1 #include <linux/kernel.h> 2 #include <traceevent/event-parse.h> 3 4 #include <byteswap.h> 5 #include <unistd.h> 6 #include <sys/types.h> 7 #include <sys/mman.h> 8 9 #include "evlist.h" 10 #include "evsel.h" 11 #include "session.h" 12 #include "tool.h" 13 #include "sort.h" 14 #include "util.h" 15 #include "cpumap.h" 16 #include "perf_regs.h" 17 #include "asm/bug.h" 18 19 static int perf_session__open(struct perf_session *session) 20 { 21 struct perf_data_file *file = session->file; 22 23 if (perf_session__read_header(session) < 0) { 24 pr_err("incompatible file format (rerun with -v to learn more)"); 25 return -1; 26 } 27 28 if (perf_data_file__is_pipe(file)) 29 return 0; 30 31 if (!perf_evlist__valid_sample_type(session->evlist)) { 32 pr_err("non matching sample_type"); 33 return -1; 34 } 35 36 if (!perf_evlist__valid_sample_id_all(session->evlist)) { 37 pr_err("non matching sample_id_all"); 38 return -1; 39 } 40 41 if (!perf_evlist__valid_read_format(session->evlist)) { 42 pr_err("non matching read_format"); 43 return -1; 44 } 45 46 return 0; 47 } 48 49 void perf_session__set_id_hdr_size(struct perf_session *session) 50 { 51 u16 id_hdr_size = perf_evlist__id_hdr_size(session->evlist); 52 53 machines__set_id_hdr_size(&session->machines, id_hdr_size); 54 } 55 56 int perf_session__create_kernel_maps(struct perf_session *session) 57 { 58 int ret = machine__create_kernel_maps(&session->machines.host); 59 60 if (ret >= 0) 61 ret = machines__create_guest_kernel_maps(&session->machines); 62 return ret; 63 } 64 65 static void perf_session__destroy_kernel_maps(struct perf_session *session) 66 { 67 machines__destroy_kernel_maps(&session->machines); 68 } 69 70 static bool perf_session__has_comm_exec(struct perf_session *session) 71 { 72 struct perf_evsel *evsel; 73 74 evlist__for_each(session->evlist, evsel) { 75 if (evsel->attr.comm_exec) 76 return true; 77 } 78 79 return false; 80 } 81 82 static void perf_session__set_comm_exec(struct perf_session *session) 83 { 84 bool comm_exec = perf_session__has_comm_exec(session); 85 86 machines__set_comm_exec(&session->machines, comm_exec); 87 } 88 89 struct perf_session *perf_session__new(struct perf_data_file *file, 90 bool repipe, struct perf_tool *tool) 91 { 92 struct perf_session *session = zalloc(sizeof(*session)); 93 94 if (!session) 95 goto out; 96 97 session->repipe = repipe; 98 ordered_events__init(&session->ordered_events); 99 machines__init(&session->machines); 100 101 if (file) { 102 if (perf_data_file__open(file)) 103 goto out_delete; 104 105 session->file = file; 106 107 if (perf_data_file__is_read(file)) { 108 if (perf_session__open(session) < 0) 109 goto out_close; 110 111 perf_session__set_id_hdr_size(session); 112 perf_session__set_comm_exec(session); 113 } 114 } 115 116 if (!file || perf_data_file__is_write(file)) { 117 /* 118 * In O_RDONLY mode this will be performed when reading the 119 * kernel MMAP event, in perf_event__process_mmap(). 120 */ 121 if (perf_session__create_kernel_maps(session) < 0) 122 pr_warning("Cannot read kernel map\n"); 123 } 124 125 if (tool && tool->ordering_requires_timestamps && 126 tool->ordered_events && !perf_evlist__sample_id_all(session->evlist)) { 127 dump_printf("WARNING: No sample_id_all support, falling back to unordered processing\n"); 128 tool->ordered_events = false; 129 } 130 131 return session; 132 133 out_close: 134 perf_data_file__close(file); 135 out_delete: 136 perf_session__delete(session); 137 out: 138 return NULL; 139 } 140 141 static void perf_session__delete_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, 512 union perf_event *event __maybe_unused, 513 struct perf_session *session) 514 { 515 struct ordered_events *oe = &session->ordered_events; 516 struct perf_evlist *evlist = session->evlist; 517 struct machines *machines = &session->machines; 518 519 return ordered_events__flush(oe, machines, evlist, tool, OE_FLUSH__ROUND); 520 } 521 522 int perf_session_queue_event(struct perf_session *s, union perf_event *event, 523 struct perf_tool *tool, struct perf_sample *sample, 524 u64 file_offset) 525 { 526 struct ordered_events *oe = &s->ordered_events; 527 struct perf_evlist *evlist = s->evlist; 528 struct machines *machines = &s->machines; 529 530 u64 timestamp = sample->time; 531 struct ordered_event *new; 532 533 if (!timestamp || timestamp == ~0ULL) 534 return -ETIME; 535 536 if (timestamp < oe->last_flush) { 537 pr_oe_time(timestamp, "out of order event\n"); 538 pr_oe_time(oe->last_flush, "last flush, last_flush_type %d\n", 539 oe->last_flush_type); 540 541 s->evlist->stats.nr_unordered_events++; 542 } 543 544 new = ordered_events__new(oe, timestamp, event); 545 if (!new) { 546 ordered_events__flush(oe, machines, evlist, tool, OE_FLUSH__HALF); 547 new = ordered_events__new(oe, timestamp, event); 548 } 549 550 if (!new) 551 return -ENOMEM; 552 553 new->file_offset = file_offset; 554 return 0; 555 } 556 557 static void callchain__lbr_callstack_printf(struct perf_sample *sample) 558 { 559 struct ip_callchain *callchain = sample->callchain; 560 struct branch_stack *lbr_stack = sample->branch_stack; 561 u64 kernel_callchain_nr = callchain->nr; 562 unsigned int i; 563 564 for (i = 0; i < kernel_callchain_nr; i++) { 565 if (callchain->ips[i] == PERF_CONTEXT_USER) 566 break; 567 } 568 569 if ((i != kernel_callchain_nr) && lbr_stack->nr) { 570 u64 total_nr; 571 /* 572 * LBR callstack can only get user call chain, 573 * i is kernel call chain number, 574 * 1 is PERF_CONTEXT_USER. 575 * 576 * The user call chain is stored in LBR registers. 