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) 20 { 21 struct perf_data_file *file = self->file; 22 23 if (perf_session__read_header(self) < 0) { 24 pr_err("incompatible file format (rerun with -v to learn more)"); 25 return -1; 26 } 27 28 if (perf_data_file__is_pipe(file)) 29 return 0; 30 31 if (!perf_evlist__valid_sample_type(self->evlist)) { 32 pr_err("non matching sample_type"); 33 return -1; 34 } 35 36 if (!perf_evlist__valid_sample_id_all(self->evlist)) { 37 pr_err("non matching sample_id_all"); 38 return -1; 39 } 40 41 if (!perf_evlist__valid_read_format(self->evlist)) { 42 pr_err("non matching read_format"); 43 return -1; 44 } 45 46 return 0; 47 } 48 49 void perf_session__set_id_hdr_size(struct perf_session *session) 50 { 51 u16 id_hdr_size = perf_evlist__id_hdr_size(session->evlist); 52 53 machines__set_id_hdr_size(&session->machines, id_hdr_size); 54 } 55 56 int perf_session__create_kernel_maps(struct perf_session *self) 57 { 58 int ret = machine__create_kernel_maps(&self->machines.host); 59 60 if (ret >= 0) 61 ret = machines__create_guest_kernel_maps(&self->machines); 62 return ret; 63 } 64 65 static void perf_session__destroy_kernel_maps(struct perf_session *self) 66 { 67 machines__destroy_kernel_maps(&self->machines); 68 } 69 70 struct perf_session *perf_session__new(struct perf_data_file *file, 71 bool repipe, struct perf_tool *tool) 72 { 73 struct perf_session *self; 74 75 self = zalloc(sizeof(*self)); 76 if (!self) 77 goto out; 78 79 self->repipe = repipe; 80 INIT_LIST_HEAD(&self->ordered_samples.samples); 81 INIT_LIST_HEAD(&self->ordered_samples.sample_cache); 82 INIT_LIST_HEAD(&self->ordered_samples.to_free); 83 machines__init(&self->machines); 84 85 if (file) { 86 if (perf_data_file__open(file)) 87 goto out_delete; 88 89 self->file = file; 90 91 if (perf_data_file__is_read(file)) { 92 if (perf_session__open(self) < 0) 93 goto out_close; 94 95 perf_session__set_id_hdr_size(self); 96 } 97 } 98 99 if (!file || perf_data_file__is_write(file)) { 100 /* 101 * In O_RDONLY mode this will be performed when reading the 102 * kernel MMAP event, in perf_event__process_mmap(). 103 */ 104 if (perf_session__create_kernel_maps(self) < 0) 105 goto out_delete; 106 } 107 108 if (tool && tool->ordering_requires_timestamps && 109 tool->ordered_samples && !perf_evlist__sample_id_all(self->evlist)) { 110 dump_printf("WARNING: No sample_id_all support, falling back to unordered processing\n"); 111 tool->ordered_samples = false; 112 } 113 114 return self; 115 116 out_close: 117 perf_data_file__close(file); 118 out_delete: 119 perf_session__delete(self); 120 out: 121 return NULL; 122 } 123 124 static void perf_session__delete_dead_threads(struct perf_session *session) 125 { 126 machine__delete_dead_threads(&session->machines.host); 127 } 128 129 static void perf_session__delete_threads(struct perf_session *session) 130 { 131 machine__delete_threads(&session->machines.host); 132 } 133 134 static void perf_session_env__delete(struct perf_session_env *env) 135 { 136 free(env->hostname); 137 free(env->os_release); 138 free(env->version); 139 free(env->arch); 140 free(env->cpu_desc); 141 free(env->cpuid); 142 143 free(env->cmdline); 144 free(env->sibling_cores); 145 free(env->sibling_threads); 146 free(env->numa_nodes); 147 free(env->pmu_mappings); 148 } 149 150 void perf_session__delete(struct perf_session *self) 151 { 152 perf_session__destroy_kernel_maps(self); 153 perf_session__delete_dead_threads(self); 154 perf_session__delete_threads(self); 155 perf_session_env__delete(&self->header.env); 156 machines__exit(&self->machines); 157 if (self->file) 158 perf_data_file__close(self->file); 159 free(self); 160 vdso__exit(); 161 } 162 163 static int process_event_synth_tracing_data_stub(struct perf_tool *tool 164 __maybe_unused, 165 union perf_event *event 166 __maybe_unused, 167 struct perf_session *session 168 __maybe_unused) 169 { 170 dump_printf(": unhandled!\n"); 171 return 0; 172 } 173 174 static int process_event_synth_attr_stub(struct perf_tool *tool __maybe_unused, 175 union perf_event *event __maybe_unused, 176 struct perf_evlist **pevlist 177 __maybe_unused) 178 { 179 dump_printf(": unhandled!\n"); 180 return 0; 181 } 182 183 static int process_event_sample_stub(struct perf_tool *tool __maybe_unused, 184 union perf_event *event __maybe_unused, 185 struct perf_sample *sample __maybe_unused, 186 struct perf_evsel *evsel __maybe_unused, 187 struct machine *machine __maybe_unused) 188 { 189 dump_printf(": unhandled!\n"); 190 return 0; 191 } 192 193 static int process_event_stub(struct perf_tool *tool __maybe_unused, 194 union perf_event *event __maybe_unused, 195 struct perf_sample *sample __maybe_unused, 196 struct machine *machine __maybe_unused) 197 { 198 dump_printf(": unhandled!\n"); 199 return 0; 200 } 201 202 static int process_finished_round_stub(struct perf_tool *tool __maybe_unused, 203 union perf_event *event __maybe_unused, 204 struct perf_session *perf_session 205 __maybe_unused) 206 { 207 dump_printf(": unhandled!\n"); 208 return 0; 209 } 210 211 static int process_finished_round(struct perf_tool *tool, 212 union perf_event *event, 213 struct perf_session *session); 214 215 void perf_tool__fill_defaults(struct perf_tool *tool) 216 { 217 if (tool->sample == NULL) 218 tool->sample = process_event_sample_stub; 219 if (tool->mmap == NULL) 220 tool->mmap = process_event_stub; 221 if (tool->mmap2 == NULL) 222 tool->mmap2 = process_event_stub; 223 if (tool->comm == NULL) 224 tool->comm = process_event_stub; 225 if (tool->fork == NULL) 226 tool->fork = process_event_stub; 227 if (tool->exit == NULL) 228 tool->exit = process_event_stub; 229 if (tool->lost == NULL) 230 tool->lost = perf_event__process_lost; 231 if (tool->read == NULL) 232 tool->read = process_event_sample_stub; 233 if (tool->throttle == NULL) 234 tool->throttle = process_event_stub; 235 if (tool->unthrottle == NULL) 236 tool->unthrottle = process_event_stub; 237 if (tool->attr == NULL) 238 tool->attr = process_event_synth_attr_stub; 239 if (tool->tracing_data == NULL) 240 tool->tracing_data = process_event_synth_tracing_data_stub; 241 if (tool->build_id == NULL) 242 tool->build_id = process_finished_round_stub; 243 if (tool->finished_round == NULL) { 244 if (tool->ordered_samples) 245 tool->finished_round = process_finished_round; 246 else 247 tool->finished_round = process_finished_round_stub; 248 } 249 } 250 251 void mem_bswap_32(void *src, int byte_size) 252 { 253 u32 *m = src; 254 while (byte_size > 0) { 255 *m = bswap_32(*m); 256 byte_size -= sizeof(u32); 257 ++m; 258 } 259 } 260 261 void mem_bswap_64(void *src, int byte_size) 262 { 263 u64 *m = src; 264 265 while (byte_size > 0) { 266 *m = bswap_64(*m); 267 byte_size -= sizeof(u64); 268 ++m; 269 } 270 } 271 272 static void swap_sample_id_all(union perf_event *event, void *data) 273 { 274 void *end = (void *) event + event->header.size; 275 int size = end - data; 276 277 BUG_ON(size % sizeof(u64)); 278 mem_bswap_64(data, size); 279 } 280 281 static void perf_event__all64_swap(union perf_event *event, 282 bool sample_id_all __maybe_unused) 283 { 284 struct perf_event_header *hdr = &event->header; 285 mem_bswap_64(hdr + 1, event->header.size - sizeof(*hdr)); 286 } 287 288 static void perf_event__comm_swap(union perf_event *event, bool sample_id_all) 289 { 290 event->comm.pid = bswap_32(event->comm.pid); 291 event->comm.tid = bswap_32(event->comm.tid); 292 293 if (sample_id_all) { 294 void *data = &event->comm.comm; 295 296 data += PERF_ALIGN(strlen(data) + 1, sizeof(u64)); 297 swap_sample_id_all(event, data); 298 } 299 } 300 301 static void perf_event__mmap_swap(union perf_event *event, 302 bool sample_id_all) 303 { 304 event->mmap.pid = bswap_32(event->mmap.pid); 305 event->mmap.tid = bswap_32(event->mmap.tid); 306 event->mmap.start = bswap_64(event->mmap.start); 307 event->mmap.len = bswap_64(event->mmap.len); 308 event->mmap.pgoff = bswap_64(event->mmap.pgoff); 309 310 if (sample_id_all) { 311 void *data = &event->mmap.filename; 312 313 data += PERF_ALIGN(strlen(data) + 1, sizeof(u64)); 314 swap_sample_id_all(event, data); 315 } 316 } 317 318 static void perf_event__mmap2_swap(union perf_event *event, 319 bool sample_id_all) 320 { 321 event->mmap2.pid = bswap_32(event->mmap2.pid); 322 event->mmap2.tid = bswap_32(event->mmap2.tid); 323 event->mmap2.start = bswap_64(event->mmap2.start); 324 event->mmap2.len = bswap_64(event->mmap2.len); 325 event->mmap2.pgoff = bswap_64(event->mmap2.pgoff); 326 event->mmap2.maj = bswap_32(event->mmap2.maj); 327 event->mmap2.min = bswap_32(event->mmap2.min); 328 event->mmap2.ino = bswap_64(event->mmap2.ino); 329 330 if (sample_id_all) { 331 void *data = &event->mmap2.filename; 332 333 data += PERF_ALIGN(strlen(data) + 1, sizeof(u64)); 334 swap_sample_id_all(event, data); 335 } 336 } 337 static void perf_event__task_swap(union perf_event *event, bool sample_id_all) 338 { 339 event->fork.pid = bswap_32(event->fork.pid); 340 event->fork.tid = bswap_32(event->fork.tid); 341 event->fork.ppid = bswap_32(event->fork.ppid); 342 event->fork.ptid = bswap_32(event->fork.ptid); 343 event->fork.time = bswap_64(event->fork.time); 344 345 if (sample_id_all) 346 swap_sample_id_all(event, &event->fork + 1); 347 } 348 349 static void perf_event__read_swap(union perf_event *event, bool sample_id_all) 350 { 351 event->read.pid = bswap_32(event->read.pid); 352 event->read.tid = bswap_32(event->read.tid); 353 event->read.value = bswap_64(event->read.value); 354 event->read.time_enabled = bswap_64(event->read.time_enabled); 355 event->read.time_running = bswap_64(event->read.time_running); 356 event->read.id = bswap_64(event->read.id); 357 358 if (sample_id_all) 359 swap_sample_id_all(event, &event->read + 1); 360 } 361 362 static void perf_event__throttle_swap(union perf_event *event, 363 bool sample_id_all) 364 { 365 event->throttle.time = bswap_64(event->throttle.time); 366 event->throttle.id = bswap_64(event->throttle.id); 367 event->throttle.stream_id = bswap_64(event->throttle.stream_id); 368 369 if (sample_id_all) 370 swap_sample_id_all(event, &event->throttle + 1); 371 } 372 373 static u8 revbyte(u8 b) 374 { 375 int rev = (b >> 4) | ((b & 0xf) << 4); 376 rev = ((rev & 0xcc) >> 2) | ((rev & 0x33) << 2); 377 rev = ((rev & 0xaa) >> 1) | ((rev & 0x55) << 1); 378 return (u8) rev; 379 } 380 381 /* 382 * XXX this is hack in attempt to carry flags bitfield 383 * throught endian village. ABI says: 384 * 385 * Bit-fields are allocated from right to left (least to most significant) 386 * on little-endian implementations and from left to right (most to least 387 * significant) on big-endian implementations. 388 * 389 * The above seems to be byte specific, so we need to reverse each 390 * byte of the bitfield. 'Internet' also says this might be implementation 391 * specific and we probably need proper fix and carry perf_event_attr 392 * bitfield flags in separate data file FEAT_ section. Thought this seems 393 * to work for now. 394 */ 395 static void swap_bitfield(u8 *p, unsigned len) 396 { 397 unsigned i; 398 399 for (i = 0; i < len; i++) { 400 *p = revbyte(*p); 401 p++; 402 } 403 } 404 405 /* exported for swapping attributes in file header */ 406 void perf_event__attr_swap(struct perf_event_attr *attr) 407 { 408 attr->type = bswap_32(attr->type); 409 attr->size = bswap_32(attr->size); 410 attr->config = bswap_64(attr->config); 411 attr->sample_period = bswap_64(attr->sample_period); 412 attr->sample_type = bswap_64(attr->sample_type); 413 attr->read_format = bswap_64(attr->read_format); 414 attr->wakeup_events = bswap_32(attr->wakeup_events); 415 attr->bp_type = bswap_32(attr->bp_type); 416 attr->bp_addr = bswap_64(attr->bp_addr); 417 attr->bp_len = bswap_64(attr->bp_len); 418 attr->branch_sample_type = bswap_64(attr->branch_sample_type); 419 attr->sample_regs_user = bswap_64(attr->sample_regs_user); 420 attr->sample_stack_user = bswap_32(attr->sample_stack_user); 421 422 swap_bitfield((u8 *) (&attr->read_format + 1), sizeof(u64)); 423 } 424 425 static void perf_event__hdr_attr_swap(union perf_event *event, 426 bool sample_id_all __maybe_unused) 427 { 428 size_t size; 429 430 perf_event__attr_swap(&event->attr.attr); 431 432 size = event->header.size; 433 size -= (void *)&event->attr.id - (void *)event; 434 mem_bswap_64(event->attr.id, size); 435 } 436 437 static void perf_event__event_type_swap(union perf_event *event, 438 bool sample_id_all __maybe_unused) 439 { 440 event->event_type.event_type.event_id = 441 bswap_64(event->event_type.event_type.event_id); 442 } 443 444 static void perf_event__tracing_data_swap(union perf_event *event, 445 bool sample_id_all __maybe_unused) 446 { 447 event->tracing_data.size = bswap_32(event->tracing_data.size); 448 } 449 450 typedef void (*perf_event__swap_op)(union perf_event *event, 451 bool sample_id_all); 452 453 static perf_event__swap_op perf_event__swap_ops[] = { 454 [PERF_RECORD_MMAP] = perf_event__mmap_swap, 455 [PERF_RECORD_MMAP2] = perf_event__mmap2_swap, 456 [PERF_RECORD_COMM] = perf_event__comm_swap, 457 [PERF_RECORD_FORK] = perf_event__task_swap, 458 [PERF_RECORD_EXIT] = perf_event__task_swap, 459 [PERF_RECORD_LOST] = perf_event__all64_swap, 460 [PERF_RECORD_READ] = perf_event__read_swap, 461 [PERF_RECORD_THROTTLE] = perf_event__throttle_swap, 462 [PERF_RECORD_UNTHROTTLE] = perf_event__throttle_swap, 463 [PERF_RECORD_SAMPLE] = perf_event__all64_swap, 464 [PERF_RECORD_HEADER_ATTR] = perf_event__hdr_attr_swap, 465 [PERF_RECORD_HEADER_EVENT_TYPE] = perf_event__event_type_swap, 466 [PERF_RECORD_HEADER_TRACING_DATA] = perf_event__tracing_data_swap, 467 [PERF_RECORD_HEADER_BUILD_ID] = NULL, 468 [PERF_RECORD_HEADER_MAX] = NULL, 469 }; 470 471 struct sample_queue { 472 u64 timestamp; 473 u64 file_offset; 474 union perf_event *event; 475 struct list_head list; 476 }; 477 478 static void perf_session_free_sample_buffers(struct perf_session *session) 479 { 480 struct ordered_samples *os = &session->ordered_samples; 481 482 while (!list_empty(&os->to_free)) { 483 struct sample_queue *sq; 484 485 sq = list_entry(os->to_free.next, struct sample_queue, list); 486 list_del(&sq->list); 487 free(sq); 488 } 489 } 490 491 static int perf_session_deliver_event(struct perf_session *session, 492 union perf_event *event, 493 struct perf_sample *sample, 494 struct perf_tool *tool, 495 u64 file_offset); 496 497 static int flush_sample_queue(struct perf_session *s, 498 struct perf_tool *tool) 499 { 500 struct ordered_samples *os = &s->ordered_samples; 501 struct list_head *head = &os->samples; 502 struct sample_queue *tmp, *iter; 503 struct perf_sample sample; 504 u64 limit = os->next_flush; 505 u64 last_ts = os->last_sample ? os->last_sample->timestamp : 0ULL; 506 unsigned idx = 0, progress_next = os->nr_samples / 16; 507 bool show_progress = limit == ULLONG_MAX; 508 int ret; 509 510 if (!tool->ordered_samples || !limit) 511 return 0; 512 513 list_for_each_entry_safe(iter, tmp, head, list) { 514 if (session_done()) 515 return 0; 516 517 if (iter->timestamp > limit) 518 break; 519 520 ret = perf_evlist__parse_sample(s->evlist, iter->event, &sample); 521 if (ret) 522 pr_err("Can't parse sample, err = %d\n", ret); 523 else { 524 ret = perf_session_deliver_event(s, iter->event, &sample, tool, 525 iter->file_offset); 526 if (ret) 527 return ret; 528 } 529 530 os->last_flush = iter->timestamp; 531 list_del(&iter->list); 532 list_add(&iter->list, &os->sample_cache); 533 if (show_progress && (++idx >= progress_next)) { 534 progress_next += os->nr_samples / 16; 535 ui_progress__update(idx, os->nr_samples, 536 "Processing time ordered events..."); 537 } 538 } 539 540 if (list_empty(head)) { 541 os->last_sample = NULL; 542 } else if (last_ts <= limit) { 543 os->last_sample = 544 list_entry(head->prev, struct sample_queue, list); 545 } 546 547 os->nr_samples = 0; 548 549 return 0; 550 } 551 552 /* 553 * When perf record finishes a pass on every buffers, it records this pseudo 554 * event. 555 * We record the max timestamp t found in the pass n. 556 * Assuming these timestamps are monotonic across cpus, we know that if 557 * a buffer still has events with timestamps below t, they will be all 558 * available and then read in the pass n + 1. 559 * Hence when we start to read the pass n + 2, we can safely flush every 560 * events with timestamps below t. 561 * 562 * ============ PASS n ================= 563 * CPU 0 | CPU 1 564 * | 565 * cnt1 timestamps | cnt2 timestamps 566 * 1 | 2 567 * 2 | 3 568 * - | 4 <--- max recorded 569 * 570 * ============ PASS n + 1 ============== 571 * CPU 0 | CPU 1 572 * | 573 * cnt1 timestamps | cnt2 timestamps 574 * 3 | 5 575 * 4 | 6 576 * 5 | 7 <---- max recorded 577 * 578 * Flush every events below timestamp 4 579 * 580 * ============ PASS n + 2 ============== 581 * CPU 0 | CPU 1 582 * | 583 * cnt1 timestamps | cnt2 timestamps 584 * 6 | 8 585 * 7 | 9 586 * - | 10 587 * 588 * Flush every events below timestamp 7 589 * etc... 590 */ 591 static int process_finished_round(struct perf_tool *tool, 592 union perf_event *event __maybe_unused, 593 struct perf_session *session) 594 { 595 int ret = flush_sample_queue(session, tool); 596 if (!ret) 597 session->ordered_samples.next_flush = session->ordered_samples.max_timestamp; 598 599 return ret; 600 } 601 602 /* The queue is ordered by time */ 603 static void __queue_event(struct sample_queue *new, struct perf_session *s) 604 { 605 struct ordered_samples *os = &s->ordered_samples; 606 struct sample_queue *sample = os->last_sample; 607 u64 timestamp = new->timestamp; 608 struct list_head *p; 609 610 ++os->nr_samples; 611 os->last_sample = new; 612 613 if (!sample) { 614 list_add(&new->list, &os->samples); 615 os->max_timestamp = timestamp; 616 return; 617 } 618 619 /* 620 * last_sample might point to some random place in the list as it's 621 * the last queued event. We expect that the new event is close to 622 * this. 623 */ 624 if (sample->timestamp <= timestamp) { 625 while (sample->timestamp <= timestamp) { 626 p = sample->list.next; 627 if (p == &os->samples) { 628 list_add_tail(&new->list, &os->samples); 629 os->max_timestamp = timestamp; 630 return; 631 } 632 sample = list_entry(p, struct sample_queue, list); 633 } 634 list_add_tail(&new->list, &sample->list); 635 } else { 636 while (sample->timestamp > timestamp) { 637 p = sample->list.prev; 638 if (p == &os->samples) { 639 list_add(&new->list, &os->samples); 640 return; 641 } 642 sample = list_entry(p, struct sample_queue, list); 643 } 644 list_add(&new->list, &sample->list); 645 } 646 } 647 648 #define MAX_SAMPLE_BUFFER (64 * 1024 / sizeof(struct sample_queue)) 649 650 int perf_session_queue_event(struct perf_session *s, union perf_event *event, 651 struct perf_sample *sample, u64 file_offset) 652 { 653 struct ordered_samples *os = &s->ordered_samples; 654 struct list_head *sc = &os->sample_cache; 655 u64 timestamp = sample->time; 656 struct sample_queue *new; 657 658 if (!timestamp || timestamp == ~0ULL) 659 return -ETIME; 660 661 if (timestamp < s->ordered_samples.last_flush) { 662 printf("Warning: Timestamp below last timeslice flush\n"); 663 return -EINVAL; 664 } 665 666 if (!list_empty(sc)) { 667 new = list_entry(sc->next, struct sample_queue, list); 668 list_del(&new->list); 669 } else if (os->sample_buffer) { 670 new = os->sample_buffer + os->sample_buffer_idx; 671 if (++os->sample_buffer_idx == MAX_SAMPLE_BUFFER) 672 os->sample_buffer = NULL; 673 } else { 674 os->sample_buffer = malloc(MAX_SAMPLE_BUFFER * sizeof(*new)); 675 if (!os->sample_buffer) 676 return -ENOMEM; 677 list_add(&os->sample_buffer->list, &os->to_free); 678 os->sample_buffer_idx = 2; 679 new = os->sample_buffer + 1; 680 } 681 682 new->timestamp = timestamp; 683 new->file_offset = file_offset; 684 new->event = event; 685 686 __queue_event(new, s); 687 688 return 0; 689 } 690 691 static void callchain__printf(struct perf_sample *sample) 692 { 693 unsigned int i; 694 695 printf("... chain: nr:%" PRIu64 "\n", sample->callchain->nr); 696 697 for (i = 0; i < sample->callchain->nr; i++) 698 printf("..... %2d: %016" PRIx64 "\n", 699 i, sample->callchain->ips[i]); 700 } 701 702 static void branch_stack__printf(struct perf_sample *sample) 703 { 704 uint64_t i; 705 706 printf("... branch stack: nr:%" PRIu64 "\n", sample->branch_stack->nr); 707 708 for (i = 0; i < sample->branch_stack->nr; i++) 709 printf("..... %2"PRIu64": %016" PRIx64 " -> %016" PRIx64 "\n", 710 i, sample->branch_stack->entries[i].from, 711 sample->branch_stack->entries[i].to); 712 } 713 714 static void regs_dump__printf(u64 mask, u64 *regs) 715 { 716 unsigned rid, i = 0; 717 718 for_each_set_bit(rid, (unsigned long *) &mask, sizeof(mask) * 8) { 719 u64 val = regs[i++]; 720 721 printf(".... %-5s 0x%" PRIx64 "\n", 722 perf_reg_name(rid), val); 723 } 724 } 725 726 static void regs_user__printf(struct perf_sample *sample, u64 mask) 727 { 728 struct regs_dump *user_regs = &sample->user_regs; 729 730 if (user_regs->regs) { 731 printf("... user regs: mask 0x%" PRIx64 "\n", mask); 732 regs_dump__printf(mask, user_regs->regs); 733 } 734 } 735 736 static void stack_user__printf(struct stack_dump *dump) 737 { 738 printf("... ustack: size %" PRIu64 ", offset 0x%x\n", 739 dump->size, dump->offset); 740 } 741 742 static void perf_session__print_tstamp(struct perf_session *session, 743 union perf_event *event, 744 struct perf_sample *sample) 745 { 746 u64 sample_type = __perf_evlist__combined_sample_type(session->evlist); 747 748 if (event->header.type != PERF_RECORD_SAMPLE && 749 !perf_evlist__sample_id_all(session->evlist)) { 750 fputs("-1 -1 ", stdout); 751 return; 752 } 753 754 if ((sample_type & PERF_SAMPLE_CPU)) 755 printf("%u ", sample->cpu); 756 757 if (sample_type & PERF_SAMPLE_TIME) 758 printf("%" PRIu64 " ", sample->time); 759 } 760 761 static void sample_read__printf(struct perf_sample *sample, u64 read_format) 762 { 763 printf("... sample_read:\n"); 764 765 if (read_format & PERF_FORMAT_TOTAL_TIME_ENABLED) 766 printf("...... time enabled %016" PRIx64 "\n", 767 sample->read.time_enabled); 768 769 if (read_format & PERF_FORMAT_TOTAL_TIME_RUNNING) 770 printf("...... time running %016" PRIx64 "\n", 771 sample->read.time_running); 772 773 if (read_format & PERF_FORMAT_GROUP) { 774 u64 i; 775 776 printf(".... group nr %" PRIu64 "\n", sample->read.group.nr); 777 778 for (i = 0; i < sample->read.group.nr; i++) { 779 struct sample_read_value *value; 780 781 value = &sample->read.group.values[i]; 782 printf("..... id %016" PRIx64 783 ", value %016" PRIx64 "\n", 784 value->id, value->value); 785 } 786 } else 787 printf("..... id %016" PRIx64 ", value %016" PRIx64 "\n", 788 sample->read.one.id, sample->read.one.value); 789 } 790 791 static void dump_event(struct perf_session *session, union perf_event *event, 792 u64 file_offset, struct perf_sample *sample) 793 { 794 if (!dump_trace) 795 return; 796 797 printf("\n%#" PRIx64 " [%#x]: event: %d\n", 798 file_offset, event->header.size, event->header.type); 799 800 trace_event(event); 801 802 if (sample) 803 perf_session__print_tstamp(session, event, sample); 804 805 printf("%#" PRIx64 " [%#x]: PERF_RECORD_%s", file_offset, 806 event->header.size, perf_event__name(event->header.type)); 807 } 808 809 static void dump_sample(struct perf_evsel *evsel, union perf_event *event, 810 struct perf_sample *sample) 811 { 812 u64 sample_type; 813 814 if (!dump_trace) 815 return; 816 817 printf("(IP, %d): %d/%d: %#" PRIx64 " period: %" PRIu64 " addr: %#" PRIx64 "\n", 818 event->header.misc, sample->pid, sample->tid, sample->ip, 819 sample->period, sample->addr); 820 821 sample_type = evsel->attr.sample_type; 822 823 if (sample_type & PERF_SAMPLE_CALLCHAIN) 824 