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 "util.h" 37 #include "arch/common.h" 38 #include "units.h" 39 #include "annotate.h" 40 #include <internal/lib.h> 41 42 #ifdef HAVE_ZSTD_SUPPORT 43 static int perf_session__process_compressed_event(struct perf_session *session, 44 union perf_event *event, u64 file_offset, 45 const char *file_path) 46 { 47 void *src; 48 size_t decomp_size, src_size; 49 u64 decomp_last_rem = 0; 50 size_t mmap_len, decomp_len = session->header.env.comp_mmap_len; 51 struct decomp *decomp, *decomp_last = session->active_decomp->decomp_last; 52 53 if (decomp_last) { 54 decomp_last_rem = decomp_last->size - decomp_last->head; 55 decomp_len += decomp_last_rem; 56 } 57 58 mmap_len = sizeof(struct decomp) + decomp_len; 59 decomp = mmap(NULL, mmap_len, PROT_READ|PROT_WRITE, 60 MAP_ANONYMOUS|MAP_PRIVATE, -1, 0); 61 if (decomp == MAP_FAILED) { 62 pr_err("Couldn't allocate memory for decompression\n"); 63 return -1; 64 } 65 66 decomp->file_pos = file_offset; 67 decomp->file_path = file_path; 68 decomp->mmap_len = mmap_len; 69 decomp->head = 0; 70 71 if (decomp_last_rem) { 72 memcpy(decomp->data, &(decomp_last->data[decomp_last->head]), decomp_last_rem); 73 decomp->size = decomp_last_rem; 74 } 75 76 src = (void *)event + sizeof(struct perf_record_compressed); 77 src_size = event->pack.header.size - sizeof(struct perf_record_compressed); 78 79 decomp_size = zstd_decompress_stream(session->active_decomp->zstd_decomp, src, src_size, 80 &(decomp->data[decomp_last_rem]), decomp_len - decomp_last_rem); 81 if (!decomp_size) { 82 munmap(decomp, mmap_len); 83 pr_err("Couldn't decompress data\n"); 84 return -1; 85 } 86 87 decomp->size += decomp_size; 88 89 if (session->active_decomp->decomp == NULL) 90 session->active_decomp->decomp = decomp; 91 else 92 session->active_decomp->decomp_last->next = decomp; 93 94 session->active_decomp->decomp_last = decomp; 95 96 pr_debug("decomp (B): %zd to %zd\n", src_size, decomp_size); 97 98 return 0; 99 } 100 #else /* !HAVE_ZSTD_SUPPORT */ 101 #define perf_session__process_compressed_event perf_session__process_compressed_event_stub 102 #endif 103 104 static int perf_session__deliver_event(struct perf_session *session, 105 union perf_event *event, 106 struct perf_tool *tool, 107 u64 file_offset, 108 const char *file_path); 109 110 static int perf_session__open(struct perf_session *session, int repipe_fd) 111 { 112 struct perf_data *data = session->data; 113 114 if (perf_session__read_header(session, repipe_fd) < 0) { 115 pr_err("incompatible file format (rerun with -v to learn more)\n"); 116 return -1; 117 } 118 119 if (perf_header__has_feat(&session->header, HEADER_AUXTRACE)) { 120 /* Auxiliary events may reference exited threads, hold onto dead ones. */ 121 symbol_conf.keep_exited_threads = true; 122 } 123 124 if (perf_data__is_pipe(data)) 125 return 0; 126 127 if (perf_header__has_feat(&session->header, HEADER_STAT)) 128 return 0; 129 130 if (!evlist__valid_sample_type(session->evlist)) { 131 pr_err("non matching sample_type\n"); 132 return -1; 133 } 134 135 if (!evlist__valid_sample_id_all(session->evlist)) { 136 pr_err("non matching sample_id_all\n"); 137 return -1; 138 } 139 140 if (!evlist__valid_read_format(session->evlist)) { 141 pr_err("non matching read_format\n"); 142 return -1; 143 } 144 145 return 0; 146 } 147 148 void perf_session__set_id_hdr_size(struct perf_session *session) 149 { 150 u16 id_hdr_size = evlist__id_hdr_size(session->evlist); 151 152 machines__set_id_hdr_size(&session->machines, id_hdr_size); 153 } 154 155 int perf_session__create_kernel_maps(struct perf_session *session) 156 { 157 int ret = machine__create_kernel_maps(&session->machines.host); 158 159 if (ret >= 0) 160 ret = machines__create_guest_kernel_maps(&session->machines); 161 return ret; 162 } 163 164 static void perf_session__destroy_kernel_maps(struct perf_session *session) 165 { 166 machines__destroy_kernel_maps(&session->machines); 167 } 168 169 static bool perf_session__has_comm_exec(struct perf_session *session) 170 { 171 struct evsel *evsel; 172 173 evlist__for_each_entry(session->evlist, evsel) { 174 if (evsel->core.attr.comm_exec) 175 return true; 176 } 177 178 return false; 179 } 180 181 static void perf_session__set_comm_exec(struct perf_session *session) 182 { 183 bool comm_exec = perf_session__has_comm_exec(session); 184 185 machines__set_comm_exec(&session->machines, comm_exec); 186 } 187 188 static int ordered_events__deliver_event(struct ordered_events *oe, 189 struct ordered_event *event) 190 { 191 struct perf_session *session = container_of(oe, struct perf_session, 192 ordered_events); 193 194 return perf_session__deliver_event(session, event->event, 195 session->tool, event->file_offset, 196 event->file_path); 197 } 198 199 struct perf_session *__perf_session__new(struct perf_data *data, 200 bool repipe, int repipe_fd, 201 struct perf_tool *tool) 202 { 203 int ret = -ENOMEM; 204 struct perf_session *session = zalloc(sizeof(*session)); 205 206 if (!session) 207 goto out; 208 209 session->repipe = repipe; 210 session->tool = tool; 211 session->decomp_data.zstd_decomp = &session->zstd_data; 212 session->active_decomp = &session->decomp_data; 213 INIT_LIST_HEAD(&session->auxtrace_index); 214 machines__init(&session->machines); 215 ordered_events__init(&session->ordered_events, 216 ordered_events__deliver_event, NULL); 217 218 perf_env__init(&session->header.env); 219 if (data) { 220 ret = perf_data__open(data); 221 if (ret < 0) 222 goto out_delete; 223 224 session->data = data; 225 226 if (perf_data__is_read(data)) { 227 ret = perf_session__open(session, repipe_fd); 228 if (ret < 0) 229 goto out_delete; 230 231 /* 232 * set session attributes that are present in perf.data 233 * but not in pipe-mode. 234 */ 235 if (!data->is_pipe) { 236 perf_session__set_id_hdr_size(session); 237 perf_session__set_comm_exec(session); 238 } 239 240 evlist__init_trace_event_sample_raw(session->evlist); 241 242 /* Open the directory data. */ 243 if (data->is_dir) { 244 ret = perf_data__open_dir(data); 245 if (ret) 246 goto out_delete; 247 } 248 249 if (!symbol_conf.kallsyms_name && 250 !symbol_conf.vmlinux_name) 251 symbol_conf.kallsyms_name = perf_data__kallsyms_name(data); 252 } 253 } else { 254 session->machines.host.env = &perf_env; 255 } 256 257 session->machines.host.single_address_space = 258 perf_env__single_address_space(session->machines.host.env); 259 260 if (!data || perf_data__is_write(data)) { 261 /* 262 * In O_RDONLY mode this will be performed when reading the 263 * kernel MMAP event, in perf_event__process_mmap(). 264 */ 265 if (perf_session__create_kernel_maps(session) < 0) 266 pr_warning("Cannot read kernel map\n"); 267 } 268 269 /* 270 * In pipe-mode, evlist is empty until PERF_RECORD_HEADER_ATTR is 271 * processed, so evlist__sample_id_all is not meaningful here. 272 */ 273 if ((!data || !data->is_pipe) && tool && tool->ordering_requires_timestamps && 274 tool->ordered_events && !evlist__sample_id_all(session->evlist)) { 275 dump_printf("WARNING: No sample_id_all support, falling back to unordered processing\n"); 276 tool->ordered_events = false; 277 } 278 279 return session; 280 281 out_delete: 282 perf_session__delete(session); 283 out: 284 return ERR_PTR(ret); 285 } 286 287 static void perf_decomp__release_events(struct decomp *next) 288 { 289 struct decomp *decomp; 290 size_t mmap_len; 291 292 do { 293 decomp = next; 294 if (decomp == NULL) 295 break; 296 next = decomp->next; 297 mmap_len = decomp->mmap_len; 298 munmap(decomp, mmap_len); 299 } while (1); 300 } 301 302 void perf_session__delete(struct perf_session *session) 303 { 304 if (session == NULL) 305 return; 306 auxtrace__free(session); 307 auxtrace_index__free(&session->auxtrace_index); 308 debuginfo_cache__delete(); 309 perf_session__destroy_kernel_maps(session); 310 perf_decomp__release_events(session->decomp_data.decomp); 311 perf_env__exit(&session->header.env); 312 machines__exit(&session->machines); 313 if (session->data) { 314 if (perf_data__is_read(session->data)) 315 evlist__delete(session->evlist); 316 perf_data__close(session->data); 317 } 318 #ifdef HAVE_LIBTRACEEVENT 319 trace_event__cleanup(&session->tevent); 320 #endif 321 free(session); 322 } 323 324 static int process_event_synth_tracing_data_stub(struct perf_session *session 325 __maybe_unused, 326 union perf_event *event 327 __maybe_unused) 328 { 329 dump_printf(": unhandled!\n"); 330 return 0; 331 } 332 333 static int process_event_synth_attr_stub(struct perf_tool *tool __maybe_unused, 334 union perf_event *event __maybe_unused, 335 struct evlist **pevlist 336 __maybe_unused) 337 { 338 dump_printf(": unhandled!\n"); 339 return 0; 340 } 341 342 static int process_event_synth_event_update_stub(struct perf_tool *tool __maybe_unused, 343 union perf_event *event __maybe_unused, 344 struct evlist **pevlist 345 __maybe_unused) 346 { 347 if (dump_trace) 348 perf_event__fprintf_event_update(event, stdout); 349 350 dump_printf(": unhandled!\n"); 351 return 0; 352 } 353 354 static int process_event_sample_stub(struct perf_tool *tool __maybe_unused, 355 union perf_event *event __maybe_unused, 356 struct perf_sample *sample __maybe_unused, 357 struct evsel *evsel __maybe_unused, 358 struct machine *machine __maybe_unused) 359 { 360 dump_printf(": unhandled!\n"); 361 return 0; 362 } 363 364 static int process_event_stub(struct perf_tool *tool __maybe_unused, 365 union perf_event *event __maybe_unused, 366 struct perf_sample *sample __maybe_unused, 367 struct machine *machine __maybe_unused) 368 { 369 dump_printf(": unhandled!\n"); 370 return 0; 371 } 372 373 static int process_finished_round_stub(struct perf_tool *tool __maybe_unused, 374 union perf_event *event __maybe_unused, 375 struct ordered_events *oe __maybe_unused) 376 { 377 dump_printf(": unhandled!\n"); 378 return 0; 379 } 380 381 static int skipn(int fd, off_t n) 382 { 383 char buf[4096]; 384 ssize_t ret; 385 386 while (n > 0) { 387 ret = read(fd, buf, min(n, (off_t)sizeof(buf))); 388 if (ret <= 0) 389 return ret; 390 n -= ret; 391 } 392 393 return 0; 394 } 395 396 static s64 process_event_auxtrace_stub(struct perf_session *session __maybe_unused, 397 union perf_event *event) 398 { 399 dump_printf(": unhandled!\n"); 400 if (perf_data__is_pipe(session->data)) 401 skipn(perf_data__fd(session->data), event->auxtrace.size); 402 return event->auxtrace.size; 403 } 404 405 static int process_event_op2_stub(struct perf_session *session __maybe_unused, 406 union perf_event *event __maybe_unused) 407 { 408 dump_printf(": unhandled!\n"); 409 return 0; 410 } 411 412 413 static 414 int process_event_thread_map_stub(struct perf_session *session __maybe_unused, 415 union perf_event *event __maybe_unused) 416 { 417 if (dump_trace) 418 perf_event__fprintf_thread_map(event, stdout); 419 420 dump_printf(": unhandled!\n"); 421 return 0; 422 } 423 424 static 425 int process_event_cpu_map_stub(struct perf_session *session __maybe_unused, 426 union perf_event *event __maybe_unused) 427 { 428 if (dump_trace) 429 perf_event__fprintf_cpu_map(event, stdout); 430 431 dump_printf(": unhandled!