1 #define _FILE_OFFSET_BITS 64 2 3 #include <linux/kernel.h> 4 5 #include <byteswap.h> 6 #include <unistd.h> 7 #include <sys/types.h> 8 #include <sys/mman.h> 9 10 #include "evlist.h" 11 #include "evsel.h" 12 #include "session.h" 13 #include "tool.h" 14 #include "sort.h" 15 #include "util.h" 16 #include "cpumap.h" 17 #include "event-parse.h" 18 19 static int perf_session__open(struct perf_session *self, bool force) 20 { 21 struct stat input_stat; 22 23 if (!strcmp(self->filename, "-")) { 24 self->fd_pipe = true; 25 self->fd = STDIN_FILENO; 26 27 if (perf_session__read_header(self, self->fd) < 0) 28 pr_err("incompatible file format (rerun with -v to learn more)"); 29 30 return 0; 31 } 32 33 self->fd = open(self->filename, O_RDONLY); 34 if (self->fd < 0) { 35 int err = errno; 36 37 pr_err("failed to open %s: %s", self->filename, strerror(err)); 38 if (err == ENOENT && !strcmp(self->filename, "perf.data")) 39 pr_err(" (try 'perf record' first)"); 40 pr_err("\n"); 41 return -errno; 42 } 43 44 if (fstat(self->fd, &input_stat) < 0) 45 goto out_close; 46 47 if (!force && input_stat.st_uid && (input_stat.st_uid != geteuid())) { 48 pr_err("file %s not owned by current user or root\n", 49 self->filename); 50 goto out_close; 51 } 52 53 if (!input_stat.st_size) { 54 pr_info("zero-sized file (%s), nothing to do!\n", 55 self->filename); 56 goto out_close; 57 } 58 59 if (perf_session__read_header(self, self->fd) < 0) { 60 pr_err("incompatible file format (rerun with -v to learn more)"); 61 goto out_close; 62 } 63 64 if (!perf_evlist__valid_sample_type(self->evlist)) { 65 pr_err("non matching sample_type"); 66 goto out_close; 67 } 68 69 if (!perf_evlist__valid_sample_id_all(self->evlist)) { 70 pr_err("non matching sample_id_all"); 71 goto out_close; 72 } 73 74 self->size = input_stat.st_size; 75 return 0; 76 77 out_close: 78 close(self->fd); 79 self->fd = -1; 80 return -1; 81 } 82 83 void perf_session__update_sample_type(struct perf_session *self) 84 { 85 self->sample_type = perf_evlist__sample_type(self->evlist); 86 self->sample_size = __perf_evsel__sample_size(self->sample_type); 87 self->sample_id_all = perf_evlist__sample_id_all(self->evlist); 88 self->id_hdr_size = perf_evlist__id_hdr_size(self->evlist); 89 self->host_machine.id_hdr_size = self->id_hdr_size; 90 } 91 92 int perf_session__create_kernel_maps(struct perf_session *self) 93 { 94 int ret = machine__create_kernel_maps(&self->host_machine); 95 96 if (ret >= 0) 97 ret = machines__create_guest_kernel_maps(&self->machines); 98 return ret; 99 } 100 101 static void perf_session__destroy_kernel_maps(struct perf_session *self) 102 { 103 machine__destroy_kernel_maps(&self->host_machine); 104 machines__destroy_guest_kernel_maps(&self->machines); 105 } 106 107 struct perf_session *perf_session__new(const char *filename, int mode, 108 bool force, bool repipe, 109 struct perf_tool *tool) 110 { 111 struct perf_session *self; 112 struct stat st; 113 size_t len; 114 115 if (!filename || !strlen(filename)) { 116 if (!fstat(STDIN_FILENO, &st) && S_ISFIFO(st.st_mode)) 117 filename = "-"; 118 else 119 filename = "perf.data"; 120 } 121 122 len = strlen(filename); 123 self = zalloc(sizeof(*self) + len); 124 125 if (self == NULL) 126 goto out; 127 128 memcpy(self->filename, filename, len); 129 /* 130 * On 64bit we can mmap the data file in one go. No need for tiny mmap 131 * slices. On 32bit we use 32MB. 132 */ 133 #if BITS_PER_LONG == 64 134 self->mmap_window = ULLONG_MAX; 135 #else 136 self->mmap_window = 32 * 1024 * 1024ULL; 137 #endif 138 self->machines = RB_ROOT; 139 self->repipe = repipe; 140 INIT_LIST_HEAD(&self->ordered_samples.samples); 141 INIT_LIST_HEAD(&self->ordered_samples.sample_cache); 142 INIT_LIST_HEAD(&self->ordered_samples.to_free); 143 machine__init(&self->host_machine, "", HOST_KERNEL_ID); 144 hists__init(&self->hists); 145 146 if (mode == O_RDONLY) { 147 if (perf_session__open(self, force) < 0) 148 goto out_delete; 149 perf_session__update_sample_type(self); 150 } else if (mode == O_WRONLY) { 151 /* 152 * In O_RDONLY mode this will be performed when reading the 153 * kernel MMAP event, in perf_event__process_mmap(). 154 */ 155 if (perf_session__create_kernel_maps(self) < 0) 156 goto out_delete; 157 } 158 159 if (tool && tool->ordering_requires_timestamps && 160 tool->ordered_samples && !self->sample_id_all) { 161 dump_printf("WARNING: No sample_id_all support, falling back to unordered processing\n"); 162 tool->ordered_samples = false; 163 } 164 165 out: 166 return self; 167 out_delete: 168 perf_session__delete(self); 169 return NULL; 170 } 171 172 static void machine__delete_dead_threads(struct machine *machine) 173 { 174 struct thread *n, *t; 175 176 list_for_each_entry_safe(t, n, &machine->dead_threads, node) { 177 list_del(&t->node); 178 thread__delete(t); 179 } 180 } 181 182 static void perf_session__delete_dead_threads(struct perf_session *session) 183 { 184 machine__delete_dead_threads(&session->host_machine); 185 } 186 187 static void machine__delete_threads(struct machine *self) 188 { 189 struct rb_node *nd = rb_first(&self->threads); 190 191 while (nd) { 192 struct thread *t = rb_entry(nd, struct thread, rb_node); 193 194 rb_erase(&t->rb_node, &self->threads); 195 nd = rb_next(nd); 196 thread__delete(t); 197 } 198 } 199 200 static void perf_session__delete_threads(struct perf_session *session) 201 { 202 machine__delete_threads(&session->host_machine); 203 } 204 205 void perf_session__delete(struct perf_session *self) 206 { 207 perf_session__destroy_kernel_maps(self); 208 perf_session__delete_dead_threads(self); 209 perf_session__delete_threads(self); 210 machine__exit(&self->host_machine); 211 close(self->fd); 212 free(self); 213 } 214 215 void machine__remove_thread(struct machine *self, struct thread *th) 216 { 217 self->last_match = NULL; 218 rb_erase(&th->rb_node, &self->threads); 219 /* 220 * We may have references to this