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 "vdso.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) < 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) < 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 if (!perf_evlist__valid_read_format(self->evlist)) { 75 pr_err("non matching read_format"); 76 goto out_close; 77 } 78 79 self->size = input_stat.st_size; 80 return 0; 81 82 out_close: 83 close(self->fd); 84 self->fd = -1; 85 return -1; 86 } 87 88 void perf_session__set_id_hdr_size(struct perf_session *session) 89 { 90 u16 id_hdr_size = perf_evlist__id_hdr_size(session->evlist); 91 92 machines__set_id_hdr_size(&session->machines, id_hdr_size); 93 } 94 95 int perf_session__create_kernel_maps(struct perf_session *self) 96 { 97 int ret = machine__create_kernel_maps(&self->machines.host); 98 99 if (ret >= 0) 100 ret = machines__create_guest_kernel_maps(&self->machines); 101 return ret; 102 } 103 104 static void perf_session__destroy_kernel_maps(struct perf_session *self) 105 { 106 machines__destroy_kernel_maps(&self->machines); 107 } 108 109 struct perf_session *perf_session__new(const char *filename, int mode, 110 bool force, bool repipe, 111 struct perf_tool *tool) 112 { 113 struct perf_session *self; 114 struct stat st; 115 size_t len; 116 117 if (!filename || !strlen(filename)) { 118 if (!fstat(STDIN_FILENO, &st) && S_ISFIFO(st.st_mode)) 119 filename = "-"; 120 else 121 filename = "perf.data"; 122 } 123 124 len = strlen(filename); 125 self = zalloc(sizeof(*self) + len); 126 127 if (self == NULL) 128 goto out; 129 130 memcpy(self->filename, filename, len); 131 self->repipe = repipe; 132 INIT_LIST_HEAD(&self->ordered_samples.samples); 133 INIT_LIST_HEAD(&self->ordered_samples.sample_cache); 134 INIT_LIST_HEAD(&self->ordered_samples.to_free); 135 machines__init(&self->machines); 136 137 if (mode == O_RDONLY) { 138 if (perf_session__open(self, force) < 0) 139 goto out_delete; 140 perf_session__set_id_hdr_size(self); 141 } else if (mode == O_WRONLY) { 142 /* 143 * In O_RDONLY mode this will be performed when reading the 144 * kernel MMAP event, in perf_event__process_mmap(). 145 */ 146 if (perf_session__create_kernel_maps(self) < 0) 147 goto out_delete; 148 } 149 150 if (tool && tool->ordering_requires_timestamps && 151 tool->ordered_samples && !perf_evlist__sample_id_all(self->evlist)) { 152 dump_printf("WARNING: No sample_id_all support, falling back to unordered processing\n"); 153 tool->ordered_samples = false; 154 } 155 156 out: 157 return self; 158 out_delete: 159 perf_session__delete(self); 160 return NULL; 161 } 162 163 static void perf_session__delete_dead_threads(struct perf_session *session) 164 { 165 machine__delete_dead_threads(&session->machines.host); 166 } 167 168 static void perf_session__delete_threads(struct perf_session *session) 169 { 170 machine__delete_threads(&session->machines.host); 171 } 172 173 static void perf_session_env__delete(struct perf_session_env *env) 174 { 175 free(env->hostname); 176 free(env->os_release); 177 free(env->version); 178 free(env->arch); 179 free(env->cpu_desc); 180 free(env->cpuid); 181 182 free(env->cmdline); 183 free(env->sibling_cores); 184 free(env->sibling_threads); 185 free(env->numa_nodes); 186 free(env->pmu_mappings); 187 } 188 189 void perf_session__delete(struct perf_session *self) 190 { 191 perf_session__destroy_kernel_maps(self); 192 perf_session__delete_dead_threads(self); 193 perf_session__delete_threads(self); 194 perf_session_env__delete(&self->header.env); 195 machines__exit(&self->machines); 196 close(self->fd); 197 free(self); 198 vdso__exit(); 199 } 200 201 static int process_event_synth_tracing_data_stub(struct perf_tool *tool 202 __maybe_unused, 203 union perf_event *event 204 __maybe_unused, 205 struct perf_session *session 206 __maybe_unused) 207 { 208 dump_printf(": unhandled!\n"); 209 return 0; 210 } 211 212 static int process_event_synth_attr_stub(struct perf_tool *tool __maybe_unused, 213 union perf_event *event __maybe_unused, 214 struct perf_evlist **pevlist 215 __maybe_unused) 216 { 217 dump_printf(": unhandled!\n"); 218 return 0; 219 } 220 221 static int process_event_sample_stub(struct perf_tool *tool __maybe_unused, 222 union perf_event *event __maybe_unused, 223 struct perf_sample *sample __maybe_unused, 224 struct perf_evsel *evsel __maybe_unused, 225 struct machine *machine __maybe_unused) 226 { 227 dump_printf(": unhandled!\n"); 228 return 0; 229 } 230 231 static int process_event_stub(struct perf_tool *tool __maybe_unused, 232 union perf_event *event __maybe_unused, 233 struct perf_sample *sample __maybe_unused, 234 struct machine *machine __maybe_unused) 235 { 236 dump_printf(": unhandled!\n"); 237 return 0; 238 } 239 240 static int process_finished_round_stub(struct perf_tool *tool __maybe_unused, 241 union perf_event *event __maybe_unused, 242 struct perf_session *perf_session 243 __maybe_unused) 244 { 245 dump_printf(": unhandled!\n"); 246 return 0; 247 } 248 249 static int process_finished_round(struct perf_tool *tool, 250 union perf_event *event, 251 struct perf_session *session); 252 253 void perf_tool__fill_defaults(struct perf_tool *tool) 254 { 255 if (tool->sample == NULL) 256 tool->sample = process_event_sample_stub; 257 if (tool->mmap == NULL) 258 tool->mmap = process_event_stub; 259 if (tool->mmap2 == NULL) 260 tool->mmap2 = process_event_stub; 261 if (tool->comm == NULL) 262 tool->comm = process_event_stub; 263 if (tool->fork == NULL) 264 tool->fork = process_event_stub; 265 if (tool->exit == NULL) 266 tool->exit = process_event_stub; 267 if (tool->lost == NULL) 268 tool->lost = perf_event__process_lost; 269 if (tool->read == NULL) 270 tool->read = process_event_sample_stub; 271 if (tool->throttle == NULL) 272 tool->throttle = process_event_stub; 273 if (tool->unthrottle == NULL) 274 tool->unthrottle = process_event_stub; 275 if (tool->attr == NULL) 276 tool->attr = process_event_synth_attr_stub; 277 if (tool->tracing_data == NULL) 278 tool->tracing_data = process_event_synth_tracing_data_stub; 279 if (tool->build_id == NULL) 280 tool->build_id = process_finished_round_stub; 281 if (tool->finished_round == NULL) { 282 if (tool->ordered_samples) 283 tool->finished_round = process_finished_round; 284 else 285 tool->finished_round = process_finished_round_stub; 286 } 287 } 288 289 void mem_bswap_32(void *src, int byte_size) 290 { 291 u32 *m = src; 292 while (byte_size > 0) { 293 *m = bswap_32(*m); 294 byte_size -= sizeof(u32); 295 ++m; 296 } 297 } 298 299 void mem_bswap_64(void *src, int byte_size) 300 { 301 u64 *m = src; 302 303 while (byte_size > 0) { 304 *m = bswap_64(*m); 305 byte_size -= sizeof(u64); 306 ++m; 307 } 308 } 309 310 static void swap_sample_id_all(union perf_event *event, void *data) 311 { 312 void *end = (void *) event + event->header.size; 313 int size = end - data; 314 315 BUG_ON(size % sizeof(u64)); 316 mem_bswap_64(data, size); 317 } 318 319 static void perf_event__all64_swap(union perf_event *event, 320 bool sample_id_all __maybe_unused) 321 { 322 struct perf_event_header *hdr = &event->header; 323 mem_bswap_64(hdr + 1, event->header.size - sizeof(*hdr)); 324 } 325 326 static void perf_event__comm_swap(union perf_event *event, bool sample_id_all) 327 { 328 event->comm.pid = bswap_32(event->comm.pid); 329 event->comm.tid = bswap_32(event->comm.tid); 330 331 if (sample_id_all) { 332 void *data = &event->comm.comm; 333 334 data += PERF_ALIGN(strlen(data) + 1, sizeof(u64)); 335 swap_sample_id_all(event, data); 336 } 337 } 338 339 static void perf_event__mmap_swap(union perf_event *event, 340 bool sample_id_all) 341 { 342 event->mmap.pid = bswap_32(event->mmap.pid); 343 event->mmap.tid = bswap_32(event->mmap.tid); 344 event->mmap.start = bswap_64(event->mmap.start); 345 event->mmap.len = bswap_64(event->mmap.len); 346 event->mmap.pgoff = bswap_64(event->mmap.pgoff); 347 348 if (sample_id_all) { 349 void *data = &event->mmap.filename; 350 351 data += PERF_ALIGN(strlen(data) + 1, sizeof(u64)); 352 swap_sample_id_all(event, data); 353 } 354 } 355 356 static void perf_event__mmap2_swap(union perf_event *event, 357 bool sample_id_all) 358 { 359 event->mmap2.pid = bswap_32(event->mmap2.pid); 360 event->mmap2.tid = bswap_32(event->mmap2.tid); 361 event->mmap2.start = bswap_64(event->mmap2.start); 362 event->mmap2.len = bswap_64(event->mmap2.len); 363 event->mmap2.pgoff = bswap_64(event->mmap2.pgoff); 364 event->mmap2.maj = bswap_32(event->mmap2.maj); 365 event->mmap2.min = bswap_32(event->mmap2.min); 366 event->mmap2.ino = bswap_64(event->mmap2.ino); 367 368 if (sample_id_all) { 369 void *data = &event->mmap2.filename; 370 371 data += PERF_ALIGN(strlen(data) + 1, sizeof(u64)); 372 swap_sample_id_all(event, data); 373 } 374 } 375 static void perf_event__task_swap(union perf_event *event, bool sample_id_all) 376 { 377 event->fork.pid = bswap_32(event->fork.pid); 378 event->fork.tid = bswap_32(event->fork.tid); 379 event->fork.ppid = bswap_32(event->fork.ppid); 380 event->fork.ptid = bswap_32(event->fork.ptid); 381 event->fork.time = bswap_64(event->fork.time); 382 383 if (sample_id_all) 384 swap_sample_id_all(event, &event->fork + 1); 385 } 386 387 static void perf_event__read_swap(union perf_event *event, bool sample_id_all) 388 { 389 event->read.pid = bswap_32(event->read.pid); 390 event->read.tid = bswap_32(event->read.tid); 391 event->read.value = bswap_64(event->read.value); 392 event->read.time_enabled = bswap_64(event->read.time_enabled); 393 event->read.time_running = bswap_64(event->read.time_running); 394 event->read.id = bswap_64(event->read.id); 395 396 if (sample_id_all) 397 swap_sample_id_all(event, &event->read + 1); 398 } 399 400 static void perf_event__throttle_swap(union perf_event *event, 401 bool sample_id_all) 402 { 403 event->throttle.time = bswap_64(event->throttle.time); 404 event->throttle.id = bswap_64(event->throttle.id); 405 event->throttle.stream_id = bswap_64(event->throttle.stream_id); 406 407 if (sample_id_all) 408 swap_sample_id_all(event, &event->throttle + 1); 409 } 410 411 static u8 revbyte(u8 b) 412 { 413 int rev = (b >> 4) | ((b & 0xf) << 4); 414 rev = ((rev & 0xcc) >> 2) | ((rev & 0x33) << 2); 415 rev = ((rev & 0xaa) >> 1) | ((rev & 0x55) << 1); 416 return (u8) rev; 417 } 418 419 /* 420 * XXX this is hack in attempt to carry flags bitfield 421 * throught endian village. ABI says: 422 * 423 * Bit-fields are allocated from right to left (least to most significant) 424 * on little-endian implementations and from left to right (most to least 425 * significant) on big-endian implementations. 426 * 427 * The above seems to be byte specific, so we need to reverse each 428 * byte of the bitfield. 'Internet' also says this might be implementation 429 * specific and we probably need proper fix and carry perf_event_attr 430 * bitfield flags in separate data file FEAT_ section. Thought this seems 431 * to work for now. 432 */ 433 static void swap_bitfield(u8 *p, unsigned len) 434 { 435 unsigned i; 436 437 for (i = 0; i < len; i++) { 438 *p = revbyte(*p); 439 p++; 440 } 441 } 442 443 /* exported for swapping attributes in file header */ 444 void perf_event__attr_swap(struct perf_event_attr *attr) 445 { 446 attr->type = bswap_32(attr->type); 447 attr->size = bswap_32(attr->size); 448 attr->config = bswap_64(attr->config); 449 attr->sample_period = bswap_64(attr->sample_period); 450 attr->sample_type = bswap_64(attr->sample_type); 451 attr->read_format = bswap_64(attr->read_format); 452 attr->wakeup_events = bswap_32(attr->wakeup_events); 453 attr->bp_type = bswap_32(attr->bp_type); 454 attr->bp_addr = bswap_64(attr->bp_addr); 455 attr->bp_len = bswap_64(attr->bp_len); 456 attr->branch_sample_type = bswap_64(attr->branch_sample_type); 457 attr->sample_regs_user = bswap_64(attr->sample_regs_user); 458 attr->sample_stack_user = bswap_32(attr->sample_stack_user); 459 460 swap_bitfield((u8 *) (&attr->read_format + 1), sizeof(u64)); 461 } 462 463 static void perf_event__hdr_attr_swap(union perf_event *event, 464 bool sample_id_all __maybe_unused) 465 { 466 size_t size; 467 468 perf_event__attr_swap(&event->attr.attr); 469 470 size = event->header.size; 471 size -= (void *)&event->attr.id - (void *)event; 472 mem_bswap_64(event->attr.id, size); 473 } 474 475 static void perf_event__event_type_swap(union perf_event *event, 476 bool sample_id_all __maybe_unused) 477 { 478 event->event_type.event_type.event_id = 479 bswap_64(event->event_type.event_type.event_id); 480 } 481 482 static void perf_event__tracing_data_swap(union perf_event *event, 483 bool sample_id_all __maybe_unused) 484 { 485 event->tracing_data.size = bswap_32(event->tracing_data.size); 486 } 487 488 typedef void (*perf_event__swap_op)(union perf_event *event, 489 bool sample_id_all); 490 491 static perf_event__swap_op perf_event__swap_ops[] = { 492 [PERF_RECORD_MMAP] = perf_event__mmap_swap, 493 [PERF_RECORD_MMAP2] = perf_event__mmap2_swap, 494 [PERF_RECORD_COMM] = perf_event__comm_swap, 495 [PERF_RECORD_FORK] = perf_event__task_swap, 496 [PERF_RECORD_EXIT] = perf_event__task_swap, 497 [PERF_RECORD_LOST] = perf_event__all64_swap, 498 [PERF_RECORD_READ] = perf_event__read_swap, 499 [PERF_RECORD_THROTTLE] = perf_event__throttle_swap, 500 [PERF_RECORD_UNTHROTTLE] = perf_event__throttle_swap, 501 [PERF_RECORD_SAMPLE] = perf_event__all64_swap, 502 [PERF_RECORD_HEADER_ATTR] = perf_event__hdr_attr_swap, 503 [PERF_RECORD_HEADER_EVENT_TYPE] = perf_event__event_type_swap, 504 [PERF_RECORD_HEADER_TRACING_DATA] = perf_event__tracing_data_swap, 505 [PERF_RECORD_HEADER_BUILD_ID] = NULL, 506 [PERF_RECORD_HEADER_MAX] = NULL, 507 }; 508 509 struct sample_queue { 510 u64 timestamp; 511 u64 file_offset; 512 union perf_event *event; 513 struct list_head list; 514 }; 515 516 static void perf_session_free_sample_buffers(struct perf_session *session) 517 { 518 struct ordered_samples *os = &session->ordered_samples; 519 520 while (!list_empty(&os->to_free)) { 521 struct sample_queue *sq; 522 523 sq = list_entry(os->to_free.next, struct sample_queue, list); 524 list_del(&sq->list); 525 free(sq); 526 } 527 } 528 529 static int perf_session_deliver_event(struct perf_session *session, 530 union perf_event *event, 531 struct perf_sample *sample, 532 struct perf_tool *tool, 533 u64 file_offset); 534 535 static int flush_sample_queue(struct perf_session *s, 536 struct perf_tool *tool) 537 { 538 struct ordered_samples *os = &s->ordered_samples; 539 struct list_head *head = &os->samples; 540 struct sample_queue *tmp, *iter; 541 struct perf_sample sample; 542 u64 limit = os->next_flush; 543 u64 last_ts = os->last_sample ? os->last_sample->timestamp : 0ULL; 544 unsigned idx = 0, progress_next = os->nr_samples / 16; 545 bool show_progress = limit == ULLONG_MAX; 546 int ret; 547 548 if (!tool->ordered_samples || !limit) 549 return 0; 550 551 list_for_each_entry_safe(iter, tmp, head, list) { 552 if (session_done()) 553 return 0; 554 555 if (iter->timestamp > limit) 556 break; 557 558 ret = perf_evlist__parse_sample(s->evlist, iter->event, &sample); 559 if (ret) 560 pr_err("Can't parse sample, err = %d\n", ret); 561 else { 562 ret = perf_session_deliver_event(s, iter->event, &sample, tool, 563 iter->file_offset); 564 if (ret) 565 return ret; 566 } 567 568 os->last_flush = iter->timestamp; 569 list_del(&iter->list); 570 list_add(&iter->list, &os->sample_cache); 571 if (show_progress && (++idx >= progress_next)) { 572 progress_next += os->nr_samples / 16; 573 ui_progress__update(idx, os->nr_samples, 574 "Processing time ordered events..."); 575 } 576 } 577 578 if (list_empty(head)) { 579 os->last_sample = NULL; 580 } else if (last_ts <= limit) { 581 os->last_sample = 582 list_entry(head->prev, struct sample_queue, list); 583 } 584 585 os->nr_samples = 0; 586 587 return 0; 588 } 589 590 /* 591 * When perf record finishes a pass on every buffers, it records this pseudo 592 * event. 593 * We record the max timestamp t found in the pass n. 594 * Assuming these timestamps are monotonic across cpus, we know that if 595 * a buffer still has events with timestamps below t, they will be all 596 * available and then read in the pass n + 1. 597 * Hence when we start to read the pass n + 2, we can safely flush every 598 * events with timestamps below t. 599 * 600 * ============ PASS n ================= 601 * CPU 0 | CPU 1 602 * | 603 * cnt1 timestamps | cnt2 timestamps 604 * 1 | 2 605 * 2 | 3 606 * - | 4 <--- max recorded 607 * 608 * ============ PASS n + 1 ============== 609 * CPU 0 | CPU 1 610 * | 611 * cnt1 timestamps | cnt2 timestamps 612 * 3 | 5 613 * 4 | 6 614 * 5 | 7 <---- max recorded 615 * 616 * Flush every events below timestamp 4 617 * 618 * ============ PASS n + 2 ============== 619 * CPU 0 | CPU 1 620 * | 621 * cnt1 timestamps | cnt2 timestamps 622 * 6 | 8 623 * 7 | 9 624 * - | 10 625 * 626 * Flush every events below timestamp 7 627 * etc... 628 */ 629 static int process_finished_round(struct perf_tool *tool, 630 union perf_event *event __maybe_unused, 631 struct perf_session *session) 632 { 633 int ret = flush_sample_queue(session, tool); 634 if (!ret) 635 session->ordered_samples.next_flush = session->ordered_samples.max_timestamp; 636 637 return ret; 638 } 639 640 /* The queue is ordered by time */ 641 static void __queue_event(struct sample_queue *new, struct perf_session *s) 642 { 643 struct ordered_samples *os = &s->ordered_samples; 644 struct sample_queue *sample = os->last_sample; 645 u64 timestamp = new->timestamp; 646 struct list_head *p; 647 648 ++os->nr_samples; 649 os->last_sample = new; 650 651 if (!sample) { 652 list_add(&new->list, &os->samples); 653 os->max_timestamp = timestamp; 654 return; 655 } 656 657 /* 658 * last_sample might point to some random place in the list as it's 659 * the last queued event. We expect that the new event is close to 660 * this. 661 */ 662 if (sample->timestamp <= timestamp) { 663 while (sample->timestamp <= timestamp) { 664 p = sample->list.next; 665 if (p == &os->samples) { 666 list_add_tail(&new->list, &os->samples); 667 os->max_timestamp = timestamp; 668 return; 669 } 670 sample = list_entry(p, struct sample_queue, list); 671 } 672 list_add_tail(&new->list, &sample->list); 673 } else { 674 while (sample->timestamp > timestamp) { 675 p = sample->list.prev; 676 if (p == &os->samples) { 677 list_add(&new->list, &os->samples); 678 return; 679 } 680 sample = list_entry(p, struct sample_queue, list); 681 } 682 list_add(&new->list, &sample->list); 683 } 684 } 685 686 #define MAX_SAMPLE_BUFFER (64 * 1024 / sizeof(struct sample_queue)) 687 688 int perf_session_queue_event(struct perf_session *s, union perf_event *event, 689 struct perf_sample *sample, u64 file_offset) 690 { 691 struct ordered_samples *os = &s->ordered_samples; 692 struct list_head *sc = &os->sample_cache; 693 u64 timestamp = sample->time; 694 struct sample_queue *new; 695 696 if (!timestamp || timestamp == ~0ULL) 697 return -ETIME; 698 699 if (timestamp < s->ordered_samples.last_flush) { 700 printf("Warning: Timestamp below last timeslice flush\n"); 701 return -EINVAL; 702 } 703 704 if (!list_empty(sc)) { 705 new = list_entry(sc->next, struct sample_queue, list); 706 list_del(&new->list); 707 } else if (os->sample_buffer) { 708 new = os->sample_buffer + os->sample_buffer_idx; 709 if (++os->sample_buffer_idx == MAX_SAMPLE_BUFFER) 710 os->sample_buffer = NULL; 711 } else { 712 os->sample_buffer = malloc(MAX_SAMPLE_BUFFER * sizeof(*new)); 713 if (!os->sample_buffer) 714 return -ENOMEM; 715 list_add(&os->sample_buffer->list, &os->to_free); 716 os->sample_buffer_idx = 2; 717 new = os->sample_buffer + 1; 718 } 719 720 new->timestamp = timestamp; 721 new->file_offset = file_offset; 722 new->event = event; 723 724 __queue_event(new, s); 725 726 return 0; 727 } 728 729 static void callchain__printf(struct perf_sample *sample) 730 { 731 unsigned int i; 732 733 printf("... chain: nr:%" PRIu64 "\n", sample->callchain->nr); 734 735 for (i = 0; i < sample->callchain->nr; i++) 736 printf("..... %2d: %016" PRIx64 "\n", 737 i, sample->callchain->ips[i]); 738 } 739 740 static void branch_stack__printf(struct perf_sample *sample) 741 { 742 uint64_t i; 743 744 printf("... branch stack: nr:%" PRIu64 "\n", sample->branch_stack->nr); 745 746 for (i = 0; i < sample->branch_stack->nr; i++) 747 printf("..... %2"PRIu64": %016" PRIx64 " -> %016" PRIx64 "\n", 748 i, sample->branch_stack->entries[i].from, 749 sample->branch_stack->entries[i].to); 750 } 751 752 static void regs_dump__printf(u64 mask, u64 *regs) 753 { 754 unsigned rid, i = 0; 755 756 for_each_set_bit(rid, (unsigned long *) &mask, sizeof(mask) * 8) { 757 u64 val = regs[i++]; 758 759 printf(".... %-5s 0x%" PRIx64 "\n", 760 perf_reg_name(rid), val); 761 } 762 } 763 764 static void regs_user__printf(struct perf_sample *sample, u64 mask) 765 { 766 struct regs_dump *user_regs = &sample->user_regs; 767 768 if (user_regs->regs) { 769 printf("... user regs: mask 0x%" PRIx64 "\n", mask); 770 regs_dump__printf(mask, user_regs->regs); 771 } 772 } 773 774 static void stack_user__printf(struct stack_dump *dump) 775 { 776 printf("... ustack: size %" PRIu64 ", offset 0x%x\n", 777 dump->size, dump->offset); 778 } 779 780 static void perf_session__print_tstamp(struct perf_session *session, 781 union perf_event *event, 782 struct perf_sample *sample) 783 { 784 u64 sample_type = __perf_evlist__combined_sample_type(session->evlist); 785 786 if (event->header.type != PERF_RECORD_SAMPLE && 787 !perf_evlist__sample_id_all(session->evlist)) { 788 fputs("-1 -1 ", stdout); 789 return; 790 } 791 792 if ((sample_type & PERF_SAMPLE_CPU)) 793 printf("%u ", sample->cpu); 794 795 if (sample_type & PERF_SAMPLE_TIME) 796 printf("%" PRIu64 " ", sample->time); 797 } 798 799 static void sample_read__printf(struct perf_sample *sample, u64 read_format) 800 { 801 printf("... sample_read:\n"); 802 803 if (read_format & PERF_FORMAT_TOTAL_TIME_ENABLED) 804 printf("...... time enabled %016" PRIx64 "\n", 805 sample->read.time_enabled); 806 807 if (read_format & PERF_FORMAT_TOTAL_TIME_RUNNING) 808 printf("...... time running %016" PRIx64 "\n", 809 sample->read.time_running); 810 811 if (read_format & PERF_FORMAT_GROUP) { 812 u64 i; 813 814 printf(".... group nr %" PRIu64 "\n", sample->read.group.nr); 815 816 for (i = 0; i < sample->read.group.nr; i++) { 817 struct sample_read_value *value; 818 819 value = &sample->read.group.values[i]; 820 printf("..... id %016" PRIx64 821 ", value %016" PRIx64 "\n", 822 value->id, value->value); 823 } 824 } else 825 printf("..... id %016" PRIx64 ", value %016" PRIx64 "\n", 826 sample->read.one.id, sample->read.one.value); 827 } 828 829 static void dump_event(struct perf_session *session, union perf_event *event, 830 u64 file_offset, struct perf_sample *sample) 831 { 832 if (!dump_trace) 833 return; 834 835 printf("\n%#" PRIx64 " [%#x]: event: %d\n", 836 file_offset, event->header.size, event->header.type); 837 838 trace_event(event); 839 840 if (sample) 841 perf_session__print_tstamp(session, event, sample); 842 843 printf("%#" PRIx64 " [%#x]: PERF_RECORD_%s", file_offset, 844 event->header.size, perf_event__name(event->header.type)); 845 } 846 847 static void dump_sample(struct perf_evsel *evsel, union perf_event *event, 848 struct perf_sample *sample) 849 { 850 u64 sample_type; 851 852 if (!dump_trace) 853 return; 854 855 printf("(IP, %d): %d/%d: %#" PRIx64 " period: %" PRIu64 " addr: %#" PRIx64 "\n", 856 event->header.misc, sample->pid, sample->tid, sample->ip, 857 sample->period, sample->addr); 858 859 sample_type = evsel->attr.sample_type; 