1 // SPDX-License-Identifier: GPL-2.0 2 #include <errno.h> 3 #include <inttypes.h> 4 #include <linux/err.h> 5 #include <linux/kernel.h> 6 #include <linux/zalloc.h> 7 #include <api/fs/fs.h> 8 9 #include <byteswap.h> 10 #include <unistd.h> 11 #include <sys/types.h> 12 #include <sys/mman.h> 13 #include <perf/cpumap.h> 14 15 #include "map_symbol.h" 16 #include "branch.h" 17 #include "debug.h" 18 #include "env.h" 19 #include "evlist.h" 20 #include "evsel.h" 21 #include "memswap.h" 22 #include "map.h" 23 #include "symbol.h" 24 #include "session.h" 25 #include "tool.h" 26 #include "perf_regs.h" 27 #include "asm/bug.h" 28 #include "auxtrace.h" 29 #include "thread.h" 30 #include "thread-stack.h" 31 #include "sample-raw.h" 32 #include "stat.h" 33 #include "tsc.h" 34 #include "ui/progress.h" 35 #include "../perf.h" 36 #include "arch/common.h" 37 #include "units.h" 38 #include <internal/lib.h> 39 40 #ifdef HAVE_ZSTD_SUPPORT 41 static int perf_session__process_compressed_event(struct perf_session *session, 42 union perf_event *event, u64 file_offset, 43 const char *file_path) 44 { 45 void *src; 46 size_t decomp_size, src_size; 47 u64 decomp_last_rem = 0; 48 size_t mmap_len, decomp_len = session->header.env.comp_mmap_len; 49 struct decomp *decomp, *decomp_last = session->active_decomp->decomp_last; 50 51 if (decomp_last) { 52 decomp_last_rem = decomp_last->size - decomp_last->head; 53 decomp_len += decomp_last_rem; 54 } 55 56 mmap_len = sizeof(struct decomp) + decomp_len; 57 decomp = mmap(NULL, mmap_len, PROT_READ|PROT_WRITE, 58 MAP_ANONYMOUS|MAP_PRIVATE, -1, 0); 59 if (decomp == MAP_FAILED) { 60 pr_err("Couldn't allocate memory for decompression\n"); 61 return -1; 62 } 63 64 decomp->file_pos = file_offset; 65 decomp->file_path = file_path; 66 decomp->mmap_len = mmap_len; 67 decomp->head = 0; 68 69 if (decomp_last_rem) { 70 memcpy(decomp->data, &(decomp_last->data[decomp_last->head]), decomp_last_rem); 71 decomp->size = decomp_last_rem; 72 } 73 74 src = (void *)event + sizeof(struct perf_record_compressed); 75 src_size = event->pack.header.size - sizeof(struct perf_record_compressed); 76 77 decomp_size = zstd_decompress_stream(session->active_decomp->zstd_decomp, src, src_size, 78 &(decomp->data[decomp_last_rem]), decomp_len - decomp_last_rem); 79 if (!decomp_size) { 80 munmap(decomp, mmap_len); 81 pr_err("Couldn't decompress data\n"); 82 return -1; 83 } 84 85 decomp->size += decomp_size; 86 87 if (session->active_decomp->decomp == NULL) 88 session->active_decomp->decomp = decomp; 89 else 90 session->active_decomp->decomp_last->next = decomp; 91 92 session->active_decomp->decomp_last = decomp; 93 94 pr_debug("decomp (B): %zd to %zd\n", src_size, decomp_size); 95 96 return 0; 97 } 98 #else /* !HAVE_ZSTD_SUPPORT */ 99 #define perf_session__process_compressed_event perf_session__process_compressed_event_stub 100 #endif 101 102 static int perf_session__deliver_event(struct perf_session *session, 103 union perf_event *event, 104 struct perf_tool *tool, 105 u64 file_offset, 106 const char *file_path); 107 108 static int perf_session__open(struct perf_session *session, int repipe_fd) 109 { 110 struct perf_data *data = session->data; 111 112 if (perf_session__read_header(session, repipe_fd) < 0) { 113 pr_err("incompatible file format (rerun with -v to learn more)\n"); 114 return -1; 115 } 116 117 if (perf_data__is_pipe(data)) 118 return 0; 119 120 if (perf_header__has_feat(&session->header, HEADER_STAT)) 121 return 0; 122 123 if (!evlist__valid_sample_type(session->evlist)) { 124 pr_err("non matching sample_type\n"); 125 return -1; 126 } 127 128 if (!evlist__valid_sample_id_all(session->evlist)) { 129 pr_err("non matching sample_id_all\n"); 130 return -1; 131 } 132 133 if (!evlist__valid_read_format(session->evlist)) { 134 pr_err("non matching read_format\n"); 135 return -1; 136 } 137 138 return 0; 139 } 140 141 void perf_session__set_id_hdr_size(struct perf_session *session) 142 { 143 u16 id_hdr_size = evlist__id_hdr_size(session->evlist); 144 145 machines__set_id_hdr_size(&session->machines, id_hdr_size); 146 } 147 148 int perf_session__create_kernel_maps(struct perf_session *session) 149 { 150 int ret = machine__create_kernel_maps(&session->machines.host); 151 152 if (ret >= 0) 153 ret = machines__create_guest_kernel_maps(&session->machines); 154 return ret; 155 } 156 157 static void perf_session__destroy_kernel_maps(struct perf_session *session) 158 { 159 machines__destroy_kernel_maps(&session->machines); 160 } 161 162 static bool perf_session__has_comm_exec(struct perf_session *session) 163 { 164 struct evsel *evsel; 165 166 evlist__for_each_entry(session->evlist, evsel) { 167 if (evsel->core.attr.comm_exec) 168 return true; 169 } 170 171 return false; 172 } 173 174 static void perf_session__set_comm_exec(struct perf_session *session) 175 { 176 bool comm_exec = perf_session__has_comm_exec(session); 177 178 machines__set_comm_exec(&session->machines, comm_exec); 179 } 180 181 static int ordered_events__deliver_event(struct ordered_events *oe, 182 struct ordered_event *event) 183 { 184 struct perf_session *session = container_of(oe, struct perf_session, 185 ordered_events); 186 187 return perf_session__deliver_event(session, event->event, 188 session->tool, event->file_offset, 189 event->file_path); 190 } 191 192 struct perf_session *__perf_session__new(struct perf_data *data, 193 bool repipe, int repipe_fd, 194 struct perf_tool *tool) 195 { 196 int ret = -ENOMEM; 197 struct perf_session *session = zalloc(sizeof(*session)); 198 199 if (!session) 200 goto out; 201 202 session->repipe = repipe; 203 session->tool = tool; 204 session->decomp_data.zstd_decomp = &session->zstd_data; 205 session->active_decomp = &session->decomp_data; 206 INIT_LIST_HEAD(&session->auxtrace_index); 207 machines__init(&session->machines); 208 ordered_events__init(&session->ordered_events, 209 ordered_events__deliver_event, NULL); 210 211 perf_env__init(&session->header.env); 212 if (data) { 213 ret = perf_data__open(data); 214 if (ret < 0) 215 goto out_delete; 216 217 session->data = data; 218 219 if (perf_data__is_read(data)) { 220 ret = perf_session__open(session, repipe_fd); 221 if (ret < 0) 222 goto out_delete; 223 224 /* 225 * set session attributes that are present in perf.data 226 * but not in pipe-mode. 227 */ 228 if (!data->is_pipe) { 229 perf_session__set_id_hdr_size(session); 230 perf_session__set_comm_exec(session); 231 } 232 233 evlist__init_trace_event_sample_raw(session->evlist); 234 235 /* Open the directory data. */ 236 if (data->is_dir) { 237 ret = perf_data__open_dir(data); 238 if (ret) 239 goto out_delete; 240 } 241 242 if (!symbol_conf.kallsyms_name && 243 !symbol_conf.vmlinux_name) 244 symbol_conf.kallsyms_name = perf_data__kallsyms_name(data); 245 } 246 } else { 247 session->machines.host.env = &perf_env; 248 } 249 250 session->machines.host.single_address_space = 251 perf_env__single_address_space(session->machines.host.env); 252 253 if (!data || perf_data__is_write(data)) { 254 /* 255 * In O_RDONLY mode this will be performed when reading the 256 * kernel MMAP event, in perf_event__process_mmap(). 257 */ 258 if (perf_session__create_kernel_maps(session) < 0) 259 pr_warning("Cannot read kernel map\n"); 260 } 261 262 /* 263 * In pipe-mode, evlist is empty until PERF_RECORD_HEADER_ATTR is 264 * processed, so evlist__sample_id_all is not meaningful here. 265 */ 266 if ((!data || !data->is_pipe) && tool && tool->ordering_requires_timestamps && 267 tool->ordered_events && !evlist__sample_id_all(session->evlist)) { 268 dump_printf("WARNING: No sample_id_all support, falling back to unordered processing\n"); 269 tool->ordered_events = false; 270 } 271 272 return session; 273 274 out_delete: 275 perf_session__delete(session); 276 out: 277 return ERR_PTR(ret); 278 } 279 280 static void perf_session__delete_threads(struct perf_session *session) 281 { 282 machine__delete_threads(&session->machines.host); 283 } 284 285 static void perf_decomp__release_events(struct decomp *next) 286 { 287 struct decomp *decomp; 288 size_t mmap_len; 289 290 do { 291 decomp = next; 292 if (decomp == NULL) 293 break; 294 next = decomp->next; 295 mmap_len = decomp->mmap_len; 296 munmap(decomp, mmap_len); 297 } while (1); 298 } 299 300 void perf_session__delete(struct perf_session *session) 301 { 302 if (session == NULL) 303 return; 304 auxtrace__free(session); 305 auxtrace_index__free(&session->auxtrace_index); 306 perf_session__destroy_kernel_maps(session); 307 perf_session__delete_threads(session); 308 perf_decomp__release_events(session->decomp_data.decomp); 309 perf_env__exit(&session->header.env); 310 machines__exit(&session->machines); 311 if (session->data) { 312 if (perf_data__is_read(session->data)) 313 evlist__delete(session->evlist); 314 perf_data__close(session->data); 315 } 316 trace_event__cleanup(&session->tevent); 317 free(session); 318 } 319 320 static int process_event_synth_tracing_data_stub(struct perf_session *session 321 __maybe_unused, 322 union perf_event *event 323 __maybe_unused) 324 { 325 dump_printf(": unhandled!\n"); 326 return 0; 327 } 328 329 static int process_event_synth_attr_stub(struct perf_tool *tool __maybe_unused, 330 union perf_event *event __maybe_unused, 331 struct evlist **pevlist 332 __maybe_unused) 333 { 334 dump_printf(": unhandled!\n"); 335 return 0; 336 } 337 338 static int process_event_synth_event_update_stub(struct perf_tool *tool __maybe_unused, 339 union perf_event *event __maybe_unused, 340 struct evlist **pevlist 341 __maybe_unused) 342 { 343 if (dump_trace) 344 perf_event__fprintf_event_update(event, stdout); 345 346 dump_printf(": unhandled!\n"); 347 return 0; 348 } 349 350 static int process_event_sample_stub(struct perf_tool *tool __maybe_unused, 351 union perf_event *event __maybe_unused, 352 struct perf_sample *sample __maybe_unused, 353 struct evsel *evsel __maybe_unused, 354 struct machine *machine __maybe_unused) 355 { 356 dump_printf(": unhandled!\n"); 357 return 0; 358 } 359 360 static int process_event_stub(struct perf_tool *tool __maybe_unused, 361 union perf_event *event __maybe_unused, 362 struct perf_sample *sample __maybe_unused, 363 struct machine *machine __maybe_unused) 364 { 365 dump_printf(": unhandled!\n"); 366 return 0; 367 } 368 369 static int process_finished_round_stub(struct perf_tool *tool __maybe_unused, 370 union perf_event *event __maybe_unused, 371 struct ordered_events *oe __maybe_unused) 372 { 373 dump_printf(": unhandled!\n"); 374 return 0; 375 } 376 377 static int skipn(int fd, off_t n) 378 { 379 char buf[4096]; 380 ssize_t ret; 381 382 while (n > 0) { 383 ret = read(fd, buf, min(n, (off_t)sizeof(buf))); 384 if (ret <= 0) 385 return ret; 386 n -= ret; 387 } 388 389 return 0; 390 } 391 392 static s64 process_event_auxtrace_stub(struct perf_session *session __maybe_unused, 393 union perf_event *event) 394 { 395 dump_printf(": unhandled!