union perf_event *event, 767 bool sample_id_all __maybe_unused) 768 { 769 event->stat.id = bswap_64(event->stat.id); 770 event->stat.thread = bswap_32(event->stat.thread); 771 event->stat.cpu = bswap_32(event->stat.cpu); 772 event->stat.val = bswap_64(event->stat.val); 773 event->stat.ena = bswap_64(event->stat.ena); 774 event->stat.run = bswap_64(event->stat.run); 775 } 776 777 static void perf_event__stat_round_swap(union perf_event *event, 778 bool sample_id_all __maybe_unused) 779 { 780 event->stat_round.type = bswap_64(event->stat_round.type); 781 event->stat_round.time = bswap_64(event->stat_round.time); 782 } 783 784 typedef void (*perf_event__swap_op)(union perf_event *event, 785 bool sample_id_all); 786 787 static perf_event__swap_op perf_event__swap_ops[] = { 788 [PERF_RECORD_MMAP] = perf_event__mmap_swap, 789 [PERF_RECORD_MMAP2] = perf_event__mmap2_swap, 790 [PERF_RECORD_COMM] = perf_event__comm_swap, 791 [PERF_RECORD_FORK] = perf_event__task_swap, 792 [PERF_RECORD_EXIT] = perf_event__task_swap, 793 [PERF_RECORD_LOST] = perf_event__all64_swap, 794 [PERF_RECORD_READ] = perf_event__read_swap, 795 [PERF_RECORD_THROTTLE] = perf_event__throttle_swap, 796 [PERF_RECORD_UNTHROTTLE] = perf_event__throttle_swap, 797 [PERF_RECORD_SAMPLE] = perf_event__all64_swap, 798 [PERF_RECORD_AUX] = perf_event__aux_swap, 799 [PERF_RECORD_ITRACE_START] = perf_event__itrace_start_swap, 800 [PERF_RECORD_LOST_SAMPLES] = perf_event__all64_swap, 801 [PERF_RECORD_SWITCH] = perf_event__switch_swap, 802 [PERF_RECORD_SWITCH_CPU_WIDE] = perf_event__switch_swap, 803 [PERF_RECORD_HEADER_ATTR] = perf_event__hdr_attr_swap, 804 [PERF_RECORD_HEADER_EVENT_TYPE] = perf_event__event_type_swap, 805 [PERF_RECORD_HEADER_TRACING_DATA] = perf_event__tracing_data_swap, 806 [PERF_RECORD_HEADER_BUILD_ID] = NULL, 807 [PERF_RECORD_ID_INDEX] = perf_event__all64_swap, 808 [PERF_RECORD_AUXTRACE_INFO] = perf_event__auxtrace_info_swap, 809 [PERF_RECORD_AUXTRACE] = perf_event__auxtrace_swap, 810 [PERF_RECORD_AUXTRACE_ERROR] = perf_event__auxtrace_error_swap, 811 [PERF_RECORD_THREAD_MAP] = perf_event__thread_map_swap, 812 [PERF_RECORD_CPU_MAP] = perf_event__cpu_map_swap, 813 [PERF_RECORD_STAT_CONFIG] = perf_event__stat_config_swap, 814 [PERF_RECORD_STAT] = perf_event__stat_swap, 815 [PERF_RECORD_STAT_ROUND] = perf_event__stat_round_swap, 816 [PERF_RECORD_EVENT_UPDATE] = perf_event__event_update_swap, 817 [PERF_RECORD_TIME_CONV] = perf_event__all64_swap, 818 [PERF_RECORD_HEADER_MAX] = NULL, 819 }; 820 821 /* 822 * When perf record finishes a pass on every buffers, it records this pseudo 823 * event. 824 * We record the max timestamp t found in the pass n. 825 * Assuming these timestamps are monotonic across cpus, we know that if 826 * a buffer still has events with timestamps below t, they will be all 827 * available and then read in the pass n + 1. 828 * Hence when we start to read the pass n + 2, we can safely flush every 829 * events with timestamps below t. 830 * 831 * ============ PASS n ================= 832 * CPU 0 | CPU 1 833 * | 834 * cnt1 timestamps | cnt2 timestamps 835 * 1 | 2 836 * 2 | 3 837 * - | 4 <--- max recorded 838 * 839 * ============ PASS n + 1 ============== 840 * CPU 0 | CPU 1 841 * | 842 * cnt1 timestamps | cnt2 timestamps 843 * 3 | 5 844 * 4 | 6 845 * 5 | 7 <---- max recorded 846 * 847 * Flush every events below timestamp 4 848 * 849 * ============ PASS n + 2 ============== 850 * CPU 0 | CPU 1 851 * | 852 * cnt1 timestamps | cnt2 timestamps 853 * 6 | 8 854 * 7 | 9 855 * - | 10 856 * 857 * Flush every events below timestamp 7 858 * etc... 859 */ 860 static int process_finished_round(struct perf_tool *tool __maybe_unused, 861 union perf_event *event __maybe_unused, 862 struct ordered_events *oe) 863 { 864 if (dump_trace) 865 fprintf(stdout, "\n"); 866 return ordered_events__flush(oe, OE_FLUSH__ROUND); 867 } 868 869 int perf_session__queue_event(struct perf_session *s, union perf_event *event, 870 u64 timestamp, u64 file_offset) 871 { 872 return ordered_events__queue(&s->ordered_events, event, timestamp, file_offset); 873 } 874 875 static void callchain__lbr_callstack_printf(struct perf_sample *sample) 876 { 877 struct ip_callchain *callchain = sample->callchain; 878 struct branch_stack *lbr_stack = sample->branch_stack; 879 u64 kernel_callchain_nr = callchain->nr; 880 unsigned int i; 881 882 for (i = 0; i < kernel_callchain_nr; i++) { 883 if (callchain->ips[i] == PERF_CONTEXT_USER) 884 break; 885 } 886 887 if ((i != kernel_callchain_nr) && lbr_stack->nr) { 888 u64 total_nr; 889 /* 890 * LBR callstack can only get user call chain, 891 * i is kernel call chain number, 892 * 1 is PERF_CONTEXT_USER. 893 * 894 * The user call chain is stored in LBR registers. 895 * LBR are pair registers. The caller is stored 896 * in "from" register, while the callee is stored 897 * in "to" register. 898 * For example, there is a call stack 899 * "A"->"B"->"C"->"D". 900 * The LBR registers will recorde like 901 * "C"->"D", "B"->"C", "A"->"B". 902 * So only the first "to" register and all "from" 903 * registers are needed to construct the whole stack. 