\n"); 432 return 0; 433 } 434 435 static 436 int process_event_stat_config_stub(struct perf_session *session __maybe_unused, 437 union perf_event *event __maybe_unused) 438 { 439 if (dump_trace) 440 perf_event__fprintf_stat_config(event, stdout); 441 442 dump_printf(": unhandled!\n"); 443 return 0; 444 } 445 446 static int process_stat_stub(struct perf_session *perf_session __maybe_unused, 447 union perf_event *event) 448 { 449 if (dump_trace) 450 perf_event__fprintf_stat(event, stdout); 451 452 dump_printf(": unhandled!\n"); 453 return 0; 454 } 455 456 static int process_stat_round_stub(struct perf_session *perf_session __maybe_unused, 457 union perf_event *event) 458 { 459 if (dump_trace) 460 perf_event__fprintf_stat_round(event, stdout); 461 462 dump_printf(": unhandled!\n"); 463 return 0; 464 } 465 466 static int process_event_time_conv_stub(struct perf_session *perf_session __maybe_unused, 467 union perf_event *event) 468 { 469 if (dump_trace) 470 perf_event__fprintf_time_conv(event, stdout); 471 472 dump_printf(": unhandled!\n"); 473 return 0; 474 } 475 476 static int perf_session__process_compressed_event_stub(struct perf_session *session __maybe_unused, 477 union perf_event *event __maybe_unused, 478 u64 file_offset __maybe_unused, 479 const char *file_path __maybe_unused) 480 { 481 dump_printf(": unhandled!\n"); 482 return 0; 483 } 484 485 void perf_tool__fill_defaults(struct perf_tool *tool) 486 { 487 if (tool->sample == NULL) 488 tool->sample = process_event_sample_stub; 489 if (tool->mmap == NULL) 490 tool->mmap = process_event_stub; 491 if (tool->mmap2 == NULL) 492 tool->mmap2 = process_event_stub; 493 if (tool->comm == NULL) 494 tool->comm = process_event_stub; 495 if (tool->namespaces == NULL) 496 tool->namespaces = process_event_stub; 497 if (tool->cgroup == NULL) 498 tool->cgroup = process_event_stub; 499 if (tool->fork == NULL) 500 tool->fork = process_event_stub; 501 if (tool->exit == NULL) 502 tool->exit = process_event_stub; 503 if (tool->lost == NULL) 504 tool->lost = perf_event__process_lost; 505 if (tool->lost_samples == NULL) 506 tool->lost_samples = perf_event__process_lost_samples; 507 if (tool->aux == NULL) 508 tool->aux = perf_event__process_aux; 509 if (tool->itrace_start == NULL) 510 tool->itrace_start = perf_event__process_itrace_start; 511 if (tool->context_switch == NULL) 512 tool->context_switch = perf_event__process_switch; 513 if (tool->ksymbol == NULL) 514 tool->ksymbol = perf_event__process_ksymbol; 515 if (tool->bpf == NULL) 516 tool->bpf = perf_event__process_bpf; 517 if (tool->text_poke == NULL) 518 tool->text_poke = perf_event__process_text_poke; 519 if (tool->aux_output_hw_id == NULL) 520 tool->aux_output_hw_id = perf_event__process_aux_output_hw_id; 521 if (tool->read == NULL) 522 tool->read = process_event_sample_stub; 523 if (tool->throttle == NULL) 524 tool->throttle = process_event_stub; 525 if (tool->unthrottle == NULL) 526 tool->unthrottle = process_event_stub; 527 if (tool->attr == NULL) 528 tool->attr = process_event_synth_attr_stub; 529 if (tool->event_update == NULL) 530 tool->event_update = process_event_synth_event_update_stub; 531 if (tool->tracing_data == NULL) 532 tool->tracing_data = process_event_synth_tracing_data_stub; 533 if (tool->build_id == NULL) 534 tool->build_id = process_event_op2_stub; 535 if (tool->finished_round == NULL) { 536 if (tool->ordered_events) 537 tool->finished_round = perf_event__process_finished_round; 538 else 539 tool->finished_round = process_finished_round_stub; 540 } 541 if (tool->id_index == NULL) 542 tool->id_index = process_event_op2_stub; 543 if (tool->auxtrace_info == NULL) 544 tool->auxtrace_info = process_event_op2_stub; 545 if (tool->auxtrace == NULL) 546 tool->auxtrace = process_event_auxtrace_stub; 547 if (tool->auxtrace_error == NULL) 548 tool->auxtrace_error = process_event_op2_stub; 549 if (tool->thread_map == NULL) 550 tool->thread_map = process_event_thread_map_stub; 551 if (tool->cpu_map == NULL) 552 tool->cpu_map = process_event_cpu_map_stub; 553 if (tool->stat_config == NULL) 554 tool->stat_config = process_event_stat_config_stub; 555 if (tool->stat == NULL) 556 tool->stat = process_stat_stub; 557 if (tool->stat_round == NULL) 558 tool->stat_round = process_stat_round_stub; 559 if (tool->time_conv == NULL) 560 tool->time_conv = process_event_time_conv_stub; 561 if (tool->feature == NULL) 562 tool->feature = process_event_op2_stub; 563 if (tool->compressed == NULL) 564 tool->compressed = perf_session__process_compressed_event; 565 if (tool->finished_init == NULL) 566 tool->finished_init = process_event_op2_stub; 567 } 568 569 static void swap_sample_id_all(union perf_event *event, void *data) 570 { 571 void *end = (void *) event + event->header.size; 572 int size = end - data; 573 574 BUG_ON(size % sizeof(u64)); 575 mem_bswap_64(data, size); 576 } 577 578 static void perf_event__all64_swap(union perf_event *event, 579 bool sample_id_all __maybe_unused) 580 { 581 struct perf_event_header *hdr = &event->header; 582 mem_bswap_64(hdr + 1, event->header.size - sizeof(*hdr)); 583 } 584 585 static void perf_event__comm_swap(union perf_event *event, bool sample_id_all) 586 { 587 event->comm.pid = bswap_32(event->comm.pid); 588 event->comm.tid = bswap_32(event->comm.tid); 589 590 if (sample_id_all) { 591 void *data = &event->comm.comm; 592 593 data += PERF_ALIGN(strlen(data) + 1, sizeof(u64)); 594 swap_sample_id_all(event, data); 595 } 596 } 597 598 static void perf_event__mmap_swap(union perf_event *event, 599 bool sample_id_all) 600 { 601 event->mmap.pid = bswap_32(event->mmap.pid); 602 event->mmap.tid = bswap_32(event->mmap.tid); 603 event->mmap.start = bswap_64(event->mmap.start); 604 event->mmap.len = bswap_64(event->mmap.len); 605 event->mmap.pgoff = bswap_64(event->mmap.pgoff); 606 607 if (sample_id_all) { 608 void *data = &event->mmap.filename; 609 610 data += PERF_ALIGN(strlen(data) + 1, sizeof(u64)); 611 swap_sample_id_all(event, data); 612 } 613 } 614 615 static void perf_event__mmap2_swap(union perf_event *event, 616 bool sample_id_all) 617 { 618 event->mmap2.pid = bswap_32(event->mmap2.pid); 619 event->mmap2.tid = bswap_32(event->mmap2.tid); 620 event->mmap2.start = bswap_64(event->mmap2.start); 621 event->mmap2.len = bswap_64(event->mmap2.len); 622 event->mmap2.pgoff = bswap_64(event->mmap2.pgoff); 623 624 if (!(event->header.misc & PERF_RECORD_MISC_MMAP_BUILD_ID)) { 625 event->mmap2.maj = bswap_32(event->mmap2.maj); 626 event->mmap2.min = bswap_32(event->mmap2.min); 627 event->mmap2.ino = bswap_64(event->mmap2.ino); 628 event->mmap2.ino_generation = bswap_64(event->mmap2.ino_generation); 629 } 630 631 if (sample_id_all) { 632 void *data = &event->mmap2.filename; 633 634 data += PERF_ALIGN(strlen(data) + 1, sizeof(u64)); 635 swap_sample_id_all(event, data); 636 } 637 } 638 static void perf_event__task_swap(union perf_event *event, bool sample_id_all) 639 { 640 event->fork.pid = bswap_32(event->fork.pid); 641 event->fork.tid = bswap_32(event->fork.tid); 642 event->fork.ppid = bswap_32(event->fork.ppid); 643 event->fork.ptid = bswap_32(event->fork.ptid); 644 event->fork.time = bswap_64(event->fork.time); 645 646 if (sample_id_all) 647 swap_sample_id_all(event, &event->fork + 1); 648 } 649 650 static void perf_event__read_swap(union perf_event *event, bool sample_id_all) 651 { 652 event->read.pid = bswap_32(event->read.pid); 653 event->read.tid = bswap_32(event->read.tid); 654 event->read.value = bswap_64(event->read.value); 655 event->read.time_enabled = bswap_64(event->read.time_enabled); 656 event->read.time_running = bswap_64(event->read.time_running); 657 event->read.id = bswap_64(event->read.id); 658 659 if (sample_id_all) 660 swap_sample_id_all(event, &event->read + 1); 661 } 662 663 static void perf_event__aux_swap(union perf_event *event, bool sample_id_all) 664 { 665 event->aux.aux_offset = bswap_64(event->aux.aux_offset); 666 event->aux.aux_size = bswap_64(event->aux.aux_size); 667 event->aux.flags = bswap_64(event->aux.flags); 668 669 if (sample_id_all) 670 swap_sample_id_all(event, &event->aux + 1); 671 } 672 673 static void perf_event__itrace_start_swap(union perf_event *event, 674 bool sample_id_all) 675 { 676 event->itrace_start.pid = bswap_32(event->itrace_start.pid); 677 event->itrace_start.tid = bswap_32(event->itrace_start.tid); 678 679 if (sample_id_all) 680 swap_sample_id_all(event, &event->itrace_start + 1); 681 } 682 683 static void perf_event__switch_swap(union perf_event *event, bool sample_id_all) 684 { 685 if (event->header.type == PERF_RECORD_SWITCH_CPU_WIDE) { 686 event->context_switch.next_prev_pid = 687 bswap_32(event->context_switch.next_prev_pid); 688 event->context_switch.next_prev_tid = 689 bswap_32(event->context_switch.next_prev_tid); 690 } 691 692 if (sample_id_all) 693 swap_sample_id_all(event, &event->context_switch + 1); 694 } 695 696 static void perf_event__text_poke_swap(union perf_event *event, bool sample_id_all) 697 { 698 event->text_poke.addr = bswap_64(event->text_poke.addr); 699 event->text_poke.old_len = bswap_16(event->text_poke.old_len); 700 event->text_poke.new_len = bswap_16(event->text_poke.new_len); 701 702 if (sample_id_all) { 703 size_t len = sizeof(event->text_poke.old_len) + 704 sizeof(event->text_poke.new_len) + 705 event->text_poke.old_len + 706 event->text_poke.new_len; 707 void *data = &event->text_poke.old_len; 708 709 data += PERF_ALIGN(len, sizeof(u64)); 710 swap_sample_id_all(event, data); 711 } 712 } 713 714 static void perf_event__throttle_swap(union perf_event *event, 715 bool sample_id_all) 716 { 717 event->throttle.time = bswap_64(event->throttle.time); 718 event->throttle.id = bswap_64(event->throttle.id); 719 event->throttle.stream_id = bswap_64(event->throttle.stream_id); 720 721 if (sample_id_all) 722 swap_sample_id_all(event, &event->throttle + 1); 723 } 724 725 static void perf_event__namespaces_swap(union perf_event *event, 726 bool sample_id_all) 727 { 728 u64 i; 729 730 event->namespaces.pid = bswap_32(event->namespaces.pid); 731 event->namespaces.tid = bswap_32(event->namespaces.tid); 732 event->namespaces.nr_namespaces = bswap_64(event->namespaces.nr_namespaces); 733 734 for (i = 0; i < event->namespaces.nr_namespaces; i++) { 735 struct perf_ns_link_info *ns = &event->namespaces.link_info[i]; 736 737 ns->dev = bswap_64(ns->dev); 738 ns->ino = bswap_64(ns->ino); 739 } 740 741 if (sample_id_all) 742 swap_sample_id_all(event, &event->namespaces.link_info[i]); 743 } 744 745 static void perf_event__cgroup_swap(union perf_event *event, bool sample_id_all) 746 { 747 event->cgroup.id = bswap_64(event->cgroup.id); 748 749 if (sample_id_all) { 750 void *data = &event->cgroup.path; 751 752 data += PERF_ALIGN(strlen(data) + 1, sizeof(u64)); 753 swap_sample_id_all(event, data); 754 } 755 } 756 757 static u8 revbyte(u8 b) 758 { 759 int rev = (b >> 4) | ((b & 0xf) << 4); 760 rev = ((rev & 0xcc) >> 2) | ((rev & 0x33) << 2); 761 rev = ((rev & 0xaa) >> 1) | ((rev & 0x55) << 1); 762 return (u8) rev; 763 } 764 765 /* 766 * XXX this is hack in attempt to carry flags bitfield 767 * through endian village. ABI says: 768 * 769 * Bit-fields are allocated from right to left (least to most significant) 770 * on little-endian implementations and from left to right (most to least 771 * significant) on big-endian implementations. 