1 // SPDX-License-Identifier: GPL-2.0 2 #include <errno.h> 3 #include <signal.h> 4 #include <inttypes.h> 5 #include <linux/err.h> 6 #include <linux/kernel.h> 7 #include <linux/zalloc.h> 8 #include <api/fs/fs.h> 9 10 #include <byteswap.h> 11 #include <unistd.h> 12 #include <sys/types.h> 13 #include <sys/mman.h> 14 #include <perf/cpumap.h> 15 16 #include "map_symbol.h" 17 #include "branch.h" 18 #include "debug.h" 19 #include "env.h" 20 #include "evlist.h" 21 #include "evsel.h" 22 #include "memswap.h" 23 #include "map.h" 24 #include "symbol.h" 25 #include "session.h" 26 #include "tool.h" 27 #include "perf_regs.h" 28 #include "asm/bug.h" 29 #include "auxtrace.h" 30 #include "thread.h" 31 #include "thread-stack.h" 32 #include "sample-raw.h" 33 #include "stat.h" 34 #include "tsc.h" 35 #include "ui/progress.h" 36 #include "../perf.h" 37 #include "arch/common.h" 38 #include "units.h" 39 #include <internal/lib.h> 40 41 #ifdef HAVE_ZSTD_SUPPORT 42 static int perf_session__process_compressed_event(struct perf_session *session, 43 union perf_event *event, u64 file_offset, 44 const char *file_path) 45 { 46 void *src; 47 size_t decomp_size, src_size; 48 u64 decomp_last_rem = 0; 49 size_t mmap_len, decomp_len = session->header.env.comp_mmap_len; 50 struct decomp *decomp, *decomp_last = session->active_decomp->decomp_last; 51 52 if (decomp_last) { 53 decomp_last_rem = decomp_last->size - decomp_last->head; 54 decomp_len += decomp_last_rem; 55 } 56 57 mmap_len = sizeof(struct decomp) + decomp_len; 58 decomp = mmap(NULL, mmap_len, PROT_READ|PROT_WRITE, 59 MAP_ANONYMOUS|MAP_PRIVATE, -1, 0); 60 if (decomp == MAP_FAILED) { 61 pr_err("Couldn't allocate memory for decompression\n"); 62 return -1; 63 } 64 65 decomp->file_pos = file_offset; 66 decomp->file_path = file_path; 67 decomp->mmap_len = mmap_len; 68 decomp->head = 0; 69 70 if (decomp_last_rem) { 71 memcpy(decomp->data, &(decomp_last->data[decomp_last->head]), decomp_last_rem); 72 decomp->size = decomp_last_rem; 73 } 74 75 src = (void *)event + sizeof(struct perf_record_compressed); 76 src_size = event->pack.header.size - sizeof(struct perf_record_compressed); 77 78 decomp_size = zstd_decompress_stream(session->active_decomp->zstd_decomp, src, src_size, 79 &(decomp->data[decomp_last_rem]), decomp_len - decomp_last_rem); 80 if (!decomp_size) { 81 munmap(decomp, mmap_len); 82 pr_err("Couldn't decompress data\n"); 83 return -1; 84 } 85 86 decomp->size += decomp_size; 87 88 if (session->active_decomp->decomp == NULL) 89 session->active_decomp->decomp = decomp; 90 else 91 session->active_decomp->decomp_last->next = decomp; 92 93 session->active_decomp->decomp_last = decomp; 94 95 pr_debug("decomp (B): %zd to %zd\n", src_size, decomp_size); 96 97 return 0; 98 } 99 #else /* !HAVE_ZSTD_SUPPORT */ 100 #define perf_session__process_compressed_event perf_session__process_compressed_event_stub 101 #endif 102 103 static int perf_session__deliver_event(struct perf_session *session, 104 union perf_event *event, 105 struct perf_tool *tool, 106 u64 file_offset, 107 const char *file_path); 108 109 static int perf_session__open(struct perf_session *session, int repipe_fd) 110 { 111 struct perf_data *data = session->data; 112 113 if (perf_session__read_header(session, repipe_fd) < 0) { 114 pr_err("incompatible file format (rerun with -v to learn more)\n"); 115 return -1; 116 } 117 118 if (perf_data__is_pipe(data)) 119 return 0; 120 121 if (perf_header__has_feat(&session->header, HEADER_STAT)) 122 return 0; 123 124 if (!evlist__valid_sample_type(session->evlist)) { 125 pr_err("non matching sample_type\n"); 126 return -1; 127 } 128 129 if (!evlist__valid_sample_id_all(session->evlist)) { 130 pr_err("non matching sample_id_all\n"); 131 return -1; 132 } 133 134 if (!evlist__valid_read_format(session->evlist)) { 135 pr_err("non matching read_format\n"); 136 return -1; 137 } 138 139 return 0; 140 } 141 142 void perf_session__set_id_hdr_size(struct perf_session *session) 143 { 144 u16 id_hdr_size = evlist__id_hdr_size(session->evlist); 145 146 machines__set_id_hdr_size(&session->machines, id_hdr_size); 147 } 148 149 int perf_session__create_kernel_maps(struct perf_session *session) 150 { 151 int ret = machine__create_kernel_maps(&session->machines.host); 152 153 if (ret >= 0) 154 ret = machines__create_guest_kernel_maps(&session->machines); 155 return ret; 156 } 157 158 static void perf_session__destroy_kernel_maps(struct perf_session *session) 159 { 160 machines__destroy_kernel_maps(&session->machines); 161 } 162 163 static bool perf_session__has_comm_exec(struct perf_session *session) 164 { 165 struct evsel *evsel; 166 167 evlist__for_each_entry(session->evlist, evsel) { 168 if (evsel->core.attr.comm_exec) 169 return true; 170 } 171 172 return false; 173 } 174 175 static void perf_session__set_comm_exec(struct perf_session *session) 176 { 177 bool comm_exec = perf_session__has_comm_exec(session); 178 179 machines__set_comm_exec(&session->machines, comm_exec); 180 } 181 182 static int ordered_events__deliver_event(struct ordered_events *oe, 183 struct ordered_event *event) 184 { 185 struct perf_session *session = container_of(oe, struct perf_session, 186 ordered_events); 187 188 return perf_session__deliver_event(session, event->event, 189 session->tool, event->file_offset, 190 event->file_path); 191 } 192 193 struct perf_session *__perf_session__new(struct perf_data *data, 194 bool repipe, int repipe_fd, 195 struct perf_tool *tool) 196 { 197 int ret = -ENOMEM; 198 struct perf_session *session = zalloc(sizeof(*session)); 199 200 if (!session) 201 goto out; 202 203 session->repipe = repipe; 204 session->tool = tool; 205 session->decomp_data.zstd_decomp = &session->zstd_data; 206 session->active_decomp = &session->decomp_data; 207 INIT_LIST_HEAD(&session->auxtrace_index); 208 machines__init(&session->machines); 209 ordered_events__init(&session->ordered_events, 210 ordered_events__deliver_event, NULL); 211 212 perf_env__init(&session->header.env); 213 if (data) { 214 ret = perf_data__open(data); 215 if (ret < 0) 216 goto out_delete; 217 218 session->data = data; 219 220 if (perf_data__is_read(data)) { 221 ret = perf_session__open(session, repipe_fd); 222 if (ret < 0) 223 goto out_delete; 224 225 /* 226 * set session attributes that are present in perf.data 227 * but not in pipe-mode. 228 */ 229 if (!data->is_pipe) { 230 perf_session__set_id_hdr_size(session); 231 perf_session__set_comm_exec(session); 232 } 233 234 evlist__init_trace_event_sample_raw(session->evlist); 235 236 /* Open the directory data. */ 237 if (data->is_dir) { 238 ret = perf_data__open_dir(data); 239 if (ret) 240 goto out_delete; 241 } 242 243 if (!symbol_conf.kallsyms_name && 244 !symbol_conf.vmlinux_name) 245 symbol_conf.kallsyms_name = perf_data__kallsyms_name(data); 246 } 247 } else { 248 session->machines.host.env = &perf_env; 249 } 250 251 session->machines.host.single_address_space = 252 perf_env__single_address_space(session->machines.host.env); 253 254 if (!data || perf_data__is_write(data)) { 255 /* 256 * In O_RDONLY mode this will be performed when reading the 257 * kernel MMAP event, in perf_event__process_mmap(). 258 */ 259 if (perf_session__create_kernel_maps(session) < 0) 260 pr_warning("Cannot read kernel map\n"); 261 } 262 263 /* 264 * In pipe-mode, evlist is empty until PERF_RECORD_HEADER_ATTR is 265 * processed, so evlist__sample_id_all is not meaningful here. 266 */ 267 if ((!data || !data->is_pipe) && tool && tool->ordering_requires_timestamps && 268 tool->ordered_events && !evlist__sample_id_all(session->evlist)) { 269 dump_printf("WARNING: No sample_id_all support, falling back to unordered processing\n"); 270 tool->ordered_events = false; 271 } 272 273 return session; 274 275 out_delete: 276 perf_session__delete(session); 277 out: 278 return ERR_PTR(ret); 279 } 280 281 static void perf_session__delete_threads(struct perf_session *session) 282 { 283 machine__delete_threads(&session->machines.host); 284 } 285 286 static void perf_decomp__release_events(struct decomp *next) 287 { 288 struct decomp *decomp; 289 size_t mmap_len; 290 291 do { 292 decomp = next; 293 if (decomp == NULL) 294 break; 295 next = decomp->next; 296 mmap_len = decomp->mmap_len; 297 munmap(decomp, mmap_len); 298 } while (1); 299 } 300 301 void perf_session__delete(struct perf_session *session) 302 { 303 if (session == NULL) 304 return; 305 auxtrace__free(session); 306 auxtrace_index__free(&session->auxtrace_index); 307 perf_session__destroy_kernel_maps(session); 308 perf_session__delete_threads(session); 309 perf_decomp__release_events(session->decomp_data.decomp); 310 perf_env__exit(&session->header.env); 311 machines__exit(&session->machines); 312 if (session->data) { 313 if (perf_data__is_read(session->data)) 314 evlist__delete(session->evlist); 315 perf_data__close(session->data); 316 } 317 #ifdef HAVE_LIBTRACEEVENT 318 trace_event__cleanup(&session->tevent); 319 #endif 320 free(session); 321 } 322 323 static int process_event_synth_tracing_data_stub(struct perf_session *session 324 __maybe_unused, 325 union perf_event *event 326 __maybe_unused) 327 { 328 dump_printf(": unhandled!\n"); 329 return 0; 330 } 331 332 static int process_event_synth_attr_stub(struct perf_tool *tool __maybe_unused, 333 union perf_event *event __maybe_unused, 334 struct evlist **pevlist 335 __maybe_unused) 336 { 337 dump_printf(": unhandled!\n"); 338 return 0; 339 } 340 341 static int process_event_synth_event_update_stub(struct perf_tool *tool __maybe_unused, 342 union perf_event *event __maybe_unused, 343 struct evlist **pevlist 344 __maybe_unused) 345 { 346 if (dump_trace) 347 perf_event__fprintf_event_update(event, stdout); 348 349 dump_printf(": unhandled!\n"); 350 return 0; 351 } 352 353 static int process_event_sample_stub(struct perf_tool *tool __maybe_unused, 354 union perf_event *event __maybe_unused, 355 struct perf_sample *sample __maybe_unused, 356 struct evsel *evsel __maybe_unused, 357 struct machine *machine __maybe_unused) 358 { 359 dump_printf(": unhandled!\n"); 360 return 0; 361 } 362 363 static int process_event_stub(struct perf_tool *tool __maybe_unused, 364 union perf_event *event __maybe_unused, 365 struct perf_sample *sample __maybe_unused, 366 struct machine *machine __maybe_unused) 367 { 368 dump_printf(": unhandled!\n"); 369 return 0; 370 } 371 372 static int process_finished_round_stub(struct perf_tool *tool __maybe_unused, 373 union perf_event *event __maybe_unused, 374 struct ordered_events *oe __maybe_unused) 375 { 376 dump_printf(": unhandled!\n"); 377 return 0; 378 } 379 380 static int skipn(int fd, off_t n) 381 { 382 char buf[4096]; 383 ssize_t ret; 384 385 while (n > 0) { 386 ret = read(fd, buf, min(n, (off_t)sizeof(buf))); 387 if (ret <= 0) 388 return ret; 389 n -= ret; 390 } 391 392 return 0; 393 } 394 395 static s64 process_event_auxtrace_stub(struct perf_session *session __maybe_unused, 396 union perf_event *event) 397 { 398 dump_printf(": unhandled!