callchain__printf(sample); 825 826 if (sample_type & PERF_SAMPLE_BRANCH_STACK) 827 branch_stack__printf(sample); 828 829 if (sample_type & PERF_SAMPLE_REGS_USER) 830 regs_user__printf(sample, evsel->attr.sample_regs_user); 831 832 if (sample_type & PERF_SAMPLE_STACK_USER) 833 stack_user__printf(&sample->user_stack); 834 835 if (sample_type & PERF_SAMPLE_WEIGHT) 836 printf("... weight: %" PRIu64 "\n", sample->weight); 837 838 if (sample_type & PERF_SAMPLE_DATA_SRC) 839 printf(" . data_src: 0x%"PRIx64"\n", sample->data_src); 840 841 if (sample_type & PERF_SAMPLE_TRANSACTION) 842 printf("... transaction: %" PRIx64 "\n", sample->transaction); 843 844 if (sample_type & PERF_SAMPLE_READ) 845 sample_read__printf(sample, evsel->attr.read_format); 846 } 847 848 static struct machine * 849 perf_session__find_machine_for_cpumode(struct perf_session *session, 850 union perf_event *event, 851 struct perf_sample *sample) 852 { 853 const u8 cpumode = event->header.misc & PERF_RECORD_MISC_CPUMODE_MASK; 854 855 if (perf_guest && 856 ((cpumode == PERF_RECORD_MISC_GUEST_KERNEL) || 857 (cpumode == PERF_RECORD_MISC_GUEST_USER))) { 858 u32 pid; 859 860 if (event->header.type == PERF_RECORD_MMAP 861 || event->header.type == PERF_RECORD_MMAP2) 862 pid = event->mmap.pid; 863 else 864 pid = sample->pid; 865 866 return perf_session__findnew_machine(session, pid); 867 } 868 869 return &session->machines.host; 870 } 871 872 static int deliver_sample_value(struct perf_session *session, 873 struct perf_tool *tool, 874 union perf_event *event, 875 struct perf_sample *sample, 876 struct sample_read_value *v, 877 struct machine *machine) 878 { 879 struct perf_sample_id *sid; 880 881 sid = perf_evlist__id2sid(session->evlist, v->id); 882 if (sid) { 883 sample->id = v->id; 884 sample->period = v->value - sid->period; 885 sid->period = v->value; 886 } 887 888 if (!sid || sid->evsel == NULL) { 889 ++session->stats.nr_unknown_id; 890 return 0; 891 } 892 893 return tool->sample(tool, event, sample, sid->evsel, machine); 894 } 895 896 static int deliver_sample_group(struct perf_session *session, 897 struct perf_tool *tool, 898 union perf_event *event, 899 struct perf_sample *sample, 900 struct machine *machine) 901 { 902 int ret = -EINVAL; 903 u64 i; 904 905 for (i = 0; i < sample->read.group.nr; i++) { 906 ret = deliver_sample_value(session, tool, event, sample, 907 &sample->read.group.values[i], 908 machine); 909 if (ret) 910 break; 911 } 912 913 return ret; 914 } 915 916 static int 917 perf_session__deliver_sample(struct perf_session *session, 918 struct perf_tool *tool, 919 union perf_event *event, 920 struct perf_sample *sample, 921 struct perf_evsel *evsel, 922 struct machine *machine) 923 { 924 /* We know evsel != NULL. */ 925 u64 sample_type = evsel->attr.sample_type; 926 u64 read_format = evsel->attr.read_format; 927 928 /* Standard sample delievery. */ 929 if (!(sample_type & PERF_SAMPLE_READ)) 930 return tool->sample(tool, event, sample, evsel, machine); 931 932 /* For PERF_SAMPLE_READ we have either single or group mode. */ 933 if (read_format & PERF_FORMAT_GROUP) 934 return deliver_sample_group(session, tool, event, sample, 935 machine); 936 else 937 return deliver_sample_value(session, tool, event, sample, 938 &sample->read.one, machine); 939 } 940 941 static int perf_session_deliver_event(struct perf_session *session, 942 union perf_event *event, 943 struct perf_sample *sample, 944 struct perf_tool *tool, 945 u64 file_offset) 946 { 947 struct perf_evsel *evsel; 948 struct machine *machine; 949 950 dump_event(session, event, file_offset, sample); 951 952 evsel = perf_evlist__id2evsel(session->evlist, sample->id); 953 if (evsel != NULL && event->header.type != PERF_RECORD_SAMPLE) { 954 /* 955 * XXX We're leaving PERF_RECORD_SAMPLE unnacounted here 956 * because the tools right now may apply filters, discarding 957 * some of the samples. For consistency, in the future we 958 * should have something like nr_filtered_samples and remove 959 * the sample->period from total_sample_period, etc, KISS for 960 * now tho. 961 * 962 * Also testing against NULL allows us to handle files without 963 * attr.sample_id_all and/or without PERF_SAMPLE_ID. In the 964 * future probably it'll be a good idea to restrict event 965 * processing via perf_session to files with both set. 966 */ 967 hists__inc_nr_events(&evsel->hists, event->header.type); 968 } 969 970 machine = perf_session__find_machine_for_cpumode(session, event, 971 sample); 972 973 switch (event->header.type) { 974 case PERF_RECORD_SAMPLE: 975 dump_sample(evsel, event, sample); 976 if (evsel == NULL) { 977 ++session->stats.nr_unknown_id; 978 return 0; 979 } 980 if (machine == NULL) { 981 ++session->stats.nr_unprocessable_samples; 982 return 0; 983 } 984 return perf_session__deliver_sample(session, tool, event, 985 sample, evsel, machine); 986 case PERF_RECORD_MMAP: 987 return tool->mmap(tool, event, sample, machine); 988 case PERF_RECORD_MMAP2: 989 return tool->mmap2(tool, event, sample, machine); 990 case PERF_RECORD_COMM: 991 return tool->comm(tool, event, sample, machine); 992 case PERF_RECORD_FORK: 993 return tool->fork(tool, event, sample, machine); 994 case PERF_RECORD_EXIT: 995 return tool->exit(tool, event, sample, machine); 996 case PERF_RECORD_LOST: 997 if (tool->lost == perf_event__process_lost) 998 session->stats.total_lost += event->lost.lost; 999 return tool->lost(tool, event, sample, machine); 1000 case PERF_RECORD_READ: 1001 return tool->read(tool, event, sample, evsel, machine); 1002 case PERF_RECORD_THROTTLE: 1003 return tool->throttle(tool, event, sample, machine); 1004 case PERF_RECORD_UNTHROTTLE: 1005 return tool->unthrottle(tool, event, sample, machine); 1006 default: 1007 ++session->stats.nr_unknown_events; 1008 return -1; 1009 } 1010 } 1011 1012 static int perf_session__process_user_event(struct perf_session *session, union perf_event *event, 1013 struct perf_tool *tool, u64 file_offset) 1014 { 1015 int fd = perf_data_file__fd(session->file); 1016 int err; 1017 1018 dump_event(session, event, file_offset, NULL); 1019 1020 /* These events are processed right away */ 1021 switch (event->header.type) { 1022 case PERF_RECORD_HEADER_ATTR: 1023 err = tool->attr(tool, event, &session->evlist); 1024 if (err == 0) 1025 perf_session__set_id_hdr_size(session); 1026 return err; 1027 case PERF_RECORD_HEADER_TRACING_DATA: 1028 /* setup for reading amidst mmap */ 1029 lseek(fd, file_offset, SEEK_SET); 1030 return tool->tracing_data(tool, event, session); 1031 case PERF_RECORD_HEADER_BUILD_ID: 1032 return tool->build_id(tool, event, session); 1033 case PERF_RECORD_FINISHED_ROUND: 1034 return tool->finished_round(tool, event, session); 1035 default: 1036 return -EINVAL; 1037 } 1038 } 1039 1040 static void event_swap(union perf_event *event, bool sample_id_all) 1041 { 1042 perf_event__swap_op swap; 1043 1044 swap = perf_event__swap_ops[event->header.type]; 1045 if (swap) 1046 swap(event, sample_id_all); 1047 } 1048 1049 static int perf_session__process_event(struct perf_session *session, 1050 union perf_event *event, 1051 struct perf_tool *tool, 1052 u64 file_offset) 1053 { 1054 struct perf_sample sample; 1055 int ret; 1056 1057 if (session->header.needs_swap) 1058 event_swap(event, perf_evlist__sample_id_all(session->evlist)); 1059 1060 if (event->header.type >= PERF_RECORD_HEADER_MAX) 1061 return -EINVAL; 1062 1063 events_stats__inc(&session->stats, event->header.type); 1064 1065 if (event->header.type >= PERF_RECORD_USER_TYPE_START) 1066 return perf_session__process_user_event(session, event, tool, file_offset); 1067 1068 /* 1069 * For all kernel events we get the sample data 1070 */ 1071 ret = perf_evlist__parse_sample(session->evlist, event, &sample); 1072 if (ret) 1073 return ret; 1074 1075 if (tool->ordered_samples) { 1076 ret = perf_session_queue_event(session, event, &sample, 1077 file_offset); 1078 if (ret != -ETIME) 1079 return ret; 1080 } 1081 1082 return perf_session_deliver_event(session, event, &sample, tool, 1083 file_offset); 1084 } 1085 1086 void perf_event_header__bswap(struct perf_event_header *self) 1087 { 1088 self->type = bswap_32(self->type); 1089 self->misc = bswap_16(self->misc); 1090 self->size = bswap_16(self->size); 1091 } 1092 1093 struct thread *perf_session__findnew(struct perf_session *session, pid_t pid) 1094 { 1095 return machine__findnew_thread(&session->machines.host, 0, pid); 1096 } 1097 1098 static struct thread *perf_session__register_idle_thread(struct perf_session *self) 1099 { 1100 struct thread *thread = perf_session__findnew(self, 0); 1101 1102 if (thread == NULL || thread__set_comm(thread, "swapper")) { 1103 pr_err("problem inserting idle task.\n"); 1104 thread = NULL; 1105 } 1106 1107 return thread; 1108 } 1109 1110 static void perf_session__warn_about_errors(const struct perf_session *session, 1111 const struct perf_tool *tool) 1112 { 1113 if (tool->lost == perf_event__process_lost && 1114 session->stats.nr_events[PERF_RECORD_LOST] != 0) { 1115 ui__warning("Processed %d events and lost %d chunks!\n\n" 1116 "Check IO/CPU overload!\n\n", 1117 session->stats.nr_events[0], 1118 session->stats.nr_events[PERF_RECORD_LOST]); 1119 } 1120 1121 if (session->stats.nr_unknown_events != 0) { 1122 ui__warning("Found %u unknown events!\n\n" 1123 "Is this an older tool processing a perf.data " 1124 "file generated by a more recent tool?\n\n" 1125 "If that is not the case, consider " 1126 "reporting to linux-kernel@vger.kernel.org.\n\n", 1127 session->stats.nr_unknown_events); 1128 } 1129 1130 if (session->stats.nr_unknown_id != 0) { 1131 ui__warning("%u samples with id not present in the header\n", 1132 session->stats.nr_unknown_id); 1133 } 1134 1135 if (session->stats.nr_invalid_chains != 0) { 1136 ui__warning("Found invalid callchains!\n\n" 1137 "%u out of %u events were discarded for this reason.\n\n" 1138 "Consider reporting to linux-kernel@vger.kernel.org.\n\n", 1139 session->stats.nr_invalid_chains, 1140 session->stats.nr_events[PERF_RECORD_SAMPLE]); 1141 } 1142 1143 if (session->stats.nr_unprocessable_samples != 0) { 1144 ui__warning("%u unprocessable samples recorded.\n" 1145 "Do you have a KVM guest running and not using 'perf kvm'?\n", 1146 session->stats.nr_unprocessable_samples); 1147 } 1148 } 1149 1150 volatile int session_done; 1151 1152 static int __perf_session__process_pipe_events(struct perf_session *self, 1153 struct perf_tool *tool) 1154 { 1155 int fd = perf_data_file__fd(self->file); 1156 union perf_event *event; 1157 uint32_t size, cur_size = 0; 1158 void *buf = NULL; 1159 int skip = 0; 1160 u64 head; 1161 int err; 1162 void *p; 1163 1164 perf_tool__fill_defaults(tool); 1165 1166 head = 0; 1167 cur_size = sizeof(union perf_event); 1168 1169 buf = malloc(cur_size); 1170 if (!buf) 1171 return -errno; 1172 more: 1173 event = buf; 1174 err = readn(fd, event, sizeof(struct perf_event_header)); 1175 if (err <= 0) { 1176 if (err == 0) 1177 goto done; 1178 1179 pr_err("failed to read event header\n"); 1180 goto out_err; 1181 } 1182 1183 if (self->header.needs_swap) 1184 perf_event_header__bswap(&event->header); 1185 1186 size = event->header.size; 1187 if (size < sizeof(struct perf_event_header)) { 1188 pr_err("bad event header size\n"); 1189 goto out_err; 1190 } 1191 1192 if (size > cur_size) { 1193 void *new = realloc(buf, size); 1194 if (!new) { 1195 pr_err("failed to allocate memory to read event\n"); 1196 goto out_err; 1197 } 1198 buf = new; 1199 cur_size = size; 1200 event = buf; 1201 } 1202 p = event; 1203 p += sizeof(struct perf_event_header); 1204 1205 if (size - sizeof(struct perf_event_header)) { 1206 err = readn(fd, p, size - sizeof(struct perf_event_header)); 1207 if (err <= 0) { 1208 if (err == 0) { 1209 pr_err("unexpected end of event stream\n"); 1210 goto done; 1211 } 1212 1213 pr_err("failed to read event data\n"); 