860 861 if (sample_type & PERF_SAMPLE_CALLCHAIN) 862 callchain__printf(sample); 863 864 if (sample_type & PERF_SAMPLE_BRANCH_STACK) 865 branch_stack__printf(sample); 866 867 if (sample_type & PERF_SAMPLE_REGS_USER) 868 regs_user__printf(sample, evsel->attr.sample_regs_user); 869 870 if (sample_type & PERF_SAMPLE_STACK_USER) 871 stack_user__printf(&sample->user_stack); 872 873 if (sample_type & PERF_SAMPLE_WEIGHT) 874 printf("... weight: %" PRIu64 "\n", sample->weight); 875 876 if (sample_type & PERF_SAMPLE_DATA_SRC) 877 printf(" . data_src: 0x%"PRIx64"\n", sample->data_src); 878 879 if (sample_type & PERF_SAMPLE_TRANSACTION) 880 printf("... transaction: %" PRIx64 "\n", sample->transaction); 881 882 if (sample_type & PERF_SAMPLE_READ) 883 sample_read__printf(sample, evsel->attr.read_format); 884 } 885 886 static struct machine * 887 perf_session__find_machine_for_cpumode(struct perf_session *session, 888 union perf_event *event, 889 struct perf_sample *sample) 890 { 891 const u8 cpumode = event->header.misc & PERF_RECORD_MISC_CPUMODE_MASK; 892 893 if (perf_guest && 894 ((cpumode == PERF_RECORD_MISC_GUEST_KERNEL) || 895 (cpumode == PERF_RECORD_MISC_GUEST_USER))) { 896 u32 pid; 897 898 if (event->header.type == PERF_RECORD_MMAP 899 || event->header.type == PERF_RECORD_MMAP2) 900 pid = event->mmap.pid; 901 else 902 pid = sample->pid; 903 904 return perf_session__findnew_machine(session, pid); 905 } 906 907 return &session->machines.host; 908 } 909 910 static int deliver_sample_value(struct perf_session *session, 911 struct perf_tool *tool, 912 union perf_event *event, 913 struct perf_sample *sample, 914 struct sample_read_value *v, 915 struct machine *machine) 916 { 917 struct perf_sample_id *sid; 918 919 sid = perf_evlist__id2sid(session->evlist, v->id); 920 if (sid) { 921 sample->id = v->id; 922 sample->period = v->value - sid->period; 923 sid->period = v->value; 924 } 925 926 if (!sid || sid->evsel == NULL) { 927 ++session->stats.nr_unknown_id; 928 return 0; 929 } 930 931 return tool->sample(tool, event, sample, sid->evsel, machine); 932 } 933 934 static int deliver_sample_group(struct perf_session *session, 935 struct perf_tool *tool, 936 union perf_event *event, 937 struct perf_sample *sample, 938 struct machine *machine) 939 { 940 int ret = -EINVAL; 941 u64 i; 942 943 for (i = 0; i < sample->read.group.nr; i++) { 944 ret = deliver_sample_value(session, tool, event, sample, 945 &sample->read.group.values[i], 946 machine); 947 if (ret) 948 break; 949 } 950 951 return ret; 952 } 953 954 static int 955 perf_session__deliver_sample(struct perf_session *session, 956 struct perf_tool *tool, 957 union perf_event *event, 958 struct perf_sample *sample, 959 struct perf_evsel *evsel, 960 struct machine *machine) 961 { 962 /* We know evsel != NULL. */ 963 u64 sample_type = evsel->attr.sample_type; 964 u64 read_format = evsel->attr.read_format; 965 966 /* Standard sample delievery. */ 967 if (!(sample_type & PERF_SAMPLE_READ)) 968 return tool->sample(tool, event, sample, evsel, machine); 969 970 /* For PERF_SAMPLE_READ we have either single or group mode. */ 971 if (read_format & PERF_FORMAT_GROUP) 972 return deliver_sample_group(session, tool, event, sample, 973 machine); 974 else 975 return deliver_sample_value(session, tool, event, sample, 976 &sample->read.one, machine); 977 } 978 979 static int perf_session_deliver_event(struct perf_session *session, 980 union perf_event *event, 981 struct perf_sample *sample, 982 struct perf_tool *tool, 983 u64 file_offset) 984 { 985 struct perf_evsel *evsel; 986 struct machine *machine; 987 988 dump_event(session, event, file_offset, sample); 989 990 evsel = perf_evlist__id2evsel(session->evlist, sample->id); 991 if (evsel != NULL && event->header.type != PERF_RECORD_SAMPLE) { 992 /* 993 * XXX We're leaving PERF_RECORD_SAMPLE unnacounted here 994 * because the tools right now may apply filters, discarding 995 * some of the samples. For consistency, in the future we 996 * should have something like nr_filtered_samples and remove 997 * the sample->period from total_sample_period, etc, KISS for 998 * now tho. 999 * 1000 * Also testing against NULL allows us to handle files without 1001 * attr.sample_id_all and/or without PERF_SAMPLE_ID. In the 1002 * future probably it'll be a good idea to restrict event 1003 * processing via perf_session to files with both set. 1004 */ 1005 hists__inc_nr_events(&evsel->hists, event->header.type); 1006 } 1007 1008 machine = perf_session__find_machine_for_cpumode(session, event, 1009 sample); 1010 1011 switch (event->header.type) { 1012 case PERF_RECORD_SAMPLE: 1013 dump_sample(evsel, event, sample); 1014 if (evsel == NULL) { 1015 ++session->stats.nr_unknown_id; 1016 return 0; 1017 } 1018 if (machine == NULL) { 1019 ++session->stats.nr_unprocessable_samples; 1020 return 0; 1021 } 1022 return perf_session__deliver_sample(session, tool, event, 1023 sample, evsel, machine); 1024 case PERF_RECORD_MMAP: 1025 return tool->mmap(tool, event, sample, machine); 1026 case PERF_RECORD_MMAP2: 1027 return tool->mmap2(tool, event, sample, machine); 1028 case PERF_RECORD_COMM: 1029 return tool->comm(tool, event, sample, machine); 1030 case PERF_RECORD_FORK: 1031 return tool->fork(tool, event, sample, machine); 1032 case PERF_RECORD_EXIT: 1033 return tool->exit(tool, event, sample, machine); 1034 case PERF_RECORD_LOST: 1035 if (tool->lost == perf_event__process_lost) 1036 session->stats.total_lost += event->lost.lost; 1037 return tool->lost(tool, event, sample, machine); 1038 case PERF_RECORD_READ: 1039 return tool->read(tool, event, sample, evsel, machine); 1040 case PERF_RECORD_THROTTLE: 1041 return tool->throttle(tool, event, sample, machine); 1042 case PERF_RECORD_UNTHROTTLE: 1043 return tool->unthrottle(tool, event, sample, machine); 1044 default: 1045 ++session->stats.nr_unknown_events; 1046 return -1; 1047 } 