\n"); 399 if (perf_data__is_pipe(session->data)) 400 skipn(perf_data__fd(session->data), event->auxtrace.size); 401 return event->auxtrace.size; 402 } 403 404 static int process_event_op2_stub(struct perf_session *session __maybe_unused, 405 union perf_event *event __maybe_unused) 406 { 407 dump_printf(": unhandled!\n"); 408 return 0; 409 } 410 411 412 static 413 int process_event_thread_map_stub(struct perf_session *session __maybe_unused, 414 union perf_event *event __maybe_unused) 415 { 416 if (dump_trace) 417 perf_event__fprintf_thread_map(event, stdout); 418 419 dump_printf(": unhandled!\n"); 420 return 0; 421 } 422 423 static 424 int process_event_cpu_map_stub(struct perf_session *session __maybe_unused, 425 union perf_event *event __maybe_unused) 426 { 427 if (dump_trace) 428 perf_event__fprintf_cpu_map(event, stdout); 429 430 dump_printf(": unhandled!\n"); 431 return 0; 432 } 433 434 static 435 int process_event_stat_config_stub(struct perf_session *session __maybe_unused, 436 union perf_event *event __maybe_unused) 437 { 438 if (dump_trace) 439 perf_event__fprintf_stat_config(event, stdout); 440 441 dump_printf(": unhandled!\n"); 442 return 0; 443 } 444 445 static int process_stat_stub(struct perf_session *perf_session __maybe_unused, 446 union perf_event *event) 447 { 448 if (dump_trace) 449 perf_event__fprintf_stat(event, stdout); 450 451 dump_printf(": unhandled!\n"); 452 return 0; 453 } 454 455 static int process_stat_round_stub(struct perf_session *perf_session __maybe_unused, 456 union perf_event *event) 457 { 458 if (dump_trace) 459 perf_event__fprintf_stat_round(event, stdout); 460 461 dump_printf(": unhandled!\n"); 462 return 0; 463 } 464 465 static int process_event_time_conv_stub(struct perf_session *perf_session __maybe_unused, 466 union perf_event *event) 467 { 468 if (dump_trace) 469 perf_event__fprintf_time_conv(event, stdout); 470 471 dump_printf(": unhandled!\n"); 472 return 0; 473 } 474 475 static int perf_session__process_compressed_event_stub(struct perf_session *session __maybe_unused, 476 union perf_event *event __maybe_unused, 477 u64 file_offset __maybe_unused, 478 const char *file_path __maybe_unused) 479 { 480 dump_printf(": unhandled!\n"); 481 return 0; 482 } 483 484 void perf_tool__fill_defaults(struct perf_tool *tool) 485 { 486 if (tool->sample == NULL) 487 tool->sample = process_event_sample_stub; 488 if (tool->mmap == NULL) 489 tool->mmap = process_event_stub; 490 if (tool->mmap2 == NULL) 491 tool->mmap2 = process_event_stub; 492 if (tool->comm == NULL) 493 tool->comm = process_event_stub; 494 if (tool->namespaces == NULL) 495 tool->namespaces = process_event_stub; 496 if (tool->cgroup == NULL) 497 tool->cgroup = process_event_stub; 498 if (tool->fork == NULL) 499 tool->fork = process_event_stub; 500 if (tool->exit == NULL) 501 tool->exit = process_event_stub; 502 if (tool->lost == NULL) 503 tool->lost = perf_event__process_lost; 504 if (tool->lost_samples == NULL) 505 tool->lost_samples = perf_event__process_lost_samples; 506 if (tool->aux == NULL) 507 tool->aux = perf_event__process_aux; 508 if (tool->itrace_start == NULL) 509 tool->itrace_start = perf_event__process_itrace_start; 510 if (tool->context_switch == NULL) 511 tool->context_switch = perf_event__process_switch; 512 if (tool->ksymbol == NULL) 513 tool->ksymbol = perf_event__process_ksymbol; 514 if (tool->bpf == NULL) 515 tool->bpf = perf_event__process_bpf; 516 if (tool->text_poke == NULL) 517 tool->text_poke = perf_event__process_text_poke; 518 if (tool->aux_output_hw_id == NULL) 519 tool->aux_output_hw_id = perf_event__process_aux_output_hw_id; 520 if (tool->read == NULL) 521 tool->read = process_event_sample_stub; 522 if (tool->throttle == NULL) 523 tool->throttle = process_event_stub; 524 if (tool->unthrottle == NULL) 525 tool->unthrottle = process_event_stub; 526 if (tool->attr == NULL) 527 tool->attr = process_event_synth_attr_stub; 528 if (tool->event_update == NULL) 529 tool->event_update = process_event_synth_event_update_stub; 530 if (tool->tracing_data == NULL) 531 tool->tracing_data = process_event_synth_tracing_data_stub; 532 if (tool->build_id == NULL) 533 tool->build_id = process_event_op2_stub; 534 if (tool->finished_round == NULL) { 535 if (tool->ordered_events) 536 tool->finished_round = perf_event__process_finished_round; 537 else 538 tool->finished_round = process_finished_round_stub; 539 } 540 if (tool->id_index == NULL) 541 tool->id_index = process_event_op2_stub; 542 if (tool->auxtrace_info == NULL) 543 tool->auxtrace_info = process_event_op2_stub; 544 if (tool->auxtrace == NULL) 545 tool->auxtrace = process_event_auxtrace_stub; 546 if (tool->auxtrace_error == NULL) 547 tool->auxtrace_error = process_event_op2_stub; 548 if (tool->thread_map == NULL) 549 tool->thread_map = process_event_thread_map_stub; 550 if (tool->cpu_map == NULL) 551 tool->cpu_map = process_event_cpu_map_stub; 552 if (tool->stat_config == NULL) 553 tool->stat_config = process_event_stat_config_stub; 554 if (tool->stat == NULL) 555 tool->stat = process_stat_stub; 556 if (tool->stat_round == NULL) 557 tool->stat_round = process_stat_round_stub; 558 if (tool->time_conv == NULL) 559 tool->time_conv = process_event_time_conv_stub; 560 if (tool->feature == NULL) 561 tool->feature = process_event_op2_stub; 562 if (tool->compressed == NULL) 563 tool->compressed = perf_session__process_compressed_event; 564 if (tool->finished_init == NULL) 565 tool->finished_init = process_event_op2_stub; 566 } 567 568 static void swap_sample_id_all(union perf_event *event, void *data) 569 { 570 void *end = (void *) event + event->header.size; 571 int size = end - data; 572 573 BUG_ON(size % sizeof(u64)); 574 mem_bswap_64(data, size); 575 } 576 577 static void perf_event__all64_swap(union perf_event *event, 578 bool sample_id_all __maybe_unused) 579 { 580 struct perf_event_header *hdr = &event->header; 581 mem_bswap_64(hdr + 1, event->header.size - sizeof(*hdr)); 582 } 583 584 static void perf_event__comm_swap(union perf_event *event, bool sample_id_all) 585 { 586 event->comm.pid = bswap_32(event->comm.pid); 587 event->comm.tid = bswap_32(event->comm.tid); 588 589 if (sample_id_all) { 590 void *data = &event->comm.comm; 591 592 data += PERF_ALIGN(strlen(data) + 1, sizeof(u64)); 593 swap_sample_id_all(event, data); 594 } 595 } 596 597 static void perf_event__mmap_swap(union perf_event *event, 598 bool sample_id_all) 599 { 600 event->mmap.pid = bswap_32(event->mmap.pid); 601 event->mmap.tid = bswap_32(event->mmap.tid); 602 event->mmap.start = bswap_64(event->mmap.start); 603 event->mmap.len = bswap_64(event->mmap.len); 604 event->mmap.pgoff = bswap_64(event->mmap.pgoff); 605 606 if (sample_id_all) { 607 void *data = &event->mmap.filename; 608 609 data += PERF_ALIGN(strlen(data) + 1, sizeof(u64)); 610 swap_sample_id_all(event, data); 611 } 612 } 613 614 static void perf_event__mmap2_swap(union perf_event *event, 615 bool sample_id_all) 616 { 617 event->mmap2.pid = bswap_32(event->mmap2.pid); 618 event->mmap2.tid = bswap_32(event->mmap2.tid); 619 event->mmap2.start = bswap_64(event->mmap2.start); 620 event->mmap2.len = bswap_64(event->mmap2.len); 621 event->mmap2.pgoff = bswap_64(event->mmap2.pgoff); 622 623 if (!(event->header.misc & PERF_RECORD_MISC_MMAP_BUILD_ID)) { 624 event->mmap2.maj = bswap_32(event->mmap2.maj); 625 event->mmap2.min = bswap_32(event->mmap2.min); 626 event->mmap2.ino = bswap_64(event->mmap2.ino); 627 event->mmap2.ino_generation = bswap_64(event->mmap2.ino_generation); 628 } 629 630 if (sample_id_all) { 631 void *data = &event->mmap2.filename; 632 633 data += PERF_ALIGN(strlen(data) + 1, sizeof(u64)); 634 swap_sample_id_all(event, data); 635 } 636 } 637 static void perf_event__task_swap(union perf_event *event, bool sample_id_all) 638 { 639 event->fork.pid = bswap_32(event->fork.pid); 640 event->fork.tid = bswap_32(event->fork.tid); 641 event->fork.ppid = bswap_32(event->fork.ppid); 642 event->fork.ptid = bswap_32(event->fork.ptid); 643 event->fork.time = bswap_64(event->fork.time); 644 645 if (sample_id_all) 646 swap_sample_id_all(event, &event->fork + 1); 647 } 648 649 static void perf_event__read_swap(union perf_event *event, bool sample_id_all) 650 { 651 event->read.pid = bswap_32(event->read.pid); 652 event->read.tid = bswap_32(event->read.tid); 653 event->read.value = bswap_64(event->read.value); 654 event->read.time_enabled = bswap_64(event->read.time_enabled); 655 event->read.time_running = bswap_64(event->read.time_running); 656 event->read.id = bswap_64(event->read.id); 657 658 if (sample_id_all) 659 swap_sample_id_all(event, &event->read + 1); 660 } 661 662 static void perf_event__aux_swap(union perf_event *event, bool sample_id_all) 663 { 664 event->aux.aux_offset = bswap_64(event->aux.aux_offset); 665 event->aux.aux_size = bswap_64(event->aux.aux_size); 666 event->aux.flags = bswap_64(event->aux.flags); 667 668 if (sample_id_all) 669 swap_sample_id_all(event, &event->aux + 1); 670 } 671 672 static void perf_event__itrace_start_swap(union perf_event *event, 673 bool sample_id_all) 674 { 675 event->itrace_start.pid = bswap_32(event->itrace_start.pid); 676 event->itrace_start.tid = bswap_32(event->itrace_start.tid); 677 678 if (sample_id_all) 679 swap_sample_id_all(event, &event->itrace_start + 1); 680 } 681 682 static void perf_event__switch_swap(union perf_event *event, bool sample_id_all) 683 { 684 if (event->header.type == PERF_RECORD_SWITCH_CPU_WIDE) { 685 event->context_switch.next_prev_pid = 686 bswap_32(event->context_switch.next_prev_pid); 687 event->context_switch.next_prev_tid = 688 bswap_32(event->context_switch.next_prev_tid); 689 } 690 691 if (sample_id_all) 692 swap_sample_id_all(event, &event->context_switch + 1); 693 } 694 695 static void perf_event__text_poke_swap(union perf_event *event, bool sample_id_all) 696 { 697 event->text_poke.addr = bswap_64(event->text_poke.addr); 698 event->text_poke.old_len = bswap_16(event->text_poke.old_len); 699 event->text_poke.new_len = bswap_16(event->text_poke.new_len); 700 701 if (sample_id_all) { 702 size_t len = sizeof(event->text_poke.old_len) + 703 sizeof(event->text_poke.new_len) + 704 event->text_poke.old_len + 705 event->text_poke.new_len; 706 void *data = &event->text_poke.old_len; 707 708 data += PERF_ALIGN(len, sizeof(u64)); 709 swap_sample_id_all(event, data); 710 } 711 } 712 713 static void perf_event__throttle_swap(union perf_event *event, 714 bool sample_id_all) 715 { 716 event->throttle.time = bswap_64(event->throttle.time); 717 event->throttle.id = bswap_64(event->throttle.id); 718 event->throttle.stream_id = bswap_64(event->throttle.stream_id); 719 720 if (sample_id_all) 721 swap_sample_id_all(event, &event->throttle + 1); 722 } 723 724 static void perf_event__namespaces_swap(union perf_event *event, 725 bool sample_id_all) 726 { 727 u64 i; 728 729 event->namespaces.pid = bswap_32(event->namespaces.pid); 730 event->namespaces.tid = bswap_32(event->namespaces.tid); 731 event->namespaces.nr_namespaces = bswap_64(event->namespaces.nr_namespaces); 732 733 for (i = 0; i < event->namespaces.nr_namespaces; i++) { 734 struct perf_ns_link_info *ns = &event->namespaces.link_info[i]; 735 736 ns->dev = bswap_64(ns->dev); 737 ns->ino = bswap_64(ns->ino); 738 } 739 740 if (sample_id_all) 741 swap_sample_id_all(event, &event->namespaces.link_info[i]); 742 } 743 744 static void perf_event__cgroup_swap(union perf_event *event, bool sample_id_all) 745 { 746 event->cgroup.id = bswap_64(event->cgroup.id); 747 748 if (sample_id_all) { 749 void *data = &event->cgroup.path; 750 751 data += PERF_ALIGN(strlen(data) + 1, sizeof(u64)); 752 swap_sample_id_all(event, data); 753 } 754 } 755 756 static u8 revbyte(u8 b) 757 { 758 int rev = (b >> 4) | ((b & 0xf) << 4); 759 rev = ((rev & 0xcc) >> 2) | ((rev & 0x33) << 2); 760 rev = ((rev & 0xaa) >> 1) | ((rev & 0x55) << 1); 761 return (u8) rev; 762 } 763 764 /* 765 * XXX this is hack in attempt to carry flags bitfield 766 * through endian village. ABI says: 767 * 768 * Bit-fields are allocated from right to left (least to most significant) 769 * on little-endian implementations and from left to right (most to least 770 * significant) on big-endian implementations. 