904 */ 905 total_nr = i + 1 + lbr_stack->nr + 1; 906 kernel_callchain_nr = i + 1; 907 908 printf("... LBR call chain: nr:%" PRIu64 "\n", total_nr); 909 910 for (i = 0; i < kernel_callchain_nr; i++) 911 printf("..... %2d: %016" PRIx64 "\n", 912 i, callchain->ips[i]); 913 914 printf("..... %2d: %016" PRIx64 "\n", 915 (int)(kernel_callchain_nr), lbr_stack->entries[0].to); 916 for (i = 0; i < lbr_stack->nr; i++) 917 printf("..... %2d: %016" PRIx64 "\n", 918 (int)(i + kernel_callchain_nr + 1), lbr_stack->entries[i].from); 919 } 920 } 921 922 static void callchain__printf(struct perf_evsel *evsel, 923 struct perf_sample *sample) 924 { 925 unsigned int i; 926 struct ip_callchain *callchain = sample->callchain; 927 928 if (perf_evsel__has_branch_callstack(evsel)) 929 callchain__lbr_callstack_printf(sample); 930 931 printf("... FP chain: nr:%" PRIu64 "\n", callchain->nr); 932 933 for (i = 0; i < callchain->nr; i++) 934 printf("..... %2d: %016" PRIx64 "\n", 935 i, callchain->ips[i]); 936 } 937 938 static void branch_stack__printf(struct perf_sample *sample) 939 { 940 uint64_t i; 941 942 printf("... branch stack: nr:%" PRIu64 "\n", sample->branch_stack->nr); 943 944 for (i = 0; i < sample->branch_stack->nr; i++) { 945 struct branch_entry *e = &sample->branch_stack->entries[i]; 946 947 printf("..... %2"PRIu64": %016" PRIx64 " -> %016" PRIx64 " %hu cycles %s%s%s%s %x\n", 948 i, e->from, e->to, 949 (unsigned short)e->flags.cycles, 950 e->flags.mispred ? "M" : " ", 951 e->flags.predicted ? "P" : " ", 952 e->flags.abort ? "A" : " ", 953 e->flags.in_tx ? "T" : " ", 954 (unsigned)e->flags.reserved); 955 } 956 } 957 958 static void regs_dump__printf(u64 mask, u64 *regs) 959 { 960 unsigned rid, i = 0; 961 962 for_each_set_bit(rid, (unsigned long *) &mask, sizeof(mask) * 8) { 963 u64 val = regs[i++]; 964 965 printf(".... %-5s 0x%" PRIx64 "\n", 966 perf_reg_name(rid), val); 967 } 968 } 969 970 static const char *regs_abi[] = { 971 [PERF_SAMPLE_REGS_ABI_NONE] = "none", 972 [PERF_SAMPLE_REGS_ABI_32] = "32-bit", 973 [PERF_SAMPLE_REGS_ABI_64] = "64-bit", 974 }; 975 976 static inline const char *regs_dump_abi(struct regs_dump *d) 977 { 978 if (d->abi > PERF_SAMPLE_REGS_ABI_64) 979 return "unknown"; 980 981 return regs_abi[d->abi]; 982 } 983 984 static void regs__printf(const char *type, struct regs_dump *regs) 985 { 986 u64 mask = regs->mask; 987 988 printf("... %s regs: mask 0x%" PRIx64 " ABI %s\n", 989 type, 990 mask, 991 regs_dump_abi(regs)); 992 993 regs_dump__printf(mask, regs->regs); 994 } 995 996 static void regs_user__printf(struct perf_sample *sample) 997 { 998 struct regs_dump *user_regs = &sample->user_regs; 999 1000 if (user_regs->regs) 1001 regs__printf("user", user_regs); 1002 } 1003 1004 static void regs_intr__printf(struct perf_sample *sample) 1005 { 1006 struct regs_dump *intr_regs = &sample->intr_regs; 1007 1008 if (intr_regs->regs) 1009 regs__printf("intr", intr_regs); 1010 } 1011 1012 static void stack_user__printf(struct stack_dump *dump) 1013 { 1014 printf("... ustack: size %" PRIu64 ", offset 0x%x\n", 1015 dump->size, dump->offset); 1016 } 1017 1018 static void perf_evlist__print_tstamp(struct perf_evlist *evlist, 1019 union perf_event *event, 1020 struct perf_sample *sample) 1021 { 1022 u64 sample_type = __perf_evlist__combined_sample_type(evlist); 1023 1024 if (event->header.type != PERF_RECORD_SAMPLE && 1025 !perf_evlist__sample_id_all(evlist)) { 1026 fputs("-1 -1 ", stdout); 1027 return; 1028 } 1029 1030 if ((sample_type & PERF_SAMPLE_CPU)) 1031 printf("%u ", sample->cpu); 1032 1033 if (sample_type & PERF_SAMPLE_TIME) 1034 printf("%" PRIu64 " ", sample->time); 1035 } 1036 1037 static void sample_read__printf(struct perf_sample *sample, u64 read_format) 1038 { 1039 printf("... sample_read:\n"); 1040 1041 if (read_format & PERF_FORMAT_TOTAL_TIME_ENABLED) 1042 printf("...... time enabled %016" PRIx64 "\n", 1043 sample->read.time_enabled); 1044 1045 if (read_format & PERF_FORMAT_TOTAL_TIME_RUNNING) 1046 printf("...... time running %016" PRIx64 "\n", 1047 sample->read.time_running); 1048 1049 if (read_format & PERF_FORMAT_GROUP) { 1050 u64 i; 1051 1052 printf(".... group nr %" PRIu64 "\n", sample->read.group.nr); 1053 1054 for (i = 0; i < sample->read.group.nr; i++) { 1055 struct sample_read_value *value; 1056 1057 value = &sample->read.group.values[i]; 1058 printf("..... id %016" PRIx64 1059 ", value %016" PRIx64 "\n", 1060 value->id, value->value); 1061 } 1062 } else 1063 printf("..... id %016" PRIx64 ", value %016" PRIx64 "\n", 1064 sample->read.one.id, sample->read.one.value); 1065 } 1066 1067 static void dump_event(struct perf_evlist *evlist, union perf_event *event, 1068 u64 file_offset, struct perf_sample *sample) 1069 { 1070 if (!dump_trace) 1071 return; 1072 1073 printf("\n%#" PRIx64 " [%#x]: event: %d\n", 1074 file_offset, event->header.size, event->header.type); 1075 1076 trace_event(event); 1077 if (event->header.type == PERF_RECORD_SAMPLE && evlist->trace_event_sample_raw) 1078 evlist->trace_event_sample_raw(evlist, event, sample); 1079 1080 if (sample) 1081 perf_evlist__print_tstamp(evlist, event, sample); 1082 1083 printf("%#" PRIx64 " [%#x]: PERF_RECORD_%s", file_offset, 1084 event->header.size, perf_event__name(event->header.type)); 1085 } 1086 1087 static void dump_sample(struct perf_evsel *evsel, union perf_event *event, 1088 struct perf_sample *sample) 1089 { 1090 u64 sample_type; 1091 1092 if (!dump_trace) 1093 return; 1094 1095 printf("(IP, 0x%x): %d/%d: %#" PRIx64 " period: %" PRIu64 " addr: %#" PRIx64 "\n", 1096 event->header.misc, sample->pid, sample->tid, sample->ip, 1097 sample->period, sample->addr); 1098 1099 sample_type = evsel->attr.sample_type; 1100 1101 if (evsel__has_callchain(evsel)) 1102 callchain__printf(evsel, sample); 1103 1104 if ((sample_type & PERF_SAMPLE_BRANCH_STACK) && !perf_evsel__has_branch_callstack(evsel)) 1105 branch_stack__printf(sample); 1106 1107 if (sample_type & PERF_SAMPLE_REGS_USER) 1108 regs_user__printf(sample); 1109 1110 if (sample_type & PERF_SAMPLE_REGS_INTR) 1111 regs_intr__printf(sample); 1112 1113 if (sample_type & PERF_SAMPLE_STACK_USER) 1114 stack_user__printf(&sample->user_stack); 1115 