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 perf_session__open(struct perf_session *session) 20 { 21 struct perf_data_file *file = session->file; 22 23 if (perf_session__read_header(session) < 0) { 24 pr_err("incompatible file format (rerun with -v to learn more)"); 25 return -1; 26 } 27 28 if (perf_data_file__is_pipe(file)) 29 return 0; 30 31 if (!perf_evlist__valid_sample_type(session->evlist)) { 32 pr_err("non matching sample_type"); 33 return -1; 34 } 35 36 if (!perf_evlist__valid_sample_id_all(session->evlist)) { 37 pr_err("non matching sample_id_all"); 38 return -1; 39 } 40 41 if (!perf_evlist__valid_read_format(session->evlist)) { 42 pr_err("non matching read_format"); 43 return -1; 44 } 45 46 return 0; 47 } 48 49 void perf_session__set_id_hdr_size(struct perf_session *session) 50 { 51 u16 id_hdr_size = perf_evlist__id_hdr_size(session->evlist); 52 53 machines__set_id_hdr_size(&session->machines, id_hdr_size); 54 } 55 56 int perf_session__create_kernel_maps(struct perf_session *session) 57 { 58 int ret = machine__create_kernel_maps(&session->machines.host); 59 60 if (ret >= 0) 61 ret = machines__create_guest_kernel_maps(&session->machines); 62 return ret; 63 } 64 65 static void perf_session__destroy_kernel_maps(struct perf_session *session) 66 { 67 machines__destroy_kernel_maps(&session->machines); 68 } 69 70 static bool perf_session__has_comm_exec(struct perf_session *session) 71 { 72 struct perf_evsel *evsel; 73 74 evlist__for_each(session->evlist, evsel) { 75 if (evsel->attr.comm_exec) 76 return true; 77 } 78 79 return false; 80 } 81 82 static void perf_session__set_comm_exec(struct perf_session *session) 83 { 84 bool comm_exec = perf_session__has_comm_exec(session); 85 86 machines__set_comm_exec(&session->machines, comm_exec); 87 } 88 89 struct perf_session *perf_session__new(struct perf_data_file *file, 90 bool repipe, struct perf_tool *tool) 91 { 92 struct perf_session *session = zalloc(sizeof(*session)); 93 94 if (!session) 95 goto out; 96 97 session->repipe = repipe; 98 ordered_events__init(&session->ordered_events); 99 machines__init(&session->machines); 100 101 if (file) { 102 if (perf_data_file__open(file)) 103 goto out_delete; 104 105 session->file = file; 106 107 if (perf_data_file__is_read(file)) { 108 if (perf_session__open(session) < 0) 109 goto out_close; 110 111 perf_session__set_id_hdr_size(session); 112 perf_session__set_comm_exec(session); 113 } 114 } 115 116 if (!file || perf_data_file__is_write(file)) { 117 /* 118 * In O_RDONLY mode this will be performed when reading the 119 * kernel MMAP event, in perf_event__process_mmap(). 120 */ 121 if (perf_session__create_kernel_maps(session) < 0) 122 pr_warning("Cannot read kernel map\n"); 123 } 124 125 if (tool && tool->ordering_requires_timestamps && 126 tool->ordered_events && !perf_evlist__sample_id_all(session->evlist)) { 127 dump_printf("WARNING: No sample_id_all support, falling back to unordered processing\n"); 128 tool->ordered_events = false; 129 } 130 131 return session; 132 133 out_close: 134 perf_data_file__close(file); 135 out_delete: 136 perf_session__delete(session); 137 out: 138 return NULL; 139 } 140 141 static void perf_session__delete_dead_threads(struct perf_session *session) 142 { 143 machine__delete_dead_threads(&session->machines.host); 144 } 145 146 static void perf_session__delete_threads(struct perf_session *session) 147 { 148 machine__delete_threads(&session->machines.host); 149 } 150 151 static void perf_session_env__delete(struct perf_session_env *env) 152 { 153 zfree(&env->hostname); 154 zfree(&env->os_release); 155 zfree(&env->version); 156 zfree(&env->arch); 157 zfree(&env->cpu_desc); 158 zfree(&env->cpuid); 159 160 zfree(&env->cmdline); 161 zfree(&env->sibling_cores); 162 zfree(&env->sibling_threads); 163 zfree(&env->numa_nodes); 164 zfree(&env->pmu_mappings); 165 } 166 167 void perf_session__delete(struct perf_session *session) 168 { 169 perf_session__destroy_kernel_maps(session); 170 perf_session__delete_dead_threads(session); 171 perf_session__delete_threads(session); 172 perf_session_env__delete(&session->header.env); 173 machines__exit(&session->machines); 174 if (session->file) 175 perf_data_file__close(session->file); 176 free(session); 177 } 178 179 static int process_event_synth_tracing_data_stub(struct perf_tool *tool 180 __maybe_unused, 181 union perf_event *event 182 __maybe_unused, 183 struct perf_session *session 184 __maybe_unused) 185 { 186 dump_printf(": unhandled!\n"); 187 return 0; 188 } 189 190 static int process_event_synth_attr_stub(struct perf_tool *tool __maybe_unused, 191 union perf_event *event __maybe_unused, 192 struct perf_evlist **pevlist 193 __maybe_unused) 194 { 195 dump_printf(": unhandled!\n"); 196 return 0; 197 } 198 199 static int process_event_sample_stub(struct perf_tool *tool __maybe_unused, 200 union perf_event *event __maybe_unused, 201 struct perf_sample *sample __maybe_unused, 202 struct perf_evsel *evsel __maybe_unused, 203 struct machine *machine __maybe_unused) 204 { 205 dump_printf(": unhandled!\n"); 206 return 0; 207 } 208 209 static int process_event_stub(struct perf_tool *tool __maybe_unused, 210 union perf_event *event __maybe_unused, 211 struct perf_sample *sample __maybe_unused, 212 struct machine *machine __maybe_unused) 213 { 214 dump_printf(": unhandled!\n"); 215 return 0; 216 } 217 218 static int process_finished_round_stub(struct perf_tool *tool __maybe_unused, 219 union perf_event *event __maybe_unused, 220 struct perf_session *perf_session 221 __maybe_unused) 222 { 223 dump_printf(": unhandled!\n"); 224 return 0; 225 } 226 227 static int process_finished_round(struct perf_tool *tool, 228 union perf_event *event, 229 struct perf_session *session); 230 231 static int process_id_index_stub(struct perf_tool *tool __maybe_unused, 232 union perf_event *event __maybe_unused, 233 struct perf_session *perf_session 234 __maybe_unused) 235 { 236 dump_printf(": unhandled!