\n"); 396 if (perf_data__is_pipe(session->data)) 397 skipn(perf_data__fd(session->data), event->auxtrace.size); 398 return event->auxtrace.size; 399 } 400 401 static int process_event_op2_stub(struct perf_session *session __maybe_unused, 402 union perf_event *event __maybe_unused) 403 { 404 dump_printf(": unhandled!\n"); 405 return 0; 406 } 407 408 409 static 410 int process_event_thread_map_stub(struct perf_session *session __maybe_unused, 411 union perf_event *event __maybe_unused) 412 { 413 if (dump_trace) 414 perf_event__fprintf_thread_map(event, stdout); 415 416 dump_printf(": unhandled!\n"); 417 return 0; 418 } 419 420 static 421 int process_event_cpu_map_stub(struct perf_session *session __maybe_unused, 422 union perf_event *event __maybe_unused) 423 { 424 if (dump_trace) 425 perf_event__fprintf_cpu_map(event, stdout); 426 427 dump_printf(": unhandled!\n"); 428 return 0; 429 } 430 431 static 432 int process_event_stat_config_stub(struct perf_session *session __maybe_unused, 433 union perf_event *event __maybe_unused) 434 { 435 if (dump_trace) 436 perf_event__fprintf_stat_config(event, stdout); 437 438 dump_printf(": unhandled!\n"); 439 return 0; 440 } 441 442 static int process_stat_stub(struct perf_session *perf_session __maybe_unused, 443 union perf_event *event) 444 { 445 if (dump_trace) 446 perf_event__fprintf_stat(event, stdout); 447 448 dump_printf(": unhandled!\n"); 449 return 0; 450 } 451 452 static int process_stat_round_stub(struct perf_session *perf_session __maybe_unused, 453 union perf_event *event) 454 { 455 if (dump_trace) 456 perf_event__fprintf_stat_round(event, stdout); 457 458 dump_printf(": unhandled!\n"); 459 return 0; 460 } 461 462 static int process_event_time_conv_stub(struct perf_session *perf_session __maybe_unused, 463 union perf_event *event) 464 { 465 if (dump_trace) 466 perf_event__fprintf_time_conv(event, stdout); 467 468 dump_printf(": unhandled!\n"); 469 return 0; 470 } 471 472 static int perf_session__process_compressed_event_stub(struct perf_session *session __maybe_unused, 473 union perf_event *event __maybe_unused, 474 u64 file_offset __maybe_unused, 475 const char *file_path __maybe_unused) 476 { 477 dump_printf(": unhandled!\n"); 478 return 0; 479 } 480 481 void perf_tool__fill_defaults(struct perf_tool *tool) 482 { 483 if (tool->sample == NULL) 484 tool->sample = process_event_sample_stub; 485 if (tool->mmap == NULL) 486 tool->mmap = process_event_stub; 487 if (tool->mmap2 == NULL) 488 tool->mmap2 = process_event_stub; 489 if (tool->comm == NULL) 490 tool->comm = process_event_stub; 491 if (tool->namespaces == NULL) 492 tool->namespaces = process_event_stub; 493 if (tool->cgroup == NULL) 494 tool->cgroup = process_event_stub; 495 if (tool->fork == NULL) 496 tool->fork = process_event_stub; 497 if (tool->exit == NULL) 498 tool->exit = process_event_stub; 499 if (tool->lost == NULL) 500 tool->lost = perf_event__process_lost; 501 if (tool->lost_samples == NULL) 502 tool->lost_samples = perf_event__process_lost_samples; 503 if (tool->aux == NULL) 504 tool->aux = perf_event__process_aux; 505 if (tool->itrace_start == NULL) 506 tool->itrace_start = perf_event__process_itrace_start; 507 if (tool->context_switch == NULL) 508 tool->context_switch = perf_event__process_switch; 509 if (tool->ksymbol == NULL) 510 tool->ksymbol = perf_event__process_ksymbol; 511 if (tool->bpf == NULL) 512 tool->bpf = perf_event__process_bpf; 513 if (tool->text_poke == NULL) 514 tool->text_poke = perf_event__process_text_poke; 515 if (tool->aux_output_hw_id == NULL) 516 tool->aux_output_hw_id = perf_event__process_aux_output_hw_id; 517 if (tool->read == NULL) 518 tool->read = process_event_sample_stub; 519 if (tool->throttle == NULL) 520 tool->throttle = process_event_stub; 521 if (tool->unthrottle == NULL) 522 tool->unthrottle = process_event_stub; 523 if (tool->attr == NULL) 524 tool->attr = process_event_synth_attr_stub; 525 if (tool->event_update == NULL) 526 tool->event_update = process_event_synth_event_update_stub; 527 if (tool->tracing_data == NULL) 528 tool->tracing_data = process_event_synth_tracing_data_stub; 529 if (tool->build_id == NULL) 530 tool->build_id = process_event_op2_stub; 531 if (tool->finished_round == NULL) { 532 if (tool->ordered_events) 533 tool->finished_round = perf_event__process_finished_round; 534 else 535 tool->finished_round = process_finished_round_stub; 536 } 537 if (tool->id_index == NULL) 538 tool->id_index = process_event_op2_stub; 539 if (tool->auxtrace_info == NULL) 540 tool->auxtrace_info = process_event_op2_stub; 541 if (tool->auxtrace == NULL) 542 tool->auxtrace = process_event_auxtrace_stub; 543 if (tool->auxtrace_error == NULL) 544 tool->auxtrace_error = process_event_op2_stub; 545 if (tool->thread_map == NULL) 546 tool->thread_map = process_event_thread_map_stub; 547 if (tool->cpu_map == NULL) 548 tool->cpu_map = process_event_cpu_map_stub; 549 if (tool->stat_config == NULL) 550 tool->stat_config = process_event_stat_config_stub; 551 if (tool->stat == NULL) 552 tool->stat = process_stat_stub; 553 if (tool->stat_round == NULL) 554 tool->stat_round = process_stat_round_stub; 555 if (tool->time_conv == NULL) 556 tool->time_conv = process_event_time_conv_stub; 557 if (tool->feature == NULL) 558 tool->feature = process_event_op2_stub; 559 if (tool->compressed == NULL) 560 tool->compressed = perf_session__process_compressed_event; 561 if (tool->finished_init == NULL) 562 tool->finished_init = process_event_op2_stub; 563 } 564 565 static void swap_sample_id_all(union perf_event *event, void *data) 566 { 567 void *end = (void *) event + event->header.size; 568 int size = end - data; 569 570 BUG_ON(size % sizeof(u64)); 571 mem_bswap_64(data, size); 572 } 573 574 static void perf_event__all64_swap(union perf_event *event, 575 bool sample_id_all __maybe_unused) 576 { 577 struct perf_event_header *hdr = &event->header; 578 mem_bswap_64(hdr + 1, event->header.size - sizeof(*hdr)); 579 } 580 581 static void perf_event__comm_swap(union perf_event *event, bool sample_id_all) 582 { 583 event->comm.pid = bswap_32(event->comm.pid); 584 event->comm.tid = bswap_32(event->comm.tid); 585 586 if (sample_id_all) { 587 void *data = &event->comm.comm; 588 589 data += PERF_ALIGN(strlen(data) + 1, sizeof(u64)); 590 swap_sample_id_all(event, data); 591 } 592 } 593 594 static void perf_event__mmap_swap(union perf_event *event, 595 bool sample_id_all) 596 { 597 event->mmap.pid = bswap_32(event->mmap.pid); 598 event->mmap.tid = bswap_32(event->mmap.tid); 599 event->mmap.start = bswap_64(event->mmap.start); 600 event->mmap.len = bswap_64(event->mmap.len); 601 event->mmap.pgoff = bswap_64(event->mmap.pgoff); 602 603 if (sample_id_all) { 604 void *data = &event->mmap.filename; 605 606 data += PERF_ALIGN(strlen(data) + 1, sizeof(u64)); 607 swap_sample_id_all(event, data); 608 } 609 } 610 611 static void perf_event__mmap2_swap(union perf_event *event, 612 bool sample_id_all) 613 { 614 event->mmap2.pid = bswap_32(event->mmap2.pid); 615 event->mmap2.tid = bswap_32(event->mmap2.tid); 616 event->mmap2.start = bswap_64(event->mmap2.start); 617 event->mmap2.len = bswap_64(event->mmap2.len); 618 event->mmap2.pgoff = bswap_64(event->mmap2.pgoff); 619 620 if (!(event->header.misc & PERF_RECORD_MISC_MMAP_BUILD_ID)) { 621 event->mmap2.maj = bswap_32(event->mmap2.maj); 622 event->mmap2.min = bswap_32(event->mmap2.min); 623 event->mmap2.ino = bswap_64(event->mmap2.ino); 624 event->mmap2.ino_generation = bswap_64(event->mmap2.ino_generation); 625 } 626 627 if (sample_id_all) { 628 void *data = &event->mmap2.filename; 629 630 data += PERF_ALIGN(strlen(data) + 1, sizeof(u64)); 631 swap_sample_id_all(event, data); 632 } 633 } 634 static void perf_event__task_swap(union perf_event *event, bool sample_id_all) 635 { 636 event->fork.pid = bswap_32(event->fork.pid); 637 event->fork.tid = bswap_32(event->fork.tid); 638 event->fork.ppid = bswap_32(event->fork.ppid); 639 event->fork.ptid = bswap_32(event->fork.ptid); 640 event->fork.time = bswap_64(event->fork.time); 641 642 if (sample_id_all) 643 swap_sample_id_all(event, &event->fork + 1); 644 } 645 646 static void perf_event__read_swap(union perf_event *event, bool sample_id_all) 647 { 648 event->read.pid = bswap_32(event->read.pid); 649 event->read.tid = bswap_32(event->read.tid); 650 event->read.value = bswap_64(event->read.value); 651 event->read.time_enabled = bswap_64(event->read.time_enabled); 652 event->read.time_running = bswap_64(event->read.time_running); 653 event->read.id = bswap_64(event->read.id); 654 655 if (sample_id_all) 656 swap_sample_id_all(event, &event->read + 1); 657 } 658 659 static void perf_event__aux_swap(union perf_event *event, bool sample_id_all) 660 { 661 event->aux.aux_offset = bswap_64(event->aux.aux_offset); 662 event->aux.aux_size = bswap_64(event->aux.aux_size); 663 event->aux.flags = bswap_64(event->aux.flags); 664 665 if (sample_id_all) 666 swap_sample_id_all(event, &event->aux + 1); 667 } 668 669 static void perf_event__itrace_start_swap(union perf_event *event, 670 bool sample_id_all) 671 { 672 event->itrace_start.pid = bswap_32(event->itrace_start.pid); 673 event->itrace_start.tid = bswap_32(event->itrace_start.tid); 674 675 if (sample_id_all) 676 swap_sample_id_all(event, &event->itrace_start + 1); 677 } 678 679 static void perf_event__switch_swap(union perf_event *event, bool sample_id_all) 680 { 681 if (event->header.type == PERF_RECORD_SWITCH_CPU_WIDE) { 682 event->context_switch.next_prev_pid = 683 bswap_32(event->context_switch.next_prev_pid); 684 event->context_switch.next_prev_tid = 685 bswap_32(event->context_switch.next_prev_tid); 686 } 687 688 if (sample_id_all) 689 swap_sample_id_all(event, &event->context_switch + 1); 690 } 691 692 static void perf_event__text_poke_swap(union perf_event *event, bool sample_id_all) 693 { 694 event->text_poke.addr = bswap_64(event->text_poke.addr); 695 event->text_poke.old_len = bswap_16(event->text_poke.old_len); 696 event->text_poke.new_len = bswap_16(event->text_poke.new_len); 697 698 if (sample_id_all) { 699 size_t len = sizeof(event->text_poke.old_len) + 700 sizeof(event->text_poke.new_len) + 701 event->text_poke.old_len + 702 event->text_poke.new_len; 703 void *data = &event->text_poke.old_len; 704 705 data += PERF_ALIGN(len, sizeof(u64)); 706 swap_sample_id_all(event, data); 707 } 708 } 709 710 static void perf_event__throttle_swap(union perf_event *event, 711 bool sample_id_all) 712 { 713 event->throttle.time = bswap_64(event->throttle.time); 714 event->throttle.id = bswap_64(event->throttle.id); 715 event->throttle.stream_id = bswap_64(event->throttle.stream_id); 716 717 if (sample_id_all) 718 swap_sample_id_all(event, &event->throttle + 1); 719 } 720 721 static void perf_event__namespaces_swap(union perf_event *event, 722 bool sample_id_all) 723 { 724 u64 i; 725 726 event->namespaces.pid = bswap_32(event->namespaces.pid); 727 event->namespaces.tid = bswap_32(event->namespaces.tid); 728 event->namespaces.nr_namespaces = bswap_64(event->namespaces.nr_namespaces); 729 730 for (i = 0; i < event->namespaces.nr_namespaces; i++) { 731 struct perf_ns_link_info *ns = &event->namespaces.link_info[i]; 732 733 ns->dev = bswap_64(ns->dev); 734 ns->ino = bswap_64(ns->ino); 735 } 736 737 if (sample_id_all) 738 swap_sample_id_all(event, &event->namespaces.link_info[i]); 739 } 740 741 static void perf_event__cgroup_swap(union perf_event *event, bool sample_id_all) 742 { 743 event->cgroup.id = bswap_64(event->cgroup.id); 744 745 if (sample_id_all) { 746 void *data = &event->cgroup.path; 747 748 data += PERF_ALIGN(strlen(data) + 1, sizeof(u64)); 749 swap_sample_id_all(event, data); 750 } 751 } 752 753 static u8 revbyte(u8 b) 754 { 755 int rev = (b >> 4) | ((b & 0xf) << 4); 756 rev = ((rev & 0xcc) >> 2) | ((rev & 0x33) << 2); 757 rev = ((rev & 0xaa) >> 1) | ((rev & 0x55) << 1); 758 return (u8) rev; 759 } 760 761 /* 762 * XXX this is hack in attempt to carry flags bitfield 763 * through endian village. ABI says: 764 * 765 * Bit-fields are allocated from right to left (least to most significant) 766 * on little-endian implementations and from left to right (most to least 767 * significant) on big-endian implementations. 