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 <subcmd/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 struct perf_evlist *evlist) 483 { 484 if (itr) 485 return itr->info_priv_size(itr, evlist); 486 return 0; 487 } 488 489 static int auxtrace_not_supported(void) 490 { 491 pr_err("AUX area tracing is not supported on this architecture\n"); 492 return -EINVAL; 493 } 494 495 int auxtrace_record__info_fill(struct auxtrace_record *itr, 496 struct perf_session *session, 497 struct auxtrace_info_event *auxtrace_info, 498 size_t priv_size) 499 { 500 if (itr) 501 return itr->info_fill(itr, session, auxtrace_info, priv_size); 502 return auxtrace_not_supported(); 503 } 504 505 void auxtrace_record__free(struct auxtrace_record *itr) 506 { 507 if (itr) 508 itr->free(itr); 509 } 510 511 int auxtrace_record__snapshot_start(struct auxtrace_record *itr) 512 { 513 if (itr && itr->snapshot_start) 514 return itr->snapshot_start(itr); 515 return 0; 516 } 517 518 int auxtrace_record__snapshot_finish(struct auxtrace_record *itr) 519 { 520 if (itr && itr->snapshot_finish) 521 return itr->snapshot_finish(itr); 522 return 0; 523 } 524 525 int auxtrace_record__find_snapshot(struct auxtrace_record *itr, int idx, 526 struct auxtrace_mmap *mm, 527 unsigned char *data, u64 *head, u64 *old) 528 { 529 if (itr && itr->find_snapshot) 530 return itr->find_snapshot(itr, idx, mm, data, head, old); 531 return 0; 532 } 533 534 int auxtrace_record__options(struct auxtrace_record *itr, 535 struct perf_evlist *evlist, 536 struct record_opts *opts) 537 { 538 if (itr) 539 return itr->recording_options(itr, evlist, opts); 540 return 0; 541 } 542 543 u64 auxtrace_record__reference(struct auxtrace_record *itr) 544 { 545 if (itr) 546 return itr->reference(itr); 547 return 0; 548 } 549 550 int auxtrace_parse_snapshot_options(struct auxtrace_record *itr, 551 struct record_opts *opts, const char *str) 552 { 553 if (!str) 554 return 0; 555 556 if (itr) 557 return itr->parse_snapshot_options(itr, opts, str); 558 559 pr_err("No AUX area tracing to snapshot\n"); 560 return -EINVAL; 561 } 562 563 struct auxtrace_record *__weak 564 auxtrace_record__init(struct perf_evlist *evlist __maybe_unused, int *err) 565 { 566 *err = 0; 567 return NULL; 568 } 569 570 static int auxtrace_index__alloc(struct list_head *head) 571 { 572 struct auxtrace_index *auxtrace_index; 573 574 auxtrace_index = malloc(sizeof(struct auxtrace_index)); 575 if (!auxtrace_index) 576 return -ENOMEM; 577 578 auxtrace_index->nr = 0; 579 INIT_LIST_HEAD(&auxtrace_index->list); 580 581 list_add_tail(&auxtrace_index->list, head); 582 583 return 0; 584 } 585 586 void auxtrace_index__free(struct list_head *head) 587 { 588 struct auxtrace_index *auxtrace_index, *n; 589 590 list_for_each_entry_safe(auxtrace_index, n, head, list) { 591 list_del(&auxtrace_index->list); 592 free(auxtrace_index); 593 } 594 } 595 596 static struct auxtrace_index *auxtrace_index__last(struct list_head *head) 597 { 598 struct auxtrace_index *auxtrace_index; 599 int err; 600 601 if (list_empty(head)) { 602 err = auxtrace_index__alloc(head); 603 if (err) 604 return NULL; 605 } 606 607 auxtrace_index = list_entry(head->prev, struct auxtrace_index, list); 608 609 if (auxtrace_index->nr >= PERF_AUXTRACE_INDEX_ENTRY_COUNT) { 610 err = auxtrace_index__alloc(head); 611 if (err) 612 return NULL; 613 auxtrace_index = list_entry(head->prev, struct auxtrace_index, 614 list); 615 } 616 617 return auxtrace_index; 618 } 619 620 int auxtrace_index__auxtrace_event(struct list_head *head, 621 union perf_event *event, off_t file_offset) 622 { 623 struct auxtrace_index *auxtrace_index; 624 size_t nr; 625 626 auxtrace_index = auxtrace_index__last(head); 627 if (!auxtrace_index) 628 return -ENOMEM; 629 630 nr = auxtrace_index->nr; 631 auxtrace_index->entries[nr].file_offset = file_offset; 632 auxtrace_index->entries[nr].sz = event->header.size; 633 auxtrace_index->nr += 1; 634 635 return 0; 636 } 637 638 static int auxtrace_index__do_write(int