772 * 773 * The above seems to be byte specific, so we need to reverse each 774 * byte of the bitfield. 'Internet' also says this might be implementation 775 * specific and we probably need proper fix and carry perf_event_attr 776 * bitfield flags in separate data file FEAT_ section. Thought this seems 777 * to work for now. 778 */ 779 static void swap_bitfield(u8 *p, unsigned len) 780 { 781 unsigned i; 782 783 for (i = 0; i < len; i++) { 784 *p = revbyte(*p); 785 p++; 786 } 787 } 788 789 /* exported for swapping attributes in file header */ 790 void perf_event__attr_swap(struct perf_event_attr *attr) 791 { 792 attr->type = bswap_32(attr->type); 793 attr->size = bswap_32(attr->size); 794 795 #define bswap_safe(f, n) \ 796 (attr->size > (offsetof(struct perf_event_attr, f) + \ 797 sizeof(attr->f) * (n))) 798 #define bswap_field(f, sz) \ 799 do { \ 800 if (bswap_safe(f, 0)) \ 801 attr->f = bswap_##sz(attr->f); \ 802 } while(0) 803 #define bswap_field_16(f) bswap_field(f, 16) 804 #define bswap_field_32(f) bswap_field(f, 32) 805 #define bswap_field_64(f) bswap_field(f, 64) 806 807 bswap_field_64(config); 808 bswap_field_64(sample_period); 809 bswap_field_64(sample_type); 810 bswap_field_64(read_format); 811 bswap_field_32(wakeup_events); 812 bswap_field_32(bp_type); 813 bswap_field_64(bp_addr); 814 bswap_field_64(bp_len); 815 bswap_field_64(branch_sample_type); 816 bswap_field_64(sample_regs_user); 817 bswap_field_32(sample_stack_user); 818 bswap_field_32(aux_watermark); 819 bswap_field_16(sample_max_stack); 820 bswap_field_32(aux_sample_size); 821 822 /* 823 * After read_format are bitfields. Check read_format because 824 * we are unable to use offsetof on bitfield. 825 */ 826 if (bswap_safe(read_format, 1)) 827 swap_bitfield((u8 *) (&attr->read_format + 1), 828 sizeof(u64)); 829 #undef bswap_field_64 830 #undef bswap_field_32 831 #undef bswap_field 832 #undef bswap_safe 833 } 834 835 static void perf_event__hdr_attr_swap(union perf_event *event, 836 bool sample_id_all __maybe_unused) 837 { 838 size_t size; 839 840 perf_event__attr_swap(&event->attr.attr); 841 842 size = event->header.size; 843 size -= perf_record_header_attr_id(event) - (void *)event; 844 mem_bswap_64(perf_record_header_attr_id(event), size); 845 } 846 847 static void perf_event__event_update_swap(union perf_event *event, 848 bool sample_id_all __maybe_unused) 849 { 850 event->event_update.type = bswap_64(event->event_update.type); 851 event->event_update.id = bswap_64(event->event_update.id); 852 } 853 854 static void perf_event__event_type_swap(union perf_event *event, 855 bool sample_id_all __maybe_unused) 856 { 857 event->event_type.event_type.event_id = 858 bswap_64(event->event_type.event_type.event_id); 859 } 860 861 static void perf_event__tracing_data_swap(union perf_event *event, 862 bool sample_id_all __maybe_unused) 863 { 864 event->tracing_data.size = bswap_32(event->tracing_data.size); 865 } 866 867 static void perf_event__auxtrace_info_swap(union perf_event *event, 868 bool sample_id_all __maybe_unused) 869 { 870 size_t size; 871 872 event->auxtrace_info.type = bswap_32(event->auxtrace_info.type); 873 874 size = event->header.size; 875 size -= (void *)&event->auxtrace_info.priv - (void *)event; 876 mem_bswap_64(event->auxtrace_info.priv, size); 877 } 878 879 static void perf_event__auxtrace_swap(union perf_event *event, 880 bool sample_id_all __maybe_unused) 881 { 882 event->auxtrace.size = bswap_64(event->auxtrace.size); 883 event->auxtrace.offset = bswap_64(event->auxtrace.offset); 884 event->auxtrace.reference = bswap_64(event->auxtrace.reference); 885 event->auxtrace.idx = bswap_32(event->auxtrace.idx); 886 event->auxtrace.tid = bswap_32(event->auxtrace.tid); 887 event->auxtrace.cpu = bswap_32(event->auxtrace.cpu); 888 } 889 890 static void perf_event__auxtrace_error_swap(union perf_event *event, 891 bool sample_id_all __maybe_unused) 892 { 893 event->auxtrace_error.type = bswap_32(event->auxtrace_error.type); 894 event->auxtrace_error.code = bswap_32(event->auxtrace_error.code); 895 event->auxtrace_error.cpu = bswap_32(event->auxtrace_error.cpu); 896 event->auxtrace_error.pid = bswap_32(event->auxtrace_error.pid); 897 event->auxtrace_error.tid = bswap_32(event->auxtrace_error.tid); 898 event->auxtrace_error.fmt = bswap_32(event->auxtrace_error.fmt); 899 event->auxtrace_error.ip = bswap_64(event->auxtrace_error.ip); 900 if (event->auxtrace_error.fmt) 901 event->auxtrace_error.time = bswap_64(event->auxtrace_error.time); 902 if (event->auxtrace_error.fmt >= 2) { 903 event->auxtrace_error.machine_pid = bswap_32(event->auxtrace_error.machine_pid); 904 event->auxtrace_error.vcpu = bswap_32(event->auxtrace_error.vcpu); 905 } 906 } 907 908 static void perf_event__thread_map_swap(union perf_event *event, 909 bool sample_id_all __maybe_unused) 910 { 911 unsigned i; 912 913 event->thread_map.nr = bswap_64(event->thread_map.nr); 914 915 for (i = 0; i < event->thread_map.nr; i++) 916 event->thread_map.entries[i].pid = bswap_64(event->thread_map.entries[i].pid); 917 } 918 919 static void perf_event__cpu_map_swap(union perf_event *event, 920 bool sample_id_all __maybe_unused) 921 { 922 struct perf_record_cpu_map_data *data = &event->cpu_map.data; 923 924 data->type = bswap_16(data->type); 925 926 switch (data->type) { 927 case PERF_CPU_MAP__CPUS: 928 data->cpus_data.nr = bswap_16(data->cpus_data.nr); 929 930 for (unsigned i = 0; i < data->cpus_data.nr; i++) 931 data->cpus_data.cpu[i] = bswap_16(data->cpus_data.cpu[i]); 932 break; 933 case PERF_CPU_MAP__MASK: 934 data->mask32_data.long_size = bswap_16(data->mask32_data.long_size); 935 936 switch (data->mask32_data.long_size) { 937 case 4: 938 data->mask32_data.nr = bswap_16(data->mask32_data.nr); 939 for (unsigned i = 0; i < data->mask32_data.nr; i++) 940 data->mask32_data.mask[i] = bswap_32(data->mask32_data.mask[i]); 941 break; 942 case 8: 943 data->mask64_data.nr = bswap_16(data->mask64_data.nr); 944 for (unsigned i = 0; i < data->mask64_data.nr; i++) 945 data->mask64_data.mask[i] = bswap_64(data->mask64_data.mask[i]); 946 break; 947 default: 948 pr_err("cpu_map swap: unsupported long size\n"); 949 } 950 break; 951 case PERF_CPU_MAP__RANGE_CPUS: 952 data->range_cpu_data.start_cpu = bswap_16(data->range_cpu_data.start_cpu); 953 data->range_cpu_data.end_cpu = bswap_16(data->range_cpu_data.end_cpu); 954 break; 955 default: 956 break; 957 } 958 } 959 960 static void perf_event__stat_config_swap(union perf_event *event, 961 bool sample_id_all __maybe_unused) 962 { 963 u64 size; 964 965 size = bswap_64(event->stat_config.nr) * sizeof(event->stat_config.data[0]); 966 size += 1; /* nr item itself */ 967 mem_bswap_64(&event->stat_config.nr, size); 968 } 969 970 static void perf_event__stat_swap(union perf_event *event, 971 bool sample_id_all __maybe_unused) 972 { 973 event->stat.id = bswap_64(event->stat.id); 974 event->stat.thread = bswap_32(event->stat.thread); 975 event->stat.cpu = bswap_32(event->stat.cpu); 976 event->stat.val = bswap_64(event->stat.val); 977 event->stat.ena = bswap_64(event->stat.ena); 978 event->stat.run = bswap_64(event->stat.run); 979 } 980 981 static void perf_event__stat_round_swap(union perf_event *event, 982 bool sample_id_all __maybe_unused) 983 { 984 event->stat_round.type = bswap_64(event->stat_round.type); 985 event->stat_round.time = bswap_64(event->stat_round.time); 986 } 987 988 static void perf_event__time_conv_swap(union perf_event *event, 989 bool sample_id_all __maybe_unused) 990 { 991 event->time_conv.time_shift = bswap_64(event->time_conv.time_shift); 992 event->time_conv.time_mult = bswap_64(event->time_conv.time_mult); 993 event->time_conv.time_zero = bswap_64(event->time_conv.time_zero); 994 995 if (event_contains(event->time_conv, time_cycles)) { 996 event->time_conv.time_cycles = bswap_64(event->time_conv.time_cycles); 997 event->time_conv.time_mask = bswap_64(event->time_conv.time_mask); 998 } 999 } 1000 1001 typedef void (*perf_event__swap_op)(union perf_event *event, 1002 bool sample_id_all); 1003 1004 static perf_event__swap_op perf_event__swap_ops[] = { 1005 [PERF_RECORD_MMAP] = perf_event__mmap_swap, 1006 [PERF_RECORD_MMAP2] = perf_event__mmap2_swap, 1007 [PERF_RECORD_COMM] = perf_event__comm_swap, 1008 [PERF_RECORD_FORK] = perf_event__task_swap, 1009 [PERF_RECORD_EXIT] = perf_event__task_swap, 1010 [PERF_RECORD_LOST] = perf_event__all64_swap, 1011 [PERF_RECORD_READ] = perf_event__read_swap, 1012 [PERF_RECORD_THROTTLE] = perf_event__throttle_swap, 1013 [PERF_RECORD_UNTHROTTLE] = perf_event__throttle_swap, 1014 [PERF_RECORD_SAMPLE] = perf_event__all64_swap, 1015 [PERF_RECORD_AUX] = perf_event__aux_swap, 1016 [PERF_RECORD_ITRACE_START] = perf_event__itrace_start_swap, 1017 [PERF_RECORD_LOST_SAMPLES] = perf_event__all64_swap, 1018 [PERF_RECORD_SWITCH] = perf_event__switch_swap, 1019 [PERF_RECORD_SWITCH_CPU_WIDE] = perf_event__switch_swap, 1020 [PERF_RECORD_NAMESPACES] = perf_event__namespaces_swap, 1021 [PERF_RECORD_CGROUP] = perf_event__cgroup_swap, 1022 [PERF_RECORD_TEXT_POKE] = perf_event__text_poke_swap, 1023 [PERF_RECORD_AUX_OUTPUT_HW_ID] = perf_event__all64_swap, 1024 [PERF_RECORD_HEADER_ATTR] = perf_event__hdr_attr_swap, 1025 [PERF_RECORD_HEADER_EVENT_TYPE] = perf_event__event_type_swap, 1026 [PERF_RECORD_HEADER_TRACING_DATA] = perf_event__tracing_data_swap, 1027 [PERF_RECORD_HEADER_BUILD_ID] = NULL, 1028 [PERF_RECORD_ID_INDEX] = perf_event__all64_swap, 1029 [PERF_RECORD_AUXTRACE_INFO] = perf_event__auxtrace_info_swap, 1030 [PERF_RECORD_AUXTRACE] = perf_event__auxtrace_swap, 1031 [PERF_RECORD_AUXTRACE_ERROR] = perf_event__auxtrace_error_swap, 1032 [PERF_RECORD_THREAD_MAP] = perf_event__thread_map_swap, 1033 [PERF_RECORD_CPU_MAP] = perf_event__cpu_map_swap, 1034 [PERF_RECORD_STAT_CONFIG] = perf_event__stat_config_swap, 1035 [PERF_RECORD_STAT] = perf_event__stat_swap, 1036 [PERF_RECORD_STAT_ROUND] = perf_event__stat_round_swap, 1037 [PERF_RECORD_EVENT_UPDATE] = perf_event__event_update_swap, 1038 [PERF_RECORD_TIME_CONV] = perf_event__time_conv_swap, 1039 [PERF_RECORD_HEADER_MAX] = NULL, 1040 }; 1041 1042 /* 1043 * When perf record finishes a pass on every buffers, it records this pseudo 1044 * event. 1045 * We record the max timestamp t found in the pass n. 1046 * Assuming these timestamps are monotonic across cpus, we know that if 1047 * a buffer still has events with timestamps below t, they will be all 1048 * available and then read in the pass n + 1. 