768 * 769 * The above seems to be byte specific, so we need to reverse each 770 * byte of the bitfield. 'Internet' also says this might be implementation 771 * specific and we probably need proper fix and carry perf_event_attr 772 * bitfield flags in separate data file FEAT_ section. Thought this seems 773 * to work for now. 774 */ 775 static void swap_bitfield(u8 *p, unsigned len) 776 { 777 unsigned i; 778 779 for (i = 0; i < len; i++) { 780 *p = revbyte(*p); 781 p++; 782 } 783 } 784 785 /* exported for swapping attributes in file header */ 786 void perf_event__attr_swap(struct perf_event_attr *attr) 787 { 788 attr->type = bswap_32(attr->type); 789 attr->size = bswap_32(attr->size); 790 791 #define bswap_safe(f, n) \ 792 (attr->size > (offsetof(struct perf_event_attr, f) + \ 793 sizeof(attr->f) * (n))) 794 #define bswap_field(f, sz) \ 795 do { \ 796 if (bswap_safe(f, 0)) \ 797 attr->f = bswap_##sz(attr->f); \ 798 } while(0) 799 #define bswap_field_16(f) bswap_field(f, 16) 800 #define bswap_field_32(f) bswap_field(f, 32) 801 #define bswap_field_64(f) bswap_field(f, 64) 802 803 bswap_field_64(config); 804 bswap_field_64(sample_period); 805 bswap_field_64(sample_type); 806 bswap_field_64(read_format); 807 bswap_field_32(wakeup_events); 808 bswap_field_32(bp_type); 809 bswap_field_64(bp_addr); 810 bswap_field_64(bp_len); 811 bswap_field_64(branch_sample_type); 812 bswap_field_64(sample_regs_user); 813 bswap_field_32(sample_stack_user); 814 bswap_field_32(aux_watermark); 815 bswap_field_16(sample_max_stack); 816 bswap_field_32(aux_sample_size); 817 818 /* 819 * After read_format are bitfields. Check read_format because 820 * we are unable to use offsetof on bitfield. 821 */ 822 if (bswap_safe(read_format, 1)) 823 swap_bitfield((u8 *) (&attr->read_format + 1), 824 sizeof(u64)); 825 #undef bswap_field_64 826 #undef bswap_field_32 827 #undef bswap_field 828 #undef bswap_safe 829 } 830 831 static void perf_event__hdr_attr_swap(union perf_event *event, 832 bool sample_id_all __maybe_unused) 833 { 834 size_t size; 835 836 perf_event__attr_swap(&event->attr.attr); 837 838 size = event->header.size; 839 size -= (void *)&event->attr.id - (void *)event; 840 mem_bswap_64(event->attr.id, size); 841 } 842 843 static void perf_event__event_update_swap(union perf_event *event, 844 bool sample_id_all __maybe_unused) 845 { 846 event->event_update.type = bswap_64(event->event_update.type); 847 event->event_update.id = bswap_64(event->event_update.id); 848 } 849 850 static void perf_event__event_type_swap(union perf_event *event, 851 bool sample_id_all __maybe_unused) 852 { 853 event->event_type.event_type.event_id = 854 bswap_64(event->event_type.event_type.event_id); 855 } 856 857 static void perf_event__tracing_data_swap(union perf_event *event, 858 bool sample_id_all __maybe_unused) 859 { 860 event->tracing_data.size = bswap_32(event->tracing_data.size); 861 } 862 863 static void perf_event__auxtrace_info_swap(union perf_event *event, 864 bool sample_id_all __maybe_unused) 865 { 866 size_t size; 867 868 event->auxtrace_info.type = bswap_32(event->auxtrace_info.type); 869 870 size = event->header.size; 871 size -= (void *)&event->auxtrace_info.priv - (void *)event; 872 mem_bswap_64(event->auxtrace_info.priv, size); 873 } 874 875 static void perf_event__auxtrace_swap(union perf_event *event, 876 bool sample_id_all __maybe_unused) 877 { 878 event->auxtrace.size = bswap_64(event->auxtrace.size); 879 event->auxtrace.offset = bswap_64(event->auxtrace.offset); 880 event->auxtrace.reference = bswap_64(event->auxtrace.reference); 881 event->auxtrace.idx = bswap_32(event->auxtrace.idx); 882 event->auxtrace.tid = bswap_32(event->auxtrace.tid); 883 event->auxtrace.cpu = bswap_32(event->auxtrace.cpu); 884 } 885 886 static void perf_event__auxtrace_error_swap(union perf_event *event, 887 bool sample_id_all __maybe_unused) 888 { 889 event->auxtrace_error.type = bswap_32(event->auxtrace_error.type); 890 event->auxtrace_error.code = bswap_32(event->auxtrace_error.code); 891 event->auxtrace_error.cpu = bswap_32(event->auxtrace_error.cpu); 892 event->auxtrace_error.pid = bswap_32(event->auxtrace_error.pid); 893 event->auxtrace_error.tid = bswap_32(event->auxtrace_error.tid); 894 event->auxtrace_error.fmt = bswap_32(event->auxtrace_error.fmt); 895 event->auxtrace_error.ip = bswap_64(event->auxtrace_error.ip); 896 if (event->auxtrace_error.fmt) 897 event->auxtrace_error.time = bswap_64(event->auxtrace_error.time); 898 } 899 900 static void perf_event__thread_map_swap(union perf_event *event, 901 bool sample_id_all __maybe_unused) 902 { 903 unsigned i; 904 905 event->thread_map.nr = bswap_64(event->thread_map.nr); 906 907 for (i = 0; i < event->thread_map.nr; i++) 908 event->thread_map.entries[i].pid = bswap_64(event->thread_map.entries[i].pid); 909 } 910 911 static void perf_event__cpu_map_swap(union perf_event *event, 912 bool sample_id_all __maybe_unused) 913 { 914 struct perf_record_cpu_map_data *data = &event->cpu_map.data; 915 struct cpu_map_entries *cpus; 916 struct perf_record_record_cpu_map *mask; 917 unsigned i; 918 919 data->type = bswap_16(data->type); 920 921 switch (data->type) { 922 case PERF_CPU_MAP__CPUS: 923 cpus = (struct cpu_map_entries *)data->data; 924 925 cpus->nr = bswap_16(cpus->nr); 926 927 for (i = 0; i < cpus->nr; i++) 928 cpus->cpu[i] = bswap_16(cpus->cpu[i]); 929 break; 930 case PERF_CPU_MAP__MASK: 931 mask = (struct perf_record_record_cpu_map *)data->data; 932 933 mask->nr = bswap_16(mask->nr); 934 mask->long_size = bswap_16(mask->long_size); 935 936 switch (mask->long_size) { 937 case 4: mem_bswap_32(&mask->mask, mask->nr); break; 938 case 8: mem_bswap_64(&mask->mask, mask->nr); break; 939 default: 940 pr_err("cpu_map swap: unsupported long size\n"); 941 } 942 default: 943 break; 944 } 945 } 946 947 static void perf_event__stat_config_swap(union perf_event *event, 948 bool sample_id_all __maybe_unused) 949 { 950 u64 size; 951 952 size = bswap_64(event->stat_config.nr) * sizeof(event->stat_config.data[0]); 953 size += 1; /* nr item itself */ 954 mem_bswap_64(&event->stat_config.nr, size); 955 } 956 957 static void perf_event__stat_swap(union perf_event *event, 958 bool sample_id_all __maybe_unused) 959 { 960 event->stat.id = bswap_64(event->stat.id); 961 event->stat.thread = bswap_32(event->stat.thread); 962 event->stat.cpu = bswap_32(event->stat.cpu); 963 event->stat.val = bswap_64(event->stat.val); 964 event->stat.ena = bswap_64(event->stat.ena); 965 event->stat.run = bswap_64(event->stat.run); 966 } 967 968 static void perf_event__stat_round_swap(union perf_event *event, 969 bool sample_id_all __maybe_unused) 970 { 971 event->stat_round.type = bswap_64(event->stat_round.type); 972 event->stat_round.time = bswap_64(event->stat_round.time); 973 } 974 975 static void perf_event__time_conv_swap(union perf_event *event, 976 bool sample_id_all __maybe_unused) 977 { 978 event->time_conv.time_shift = bswap_64(event->time_conv.time_shift); 979 event->time_conv.time_mult = bswap_64(event->time_conv.time_mult); 980 event->time_conv.time_zero = bswap_64(event->time_conv.time_zero); 981 982 if (event_contains(event->time_conv, time_cycles)) { 983 event->time_conv.time_cycles = bswap_64(event->time_conv.time_cycles); 984 event->time_conv.time_mask = bswap_64(event->time_conv.time_mask); 985 } 986 } 987 988 typedef void (*perf_event__swap_op)(union perf_event *event, 989 bool sample_id_all); 990 991 static perf_event__swap_op perf_event__swap_ops[] = { 992 [PERF_RECORD_MMAP] = perf_event__mmap_swap, 993 [PERF_RECORD_MMAP2] = perf_event__mmap2_swap, 994 [PERF_RECORD_COMM] = perf_event__comm_swap, 995 [PERF_RECORD_FORK] = perf_event__task_swap, 996 [PERF_RECORD_EXIT] = perf_event__task_swap, 997 [PERF_RECORD_LOST] = perf_event__all64_swap, 998 [PERF_RECORD_READ] = perf_event__read_swap, 999 [PERF_RECORD_THROTTLE] = perf_event__throttle_swap, 1000 [PERF_RECORD_UNTHROTTLE] = perf_event__throttle_swap, 1001 [PERF_RECORD_SAMPLE] = perf_event__all64_swap, 1002 [PERF_RECORD_AUX] = perf_event__aux_swap, 1003 [PERF_RECORD_ITRACE_START] = perf_event__itrace_start_swap, 1004 [PERF_RECORD_LOST_SAMPLES] = perf_event__all64_swap, 1005 [PERF_RECORD_SWITCH] = perf_event__switch_swap, 1006 [PERF_RECORD_SWITCH_CPU_WIDE] = perf_event__switch_swap, 1007 [PERF_RECORD_NAMESPACES] = perf_event__namespaces_swap, 1008 [PERF_RECORD_CGROUP] = perf_event__cgroup_swap, 1009 [PERF_RECORD_TEXT_POKE] = perf_event__text_poke_swap, 1010 [PERF_RECORD_AUX_OUTPUT_HW_ID] = perf_event__all64_swap, 1011 [PERF_RECORD_HEADER_ATTR] = perf_event__hdr_attr_swap, 1012 [PERF_RECORD_HEADER_EVENT_TYPE] = perf_event__event_type_swap, 1013 [PERF_RECORD_HEADER_TRACING_DATA] = perf_event__tracing_data_swap, 1014 [PERF_RECORD_HEADER_BUILD_ID] = NULL, 1015 [PERF_RECORD_ID_INDEX] = perf_event__all64_swap, 1016 [PERF_RECORD_AUXTRACE_INFO] = perf_event__auxtrace_info_swap, 1017 [PERF_RECORD_AUXTRACE] = perf_event__auxtrace_swap, 1018 [PERF_RECORD_AUXTRACE_ERROR] = perf_event__auxtrace_error_swap, 1019 [PERF_RECORD_THREAD_MAP] = perf_event__thread_map_swap, 1020 [PERF_RECORD_CPU_MAP] = perf_event__cpu_map_swap, 1021 [PERF_RECORD_STAT_CONFIG] = perf_event__stat_config_swap, 1022 [PERF_RECORD_STAT] = perf_event__stat_swap, 1023 [PERF_RECORD_STAT_ROUND] = perf_event__stat_round_swap, 1024 [PERF_RECORD_EVENT_UPDATE] = perf_event__event_update_swap, 1025 [PERF_RECORD_TIME_CONV] = perf_event__time_conv_swap, 1026 [PERF_RECORD_HEADER_MAX] = NULL, 1027 }; 1028 1029 /* 1030 * When perf record finishes a pass on every buffers, it records this pseudo 1031 * event. 