1 #include <linux/kernel.h> 2 #include <traceevent/event-parse.h> 3 4 #include <byteswap.h> 5 #include <unistd.h> 6 #include <sys/types.h> 7 #include <sys/mman.h> 8 9 #include "evlist.h" 10 #include "evsel.h" 11 #include "session.h" 12 #include "tool.h" 13 #include "sort.h" 14 #include "util.h" 15 #include "cpumap.h" 16 #include "perf_regs.h" 17 #include "asm/bug.h" 18 #include "auxtrace.h" 19 #include "thread-stack.h" 20 21 static int perf_session__deliver_event(struct perf_session *session, 22 union perf_event *event, 23 struct perf_sample *sample, 24 struct perf_tool *tool, 25 u64 file_offset); 26 27 static int perf_session__open(struct perf_session *session) 28 { 29 struct perf_data_file *file = session->file; 30 31 if (perf_session__read_header(session) < 0) { 32 pr_err("incompatible file format (rerun with -v to learn more)"); 33 return -1; 34 } 35 36 if (perf_data_file__is_pipe(file)) 37 return 0; 38 39 if (!perf_evlist__valid_sample_type(session->evlist)) { 40 pr_err("non matching sample_type"); 41 return -1; 42 } 43 44 if (!perf_evlist__valid_sample_id_all(session->evlist)) { 45 pr_err("non matching sample_id_all"); 46 return -1; 47 } 48 49 if (!perf_evlist__valid_read_format(session->evlist)) { 50 pr_err("non matching read_format"); 51 return -1; 52 } 53 54 return 0; 55 } 56 57 void perf_session__set_id_hdr_size(struct perf_session *session) 58 { 59 u16 id_hdr_size = perf_evlist__id_hdr_size(session->evlist); 60 61 machines__set_id_hdr_size(&session->machines, id_hdr_size); 62 } 63 64 int perf_session__create_kernel_maps(struct perf_session *session) 65 { 66 int ret = machine__create_kernel_maps(&session->machines.host); 67 68 if (ret >= 0) 69 ret = machines__create_guest_kernel_maps(&session->machines); 70 return ret; 71 } 72 73 static void perf_session__destroy_kernel_maps(struct perf_session *session) 74 { 75 machines__destroy_kernel_maps(&session->machines); 76 } 77 78 static bool perf_session__has_comm_exec(struct perf_session *session) 79 { 80 struct perf_evsel *evsel; 81 82 evlist__for_each(session->evlist, evsel) { 83 if (evsel->attr.comm_exec) 84 return true; 85 } 86 87 return false; 88 } 89 90 static void perf_session__set_comm_exec(struct perf_session *session) 91 { 92 bool comm_exec = perf_session__has_comm_exec(session); 93 94 machines__set_comm_exec(&session->machines, comm_exec); 95 } 96 97 static int ordered_events__deliver_event(struct ordered_events *oe, 98 struct ordered_event *event) 99 { 100 struct perf_sample sample; 101 struct perf_session *session = container_of(oe, struct perf_session, 102 ordered_events); 103 int ret = perf_evlist__parse_sample(session->evlist, event->event, &sample); 104 105 if (ret) { 106 pr_err("Can't parse sample, err = %d\n", ret); 107 return ret; 108 } 109 110 return perf_session__deliver_event(session, event->event, &sample, 111 session->tool, event->file_offset); 112 } 113 114 struct perf_session *perf_session__new(struct perf_data_file *file, 115 bool repipe, struct perf_tool *tool) 116 { 117 struct perf_session *session = zalloc(sizeof(*session)); 118 119 if (!session) 120 goto out; 121 122 session->repipe = repipe; 123 session->tool = tool; 124 INIT_LIST_HEAD(&session->auxtrace_index); 125 machines__init(&session->machines); 126 ordered_events__init(&session->ordered_events, ordered_events__deliver_event); 127 128 if (file) { 129 if (perf_data_file__open(file)) 130 goto out_delete; 131 132 session->file = file; 133 134 if (perf_data_file__is_read(file)) { 135 if (perf_session__open(session) < 0) 136 goto out_close; 137 138 perf_session__set_id_hdr_size(session); 139 perf_session__set_comm_exec(session); 140 } 141 } 142 143 if (!file || perf_data_file__is_write(file)) { 144 /* 145 * In O_RDONLY mode this will be performed when reading the 146 * kernel MMAP event, in perf_event__process_mmap(). 147 */ 148 if (perf_session__create_kernel_maps(session) < 0) 149 pr_warning("Cannot read kernel map\n"); 150 } 151 152 if (tool && tool->ordering_requires_timestamps && 153 tool->ordered_events && !perf_evlist__sample_id_all(session->evlist)) { 154 dump_printf("WARNING: No sample_id_all support, falling back to unordered processing\n"); 155 tool->ordered_events = false; 156 } 157 158 return session; 159 160 out_close: 161 perf_data_file__close(file); 162 out_delete: 163 perf_session__delete(session); 164 out: 165 return NULL; 166 } 167 168 static void perf_session__delete_threads(struct perf_session *session) 169 { 170 machine__delete_threads(&session->machines.host); 171 } 172 173 static void perf_session_env__delete(struct perf_session_env *env) 174 { 175 zfree(&env->hostname); 176 zfree(&env->os_release); 177 zfree(&env->version); 178 zfree(&env->arch); 179 zfree(&env->cpu_desc); 180 zfree(&env->cpuid); 181 182 zfree(&env->cmdline); 183 zfree(&env->sibling_cores); 184 zfree(&env->sibling_threads); 185 zfree(&env->numa_nodes); 186 zfree(&env->pmu_mappings); 187 } 188 189 void perf_session__delete(struct perf_session *session) 190 { 191 auxtrace__free(session); 192 auxtrace_index__free(&session->auxtrace_index); 193 perf_session__destroy_kernel_maps(session); 194 perf_session__delete_threads(session); 195 perf_session_env__delete(&session->header.env); 196 machines__exit(&session->machines); 197 if (session->file) 198 perf_data_file__close(session->file); 199 free(session); 200 } 201 202 static int process_event_synth_tracing_data_stub(struct perf_tool *tool 203 __maybe_unused, 204 union perf_event *event 205 __maybe_unused, 206 struct perf_session *session 207 __maybe_unused) 208 { 209 dump_printf(": unhandled!\n"); 210 return 0; 211 } 212 213 static int process_event_synth_attr_stub(struct perf_tool *tool __maybe_unused, 214 union perf_event *event __maybe_unused, 215 struct perf_evlist **pevlist 216 __maybe_unused) 217 { 218 dump_printf(": unhandled!\n"); 219 return 0; 220 } 221 222 static int process_event_sample_stub(struct perf_tool *tool __maybe_unused, 223 union perf_event *event __maybe_unused, 224 struct perf_sample *sample __maybe_unused, 225 struct perf_evsel *evsel __maybe_unused, 226 struct machine *machine __maybe_unused) 227 { 228 dump_printf(": unhandled!\n"); 229 return 0; 230 } 231 232 static int process_event_stub(struct perf_tool *tool __maybe_unused, 233 union perf_event *event __maybe_unused, 234 struct perf_sample *sample __maybe_unused, 235 struct machine *machine __maybe_unused) 236 { 237 dump_printf(": unhandled!\n"); 238 return 0; 239 } 240 241 static int process_build_id_stub(struct perf_tool *tool __maybe_unused, 242 union perf_event *event __maybe_unused, 243 struct perf_session *session __maybe_unused) 244 { 245 dump_printf(": unhandled!\n"); 246 return 0; 247 } 248 249 static int process_finished_round_stub(struct perf_tool *tool __maybe_unused, 250 union perf_event *event __maybe_unused, 251 struct ordered_events *oe __maybe_unused) 252 { 253 dump_printf(": unhandled!\n"); 254 return 0; 255 } 256 257 static int process_finished_round(struct perf_tool *tool, 258 union perf_event *event, 259 struct ordered_events *oe); 260 261 static int process_id_index_stub(struct perf_tool *tool __maybe_unused, 262 union perf_event *event __maybe_unused, 263 struct perf_session *perf_session 264 __maybe_unused) 265 { 266 dump_printf(": unhandled!\n"); 267 return 0; 268 } 269 270 static int process_event_auxtrace_info_stub(struct perf_tool *tool __maybe_unused, 271 union perf_event *event __maybe_unused, 272 struct perf_session *session __maybe_unused) 273 { 274 dump_printf(": unhandled!