1214 goto out_err; 1215 } 1216 } 1217 1218 if ((skip = perf_session__process_event(self, event, tool, head)) < 0) { 1219 pr_err("%#" PRIx64 " [%#x]: failed to process type: %d\n", 1220 head, event->header.size, event->header.type); 1221 err = -EINVAL; 1222 goto out_err; 1223 } 1224 1225 head += size; 1226 1227 if (skip > 0) 1228 head += skip; 1229 1230 if (!session_done()) 1231 goto more; 1232 done: 1233 /* do the final flush for ordered samples */ 1234 self->ordered_samples.next_flush = ULLONG_MAX; 1235 err = flush_sample_queue(self, tool); 1236 out_err: 1237 free(buf); 1238 perf_session__warn_about_errors(self, tool); 1239 perf_session_free_sample_buffers(self); 1240 return err; 1241 } 1242 1243 static union perf_event * 1244 fetch_mmaped_event(struct perf_session *session, 1245 u64 head, size_t mmap_size, char *buf) 1246 { 1247 union perf_event *event; 1248 1249 /* 1250 * Ensure we have enough space remaining to read 1251 * the size of the event in the headers. 1252 */ 1253 if (head + sizeof(event->header) > mmap_size) 1254 return NULL; 1255 1256 event = (union perf_event *)(buf + head); 1257 1258 if (session->header.needs_swap) 1259 perf_event_header__bswap(&event->header); 1260 1261 if (head + event->header.size > mmap_size) { 1262 /* We're not fetching the event so swap back again */ 1263 if (session->header.needs_swap) 1264 perf_event_header__bswap(&event->header); 1265 return NULL; 1266 } 1267 1268 return event; 1269 } 1270 1271 /* 1272 * On 64bit we can mmap the data file in one go. No need for tiny mmap 1273 * slices. On 32bit we use 32MB. 1274 */ 1275 #if BITS_PER_LONG == 64 1276 #define MMAP_SIZE ULLONG_MAX 1277 #define NUM_MMAPS 1 1278 #else 1279 #define MMAP_SIZE (32 * 1024 * 1024ULL) 1280 #define NUM_MMAPS 128 1281 #endif 1282 1283 int __perf_session__process_events(struct perf_session *session, 1284 u64 data_offset, u64 data_size, 1285 u64 file_size, struct perf_tool *tool) 1286 { 1287 int fd = perf_data_file__fd(session->file); 1288 u64 head, page_offset, file_offset, file_pos, progress_next; 1289 int err, mmap_prot, mmap_flags, map_idx = 0; 1290 size_t mmap_size; 1291 char *buf, *mmaps[NUM_MMAPS]; 1292 union perf_event *event; 1293 uint32_t size; 1294 1295 perf_tool__fill_defaults(tool); 1296 1297 page_offset = page_size * (data_offset / page_size); 1298 file_offset = page_offset; 1299 head = data_offset - page_offset; 1300 1301 if (data_size && (data_offset + data_size < file_size)) 1302 file_size = data_offset + data_size; 1303 1304 progress_next = file_size / 16; 1305 1306 mmap_size = MMAP_SIZE; 1307 if (mmap_size > file_size) 1308 mmap_size = file_size; 1309 1310 memset(mmaps, 0, sizeof(mmaps)); 1311 1312 mmap_prot = PROT_READ; 1313 mmap_flags = MAP_SHARED; 1314 1315 if (session->header.needs_swap) { 1316 mmap_prot |= PROT_WRITE; 1317 mmap_flags = MAP_PRIVATE; 1318 } 1319 remap: 1320 buf = mmap(NULL, mmap_size, mmap_prot, mmap_flags, fd, 1321 file_offset); 1322 if (buf == MAP_FAILED) { 1323 pr_err("failed to mmap file\n"); 1324 err = -errno; 1325 goto out_err; 1326 } 1327 mmaps[map_idx] = buf; 1328 map_idx = (map_idx + 1) & (ARRAY_SIZE(mmaps) - 1); 1329 file_pos = file_offset + head; 1330 1331 more: 1332 event = fetch_mmaped_event(session, head, mmap_size, buf); 1333 if (!event) { 1334 if (mmaps[map_idx]) { 1335 munmap(mmaps[map_idx], mmap_size); 1336 mmaps[map_idx] = NULL; 1337 } 1338 1339 page_offset = page_size * (head / page_size); 1340 file_offset += page_offset; 1341 head -= page_offset; 1342 goto remap; 1343 } 1344 1345 size = event->header.size; 1346 1347 if (size < sizeof(struct perf_event_header) || 1348 perf_session__process_event(session, event, tool, file_pos) < 0) { 1349 pr_err("%#" PRIx64 " [%#x]: failed to process type: %d\n", 1350 file_offset + head, event->header.size, 1351 event->header.type); 1352 err = -EINVAL; 1353 goto out_err; 1354 } 1355 1356 head += size; 1357 file_pos += size; 1358 1359 if (file_pos >= progress_next) { 1360 progress_next += file_size / 16; 1361 ui_progress__update(file_pos, file_size, 1362 "Processing events..."); 1363 } 1364 1365 if (session_done()) 1366 goto out; 1367 1368 if (file_pos < file_size) 1369 goto more; 1370 1371 out: 1372 /* do the final flush for ordered samples */ 1373 session->ordered_samples.next_flush = ULLONG_MAX; 1374 err = flush_sample_queue(session, tool); 1375 out_err: 1376 ui_progress__finish(); 1377 perf_session__warn_about_errors(session, tool); 1378 perf_session_free_sample_buffers(session); 1379 return err; 1380 } 1381 1382 int perf_session__process_events(struct perf_session *self, 1383 struct perf_tool *tool) 1384 { 1385 u64 size = perf_data_file__size(self->file); 1386 int err; 1387 1388 if (perf_session__register_idle_thread(self) == NULL) 1389 return -ENOMEM; 1390 1391 if (!perf_data_file__is_pipe(self->file)) 1392 err = __perf_session__process_events(self, 1393 self->header.data_offset, 1394 self->header.data_size, 1395 size, tool); 1396 else 1397 err = __perf_session__process_pipe_events(self, tool); 1398 1399 return err; 1400 } 1401 1402 bool perf_session__has_traces(struct perf_session *session, const char *msg) 1403 { 1404 struct perf_evsel *evsel; 1405 1406 list_for_each_entry(evsel, &session->evlist->entries, node) { 1407 if (evsel->attr.type == PERF_TYPE_TRACEPOINT) 1408 return true; 1409 } 1410 1411 pr_err("No trace sample to read. Did you call 'perf %s'?