1032 * We record the max timestamp t found in the pass n. 1033 * Assuming these timestamps are monotonic across cpus, we know that if 1034 * a buffer still has events with timestamps below t, they will be all 1035 * available and then read in the pass n + 1. 1036 * Hence when we start to read the pass n + 2, we can safely flush every 1037 * events with timestamps below t. 1038 * 1039 * ============ PASS n ================= 1040 * CPU 0 | CPU 1 1041 * | 1042 * cnt1 timestamps | cnt2 timestamps 1043 * 1 | 2 1044 * 2 | 3 1045 * - | 4 <--- max recorded 1046 * 1047 * ============ PASS n + 1 ============== 1048 * CPU 0 | CPU 1 1049 * | 1050 * cnt1 timestamps | cnt2 timestamps 1051 * 3 | 5 1052 * 4 | 6 1053 * 5 | 7 <---- max recorded 1054 * 1055 * Flush every events below timestamp 4 1056 * 1057 * ============ PASS n + 2 ============== 1058 * CPU 0 | CPU 1 1059 * | 1060 * cnt1 timestamps | cnt2 timestamps 1061 * 6 | 8 1062 * 7 | 9 1063 * - | 10 1064 * 1065 * Flush every events below timestamp 7 1066 * etc... 1067 */ 1068 int perf_event__process_finished_round(struct perf_tool *tool __maybe_unused, 1069 union perf_event *event __maybe_unused, 1070 struct ordered_events *oe) 1071 { 1072 if (dump_trace) 1073 fprintf(stdout, "\n"); 1074 return ordered_events__flush(oe, OE_FLUSH__ROUND); 1075 } 1076 1077 int perf_session__queue_event(struct perf_session *s, union perf_event *event, 1078 u64 timestamp, u64 file_offset, const char *file_path) 1079 { 1080 return ordered_events__queue(&s->ordered_events, event, timestamp, file_offset, file_path); 1081 } 1082 1083 static void callchain__lbr_callstack_printf(struct perf_sample *sample) 1084 { 1085 struct ip_callchain *callchain = sample->callchain; 1086 struct branch_stack *lbr_stack = sample->branch_stack; 1087 struct branch_entry *entries = perf_sample__branch_entries(sample); 1088 u64 kernel_callchain_nr = callchain->nr; 1089 unsigned int i; 1090 1091 for (i = 0; i < kernel_callchain_nr; i++) { 1092 if (callchain->ips[i] == PERF_CONTEXT_USER) 1093 break; 1094 } 1095 1096 if ((i != kernel_callchain_nr) && lbr_stack->nr) { 1097 u64 total_nr; 1098 /* 1099 * LBR callstack can only get user call chain, 1100 * i is kernel call chain number, 1101 * 1 is PERF_CONTEXT_USER. 1102 * 1103 * The user call chain is stored in LBR registers. 1104 * LBR are pair registers. The caller is stored 1105 * in "from" register, while the callee is stored 1106 * in "to" register. 1107 * For example, there is a call stack 1108 * "A"->"B"->"C"->"D". 1109 * The LBR registers will be recorded like 1110 * "C"->"D", "B"->"C", "A"->"B". 1111 * So only the first "to" register and all "from" 1112 * registers are needed to construct the whole stack. 1113 */ 1114 total_nr = i + 1 + lbr_stack->nr + 1; 1115 kernel_callchain_nr = i + 1; 1116 1117 printf("... LBR call chain: nr:%" PRIu64 "\n", total_nr); 1118 1119 for (i = 0; i < kernel_callchain_nr; i++) 1120 printf("..... %2d: %016" PRIx64 "\n", 1121 i, callchain->ips[i]); 1122 1123 printf("..... %2d: %016" PRIx64 "\n", 1124 (int)(kernel_callchain_nr), entries[0].to); 1125 for (i = 0; i < lbr_stack->nr; i++) 1126 printf("..... %2d: %016" PRIx64 "\n", 1127 (int)(i + kernel_callchain_nr + 1), entries[i].from); 1128 } 1129 } 1130 1131 static void callchain__printf(struct evsel *evsel, 1132 struct perf_sample *sample) 1133 { 1134 unsigned int i; 1135 struct ip_callchain *callchain = sample->callchain; 1136 1137 if (evsel__has_branch_callstack(evsel)) 1138 callchain__lbr_callstack_printf(sample); 1139 1140 printf("... FP chain: nr:%" PRIu64 "\n", callchain->nr); 1141 1142 for (i = 0; i < callchain->nr; i++) 1143 printf("..... %2d: %016" PRIx64 "\n", 1144 i, callchain->ips[i]); 1145 } 1146 1147 static void branch_stack__printf(struct perf_sample *sample, bool callstack) 1148 { 1149 struct branch_entry *entries = perf_sample__branch_entries(sample); 1150 uint64_t i; 1151 1152 if (!callstack) { 1153 printf("%s: nr:%" PRIu64 "\n", "... branch stack", sample->branch_stack->nr); 1154 } else { 1155 /* the reason of adding 1 to nr is because after expanding 1156 * branch stack it generates nr + 1 callstack records. e.g., 1157 * B()->C() 1158 * A()->B() 1159 * the final callstack should be: 1160 * C() 1161 * B() 1162 * A() 1163 */ 1164 printf("%s: nr:%" PRIu64 "\n", "... branch callstack", sample->branch_stack->nr+1); 1165 } 1166 1167 for (i = 0; i < sample->branch_stack->nr; i++) { 1168 struct branch_entry *e = &entries[i]; 1169 1170 if (!callstack) { 1171 printf("..... %2"PRIu64": %016" PRIx64 " -> %016" PRIx64 " %hu cycles %s%s%s%s %x %s\n", 1172 i, e->from, e->to, 1173 (unsigned short)e->flags.cycles, 1174 e->flags.mispred ? "M" : " ", 1175 e->flags.predicted ? "P" : " ", 1176 e->flags.abort ? "A" : " ", 1177 e->flags.in_tx ? "T" : " ", 1178 (unsigned)e->flags.reserved, 1179 e->flags.type ? branch_type_name(e->flags.type) : ""); 1180 } else { 1181 if (i == 0) { 1182 printf("..... %2"PRIu64": %016" PRIx64 "\n" 1183 "..... %2"PRIu64": %016" PRIx64 "\n", 1184 i, e->to, i+1, e->from); 1185 } else { 1186 printf("..... %2"PRIu64": %016" PRIx64 "\n", i+1, e->from); 1187 } 1188 } 1189 } 1190 } 1191 1192 static void regs_dump__printf(u64 mask, u64 *regs, const char *arch) 1193 { 1194 unsigned rid, i = 0; 1195 1196 for_each_set_bit(rid, (unsigned long *) &mask, sizeof(mask) * 8) { 1197 u64 val = regs[i++]; 1198 1199 printf(".... %-5s 0x%016" PRIx64 "\n", 1200 perf_reg_name(rid, arch), val); 1201 } 1202 } 1203 1204 static const char *regs_abi[] = { 1205 [PERF_SAMPLE_REGS_ABI_NONE] = "none", 1206 [PERF_SAMPLE_REGS_ABI_32] = "32-bit", 1207 [PERF_SAMPLE_REGS_ABI_64] = "64-bit", 1208 }; 1209 1210 static inline const char *regs_dump_abi(struct regs_dump *d) 1211 { 1212 if (d->abi > PERF_SAMPLE_REGS_ABI_64) 1213 return "unknown"; 1214 1215 return regs_abi[d->abi]; 1216 } 1217 1218 static void regs__printf(const char *type, struct regs_dump *regs, const char *arch) 1219 { 1220 u64 mask = regs->mask; 1221 1222 printf("... %s regs: mask 0x%" PRIx64 " ABI %s\n", 1223 type, 1224 mask, 1225 regs_dump_abi(regs)); 1226 1227 regs_dump__printf(mask, regs->regs, arch); 1228 } 1229 1230 static void regs_user__printf(struct perf_sample *sample, const char *arch) 1231 { 1232 struct regs_dump *user_regs = &sample->user_regs; 1233 1234 if (user_regs->regs) 1235 regs__printf("user", user_regs, arch); 1236 } 1237 1238 static void regs_intr__printf(struct perf_sample *sample, const char *arch) 1239 { 1240 struct regs_dump *intr_regs = &sample->intr_regs; 1241 1242 if (intr_regs->regs) 1243 regs__printf("intr", intr_regs, arch); 1244 } 1245 1246 static void stack_user__printf(struct stack_dump *dump) 1247 { 1248 printf("... ustack: size %" PRIu64 ", offset 0x%x\n", 1249 dump->size, dump->offset); 1250 } 1251 1252 static void evlist__print_tstamp(struct evlist *evlist, union perf_event *event, struct perf_sample *sample) 1253 { 1254 u64 sample_type = __evlist__combined_sample_type(evlist); 1255 1256 if (event->header.type != PERF_RECORD_SAMPLE && 1257 !evlist__sample_id_all(evlist)) { 1258 fputs("-1 -1 ", stdout); 1259 return; 1260 } 1261 1262 if ((sample_type & PERF_SAMPLE_CPU)) 1263 printf("%u ", sample->cpu); 1264 1265 if (sample_type & PERF_SAMPLE_TIME) 1266 printf("%" PRIu64 " ", sample->time); 1267 } 1268 1269 static void sample_read__printf(struct perf_sample *sample, u64 read_format) 1270 { 1271 printf("... sample_read:\n"); 1272 1273 if (read_format & PERF_FORMAT_TOTAL_TIME_ENABLED) 1274 printf("...... time enabled %016" PRIx64 "\n", 1275 sample->read.time_enabled); 1276 1277 if (read_format & PERF_FORMAT_TOTAL_TIME_RUNNING) 1278 printf("...... time running %016" PRIx64 "\n", 1279 sample->read.time_running); 1280 1281 if (read_format & PERF_FORMAT_GROUP) { 1282 u64 i; 1283 1284 printf(".... group nr %" PRIu64 "\n", sample->read.group.nr); 1285 1286 for (i = 0; i < sample->read.group.nr; i++) { 1287 struct sample_read_value *value; 1288 1289 value = &sample->read.group.values[i]; 1290 printf("..... id %016" PRIx64 1291 ", value %016" PRIx64 "\n", 1292 value->id, value->value); 1293 } 1294 } else 1295 printf("..... id %016" PRIx64 ", value %016" PRIx64 "\n", 1296 sample->read.one.id, sample->read.one.value); 1297 } 1298 1299 static void dump_event(struct evlist *evlist, union perf_event *event, 1300 u64 file_offset, struct perf_sample *sample, 1301 const char *file_path) 1302 { 1303 if (!dump_trace) 1304 return; 1305 1306 printf("\n%#" PRIx64 "@%s [%#x]: event: %d\n", 1307 file_offset, file_path, event->header.size, event->header.type); 1308 1309 trace_event(event); 1310 if (event->header.type == PERF_RECORD_SAMPLE && evlist->trace_event_sample_raw) 1311 evlist->trace_event_sample_raw(evlist, event, sample); 1312 1313 if (sample) 1314 evlist__print_tstamp(evlist, event, sample); 1315 1316 printf("%#" PRIx64 " [%#x]: PERF_RECORD_%s", file_offset, 1317 event->header.size, perf_event__name(event->header.type)); 1318 } 1319 1320 char *get_page_size_name(u64 size, char *str) 1321 { 1322 if (!size || !unit_number__scnprintf(str, PAGE_SIZE_NAME_LEN, size)) 1323 snprintf(str, PAGE_SIZE_NAME_LEN, "%s", "N/A"); 1324 1325 return str; 1326 } 1327 1328 static void dump_sample(struct evsel *evsel, union perf_event *event, 1329 struct perf_sample *sample, const char *arch) 1330 { 1331 u64 sample_type; 1332 char str[PAGE_SIZE_NAME_LEN]; 1333 1334 if (!dump_trace) 1335 return; 1336 1337 printf("(IP, 0x%x): %d/%d: %#" PRIx64 " period: %" PRIu64 " addr: %#" PRIx64 "\n", 1338 event->header.misc, sample->pid, sample->tid, sample->ip, 1339 sample->period, sample->addr); 1340 1341 sample_type = evsel->core.attr.sample_type; 1342 1343 if (evsel__has_callchain(evsel)) 1344 callchain__printf(evsel, sample); 1345 1346 if (evsel__has_br_stack(evsel)) 1347 branch_stack__printf(sample, evsel__has_branch_callstack(evsel)); 1348 1349 if (sample_type & PERF_SAMPLE_REGS_USER) 1350 regs_user__printf(sample, arch); 1351 1352 if (sample_type & PERF_SAMPLE_REGS_INTR) 1353 regs_intr__printf(sample, arch); 1354 1355 if (sample_type & PERF_SAMPLE_STACK_USER) 1356 stack_user__printf(&sample->user_stack); 1357 1358 if (sample_type & PERF_SAMPLE_WEIGHT_TYPE) { 1359 printf("... weight: %" PRIu64 "", sample->weight); 1360 if (sample_type & PERF_SAMPLE_WEIGHT_STRUCT) { 1361 printf(",0x%"PRIx16"", sample->ins_lat); 1362 printf(",0x%"PRIx16"", sample->p_stage_cyc); 1363 } 1364 printf("\n"); 1365 } 1366 1367 if (sample_type & PERF_SAMPLE_DATA_SRC) 1368 printf(" . data_src: 0x%"PRIx64"\n", sample->data_src); 1369 1370 if (sample_type & PERF_SAMPLE_PHYS_ADDR) 1371 printf(" .. phys_addr: 0x%"PRIx64"\n", sample->phys_addr); 1372 1373 if (sample_type & PERF_SAMPLE_DATA_PAGE_SIZE) 1374 printf(" .. data page size: %s\n", get_page_size_name(sample->data_page_size, str)); 1375 1376 if (sample_type & PERF_SAMPLE_CODE_PAGE_SIZE) 1377 printf(" .. code page size: %s\n", get_page_size_name(sample->code_page_size, str)); 1378 1379 if (sample_type & PERF_SAMPLE_TRANSACTION) 1380 printf("... transaction: %" PRIx64 "\n", sample->transaction); 1381 1382 if (sample_type & PERF_SAMPLE_READ) 1383 sample_read__printf(sample, evsel->core.attr.read_format); 1384 } 1385 1386 static void dump_read(struct evsel *evsel, union perf_event *event) 1387 { 1388 struct perf_record_read *read_event = &event->read; 1389 u64 read_format; 1390 1391 if (!dump_trace) 1392 return; 1393 1394 printf(": %d %d %s %" PRI_lu64 "\n", event->read.pid, event->read.tid, 1395 evsel__name(evsel), event->read.value); 1396 1397 if (!evsel) 1398 return; 1399 1400 read_format = evsel->core.attr.read_format; 1401 1402 if (read_format & PERF_FORMAT_TOTAL_TIME_ENABLED) 1403 printf("... time enabled : %" PRI_lu64 "\n", read_event->time_enabled); 1404 1405 if (read_format & PERF_FORMAT_TOTAL_TIME_RUNNING) 1406 printf("... time running : %" PRI_lu64 "\n", read_event->time_running); 1407 1408 if (read_format & PERF_FORMAT_ID) 1409 printf("... id : %" PRI_lu64 "\n", read_event->id); 1410 } 1411 1412 static struct machine *machines__find_for_cpumode(struct machines *machines, 1413 union perf_event *event, 1414 struct perf_sample *sample) 1415 { 1416 if (perf_guest && 1417 ((sample->cpumode == PERF_RECORD_MISC_GUEST_KERNEL) || 1418 (sample->cpumode == PERF_RECORD_MISC_GUEST_USER))) { 1419 u32 pid; 1420 1421 if (sample->machine_pid) 1422 pid = sample->machine_pid; 1423 else if (event->header.type == PERF_RECORD_MMAP 1424 || event->header.type == PERF_RECORD_MMAP2) 1425 pid = event->mmap.pid; 1426 else 1427 pid = sample->pid; 1428 1429 /* 1430 * Guest code machine is created as needed and does not use 1431 * DEFAULT_GUEST_KERNEL_ID. 1432 */ 1433 if (symbol_conf.guest_code) 1434 return machines__findnew(machines, pid); 1435 1436 return machines__find_guest(machines, pid); 1437 } 1438 1439 return &machines->host; 1440 } 1441 1442 static int deliver_sample_value(struct evlist *evlist, 1443 struct perf_tool *tool, 1444 union perf_event *event, 1445 struct perf_sample *sample, 1446 struct sample_read_value *v, 1447 struct machine *machine) 1448 { 1449 struct perf_sample_id *sid = evlist__id2sid(evlist, v->id); 1450 struct evsel *evsel; 1451 1452 if (sid) { 1453 sample->id = v->id; 1454 sample->period = v->value - sid->period; 1455 sid->period = v->value; 1456 } 1457 1458 if (!sid || sid->evsel == NULL) { 1459 ++evlist->stats.nr_unknown_id; 1460 return 0; 1461 } 1462 1463 /* 1464 * There's no reason to deliver sample 1465 * for zero period, bail out. 1466 */ 1467 if (!sample->period) 1468 return 0; 1469 1470 evsel = container_of(sid->evsel, struct evsel, core); 1471 return tool->sample(tool, event, sample, evsel, machine); 1472 } 1473 1474 static int deliver_sample_group(struct evlist *evlist, 1475 struct perf_tool *tool, 1476 union perf_event *event, 1477 struct perf_sample *sample, 1478 struct machine *machine) 1479 { 1480 int ret = -EINVAL; 1481 u64 i; 1482 1483 for (i = 0; i < sample->read.group.nr; i++) { 1484 ret = deliver_sample_value(evlist, tool, event, sample, 1485 &sample->read.group.values[i], 1486 machine); 1487 if (ret) 1488 break; 1489 } 1490 1491 return ret; 1492 } 1493 1494 static int evlist__deliver_sample(struct evlist *evlist, struct perf_tool *tool, 1495 union perf_event *event, struct perf_sample *sample, 1496 struct evsel *evsel, struct machine *machine) 1497 { 1498 /* We know evsel != NULL. */ 1499 u64 sample_type = evsel->core.attr.sample_type; 1500 u64 read_format = evsel->core.attr.read_format; 1501 1502 /* Standard sample delivery. */ 1503 if (!(sample_type & PERF_SAMPLE_READ)) 1504 return tool->sample(tool, event, sample, evsel, machine); 1505 1506 /* For PERF_SAMPLE_READ we have either single or group mode. */ 1507 if (read_format & PERF_FORMAT_GROUP) 1508 return deliver_sample_group(evlist, tool, event, sample, 1509 machine); 1510 else 1511 return deliver_sample_value(evlist, tool, event, sample, 1512 &sample->read.one, machine); 1513 } 1514 1515 static int machines__deliver_event(struct machines *machines, 1516 struct evlist *evlist, 1517 union perf_event *event, 1518 struct perf_sample *sample, 1519 struct perf_tool *tool, u64 file_offset, 1520 const char *file_path) 1521 { 1522 struct evsel *evsel; 1523 struct machine *machine; 1524 1525 dump_event(evlist, event, file_offset, sample, file_path); 1526 1527 evsel = evlist__id2evsel(evlist, sample->id); 1528 1529 machine = machines__find_for_cpumode(machines, event, sample); 1530 1531 switch (event->header.type) { 1532 case PERF_RECORD_SAMPLE: 1533 if (evsel == NULL) { 1534 ++evlist->stats.nr_unknown_id; 1535 return 0; 1536 } 1537 if (machine == NULL) { 1538 ++evlist->stats.nr_unprocessable_samples; 1539 dump_sample(evsel, event, sample, perf_env__arch(NULL)); 1540 return 0; 1541 } 1542 dump_sample(evsel, event, sample, perf_env__arch(machine->env)); 1543 return evlist__deliver_sample(evlist, tool, event, sample, evsel, machine); 1544 case PERF_RECORD_MMAP: 1545 return tool->mmap(tool, event, sample, machine); 1546 case PERF_RECORD_MMAP2: 1547 if (event->header.misc & PERF_RECORD_MISC_PROC_MAP_PARSE_TIMEOUT) 1548 ++evlist->stats.nr_proc_map_timeout; 1549 return tool->mmap2(tool, event, sample, machine); 1550 case PERF_RECORD_COMM: 1551 return tool->comm(tool, event, sample, machine); 1552 case PERF_RECORD_NAMESPACES: 1553 return tool->namespaces(tool, event, sample, machine); 1554 case PERF_RECORD_CGROUP: 1555 return tool->cgroup(tool, event, sample, machine); 1556 case PERF_RECORD_FORK: 1557 return tool->fork(tool, event, sample, machine); 1558 case PERF_RECORD_EXIT: 1559 return tool->exit(tool, event, sample, machine); 1560 case PERF_RECORD_LOST: 1561 if (tool->lost == perf_event__process_lost) 1562 evlist->stats.total_lost += event->lost.lost; 1563 return tool->lost(tool, event, sample, machine); 1564 case PERF_RECORD_LOST_SAMPLES: 1565 if (tool->lost_samples == perf_event__process_lost_samples) 1566 evlist->stats.total_lost_samples += event->lost_samples.lost; 1567 return tool->lost_samples(tool, event, sample, machine); 1568 case PERF_RECORD_READ: 1569 dump_read(evsel, event); 1570 return tool->read(tool, event, sample, evsel, machine); 1571 case PERF_RECORD_THROTTLE: 1572 return tool->throttle(tool, event, sample, machine); 1573 case PERF_RECORD_UNTHROTTLE: 1574 return tool->unthrottle(tool, event, sample, machine); 1575 case PERF_RECORD_AUX: 1576 if (tool->aux == perf_event__process_aux) { 1577 if (event->aux.flags & PERF_AUX_FLAG_TRUNCATED) 1578 evlist->stats.total_aux_lost += 1; 1579 if (event->aux.flags & PERF_AUX_FLAG_PARTIAL) 1580 evlist->stats.total_aux_partial += 1; 1581 if (event->aux.flags & PERF_AUX_FLAG_COLLISION) 1582 evlist->stats.total_aux_collision += 1; 1583 } 1584 return tool->aux(tool, event, sample, machine); 1585 case PERF_RECORD_ITRACE_START: 1586 return tool->itrace_start(tool, event, sample, machine); 1587 case PERF_RECORD_SWITCH: 1588 case PERF_RECORD_SWITCH_CPU_WIDE: 1589 return tool->context_switch(tool, event, sample, machine); 1590 case PERF_RECORD_KSYMBOL: 1591 return tool->ksymbol(tool, event, sample, machine); 1592 case PERF_RECORD_BPF_EVENT: 1593 return tool->bpf(tool, event, sample, machine); 1594 case PERF_RECORD_TEXT_POKE: 1595 return tool->text_poke(tool, event, sample, machine); 1596 case PERF_RECORD_AUX_OUTPUT_HW_ID: 1597 return tool->aux_output_hw_id(tool, event, sample, machine); 1598 default: 1599 ++evlist->stats.nr_unknown_events; 1600 return -1; 1601 } 1602 } 1603 1604 static int perf_session__deliver_event(struct perf_session *session, 1605 union perf_event *event, 1606 struct perf_tool *tool, 1607 u64 file_offset, 1608 const char *file_path) 1609 { 1610 struct perf_sample sample; 1611 int ret = evlist__parse_sample(session->evlist, event, &sample); 1612 1613 if (ret) { 1614 pr_err("Can't parse sample, err = %d\n", ret); 1615 return ret; 1616 } 1617 1618 ret = auxtrace__process_event(session, event, &sample, tool); 1619 if (ret < 0) 1620 return ret; 1621 if (ret > 0) 1622 return 0; 1623 1624 ret = machines__deliver_event(&session->machines, session->evlist, 1625 event, &sample, tool, file_offset, file_path); 1626 1627 if (dump_trace && sample.aux_sample.size) 1628 auxtrace__dump_auxtrace_sample(session, &sample); 1629 1630 return ret; 1631 } 1632 1633 static s64 perf_session__process_user_event(struct perf_session *session, 1634 union perf_event *event, 1635 u64 file_offset, 1636 const char *file_path) 1637 { 1638 struct ordered_events *oe = &session->ordered_events; 1639 struct perf_tool *tool = session->tool; 1640 struct perf_sample sample = { .time = 0, }; 1641 int fd = perf_data__fd(session->data); 1642 int err; 1643 1644 if (event->header.type != PERF_RECORD_COMPRESSED || 1645 tool->compressed == perf_session__process_compressed_event_stub) 1646 dump_event(session->evlist, event, file_offset, &sample, file_path); 1647 1648 /* These events are processed right away */ 1649 switch (event->header.type) { 1650 case PERF_RECORD_HEADER_ATTR: 1651 err = tool->attr(tool, event, &session->evlist); 1652 if (err == 0) { 1653 perf_session__set_id_hdr_size(session); 1654 perf_session__set_comm_exec(session); 1655 } 1656 return err; 1657 case PERF_RECORD_EVENT_UPDATE: 1658 return tool->event_update(tool, event, &session->evlist); 1659 case PERF_RECORD_HEADER_EVENT_TYPE: 1660 /* 1661 * Deprecated, but we need to handle it for sake 1662 * of old data files create in pipe mode. 1663 */ 1664 return 0; 1665 case PERF_RECORD_HEADER_TRACING_DATA: 1666 /* 1667 * Setup for reading amidst mmap, but only when we 1668 * are in 'file' mode. The 'pipe' fd is in proper 1669 * place already. 