577 * LBR are pair registers. The caller is stored 578 * in "from" register, while the callee is stored 579 * in "to" register. 580 * For example, there is a call stack 581 * "A"->"B"->"C"->"D". 582 * The LBR registers will recorde like 583 * "C"->"D", "B"->"C", "A"->"B". 584 * So only the first "to" register and all "from" 585 * registers are needed to construct the whole stack. 586 */ 587 total_nr = i + 1 + lbr_stack->nr + 1; 588 kernel_callchain_nr = i + 1; 589 590 printf("... LBR call chain: nr:%" PRIu64 "\n", total_nr); 591 592 for (i = 0; i < kernel_callchain_nr; i++) 593 printf("..... %2d: %016" PRIx64 "\n", 594 i, callchain->ips[i]); 595 596 printf("..... %2d: %016" PRIx64 "\n", 597 (int)(kernel_callchain_nr), lbr_stack->entries[0].to); 598 for (i = 0; i < lbr_stack->nr; i++) 599 printf("..... %2d: %016" PRIx64 "\n", 600 (int)(i + kernel_callchain_nr + 1), lbr_stack->entries[i].from); 601 } 602 } 603 604 static void callchain__printf(struct perf_evsel *evsel, 605 struct perf_sample *sample) 606 { 607 unsigned int i; 608 struct ip_callchain *callchain = sample->callchain; 609 610 if (has_branch_callstack(evsel)) 611 callchain__lbr_callstack_printf(sample); 612 613 printf("... FP chain: nr:%" PRIu64 "\n", callchain->nr); 614 615 for (i = 0; i < callchain->nr; i++) 616 printf("..... %2d: %016" PRIx64 "\n", 617 i, callchain->ips[i]); 618 } 619 620 static void branch_stack__printf(struct perf_sample *sample) 621 { 622 uint64_t i; 623 624 printf("... branch stack: nr:%" PRIu64 "\n", sample->branch_stack->nr); 625 626 for (i = 0; i < sample->branch_stack->nr; i++) 627 printf("..... %2"PRIu64": %016" PRIx64 " -> %016" PRIx64 "\n", 628 i, sample->branch_stack->entries[i].from, 629 sample->branch_stack->entries[i].to); 630 } 631 632 static void regs_dump__printf(u64 mask, u64 *regs) 633 { 634 unsigned rid, i = 0; 635 636 for_each_set_bit(rid, (unsigned long *) &mask, sizeof(mask) * 8) { 637 u64 val = regs[i++]; 638 639 printf(".... %-5s 0x%" PRIx64 "\n", 640 perf_reg_name(rid), val); 641 } 642 } 643 644 static const char *regs_abi[] = { 645 [PERF_SAMPLE_REGS_ABI_NONE] = "none", 646 [PERF_SAMPLE_REGS_ABI_32] = "32-bit", 647 [PERF_SAMPLE_REGS_ABI_64] = "64-bit", 648 }; 649 650 static inline const char *regs_dump_abi(struct regs_dump *d) 651 { 652 if (d->abi > PERF_SAMPLE_REGS_ABI_64) 653 return "unknown"; 654 655 return regs_abi[d->abi]; 656 } 657 658 static void regs__printf(const char *type, struct regs_dump *regs) 659 { 660 u64 mask = regs->mask; 661 662 printf("... %s regs: mask 0x%" PRIx64 " ABI %s\n", 663 type, 664 mask, 665 regs_dump_abi(regs)); 666 667 regs_dump__printf(mask, regs->regs); 668 } 669 670 static void regs_user__printf(struct perf_sample *sample) 671 { 672 struct regs_dump *user_regs = &sample->user_regs; 673 674 if (user_regs->regs) 675 regs__printf("user", user_regs); 676 } 677 678 static void regs_intr__printf(struct perf_sample *sample) 679 { 680 struct regs_dump *intr_regs = &sample->intr_regs; 681 682 if (intr_regs->regs) 683 regs__printf("intr", intr_regs); 684 } 685 686 static void stack_user__printf(struct stack_dump *dump) 687 { 688 printf("... ustack: size %" PRIu64 ", offset 0x%x\n", 689 dump->size, dump->offset); 690 } 691 692 static void perf_evlist__print_tstamp(struct perf_evlist *evlist, 693 union perf_event *event, 694 struct perf_sample *sample) 695 { 696 u64 sample_type = __perf_evlist__combined_sample_type(evlist); 697 698 if (event->header.type != PERF_RECORD_SAMPLE && 699 !perf_evlist__sample_id_all(evlist)) { 700 fputs("-1 -1 ", stdout); 701 return; 702 } 703 704 if ((sample_type & PERF_SAMPLE_CPU)) 705 printf("%u ", sample->cpu); 706 707 if (sample_type & PERF_SAMPLE_TIME) 708 printf("%" PRIu64 " ", sample->time); 709 } 710 711 static void sample_read__printf(struct perf_sample *sample, u64 read_format) 712 { 713 printf("... sample_read:\n"); 714 715 if (read_format & PERF_FORMAT_TOTAL_TIME_ENABLED) 716 printf("...... time enabled %016" PRIx64 "\n", 717 sample->read.time_enabled); 718 719 if (read_format & PERF_FORMAT_TOTAL_TIME_RUNNING) 720 printf("...... time running %016" PRIx64 "\n", 721 sample->read.time_running); 722 723 if (read_format & PERF_FORMAT_GROUP) { 724 u64 i; 725 726 printf(".... group nr %" PRIu64 "\n", sample->read.group.nr); 727 728 for (i = 0; i < sample->read.group.nr; i++) { 729 struct sample_read_value *value; 730 731 value = &sample->read.group.values[i]; 732 printf("..... id %016" PRIx64 733 ", value %016" PRIx64 "\n", 734 value->id, value->value); 735 } 736 } else 737 printf("..... id %016" PRIx64 ", value %016" PRIx64 "\n", 738 sample->read.one.id, sample->read.one.value); 739 } 740 741 static void dump_event(struct perf_evlist *evlist, union perf_event *event, 742 u64 file_offset, struct perf_sample *sample) 743 { 744 if (!dump_trace) 745 return; 746 747 printf("\n%#" PRIx64 " [%#x]: event: %d\n", 748 file_offset, event->header.size, event->header.type); 749 750 trace_event(event); 751 752 if (sample) 753 perf_evlist__print_tstamp(evlist, event, sample); 754 755 printf("%#" PRIx64 " [%#x]: PERF_RECORD_%s", file_offset, 756 event->header.size, perf_event__name(event->header.type)); 757 } 758 759 static void dump_sample(struct perf_evsel *evsel, union perf_event *event, 760 struct perf_sample *sample) 761 { 762 u64 sample_type; 763 764 if (!dump_trace) 765 return; 766 767 printf("(IP, 0x%x): %d/%d: %#" PRIx64 " period: %" PRIu64 " addr: %#" PRIx64 "\n", 768 event->header.misc, sample->pid, sample->tid, sample->ip, 769 sample->period, sample->addr); 770 771 sample_type = evsel->attr.sample_type; 772 773 if (sample_type & PERF_SAMPLE_CALLCHAIN) 774 callchain__printf(evsel, sample); 775 776 if ((sample_type & PERF_SAMPLE_BRANCH_STACK) && !has_branch_callstack(evsel)) 777 branch_stack__printf(sample); 778 779 if (sample_type & PERF_SAMPLE_REGS_USER) 780 regs_user__printf(sample); 781 782 if (sample_type & PERF_SAMPLE_REGS_INTR) 783 regs_intr__printf(sample); 784 785 if (sample_type & PERF_SAMPLE_STACK_USER) 786 stack_user__printf(&sample->user_stack); 787 788 if (sample_type & PERF_SAMPLE_WEIGHT) 789 printf("... weight: %" PRIu64 "\n", sample->weight); 790 791 if (sample_type & PERF_SAMPLE_DATA_SRC) 792 printf(" . data_src: 0x%"PRIx64"\n", sample->data_src); 793 794 if (sample_type & PERF_SAMPLE_TRANSACTION) 795 printf("... transaction: %" PRIx64 "\n", sample->transaction); 796 797 if (sample_type & PERF_SAMPLE_READ) 798 sample_read__printf(sample, evsel->attr.read_format); 799 } 800 801 static struct machine *machines__find_for_cpumode(struct machines *machines, 802 union perf_event *event, 803 struct perf_sample *sample) 804 { 805 const u8 cpumode = event->header.misc & PERF_RECORD_MISC_CPUMODE_MASK; 806 struct machine *machine; 807 808 if (perf_guest && 809 ((cpumode == PERF_RECORD_MISC_GUEST_KERNEL) || 810 (cpumode == PERF_RECORD_MISC_GUEST_USER))) { 811 u32 pid; 812 813 if (event->header.type == PERF_RECORD_MMAP 814 || event->header.type == PERF_RECORD_MMAP2) 815 pid = event->mmap.pid; 816 else 817 pid = sample->pid; 818 819 machine = machines__find(machines, pid); 820 if (!machine) 821 machine = machines__find(machines, DEFAULT_GUEST_KERNEL_ID); 822 return machine; 823 } 824 825 return &machines->host; 826 } 827 828 static int deliver_sample_value(struct perf_evlist *evlist, 829 struct perf_tool *tool, 830 union perf_event *event, 831 struct perf_sample *sample, 832 struct sample_read_value *v, 833 struct machine *machine) 834 { 835 struct perf_sample_id *sid = perf_evlist__id2sid(evlist, v->id); 836 837 if (sid) { 838 sample->id = v->id; 839 sample->period = v->value - sid->period; 840 sid->period = v->value; 841 } 842 843 if (!sid || sid->evsel == NULL) { 844 ++evlist->stats.nr_unknown_id; 845 return 0; 846 } 847 848 return tool->sample(tool, event, sample, sid->evsel, machine); 849 } 850 851 static int deliver_sample_group(struct perf_evlist *evlist, 852 struct perf_tool *tool, 853 union perf_event *event, 854 struct perf_sample *sample, 855 struct machine *machine) 856 { 857 int ret = -EINVAL; 858 u64 i; 859 860 for (i = 0; i < sample->read.group.nr; i++) { 861 ret = deliver_sample_value(evlist, tool, event, sample, 862 &sample->read.group.values[i], 863 machine); 864 if (ret) 865 break; 866 } 867 868 return ret; 869 } 870 871 static int 872 perf_evlist__deliver_sample(struct perf_evlist *evlist, 873 struct perf_tool *tool, 874 union perf_event *event, 875 struct perf_sample *sample, 876 struct perf_evsel *evsel, 877 struct machine *machine) 878 { 879 /* We know evsel != NULL. */ 880 u64 sample_type = evsel->attr.sample_type; 881 u64 read_format = evsel->attr.read_format; 882 883 /* Standard sample delievery. */ 884 if (!(sample_type & PERF_SAMPLE_READ)) 885 return tool->sample(tool, event, sample, evsel, machine); 886 887 /* For PERF_SAMPLE_READ we have either single or group mode. */ 888 if (read_format & PERF_FORMAT_GROUP) 889 return deliver_sample_group(evlist, tool, event, sample, 890 machine); 891 else 892 return deliver_sample_value(evlist, tool, event, sample, 893 &sample->read.one, machine); 894 } 895 896 int machines__deliver_event(struct machines *machines, 897 struct perf_evlist *evlist, 898 union perf_event *event, 899 struct perf_sample *sample, 900 struct perf_tool *tool, u64 file_offset) 901 { 902 struct perf_evsel *evsel; 903 struct machine *machine; 904 905 dump_event(evlist, event, file_offset, sample); 906 907 evsel = perf_evlist__id2evsel(evlist, sample->id); 908 909 