771 * 772 * The above seems to be byte specific, so we need to reverse each 773 * byte of the bitfield. 'Internet' also says this might be implementation 774 * specific and we probably need proper fix and carry perf_event_attr 775 * bitfield flags in separate data file FEAT_ section. Thought this seems 776 * to work for now. 777 */ 778 static void swap_bitfield(u8 *p, unsigned len) 779 { 780 unsigned i; 781 782 for (i = 0; i < len; i++) { 783 *p = revbyte(*p); 784 p++; 785 } 786 } 787 788 /* exported for swapping attributes in file header */ 789 void perf_event__attr_swap(struct perf_event_attr *attr) 790 { 791 attr->type = bswap_32(attr->type); 792 attr->size = bswap_32(attr->size); 793 794 #define bswap_safe(f, n) \ 795 (attr->size > (offsetof(struct perf_event_attr, f) + \ 796 sizeof(attr->f) * (n))) 797 #define bswap_field(f, sz) \ 798 do { \ 799 if (bswap_safe(f, 0)) \ 800 attr->f = bswap_##sz(attr->f); \ 801 } while(0) 802 #define bswap_field_16(f) bswap_field(f, 16) 803 #define bswap_field_32(f) bswap_field(f, 32) 804 #define bswap_field_64(f) bswap_field(f, 64) 805 806 bswap_field_64(config); 807 bswap_field_64(sample_period); 808 bswap_field_64(sample_type); 809 bswap_field_64(read_format); 810 bswap_field_32(wakeup_events); 811 bswap_field_32(bp_type); 812 bswap_field_64(bp_addr); 813 bswap_field_64(bp_len); 814 bswap_field_64(branch_sample_type); 815 bswap_field_64(sample_regs_user); 816 bswap_field_32(sample_stack_user); 817 bswap_field_32(aux_watermark); 818 bswap_field_16(sample_max_stack); 819 bswap_field_32(aux_sample_size); 820 821 /* 822 * After read_format are bitfields. Check read_format because 823 * we are unable to use offsetof on bitfield. 824 */ 825 if (bswap_safe(read_format, 1)) 826 swap_bitfield((u8 *) (&attr->read_format + 1), 827 sizeof(u64)); 828 #undef bswap_field_64 829 #undef bswap_field_32 830 #undef bswap_field 831 #undef bswap_safe 832 } 833 834 static void perf_event__hdr_attr_swap(union perf_event *event, 835 bool sample_id_all __maybe_unused) 836 { 837 size_t size; 838 839 perf_event__attr_swap(&event->attr.attr); 840 841 size = event->header.size; 842 size -= (void *)&event->attr.id - (void *)event; 843 mem_bswap_64(event->attr.id, size); 844 } 845 846 static void perf_event__event_update_swap(union perf_event *event, 847 bool sample_id_all __maybe_unused) 848 { 849 event->event_update.type = bswap_64(event->event_update.type); 850 event->event_update.id = bswap_64(event->event_update.id); 851 } 852 853 static void perf_event__event_type_swap(union perf_event *event, 854 bool sample_id_all __maybe_unused) 855 { 856 event->event_type.event_type.event_id = 857 bswap_64(event->event_type.event_type.event_id); 858 } 859 860 static void perf_event__tracing_data_swap(union perf_event *event, 861 bool sample_id_all __maybe_unused) 862 { 863 event->tracing_data.size = bswap_32(event->tracing_data.size); 864 } 865 866 static void perf_event__auxtrace_info_swap(union perf_event *event, 867 bool sample_id_all __maybe_unused) 868 { 869 size_t size; 870 871 event->auxtrace_info.type = bswap_32(event->auxtrace_info.type); 872 873 size = event->header.size; 874 size -= (void *)&event->auxtrace_info.priv - (void *)event; 875 mem_bswap_64(event->auxtrace_info.priv, size); 876 } 877 878 static void perf_event__auxtrace_swap(union perf_event *event, 879 bool sample_id_all __maybe_unused) 880 { 881 event->auxtrace.size = bswap_64(event->auxtrace.size); 882 event->auxtrace.offset = bswap_64(event->auxtrace.offset); 883 event->auxtrace.reference = bswap_64(event->auxtrace.reference); 884 event->auxtrace.idx = bswap_32(event->auxtrace.idx); 885 event->auxtrace.tid = bswap_32(event->auxtrace.tid); 886 event->auxtrace.cpu = bswap_32(event->auxtrace.cpu); 887 } 888 889 static void perf_event__auxtrace_error_swap(union perf_event *event, 890 bool sample_id_all __maybe_unused) 891 { 892 event->auxtrace_error.type = bswap_32(event->auxtrace_error.type); 893 event->auxtrace_error.code = bswap_32(event->auxtrace_error.code); 894 event->auxtrace_error.cpu = bswap_32(event->auxtrace_error.cpu); 895 event->auxtrace_error.pid = bswap_32(event->auxtrace_error.pid); 896 event->auxtrace_error.tid = bswap_32(event->auxtrace_error.tid); 897 event->auxtrace_error.fmt = bswap_32(event->auxtrace_error.fmt); 898 event->auxtrace_error.ip = bswap_64(event->auxtrace_error.ip); 899 if (event->auxtrace_error.fmt) 900 event->auxtrace_error.time = bswap_64(event->auxtrace_error.time); 901 if (event->auxtrace_error.fmt >= 2) { 902 event->auxtrace_error.machine_pid = bswap_32(event->auxtrace_error.machine_pid); 903 event->auxtrace_error.vcpu = bswap_32(event->auxtrace_error.vcpu); 904 } 905 } 906 907 static void perf_event__thread_map_swap(union perf_event *event, 908 bool sample_id_all __maybe_unused) 909 { 910 unsigned i; 911 912 event->thread_map.nr = bswap_64(event->thread_map.nr); 913 914 for (i = 0; i < event->thread_map.nr; i++) 915 event->thread_map.entries[i].pid = bswap_64(event->thread_map.entries[i].pid); 916 } 917 918 static void perf_event__cpu_map_swap(union perf_event *event, 919 bool sample_id_all __maybe_unused) 920 { 921 struct perf_record_cpu_map_data *data = &event->cpu_map.data; 922 923 data->type = bswap_16(data->type); 924 925 switch (data->type) { 926 case PERF_CPU_MAP__CPUS: 927 data->cpus_data.nr = bswap_16(data->cpus_data.nr); 928 929 for (unsigned i = 0; i < data->cpus_data.nr; i++) 930 data->cpus_data.cpu[i] = bswap_16(data->cpus_data.cpu[i]); 931 break; 932 case PERF_CPU_MAP__MASK: 933 data->mask32_data.long_size = bswap_16(data->mask32_data.long_size); 934 935 switch (data->mask32_data.long_size) { 936 case 4: 937 data->mask32_data.nr = bswap_16(data->mask32_data.nr); 938 for (unsigned i = 0; i < data->mask32_data.nr; i++) 939 data->mask32_data.mask[i] = bswap_32(data->mask32_data.mask[i]); 940 break; 941 case 8: 942 data->mask64_data.nr = bswap_16(data->mask64_data.nr); 943 for (unsigned i = 0; i < data->mask64_data.nr; i++) 944 data->mask64_data.mask[i] = bswap_64(data->mask64_data.mask[i]); 945 break; 946 default: 947 pr_err("cpu_map swap: unsupported long size\n"); 948 } 949 break; 950 case PERF_CPU_MAP__RANGE_CPUS: 951 data->range_cpu_data.start_cpu = bswap_16(data->range_cpu_data.start_cpu); 952 data->range_cpu_data.end_cpu = bswap_16(data->range_cpu_data.end_cpu); 953 break; 954 default: 955 break; 956 } 957 } 958 959 static void perf_event__stat_config_swap(union perf_event *event, 960 bool sample_id_all __maybe_unused) 961 { 962 u64 size; 963 964 size = bswap_64(event->stat_config.nr) * sizeof(event->stat_config.data[0]); 965 size += 1; /* nr item itself */ 966 mem_bswap_64(&event->stat_config.nr, size); 967 } 968 969 static void perf_event__stat_swap(union perf_event *event, 970 bool sample_id_all __maybe_unused) 971 { 972 event->stat.id = bswap_64(event->stat.id); 973 event->stat.thread = bswap_32(event->stat.thread); 974 event->stat.cpu = bswap_32(event->stat.cpu); 975 event->stat.val = bswap_64(event->stat.val); 976 event->stat.ena = bswap_64(event->stat.ena); 977 event->stat.run = bswap_64(event->stat.run); 978 } 979 980 static void perf_event__stat_round_swap(union perf_event *event, 981 bool sample_id_all __maybe_unused) 982 { 983 event->stat_round.type = bswap_64(event->stat_round.type); 984 event->stat_round.time = bswap_64(event->stat_round.time); 985 } 986 987 static void perf_event__time_conv_swap(union perf_event *event, 988 bool sample_id_all __maybe_unused) 989 { 990 event->time_conv.time_shift = bswap_64(event->time_conv.time_shift); 991 event->time_conv.time_mult = bswap_64(event->time_conv.time_mult); 992 event->time_conv.time_zero = bswap_64(event->time_conv.time_zero); 993 994 if (event_contains(event->time_conv, time_cycles)) { 995 event->time_conv.time_cycles = bswap_64(event->time_conv.time_cycles); 996 event->time_conv.time_mask = bswap_64(event->time_conv.time_mask); 997 } 998 } 999 1000 typedef void (*perf_event__swap_op)(union perf_event *event, 1001 bool sample_id_all); 1002 1003 static perf_event__swap_op perf_event__swap_ops[] = { 1004 [PERF_RECORD_MMAP] = perf_event__mmap_swap, 1005 [PERF_RECORD_MMAP2] = perf_event__mmap2_swap, 1006 [PERF_RECORD_COMM] = perf_event__comm_swap, 1007 [PERF_RECORD_FORK] = perf_event__task_swap, 1008 [PERF_RECORD_EXIT] = perf_event__task_swap, 1009 [PERF_RECORD_LOST] = perf_event__all64_swap, 1010 [PERF_RECORD_READ] = perf_event__read_swap, 1011 [PERF_RECORD_THROTTLE] = perf_event__throttle_swap, 1012 [PERF_RECORD_UNTHROTTLE] = perf_event__throttle_swap, 1013 [PERF_RECORD_SAMPLE] = perf_event__all64_swap, 1014 [PERF_RECORD_AUX] = perf_event__aux_swap, 1015 [PERF_RECORD_ITRACE_START] = perf_event__itrace_start_swap, 1016 [PERF_RECORD_LOST_SAMPLES] = perf_event__all64_swap, 1017 [PERF_RECORD_SWITCH] = perf_event__switch_swap, 1018 [PERF_RECORD_SWITCH_CPU_WIDE] = perf_event__switch_swap, 1019 [PERF_RECORD_NAMESPACES] = perf_event__namespaces_swap, 1020 [PERF_RECORD_CGROUP] = perf_event__cgroup_swap, 1021 [PERF_RECORD_TEXT_POKE] = perf_event__text_poke_swap, 1022 [PERF_RECORD_AUX_OUTPUT_HW_ID] = perf_event__all64_swap, 1023 [PERF_RECORD_HEADER_ATTR] = perf_event__hdr_attr_swap, 1024 [PERF_RECORD_HEADER_EVENT_TYPE] = perf_event__event_type_swap, 1025 [PERF_RECORD_HEADER_TRACING_DATA] = perf_event__tracing_data_swap, 1026 [PERF_RECORD_HEADER_BUILD_ID] = NULL, 1027 [PERF_RECORD_ID_INDEX] = perf_event__all64_swap, 1028 [PERF_RECORD_AUXTRACE_INFO] = perf_event__auxtrace_info_swap, 1029 [PERF_RECORD_AUXTRACE] = perf_event__auxtrace_swap, 1030 [PERF_RECORD_AUXTRACE_ERROR] = perf_event__auxtrace_error_swap, 1031 [PERF_RECORD_THREAD_MAP] = perf_event__thread_map_swap, 1032 [PERF_RECORD_CPU_MAP] = perf_event__cpu_map_swap, 1033 [PERF_RECORD_STAT_CONFIG] = perf_event__stat_config_swap, 1034 [PERF_RECORD_STAT] = perf_event__stat_swap, 1035 [PERF_RECORD_STAT_ROUND] = perf_event__stat_round_swap, 1036 [PERF_RECORD_EVENT_UPDATE] = perf_event__event_update_swap, 1037 [PERF_RECORD_TIME_CONV] = perf_event__time_conv_swap, 1038 [PERF_RECORD_HEADER_MAX] = NULL, 1039 }; 1040 1041 /* 1042 * When perf record finishes a pass on every buffers, it records this pseudo 1043 * event. 