1116 if (sample_type & PERF_SAMPLE_WEIGHT) 1117 printf("... weight: %" PRIu64 "\n", sample->weight); 1118 1119 if (sample_type & PERF_SAMPLE_DATA_SRC) 1120 printf(" . data_src: 0x%"PRIx64"\n", sample->data_src); 1121 1122 if (sample_type & PERF_SAMPLE_PHYS_ADDR) 1123 printf(" .. phys_addr: 0x%"PRIx64"\n", sample->phys_addr); 1124 1125 if (sample_type & PERF_SAMPLE_TRANSACTION) 1126 printf("... transaction: %" PRIx64 "\n", sample->transaction); 1127 1128 if (sample_type & PERF_SAMPLE_READ) 1129 sample_read__printf(sample, evsel->attr.read_format); 1130 } 1131 1132 static void dump_read(struct perf_evsel *evsel, union perf_event *event) 1133 { 1134 struct read_event *read_event = &event->read; 1135 u64 read_format; 1136 1137 if (!dump_trace) 1138 return; 1139 1140 printf(": %d %d %s %" PRIu64 "\n", event->read.pid, event->read.tid, 1141 evsel ? perf_evsel__name(evsel) : "FAIL", 1142 event->read.value); 1143 1144 read_format = evsel->attr.read_format; 1145 1146 if (read_format & PERF_FORMAT_TOTAL_TIME_ENABLED) 1147 printf("... time enabled : %" PRIu64 "\n", read_event->time_enabled); 1148 1149 if (read_format & PERF_FORMAT_TOTAL_TIME_RUNNING) 1150 printf("... time running : %" PRIu64 "\n", read_event->time_running); 1151 1152 if (read_format & PERF_FORMAT_ID) 1153 printf("... id : %" PRIu64 "\n", read_event->id); 1154 } 1155 1156 static struct machine *machines__find_for_cpumode(struct machines *machines, 1157 union perf_event *event, 1158 struct perf_sample *sample) 1159 { 1160 struct machine *machine; 1161 1162 if (perf_guest && 1163 ((sample->cpumode == PERF_RECORD_MISC_GUEST_KERNEL) || 1164 (sample->cpumode == PERF_RECORD_MISC_GUEST_USER))) { 1165 u32 pid; 1166 1167 if (event->header.type == PERF_RECORD_MMAP 1168 || event->header.type == PERF_RECORD_MMAP2) 1169 pid = event->mmap.pid; 1170 else 1171 pid = sample->pid; 1172 1173 machine = machines__find(machines, pid); 1174 if (!machine) 1175 machine = machines__findnew(machines, DEFAULT_GUEST_KERNEL_ID); 1176 return machine; 1177 } 1178 1179 return &machines->host; 1180 } 1181 1182 static int deliver_sample_value(struct perf_evlist *evlist, 1183 struct perf_tool *tool, 1184 union perf_event *event, 1185 struct perf_sample *sample, 1186 struct sample_read_value *v, 1187 struct machine *machine) 1188 { 1189 struct perf_sample_id *sid = perf_evlist__id2sid(evlist, v->id); 1190 1191 if (sid) { 1192 sample->id = v->id; 1193 sample->period = v->value - sid->period; 1194 sid->period = v->value; 1195 } 1196 1197 if (!sid || sid->evsel == NULL) { 1198 ++evlist->stats.nr_unknown_id; 1199 return 0; 1200 } 1201 1202 return tool->sample(tool, event, sample, sid->evsel, machine); 1203 } 1204 1205 static int deliver_sample_group(struct perf_evlist *evlist, 1206 struct perf_tool *tool, 1207 union perf_event *event, 1208 struct perf_sample *sample, 1209 struct machine *machine) 1210 { 1211 int ret = -EINVAL; 1212 u64 i; 1213 1214 for (i = 0; i < sample->read.group.nr; i++) { 1215 ret = deliver_sample_value(evlist, tool, event, sample, 1216 &sample->read.group.values[i], 1217 machine); 1218 if (ret) 1219 break; 1220 } 1221 1222 return ret; 1223 } 1224 1225 static int 1226 perf_evlist__deliver_sample(struct perf_evlist *evlist, 1227 struct perf_tool *tool, 1228 union perf_event *event, 1229 struct perf_sample *sample, 1230 struct perf_evsel *evsel, 1231 struct machine *machine) 1232 { 1233 /* We know evsel != NULL. */ 1234 u64 sample_type = evsel->attr.sample_type; 1235 u64 read_format = evsel->attr.read_format; 1236 1237 /* Standard sample delivery. */ 1238 if (!(sample_type & PERF_SAMPLE_READ)) 1239 return tool->sample(tool, event, sample, evsel, machine); 1240 1241 /* For PERF_SAMPLE_READ we have either single or group mode. */ 1242 if (read_format & PERF_FORMAT_GROUP) 1243 return deliver_sample_group(evlist, tool, event, sample, 1244 machine); 1245 else 1246 return deliver_sample_value(evlist, tool, event, sample, 1247 &sample->read.one, machine); 1248 } 1249 1250 static int machines__deliver_event(struct machines *machines, 1251 struct perf_evlist *evlist, 1252 union perf_event *event, 1253 struct perf_sample *sample, 1254 struct perf_tool *tool, u64 file_offset) 1255 { 1256 struct perf_evsel *evsel; 1257 struct machine *machine; 1258 1259 dump_event(evlist, event, file_offset, sample); 1260 1261 evsel = perf_evlist__id2evsel(evlist, sample->id); 1262 1263 machine = machines__find_for_cpumode(machines, event, sample); 1264 1265 switch (event->header.type) { 1266 case PERF_RECORD_SAMPLE: 1267 if (evsel == NULL) { 1268 ++evlist->stats.nr_unknown_id; 1269 return 0; 1270 } 1271 dump_sample(evsel, event, sample); 1272 if (machine == NULL) { 1273 ++evlist->stats.nr_unprocessable_samples; 1274 return 0; 1275 } 1276 return perf_evlist__deliver_sample(evlist, tool, event, sample, evsel, machine); 1277 case PERF_RECORD_MMAP: 1278 return tool->mmap(tool, event, sample, machine); 1279 case PERF_RECORD_MMAP2: 1280 if (event->header.misc & PERF_RECORD_MISC_PROC_MAP_PARSE_TIMEOUT) 1281 ++evlist->stats.nr_proc_map_timeout; 1282 return tool->mmap2(tool, event, sample, machine); 1283 case PERF_RECORD_COMM: 1284 return tool->comm(tool, event, sample, machine); 1285 case PERF_RECORD_NAMESPACES: 1286 return tool->namespaces(tool, event, sample, machine); 1287 case PERF_RECORD_FORK: 1288 return tool->fork(tool, event, sample, machine); 1289 case PERF_RECORD_EXIT: 1290 return tool->exit(tool, event, sample, machine); 1291 case PERF_RECORD_LOST: 1292 if (tool->lost == perf_event__process_lost) 1293 evlist->stats.total_lost += event->lost.lost; 1294 return tool->lost(tool, event, sample, machine); 1295 case