machine = machines__find_for_cpumode(machines, event, sample); 910 911 switch (event->header.type) { 912 case PERF_RECORD_SAMPLE: 913 dump_sample(evsel, event, sample); 914 if (evsel == NULL) { 915 ++evlist->stats.nr_unknown_id; 916 return 0; 917 } 918 if (machine == NULL) { 919 ++evlist->stats.nr_unprocessable_samples; 920 return 0; 921 } 922 return perf_evlist__deliver_sample(evlist, tool, event, sample, evsel, machine); 923 case PERF_RECORD_MMAP: 924 return tool->mmap(tool, event, sample, machine); 925 case PERF_RECORD_MMAP2: 926 return tool->mmap2(tool, event, sample, machine); 927 case PERF_RECORD_COMM: 928 return tool->comm(tool, event, sample, machine); 929 case PERF_RECORD_FORK: 930 return tool->fork(tool, event, sample, machine); 931 case PERF_RECORD_EXIT: 932 return tool->exit(tool, event, sample, machine); 933 case PERF_RECORD_LOST: 934 if (tool->lost == perf_event__process_lost) 935 evlist->stats.total_lost += event->lost.lost; 936 return tool->lost(tool, event, sample, machine); 937 case PERF_RECORD_READ: 938 return tool->read(tool, event, sample, evsel, machine); 939 case PERF_RECORD_THROTTLE: 940 return tool->throttle(tool, event, sample, machine); 941 case PERF_RECORD_UNTHROTTLE: 942 return tool->unthrottle(tool, event, sample, machine); 943 default: 944 ++evlist->stats.nr_unknown_events; 945 return -1; 946 } 947 } 948 949 static s64 perf_session__process_user_event(struct perf_session *session, 950 union perf_event *event, 951 struct perf_tool *tool, 952 u64 file_offset) 953 { 954 int fd = perf_data_file__fd(session->file); 955 int err; 956 957 dump_event(session->evlist, event, file_offset, NULL); 958 959 /* These events are processed right away */ 960 switch (event->header.type) { 961 case PERF_RECORD_HEADER_ATTR: 962 err = tool->attr(tool, event, &session->evlist); 963 if (err == 0) { 964 perf_session__set_id_hdr_size(session); 965 perf_session__set_comm_exec(session); 966 } 967 return err; 968 case PERF_RECORD_HEADER_EVENT_TYPE: 969 /* 970 * Depreceated, but we need to handle it for sake 971 * of old data files create in pipe mode. 972 */ 973 return 0; 974 case PERF_RECORD_HEADER_TRACING_DATA: 975 /* setup for reading amidst mmap */ 976 lseek(fd, file_offset, SEEK_SET); 977 return tool->tracing_data(tool, event, session); 978 case PERF_RECORD_HEADER_BUILD_ID: 979 return tool->build_id(tool, event, session); 980 case PERF_RECORD_FINISHED_ROUND: 981 return tool->finished_round(tool, event, session); 982 case PERF_RECORD_ID_INDEX: 983 return tool->id_index(tool, event, session); 984 default: 985 return -EINVAL; 986 } 987 } 988 989 int perf_session__deliver_synth_event(struct perf_session *session, 990 union perf_event *event, 991 struct perf_sample *sample, 992 struct perf_tool *tool) 993 { 994 struct perf_evlist *evlist = session->evlist; 995 996 events_stats__inc(&evlist->stats, event->header.type); 997 998 if (event->header.type >= PERF_RECORD_USER_TYPE_START) 999 return perf_session__process_user_event(session, event, tool, 0); 1000 1001 return machines__deliver_event(&session->machines, evlist, event, sample, tool, 0); 1002 } 1003 1004 static void event_swap(union perf_event *event, bool sample_id_all) 1005 { 1006 perf_event__swap_op swap; 1007 1008 swap = perf_event__swap_ops[event->header.type]; 1009 if (swap) 1010 swap(event, sample_id_all); 1011 } 1012 1013 int perf_session__peek_event(struct perf_session *session, off_t file_offset, 1014 void *buf, size_t buf_sz, 1015 union perf_event **event_ptr, 1016 struct perf_sample *sample) 1017 { 1018 union perf_event *event; 1019 size_t hdr_sz, rest; 1020 int fd; 1021 1022 if (session->one_mmap && !session->header.needs_swap) { 1023 event = file_offset - session->one_mmap_offset + 1024 session->one_mmap_addr; 1025 goto out_parse_sample; 1026 } 1027 1028 if (perf_data_file__is_pipe(session->file)) 1029 return -1; 1030 1031 fd = perf_data_file__fd(session->file); 1032 hdr_sz = sizeof(struct perf_event_header); 1033 1034 if (buf_sz < hdr_sz) 1035 return -1; 1036 1037 if (lseek(fd, file_offset, SEEK_SET) == (off_t)-1 || 1038 readn(fd, &buf, hdr_sz) != (ssize_t)hdr_sz) 1039 return -1; 1040 1041 event = (union perf_event *)buf; 1042 1043 if (session->header.needs_swap) 1044 perf_event_header__bswap(&event->header); 1045 1046 if (event->header.size < hdr_sz) 1047 return -1; 1048 1049 rest = event->header.size - hdr_sz; 1050 1051 if (readn(fd, &buf, rest) != (ssize_t)rest) 1052 return -1; 1053 1054 if (session->header.needs_swap) 1055 event_swap(event, perf_evlist__sample_id_all(session->evlist)); 1056 1057 out_parse_sample: 1058 1059 if (sample && event->header.type < PERF_RECORD_USER_TYPE_START && 1060 perf_evlist__parse_sample(session->evlist, event, sample)) 1061 return -1; 1062 1063 *event_ptr = event; 1064 1065 return 0; 1066 } 1067 1068 static s64 perf_session__process_event(struct perf_session *session, 1069 union perf_event *event, 1070 struct perf_tool *tool, 1071 u64 file_offset) 1072 { 1073 struct perf_evlist *evlist = session->evlist; 