1048 } 1049 1050 static int perf_session__process_user_event(struct perf_session *session, union perf_event *event, 1051 struct perf_tool *tool, u64 file_offset) 1052 { 1053 int err; 1054 1055 dump_event(session, event, file_offset, NULL); 1056 1057 /* These events are processed right away */ 1058 switch (event->header.type) { 1059 case PERF_RECORD_HEADER_ATTR: 1060 err = tool->attr(tool, event, &session->evlist); 1061 if (err == 0) 1062 perf_session__set_id_hdr_size(session); 1063 return err; 1064 case PERF_RECORD_HEADER_TRACING_DATA: 1065 /* setup for reading amidst mmap */ 1066 lseek(session->fd, file_offset, SEEK_SET); 1067 return tool->tracing_data(tool, event, session); 1068 case PERF_RECORD_HEADER_BUILD_ID: 1069 return tool->build_id(tool, event, session); 1070 case PERF_RECORD_FINISHED_ROUND: 1071 return tool->finished_round(tool, event, session); 1072 default: 1073 return -EINVAL; 1074 } 1075 } 1076 1077 static void event_swap(union perf_event *event, bool sample_id_all) 1078 { 1079 perf_event__swap_op swap; 1080 1081 swap = perf_event__swap_ops[event->header.type]; 1082 if (swap) 1083 swap(event, sample_id_all); 1084 } 1085 1086 static int perf_session__process_event(struct perf_session *session, 1087 union perf_event *event, 1088 struct perf_tool *tool, 1089 u64 file_offset) 1090 { 1091 struct perf_sample sample; 1092 int ret; 1093 1094 if (session->header.needs_swap) 1095 event_swap(event, perf_evlist__sample_id_all(session->evlist)); 1096 1097 if (event->header.type >= PERF_RECORD_HEADER_MAX) 1098 return -EINVAL; 1099 1100 events_stats__inc(&session->stats, event->header.type); 1101 1102 if (event->header.type >= PERF_RECORD_USER_TYPE_START) 1103 return perf_session__process_user_event(session, event, tool, file_offset); 1104 1105 /* 1106 * For all kernel events we get the sample data 1107 */ 1108 ret = perf_evlist__parse_sample(session->evlist, event, &sample); 1109 if (ret) 1110 return ret; 1111 1112 if (tool->ordered_samples) { 1113 ret = perf_session_queue_event(session, event, &sample, 1114 file_offset); 1115 if (ret != -ETIME) 1116 return ret; 1117 } 1118 1119 return perf_session_deliver_event(session, event, &sample, tool, 1120 file_offset); 1121 } 1122 1123 void perf_event_header__bswap(struct perf_event_header *self) 1124 { 1125 self->type = bswap_32(self->type); 1126 self->misc = bswap_16(self->misc); 1127 self->size = bswap_16(self->size); 1128 } 1129 1130 struct thread *perf_session__findnew(struct perf_session *session, pid_t pid) 1131 { 1132 return machine__findnew_thread(&session->machines.host, 0, pid); 1133 } 1134 1135 static struct thread *perf_session__register_idle_thread(struct perf_session *self) 1136 { 1137 struct thread *thread = perf_session__findnew(self, 0); 1138 1139 if (thread == NULL || thread__set_comm(thread, "swapper")) { 1140 pr_err("problem inserting idle task.\n"); 1141 thread = NULL; 1142 } 1143 1144 return thread; 1145 } 1146 1147 static void perf_session__warn_about_errors(const struct perf_session *session, 1148 const struct perf_tool *tool) 1149 { 1150 if (tool->lost == perf_event__process_lost && 1151 session->stats.nr_events[PERF_RECORD_LOST] != 0) { 1152 ui__warning("Processed %d events and lost %d chunks!\n\n" 1153 "Check IO/CPU overload!\n\n", 1154 session->stats.nr_events[0], 1155 session->stats.nr_events[PERF_RECORD_LOST]); 1156 } 1157 1158 if (session->stats.nr_unknown_events != 0) { 1159 ui__warning("Found %u unknown events!\n\n" 1160 "Is this an older tool processing a perf.data " 1161 "file generated by a more recent tool?\n\n" 1162 "If that is not the case, consider " 1163 "reporting to linux-kernel@vger.kernel.org.\n\n", 1164 session->stats.nr_unknown_events); 1165 } 1166 1167 if (session->stats.nr_unknown_id != 0) { 1168 ui__warning("%u samples with id not present in the header\n", 1169 session->stats.nr_unknown_id); 1170 } 1171 1172 if (session->stats.nr_invalid_chains != 0) { 1173 ui__warning("Found invalid callchains!\n\n" 1174 "%u out of %u events were discarded for this reason.\n\n" 1175 "Consider reporting to linux-kernel@vger.kernel.org.\n\n", 1176 session->stats.nr_invalid_chains, 1177 session->stats.nr_events[PERF_RECORD_SAMPLE]); 1178 } 1179 1180 if (session->stats.nr_unprocessable_samples != 0) { 1181 ui__warning("%u unprocessable samples recorded.\n" 1182 "Do you have a KVM guest running and not using 'perf kvm'?\n", 1183 session->stats.nr_unprocessable_samples); 1184 } 1185 } 1186 1187 volatile int session_done; 1188 1189 static int __perf_session__process_pipe_events(struct perf_session *self, 1190 struct perf_tool *tool) 1191 { 1192 union perf_event *event; 1193 uint32_t size, cur_size = 0; 1194 void *buf = NULL; 1195 int skip = 0; 1196 u64 head; 1197 int err; 1198 void *p; 1199 1200 perf_tool__fill_defaults(tool); 1201 1202 head = 0; 1203 cur_size = sizeof(union perf_event); 1204 1205 buf = malloc(cur_size); 1206 if (!buf) 1207 return -errno; 1208 more: 1209 event = buf; 1210 err = readn(self->fd, event, sizeof(struct perf_event_header)); 1211 if (err <= 0) { 1212 if (err == 0) 1213 goto done; 1214 1215 pr_err("failed to read event header\n"); 1216 goto out_err; 1217 } 1218 1219 if (self->header.needs_swap) 1220 perf_event_header__bswap(&event->header); 1221 1222 size = event->header.size; 1223 if (size < sizeof(struct perf_event_header)) { 1224 pr_err("bad event header size\n"); 1225 goto out_err; 1226 } 1227 1228 if (size > cur_size) { 1229 void *new = realloc(buf, size); 1230 if (!new) { 1231 pr_err("failed to allocate memory to read event\n"); 1232 goto out_err; 1233 } 1234 buf = new; 1235 cur_size = size; 1236 event = buf; 1237 } 1238 p = event; 1239 p += sizeof(struct perf_event_header); 1240 1241 if (size - sizeof(struct perf_event_header)) { 1242 err = readn(self->fd, p, size - sizeof(struct perf_event_header)); 1243 if (err <= 0) { 1244 if (err == 0) { 1245 pr_err("unexpected