1044 * We record the max timestamp t found in the pass n. 1045 * Assuming these timestamps are monotonic across cpus, we know that if 1046 * a buffer still has events with timestamps below t, they will be all 1047 * available and then read in the pass n + 1. 1048 * Hence when we start to read the pass n + 2, we can safely flush every 1049 * events with timestamps below t. 1050 * 1051 * ============ PASS n ================= 1052 * CPU 0 | CPU 1 1053 * | 1054 * cnt1 timestamps | cnt2 timestamps 1055 * 1 | 2 1056 * 2 | 3 1057 * - | 4 <--- max recorded 1058 * 1059 * ============ PASS n + 1 ============== 1060 * CPU 0 | CPU 1 1061 * | 1062 * cnt1 timestamps | cnt2 timestamps 1063 * 3 | 5 1064 * 4 | 6 1065 * 5 | 7 <---- max recorded 1066 * 1067 * Flush every events below timestamp 4 1068 * 1069 * ============ PASS n + 2 ============== 1070 * CPU 0 | CPU 1 1071 * | 1072 * cnt1 timestamps | cnt2 timestamps 1073 * 6 | 8 1074 * 7 | 9 1075 * - | 10 1076 * 1077 * Flush every events below timestamp 7 1078 * etc... 1079 */ 1080 int perf_event__process_finished_round(struct perf_tool *tool __maybe_unused, 1081 union perf_event *event __maybe_unused, 1082 struct ordered_events *oe) 1083 { 1084 if (dump_trace) 1085 fprintf(stdout, "\n"); 1086 return ordered_events__flush(oe, OE_FLUSH__ROUND); 1087 } 1088 1089 int perf_session__queue_event(struct perf_session *s, union perf_event *event, 1090 u64 timestamp, u64 file_offset, const char *file_path) 1091 { 1092 return ordered_events__queue(&s->ordered_events, event, timestamp, file_offset, file_path); 1093 } 1094 1095 static void callchain__lbr_callstack_printf(struct perf_sample *sample) 1096 { 1097 struct ip_callchain *callchain = sample->callchain; 1098 struct branch_stack *lbr_stack = sample->branch_stack; 1099 struct branch_entry *entries = perf_sample__branch_entries(sample); 1100 u64 kernel_callchain_nr = callchain->nr; 1101 unsigned int i; 1102 1103 for (i = 0; i < kernel_callchain_nr; i++) { 1104 if (callchain->ips[i] == PERF_CONTEXT_USER) 1105 break; 1106 } 1107 1108 if ((i != kernel_callchain_nr) && lbr_stack->nr) { 1109 u64 total_nr; 1110 /* 1111 * LBR callstack can only get user call chain, 1112 * i is kernel call chain number, 1113 * 1 is PERF_CONTEXT_USER. 1114 * 1115 * The user call chain is stored in LBR registers. 1116 * LBR are pair registers. The caller is stored 1117 * in "from" register, while the callee is stored 1118 * in "to" register. 1119 * For example, there is a call stack 1120 * "A"->"B"->"C"->"D". 1121 * The LBR registers will be recorded like 1122 * "C"->"D", "B"->"C", "A"->"B". 1123 * So only the first "to" register and all "from" 1124 * registers are needed to construct the whole stack. 1125 */ 1126 total_nr = i + 1 + lbr_stack->nr + 1; 1127 kernel_callchain_nr = i + 1; 1128 1129 printf("... LBR call chain: nr:%" PRIu64 "\n", total_nr); 1130 1131 for (i = 0; i < kernel_callchain_nr; i++) 1132 printf("..... %2d: %016" PRIx64 "\n", 1133 i, callchain->ips[i]); 1134 1135 printf("..... %2d: %016" PRIx64 "\n", 1136 (int)(kernel_callchain_nr), entries[0].to); 1137 for (i = 0; i < lbr_stack->nr; i++) 1138 printf("..... %2d: %016" PRIx64 "\n", 1139 (int)(i + kernel_callchain_nr + 1), entries[i].from); 1140 } 1141 } 1142 1143 static void callchain__printf(struct evsel *evsel, 1144 struct perf_sample *sample) 1145 { 1146 unsigned int i; 1147 struct ip_callchain *callchain = sample->callchain; 1148 1149 if (evsel__has_branch_callstack(evsel)) 1150 callchain__lbr_callstack_printf(sample); 1151 1152 printf("... FP chain: nr:%" PRIu64 "\n", callchain->nr); 1153 1154 for (i = 0; i < callchain->nr; i++) 1155 printf("..... %2d: %016" PRIx64 "\n", 1156 i, callchain->ips[i]); 1157 } 1158 1159 static void branch_stack__printf(struct perf_sample *sample, bool callstack) 1160 { 1161 struct branch_entry *entries = perf_sample__branch_entries(sample); 1162 uint64_t i; 1163 1164 if (!callstack) { 1165 printf("%s: nr:%" PRIu64 "\n", "... branch stack", sample->branch_stack->nr); 1166 } else { 1167 /* the reason of adding 1 to nr is because after expanding 1168 * branch stack it generates nr + 1 callstack records. e.g., 1169 * B()->C() 1170 * A()->B() 1171 * the final callstack should be: 1172 * C() 1173 * B() 1174 * A() 1175 */ 1176 printf("%s: nr:%" PRIu64 "\n", "... branch callstack", sample->branch_stack->nr+1); 1177 } 1178 1179 for (i = 0; i < sample->branch_stack->nr; i++) { 1180 struct branch_entry *e = &entries[i]; 1181 1182 if (!callstack) { 1183 printf("..... %2"PRIu64": %016" PRIx64 " -> %016" PRIx64 " %hu cycles %s%s%s%s %x %s\n", 1184 i, e->from, e->to, 1185 (unsigned short)e->flags.cycles, 1186 e->flags.mispred ? "M" : " ", 1187 e->flags.predicted ? "P" : " ", 1188 e->flags.abort ? "A" : " ", 1189 e->flags.in_tx ? "T" : " ", 1190 (unsigned)e->flags.reserved, 1191 get_branch_type(e)); 1192 } else { 1193 if (i == 0) { 1194 printf("..... %2"PRIu64": %016" PRIx64 "\n" 1195 "..... %2"PRIu64": %016" PRIx64 "\n", 1196 i, e->to, i+1, e->from); 1197 } else { 1198 printf("..... %2"PRIu64": %016" PRIx64 "\n", i+1, e->from); 1199 } 1200 } 1201 } 1202 } 1203 1204 static void regs_dump__printf(u64 mask, u64 *regs, const char *arch) 1205 { 1206 unsigned rid, i = 0; 1207 1208 for_each_set_bit(rid, (unsigned long *) &mask, sizeof(mask) * 8) { 1209 u64 val = regs[i++]; 1210 1211 printf(".... %-5s 0x%016" PRIx64 "\n", 1212 perf_reg_name(rid, arch), val); 1213 } 1214 } 1215 1216 static const char *regs_abi[] = { 1217 [PERF_SAMPLE_REGS_ABI_NONE] = "none", 1218 [PERF_SAMPLE_REGS_ABI_32] = "32-bit", 1219 [PERF_SAMPLE_REGS_ABI_64] = "64-bit", 1220 }; 1221 1222 static inline const char *regs_dump_abi(struct regs_dump *d) 1223 { 1224 if (d->abi > PERF_SAMPLE_REGS_ABI_64) 1225 return "unknown"; 1226 1227 return regs_abi[d->abi]; 1228 } 1229 1230 static void regs__printf(const char *type, struct regs_dump *regs, const char *arch) 1231 { 1232 u64 mask = regs->mask; 1233 1234 printf("... %s regs: mask 0x%" PRIx64 " ABI %s\n", 1235 type, 1236 mask, 1237 regs_dump_abi(regs)); 1238 1239 regs_dump__printf(mask, regs->regs, arch); 1240 } 1241 1242 static void regs_user__printf(struct perf_sample *sample, const char *arch) 1243 { 1244 struct regs_dump *user_regs = &sample->user_regs; 1245 1246 if (user_regs->regs) 1247 regs__printf("user", user_regs, arch); 1248 } 1249 1250 static void regs_intr__printf(struct perf_sample *sample, const char *arch) 1251 { 1252 struct regs_dump *intr_regs = &sample->intr_regs; 1253 1254 if (intr_regs->regs) 1255 regs__printf("intr", intr_regs, arch); 1256 } 1257 1258 static void stack_user__printf(struct stack_dump *dump) 1259 { 1260 printf("... ustack: size %" PRIu64 ", offset 0x%x\n", 1261 dump->size, dump->offset); 1262 } 1263 1264 static void evlist__print_tstamp(struct evlist *evlist, union perf_event *event, struct perf_sample *sample) 1265 { 1266 u64 sample_type = __evlist__combined_sample_type(evlist); 1267 1268 if (event->header.type != PERF_RECORD_SAMPLE && 1269 !evlist__sample_id_all(evlist)) { 1270 fputs("-1 -1 ", stdout); 1271 return; 1272 } 1273 1274 if ((sample_type & PERF_SAMPLE_CPU)) 1275 printf("%u ", sample->cpu); 1276 1277 if (sample_type & PERF_SAMPLE_TIME) 1278 printf("%" PRIu64 " ", sample->time); 1279 } 1280 1281 static void sample_read__printf(struct perf_sample *sample, u64 read_format) 1282 { 1283 printf("... sample_read:\n"); 1284 1285 if (read_format & PERF_FORMAT_TOTAL_TIME_ENABLED) 1286 printf("...... time enabled %016" PRIx64 "\n", 1287 sample->read.time_enabled); 1288 1289 if (read_format & PERF_FORMAT_TOTAL_TIME_RUNNING) 1290 printf("...... time running %016" PRIx64 "\n", 1291 sample->read.time_running); 1292 1293 if (read_format & PERF_FORMAT_GROUP) { 1294 struct sample_read_value *value = sample->read.group.values; 1295 1296 printf(".... group nr %" PRIu64 "\n", sample->read.group.nr); 1297 1298 sample_read_group__for_each(value, sample->read.group.nr, read_format) { 1299 printf("..... id %016" PRIx64 1300 ", value %016" PRIx64, 1301 value->id, value->value); 1302 if (read_format & PERF_FORMAT_LOST) 1303 printf(", lost %" PRIu64, value->lost); 1304 printf("\n"); 1305 } 1306 } else { 1307 printf("..... id %016" PRIx64 ", value %016" PRIx64, 1308 sample->read.one.id, sample->read.one.value); 1309 if (read_format & PERF_FORMAT_LOST) 1310 printf(", lost %" PRIu64, sample->read.one.lost); 1311 printf("\n"); 1312 } 1313 } 1314 1315 static void dump_event(struct evlist *evlist, union perf_event *event, 1316 u64 file_offset, struct perf_sample *sample, 1317 const char *file_path) 1318 { 1319 if (!dump_trace) 1320 return; 1321 1322 printf("\n%#" PRIx64 "@%s [%#x]: event: %d\n", 1323 file_offset, file_path, event->header.size, event->header.type); 1324 1325 trace_event(event); 1326 if (event->header.type == PERF_RECORD_SAMPLE && evlist->trace_event_sample_raw) 1327 evlist->trace_event_sample_raw(evlist, event, sample); 1328 1329 if (sample) 1330 evlist__print_tstamp(evlist, event, sample); 1331 1332 printf("%#" PRIx64 " [%#x]: PERF_RECORD_%s", file_offset, 1333 event->header.size, perf_event__name(event->header.type)); 1334 } 1335 1336 char *get_page_size_name(u64 size, char *str) 1337 { 1338 if (!size || !unit_number__scnprintf(str, PAGE_SIZE_NAME_LEN, size)) 1339 snprintf(str, PAGE_SIZE_NAME_LEN, "%s", "N/A"); 1340 1341 return str; 1342 } 1343 1344 static void dump_sample(struct evsel *evsel, union perf_event *event, 1345 struct perf_sample *sample, const char *arch) 1346 { 1347 u64 sample_type; 1348 char str[PAGE_SIZE_NAME_LEN]; 1349 1350 if (!dump_trace) 1351 return; 1352 1353 printf("(IP, 0x%x): %d/%d: %#" PRIx64 " period: %" PRIu64 " addr: %#" PRIx64 "\n", 1354 event->header.misc, sample->pid, sample->tid, sample->ip, 1355 sample->period, sample->addr); 1356 1357 sample_type = evsel->core.attr.sample_type; 1358 1359 if (evsel__has_callchain(evsel)) 1360 callchain__printf(evsel, sample); 1361 1362 if (evsel__has_br_stack(evsel)) 1363 branch_stack__printf(sample, evsel__has_branch_callstack(evsel)); 1364 1365 if (sample_type & PERF_SAMPLE_REGS_USER) 1366 regs_user__printf(sample, arch); 1367 1368 if (sample_type & PERF_SAMPLE_REGS_INTR) 1369 regs_intr__printf(sample, arch); 1370 1371 if (sample_type & PERF_SAMPLE_STACK_USER) 1372 stack_user__printf(&sample->user_stack); 1373 1374 if (sample_type & PERF_SAMPLE_WEIGHT_TYPE) { 1375 printf("... weight: %" PRIu64 "", sample->weight); 1376 if (sample_type & PERF_SAMPLE_WEIGHT_STRUCT) { 1377 printf(",0x%"PRIx16"", sample->ins_lat); 1378 printf(",0x%"PRIx16"", sample->p_stage_cyc); 1379 } 1380 printf("\n"); 1381 } 1382 1383 if (sample_type & PERF_SAMPLE_DATA_SRC) 1384 printf(" . data_src: 0x%"PRIx64"\n", sample->data_src); 1385 1386 if (sample_type & PERF_SAMPLE_PHYS_ADDR) 1387 printf(" .. phys_addr: 