\n"); 237 return 0; 238 } 239 240 void perf_tool__fill_defaults(struct perf_tool *tool) 241 { 242 if (tool->sample == NULL) 243 tool->sample = process_event_sample_stub; 244 if (tool->mmap == NULL) 245 tool->mmap = process_event_stub; 246 if (tool->mmap2 == NULL) 247 tool->mmap2 = process_event_stub; 248 if (tool->comm == NULL) 249 tool->comm = process_event_stub; 250 if (tool->fork == NULL) 251 tool->fork = process_event_stub; 252 if (tool->exit == NULL) 253 tool->exit = process_event_stub; 254 if (tool->lost == NULL) 255 tool->lost = perf_event__process_lost; 256 if (tool->read == NULL) 257 tool->read = process_event_sample_stub; 258 if (tool->throttle == NULL) 259 tool->throttle = process_event_stub; 260 if (tool->unthrottle == NULL) 261 tool->unthrottle = process_event_stub; 262 if (tool->attr == NULL) 263 tool->attr = process_event_synth_attr_stub; 264 if (tool->tracing_data == NULL) 265 tool->tracing_data = process_event_synth_tracing_data_stub; 266 if (tool->build_id == NULL) 267 tool->build_id = process_finished_round_stub; 268 if (tool->finished_round == NULL) { 269 if (tool->ordered_events) 270 tool->finished_round = process_finished_round; 271 else 272 tool->finished_round = process_finished_round_stub; 273 } 274 if (tool->id_index == NULL) 275 tool->id_index = process_id_index_stub; 276 } 277 278 static void swap_sample_id_all(union perf_event *event, void *data) 279 { 280 void *end = (void *) event + event->header.size; 281 int size = end - data; 282 283 BUG_ON(size % sizeof(u64)); 284 mem_bswap_64(data, size); 285 } 286 287 static void perf_event__all64_swap(union perf_event *event, 288 bool sample_id_all __maybe_unused) 289 { 290 struct perf_event_header *hdr = &event->header; 291 mem_bswap_64(hdr + 1, event->header.size - sizeof(*hdr)); 292 } 293 294 static void perf_event__comm_swap(union perf_event *event, bool sample_id_all) 295 { 296 event->comm.pid = bswap_32(event->comm.pid); 297 event->comm.tid = bswap_32(event->comm.tid); 298 299 if (sample_id_all) { 300 void *data = &event->comm.comm; 301 302 data += PERF_ALIGN(strlen(data) + 1, sizeof(u64)); 303 swap_sample_id_all(event, data); 304 } 305 } 306 307 static void perf_event__mmap_swap(union perf_event *event, 308 bool sample_id_all) 309 { 310 event->mmap.pid = bswap_32(event->mmap.pid); 311 event->mmap.tid = bswap_32(event->mmap.tid); 312 event->mmap.start = bswap_64(event->mmap.start); 313 event->mmap.len = bswap_64(event->mmap.len); 314 event->mmap.pgoff = bswap_64(event->mmap.pgoff); 315 316 if (sample_id_all) { 317 void *data = &event->mmap.filename; 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__mmap2_swap(union perf_event *event, 325 bool sample_id_all) 326 { 327 event->mmap2.pid = bswap_32(event->mmap2.pid); 328 event->mmap2.tid = bswap_32(event->mmap2.tid); 329 event->mmap2.start = bswap_64(event->mmap2.start); 330 event->mmap2.len = bswap_64(event->mmap2.len); 331 event->mmap2.pgoff = bswap_64(event->mmap2.pgoff); 332 event->mmap2.maj = bswap_32(event->mmap2.maj); 333 event->mmap2.min = bswap_32(event->mmap2.min); 334 event->mmap2.ino = bswap_64(event->mmap2.ino); 335 336 if (sample_id_all) { 337 void *data = &event->mmap2.filename; 338 339 data += PERF_ALIGN(strlen(data) + 1, sizeof(u64)); 340 swap_sample_id_all(event, data); 341 } 342 } 343 static void perf_event__task_swap(union perf_event *event, bool sample_id_all) 344 { 345 event->fork.pid = bswap_32(event->fork.pid); 346 event->fork.tid = bswap_32(event->fork.tid); 347 event->fork.ppid = bswap_32(event->fork.ppid); 348 event->fork.ptid = bswap_32(event->fork.ptid); 349 event->fork.time = bswap_64(event->fork.time); 350 351 if (sample_id_all) 352 swap_sample_id_all(event, &event->fork + 1); 353 } 354 355 static void perf_event__read_swap(union perf_event *event, bool sample_id_all) 356 { 357 event->read.pid = bswap_32(event->read.pid); 358 event->read.tid = bswap_32(event->read.tid); 359 event->read.value = bswap_64(event->read.value); 360 event->read.time_enabled = bswap_64(event->read.time_enabled); 361 event->read.time_running = bswap_64(event->read.time_running); 362 event->read.id = bswap_64(event->read.id); 363 364 if (sample_id_all) 365 swap_sample_id_all(event, &event->read + 1); 366 } 367 368 static void perf_event__throttle_swap(union perf_event *event, 369 bool sample_id_all) 370 { 371 event->throttle.time = bswap_64(event->throttle.time); 372 event->throttle.id = bswap_64(event->throttle.id); 373 event->throttle.stream_id = bswap_64(event->throttle.stream_id); 374 375 if (sample_id_all) 376 swap_sample_id_all(event, &event->throttle + 1); 377 } 378 379 static u8 revbyte(u8 b) 380 { 381 int rev = (b >> 4) | ((b & 0xf) << 4); 382 rev = ((rev & 0xcc) >> 2) | ((rev & 0x33) << 2); 383 rev = ((rev & 0xaa) >> 1) | ((rev & 0x55) << 1); 384 return (u8) rev; 385 } 386 387 /* 388 * XXX this is hack in attempt to carry flags bitfield 389 * throught endian village. ABI says: 390 * 391 * Bit-fields are allocated from right to left (least to most significant) 392 * on little-endian implementations and from left to right (most to least 393 * significant) on big-endian implementations. 394 * 395 * The above seems to be byte specific, so we need to reverse each 396 * byte of the bitfield. 'Internet' also says this might be implementation 397 * specific and we probably need proper fix and carry perf_event_attr 398 * bitfield flags in separate data file FEAT_ section. Thought this seems 399 * to work for now. 400 */ 401 static void swap_bitfield(u8 *p, unsigned len) 402 { 403 unsigned i; 404 405 for (i = 0; i < len; i++) { 406 *p = revbyte(*p); 407 p++; 408 } 409 } 410 411 /* exported for swapping attributes in file header */ 412 void perf_event__attr_swap(struct perf_event_attr *attr) 413 { 414 attr->type = bswap_32(attr->type); 415 attr->size = bswap_32(attr->size); 416 attr->config = bswap_64(attr->config); 417 attr->sample_period = bswap_64(attr->sample_period); 418 attr->sample_type = bswap_64(attr->sample_type); 419 attr->read_format = bswap_64(attr->read_format); 420 attr->wakeup_events = bswap_32(attr->wakeup_events); 421 attr->bp_type = bswap_32(attr->bp_type); 422 attr->bp_addr = bswap_64(attr->bp_addr); 423 attr->bp_len = bswap_64(attr->bp_len); 424 attr->branch_sample_type = bswap_64(attr->branch_sample_type); 425 attr->sample_regs_user = bswap_64(attr->sample_regs_user); 426 attr->sample_stack_user = bswap_32(attr->sample_stack_user); 427 428 swap_bitfield((u8 *) (&attr->read_format + 1), sizeof(u64)); 429 } 430 431 static void perf_event__hdr_attr_swap(union perf_event *event, 432 bool sample_id_all __maybe_unused) 433 { 434 size_t size; 435 436 perf_event__attr_swap(&event->attr.attr); 437 438 size = event->header.size; 439 size -= (void *)&event->attr.id - (void *)event; 440 mem_bswap_64(event->attr.id, size); 441 } 442 443 static void perf_event__event_type_swap(union