\n"); 275 return 0; 276 } 277 278 static int skipn(int fd, off_t n) 279 { 280 char buf[4096]; 281 ssize_t ret; 282 283 while (n > 0) { 284 ret = read(fd, buf, min(n, (off_t)sizeof(buf))); 285 if (ret <= 0) 286 return ret; 287 n -= ret; 288 } 289 290 return 0; 291 } 292 293 static s64 process_event_auxtrace_stub(struct perf_tool *tool __maybe_unused, 294 union perf_event *event, 295 struct perf_session *session 296 __maybe_unused) 297 { 298 dump_printf(": unhandled!\n"); 299 if (perf_data_file__is_pipe(session->file)) 300 skipn(perf_data_file__fd(session->file), event->auxtrace.size); 301 return event->auxtrace.size; 302 } 303 304 static 305 int process_event_auxtrace_error_stub(struct perf_tool *tool __maybe_unused, 306 union perf_event *event __maybe_unused, 307 struct perf_session *session __maybe_unused) 308 { 309 dump_printf(": unhandled!\n"); 310 return 0; 311 } 312 313 void perf_tool__fill_defaults(struct perf_tool *tool) 314 { 315 if (tool->sample == NULL) 316 tool->sample = process_event_sample_stub; 317 if (tool->mmap == NULL) 318 tool->mmap = process_event_stub; 319 if (tool->mmap2 == NULL) 320 tool->mmap2 = process_event_stub; 321 if (tool->comm == NULL) 322 tool->comm = process_event_stub; 323 if (tool->fork == NULL) 324 tool->fork = process_event_stub; 325 if (tool->exit == NULL) 326 tool->exit = process_event_stub; 327 if (tool->lost == NULL) 328 tool->lost = perf_event__process_lost; 329 if (tool->lost_samples == NULL) 330 tool->lost_samples = perf_event__process_lost_samples; 331 if (tool->aux == NULL) 332 tool->aux = perf_event__process_aux; 333 if (tool->itrace_start == NULL) 334 tool->itrace_start = perf_event__process_itrace_start; 335 if (tool->read == NULL) 336 tool->read = process_event_sample_stub; 337 if (tool->throttle == NULL) 338 tool->throttle = process_event_stub; 339 if (tool->unthrottle == NULL) 340 tool->unthrottle = process_event_stub; 341 if (tool->attr == NULL) 342 tool->attr = process_event_synth_attr_stub; 343 if (tool->tracing_data == NULL) 344 tool->tracing_data = process_event_synth_tracing_data_stub; 345 if (tool->build_id == NULL) 346 tool->build_id = process_build_id_stub; 347 if (tool->finished_round == NULL) { 348 if (tool->ordered_events) 349 tool->finished_round = process_finished_round; 350 else 351 tool->finished_round = process_finished_round_stub; 352 } 353 if (tool->id_index == NULL) 354 tool->id_index = process_id_index_stub; 355 if (tool->auxtrace_info == NULL) 356 tool->auxtrace_info = process_event_auxtrace_info_stub; 357 if (tool->auxtrace == NULL) 358 tool->auxtrace = process_event_auxtrace_stub; 359 if (tool->auxtrace_error == NULL) 360 tool->auxtrace_error = process_event_auxtrace_error_stub; 361 } 362 363 static void swap_sample_id_all(union perf_event *event, void *data) 364 { 365 void *end = (void *) event + event->header.size; 366 int size = end - data; 367 368 BUG_ON(size % sizeof(u64)); 369 mem_bswap_64(data, size); 370 } 371 372 static void perf_event__all64_swap(union perf_event *event, 373 bool sample_id_all __maybe_unused) 374 { 375 struct perf_event_header *hdr = &event->header; 376 mem_bswap_64(hdr + 1, event->header.size - sizeof(*hdr)); 377 } 378 379 static void perf_event__comm_swap(union perf_event *event, bool sample_id_all) 380 { 381 event->comm.pid = bswap_32(event->comm.pid); 382 event->comm.tid = bswap_32(event->comm.tid); 383 384 if (sample_id_all) { 385 void *data = &event->comm.comm; 386 387 data += PERF_ALIGN(strlen(data) + 1, sizeof(u64)); 388 swap_sample_id_all(event, data); 389 } 390 } 391 392 static void perf_event__mmap_swap(union perf_event *event, 393 bool sample_id_all) 394 { 395 event->mmap.pid = bswap_32(event->mmap.pid); 396 event->mmap.tid = bswap_32(event->mmap.tid); 397 event->mmap.start = bswap_64(event->mmap.start); 398 event->mmap.len = bswap_64(event->mmap.len); 399 event->mmap.pgoff = bswap_64(event->mmap.pgoff); 400 401 if (sample_id_all) { 402 void *data = &event->mmap.filename; 403 404 data += PERF_ALIGN(strlen(data) + 1, sizeof(u64)); 405 swap_sample_id_all(event, data); 406 } 407 } 408 409 static void perf_event__mmap2_swap(union perf_event *event, 410 bool sample_id_all) 411 { 412 event->mmap2.pid = bswap_32(event->mmap2.pid); 413 event->mmap2.tid = bswap_32(event->mmap2.tid); 414 event->mmap2.start = bswap_64(event->mmap2.start); 415 event->mmap2.len = bswap_64(event->mmap2.len); 416 event->mmap2.pgoff = bswap_64(event->mmap2.pgoff); 417 event->mmap2.maj = bswap_32(event->mmap2.maj); 418 event->mmap2.min = bswap_32(event->mmap2.min); 419 event->mmap2.ino = bswap_64(event->mmap2.ino); 420 421 if (sample_id_all) { 422 void *data = &event->mmap2.filename; 423 424 data += PERF_ALIGN(strlen(data) + 1, sizeof(u64)); 425 swap_sample_id_all(event, data); 426 } 427 } 428 static void perf_event__task_swap(union perf_event *event, bool sample_id_all) 429 { 430 event->fork.pid = bswap_32(event->fork.pid); 431 event->fork.tid = bswap_32(event->fork.tid); 432 event->fork.ppid = bswap_32(event->fork.ppid); 433 event->fork.ptid = bswap_32(event->fork.ptid); 434 event->fork.time = bswap_64(event->fork.time); 435 436 if (sample_id_all) 437 swap_sample_id_all(event, &event->fork + 1); 438 } 439 440 static void perf_event__read_swap(union perf_event *event, bool sample_id_all) 441 { 442 event->read.pid = bswap_32(event->read.pid); 443 event->read.tid = bswap_32(event->read.tid); 444 event->read.value = bswap_64(event->read.value); 445 event->read.time_enabled = bswap_64(event->read.time_enabled); 446 event->read.time_running = bswap_64(event->read.time_running); 447 event->read.id = bswap_64(event->read.id); 448 449 if (sample_id_all) 450 swap_sample_id_all(event, &event->read + 1); 451 } 452 453 static void perf_event__aux_swap(union perf_event *event, bool sample_id_all) 454 { 455 event->aux.aux_offset = bswap_64(event->aux.aux_offset); 456 event->aux.aux_size = bswap_64(event->aux.aux_size); 457 event->aux.flags = bswap_64(event->aux.flags); 458 459 if (sample_id_all) 460 swap_sample_id_all(event, &event->aux + 1); 461 } 462 463 static void perf_event__itrace_start_swap(union perf_event *event, 464 bool sample_id_all) 465 { 466 event->itrace_start.pid = bswap_32(event->itrace_start.pid); 467 event->itrace_start.tid = bswap_32(event->itrace_start.tid); 468 469 if (sample_id_all) 470 swap_sample_id_all(event, &event->itrace_start + 1); 471 } 472 473 static void perf_event__throttle_swap(union perf_event *event, 474 bool sample_id_all) 475 { 476 event->throttle.time = bswap_64(event->throttle.time); 477 event->throttle.id = bswap_64(event->throttle.id); 478 event->throttle.stream_id = bswap_64(event->throttle.stream_id); 479 480 if (sample_id_all) 481 swap_sample_id_all(event, &event->throttle + 1); 482 } 483 484 static u8 revbyte(u8 b) 485 { 486 int rev = (b >> 4) | ((b & 0xf) << 4); 487 rev = ((rev & 0xcc) >> 2) | ((rev & 0x33) << 2); 488 rev = ((rev & 0xaa) >> 1) | ((rev & 0x55) << 1); 489 return (u8) rev; 490 } 491 492 /* 493 * XXX this is hack in attempt to carry flags bitfield 494 * throught endian village. ABI says: 495 * 496 * Bit-fields are allocated from right to left (least to most significant) 497 * on little-endian implementations and from left to right (most to least 498 * significant) on big-endian implementations. 499 * 500 * The above seems to be byte specific, so we need to reverse each 501 * byte of the bitfield. 