fd, 639 struct auxtrace_index *auxtrace_index) 640 { 641 struct auxtrace_index_entry ent; 642 size_t i; 643 644 for (i = 0; i < auxtrace_index->nr; i++) { 645 ent.file_offset = auxtrace_index->entries[i].file_offset; 646 ent.sz = auxtrace_index->entries[i].sz; 647 if (writen(fd, &ent, sizeof(ent)) != sizeof(ent)) 648 return -errno; 649 } 650 return 0; 651 } 652 653 int auxtrace_index__write(int fd, struct list_head *head) 654 { 655 struct auxtrace_index *auxtrace_index; 656 u64 total = 0; 657 int err; 658 659 list_for_each_entry(auxtrace_index, head, list) 660 total += auxtrace_index->nr; 661 662 if (writen(fd, &total, sizeof(total)) != sizeof(total)) 663 return -errno; 664 665 list_for_each_entry(auxtrace_index, head, list) { 666 err = auxtrace_index__do_write(fd, auxtrace_index); 667 if (err) 668 return err; 669 } 670 671 return 0; 672 } 673 674 static int auxtrace_index__process_entry(int fd, struct list_head *head, 675 bool needs_swap) 676 { 677 struct auxtrace_index *auxtrace_index; 678 struct auxtrace_index_entry ent; 679 size_t nr; 680 681 if (readn(fd, &ent, sizeof(ent)) != sizeof(ent)) 682 return -1; 683 684 auxtrace_index = auxtrace_index__last(head); 685 if (!auxtrace_index) 686 return -1; 687 688 nr = auxtrace_index->nr; 689 if (needs_swap) { 690 auxtrace_index->entries[nr].file_offset = 691 bswap_64(ent.file_offset); 692 auxtrace_index->entries[nr].sz = bswap_64(ent.sz); 693 } else { 694 auxtrace_index->entries[nr].file_offset = ent.file_offset; 695 auxtrace_index->entries[nr].sz = ent.sz; 696 } 697 698 auxtrace_index->nr = nr + 1; 699 700 return 0; 701 } 702 703 int auxtrace_index__process(int fd, u64 size, struct perf_session *session, 704 bool needs_swap) 705 { 706 struct list_head *head = &session->auxtrace_index; 707 u64 nr; 708 709 if (readn(fd, &nr, sizeof(u64)) != sizeof(u64)) 710 return -1; 711 712 if (needs_swap) 713 nr = bswap_64(nr); 714 715 if (sizeof(u64) + nr * sizeof(struct auxtrace_index_entry) > size) 716 return -1; 717 718 while (nr--) { 719 int err; 720 721 err = auxtrace_index__process_entry(fd, head, needs_swap); 722 if (err) 723 return -1; 724 } 725 726 return 0; 727 } 728 729 static int auxtrace_queues__process_index_entry(struct auxtrace_queues *queues, 730 struct perf_session *session, 731 struct auxtrace_index_entry *ent) 732 { 733 return auxtrace_queues__add_indexed_event(queues, session, 734 ent->file_offset, ent->sz); 735 } 736 737 int auxtrace_queues__process_index(struct auxtrace_queues *queues, 738 struct perf_session *session) 739 { 740 struct auxtrace_index *auxtrace_index; 741 struct auxtrace_index_entry *ent; 742 size_t i; 743 int err; 744 745 list_for_each_entry(auxtrace_index, &session->auxtrace_index, list) { 746 for (i = 0; i < auxtrace_index->nr; i++) { 747 ent = &auxtrace_index->entries[i]; 748 err = auxtrace_queues__process_index_entry(queues, 749 session, 750 ent); 751 if (err) 752 return err; 753 } 754 } 755 return 0; 756 } 757 758 struct auxtrace_buffer *auxtrace_buffer__next(struct auxtrace_queue *queue, 759 struct auxtrace_buffer *buffer) 760 { 761 if (buffer) { 762 if (list_is_last(&buffer->list, &queue->head)) 763 return NULL; 764 return list_entry(buffer->list.next, struct auxtrace_buffer, 765 list); 766 } else { 767 if (list_empty(&queue->head)) 768 return NULL; 769 return list_entry(queue->head.next, struct auxtrace_buffer, 770 list); 771 } 772 } 773 774 void *auxtrace_buffer__get_data(struct auxtrace_buffer *buffer, int fd) 775 { 776 size_t adj = buffer->data_offset & (page_size - 1); 777 size_t size = buffer->size + adj; 778 off_t file_offset = buffer->data_offset - adj; 779 void *addr; 780 781 if (buffer->data) 782 return buffer->data; 783 784 addr = mmap(NULL, size, PROT_READ, MAP_SHARED, fd, file_offset); 785 if (addr == MAP_FAILED) 786 return NULL; 787 788 buffer->mmap_addr = addr; 789 buffer->mmap_size = size; 790 791 buffer->data = addr + adj; 792 793 return buffer->data; 794 } 795 796 void auxtrace_buffer__put_data(struct