1 /* 2 * auxtrace.c: AUX area trace support 3 * Copyright (c) 2013-2015, Intel Corporation. 4 * 5 * This program is free software; you can redistribute it and/or modify it 6 * under the terms and conditions of the GNU General Public License, 7 * version 2, as published by the Free Software Foundation. 8 * 9 * This program is distributed in the hope it will be useful, but WITHOUT 10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or 11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for 12 * more details. 13 * 14 */ 15 16 #include <sys/types.h> 17 #include <sys/mman.h> 18 #include <stdbool.h> 19 20 #include <linux/kernel.h> 21 #include <linux/perf_event.h> 22 #include <linux/types.h> 23 #include <linux/bitops.h> 24 #include <linux/log2.h> 25 #include <linux/string.h> 26 27 #include <sys/param.h> 28 #include <stdlib.h> 29 #include <stdio.h> 30 #include <string.h> 31 #include <limits.h> 32 #include <errno.h> 33 #include <linux/list.h> 34 35 #include "../perf.h" 36 #include "util.h" 37 #include "evlist.h" 38 #include "cpumap.h" 39 #include "thread_map.h" 40 #include "asm/bug.h" 41 #include "auxtrace.h" 42 43 #include <linux/hash.h> 44 45 #include "event.h" 46 #include "session.h" 47 #include "debug.h" 48 #include "parse-options.h" 49 50 #include "intel-pt.h" 51 #include "intel-bts.h" 52 53 int auxtrace_mmap__mmap(struct auxtrace_mmap *mm, 54 struct auxtrace_mmap_params *mp, 55 void *userpg, int fd) 56 { 57 struct perf_event_mmap_page *pc = userpg; 58 59 WARN_ONCE(mm->base, "Uninitialized auxtrace_mmap\n"); 60 61 mm->userpg = userpg; 62 mm->mask = mp->mask; 63 mm->len = mp->len; 64 mm->prev = 0; 65 mm->idx = mp->idx; 66 mm->tid = mp->tid; 67 mm->cpu = mp->cpu; 68 69 if (!mp->len) { 70 mm->base = NULL; 71 return 0; 72 } 73 74 #if BITS_PER_LONG != 64 && !defined(HAVE_SYNC_COMPARE_AND_SWAP_SUPPORT) 75 pr_err("Cannot use AUX area tracing mmaps\n"); 76 return -1; 77 #endif 78 79 pc->aux_offset = mp->offset; 80 pc->aux_size = mp->len; 81 82 mm->base = mmap(NULL, mp->len, mp->prot, MAP_SHARED, fd, mp->offset); 83 if (mm->base == MAP_FAILED) { 84 pr_debug2("failed to mmap AUX area\n"); 85 mm->base = NULL; 86 return -1; 87 } 88 89 return 0; 90 } 91 92 void auxtrace_mmap__munmap(struct auxtrace_mmap *mm) 93 { 94 if (mm->base) { 95 munmap(mm->base, mm->len); 96 mm->base = NULL; 97 } 98 } 99 100 void auxtrace_mmap_params__init(struct auxtrace_mmap_params *mp, 101 off_t auxtrace_offset, 102 unsigned int auxtrace_pages, 103 bool auxtrace_overwrite) 104 { 105 if (auxtrace_pages) { 106 mp->offset = auxtrace_offset; 107 mp->len = auxtrace_pages * (size_t)page_size; 108 mp->mask = is_power_of_2(mp->len) ? mp->len - 1 : 0; 109 mp->prot = PROT_READ | (auxtrace_overwrite ? 0 : PROT_WRITE); 110 pr_debug2("AUX area mmap length %zu\n", mp->len); 111 } else { 112 mp->len = 0; 113 } 114 } 115 116 void auxtrace_mmap_params__set_idx(struct auxtrace_mmap_params *mp, 117 struct perf_evlist *evlist, int idx, 118 bool per_cpu) 119 { 120 mp->idx = idx; 121 122 if (per_cpu) { 123 mp->cpu = evlist->cpus->map[idx]; 124 if (evlist->threads) 125 mp->tid = thread_map__pid(evlist->threads, 0); 126 else 127 mp->tid = -1; 128 } else { 129 mp->cpu = -1; 130 mp->tid = thread_map__pid(evlist->threads, idx); 131 } 132 } 133 134 #define AUXTRACE_INIT_NR_QUEUES 32 135 136 static struct auxtrace_queue *auxtrace_alloc_queue_array(unsigned int nr_queues) 137 { 138 struct auxtrace_queue *queue_array; 139 unsigned int max_nr_queues, i; 140 141 max_nr_queues = UINT_MAX / sizeof(struct auxtrace_queue); 142 if (nr_queues > max_nr_queues) 143 return NULL; 144 145 queue_array = calloc(nr_queues, sizeof(struct auxtrace_queue)); 146 if (!queue_array) 147 return NULL; 148 149 for (i = 0; i < nr_queues; i++) { 150 INIT_LIST_HEAD(&queue_array[i].head); 151 queue_array[i].priv = NULL; 152 } 153 154 return queue_array; 155 } 156 157 int auxtrace_queues__init(struct auxtrace_queues *queues) 158 { 159 queues->nr_queues = AUXTRACE_INIT_NR_QUEUES; 160 queues->queue_array = auxtrace_alloc_queue_array(queues->nr_queues); 161 if (!queues->queue_array) 162 return -ENOMEM; 163 return 0; 164 } 165 166 static int auxtrace_queues__grow(struct auxtrace_queues *queues, 167 unsigned int new_nr_queues) 168 { 169 unsigned int nr_queues = queues->nr_queues; 170 struct auxtrace_queue *queue_array; 171 unsigned int i; 172 173 if (!nr_queues) 174 nr_queues = AUXTRACE_INIT_NR_QUEUES; 175 176 while (nr_queues && nr_queues < new_nr_queues) 177 nr_queues <<= 1; 178 