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 return ordered_events__queue(&s->ordered_events, event, sample, file_offset); 545 } 546 547 static void callchain__lbr_callstack_printf(struct perf_sample *sample) 548 { 549 struct ip_callchain *callchain = sample->callchain; 550 struct branch_stack *lbr_stack = sample->branch_stack; 551 u64 kernel_callchain_nr = callchain->nr; 552 unsigned int i; 553 554 for (i = 0; i < kernel_callchain_nr; i++) { 555 if (callchain->ips[i] == PERF_CONTEXT_USER) 556 break; 557 } 558 559 if ((i != kernel_callchain_nr) && lbr_stack->nr) { 560 u64 total_nr; 561 /* 562 * LBR callstack can only get user call chain, 563 * i is kernel call chain number, 564 * 1 is PERF_CONTEXT_USER. 565 * 566 * The user call chain is stored in LBR registers. 567 * LBR are pair registers. The caller is stored 568 * in "from" register, while the callee is stored 569 * in "to" register. 570 * For example, there is a call stack 571 * "A"->"B"->"C"->"D". 572 * The LBR registers will recorde like 573 * "C"->"D", "B"->"C", "A"->"B". 574 * So only the first "to" register and all "from" 575 * registers are needed to construct the whole stack. 576 */ 577 total_nr = i + 1 + lbr_stack->nr + 1; 578 kernel_callchain_nr = i + 1; 579 580 printf("... LBR call chain: nr:%" PRIu64 "\n", total_nr); 581 582 for (i = 0; i < kernel_callchain_nr; i++) 583 printf("..... %2d: %016" PRIx64 "\n", 584 i, callchain->ips[i]); 585 586 printf("..... %2d: %016" PRIx64 "\n", 587 (int)(kernel_callchain_nr), lbr_stack->entries[0].to); 588 for (i = 0; i < lbr_stack->nr; i++) 589 printf("..... %2d: %016" PRIx64 "\n", 590 (int)(i + kernel_callchain_nr + 1), lbr_stack->entries[i].from); 591 } 592 } 593 594 static void callchain__printf(struct perf_evsel *evsel, 595 struct perf_sample *sample) 596 { 597 unsigned int i; 598 struct ip_callchain *callchain = sample->callchain; 599 600 if (has_branch_callstack(evsel)) 601 callchain__lbr_callstack_printf(sample); 602 603 printf("... FP chain: nr:%" PRIu64 "\n", callchain->nr); 604 605 for (i = 0; i < callchain->nr; i++) 606 printf("..... %2d: %016" PRIx64 "\n", 607 i, callchain->ips[i]); 608 } 609 610 static void branch_stack__printf(struct perf_sample *sample) 611 { 612 uint64_t i; 613 614 printf("... branch stack: nr:%" PRIu64 "\n", sample->branch_stack->nr); 615 616 for (i = 0; i < sample->branch_stack->nr; i++) 617 printf("..... %2"PRIu64": %016" PRIx64 " -> %016" PRIx64 "\n", 618 i, sample->branch_stack->entries[i].from, 619 sample->branch_stack->entries[i].to); 620 } 621 622 static void regs_dump__printf(u64 mask, u64 *regs) 623 { 624 unsigned rid, i = 0; 625 626 for_each_set_bit(rid, (unsigned long *) &mask, sizeof(mask) * 8) { 627 u64 val = regs[i++]; 628 629 printf(".... %-5s 0x%" PRIx64 "\n", 630 perf_reg_name(rid), val); 631 } 632 } 633 634 static const char *regs_abi[] = { 635 [PERF_SAMPLE_REGS_ABI_NONE] = "none", 636 [PERF_SAMPLE_REGS_ABI_32] = "32-bit", 637 [PERF_SAMPLE_REGS_ABI_64] = "64-bit", 638 }; 639 640 static inline const char *regs_dump_abi(struct regs_dump *d) 641 { 642 if (d->abi > PERF_SAMPLE_REGS_ABI_64) 643 return "unknown"; 644 645 return regs_abi[d->abi]; 646 } 647 648 static void regs__printf(const char *type, struct regs_dump *regs) 649 { 650 u64 mask = regs->mask; 651 652 printf("... %s regs: mask 0x%" PRIx64 " ABI %s\n", 653 type, 654 mask, 655 regs_dump_abi(regs)); 656 657 regs_dump__printf(mask, regs->regs); 658 } 659 660 static void regs_user__printf(struct perf_sample *sample) 661 { 662 struct regs_dump *user_regs = &sample->user_regs; 663 664 if (user_regs->regs) 665 regs__printf("user", user_regs); 666 } 667 668 static void regs_intr__printf(struct perf_sample *sample) 669 { 670 struct regs_dump *intr_regs = &sample->intr_regs; 671 672 if (intr_regs->regs) 673 regs__printf("intr", intr_regs); 674 } 675 676 static void stack_user__printf(struct stack_dump *dump) 677 { 678 printf("... ustack: size %" PRIu64 ", offset 0x%x\n", 679 dump->size, dump->offset); 680 } 681 682 static void perf_evlist__print_tstamp(struct perf_evlist *evlist, 683 union perf_event *event, 684 struct perf_sample *sample) 685 { 686 u64 sample_type = __perf_evlist__combined_sample_type(evlist); 687 688 if (event->header.type != PERF_RECORD_SAMPLE && 689 !perf_evlist__sample_id_all(evlist)) { 690 fputs("-1 -1 ", stdout); 691 return; 692 } 693 694 if ((sample_type & PERF_SAMPLE_CPU)) 695 printf("%u ", sample->cpu); 696 697 if (sample_type & PERF_SAMPLE_TIME) 698 printf("%" PRIu64 " ", sample->time); 699 } 700 701 static void sample_read__printf(struct perf_sample *sample, u64 read_format) 702 { 703 printf("... sample_read:\n"); 704 705 if (read_format & PERF_FORMAT_TOTAL_TIME_ENABLED) 706 printf("...... time enabled %016" PRIx64 "\n", 707 sample->read.time_enabled); 708 709 if (read_format & PERF_FORMAT_TOTAL_TIME_RUNNING) 710 printf("...... time running %016" PRIx64 "\n", 711 sample->read.time_running); 712 713 if (read_format & PERF_FORMAT_GROUP) { 714 u64 i; 715 716 printf(".... group nr %" PRIu64 "\n", sample->read.group.nr); 717 718 for (i = 0; i < sample->read.group.nr; i++) { 719 struct sample_read_value *value; 720 721 value = &sample->read.group.values[i]; 722 printf("..... id %016" PRIx64 723 ", value %016" PRIx64 "\n", 724 value->id, value->value); 725 } 726 } else 727 printf("..... id %016" PRIx64 ", value %016" PRIx64 "\n", 728 sample->read.one.id, sample->read.one.value); 729 } 730 731 static void dump_event(struct perf_evlist *evlist, union perf_event *event, 732 u64 file_offset, struct perf_sample *sample) 733 { 734 if (!dump_trace) 735 return; 736 737 printf("\n%#" PRIx64 " [%#x]: event: %d\n", 738 file_offset, event->header.size, event->header.type); 739 740 trace_event(event); 741 742 if (sample) 743 perf_evlist__print_tstamp(evlist, event, sample); 744 745 printf("%#" PRIx64 " [%#x]: PERF_RECORD_%s", file_offset, 746 event->header.size, perf_event__name(event->header.type)); 747 } 748 749 static void dump_sample(struct perf_evsel *evsel, union perf_event *event, 750 struct perf_sample *sample) 751 { 752 u64 sample_type; 753 754 if (!dump_trace) 755 return; 756 757 printf("(IP, 0x%x): %d/%d: %#" PRIx64 " period: %" PRIu64 " addr: %#" PRIx64 "\n", 758 event->header.misc, sample->pid, sample->tid, sample->ip, 759 sample->period, sample->addr); 760 761 sample_type = evsel->attr.sample_type; 762 763 if (sample_type & PERF_SAMPLE_CALLCHAIN) 764 callchain__printf(evsel, sample); 765 766 if ((sample_type & PERF_SAMPLE_BRANCH_STACK) && !has_branch_callstack(evsel)) 767 branch_stack__printf(sample); 768 769 if (sample_type & PERF_SAMPLE_REGS_USER) 770 regs_user__printf(sample); 771 772 if (sample_type & PERF_SAMPLE_REGS_INTR) 773 regs_intr__printf(sample); 774 775 if (sample_type & PERF_SAMPLE_STACK_USER) 776 stack_user__printf(&sample->user_stack); 777 778 if (sample_type & PERF_SAMPLE_WEIGHT) 779 printf("... weight: %" PRIu64 "\n", sample->weight); 780 781 if (sample_type & PERF_SAMPLE_DATA_SRC) 782 printf(" . data_src: 0x%"PRIx64"\n", sample->data_src); 783 784 if (sample_type & PERF_SAMPLE_TRANSACTION) 785 printf("... transaction: %" PRIx64 "\n", sample->transaction); 786 787 if (sample_type & PERF_SAMPLE_READ) 788 sample_read__printf(sample, evsel->attr.read_format); 789 } 790 791 static struct machine *machines__find_for_cpumode(struct machines *machines, 792 union perf_event *event, 793 struct perf_sample *sample) 794 { 795 const u8 cpumode = event->header.misc & PERF_RECORD_MISC_CPUMODE_MASK; 796 struct machine *machine; 797 798 if (perf_guest && 799 ((cpumode == PERF_RECORD_MISC_GUEST_KERNEL) || 800 (cpumode == PERF_RECORD_MISC_GUEST_USER))) { 801 u32 pid; 802 803 if (event->header.type == PERF_RECORD_MMAP 804 || event->header.type == PERF_RECORD_MMAP2) 805 pid = event->mmap.pid; 806 else 807 pid = sample->pid; 808 809 machine = machines__find(machines, pid); 810 if (!machine) 811 machine = machines__find(machines, DEFAULT_GUEST_KERNEL_ID); 812 return machine; 813 } 814 815 return &machines->host; 816 } 817 818 static int deliver_sample_value(struct perf_evlist *evlist, 819 struct perf_tool *tool, 820 union perf_event *event, 821 struct perf_sample *sample, 822 struct sample_read_value *v, 823 struct machine *machine) 824 { 825 struct perf_sample_id *sid = perf_evlist__id2sid(evlist, v->id); 826 827 if (sid) { 828 sample->id = v->id; 829 sample->period = v->value - sid->period; 830 sid->period = v->value; 831 } 832 833 if (!sid || sid->evsel == NULL) { 834 ++evlist->stats.nr_unknown_id; 835 return 0; 836 } 837 838 return tool->sample(tool, event, sample, sid->evsel, machine); 839 } 840 841 static int deliver_sample_group(struct perf_evlist *evlist, 842 struct perf_tool *tool, 843 union perf_event *event, 844 struct perf_sample *sample, 845 