'Internet' also says this might be implementation 502 * specific and we probably need proper fix and carry perf_event_attr 503 * bitfield flags in separate data file FEAT_ section. Thought this seems 504 * to work for now. 505 */ 506 static void swap_bitfield(u8 *p, unsigned len) 507 { 508 unsigned i; 509 510 for (i = 0; i < len; i++) { 511 *p = revbyte(*p); 512 p++; 513 } 514 } 515 516 /* exported for swapping attributes in file header */ 517 void perf_event__attr_swap(struct perf_event_attr *attr) 518 { 519 attr->type = bswap_32(attr->type); 520 attr->size = bswap_32(attr->size); 521 522 #define bswap_safe(f, n) \ 523 (attr->size > (offsetof(struct perf_event_attr, f) + \ 524 sizeof(attr->f) * (n))) 525 #define bswap_field(f, sz) \ 526 do { \ 527 if (bswap_safe(f, 0)) \ 528 attr->f = bswap_##sz(attr->f); \ 529 } while(0) 530 #define bswap_field_32(f) bswap_field(f, 32) 531 #define bswap_field_64(f) bswap_field(f, 64) 532 533 bswap_field_64(config); 534 bswap_field_64(sample_period); 535 bswap_field_64(sample_type); 536 bswap_field_64(read_format); 537 bswap_field_32(wakeup_events); 538 bswap_field_32(bp_type); 539 bswap_field_64(bp_addr); 540 bswap_field_64(bp_len); 541 bswap_field_64(branch_sample_type); 542 bswap_field_64(sample_regs_user); 543 bswap_field_32(sample_stack_user); 544 bswap_field_32(aux_watermark); 545 546 /* 547 * After read_format are bitfields. Check read_format because 548 * we are unable to use offsetof on bitfield. 549 */ 550 if (bswap_safe(read_format, 1)) 551 swap_bitfield((u8 *) (&attr->read_format + 1), 552 sizeof(u64)); 553 #undef bswap_field_64 554 #undef bswap_field_32 555 #undef bswap_field 556 #undef bswap_safe 557 } 558 559 static void perf_event__hdr_attr_swap(union perf_event *event, 560 bool sample_id_all __maybe_unused) 561 { 562 size_t size; 563 564 perf_event__attr_swap(&event->attr.attr); 565 566 size = event->header.size; 567 size -= (void *)&event->attr.id - (void *)event; 568 mem_bswap_64(event->attr.id, size); 569 } 570 571 static void perf_event__event_type_swap(union perf_event *event, 572 bool sample_id_all __maybe_unused) 573 { 574 event->event_type.event_type.event_id = 575 bswap_64(event->event_type.event_type.event_id); 576 } 577 578 static void perf_event__tracing_data_swap(union perf_event *event, 579 bool sample_id_all __maybe_unused) 580 { 581 event->tracing_data.size = bswap_32(event->tracing_data.size); 582 } 583 584 static void perf_event__auxtrace_info_swap(union perf_event *event, 585 bool sample_id_all __maybe_unused) 586 { 587 size_t size; 588 589 event->auxtrace_info.type = bswap_32(event->auxtrace_info.type); 590 591 size = event->header.size; 592 size -= (void *)&event->auxtrace_info.priv - (void *)event; 593 mem_bswap_64(event->auxtrace_info.priv, size); 594 } 595 596 static void perf_event__auxtrace_swap(union perf_event *event, 597 bool sample_id_all __maybe_unused) 598 { 599 event->auxtrace.size = bswap_64(event->auxtrace.size); 600 event->auxtrace.offset = bswap_64(event->auxtrace.offset); 601 event->auxtrace.reference = bswap_64(event->auxtrace.reference); 602 event->auxtrace.idx = bswap_32(event->auxtrace.idx); 603 event->auxtrace.tid = bswap_32(event->auxtrace.tid); 604 event->auxtrace.cpu = bswap_32(event->auxtrace.cpu); 605 } 606 607 static void perf_event__auxtrace_error_swap(union perf_event *event, 608 bool sample_id_all __maybe_unused) 609 { 610 event->auxtrace_error.type = bswap_32(event->auxtrace_error.type); 611 event->auxtrace_error.code = bswap_32(event->auxtrace_error.code); 612 event->auxtrace_error.cpu = bswap_32(event->auxtrace_error.cpu); 613 event->auxtrace_error.pid = bswap_32(event->auxtrace_error.pid); 614 event->auxtrace_error.tid = bswap_32(event->auxtrace_error.tid); 615 event->auxtrace_error.ip = bswap_64(event->auxtrace_error.ip); 616 } 617 618 typedef void (*perf_event__swap_op)(union perf_event *event, 619 bool sample_id_all); 620 621 static perf_event__swap_op perf_event__swap_ops[] = { 622 [PERF_RECORD_MMAP] = perf_event__mmap_swap, 623 [PERF_RECORD_MMAP2] = perf_event__mmap2_swap, 624 [PERF_RECORD_COMM] = perf_event__comm_swap, 625 [PERF_RECORD_FORK] = perf_event__task_swap, 626 [PERF_RECORD_EXIT] = perf_event__task_swap, 627 [PERF_RECORD_LOST] = perf_event__all64_swap, 628 [PERF_RECORD_READ] = perf_event__read_swap, 629 [PERF_RECORD_THROTTLE] = perf_event__throttle_swap, 630 [PERF_RECORD_UNTHROTTLE] = perf_event__throttle_swap, 631 [PERF_RECORD_SAMPLE] = perf_event__all64_swap, 632 [PERF_RECORD_AUX] = perf_event__aux_swap, 633 [PERF_RECORD_ITRACE_START] = perf_event__itrace_start_swap, 634 [PERF_RECORD_LOST_SAMPLES] = perf_event__all64_swap, 635 [PERF_RECORD_HEADER_ATTR] = perf_event__hdr_attr_swap, 636 [PERF_RECORD_HEADER_EVENT_TYPE] = perf_event__event_type_swap, 637 [PERF_RECORD_HEADER_TRACING_DATA] = perf_event__tracing_data_swap, 638 [PERF_RECORD_HEADER_BUILD_ID] = NULL, 639 [PERF_RECORD_ID_INDEX] = perf_event__all64_swap, 640 [PERF_RECORD_AUXTRACE_INFO] = perf_event__auxtrace_info_swap, 641 [PERF_RECORD_AUXTRACE] = perf_event__auxtrace_swap, 642 [PERF_RECORD_AUXTRACE_ERROR] = perf_event__auxtrace_error_swap, 643 [PERF_RECORD_HEADER_MAX] = NULL, 644 }; 645 646 /* 647 * When perf record finishes a pass on every buffers, it records this pseudo 648 * event. 649 * We record the max timestamp t found in the pass n. 650 * Assuming these timestamps are monotonic across cpus, we know that if 651 * a buffer still has events with timestamps below t, they will be all 652 * available and then read in the pass n + 1. 653 * Hence when we start to read the pass n + 2, we can safely flush every 654 * events with timestamps below t. 655 * 656 * ============ PASS n ================= 657 * CPU 0 | CPU 1 658 * | 659 * cnt1 timestamps | cnt2 timestamps 660 * 1 | 2 661 * 2 | 3 662 * - | 4 <--- max recorded 663 * 664 * ============ PASS n + 1 ============== 665 * CPU 0 | CPU 1 666 * | 667 * cnt1 timestamps | cnt2 timestamps 668 * 3 | 5 669 * 4 | 6 670 * 5 | 7 <---- max recorded 671 * 672 * Flush every events below timestamp 4 673 * 674 * ============ PASS n + 2 ============== 675 * CPU 0 | CPU 1 676 * | 677 * cnt1 timestamps | cnt2 timestamps 678 * 6 | 8 679 * 7 | 9 680 * - | 10 681 * 682 * Flush every events below timestamp 7 683 * etc... 684 */ 685 static int process_finished_round(struct perf_tool *tool __maybe_unused, 686 union perf_event *event __maybe_unused, 687 struct ordered_events *oe) 688 { 689 return ordered_events__flush(oe, OE_FLUSH__ROUND); 690 } 691 692 int perf_session__queue_event(struct perf_session *s, union perf_event *event, 693 struct perf_sample *sample, u64 file_offset) 694 { 695 return ordered_events__queue(&s->ordered_events, event, sample, file_offset); 696 } 697 698 static void callchain__lbr_callstack_printf(struct perf_sample *sample) 699 { 700 struct ip_callchain *callchain = sample->callchain; 701 struct branch_stack *lbr_stack = sample->branch_stack; 702 u64 kernel_callchain_nr = callchain->nr; 703 unsigned int i; 704 705 for (i = 0; i < kernel_callchain_nr; i++) { 706 if (callchain->ips[i] == PERF_CONTEXT_USER) 707 break; 708 } 709 710 if ((i != kernel_callchain_nr) && lbr_stack->nr) { 711 u64 total_nr; 712 /* 713 * LBR callstack can only get user call chain, 714 * i is kernel call chain number, 715 * 1 is PERF_CONTEXT_USER. 716 * 717 * The user call chain is stored in LBR registers. 718 * LBR are pair registers. The caller is stored 719 * in "from" register, while the callee is stored 720 * in "to" register. 721 * For example, there is a call stack 722 * "A"->"B"->"C"->"D". 723 * The LBR registers will recorde like 724 * "C"->"D", "B"->"C", "A"->"B". 