\n", msg); 1412 return false; 1413 } 1414 1415 int maps__set_kallsyms_ref_reloc_sym(struct map **maps, 1416 const char *symbol_name, u64 addr) 1417 { 1418 char *bracket; 1419 enum map_type i; 1420 struct ref_reloc_sym *ref; 1421 1422 ref = zalloc(sizeof(struct ref_reloc_sym)); 1423 if (ref == NULL) 1424 return -ENOMEM; 1425 1426 ref->name = strdup(symbol_name); 1427 if (ref->name == NULL) { 1428 free(ref); 1429 return -ENOMEM; 1430 } 1431 1432 bracket = strchr(ref->name, ']'); 1433 if (bracket) 1434 *bracket = '\0'; 1435 1436 ref->addr = addr; 1437 1438 for (i = 0; i < MAP__NR_TYPES; ++i) { 1439 struct kmap *kmap = map__kmap(maps[i]); 1440 kmap->ref_reloc_sym = ref; 1441 } 1442 1443 return 0; 1444 } 1445 1446 size_t perf_session__fprintf_dsos(struct perf_session *self, FILE *fp) 1447 { 1448 return machines__fprintf_dsos(&self->machines, fp); 1449 } 1450 1451 size_t perf_session__fprintf_dsos_buildid(struct perf_session *self, FILE *fp, 1452 bool (skip)(struct dso *dso, int parm), int parm) 1453 { 1454 return machines__fprintf_dsos_buildid(&self->machines, fp, skip, parm); 1455 } 1456 1457 size_t perf_session__fprintf_nr_events(struct perf_session *session, FILE *fp) 1458 { 1459 struct perf_evsel *pos; 1460 size_t ret = fprintf(fp, "Aggregated stats:\n"); 1461 1462 ret += events_stats__fprintf(&session->stats, fp); 1463 1464 list_for_each_entry(pos, &session->evlist->entries, node) { 1465 ret += fprintf(fp, "%s stats:\n", perf_evsel__name(pos)); 1466 ret += events_stats__fprintf(&pos->hists.stats, fp); 1467 } 1468 1469 return ret; 1470 } 1471 1472 size_t perf_session__fprintf(struct perf_session *session, FILE *fp) 1473 { 1474 /* 1475 * FIXME: Here we have to actually print all the machines in this 1476 * session, not just the host... 1477 */ 1478 return machine__fprintf(&session->machines.host, fp); 1479 } 1480 1481 struct perf_evsel *perf_session__find_first_evtype(struct perf_session *session, 1482 unsigned int type) 1483 { 1484 struct perf_evsel *pos; 1485 1486 list_for_each_entry(pos, &session->evlist->entries, node) { 1487 if (pos->attr.type == type) 1488 return pos; 1489 } 1490 return NULL; 1491 } 1492 1493 void perf_evsel__print_ip(struct perf_evsel *evsel, union perf_event *event, 1494 struct perf_sample *sample, struct machine *machine, 1495 unsigned int print_opts, unsigned int stack_depth) 1496 { 1497 struct addr_location al; 1498 struct callchain_cursor_node *node; 1499 int print_ip = print_opts & PRINT_IP_OPT_IP; 1500 int print_sym = print_opts & PRINT_IP_OPT_SYM; 1501 int print_dso = print_opts & PRINT_IP_OPT_DSO; 1502 int print_symoffset = print_opts & PRINT_IP_OPT_SYMOFFSET; 1503 int print_oneline = print_opts & PRINT_IP_OPT_ONELINE; 1504 char s = print_oneline ? ' ' : '\t'; 1505 1506 if (perf_event__preprocess_sample(event, machine, &al, sample) < 0) { 1507 error("problem processing %d event, skipping it.\n", 1508 event->header.type); 1509 return; 1510 } 1511 1512 if (symbol_conf.use_callchain && sample->callchain) { 1513 1514 if (machine__resolve_callchain(machine, evsel, al.thread, 1515 sample, NULL, NULL, 1516 PERF_MAX_STACK_DEPTH) != 0) { 1517 if (verbose) 1518 error("Failed to resolve callchain. Skipping\n"); 1519 return; 1520 } 1521 callchain_cursor_commit(&callchain_cursor); 1522 1523 while (stack_depth) { 1524 node = callchain_cursor_current(&callchain_cursor); 1525 if (!node) 1526 break; 1527 1528 if (print_ip) 1529 printf("%c%16" PRIx64, s, node->ip); 1530 1531 if (print_sym) { 1532 printf(" "); 1533 if (print_symoffset) { 1534 al.addr = node->ip; 1535 al.map = node->map; 1536 symbol__fprintf_symname_offs(node->sym, &al, stdout); 1537 } else 1538 symbol__fprintf_symname(node->sym, stdout); 1539 } 1540 1541 if (print_dso) { 1542 printf(" ("); 1543 map__fprintf_dsoname(node->map, stdout); 1544 printf(")"); 1545 } 1546 1547 if (!print_oneline) 1548 printf("\n"); 1549 1550 callchain_cursor_advance(&callchain_cursor); 1551 1552 stack_depth--; 1553 } 1554 1555 } else { 1556 if (print_ip) 1557 printf("%16" PRIx64, sample->ip); 1558 1559 if (print_sym) { 1560 printf(" "); 1561 if (print_symoffset) 1562 symbol__fprintf_symname_offs(al.sym, &al, 1563 stdout); 1564 else 1565 symbol__fprintf_symname(al.sym, stdout); 1566 } 1567 1568 if (print_dso) { 1569 printf(" ("); 1570 map__fprintf_dsoname(al.map, stdout); 1571 printf(")"); 1572 } 1573 } 1574 } 1575 1576 int perf_session__cpu_bitmap(struct perf_session *session, 1577 const char *cpu_list, unsigned long *cpu_bitmap) 1578 { 1579 int i; 1580 struct cpu_map *map; 1581 1582 for (i = 0; i < PERF_TYPE_MAX; ++i) { 1583 struct perf_evsel *evsel; 1584 1585 evsel = perf_session__find_first_evtype(session, i); 1586 if (!evsel) 1587 continue; 1588 1589 if (!(evsel->attr.sample_type & PERF_SAMPLE_CPU)) { 1590 pr_err("File does not contain CPU events. " 1591 "Remove -c option to proceed.\n"); 1592 return -1; 1593 } 1594 } 1595 1596 map = cpu_map__new(cpu_list); 1597 if (map == NULL) { 1598 pr_err("Invalid cpu_list\n"); 1599 return -1; 1600 } 1601 1602 for (i = 0; i < map->nr; i++) { 1603 int cpu = map->map[i]; 1604 1605 if (cpu >= MAX_NR_CPUS) { 1606 pr_err("Requested CPU %d too large. " 1607 "Consider raising MAX_NR_CPUS\n", cpu); 1608 return -1; 1609 } 1610 1611 set_bit(cpu, cpu_bitmap); 1612 } 1613 1614 return 0; 1615 } 1616 1617 void perf_session__fprintf_info(struct perf_session *session, FILE *fp, 1618 bool full) 1619 { 1620 int fd = perf_data_file__fd(session->file); 1621 struct stat st; 1622 int ret; 1623 1624 if (session == NULL || fp == NULL) 1625 return; 1626 1627 ret = fstat(fd, &st); 1628 if (ret == -1) 1629 return; 1630 1631 fprintf(fp, "# ========\n"); 1632 fprintf(fp, "# captured on: %s", ctime(&st.st_ctime)); 1633 perf_header__fprintf_info(session, fp, full); 1634 fprintf(fp, "# ========\n#\n"); 1635 } 1636 1637 1638 int __perf_session__set_tracepoints_handlers(struct perf_session *session, 1639 const struct perf_evsel_str_handler *assocs, 1640 size_t nr_assocs) 1641 { 1642 struct perf_evsel *evsel; 1643 size_t i; 1644 int err; 1645 1646 for (i = 0; i < nr_assocs; i++) { 1647 /* 1648 * Adding a handler for an event not in the session, 1649 * just ignore it. 1650 */ 1651 evsel = perf_evlist__find_tracepoint_by_name(session->evlist, assocs[i].name); 1652 if (evsel == NULL) 1653 continue; 1654 1655 err = -EEXIST; 1656 if (evsel->handler.func != NULL) 1657 goto out; 1658 evsel->handler.func = assocs[i].handler; 1659 } 1660 1661 err = 0; 1662 out: 1663 return err; 1664 } 1665