PERF_RECORD_LOST_SAMPLES: 1296 if (tool->lost_samples == perf_event__process_lost_samples) 1297 evlist->stats.total_lost_samples += event->lost_samples.lost; 1298 return tool->lost_samples(tool, event, sample, machine); 1299 case PERF_RECORD_READ: 1300 dump_read(evsel, event); 1301 return tool->read(tool, event, sample, evsel, machine); 1302 case PERF_RECORD_THROTTLE: 1303 return tool->throttle(tool, event, sample, machine); 1304 case PERF_RECORD_UNTHROTTLE: 1305 return tool->unthrottle(tool, event, sample, machine); 1306 case PERF_RECORD_AUX: 1307 if (tool->aux == perf_event__process_aux) { 1308 if (event->aux.flags & PERF_AUX_FLAG_TRUNCATED) 1309 evlist->stats.total_aux_lost += 1; 1310 if (event->aux.flags & PERF_AUX_FLAG_PARTIAL) 1311 evlist->stats.total_aux_partial += 1; 1312 } 1313 return tool->aux(tool, event, sample, machine); 1314 case PERF_RECORD_ITRACE_START: 1315 return tool->itrace_start(tool, event, sample, machine); 1316 case PERF_RECORD_SWITCH: 1317 case PERF_RECORD_SWITCH_CPU_WIDE: 1318 return tool->context_switch(tool, event, sample, machine); 1319 case PERF_RECORD_KSYMBOL: 1320 return tool->ksymbol(tool, event, sample, machine); 1321 case PERF_RECORD_BPF_EVENT: 1322 return tool->bpf_event(tool, event, sample, machine); 1323 default: 1324 ++evlist->stats.nr_unknown_events; 1325 return -1; 1326 } 1327 } 1328 1329 static int perf_session__deliver_event(struct perf_session *session, 1330 union perf_event *event, 1331 struct perf_tool *tool, 1332 u64 file_offset) 1333 { 1334 struct perf_sample sample; 1335 int ret; 1336 1337 ret = perf_evlist__parse_sample(session->evlist, event, &sample); 1338 if (ret) { 1339 pr_err("Can't parse sample, err = %d\n", ret); 1340 return ret; 1341 } 1342 1343 ret = auxtrace__process_event(session, event, &sample, tool); 1344 if (ret < 0) 1345 return ret; 1346 if (ret > 0) 1347 return 0; 1348 1349 return machines__deliver_event(&session->machines, session->evlist, 1350 event, &sample, tool, file_offset); 1351 } 1352 1353 static s64 perf_session__process_user_event(struct perf_session *session, 1354 union perf_event *event, 1355 u64 file_offset) 1356 { 1357 struct ordered_events *oe = &session->ordered_events; 1358 struct perf_tool *tool = session->tool; 1359 struct perf_sample sample = { .time = 0, }; 1360 int fd = perf_data__fd(session->data); 1361 int err; 1362 1363 dump_event(session->evlist, event, file_offset, &sample); 1364 1365 /* These events are processed right away */ 1366 switch (event->header.type) { 1367 case PERF_RECORD_HEADER_ATTR: 1368 err = tool->attr(tool, event, &session->evlist); 1369 if (err == 0) { 1370 perf_session__set_id_hdr_size(session); 1371 perf_session__set_comm_exec(session); 1372 } 1373 return err; 1374 case PERF_RECORD_EVENT_UPDATE: 1375 return tool->event_update(tool, event, &session->evlist); 1376 case PERF_RECORD_HEADER_EVENT_TYPE: 1377 /* 1378 * Depreceated, but we need to handle it for sake 1379 * of old data files create in pipe mode. 1380 */ 1381 return 0; 1382 case PERF_RECORD_HEADER_TRACING_DATA: 1383 /* setup for reading amidst mmap */ 1384 lseek(fd, file_offset, SEEK_SET); 1385 return tool->tracing_data(session, event); 1386 case PERF_RECORD_HEADER_BUILD_ID: 1387 return tool->build_id(session, event); 1388 case PERF_RECORD_FINISHED_ROUND: 1389 return tool->finished_round(tool, event, oe); 1390 case PERF_RECORD_ID_INDEX: 1391 return tool->id_index(session, event); 1392 case PERF_RECORD_AUXTRACE_INFO: 1393 return tool->auxtrace_info(session, event); 1394 case PERF_RECORD_AUXTRACE: 1395 /* setup for reading amidst mmap */ 1396 lseek(fd, file_offset + event->header.size, SEEK_SET); 1397 return tool->auxtrace(session, event); 1398 case PERF_RECORD_AUXTRACE_ERROR: 1399 perf_session__auxtrace_error_inc(session, event); 1400 return tool->auxtrace_error(session, event); 1401 case PERF_RECORD_THREAD_MAP: 1402 return tool->thread_map(session, event); 1403 case PERF_RECORD_CPU_MAP: 1404 return tool->cpu_map(session, event); 1405 case PERF_RECORD_STAT_CONFIG: 1406 return tool->stat_config(session, event); 1407 case PERF_RECORD_STAT: 1408 return tool->stat(session, event); 1409 case PERF_RECORD_STAT_ROUND: 1410 return tool->stat_round(session, event); 1411 case PERF_RECORD_TIME_CONV: 1412 session->time_conv = event->time_conv; 1413 return tool->time_conv(session, event); 1414 case PERF_RECORD_HEADER_FEATURE: 1415 return tool->feature(session, event); 1416 default: 1417 return -EINVAL; 1418 } 1419 } 1420 1421 int perf_session__deliver_synth_event(struct perf_session *session, 1422 union perf_event *event, 1423 struct perf_sample *sample) 1424 { 1425 struct perf_evlist *evlist = session->evlist; 1426 struct perf_tool *tool = session->tool; 1427 1428 events_stats__inc(&evlist->stats, event->header.type); 1429 1430 if (event->header.type >= PERF_RECORD_USER_TYPE_START) 1431 return perf_session__process_user_event(session, event, 0); 1432 1433 return machines__deliver_event(&session->machines, evlist, event, sample, tool, 0); 1434 } 1435 1436 static void event_swap(union perf_event *event, bool sample_id_all) 1437 { 1438 perf_event__swap_op swap; 1439 1440 swap = perf_event__swap_ops[event->header.type]; 1441 if (swap) 1442 swap(event, sample_id_all); 1443 } 1444 1445 int perf_session__peek_event(struct perf_session *session, off_t file_offset, 1446 void *buf, size_t buf_sz, 1447 union perf_event **event_ptr, 1448 struct perf_sample *sample) 1449 { 1450 union perf_event *event; 1451 size_t hdr_sz, rest; 1452 int fd; 1453 1454 if (session->one_mmap && !session->header.needs_swap) { 1455 event = file_offset - session->one_mmap_offset + 1456 session->one_mmap_addr; 1457 