179 if (nr_queues < queues->nr_queues || nr_queues < new_nr_queues) 180 return -EINVAL; 181 182 queue_array = auxtrace_alloc_queue_array(nr_queues); 183 if (!queue_array) 184 return -ENOMEM; 185 186 for (i = 0; i < queues->nr_queues; i++) { 187 list_splice_tail(&queues->queue_array[i].head, 188 &queue_array[i].head); 189 queue_array[i].priv = queues->queue_array[i].priv; 190 } 191 192 queues->nr_queues = nr_queues; 193 queues->queue_array = queue_array; 194 195 return 0; 196 } 197 198 static void *auxtrace_copy_data(u64 size, struct perf_session *session) 199 { 200 int fd = perf_data_file__fd(session->file); 201 void *p; 202 ssize_t ret; 203 204 if (size > SSIZE_MAX) 205 return NULL; 206 207 p = malloc(size); 208 if (!p) 209 return NULL; 210 211 ret = readn(fd, p, size); 212 if (ret != (ssize_t)size) { 213 free(p); 214 return NULL; 215 } 216 217 return p; 218 } 219 220 static int auxtrace_queues__add_buffer(struct auxtrace_queues *queues, 221 unsigned int idx, 222 struct auxtrace_buffer *buffer) 223 { 224 struct auxtrace_queue *queue; 225 int err; 226 227 if (idx >= queues->nr_queues) { 228 err = auxtrace_queues__grow(queues, idx + 1); 229 if (err) 230 return err; 231 } 232 233 queue = &queues->queue_array[idx]; 234 235 if (!queue->set) { 236 queue->set = true; 237 queue->tid = buffer->tid; 238 queue->cpu = buffer->cpu; 239 } else if (buffer->cpu != queue->cpu || buffer->tid != queue->tid) { 240 pr_err("auxtrace queue conflict: cpu %d, tid %d vs cpu %d, tid %d\n", 241 queue->cpu, queue->tid, buffer->cpu, buffer->tid); 242 return -EINVAL; 243 } 244 245 buffer->buffer_nr = queues->next_buffer_nr++; 246 247 list_add_tail(&buffer->list, &queue->head); 248 249 queues->new_data = true; 250 queues->populated = true; 251 252 return 0; 253 } 254 255 /* Limit buffers to 32MiB on 32-bit */ 256 #define BUFFER_LIMIT_FOR_32_BIT (32 * 1024 * 1024) 257 258 static int auxtrace_queues__split_buffer(struct auxtrace_queues *queues, 259 unsigned int idx, 260 struct auxtrace_buffer *buffer) 261 { 262 u64 sz = buffer->size; 263 bool consecutive = false; 264 struct auxtrace_buffer *b; 265 int err; 266 267 while (sz > BUFFER_LIMIT_FOR_32_BIT) { 268 b = memdup(buffer, sizeof(struct auxtrace_buffer)); 269 if (!b) 270 return -ENOMEM; 271 b->size = BUFFER_LIMIT_FOR_32_BIT; 272 b->consecutive = consecutive; 273 err = auxtrace_queues__add_buffer(queues, idx, b); 274 if (err) { 275 auxtrace_buffer__free(b); 276 return err; 277 } 278 buffer->data_offset += BUFFER_LIMIT_FOR_32_BIT; 279 sz -= BUFFER_LIMIT_FOR_32_BIT; 280 consecutive = true; 281 } 282 283 buffer->size = sz; 284 buffer->consecutive = consecutive; 285 286 return 0; 287 } 288 289 static int auxtrace_queues__add_event_buffer(struct auxtrace_queues *queues, 290 struct perf_session *session, 291 unsigned int idx, 292 struct auxtrace_buffer *buffer) 293 { 294 if (session->one_mmap) { 295 buffer->data = buffer->data_offset - session->one_mmap_offset + 296 session->one_mmap_addr; 297 } else if (perf_data_file__is_pipe(session->file)) { 298 buffer->data = auxtrace_copy_data(buffer->size, session); 299 if (!buffer->data) 300 return -ENOMEM; 301 buffer->data_needs_freeing = true; 302 } else if (BITS_PER_LONG == 32 && 303 buffer->size > BUFFER_LIMIT_FOR_32_BIT) { 304 int err; 305 306 err = auxtrace_queues__split_buffer(queues, idx, buffer); 307 if (err) 308 return err; 309 } 310 311 return auxtrace_queues__add_buffer(queues, idx, buffer); 312 } 313 314 int auxtrace_queues__add_event(struct auxtrace_queues *queues, 315 struct perf_session *session, 316 union perf_event *event, off_t data_offset, 317 struct auxtrace_buffer **buffer_ptr) 318 { 319 struct auxtrace_buffer *buffer; 320 unsigned int idx; 321 int err; 322 323 buffer = zalloc(sizeof(struct auxtrace_buffer)); 324 if (!buffer) 325 return -ENOMEM; 326 327 buffer->pid = -1; 328 buffer->tid = event->auxtrace.tid; 329 buffer->cpu = event->auxtrace.cpu; 330 buffer->data_offset = data_offset; 331 buffer->offset = event->auxtrace.offset; 332 buffer->reference = event->auxtrace.reference; 333 buffer->size = event->auxtrace.size; 334 idx = event->auxtrace.idx; 335 336 err = auxtrace_queues__add_event_buffer(queues, session, idx, buffer); 337 if (err) 338 goto out_err; 339 340 if (buffer_ptr) 341 *buffer_ptr = buffer; 342 343 return 0; 344 345 out_err: 346 auxtrace_buffer__free(buffer); 347 return err; 348 } 349 350 static int auxtrace_queues__add_indexed_event(struct auxtrace_queues *queues, 351 struct perf_session *session, 352 off_t file_offset, size_t sz) 353 { 354 union perf_event *event; 355 int err; 356 char buf[PERF_SAMPLE_MAX_SIZE]; 357 358 err = perf_session__peek_event(session, file_offset, buf, 359 PERF_SAMPLE_MAX_SIZE, &event, NULL); 360 if (err) 361 return err; 362 363 if (event->header.type == PERF_RECORD_AUXTRACE) { 364 if (event->header.size < sizeof(struct auxtrace_event) || 365 event->header.size != sz) { 366 err = -EINVAL; 367 goto out; 368 } 369 file_offset += event->header.size; 370 err = auxtrace_queues__add_event(queues, session, event, 371 file_offset, NULL); 372 } 373 out: 374 return err; 375 } 376 377 void auxtrace_queues__free(struct auxtrace_queues *queues) 378 { 379 unsigned int i; 380 381 for (i = 0; i < queues->nr_queues; i++) { 382 while (!list_empty(&queues->queue_array[i].head)) { 383 struct auxtrace_buffer *buffer; 384 385 buffer = list_entry(queues->queue_array[i].head.next, 386 struct auxtrace_buffer, list); 387 list_del(&buffer->list); 388 auxtrace_buffer__free(buffer); 389 } 390 } 391 392 zfree(&queues->queue_array); 393 queues->nr_queues = 0; 394 } 395 396 static void auxtrace_heapify(struct auxtrace_heap_item *heap_array, 397 unsigned int pos, unsigned int queue_nr, 398 u64 ordinal) 399 { 400 unsigned int parent; 401 402 while (pos) { 403 parent = (pos - 1) >> 1; 404 if (heap_array[parent].ordinal <= ordinal) 405 break; 406 heap_array[pos] = heap_array[parent]; 407 pos = parent; 408 } 409 heap_array[pos].queue_nr = queue_nr; 410 heap_array[pos].ordinal = ordinal; 411 } 412 413 int auxtrace_heap__add(struct auxtrace_heap *heap, unsigned int queue_nr, 414 u64 ordinal) 415 { 416 struct auxtrace_heap_item *heap_array; 417 418 if (queue_nr >= heap->heap_sz) { 419 unsigned int heap_sz = AUXTRACE_INIT_NR_QUEUES; 420 421 while (heap_sz <= queue_nr) 422 heap_sz <<= 1; 423 heap_array = realloc(heap->heap_array, 424 heap_sz * sizeof(struct auxtrace_heap_item)); 425 if (!heap_array) 426 return -ENOMEM; 427 heap->heap_array = heap_array; 428 heap->heap_sz = heap_sz; 429 } 430 431 auxtrace_heapify(heap->heap_array, heap->heap_cnt++, queue_nr, ordinal); 432 433 return 0; 434 } 435 436 void auxtrace_heap__free(struct auxtrace_heap *heap) 437 { 438 zfree(&heap->heap_array); 439 heap->heap_cnt = 0; 440 heap->heap_sz = 0; 441 } 442 443 void auxtrace_heap__pop(struct auxtrace_heap *heap) 444 { 445 unsigned int pos, last, heap_cnt = heap->heap_cnt; 446 struct auxtrace_heap_item *heap_array; 447 448 if (!heap_cnt) 449 return; 450 451 heap->heap_cnt -= 1; 452 453 heap_array = heap->heap_array; 454 455 pos = 0; 456 while (1) { 457 unsigned int left, right; 458 459 left = (pos << 1) + 1; 460 if (left >= heap_cnt) 461 break; 462 right = left + 1; 463 if (right >= heap_cnt) { 464 heap_array[pos] = heap_array[left]; 465 return; 466 } 467 if (heap_array[left].ordinal < heap_array[right].ordinal) { 468 heap_array[pos] = heap_array[left]; 469 pos = left; 470 } else { 471 heap_array[pos] = heap_array[right]; 472 pos = right; 473 } 474 } 475 476 last = heap_cnt - 1; 477 auxtrace_heapify(heap_array, pos, heap_array[last].queue_nr, 478 heap_array[last].ordinal); 479 } 480 481 size_t auxtrace_record__info_priv_size(struct auxtrace_record *itr) 482 { 483 if (itr) 484 return itr->info_priv_size(itr); 485 return 0; 486 } 487 488 static int auxtrace_not_supported(void) 489 { 490 pr_err("AUX area tracing is not supported on this architecture\n"); 491 return -EINVAL; 492 } 493 494 int auxtrace_record__info_fill(struct auxtrace_record *itr, 495 struct perf_session *session, 496 struct auxtrace_info_event *auxtrace_info, 497 size_t priv_size) 498 { 499 if (itr) 500 return itr->info_fill(itr, session, auxtrace_info, priv_size); 501 return auxtrace_not_supported(); 502 } 503 504 void auxtrace_record__free(struct auxtrace_record *itr) 505 { 506 if (itr) 507 itr->free(itr); 508 } 509 510 int auxtrace_record__snapshot_start(struct auxtrace_record *itr) 511 { 512 if (itr && itr->snapshot_start) 513 return itr->snapshot_start(itr); 514 return 0; 515 } 516 517 int auxtrace_record__snapshot_finish(struct auxtrace_record *itr) 518 { 519 if (itr && itr->snapshot_finish) 520 return itr->snapshot_finish(itr); 521 return 0; 522 } 523 524 int auxtrace_record__find_snapshot(struct auxtrace_record *itr, int idx, 525 struct auxtrace_mmap *mm, 526 unsigned char *data, u64 *head, u64 *old) 527 { 528 if (itr && itr->find_snapshot) 529 return itr->find_snapshot(itr, idx, mm, data, head, old); 530 return 0; 531 } 532 