struct machine *machine) 846 { 847 int ret = -EINVAL; 848 u64 i; 849 850 for (i = 0; i < sample->read.group.nr; i++) { 851 ret = deliver_sample_value(evlist, tool, event, sample, 852 &sample->read.group.values[i], 853 machine); 854 if (ret) 855 break; 856 } 857 858 return ret; 859 } 860 861 static int 862 perf_evlist__deliver_sample(struct perf_evlist *evlist, 863 struct perf_tool *tool, 864 union perf_event *event, 865 struct perf_sample *sample, 866 struct perf_evsel *evsel, 867 struct machine *machine) 868 { 869 /* We know evsel != NULL. */ 870 u64 sample_type = evsel->attr.sample_type; 871 u64 read_format = evsel->attr.read_format; 872 873 /* Standard sample delievery. */ 874 if (!(sample_type & PERF_SAMPLE_READ)) 875 return tool->sample(tool, event, sample, evsel, machine); 876 877 /* For PERF_SAMPLE_READ we have either single or group mode. */ 878 if (read_format & PERF_FORMAT_GROUP) 879 return deliver_sample_group(evlist, tool, event, sample, 880 machine); 881 else 882 return deliver_sample_value(evlist, tool, event, sample, 883 &sample->read.one, machine); 884 } 885 886 static int machines__deliver_event(struct machines *machines, 887 struct perf_evlist *evlist, 888 union perf_event *event, 889 struct perf_sample *sample, 890 struct perf_tool *tool, u64 file_offset) 891 { 892 struct perf_evsel *evsel; 893 struct machine *machine; 894 895 dump_event(evlist, event, file_offset, sample); 896 897 evsel = perf_evlist__id2evsel(evlist, sample->id); 898 899 machine = machines__find_for_cpumode(machines, event, sample); 900 901 switch (event->header.type) { 902 case PERF_RECORD_SAMPLE: 903 dump_sample(evsel, event, sample); 904 if (evsel == NULL) { 905 ++evlist->stats.nr_unknown_id; 906 return 0; 907 } 908 if (machine == NULL) { 909 ++evlist->stats.nr_unprocessable_samples; 910 return 0; 911 } 912 return perf_evlist__deliver_sample(evlist, tool, event, sample, evsel, machine); 913 case PERF_RECORD_MMAP: 914 return tool->mmap(tool, event, sample, machine); 915 case PERF_RECORD_MMAP2: 916 return tool->mmap2(tool, event, sample, machine); 917 case PERF_RECORD_COMM: 918 return tool->comm(tool, event, sample, machine); 919 case PERF_RECORD_FORK: 920 return tool->fork(tool, event, sample, machine); 921 case PERF_RECORD_EXIT: 922 return tool->exit(tool, event, sample, machine); 923 case PERF_RECORD_LOST: 924 if (tool->lost == perf_event__process_lost) 925 evlist->stats.total_lost += event->lost.lost; 926 return tool->lost(tool, event, sample, machine); 927 case PERF_RECORD_READ: 928 return tool->read(tool, event, sample, evsel, machine); 929 case PERF_RECORD_THROTTLE: 930 return tool->throttle(tool, event, sample, machine); 931 case PERF_RECORD_UNTHROTTLE: 932 return tool->unthrottle(tool, event, sample, machine); 933 default: 934 ++evlist->stats.nr_unknown_events; 935 return -1; 936 } 937 } 938 939 static s64 perf_session__process_user_event(struct perf_session *session, 940 union perf_event *event, 941 u64 file_offset) 942 { 943 struct ordered_events *oe = &session->ordered_events; 944 struct perf_tool *tool = oe->tool; 945 int fd = perf_data_file__fd(session->file); 946 int err; 947 948 dump_event(session->evlist, event, file_offset, NULL); 949 950 /* These events are processed right away */ 951 switch (event->header.type) { 952 case PERF_RECORD_HEADER_ATTR: 953 err = tool->attr(tool, event, &session->evlist); 954 if (err == 0) { 955 perf_session__set_id_hdr_size(session); 956 perf_session__set_comm_exec(session); 957 } 958 return err; 959 case PERF_RECORD_HEADER_EVENT_TYPE: 960 /* 961 * Depreceated, but we need to handle it for sake 962 * of old data files create in pipe mode. 963 */ 964 return 0; 965 case PERF_RECORD_HEADER_TRACING_DATA: 966 /* setup for reading amidst mmap */ 967 lseek(fd, file_offset, SEEK_SET); 968 return tool->tracing_data(tool, event, session); 969 case PERF_RECORD_HEADER_BUILD_ID: 970 return tool->build_id(tool, event, session); 971 case PERF_RECORD_FINISHED_ROUND: 972 return tool->finished_round(tool, event, oe); 973 case PERF_RECORD_ID_INDEX: 974 return tool->id_index(tool, event, session); 975 default: 976 return -EINVAL; 977 } 978 } 979 980 int perf_session__deliver_synth_event(struct perf_session *session, 981 union perf_event *event, 982 struct perf_sample *sample) 983 { 984 struct perf_evlist *evlist = session->evlist; 985 struct perf_tool *tool = session->ordered_events.tool; 