1074 struct perf_sample sample; 1075 int ret; 1076 1077 if (session->header.needs_swap) 1078 event_swap(event, perf_evlist__sample_id_all(evlist)); 1079 1080 if (event->header.type >= PERF_RECORD_HEADER_MAX) 1081 return -EINVAL; 1082 1083 events_stats__inc(&evlist->stats, event->header.type); 1084 1085 if (event->header.type >= PERF_RECORD_USER_TYPE_START) 1086 return perf_session__process_user_event(session, event, tool, file_offset); 1087 1088 /* 1089 * For all kernel events we get the sample data 1090 */ 1091 ret = perf_evlist__parse_sample(evlist, event, &sample); 1092 if (ret) 1093 return ret; 1094 1095 if (tool->ordered_events) { 1096 ret = perf_session_queue_event(session, event, tool, &sample, 1097 file_offset); 1098 if (ret != -ETIME) 1099 return ret; 1100 } 1101 1102 return machines__deliver_event(&session->machines, evlist, event, 1103 &sample, tool, file_offset); 1104 } 1105 1106 void perf_event_header__bswap(struct perf_event_header *hdr) 1107 { 1108 hdr->type = bswap_32(hdr->type); 1109 hdr->misc = bswap_16(hdr->misc); 1110 hdr->size = bswap_16(hdr->size); 1111 } 1112 1113 struct thread *perf_session__findnew(struct perf_session *session, pid_t pid) 1114 { 1115 return machine__findnew_thread(&session->machines.host, -1, pid); 1116 } 1117 1118 static struct thread *perf_session__register_idle_thread(struct perf_session *session) 1119 { 1120 struct thread *thread; 1121 1122 thread = machine__findnew_thread(&session->machines.host, 0, 0); 1123 if (thread == NULL || thread__set_comm(thread, "swapper", 0)) { 1124 pr_err("problem inserting idle task.\n"); 1125 thread = NULL; 1126 } 1127 1128 return thread; 1129 } 1130 1131 static void perf_tool__warn_about_errors(const struct perf_tool *tool, 1132 const struct events_stats *stats) 1133 { 1134 if (tool->lost == perf_event__process_lost && 1135 stats->nr_events[PERF_RECORD_LOST] != 0) { 1136 ui__warning("Processed %d events and lost %d chunks!\n\n" 1137 "Check IO/CPU overload!\n\n", 1138 stats->nr_events[0], 1139 stats->nr_events[PERF_RECORD_LOST]); 1140 } 1141 1142 if (stats->nr_unknown_events != 0) { 1143 ui__warning("Found %u unknown events!\n\n" 1144 "Is this an older tool processing a perf.data " 1145 "file generated by a more recent tool?\n\n" 1146 "If that is not the case, consider " 1147 "reporting to linux-kernel@vger.kernel.org.\n\n", 1148 stats->nr_unknown_events); 1149 } 1150 1151 if (stats->nr_unknown_id != 0) { 1152 ui__warning("%u samples with id not present in the header\n", 1153 stats->nr_unknown_id); 1154 } 1155 1156 if (stats->nr_invalid_chains != 0) { 1157 ui__warning("Found invalid callchains!\n\n" 1158 "%u out of %u events were discarded for this reason.\n\n" 1159 "Consider reporting to linux-kernel@vger.kernel.org.\n\n", 1160 stats->nr_invalid_chains, 1161 stats->nr_events[PERF_RECORD_SAMPLE]); 1162 } 1163 1164 if (stats->nr_unprocessable_samples != 0) { 1165 ui__warning("%u unprocessable samples recorded.\n" 1166 "Do you have a KVM guest running and not using 'perf kvm'?\n", 1167 stats->nr_unprocessable_samples); 1168 } 1169 1170 if (stats->nr_unordered_events != 0) 1171 ui__warning("%u out of order events recorded.\n", stats->nr_unordered_events); 1172 } 1173 1174 volatile int session_done; 1175 1176 static int __perf_session__process_pipe_events(struct perf_session *session, 1177 struct perf_tool *tool) 1178 { 1179 struct ordered_events *oe = &session->ordered_events; 1180 struct perf_evlist *evlist = session->evlist; 1181 struct machines *machines = &session->machines; 1182 int fd = perf_data_file__fd(session->file); 1183 union perf_event *event; 1184 uint32_t size, cur_size = 0; 1185 void *buf = NULL; 1186 s64 skip = 0; 1187 u64 head; 1188 ssize_t err; 1189 void *p; 1190 1191 perf_tool__fill_defaults(tool); 1192 1193 head = 0; 1194 cur_size = sizeof(union perf_event); 1195 1196 buf = malloc(cur_size); 1197 if (!buf) 1198 return -errno; 1199 more: 1200 event = buf; 1201 err = readn(fd, event, sizeof(struct perf_event_header)); 1202 if (err <= 0) { 1203 if (err == 0) 1204 goto done; 1205 1206 pr_err("failed to read event header\n"); 1207 goto out_err; 1208 } 1209 1210 if (session->header.needs_swap) 1211 perf_event_header__bswap(&event->header); 1212 1213 size = event->header.size; 1214 if (size < sizeof(struct perf_event_header)) { 1215 pr_err("bad event header size\n"); 1216 goto out_err; 1217 } 1218 1219 if (size > cur_size) { 1220 void *new = realloc(buf, size); 1221 if (!new) { 1222 pr_err("failed to allocate memory to read event\n"); 1223 goto out_err; 1224 } 1225 buf = new; 1226 cur_size = size; 1227 event = buf; 1228 } 1229 p = event; 1230 p += sizeof(struct perf_event_header); 1231 1232 if (size - sizeof(struct perf_event_header)) { 1233 err = readn(fd, p, size - sizeof(struct perf_event_header)); 1234 if (err <= 0) { 1235 if (err == 0) { 1236 pr_err("unexpected end of event stream\n"); 1237 goto done; 1238 } 1239 1240 pr_err("failed to read event data\n"); 1241 goto out_err; 1242 } 1243 } 1244 1245 if ((skip = perf_session__process_event(session, event, tool, head)) < 0) { 1246 pr_err("%#" PRIx64 " [%#x]: failed to process type: %d\n", 1247 head, event->header.size, event->header.type); 1248 err = -EINVAL; 1249 goto out_err; 1250 } 1251 1252 head += size; 1253 1254 if (skip > 0) 1255 head += skip; 1256 1257 if (!session_done()) 1258 goto more; 1259 done: 1260 /* do the final flush for ordered samples */ 1261 err = ordered_events__flush(oe, machines, evlist, tool, OE_FLUSH__FINAL); 1262 out_err: 1263 free(buf); 1264 perf_tool__warn_about_errors(tool, &session->evlist->stats); 1265 ordered_events__free(&session->ordered_events); 1266 return err; 1267 } 1268 1269 static union perf_event * 1270 fetch_mmaped_event(struct perf_session *session, 1271 u64 head, size_t mmap_size, char *buf) 1272 { 1273 union perf_event *event; 1274 1275 /* 1276 * Ensure we have enough space remaining to read 1277 * the size of the event in the headers. 1278 */ 1279 if (head + sizeof(event->header) > mmap_size) 1280 return NULL; 1281 1282 event = (union perf_event *)(buf + head); 1283 1284 if (session->header.needs_swap) 1285 perf_event_header__bswap(&event->header); 1286 1287 if (head + event->header.size > mmap_size) { 1288 /* We're not fetching the event so swap back again */ 1289 if (session->header.needs_swap) 1290 perf_event_header__bswap(&event->header); 1291 return NULL; 1292 } 1293 1294 return event; 1295 } 1296 1297 /* 1298 * On 64bit we can mmap the data file in one go. No need for tiny mmap 1299 * slices. On 32bit we use 32MB. 1300 */ 1301 #if BITS_PER_LONG == 64 1302 #define MMAP_SIZE ULLONG_MAX 1303 #define NUM_MMAPS 1 1304 #else 1305 #define MMAP_SIZE (32 * 1024 * 1024ULL) 1306 #define NUM_MMAPS 128 1307 #endif 1308 1309 static int __perf_session__process_events(struct perf_session *session, 1310 u64 data_offset, u64 data_size, 1311 u64 file_size, struct perf_tool *tool) 1312 { 1313 struct ordered_events *oe = &session->ordered_events; 1314 struct perf_evlist *evlist = session->evlist; 1315 struct machines *machines = &session->machines; 1316 int fd = perf_data_file__fd(session->file); 1317 u64 head, page_offset, file_offset, file_pos, size; 1318 int err, mmap_prot, mmap_flags, map_idx = 0; 1319 size_t mmap_size; 1320 char *buf, *mmaps[NUM_MMAPS]; 1321 union perf_event *event; 1322 struct ui_progress prog; 1323 s64 skip; 1324 1325 perf_tool__fill_defaults(tool); 1326 1327 page_offset = page_size * (data_offset / page_size); 1328 file_offset = page_offset; 1329 head = data_offset - page_offset; 1330 1331 if (data_size && (data_offset + data_size < file_size)) 1332 file_size = data_offset + data_size; 1333 1334 ui_progress__init(&prog, file_size, "Processing events..."); 1335 1336 mmap_size = MMAP_SIZE; 1337 if (mmap_size > file_size) { 1338 mmap_size = file_size; 1339 session->one_mmap = true; 1340 } 1341 1342 memset(mmaps, 0, sizeof(mmaps)); 1343 1344 mmap_prot = PROT_READ; 1345 mmap_flags = MAP_SHARED; 1346 1347 if (session->header.needs_swap) { 1348 mmap_prot |= PROT_WRITE; 1349 mmap_flags = MAP_PRIVATE; 1350 } 1351 remap: 1352 buf = mmap(NULL, mmap_size, mmap_prot, mmap_flags, fd, 1353 file_offset); 1354 if (buf == MAP_FAILED) { 1355 pr_err("failed to mmap file\n"); 1356 err = -errno; 1357 goto out_err; 1358 } 1359 mmaps[map_idx] = buf; 1360 map_idx = (map_idx + 1) & (ARRAY_SIZE(mmaps) - 1); 1361 file_pos = file_offset + head; 1362 if (session->one_mmap) { 1363 session->one_mmap_addr = buf; 1364 session->one_mmap_offset = file_offset; 1365 } 1366 1367 more: 1368 event = fetch_mmaped_event(session, head, mmap_size, buf); 1369 if (!event) { 1370 if (mmaps[map_idx]) { 1371 munmap(mmaps[map_idx], mmap_size); 1372 mmaps[map_idx] = NULL; 1373 } 1374 1375 page_offset = page_size * (head / page_size); 1376 file_offset += page_offset; 1377 head -= page_offset; 1378 goto remap; 1379 } 1380 1381 size = event->header.size; 1382 1383 if (size < sizeof(struct perf_event_header) || 1384 (skip = perf_session__process_event(session, event, tool, file_pos)) 1385 < 0) { 1386 pr_err("%#" PRIx64 " [%#x]: failed to process type: %d\n", 1387 file_offset + head, event->header.size, 1388 event->header.type); 1389 err = -EINVAL; 1390 goto out_err; 1391 } 1392 1393 if (skip) 1394 size += skip; 1395 1396 head += size; 1397 file_pos += size; 1398 1399 ui_progress__update(&prog, size); 1400 1401 if (session_done()) 1402 goto out; 1403 1404 if (file_pos < file_size) 1405 goto more; 1406 1407 out: 1408 /* do the final flush for ordered samples */ 1409 err = ordered_events__flush(oe, machines, evlist, tool, OE_FLUSH__FINAL); 1410 out_err: 1411 ui_progress__finish(); 1412 perf_tool__warn_about_errors(tool, &session->evlist->stats); 1413 ordered_events__free(&session->ordered_events); 1414 session->one_mmap = false; 1415 return err; 1416 } 1417 1418 int perf_session__process_events(struct perf_session *session, 1419 struct perf_tool *tool) 1420 { 1421 u64 size = perf_data_file__size(session->file); 1422 int err; 1423 1424 if (perf_session__register_idle_thread(session) == NULL) 1425 return -ENOMEM; 1426 1427 if (!perf_data_file__is_pipe(session->file)) 1428 err = __perf_session__process_events(session, 1429 session->header.data_offset, 1430 session->header.data_size, 1431 size, tool); 1432 else 1433 err = __perf_session__process_pipe_events(session, tool); 1434 1435 return err; 1436 } 1437 1438 bool perf_session__has_traces(struct perf_session *session, const char *msg) 1439 { 1440 struct perf_evsel *evsel; 1441 1442 evlist__for_each(session->evlist, evsel) { 1443 if (evsel->attr.type == PERF_TYPE_TRACEPOINT) 1444 return true; 1445 } 1446 1447 pr_err("No trace sample to read. Did you call 'perf %s'?