perf_event *event, 444 bool sample_id_all __maybe_unused) 445 { 446 event->event_type.event_type.event_id = 447 bswap_64(event->event_type.event_type.event_id); 448 } 449 450 static void perf_event__tracing_data_swap(union perf_event *event, 451 bool sample_id_all __maybe_unused) 452 { 453 event->tracing_data.size = bswap_32(event->tracing_data.size); 454 } 455 456 typedef void (*perf_event__swap_op)(union perf_event *event, 457 bool sample_id_all); 458 459 static perf_event__swap_op perf_event__swap_ops[] = { 460 [PERF_RECORD_MMAP] = perf_event__mmap_swap, 461 [PERF_RECORD_MMAP2] = perf_event__mmap2_swap, 462 [PERF_RECORD_COMM] = perf_event__comm_swap, 463 [PERF_RECORD_FORK] = perf_event__task_swap, 464 [PERF_RECORD_EXIT] = perf_event__task_swap, 465 [PERF_RECORD_LOST] = perf_event__all64_swap, 466 [PERF_RECORD_READ] = perf_event__read_swap, 467 [PERF_RECORD_THROTTLE] = perf_event__throttle_swap, 468 [PERF_RECORD_UNTHROTTLE] = perf_event__throttle_swap, 469 [PERF_RECORD_SAMPLE] = perf_event__all64_swap, 470 [PERF_RECORD_HEADER_ATTR] = perf_event__hdr_attr_swap, 471 [PERF_RECORD_HEADER_EVENT_TYPE] = perf_event__event_type_swap, 472 [PERF_RECORD_HEADER_TRACING_DATA] = perf_event__tracing_data_swap, 473 [PERF_RECORD_HEADER_BUILD_ID] = NULL, 474 [PERF_RECORD_ID_INDEX] = perf_event__all64_swap, 475 [PERF_RECORD_HEADER_MAX] = NULL, 476 }; 477 478 /* 479 * When perf record finishes a pass on every buffers, it records this pseudo 480 * event. 481 * We record the max timestamp t found in the pass n. 482 * Assuming these timestamps are monotonic across cpus, we know that if 483 * a buffer still has events with timestamps below t, they will be all 484 * available and then read in the pass n + 1. 485 * Hence when we start to read the pass n + 2, we can safely flush every 486 * events with timestamps below t. 487 * 488 * ============ PASS n ================= 489 * CPU 0 | CPU 1 490 * | 491 * cnt1 timestamps | cnt2 timestamps 492 * 1 | 2 493 * 2 | 3 494 * - | 4 <--- max recorded 495 * 496 * ============ PASS n + 1 ============== 497 * CPU 0 | CPU 1 498 * | 499 * cnt1 timestamps | cnt2 timestamps 500 * 3 | 5 501 * 4 | 6 502 * 5 | 7 <---- max recorded 503 * 504 * Flush every events below timestamp 4 505 * 506 * ============ PASS n + 2 ============== 507 * CPU 0 | CPU 1 508 * | 509 * cnt1 timestamps | cnt2 timestamps 510 * 6 | 8 511 * 7 | 9 512 * - | 10 513 * 514 * Flush every events below timestamp 7 515 * etc... 516 */ 517 static int process_finished_round(struct perf_tool *tool, 518 union perf_event *event __maybe_unused, 519 struct perf_session *session) 520 { 521 return ordered_events__flush(session, tool, OE_FLUSH__ROUND); 522 } 523 524 int perf_session_queue_event(struct perf_session *s, union perf_event *event, 525 struct perf_tool *tool, struct perf_sample *sample, 526 u64 file_offset) 527 { 528 struct ordered_events *oe = &s->ordered_events; 529 u64 timestamp = sample->time; 530 struct ordered_event *new; 531 532 if (!timestamp || timestamp == ~0ULL) 533 return -ETIME; 534 535 if (timestamp < oe->last_flush) { 536 pr_oe_time(timestamp, "out of order event\n"); 537 pr_oe_time(oe->last_flush, "last flush, last_flush_type %d\n", 538 oe->last_flush_type); 539 540 s->stats.nr_unordered_events++; 541 } 542 543 new = ordered_events__new(oe, timestamp, event); 544 if (!new) { 545 ordered_events__flush(s, tool, OE_FLUSH__HALF); 546 new = ordered_events__new(oe, timestamp, event); 547 } 548 549 if (!new) 550 return -ENOMEM; 551 552 new->file_offset = file_offset; 553 return 0; 554 } 555 556 static void callchain__lbr_callstack_printf(struct perf_sample *sample) 557 { 558 struct ip_callchain *callchain = sample->callchain; 559 struct branch_stack *lbr_stack = sample->branch_stack; 560 u64 kernel_callchain_nr = callchain->nr; 561 unsigned int i; 562 563 for (i = 0; i < kernel_callchain_nr; i++) { 564 if (callchain->ips[i] == PERF_CONTEXT_USER) 565 break; 566 } 567 568 if ((i != kernel_callchain_nr) && lbr_stack->nr) { 569 u64 total_nr; 570 /* 571 * LBR callstack can only get user call chain, 572 * i is kernel call chain number, 573 * 1 is PERF_CONTEXT_USER. 574 * 575 * The user call chain is stored in LBR registers. 576 * LBR are pair registers. The caller is stored 577 * in "from" register, while the callee is stored 578 * in "to" register. 579 * For example, there is a call stack 580 * "A"->"B"->"C"->"D". 581 * The LBR registers will recorde like 582 * "C"->"D", "B"->"C", "A"->"B". 583 * So only the first "to" register and all "from" 584 * registers are needed to construct the whole stack. 585 */ 586 total_nr = i + 1 + lbr_stack->nr + 1; 587 kernel_callchain_nr = i + 1; 588 589 printf("... LBR call chain: nr:%" PRIu64 "\n", total_nr); 590 591 for (i = 0; i < kernel_callchain_nr; i++) 592 printf("..... %2d: %016" PRIx64 "\n", 593 i, callchain->ips[i]); 594 595 printf("..... %2d: %016" PRIx64 "\n", 596 (int)(kernel_callchain_nr), lbr_stack->entries[0].to); 597 for (i = 0; i < lbr_stack->nr; i++) 598 printf("..... %2d: %016" PRIx64 "\n", 599 (int)(i + kernel_callchain_nr + 1), lbr_stack->entries[i].from); 600 } 601 } 602 603 static void callchain__printf(struct perf_evsel *evsel, 604 struct perf_sample *sample) 605 { 606 unsigned int i; 607 struct ip_callchain *callchain = sample->callchain; 608 609 if (has_branch_callstack(evsel)) 610 callchain__lbr_callstack_printf(sample); 611 612 printf("... FP chain: nr:%" PRIu64 "\n", callchain->nr); 613 614 for (i = 0; i < callchain->nr; i++) 615 printf("..... %2d: %016" PRIx64 "\n", 616 i, callchain->ips[i]); 617 } 618 619 static void branch_stack__printf(struct perf_sample *sample) 620 { 621 uint64_t i; 622 623 printf("... branch stack: nr:%" PRIu64 "\n", sample->branch_stack->nr); 624 625 for (i = 0; i < sample->branch_stack->nr; i++) 626 printf("..... %2"PRIu64": %016" PRIx64 " -> %016" PRIx64 "\n", 627 i, sample->branch_stack->entries[i].from, 628 sample->branch_stack->entries[i].to); 629 } 630 631 static void regs_dump__printf(u64 mask, u64 *regs) 632 { 633 unsigned rid, i = 0; 634 635 for_each_set_bit(rid, (unsigned long *) &mask, sizeof(mask) * 8) { 636 u64 val = regs[i++]; 637 638 printf(".... %-5s 0x%" PRIx64 "\n", 639 perf_reg_name(rid), val); 640 } 641 } 642 643 static const char *regs_abi[] = { 644 [PERF_SAMPLE_REGS_ABI_NONE] = "none", 645 [PERF_SAMPLE_REGS_ABI_32] = "32-bit", 646 [PERF_SAMPLE_REGS_ABI_64] = "64-bit", 