533 int auxtrace_record__options(struct auxtrace_record *itr, 534 struct perf_evlist *evlist, 535 struct record_opts *opts) 536 { 537 if (itr) 538 return itr->recording_options(itr, evlist, opts); 539 return 0; 540 } 541 542 u64 auxtrace_record__reference(struct auxtrace_record *itr) 543 { 544 if (itr) 545 return itr->reference(itr); 546 return 0; 547 } 548 549 int auxtrace_parse_snapshot_options(struct auxtrace_record *itr, 550 struct record_opts *opts, const char *str) 551 { 552 if (!str) 553 return 0; 554 555 if (itr) 556 return itr->parse_snapshot_options(itr, opts, str); 557 558 pr_err("No AUX area tracing to snapshot\n"); 559 return -EINVAL; 560 } 561 562 struct auxtrace_record *__weak 563 auxtrace_record__init(struct perf_evlist *evlist __maybe_unused, int *err) 564 { 565 *err = 0; 566 return NULL; 567 } 568 569 static int auxtrace_index__alloc(struct list_head *head) 570 { 571 struct auxtrace_index *auxtrace_index; 572 573 auxtrace_index = malloc(sizeof(struct auxtrace_index)); 574 if (!auxtrace_index) 575 return -ENOMEM; 576 577 auxtrace_index->nr = 0; 578 INIT_LIST_HEAD(&auxtrace_index->list); 579 580 list_add_tail(&auxtrace_index->list, head); 581 582 return 0; 583 } 584 585 void auxtrace_index__free(struct list_head *head) 586 { 587 struct auxtrace_index *auxtrace_index, *n; 588 589 list_for_each_entry_safe(auxtrace_index, n, head, list) { 590 list_del(&auxtrace_index->list); 591 free(auxtrace_index); 592 } 593 } 594 595 static struct auxtrace_index *auxtrace_index__last(struct list_head *head) 596 { 597 struct auxtrace_index *auxtrace_index; 598 int err; 599 600 if (list_empty(head)) { 601 err = auxtrace_index__alloc(head); 602 if (err) 603 return NULL; 604 } 605 606 auxtrace_index = list_entry(head->prev, struct auxtrace_index, list); 607 608 if (auxtrace_index->nr >= PERF_AUXTRACE_INDEX_ENTRY_COUNT) { 609 err = auxtrace_index__alloc(head); 610 if (err) 611 return NULL; 612 auxtrace_index = list_entry(head->prev, struct auxtrace_index, 613 list); 614 } 615 616 return auxtrace_index; 617 } 618 619 int auxtrace_index__auxtrace_event(struct list_head *head, 620 union perf_event *event, off_t file_offset) 621 { 622 struct auxtrace_index *auxtrace_index; 623 size_t nr; 624 625 auxtrace_index = auxtrace_index__last(head); 626 if (!auxtrace_index) 627 return -ENOMEM; 628 629 nr = auxtrace_index->nr; 630 auxtrace_index->entries[nr].file_offset = file_offset; 631 auxtrace_index->entries[nr].sz = event->header.size; 632 auxtrace_index->nr += 1; 633 634 return 0; 635 } 636 637 static int auxtrace_index__do_write(int fd, 638 struct auxtrace_index *auxtrace_index) 639 { 640 struct auxtrace_index_entry ent; 641 size_t i; 642 643 for (i = 0; i < auxtrace_index->nr; i++) { 644 ent.file_offset = auxtrace_index->entries[i].file_offset; 645 ent.sz = auxtrace_index->entries[i].sz; 646 if (writen(fd, &ent, sizeof(ent)) != sizeof(ent)) 647 return -errno; 648 } 649 return 0; 650 } 651 652 int auxtrace_index__write(int fd, struct list_head *head) 653 { 654 struct auxtrace_index *auxtrace_index; 655 u64 total = 0; 656 int err; 657 658 list_for_each_entry(auxtrace_index, head, list) 659 total += auxtrace_index->nr; 660 661 if (writen(fd, &total, sizeof(total)) != sizeof(total)) 662 return -errno; 663 664 list_for_each_entry(auxtrace_index, head, list) { 665 err = auxtrace_index__do_write(fd, auxtrace_index); 666 if (err) 667 return err; 668 } 669 670 return 0; 671 } 672 673 static int auxtrace_index__process_entry(int fd, struct list_head *head, 674 bool needs_swap) 675 { 676 struct auxtrace_index *auxtrace_index; 677 struct auxtrace_index_entry ent; 678 size_t nr; 679 680 if (readn(fd, &ent, sizeof(ent)) != sizeof(ent)) 681 return -1; 682 683 auxtrace_index = auxtrace_index__last(head); 684 if (!auxtrace_index) 685 return -1; 686 687 nr = auxtrace_index->nr; 688 if (needs_swap) { 689 auxtrace_index->entries[nr].file_offset = 690 bswap_64(ent.file_offset); 691 auxtrace_index->entries[nr].sz = bswap_64(ent.sz); 692 } else { 693 auxtrace_index->entries[nr].file_offset = ent.file_offset; 694 auxtrace_index->entries[nr].sz = ent.sz; 695 } 696 697 auxtrace_index->nr = nr + 1; 698 699 return 0; 700 } 701 702 int auxtrace_index__process(int fd, u64 size, struct perf_session *session, 703 bool needs_swap) 704 { 705 struct list_head *head = &session->auxtrace_index; 706 u64 nr; 707 708 if (readn(fd, &nr, sizeof(u64)) != sizeof(u64)) 709 return -1; 710 711 if (needs_swap) 712 nr = bswap_64(nr); 713 714 