1670 */ 1671 if (!perf_data__is_pipe(session->data)) 1672 lseek(fd, file_offset, SEEK_SET); 1673 return tool->tracing_data(session, event); 1674 case PERF_RECORD_HEADER_BUILD_ID: 1675 return tool->build_id(session, event); 1676 case PERF_RECORD_FINISHED_ROUND: 1677 return tool->finished_round(tool, event, oe); 1678 case PERF_RECORD_ID_INDEX: 1679 return tool->id_index(session, event); 1680 case PERF_RECORD_AUXTRACE_INFO: 1681 return tool->auxtrace_info(session, event); 1682 case PERF_RECORD_AUXTRACE: 1683 /* setup for reading amidst mmap */ 1684 lseek(fd, file_offset + event->header.size, SEEK_SET); 1685 return tool->auxtrace(session, event); 1686 case PERF_RECORD_AUXTRACE_ERROR: 1687 perf_session__auxtrace_error_inc(session, event); 1688 return tool->auxtrace_error(session, event); 1689 case PERF_RECORD_THREAD_MAP: 1690 return tool->thread_map(session, event); 1691 case PERF_RECORD_CPU_MAP: 1692 return tool->cpu_map(session, event); 1693 case PERF_RECORD_STAT_CONFIG: 1694 return tool->stat_config(session, event); 1695 case PERF_RECORD_STAT: 1696 return tool->stat(session, event); 1697 case PERF_RECORD_STAT_ROUND: 1698 return tool->stat_round(session, event); 1699 case PERF_RECORD_TIME_CONV: 1700 session->time_conv = event->time_conv; 1701 return tool->time_conv(session, event); 1702 case PERF_RECORD_HEADER_FEATURE: 1703 return tool->feature(session, event); 1704 case PERF_RECORD_COMPRESSED: 1705 err = tool->compressed(session, event, file_offset, file_path); 1706 if (err) 1707 dump_event(session->evlist, event, file_offset, &sample, file_path); 1708 return err; 1709 case PERF_RECORD_FINISHED_INIT: 1710 return tool->finished_init(session, event); 1711 default: 1712 return -EINVAL; 1713 } 1714 } 1715 1716 int perf_session__deliver_synth_event(struct perf_session *session, 1717 union perf_event *event, 1718 struct perf_sample *sample) 1719 { 1720 struct evlist *evlist = session->evlist; 1721 struct perf_tool *tool = session->tool; 1722 1723 events_stats__inc(&evlist->stats, event->header.type); 1724 1725 if (event->header.type >= PERF_RECORD_USER_TYPE_START) 1726 return perf_session__process_user_event(session, event, 0, NULL); 1727 1728 return machines__deliver_event(&session->machines, evlist, event, sample, tool, 0, NULL); 1729 } 1730 1731 static void event_swap(union perf_event *event, bool sample_id_all) 1732 { 1733 perf_event__swap_op swap; 1734 1735 swap = perf_event__swap_ops[event->header.type]; 1736 if (swap) 1737 swap(event, sample_id_all); 1738 } 1739 1740 int perf_session__peek_event(struct perf_session *session, off_t file_offset, 1741 void *buf, size_t buf_sz, 1742 union perf_event **event_ptr, 1743 struct perf_sample *sample) 1744 { 1745 union perf_event *event; 1746 size_t hdr_sz, rest; 1747 int fd; 1748 1749 if (session->one_mmap && !session->header.needs_swap) { 1750 event = file_offset - session->one_mmap_offset + 1751 session->one_mmap_addr; 1752 goto out_parse_sample; 1753 } 1754 1755 if (perf_data__is_pipe(session->data)) 1756 return -1; 1757 1758 fd = perf_data__fd(session->data); 1759 hdr_sz = sizeof(struct perf_event_header); 1760 1761 if (buf_sz < hdr_sz) 1762 return -1; 1763 1764 if (lseek(fd, file_offset, SEEK_SET) == (off_t)-1 || 1765 readn(fd, buf, hdr_sz) != (ssize_t)hdr_sz) 1766 return -1; 1767 1768 event = (union perf_event *)buf; 1769 1770 if (session->header.needs_swap) 1771 perf_event_header__bswap(&event->header); 1772 1773 if (event->header.size < hdr_sz || event->header.size > buf_sz) 1774 return -1; 1775 1776 buf += hdr_sz; 1777 rest = event->header.size - hdr_sz; 1778 1779 if (readn(fd, buf, rest) != (ssize_t)rest) 1780 return -1; 1781 1782 if (session->header.needs_swap) 1783 event_swap(event, evlist__sample_id_all(session->evlist)); 1784 1785 out_parse_sample: 1786 1787 if (sample && event->header.type < PERF_RECORD_USER_TYPE_START && 1788 evlist__parse_sample(session->evlist, event, sample)) 1789 return -1; 1790 1791 *event_ptr = event; 1792 1793 return 0; 1794 } 1795 1796 int perf_session__peek_events(struct perf_session *session, u64 offset, 1797 u64 size, peek_events_cb_t cb, void *data) 1798 { 1799 u64 max_offset = offset + size; 1800 char buf[PERF_SAMPLE_MAX_SIZE]; 1801 union perf_event *event; 1802 int err; 1803 1804 do { 1805 err = perf_session__peek_event(session, offset, buf, 1806 PERF_SAMPLE_MAX_SIZE, &event, 1807 NULL); 1808 if (err) 1809 return err; 1810 1811 err = cb(session, event, offset, data); 1812 if (err) 1813 return err; 1814 1815 offset += event->header.size; 1816 if (event->header.type == PERF_RECORD_AUXTRACE) 1817 offset += event->auxtrace.size; 1818 1819 } while (offset < max_offset); 1820 1821 return err; 1822 } 1823 1824 static s64 perf_session__process_event(struct perf_session *session, 1825 union perf_event *event, u64 file_offset, 1826 const char *file_path) 1827 { 1828 struct evlist *evlist = session->evlist; 1829 struct perf_tool *tool = session->tool; 1830 int ret; 1831 1832 if (session->header.needs_swap) 1833 event_swap(event, evlist__sample_id_all(evlist)); 1834 1835 if (event->header.type >= PERF_RECORD_HEADER_MAX) 1836 return -EINVAL; 1837 1838 events_stats__inc(&evlist->stats, event->header.type); 1839 1840 if (event->header.type >= PERF_RECORD_USER_TYPE_START) 1841 return perf_session__process_user_event(session, event, file_offset, file_path); 1842 1843 if (tool->ordered_events) { 1844 u64 timestamp = -1ULL; 1845 1846 ret = evlist__parse_sample_timestamp(evlist, event, ×tamp); 1847 if (ret && ret != -1) 1848 return ret; 1849 1850 ret = perf_session__queue_event(session, event, timestamp, file_offset, file_path); 1851 if (ret != -ETIME) 1852 return ret; 1853 } 1854 1855 return perf_session__deliver_event(session, event, tool, file_offset, file_path); 1856 } 1857 1858 void perf_event_header__bswap(struct perf_event_header *hdr) 1859 { 1860 hdr->type = bswap_32(hdr->type); 1861 hdr->misc = bswap_16(hdr->misc); 1862 hdr->size = bswap_16(hdr->size); 1863 } 1864 1865 struct thread *perf_session__findnew(struct perf_session *session, pid_t pid) 1866 { 1867 return machine__findnew_thread(&session->machines.host, -1, pid); 1868 } 1869 1870 int perf_session__register_idle_thread(struct perf_session *session) 1871 { 1872 struct thread *thread = machine__idle_thread(&session->machines.host); 1873 1874 /* machine__idle_thread() got the thread, so put it */ 1875 thread__put(thread); 1876 return thread ? 0 : -1; 1877 } 1878 1879 static void 1880 perf_session__warn_order(const struct perf_session *session) 1881 { 1882 const struct ordered_events *oe = &session->ordered_events; 1883 struct evsel *evsel; 1884 bool should_warn = true; 1885 1886 evlist__for_each_entry(session->evlist, evsel) { 1887 if (evsel->core.attr.write_backward) 1888 should_warn = false; 1889 } 1890 1891 if (!should_warn) 1892 return; 1893 if (oe->nr_unordered_events != 0) 1894 ui__warning("%u out of order events recorded.\n", oe->nr_unordered_events); 1895 } 1896 1897 static void perf_session__warn_about_errors(const struct perf_session *session) 1898 { 1899 const struct events_stats *stats = &session->evlist->stats; 1900 1901 if (session->tool->lost == perf_event__process_lost && 1902 stats->nr_events[PERF_RECORD_LOST] != 0) { 1903 ui__warning("Processed %d events and lost %d chunks!\n\n" 1904 "Check IO/CPU overload!\n\n", 1905 stats->nr_events[0], 1906 stats->nr_events[PERF_RECORD_LOST]); 1907 } 1908 1909 if (session->tool->lost_samples == perf_event__process_lost_samples) { 1910 double drop_rate; 1911 1912 drop_rate = (double)stats->total_lost_samples / 1913 (double) (stats->nr_events[PERF_RECORD_SAMPLE] + stats->total_lost_samples); 1914 if (drop_rate > 0.05) { 1915 ui__warning("Processed %" PRIu64 " samples and lost %3.2f%%!\n\n", 1916 stats->nr_events[PERF_RECORD_SAMPLE] + stats->total_lost_samples, 1917 drop_rate * 100.0); 1918 } 1919 } 1920 1921 if (session->tool->aux == perf_event__process_aux && 1922 stats->total_aux_lost != 0) { 1923 ui__warning("AUX data lost %" PRIu64 " times out of %u!\n\n", 1924 stats->total_aux_lost, 1925 stats->nr_events[PERF_RECORD_AUX]); 1926 } 1927 1928 if (session->tool->aux == perf_event__process_aux && 1929 stats->total_aux_partial != 0) { 1930 bool vmm_exclusive = false; 1931 1932 (void)sysfs__read_bool("module/kvm_intel/parameters/vmm_exclusive", 1933 &vmm_exclusive); 1934 1935 ui__warning("AUX data had gaps in it %" PRIu64 " times out of %u!\n\n" 1936 "Are you running a KVM guest in the background?%s\n\n", 1937 stats->total_aux_partial, 1938 stats->nr_events[PERF_RECORD_AUX], 1939 vmm_exclusive ? 1940 "\nReloading kvm_intel module with vmm_exclusive=0\n" 1941 "will reduce the gaps to only guest's timeslices." : 1942 ""); 1943 } 1944 1945 if (session->tool->aux == perf_event__process_aux && 1946 stats->total_aux_collision != 0) { 1947 ui__warning("AUX data detected collision %" PRIu64 " times out of %u!\n\n", 1948 stats->total_aux_collision, 1949 stats->nr_events[PERF_RECORD_AUX]); 1950 } 1951 1952 if (stats->nr_unknown_events != 0) { 1953 ui__warning("Found %u unknown events!\n\n" 1954 "Is this an older tool processing a perf.data " 1955 "file generated by a more recent tool?\n\n" 1956 "If that is not the case, consider " 1957 "reporting to linux-kernel@vger.kernel.org.\n\n", 1958 stats->nr_unknown_events); 1959 } 1960 1961 if (stats->nr_unknown_id != 0) { 1962 ui__warning("%u samples with id not present in the header\n", 1963 stats->nr_unknown_id); 1964 } 1965 1966 if (stats->nr_invalid_chains != 0) { 1967 ui__warning("Found invalid callchains!\n\n" 1968 "%u out of %u events were discarded for this reason.\n\n" 1969 "Consider reporting to linux-kernel@vger.kernel.org.\n\n", 1970 stats->nr_invalid_chains, 1971 stats->nr_events[PERF_RECORD_SAMPLE]); 1972 } 1973 1974 if (stats->nr_unprocessable_samples != 0) { 1975 ui__warning("%u unprocessable samples recorded.\n" 1976 "Do you have a KVM guest running and not using 'perf kvm'?\n", 1977 stats->nr_unprocessable_samples); 1978 } 1979 1980 perf_session__warn_order(session); 1981 1982 events_stats__auxtrace_error_warn(stats); 1983 1984 if (stats->nr_proc_map_timeout != 0) { 1985 ui__warning("%d map information files for pre-existing threads were\n" 1986 "not processed, if there are samples for addresses they\n" 1987 "will not be resolved, you may find out which are these\n" 1988 "threads by running with -v and redirecting the output\n" 1989 "to a file.\n" 1990 "The time limit to process proc map is too short?