thread, for instance in some hist_entry 221 * instances, so just move them to a separate list. 222 */ 223 list_add_tail(&th->node, &self->dead_threads); 224 } 225 226 static bool symbol__match_parent_regex(struct symbol *sym) 227 { 228 if (sym->name && !regexec(&parent_regex, sym->name, 0, NULL, 0)) 229 return 1; 230 231 return 0; 232 } 233 234 static const u8 cpumodes[] = { 235 PERF_RECORD_MISC_USER, 236 PERF_RECORD_MISC_KERNEL, 237 PERF_RECORD_MISC_GUEST_USER, 238 PERF_RECORD_MISC_GUEST_KERNEL 239 }; 240 #define NCPUMODES (sizeof(cpumodes)/sizeof(u8)) 241 242 static void ip__resolve_ams(struct machine *self, struct thread *thread, 243 struct addr_map_symbol *ams, 244 u64 ip) 245 { 246 struct addr_location al; 247 size_t i; 248 u8 m; 249 250 memset(&al, 0, sizeof(al)); 251 252 for (i = 0; i < NCPUMODES; i++) { 253 m = cpumodes[i]; 254 /* 255 * We cannot use the header.misc hint to determine whether a 256 * branch stack address is user, kernel, guest, hypervisor. 257 * Branches may straddle the kernel/user/hypervisor boundaries. 258 * Thus, we have to try consecutively until we find a match 259 * or else, the symbol is unknown 260 */ 261 thread__find_addr_location(thread, self, m, MAP__FUNCTION, 262 ip, &al, NULL); 263 if (al.sym) 264 goto found; 265 } 266 found: 267 ams->addr = ip; 268 ams->al_addr = al.addr; 269 ams->sym = al.sym; 270 ams->map = al.map; 271 } 272 273 struct branch_info *machine__resolve_bstack(struct machine *self, 274 struct thread *thr, 275 struct branch_stack *bs) 276 { 277 struct branch_info *bi; 278 unsigned int i; 279 280 bi = calloc(bs->nr, sizeof(struct branch_info)); 281 if (!bi) 282 return NULL; 283 284 for (i = 0; i < bs->nr; i++) { 285 ip__resolve_ams(self, thr, &bi[i].to, bs->entries[i].to); 286 ip__resolve_ams(self, thr, &bi[i].from, bs->entries[i].from); 287 bi[i].flags = bs->entries[i].flags; 288 } 289 return bi; 290 } 291 292 int machine__resolve_callchain(struct machine *self, 293 struct thread *thread, 294 struct ip_callchain *chain, 295 struct symbol **parent) 296 { 297 u8 cpumode = PERF_RECORD_MISC_USER; 298 unsigned int i; 299 int err; 300 301 callchain_cursor_reset(&callchain_cursor); 302 303 if (chain->nr > PERF_MAX_STACK_DEPTH) { 304 pr_warning("corrupted callchain. skipping...\n"); 305 return 0; 306 } 307 308 for (i = 0; i < chain->nr; i++) { 309 u64 ip; 310 struct addr_location al; 311 312 if (callchain_param.order == ORDER_CALLEE) 313 ip = chain->ips[i]; 314 else 315 ip = chain->ips[chain->nr - i - 1]; 316 317 if (ip >= PERF_CONTEXT_MAX) { 318 switch (ip) { 319 case PERF_CONTEXT_HV: 320 cpumode = PERF_RECORD_MISC_HYPERVISOR; break; 321 case PERF_CONTEXT_KERNEL: 322 cpumode = PERF_RECORD_MISC_KERNEL; break; 323 case PERF_CONTEXT_USER: 324 cpumode = PERF_RECORD_MISC_USER; break; 325 default: 326 pr_debug("invalid callchain context: " 327 "%"PRId64"\n", (s64) ip); 328 /* 329 * It seems the callchain is corrupted. 330 * Discard all. 331 */ 332 callchain_cursor_reset(&callchain_cursor); 333 return 0; 334 } 335 continue; 336 } 337 338 al.filtered = false; 339 thread__find_addr_location(thread, self, cpumode, 340 MAP__FUNCTION, ip, &al, NULL); 341 if (al.sym != NULL) { 342 if (sort__has_parent && !*parent && 343 symbol__match_parent_regex(al.sym)) 344 *parent = al.sym; 345 if (!symbol_conf.use_callchain) 346 break; 347 } 348 349 err = callchain_cursor_append(&callchain_cursor, 350 ip, al.map, al.sym); 351 if (err) 352 return err; 353 } 354 355 return 0; 356 } 357 358 static int process_event_synth_tracing_data_stub(union perf_event *event __used, 359 struct perf_session *session __used) 360 { 361 dump_printf(": unhandled!\n"); 362 return 0; 363 } 364 365 static int process_event_synth_attr_stub(union perf_event *event __used, 366 struct perf_evlist **pevlist __used) 367 { 368 dump_printf(": unhandled!\n"); 369 return 0; 370 } 371 372 static int process_event_sample_stub(struct perf_tool *tool __used, 373 union perf_event *event __used, 374 struct perf_sample *sample __used, 375 struct perf_evsel *evsel __used, 376 struct machine *machine __used) 377 { 378 dump_printf(": unhandled!\n"); 379 return 0; 380 } 381 382 static int process_event_stub(struct perf_tool *tool __used, 383 union perf_event *event __used, 384 struct perf_sample *sample __used, 385 struct machine *machine __used) 386 { 387 dump_printf(": unhandled!\n"); 388 return 0; 389 } 390 391 static int process_finished_round_stub(struct perf_tool *tool __used, 392 union perf_event *event __used, 393 struct perf_session *perf_session __used) 394 { 395 dump_printf(": unhandled!\n"); 396 return 0; 397 } 398 399 static int process_event_type_stub(struct perf_tool *tool __used, 400 union perf_event *event __used) 401 { 402 dump_printf(": unhandled!\n"); 403 return 0; 404 } 405 406 static int process_finished_round(struct perf_tool *tool, 407 union perf_event *event, 408 struct perf_session *session); 409 410 static void perf_tool__fill_defaults(struct perf_tool *tool) 411 { 412 if (tool->sample == NULL) 413 tool->sample = process_event_sample_stub; 414 if (tool->mmap == NULL) 415 tool->mmap = process_event_stub; 416 if (tool->comm == NULL) 417 tool->comm = process_event_stub; 418 if (tool->fork == NULL) 419 tool->fork = process_event_stub; 420 if (tool->exit == NULL) 421 tool->exit = process_event_stub; 422 if (tool->lost == NULL) 423 tool->lost = perf_event__process_lost; 424 if (tool->read == NULL) 425 tool->read = process_event_sample_stub; 426 if (tool->throttle == NULL) 427 tool->throttle = process_event_stub; 428 if (tool->unthrottle == NULL) 429 tool->unthrottle = process_event_stub; 430 if (tool->attr == NULL) 431 tool->attr = process_event_synth_attr_stub; 432 if (tool->event_type == NULL) 433 tool->event_type = process_event_type_stub; 434 if (tool->tracing_data == NULL) 435 tool->tracing_data = process_event_synth_tracing_data_stub; 436 if (tool->build_id == NULL) 437 tool->build_id = process_finished_round_stub; 438 if (tool->finished_round == NULL) { 439 if (tool->ordered_samples) 440 tool->finished_round = process_finished_round; 441 else 442 tool->finished_round = process_finished_round_stub; 443 } 444 } 445 446 void mem_bswap_32(void *src, int byte_size) 447 { 448 u32 *m = src; 449 while (byte_size > 0) { 450 *m = bswap_32(*m); 451 byte_size -= sizeof(u32); 452 ++m; 453 } 454 } 455 456 void mem_bswap_64(void *src, int byte_size) 457 { 458 u64 *m = src; 459 460 while (byte_size > 0) { 461 *m = bswap_64(*m); 462 byte_size -= sizeof(u64); 463 ++m; 464 } 465 } 466 467 static void swap_sample_id_all(union perf_event *event, void *data) 468 { 469 void *end = (void *) event + event->header.size; 470 int size = end - data; 471 472 BUG_ON(size % sizeof(u64)); 473 mem_bswap_64(data, size); 474 } 475 476 static void perf_event__all64_swap(union perf_event *event, 477 bool sample_id_all __used) 478 { 479 struct perf_event_header *hdr = &event->header; 480 mem_bswap_64(hdr + 1, event->header.size - sizeof(*hdr)); 481 } 482 483 static void perf_event__comm_swap(union perf_event *event, bool sample_id_all) 484 { 485 event->comm.pid = bswap_32(event->comm.pid); 486 event->comm.tid = bswap_32(event->comm.tid); 487 488 if (sample_id_all) { 489 void *data = &event->comm.comm; 490 491 data += ALIGN(strlen(data) + 1, sizeof(u64)); 492 swap_sample_id_all(event, data); 493 } 494 } 495 496 static void perf_event__mmap_swap(union perf_event *event, 497 bool sample_id_all) 498 { 499 event->mmap.pid = bswap_32(event->mmap.pid); 500 event->mmap.tid = bswap_32(event->mmap.tid); 501 event->mmap.start = bswap_64(event->mmap.start); 502 event->mmap.len = bswap_64(event->mmap.len); 503 event->mmap.pgoff = bswap_64(event->mmap.pgoff); 504 505 if (sample_id_all) { 506 void *data = &event->mmap.filename; 507 508 data += ALIGN(strlen(data) + 1, sizeof(u64)); 509 swap_sample_id_all(event, data); 510 } 511 } 512 513 static void perf_event__task_swap(union perf_event *event, bool sample_id_all) 514 { 515 event->fork.pid = bswap_32(event->fork.pid); 516 event->fork.tid = bswap_32(event->fork.tid); 517 event->fork.ppid = bswap_32(event->fork.ppid); 518 event->fork.ptid = bswap_32(event->fork.ptid); 519 event->fork.time = bswap_64(event->fork.time); 520 521 if (sample_id_all) 522 swap_sample_id_all(event, &event->fork + 1); 523 } 524 525 static void perf_event__read_swap(union perf_event *event, bool sample_id_all) 526 { 527 event->read.pid = bswap_32(event->read.pid); 528 event->read.tid = bswap_32(event->read.tid); 529 event->read.value = bswap_64(event->read.value); 530 event->read.time_enabled = bswap_64(event->read.time_enabled); 531 event->read.time_running = bswap_64(event->read.time_running); 532 event->read.id = bswap_64(event->read.id); 533 534 if (sample_id_all) 535 swap_sample_id_all(event, &event->read + 1); 536 } 537 538 static u8 revbyte(u8 b) 539 { 540 int rev = (b >> 4) | ((b & 0xf) << 4); 541 rev = ((rev & 0xcc) >> 2) | ((rev & 0x33) << 2); 542 rev = ((rev & 0xaa) >> 1) | ((rev & 0x55) << 1); 543 return (u8) rev; 544 } 545 546 /* 547 * XXX this is hack in attempt to carry flags bitfield 548 * throught endian village. ABI says: 549 * 550 * Bit-fields are allocated from right to left (least to most significant) 551 * on little-endian implementations and from left to right (most to least 552 * significant) on big-endian implementations. 553 * 554 * The above seems to be byte specific, so we need to reverse each 555 * byte of the bitfield. 'Internet' also says this might be implementation 556 * specific and we probably need proper fix and carry perf_event_attr 557 * bitfield flags in separate data file FEAT_ section. Thought this seems 558 * to work for now. 559 */ 560 static void swap_bitfield(u8 *p, unsigned len) 561 { 562 unsigned i; 563 564 for (i = 0; i < len; i++) { 565 *p = revbyte(*p); 566 p++; 567 } 568 } 569 570 /* exported for swapping attributes in file header */ 571 void perf_event__attr_swap(struct perf_event_attr *attr) 572 { 573 attr->type = bswap_32(attr->type); 574 attr->size = bswap_32(attr->size); 575 attr->config = bswap_64(attr->config); 576 attr->sample_period = bswap_64(attr->sample_period); 577 attr->sample_type = bswap_64(attr->sample_type); 578 attr->read_format = bswap_64(attr->read_format); 579 attr->wakeup_events = bswap_32(attr->wakeup_events); 580 attr->bp_type = bswap_32(attr->bp_type); 581 attr->bp_addr = bswap_64(attr->bp_addr); 582 attr->bp_len = bswap_64(attr->bp_len); 583 584 swap_bitfield((u8 *) (&attr->read_format + 1), sizeof(u64)); 585 } 586 587 static void perf_event__hdr_attr_swap(union perf_event *event, 588 bool sample_id_all __used) 589 { 590 size_t size; 591 592 perf_event__attr_swap(&event->attr.attr); 593 594 size = event->header.size; 595 size -= (void *)&event->attr.id - (void *)event; 596 mem_bswap_64(event->attr.id, size); 597 } 598 599 static void perf_event__event_type_swap(union perf_event *event, 600 bool sample_id_all __used) 601 { 602 event->event_type.event_type.event_id = 603 bswap_64(event->event_type.event_type.event_id); 604 } 605 606 static void perf_event__tracing_data_swap(union perf_event *event, 607 bool sample_id_all __used) 608 { 609 event->tracing_data.size = bswap_32(event->tracing_data.size); 610 } 611 612 typedef void (*perf_event__swap_op)(union perf_event *event, 613 bool sample_id_all); 614 615 static perf_event__swap_op perf_event__swap_ops[] = { 616 [PERF_RECORD_MMAP] = perf_event__mmap_swap, 617 [PERF_RECORD_COMM] = perf_event__comm_swap, 618 [PERF_RECORD_FORK] = perf_event__task_swap, 619 [PERF_RECORD_EXIT] = perf_event__task_swap, 620 [PERF_RECORD_LOST] = perf_event__all64_swap, 621 [PERF_RECORD_READ] = perf_event__read_swap, 622 [PERF_RECORD_SAMPLE] = perf_event__all64_swap, 623 [PERF_RECORD_HEADER_ATTR] = perf_event__hdr_attr_swap, 624 [PERF_RECORD_HEADER_EVENT_TYPE] = perf_event__event_type_swap, 625 [PERF_RECORD_HEADER_TRACING_DATA] = perf_event__tracing_data_swap, 626 [PERF_RECORD_HEADER_BUILD_ID] = NULL, 627 [PERF_RECORD_HEADER_MAX] = NULL, 628 }; 629 630 struct sample_queue { 631 u64 timestamp; 632 u64 file_offset; 633 union perf_event *event; 634 struct list_head list; 635 }; 636 637 static void perf_session_free_sample_buffers(struct perf_session *session) 638 { 639 struct ordered_samples *os = &session->ordered_samples; 640 641 while (!list_empty(&os->to_free)) { 642 struct sample_queue *sq; 643 644 sq = list_entry(os->to_free.next, struct sample_queue, list); 645 list_del(&sq->list); 646 free(sq); 647 } 648 } 649 650 static int perf_session_deliver_event(struct perf_session *session, 651 union perf_event *event, 652 struct perf_sample *sample, 653 struct perf_tool *tool, 654 u64 file_offset); 655 656 static void flush_sample_queue(struct perf_session *s, 657 struct perf_tool *tool) 658 { 659 struct ordered_samples *os = &s->ordered_samples; 660 struct list_head *head = &os->samples; 661 struct sample_queue *tmp, *iter; 662 struct perf_sample sample; 663 u64 limit = os->next_flush; 664 u64 last_ts = os->last_sample ? os->last_sample->timestamp : 0ULL; 665 unsigned idx = 0, progress_next = os->nr_samples / 16; 666 int ret; 667 668 if (!tool->ordered_samples || !limit) 669 return; 670 671 list_for_each_entry_safe(iter, tmp, head, list) { 672 if (iter->timestamp > limit) 673 break; 674 675 ret = perf_session__parse_sample(s, iter->event, &sample); 676 if (ret) 677 pr_err("Can't parse sample, err = %d\n", ret); 678 else 679 perf_session_deliver_event(s, iter->event, &sample, tool, 680 iter->file_offset); 681 682 os->last_flush = iter->timestamp; 683 list_del(&iter->list); 684 list_add(&iter->list, &os->sample_cache); 685 if (++idx >= progress_next) { 686 progress_next += os->nr_samples / 16; 687 ui_progress__update(idx, os->nr_samples, 688 "Processing time ordered events..."); 689 } 690 } 691 692 if (list_empty(head)) { 693 os->last_sample = NULL; 694 } else if (last_ts <= limit) { 695 os->last_sample = 696 list_entry(head->prev, struct sample_queue, list); 697 } 698 699 os->nr_samples = 0; 700 } 701 702 /* 703 * When perf record finishes a pass on every buffers, it records this pseudo 704 * event. 705 * We record the max timestamp t found in the pass n. 706 * Assuming these timestamps are monotonic across cpus, we know that if 707 * a buffer still has events with timestamps below t, they will be all 708 * available and then read in the pass n + 1. 709 * Hence when we start to read the pass n + 2, we can safely flush every 710 * events with timestamps below t. 711 * 712 * ============ PASS n ================= 713 * CPU 0 | CPU 1 714 * | 715 * cnt1 timestamps | cnt2 timestamps 716 * 1 | 2 717 * 2 | 3 718 * - | 4 <--- max recorded 719 * 720 * ============ PASS n + 1 ============== 721 * CPU 0 | CPU 1 722 * | 723 * cnt1 timestamps | cnt2 timestamps 724 * 3 | 5 725 * 4 | 6 726 * 5 | 7 <---- max recorded 727 * 728 * Flush every events below timestamp 4 729 * 730 * ============ PASS n + 2 ============== 731 * CPU 0 | CPU 1 732 * | 733 * cnt1 timestamps | cnt2 timestamps 734 * 6 | 8 735 * 7 | 9 736 * - | 10 737 * 738 * Flush every events below timestamp 7 739 * etc... 740 */ 741 static int process_finished_round(struct perf_tool *tool, 742 union perf_event *event __used, 743 struct perf_session *session) 744 { 745 flush_sample_queue(session, tool); 746 session->ordered_samples.next_flush = session->ordered_samples.max_timestamp; 747 748 return 0; 749 } 750 751 /* The queue is ordered by time */ 752 static void __queue_event(struct sample_queue *new, struct perf_session *s) 753 { 754 struct ordered_samples *os = &s->ordered_samples; 755 struct sample_queue *sample = os->last_sample; 756 u64 timestamp = new->timestamp; 757 struct list_head *p; 758 759 ++os->nr_samples; 760 os->last_sample = new; 761 762 if (!sample) { 763 list_add(&new->list, &os->samples); 764 os->max_timestamp = timestamp; 765 return; 766 } 767 768 /* 769 * last_sample might point to some random place in the list as it's 770 * the last queued event. We expect that the new event is close to 771 * this. 772 */ 773 if (sample->timestamp <= timestamp) { 774 while (sample->timestamp <= timestamp) { 775 p = sample->list.next; 776 if (p == &os->samples) { 777 list_add_tail(&new->list, &os->samples); 778 os->max_timestamp = timestamp; 779 return; 780 } 781 sample = list_entry(p, struct sample_queue, list); 782 } 783 list_add_tail(&new->list, &sample->list); 784 } else { 785 while (sample->timestamp > timestamp) { 786 p = sample->list.prev; 787 if (p == &os->samples) { 788 list_add(&new->list, &os->samples); 789 return; 790 } 791 sample = list_entry(p, struct sample_queue, list); 792 } 793 list_add(&new->list, &sample->list); 794 } 795 } 796 797 #define MAX_SAMPLE_BUFFER (64 * 1024 / sizeof(struct sample_queue)) 798 799 static int perf_session_queue_event(struct perf_session *s, union perf_event *event, 800 struct perf_sample *sample, u64 file_offset) 801 { 802 struct ordered_samples *os = &s->ordered_samples; 803 struct list_head *sc = &os->sample_cache; 804 u64 timestamp = sample->time; 805 struct sample_queue *new; 806 807 if (!timestamp || timestamp == ~0ULL) 808 return -ETIME; 809 810 if (timestamp < s->ordered_samples.last_flush) { 811 printf("Warning: Timestamp below last timeslice flush\n"); 812 return -EINVAL; 813 } 814 815 if (!list_empty(sc)) { 816 new = list_entry(sc->next, struct sample_queue, list); 817 list_del(&new->list); 818 } else if (os->sample_buffer) { 819 new = os->sample_buffer + os->sample_buffer_idx; 820 if (++os->sample_buffer_idx == MAX_SAMPLE_BUFFER) 821 os->sample_buffer = NULL; 822 } else { 823 os->sample_buffer = malloc(MAX_SAMPLE_BUFFER * sizeof(*new)); 824 if (!os->sample_buffer) 825 return -ENOMEM; 826 list_add(&os->sample_buffer->list, &os->to_free); 827 os->sample_buffer_idx = 2; 828 new = os->sample_buffer + 1; 829 } 830 831 new->timestamp = timestamp; 832 new->file_offset = file_offset; 833 new->event = event; 834 835 __queue_event(new, s); 836 837 return 0; 838 } 839 840 static void callchain__printf(struct perf_sample *sample) 841 { 842 unsigned int i; 843 844 printf("... chain: nr:%" PRIu64 "\n", sample->callchain->nr); 845 846 for (i = 0; i < sample->callchain->nr; i++) 847 printf("..... %2d: %016" PRIx64 "\n", 848 i, sample->callchain->ips[i]); 849 } 850 851 static void branch_stack__printf(struct perf_sample *sample) 852 { 853 uint64_t i; 854 855 printf("... branch stack: nr:%" PRIu64 "\n", sample->branch_stack->nr); 856 857 for (i = 0; i < sample->branch_stack->nr; i++) 858 printf("..... %2"PRIu64": %016" PRIx64 " -> %016" PRIx64 "\n", 859 i, sample->branch_stack->entries[i].from, 860 sample->branch_stack->entries[i].to); 861 } 862 863 static void perf_session__print_tstamp(struct perf_session *session, 864 union perf_event *event, 865 struct perf_sample *sample) 866 { 867 if (event->header.type != PERF_RECORD_SAMPLE && 868 !session->sample_id_all) { 869 fputs("-1 -1 ", stdout); 870 return; 871 } 872 873 if ((session->sample_type & PERF_SAMPLE_CPU)) 874 printf("%u ", sample->cpu); 875 876 if (session->sample_type & PERF_SAMPLE_TIME) 877 printf("%" PRIu64 " ", sample->time); 878 } 879 880 static void dump_event(struct perf_session *session, union perf_event *event, 881 u64 file_offset, struct perf_sample *sample) 882 { 883 if (!dump_trace) 884 return; 885 886 printf("\n%#" PRIx64 " [%#x]: event: %d\n", 887 file_offset, event->header.size, event->header.type); 888 889 trace_event(event); 890 891 if (sample) 892 perf_session__print_tstamp(session, event, sample); 893 894 printf("%#" PRIx64 " [%#x]: PERF_RECORD_%s", file_offset, 895 event->header.size, perf_event__name(event->header.type)); 896 } 897 898 static void dump_sample(struct perf_session *session, union perf_event *event, 899 struct perf_sample *sample) 900 { 901 if (!dump_trace) 902 return; 903 904 printf("(IP, %d): %d/%d: %#" PRIx64 " period: %" PRIu64 " addr: %#" PRIx64 "\n", 905 event->header.misc, sample->pid, sample->tid, sample->ip, 906 sample->period, sample->addr); 907 908 if (session->sample_type & PERF_SAMPLE_CALLCHAIN) 909 callchain__printf(sample); 910 911 if (session->sample_type & PERF_SAMPLE_BRANCH_STACK) 912 branch_stack__printf(sample); 913 } 914 915 static struct machine * 916 perf_session__find_machine_for_cpumode(struct perf_session *session, 917 union perf_event *event) 918 { 919 const u8 cpumode = event->header.misc & PERF_RECORD_MISC_CPUMODE_MASK; 920 921 if (perf_guest && 922 ((cpumode == PERF_RECORD_MISC_GUEST_KERNEL) || 923 (cpumode == PERF_RECORD_MISC_GUEST_USER))) { 924 u32 pid; 925 926 if (event->header.type == PERF_RECORD_MMAP) 927 pid = event->mmap.pid; 928 else 929 pid = event->ip.pid; 930 931 return perf_session__findnew_machine(session, pid); 932 } 933 934 return perf_session__find_host_machine(session); 935 } 936 937 static int perf_session_deliver_event(struct perf_session *session, 938 union perf_event *event, 939 struct perf_sample *sample, 940 struct perf_tool *tool, 941 u64 file_offset) 942 { 943 struct perf_evsel *evsel; 944 struct machine *machine; 945 946 dump_event(session, event, file_offset, sample); 947 948 evsel = perf_evlist__id2evsel(session->evlist, sample->id); 949 if (evsel != NULL && event->header.type != PERF_RECORD_SAMPLE) { 950 /* 951 * XXX We're leaving PERF_RECORD_SAMPLE unnacounted here 952 * because the tools right now may apply filters, discarding 953 * some of the samples. For consistency, in the future we 954 * should have something like nr_filtered_samples and remove 955 * the sample->period from total_sample_period, etc, KISS for 956 * now tho. 957 * 958 * Also testing against NULL allows us to handle files without 959 * attr.sample_id_all and/or without PERF_SAMPLE_ID. In the 960 * future probably it'll be a good idea to restrict event 961 * processing via perf_session to files with both set. 962 */ 963 hists__inc_nr_events(&evsel->hists, event->header.type); 964 } 965 966 machine = perf_session__find_machine_for_cpumode(session, event); 967 968 switch (event->header.type) { 969 case PERF_RECORD_SAMPLE: 970 dump_sample(session, event, sample); 971 if (evsel == NULL) { 972 ++session->hists.stats.nr_unknown_id; 973 return 0; 974 } 975 if (machine == NULL) { 976 ++session->hists.stats.nr_unprocessable_samples; 977 return 0; 978 } 979 return tool->sample(tool, event, sample, evsel, machine); 980 case PERF_RECORD_MMAP: 981 return tool->mmap(tool, event, sample, machine); 982 case PERF_RECORD_COMM: 983 return tool->comm(tool, event, sample, machine); 984 case PERF_RECORD_FORK: 985 return tool->fork(tool, event, sample, machine); 986 case PERF_RECORD_EXIT: 987 return tool->exit(tool, event, sample, machine); 988 case PERF_RECORD_LOST: 989 if (tool->lost == perf_event__process_lost) 990 session->hists.stats.total_lost += event->lost.lost; 991 return tool->lost(tool, event, sample, machine); 992 case PERF_RECORD_READ: 993 return tool->read(tool, event, sample, evsel, machine); 994 case PERF_RECORD_THROTTLE: 995 return tool->throttle(tool, event, sample, machine); 996 case PERF_RECORD_UNTHROTTLE: 997 return tool->unthrottle(tool, event, sample, machine); 998 default: 999 ++session->hists.stats.nr_unknown_events; 1000 return -1; 1001 } 1002 } 1003 1004 static int perf_session__preprocess_sample(struct perf_session *session, 1005 union perf_event *event, struct perf_sample *sample) 1006 { 1007 if (event->header.type != PERF_RECORD_SAMPLE || 1008 !(session->sample_type & PERF_SAMPLE_CALLCHAIN)) 1009 return 0; 1010 1011 if (!ip_callchain__valid(sample->callchain, event)) { 1012 pr_debug("call-chain problem with event, skipping it.