986 987 events_stats__inc(&evlist->stats, event->header.type); 988 989 if (event->header.type >= PERF_RECORD_USER_TYPE_START) 990 return perf_session__process_user_event(session, event, 0); 991 992 return machines__deliver_event(&session->machines, evlist, event, sample, tool, 0); 993 } 994 995 static void event_swap(union perf_event *event, bool sample_id_all) 996 { 997 perf_event__swap_op swap; 998 999 swap = perf_event__swap_ops[event->header.type]; 1000 if (swap) 1001 swap(event, sample_id_all); 1002 } 1003 1004 int perf_session__peek_event(struct perf_session *session, off_t file_offset, 1005 void *buf, size_t buf_sz, 1006 union perf_event **event_ptr, 1007 struct perf_sample *sample) 1008 { 1009 union perf_event *event; 1010 size_t hdr_sz, rest; 1011 int fd; 1012 1013 if (session->one_mmap && !session->header.needs_swap) { 1014 event = file_offset - session->one_mmap_offset + 1015 session->one_mmap_addr; 1016 goto out_parse_sample; 1017 } 1018 1019 if (perf_data_file__is_pipe(session->file)) 1020 return -1; 1021 1022 fd = perf_data_file__fd(session->file); 1023 hdr_sz = sizeof(struct perf_event_header); 1024 1025 if (buf_sz < hdr_sz) 1026 return -1; 1027 1028 if (lseek(fd, file_offset, SEEK_SET) == (off_t)-1 || 1029 readn(fd, &buf, hdr_sz) != (ssize_t)hdr_sz) 1030 return -1; 1031 1032 event = (union perf_event *)buf; 1033 1034 if (session->header.needs_swap) 1035 perf_event_header__bswap(&event->header); 1036 1037 if (event->header.size < hdr_sz) 1038 return -1; 1039 1040 rest = event->header.size - hdr_sz; 1041 1042 if (readn(fd, &buf, rest) != (ssize_t)rest) 1043 return -1; 1044 1045 if (session->header.needs_swap) 1046 event_swap(event, perf_evlist__sample_id_all(session->evlist)); 1047 1048 out_parse_sample: 1049 1050 if (sample && event->header.type < PERF_RECORD_USER_TYPE_START && 1051 perf_evlist__parse_sample(session->evlist, event, sample)) 1052 return -1; 1053 1054 *event_ptr = event; 1055 1056 return 0; 1057 } 1058 1059 static s64 perf_session__process_event(struct perf_session *session, 1060 union perf_event *event, u64 file_offset) 1061 { 1062 struct perf_evlist *evlist = session->evlist; 1063 struct perf_tool *tool = session->ordered_events.tool; 1064 struct perf_sample sample; 1065 int ret; 1066 1067 if (session->header.needs_swap) 1068 event_swap(event, perf_evlist__sample_id_all(evlist)); 1069 1070 if (event->header.type >= PERF_RECORD_HEADER_MAX) 1071 return -EINVAL; 1072 1073 events_stats__inc(&evlist->stats, event->header.type); 1074 1075 if (event->header.type >= PERF_RECORD_USER_TYPE_START) 1076 return perf_session__process_user_event(session, event, file_offset); 1077 1078 /* 1079 * For all kernel events we get the sample data 1080 */ 1081 ret = perf_evlist__parse_sample(evlist, event, &sample); 1082 if (ret) 1083 return ret; 1084 1085 if (tool->ordered_events) { 1086 ret = perf_session__queue_event(session, event, &sample, file_offset); 1087 if (ret != -ETIME) 1088 return ret; 1089 } 1090 1091 return machines__deliver_event(&session->machines, evlist, event, 1092 &sample, tool, file_offset); 1093 } 1094 1095 void perf_event_header__bswap(struct perf_event_header *hdr) 1096 { 1097 hdr->type = bswap_32(hdr->type); 1098 hdr->misc = bswap_16(hdr->misc); 1099 hdr->size = bswap_16(hdr->size); 1100 } 1101 1102 struct thread *perf_session__findnew(struct perf_session *session, pid_t pid) 1103 { 1104 return machine__findnew_thread(&session->machines.host, -1, pid); 1105 } 1106 1107 static struct thread *perf_session__register_idle_thread(struct perf_session *session) 1108 { 1109 struct thread *thread; 1110 1111 thread = machine__findnew_thread(&session->machines.host, 0, 0); 1112 if (thread == NULL || thread__set_comm(thread, "swapper", 0)) { 1113 pr_err("problem inserting idle task.\n"); 1114 thread = NULL; 1115 } 1116 1117 return thread; 1118 } 1119 1120 static void perf_tool__warn_about_errors(const struct perf_tool *tool, 1121 const struct events_stats *stats) 1122 { 1123 if (tool->lost == perf_event__process_lost && 1124 stats->nr_events[PERF_RECORD_LOST] != 0) { 1125 ui__warning("Processed %d events and lost %d chunks!\n\n" 1126 "Check IO/CPU overload!\n\n", 1127 stats->nr_events[0], 1128 stats->nr_events[PERF_RECORD_LOST]); 1129 } 1130 1131 if (stats->nr_unknown_events != 0) { 1132 ui__warning("Found %u unknown events!