goto out_parse_sample; 1458 } 1459 1460 if (perf_data__is_pipe(session->data)) 1461 return -1; 1462 1463 fd = perf_data__fd(session->data); 1464 hdr_sz = sizeof(struct perf_event_header); 1465 1466 if (buf_sz < hdr_sz) 1467 return -1; 1468 1469 if (lseek(fd, file_offset, SEEK_SET) == (off_t)-1 || 1470 readn(fd, buf, hdr_sz) != (ssize_t)hdr_sz) 1471 return -1; 1472 1473 event = (union perf_event *)buf; 1474 1475 if (session->header.needs_swap) 1476 perf_event_header__bswap(&event->header); 1477 1478 if (event->header.size < hdr_sz || event->header.size > buf_sz) 1479 return -1; 1480 1481 rest = event->header.size - hdr_sz; 1482 1483 if (readn(fd, buf, rest) != (ssize_t)rest) 1484 return -1; 1485 1486 if (session->header.needs_swap) 1487 event_swap(event, perf_evlist__sample_id_all(session->evlist)); 1488 1489 out_parse_sample: 1490 1491 if (sample && event->header.type < PERF_RECORD_USER_TYPE_START && 1492 perf_evlist__parse_sample(session->evlist, event, sample)) 1493 return -1; 1494 1495 *event_ptr = event; 1496 1497 return 0; 1498 } 1499 1500 static s64 perf_session__process_event(struct perf_session *session, 1501 union perf_event *event, u64 file_offset) 1502 { 1503 struct perf_evlist *evlist = session->evlist; 1504 struct perf_tool *tool = session->tool; 1505 int ret; 1506 1507 if (session->header.needs_swap) 1508 event_swap(event, perf_evlist__sample_id_all(evlist)); 1509 1510 if (event->header.type >= PERF_RECORD_HEADER_MAX) 1511 return -EINVAL; 1512 1513 events_stats__inc(&evlist->stats, event->header.type); 1514 1515 if (event->header.type >= PERF_RECORD_USER_TYPE_START) 1516 return perf_session__process_user_event(session, event, file_offset); 1517 1518 if (tool->ordered_events) { 1519 u64 timestamp = -1ULL; 1520 1521 ret = perf_evlist__parse_sample_timestamp(evlist, event, ×tamp); 1522 if (ret && ret != -1) 1523 return ret; 1524 1525 ret = perf_session__queue_event(session, event, timestamp, file_offset); 1526 if (ret != -ETIME) 1527 return ret; 1528 } 1529 1530 return perf_session__deliver_event(session, event, tool, file_offset); 1531 } 1532 1533 void perf_event_header__bswap(struct perf_event_header *hdr) 1534 { 1535 hdr->type = bswap_32(hdr->type); 1536 hdr->misc = bswap_16(hdr->misc); 1537 hdr->size = bswap_16(hdr->size); 1538 } 1539 1540 struct thread *perf_session__findnew(struct perf_session *session, pid_t pid) 1541 { 1542 return machine__findnew_thread(&session->machines.host, -1, pid); 1543 } 1544 1545 /* 1546 * Threads are identified by pid and tid, and the idle task has pid == tid == 0. 1547 * So here a single thread is created for that, but actually there is a separate 1548 * idle task per cpu, so there should be one 'struct thread' per cpu, but there 1549 * is only 1. That causes problems for some tools, requiring workarounds. For 1550 * example get_idle_thread() in builtin-sched.c, or thread_stack__per_cpu(). 1551 */ 1552 int perf_session__register_idle_thread(struct perf_session *session) 1553 { 1554 struct thread *thread; 1555 int err = 0; 1556 1557 thread = machine__findnew_thread(&session->machines.host, 0, 0); 1558 if (thread == NULL || thread__set_comm(thread, "swapper", 0)) { 1559 pr_err("problem inserting idle task.\n"); 1560 err = -1; 1561 } 1562 1563 if (thread == NULL || thread__set_namespaces(thread, 0, NULL)) { 1564 pr_err("problem inserting idle task.\n"); 1565 err = -1; 1566 } 1567 1568 /* machine__findnew_thread() got the thread, so put it */ 1569 thread__put(thread); 1570 return err; 1571 } 1572 1573 static void 1574 perf_session__warn_order(const struct perf_session *session) 1575 { 1576 const struct ordered_events *oe = &session->ordered_events; 1577 struct perf_evsel *evsel; 1578 bool should_warn = true; 1579 1580 evlist__for_each_entry(session->evlist, evsel) { 1581 if (evsel->attr.write_backward) 1582 should_warn = false; 1583 } 1584 1585 if (!should_warn) 1586 return; 1587 if (oe->nr_unordered_events != 0) 1588 ui__warning("%u out of order events recorded.\n", oe->nr_unordered_events); 1589 } 1590 1591 static void perf_session__warn_about_errors(const struct perf_session *session) 1592 { 1593 const struct events_stats *stats = &session->evlist->stats; 1594 1595 if (session->tool->lost == perf_event__process_lost && 1596 stats->nr_events[PERF_RECORD_LOST] != 0) { 1597 ui__warning("Processed %d events and lost %d chunks!\n\n" 1598 "Check IO/CPU overload!\n\n", 1599 stats->nr_events[0], 1600 stats->nr_events[PERF_RECORD_LOST]); 1601 } 1602 1603 if (session->tool->lost_samples == perf_event__process_lost_samples) { 1604 double drop_rate; 1605 1606 drop_rate = (double)stats->total_lost_samples / 1607 (double) (stats->nr_events[PERF_RECORD_SAMPLE] + stats->total_lost_samples); 1608 if (drop_rate > 0.05) { 1609 ui__warning("Processed %" PRIu64 " samples and lost %3.2f%%!\n\n", 1610 stats->nr_events[PERF_RECORD_SAMPLE] + stats->total_lost_samples, 1611 drop_rate * 100.0); 1612 } 1613 } 1614 1615 if (session->tool->aux == perf_event__process_aux && 1616 stats->total_aux_lost != 0) { 1617 ui__warning("AUX data lost %" PRIu64 " times out of %u!\n\n", 1618 stats->total_aux_lost, 1619 stats->nr_events[PERF_RECORD_AUX]); 1620 } 1621 1622 if (session->tool->aux == perf_event__process_aux && 1623 stats->total_aux_partial != 0) { 1624 bool vmm_exclusive = false; 1625 1626 (void)sysfs__read_bool("module/kvm_intel/parameters/vmm_exclusive", 1627 &vmm_exclusive); 1628 1629 ui__warning("AUX data had gaps in it %" PRIu64 " times out of %u!\n\n" 1630 "Are you running a KVM guest in the background?%s\n\n", 1631 stats->total_aux_partial, 1632 stats->nr_events[PERF_RECORD_AUX], 1633 vmm_exclusive ? 