end of event stream\n"); 1246 goto done; 1247 } 1248 1249 pr_err("failed to read event data\n"); 1250 goto out_err; 1251 } 1252 } 1253 1254 if ((skip = perf_session__process_event(self, event, tool, head)) < 0) { 1255 pr_err("%#" PRIx64 " [%#x]: failed to process type: %d\n", 1256 head, event->header.size, event->header.type); 1257 err = -EINVAL; 1258 goto out_err; 1259 } 1260 1261 head += size; 1262 1263 if (skip > 0) 1264 head += skip; 1265 1266 if (!session_done()) 1267 goto more; 1268 done: 1269 /* do the final flush for ordered samples */ 1270 self->ordered_samples.next_flush = ULLONG_MAX; 1271 err = flush_sample_queue(self, tool); 1272 out_err: 1273 free(buf); 1274 perf_session__warn_about_errors(self, tool); 1275 perf_session_free_sample_buffers(self); 1276 return err; 1277 } 1278 1279 static union perf_event * 1280 fetch_mmaped_event(struct perf_session *session, 1281 u64 head, size_t mmap_size, char *buf) 1282 { 1283 union perf_event *event; 1284 1285 /* 1286 * Ensure we have enough space remaining to read 1287 * the size of the event in the headers. 1288 */ 1289 if (head + sizeof(event->header) > mmap_size) 1290 return NULL; 1291 1292 event = (union perf_event *)(buf + head); 1293 1294 if (session->header.needs_swap) 1295 perf_event_header__bswap(&event->header); 1296 1297 if (head + event->header.size > mmap_size) { 1298 /* We're not fetching the event so swap back again */ 1299 if (session->header.needs_swap) 1300 perf_event_header__bswap(&event->header); 1301 return NULL; 1302 } 1303 1304 return event; 1305 } 1306 1307 /* 1308 * On 64bit we can mmap the data file in one go. No need for tiny mmap 1309 * slices. On 32bit we use 32MB. 1310 */ 1311 #if BITS_PER_LONG == 64 1312 #define MMAP_SIZE ULLONG_MAX 1313 #define NUM_MMAPS 1 1314 #else 1315 #define MMAP_SIZE (32 * 1024 * 1024ULL) 1316 #define NUM_MMAPS 128 1317 #endif 1318 1319 int __perf_session__process_events(struct perf_session *session, 1320 u64 data_offset, u64 data_size, 1321 u64 file_size, struct perf_tool *tool) 1322 { 1323 u64 head, page_offset, file_offset, file_pos, progress_next; 1324 int err, mmap_prot, mmap_flags, map_idx = 0; 1325 size_t mmap_size; 1326 char *buf, *mmaps[NUM_MMAPS]; 1327 union perf_event *event; 1328 uint32_t size; 1329 1330 perf_tool__fill_defaults(tool); 1331 1332 page_offset = page_size * (data_offset / page_size); 1333 file_offset = page_offset; 1334 head = data_offset - page_offset; 1335 1336 if (data_size && (data_offset + data_size < file_size)) 1337 file_size = data_offset + data_size; 1338 1339 progress_next = file_size / 16; 1340 1341 mmap_size = MMAP_SIZE; 1342 if (mmap_size > file_size) 1343 mmap_size = file_size; 1344 1345 memset(mmaps, 0, sizeof(mmaps)); 1346 1347 mmap_prot = PROT_READ; 1348 mmap_flags = MAP_SHARED; 1349 1350 if (session->header.needs_swap) { 1351 mmap_prot |= PROT_WRITE; 1352 mmap_flags = MAP_PRIVATE; 1353 } 1354 remap: 1355 buf = mmap(NULL, mmap_size, mmap_prot, mmap_flags, session->fd, 1356 file_offset); 1357 if (buf == MAP_FAILED) { 1358 pr_err("failed to mmap file\n"); 1359 err = -errno; 1360 goto out_err; 1361 } 1362 mmaps[map_idx] = buf; 1363 map_idx = (map_idx + 1) & (ARRAY_SIZE(mmaps) - 1); 1364 file_pos = file_offset + head; 1365 1366 more: 1367 event = fetch_mmaped_event(session, head, mmap_size, buf); 1368 if (!event) { 1369 if (mmaps[map_idx]) { 1370 munmap(mmaps[map_idx], mmap_size); 1371 mmaps[map_idx] = NULL; 1372 } 1373 1374 page_offset = page_size * (head / page_size); 1375 file_offset += page_offset; 1376 head -= page_offset; 1377 goto remap; 1378 } 1379 1380 size = event->header.size; 1381 1382 if (size < sizeof(struct perf_event_header) || 1383 perf_session__process_event(session, event, tool, file_pos) < 0) { 1384 pr_err("%#" PRIx64 " [%#x]: failed to process type: %d\n", 1385 file_offset + head, event->header.size, 1386 event->header.type); 1387 err = -EINVAL; 1388 goto out_err; 1389 } 1390 1391 head += size; 1392 file_pos += size; 1393 1394 if (file_pos >= progress_next) { 1395 progress_next += file_size / 16; 1396 ui_progress__update(file_pos, file_size, 1397 "Processing events..."); 1398 } 1399 1400 if (session_done()) 1401 goto out; 1402 1403 if (file_pos < file_size) 1404 goto more; 1405 1406 out: 1407 /* do the final flush for ordered samples */ 1408 session->ordered_samples.next_flush = ULLONG_MAX; 1409 err = flush_sample_queue(session, tool); 1410 out_err: 1411 ui_progress__finish(); 1412 perf_session__warn_about_errors(session, tool); 1413 perf_session_free_sample_buffers(session); 1414 return err; 1415 } 1416 1417 int perf_session__process_events(struct perf_session *self, 1418 struct perf_tool *tool) 1419 { 1420 int err; 1421 1422 if (perf_session__register_idle_thread(self) == NULL) 1423 return -ENOMEM; 1424 1425 if (!self->fd_pipe) 1426 err = __perf_session__process_events(self, 1427 self->header.data_offset, 1428 self->header.data_size, 1429 self->size, tool); 1430 else 1431 err = __perf_session__process_pipe_events(self, tool); 1432 1433 return err; 1434 } 1435 1436 bool perf_session__has_traces(struct perf_session *session, const char *msg) 1437 { 1438 struct perf_evsel *evsel; 1439 1440 list_for_each_entry(evsel, &session->evlist->entries, node) { 1441 if (evsel->attr.type == PERF_TYPE_TRACEPOINT) 1442 return true; 1443 } 1444 1445 pr_err("No trace sample to read. Did you call 'perf %s'?