725 * So only the first "to" register and all "from" 726 * registers are needed to construct the whole stack. 727 */ 728 total_nr = i + 1 + lbr_stack->nr + 1; 729 kernel_callchain_nr = i + 1; 730 731 printf("... LBR call chain: nr:%" PRIu64 "\n", total_nr); 732 733 for (i = 0; i < kernel_callchain_nr; i++) 734 printf("..... %2d: %016" PRIx64 "\n", 735 i, callchain->ips[i]); 736 737 printf("..... %2d: %016" PRIx64 "\n", 738 (int)(kernel_callchain_nr), lbr_stack->entries[0].to); 739 for (i = 0; i < lbr_stack->nr; i++) 740 printf("..... %2d: %016" PRIx64 "\n", 741 (int)(i + kernel_callchain_nr + 1), lbr_stack->entries[i].from); 742 } 743 } 744 745 static void callchain__printf(struct perf_evsel *evsel, 746 struct perf_sample *sample) 747 { 748 unsigned int i; 749 struct ip_callchain *callchain = sample->callchain; 750 751 if (has_branch_callstack(evsel)) 752 callchain__lbr_callstack_printf(sample); 753 754 printf("... FP chain: nr:%" PRIu64 "\n", callchain->nr); 755 756 for (i = 0; i < callchain->nr; i++) 757 printf("..... %2d: %016" PRIx64 "\n", 758 i, callchain->ips[i]); 759 } 760 761 static void branch_stack__printf(struct perf_sample *sample) 762 { 763 uint64_t i; 764 765 printf("... branch stack: nr:%" PRIu64 "\n", sample->branch_stack->nr); 766 767 for (i = 0; i < sample->branch_stack->nr; i++) 768 printf("..... %2"PRIu64": %016" PRIx64 " -> %016" PRIx64 "\n", 769 i, sample->branch_stack->entries[i].from, 770 sample->branch_stack->entries[i].to); 771 } 772 773 static void regs_dump__printf(u64 mask, u64 *regs) 774 { 775 unsigned rid, i = 0; 776 777 for_each_set_bit(rid, (unsigned long *) &mask, sizeof(mask) * 8) { 778 u64 val = regs[i++]; 779 780 printf(".... %-5s 0x%" PRIx64 "\n", 781 perf_reg_name(rid), val); 782 } 783 } 784 785 static const char *regs_abi[] = { 786 [PERF_SAMPLE_REGS_ABI_NONE] = "none", 787 [PERF_SAMPLE_REGS_ABI_32] = "32-bit", 788 [PERF_SAMPLE_REGS_ABI_64] = "64-bit", 789 }; 790 791 static inline const char *regs_dump_abi(struct regs_dump *d) 792 { 793 if (d->abi > PERF_SAMPLE_REGS_ABI_64) 794 return "unknown"; 795 796 return regs_abi[d->abi]; 797 } 798 799 static void regs__printf(const char *type, struct regs_dump *regs) 800 { 801 u64 mask = regs->mask; 802 803 printf("... %s regs: mask 0x%" PRIx64 " ABI %s\n", 804 type, 805 mask, 806 regs_dump_abi(regs)); 807 808 regs_dump__printf(mask, regs->regs); 809 } 810 811 static void regs_user__printf(struct perf_sample *sample) 812 { 813 struct regs_dump *user_regs = &sample->user_regs; 814 815 if (user_regs->regs) 816 regs__printf("user", user_regs); 817 } 818 819 static void regs_intr__printf(struct perf_sample *sample) 820 { 821 struct regs_dump *intr_regs = &sample->intr_regs; 822 823 if (intr_regs->regs) 824 regs__printf("intr", intr_regs); 825 } 826 827 static void stack_user__printf(struct stack_dump *dump) 828 { 829 printf("... ustack: size %" PRIu64 ", offset 0x%x\n", 830 dump->size, dump->offset); 831 } 832 833 static void perf_evlist__print_tstamp(struct perf_evlist *evlist, 834 union perf_event *event, 835 struct perf_sample *sample) 836 { 837 u64 sample_type = __perf_evlist__combined_sample_type(evlist); 838 839 if (event->header.type != PERF_RECORD_SAMPLE && 840 !perf_evlist__sample_id_all(evlist)) { 841 fputs("-1 -1 ", stdout); 842 return; 843 } 844 845 if ((sample_type & PERF_SAMPLE_CPU)) 846 printf("%u ", sample->cpu); 847 848 if (sample_type & PERF_SAMPLE_TIME) 849 printf("%" PRIu64 " ", sample->time); 850 } 851 852 static void sample_read__printf(struct perf_sample *sample, u64 read_format) 853 { 854 printf("... sample_read:\n"); 855 856 if (read_format & PERF_FORMAT_TOTAL_TIME_ENABLED) 857 printf("...... time enabled %016" PRIx64 "\n", 858 sample->read.time_enabled); 859 860 if (read_format & PERF_FORMAT_TOTAL_TIME_RUNNING) 861 printf("...... time running %016" PRIx64 "\n", 862 sample->read.time_running); 863 864 if (read_format & PERF_FORMAT_GROUP) { 865 u64 i; 866 867 printf(".... group nr %" PRIu64 "\n", sample->read.group.nr); 868 869 for (i = 0; i < sample->read.group.nr; i++) { 870 struct sample_read_value *value; 871 872 value = &sample->read.group.values[i]; 873 printf("..... id %016" PRIx64 874 ", value %016" PRIx64 "\n", 875 value->id, value->value); 876 } 877 } else 878 printf("..... id %016" PRIx64 ", value %016" PRIx64 "\n", 879 sample->read.one.id, sample->read.one.value); 880 } 881 882 static void dump_event(struct perf_evlist *evlist, union perf_event *event, 883 u64 file_offset, struct perf_sample *sample) 884 { 885 if (!dump_trace) 886 return; 887 888 printf("\n%#" PRIx64 " [%#x]: event: %d\n", 889 file_offset, event->header.size, event->header.type); 890 891 trace_event(event); 892 893 if (sample) 894 perf_evlist__print_tstamp(evlist, event, sample); 895 896 printf("%#" PRIx64 " [%#x]: PERF_RECORD_%s", file_offset, 897 event->header.size, perf_event__name(event->header.type)); 898 } 899 900 static void dump_sample(struct perf_evsel *evsel, union perf_event *event, 901 struct perf_sample *sample) 902 { 903 u64 sample_type; 904 905 if (!dump_trace) 906 return; 907 908 printf("(IP, 0x%x): %d/%d: %#" PRIx64 " period: %" PRIu64 " addr: %#" PRIx64 "\n", 909 event->header.misc, sample->pid, sample->tid, sample->ip, 910 sample->period, sample->addr); 911 912 sample_type = evsel->attr.sample_type; 913 914 if (sample_type & PERF_SAMPLE_CALLCHAIN) 915 callchain__printf(evsel, sample); 916 917 if ((sample_type & PERF_SAMPLE_BRANCH_STACK) && !has_branch_callstack(evsel)) 918 branch_stack__printf(sample); 919 920 if (sample_type & PERF_SAMPLE_REGS_USER) 921 regs_user__printf(sample); 922 923 if (sample_type & PERF_SAMPLE_REGS_INTR) 924 regs_intr__printf(sample); 925 926 if (sample_type & PERF_SAMPLE_STACK_USER) 927 stack_user__printf(&sample->user_stack); 928 929 if (sample_type & PERF_SAMPLE_WEIGHT) 930 printf("... weight: %" PRIu64 "\n", sample->weight); 931 932 if (sample_type & PERF_SAMPLE_DATA_SRC) 933 printf(" . data_src: 0x%"PRIx64"\n", sample->data_src); 934 935 if (sample_type & PERF_SAMPLE_TRANSACTION) 936 printf("... transaction: %" PRIx64 "\n", sample->transaction); 937 938 if (sample_type & PERF_SAMPLE_READ) 939 sample_read__printf(sample, evsel->attr.read_format); 940 } 941 942 static struct machine *machines__find_for_cpumode(struct machines *machines, 943 union perf_event *event, 944 struct perf_sample *sample) 945 { 946 const u8 cpumode = event->header.misc & PERF_RECORD_MISC_CPUMODE_MASK; 947 struct machine *machine; 948 949 if (perf_guest && 950 ((cpumode == PERF_RECORD_MISC_GUEST_KERNEL) || 951 (cpumode == PERF_RECORD_MISC_GUEST_USER))) { 952 u32 pid; 953 954 if (event->header.type == PERF_RECORD_MMAP 955 || event->header.type == PERF_RECORD_MMAP2) 956 pid = event->mmap.pid; 957 else 958 pid = sample->pid; 959 960 machine = machines__find(machines, pid); 961 if (!machine) 962 machine = machines__find(machines, DEFAULT_GUEST_KERNEL_ID); 963 return machine; 964 } 965 966 return &machines->host; 967 } 968 969 static int deliver_sample_value(struct perf_evlist *evlist, 970 struct perf_tool *tool, 971 union perf_event *event, 972 struct perf_sample *sample, 973 struct sample_read_value *v, 974 struct machine *machine) 975 { 976 struct perf_sample_id *sid = perf_evlist__id2sid(evlist, v->id); 977 978 if (sid) { 979 sample->id = v->id; 980 sample->period = v->value - sid->period; 981 sid->period = v->value; 982 } 983 984 if (!sid || sid->evsel == NULL) { 985 ++evlist->stats.nr_unknown_id; 986 return 0; 987 } 988 989 return tool->sample(tool, event, sample, sid->evsel, machine); 990 } 991 992 static int deliver_sample_group(struct perf_evlist *evlist, 993 struct perf_tool *tool, 994 union perf_event *event, 995 struct perf_sample *sample, 996 struct machine *machine) 997 { 998 int ret = -EINVAL; 999 u64 i; 1000 1001 for (i = 0; i < sample->read.group.nr; i++) { 1002 ret = deliver_sample_value(evlist, tool, event, sample, 1003 &sample->read.group.values[i], 1004 machine); 1005 if (ret) 1006 break; 1007 } 1008 1009 return ret; 1010 } 1011 1012 static int 1013 perf_evlist__deliver_sample(struct perf_evlist *evlist, 1014 struct perf_tool *tool, 1015 union perf_event *event, 1016 struct perf_sample *sample, 1017 struct perf_evsel *evsel, 1018 struct machine *machine) 1019 { 1020 /* We know evsel != NULL. */ 1021 u64 sample_type = evsel->attr.sample_type; 1022 u64 read_format = evsel->attr.read_format; 1023 1024 /* Standard sample delievery. */ 1025 if (!