647 }; 648 649 static inline const char *regs_dump_abi(struct regs_dump *d) 650 { 651 if (d->abi > PERF_SAMPLE_REGS_ABI_64) 652 return "unknown"; 653 654 return regs_abi[d->abi]; 655 } 656 657 static void regs__printf(const char *type, struct regs_dump *regs) 658 { 659 u64 mask = regs->mask; 660 661 printf("... %s regs: mask 0x%" PRIx64 " ABI %s\n", 662 type, 663 mask, 664 regs_dump_abi(regs)); 665 666 regs_dump__printf(mask, regs->regs); 667 } 668 669 static void regs_user__printf(struct perf_sample *sample) 670 { 671 struct regs_dump *user_regs = &sample->user_regs; 672 673 if (user_regs->regs) 674 regs__printf("user", user_regs); 675 } 676 677 static void regs_intr__printf(struct perf_sample *sample) 678 { 679 struct regs_dump *intr_regs = &sample->intr_regs; 680 681 if (intr_regs->regs) 682 regs__printf("intr", intr_regs); 683 } 684 685 static void stack_user__printf(struct stack_dump *dump) 686 { 687 printf("... ustack: size %" PRIu64 ", offset 0x%x\n", 688 dump->size, dump->offset); 689 } 690 691 static void perf_session__print_tstamp(struct perf_session *session, 692 union perf_event *event, 693 struct perf_sample *sample) 694 { 695 u64 sample_type = __perf_evlist__combined_sample_type(session->evlist); 696 697 if (event->header.type != PERF_RECORD_SAMPLE && 698 !perf_evlist__sample_id_all(session->evlist)) { 699 fputs("-1 -1 ", stdout); 700 return; 701 } 702 703 if ((sample_type & PERF_SAMPLE_CPU)) 704 printf("%u ", sample->cpu); 705 706 if (sample_type & PERF_SAMPLE_TIME) 707 printf("%" PRIu64 " ", sample->time); 708 } 709 710 static void sample_read__printf(struct perf_sample *sample, u64 read_format) 711 { 712 printf("... sample_read:\n"); 713 714 if (read_format & PERF_FORMAT_TOTAL_TIME_ENABLED) 715 printf("...... time enabled %016" PRIx64 "\n", 716 sample->read.time_enabled); 717 718 if (read_format & PERF_FORMAT_TOTAL_TIME_RUNNING) 719 printf("...... time running %016" PRIx64 "\n", 720 sample->read.time_running); 721 722 if (read_format & PERF_FORMAT_GROUP) { 723 u64 i; 724 725 printf(".... group nr %" PRIu64 "\n", sample->read.group.nr); 726 727 for (i = 0; i < sample->read.group.nr; i++) { 728 struct sample_read_value *value; 729 730 value = &sample->read.group.values[i]; 731 printf("..... id %016" PRIx64 732 ", value %016" PRIx64 "\n", 733 value->id, value->value); 734 } 735 } else 736 printf("..... id %016" PRIx64 ", value %016" PRIx64 "\n", 737 sample->read.one.id, sample->read.one.value); 738 } 739 740 static void dump_event(struct perf_session *session, union perf_event *event, 741 u64 file_offset, struct perf_sample *sample) 742 { 743 if (!dump_trace) 744 return; 745 746 printf("\n%#" PRIx64 " [%#x]: event: %d\n", 747 file_offset, event->header.size, event->header.type); 748 749 trace_event(event); 750 751 if (sample) 752 perf_session__print_tstamp(session, event, sample); 753 754 printf("%#" PRIx64 " [%#x]: PERF_RECORD_%s", file_offset, 755 event->header.size, perf_event__name(event->header.type)); 756 } 757 758 static void dump_sample(struct perf_evsel *evsel, union perf_event *event, 759 struct perf_sample *sample) 760 { 761 u64 sample_type; 762 763 if (!dump_trace) 764 return; 765 766 printf("(IP, 0x%x): %d/%d: %#" PRIx64 " period: %" PRIu64 " addr: %#" PRIx64 "\n", 767 event->header.misc, sample->pid, sample->tid, sample->ip, 768 sample->period, sample->addr); 769 770 sample_type = evsel->attr.sample_type; 771 772 if (sample_type & PERF_SAMPLE_CALLCHAIN) 773 callchain__printf(evsel, sample); 774 775 if ((sample_type & PERF_SAMPLE_BRANCH_STACK) && !has_branch_callstack(evsel)) 776 branch_stack__printf(sample); 777 778 if (sample_type & PERF_SAMPLE_REGS_USER) 779 regs_user__printf(sample); 780 781 if (sample_type & PERF_SAMPLE_REGS_INTR) 782 regs_intr__printf(sample); 783 784 if (sample_type & PERF_SAMPLE_STACK_USER) 785 stack_user__printf(&sample->user_stack); 786 787 if (sample_type & PERF_SAMPLE_WEIGHT) 788 printf("... weight: %" PRIu64 "\n", sample->weight); 789 790 if (sample_type & PERF_SAMPLE_DATA_SRC) 791 printf(" . data_src: 0x%"PRIx64"\n", sample->data_src); 792 793 if (sample_type & PERF_SAMPLE_TRANSACTION) 794 printf("... transaction: %" PRIx64 "\n", sample->transaction); 795 796 if (sample_type & PERF_SAMPLE_READ) 797 sample_read__printf(sample, evsel->attr.read_format); 798 } 799 800 static struct machine * 801 perf_session__find_machine_for_cpumode(struct perf_session *session, 802 union perf_event *event, 803 struct perf_sample *sample) 804 { 805 const u8 cpumode = event->header.misc & PERF_RECORD_MISC_CPUMODE_MASK; 806 struct machine *machine; 807 808 if (perf_guest && 809 ((cpumode == PERF_RECORD_MISC_GUEST_KERNEL) || 810 (cpumode == PERF_RECORD_MISC_GUEST_USER))) { 811 u32 pid; 812 813 if (event->header.type == PERF_RECORD_MMAP 814 || event->header.type == PERF_RECORD_MMAP2) 815 pid = event->mmap.pid; 816 else 817 pid = sample->pid; 818 819 machine = perf_session__find_machine(session, pid); 820 if (!machine) 821 machine = perf_session__findnew_machine(session, 822 DEFAULT_GUEST_KERNEL_ID); 823 return machine; 824 } 825 826 return &session->machines.host; 827 } 828 829 static int deliver_sample_value(struct perf_session *session, 830 struct perf_tool *tool, 831 union perf_event *event, 832 struct perf_sample *sample, 833 struct sample_read_value *v, 834 struct machine *machine) 835 { 836 struct perf_sample_id *sid; 837 838 sid = perf_evlist__id2sid(session->evlist, v->id); 839 if (sid) { 840 sample->id = v->id; 841 sample->period = v->value - sid->period; 842 sid->period = v->value; 843 } 844 845 if (!sid || sid->evsel == NULL) { 846 ++session->stats.nr_unknown_id; 847 return 0; 848 } 849 850 return tool->sample(tool, event, sample, sid->evsel, machine); 851 } 852 853 static int deliver_sample_group(struct perf_session *session, 854 struct perf_tool *tool, 855 union perf_event *event, 856 struct perf_sample *sample, 857 struct machine *machine) 858 { 859 int ret = -EINVAL; 860 u64 i; 861 862 for (i = 0; i < sample->read.group.nr; i++) { 863 ret = deliver_sample_value(session, tool, event, sample, 864 &sample->read.group.values[i], 865 machine); 866 if (ret) 867 break; 868 } 869 870 return ret; 871 } 872 873 static int 874 perf_session__deliver_sample(struct perf_session *session, 875 struct perf_tool *tool, 876 union perf_event *event, 877 struct perf_sample *sample, 878 struct perf_evsel *evsel, 879 struct machine *machine) 880 { 881 /* We know evsel != NULL. */ 882 u64 sample_type = evsel->attr.sample_type; 883 u64 read_format = evsel->attr.read_format; 884 885 /* Standard sample delievery. */ 886 if (!