\n\n" 1133 "Is this an older tool processing a perf.data " 1134 "file generated by a more recent tool?\n\n" 1135 "If that is not the case, consider " 1136 "reporting to linux-kernel@vger.kernel.org.\n\n", 1137 stats->nr_unknown_events); 1138 } 1139 1140 if (stats->nr_unknown_id != 0) { 1141 ui__warning("%u samples with id not present in the header\n", 1142 stats->nr_unknown_id); 1143 } 1144 1145 if (stats->nr_invalid_chains != 0) { 1146 ui__warning("Found invalid callchains!\n\n" 1147 "%u out of %u events were discarded for this reason.\n\n" 1148 "Consider reporting to linux-kernel@vger.kernel.org.\n\n", 1149 stats->nr_invalid_chains, 1150 stats->nr_events[PERF_RECORD_SAMPLE]); 1151 } 1152 1153 if (stats->nr_unprocessable_samples != 0) { 1154 ui__warning("%u unprocessable samples recorded.\n" 1155 "Do you have a KVM guest running and not using 'perf kvm'?\n", 1156 stats->nr_unprocessable_samples); 1157 } 1158 1159 if (stats->nr_unordered_events != 0) 1160 ui__warning("%u out of order events recorded.\n", stats->nr_unordered_events); 1161 } 1162 1163 volatile int session_done; 1164 1165 static int __perf_session__process_pipe_events(struct perf_session *session) 1166 { 1167 struct ordered_events *oe = &session->ordered_events; 1168 struct perf_tool *tool = oe->tool; 1169 int fd = perf_data_file__fd(session->file); 1170 union perf_event *event; 1171 uint32_t size, cur_size = 0; 1172 void *buf = NULL; 1173 s64 skip = 0; 1174 u64 head; 1175 ssize_t err; 1176 void *p; 1177 1178 perf_tool__fill_defaults(tool); 1179 1180 head = 0; 1181 cur_size = sizeof(union perf_event); 1182 1183 buf = malloc(cur_size); 1184 if (!buf) 1185 return -errno; 1186 more: 1187 event = buf; 1188 err = readn(fd, event, sizeof(struct perf_event_header)); 1189 if (err <= 0) { 1190 if (err == 0) 1191 goto done; 1192 1193 pr_err("failed to read event header\n"); 1194 goto out_err; 1195 } 1196 1197 if (session->header.needs_swap) 1198 perf_event_header__bswap(&event->header); 1199 1200 size = event->header.size; 1201 if (size < sizeof(struct perf_event_header)) { 1202 pr_err("bad event header size\n"); 1203 goto out_err; 1204 } 1205 1206 if (size > cur_size) { 1207 void *new = realloc(buf, size); 1208 if (!new) { 1209 pr_err("failed to allocate memory to read event\n"); 1210 goto out_err; 1211 } 1212 buf = new; 1213 cur_size = size; 1214 event = buf; 1215 } 1216 p = event; 1217 p += sizeof(struct perf_event_header); 1218 1219 if (size - sizeof(struct perf_event_header)) { 1220 err = readn(fd, p, size - sizeof(struct perf_event_header)); 1221 if (err <= 0) { 1222 if (err == 0) { 1223 pr_err("unexpected end of event stream\n"); 1224 goto done; 1225 } 1226 1227 pr_err("failed to read event data\n"); 1228 goto out_err; 1229 } 1230 } 1231 1232 if ((skip = perf_session__process_event(session, event, head)) < 0) { 1233 pr_err("%#" PRIx64 " [%#x]: failed to process type: %d\n", 1234 head, event->header.size, event->header.type); 1235 err = -EINVAL; 1236 goto out_err; 1237 } 1238 1239 head += size; 1240 1241 if (skip > 0) 1242 head += skip; 1243 1244 if (!session_done()) 1245 goto more; 1246 done: 1247 /* do the final flush for ordered samples */ 1248 err = ordered_events__flush(oe, OE_FLUSH__FINAL); 1249 out_err: 1250 free(buf); 1251 perf_tool__warn_about_errors(tool, &session->evlist->stats); 1252 ordered_events__free(&session->ordered_events); 1253 return err; 1254 } 1255 1256 static union perf_event * 1257 fetch_mmaped_event(struct perf_session *session, 1258 u64 head, size_t mmap_size, char *buf) 1259 { 1260 union perf_event *event; 1261 1262 /* 1263 * Ensure we have enough space remaining to read 1264 * the size of the event in the headers. 1265 */ 1266 if (head + sizeof(event->header) > mmap_size) 1267 return NULL; 1268 1269 event = (union perf_event *)(buf + head); 1270 1271 if (session->header.needs_swap) 1272 perf_event_header__bswap(&event->header); 1273 1274 if (head + event->header.size > mmap_size) { 1275 /* We're not fetching the event so swap back again */ 1276 if (session->header.needs_swap) 1277 perf_event_header__bswap(&event->header); 1278 return NULL; 1279 } 1280 1281 return event; 1282 } 1283 1284 /* 1285 * On 64bit we can mmap the data file in one go. No need for tiny mmap 1286 * slices. On 32bit we use 32MB. 1287 */ 1288 #if BITS_PER_LONG == 64 1289 #define MMAP_SIZE ULLONG_MAX 1290 #define NUM_MMAPS 1 1291 #else 1292 #define MMAP_SIZE (32 * 1024 * 1024ULL) 