0x%"PRIx64"\n", sample->phys_addr); 1388 1389 if (sample_type & PERF_SAMPLE_DATA_PAGE_SIZE) 1390 printf(" .. data page size: %s\n", get_page_size_name(sample->data_page_size, str)); 1391 1392 if (sample_type & PERF_SAMPLE_CODE_PAGE_SIZE) 1393 printf(" .. code page size: %s\n", get_page_size_name(sample->code_page_size, str)); 1394 1395 if (sample_type & PERF_SAMPLE_TRANSACTION) 1396 printf("... transaction: %" PRIx64 "\n", sample->transaction); 1397 1398 if (sample_type & PERF_SAMPLE_READ) 1399 sample_read__printf(sample, evsel->core.attr.read_format); 1400 } 1401 1402 static void dump_read(struct evsel *evsel, union perf_event *event) 1403 { 1404 struct perf_record_read *read_event = &event->read; 1405 u64 read_format; 1406 1407 if (!dump_trace) 1408 return; 1409 1410 printf(": %d %d %s %" PRI_lu64 "\n", event->read.pid, event->read.tid, 1411 evsel__name(evsel), event->read.value); 1412 1413 if (!evsel) 1414 return; 1415 1416 read_format = evsel->core.attr.read_format; 1417 1418 if (read_format & PERF_FORMAT_TOTAL_TIME_ENABLED) 1419 printf("... time enabled : %" PRI_lu64 "\n", read_event->time_enabled); 1420 1421 if (read_format & PERF_FORMAT_TOTAL_TIME_RUNNING) 1422 printf("... time running : %" PRI_lu64 "\n", read_event->time_running); 1423 1424 if (read_format & PERF_FORMAT_ID) 1425 printf("... id : %" PRI_lu64 "\n", read_event->id); 1426 1427 if (read_format & PERF_FORMAT_LOST) 1428 printf("... lost : %" PRI_lu64 "\n", read_event->lost); 1429 } 1430 1431 static struct machine *machines__find_for_cpumode(struct machines *machines, 1432 union perf_event *event, 1433 struct perf_sample *sample) 1434 { 1435 if (perf_guest && 1436 ((sample->cpumode == PERF_RECORD_MISC_GUEST_KERNEL) || 1437 (sample->cpumode == PERF_RECORD_MISC_GUEST_USER))) { 1438 u32 pid; 1439 1440 if (sample->machine_pid) 1441 pid = sample->machine_pid; 1442 else if (event->header.type == PERF_RECORD_MMAP 1443 || event->header.type == PERF_RECORD_MMAP2) 1444 pid = event->mmap.pid; 1445 else 1446 pid = sample->pid; 1447 1448 /* 1449 * Guest code machine is created as needed and does not use 1450 * DEFAULT_GUEST_KERNEL_ID. 1451 */ 1452 if (symbol_conf.guest_code) 1453 return machines__findnew(machines, pid); 1454 1455 return machines__find_guest(machines, pid); 1456 } 1457 1458 return &machines->host; 1459 } 1460 1461 static int deliver_sample_value(struct evlist *evlist, 1462 struct perf_tool *tool, 1463 union perf_event *event, 1464 struct perf_sample *sample, 1465 struct sample_read_value *v, 1466 struct machine *machine) 1467 { 1468 struct perf_sample_id *sid = evlist__id2sid(evlist, v->id); 1469 struct evsel *evsel; 1470 1471 if (sid) { 1472 sample->id = v->id; 1473 sample->period = v->value - sid->period; 1474 sid->period = v->value; 1475 } 1476 1477 if (!sid || sid->evsel == NULL) { 1478 ++evlist->stats.nr_unknown_id; 1479 return 0; 1480 } 1481 1482 /* 1483 * There's no reason to deliver sample 1484 * for zero period, bail out. 1485 */ 1486 if (!sample->period) 1487 return 0; 1488 1489 evsel = container_of(sid->evsel, struct evsel, core); 1490 return tool->sample(tool, event, sample, evsel, machine); 1491 } 1492 1493 static int deliver_sample_group(struct evlist *evlist, 1494 struct perf_tool *tool, 1495 union perf_event *event, 1496 struct perf_sample *sample, 1497 struct machine *machine, 1498 u64 read_format) 1499 { 1500 int ret = -EINVAL; 1501 struct sample_read_value *v = sample->read.group.values; 1502 1503 sample_read_group__for_each(v, sample->read.group.nr, read_format) { 1504 ret = deliver_sample_value(evlist, tool, event, sample, v, 1505 machine); 1506 if (ret) 1507 break; 1508 } 1509 1510 return ret; 1511 } 1512 1513 static int evlist__deliver_sample(struct evlist *evlist, struct perf_tool *tool, 1514 union perf_event *event, struct perf_sample *sample, 1515 struct evsel *evsel, struct machine *machine) 1516 { 1517 /* We know evsel != NULL. */ 1518 u64 sample_type = evsel->core.attr.sample_type; 1519 u64 read_format = evsel->core.attr.read_format; 1520 1521 /* Standard sample delivery. */ 1522 if (!(sample_type & PERF_SAMPLE_READ)) 1523 return tool->sample(tool, event, sample, evsel, machine); 1524 1525 /* For PERF_SAMPLE_READ we have either single or group mode. */ 1526 if (read_format & PERF_FORMAT_GROUP) 1527 return deliver_sample_group(evlist, tool, event, sample, 1528 machine, read_format); 1529 else 1530 return deliver_sample_value(evlist, tool, event, sample, 1531 &sample->read.one, machine); 1532 } 1533 1534 static int machines__deliver_event(struct machines *machines, 1535 struct evlist *evlist, 1536 union perf_event *event, 1537 struct perf_sample *sample, 1538 struct perf_tool *tool, u64 file_offset, 1539 const char *file_path) 1540 { 1541 struct evsel *evsel; 1542 struct machine *machine; 1543 1544 dump_event(evlist, event, file_offset, sample, file_path); 1545 1546 evsel = evlist__id2evsel(evlist, sample->id); 1547 1548 machine = machines__find_for_cpumode(machines, event, sample); 1549 1550 switch (event->header.type) { 1551 case PERF_RECORD_SAMPLE: 1552 if (evsel == NULL) { 1553 ++evlist->stats.nr_unknown_id; 1554 return 0; 1555 } 1556 if (machine == NULL) { 1557 ++evlist->stats.nr_unprocessable_samples; 1558 dump_sample(evsel, event, sample, perf_env__arch(NULL)); 1559 return 0; 1560 } 1561 dump_sample(evsel, event, sample, perf_env__arch(machine->env)); 1562 return evlist__deliver_sample(evlist, tool, event, sample, evsel, machine); 1563 case PERF_RECORD_MMAP: 1564 return tool->mmap(tool, event, sample, machine); 1565 case PERF_RECORD_MMAP2: 1566 if (event->header.misc & PERF_RECORD_MISC_PROC_MAP_PARSE_TIMEOUT) 1567 ++evlist->stats.nr_proc_map_timeout; 1568 return tool->mmap2(tool, event, sample, machine); 1569 case PERF_RECORD_COMM: 1570 return tool->comm(tool, event, sample, machine); 1571 case PERF_RECORD_NAMESPACES: 1572 return tool->namespaces(tool, event, sample, machine); 1573 case PERF_RECORD_CGROUP: 1574 return tool->cgroup(tool, event, sample, machine); 1575 case PERF_RECORD_FORK: 1576 return tool->fork(tool, event, sample, machine); 1577 case PERF_RECORD_EXIT: 1578 return tool->exit(tool, event, sample, machine); 1579 case PERF_RECORD_LOST: 1580 if (tool->lost == perf_event__process_lost) 1581 evlist->stats.total_lost += event->lost.lost; 1582 return tool->lost(tool, event, sample, machine); 1583 case PERF_RECORD_LOST_SAMPLES: 1584 if (tool->lost_samples == perf_event__process_lost_samples) 1585 evlist->stats.total_lost_samples += event->lost_samples.lost; 1586 return tool->lost_samples(tool, event, sample, machine); 1587 case PERF_RECORD_READ: 1588 dump_read(evsel, event); 1589 return tool->read(tool, event, sample, evsel, machine); 1590 case PERF_RECORD_THROTTLE: 1591 return tool->throttle(tool, event, sample, machine); 1592 case PERF_RECORD_UNTHROTTLE: 1593 return tool->unthrottle(tool, event, sample, machine); 1594 case PERF_RECORD_AUX: 1595 if (tool->aux == perf_event__process_aux) { 1596 if (event->aux.flags & PERF_AUX_FLAG_TRUNCATED) 1597 evlist->stats.total_aux_lost += 1; 1598 if (event->aux.flags & PERF_AUX_FLAG_PARTIAL) 1599 evlist->stats.total_aux_partial += 1; 1600 if (event->aux.flags & PERF_AUX_FLAG_COLLISION) 1601 evlist->stats.total_aux_collision += 1; 1602 } 1603 return tool->aux(tool, event, sample, machine); 1604 case PERF_RECORD_ITRACE_START: 1605 return tool->itrace_start(tool, event, sample, machine); 1606 case PERF_RECORD_SWITCH: 1607 case PERF_RECORD_SWITCH_CPU_WIDE: 1608 return tool->context_switch(tool, event, sample, machine); 1609 case PERF_RECORD_KSYMBOL: 1610 return tool->ksymbol(tool, event, sample, machine); 1611 case PERF_RECORD_BPF_EVENT: 1612 return tool->bpf(tool, event, sample, machine); 1613 case PERF_RECORD_TEXT_POKE: 1614 return tool->text_poke(tool, event, sample, machine); 1615 case PERF_RECORD_AUX_OUTPUT_HW_ID: 1616 return tool->aux_output_hw_id(tool, event, sample, machine); 1617 default: 1618 ++evlist->stats.nr_unknown_events; 1619 return -1; 1620 } 1621 } 1622 1623 static int perf_session__deliver_event(struct perf_session *session, 1624 union perf_event *event, 1625 struct perf_tool *tool, 1626 u64 file_offset, 1627 const char *file_path) 1628 { 1629 struct perf_sample sample; 1630 int ret = evlist__parse_sample(session->evlist, event, &sample); 1631 1632 if (ret) { 1633 pr_err("Can't parse sample, err = %d\n", ret); 1634 return ret; 1635 } 1636 1637 ret = auxtrace__process_event(session, event, &sample, tool); 1638 if (ret < 0) 1639 return ret; 1640 if (ret > 0) 1641 return 0; 1642 1643 ret = machines__deliver_event(&session->machines, session->evlist, 1644 event, &sample, tool, file_offset, file_path); 1645 1646 if (dump_trace && sample.aux_sample.size) 1647 auxtrace__dump_auxtrace_sample(session, &sample); 1648 1649 return ret; 1650 } 1651 1652 static s64 perf_session__process_user_event(struct perf_session *session, 1653 union perf_event *event, 1654 u64 file_offset, 1655 const char *file_path) 1656 { 1657 struct ordered_events *oe = &session->ordered_events; 1658 struct perf_tool *tool = session->tool; 1659 struct perf_sample sample = { .time = 0, }; 1660 int fd = perf_data__fd(session->data); 1661 int err; 1662 1663 if (event->header.type != PERF_RECORD_COMPRESSED || 1664 tool->compressed == perf_session__process_compressed_event_stub) 1665 dump_event(session->evlist, event, file_offset, &sample, file_path); 1666 1667 /* These events are processed right away */ 1668 switch (event->header.type) { 1669 case PERF_RECORD_HEADER_ATTR: 1670 err = tool->attr(tool, event, &session->evlist); 1671 if (err == 0) { 1672 perf_session__set_id_hdr_size(session); 1673 perf_session__set_comm_exec(session); 1674 } 1675 return err; 1676 case PERF_RECORD_EVENT_UPDATE: 1677 return tool->event_update(tool, event, &session->evlist); 1678 case PERF_RECORD_HEADER_EVENT_TYPE: 1679 /* 1680 * Deprecated, but we need to handle it for sake 1681 * of old data files create in pipe mode. 1682 */ 1683 return 0; 1684 case PERF_RECORD_HEADER_TRACING_DATA: 1685 /* 1686 * Setup for reading amidst mmap, but only when we 1687 * are in 'file' mode. The 'pipe' fd is in proper 1688 * place already. 1689 */ 1690 if (!perf_data__is_pipe(session->data)) 1691 lseek(fd, file_offset, SEEK_SET); 1692 return tool->tracing_data(session, event); 1693 case PERF_RECORD_HEADER_BUILD_ID: 1694 return tool->build_id(session, event); 1695 case PERF_RECORD_FINISHED_ROUND: 1696 return tool->finished_round(tool, event, oe); 1697 case PERF_RECORD_ID_INDEX: 1698 return tool->id_index(session, event); 1699 case PERF_RECORD_AUXTRACE_INFO: 1700 return tool->auxtrace_info(session, event); 1701 case PERF_RECORD_AUXTRACE: 1702 /* 1703 * Setup for reading amidst mmap, but only when we 1704 * are in 'file' mode. The 'pipe' fd is in proper 1705 * place already. 