if (sizeof(u64) + nr * sizeof(struct auxtrace_index_entry) > size) 715 return -1; 716 717 while (nr--) { 718 int err; 719 720 err = auxtrace_index__process_entry(fd, head, needs_swap); 721 if (err) 722 return -1; 723 } 724 725 return 0; 726 } 727 728 static int auxtrace_queues__process_index_entry(struct auxtrace_queues *queues, 729 struct perf_session *session, 730 struct auxtrace_index_entry *ent) 731 { 732 return auxtrace_queues__add_indexed_event(queues, session, 733 ent->file_offset, ent->sz); 734 } 735 736 int auxtrace_queues__process_index(struct auxtrace_queues *queues, 737 struct perf_session *session) 738 { 739 struct auxtrace_index *auxtrace_index; 740 struct auxtrace_index_entry *ent; 741 size_t i; 742 int err; 743 744 list_for_each_entry(auxtrace_index, &session->auxtrace_index, list) { 745 for (i = 0; i < auxtrace_index->nr; i++) { 746 ent = &auxtrace_index->entries[i]; 747 err = auxtrace_queues__process_index_entry(queues, 748 session, 749 ent); 750 if (err) 751 return err; 752 } 753 } 754 return 0; 755 } 756 757 struct auxtrace_buffer *auxtrace_buffer__next(struct auxtrace_queue *queue, 758 struct auxtrace_buffer *buffer) 759 { 760 if (buffer) { 761 if (list_is_last(&buffer->list, &queue->head)) 762 return NULL; 763 return list_entry(buffer->list.next, struct auxtrace_buffer, 764 list); 765 } else { 766 if (list_empty(&queue->head)) 767 return NULL; 768 return list_entry(queue->head.next, struct auxtrace_buffer, 769 list); 770 } 771 } 772 773 void *auxtrace_buffer__get_data(struct auxtrace_buffer *buffer, int fd) 774 { 775 size_t adj = buffer->data_offset & (page_size - 1); 776 size_t size = buffer->size + adj; 777 off_t file_offset = buffer->data_offset - adj; 778 void *addr; 779 780 if (buffer->data) 781 return buffer->data; 782 783 addr = mmap(NULL, size, PROT_READ, MAP_SHARED, fd, file_offset); 784 if (addr == MAP_FAILED) 785 return NULL; 786 787 buffer->mmap_addr = addr; 788 buffer->mmap_size = size; 789 790 buffer->data = addr + adj; 791 792 return buffer->data; 793 } 794 795 void auxtrace_buffer__put_data(struct auxtrace_buffer *buffer) 796 { 797 if (!buffer->data || !buffer->mmap_addr) 798 return; 799 munmap(buffer->mmap_addr, buffer->mmap_size); 800 buffer->mmap_addr = NULL; 801 buffer->mmap_size = 0; 802 buffer->data = NULL; 803 buffer->use_data = NULL; 804 } 805 806 void auxtrace_buffer__drop_data(struct auxtrace_buffer *buffer) 807 { 808 auxtrace_buffer__put_data(buffer); 809 if (buffer->data_needs_freeing) { 810 buffer->data_needs_freeing = false; 811 zfree(&buffer->data); 812 buffer->use_data = NULL; 813 buffer->size = 0; 814 } 815 } 816 817 void auxtrace_buffer__free(struct auxtrace_buffer *buffer) 818 { 819 auxtrace_buffer__drop_data(buffer); 820 free(buffer); 821 } 822 823 void auxtrace_synth_error(struct auxtrace_error_event *auxtrace_error, int type, 824 int code, int cpu, pid_t pid, pid_t tid, u64 ip, 825 const char *msg) 826 { 827 size_t size; 828 829 memset(auxtrace_error, 0, sizeof(struct auxtrace_error_event)); 830 831 auxtrace_error->header.type = PERF_RECORD_AUXTRACE_ERROR; 832 auxtrace_error->type = type; 833 auxtrace_error->code = code; 834 auxtrace_error->cpu = cpu; 835 auxtrace_error->pid = pid; 836 auxtrace_error->tid = tid; 837 auxtrace_error->ip = ip; 838 strlcpy(auxtrace_error->msg, msg, MAX_AUXTRACE_ERROR_MSG); 839 840 size = (void *)auxtrace_error->msg - (void *)auxtrace_error + 841 strlen(auxtrace_error->msg) + 1; 842 auxtrace_error->header.size = PERF_ALIGN(size, sizeof(u64)); 843 } 844 845 int perf_event__synthesize_auxtrace_info(struct auxtrace_record *itr, 846 struct perf_tool *tool, 847 struct perf_session *session, 848 perf_event__handler_t process) 849 { 850 union perf_event *ev; 851 size_t priv_size; 852 int err; 853 854 pr_debug2("Synthesizing auxtrace information\n"); 855 priv_size = auxtrace_record__info_priv_size(itr); 856 ev = zalloc(sizeof(struct auxtrace_info_event) + priv_size); 857 if (!ev) 858 return -ENOMEM; 859 860 ev->auxtrace_info.header.type = PERF_RECORD_AUXTRACE_INFO; 861 ev->auxtrace_info.header.size = sizeof(struct auxtrace_info_event) + 862 priv_size; 863 err = auxtrace_record__info_fill(itr, session, &ev->auxtrace_info, 864 priv_size); 865 if (err) 866 goto out_free; 867 868 err = process(tool, ev, NULL, NULL); 869 out_free: 870 free(ev); 871 return err; 872 } 873 874 static bool auxtrace__dont_decode(struct perf_session *session) 875 { 876 return !session->itrace_synth_opts || 877 