1293 #define NUM_MMAPS 128 1294 #endif 1295 1296 static int __perf_session__process_events(struct perf_session *session, 1297 u64 data_offset, u64 data_size, 1298 u64 file_size) 1299 { 1300 struct ordered_events *oe = &session->ordered_events; 1301 struct perf_tool *tool = oe->tool; 1302 int fd = perf_data_file__fd(session->file); 1303 u64 head, page_offset, file_offset, file_pos, size; 1304 int err, mmap_prot, mmap_flags, map_idx = 0; 1305 size_t mmap_size; 1306 char *buf, *mmaps[NUM_MMAPS]; 1307 union perf_event *event; 1308 struct ui_progress prog; 1309 s64 skip; 1310 1311 perf_tool__fill_defaults(tool); 1312 1313 page_offset = page_size * (data_offset / page_size); 1314 file_offset = page_offset; 1315 head = data_offset - page_offset; 1316 1317 if (data_size && (data_offset + data_size < file_size)) 1318 file_size = data_offset + data_size; 1319 1320 ui_progress__init(&prog, file_size, "Processing events..."); 1321 1322 mmap_size = MMAP_SIZE; 1323 if (mmap_size > file_size) { 1324 mmap_size = file_size; 1325 session->one_mmap = true; 1326 } 1327 1328 memset(mmaps, 0, sizeof(mmaps)); 1329 1330 mmap_prot = PROT_READ; 1331 mmap_flags = MAP_SHARED; 1332 1333 if (session->header.needs_swap) { 1334 mmap_prot |= PROT_WRITE; 1335 mmap_flags = MAP_PRIVATE; 1336 } 1337 remap: 1338 buf = mmap(NULL, mmap_size, mmap_prot, mmap_flags, fd, 1339 file_offset); 1340 if (buf == MAP_FAILED) { 1341 pr_err("failed to mmap file\n"); 1342 err = -errno; 1343 goto out_err; 1344 } 1345 mmaps[map_idx] = buf; 1346 map_idx = (map_idx + 1) & (ARRAY_SIZE(mmaps) - 1); 1347 file_pos = file_offset + head; 1348 if (session->one_mmap) { 1349 session->one_mmap_addr = buf; 1350 session->one_mmap_offset = file_offset; 1351 } 1352 1353 more: 1354 event = fetch_mmaped_event(session, head, mmap_size, buf); 1355 if (!event) { 1356 if (mmaps[map_idx]) { 1357 munmap(mmaps[map_idx], mmap_size); 1358 mmaps[map_idx] = NULL; 1359 } 1360 1361 page_offset = page_size * (head / page_size); 1362 file_offset += page_offset; 1363 head -= page_offset; 1364 goto remap; 1365 } 1366 1367 size = event->header.size; 1368 1369 if (size < sizeof(struct perf_event_header) || 1370 (skip = perf_session__process_event(session, event, file_pos)) < 0) { 1371 pr_err("%#" PRIx64 " [%#x]: failed to process type: %d\n", 1372 file_offset + head, event->header.size, 1373 event->header.type); 1374 err = -EINVAL; 1375 goto out_err; 1376 } 1377 1378 if (skip) 1379 size += skip; 1380 1381 head += size; 1382 file_pos += size; 1383 1384 ui_progress__update(&prog, size); 1385 1386 if (session_done()) 1387 goto out; 1388 1389 if (file_pos < file_size) 1390 goto more; 1391 1392 out: 1393 /* do the final flush for ordered samples */ 1394 err = ordered_events__flush(oe, OE_FLUSH__FINAL); 1395 out_err: 1396 ui_progress__finish(); 1397 perf_tool__warn_about_errors(tool, &session->evlist->stats); 1398 ordered_events__free(&session->ordered_events); 1399 session->one_mmap = false; 1400 return err; 1401 } 1402 1403 int perf_session__process_events(struct perf_session *session) 1404 { 1405 u64 size = perf_data_file__size(session->file); 1406 int err; 1407 1408 if (perf_session__register_idle_thread(session) == NULL) 1409 return -ENOMEM; 1410 1411 if (!perf_data_file__is_pipe(session->file)) 1412 err = __perf_session__process_events(session, 1413 session->header.data_offset, 1414 session->header.data_size, size); 1415 else 1416 err = __perf_session__process_pipe_events(session); 1417 1418 return err; 1419 } 1420 1421 bool perf_session__has_traces(struct perf_session *session, const char *msg) 1422 { 1423 struct perf_evsel *evsel; 1424 1425 evlist__for_each(session->evlist, evsel) { 1426 if (evsel->attr.type == PERF_TYPE_TRACEPOINT) 1427 return true; 1428 } 1429 1430 pr_err("No trace sample to read. Did you call 'perf %s'?