(sample_type & PERF_SAMPLE_READ)) 1026 return tool->sample(tool, event, sample, evsel, machine); 1027 1028 /* For PERF_SAMPLE_READ we have either single or group mode. */ 1029 if (read_format & PERF_FORMAT_GROUP) 1030 return deliver_sample_group(evlist, tool, event, sample, 1031 machine); 1032 else 1033 return deliver_sample_value(evlist, tool, event, sample, 1034 &sample->read.one, machine); 1035 } 1036 1037 static int machines__deliver_event(struct machines *machines, 1038 struct perf_evlist *evlist, 1039 union perf_event *event, 1040 struct perf_sample *sample, 1041 struct perf_tool *tool, u64 file_offset) 1042 { 1043 struct perf_evsel *evsel; 1044 struct machine *machine; 1045 1046 dump_event(evlist, event, file_offset, sample); 1047 1048 evsel = perf_evlist__id2evsel(evlist, sample->id); 1049 1050 machine = machines__find_for_cpumode(machines, event, sample); 1051 1052 switch (event->header.type) { 1053 case PERF_RECORD_SAMPLE: 1054 dump_sample(evsel, event, sample); 1055 if (evsel == NULL) { 1056 ++evlist->stats.nr_unknown_id; 1057 return 0; 1058 } 1059 if (machine == NULL) { 1060 ++evlist->stats.nr_unprocessable_samples; 1061 return 0; 1062 } 1063 return perf_evlist__deliver_sample(evlist, tool, event, sample, evsel, machine); 1064 case PERF_RECORD_MMAP: 1065 return tool->mmap(tool, event, sample, machine); 1066 case PERF_RECORD_MMAP2: 1067 if (event->header.misc & PERF_RECORD_MISC_PROC_MAP_PARSE_TIMEOUT) 1068 ++evlist->stats.nr_proc_map_timeout; 1069 return tool->mmap2(tool, event, sample, machine); 1070 case PERF_RECORD_COMM: 1071 return tool->comm(tool, event, sample, machine); 1072 case PERF_RECORD_FORK: 1073 return tool->fork(tool, event, sample, machine); 1074 case PERF_RECORD_EXIT: 1075 return tool->exit(tool, event, sample, machine); 1076 case PERF_RECORD_LOST: 1077 if (tool->lost == perf_event__process_lost) 1078 evlist->stats.total_lost += event->lost.lost; 1079 return tool->lost(tool, event, sample, machine); 1080 case PERF_RECORD_LOST_SAMPLES: 1081 if (tool->lost_samples == perf_event__process_lost_samples) 1082 evlist->stats.total_lost_samples += event->lost_samples.lost; 1083 return tool->lost_samples(tool, event, sample, machine); 1084 case PERF_RECORD_READ: 1085 return tool->read(tool, event, sample, evsel, machine); 1086 case PERF_RECORD_THROTTLE: 1087 return tool->throttle(tool, event, sample, machine); 1088 case PERF_RECORD_UNTHROTTLE: 1089 return tool->unthrottle(tool, event, sample, machine); 1090 case PERF_RECORD_AUX: 1091 return tool->aux(tool, event, sample, machine); 1092 case PERF_RECORD_ITRACE_START: 1093 return tool->itrace_start(tool, event, sample, machine); 1094 default: 1095 ++evlist->stats.nr_unknown_events; 1096 return -1; 1097 } 1098 } 1099 1100 static int perf_session__deliver_event(struct perf_session *session, 1101 union perf_event *event, 1102 struct perf_sample *sample, 1103 struct perf_tool *tool, 1104 u64 file_offset) 1105 { 1106 int ret; 1107 1108 ret = auxtrace__process_event(session, event, sample, tool); 1109 if (ret < 0) 1110 return ret; 1111 if (ret > 0) 1112 return 0; 1113 1114 return machines__deliver_event(&session->machines, session->evlist, 1115 event, sample, tool, file_offset); 1116 } 1117 1118 static s64 perf_session__process_user_event(struct perf_session *session, 1119 union perf_event *event, 1120 u64 file_offset) 1121 { 1122 struct ordered_events *oe = &session->ordered_events; 1123 struct perf_tool *tool = session->tool; 1124 int fd = perf_data_file__fd(session->file); 1125 int err; 1126 1127 dump_event(session->evlist, event, file_offset, NULL); 1128 1129 /* These events are processed right away */ 1130 switch (event->header.type) { 1131 case PERF_RECORD_HEADER_ATTR: 1132 err = tool->attr(tool, event, &session->evlist); 1133 if (err == 0) { 1134 perf_session__set_id_hdr_size(session); 1135 perf_session__set_comm_exec(session); 1136 } 1137 return err; 1138 case PERF_RECORD_HEADER_EVENT_TYPE: 1139 /* 1140 * Depreceated, but we need to handle it for sake 1141 * of old data files create in pipe mode. 1142 */ 1143 return 0; 1144 case PERF_RECORD_HEADER_TRACING_DATA: 1145 /* setup for reading amidst mmap */ 1146 lseek(fd, file_offset, SEEK_SET); 1147 return tool->tracing_data(tool, event, session); 1148 case PERF_RECORD_HEADER_BUILD_ID: 1149 return tool->build_id(tool, event, session); 1150 case PERF_RECORD_FINISHED_ROUND: 1151 return tool->finished_round(tool, event, oe); 1152 case PERF_RECORD_ID_INDEX: 1153 return tool->id_index(tool, event, session); 1154 case PERF_RECORD_AUXTRACE_INFO: 1155 return tool->auxtrace_info(tool, event, session); 1156 case PERF_RECORD_AUXTRACE: 1157 /* setup for reading amidst mmap */ 1158 lseek(fd, file_offset + event->header.size, SEEK_SET); 1159 return tool->auxtrace(tool, event, session); 1160 case PERF_RECORD_AUXTRACE_ERROR: 1161 perf_session__auxtrace_error_inc(session, event); 1162 return tool->auxtrace_error(tool, event, session); 1163 default: 1164 return -EINVAL; 1165 } 1166 } 1167 1168 int perf_session__deliver_synth_event(struct perf_session *session, 1169 union perf_event *event, 1170 struct perf_sample *sample) 1171 { 1172 struct perf_evlist *evlist = session->evlist; 1173 struct perf_tool *tool = session->tool; 1174 1175 events_stats__inc(&evlist->stats, event->header.type); 1176 1177 if (event->header.type >= PERF_RECORD_USER_TYPE_START) 1178 return perf_session__process_user_event(session, event, 0); 1179 1180 return machines__deliver_event(&session->machines, evlist, event, sample, tool, 0); 1181 } 1182 1183 static void event_swap(union perf_event *event, bool sample_id_all) 1184 { 1185 perf_event__swap_op swap; 1186 1187 swap = perf_event__swap_ops[event->header.type]; 1188 if (swap) 1189 swap(event, sample_id_all); 1190 } 1191 1192 int perf_session__peek_event(struct perf_session *session, off_t file_offset, 1193 void *buf, size_t buf_sz, 1194 union perf_event **event_ptr, 1195 struct perf_sample *sample) 1196 { 1197 union perf_event *event; 1198 size_t hdr_sz, rest; 1199 int fd; 1200 1201 if (session->one_mmap && !session->header.needs_swap) { 1202 event = file_offset - session->one_mmap_offset + 1203 session->one_mmap_addr; 1204 goto out_parse_sample; 1205 } 1206 1207 if (perf_data_file__is_pipe(session->file)) 1208 return -1; 1209 1210 fd = perf_data_file__fd(session->file); 1211 hdr_sz = sizeof(struct perf_event_header); 1212 1213 if (buf_sz < hdr_sz) 1214 return -1; 1215 1216 if (lseek(fd, file_offset, SEEK_SET) == (off_t)-1 || 1217 readn(fd, buf, hdr_sz) != (ssize_t)hdr_sz) 1218 return -1; 1219 1220 event = (union perf_event *)buf; 1221 1222 if (session->header.needs_swap) 