(sample_type & PERF_SAMPLE_READ)) 887 return tool->sample(tool, event, sample, evsel, machine); 888 889 /* For PERF_SAMPLE_READ we have either single or group mode. */ 890 if (read_format & PERF_FORMAT_GROUP) 891 return deliver_sample_group(session, tool, event, sample, 892 machine); 893 else 894 return deliver_sample_value(session, tool, event, sample, 895 &sample->read.one, machine); 896 } 897 898 int perf_session__deliver_event(struct perf_session *session, 899 union perf_event *event, 900 struct perf_sample *sample, 901 struct perf_tool *tool, u64 file_offset) 902 { 903 struct perf_evsel *evsel; 904 struct machine *machine; 905 906 dump_event(session, event, file_offset, sample); 907 908 evsel = perf_evlist__id2evsel(session->evlist, sample->id); 909 910 machine = perf_session__find_machine_for_cpumode(session, event, 911 sample); 912 913 switch (event->header.type) { 914 case PERF_RECORD_SAMPLE: 915 dump_sample(evsel, event, sample); 916 if (evsel == NULL) { 917 ++session->stats.nr_unknown_id; 918 return 0; 919 } 920 if (machine == NULL) { 921 ++session->stats.nr_unprocessable_samples; 922 return 0; 923 } 924 return perf_session__deliver_sample(session, tool, event, 925 sample, evsel, machine); 926 case PERF_RECORD_MMAP: 927 return tool->mmap(tool, event, sample, machine); 928 case PERF_RECORD_MMAP2: 929 return tool->mmap2(tool, event, sample, machine); 930 case PERF_RECORD_COMM: 931 return tool->comm(tool, event, sample, machine); 932 case PERF_RECORD_FORK: 933 return tool->fork(tool, event, sample, machine); 934 case PERF_RECORD_EXIT: 935 return tool->exit(tool, event, sample, machine); 936 case PERF_RECORD_LOST: 937 if (tool->lost == perf_event__process_lost) 938 session->stats.total_lost += event->lost.lost; 939 return tool->lost(tool, event, sample, machine); 940 case PERF_RECORD_READ: 941 return tool->read(tool, event, sample, evsel, machine); 942 case PERF_RECORD_THROTTLE: 943 return tool->throttle(tool, event, sample, machine); 944 case PERF_RECORD_UNTHROTTLE: 945 return tool->unthrottle(tool, event, sample, machine); 946 default: 947 ++session->stats.nr_unknown_events; 948 return -1; 949 } 950 } 951 952 static s64 perf_session__process_user_event(struct perf_session *session, 953 union perf_event *event, 954 struct perf_tool *tool, 955 u64 file_offset) 956 { 957 int fd = perf_data_file__fd(session->file); 958 int err; 959 960 dump_event(session, event, file_offset, NULL); 961 962 /* These events are processed right away */ 963 switch (event->header.type) { 964 case PERF_RECORD_HEADER_ATTR: 965 err = tool->attr(tool, event, &session->evlist); 966 if (err == 0) { 967 perf_session__set_id_hdr_size(session); 968 perf_session__set_comm_exec(session); 969 } 970 return err; 971 case PERF_RECORD_HEADER_EVENT_TYPE: 972 /* 973 * Depreceated, but we need to handle it for sake 974 * of old data files create in pipe mode. 975 */ 976 return 0; 977 case PERF_RECORD_HEADER_TRACING_DATA: 978 /* setup for reading amidst mmap */ 979 lseek(fd, file_offset, SEEK_SET); 980 return tool->tracing_data(tool, event, session); 981 case PERF_RECORD_HEADER_BUILD_ID: 982 return tool->build_id(tool, event, session); 983 case PERF_RECORD_FINISHED_ROUND: 984 return tool->finished_round(tool, event, session); 985 case PERF_RECORD_ID_INDEX: 986 return tool->id_index(tool, event, session); 987 default: 988 return -EINVAL; 989 } 990 } 991 992 int perf_session__deliver_synth_event(struct perf_session *session, 993 union perf_event *event, 994 struct perf_sample *sample, 995 struct perf_tool *tool) 996 { 997 events_stats__inc(&session->stats, event->header.type); 998 999 if (event->header.type >= PERF_RECORD_USER_TYPE_START) 1000 return perf_session__process_user_event(session, event, tool, 0); 1001 1002 return perf_session__deliver_event(session, event, sample, tool, 0); 1003 } 1004 1005 static void event_swap(union perf_event *event, bool sample_id_all) 1006 { 1007 perf_event__swap_op swap; 1008 1009 swap = perf_event__swap_ops[event->header.type]; 1010 if (swap) 1011 swap(event, sample_id_all); 1012 } 1013 1014 int perf_session__peek_event(struct perf_session *session, off_t file_offset, 1015 void *buf, size_t buf_sz, 1016 union perf_event **event_ptr, 1017 struct perf_sample *sample) 1018 { 1019 union perf_event *event; 1020 size_t hdr_sz, rest; 1021 int fd; 1022 1023 if (session->one_mmap && !session->header.needs_swap) { 1024 event = file_offset - session->one_mmap_offset + 1025 session->one_mmap_addr; 1026 goto out_parse_sample; 1027 } 1028 1029 if (perf_data_file__is_pipe(session->file)) 1030 return -1; 1031 1032 fd = perf_data_file__fd(session->file); 1033 hdr_sz = sizeof(struct perf_event_header); 1034 1035 if (buf_sz < hdr_sz) 1036 return -1; 1037 1038 if (lseek(fd, file_offset, SEEK_SET) == (off_t)-1 || 1039 readn(fd, &buf, hdr_sz) != (ssize_t)hdr_sz) 1040 return -1; 1041 1042 event = (union perf_event *)buf; 1043 1044 if (session->header.needs_swap) 1045 perf_event_header__bswap(&event->header); 1046 1047 if (event->header.size < hdr_sz) 1048 return -1; 1049 1050 rest = event->header.size - hdr_sz; 1051 1052 if (readn(fd, &buf, rest) != (ssize_t)rest) 1053 return -1; 1054 1055 if (session->header.needs_swap) 1056 event_swap(event, perf_evlist__sample_id_all(session->evlist)); 1057 1058 out_parse_sample: 1059 1060 if (sample && event->header.type < PERF_RECORD_USER_TYPE_START && 1061 perf_evlist__parse_sample(session->evlist, event, sample)) 1062 return -1; 1063 1064 *event_ptr = event; 1065 1066 return 0; 1067 } 1068 1069 static s64 perf_session__process_event(struct perf_session *session, 1070 union perf_event *event, 1071 struct perf_tool *tool, 1072 u64 file_offset) 1073 { 1074 struct perf_sample sample; 1075 int ret; 1076 1077 if (session->header.needs_swap) 1078 event_swap(event, perf_evlist__sample_id_all(session->evlist)); 1079 1080 if (event->header.type >= PERF_RECORD_HEADER_MAX) 1081 return -EINVAL; 1082 1083 events_stats__inc(&session->stats, event->header.type); 1084 1085 if (event->header.type >= PERF_RECORD_USER_TYPE_START) 1086 return perf_session__process_user_event(session, event, tool, file_offset); 1087 1088 /* 1089 * For all kernel events we get the sample data 1090 */ 1091 ret = perf_evlist__parse_sample(session->evlist, event, &sample); 1092 if (ret) 1093 return ret; 1094 1095 if (tool->ordered_events) { 1096 ret = perf_session_queue_event(session, event, tool, &sample, 1097 file_offset); 1098 if (ret != -ETIME) 1099 return ret; 1100 } 1101 1102 return perf_session__deliver_event(session, event, &sample, tool, 1103 file_offset); 1104 } 1105 1106 void perf_event_header__bswap(struct perf_event_header *hdr) 1107 { 1108 hdr->type = bswap_32(hdr->type); 1109 hdr->misc = bswap_16(hdr->misc); 1110 hdr->size = bswap_16(hdr->size); 1111 } 1112 1113 struct thread *perf_session__findnew(struct perf_session *session, pid_t pid) 1114 { 1115 return machine__findnew_thread(&session->machines.host, -1, pid); 1116 } 1117 1118 static struct thread *perf_session__register_idle_thread(struct perf_session *session) 1119 { 1120 struct thread *thread; 1121 1122 thread = machine__findnew_thread(&session->machines.host, 0, 0); 1123 if (thread == NULL || thread__set_comm(thread, "swapper", 0)) { 1124 pr_err("problem inserting idle task.\n"); 1125 thread = NULL; 1126 } 1127 1128 return thread; 1129 } 1130 1131 static void perf_session__warn_about_errors(const struct perf_session *session, 1132 const struct perf_tool *tool) 1133 { 1134 if (tool->lost == perf_event__process_lost && 1135 session->stats.nr_events[PERF_RECORD_LOST] != 0) { 1136 ui__warning("Processed %d events and lost %d chunks!\n\n" 1137 "Check IO/CPU overload!\n\n", 1138 session->stats.nr_events[0], 1139 session->stats.nr_events[PERF_RECORD_LOST]); 1140 } 1141 1142 if (session->stats.nr_unknown_events != 0) { 1143 ui__warning("Found %u unknown events!\n\n" 1144 "Is this an older tool processing a perf.data " 1145 "file generated by a more recent tool?\n\n" 1146 "If that is not the case, consider " 1147 "reporting to linux-kernel@vger.kernel.org.\n\n", 1148 session->stats.nr_unknown_events); 1149 } 1150 1151 if (session->stats.nr_unknown_id != 0) { 1152 ui__warning("%u samples with id not present in the header\n", 1153 session->stats.nr_unknown_id); 1154 } 1155 1156 if (session->stats.nr_invalid_chains != 0) { 1157 ui__warning("Found invalid callchains!\n\n" 1158 "%u out of %u events were discarded for this reason.\n\n" 1159 "Consider reporting to linux-kernel@vger.kernel.org.\n\n", 1160 session->stats.nr_invalid_chains, 1161 session->stats.nr_events[PERF_RECORD_SAMPLE]); 1162 } 1163 1164 if (session->stats.nr_unprocessable_samples != 0) { 1165 ui__warning("%u unprocessable samples recorded.\n" 1166 "Do you have a KVM guest running and not using 'perf kvm'?\n", 1167 session->stats.nr_unprocessable_samples); 1168 } 1169 1170 if (session->stats.nr_unordered_events != 0) 1171 ui__warning("%u out of order events recorded.\n", session->stats.nr_unordered_events); 1172 } 1173 1174 volatile int session_done; 1175 1176 static int __perf_session__process_pipe_events(struct perf_session *session, 1177 struct perf_tool *tool) 1178 { 1179 int fd = perf_data_file__fd(session->file); 1180 union perf_event *event; 1181 uint32_t size, cur_size = 0; 1182 void *buf = NULL; 1183 s64 skip = 0; 1184 u64 head; 1185 ssize_t err; 1186 void *p; 1187 1188 perf_tool__fill_defaults(tool); 1189 1190 head = 0; 1191 cur_size = sizeof(union perf_event); 1192 1193 buf = malloc(cur_size); 1194 if (!buf) 1195 return -errno; 1196 more: 1197 event = buf; 1198 err = readn(fd, event, sizeof(struct perf_event_header)); 1199 if (err <= 0) { 1200 if (err == 0) 1201 goto done; 1202 1203 pr_err("failed to read event header\n"); 1204 goto out_err; 1205 } 1206 1207 if (session->header.needs_swap) 1208 perf_event_header__bswap(&event->header); 1209 1210 size = event->header.size; 1211 if (size < sizeof(struct perf_event_header)) { 1212 pr_err("bad event header size\n"); 1213 goto out_err; 1214 } 1215 1216 if (size > cur_size) { 1217 void *new = realloc(buf, size); 1218 if (!new) { 1219 pr_err("failed to allocate memory to read event\n"); 1220 goto out_err; 1221 } 1222 buf = new; 1223 cur_size = size; 1224 event = buf; 1225 } 1226 p = event; 1227 p += sizeof(struct perf_event_header); 1228 1229 if (size - sizeof(struct perf_event_header)) { 1230 err = readn(fd, p, size - sizeof(struct perf_event_header)); 1231 if (err <= 0) { 1232 if (err == 0) { 1233 pr_err("unexpected end of event stream\n"); 1234 goto done; 1235 } 1236 1237 pr_err("failed to read event data\n"); 1238 goto out_err; 1239 } 1240 } 1241 1242 if ((skip = perf_session__process_event(session, event, tool, head)) < 0) { 1243 pr_err("%#" PRIx64 " [%#x]: failed to process type: %d\n", 1244 head, event->header.size, event->header.type); 1245 err = -EINVAL; 1246 goto out_err; 1247 } 1248 1249 head += size; 1250 1251 if (skip > 0) 1252 head += skip; 1253 1254 if (!session_done()) 1255 goto more; 1256 done: 1257 /* do the final flush for ordered samples */ 1258 err = ordered_events__flush(session, tool, OE_FLUSH__FINAL); 1259 out_err: 1260 free(buf); 1261 perf_session__warn_about_errors(session, tool); 1262 ordered_events__free(&session->ordered_events); 1263 return err; 1264 } 1265 1266 static union perf_event * 1267 fetch_mmaped_event(struct perf_session *session, 1268 u64 head, size_t mmap_size, char *buf) 1269 { 1270 union perf_event *event; 1271 1272 /* 1273 * Ensure we have enough space remaining to read 1274 * the size of the event in the headers. 1275 */ 1276 if (head + sizeof(event->header) > mmap_size) 1277 return NULL; 1278 1279 event = (union perf_event *)(buf + head); 1280 1281 if (session->header.needs_swap) 1282 perf_event_header__bswap(&event->header); 1283 1284 if (head + event->header.size > mmap_size) { 1285 /* We're not fetching the event so swap back again */ 1286 if (session->header.needs_swap) 1287 perf_event_header__bswap(&event->header); 1288 return NULL; 1289 } 1290 1291 return event; 1292 } 1293 1294 /* 1295 * On 64bit we can mmap the data file in one go. No need for tiny mmap 1296 * slices. On 32bit we use 32MB. 1297 */ 1298 #if BITS_PER_LONG == 64 1299 #define MMAP_SIZE ULLONG_MAX 1300 #define NUM_MMAPS 1 1301 #else 1302 #define MMAP_SIZE (32 * 1024 * 1024ULL) 1303 #define NUM_MMAPS 128 1304 #endif 1305 1306 static int __perf_session__process_events(struct perf_session *session, 1307 u64 data_offset, u64 data_size, 1308 u64 file_size, struct perf_tool *tool) 1309 { 1310 int fd = perf_data_file__fd(session->file); 1311 u64 head, page_offset, file_offset, file_pos, size; 