1049 * Hence when we start to read the pass n + 2, we can safely flush every 1050 * events with timestamps below t. 1051 * 1052 * ============ PASS n ================= 1053 * CPU 0 | CPU 1 1054 * | 1055 * cnt1 timestamps | cnt2 timestamps 1056 * 1 | 2 1057 * 2 | 3 1058 * - | 4 <--- max recorded 1059 * 1060 * ============ PASS n + 1 ============== 1061 * CPU 0 | CPU 1 1062 * | 1063 * cnt1 timestamps | cnt2 timestamps 1064 * 3 | 5 1065 * 4 | 6 1066 * 5 | 7 <---- max recorded 1067 * 1068 * Flush every events below timestamp 4 1069 * 1070 * ============ PASS n + 2 ============== 1071 * CPU 0 | CPU 1 1072 * | 1073 * cnt1 timestamps | cnt2 timestamps 1074 * 6 | 8 1075 * 7 | 9 1076 * - | 10 1077 * 1078 * Flush every events below timestamp 7 1079 * etc... 1080 */ 1081 int perf_event__process_finished_round(struct perf_tool *tool __maybe_unused, 1082 union perf_event *event __maybe_unused, 1083 struct ordered_events *oe) 1084 { 1085 if (dump_trace) 1086 fprintf(stdout, "\n"); 1087 return ordered_events__flush(oe, OE_FLUSH__ROUND); 1088 } 1089 1090 int perf_session__queue_event(struct perf_session *s, union perf_event *event, 1091 u64 timestamp, u64 file_offset, const char *file_path) 1092 { 1093 return ordered_events__queue(&s->ordered_events, event, timestamp, file_offset, file_path); 1094 } 1095 1096 static void callchain__lbr_callstack_printf(struct perf_sample *sample) 1097 { 1098 struct ip_callchain *callchain = sample->callchain; 1099 struct branch_stack *lbr_stack = sample->branch_stack; 1100 struct branch_entry *entries = perf_sample__branch_entries(sample); 1101 u64 kernel_callchain_nr = callchain->nr; 1102 unsigned int i; 1103 1104 for (i = 0; i < kernel_callchain_nr; i++) { 1105 if (callchain->ips[i] == PERF_CONTEXT_USER) 1106 break; 1107 } 1108 1109 if ((i != kernel_callchain_nr) && lbr_stack->nr) { 1110 u64 total_nr; 1111 /* 1112 * LBR callstack can only get user call chain, 1113 * i is kernel call chain number, 1114 * 1 is PERF_CONTEXT_USER. 1115 * 1116 * The user call chain is stored in LBR registers. 1117 * LBR are pair registers. The caller is stored 1118 * in "from" register, while the callee is stored 1119 * in "to" register. 1120 * For example, there is a call stack 1121 * "A"->"B"->"C"->"D". 1122 * The LBR registers will be recorded like 1123 * "C"->"D", "B"->"C", "A"->"B". 1124 * So only the first "to" register and all "from" 1125 * registers are needed to construct the whole stack. 1126 */ 1127 total_nr = i + 1 + lbr_stack->nr + 1; 1128 kernel_callchain_nr = i + 1; 1129 1130 printf("... LBR call chain: nr:%" PRIu64 "\n", total_nr); 1131 1132 for (i = 0; i < kernel_callchain_nr; i++) 1133 printf("..... %2d: %016" PRIx64 "\n", 1134 i, callchain->ips[i]); 1135 1136 printf("..... %2d: %016" PRIx64 "\n", 1137 (int)(kernel_callchain_nr), entries[0].to); 1138 for (i = 0; i < lbr_stack->nr; i++) 1139 printf("..... %2d: %016" PRIx64 "\n", 1140 (int)(i + kernel_callchain_nr + 1), entries[i].from); 1141 } 1142 } 1143 1144 static void callchain__printf(struct evsel *evsel, 1145 struct perf_sample *sample) 1146 { 1147 unsigned int i; 1148 struct ip_callchain *callchain = sample->callchain; 1149 1150 if (evsel__has_branch_callstack(evsel)) 1151 callchain__lbr_callstack_printf(sample); 1152 1153 printf("... FP chain: nr:%" PRIu64 "\n", callchain->nr); 1154 1155 for (i = 0; i < callchain->nr; i++) 1156 printf("..... %2d: %016" PRIx64 "\n", 1157 i, callchain->ips[i]); 1158 } 1159 1160 static void branch_stack__printf(struct perf_sample *sample, 1161 struct evsel *evsel) 1162 { 1163 struct branch_entry *entries = perf_sample__branch_entries(sample); 1164 bool callstack = evsel__has_branch_callstack(evsel); 1165 u64 *branch_stack_cntr = sample->branch_stack_cntr; 1166 struct perf_env *env = evsel__env(evsel); 1167 uint64_t i; 1168 1169 if (!callstack) { 1170 printf("%s: nr:%" PRIu64 "\n", "... branch stack", sample->branch_stack->nr); 1171 } else { 1172 /* the reason of adding 1 to nr is because after expanding 1173 * branch stack it generates nr + 1 callstack records. e.g., 1174 * B()->C() 1175 * A()->B() 1176 * the final callstack should be: 1177 * C() 1178 * B() 1179 * A() 1180 */ 1181 printf("%s: nr:%" PRIu64 "\n", "... branch callstack", sample->branch_stack->nr+1); 1182 } 1183 1184 for (i = 0; i < sample->branch_stack->nr; i++) { 1185 struct branch_entry *e = &entries[i]; 1186 1187 if (!callstack) { 1188 printf("..... %2"PRIu64": %016" PRIx64 " -> %016" PRIx64 " %hu cycles %s%s%s%s %x %s %s\n", 1189 i, e->from, e->to, 1190 (unsigned short)e->flags.cycles, 1191 e->flags.mispred ? "M" : " ", 1192 e->flags.predicted ? "P" : " ", 1193 e->flags.abort ? "A" : " ", 1194 e->flags.in_tx ? "T" : " ", 1195 (unsigned)e->flags.reserved, 1196 get_branch_type(e), 1197 e->flags.spec ? branch_spec_desc(e->flags.spec) : ""); 1198 } else { 1199 if (i == 0) { 1200 printf("..... %2"PRIu64": %016" PRIx64 "\n" 1201 "..... %2"PRIu64": %016" PRIx64 "\n", 1202 i, e->to, i+1, e->from); 1203 } else { 1204 printf("..... %2"PRIu64": %016" PRIx64 "\n", i+1, e->from); 1205 } 1206 } 1207 } 1208 1209 if (branch_stack_cntr) { 1210 printf("... branch stack counters: nr:%" PRIu64 " (counter width: %u max counter nr:%u)\n", 1211 sample->branch_stack->nr, env->br_cntr_width, env->br_cntr_nr); 1212 for (i = 0; i < sample->branch_stack->nr; i++) 1213 printf("..... %2"PRIu64": %016" PRIx64 "\n", i, branch_stack_cntr[i]); 1214 } 1215 } 1216 1217 static void regs_dump__printf(u64 mask, u64 *regs, const char *arch) 1218 { 1219 unsigned rid, i = 0; 1220 1221 for_each_set_bit(rid, (unsigned long *) &mask, sizeof(mask) * 8) { 1222 u64 val = regs[i++]; 1223 1224 printf(".... %-5s 0x%016" PRIx64 "\n", 1225 perf_reg_name(rid, arch), val); 1226 } 1227 } 1228 1229 static const char *regs_abi[] = { 1230 [PERF_SAMPLE_REGS_ABI_NONE] = "none", 1231 [PERF_SAMPLE_REGS_ABI_32] = "32-bit", 1232 [PERF_SAMPLE_REGS_ABI_64] = "64-bit", 1233 }; 1234 1235 static inline const char *regs_dump_abi(struct regs_dump *d) 1236 { 1237 if (d->abi > PERF_SAMPLE_REGS_ABI_64) 1238 return "unknown"; 1239 1240 return regs_abi[d->abi]; 1241 } 1242 1243 static void regs__printf(const char *type, struct regs_dump *regs, const char *arch) 1244 { 1245 u64 mask = regs->mask; 1246 1247 printf("... %s regs: mask 0x%" PRIx64 " ABI %s\n", 1248 type, 1249 mask, 1250 regs_dump_abi(regs)); 1251 1252 regs_dump__printf(mask, regs->regs, arch); 1253 } 1254 1255 static void regs_user__printf(struct perf_sample *sample, const char *arch) 1256 { 1257 struct regs_dump *user_regs = &sample->user_regs; 1258 1259 if (user_regs->regs) 1260 regs__printf("user", user_regs, arch); 1261 } 1262 1263 static void regs_intr__printf(struct perf_sample *sample, const char *arch) 1264 { 1265 struct regs_dump *intr_regs = &sample->intr_regs; 1266 1267 if (intr_regs->regs) 1268 regs__printf("intr", intr_regs, arch); 1269 } 1270 1271 static void stack_user__printf(struct stack_dump *dump) 1272 { 1273 printf("... ustack: size %" PRIu64 ", offset 0x%x\n", 1274 dump->size, dump->offset); 1275 } 1276 1277 static void evlist__print_tstamp(struct evlist *evlist, union perf_event *event, struct perf_sample *sample) 1278 { 1279 u64 sample_type = __evlist__combined_sample_type(evlist); 1280 1281 if (event->header.type != PERF_RECORD_SAMPLE && 1282 !evlist__sample_id_all(evlist)) { 1283 fputs("-1 -1 ", stdout); 1284 return; 1285 } 1286 1287 if ((sample_type & PERF_SAMPLE_CPU)) 1288 printf("%u ", sample->cpu); 1289 1290 if (sample_type & PERF_SAMPLE_TIME) 1291 printf("%" PRIu64 " ", sample->time); 1292 } 1293 1294 static void sample_read__printf(struct perf_sample *sample, u64 read_format) 1295 { 1296 printf("... sample_read:\n"); 1297 1298 if (read_format & PERF_FORMAT_TOTAL_TIME_ENABLED) 1299 printf("...... time enabled %016" PRIx64 "\n", 1300 sample->read.time_enabled); 1301 1302 if (read_format & PERF_FORMAT_TOTAL_TIME_RUNNING) 1303 printf("...... time running %016" PRIx64 "\n", 1304 sample->read.time_running); 1305 1306 if (read_format & PERF_FORMAT_GROUP) { 1307 struct sample_read_value *value = sample->read.group.values; 1308 1309 printf(".... group nr %" PRIu64 "\n", sample->read.group.nr); 1310 1311 sample_read_group__for_each(value, sample->read.group.nr, read_format) { 1312 printf("..... id %016" PRIx64 1313 ", value %016" PRIx64, 1314 value->id, value->value); 1315 if (read_format & PERF_FORMAT_LOST) 1316 printf(", lost %" PRIu64, value->lost); 1317 printf("\n"); 1318 } 1319 } else { 1320 printf("..... id %016" PRIx64 ", value %016" PRIx64, 1321 sample->read.one.id, sample->read.one.value); 1322 if (read_format & PERF_FORMAT_LOST) 1323 printf(", lost %" PRIu64, sample->read.one.lost); 1324 printf("\n"); 1325 } 1326 } 1327 1328 static void dump_event(struct evlist *evlist, union perf_event *event, 1329 u64 file_offset, struct perf_sample *sample, 1330 const char *file_path) 1331 { 1332 if (!dump_trace) 1333 return; 1334 1335 printf("\n%#" PRIx64 "@%s [%#x]: event: %d\n", 1336 file_offset, file_path, event->header.size, event->header.type); 1337 1338 trace_event(event); 1339 if (event->header.type == PERF_RECORD_SAMPLE && evlist->trace_event_sample_raw) 1340 evlist->trace_event_sample_raw(evlist, event, sample); 1341 1342 if (sample) 1343 evlist__print_tstamp(evlist, event, sample); 1344 1345 printf("%#" PRIx64 " [%#x]: PERF_RECORD_%s", file_offset, 1346 event->header.size, perf_event__name(event->header.type)); 1347 } 1348 1349 char *get_page_size_name(u64 size, char *str) 1350 { 1351 if (!size || !unit_number__scnprintf(str, PAGE_SIZE_NAME_LEN, size)) 1352 snprintf(str, PAGE_SIZE_NAME_LEN, "%s", "N/A"); 1353 1354 return str; 1355 } 1356 1357 static void dump_sample(struct evsel *evsel, union perf_event *event, 1358 struct perf_sample *sample, const char *arch) 1359 { 1360 u64 sample_type; 1361 char str[PAGE_SIZE_NAME_LEN]; 1362 1363 if (!dump_trace) 1364 return; 1365 1366 printf("(IP, 0x%x): %d/%d: %#" PRIx64 " period: %" PRIu64 " addr: %#" PRIx64 "\n", 1367 event->header.misc, sample->pid, sample->tid, sample->ip, 1368 sample->period, sample->addr); 1369 1370 sample_type = evsel->core.attr.sample_type; 1371 1372 if (evsel__has_callchain(evsel)) 1373 callchain__printf(evsel, sample); 1374 1375 if (evsel__has_br_stack(evsel)) 1376 branch_stack__printf(sample, evsel); 1377 1378 if (sample_type & PERF_SAMPLE_REGS_USER) 1379 regs_user__printf(sample, arch); 1380 1381 if (sample_type & PERF_SAMPLE_REGS_INTR) 1382 regs_intr__printf(sample, arch); 1383 1384 if (sample_type & PERF_SAMPLE_STACK_USER) 1385 stack_user__printf(&sample->user_stack); 1386 1387 if (sample_type & PERF_SAMPLE_WEIGHT_TYPE) { 1388 printf("... weight: %" PRIu64 "", sample->weight); 1389 if (sample_type & PERF_SAMPLE_WEIGHT_STRUCT) { 1390 printf(",0x%"PRIx16"", sample->ins_lat); 1391 printf(",0x%"PRIx16"", sample->p_stage_cyc); 1392 } 1393 printf("\n"); 1394 } 1395 1396 if (sample_type & PERF_SAMPLE_DATA_SRC) 1397 printf(" . data_src: 0x%"PRIx64"\n", sample->data_src); 1398 1399 if (sample_type & PERF_SAMPLE_PHYS_ADDR) 1400 printf(" .. phys_addr: 0x%"PRIx64"\n", sample->phys_addr); 1401 1402 if (sample_type & PERF_SAMPLE_DATA_PAGE_SIZE) 1403 printf(" .. data page size: %s\n", get_page_size_name(sample->data_page_size, str)); 1404 1405 if (sample_type & PERF_SAMPLE_CODE_PAGE_SIZE) 1406 printf(" .. code page size: %s\n", get_page_size_name(sample->code_page_size, str)); 1407 1408 if (sample_type & PERF_SAMPLE_TRANSACTION) 1409 printf("... transaction: %" PRIx64 "\n", sample->transaction); 1410 1411 if (sample_type & PERF_SAMPLE_READ) 1412 sample_read__printf(sample, evsel->core.attr.read_format); 1413 } 1414 1415 static void dump_read(struct evsel *evsel, union perf_event *event) 1416 { 1417 struct perf_record_read *read_event = &event->read; 1418 u64 read_format; 1419 1420 if (!dump_trace) 1421 return; 1422 1423 printf(": %d %d %s %" PRI_lu64 "\n", event->read.pid, event->read.tid, 1424 evsel__name(evsel), event->read.value); 1425 1426 if (!evsel) 1427 return; 1428 1429 read_format = evsel->core.attr.read_format; 1430 1431 if (read_format & PERF_FORMAT_TOTAL_TIME_ENABLED) 1432 printf("... time enabled : %" PRI_lu64 "\n", read_event->time_enabled); 1433 1434 if (read_format & PERF_FORMAT_TOTAL_TIME_RUNNING) 1435 printf("... time running : %" PRI_lu64 "\n", read_event->time_running); 1436 1437 if (read_format & PERF_FORMAT_ID) 1438 printf("... id : %" PRI_lu64 "\n", read_event->id); 1439 1440 if (read_format & PERF_FORMAT_LOST) 1441 printf("... lost : %" PRI_lu64 "\n", read_event->lost); 1442 } 1443 1444 static struct machine *machines__find_for_cpumode(struct machines *machines, 1445 union perf_event *event, 1446 struct perf_sample *sample) 1447 { 1448 if (perf_guest && 1449 ((sample->cpumode == PERF_RECORD_MISC_GUEST_KERNEL) || 1450 (sample->cpumode == PERF_RECORD_MISC_GUEST_USER))) { 1451 u32 pid; 1452 1453 if (sample->machine_pid) 1454 pid = sample->machine_pid; 1455 else if (event->header.type == PERF_RECORD_MMAP 1456 || event->header.type == PERF_RECORD_MMAP2) 1457 pid = event->mmap.pid; 1458 else 1459 pid = sample->pid; 1460 1461 /* 1462 * Guest code machine is created as needed and does not use 1463 * DEFAULT_GUEST_KERNEL_ID. 1464 */ 1465 if (symbol_conf.guest_code) 1466 return machines__findnew(machines, pid); 1467 1468 return machines__find_guest(machines, pid); 1469 } 1470 1471 return &machines->host; 1472 } 1473 1474 static int deliver_sample_value(struct evlist *evlist, 1475 struct perf_tool *tool, 1476 union perf_event *event, 1477 struct perf_sample *sample, 1478 struct sample_read_value *v, 1479 struct machine *machine) 1480 { 1481 struct perf_sample_id *sid = evlist__id2sid(evlist, v->id); 1482 struct evsel *evsel; 1483 1484 if (sid) { 1485 sample->id = v->id; 1486 sample->period = v->value - sid->period; 1487 sid->period = v->value; 1488 } 1489 1490 if (!sid || sid->evsel == NULL) { 1491 ++evlist->stats.nr_unknown_id; 1492 return 0; 1493 } 1494 1495 /* 1496 * There's no reason to deliver sample 1497 * for zero period, bail out. 1498 */ 1499 if (!sample->period) 1500 return 0; 1501 1502 evsel = container_of(sid->evsel, struct evsel, core); 1503 return tool->sample(tool, event, sample, evsel, machine); 1504 } 1505 1506 static int deliver_sample_group(struct evlist *evlist, 1507 struct perf_tool *tool, 1508 union perf_event *event, 1509 struct perf_sample *sample, 1510 struct machine *machine, 1511 u64 read_format) 1512 { 1513 int ret = -EINVAL; 1514 struct sample_read_value *v = sample->read.group.values; 1515 1516 sample_read_group__for_each(v, sample->read.group.nr, read_format) { 1517 ret = deliver_sample_value(evlist, tool, event, sample, v, 1518 machine); 1519 if (ret) 1520 break; 1521 } 1522 1523 return ret; 1524 } 1525 1526 static int evlist__deliver_sample(struct evlist *evlist, struct perf_tool *tool, 1527 union perf_event *event, struct perf_sample *sample, 1528 struct evsel *evsel, struct machine *machine) 1529 { 1530 /* We know evsel != NULL. */ 1531 u64 sample_type = evsel->core.attr.sample_type; 1532 u64 read_format = evsel->core.attr.read_format; 1533 1534 /* Standard sample delivery. */ 1535 if (!(sample_type & PERF_SAMPLE_READ)) 1536 return tool->sample(tool, event, sample, evsel, machine); 1537 1538 /* For PERF_SAMPLE_READ we have either single or group mode. */ 1539 if (read_format & PERF_FORMAT_GROUP) 1540 return deliver_sample_group(evlist, tool, event, sample, 1541 machine, read_format); 1542 else 1543 return deliver_sample_value(evlist, tool, event, sample, 1544 &sample->read.one, machine); 1545 } 1546 1547 static int machines__deliver_event(struct machines *machines, 1548 struct evlist *evlist, 1549 union perf_event *event, 1550 struct perf_sample *sample, 1551 struct perf_tool *tool, u64 file_offset, 1552 const char *file_path) 1553 { 1554 struct evsel *evsel; 1555 struct machine *machine; 1556 1557 dump_event(evlist, event, file_offset, sample, file_path); 1558 1559 evsel = evlist__id2evsel(evlist, sample->id); 1560 1561 machine = machines__find_for_cpumode(machines, event, sample); 1562 1563 switch (event->header.type) { 1564 case PERF_RECORD_SAMPLE: 1565 if (evsel == NULL) { 1566 ++evlist->stats.nr_unknown_id; 1567 return 0; 1568 } 1569 if (machine == NULL) { 1570 ++evlist->stats.nr_unprocessable_samples; 1571 dump_sample(evsel, event, sample, perf_env__arch(NULL)); 1572 return 0; 1573 } 1574 dump_sample(evsel, event, sample, perf_env__arch(machine->env)); 1575 return evlist__deliver_sample(evlist, tool, event, sample, evsel, machine); 1576 case PERF_RECORD_MMAP: 1577 return tool->mmap(tool, event, sample, machine); 1578 case PERF_RECORD_MMAP2: 1579 if (event->header.misc & PERF_RECORD_MISC_PROC_MAP_PARSE_TIMEOUT) 1580 ++evlist->stats.nr_proc_map_timeout; 1581 return tool->mmap2(tool, event, sample, machine); 1582 case PERF_RECORD_COMM: 1583 return tool->comm(tool, event, sample, machine); 1584 case PERF_RECORD_NAMESPACES: 1585 return tool->namespaces(tool, event, sample, machine); 1586 case PERF_RECORD_CGROUP: 1587 return tool->cgroup(tool, event, sample, machine); 1588 case PERF_RECORD_FORK: 1589 return tool->fork(tool, event, sample, machine); 1590 case PERF_RECORD_EXIT: 1591 return tool->exit(tool, event, sample, machine); 1592 case PERF_RECORD_LOST: 1593 if (tool->lost == perf_event__process_lost) 1594 evlist->stats.total_lost += event->lost.lost; 1595 return tool->lost(tool, event, sample, machine); 1596 case PERF_RECORD_LOST_SAMPLES: 1597 if (tool->lost_samples == perf_event__process_lost_samples && 1598 !(event->header.misc & PERF_RECORD_MISC_LOST_SAMPLES_BPF)) 1599 evlist->stats.total_lost_samples += event->lost_samples.lost; 1600 return tool->lost_samples(tool, event, sample, machine); 1601 case PERF_RECORD_READ: 1602 dump_read(evsel, event); 1603 return tool->read(tool, event, sample, evsel, machine); 1604 case PERF_RECORD_THROTTLE: 1605 return tool->throttle(tool, event, sample, machine); 1606 case PERF_RECORD_UNTHROTTLE: 1607 return tool->unthrottle(tool, event, sample, machine); 1608 case PERF_RECORD_AUX: 1609 if (tool->aux == perf_event__process_aux) { 1610 if (event->aux.flags & PERF_AUX_FLAG_TRUNCATED) 1611 evlist->stats.total_aux_lost += 1; 1612 if (event->aux.flags & PERF_AUX_FLAG_PARTIAL) 1613 evlist->stats.total_aux_partial += 1; 1614 if (event->aux.flags & PERF_AUX_FLAG_COLLISION) 1615 evlist->stats.total_aux_collision += 1; 1616 } 1617 return tool->aux(tool, event, sample, machine); 1618 case PERF_RECORD_ITRACE_START: 1619 return tool->itrace_start(tool, event, sample, machine); 1620 case PERF_RECORD_SWITCH: 1621 case PERF_RECORD_SWITCH_CPU_WIDE: 1622 return tool->context_switch(tool, event, sample, machine); 1623 case PERF_RECORD_KSYMBOL: 1624 return tool->ksymbol(tool, event, sample, machine); 1625 case PERF_RECORD_BPF_EVENT: 1626 return tool->bpf(tool, event, sample, machine); 1627 case PERF_RECORD_TEXT_POKE: 1628 return tool->text_poke(tool, event, sample, machine); 1629 case PERF_RECORD_AUX_OUTPUT_HW_ID: 1630 return tool->aux_output_hw_id(tool, event, sample, machine); 1631 default: 1632 ++evlist->stats.nr_unknown_events; 1633 return -1; 1634 } 1635 } 1636 1637 static int perf_session__deliver_event(struct perf_session *session, 1638 union perf_event *event, 1639 struct perf_tool *tool, 1640 u64 file_offset, 1641 const char *file_path) 1642 { 1643 struct perf_sample sample; 1644 int ret = evlist__parse_sample(session->evlist, event, &sample); 1645 1646 if (ret) { 1647 pr_err("Can't parse sample, err = %d\n", ret); 1648 return ret; 1649 } 1650 1651 ret = auxtrace__process_event(session, event, &sample, tool); 1652 if (ret < 0) 1653 return ret; 1654 if (ret > 0) 1655 return 0; 1656 1657 ret = machines__deliver_event(&session->machines, session->evlist, 1658 event, &sample, tool, file_offset, file_path); 1659 1660 if (dump_trace && sample.aux_sample.size) 1661 auxtrace__dump_auxtrace_sample(session, &sample); 1662 1663 return ret; 1664 } 1665 1666 static s64 perf_session__process_user_event(struct perf_session *session, 1667 union perf_event *event, 1668 u64 file_offset, 1669 const char *file_path) 1670 { 1671 struct ordered_events *oe = &session->ordered_events; 1672 struct perf_tool *tool = session->tool; 1673 struct perf_sample sample = { .time = 0, }; 1674 int fd = perf_data__fd(session->data); 1675 int err; 1676 1677 if (event->header.type != PERF_RECORD_COMPRESSED || 1678 tool->compressed == perf_session__process_compressed_event_stub) 1679 dump_event(session->evlist, event, file_offset, &sample, file_path); 1680 1681 /* These events are processed right away */ 1682 switch (event->header.type) { 1683 case PERF_RECORD_HEADER_ATTR: 1684 err = tool->attr(tool, event, &session->evlist); 1685 if (err == 0) { 1686 perf_session__set_id_hdr_size(session); 1687 perf_session__set_comm_exec(session); 1688 } 1689 return err; 1690 case PERF_RECORD_EVENT_UPDATE: 1691 return tool->event_update(tool, event, &session->evlist); 1692 case PERF_RECORD_HEADER_EVENT_TYPE: 1693 /* 1694 * Deprecated, but we need to handle it for sake 1695 * of old data files create in pipe mode. 1696 */ 1697 return 0; 1698 case PERF_RECORD_HEADER_TRACING_DATA: 1699 /* 1700 * Setup for reading amidst mmap, but only when we 1701 * are in 'file' mode. The 'pipe' fd is in proper 1702 * place already. 1703 */ 1704 if (!perf_data__is_pipe(session->data)) 1705 lseek(fd, file_offset, SEEK_SET); 1706 return tool->tracing_data(session, event); 1707 case PERF_RECORD_HEADER_BUILD_ID: 1708 return tool->build_id(session, event); 1709 case PERF_RECORD_FINISHED_ROUND: 1710 return tool->finished_round(tool, event, oe); 1711 case PERF_RECORD_ID_INDEX: 1712 return tool->id_index(session, event); 1713 case PERF_RECORD_AUXTRACE_INFO: 1714 return tool->auxtrace_info(session, event); 1715 case PERF_RECORD_AUXTRACE: 1716 /* 1717 * Setup for reading amidst mmap, but only when we 1718 * are in 'file' mode. The 'pipe' fd is in proper 1719 * place already. 