1223 perf_event_header__bswap(&event->header); 1224 1225 if (event->header.size < hdr_sz || event->header.size > buf_sz) 1226 return -1; 1227 1228 rest = event->header.size - hdr_sz; 1229 1230 if (readn(fd, buf, rest) != (ssize_t)rest) 1231 return -1; 1232 1233 if (session->header.needs_swap) 1234 event_swap(event, perf_evlist__sample_id_all(session->evlist)); 1235 1236 out_parse_sample: 1237 1238 if (sample && event->header.type < PERF_RECORD_USER_TYPE_START && 1239 perf_evlist__parse_sample(session->evlist, event, sample)) 1240 return -1; 1241 1242 *event_ptr = event; 1243 1244 return 0; 1245 } 1246 1247 static s64 perf_session__process_event(struct perf_session *session, 1248 union perf_event *event, u64 file_offset) 1249 { 1250 struct perf_evlist *evlist = session->evlist; 1251 struct perf_tool *tool = session->tool; 1252 struct perf_sample sample; 1253 int ret; 1254 1255 if (session->header.needs_swap) 1256 event_swap(event, perf_evlist__sample_id_all(evlist)); 1257 1258 if (event->header.type >= PERF_RECORD_HEADER_MAX) 1259 return -EINVAL; 1260 1261 events_stats__inc(&evlist->stats, event->header.type); 1262 1263 if (event->header.type >= PERF_RECORD_USER_TYPE_START) 1264 return perf_session__process_user_event(session, event, file_offset); 1265 1266 /* 1267 * For all kernel events we get the sample data 1268 */ 1269 ret = perf_evlist__parse_sample(evlist, event, &sample); 1270 if (ret) 1271 return ret; 1272 1273 if (tool->ordered_events) { 1274 ret = perf_session__queue_event(session, event, &sample, file_offset); 1275 if (ret != -ETIME) 1276 return ret; 1277 } 1278 1279 return perf_session__deliver_event(session, event, &sample, tool, 1280 file_offset); 1281 } 1282 1283 void perf_event_header__bswap(struct perf_event_header *hdr) 1284 { 1285 hdr->type = bswap_32(hdr->type); 1286 hdr->misc = bswap_16(hdr->misc); 1287 hdr->size = bswap_16(hdr->size); 1288 } 1289 1290 struct thread *perf_session__findnew(struct perf_session *session, pid_t pid) 1291 { 1292 return machine__findnew_thread(&session->machines.host, -1, pid); 1293 } 1294 1295 static struct thread *perf_session__register_idle_thread(struct perf_session *session) 1296 { 1297 struct thread *thread; 1298 1299 thread = machine__findnew_thread(&session->machines.host, 0, 0); 1300 if (thread == NULL || thread__set_comm(thread, "swapper", 0)) { 1301 pr_err("problem inserting idle task.\n"); 1302 thread = NULL; 1303 } 1304 1305 return thread; 1306 } 1307 1308 static void perf_session__warn_about_errors(const struct perf_session *session) 1309 { 1310 const struct events_stats *stats = &session->evlist->stats; 1311 const struct ordered_events *oe = &session->ordered_events; 1312 1313 if (session->tool->lost == perf_event__process_lost && 1314 stats->nr_events[PERF_RECORD_LOST] != 0) { 1315 ui__warning("Processed %d events and lost %d chunks!\n\n" 1316 "Check IO/CPU overload!\n\n", 1317 stats->nr_events[0], 1318 stats->nr_events[PERF_RECORD_LOST]); 1319 } 1320 1321 if (session->tool->lost_samples == perf_event__process_lost_samples) { 1322 double drop_rate; 1323 1324 drop_rate = (double)stats->total_lost_samples / 1325 (double) (stats->nr_events[PERF_RECORD_SAMPLE] + stats->total_lost_samples); 1326 if (drop_rate > 0.05) { 1327 ui__warning("Processed %" PRIu64 " samples and lost %3.2f%% samples!\n\n", 1328 stats->nr_events[PERF_RECORD_SAMPLE] + stats->total_lost_samples, 1329 drop_rate * 100.0); 1330 } 1331 } 1332 1333 if (stats->nr_unknown_events != 0) { 1334 ui__warning("Found %u unknown events!\n\n" 1335 "Is this an older tool processing a perf.data " 1336 "file generated by a more recent tool?\n\n" 1337 "If that is not the case, consider " 1338 "reporting to linux-kernel@vger.kernel.org.\n\n", 1339 stats->nr_unknown_events); 1340 } 1341 1342 if (stats->nr_unknown_id != 0) { 1343 ui__warning("%u samples with id not present in the header\n", 1344 stats->nr_unknown_id); 1345 } 1346 1347 if (stats->nr_invalid_chains != 0) { 1348 ui__warning("Found invalid callchains!\n\n" 1349 "%u out of %u events were discarded for this reason.\n\n" 1350 "Consider reporting to linux-kernel@vger.kernel.org.\n\n", 1351 stats->nr_invalid_chains, 1352 stats->nr_events[PERF_RECORD_SAMPLE]); 1353 } 1354 1355 if (stats->nr_unprocessable_samples != 0) { 1356 ui__warning("%u unprocessable samples recorded.\n" 1357 "Do you have a KVM guest running and not using 'perf kvm'?\n", 1358 stats->nr_unprocessable_samples); 1359 } 1360 1361 if (oe->nr_unordered_events != 0) 1362 ui__warning("%u out of order events recorded.\n", oe->nr_unordered_events); 1363 1364 events_stats__auxtrace_error_warn(stats); 1365 1366 if (stats->nr_proc_map_timeout != 0) { 1367 ui__warning("%d map information files for pre-existing threads were\n" 1368 "not processed, if there are samples for addresses they\n" 1369 "will not be resolved, you may find out which are these\n" 1370 "threads by running with -v and redirecting the output\n" 1371 "to a file.\n", 1372 stats->nr_proc_map_timeout); 1373 } 1374 } 1375 1376 static int perf_session__flush_thread_stack(struct thread *thread, 1377 void *p __maybe_unused) 1378 { 1379 return thread_stack__flush(thread); 1380 } 1381 1382 static int perf_session__flush_thread_stacks(struct perf_session *session) 1383 { 1384 return machines__for_each_thread(&session->machines, 1385 perf_session__flush_thread_stack, 1386 NULL); 1387 } 1388 1389 volatile int session_done; 1390 1391 static int __perf_session__process_pipe_events(struct perf_session *session) 1392 { 1393 struct ordered_events *oe = &session->ordered_events; 1394 struct perf_tool *tool = session->tool; 1395 int fd = perf_data_file__fd(session->file); 1396 union perf_event *event; 1397 uint32_t size, cur_size = 0; 1398 void *buf = NULL; 1399 s64 skip = 0; 1400 u64 head; 1401 ssize_t err; 1402 void *p; 1403 1404 perf_tool__fill_defaults(tool); 1405 1406 head = 0; 1407 cur_size = sizeof(union perf_event); 1408 1409 buf = malloc(cur_size); 1410 if (!buf) 1411 return -errno; 1412 more: 1413 event = buf; 1414 err = readn(fd, event, sizeof(struct perf_event_header)); 1415 if (err <= 0) { 1416 if (err == 0) 1417 goto done; 1418 1419 pr_err("failed to read event header\n"); 1420 goto out_err; 1421 } 1422 1423 if (session->header.needs_swap) 1424 perf_event_header__bswap(&event->header); 1425 1426 size = event->header.size; 1427 if (size < sizeof(struct perf_event_header)) { 1428 pr_err("bad event header size\n"); 1429 goto out_err; 1430 } 1431 1432 if (size > cur_size) { 1433 void *new = realloc(buf, size); 1434 if (!new) { 1435 pr_err("failed to allocate memory to read event\n"); 1436 goto out_err; 1437 } 1438 buf = new; 1439 cur_size = size; 1440 event = buf; 1441 } 1442 p = event; 1443 p += sizeof(struct perf_event_header); 1444 1445 if (size - sizeof(struct perf_event_header)) { 1446 err = readn(fd, p, size - sizeof(struct perf_event_header)); 1447 if (err <= 0) { 1448 if (err == 0) { 1449 pr_err("unexpected end of event stream\n"); 1450 goto done; 1451 } 1452 1453 pr_err("failed to read event data\n"); 1454 goto out_err; 1455 } 1456 } 1457 1458 if ((skip = perf_session__process_event(session, event, head)) < 0) { 1459 pr_err("%#" PRIx64 " [%#x]: failed to process type: %d\n", 1460 head, event->header.size, event->header.type); 1461 err = -EINVAL; 1462 goto out_err; 1463 } 1464 1465 head += size; 1466 1467 if (skip > 0) 1468 head += skip; 1469 1470 if (!session_done()) 1471 goto more; 1472 done: 1473 /* do the final flush for ordered samples */ 1474 err = ordered_events__flush(oe, OE_FLUSH__FINAL); 1475 if (err) 1476 goto out_err; 1477 err = auxtrace__flush_events(session, tool); 1478 if (err) 1479 goto out_err; 1480 err = perf_session__flush_thread_stacks(session); 1481 out_err: 1482 free(buf); 1483 perf_session__warn_about_errors(session); 1484 ordered_events__free(&session->ordered_events); 1485 auxtrace__free_events(session); 1486 return err; 1487 } 1488 1489 static union perf_event * 1490 fetch_mmaped_event(struct perf_session *session, 1491 u64 head, size_t mmap_size, char *buf) 1492 { 1493 union perf_event *event; 1494 1495 /* 1496 * Ensure we have enough space remaining to read 1497 * the size of the event in the headers. 