\n" 1991 "Increase it by --proc-map-timeout\n", 1992 stats->nr_proc_map_timeout); 1993 } 1994 } 1995 1996 static int perf_session__flush_thread_stack(struct thread *thread, 1997 void *p __maybe_unused) 1998 { 1999 return thread_stack__flush(thread); 2000 } 2001 2002 static int perf_session__flush_thread_stacks(struct perf_session *session) 2003 { 2004 return machines__for_each_thread(&session->machines, 2005 perf_session__flush_thread_stack, 2006 NULL); 2007 } 2008 2009 volatile int session_done; 2010 2011 static int __perf_session__process_decomp_events(struct perf_session *session); 2012 2013 static int __perf_session__process_pipe_events(struct perf_session *session) 2014 { 2015 struct ordered_events *oe = &session->ordered_events; 2016 struct perf_tool *tool = session->tool; 2017 union perf_event *event; 2018 uint32_t size, cur_size = 0; 2019 void *buf = NULL; 2020 s64 skip = 0; 2021 u64 head; 2022 ssize_t err; 2023 void *p; 2024 2025 perf_tool__fill_defaults(tool); 2026 2027 head = 0; 2028 cur_size = sizeof(union perf_event); 2029 2030 buf = malloc(cur_size); 2031 if (!buf) 2032 return -errno; 2033 ordered_events__set_copy_on_queue(oe, true); 2034 more: 2035 event = buf; 2036 err = perf_data__read(session->data, event, 2037 sizeof(struct perf_event_header)); 2038 if (err <= 0) { 2039 if (err == 0) 2040 goto done; 2041 2042 pr_err("failed to read event header\n"); 2043 goto out_err; 2044 } 2045 2046 if (session->header.needs_swap) 2047 perf_event_header__bswap(&event->header); 2048 2049 size = event->header.size; 2050 if (size < sizeof(struct perf_event_header)) { 2051 pr_err("bad event header size\n"); 2052 goto out_err; 2053 } 2054 2055 if (size > cur_size) { 2056 void *new = realloc(buf, size); 2057 if (!new) { 2058 pr_err("failed to allocate memory to read event\n"); 2059 goto out_err; 2060 } 2061 buf = new; 2062 cur_size = size; 2063 event = buf; 2064 } 2065 p = event; 2066 p += sizeof(struct perf_event_header); 2067 2068 if (size - sizeof(struct perf_event_header)) { 2069 err = perf_data__read(session->data, p, 2070 size - sizeof(struct perf_event_header)); 2071 if (err <= 0) { 2072 if (err == 0) { 2073 pr_err("unexpected end of event stream\n"); 2074 goto done; 2075 } 2076 2077 pr_err("failed to read event data\n"); 2078 goto out_err; 2079 } 2080 } 2081 2082 if ((skip = perf_session__process_event(session, event, head, "pipe")) < 0) { 2083 pr_err("%#" PRIx64 " [%#x]: failed to process type: %d\n", 2084 head, event->header.size, event->header.type); 2085 err = -EINVAL; 2086 goto out_err; 2087 } 2088 2089 head += size; 2090 2091 if (skip > 0) 2092 head += skip; 2093 2094 err = __perf_session__process_decomp_events(session); 2095 if (err) 2096 goto out_err; 2097 2098 if (!session_done()) 2099 goto more; 2100 done: 2101 /* do the final flush for ordered samples */ 2102 err = ordered_events__flush(oe, OE_FLUSH__FINAL); 2103 if (err) 2104 goto out_err; 2105 err = auxtrace__flush_events(session, tool); 2106 if (err) 2107 goto out_err; 2108 err = perf_session__flush_thread_stacks(session); 2109 out_err: 2110 free(buf); 2111 if (!tool->no_warn) 2112 perf_session__warn_about_errors(session); 2113 ordered_events__free(&session->ordered_events); 2114 auxtrace__free_events(session); 2115 return err; 2116 } 2117 2118 static union perf_event * 2119 prefetch_event(char *buf, u64 head, size_t mmap_size, 2120 bool needs_swap, union perf_event *error) 2121 { 2122 union perf_event *event; 2123 u16 event_size; 2124 2125 /* 2126 * Ensure we have enough space remaining to read 2127 * the size of the event in the headers. 2128 */ 2129 if (head + sizeof(event->header) > mmap_size) 2130 return NULL; 2131 2132 event = (union perf_event *)(buf + head); 2133 if (needs_swap) 2134 perf_event_header__bswap(&event->header); 2135 2136 event_size = event->header.size; 2137 if (head + event_size <= mmap_size) 2138 return event; 2139 2140 /* We're not fetching the event so swap back again */ 2141 if (needs_swap) 2142 perf_event_header__bswap(&event->header); 2143 2144 /* Check if the event fits into the next mmapped buf. */ 2145 if (event_size <= mmap_size - head % page_size) { 2146 /* Remap buf and fetch again. */ 2147 return NULL; 2148 } 2149 2150 /* Invalid input. Event size should never exceed mmap_size. */ 2151 pr_debug("%s: head=%#" PRIx64 " event->header.size=%#x, mmap_size=%#zx:" 2152 " fuzzed or compressed perf.data?\n", __func__, head, event_size, mmap_size); 2153 2154 return error; 2155 } 2156 2157 static union perf_event * 2158 fetch_mmaped_event(u64 head, size_t mmap_size, char *buf, bool needs_swap) 2159 { 2160 return prefetch_event(buf, head, mmap_size, needs_swap, ERR_PTR(-EINVAL)); 2161 } 2162 2163 static union perf_event * 2164 fetch_decomp_event(u64 head, size_t mmap_size, char *buf, bool needs_swap) 2165 { 2166 return prefetch_event(buf, head, mmap_size, needs_swap, NULL); 2167 } 2168 2169 static int __perf_session__process_decomp_events(struct perf_session *session) 2170 { 2171 s64 skip; 2172 u64 size; 2173 struct decomp *decomp = session->active_decomp->decomp_last; 2174 2175 if (!decomp) 2176 return 0; 2177 2178 while (decomp->head < decomp->size && !session_done()) { 2179 union perf_event *event = fetch_decomp_event(decomp->head, decomp->size, decomp->data, 2180 session->header.needs_swap); 2181 2182 if (!event) 2183 break; 2184 2185 size = event->header.size; 2186 2187 if (size < sizeof(struct perf_event_header) || 2188 (skip = perf_session__process_event(session, event, decomp->file_pos, 2189 decomp->file_path)) < 0) { 2190 pr_err("%#" PRIx64 " [%#x]: failed to process type: %d\n", 2191 decomp->file_pos + decomp->head, event->header.size, event->header.type); 2192 return -EINVAL; 2193 } 2194 2195 if (skip) 2196 size += skip; 2197 2198 decomp->head += size; 2199 } 2200 2201 return 0; 2202 } 2203 2204 /* 2205 * On 64bit we can mmap the data file in one go. No need for tiny mmap 2206 * slices. On 32bit we use 32MB. 2207 */ 2208 #if BITS_PER_LONG == 64 2209 #define MMAP_SIZE ULLONG_MAX 2210 #define NUM_MMAPS 1 2211 #else 2212 #define MMAP_SIZE (32 * 1024 * 1024ULL) 2213 #define NUM_MMAPS 128 2214 #endif 2215 2216 struct reader; 2217 2218 typedef s64 (*reader_cb_t)(struct perf_session *session, 2219 union perf_event *event, 2220 u64 file_offset, 2221 const char *file_path); 2222 2223 struct reader { 2224 int fd; 2225 const char *path; 2226 u64 data_size; 2227 u64 data_offset; 2228 reader_cb_t process; 2229 bool in_place_update; 2230 char *mmaps[NUM_MMAPS]; 2231 size_t mmap_size; 2232 int mmap_idx; 2233 char *mmap_cur; 2234 u64 file_pos; 2235 u64 file_offset; 2236 u64 head; 2237 u64 size; 2238 bool done; 2239 struct zstd_data zstd_data; 2240 struct decomp_data decomp_data; 2241 }; 2242 2243 static int 2244 reader__init(struct reader *rd, bool *one_mmap) 2245 { 2246 u64 data_size = rd->data_size; 2247 char **mmaps = rd->mmaps; 2248 2249 rd->head = rd->data_offset; 2250 data_size += rd->data_offset; 2251 2252 rd->mmap_size = MMAP_SIZE; 2253 if (rd->mmap_size > data_size) { 2254 rd->mmap_size = data_size; 2255 if (one_mmap) 2256 *one_mmap = true; 2257 } 2258 2259 memset(mmaps, 0, sizeof(rd->mmaps)); 2260 2261 if (zstd_init(&rd->zstd_data, 0)) 2262 return -1; 2263 rd->decomp_data.zstd_decomp = &rd->zstd_data; 2264 2265 return 0; 2266 } 2267 2268 static void 2269 reader__release_decomp(struct reader *rd) 2270 { 2271 perf_decomp__release_events(rd->decomp_data.decomp); 2272 zstd_fini(&rd->zstd_data); 2273 } 2274 2275 static int 2276 reader__mmap(struct reader *rd, struct perf_session *session) 2277 { 2278 int mmap_prot, mmap_flags; 2279 char *buf, **mmaps = rd->mmaps; 2280 u64 page_offset; 2281 2282 mmap_prot = PROT_READ; 2283 mmap_flags = MAP_SHARED; 2284 2285 if (rd->in_place_update) { 2286 mmap_prot |= PROT_WRITE; 2287 } else if (session->header.needs_swap) { 2288 mmap_prot |= PROT_WRITE; 2289 mmap_flags = MAP_PRIVATE; 2290 } 2291 2292 if (mmaps[rd->mmap_idx]) { 2293 munmap(mmaps[rd->mmap_idx], rd->mmap_size); 2294 mmaps[rd->mmap_idx] = NULL; 2295 } 2296 2297 page_offset = page_size * (rd->head / page_size); 2298 rd->file_offset += page_offset; 2299 rd->head -= page_offset; 2300 2301 buf = mmap(NULL, rd->mmap_size, mmap_prot, mmap_flags, rd->fd, 2302 rd->file_offset); 2303 if (buf == MAP_FAILED) { 2304 pr_err("failed to mmap file\n"); 2305 return -errno; 2306 } 2307 mmaps[rd->mmap_idx] = rd->mmap_cur = buf; 2308 rd->mmap_idx = (rd->mmap_idx + 1) & (ARRAY_SIZE(rd->mmaps) - 1); 2309 rd->file_pos = rd->file_offset + rd->head; 2310 if (session->one_mmap) { 2311 session->one_mmap_addr = buf; 2312 session->one_mmap_offset = rd->file_offset; 2313 } 2314 2315 return 0; 2316 } 2317 2318 enum { 2319 READER_OK, 2320 READER_NODATA, 2321 }; 2322 2323 static int 2324 reader__read_event(struct reader *rd, struct perf_session *session, 2325 struct ui_progress *prog) 2326 { 2327 u64 size; 2328 int err = READER_OK; 2329 union perf_event *event; 2330 s64 skip; 2331 2332 event = fetch_mmaped_event(rd->head, rd->mmap_size, rd->mmap_cur, 2333 session->header.needs_swap); 2334 if (IS_ERR(event)) 2335 return PTR_ERR(event); 2336 2337 if (!event) 2338 return READER_NODATA; 2339 2340 size = event->header.size; 2341 2342 skip = -EINVAL; 2343 2344 if (size < sizeof(struct perf_event_header) || 2345 (skip = rd->process(session, event, rd->file_pos, rd->path)) < 0) { 2346 pr_err("%#" PRIx64 " [%#x]: failed to process type: %d [%s]\n", 2347 rd->file_offset + rd->head, event->header.size, 2348 event->header.type, strerror(-skip)); 2349 err = skip; 2350 goto out; 2351 } 2352 2353 if (skip) 2354 size += skip; 2355 2356 rd->size += size; 2357 rd->head += size; 2358 rd->file_pos += size; 2359 2360 err = __perf_session__process_decomp_events(session); 2361 if (err) 2362 goto out; 2363 2364 ui_progress__update(prog, size); 2365 2366 out: 2367 return err; 2368 } 2369 2370 static inline bool 2371 reader__eof(struct reader *rd) 2372 { 2373 return (rd->file_pos >= rd->data_size + rd->data_offset); 2374 } 2375 2376 static int 2377 reader__process_events(struct reader *rd, struct perf_session *session, 2378 struct ui_progress *prog) 2379 { 2380 int err; 2381 2382 err = reader__init(rd, &session->one_mmap); 2383 if (err) 2384 goto out; 2385 2386 session->active_decomp = &rd->decomp_data; 2387 2388 remap: 2389 err = reader__mmap(rd, session); 2390 if (err) 2391 goto out; 2392 2393 more: 2394 err = reader__read_event(rd, session, prog); 2395 if (err < 0) 2396 goto out; 2397 else if (err == READER_NODATA) 2398 goto remap; 2399 2400 if (session_done()) 2401 goto out; 2402 2403 if (!reader__eof(rd)) 2404 goto more; 2405 2406 out: 2407 session->active_decomp = &session->decomp_data; 2408 return err; 2409 } 2410 2411 static s64 process_simple(struct perf_session *session, 2412 union perf_event *event, 2413 u64 file_offset, 2414 const char *file_path) 2415 { 2416 return perf_session__process_event(session, event, file_offset, file_path); 2417 } 2418 2419 static int __perf_session__process_events(struct perf_session *session) 2420 { 2421 struct reader rd = { 2422 .fd = perf_data__fd(session->data), 2423 .path = session->data->file.path, 2424 .data_size = session->header.data_size, 2425 .data_offset = session->header.data_offset, 2426 .process = process_simple, 2427 .in_place_update = session->data->in_place_update, 2428 }; 2429 struct ordered_events *oe = &session->ordered_events; 2430 struct perf_tool *tool = session->tool; 2431 struct ui_progress prog; 2432 int err; 2433 2434 perf_tool__fill_defaults(tool); 2435 2436 if (rd.data_size == 0) 2437 return -1; 2438 2439 ui_progress__init_size(&prog, rd.data_size, "Processing events..."); 2440 2441 err = reader__process_events(&rd, session, &prog); 2442 if (err) 2443 goto out_err; 2444 /* do the final flush for ordered samples */ 2445 err = ordered_events__flush(oe, OE_FLUSH__FINAL); 2446 if (err) 2447 goto out_err; 2448 err = auxtrace__flush_events(session, tool); 2449 if (err) 2450 goto out_err; 2451 err = perf_session__flush_thread_stacks(session); 2452 out_err: 2453 ui_progress__finish(); 2454 if (!tool->no_warn) 2455 perf_session__warn_about_errors(session); 2456 /* 2457 * We may switching perf.data output, make ordered_events 2458 * reusable. 2459 */ 2460 ordered_events__reinit(&session->ordered_events); 2461 auxtrace__free_events(session); 2462 reader__release_decomp(&rd); 2463 session->one_mmap = false; 2464 return err; 2465 } 2466 2467 /* 2468 * Processing 2 MB of data from each reader in sequence, 2469 * because that's the way the ordered events sorting works 2470 * most efficiently. 