\n", msg); 1446 return false; 1447 } 1448 1449 int maps__set_kallsyms_ref_reloc_sym(struct map **maps, 1450 const char *symbol_name, u64 addr) 1451 { 1452 char *bracket; 1453 enum map_type i; 1454 struct ref_reloc_sym *ref; 1455 1456 ref = zalloc(sizeof(struct ref_reloc_sym)); 1457 if (ref == NULL) 1458 return -ENOMEM; 1459 1460 ref->name = strdup(symbol_name); 1461 if (ref->name == NULL) { 1462 free(ref); 1463 return -ENOMEM; 1464 } 1465 1466 bracket = strchr(ref->name, ']'); 1467 if (bracket) 1468 *bracket = '\0'; 1469 1470 ref->addr = addr; 1471 1472 for (i = 0; i < MAP__NR_TYPES; ++i) { 1473 struct kmap *kmap = map__kmap(maps[i]); 1474 kmap->ref_reloc_sym = ref; 1475 } 1476 1477 return 0; 1478 } 1479 1480 size_t perf_session__fprintf_dsos(struct perf_session *self, FILE *fp) 1481 { 1482 return machines__fprintf_dsos(&self->machines, fp); 1483 } 1484 1485 size_t perf_session__fprintf_dsos_buildid(struct perf_session *self, FILE *fp, 1486 bool (skip)(struct dso *dso, int parm), int parm) 1487 { 1488 return machines__fprintf_dsos_buildid(&self->machines, fp, skip, parm); 1489 } 1490 1491 size_t perf_session__fprintf_nr_events(struct perf_session *session, FILE *fp) 1492 { 1493 struct perf_evsel *pos; 1494 size_t ret = fprintf(fp, "Aggregated stats:\n"); 1495 1496 ret += events_stats__fprintf(&session->stats, fp); 1497 1498 list_for_each_entry(pos, &session->evlist->entries, node) { 1499 ret += fprintf(fp, "%s stats:\n", perf_evsel__name(pos)); 1500 ret += events_stats__fprintf(&pos->hists.stats, fp); 1501 } 1502 1503 return ret; 1504 } 1505 1506 size_t perf_session__fprintf(struct perf_session *session, FILE *fp) 1507 { 1508 /* 1509 * FIXME: Here we have to actually print all the machines in this 1510 * session, not just the host... 1511 */ 1512 return machine__fprintf(&session->machines.host, fp); 1513 } 1514 1515 struct perf_evsel *perf_session__find_first_evtype(struct perf_session *session, 1516 unsigned int type) 1517 { 1518 struct perf_evsel *pos; 1519 1520 list_for_each_entry(pos, &session->evlist->entries, node) { 1521 if (pos->attr.type == type) 1522 return pos; 1523 } 1524 return NULL; 1525 } 1526 1527 void perf_evsel__print_ip(struct perf_evsel *evsel, union perf_event *event, 1528 struct perf_sample *sample, struct machine *machine, 1529 unsigned int print_opts, unsigned int stack_depth) 1530 { 1531 struct addr_location al; 1532 struct callchain_cursor_node *node; 1533 int print_ip = print_opts & PRINT_IP_OPT_IP; 1534 int print_sym = print_opts & PRINT_IP_OPT_SYM; 1535 int print_dso = print_opts & PRINT_IP_OPT_DSO; 1536 int print_symoffset = print_opts & PRINT_IP_OPT_SYMOFFSET; 1537 int print_oneline = print_opts & PRINT_IP_OPT_ONELINE; 1538 char s = print_oneline ? ' ' : '\t'; 1539 1540 if (perf_event__preprocess_sample(event, machine, &al, sample) < 0) { 1541 error("problem processing %d event, skipping it.\n", 1542 event->header.type); 1543 return; 1544 } 1545 1546 if (symbol_conf.use_callchain && sample->callchain) { 1547 1548 if (machine__resolve_callchain(machine, evsel, al.thread, 1549 sample, NULL, NULL) != 0) { 1550 if (verbose) 1551 error("Failed to resolve callchain. Skipping\n"); 1552 return; 1553 } 1554 callchain_cursor_commit(&callchain_cursor); 1555 1556 while (stack_depth) { 1557 node = callchain_cursor_current(&callchain_cursor); 1558 if (!node) 1559 break; 1560 1561 if (print_ip) 1562 printf("%c%16" PRIx64, s, node->ip); 1563 1564 if (print_sym) { 1565 printf(" "); 1566 if (print_symoffset) { 1567 al.addr = node->ip; 1568 al.map = node->map; 1569 symbol__fprintf_symname_offs(node->sym, &al, stdout); 1570 } else 1571 symbol__fprintf_symname(node->sym, stdout); 1572 } 1573 1574 if (print_dso) { 1575 printf(" ("); 1576 map__fprintf_dsoname(node->map, stdout); 1577 printf(")"); 1578 } 1579 1580 if (!print_oneline) 1581 printf("\n"); 1582 1583 callchain_cursor_advance(&callchain_cursor); 1584 1585 stack_depth--; 1586 } 1587 1588 } else { 1589 if (print_ip) 1590 printf("%16" PRIx64, sample->ip); 1591 1592 if (print_sym) { 1593 printf(" "); 1594 if (print_symoffset) 1595 symbol__fprintf_symname_offs(al.sym, &al, 1596 stdout); 1597 else 1598 symbol__fprintf_symname(al.sym, stdout); 1599 } 1600 1601 if (print_dso) { 1602 printf(" ("); 1603 map__fprintf_dsoname(al.map, stdout); 1604 printf(")"); 1605 } 1606 } 1607 } 1608 1609 int perf_session__cpu_bitmap(struct perf_session *session, 1610 const char *cpu_list, unsigned long *cpu_bitmap) 1611 { 1612 int i; 1613 struct cpu_map *map; 1614 1615 for (i = 0; i < PERF_TYPE_MAX; ++i) { 1616 struct perf_evsel *evsel; 1617 1618 evsel = perf_session__find_first_evtype(session, i); 1619 if (!evsel) 1620 continue; 1621 1622 if (!(evsel->attr.sample_type & PERF_SAMPLE_CPU)) { 1623 pr_err("File does not contain CPU events. " 1624 "Remove -c option to proceed.\n"); 1625 return -1; 1626 } 1627 } 1628 1629 map = cpu_map__new(cpu_list); 1630 if (map == NULL) { 1631 pr_err("Invalid cpu_list\n"); 1632 return -1; 1633 } 1634 1635 for (i = 0; i < map->nr; i++) { 1636 int cpu = map->map[i]; 1637 1638 if (cpu >= MAX_NR_CPUS) { 1639 pr_err("Requested CPU %d too large. " 1640 "Consider raising MAX_NR_CPUS\n", cpu); 1641 return -1; 1642 } 1643 1644 set_bit(cpu, cpu_bitmap); 1645 } 1646 1647 return 0; 1648 } 1649 1650 void perf_session__fprintf_info(struct perf_session *session, FILE *fp, 1651 bool full) 1652 { 1653 struct stat st; 1654 int ret; 1655 1656 if (session == NULL || fp == NULL) 1657 return; 1658 1659 ret = fstat(session->fd, &st); 1660 if (ret == -1) 1661 return; 1662 1663 fprintf(fp, "# ========\n"); 1664 fprintf(fp, "# captured on: %s", ctime(&st.st_ctime)); 1665 perf_header__fprintf_info(session, fp, full); 1666 fprintf(fp, "# ========\n#\n"); 1667 } 1668 1669 1670 int __perf_session__set_tracepoints_handlers(struct perf_session *session, 1671 const struct perf_evsel_str_handler *assocs, 1672 size_t nr_assocs) 1673 { 1674 struct perf_evsel *evsel; 1675 size_t i; 1676 int err; 1677 1678 for (i = 0; i < nr_assocs; i++) { 1679 /* 1680 * Adding a handler for an event not in the session, 1681 * just ignore it. 1682 */ 1683 evsel = perf_evlist__find_tracepoint_by_name(session->evlist, assocs[i].name); 1684 if (evsel == NULL) 1685 continue; 1686 1687 err = -EEXIST; 1688 if (evsel->handler.func != NULL) 1689 goto out; 1690 evsel->handler.func = assocs[i].handler; 1691 } 1692 1693 err = 0; 1694 out: 1695 return err; 1696 } 1697