\n", msg); 1431 return false; 1432 } 1433 1434 int maps__set_kallsyms_ref_reloc_sym(struct map **maps, 1435 const char *symbol_name, u64 addr) 1436 { 1437 char *bracket; 1438 enum map_type i; 1439 struct ref_reloc_sym *ref; 1440 1441 ref = zalloc(sizeof(struct ref_reloc_sym)); 1442 if (ref == NULL) 1443 return -ENOMEM; 1444 1445 ref->name = strdup(symbol_name); 1446 if (ref->name == NULL) { 1447 free(ref); 1448 return -ENOMEM; 1449 } 1450 1451 bracket = strchr(ref->name, ']'); 1452 if (bracket) 1453 *bracket = '\0'; 1454 1455 ref->addr = addr; 1456 1457 for (i = 0; i < MAP__NR_TYPES; ++i) { 1458 struct kmap *kmap = map__kmap(maps[i]); 1459 kmap->ref_reloc_sym = ref; 1460 } 1461 1462 return 0; 1463 } 1464 1465 size_t perf_session__fprintf_dsos(struct perf_session *session, FILE *fp) 1466 { 1467 return machines__fprintf_dsos(&session->machines, fp); 1468 } 1469 1470 size_t perf_session__fprintf_dsos_buildid(struct perf_session *session, FILE *fp, 1471 bool (skip)(struct dso *dso, int parm), int parm) 1472 { 1473 return machines__fprintf_dsos_buildid(&session->machines, fp, skip, parm); 1474 } 1475 1476 size_t perf_session__fprintf_nr_events(struct perf_session *session, FILE *fp) 1477 { 1478 size_t ret = fprintf(fp, "Aggregated stats:\n"); 1479 1480 ret += events_stats__fprintf(&session->evlist->stats, fp); 1481 return ret; 1482 } 1483 1484 size_t perf_session__fprintf(struct perf_session *session, FILE *fp) 1485 { 1486 /* 1487 * FIXME: Here we have to actually print all the machines in this 1488 * session, not just the host... 1489 */ 1490 return machine__fprintf(&session->machines.host, fp); 1491 } 1492 1493 struct perf_evsel *perf_session__find_first_evtype(struct perf_session *session, 1494 unsigned int type) 1495 { 1496 struct perf_evsel *pos; 1497 1498 evlist__for_each(session->evlist, pos) { 1499 if (pos->attr.type == type) 1500 return pos; 1501 } 1502 return NULL; 1503 } 1504 1505 void perf_evsel__print_ip(struct perf_evsel *evsel, struct perf_sample *sample, 1506 struct addr_location *al, 1507 unsigned int print_opts, unsigned int stack_depth) 1508 { 1509 struct callchain_cursor_node *node; 1510 int print_ip = print_opts & PRINT_IP_OPT_IP; 1511 int print_sym = print_opts & PRINT_IP_OPT_SYM; 1512 int print_dso = print_opts & PRINT_IP_OPT_DSO; 1513 int print_symoffset = print_opts & PRINT_IP_OPT_SYMOFFSET; 1514 int print_oneline = print_opts & PRINT_IP_OPT_ONELINE; 1515 int print_srcline = print_opts & PRINT_IP_OPT_SRCLINE; 1516 char s = print_oneline ? ' ' : '\t'; 1517 1518 if (symbol_conf.use_callchain && sample->callchain) { 1519 struct addr_location node_al; 1520 1521 if (thread__resolve_callchain(al->thread, evsel, 1522 sample, NULL, NULL, 1523 PERF_MAX_STACK_DEPTH) != 0) { 1524 if (verbose) 1525 error("Failed to resolve callchain. Skipping\n"); 1526 return; 1527 } 1528 callchain_cursor_commit(&callchain_cursor); 1529 1530 if (print_symoffset) 1531 node_al = *al; 1532 1533 while (stack_depth) { 1534 u64 addr = 0; 1535 1536 node = callchain_cursor_current(&callchain_cursor); 1537 if (!node) 1538 break; 1539 1540 if (node->sym && node->sym->ignore) 1541 goto next; 1542 1543 if (print_ip) 1544 printf("%c%16" PRIx64, s, node->ip); 1545 1546 if (node->map) 1547 addr = node->map->map_ip(node->map, node->ip); 1548 1549 if (print_sym) { 1550 printf(" "); 1551 if (print_symoffset) { 1552 node_al.addr = addr; 1553 node_al.map = node->map; 1554 symbol__fprintf_symname_offs(node->sym, &node_al, stdout); 1555 } else 1556 symbol__fprintf_symname(node->sym, stdout); 1557 } 1558 1559 if (print_dso) { 1560 printf(" ("); 1561 map__fprintf_dsoname(node->map, stdout); 1562 printf(")"); 1563 } 1564 1565 if (print_srcline) 1566 map__fprintf_srcline(node->map, addr, "\n ", 1567 stdout); 1568 1569 if (!print_oneline) 1570 printf("\n"); 1571 1572 stack_depth--; 1573 next: 1574 callchain_cursor_advance(&callchain_cursor); 1575 } 1576 1577 } else { 1578 if (al->sym && al->sym->ignore) 1579 return; 1580 1581 if (print_ip) 1582 printf("%16" PRIx64, sample->ip); 1583 1584 if (print_sym) { 1585 printf(" "); 1586 if (print_symoffset) 1587 symbol__fprintf_symname_offs(al->sym, al, 1588 stdout); 1589 else 1590 symbol__fprintf_symname(al->sym, stdout); 1591 } 1592 1593 if (print_dso) { 1594 printf(" ("); 1595 map__fprintf_dsoname(al->map, stdout); 1596 printf(")"); 1597 } 1598 1599 if (print_srcline) 1600 map__fprintf_srcline(al->map, al->addr, "\n ", stdout); 1601 } 1602 } 1603 1604 int perf_session__cpu_bitmap(struct perf_session *session, 1605 const char *cpu_list, unsigned long *cpu_bitmap) 1606 { 1607 int i, err = -1; 1608 struct cpu_map *map; 1609 1610 for (i = 0; i < PERF_TYPE_MAX; ++i) { 1611 struct perf_evsel *evsel; 1612 1613 evsel = perf_session__find_first_evtype(session, i); 1614 if (!evsel) 1615 continue; 1616 1617 if (!(evsel->attr.sample_type & PERF_SAMPLE_CPU)) { 1618 pr_err("File does not contain CPU events. " 1619 "Remove -c option to proceed.