1498 */ 1499 if (head + sizeof(event->header) > mmap_size) 1500 return NULL; 1501 1502 event = (union perf_event *)(buf + head); 1503 1504 if (session->header.needs_swap) 1505 perf_event_header__bswap(&event->header); 1506 1507 if (head + event->header.size > mmap_size) { 1508 /* We're not fetching the event so swap back again */ 1509 if (session->header.needs_swap) 1510 perf_event_header__bswap(&event->header); 1511 return NULL; 1512 } 1513 1514 return event; 1515 } 1516 1517 /* 1518 * On 64bit we can mmap the data file in one go. No need for tiny mmap 1519 * slices. On 32bit we use 32MB. 1520 */ 1521 #if BITS_PER_LONG == 64 1522 #define MMAP_SIZE ULLONG_MAX 1523 #define NUM_MMAPS 1 1524 #else 1525 #define MMAP_SIZE (32 * 1024 * 1024ULL) 1526 #define NUM_MMAPS 128 1527 #endif 1528 1529 static int __perf_session__process_events(struct perf_session *session, 1530 u64 data_offset, u64 data_size, 1531 u64 file_size) 1532 { 1533 struct ordered_events *oe = &session->ordered_events; 1534 struct perf_tool *tool = session->tool; 1535 int fd = perf_data_file__fd(session->file); 1536 u64 head, page_offset, file_offset, file_pos, size; 1537 int err, mmap_prot, mmap_flags, map_idx = 0; 1538 size_t mmap_size; 1539 char *buf, *mmaps[NUM_MMAPS]; 1540 union perf_event *event; 1541 struct ui_progress prog; 1542 s64 skip; 1543 1544 perf_tool__fill_defaults(tool); 1545 1546 page_offset = page_size * (data_offset / page_size); 1547 file_offset = page_offset; 1548 head = data_offset - page_offset; 1549 1550 if (data_size && (data_offset + data_size < file_size)) 1551 file_size = data_offset + data_size; 1552 1553 ui_progress__init(&prog, file_size, "Processing events..."); 1554 1555 mmap_size = MMAP_SIZE; 1556 if (mmap_size > file_size) { 1557 mmap_size = file_size; 1558 session->one_mmap = true; 1559 } 1560 1561 memset(mmaps, 0, sizeof(mmaps)); 1562 1563 mmap_prot = PROT_READ; 1564 mmap_flags = MAP_SHARED; 1565 1566 if (session->header.needs_swap) { 1567 mmap_prot |= PROT_WRITE; 1568 mmap_flags = MAP_PRIVATE; 1569 } 1570 remap: 1571 buf = mmap(NULL, mmap_size, mmap_prot, mmap_flags, fd, 1572 file_offset); 1573 if (buf == MAP_FAILED) { 1574 pr_err("failed to mmap file\n"); 1575 err = -errno; 1576 goto out_err; 1577 } 1578 mmaps[map_idx] = buf; 1579 map_idx = (map_idx + 1) & (ARRAY_SIZE(mmaps) - 1); 1580 file_pos = file_offset + head; 1581 if (session->one_mmap) { 1582 session->one_mmap_addr = buf; 1583 session->one_mmap_offset = file_offset; 1584 } 1585 1586 more: 1587 event = fetch_mmaped_event(session, head, mmap_size, buf); 1588 if (!event) { 1589 if (mmaps[map_idx]) { 1590 munmap(mmaps[map_idx], mmap_size); 1591 mmaps[map_idx] = NULL; 1592 } 1593 1594 page_offset = page_size * (head / page_size); 1595 file_offset += page_offset; 1596 head -= page_offset; 1597 goto remap; 1598 } 1599 1600 size = event->header.size; 1601 1602 if (size < sizeof(struct perf_event_header) || 1603 (skip = perf_session__process_event(session, event, file_pos)) < 0) { 1604 pr_err("%#" PRIx64 " [%#x]: failed to process type: %d\n", 1605 file_offset + head, event->header.size, 1606 event->header.type); 1607 err = -EINVAL; 1608 goto out_err; 1609 } 1610 1611 if (skip) 1612 size += skip; 1613 1614 head += size; 1615 file_pos += size; 1616 1617 ui_progress__update(&prog, size); 1618 1619 if (session_done()) 1620 goto out; 1621 1622 if (file_pos < file_size) 1623 goto more; 1624 1625 out: 1626 /* do the final flush for ordered samples */ 1627 err = ordered_events__flush(oe, OE_FLUSH__FINAL); 1628 if (err) 1629 goto out_err; 1630 err = auxtrace__flush_events(session, tool); 1631 if (err) 1632 goto out_err; 1633 err = perf_session__flush_thread_stacks(session); 1634 out_err: 1635 ui_progress__finish(); 1636 perf_session__warn_about_errors(session); 1637 ordered_events__free(&session->ordered_events); 1638 auxtrace__free_events(session); 1639 session->one_mmap = false; 1640 return err; 1641 } 1642 1643 int perf_session__process_events(struct perf_session *session) 1644 { 1645 u64 size = perf_data_file__size(session->file); 1646 int err; 1647 1648 if (perf_session__register_idle_thread(session) == NULL) 1649 return -ENOMEM; 1650 1651 if (!perf_data_file__is_pipe(session->file)) 1652 err = __perf_session__process_events(session, 1653 session->header.data_offset, 1654 session->header.data_size, size); 1655 else 1656 err = __perf_session__process_pipe_events(session); 1657 1658 return err; 1659 } 1660 1661 bool perf_session__has_traces(struct perf_session *session, const char *msg) 1662 { 1663 struct perf_evsel *evsel; 1664 1665 evlist__for_each(session->evlist, evsel) { 1666 if (evsel->attr.type == PERF_TYPE_TRACEPOINT) 1667 return true; 1668 } 1669 1670 pr_err("No trace sample to read. Did you call 'perf %s'?\n", msg); 1671 return false; 1672 } 1673 1674 int maps__set_kallsyms_ref_reloc_sym(struct map **maps, 1675 const char *symbol_name, u64 addr) 1676 { 1677 char *bracket; 1678 enum map_type i; 1679 struct ref_reloc_sym *ref; 1680 1681 ref = zalloc(sizeof(struct ref_reloc_sym)); 1682 if (ref == NULL) 1683 return -ENOMEM; 1684 1685 ref->name = strdup(symbol_name); 1686 if (ref->name == NULL) { 1687 free(ref); 1688 return -ENOMEM; 1689 } 1690 1691 bracket = strchr(ref->name, ']'); 1692 if (bracket) 1693 *bracket = '\0'; 1694 1695 ref->addr = addr; 1696 1697 for (i = 0; i < MAP__NR_TYPES; ++i) { 1698 struct kmap *kmap = map__kmap(maps[i]); 1699 1700 if (!kmap) 1701 continue; 1702 kmap->ref_reloc_sym = ref; 1703 } 1704 1705 return 0; 1706 } 1707 1708 size_t perf_session__fprintf_dsos(struct perf_session *session, FILE *fp) 1709 { 1710 return machines__fprintf_dsos(&session->machines, fp); 1711 } 1712 1713 size_t perf_session__fprintf_dsos_buildid(struct perf_session *session, FILE *fp, 1714 bool (skip)(struct dso *dso, int parm), int parm) 1715 { 1716 return machines__fprintf_dsos_buildid(&session->machines, fp, skip, parm); 1717 } 1718 1719 size_t perf_session__fprintf_nr_events(struct perf_session *session, FILE *fp) 1720 { 1721 size_t ret; 1722 const char *msg = ""; 1723 1724 if (perf_header__has_feat(&session->header, HEADER_AUXTRACE)) 1725 msg = " (excludes AUX area (e.g. instruction trace) decoded / synthesized events)"; 1726 1727 ret = fprintf(fp, "Aggregated stats:%s\n", msg); 1728 1729 ret += events_stats__fprintf(&session->evlist->stats, fp); 1730 return ret; 1731 } 1732 1733 size_t perf_session__fprintf(struct perf_session *session, FILE *fp) 1734 { 1735 /* 1736 * FIXME: Here we have to actually print all the machines in this 1737 * session, not just the host... 1738 */ 1739 return machine__fprintf(&session->machines.host, fp); 1740 } 1741 1742 struct perf_evsel *perf_session__find_first_evtype(struct perf_session *session, 1743 unsigned int type) 1744 { 1745 struct perf_evsel *pos; 1746 1747 evlist__for_each(session->evlist, pos) { 