session->itrace_synth_opts->dont_decode; 878 } 879 880 int perf_event__process_auxtrace_info(struct perf_tool *tool __maybe_unused, 881 union perf_event *event, 882 struct perf_session *session) 883 { 884 enum auxtrace_type type = event->auxtrace_info.type; 885 886 if (dump_trace) 887 fprintf(stdout, " type: %u\n", type); 888 889 switch (type) { 890 case PERF_AUXTRACE_INTEL_PT: 891 return intel_pt_process_auxtrace_info(event, session); 892 case PERF_AUXTRACE_INTEL_BTS: 893 return intel_bts_process_auxtrace_info(event, session); 894 case PERF_AUXTRACE_UNKNOWN: 895 default: 896 return -EINVAL; 897 } 898 } 899 900 s64 perf_event__process_auxtrace(struct perf_tool *tool, 901 union perf_event *event, 902 struct perf_session *session) 903 { 904 s64 err; 905 906 if (dump_trace) 907 fprintf(stdout, " size: %#"PRIx64" offset: %#"PRIx64" ref: %#"PRIx64" idx: %u tid: %d cpu: %d\n", 908 event->auxtrace.size, event->auxtrace.offset, 909 event->auxtrace.reference, event->auxtrace.idx, 910 event->auxtrace.tid, event->auxtrace.cpu); 911 912 if (auxtrace__dont_decode(session)) 913 return event->auxtrace.size; 914 915 if (!session->auxtrace || event->header.type != PERF_RECORD_AUXTRACE) 916 return -EINVAL; 917 918 err = session->auxtrace->process_auxtrace_event(session, event, tool); 919 if (err < 0) 920 return err; 921 922 return event->auxtrace.size; 923 } 924 925 #define PERF_ITRACE_DEFAULT_PERIOD_TYPE PERF_ITRACE_PERIOD_NANOSECS 926 #define PERF_ITRACE_DEFAULT_PERIOD 100000 927 #define PERF_ITRACE_DEFAULT_CALLCHAIN_SZ 16 928 #define PERF_ITRACE_MAX_CALLCHAIN_SZ 1024 929 #define PERF_ITRACE_DEFAULT_LAST_BRANCH_SZ 64 930 #define PERF_ITRACE_MAX_LAST_BRANCH_SZ 1024 931 932 void itrace_synth_opts__set_default(struct itrace_synth_opts *synth_opts) 933 { 934 synth_opts->instructions = true; 935 synth_opts->branches = true; 936 synth_opts->transactions = true; 937 synth_opts->errors = true; 938 synth_opts->period_type = PERF_ITRACE_DEFAULT_PERIOD_TYPE; 939 synth_opts->period = PERF_ITRACE_DEFAULT_PERIOD; 940 synth_opts->callchain_sz = PERF_ITRACE_DEFAULT_CALLCHAIN_SZ; 941 synth_opts->last_branch_sz = PERF_ITRACE_DEFAULT_LAST_BRANCH_SZ; 942 } 943 944 /* 945 * Please check tools/perf/Documentation/perf-script.txt for information 946 * about the options parsed here, which is introduced after this cset, 947 * when support in 'perf script' for these options is introduced. 948 */ 949 int itrace_parse_synth_opts(const struct option *opt, const char *str, 950 int unset) 951 { 952 struct itrace_synth_opts *synth_opts = opt->value; 953 const char *p; 954 char *endptr; 955 bool period_type_set = false; 956 bool period_set = false; 957 958 synth_opts->set = true; 959 960 if (unset) { 961 synth_opts->dont_decode = true; 962 return 0; 963 } 964 965 if (!str) { 966 itrace_synth_opts__set_default(synth_opts); 967 return 0; 968 } 969 970 for (p = str; *p;) { 971 switch (*p++) { 972 case 'i': 973 synth_opts->instructions = true; 974 while (*p == ' ' || *p == ',') 975 p += 1; 976 if (isdigit(*p)) { 977 synth_opts->period = strtoull(p, &endptr, 10); 978 period_set = true; 979 p = endptr; 980 while (*p == ' ' || *p == ',') 981 p += 1; 982 switch (*p++) { 983 case 'i': 984 synth_opts->period_type = 985 PERF_ITRACE_PERIOD_INSTRUCTIONS; 986 period_type_set = true; 987 break; 988 case 't': 989 synth_opts->period_type = 990 PERF_ITRACE_PERIOD_TICKS; 991 period_type_set = true; 992 break; 993 case 'm': 994 synth_opts->period *= 1000; 995 /* Fall through */ 996 case 'u': 997 synth_opts->period *= 1000; 998 /* Fall through */ 999 case 'n': 1000 if (*p++ != 's') 1001 goto out_err; 1002 synth_opts->period_type = 1003 PERF_ITRACE_PERIOD_NANOSECS; 1004 period_type_set = true; 1005 break; 1006 case '\0': 1007 goto out; 1008 default: 1009 goto out_err; 1010 } 1011 } 1012 break; 1013 case 'b': 1014 synth_opts->branches = true; 1015 break; 1016 case 'x': 1017 synth_opts->transactions = true; 1018 break; 1019 case 'e': 1020 synth_opts->errors = true; 1021 break; 1022 case 'd': 1023 synth_opts->log = true; 1024 break; 1025 case 'c': 1026 synth_opts->branches = true; 1027 synth_opts->calls = true; 1028 break; 1029 case 'r': 1030 synth_opts->branches = true; 1031 synth_opts->returns = true; 1032 break; 1033 case 'g': 1034 synth_opts->callchain = true; 1035 synth_opts->callchain_sz = 1036 PERF_ITRACE_DEFAULT_CALLCHAIN_SZ; 1037 while (*p == ' ' || *p == ',') 1038 p += 1; 1039 if (isdigit(*p)) { 1040 unsigned int val; 1041 1042 val = strtoul(p, &endptr, 10); 1043 