1720 */ 1721 if (!perf_data__is_pipe(session->data)) 1722 lseek(fd, file_offset + event->header.size, SEEK_SET); 1723 return tool->auxtrace(session, event); 1724 case PERF_RECORD_AUXTRACE_ERROR: 1725 perf_session__auxtrace_error_inc(session, event); 1726 return tool->auxtrace_error(session, event); 1727 case PERF_RECORD_THREAD_MAP: 1728 return tool->thread_map(session, event); 1729 case PERF_RECORD_CPU_MAP: 1730 return tool->cpu_map(session, event); 1731 case PERF_RECORD_STAT_CONFIG: 1732 return tool->stat_config(session, event); 1733 case PERF_RECORD_STAT: 1734 return tool->stat(session, event); 1735 case PERF_RECORD_STAT_ROUND: 1736 return tool->stat_round(session, event); 1737 case PERF_RECORD_TIME_CONV: 1738 session->time_conv = event->time_conv; 1739 return tool->time_conv(session, event); 1740 case PERF_RECORD_HEADER_FEATURE: 1741 return tool->feature(session, event); 1742 case PERF_RECORD_COMPRESSED: 1743 err = tool->compressed(session, event, file_offset, file_path); 1744 if (err) 1745 dump_event(session->evlist, event, file_offset, &sample, file_path); 1746 return err; 1747 case PERF_RECORD_FINISHED_INIT: 1748 return tool->finished_init(session, event); 1749 default: 1750 return -EINVAL; 1751 } 1752 } 1753 1754 int perf_session__deliver_synth_event(struct perf_session *session, 1755 union perf_event *event, 1756 struct perf_sample *sample) 1757 { 1758 struct evlist *evlist = session->evlist; 1759 struct perf_tool *tool = session->tool; 1760 1761 events_stats__inc(&evlist->stats, event->header.type); 1762 1763 if (event->header.type >= PERF_RECORD_USER_TYPE_START) 1764 return perf_session__process_user_event(session, event, 0, NULL); 1765 1766 return machines__deliver_event(&session->machines, evlist, event, sample, tool, 0, NULL); 1767 } 1768 1769 static void event_swap(union perf_event *event, bool sample_id_all) 1770 { 1771 perf_event__swap_op swap; 1772 1773 swap = perf_event__swap_ops[event->header.type]; 1774 if (swap) 1775 swap(event, sample_id_all); 1776 } 1777 1778 int perf_session__peek_event(struct perf_session *session, off_t file_offset, 1779 void *buf, size_t buf_sz, 1780 union perf_event **event_ptr, 1781 struct perf_sample *sample) 1782 { 1783 union perf_event *event; 1784 size_t hdr_sz, rest; 1785 int fd; 1786 1787 if (session->one_mmap && !session->header.needs_swap) { 1788 event = file_offset - session->one_mmap_offset + 1789 session->one_mmap_addr; 1790 goto out_parse_sample; 1791 } 1792 1793 if (perf_data__is_pipe(session->data)) 1794 return -1; 1795 1796 fd = perf_data__fd(session->data); 1797 hdr_sz = sizeof(struct perf_event_header); 1798 1799 if (buf_sz < hdr_sz) 1800 return -1; 1801 1802 if (lseek(fd, file_offset, SEEK_SET) == (off_t)-1 || 1803 readn(fd, buf, hdr_sz) != (ssize_t)hdr_sz) 1804 return -1; 1805 1806 event = (union perf_event *)buf; 1807 1808 if (session->header.needs_swap) 1809 perf_event_header__bswap(&event->header); 1810 1811 if (event->header.size < hdr_sz || event->header.size > buf_sz) 1812 return -1; 1813 1814 buf += hdr_sz; 1815 rest = event->header.size - hdr_sz; 1816 1817 if (readn(fd, buf, rest) != (ssize_t)rest) 1818 return -1; 1819 1820 if (session->header.needs_swap) 1821 event_swap(event, evlist__sample_id_all(session->evlist)); 1822 1823 out_parse_sample: 1824 1825 if (sample && event->header.type < PERF_RECORD_USER_TYPE_START && 1826 evlist__parse_sample(session->evlist, event, sample)) 1827 return -1; 1828 1829 *event_ptr = event; 1830 1831 return 0; 1832 } 1833 1834 int perf_session__peek_events(struct perf_session *session, u64 offset, 1835 u64 size, peek_events_cb_t cb, void *data) 1836 { 1837 u64 max_offset = offset + size; 1838 char buf[PERF_SAMPLE_MAX_SIZE]; 1839 union perf_event *event; 1840 int err; 1841 1842 do { 1843 err = perf_session__peek_event(session, offset, buf, 1844 PERF_SAMPLE_MAX_SIZE, &event, 1845 NULL); 1846 if (err) 1847 return err; 1848 1849 err = cb(session, event, offset, data); 1850 if (err) 1851 return err; 1852 1853 offset += event->header.size; 1854 if (event->header.type == PERF_RECORD_AUXTRACE) 1855 offset += event->auxtrace.size; 1856 1857 } while (offset < max_offset); 1858 1859 return err; 1860 } 1861 1862 static s64 perf_session__process_event(struct perf_session *session, 1863 union perf_event *event, u64 file_offset, 1864 const char *file_path) 1865 { 1866 struct evlist *evlist = session->evlist; 1867 struct perf_tool *tool = session->tool; 1868 int ret; 1869 1870 if (session->header.needs_swap) 1871 event_swap(event, evlist__sample_id_all(evlist)); 1872 1873 if (event->header.type >= PERF_RECORD_HEADER_MAX) 1874 return -EINVAL; 1875 1876 events_stats__inc(&evlist->stats, event->header.type); 1877 1878 if (event->header.type >= PERF_RECORD_USER_TYPE_START) 1879 return perf_session__process_user_event(session, event, file_offset, file_path); 1880 1881 if (tool->ordered_events) { 1882 u64 timestamp = -1ULL; 1883 1884 ret = evlist__parse_sample_timestamp(evlist, event, ×tamp); 1885 if (ret && ret != -1) 1886 return ret; 1887 1888 ret = perf_session__queue_event(session, event, timestamp, file_offset, file_path); 1889 if (ret != -ETIME) 1890 return ret; 1891 } 1892 1893 return perf_session__deliver_event(session, event, tool, file_offset, file_path); 1894 } 1895 1896 void perf_event_header__bswap(struct perf_event_header *hdr) 1897 { 1898 hdr->type = bswap_32(hdr->type); 1899 hdr->misc = bswap_16(hdr->misc); 1900 hdr->size = bswap_16(hdr->size); 1901 } 1902 1903 struct thread *perf_session__findnew(struct perf_session *session, pid_t pid) 1904 { 1905 return machine__findnew_thread(&session->machines.host, -1, pid); 1906 } 1907 1908 int perf_session__register_idle_thread(struct perf_session *session) 1909 { 1910 struct thread *thread = machine__idle_thread(&session->machines.host); 1911 1912 /* machine__idle_thread() got the thread, so put it */ 1913 thread__put(thread); 1914 return thread ? 0 : -1; 1915 } 1916 1917 static void 1918 perf_session__warn_order(const struct perf_session *session) 1919 { 1920 const struct ordered_events *oe = &session->ordered_events; 1921 struct evsel *evsel; 1922 bool should_warn = true; 1923 1924 evlist__for_each_entry(session->evlist, evsel) { 1925 if (evsel->core.attr.write_backward) 1926 should_warn = false; 1927 } 1928 1929 if (!should_warn) 1930 return; 1931 if (oe->nr_unordered_events != 0) 1932 ui__warning("%u out of order events recorded.\n", oe->nr_unordered_events); 1933 } 1934 1935 static void perf_session__warn_about_errors(const struct perf_session *session) 1936 { 1937 const struct events_stats *stats = &session->evlist->stats; 1938 1939 if (session->tool->lost == perf_event__process_lost && 1940 stats->nr_events[PERF_RECORD_LOST] != 0) { 1941 ui__warning("Processed %d events and lost %d chunks!\n\n" 1942 "Check IO/CPU overload!\n\n", 1943 stats->nr_events[0], 1944 stats->nr_events[PERF_RECORD_LOST]); 1945 } 1946 1947 if (session->tool->lost_samples == perf_event__process_lost_samples) { 1948 double drop_rate; 1949 1950 drop_rate = (double)stats->total_lost_samples / 1951 (double) (stats->nr_events[PERF_RECORD_SAMPLE] + stats->total_lost_samples); 1952 if (drop_rate > 0.05) { 1953 ui__warning("Processed %" PRIu64 " samples and lost %3.2f%%!\n\n", 1954 stats->nr_events[PERF_RECORD_SAMPLE] + stats->total_lost_samples, 1955 drop_rate * 100.0); 1956 } 1957 } 1958 1959 if (session->tool->aux == perf_event__process_aux && 1960 stats->total_aux_lost != 0) { 1961 ui__warning("AUX data lost %" PRIu64 " times out of %u!\n\n", 1962 stats->total_aux_lost, 1963 stats->nr_events[PERF_RECORD_AUX]); 1964 } 1965 1966 if (session->tool->aux == perf_event__process_aux && 1967 stats->total_aux_partial != 0) { 1968 bool vmm_exclusive = false; 1969 1970 (void)sysfs__read_bool("module/kvm_intel/parameters/vmm_exclusive", 1971 &vmm_exclusive); 1972 1973 ui__warning("AUX data had gaps in it %" PRIu64 " times out of %u!\n\n" 1974 "Are you running a KVM guest in the background?%s\n\n", 1975 stats->total_aux_partial, 1976 stats->nr_events[PERF_RECORD_AUX], 1977 vmm_exclusive ? 1978 "\nReloading kvm_intel module with vmm_exclusive=0\n" 1979 "will reduce the gaps to only guest's timeslices." : 1980 ""); 1981 } 1982 1983 if (session->tool->aux == perf_event__process_aux && 1984 stats->total_aux_collision != 0) { 1985 ui__warning("AUX data detected collision %" PRIu64 " times out of %u!\n\n", 1986 stats->total_aux_collision, 1987 stats->nr_events[PERF_RECORD_AUX]); 1988 } 1989 1990 if (stats->nr_unknown_events != 0) { 1991 ui__warning("Found %u unknown events!\n\n" 1992 "Is this an older tool processing a perf.data " 1993 "file generated by a more recent tool?\n\n" 1994 "If that is not the case, consider " 1995 "reporting to linux-kernel@vger.kernel.org.\n\n", 1996 stats->nr_unknown_events); 1997 } 1998 1999 if (stats->nr_unknown_id != 0) { 2000 ui__warning("%u samples with id not present in the header\n", 2001 stats->nr_unknown_id); 2002 } 2003 2004 if (stats->nr_invalid_chains != 0) { 2005 ui__warning("Found invalid callchains!\n\n" 2006 "%u out of %u events were discarded for this reason.\n\n" 2007 "Consider reporting to linux-kernel@vger.kernel.org.\n\n", 2008 stats->nr_invalid_chains, 2009 stats->nr_events[PERF_RECORD_SAMPLE]); 2010 } 2011 2012 if (stats->nr_unprocessable_samples != 0) { 2013 ui__warning("%u unprocessable samples recorded.\n" 2014 "Do you have a KVM guest running and not using 'perf kvm'?\n", 2015 stats->nr_unprocessable_samples); 2016 } 2017 2018 perf_session__warn_order(session); 2019 2020 events_stats__auxtrace_error_warn(stats); 2021 2022 if (stats->nr_proc_map_timeout != 0) { 2023 ui__warning("%d map information files for pre-existing threads were\n" 2024 "not processed, if there are samples for addresses they\n" 2025 "will not be resolved, you may find out which are these\n" 2026 "threads by running with -v and redirecting the output\n" 2027 "to a file.\n" 2028 "The time limit to process proc map is too short?\n" 2029 "Increase it by --proc-map-timeout\n", 2030 stats->nr_proc_map_timeout); 2031 } 2032 } 2033 2034 static int perf_session__flush_thread_stack(struct thread *thread, 2035 void *p __maybe_unused) 2036 { 2037 return thread_stack__flush(thread); 2038 } 2039 2040 static int perf_session__flush_thread_stacks(struct perf_session *session) 2041 { 2042 return machines__for_each_thread(&session->machines, 2043 perf_session__flush_thread_stack, 2044 NULL); 2045 } 2046 2047 volatile sig_atomic_t session_done; 2048 2049 static int __perf_session__process_decomp_events(struct perf_session *session); 2050 2051 static int __perf_session__process_pipe_events(struct perf_session *session) 2052 { 2053 struct ordered_events *oe = &session->ordered_events; 2054 struct perf_tool *tool = session->tool; 2055 struct ui_progress prog; 2056 union perf_event *event; 2057 uint32_t size, cur_size = 0; 2058 void *buf = NULL; 2059 s64 skip = 0; 2060 u64 head; 2061 ssize_t err; 2062 void *p; 2063 bool update_prog = false; 2064 2065 perf_tool__fill_defaults(tool); 2066 2067 /* 2068 * If it's from a file saving pipe data (by redirection), it would have 2069 * a file name other than "-". Then we can get the total size and show 2070 * the progress. 