1748 if (pos->attr.type == type) 1749 return pos; 1750 } 1751 return NULL; 1752 } 1753 1754 void perf_evsel__print_ip(struct perf_evsel *evsel, struct perf_sample *sample, 1755 struct addr_location *al, 1756 unsigned int print_opts, unsigned int stack_depth) 1757 { 1758 struct callchain_cursor_node *node; 1759 int print_ip = print_opts & PRINT_IP_OPT_IP; 1760 int print_sym = print_opts & PRINT_IP_OPT_SYM; 1761 int print_dso = print_opts & PRINT_IP_OPT_DSO; 1762 int print_symoffset = print_opts & PRINT_IP_OPT_SYMOFFSET; 1763 int print_oneline = print_opts & PRINT_IP_OPT_ONELINE; 1764 int print_srcline = print_opts & PRINT_IP_OPT_SRCLINE; 1765 char s = print_oneline ? ' ' : '\t'; 1766 1767 if (symbol_conf.use_callchain && sample->callchain) { 1768 struct addr_location node_al; 1769 1770 if (thread__resolve_callchain(al->thread, evsel, 1771 sample, NULL, NULL, 1772 PERF_MAX_STACK_DEPTH) != 0) { 1773 if (verbose) 1774 error("Failed to resolve callchain. Skipping\n"); 1775 return; 1776 } 1777 callchain_cursor_commit(&callchain_cursor); 1778 1779 if (print_symoffset) 1780 node_al = *al; 1781 1782 while (stack_depth) { 1783 u64 addr = 0; 1784 1785 node = callchain_cursor_current(&callchain_cursor); 1786 if (!node) 1787 break; 1788 1789 if (node->sym && node->sym->ignore) 1790 goto next; 1791 1792 if (print_ip) 1793 printf("%c%16" PRIx64, s, node->ip); 1794 1795 if (node->map) 1796 addr = node->map->map_ip(node->map, node->ip); 1797 1798 if (print_sym) { 1799 printf(" "); 1800 if (print_symoffset) { 1801 node_al.addr = addr; 1802 node_al.map = node->map; 1803 symbol__fprintf_symname_offs(node->sym, &node_al, stdout); 1804 } else 1805 symbol__fprintf_symname(node->sym, stdout); 1806 } 1807 1808 if (print_dso) { 1809 printf(" ("); 1810 map__fprintf_dsoname(node->map, stdout); 1811 printf(")"); 1812 } 1813 1814 if (print_srcline) 1815 map__fprintf_srcline(node->map, addr, "\n ", 1816 stdout); 1817 1818 if (!print_oneline) 1819 printf("\n"); 1820 1821 stack_depth--; 1822 next: 1823 callchain_cursor_advance(&callchain_cursor); 1824 } 1825 1826 } else { 1827 if (al->sym && al->sym->ignore) 1828 return; 1829 1830 if (print_ip) 1831 printf("%16" PRIx64, sample->ip); 1832 1833 if (print_sym) { 1834 printf(" "); 1835 if (print_symoffset) 1836 symbol__fprintf_symname_offs(al->sym, al, 1837 stdout); 1838 else 1839 symbol__fprintf_symname(al->sym, stdout); 1840 } 1841 1842 if (print_dso) { 1843 printf(" ("); 1844 map__fprintf_dsoname(al->map, stdout); 1845 printf(")"); 1846 } 1847 1848 if (print_srcline) 1849 map__fprintf_srcline(al->map, al->addr, "\n ", stdout); 1850 } 1851 } 1852 1853 int perf_session__cpu_bitmap(struct perf_session *session, 1854 const char *cpu_list, unsigned long *cpu_bitmap) 1855 { 1856 int i, err = -1; 1857 struct cpu_map *map; 1858 1859 for (i = 0; i < PERF_TYPE_MAX; ++i) { 1860 struct perf_evsel *evsel; 1861 1862 evsel = perf_session__find_first_evtype(session, i); 1863 if (!evsel) 1864 continue; 1865 1866 if (!(evsel->attr.sample_type & PERF_SAMPLE_CPU)) { 1867 pr_err("File does not contain CPU events. " 1868 "Remove -c option to proceed.\n"); 1869 return -1; 1870 } 1871 } 1872 1873 map = cpu_map__new(cpu_list); 1874 if (map == NULL) { 1875 pr_err("Invalid cpu_list\n"); 1876 return -1; 1877 } 1878 1879 for (i = 0; i < map->nr; i++) { 1880 int cpu = map->map[i]; 1881 1882 if (cpu >= MAX_NR_CPUS) { 1883 pr_err("Requested CPU %d too large. " 1884 "Consider raising MAX_NR_CPUS\n", cpu); 1885 goto out_delete_map; 1886 } 1887 1888 set_bit(cpu, cpu_bitmap); 1889 } 1890 1891 err = 0; 1892 1893 out_delete_map: 1894 cpu_map__delete(map); 1895 return err; 1896 } 1897 1898 void perf_session__fprintf_info(struct perf_session *session, FILE *fp, 1899 bool full) 1900 { 1901 struct stat st; 1902 int fd, ret; 1903 1904 if (session == NULL || fp == NULL) 1905 return; 1906 1907 fd = perf_data_file__fd(session->file); 1908 1909 ret = fstat(fd, &st); 1910 if (ret == -1) 1911 return; 1912 1913 fprintf(fp, "# ========\n"); 1914 fprintf(fp, "# captured on: %s", ctime(&st.st_ctime)); 1915 perf_header__fprintf_info(session, fp, full); 1916 fprintf(fp, "# ========\n#\n"); 1917 } 1918 1919 1920 int __perf_session__set_tracepoints_handlers(struct perf_session *session, 1921 const struct perf_evsel_str_handler *assocs, 1922 size_t nr_assocs) 1923 { 1924 struct perf_evsel *evsel; 1925 size_t i; 1926 int err; 1927 1928 for (i = 0; i < nr_assocs; i++) { 1929 /* 1930 * Adding a handler for an event not in the session, 1931 * just ignore it. 1932 */ 1933 evsel = perf_evlist__find_tracepoint_by_name(session->evlist, assocs[i].name); 1934 if (evsel == NULL) 1935 continue; 1936 1937 err = -EEXIST; 1938 if (evsel->handler != NULL) 1939 goto out; 1940 evsel->handler = assocs[i].handler; 1941 } 1942 1943 err = 0; 1944 out: 1945 return err; 1946 } 1947 1948 int perf_event__process_id_index(struct perf_tool *tool __maybe_unused, 1949 union perf_event *event, 1950 struct perf_session *session) 1951 { 1952 struct perf_evlist *evlist = session->evlist; 1953 struct id_index_event *ie = &event->id_index; 1954 size_t i, nr, max_nr; 1955 1956 max_nr = (ie->header.size - sizeof(struct id_index_event)) / 1957 sizeof(struct id_index_entry); 1958 nr = ie->nr; 1959 if (nr > max_nr) 1960 return -EINVAL; 1961 1962 if (dump_trace) 1963 fprintf(stdout, " nr: %zu\n", nr); 1964 1965 for (i = 0; i < nr; i++) { 1966 struct id_index_entry *e = &ie->entries[i]; 1967 struct perf_sample_id *sid; 1968 1969 if (dump_trace) { 1970 fprintf(stdout, " ... id: %"PRIu64, e->id); 1971 fprintf(stdout, " idx: %"PRIu64, e->idx); 1972 fprintf(stdout, " cpu: %"PRId64, e->cpu); 1973 fprintf(stdout, " tid: %"PRId64"\n", e->tid); 1974 } 1975 1976 sid = perf_evlist__id2sid(evlist, e->id); 1977 if (!sid) 1978 return -ENOENT; 1979 sid->idx = e->idx; 1980 sid->cpu = e->cpu; 1981 sid->tid = e->tid; 1982 } 1983 return 0; 1984 } 1985 1986 int perf_event__synthesize_id_index(struct perf_tool *tool, 1987 perf_event__handler_t process, 1988 struct perf_evlist *evlist, 1989 struct machine *machine) 1990 { 1991 union perf_event *ev; 1992 struct perf_evsel *evsel; 1993 size_t nr = 0, i = 0, sz, max_nr, n; 1994 int err; 1995 1996 pr_debug2("Synthesizing id index\n"); 1997 1998 max_nr = (UINT16_MAX - sizeof(struct id_index_event)) / 1999 sizeof(struct id_index_entry); 2000 2001 evlist__for_each(evlist, evsel) 2002 nr += evsel->ids; 2003 2004 n = nr > max_nr ? max_nr : nr; 2005 sz = sizeof(struct id_index_event) + n * sizeof(struct id_index_entry); 2006 ev = zalloc(sz); 2007 if (!ev) 2008 return -ENOMEM; 2009 2010 ev->id_index.header.type = PERF_RECORD_ID_INDEX; 2011 ev->id_index.header.size = sz; 2012 ev->id_index.nr = n; 2013 2014 evlist__for_each(evlist, evsel) { 2015 u32 j; 2016 2017 for (j = 0; j < evsel->ids; j++) { 2018 struct id_index_entry *e; 2019 struct perf_sample_id *sid; 2020 2021 if (i >= n) { 2022 err = process(tool, ev, NULL, machine); 2023 if (err) 2024 goto out_err; 2025 nr -= n; 2026 i = 0; 2027 } 2028 2029 e = &ev->id_index.entries[i++]; 2030 2031 e->id = evsel->id[j]; 2032 2033 sid = perf_evlist__id2sid(evlist, e->id); 2034 if (!sid) { 2035 free(ev); 2036 return -ENOENT; 2037 } 2038 2039 e->idx = sid->idx; 2040 e->cpu = sid->cpu; 2041 e->tid = sid->tid; 2042 } 2043 } 2044 2045 sz = sizeof(struct id_index_event) + nr * sizeof(struct id_index_entry); 2046 ev->id_index.header.size = sz; 2047 ev->id_index.nr = nr; 2048 2049 err = process(tool, ev, NULL, machine); 2050 out_err: 2051 free(ev); 2052 2053 return err; 2054 } 2055