1634 "\nReloading kvm_intel module with vmm_exclusive=0\n" 1635 "will reduce the gaps to only guest's timeslices." : 1636 ""); 1637 } 1638 1639 if (stats->nr_unknown_events != 0) { 1640 ui__warning("Found %u unknown events!\n\n" 1641 "Is this an older tool processing a perf.data " 1642 "file generated by a more recent tool?\n\n" 1643 "If that is not the case, consider " 1644 "reporting to linux-kernel@vger.kernel.org.\n\n", 1645 stats->nr_unknown_events); 1646 } 1647 1648 if (stats->nr_unknown_id != 0) { 1649 ui__warning("%u samples with id not present in the header\n", 1650 stats->nr_unknown_id); 1651 } 1652 1653 if (stats->nr_invalid_chains != 0) { 1654 ui__warning("Found invalid callchains!\n\n" 1655 "%u out of %u events were discarded for this reason.\n\n" 1656 "Consider reporting to linux-kernel@vger.kernel.org.\n\n", 1657 stats->nr_invalid_chains, 1658 stats->nr_events[PERF_RECORD_SAMPLE]); 1659 } 1660 1661 if (stats->nr_unprocessable_samples != 0) { 1662 ui__warning("%u unprocessable samples recorded.\n" 1663 "Do you have a KVM guest running and not using 'perf kvm'?\n", 1664 stats->nr_unprocessable_samples); 1665 } 1666 1667 perf_session__warn_order(session); 1668 1669 events_stats__auxtrace_error_warn(stats); 1670 1671 if (stats->nr_proc_map_timeout != 0) { 1672 ui__warning("%d map information files for pre-existing threads were\n" 1673 "not processed, if there are samples for addresses they\n" 1674 "will not be resolved, you may find out which are these\n" 1675 "threads by running with -v and redirecting the output\n" 1676 "to a file.\n" 1677 "The time limit to process proc map is too short?\n" 1678 "Increase it by --proc-map-timeout\n", 1679 stats->nr_proc_map_timeout); 1680 } 1681 } 1682 1683 static int perf_session__flush_thread_stack(struct thread *thread, 1684 void *p __maybe_unused) 1685 { 1686 return thread_stack__flush(thread); 1687 } 1688 1689 static int perf_session__flush_thread_stacks(struct perf_session *session) 1690 { 1691 return machines__for_each_thread(&session->machines, 1692 perf_session__flush_thread_stack, 1693 NULL); 1694 } 1695 1696 volatile int session_done; 1697 1698 static int __perf_session__process_pipe_events(struct perf_session *session) 1699 { 1700 struct ordered_events *oe = &session->ordered_events; 1701 struct perf_tool *tool = session->tool; 1702 int fd = perf_data__fd(session->data); 1703 union perf_event *event; 1704 uint32_t size, cur_size = 0; 1705 void *buf = NULL; 1706 s64 skip = 0; 1707 u64 head; 1708 ssize_t err; 1709 void *p; 1710 1711 perf_tool__fill_defaults(tool); 1712 1713 head = 0; 1714 cur_size = sizeof(union perf_event); 1715 1716 buf = malloc(cur_size); 1717 if (!buf) 1718 return -errno; 1719 ordered_events__set_copy_on_queue(oe, true); 1720 more: 1721 event = buf; 1722 err = readn(fd, event, sizeof(struct perf_event_header)); 1723 if (err <= 0) { 1724 if (err == 0) 1725 goto done; 1726 1727 pr_err("failed to read event header\n"); 1728 goto out_err; 1729 } 1730 1731 if (session->header.needs_swap) 1732 perf_event_header__bswap(&event->header); 1733 1734 size = event->header.size; 1735 if (size < sizeof(struct perf_event_header)) { 1736 pr_err("bad event header size\n"); 1737 goto out_err; 1738 } 1739 1740 if (size > cur_size) { 1741 void *new = realloc(buf, size); 1742 if (!new) { 1743 pr_err("failed to allocate memory to read event\n"); 1744 goto out_err; 1745 } 1746 buf = new; 1747 cur_size = size; 1748 event = buf; 1749 } 1750 p = event; 1751 p += sizeof(struct perf_event_header); 1752 1753 if (size - sizeof(struct perf_event_header)) { 1754 err = readn(fd, p, size - sizeof(struct perf_event_header)); 1755 if (err <= 0) { 1756 if (err == 0) { 1757 pr_err("unexpected end of event stream\n"); 1758 goto done; 1759 } 1760 1761 pr_err("failed to read event data\n"); 1762 goto out_err; 1763 } 1764 } 1765 1766 if ((skip = perf_session__process_event(session, event, head)) < 0) { 1767 pr_err("%#" PRIx64 " [%#x]: failed to process type: %d\n", 1768 head, event->header.size, event->header.type); 1769 err = -EINVAL; 1770 goto out_err; 1771 } 1772 1773 head += size; 1774 1775 if (skip > 0) 1776 head += skip; 1777 1778 if (!session_done()) 1779 goto more; 1780 done: 1781 /* do the final flush for ordered samples */ 1782 err = ordered_events__flush(oe, OE_FLUSH__FINAL); 1783 if (err) 1784 goto out_err; 1785 err = auxtrace__flush_events(session, tool); 1786 if (err) 1787 goto out_err; 1788 err = perf_session__flush_thread_stacks(session); 1789 out_err: 1790 free(buf); 1791 if (!tool->no_warn) 1792 perf_session__warn_about_errors(session); 1793 ordered_events__free(&session->ordered_events); 1794 auxtrace__free_events(session); 1795 return err; 1796 } 1797 1798 static union perf_event * 1799 fetch_mmaped_event(struct perf_session *session, 1800 u64 head, size_t mmap_size, char *buf) 1801 { 1802 union perf_event *event; 1803 1804 /* 1805 * Ensure we have enough space remaining to read 1806 * the size of the event in the headers. 