2071 */ 2072 if (strcmp(session->data->path, "-") && session->data->file.size) { 2073 ui_progress__init_size(&prog, session->data->file.size, 2074 "Processing events..."); 2075 update_prog = true; 2076 } 2077 2078 head = 0; 2079 cur_size = sizeof(union perf_event); 2080 2081 buf = malloc(cur_size); 2082 if (!buf) 2083 return -errno; 2084 ordered_events__set_copy_on_queue(oe, true); 2085 more: 2086 event = buf; 2087 err = perf_data__read(session->data, event, 2088 sizeof(struct perf_event_header)); 2089 if (err <= 0) { 2090 if (err == 0) 2091 goto done; 2092 2093 pr_err("failed to read event header\n"); 2094 goto out_err; 2095 } 2096 2097 if (session->header.needs_swap) 2098 perf_event_header__bswap(&event->header); 2099 2100 size = event->header.size; 2101 if (size < sizeof(struct perf_event_header)) { 2102 pr_err("bad event header size\n"); 2103 goto out_err; 2104 } 2105 2106 if (size > cur_size) { 2107 void *new = realloc(buf, size); 2108 if (!new) { 2109 pr_err("failed to allocate memory to read event\n"); 2110 goto out_err; 2111 } 2112 buf = new; 2113 cur_size = size; 2114 event = buf; 2115 } 2116 p = event; 2117 p += sizeof(struct perf_event_header); 2118 2119 if (size - sizeof(struct perf_event_header)) { 2120 err = perf_data__read(session->data, p, 2121 size - sizeof(struct perf_event_header)); 2122 if (err <= 0) { 2123 if (err == 0) { 2124 pr_err("unexpected end of event stream\n"); 2125 goto done; 2126 } 2127 2128 pr_err("failed to read event data\n"); 2129 goto out_err; 2130 } 2131 } 2132 2133 if ((skip = perf_session__process_event(session, event, head, "pipe")) < 0) { 2134 pr_err("%#" PRIx64 " [%#x]: failed to process type: %d\n", 2135 head, event->header.size, event->header.type); 2136 err = -EINVAL; 2137 goto out_err; 2138 } 2139 2140 head += size; 2141 2142 if (skip > 0) 2143 head += skip; 2144 2145 err = __perf_session__process_decomp_events(session); 2146 if (err) 2147 goto out_err; 2148 2149 if (update_prog) 2150 ui_progress__update(&prog, size); 2151 2152 if (!session_done()) 2153 goto more; 2154 done: 2155 /* do the final flush for ordered samples */ 2156 err = ordered_events__flush(oe, OE_FLUSH__FINAL); 2157 if (err) 2158 goto out_err; 2159 err = auxtrace__flush_events(session, tool); 2160 if (err) 2161 goto out_err; 2162 err = perf_session__flush_thread_stacks(session); 2163 out_err: 2164 free(buf); 2165 if (update_prog) 2166 ui_progress__finish(); 2167 if (!tool->no_warn) 2168 perf_session__warn_about_errors(session); 2169 ordered_events__free(&session->ordered_events); 2170 auxtrace__free_events(session); 2171 return err; 2172 } 2173 2174 static union perf_event * 2175 prefetch_event(char *buf, u64 head, size_t mmap_size, 2176 bool needs_swap, union perf_event *error) 2177 { 2178 union perf_event *event; 2179 u16 event_size; 2180 2181 /* 2182 * Ensure we have enough space remaining to read 2183 * the size of the event in the headers. 2184 */ 2185 if (head + sizeof(event->header) > mmap_size) 2186 return NULL; 2187 2188 event = (union perf_event *)(buf + head); 2189 if (needs_swap) 2190 perf_event_header__bswap(&event->header); 2191 2192 event_size = event->header.size; 2193 if (head + event_size <= mmap_size) 2194 return event; 2195 2196 /* We're not fetching the event so swap back again */ 2197 if (needs_swap) 2198 perf_event_header__bswap(&event->header); 2199 2200 /* Check if the event fits into the next mmapped buf. */ 2201 if (event_size <= mmap_size - head % page_size) { 2202 /* Remap buf and fetch again. */ 2203 return NULL; 2204 } 2205 2206 /* Invalid input. Event size should never exceed mmap_size. */ 2207 pr_debug("%s: head=%#" PRIx64 " event->header.size=%#x, mmap_size=%#zx:" 2208 " fuzzed or compressed perf.data?\n", __func__, head, event_size, mmap_size); 2209 2210 return error; 2211 } 2212 2213 static union perf_event * 2214 fetch_mmaped_event(u64 head, size_t mmap_size, char *buf, bool needs_swap) 2215 { 2216 return prefetch_event(buf, head, mmap_size, needs_swap, ERR_PTR(-EINVAL)); 2217 } 2218 2219 static union perf_event * 2220 fetch_decomp_event(u64 head, size_t mmap_size, char *buf, bool needs_swap) 2221 { 2222 return prefetch_event(buf, head, mmap_size, needs_swap, NULL); 2223 } 2224 2225 static int __perf_session__process_decomp_events(struct perf_session *session) 2226 { 2227 s64 skip; 2228 u64 size; 2229 struct decomp *decomp = session->active_decomp->decomp_last; 2230 2231 if (!decomp) 2232 return 0; 2233 2234 while (decomp->head < decomp->size && !session_done()) { 2235 union perf_event *event = fetch_decomp_event(decomp->head, decomp->size, decomp->data, 2236 session->header.needs_swap); 2237 2238 if (!event) 2239 break; 2240 2241 size = event->header.size; 2242 2243 if (size < sizeof(struct perf_event_header) || 2244 (skip = perf_session__process_event(session, event, decomp->file_pos, 2245 decomp->file_path)) < 0) { 2246 pr_err("%#" PRIx64 " [%#x]: failed to process type: %d\n", 2247 decomp->file_pos + decomp->head, event->header.size, event->header.type); 2248 return -EINVAL; 2249 } 2250 2251 if (skip) 2252 size += skip; 2253 2254 decomp->head += size; 2255 } 2256 2257 return 0; 2258 } 2259 2260 /* 2261 * On 64bit we can mmap the data file in one go. No need for tiny mmap 2262 * slices. On 32bit we use 32MB. 2263 */ 2264 #if BITS_PER_LONG == 64 2265 #define MMAP_SIZE ULLONG_MAX 2266 #define NUM_MMAPS 1 2267 #else 2268 #define MMAP_SIZE (32 * 1024 * 1024ULL) 2269 #define NUM_MMAPS 128 2270 #endif 2271 2272 struct reader; 2273 2274 typedef s64 (*reader_cb_t)(struct perf_session *session, 2275 union perf_event *event, 2276 u64 file_offset, 2277 const char *file_path); 2278 2279 struct reader { 2280 int fd; 2281 const char *path; 2282 u64 data_size; 2283 u64 data_offset; 2284 reader_cb_t process; 2285 bool in_place_update; 2286 char *mmaps[NUM_MMAPS]; 2287 size_t mmap_size; 2288 int mmap_idx; 2289 char *mmap_cur; 2290 u64 file_pos; 2291 u64 file_offset; 2292 u64 head; 2293 u64 size; 2294 bool done; 2295 struct zstd_data zstd_data; 2296 struct decomp_data decomp_data; 2297 }; 2298 2299 static int 2300 reader__init(struct reader *rd, bool *one_mmap) 2301 { 2302 u64 data_size = rd->data_size; 2303 char **mmaps = rd->mmaps; 2304 2305 rd->head = rd->data_offset; 2306 data_size += rd->data_offset; 2307 2308 rd->mmap_size = MMAP_SIZE; 2309 if (rd->mmap_size > data_size) { 2310 rd->mmap_size = data_size; 2311 if (one_mmap) 2312 *one_mmap = true; 2313 } 2314 2315 memset(mmaps, 0, sizeof(rd->mmaps)); 2316 2317 if (zstd_init(&rd->zstd_data, 0)) 2318 return -1; 2319 rd->decomp_data.zstd_decomp = &rd->zstd_data; 2320 2321 return 0; 2322 } 2323 2324 static void 2325 reader__release_decomp(struct reader *rd) 2326 { 2327 perf_decomp__release_events(rd->decomp_data.decomp); 2328 zstd_fini(&rd->zstd_data); 2329 } 2330 2331 static int 2332 reader__mmap(struct reader *rd, struct perf_session *session) 2333 { 2334 int mmap_prot, mmap_flags; 2335 char *buf, **mmaps = rd->mmaps; 2336 u64 page_offset; 2337 2338 mmap_prot = PROT_READ; 2339 mmap_flags = MAP_SHARED; 2340 2341 if (rd->in_place_update) { 2342 mmap_prot |= PROT_WRITE; 2343 } else if (session->header.needs_swap) { 2344 mmap_prot |= PROT_WRITE; 2345 mmap_flags = MAP_PRIVATE; 2346 } 2347 2348 if (mmaps[rd->mmap_idx]) { 2349 munmap(mmaps[rd->mmap_idx], rd->mmap_size); 2350 mmaps[rd->mmap_idx] = NULL; 2351 } 2352 2353 page_offset = page_size * (rd->head / page_size); 2354 rd->file_offset += page_offset; 2355 rd->head -= page_offset; 2356 2357 buf = mmap(NULL, rd->mmap_size, mmap_prot, mmap_flags, rd->fd, 2358 rd->file_offset); 2359 if (buf == MAP_FAILED) { 2360 pr_err("failed to mmap file\n"); 2361 return -errno; 2362 } 2363 mmaps[rd->mmap_idx] = rd->mmap_cur = buf; 2364 rd->mmap_idx = (rd->mmap_idx + 1) & (ARRAY_SIZE(rd->mmaps) - 1); 2365 rd->file_pos = rd->file_offset + rd->head; 2366 if (session->one_mmap) { 2367 session->one_mmap_addr = buf; 2368 session->one_mmap_offset = rd->file_offset; 2369 } 2370 2371 return 0; 2372 } 2373 2374 enum { 2375 READER_OK, 2376 READER_NODATA, 2377 }; 2378 2379 static int 2380 reader__read_event(struct reader *rd, struct perf_session *session, 2381 struct ui_progress *prog) 2382 { 2383 u64 size; 2384 int err = READER_OK; 2385 union perf_event *event; 2386 s64 skip; 2387 2388 event = fetch_mmaped_event(rd->head, rd->mmap_size, rd->mmap_cur, 2389 session->header.needs_swap); 2390 if (IS_ERR(event)) 2391 return PTR_ERR(event); 2392 2393 if (!event) 2394 return READER_NODATA; 2395 2396 size = event->header.size; 2397 2398 skip = -EINVAL; 2399 2400 if (size < sizeof(struct perf_event_header) || 2401 (skip = rd->process(session, event, rd->file_pos, rd->path)) < 0) { 2402 pr_err("%#" PRIx64 " [%#x]: failed to process type: %d [%s]\n", 2403 rd->file_offset + rd->head, event->header.size, 2404 event->header.type, strerror(-skip)); 2405 err = skip; 2406 goto out; 2407 } 2408 2409 if (skip) 2410 size += skip; 2411 2412 rd->size += size; 2413 rd->head += size; 2414 rd->file_pos += size; 2415 2416 err = __perf_session__process_decomp_events(session); 2417 if (err) 2418 goto out; 2419 2420 ui_progress__update(prog, size); 2421 2422 out: 2423 return err; 2424 } 2425 2426 static inline bool 2427 reader__eof(struct reader *rd) 2428 { 2429 return (rd->file_pos >= rd->data_size + rd->data_offset); 2430 } 2431 2432 static int 2433 reader__process_events(struct reader *rd, struct perf_session *session, 2434 struct ui_progress *prog) 2435 { 2436 int err; 2437 2438 err = reader__init(rd, &session->one_mmap); 2439 if (err) 2440 goto out; 2441 2442 session->active_decomp = &rd->decomp_data; 2443 2444 remap: 2445 err = reader__mmap(rd, session); 2446 if (err) 2447 goto out; 2448 2449 more: 2450 err = reader__read_event(rd, session, prog); 2451 if (err < 0) 2452 goto out; 2453 else if (err == READER_NODATA) 2454 goto remap; 2455 2456 if (session_done()) 2457 goto out; 2458 2459 if (!reader__eof(rd)) 2460 goto more; 2461 2462 out: 2463 session->active_decomp = &session->decomp_data; 2464 return err; 2465 } 2466 2467 static s64 process_simple(struct perf_session *session, 2468 union perf_event *event, 2469 u64 file_offset, 2470 const char *file_path) 2471 { 2472 return perf_session__process_event(session, event, file_offset, file_path); 2473 } 2474 2475 static int __perf_session__process_events(struct perf_session *session) 2476 { 2477 struct reader rd = { 2478 .fd = perf_data__fd(session->data), 2479 .path = session->data->file.path, 2480 .data_size = session->header.data_size, 2481 .data_offset = session->header.data_offset, 2482 .process = process_simple, 2483 .in_place_update = session->data->in_place_update, 2484 }; 2485 struct ordered_events *oe = &session->ordered_events; 2486 struct perf_tool *tool = session->tool; 2487 struct ui_progress prog; 2488 int err; 2489 2490 perf_tool__fill_defaults(tool); 2491 2492 if (rd.data_size == 0) 2493 return -1; 2494 2495 ui_progress__init_size(&prog, rd.data_size, "Processing events..."); 2496 2497 err = reader__process_events(&rd, session, &prog); 2498 if (err) 2499 goto out_err; 2500 /* do the final flush for ordered samples */ 2501 err = ordered_events__flush(oe, OE_FLUSH__FINAL); 2502 if (err) 2503 goto out_err; 2504 err = auxtrace__flush_events(session, tool); 2505 if (err) 2506 goto out_err; 2507 err = perf_session__flush_thread_stacks(session); 2508 out_err: 2509 ui_progress__finish(); 2510 if (!tool->no_warn) 2511 perf_session__warn_about_errors(session); 2512 /* 2513 * We may switching perf.data output, make ordered_events 2514 * reusable. 2515 */ 2516 ordered_events__reinit(&session->ordered_events); 2517 auxtrace__free_events(session); 2518 reader__release_decomp(&rd); 2519 session->one_mmap = false; 2520 return err; 2521 } 2522 2523 /* 2524 * Processing 2 MB of data from each reader in sequence, 2525 * because that's the way the ordered events sorting works 2526 * most efficiently. 2527 */ 2528 #define READER_MAX_SIZE (2 * 1024 * 1024) 2529 2530 /* 2531 * This function reads, merge and process directory data. 2532 * It assumens the version 1 of directory data, where each 2533 * data file holds per-cpu data, already sorted by kernel. 