1312 int err, mmap_prot, mmap_flags, map_idx = 0; 1313 size_t mmap_size; 1314 char *buf, *mmaps[NUM_MMAPS]; 1315 union perf_event *event; 1316 struct ui_progress prog; 1317 s64 skip; 1318 1319 perf_tool__fill_defaults(tool); 1320 1321 page_offset = page_size * (data_offset / page_size); 1322 file_offset = page_offset; 1323 head = data_offset - page_offset; 1324 1325 if (data_size && (data_offset + data_size < file_size)) 1326 file_size = data_offset + data_size; 1327 1328 ui_progress__init(&prog, file_size, "Processing events..."); 1329 1330 mmap_size = MMAP_SIZE; 1331 if (mmap_size > file_size) { 1332 mmap_size = file_size; 1333 session->one_mmap = true; 1334 } 1335 1336 memset(mmaps, 0, sizeof(mmaps)); 1337 1338 mmap_prot = PROT_READ; 1339 mmap_flags = MAP_SHARED; 1340 1341 if (session->header.needs_swap) { 1342 mmap_prot |= PROT_WRITE; 1343 mmap_flags = MAP_PRIVATE; 1344 } 1345 remap: 1346 buf = mmap(NULL, mmap_size, mmap_prot, mmap_flags, fd, 1347 file_offset); 1348 if (buf == MAP_FAILED) { 1349 pr_err("failed to mmap file\n"); 1350 err = -errno; 1351 goto out_err; 1352 } 1353 mmaps[map_idx] = buf; 1354 map_idx = (map_idx + 1) & (ARRAY_SIZE(mmaps) - 1); 1355 file_pos = file_offset + head; 1356 if (session->one_mmap) { 1357 session->one_mmap_addr = buf; 1358 session->one_mmap_offset = file_offset; 1359 } 1360 1361 more: 1362 event = fetch_mmaped_event(session, head, mmap_size, buf); 1363 if (!event) { 1364 if (mmaps[map_idx]) { 1365 munmap(mmaps[map_idx], mmap_size); 1366 mmaps[map_idx] = NULL; 1367 } 1368 1369 page_offset = page_size * (head / page_size); 1370 file_offset += page_offset; 1371 head -= page_offset; 1372 goto remap; 1373 } 1374 1375 size = event->header.size; 1376 1377 if (size < sizeof(struct perf_event_header) || 1378 (skip = perf_session__process_event(session, event, tool, file_pos)) 1379 < 0) { 1380 pr_err("%#" PRIx64 " [%#x]: failed to process type: %d\n", 1381 file_offset + head, event->header.size, 1382 event->header.type); 1383 err = -EINVAL; 1384 goto out_err; 1385 } 1386 1387 if (skip) 1388 size += skip; 1389 1390 head += size; 1391 file_pos += size; 1392 1393 ui_progress__update(&prog, size); 1394 1395 if (session_done()) 1396 goto out; 1397 1398 if (file_pos < file_size) 1399 goto more; 1400 1401 out: 1402 /* do the final flush for ordered samples */ 1403 err = ordered_events__flush(session, tool, OE_FLUSH__FINAL); 1404 out_err: 1405 ui_progress__finish(); 1406 perf_session__warn_about_errors(session, tool); 1407 ordered_events__free(&session->ordered_events); 1408 session->one_mmap = false; 1409 return err; 1410 } 1411 1412 int perf_session__process_events(struct perf_session *session, 1413 struct perf_tool *tool) 1414 { 1415 u64 size = perf_data_file__size(session->file); 1416 int err; 1417 1418 if (perf_session__register_idle_thread(session) == NULL) 1419 return -ENOMEM; 1420 1421 if (!perf_data_file__is_pipe(session->file)) 1422 err = __perf_session__process_events(session, 1423 session->header.data_offset, 1424 session->header.data_size, 1425 size, tool); 1426 else 1427 err = __perf_session__process_pipe_events(session, tool); 1428 1429 return err; 1430 } 1431 1432 bool perf_session__has_traces(struct perf_session *session, const char *msg) 1433 { 1434 struct perf_evsel *evsel; 1435 1436 evlist__for_each(session->evlist, evsel) { 1437 if (evsel->attr.type == PERF_TYPE_TRACEPOINT) 1438 return true; 1439 } 1440 1441 pr_err("No trace sample to read. Did you call 'perf %s'?\n", msg); 1442 return false; 1443 } 1444 1445 int maps__set_kallsyms_ref_reloc_sym(struct map **maps, 1446 const char *symbol_name, u64 addr) 1447 { 1448 char *bracket; 1449 enum map_type i; 1450 struct ref_reloc_sym *ref; 1451 1452 ref = zalloc(sizeof(struct ref_reloc_sym)); 1453 if (ref == NULL) 1454 return -ENOMEM; 1455 1456 ref->name = strdup(symbol_name); 1457 if (ref->name == NULL) { 1458 free(ref); 1459 return -ENOMEM; 1460 } 1461 1462 bracket = strchr(ref->name, ']'); 1463 if (bracket) 1464 *bracket = '\0'; 1465 1466 ref->addr = addr; 1467 1468 for (i = 0; i < MAP__NR_TYPES; ++i) { 1469 struct kmap *kmap = map__kmap(maps[i]); 1470 kmap->ref_reloc_sym = ref; 1471 } 1472 1473 return 0; 1474 } 1475 1476 size_t perf_session__fprintf_dsos(struct perf_session *session, FILE *fp) 1477 { 1478 return machines__fprintf_dsos(&session->machines, fp); 1479 } 1480 1481 size_t perf_session__fprintf_dsos_buildid(struct perf_session *session, FILE *fp, 1482 bool (skip)(struct dso *dso, int parm), int parm) 1483 { 1484 return machines__fprintf_dsos_buildid(&session->machines, fp, skip, parm); 1485 } 1486 1487 size_t perf_session__fprintf_nr_events(struct perf_session *session, FILE *fp) 1488 { 1489 size_t ret = fprintf(fp, "Aggregated stats:\n"); 1490 1491 ret += events_stats__fprintf(&session->stats, fp); 1492 return ret; 1493 } 1494 1495 size_t perf_session__fprintf(struct perf_session *session, FILE *fp) 1496 { 1497 /* 1498 * FIXME: Here we have to actually print all the machines in this 1499 * session, not just the host... 1500 */ 1501 return machine__fprintf(&session->machines.host, fp); 1502 } 1503 1504 struct perf_evsel *perf_session__find_first_evtype(struct perf_session *session, 1505 unsigned int type) 1506 { 1507 struct perf_evsel *pos; 1508 1509 evlist__for_each(session->evlist, pos) { 1510 if (pos->attr.type == type) 1511 return pos; 1512 } 1513 return NULL; 1514 } 1515 1516 void perf_evsel__print_ip(struct perf_evsel *evsel, struct perf_sample *sample, 1517 struct addr_location *al, 1518 unsigned int print_opts, unsigned int stack_depth) 1519 { 1520 struct callchain_cursor_node *node; 1521 int print_ip = print_opts & PRINT_IP_OPT_IP; 1522 int print_sym = print_opts & PRINT_IP_OPT_SYM; 1523 int print_dso = print_opts & PRINT_IP_OPT_DSO; 1524 int print_symoffset = print_opts & PRINT_IP_OPT_SYMOFFSET; 1525 int print_oneline = print_opts & PRINT_IP_OPT_ONELINE; 1526 int print_srcline = print_opts & PRINT_IP_OPT_SRCLINE; 1527 char s = print_oneline ? ' ' : '\t'; 1528 1529 if (symbol_conf.use_callchain && sample->callchain) { 1530 struct addr_location node_al; 1531 1532 if (thread__resolve_callchain(al->thread, evsel, 1533 sample, NULL, NULL, 1534 PERF_MAX_STACK_DEPTH) != 0) { 1535 if (verbose) 1536 error("Failed to resolve callchain. Skipping\n"); 1537 return; 1538 } 1539 callchain_cursor_commit(&callchain_cursor); 1540 1541 if (print_symoffset) 1542 node_al = *al; 1543 