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