\n"); 1620 return -1; 1621 } 1622 } 1623 1624 map = cpu_map__new(cpu_list); 1625 if (map == NULL) { 1626 pr_err("Invalid cpu_list\n"); 1627 return -1; 1628 } 1629 1630 for (i = 0; i < map->nr; i++) { 1631 int cpu = map->map[i]; 1632 1633 if (cpu >= MAX_NR_CPUS) { 1634 pr_err("Requested CPU %d too large. " 1635 "Consider raising MAX_NR_CPUS\n", cpu); 1636 goto out_delete_map; 1637 } 1638 1639 set_bit(cpu, cpu_bitmap); 1640 } 1641 1642 err = 0; 1643 1644 out_delete_map: 1645 cpu_map__delete(map); 1646 return err; 1647 } 1648 1649 void perf_session__fprintf_info(struct perf_session *session, FILE *fp, 1650 bool full) 1651 { 1652 struct stat st; 1653 int fd, ret; 1654 1655 if (session == NULL || fp == NULL) 1656 return; 1657 1658 fd = perf_data_file__fd(session->file); 1659 1660 ret = fstat(fd, &st); 1661 if (ret == -1) 1662 return; 1663 1664 fprintf(fp, "# ========\n"); 1665 fprintf(fp, "# captured on: %s", ctime(&st.st_ctime)); 1666 perf_header__fprintf_info(session, fp, full); 1667 fprintf(fp, "# ========\n#\n"); 1668 } 1669 1670 1671 int __perf_session__set_tracepoints_handlers(struct perf_session *session, 1672 const struct perf_evsel_str_handler *assocs, 1673 size_t nr_assocs) 1674 { 1675 struct perf_evsel *evsel; 1676 size_t i; 1677 int err; 1678 1679 for (i = 0; i < nr_assocs; i++) { 1680 /* 1681 * Adding a handler for an event not in the session, 1682 * just ignore it. 1683 */ 1684 evsel = perf_evlist__find_tracepoint_by_name(session->evlist, assocs[i].name); 1685 if (evsel == NULL) 1686 continue; 1687 1688 err = -EEXIST; 1689 if (evsel->handler != NULL) 1690 goto out; 1691 evsel->handler = assocs[i].handler; 1692 } 1693 1694 err = 0; 1695 out: 1696 return err; 1697 } 1698 1699 int perf_event__process_id_index(struct perf_tool *tool __maybe_unused, 1700 union perf_event *event, 1701 struct perf_session *session) 1702 { 1703 struct perf_evlist *evlist = session->evlist; 1704 struct id_index_event *ie = &event->id_index; 1705 size_t i, nr, max_nr; 1706 1707 max_nr = (ie->header.size - sizeof(struct id_index_event)) / 1708 sizeof(struct id_index_entry); 1709 nr = ie->nr; 1710 if (nr > max_nr) 1711 return -EINVAL; 1712 1713 if (dump_trace) 1714 fprintf(stdout, " nr: %zu\n", nr); 1715 1716 for (i = 0; i < nr; i++) { 1717 struct id_index_entry *e = &ie->entries[i]; 1718 struct perf_sample_id *sid; 1719 1720 if (dump_trace) { 1721 fprintf(stdout, " ... id: %"PRIu64, e->id); 1722 fprintf(stdout, " idx: %"PRIu64, e->idx); 1723 fprintf(stdout, " cpu: %"PRId64, e->cpu); 1724 fprintf(stdout, " tid: %"PRId64"\n", e->tid); 1725 } 1726 1727 sid = perf_evlist__id2sid(evlist, e->id); 1728 if (!sid) 1729 return -ENOENT; 1730 sid->idx = e->idx; 1731 sid->cpu = e->cpu; 1732 sid->tid = e->tid; 1733 } 1734 return 0; 1735 } 1736 1737 int perf_event__synthesize_id_index(struct perf_tool *tool, 1738 perf_event__handler_t process, 1739 struct perf_evlist *evlist, 1740 struct machine *machine) 1741 { 1742 union perf_event *ev; 1743 struct perf_evsel *evsel; 1744 size_t nr = 0, i = 0, sz, max_nr, n; 1745 int err; 1746 1747 pr_debug2("Synthesizing id index\n"); 1748 1749 max_nr = (UINT16_MAX - sizeof(struct id_index_event)) / 1750 sizeof(struct id_index_entry); 1751 1752 evlist__for_each(evlist, evsel) 1753 nr += evsel->ids; 1754 1755 n = nr > max_nr ? max_nr : nr; 1756 sz = sizeof(struct id_index_event) + n * sizeof(struct id_index_entry); 1757 ev = zalloc(sz); 1758 if (!ev) 1759 return -ENOMEM; 1760 1761 ev->id_index.header.type = PERF_RECORD_ID_INDEX; 1762 ev->id_index.header.size = sz; 1763 ev->id_index.nr = n; 1764 1765 evlist__for_each(evlist, evsel) { 1766 u32 j; 1767 1768 for (j = 0; j < evsel->ids; j++) { 1769 struct id_index_entry *e; 1770 struct perf_sample_id *sid; 1771 1772 if (i >= n) { 1773 err = process(tool, ev, NULL, machine); 1774 if (err) 1775 goto out_err; 1776 nr -= n; 1777 i = 0; 1778 } 1779 1780 e = &ev->id_index.entries[i++]; 1781 1782 e->id = evsel->id[j]; 1783 1784 sid = perf_evlist__id2sid(evlist, e->id); 1785 if (!sid) { 1786 free(ev); 1787 return -ENOENT; 1788 } 1789 1790 e->idx = sid->idx; 1791 e->cpu = sid->cpu; 1792 e->tid = sid->tid; 1793 } 1794 } 1795 1796 sz = sizeof(struct id_index_event) + nr * sizeof(struct id_index_entry); 1797 ev->id_index.header.size = sz; 1798 ev->id_index.nr = nr; 1799 1800 err = process(tool, ev, NULL, machine); 1801 out_err: 1802 free(ev); 1803 1804 return err; 1805 } 1806