1706 */ 1707 if (!perf_data__is_pipe(session->data)) 1708 lseek(fd, file_offset + event->header.size, SEEK_SET); 1709 return tool->auxtrace(session, event); 1710 case PERF_RECORD_AUXTRACE_ERROR: 1711 perf_session__auxtrace_error_inc(session, event); 1712 return tool->auxtrace_error(session, event); 1713 case PERF_RECORD_THREAD_MAP: 1714 return tool->thread_map(session, event); 1715 case PERF_RECORD_CPU_MAP: 1716 return tool->cpu_map(session, event); 1717 case PERF_RECORD_STAT_CONFIG: 1718 return tool->stat_config(session, event); 1719 case PERF_RECORD_STAT: 1720 return tool->stat(session, event); 1721 case PERF_RECORD_STAT_ROUND: 1722 return tool->stat_round(session, event); 1723 case PERF_RECORD_TIME_CONV: 1724 session->time_conv = event->time_conv; 1725 return tool->time_conv(session, event); 1726 case PERF_RECORD_HEADER_FEATURE: 1727 return tool->feature(session, event); 1728 case PERF_RECORD_COMPRESSED: 1729 err = tool->compressed(session, event, file_offset, file_path); 1730 if (err) 1731 dump_event(session->evlist, event, file_offset, &sample, file_path); 1732 return err; 1733 case PERF_RECORD_FINISHED_INIT: 1734 return tool->finished_init(session, event); 1735 default: 1736 return -EINVAL; 1737 } 1738 } 1739 1740 int perf_session__deliver_synth_event(struct perf_session *session, 1741 union perf_event *event, 1742 struct perf_sample *sample) 1743 { 1744 struct evlist *evlist = session->evlist; 1745 struct perf_tool *tool = session->tool; 1746 1747 events_stats__inc(&evlist->stats, event->header.type); 1748 1749 if (event->header.type >= PERF_RECORD_USER_TYPE_START) 1750 return perf_session__process_user_event(session, event, 0, NULL); 1751 1752 return machines__deliver_event(&session->machines, evlist, event, sample, tool, 0, NULL); 1753 } 1754 1755 static void event_swap(union perf_event *event, bool sample_id_all) 1756 { 1757 perf_event__swap_op swap; 1758 1759 swap = perf_event__swap_ops[event->header.type]; 1760 if (swap) 1761 swap(event, sample_id_all); 1762 } 1763 1764 int perf_session__peek_event(struct perf_session *session, off_t file_offset, 1765 void *buf, size_t buf_sz, 1766 union perf_event **event_ptr, 1767 struct perf_sample *sample) 1768 { 1769 union perf_event *event; 1770 size_t hdr_sz, rest; 1771 int fd; 1772 1773 if (session->one_mmap && !session->header.needs_swap) { 1774 event = file_offset - session->one_mmap_offset + 1775 session->one_mmap_addr; 1776 goto out_parse_sample; 1777 } 1778 1779 if (perf_data__is_pipe(session->data)) 1780 return -1; 1781 1782 fd = perf_data__fd(session->data); 1783 hdr_sz = sizeof(struct perf_event_header); 1784 1785 if (buf_sz < hdr_sz) 1786 return -1; 1787 1788 if (lseek(fd, file_offset, SEEK_SET) == (off_t)-1 || 1789 readn(fd, buf, hdr_sz) != (ssize_t)hdr_sz) 1790 return -1; 1791 1792 event = (union perf_event *)buf; 1793 1794 if (session->header.needs_swap) 1795 perf_event_header__bswap(&event->header); 1796 1797 if (event->header.size < hdr_sz || event->header.size > buf_sz) 1798 return -1; 1799 1800 buf += hdr_sz; 1801 rest = event->header.size - hdr_sz; 1802 1803 if (readn(fd, buf, rest) != (ssize_t)rest) 1804 return -1; 1805 1806 if (session->header.needs_swap) 1807 event_swap(event, evlist__sample_id_all(session->evlist)); 1808 1809 out_parse_sample: 1810 1811 if (sample && event->header.type < PERF_RECORD_USER_TYPE_START && 1812 evlist__parse_sample(session->evlist, event, sample)) 1813 return -1; 1814 1815 *event_ptr = event; 1816 1817 return 0; 1818 } 1819 1820 int perf_session__peek_events(struct perf_session *session, u64 offset, 1821 u64 size, peek_events_cb_t cb, void *data) 1822 { 1823 u64 max_offset = offset + size; 1824 char buf[PERF_SAMPLE_MAX_SIZE]; 1825 union perf_event *event; 1826 int err; 1827 1828 do { 1829 err = perf_session__peek_event(session, offset, buf, 1830 PERF_SAMPLE_MAX_SIZE, &event, 1831 NULL); 1832 if (err) 1833 return err; 1834 1835 err = cb(session, event, offset, data); 1836 if (err) 1837 return err; 1838 1839 offset += event->header.size; 1840 if (event->header.type == PERF_RECORD_AUXTRACE) 1841 offset += event->auxtrace.size; 1842 1843 } while (offset < max_offset); 1844 1845 return err; 1846 } 1847 1848 static s64 perf_session__process_event(struct perf_session *session, 1849 union perf_event *event, u64 file_offset, 1850 const char *file_path) 1851 { 1852 struct evlist *evlist = session->evlist; 1853 struct perf_tool *tool = session->tool; 1854 int ret; 1855 1856 if (session->header.needs_swap) 1857 event_swap(event, evlist__sample_id_all(evlist)); 1858 1859 if (event->header.type >= PERF_RECORD_HEADER_MAX) 1860 return -EINVAL; 1861 1862 events_stats__inc(&evlist->stats, event->header.type); 1863 1864 if (event->header.type >= PERF_RECORD_USER_TYPE_START) 1865 return perf_session__process_user_event(session, event, file_offset, file_path); 1866 1867 if (tool->ordered_events) { 1868 u64 timestamp = -1ULL; 1869 1870 ret = evlist__parse_sample_timestamp(evlist, event, ×tamp); 1871 if (ret && ret != -1) 1872 return ret; 1873 1874 ret = perf_session__queue_event(session, event, timestamp, file_offset, file_path); 1875 if (ret != -ETIME) 1876 return ret; 1877 } 1878 1879 return perf_session__deliver_event(session, event, tool, file_offset, file_path); 1880 } 1881 1882 void perf_event_header__bswap(struct perf_event_header *hdr) 1883 { 1884 hdr->type = bswap_32(hdr->type); 1885 hdr->misc = bswap_16(hdr->misc); 1886 hdr->size = bswap_16(hdr->size); 1887 } 1888 1889 struct thread *perf_session__findnew(struct perf_session *session, pid_t pid) 1890 { 1891 return machine__findnew_thread(&session->machines.host, -1, pid); 1892 } 1893 1894 int perf_session__register_idle_thread(struct perf_session *session) 1895 { 1896 struct thread *thread = machine__idle_thread(&session->machines.host); 1897 1898 /* machine__idle_thread() got the thread, so put it */ 1899 thread__put(thread); 1900 return thread ? 0 : -1; 1901 } 1902 1903 static void 1904 perf_session__warn_order(const struct perf_session *session) 1905 { 1906 const struct ordered_events *oe = &session->ordered_events; 1907 struct evsel *evsel; 1908 bool should_warn = true; 1909 1910 evlist__for_each_entry(session->evlist, evsel) { 1911 if (evsel->core.attr.write_backward) 1912 should_warn = false; 1913 } 1914 1915 if (!should_warn) 1916 return; 1917 if (oe->nr_unordered_events != 0) 1918 ui__warning("%u out of order events recorded.\n", oe->nr_unordered_events); 1919 } 1920 1921 static void perf_session__warn_about_errors(const struct perf_session *session) 1922 { 1923 const struct events_stats *stats = &session->evlist->stats; 1924 1925 if (session->tool->lost == perf_event__process_lost && 1926 stats->nr_events[PERF_RECORD_LOST] != 0) { 1927 ui__warning("Processed %d events and lost %d chunks!\n\n" 1928 "Check IO/CPU overload!\n\n", 1929 stats->nr_events[0], 1930 stats->nr_events[PERF_RECORD_LOST]); 1931 } 1932 1933 if (session->tool->lost_samples == perf_event__process_lost_samples) { 1934 double drop_rate; 1935 1936 drop_rate = (double)stats->total_lost_samples / 1937 (double) (stats->nr_events[PERF_RECORD_SAMPLE] + stats->total_lost_samples); 1938 if (drop_rate > 0.05) { 1939 ui__warning("Processed %" PRIu64 " samples and lost %3.2f%%!\n\n", 1940 stats->nr_events[PERF_RECORD_SAMPLE] + stats->total_lost_samples, 1941 drop_rate * 100.0); 1942 } 1943 } 1944 1945 if (session->tool->aux == perf_event__process_aux && 1946 stats->total_aux_lost != 0) { 1947 ui__warning("AUX data lost %" PRIu64 " times out of %u!\n\n", 1948 stats->total_aux_lost, 1949 stats->nr_events[PERF_RECORD_AUX]); 1950 } 1951 1952 if (session->tool->aux == perf_event__process_aux && 1953 stats->total_aux_partial != 0) { 1954 bool vmm_exclusive = false; 1955 1956 (void)sysfs__read_bool("module/kvm_intel/parameters/vmm_exclusive", 1957 &vmm_exclusive); 1958 1959 ui__warning("AUX data had gaps in it %" PRIu64 " times out of %u!\n\n" 1960 "Are you running a KVM guest in the background?%s\n\n", 1961 stats->total_aux_partial, 1962 stats->nr_events[PERF_RECORD_AUX], 1963 vmm_exclusive ? 1964 "\nReloading kvm_intel module with vmm_exclusive=0\n" 1965 "will reduce the gaps to only guest's timeslices." : 1966 ""); 1967 } 1968 1969 if (session->tool->aux == perf_event__process_aux && 1970 stats->total_aux_collision != 0) { 1971 ui__warning("AUX data detected collision %" PRIu64 " times out of %u!\n\n", 1972 stats->total_aux_collision, 1973 stats->nr_events[PERF_RECORD_AUX]); 1974 } 1975 1976 if (stats->nr_unknown_events != 0) { 1977 ui__warning("Found %u unknown events!\n\n" 1978 "Is this an older tool processing a perf.data " 1979 "file generated by a more recent tool?\n\n" 1980 "If that is not the case, consider " 1981 "reporting to linux-kernel@vger.kernel.org.\n\n", 1982 stats->nr_unknown_events); 1983 } 1984 1985 if (stats->nr_unknown_id != 0) { 1986 ui__warning("%u samples with id not present in the header\n", 1987 stats->nr_unknown_id); 1988 } 1989 1990 if (stats->nr_invalid_chains != 0) { 1991 ui__warning("Found invalid callchains!\n\n" 1992 "%u out of %u events were discarded for this reason.\n\n" 1993 "Consider reporting to linux-kernel@vger.kernel.org.\n\n", 1994 stats->nr_invalid_chains, 1995 stats->nr_events[PERF_RECORD_SAMPLE]); 1996 } 1997 1998 if (stats->nr_unprocessable_samples != 0) { 1999 ui__warning("%u unprocessable samples recorded.\n" 2000 "Do you have a KVM guest running and not using 'perf kvm'?\n", 2001 stats->nr_unprocessable_samples); 2002 } 2003 2004 perf_session__warn_order(session); 2005 2006 events_stats__auxtrace_error_warn(stats); 2007 2008 if (stats->nr_proc_map_timeout != 0) { 2009 ui__warning("%d map information files for pre-existing threads were\n" 2010 "not processed, if there are samples for addresses they\n" 2011 "will not be resolved, you may find out which are these\n" 2012 "threads by running with -v and redirecting the output\n" 2013 "to a file.\n" 2014 "The time limit to process proc map is too short?