\n"); 1013 ++session->hists.stats.nr_invalid_chains; 1014 session->hists.stats.total_invalid_chains += sample->period; 1015 return -EINVAL; 1016 } 1017 return 0; 1018 } 1019 1020 static int perf_session__process_user_event(struct perf_session *session, union perf_event *event, 1021 struct perf_tool *tool, u64 file_offset) 1022 { 1023 int err; 1024 1025 dump_event(session, event, file_offset, NULL); 1026 1027 /* These events are processed right away */ 1028 switch (event->header.type) { 1029 case PERF_RECORD_HEADER_ATTR: 1030 err = tool->attr(event, &session->evlist); 1031 if (err == 0) 1032 perf_session__update_sample_type(session); 1033 return err; 1034 case PERF_RECORD_HEADER_EVENT_TYPE: 1035 return tool->event_type(tool, event); 1036 case PERF_RECORD_HEADER_TRACING_DATA: 1037 /* setup for reading amidst mmap */ 1038 lseek(session->fd, file_offset, SEEK_SET); 1039 return tool->tracing_data(event, session); 1040 case PERF_RECORD_HEADER_BUILD_ID: 1041 return tool->build_id(tool, event, session); 1042 case PERF_RECORD_FINISHED_ROUND: 1043 return tool->finished_round(tool, event, session); 1044 default: 1045 return -EINVAL; 1046 } 1047 } 1048 1049 static void event_swap(union perf_event *event, bool sample_id_all) 1050 { 1051 perf_event__swap_op swap; 1052 1053 swap = perf_event__swap_ops[event->header.type]; 1054 if (swap) 1055 swap(event, sample_id_all); 1056 } 1057 1058 static int perf_session__process_event(struct perf_session *session, 1059 union perf_event *event, 1060 struct perf_tool *tool, 1061 u64 file_offset) 1062 { 1063 struct perf_sample sample; 1064 int ret; 1065 1066 if (session->header.needs_swap) 1067 event_swap(event, session->sample_id_all); 1068 1069 if (event->header.type >= PERF_RECORD_HEADER_MAX) 1070 return -EINVAL; 1071 1072 hists__inc_nr_events(&session->hists, event->header.type); 1073 1074 if (event->header.type >= PERF_RECORD_USER_TYPE_START) 1075 return perf_session__process_user_event(session, event, tool, file_offset); 1076 1077 /* 1078 * For all kernel events we get the sample data 1079 */ 1080 ret = perf_session__parse_sample(session, event, &sample); 1081 if (ret) 1082 return ret; 1083 1084 /* Preprocess sample records - precheck callchains */ 1085 if (perf_session__preprocess_sample(session, event, &sample)) 1086 return 0; 1087 1088 if (tool->ordered_samples) { 1089 ret = perf_session_queue_event(session, event, &sample, 1090 file_offset); 1091 if (ret != -ETIME) 1092 return ret; 1093 } 1094 1095 return perf_session_deliver_event(session, event, &sample, tool, 1096 file_offset); 1097 } 1098 1099 void perf_event_header__bswap(struct perf_event_header *self) 1100 { 1101 self->type = bswap_32(self->type); 1102 self->misc = bswap_16(self->misc); 1103 self->size = bswap_16(self->size); 1104 } 1105 1106 struct thread *perf_session__findnew(struct perf_session *session, pid_t pid) 1107 { 1108 return machine__findnew_thread(&session->host_machine, pid); 1109 } 1110 1111 static struct thread *perf_session__register_idle_thread(struct perf_session *self) 1112 { 1113 struct thread *thread = perf_session__findnew(self, 0); 1114 1115 if (thread == NULL || thread__set_comm(thread, "swapper")) { 1116 pr_err("problem inserting idle task.\n"); 1117 thread = NULL; 1118 } 1119 1120 return thread; 1121 } 1122 1123 static void perf_session__warn_about_errors(const struct perf_session *session, 1124 const struct perf_tool *tool) 1125 { 1126 if (tool->lost == perf_event__process_lost && 1127 session->hists.stats.nr_events[PERF_RECORD_LOST] != 0) { 1128 ui__warning("Processed %d events and lost %d chunks!\n\n" 1129 "Check IO/CPU overload!\n\n", 1130 session->hists.stats.nr_events[0], 1131 session->hists.stats.nr_events[PERF_RECORD_LOST]); 1132 } 1133 1134 if (session->hists.stats.nr_unknown_events != 0) { 1135 ui__warning("Found %u unknown events!\n\n" 1136 "Is this an older tool processing a perf.data " 1137 "file generated by a more recent tool?\n\n" 1138 "If that is not the case, consider " 1139 "reporting to linux-kernel@vger.kernel.org.\n\n", 1140 session->hists.stats.nr_unknown_events); 1141 } 1142 1143 if (session->hists.stats.nr_unknown_id != 0) { 1144 ui__warning("%u samples with id not present in the header\n", 1145 session->hists.stats.nr_unknown_id); 1146 } 1147 1148 if (session->hists.stats.nr_invalid_chains != 0) { 1149 ui__warning("Found invalid callchains!\n\n" 1150 "%u out of %u events were discarded for this reason.\n\n" 1151 "Consider reporting to linux-kernel@vger.kernel.org.\n\n", 1152 session->hists.stats.nr_invalid_chains, 1153 session->hists.stats.nr_events[PERF_RECORD_SAMPLE]); 1154 } 1155 1156 if (session->hists.stats.nr_unprocessable_samples != 0) { 1157 ui__warning("%u unprocessable samples recorded.\n" 1158 "Do you have a KVM guest running and not using 'perf kvm'?\n", 1159 session->hists.stats.nr_unprocessable_samples); 1160 } 1161 } 1162 1163 #define session_done() (*(volatile int *)(&session_done)) 1164 volatile int session_done; 1165 1166 static int __perf_session__process_pipe_events(struct perf_session *self, 1167 struct perf_tool *tool) 1168 { 1169 union perf_event *event; 1170 uint32_t size, cur_size = 0; 1171 void *buf = NULL; 1172 int skip = 0; 1173 u64 head; 1174 int err; 1175 void *p; 1176 1177 perf_tool__fill_defaults(tool); 1178 1179 head = 0; 1180 cur_size = sizeof(union perf_event); 1181 1182 buf = malloc(cur_size); 1183 if (!buf) 1184 return -errno; 1185 more: 1186 event = buf; 1187 err = readn(self->fd, event, sizeof(struct perf_event_header)); 1188 if (err <= 0) { 1189 if (err == 0) 1190 goto done; 1191 1192 pr_err("failed to read event header\n"); 1193 goto out_err; 1194 } 1195 1196 if (self->header.needs_swap) 1197 perf_event_header__bswap(&event->header); 1198 1199 size = event->header.size; 1200 if (size == 0) 1201 size = 8; 1202 1203 if (size > cur_size) { 1204 void *new = realloc(buf, size); 1205 if (!new) { 1206 pr_err("failed to allocate memory to read event\n"); 1207 goto out_err; 1208 } 1209 buf = new; 1210 cur_size = size; 1211 event = buf; 1212 } 1213 p = event; 1214 p += sizeof(struct perf_event_header); 1215 1216 if (size - sizeof(struct perf_event_header)) { 1217 err = readn(self->fd, p, size - sizeof(struct perf_event_header)); 1218 if (err <= 0) { 1219 if (err == 0) { 1220 pr_err("unexpected end of event stream\n"); 1221 goto done; 1222 } 1223 1224 pr_err("failed to read event data\n"); 1225 goto out_err; 1226 } 1227 } 1228 1229 if ((skip = perf_session__process_event(self, event, tool, head)) < 0) { 1230 pr_err("%#" PRIx64 " [%#x]: failed to process type: %d\n", 1231 head, event->header.size, event->header.type); 1232 err = -EINVAL; 1233 goto out_err; 1234 } 1235 1236 head += size; 1237 1238 if (skip > 0) 1239 head += skip; 1240 1241 if (!session_done()) 1242 goto more; 1243 done: 1244 err = 0; 1245 out_err: 1246 free(buf); 1247 perf_session__warn_about_errors(self, tool); 1248 perf_session_free_sample_buffers(self); 1249 return err; 1250 } 1251 1252 static union perf_event * 1253 fetch_mmaped_event(struct perf_session *session, 1254 u64 head, size_t mmap_size, char *buf) 1255 { 1256 union perf_event *event; 1257 1258 /* 1259 * Ensure we have enough space remaining to read 1260 * the size of the event in the headers. 1261 */ 1262 if (head + sizeof(event->header) > mmap_size) 1263 return NULL; 1264 1265 event = (union perf_event *)(buf + head); 1266 1267 if (session->header.needs_swap) 1268 perf_event_header__bswap(&event->header); 1269 1270 if (head + event->header.size > mmap_size) 1271 return NULL; 1272 1273 return event; 1274 } 1275 1276 int __perf_session__process_events(struct perf_session *session, 1277 u64 data_offset, u64 data_size, 1278 u64 file_size, struct perf_tool *tool) 1279 { 1280 u64 head, page_offset, file_offset, file_pos, progress_next; 1281 int err, mmap_prot, mmap_flags, map_idx = 0; 1282 size_t page_size, mmap_size; 1283 char *buf, *mmaps[8]; 1284 union perf_event *event; 1285 uint32_t size; 1286 1287 perf_tool__fill_defaults(tool); 1288 1289 page_size = sysconf(_SC_PAGESIZE); 1290 1291 page_offset = page_size * (data_offset / page_size); 1292 file_offset = page_offset; 1293 head = data_offset - page_offset; 1294 1295 if (data_offset + data_size < file_size) 1296 file_size = data_offset + data_size; 1297 1298 progress_next = file_size / 16; 1299 1300 mmap_size = session->mmap_window; 1301 if (mmap_size > file_size) 1302 mmap_size = file_size; 1303 1304 memset(mmaps, 0, sizeof(mmaps)); 1305 1306 mmap_prot = PROT_READ; 1307 mmap_flags = MAP_SHARED; 1308 1309 if (session->header.needs_swap) { 1310 mmap_prot |= PROT_WRITE; 1311 mmap_flags = MAP_PRIVATE; 1312 } 1313 remap: 1314 buf = mmap(NULL, mmap_size, mmap_prot, mmap_flags, session->fd, 1315 file_offset); 1316 if (buf == MAP_FAILED) { 1317 pr_err("failed to mmap file\n"); 1318 err = -errno; 1319 goto out_err; 1320 } 1321 mmaps[map_idx] = buf; 1322 map_idx = (map_idx + 1) & (ARRAY_SIZE(mmaps) - 1); 1323 file_pos = file_offset + head; 1324 1325 more: 1326 event = fetch_mmaped_event(session, head, mmap_size, buf); 1327 if (!event) { 1328 if (mmaps[map_idx]) { 1329 munmap(mmaps[map_idx], mmap_size); 1330 mmaps[map_idx] = NULL; 1331 } 1332 1333 page_offset = page_size * (head / page_size); 1334 file_offset += page_offset; 1335 head -= page_offset; 1336 goto remap; 1337 } 1338 1339 size = event->header.size; 1340 1341 if (size == 0 || 1342 perf_session__process_event(session, event, tool, file_pos) < 0) { 1343 pr_err("%#" PRIx64 " [%#x]: failed to process type: %d\n", 1344 file_offset + head, event->header.size, 1345 event->header.type); 1346 err = -EINVAL; 1347 goto out_err; 1348 } 1349 1350 head += size; 1351 file_pos += size; 1352 1353 if (file_pos >= progress_next) { 1354 progress_next += file_size / 16; 1355 ui_progress__update(file_pos, file_size, 1356 "Processing events..."); 1357 } 1358 1359 if (file_pos < file_size) 1360 goto more; 1361 1362 err = 0; 1363 /* do the final flush for ordered samples */ 1364 session->ordered_samples.next_flush = ULLONG_MAX; 1365 flush_sample_queue(session, tool); 1366 out_err: 1367 perf_session__warn_about_errors(session, tool); 1368 perf_session_free_sample_buffers(session); 1369 return err; 1370 } 1371 1372 int perf_session__process_events(struct perf_session *self, 1373 struct perf_tool *tool) 1374 { 1375 int err; 1376 1377 if (perf_session__register_idle_thread(self) == NULL) 1378 return -ENOMEM; 1379 1380 if (!self->fd_pipe) 1381 err = __perf_session__process_events(self, 1382 self->header.data_offset, 1383 self->header.data_size, 1384 self->size, tool); 1385 else 1386 err = __perf_session__process_pipe_events(self, tool); 1387 1388 return err; 1389 } 1390 1391 bool perf_session__has_traces(struct perf_session *self, const char *msg) 1392 { 1393 if (!(self->sample_type & PERF_SAMPLE_RAW)) { 1394 pr_err("No trace sample to read. Did you call 'perf %s'?\n", msg); 1395 return false; 1396 } 1397 1398 return true; 1399 } 1400 1401 int maps__set_kallsyms_ref_reloc_sym(struct map **maps, 1402 const char *symbol_name, u64 addr) 1403 { 1404 char *bracket; 1405 enum map_type i; 1406 struct ref_reloc_sym *ref; 1407 1408 ref = zalloc(sizeof(struct ref_reloc_sym)); 1409 if (ref == NULL) 1410 return -ENOMEM; 1411 1412 ref->name = strdup(symbol_name); 1413 if (ref->name == NULL) { 1414 free(ref); 1415 return -ENOMEM; 1416 } 1417 1418 bracket = strchr(ref->name, ']'); 1419 if (bracket) 1420 *bracket = '\0'; 1421 1422 ref->addr = addr; 1423 1424 for (i = 0; i < MAP__NR_TYPES; ++i) { 1425 struct kmap *kmap = map__kmap(maps[i]); 1426 kmap->ref_reloc_sym = ref; 1427 } 1428 1429 return 0; 1430 } 1431 1432 size_t perf_session__fprintf_dsos(struct perf_session *self, FILE *fp) 1433 { 1434 return __dsos__fprintf(&self->host_machine.kernel_dsos, fp) + 1435 __dsos__fprintf(&self->host_machine.user_dsos, fp) + 1436 machines__fprintf_dsos(&self->machines, fp); 1437 } 1438 1439 size_t perf_session__fprintf_dsos_buildid(struct perf_session *self, FILE *fp, 