\n", msg); 1448 return false; 1449 } 1450 1451 int maps__set_kallsyms_ref_reloc_sym(struct map **maps, 1452 const char *symbol_name, u64 addr) 1453 { 1454 char *bracket; 1455 enum map_type i; 1456 struct ref_reloc_sym *ref; 1457 1458 ref = zalloc(sizeof(struct ref_reloc_sym)); 1459 if (ref == NULL) 1460 return -ENOMEM; 1461 1462 ref->name = strdup(symbol_name); 1463 if (ref->name == NULL) { 1464 free(ref); 1465 return -ENOMEM; 1466 } 1467 1468 bracket = strchr(ref->name, ']'); 1469 if (bracket) 1470 *bracket = '\0'; 1471 1472 ref->addr = addr; 1473 1474 for (i = 0; i < MAP__NR_TYPES; ++i) { 1475 struct kmap *kmap = map__kmap(maps[i]); 1476 kmap->ref_reloc_sym = ref; 1477 } 1478 1479 return 0; 1480 } 1481 1482 size_t perf_session__fprintf_dsos(struct perf_session *session, FILE *fp) 1483 { 1484 return machines__fprintf_dsos(&session->machines, fp); 1485 } 1486 1487 size_t perf_session__fprintf_dsos_buildid(struct perf_session *session, FILE *fp, 1488 bool (skip)(struct dso *dso, int parm), int parm) 1489 { 1490 return machines__fprintf_dsos_buildid(&session->machines, fp, skip, parm); 1491 } 1492 1493 size_t perf_session__fprintf_nr_events(struct perf_session *session, FILE *fp) 1494 { 1495 size_t ret = fprintf(fp, "Aggregated stats:\n"); 1496 1497 ret += events_stats__fprintf(&session->evlist->stats, fp); 1498 return ret; 1499 } 1500 1501 size_t perf_session__fprintf(struct perf_session *session, FILE *fp) 1502 { 1503 /* 1504 * FIXME: Here we have to actually print all the machines in this 1505 * session, not just the host... 1506 */ 1507 return machine__fprintf(&session->machines.host, fp); 1508 } 1509 1510 struct perf_evsel *perf_session__find_first_evtype(struct perf_session *session, 1511 unsigned int type) 1512 { 1513 struct perf_evsel *pos; 1514 1515 evlist__for_each(session->evlist, pos) { 1516 if (pos->attr.type == type) 1517 return pos; 1518 } 1519 return NULL; 1520 } 1521 1522 void perf_evsel__print_ip(struct perf_evsel *evsel, struct perf_sample *sample, 1523 struct addr_location *al, 1524 unsigned int print_opts, unsigned int stack_depth) 1525 { 1526 struct callchain_cursor_node *node; 1527 int print_ip = print_opts & PRINT_IP_OPT_IP; 1528 int print_sym = print_opts & PRINT_IP_OPT_SYM; 1529 int print_dso = print_opts & PRINT_IP_OPT_DSO; 1530 int print_symoffset = print_opts & PRINT_IP_OPT_SYMOFFSET; 1531 int print_oneline = print_opts & PRINT_IP_OPT_ONELINE; 1532 int print_srcline = print_opts & PRINT_IP_OPT_SRCLINE; 1533 char s = print_oneline ? ' ' : '\t'; 1534 1535 if (symbol_conf.use_callchain && sample->callchain) { 1536 struct addr_location node_al; 1537 1538 if (thread__resolve_callchain(al->thread, evsel, 1539 sample, NULL, NULL, 1540 PERF_MAX_STACK_DEPTH) != 0) { 1541 if (verbose) 1542 error("Failed to resolve callchain. Skipping\n"); 1543 return; 1544 } 1545 callchain_cursor_commit(&callchain_cursor); 1546 1547 if (print_symoffset) 1548 node_al = *al; 1549 1550 while (stack_depth) { 1551 u64 addr = 0; 1552 1553 node = callchain_cursor_current(&callchain_cursor); 1554 if (!node) 1555 break; 1556 1557 if (node->sym && node->sym->ignore) 1558 goto next; 1559 1560 if (print_ip) 1561 printf("%c%16" PRIx64, s, node->ip); 1562 1563 if (node->map) 1564 addr = node->map->map_ip(node->map, node->ip); 1565 1566 if (print_sym) { 1567 printf(" "); 1568 if (print_symoffset) { 1569 node_al.addr = addr; 1570 node_al.map = node->map; 1571 symbol__fprintf_symname_offs(node->sym, &node_al, stdout); 1572 } else 1573 symbol__fprintf_symname(node->sym, stdout); 1574 } 1575 1576 if (print_dso) { 1577 printf(" ("); 1578 map__fprintf_dsoname(node->map, stdout); 1579 printf(")"); 1580 } 1581 1582 if (print_srcline) 1583 map__fprintf_srcline(node->map, addr, "\n ", 1584 stdout); 1585 1586 if (!print_oneline) 1587 printf("\n"); 1588 1589 stack_depth--; 1590 next: 1591 callchain_cursor_advance(&callchain_cursor); 1592 } 1593 1594 } else { 1595 if (al->sym && al->sym->ignore) 1596 return; 1597 1598 if (print_ip) 1599 printf("%16" PRIx64, sample->ip); 1600 1601 if (print_sym) { 1602 printf(" "); 1603 if (print_symoffset) 1604 symbol__fprintf_symname_offs(al->sym, al, 1605 stdout); 1606 else 1607 symbol__fprintf_symname(al->sym, stdout); 1608 } 1609 1610 if (print_dso) { 1611 printf(" ("); 1612 map__fprintf_dsoname(al->map, stdout); 1613 printf(")"); 1614 } 1615 1616 if (print_srcline) 1617 map__fprintf_srcline(al->map, al->addr, "\n ", stdout); 1618 } 1619 } 1620 1621 int perf_session__cpu_bitmap(struct perf_session *session, 1622 const char *cpu_list, unsigned long *cpu_bitmap) 1623 { 1624 int i, err = -1; 1625 struct cpu_map *map; 1626 1627 for (i = 0; i < PERF_TYPE_MAX; ++i) { 1628 struct perf_evsel *evsel; 1629 1630 evsel = perf_session__find_first_evtype(session, i); 1631 if (!evsel) 1632 continue; 1633 1634 if (!