\n" 2015 "Increase it by --proc-map-timeout\n", 2016 stats->nr_proc_map_timeout); 2017 } 2018 } 2019 2020 static int perf_session__flush_thread_stack(struct thread *thread, 2021 void *p __maybe_unused) 2022 { 2023 return thread_stack__flush(thread); 2024 } 2025 2026 static int perf_session__flush_thread_stacks(struct perf_session *session) 2027 { 2028 return machines__for_each_thread(&session->machines, 2029 perf_session__flush_thread_stack, 2030 NULL); 2031 } 2032 2033 volatile sig_atomic_t session_done; 2034 2035 static int __perf_session__process_decomp_events(struct perf_session *session); 2036 2037 static int __perf_session__process_pipe_events(struct perf_session *session) 2038 { 2039 struct ordered_events *oe = &session->ordered_events; 2040 struct perf_tool *tool = session->tool; 2041 union perf_event *event; 2042 uint32_t size, cur_size = 0; 2043 void *buf = NULL; 2044 s64 skip = 0; 2045 u64 head; 2046 ssize_t err; 2047 void *p; 2048 2049 perf_tool__fill_defaults(tool); 2050 2051 head = 0; 2052 cur_size = sizeof(union perf_event); 2053 2054 buf = malloc(cur_size); 2055 if (!buf) 2056 return -errno; 2057 ordered_events__set_copy_on_queue(oe, true); 2058 more: 2059 event = buf; 2060 err = perf_data__read(session->data, event, 2061 sizeof(struct perf_event_header)); 2062 if (err <= 0) { 2063 if (err == 0) 2064 goto done; 2065 2066 pr_err("failed to read event header\n"); 2067 goto out_err; 2068 } 2069 2070 if (session->header.needs_swap) 2071 perf_event_header__bswap(&event->header); 2072 2073 size = event->header.size; 2074 if (size < sizeof(struct perf_event_header)) { 2075 pr_err("bad event header size\n"); 2076 goto out_err; 2077 } 2078 2079 if (size > cur_size) { 2080 void *new = realloc(buf, size); 2081 if (!new) { 2082 pr_err("failed to allocate memory to read event\n"); 2083 goto out_err; 2084 } 2085 buf = new; 2086 cur_size = size; 2087 event = buf; 2088 } 2089 p = event; 2090 p += sizeof(struct perf_event_header); 2091 2092 if (size - sizeof(struct perf_event_header)) { 2093 err = perf_data__read(session->data, p, 2094 size - sizeof(struct perf_event_header)); 2095 if (err <= 0) { 2096 if (err == 0) { 2097 pr_err("unexpected end of event stream\n"); 2098 goto done; 2099 } 2100 2101 pr_err("failed to read event data\n"); 2102 goto out_err; 2103 } 2104 } 2105 2106 if ((skip = perf_session__process_event(session, event, head, "pipe")) < 0) { 2107 pr_err("%#" PRIx64 " [%#x]: failed to process type: %d\n", 2108 head, event->header.size, event->header.type); 2109 err = -EINVAL; 2110 goto out_err; 2111 } 2112 2113 head += size; 2114 2115 if (skip > 0) 2116 head += skip; 2117 2118 err = __perf_session__process_decomp_events(session); 2119 if (err) 2120 goto out_err; 2121 2122 if (!session_done()) 2123 goto more; 2124 done: 2125 /* do the final flush for ordered samples */ 2126 err = ordered_events__flush(oe, OE_FLUSH__FINAL); 2127 if (err) 2128 goto out_err; 2129 err = auxtrace__flush_events(session, tool); 2130 if (err) 2131 goto out_err; 2132 err = perf_session__flush_thread_stacks(session); 2133 out_err: 2134 free(buf); 2135 if (!tool->no_warn) 2136 perf_session__warn_about_errors(session); 2137 ordered_events__free(&session->ordered_events); 2138 auxtrace__free_events(session); 2139 return err; 2140 } 2141 2142 static union perf_event * 2143 prefetch_event(char *buf, u64 head, size_t mmap_size, 2144 bool needs_swap, union perf_event *error) 2145 { 2146 union perf_event *event; 2147 u16 event_size; 2148 2149 /* 2150 * Ensure we have enough space remaining to read 2151 * the size of the event in the headers. 2152 */ 2153 if (head + sizeof(event->header) > mmap_size) 2154 return NULL; 2155 2156 event = (union perf_event *)(buf + head); 2157 if (needs_swap) 2158 perf_event_header__bswap(&event->header); 2159 2160 event_size = event->header.size; 2161 if (head + event_size <= mmap_size) 2162 return event; 2163 2164 /* We're not fetching the event so swap back again */ 2165 if (needs_swap) 2166 perf_event_header__bswap(&event->header); 2167 2168 /* Check if the event fits into the next mmapped buf. */ 2169 if (event_size <= mmap_size - head % page_size) { 2170 /* Remap buf and fetch again. */ 2171 return NULL; 2172 } 2173 2174 /* Invalid input. Event size should never exceed mmap_size. */ 2175 pr_debug("%s: head=%#" PRIx64 " event->header.size=%#x, mmap_size=%#zx:" 2176 " fuzzed or compressed perf.data?\n", __func__, head, event_size, mmap_size); 2177 2178 return error; 2179 } 2180 2181 static union perf_event * 2182 fetch_mmaped_event(u64 head, size_t mmap_size, char *buf, bool needs_swap) 2183 { 2184 return prefetch_event(buf, head, mmap_size, needs_swap, ERR_PTR(-EINVAL)); 2185 } 2186 2187 static union perf_event * 2188 fetch_decomp_event(u64 head, size_t mmap_size, char *buf, bool needs_swap) 2189 { 2190 return prefetch_event(buf, head, mmap_size, needs_swap, NULL); 2191 } 2192 2193 static int __perf_session__process_decomp_events(struct perf_session *session) 2194 { 2195 s64 skip; 2196 u64 size; 2197 struct decomp *decomp = session->active_decomp->decomp_last; 2198 2199 if (!decomp) 2200 return 0; 2201 2202 while (decomp->head < decomp->size && !session_done()) { 2203 union perf_event *event = fetch_decomp_event(decomp->head, decomp->size, decomp->data, 2204 session->header.needs_swap); 2205 2206 if (!event) 2207 break; 2208 2209 size = event->header.size; 2210 2211 if (size < sizeof(struct perf_event_header) || 2212 (skip = perf_session__process_event(session, event, decomp->file_pos, 2213 decomp->file_path)) < 0) { 2214 pr_err("%#" PRIx64 " [%#x]: failed to process type: %d\n", 2215 decomp->file_pos + decomp->head, event->header.size, event->header.type); 2216 return -EINVAL; 2217 } 2218 2219 if (skip) 2220 size += skip; 2221 2222 decomp->head += size; 2223 } 2224 2225 return 0; 2226 } 2227 2228 /* 2229 * On 64bit we can mmap the data file in one go. No need for tiny mmap 2230 * slices. On 32bit we use 32MB. 2231 */ 2232 #if BITS_PER_LONG == 64 2233 #define MMAP_SIZE ULLONG_MAX 2234 #define NUM_MMAPS 1 2235 #else 2236 #define MMAP_SIZE (32 * 1024 * 1024ULL) 2237 #define NUM_MMAPS 128 2238 #endif 2239 2240 struct reader; 2241 2242 typedef s64 (*reader_cb_t)(struct perf_session *session, 2243 union perf_event *event, 2244 u64 file_offset, 2245 const char *file_path); 2246 2247 struct reader { 2248 int fd; 2249 const char *path; 2250 u64 data_size; 2251 u64 data_offset; 2252 reader_cb_t process; 2253 bool in_place_update; 2254 char *mmaps[NUM_MMAPS]; 2255 size_t mmap_size; 2256 int mmap_idx; 2257 char *mmap_cur; 2258 u64 file_pos; 2259 u64 file_offset; 2260 u64 head; 2261 u64 size; 2262 bool done; 2263 struct zstd_data zstd_data; 2264 struct decomp_data decomp_data; 2265 }; 2266 2267 static int 2268 reader__init(struct reader *rd, bool *one_mmap) 2269 { 2270 u64 data_size = rd->data_size; 2271 char **mmaps = rd->mmaps; 2272 2273 rd->head = rd->data_offset; 2274 data_size += rd->data_offset; 2275 2276 rd->mmap_size = MMAP_SIZE; 2277 if (rd->mmap_size > data_size) { 2278 rd->mmap_size = data_size; 2279 if (one_mmap) 2280 *one_mmap = true; 2281 } 2282 2283 memset(mmaps, 0, sizeof(rd->mmaps)); 2284 2285 if (zstd_init(&rd->zstd_data, 0)) 2286 return -1; 2287 rd->decomp_data.zstd_decomp = &rd->zstd_data; 2288 2289 return 0; 2290 } 2291 2292 static void 2293 reader__release_decomp(struct reader *rd) 2294 { 2295 perf_decomp__release_events(rd->decomp_data.decomp); 2296 zstd_fini(&rd->zstd_data); 2297 } 2298 2299 static int 2300 reader__mmap(struct reader *rd, struct perf_session *session) 2301 { 2302 int mmap_prot, mmap_flags; 2303 char *buf, **mmaps = rd->mmaps; 2304 u64 page_offset; 2305 2306 mmap_prot = PROT_READ; 2307 mmap_flags = MAP_SHARED; 2308 2309 if (rd->in_place_update) { 2310 mmap_prot |= PROT_WRITE; 2311 } else if (session->header.needs_swap) { 2312 mmap_prot |= PROT_WRITE; 2313 mmap_flags = MAP_PRIVATE; 2314 } 2315 2316 if (mmaps[rd->mmap_idx]) { 2317 munmap(mmaps[rd->mmap_idx], rd->mmap_size); 2318 mmaps[rd->mmap_idx] = NULL; 2319 } 2320 2321 page_offset = page_size * (rd->head / page_size); 2322 rd->file_offset += page_offset; 2323 rd->head -= page_offset; 2324 2325 buf = mmap(NULL, rd->mmap_size, mmap_prot, mmap_flags, rd->fd, 2326 rd->file_offset); 2327 if (buf == MAP_FAILED) { 2328 pr_err("failed to mmap file\n"); 2329 return -errno; 2330 } 2331 mmaps[rd->mmap_idx] = rd->mmap_cur = buf; 2332 rd->mmap_idx = (rd->mmap_idx + 1) & (ARRAY_SIZE(rd->mmaps) - 1); 2333 rd->file_pos = rd->file_offset + rd->head; 2334 if (session->one_mmap) { 2335 session->one_mmap_addr = buf; 2336 session->one_mmap_offset = rd->file_offset; 2337 } 2338 2339 return 0; 2340 } 2341 2342 enum { 2343 READER_OK, 2344 READER_NODATA, 2345 }; 2346 2347 static int 2348 reader__read_event(struct reader *rd, struct perf_session *session, 2349 struct ui_progress *prog) 2350 { 2351 u64 size; 2352 int err = READER_OK; 2353 union perf_event *event; 2354 s64 skip; 2355 2356 event = fetch_mmaped_event(rd->head, rd->mmap_size, rd->mmap_cur, 2357 session->header.needs_swap); 2358 if (IS_ERR(event)) 2359 return PTR_ERR(event); 2360 2361 if (!event) 2362 return READER_NODATA; 2363 2364 size = event->header.size; 2365 2366 skip = -EINVAL; 2367 2368 if (size < sizeof(struct perf_event_header) || 2369 (skip = rd->process(session, event, rd->file_pos, rd->path)) < 0) { 2370 pr_err("%#" PRIx64 " [%#x]: failed to process type: %d [%s]\n", 2371 rd->file_offset + rd->head, event->header.size, 2372 event->header.type, strerror(-skip)); 2373 err = skip; 2374 goto out; 2375 } 2376 2377 if (skip) 2378 size += skip; 2379 2380 rd->size += size; 2381 rd->head += size; 2382 rd->file_pos += size; 2383 2384 err = __perf_session__process_decomp_events(session); 2385 if (err) 2386 goto out; 2387 2388 ui_progress__update(prog, size); 2389 2390 out: 2391 return err; 2392 } 2393 2394 static inline bool 2395 reader__eof(struct reader *rd) 2396 { 2397 return (rd->file_pos >= rd->data_size + rd->data_offset); 2398 } 2399 2400 static int 2401 reader__process_events(struct reader *rd, struct perf_session *session, 2402 struct ui_progress *prog) 2403 { 2404 int err; 2405 2406 err = reader__init(rd, &session->one_mmap); 2407 if (err) 2408 goto out; 2409 2410 session->active_decomp = &rd->decomp_data; 2411 2412 remap: 2413 err = reader__mmap(rd, session); 2414 if (err) 2415 goto out; 2416 2417 more: 2418 err = reader__read_event(rd, session, prog); 2419 if (err < 0) 2420 goto out; 2421 else if (err == READER_NODATA) 2422 goto remap; 2423 2424 if (session_done()) 2425 goto out; 2426 2427 if (!reader__eof(rd)) 2428 goto more; 2429 2430 out: 2431 session->active_decomp = &session->decomp_data; 2432 return err; 2433 } 2434 2435 static s64 process_simple(struct perf_session *session, 2436 union perf_event *event, 2437 u64 file_offset, 2438 const char *file_path) 2439 { 2440 return perf_session__process_event(session, event, file_offset, file_path); 2441 } 2442 2443 static int __perf_session__process_events(struct perf_session *session) 2444 { 2445 struct reader rd = { 2446 .fd = perf_data__fd(session->data), 2447 .path = session->data->file.path, 2448 .data_size = session->header.data_size, 2449 .data_offset = session->header.data_offset, 2450 .process = process_simple, 2451 .in_place_update = session->data->in_place_update, 2452 }; 2453 struct ordered_events *oe = &session->ordered_events; 2454 struct perf_tool *tool = session->tool; 2455 struct ui_progress prog; 2456 int err; 2457 2458 perf_tool__fill_defaults(tool); 2459 2460 if (rd.data_size == 0) 2461 return -1; 2462 2463 ui_progress__init_size(&prog, rd.data_size, "Processing events..."); 2464 2465 err = reader__process_events(&rd, session, &prog); 2466 if (err) 2467 goto out_err; 2468 /* do the final flush for ordered samples */ 2469 err = ordered_events__flush(oe, OE_FLUSH__FINAL); 2470 if (err) 2471 goto out_err; 2472 err = auxtrace__flush_events(session, tool); 2473 if (err) 2474 goto out_err; 2475 err = perf_session__flush_thread_stacks(session); 2476 out_err: 2477 ui_progress__finish(); 2478 if (!tool->no_warn) 2479 perf_session__warn_about_errors(session); 2480 /* 2481 * We may switching perf.data output, make ordered_events 2482 * reusable. 