1440 bool with_hits) 1441 { 1442 size_t ret = machine__fprintf_dsos_buildid(&self->host_machine, fp, with_hits); 1443 return ret + machines__fprintf_dsos_buildid(&self->machines, fp, with_hits); 1444 } 1445 1446 size_t perf_session__fprintf_nr_events(struct perf_session *session, FILE *fp) 1447 { 1448 struct perf_evsel *pos; 1449 size_t ret = fprintf(fp, "Aggregated stats:\n"); 1450 1451 ret += hists__fprintf_nr_events(&session->hists, fp); 1452 1453 list_for_each_entry(pos, &session->evlist->entries, node) { 1454 ret += fprintf(fp, "%s stats:\n", perf_evsel__name(pos)); 1455 ret += hists__fprintf_nr_events(&pos->hists, fp); 1456 } 1457 1458 return ret; 1459 } 1460 1461 size_t perf_session__fprintf(struct perf_session *session, FILE *fp) 1462 { 1463 /* 1464 * FIXME: Here we have to actually print all the machines in this 1465 * session, not just the host... 1466 */ 1467 return machine__fprintf(&session->host_machine, fp); 1468 } 1469 1470 void perf_session__remove_thread(struct perf_session *session, 1471 struct thread *th) 1472 { 1473 /* 1474 * FIXME: This one makes no sense, we need to remove the thread from 1475 * the machine it belongs to, perf_session can have many machines, so 1476 * doing it always on ->host_machine is wrong. Fix when auditing all 1477 * the 'perf kvm' code. 1478 */ 1479 machine__remove_thread(&session->host_machine, th); 1480 } 1481 1482 struct perf_evsel *perf_session__find_first_evtype(struct perf_session *session, 1483 unsigned int type) 1484 { 1485 struct perf_evsel *pos; 1486 1487 list_for_each_entry(pos, &session->evlist->entries, node) { 1488 if (pos->attr.type == type) 1489 return pos; 1490 } 1491 return NULL; 1492 } 1493 1494 void perf_event__print_ip(union perf_event *event, struct perf_sample *sample, 1495 struct machine *machine, int print_sym, 1496 int print_dso, int print_symoffset) 1497 { 1498 struct addr_location al; 1499 struct callchain_cursor_node *node; 1500 1501 if (perf_event__preprocess_sample(event, machine, &al, sample, 1502 NULL) < 0) { 1503 error("problem processing %d event, skipping it.\n", 1504 event->header.type); 1505 return; 1506 } 1507 1508 if (symbol_conf.use_callchain && sample->callchain) { 1509 1510 if (machine__resolve_callchain(machine, al.thread, 1511 sample->callchain, NULL) != 0) { 1512 if (verbose) 1513 error("Failed to resolve callchain. Skipping\n"); 1514 return; 1515 } 1516 callchain_cursor_commit(&callchain_cursor); 1517 1518 while (1) { 1519 node = callchain_cursor_current(&callchain_cursor); 1520 if (!node) 1521 break; 1522 1523 printf("\t%16" PRIx64, node->ip); 1524 if (print_sym) { 1525 printf(" "); 1526 symbol__fprintf_symname(node->sym, stdout); 1527 } 1528 if (print_dso) { 1529 printf(" ("); 1530 map__fprintf_dsoname(node->map, stdout); 1531 printf(")"); 1532 } 1533 printf("\n"); 1534 1535 callchain_cursor_advance(&callchain_cursor); 1536 } 1537 1538 } else { 1539 printf("%16" PRIx64, sample->ip); 1540 if (print_sym) { 1541 printf(" "); 1542 if (print_symoffset) 1543 symbol__fprintf_symname_offs(al.sym, &al, 1544 stdout); 1545 else 1546 symbol__fprintf_symname(al.sym, stdout); 1547 } 1548 1549 if (print_dso) { 1550 printf(" ("); 1551 map__fprintf_dsoname(al.map, stdout); 1552 printf(")"); 1553 } 1554 } 1555 } 1556 1557 int perf_session__cpu_bitmap(struct perf_session *session, 1558 const char *cpu_list, unsigned long *cpu_bitmap) 1559 { 1560 int i; 1561 struct cpu_map *map; 1562 1563 for (i = 0; i < PERF_TYPE_MAX; ++i) { 1564 struct perf_evsel *evsel; 1565 1566 evsel = perf_session__find_first_evtype(session, i); 1567 if (!evsel) 1568 continue; 1569 1570 if (!(evsel->attr.sample_type & PERF_SAMPLE_CPU)) { 1571 pr_err("File does not contain CPU events. " 1572 "Remove -c option to proceed.\n"); 1573 return -1; 1574 } 1575 } 1576 1577 map = cpu_map__new(cpu_list); 1578 if (map == NULL) { 1579 pr_err("Invalid cpu_list\n"); 1580 return -1; 1581 } 1582 1583 for (i = 0; i < map->nr; i++) { 1584 int cpu = map->map[i]; 1585 1586 if (cpu >= MAX_NR_CPUS) { 1587 pr_err("Requested CPU %d too large. " 1588 "Consider raising MAX_NR_CPUS\n", cpu); 1589 return -1; 1590 } 1591 1592 set_bit(cpu, cpu_bitmap); 1593 } 1594 1595 return 0; 1596 } 1597 1598 void perf_session__fprintf_info(struct perf_session *session, FILE *fp, 1599 bool full) 1600 { 1601 struct stat st; 1602 int ret; 1603 1604 if (session == NULL || fp == NULL) 1605 return; 1606 1607 ret = fstat(session->fd, &st); 1608 if (ret == -1) 1609 return; 1610 1611 fprintf(fp, "# ========\n"); 1612 fprintf(fp, "# captured on: %s", ctime(&st.st_ctime)); 1613 perf_header__fprintf_info(session, fp, full); 1614 fprintf(fp, "# ========\n#\n"); 1615 } 1616 1617 1618 int __perf_session__set_tracepoints_handlers(struct perf_session *session, 1619 const struct perf_evsel_str_handler *assocs, 1620 size_t nr_assocs) 1621 { 1622 struct perf_evlist *evlist = session->evlist; 1623 struct event_format *format; 1624 struct perf_evsel *evsel; 1625 char *tracepoint, *name; 1626 size_t i; 1627 int err; 1628 1629 for (i = 0; i < nr_assocs; i++) { 1630 err = -ENOMEM; 1631 tracepoint = strdup(assocs[i].name); 1632 if (tracepoint == NULL) 1633 goto out; 1634 1635 err = -ENOENT; 1636 name = strchr(tracepoint, ':'); 1637 if (name == NULL) 1638 goto out_free; 1639 1640 *name++ = '\0'; 1641 format = pevent_find_event_by_name(session->pevent, 1642 tracepoint, name); 1643 if (format == NULL) { 1644 /* 1645 * Adding a handler for an event not in the session, 1646 * just ignore it. 1647 */ 1648 goto next; 1649 } 1650 1651 evsel = perf_evlist__find_tracepoint_by_id(evlist, format->id); 1652 if (evsel == NULL) 1653 goto next; 1654 1655 err = -EEXIST; 1656 if (evsel->handler.func != NULL) 1657 goto out_free; 1658 evsel->handler.func = assocs[i].handler; 1659 next: 1660 free(tracepoint); 1661 } 1662 1663 err = 0; 1664 out: 1665 return err; 1666 1667 out_free: 1668 free(tracepoint); 1669 goto out; 1670 } 1671