(evsel->attr.sample_type & PERF_SAMPLE_CPU)) { 1635 pr_err("File does not contain CPU events. " 1636 "Remove -c option to proceed.\n"); 1637 return -1; 1638 } 1639 } 1640 1641 map = cpu_map__new(cpu_list); 1642 if (map == NULL) { 1643 pr_err("Invalid cpu_list\n"); 1644 return -1; 1645 } 1646 1647 for (i = 0; i < map->nr; i++) { 1648 int cpu = map->map[i]; 1649 1650 if (cpu >= MAX_NR_CPUS) { 1651 pr_err("Requested CPU %d too large. " 1652 "Consider raising MAX_NR_CPUS\n", cpu); 1653 goto out_delete_map; 1654 } 1655 1656 set_bit(cpu, cpu_bitmap); 1657 } 1658 1659 err = 0; 1660 1661 out_delete_map: 1662 cpu_map__delete(map); 1663 return err; 1664 } 1665 1666 void perf_session__fprintf_info(struct perf_session *session, FILE *fp, 1667 bool full) 1668 { 1669 struct stat st; 1670 int fd, ret; 1671 1672 if (session == NULL || fp == NULL) 1673 return; 1674 1675 fd = perf_data_file__fd(session->file); 1676 1677 ret = fstat(fd, &st); 1678 if (ret == -1) 1679 return; 1680 1681 fprintf(fp, "# ========\n"); 1682 fprintf(fp, "# captured on: %s", ctime(&st.st_ctime)); 1683 perf_header__fprintf_info(session, fp, full); 1684 fprintf(fp, "# ========\n#\n"); 1685 } 1686 1687 1688 int __perf_session__set_tracepoints_handlers(struct perf_session *session, 1689 const struct perf_evsel_str_handler *assocs, 1690 size_t nr_assocs) 1691 { 1692 struct perf_evsel *evsel; 1693 size_t i; 1694 int err; 1695 1696 for (i = 0; i < nr_assocs; i++) { 1697 /* 1698 * Adding a handler for an event not in the session, 1699 * just ignore it. 1700 */ 1701 evsel = perf_evlist__find_tracepoint_by_name(session->evlist, assocs[i].name); 1702 if (evsel == NULL) 1703 continue; 1704 1705 err = -EEXIST; 1706 if (evsel->handler != NULL) 1707 goto out; 1708 evsel->handler = assocs[i].handler; 1709 } 1710 1711 err = 0; 1712 out: 1713 return err; 1714 } 1715 1716 int perf_event__process_id_index(struct perf_tool *tool __maybe_unused, 1717 union perf_event *event, 1718 struct perf_session *session) 1719 { 1720 struct perf_evlist *evlist = session->evlist; 1721 struct id_index_event *ie = &event->id_index; 1722 size_t i, nr, max_nr; 1723 1724 max_nr = (ie->header.size - sizeof(struct id_index_event)) / 1725 sizeof(struct id_index_entry); 1726 nr = ie->nr; 1727 if (nr > max_nr) 1728 return -EINVAL; 1729 1730 if (dump_trace) 1731 fprintf(stdout, " nr: %zu\n", nr); 1732 1733 for (i = 0; i < nr; i++) { 1734 struct id_index_entry *e = &ie->entries[i]; 1735 struct perf_sample_id *sid; 1736 1737 if (dump_trace) { 1738 fprintf(stdout, " ... id: %"PRIu64, e->id); 1739 fprintf(stdout, " idx: %"PRIu64, e->idx); 1740 fprintf(stdout, " cpu: %"PRId64, e->cpu); 1741 fprintf(stdout, " tid: %"PRId64"\n", e->tid); 1742 } 1743 1744 sid = perf_evlist__id2sid(evlist, e->id); 1745 if (!sid) 1746 return -ENOENT; 1747 sid->idx = e->idx; 1748 sid->cpu = e->cpu; 1749 sid->tid = e->tid; 1750 } 1751 return 0; 1752 } 1753 1754 int perf_event__synthesize_id_index(struct perf_tool *tool, 1755 perf_event__handler_t process, 1756 struct perf_evlist *evlist, 1757 struct machine *machine) 1758 { 1759 union perf_event *ev; 1760 struct perf_evsel *evsel; 1761 size_t nr = 0, i = 0, sz, max_nr, n; 1762 int err; 1763 1764 pr_debug2("Synthesizing id index\n"); 1765 1766 max_nr = (UINT16_MAX - sizeof(struct id_index_event)) / 1767 sizeof(struct id_index_entry); 1768 1769 evlist__for_each(evlist, evsel) 1770 nr += evsel->ids; 1771 1772 n = nr > max_nr ? max_nr : nr; 1773 sz = sizeof(struct id_index_event) + n * sizeof(struct id_index_entry); 1774 ev = zalloc(sz); 1775 if (!ev) 1776 return -ENOMEM; 1777 1778 ev->id_index.header.type = PERF_RECORD_ID_INDEX; 1779 ev->id_index.header.size = sz; 1780 ev->id_index.nr = n; 1781 1782 evlist__for_each(evlist, evsel) { 1783 u32 j; 1784 1785 for (j = 0; j < evsel->ids; j++) { 1786 struct id_index_entry *e; 1787 struct perf_sample_id *sid; 1788 1789 if (i >= n) { 1790 err = process(tool, ev, NULL, machine); 1791 if (err) 1792 goto out_err; 1793 nr -= n; 1794 i = 0; 1795 } 1796 1797 e = &ev->id_index.entries[i++]; 1798 1799 e->id = evsel->id[j]; 1800 1801 sid = perf_evlist__id2sid(evlist, e->id); 1802 if (!sid) { 1803 free(ev); 1804 return -ENOENT; 1805 } 1806 1807 e->idx = sid->idx; 1808 e->cpu = sid->cpu; 1809 e->tid = sid->tid; 1810 } 1811 } 1812 1813 sz = sizeof(struct id_index_event) + nr * sizeof(struct id_index_entry); 1814 ev->id_index.header.size = sz; 1815 ev->id_index.nr = nr; 1816 1817 err = process(tool, ev, NULL, machine); 1818 out_err: 1819 free(ev); 1820 1821 return err; 1822 } 1823