2534 */ 2535 static int __perf_session__process_dir_events(struct perf_session *session) 2536 { 2537 struct perf_data *data = session->data; 2538 struct perf_tool *tool = session->tool; 2539 int i, ret, readers, nr_readers; 2540 struct ui_progress prog; 2541 u64 total_size = perf_data__size(session->data); 2542 struct reader *rd; 2543 2544 perf_tool__fill_defaults(tool); 2545 2546 ui_progress__init_size(&prog, total_size, "Processing events..."); 2547 2548 nr_readers = 1; 2549 for (i = 0; i < data->dir.nr; i++) { 2550 if (data->dir.files[i].size) 2551 nr_readers++; 2552 } 2553 2554 rd = zalloc(nr_readers * sizeof(struct reader)); 2555 if (!rd) 2556 return -ENOMEM; 2557 2558 rd[0] = (struct reader) { 2559 .fd = perf_data__fd(session->data), 2560 .path = session->data->file.path, 2561 .data_size = session->header.data_size, 2562 .data_offset = session->header.data_offset, 2563 .process = process_simple, 2564 .in_place_update = session->data->in_place_update, 2565 }; 2566 ret = reader__init(&rd[0], NULL); 2567 if (ret) 2568 goto out_err; 2569 ret = reader__mmap(&rd[0], session); 2570 if (ret) 2571 goto out_err; 2572 readers = 1; 2573 2574 for (i = 0; i < data->dir.nr; i++) { 2575 if (!data->dir.files[i].size) 2576 continue; 2577 rd[readers] = (struct reader) { 2578 .fd = data->dir.files[i].fd, 2579 .path = data->dir.files[i].path, 2580 .data_size = data->dir.files[i].size, 2581 .data_offset = 0, 2582 .process = process_simple, 2583 .in_place_update = session->data->in_place_update, 2584 }; 2585 ret = reader__init(&rd[readers], NULL); 2586 if (ret) 2587 goto out_err; 2588 ret = reader__mmap(&rd[readers], session); 2589 if (ret) 2590 goto out_err; 2591 readers++; 2592 } 2593 2594 i = 0; 2595 while (readers) { 2596 if (session_done()) 2597 break; 2598 2599 if (rd[i].done) { 2600 i = (i + 1) % nr_readers; 2601 continue; 2602 } 2603 if (reader__eof(&rd[i])) { 2604 rd[i].done = true; 2605 readers--; 2606 continue; 2607 } 2608 2609 session->active_decomp = &rd[i].decomp_data; 2610 ret = reader__read_event(&rd[i], session, &prog); 2611 if (ret < 0) { 2612 goto out_err; 2613 } else if (ret == READER_NODATA) { 2614 ret = reader__mmap(&rd[i], session); 2615 if (ret) 2616 goto out_err; 2617 } 2618 2619 if (rd[i].size >= READER_MAX_SIZE) { 2620 rd[i].size = 0; 2621 i = (i + 1) % nr_readers; 2622 } 2623 } 2624 2625 ret = ordered_events__flush(&session->ordered_events, OE_FLUSH__FINAL); 2626 if (ret) 2627 goto out_err; 2628 2629 ret = perf_session__flush_thread_stacks(session); 2630 out_err: 2631 ui_progress__finish(); 2632 2633 if (!tool->no_warn) 2634 perf_session__warn_about_errors(session); 2635 2636 /* 2637 * We may switching perf.data output, make ordered_events 2638 * reusable. 2639 */ 2640 ordered_events__reinit(&session->ordered_events); 2641 2642 session->one_mmap = false; 2643 2644 session->active_decomp = &session->decomp_data; 2645 for (i = 0; i < nr_readers; i++) 2646 reader__release_decomp(&rd[i]); 2647 zfree(&rd); 2648 2649 return ret; 2650 } 2651 2652 int perf_session__process_events(struct perf_session *session) 2653 { 2654 if (perf_session__register_idle_thread(session) < 0) 2655 return -ENOMEM; 2656 2657 if (perf_data__is_pipe(session->data)) 2658 return __perf_session__process_pipe_events(session); 2659 2660 if (perf_data__is_dir(session->data) && session->data->dir.nr) 2661 return __perf_session__process_dir_events(session); 2662 2663 return __perf_session__process_events(session); 2664 } 2665 2666 bool perf_session__has_traces(struct perf_session *session, const char *msg) 2667 { 2668 struct evsel *evsel; 2669 2670 evlist__for_each_entry(session->evlist, evsel) { 2671 if (evsel->core.attr.type == PERF_TYPE_TRACEPOINT) 2672 return true; 2673 } 2674 2675 pr_err("No trace sample to read. Did you call 'perf %s'?\n", msg); 2676 return false; 2677 } 2678 2679 int map__set_kallsyms_ref_reloc_sym(struct map *map, const char *symbol_name, u64 addr) 2680 { 2681 char *bracket; 2682 struct ref_reloc_sym *ref; 2683 struct kmap *kmap; 2684 2685 ref = zalloc(sizeof(struct ref_reloc_sym)); 2686 if (ref == NULL) 2687 return -ENOMEM; 2688 2689 ref->name = strdup(symbol_name); 2690 if (ref->name == NULL) { 2691 free(ref); 2692 return -ENOMEM; 2693 } 2694 2695 bracket = strchr(ref->name, ']'); 2696 if (bracket) 2697 *bracket = '\0'; 2698 2699 ref->addr = addr; 2700 2701 kmap = map__kmap(map); 2702 if (kmap) 2703 kmap->ref_reloc_sym = ref; 2704 2705 return 0; 2706 } 2707 2708 size_t perf_session__fprintf_dsos(struct perf_session *session, FILE *fp) 2709 { 2710 return machines__fprintf_dsos(&session->machines, fp); 2711 } 2712 2713 size_t perf_session__fprintf_dsos_buildid(struct perf_session *session, FILE *fp, 2714 bool (skip)(struct dso *dso, int parm), int parm) 2715 { 2716 return machines__fprintf_dsos_buildid(&session->machines, fp, skip, parm); 2717 } 2718 2719 size_t perf_session__fprintf_nr_events(struct perf_session *session, FILE *fp) 2720 { 2721 size_t ret; 2722 const char *msg = ""; 2723 2724 if (perf_header__has_feat(&session->header, HEADER_AUXTRACE)) 2725 msg = " (excludes AUX area (e.g. instruction trace) decoded / synthesized events)"; 2726 2727 ret = fprintf(fp, "\nAggregated stats:%s\n", msg); 2728 2729 ret += events_stats__fprintf(&session->evlist->stats, fp); 2730 return ret; 2731 } 2732 2733 size_t perf_session__fprintf(struct perf_session *session, FILE *fp) 2734 { 2735 /* 2736 * FIXME: Here we have to actually print all the machines in this 2737 * session, not just the host... 2738 */ 2739 return machine__fprintf(&session->machines.host, fp); 2740 } 2741 2742 void perf_session__dump_kmaps(struct perf_session *session) 2743 { 2744 int save_verbose = verbose; 2745 2746 fflush(stdout); 2747 fprintf(stderr, "Kernel and module maps:\n"); 2748 verbose = 0; /* Suppress verbose to print a summary only */ 2749 maps__fprintf(machine__kernel_maps(&session->machines.host), stderr); 2750 verbose = save_verbose; 2751 } 2752 2753 struct evsel *perf_session__find_first_evtype(struct perf_session *session, 2754 unsigned int type) 2755 { 2756 struct evsel *pos; 2757 2758 evlist__for_each_entry(session->evlist, pos) { 2759 if (pos->core.attr.type == type) 2760 return pos; 2761 } 2762 return NULL; 2763 } 2764 2765 int perf_session__cpu_bitmap(struct perf_session *session, 2766 const char *cpu_list, unsigned long *cpu_bitmap) 2767 { 2768 int i, err = -1; 2769 struct perf_cpu_map *map; 2770 int nr_cpus = min(session->header.env.nr_cpus_avail, MAX_NR_CPUS); 2771 struct perf_cpu cpu; 2772 2773 for (i = 0; i < PERF_TYPE_MAX; ++i) { 2774 struct evsel *evsel; 2775 2776 evsel = perf_session__find_first_evtype(session, i); 2777 if (!evsel) 2778 continue; 2779 2780 if (!(evsel->core.attr.sample_type & PERF_SAMPLE_CPU)) { 2781 pr_err("File does not contain CPU events. " 2782 "Remove -C option to proceed.\n"); 2783 return -1; 2784 } 2785 } 2786 2787 map = perf_cpu_map__new(cpu_list); 2788 if (map == NULL) { 2789 pr_err("Invalid cpu_list\n"); 2790 return -1; 2791 } 2792 2793 perf_cpu_map__for_each_cpu(cpu, i, map) { 2794 if (cpu.cpu >= nr_cpus) { 2795 pr_err("Requested CPU %d too large. " 2796 "Consider raising MAX_NR_CPUS\n", cpu.cpu); 2797 goto out_delete_map; 2798 } 2799 2800 __set_bit(cpu.cpu, cpu_bitmap); 2801 } 2802 2803 err = 0; 2804 2805 out_delete_map: 2806 perf_cpu_map__put(map); 2807 return err; 2808 } 2809 2810 void perf_session__fprintf_info(struct perf_session *session, FILE *fp, 2811 bool full) 2812 { 2813 if (session == NULL || fp == NULL) 2814 return; 2815 2816 fprintf(fp, "# ========\n"); 2817 perf_header__fprintf_info(session, fp, full); 2818 fprintf(fp, "# ========\n#\n"); 2819 } 2820 2821 static int perf_session__register_guest(struct perf_session *session, pid_t machine_pid) 2822 { 2823 struct machine *machine = machines__findnew(&session->machines, machine_pid); 2824 struct thread *thread; 2825 2826 if (!machine) 2827 return -ENOMEM; 2828 2829 machine->single_address_space = session->machines.host.single_address_space; 2830 2831 thread = machine__idle_thread(machine); 2832 if (!thread) 2833 return -ENOMEM; 2834 thread__put(thread); 2835 2836 machine->kallsyms_filename = perf_data__guest_kallsyms_name(session->data, machine_pid); 2837 2838 return 0; 2839 } 2840 2841 static int perf_session__set_guest_cpu(struct perf_session *session, pid_t pid, 2842 pid_t tid, int guest_cpu) 2843 { 2844 struct machine *machine = &session->machines.host; 2845 struct thread *thread = machine__findnew_thread(machine, pid, tid); 2846 2847 if (!thread) 2848 return -ENOMEM; 2849 thread__set_guest_cpu(thread, guest_cpu); 2850 thread__put(thread); 2851 2852 return 0; 2853 } 2854 2855 int perf_event__process_id_index(struct perf_session *session, 2856 union perf_event *event) 2857 { 2858 struct evlist *evlist = session->evlist; 2859 struct perf_record_id_index *ie = &event->id_index; 2860 size_t sz = ie->header.size - sizeof(*ie); 2861 size_t i, nr, max_nr; 2862 size_t e1_sz = sizeof(struct id_index_entry); 2863 size_t e2_sz = sizeof(struct id_index_entry_2); 2864 size_t etot_sz = e1_sz + e2_sz; 2865 struct id_index_entry_2 *e2; 2866 pid_t last_pid = 0; 2867 2868 max_nr = sz / e1_sz; 2869 nr = ie->nr; 2870 if (nr > max_nr) { 2871 printf("Too big: nr %zu max_nr %zu\n", nr, max_nr); 2872 return -EINVAL; 2873 } 2874 2875 if (sz >= nr * etot_sz) { 2876 max_nr = sz / etot_sz; 2877 if (nr > max_nr) { 2878 printf("Too big2: nr %zu max_nr %zu\n", nr, max_nr); 2879 return -EINVAL; 2880 } 2881 e2 = (void *)ie + sizeof(*ie) + nr * e1_sz; 2882 } else { 2883 e2 = NULL; 2884 } 2885 2886 if (dump_trace) 2887 fprintf(stdout, " nr: %zu\n", nr); 2888 2889 for (i = 0; i < nr; i++, (e2 ? e2++ : 0)) { 2890 struct id_index_entry *e = &ie->entries[i]; 2891 struct perf_sample_id *sid; 2892 int ret; 2893 2894 if (dump_trace) { 2895 fprintf(stdout, " ... id: %"PRI_lu64, e->id); 2896 fprintf(stdout, " idx: %"PRI_lu64, e->idx); 2897 fprintf(stdout, " cpu: %"PRI_ld64, e->cpu); 2898 fprintf(stdout, " tid: %"PRI_ld64, e->tid); 2899 if (e2) { 2900 fprintf(stdout, " machine_pid: %"PRI_ld64, e2->machine_pid); 2901 fprintf(stdout, " vcpu: %"PRI_lu64"\n", e2->vcpu); 2902 } else { 2903 fprintf(stdout, "\n"); 2904 } 2905 } 2906 2907 sid = evlist__id2sid(evlist, e->id); 2908 if (!sid) 2909 return -ENOENT; 2910 2911 sid->idx = e->idx; 2912 sid->cpu.cpu = e->cpu; 2913 sid->tid = e->tid; 2914 2915 if (!e2) 2916 continue; 2917 2918 sid->machine_pid = e2->machine_pid; 2919 sid->vcpu.cpu = e2->vcpu; 2920 2921 if (!sid->machine_pid) 2922 continue; 2923 2924 if (sid->machine_pid != last_pid) { 2925 ret = perf_session__register_guest(session, sid->machine_pid); 2926 if (ret) 2927 return ret; 2928 last_pid = sid->machine_pid; 2929 perf_guest = true; 2930 } 2931 2932 ret = perf_session__set_guest_cpu(session, sid->machine_pid, e->tid, e2->vcpu); 2933 if (ret) 2934 return ret; 2935 } 2936 return 0; 2937 } 2938 2939 int perf_session__dsos_hit_all(struct perf_session *session) 2940 { 2941 struct rb_node *nd; 2942 int err; 2943 2944 err = machine__hit_all_dsos(&session->machines.host); 2945 if (err) 2946 return err; 2947 2948 for (nd = rb_first_cached(&session->machines.guests); nd; 2949 nd = rb_next(nd)) { 2950 struct machine *pos = rb_entry(nd, struct machine, rb_node); 2951 2952 err = machine__hit_all_dsos(pos); 2953 if (err) 2954 return err; 2955 } 2956 2957 return 0; 2958 } 2959