1544 while (stack_depth) { 1545 u64 addr = 0; 1546 1547 node = callchain_cursor_current(&callchain_cursor); 1548 if (!node) 1549 break; 1550 1551 if (node->sym && node->sym->ignore) 1552 goto next; 1553 1554 if (print_ip) 1555 printf("%c%16" PRIx64, s, node->ip); 1556 1557 if (node->map) 1558 addr = node->map->map_ip(node->map, node->ip); 1559 1560 if (print_sym) { 1561 printf(" "); 1562 if (print_symoffset) { 1563 node_al.addr = addr; 1564 node_al.map = node->map; 1565 symbol__fprintf_symname_offs(node->sym, &node_al, stdout); 1566 } else 1567 symbol__fprintf_symname(node->sym, stdout); 1568 } 1569 1570 if (print_dso) { 1571 printf(" ("); 1572 map__fprintf_dsoname(node->map, stdout); 1573 printf(")"); 1574 } 1575 1576 if (print_srcline) 1577 map__fprintf_srcline(node->map, addr, "\n ", 1578 stdout); 1579 1580 if (!print_oneline) 1581 printf("\n"); 1582 1583 stack_depth--; 1584 next: 1585 callchain_cursor_advance(&callchain_cursor); 1586 } 1587 1588 } else { 1589 if (al->sym && al->sym->ignore) 1590 return; 1591 1592 if (print_ip) 1593 printf("%16" PRIx64, sample->ip); 1594 1595 if (print_sym) { 1596 printf(" "); 1597 if (print_symoffset) 1598 symbol__fprintf_symname_offs(al->sym, al, 1599 stdout); 1600 else 1601 symbol__fprintf_symname(al->sym, stdout); 1602 } 1603 1604 if (print_dso) { 1605 printf(" ("); 1606 map__fprintf_dsoname(al->map, stdout); 1607 printf(")"); 1608 } 1609 1610 if (print_srcline) 1611 map__fprintf_srcline(al->map, al->addr, "\n ", stdout); 1612 } 1613 } 1614 1615 int perf_session__cpu_bitmap(struct perf_session *session, 1616 const char *cpu_list, unsigned long *cpu_bitmap) 1617 { 1618 int i, err = -1; 1619 struct cpu_map *map; 1620 1621 for (i = 0; i < PERF_TYPE_MAX; ++i) { 1622 struct perf_evsel *evsel; 1623 1624 evsel = perf_session__find_first_evtype(session, i); 1625 if (!evsel) 1626 continue; 1627 1628 if (!(evsel->attr.sample_type & PERF_SAMPLE_CPU)) { 1629 pr_err("File does not contain CPU events. " 1630 "Remove -c option to proceed.\n"); 1631 return -1; 1632 } 1633 } 1634 1635 map = cpu_map__new(cpu_list); 1636 if (map == NULL) { 1637 pr_err("Invalid cpu_list\n"); 1638 return -1; 1639 } 1640 1641 for (i = 0; i < map->nr; i++) { 1642 int cpu = map->map[i]; 1643 1644 if (cpu >= MAX_NR_CPUS) { 1645 pr_err("Requested CPU %d too large. " 1646 "Consider raising MAX_NR_CPUS\n", cpu); 1647 goto out_delete_map; 1648 } 1649 1650 set_bit(cpu, cpu_bitmap); 1651 } 1652 1653 err = 0; 1654 1655 out_delete_map: 1656 cpu_map__delete(map); 1657 return err; 1658 } 1659 1660 void perf_session__fprintf_info(struct perf_session *session, FILE *fp, 1661 bool full) 1662 { 1663 struct stat st; 1664 int fd, ret; 1665 1666 if (session == NULL || fp == NULL) 1667 return; 1668 1669 fd = perf_data_file__fd(session->file); 1670 1671 ret = fstat(fd, &st); 1672 if (ret == -1) 1673 return; 1674 1675 fprintf(fp, "# ========\n"); 1676 fprintf(fp, "# captured on: %s", ctime(&st.st_ctime)); 1677 perf_header__fprintf_info(session, fp, full); 1678 fprintf(fp, "# ========\n#\n"); 1679 } 1680 1681 1682 int __perf_session__set_tracepoints_handlers(struct perf_session *session, 1683 const struct perf_evsel_str_handler *assocs, 1684 size_t nr_assocs) 1685 { 1686 struct perf_evsel *evsel; 1687 size_t i; 1688 int err; 1689 1690 for (i = 0; i < nr_assocs; i++) { 1691 /* 1692 * Adding a handler for an event not in the session, 1693 * just ignore it. 1694 */ 1695 evsel = perf_evlist__find_tracepoint_by_name(session->evlist, assocs[i].name); 1696 if (evsel == NULL) 1697 continue; 1698 1699 err = -EEXIST; 1700 if (evsel->handler != NULL) 1701 goto out; 1702 evsel->handler = assocs[i].handler; 1703 } 1704 1705 err = 0; 1706 out: 1707 return err; 1708 } 1709 1710 int perf_event__process_id_index(struct perf_tool *tool __maybe_unused, 1711 union perf_event *event, 1712 struct perf_session *session) 1713 { 1714 struct perf_evlist *evlist = session->evlist; 1715 struct id_index_event *ie = &event->id_index; 1716 size_t i, nr, max_nr; 1717 1718 max_nr = (ie->header.size - sizeof(struct id_index_event)) / 1719 sizeof(struct id_index_entry); 1720 nr = ie->nr; 1721 if (nr > max_nr) 1722 return -EINVAL; 1723 1724 if (dump_trace) 1725 fprintf(stdout, " nr: %zu\n", nr); 1726 1727 for (i = 0; i < nr; i++) { 1728 struct id_index_entry *e = &ie->entries[i]; 1729 struct perf_sample_id *sid; 1730 1731 if (dump_trace) { 1732 fprintf(stdout, " ... id: %"PRIu64, e->id); 1733 fprintf(stdout, " idx: %"PRIu64, e->idx); 1734 fprintf(stdout, " cpu: %"PRId64, e->cpu); 1735 fprintf(stdout, " tid: %"PRId64"\n", e->tid); 1736 } 1737 1738 sid = perf_evlist__id2sid(evlist, e->id); 1739 if (!sid) 1740 return -ENOENT; 1741 sid->idx = e->idx; 1742 sid->cpu = e->cpu; 1743 sid->tid = e->tid; 1744 } 1745 return 0; 1746 } 1747 1748 int perf_event__synthesize_id_index(struct perf_tool *tool, 1749 perf_event__handler_t process, 1750 struct perf_evlist *evlist, 1751 struct machine *machine) 1752 { 1753 union perf_event *ev; 1754 struct perf_evsel *evsel; 1755 size_t nr = 0, i = 0, sz, max_nr, n; 1756 int err; 1757 1758 pr_debug2("Synthesizing id index\n"); 1759 1760 max_nr = (UINT16_MAX - sizeof(struct id_index_event)) / 1761 sizeof(struct id_index_entry); 1762 1763 evlist__for_each(evlist, evsel) 1764 nr += evsel->ids; 1765 1766 n = nr > max_nr ? max_nr : nr; 1767 sz = sizeof(struct id_index_event) + n * sizeof(struct id_index_entry); 1768 ev = zalloc(sz); 1769 if (!ev) 1770 return -ENOMEM; 1771 1772 ev->id_index.header.type = PERF_RECORD_ID_INDEX; 1773 ev->id_index.header.size = sz; 1774 ev->id_index.nr = n; 1775 1776 evlist__for_each(evlist, evsel) { 1777 u32 j; 1778 1779 for (j = 0; j < evsel->ids; j++) { 1780 struct id_index_entry *e; 1781 struct perf_sample_id *sid; 1782 1783 if (i >= n) { 1784 err = process(tool, ev, NULL, machine); 1785 if (err) 1786 goto out_err; 1787 nr -= n; 1788 i = 0; 1789 } 1790 1791 e = &ev->id_index.entries[i++]; 1792 1793 e->id = evsel->id[j]; 1794 1795 sid = perf_evlist__id2sid(evlist, e->id); 1796 if (!sid) { 1797 free(ev); 1798 return -ENOENT; 1799 } 1800 1801 e->idx = sid->idx; 1802 e->cpu = sid->cpu; 1803 e->tid = sid->tid; 1804 } 1805 } 1806 1807 sz = sizeof(struct id_index_event) + nr * sizeof(struct id_index_entry); 1808 ev->id_index.header.size = sz; 1809 ev->id_index.nr = nr; 1810 1811 err = process(tool, ev, NULL, machine); 1812 out_err: 1813 free(ev); 1814 1815 return err; 1816 } 1817