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