p = endptr; 1044 if (!val || val > PERF_ITRACE_MAX_CALLCHAIN_SZ) 1045 goto out_err; 1046 synth_opts->callchain_sz = val; 1047 } 1048 break; 1049 case 'l': 1050 synth_opts->last_branch = true; 1051 synth_opts->last_branch_sz = 1052 PERF_ITRACE_DEFAULT_LAST_BRANCH_SZ; 1053 while (*p == ' ' || *p == ',') 1054 p += 1; 1055 if (isdigit(*p)) { 1056 unsigned int val; 1057 1058 val = strtoul(p, &endptr, 10); 1059 p = endptr; 1060 if (!val || 1061 val > PERF_ITRACE_MAX_LAST_BRANCH_SZ) 1062 goto out_err; 1063 synth_opts->last_branch_sz = val; 1064 } 1065 break; 1066 case ' ': 1067 case ',': 1068 break; 1069 default: 1070 goto out_err; 1071 } 1072 } 1073 out: 1074 if (synth_opts->instructions) { 1075 if (!period_type_set) 1076 synth_opts->period_type = 1077 PERF_ITRACE_DEFAULT_PERIOD_TYPE; 1078 if (!period_set) 1079 synth_opts->period = PERF_ITRACE_DEFAULT_PERIOD; 1080 } 1081 1082 return 0; 1083 1084 out_err: 1085 pr_err("Bad Instruction Tracing options '%s'\n", str); 1086 return -EINVAL; 1087 } 1088 1089 static const char * const auxtrace_error_type_name[] = { 1090 [PERF_AUXTRACE_ERROR_ITRACE] = "instruction trace", 1091 }; 1092 1093 static const char *auxtrace_error_name(int type) 1094 { 1095 const char *error_type_name = NULL; 1096 1097 if (type < PERF_AUXTRACE_ERROR_MAX) 1098 error_type_name = auxtrace_error_type_name[type]; 1099 if (!error_type_name) 1100 error_type_name = "unknown AUX"; 1101 return error_type_name; 1102 } 1103 1104 size_t perf_event__fprintf_auxtrace_error(union perf_event *event, FILE *fp) 1105 { 1106 struct auxtrace_error_event *e = &event->auxtrace_error; 1107 int ret; 1108 1109 ret = fprintf(fp, " %s error type %u", 1110 auxtrace_error_name(e->type), e->type); 1111 ret += fprintf(fp, " cpu %d pid %d tid %d ip %#"PRIx64" code %u: %s\n", 1112 e->cpu, e->pid, e->tid, e->ip, e->code, e->msg); 1113 return ret; 1114 } 1115 1116 void perf_session__auxtrace_error_inc(struct perf_session *session, 1117 union perf_event *event) 1118 { 1119 struct auxtrace_error_event *e = &event->auxtrace_error; 1120 1121 if (e->type < PERF_AUXTRACE_ERROR_MAX) 1122 session->evlist->stats.nr_auxtrace_errors[e->type] += 1; 1123 } 1124 1125 void events_stats__auxtrace_error_warn(const struct events_stats *stats) 1126 { 1127 int i; 1128 1129 for (i = 0; i < PERF_AUXTRACE_ERROR_MAX; i++) { 1130 if (!stats->nr_auxtrace_errors[i]) 1131 continue; 1132 ui__warning("%u %s errors\n", 1133 stats->nr_auxtrace_errors[i], 1134 auxtrace_error_name(i)); 1135 } 1136 } 1137 1138 int perf_event__process_auxtrace_error(struct perf_tool *tool __maybe_unused, 1139 union perf_event *event, 1140 struct perf_session *session) 1141 { 1142 if (auxtrace__dont_decode(session)) 1143 return 0; 1144 1145 perf_event__fprintf_auxtrace_error(event, stdout); 1146 return 0; 1147 } 1148 1149 static int __auxtrace_mmap__read(struct auxtrace_mmap *mm, 1150 struct auxtrace_record *itr, 1151 struct perf_tool *tool, process_auxtrace_t fn, 1152 bool snapshot, size_t snapshot_size) 1153 { 1154 u64 head, old = mm->prev, offset, ref; 1155 unsigned char *data = mm->base; 1156 size_t size, head_off, old_off, len1, len2, padding; 1157 union perf_event ev; 1158 void *data1, *data2; 1159 1160 if (snapshot) { 1161 head = auxtrace_mmap__read_snapshot_head(mm); 1162 if (auxtrace_record__find_snapshot(itr, mm->idx, mm, data, 1163 &head, &old)) 1164 return -1; 1165 } else { 1166 head = auxtrace_mmap__read_head(mm); 1167 } 1168 1169 if (old == head) 1170 return 0; 1171 1172 pr_debug3("auxtrace idx %d old %#"PRIx64" head %#"PRIx64" diff %#"PRIx64"\n", 1173 mm->idx, old, head, head - old); 1174 1175 if (mm->mask) { 1176 head_off = head & mm->mask; 1177 old_off = old & mm->mask; 1178 } else { 1179 head_off = head % mm->len; 1180 old_off = old % mm->len; 1181 } 1182 1183 if (head_off > old_off) 1184 size = head_off - old_off; 1185 else 1186 size = mm->len - (old_off - head_off); 1187 1188 if (snapshot && size > snapshot_size) 1189 size = snapshot_size; 1190 1191 ref = auxtrace_record__reference(itr); 1192 1193 if (head > old || size <= head || mm->mask) { 1194 offset = head - size; 1195 } else { 1196 /* 1197 * When the buffer size is not a power of 2, 'head' wraps at the 1198 * highest multiple of the buffer size, so we have to subtract 1199 * the remainder here. 1200 */ 1201 u64 rem = (0ULL - mm->len) % mm->len; 1202 1203 offset = head - size - rem; 1204 } 1205 1206 if (size > head_off) { 1207 len1 = size - head_off; 1208 data1 = &data[mm->len - len1]; 