2471 */ 2472 #define READER_MAX_SIZE (2 * 1024 * 1024) 2473 2474 /* 2475 * This function reads, merge and process directory data. 2476 * It assumens the version 1 of directory data, where each 2477 * data file holds per-cpu data, already sorted by kernel. 2478 */ 2479 static int __perf_session__process_dir_events(struct perf_session *session) 2480 { 2481 struct perf_data *data = session->data; 2482 struct perf_tool *tool = session->tool; 2483 int i, ret, readers, nr_readers; 2484 struct ui_progress prog; 2485 u64 total_size = perf_data__size(session->data); 2486 struct reader *rd; 2487 2488 perf_tool__fill_defaults(tool); 2489 2490 ui_progress__init_size(&prog, total_size, "Sorting events..."); 2491 2492 nr_readers = 1; 2493 for (i = 0; i < data->dir.nr; i++) { 2494 if (data->dir.files[i].size) 2495 nr_readers++; 2496 } 2497 2498 rd = zalloc(nr_readers * sizeof(struct reader)); 2499 if (!rd) 2500 return -ENOMEM; 2501 2502 rd[0] = (struct reader) { 2503 .fd = perf_data__fd(session->data), 2504 .path = session->data->file.path, 2505 .data_size = session->header.data_size, 2506 .data_offset = session->header.data_offset, 2507 .process = process_simple, 2508 .in_place_update = session->data->in_place_update, 2509 }; 2510 ret = reader__init(&rd[0], NULL); 2511 if (ret) 2512 goto out_err; 2513 ret = reader__mmap(&rd[0], session); 2514 if (ret) 2515 goto out_err; 2516 readers = 1; 2517 2518 for (i = 0; i < data->dir.nr; i++) { 2519 if (!data->dir.files[i].size) 2520 continue; 2521 rd[readers] = (struct reader) { 2522 .fd = data->dir.files[i].fd, 2523 .path = data->dir.files[i].path, 2524 .data_size = data->dir.files[i].size, 2525 .data_offset = 0, 2526 .process = process_simple, 2527 .in_place_update = session->data->in_place_update, 2528 }; 2529 ret = reader__init(&rd[readers], NULL); 2530 if (ret) 2531 goto out_err; 2532 ret = reader__mmap(&rd[readers], session); 2533 if (ret) 2534 goto out_err; 2535 readers++; 2536 } 2537 2538 i = 0; 2539 while (readers) { 2540 if (session_done()) 2541 break; 2542 2543 if (rd[i].done) { 2544 i = (i + 1) % nr_readers; 2545 continue; 2546 } 2547 if (reader__eof(&rd[i])) { 2548 rd[i].done = true; 2549 readers--; 2550 continue; 2551 } 2552 2553 session->active_decomp = &rd[i].decomp_data; 2554 ret = reader__read_event(&rd[i], session, &prog); 2555 if (ret < 0) { 2556 goto out_err; 2557 } else if (ret == READER_NODATA) { 2558 ret = reader__mmap(&rd[i], session); 2559 if (ret) 2560 goto out_err; 2561 } 2562 2563 if (rd[i].size >= READER_MAX_SIZE) { 2564 rd[i].size = 0; 2565 i = (i + 1) % nr_readers; 2566 } 2567 } 2568 2569 ret = ordered_events__flush(&session->ordered_events, OE_FLUSH__FINAL); 2570 if (ret) 2571 goto out_err; 2572 2573 ret = perf_session__flush_thread_stacks(session); 2574 out_err: 2575 ui_progress__finish(); 2576 2577 if (!tool->no_warn) 2578 perf_session__warn_about_errors(session); 2579 2580 /* 2581 * We may switching perf.data output, make ordered_events 2582 * reusable. 2583 */ 2584 ordered_events__reinit(&session->ordered_events); 2585 2586 session->one_mmap = false; 2587 2588 session->active_decomp = &session->decomp_data; 2589 for (i = 0; i < nr_readers; i++) 2590 reader__release_decomp(&rd[i]); 2591 zfree(&rd); 2592 2593 return ret; 2594 } 2595 2596 int perf_session__process_events(struct perf_session *session) 2597 { 2598 if (perf_session__register_idle_thread(session) < 0) 2599 return -ENOMEM; 2600 2601 if (perf_data__is_pipe(session->data)) 2602 return __perf_session__process_pipe_events(session); 2603 2604 if (perf_data__is_dir(session->data) && session->data->dir.nr) 2605 return __perf_session__process_dir_events(session); 2606 2607 return __perf_session__process_events(session); 2608 } 2609 2610 bool perf_session__has_traces(struct perf_session *session, const char *msg) 2611 { 2612 struct evsel *evsel; 2613 2614 evlist__for_each_entry(session->evlist, evsel) { 2615 if (evsel->core.attr.type == PERF_TYPE_TRACEPOINT) 2616 return true; 2617 } 2618 2619 pr_err("No trace sample to read. Did you call 'perf %s'?\n", msg); 2620 return false; 2621 } 2622 2623 int map__set_kallsyms_ref_reloc_sym(struct map *map, const char *symbol_name, u64 addr) 2624 { 2625 char *bracket; 2626 struct ref_reloc_sym *ref; 2627 struct kmap *kmap; 2628 2629 ref = zalloc(sizeof(struct ref_reloc_sym)); 2630 if (ref == NULL) 2631 return -ENOMEM; 2632 2633 ref->name = strdup(symbol_name); 2634 if (ref->name == NULL) { 2635 free(ref); 2636 return -ENOMEM; 2637 } 2638 2639 bracket = strchr(ref->name, ']'); 2640 if (bracket) 2641 *bracket = '\0'; 2642 2643 ref->addr = addr; 2644 2645 kmap = map__kmap(map); 2646 if (kmap) 2647 kmap->ref_reloc_sym = ref; 2648 2649 return 0; 2650 } 2651 2652 size_t perf_session__fprintf_dsos(struct perf_session *session, FILE *fp) 2653 { 2654 return machines__fprintf_dsos(&session->machines, fp); 2655 } 2656 2657 size_t perf_session__fprintf_dsos_buildid(struct perf_session *session, FILE *fp, 2658 bool (skip)(struct dso *dso, int parm), int parm) 2659 { 2660 return machines__fprintf_dsos_buildid(&session->machines, fp, skip, parm); 2661 } 2662 2663 size_t perf_session__fprintf_nr_events(struct perf_session *session, FILE *fp, 2664 bool skip_empty) 2665 { 2666 size_t ret; 2667 const char *msg = ""; 2668 2669 if (perf_header__has_feat(&session->header, HEADER_AUXTRACE)) 2670 msg = " (excludes AUX area (e.g. instruction trace) decoded / synthesized events)"; 2671 2672 ret = fprintf(fp, "\nAggregated stats:%s\n", msg); 2673 2674 ret += events_stats__fprintf(&session->evlist->stats, fp, skip_empty); 2675 return ret; 2676 } 2677 2678 size_t perf_session__fprintf(struct perf_session *session, FILE *fp) 2679 { 2680 /* 2681 * FIXME: Here we have to actually print all the machines in this 2682 * session, not just the host... 2683 */ 2684 return machine__fprintf(&session->machines.host, fp); 2685 } 2686 2687 struct evsel *perf_session__find_first_evtype(struct perf_session *session, 2688 unsigned int type) 2689 { 2690 struct evsel *pos; 2691 2692 evlist__for_each_entry(session->evlist, pos) { 2693 if (pos->core.attr.type == type) 2694 return pos; 2695 } 2696 return NULL; 2697 } 2698 2699 int perf_session__cpu_bitmap(struct perf_session *session, 2700 const char *cpu_list, unsigned long *cpu_bitmap) 2701 { 2702 int i, err = -1; 2703 struct perf_cpu_map *map; 2704 int nr_cpus = min(session->header.env.nr_cpus_avail, MAX_NR_CPUS); 2705 2706 for (i = 0; i < PERF_TYPE_MAX; ++i) { 2707 struct evsel *evsel; 2708 2709 evsel = perf_session__find_first_evtype(session, i); 2710 if (!evsel) 2711 continue; 2712 2713 if (!(evsel->core.attr.sample_type & PERF_SAMPLE_CPU)) { 2714 pr_err("File does not contain CPU events. " 2715 "Remove -C option to proceed.\n"); 2716 return -1; 2717 } 2718 } 2719 2720 map = perf_cpu_map__new(cpu_list); 2721 if (map == NULL) { 2722 pr_err("Invalid cpu_list\n"); 2723 return -1; 2724 } 2725 2726 for (i = 0; i < perf_cpu_map__nr(map); i++) { 2727 struct perf_cpu cpu = perf_cpu_map__cpu(map, i); 2728 2729 if (cpu.cpu >= nr_cpus) { 2730 pr_err("Requested CPU %d too large. " 2731 "Consider raising MAX_NR_CPUS\n", cpu.cpu); 2732 goto out_delete_map; 2733 } 2734 2735 set_bit(cpu.cpu, cpu_bitmap); 2736 } 2737 2738 err = 0; 2739 2740 out_delete_map: 2741 perf_cpu_map__put(map); 2742 return err; 2743 } 2744 2745 void perf_session__fprintf_info(struct perf_session *session, FILE *fp, 2746 bool full) 2747 { 2748 if (session == NULL || fp == NULL) 2749 return; 2750 2751 fprintf(fp, "# ========\n"); 2752 perf_header__fprintf_info(session, fp, full); 2753 fprintf(fp, "# ========\n#\n"); 2754 } 2755 2756 static int perf_session__register_guest(struct perf_session *session, pid_t machine_pid) 2757 { 2758 struct machine *machine = machines__findnew(&session->machines, machine_pid); 2759 struct thread *thread; 2760 2761 if (!machine) 2762 return -ENOMEM; 2763 2764 machine->single_address_space = session->machines.host.single_address_space; 2765 2766 thread = machine__idle_thread(machine); 2767 if (!thread) 2768 return -ENOMEM; 2769 thread__put(thread); 2770 2771 return 0; 2772 } 2773 2774 static int perf_session__set_guest_cpu(struct perf_session *session, pid_t pid, 2775 pid_t tid, int guest_cpu) 2776 { 2777 struct machine *machine = &session->machines.host; 2778 struct thread *thread = machine__findnew_thread(machine, pid, tid); 2779 2780 if (!thread) 2781 return -ENOMEM; 2782 thread->guest_cpu = guest_cpu; 2783 thread__put(thread); 2784 2785 return 0; 2786 } 2787 2788 int perf_event__process_id_index(struct perf_session *session, 2789 union perf_event *event) 2790 { 2791 struct evlist *evlist = session->evlist; 2792 struct perf_record_id_index *ie = &event->id_index; 2793 size_t sz = ie->header.size - sizeof(*ie); 2794 size_t i, nr, max_nr; 2795 size_t e1_sz = sizeof(struct id_index_entry); 2796 size_t e2_sz = sizeof(struct id_index_entry_2); 2797 size_t etot_sz = e1_sz + e2_sz; 2798 struct id_index_entry_2 *e2; 2799 pid_t last_pid = 0; 2800 2801 max_nr = sz / e1_sz; 2802 nr = ie->nr; 2803 if (nr > max_nr) { 2804 printf("Too big: nr %zu max_nr %zu\n", nr, max_nr); 2805 return -EINVAL; 2806 } 2807 2808 if (sz >= nr * etot_sz) { 2809 max_nr = sz / etot_sz; 2810 if (nr > max_nr) { 2811 printf("Too big2: nr %zu max_nr %zu\n", nr, max_nr); 2812 return -EINVAL; 2813 } 2814 e2 = (void *)ie + sizeof(*ie) + nr * e1_sz; 2815 } else { 2816 e2 = NULL; 2817 } 2818 2819 if (dump_trace) 2820 fprintf(stdout, " nr: %zu\n", nr); 2821 2822 for (i = 0; i < nr; i++, (e2 ? e2++ : 0)) { 2823 struct id_index_entry *e = &ie->entries[i]; 2824 struct perf_sample_id *sid; 2825 int ret; 2826 2827 if (dump_trace) { 2828 fprintf(stdout, " ... id: %"PRI_lu64, e->id); 2829 fprintf(stdout, " idx: %"PRI_lu64, e->idx); 2830 fprintf(stdout, " cpu: %"PRI_ld64, e->cpu); 2831 fprintf(stdout, " tid: %"PRI_ld64, e->tid); 2832 if (e2) { 2833 fprintf(stdout, " machine_pid: %"PRI_ld64, e2->machine_pid); 2834 fprintf(stdout, " vcpu: %"PRI_lu64"\n", e2->vcpu); 2835 } else { 2836 fprintf(stdout, "\n"); 2837 } 2838 } 2839 2840 sid = evlist__id2sid(evlist, e->id); 2841 if (!sid) 2842 return -ENOENT; 2843 2844 sid->idx = e->idx; 2845 sid->cpu.cpu = e->cpu; 2846 sid->tid = e->tid; 2847 2848 if (!e2) 2849 continue; 2850 2851 sid->machine_pid = e2->machine_pid; 2852 sid->vcpu.cpu = e2->vcpu; 2853 2854 if (!sid->machine_pid) 2855 continue; 2856 2857 if (sid->machine_pid != last_pid) { 2858 ret = perf_session__register_guest(session, sid->machine_pid); 2859 if (ret) 2860 return ret; 2861 last_pid = sid->machine_pid; 2862 perf_guest = true; 2863 } 2864 2865 ret = perf_session__set_guest_cpu(session, sid->machine_pid, e->tid, e2->vcpu); 2866 if (ret) 2867 return ret; 2868 } 2869 return 0; 2870 } 2871