1807 */ 1808 if (head + sizeof(event->header) > mmap_size) 1809 return NULL; 1810 1811 event = (union perf_event *)(buf + head); 1812 1813 if (session->header.needs_swap) 1814 perf_event_header__bswap(&event->header); 1815 1816 if (head + event->header.size > mmap_size) { 1817 /* We're not fetching the event so swap back again */ 1818 if (session->header.needs_swap) 1819 perf_event_header__bswap(&event->header); 1820 return NULL; 1821 } 1822 1823 return event; 1824 } 1825 1826 /* 1827 * On 64bit we can mmap the data file in one go. No need for tiny mmap 1828 * slices. On 32bit we use 32MB. 1829 */ 1830 #if BITS_PER_LONG == 64 1831 #define MMAP_SIZE ULLONG_MAX 1832 #define NUM_MMAPS 1 1833 #else 1834 #define MMAP_SIZE (32 * 1024 * 1024ULL) 1835 #define NUM_MMAPS 128 1836 #endif 1837 1838 struct reader { 1839 int fd; 1840 u64 data_size; 1841 u64 data_offset; 1842 }; 1843 1844 static int 1845 reader__process_events(struct reader *rd, struct perf_session *session, 1846 struct ui_progress *prog) 1847 { 1848 u64 data_size = rd->data_size; 1849 u64 head, page_offset, file_offset, file_pos, size; 1850 int err = 0, mmap_prot, mmap_flags, map_idx = 0; 1851 size_t mmap_size; 1852 char *buf, *mmaps[NUM_MMAPS]; 1853 union perf_event *event; 1854 s64 skip; 1855 1856 page_offset = page_size * (rd->data_offset / page_size); 1857 file_offset = page_offset; 1858 head = rd->data_offset - page_offset; 1859 1860 ui_progress__init_size(prog, data_size, "Processing events..."); 1861 1862 data_size += rd->data_offset; 1863 1864 mmap_size = MMAP_SIZE; 1865 if (mmap_size > data_size) { 1866 mmap_size = data_size; 1867 session->one_mmap = true; 1868 } 1869 1870 memset(mmaps, 0, sizeof(mmaps)); 1871 1872 mmap_prot = PROT_READ; 1873 mmap_flags = MAP_SHARED; 1874 1875 if (session->header.needs_swap) { 1876 mmap_prot |= PROT_WRITE; 1877 mmap_flags = MAP_PRIVATE; 1878 } 1879 remap: 1880 buf = mmap(NULL, mmap_size, mmap_prot, mmap_flags, rd->fd, 1881 file_offset); 1882 if (buf == MAP_FAILED) { 1883 pr_err("failed to mmap file\n"); 1884 err = -errno; 1885 goto out; 1886 } 1887 mmaps[map_idx] = buf; 1888 map_idx = (map_idx + 1) & (ARRAY_SIZE(mmaps) - 1); 1889 file_pos = file_offset + head; 1890 if (session->one_mmap) { 1891 session->one_mmap_addr = buf; 1892 session->one_mmap_offset = file_offset; 1893 } 1894 1895 more: 1896 event = fetch_mmaped_event(session, head, mmap_size, buf); 1897 if (!event) { 1898 if (mmaps[map_idx]) { 1899 munmap(mmaps[map_idx], mmap_size); 1900 mmaps[map_idx] = NULL; 1901 } 1902 1903 page_offset = page_size * (head / page_size); 1904 file_offset += page_offset; 1905 head -= page_offset; 1906 goto remap; 1907 } 1908 1909 size = event->header.size; 1910 1911 if (size < sizeof(struct perf_event_header) || 1912 (skip = perf_session__process_event(session, event, file_pos)) < 0) { 1913 pr_err("%#" PRIx64 " [%#x]: failed to process type: %d\n", 1914 file_offset + head, event->header.size, 1915 event->header.type); 1916 err = -EINVAL; 1917 goto out; 1918 } 1919 1920 if (skip) 1921 size += skip; 1922 1923 head += size; 1924 file_pos += size; 1925 1926 ui_progress__update(prog, size); 1927 1928 if (session_done()) 1929 goto out; 1930 1931 if (file_pos < data_size) 1932 goto more; 1933 1934 out: 1935 return err; 1936 } 1937 1938 static int __perf_session__process_events(struct perf_session *session) 1939 { 1940 struct reader rd = { 1941 .fd = perf_data__fd(session->data), 1942 .data_size = session->header.data_size, 1943 .data_offset = session->header.data_offset, 1944 }; 1945 struct ordered_events *oe = &session->ordered_events; 1946 struct perf_tool *tool = session->tool; 1947 struct ui_progress prog; 1948 int err; 1949 1950 perf_tool__fill_defaults(tool); 1951 1952 if (rd.data_size == 0) 1953 return -1; 1954 1955 ui_progress__init_size(&prog, rd.data_size, "Processing events..."); 1956 1957 err = reader__process_events(&rd, session, &prog); 1958 if (err) 1959 goto out_err; 1960 /* do the final flush for ordered samples */ 1961 err = ordered_events__flush(oe, OE_FLUSH__FINAL); 1962 if (err) 1963 goto out_err; 1964 err = auxtrace__flush_events(session, tool); 1965 if (err) 1966 goto out_err; 1967 err = perf_session__flush_thread_stacks(session); 1968 out_err: 1969 ui_progress__finish(); 1970 if (!tool->no_warn) 1971 perf_session__warn_about_errors(session); 1972 /* 1973 * We may switching perf.data output, make ordered_events 1974 * reusable. 1975 */ 1976 ordered_events__reinit(&session->ordered_events); 1977 auxtrace__free_events(session); 1978 session->one_mmap = false; 1979 return err; 1980 } 1981 1982 int perf_session__process_events(struct perf_session *session) 1983 { 1984 if (perf_session__register_idle_thread(session) < 0) 1985 return -ENOMEM; 1986 1987 if (perf_data__is_pipe(session->data)) 1988 return __perf_session__process_pipe_events(session); 1989 1990 return __perf_session__process_events(session); 1991 } 1992 1993 bool perf_session__has_traces(struct perf_session *session, const char *msg) 1994 { 1995 struct perf_evsel *evsel; 1996 1997 evlist__for_each_entry(session->evlist, evsel) { 1998 if (evsel->attr.type == PERF_TYPE_TRACEPOINT) 1999 return true; 2000 } 2001 2002 pr_err("No trace sample to read. Did you call 'perf %s'?\n", msg); 2003 return false; 2004 } 2005 2006 int map__set_kallsyms_ref_reloc_sym(struct map *map, const char *symbol_name, u64 addr) 2007 { 2008 char *bracket; 2009 struct ref_reloc_sym *ref; 2010 struct kmap *kmap; 2011 2012 ref = zalloc(sizeof(struct ref_reloc_sym)); 2013 if (ref == NULL) 2014 return -ENOMEM; 2015 2016 ref->name = strdup(symbol_name); 2017 if (ref->name == NULL) { 2018 free(ref); 2019 return -ENOMEM; 2020 } 2021 2022 bracket = strchr(ref->name, ']'); 2023 if (bracket) 2024 *bracket = '\0'; 2025 2026 ref->addr = addr; 2027 2028 kmap = map__kmap(map); 2029 if (kmap) 2030 kmap->ref_reloc_sym = ref; 2031 2032 return 0; 2033 } 2034 2035 size_t perf_session__fprintf_dsos(struct perf_session *session, FILE *fp) 2036 { 2037 return machines__fprintf_dsos(&session->machines, fp); 2038 } 2039 2040 size_t perf_session__fprintf_dsos_buildid(struct perf_session *session, FILE *fp, 2041 bool (skip)(struct dso *dso, int parm), int parm) 2042 { 2043 return machines__fprintf_dsos_buildid(&session->machines, fp, skip, parm); 2044 } 2045 2046 size_t