2483 */ 2484 ordered_events__reinit(&session->ordered_events); 2485 auxtrace__free_events(session); 2486 reader__release_decomp(&rd); 2487 session->one_mmap = false; 2488 return err; 2489 } 2490 2491 /* 2492 * Processing 2 MB of data from each reader in sequence, 2493 * because that's the way the ordered events sorting works 2494 * most efficiently. 2495 */ 2496 #define READER_MAX_SIZE (2 * 1024 * 1024) 2497 2498 /* 2499 * This function reads, merge and process directory data. 2500 * It assumens the version 1 of directory data, where each 2501 * data file holds per-cpu data, already sorted by kernel. 2502 */ 2503 static int __perf_session__process_dir_events(struct perf_session *session) 2504 { 2505 struct perf_data *data = session->data; 2506 struct perf_tool *tool = session->tool; 2507 int i, ret, readers, nr_readers; 2508 struct ui_progress prog; 2509 u64 total_size = perf_data__size(session->data); 2510 struct reader *rd; 2511 2512 perf_tool__fill_defaults(tool); 2513 2514 ui_progress__init_size(&prog, total_size, "Sorting events..."); 2515 2516 nr_readers = 1; 2517 for (i = 0; i < data->dir.nr; i++) { 2518 if (data->dir.files[i].size) 2519 nr_readers++; 2520 } 2521 2522 rd = zalloc(nr_readers * sizeof(struct reader)); 2523 if (!rd) 2524 return -ENOMEM; 2525 2526 rd[0] = (struct reader) { 2527 .fd = perf_data__fd(session->data), 2528 .path = session->data->file.path, 2529 .data_size = session->header.data_size, 2530 .data_offset = session->header.data_offset, 2531 .process = process_simple, 2532 .in_place_update = session->data->in_place_update, 2533 }; 2534 ret = reader__init(&rd[0], NULL); 2535 if (ret) 2536 goto out_err; 2537 ret = reader__mmap(&rd[0], session); 2538 if (ret) 2539 goto out_err; 2540 readers = 1; 2541 2542 for (i = 0; i < data->dir.nr; i++) { 2543 if (!data->dir.files[i].size) 2544 continue; 2545 rd[readers] = (struct reader) { 2546 .fd = data->dir.files[i].fd, 2547 .path = data->dir.files[i].path, 2548 .data_size = data->dir.files[i].size, 2549 .data_offset = 0, 2550 .process = process_simple, 2551 .in_place_update = session->data->in_place_update, 2552 }; 2553 ret = reader__init(&rd[readers], NULL); 2554 if (ret) 2555 goto out_err; 2556 ret = reader__mmap(&rd[readers], session); 2557 if (ret) 2558 goto out_err; 2559 readers++; 2560 } 2561 2562 i = 0; 2563 while (readers) { 2564 if (session_done()) 2565 break; 2566 2567 if (rd[i].done) { 2568 i = (i + 1) % nr_readers; 2569 continue; 2570 } 2571 if (reader__eof(&rd[i])) { 2572 rd[i].done = true; 2573 readers--; 2574 continue; 2575 } 2576 2577 session->active_decomp = &rd[i].decomp_data; 2578 ret = reader__read_event(&rd[i], session, &prog); 2579 if (ret < 0) { 2580 goto out_err; 2581 } else if (ret == READER_NODATA) { 2582 ret = reader__mmap(&rd[i], session); 2583 if (ret) 2584 goto out_err; 2585 } 2586 2587 if (rd[i].size >= READER_MAX_SIZE) { 2588 rd[i].size = 0; 2589 i = (i + 1) % nr_readers; 2590 } 2591 } 2592 2593 ret = ordered_events__flush(&session->ordered_events, OE_FLUSH__FINAL); 2594 if (ret) 2595 goto out_err; 2596 2597 ret = perf_session__flush_thread_stacks(session); 2598 out_err: 2599 ui_progress__finish(); 2600 2601 if (!tool->no_warn) 2602 perf_session__warn_about_errors(session); 2603 2604 /* 2605 * We may switching perf.data output, make ordered_events 2606 * reusable. 2607 */ 2608 ordered_events__reinit(&session->ordered_events); 2609 2610 session->one_mmap = false; 2611 2612 session->active_decomp = &session->decomp_data; 2613 for (i = 0; i < nr_readers; i++) 2614 reader__release_decomp(&rd[i]); 2615 zfree(&rd); 2616 2617 return ret; 2618 } 2619 2620 int perf_session__process_events(struct perf_session *session) 2621 { 2622 if (perf_session__register_idle_thread(session) < 0) 2623 return -ENOMEM; 2624 2625 if (perf_data__is_pipe(session->data)) 2626 return __perf_session__process_pipe_events(session); 2627 2628 if (perf_data__is_dir(session->data) && session->data->dir.nr) 2629 return __perf_session__process_dir_events(session); 2630 2631 return __perf_session__process_events(session); 2632 } 2633 2634 bool perf_session__has_traces(struct perf_session *session, const char *msg) 2635 { 2636 struct evsel *evsel; 2637 2638 evlist__for_each_entry(session->evlist, evsel) { 2639 if (evsel->core.attr.type == PERF_TYPE_TRACEPOINT) 2640 return true; 2641 } 2642 2643 pr_err("No trace sample to read. Did you call 'perf %s'?\n", msg); 2644 return false; 2645 } 2646 2647 int map__set_kallsyms_ref_reloc_sym(struct map *map, const char *symbol_name, u64 addr) 2648 { 2649 char *bracket; 2650 struct ref_reloc_sym *ref; 2651 struct kmap *kmap; 2652 2653 ref = zalloc(sizeof(struct ref_reloc_sym)); 2654 if (ref == NULL) 2655 return -ENOMEM; 2656 2657 ref->name = strdup(symbol_name); 2658 if (ref->name == NULL) { 2659 free(ref); 2660 return -ENOMEM; 2661 } 2662 2663 bracket = strchr(ref->name, ']'); 2664 if (bracket) 2665 *bracket = '\0'; 2666 2667 ref->addr = addr; 2668 2669 kmap = map__kmap(map); 2670 if (kmap) 2671 kmap->ref_reloc_sym = ref; 2672 2673 return 0; 2674 } 2675 2676 size_t perf_session__fprintf_dsos(struct perf_session *session, FILE *fp) 2677 { 2678 return machines__fprintf_dsos(&session->machines, fp); 2679 } 2680 2681 size_t perf_session__fprintf_dsos_buildid(struct perf_session *session, FILE *fp, 2682 bool (skip)(struct dso *dso, int parm), int parm) 2683 { 2684 return machines__fprintf_dsos_buildid(&session->machines, fp, skip, parm); 2685 } 2686 2687 size_t perf_session__fprintf_nr_events(struct perf_session *session, FILE *fp, 2688 bool skip_empty) 2689 { 2690 size_t ret; 2691 const char *msg = ""; 2692 2693 if (perf_header__has_feat(&session->header, HEADER_AUXTRACE)) 2694 msg = " (excludes AUX area (e.g. instruction trace) decoded / synthesized events)"; 2695 2696 ret = fprintf(fp, "\nAggregated stats:%s\n", msg); 2697 2698 ret += events_stats__fprintf(&session->evlist->stats, fp, skip_empty); 2699 return ret; 2700 } 2701 2702 size_t perf_session__fprintf(struct perf_session *session, FILE *fp) 2703 { 2704 /* 2705 * FIXME: Here we have to actually print all the machines in this 2706 * session, not just the host... 2707 */ 2708 return machine__fprintf(&session->machines.host, fp); 2709 } 2710 2711 struct evsel *perf_session__find_first_evtype(struct perf_session *session, 2712 unsigned int type) 2713 { 2714 struct evsel *pos; 2715 2716 evlist__for_each_entry(session->evlist, pos) { 2717 if (pos->core.attr.type == type) 2718 return pos; 2719 } 2720 return NULL; 2721 } 2722 2723 int perf_session__cpu_bitmap(struct perf_session *session, 2724 const char *cpu_list, unsigned long *cpu_bitmap) 2725 { 2726 int i, err = -1; 2727 struct perf_cpu_map *map; 2728 int nr_cpus = min(session->header.env.nr_cpus_avail, MAX_NR_CPUS); 2729 2730 for (i = 0; i < PERF_TYPE_MAX; ++i) { 2731 struct evsel *evsel; 2732 2733 evsel = perf_session__find_first_evtype(session, i); 2734 if (!evsel) 2735 continue; 2736 2737 if (!(evsel->core.attr.sample_type & PERF_SAMPLE_CPU)) { 2738 pr_err("File does not contain CPU events. " 2739 "Remove -C option to proceed.\n"); 2740 return -1; 2741 } 2742 } 2743 2744 map = perf_cpu_map__new(cpu_list); 2745 if (map == NULL) { 2746 pr_err("Invalid cpu_list\n"); 2747 return -1; 2748 } 2749 2750 for (i = 0; i < perf_cpu_map__nr(map); i++) { 2751 struct perf_cpu cpu = perf_cpu_map__cpu(map, i); 2752 2753 if (cpu.cpu >= nr_cpus) { 2754 pr_err("Requested CPU %d too large. " 2755 "Consider raising MAX_NR_CPUS\n", cpu.cpu); 2756 goto out_delete_map; 2757 } 2758 2759 __set_bit(cpu.cpu, cpu_bitmap); 2760 } 2761 2762 err = 0; 2763 2764 out_delete_map: 2765 perf_cpu_map__put(map); 2766 return err; 2767 } 2768 2769 void perf_session__fprintf_info(struct perf_session *session, FILE *fp, 2770 bool full) 2771 { 2772 if (session == NULL || fp == NULL) 2773 return; 2774 2775 fprintf(fp, "# ========\n"); 2776 perf_header__fprintf_info(session, fp, full); 2777 fprintf(fp, "# ========\n#\n"); 2778 } 2779 2780 static int perf_session__register_guest(struct perf_session *session, pid_t machine_pid) 2781 { 2782 struct machine *machine = machines__findnew(&session->machines, machine_pid); 2783 struct thread *thread; 2784 2785 if (!machine) 2786 return -ENOMEM; 2787 2788 machine->single_address_space = session->machines.host.single_address_space; 2789 2790 thread = machine__idle_thread(machine); 2791 if (!thread) 2792 return -ENOMEM; 2793 thread__put(thread); 2794 2795 machine->kallsyms_filename = perf_data__guest_kallsyms_name(session->data, machine_pid); 2796 2797 return 0; 2798 } 2799 2800 static int perf_session__set_guest_cpu(struct perf_session *session, pid_t pid, 2801 pid_t tid, int guest_cpu) 2802 { 2803 struct machine *machine = &session->machines.host; 2804 struct thread *thread = machine__findnew_thread(machine, pid, tid); 2805 2806 if (!thread) 2807 return -ENOMEM; 2808 thread->guest_cpu = guest_cpu; 2809 thread__put(thread); 2810 2811 return 0; 2812 } 2813 2814 int perf_event__process_id_index(struct perf_session *session, 2815 union perf_event *event) 2816 { 2817 struct evlist *evlist = session->evlist; 2818 struct perf_record_id_index *ie = &event->id_index; 2819 size_t sz = ie->header.size - sizeof(*ie); 2820 size_t i, nr, max_nr; 2821 size_t e1_sz = sizeof(struct id_index_entry); 2822 size_t e2_sz = sizeof(struct id_index_entry_2); 2823 size_t etot_sz = e1_sz + e2_sz; 2824 struct id_index_entry_2 *e2; 2825 pid_t last_pid = 0; 2826 2827 max_nr = sz / e1_sz; 2828 nr = ie->nr; 2829 if (nr > max_nr) { 2830 printf("Too big: nr %zu max_nr %zu\n", nr, max_nr); 2831 return -EINVAL; 2832 } 2833 2834 if (sz >= nr * etot_sz) { 2835 max_nr = sz / etot_sz; 2836 if (nr > max_nr) { 2837 printf("Too big2: nr %zu max_nr %zu\n", nr, max_nr); 2838 return -EINVAL; 2839 } 2840 e2 = (void *)ie + sizeof(*ie) + nr * e1_sz; 2841 } else { 2842 e2 = NULL; 2843 } 2844 2845 if (dump_trace) 2846 fprintf(stdout, " nr: %zu\n", nr); 2847 2848 for (i = 0; i < nr; i++, (e2 ? e2++ : 0)) { 2849 struct id_index_entry *e = &ie->entries[i]; 2850 struct perf_sample_id *sid; 2851 int ret; 2852 2853 if (dump_trace) { 2854 fprintf(stdout, " ... id: %"PRI_lu64, e->id); 2855 fprintf(stdout, " idx: %"PRI_lu64, e->idx); 2856 fprintf(stdout, " cpu: %"PRI_ld64, e->cpu); 2857 fprintf(stdout, " tid: %"PRI_ld64, e->tid); 2858 if (e2) { 2859 fprintf(stdout, " machine_pid: %"PRI_ld64, e2->machine_pid); 2860 fprintf(stdout, " vcpu: %"PRI_lu64"\n", e2->vcpu); 2861 } else { 2862 fprintf(stdout, "\n"); 2863 } 2864 } 2865 2866 sid = evlist__id2sid(evlist, e->id); 2867 if (!sid) 2868 return -ENOENT; 2869 2870 sid->idx = e->idx; 2871 sid->cpu.cpu = e->cpu; 2872 sid->tid = e->tid; 2873 2874 if (!e2) 2875 continue; 2876 2877 sid->machine_pid = e2->machine_pid; 2878 sid->vcpu.cpu = e2->vcpu; 2879 2880 if (!sid->machine_pid) 2881 continue; 2882 2883 if (sid->machine_pid != last_pid) { 2884 ret = perf_session__register_guest(session, sid->machine_pid); 2885 if (ret) 2886 return ret; 2887 last_pid = sid->machine_pid; 2888 perf_guest = true; 2889 } 2890 2891 ret = perf_session__set_guest_cpu(session, sid->machine_pid, e->tid, e2->vcpu); 2892 if (ret) 2893 return ret; 2894 } 2895 return 0; 2896 } 2897