1209 len2 = head_off; 1210 data2 = &data[0]; 1211 } else { 1212 len1 = size; 1213 data1 = &data[head_off - len1]; 1214 len2 = 0; 1215 data2 = NULL; 1216 } 1217 1218 if (itr->alignment) { 1219 unsigned int unwanted = len1 % itr->alignment; 1220 1221 len1 -= unwanted; 1222 size -= unwanted; 1223 } 1224 1225 /* padding must be written by fn() e.g. record__process_auxtrace() */ 1226 padding = size & 7; 1227 if (padding) 1228 padding = 8 - padding; 1229 1230 memset(&ev, 0, sizeof(ev)); 1231 ev.auxtrace.header.type = PERF_RECORD_AUXTRACE; 1232 ev.auxtrace.header.size = sizeof(ev.auxtrace); 1233 ev.auxtrace.size = size + padding; 1234 ev.auxtrace.offset = offset; 1235 ev.auxtrace.reference = ref; 1236 ev.auxtrace.idx = mm->idx; 1237 ev.auxtrace.tid = mm->tid; 1238 ev.auxtrace.cpu = mm->cpu; 1239 1240 if (fn(tool, &ev, data1, len1, data2, len2)) 1241 return -1; 1242 1243 mm->prev = head; 1244 1245 if (!snapshot) { 1246 auxtrace_mmap__write_tail(mm, head); 1247 if (itr->read_finish) { 1248 int err; 1249 1250 err = itr->read_finish(itr, mm->idx); 1251 if (err < 0) 1252 return err; 1253 } 1254 } 1255 1256 return 1; 1257 } 1258 1259 int auxtrace_mmap__read(struct auxtrace_mmap *mm, struct auxtrace_record *itr, 1260 struct perf_tool *tool, process_auxtrace_t fn) 1261 { 1262 return __auxtrace_mmap__read(mm, itr, tool, fn, false, 0); 1263 } 1264 1265 int auxtrace_mmap__read_snapshot(struct auxtrace_mmap *mm, 1266 struct auxtrace_record *itr, 1267 struct perf_tool *tool, process_auxtrace_t fn, 1268 size_t snapshot_size) 1269 { 1270 return __auxtrace_mmap__read(mm, itr, tool, fn, true, snapshot_size); 1271 } 1272 1273 /** 1274 * struct auxtrace_cache - hash table to implement a cache 1275 * @hashtable: the hashtable 1276 * @sz: hashtable size (number of hlists) 1277 * @entry_size: size of an entry 1278 * @limit: limit the number of entries to this maximum, when reached the cache 1279 * is dropped and caching begins again with an empty cache 1280 * @cnt: current number of entries 1281 * @bits: hashtable size (@sz = 2^@bits) 1282 */ 1283 struct auxtrace_cache { 1284 struct hlist_head *hashtable; 1285 size_t sz; 1286 size_t entry_size; 1287 size_t limit; 1288 size_t cnt; 1289 unsigned int bits; 1290 }; 1291 1292 struct auxtrace_cache *auxtrace_cache__new(unsigned int bits, size_t entry_size, 1293 unsigned int limit_percent) 1294 { 1295 struct auxtrace_cache *c; 1296 struct hlist_head *ht; 1297 size_t sz, i; 1298 1299 c = zalloc(sizeof(struct auxtrace_cache)); 1300 if (!c) 1301 return NULL; 1302 1303 sz = 1UL << bits; 1304 1305 ht = calloc(sz, sizeof(struct hlist_head)); 1306 if (!ht) 1307 goto out_free; 1308 1309 for (i = 0; i < sz; i++) 1310 INIT_HLIST_HEAD(&ht[i]); 1311 1312 c->hashtable = ht; 1313 c->sz = sz; 1314 c->entry_size = entry_size; 1315 c->limit = (c->sz * limit_percent) / 100; 1316 c->bits = bits; 1317 1318 return c; 1319 1320 out_free: 1321 free(c); 1322 return NULL; 1323 } 1324 1325 static void auxtrace_cache__drop(struct auxtrace_cache *c) 1326 { 1327 struct auxtrace_cache_entry *entry; 1328 struct hlist_node *tmp; 1329 size_t i; 1330 1331 if (!c) 1332 return; 1333 1334 for (i = 0; i < c->sz; i++) { 1335 hlist_for_each_entry_safe(entry, tmp, &c->hashtable[i], hash) { 1336 hlist_del(&entry->hash); 1337 auxtrace_cache__free_entry(c, entry); 1338 } 1339 } 1340 1341 c->cnt = 0; 1342 } 1343 1344 void auxtrace_cache__free(struct auxtrace_cache *c) 1345 { 1346 if (!c) 1347 return; 1348 1349 auxtrace_cache__drop(c); 1350 free(c->hashtable); 1351 free(c); 1352 } 1353 1354 void *auxtrace_cache__alloc_entry(struct auxtrace_cache *c) 1355 { 1356 return malloc(c->entry_size); 1357 } 1358 1359 void auxtrace_cache__free_entry(struct auxtrace_cache *c __maybe_unused, 1360 void *entry) 1361 { 1362 free(entry); 1363 } 1364 1365 int auxtrace_cache__add(struct auxtrace_cache *c, u32 key, 1366 struct auxtrace_cache_entry *entry) 1367 { 1368 if (c->limit && ++c->cnt > c->limit) 1369 auxtrace_cache__drop(c); 1370 1371 entry->key = key; 1372 hlist_add_head(&entry->hash, &c->hashtable[hash_32(key, c->bits)]); 1373 1374 return 0; 1375 } 1376 1377 void *auxtrace_cache__lookup(struct auxtrace_cache *c, u32 key) 1378 { 1379 struct auxtrace_cache_entry *entry; 1380 struct hlist_head *hlist; 1381 1382 if (!c) 1383 return NULL; 1384 1385 hlist = &c->hashtable[hash_32(key, c->bits)]; 1386 hlist_for_each_entry(entry, hlist, hash) { 1387 if (entry->key == key) 1388 return entry; 1389 } 1390 1391 return NULL; 1392 } 1393