perf_session__fprintf_nr_events(struct perf_session *session, FILE *fp) 2047 { 2048 size_t ret; 2049 const char *msg = ""; 2050 2051 if (perf_header__has_feat(&session->header, HEADER_AUXTRACE)) 2052 msg = " (excludes AUX area (e.g. instruction trace) decoded / synthesized events)"; 2053 2054 ret = fprintf(fp, "\nAggregated stats:%s\n", msg); 2055 2056 ret += events_stats__fprintf(&session->evlist->stats, fp); 2057 return ret; 2058 } 2059 2060 size_t perf_session__fprintf(struct perf_session *session, FILE *fp) 2061 { 2062 /* 2063 * FIXME: Here we have to actually print all the machines in this 2064 * session, not just the host... 2065 */ 2066 return machine__fprintf(&session->machines.host, fp); 2067 } 2068 2069 struct perf_evsel *perf_session__find_first_evtype(struct perf_session *session, 2070 unsigned int type) 2071 { 2072 struct perf_evsel *pos; 2073 2074 evlist__for_each_entry(session->evlist, pos) { 2075 if (pos->attr.type == type) 2076 return pos; 2077 } 2078 return NULL; 2079 } 2080 2081 int perf_session__cpu_bitmap(struct perf_session *session, 2082 const char *cpu_list, unsigned long *cpu_bitmap) 2083 { 2084 int i, err = -1; 2085 struct cpu_map *map; 2086 2087 for (i = 0; i < PERF_TYPE_MAX; ++i) { 2088 struct perf_evsel *evsel; 2089 2090 evsel = perf_session__find_first_evtype(session, i); 2091 if (!evsel) 2092 continue; 2093 2094 if (!(evsel->attr.sample_type & PERF_SAMPLE_CPU)) { 2095 pr_err("File does not contain CPU events. " 2096 "Remove -C option to proceed.\n"); 2097 return -1; 2098 } 2099 } 2100 2101 map = cpu_map__new(cpu_list); 2102 if (map == NULL) { 2103 pr_err("Invalid cpu_list\n"); 2104 return -1; 2105 } 2106 2107 for (i = 0; i < map->nr; i++) { 2108 int cpu = map->map[i]; 2109 2110 if (cpu >= MAX_NR_CPUS) { 2111 pr_err("Requested CPU %d too large. " 2112 "Consider raising MAX_NR_CPUS\n", cpu); 2113 goto out_delete_map; 2114 } 2115 2116 set_bit(cpu, cpu_bitmap); 2117 } 2118 2119 err = 0; 2120 2121 out_delete_map: 2122 cpu_map__put(map); 2123 return err; 2124 } 2125 2126 void perf_session__fprintf_info(struct perf_session *session, FILE *fp, 2127 bool full) 2128 { 2129 if (session == NULL || fp == NULL) 2130 return; 2131 2132 fprintf(fp, "# ========\n"); 2133 perf_header__fprintf_info(session, fp, full); 2134 fprintf(fp, "# ========\n#\n"); 2135 } 2136 2137 2138 int __perf_session__set_tracepoints_handlers(struct perf_session *session, 2139 const struct perf_evsel_str_handler *assocs, 2140 size_t nr_assocs) 2141 { 2142 struct perf_evsel *evsel; 2143 size_t i; 2144 int err; 2145 2146 for (i = 0; i < nr_assocs; i++) { 2147 /* 2148 * Adding a handler for an event not in the session, 2149 * just ignore it. 2150 */ 2151 evsel = perf_evlist__find_tracepoint_by_name(session->evlist, assocs[i].name); 2152 if (evsel == NULL) 2153 continue; 2154 2155 err = -EEXIST; 2156 if (evsel->handler != NULL) 2157 goto out; 2158 evsel->handler = assocs[i].handler; 2159 } 2160 2161 err = 0; 2162 out: 2163 return err; 2164 } 2165 2166 int perf_event__process_id_index(struct perf_session *session, 2167 union perf_event *event) 2168 { 2169 struct perf_evlist *evlist = session->evlist; 2170 struct id_index_event *ie = &event->id_index; 2171 size_t i, nr, max_nr; 2172 2173 max_nr = (ie->header.size - sizeof(struct id_index_event)) / 2174 sizeof(struct id_index_entry); 2175 nr = ie->nr; 2176 if (nr > max_nr) 2177 return -EINVAL; 2178 2179 if (dump_trace) 2180 fprintf(stdout, " nr: %zu\n", nr); 2181 2182 for (i = 0; i < nr; i++) { 2183 struct id_index_entry *e = &ie->entries[i]; 2184 struct perf_sample_id *sid; 2185 2186 if (dump_trace) { 2187 fprintf(stdout, " ... id: %"PRIu64, e->id); 2188 fprintf(stdout, " idx: %"PRIu64, e->idx); 2189 fprintf(stdout, " cpu: %"PRId64, e->cpu); 2190 fprintf(stdout, " tid: %"PRId64"\n", e->tid); 2191 } 2192 2193 sid = perf_evlist__id2sid(evlist, e->id); 2194 if (!sid) 2195 return -ENOENT; 2196 sid->idx = e->idx; 2197 sid->cpu = e->cpu; 2198 sid->tid = e->tid; 2199 } 2200 return 0; 2201 } 2202 2203 int perf_event__synthesize_id_index(struct perf_tool *tool, 2204 perf_event__handler_t process, 2205 struct perf_evlist *evlist, 2206 struct machine *machine) 2207 { 2208 union perf_event *ev; 2209 struct perf_evsel *evsel; 2210 size_t nr = 0, i = 0, sz, max_nr, n; 2211 int err; 2212 2213 pr_debug2("Synthesizing id index\n"); 2214 2215 max_nr = (UINT16_MAX - sizeof(struct id_index_event)) / 2216 sizeof(struct id_index_entry); 2217 2218 evlist__for_each_entry(evlist, evsel) 2219 nr += evsel->ids; 2220 2221 n = nr > max_nr ? max_nr : nr; 2222 sz = sizeof(struct id_index_event) + n * sizeof(struct id_index_entry); 2223 ev = zalloc(sz); 2224 if (!ev) 2225 return -ENOMEM; 2226 2227 ev->id_index.header.type = PERF_RECORD_ID_INDEX; 2228 ev->id_index.header.size = sz; 2229 ev->id_index.nr = n; 2230 2231 evlist__for_each_entry(evlist, evsel) { 2232 u32 j; 2233 2234 for (j = 0; j < evsel->ids; j++) { 2235 struct id_index_entry *e; 2236 struct perf_sample_id *sid; 2237 2238 if (i >= n) { 2239 err = process(tool, ev, NULL, machine); 2240 if (err) 2241 goto out_err; 2242 nr -= n; 2243 i = 0; 2244 } 2245 2246 e = &ev->id_index.entries[i++]; 2247 2248 e->id = evsel->id[j]; 2249 2250 sid = perf_evlist__id2sid(evlist, e->id); 2251 if (!sid) { 2252 free(ev); 2253 return -ENOENT; 2254 } 2255 2256 e->idx = sid->idx; 2257 e->cpu = sid->cpu; 2258 e->tid = sid->tid; 2259 } 2260 } 2261 2262 sz = sizeof(struct id_index_event) + nr * sizeof(struct id_index_entry); 2263 ev->id_index.header.size = sz; 2264 ev->id_index.nr = nr; 2265 2266 err = process(tool, ev, NULL, machine); 2267 out_err: 2268 free(ev); 2269 2270 return err; 2271 } 2272