auxtrace_buffer *buffer) 797 { 798 if (!buffer->data || !buffer->mmap_addr) 799 return; 800 munmap(buffer->mmap_addr, buffer->mmap_size); 801 buffer->mmap_addr = NULL; 802 buffer->mmap_size = 0; 803 buffer->data = NULL; 804 buffer->use_data = NULL; 805 } 806 807 void auxtrace_buffer__drop_data(struct auxtrace_buffer *buffer) 808 { 809 auxtrace_buffer__put_data(buffer); 810 if (buffer->data_needs_freeing) { 811 buffer->data_needs_freeing = false; 812 zfree(&buffer->data); 813 buffer->use_data = NULL; 814 buffer->size = 0; 815 } 816 } 817 818 void auxtrace_buffer__free(struct auxtrace_buffer *buffer) 819 { 820 auxtrace_buffer__drop_data(buffer); 821 free(buffer); 822 } 823 824 void auxtrace_synth_error(struct auxtrace_error_event *auxtrace_error, int type, 825 int code, int cpu, pid_t pid, pid_t tid, u64 ip, 826 const char *msg) 827 { 828 size_t size; 829 830 memset(auxtrace_error, 0, sizeof(struct auxtrace_error_event)); 831 832 auxtrace_error->header.type = PERF_RECORD_AUXTRACE_ERROR; 833 auxtrace_error->type = type; 834 auxtrace_error->code = code; 835 auxtrace_error->cpu = cpu; 836 auxtrace_error->pid = pid; 837 auxtrace_error->tid = tid; 838 auxtrace_error->ip = ip; 839 strlcpy(auxtrace_error->msg, msg, MAX_AUXTRACE_ERROR_MSG); 840 841 size = (void *)auxtrace_error->msg - (void *)auxtrace_error + 842 strlen(auxtrace_error->msg) + 1; 843 auxtrace_error->header.size = PERF_ALIGN(size, sizeof(u64)); 844 } 845 846 int perf_event__synthesize_auxtrace_info(struct auxtrace_record *itr, 847 struct perf_tool *tool, 848 struct perf_session *session, 849 perf_event__handler_t process) 850 { 851 union perf_event *ev; 852 size_t priv_size; 853 int err; 854 855 pr_debug2("Synthesizing auxtrace information\n"); 856 priv_size = auxtrace_record__info_priv_size(itr, session->evlist); 857 ev = zalloc(sizeof(struct auxtrace_info_event) + priv_size); 858 if (!ev) 859 return -ENOMEM; 860 861 ev->auxtrace_info.header.type = PERF_RECORD_AUXTRACE_INFO; 862 ev->auxtrace_info.header.size = sizeof(struct auxtrace_info_event) + 863 priv_size; 864 err = auxtrace_record__info_fill(itr, session, &ev->auxtrace_info, 865 priv_size); 866 if (err) 867 goto out_free; 868 869 err = process(tool, ev, NULL, NULL); 870 out_free: 871 free(ev); 872 return err; 873 } 874 875 static bool auxtrace__dont_decode(struct perf_session *session) 876 { 877 return !session->itrace_synth_opts || 878 session->itrace_synth_opts->dont_decode; 879 } 880 881 int perf_event__process_auxtrace_info(struct perf_tool *tool __maybe_unused, 882 union perf_event *event, 883 struct perf_session *session) 884 { 885 enum auxtrace_type type = event->auxtrace_info.type; 886 887 if (dump_trace) 888 fprintf(stdout, " type: %u\n", type); 889 890 switch (type) { 891 case PERF_AUXTRACE_INTEL_PT: 892 return intel_pt_process_auxtrace_info(event, session); 893 case PERF_AUXTRACE_INTEL_BTS: 894 return intel_bts_process_auxtrace_info(event, session); 895 case PERF_AUXTRACE_UNKNOWN: 896 default: 897 return -EINVAL; 898 } 899 } 900 901 s64 perf_event__process_auxtrace(struct perf_tool *tool, 902 union perf_event *event, 903 struct perf_session *session) 904 { 905 s64 err; 906 907 if (dump_trace) 908 fprintf(stdout, " size: %#"PRIx64" offset: %#"PRIx64" ref: %#"PRIx64" idx: %u tid: %d cpu: %d\n", 909 event->auxtrace.size, event->auxtrace.offset, 910 event->auxtrace.reference, event->auxtrace.idx, 911 event->auxtrace.tid, event->auxtrace.cpu); 912 913 if (auxtrace__dont_decode(session)) 914 return event->auxtrace.size; 915 916 if (!session->auxtrace || event->header.type != PERF_RECORD_AUXTRACE) 917 return -EINVAL; 918 919 err = session->auxtrace->process_auxtrace_event(session, event, tool); 920 if (err < 0) 921 return err; 922 923 return event->auxtrace.size; 924 } 925 926 #define PERF_ITRACE_DEFAULT_PERIOD_TYPE PERF_ITRACE_PERIOD_NANOSECS 927 #define PERF_ITRACE_DEFAULT_PERIOD 100000 928 #define PERF_ITRACE_DEFAULT_CALLCHAIN_SZ 16 929 #define PERF_ITRACE_MAX_CALLCHAIN_SZ 1024 930 #define PERF_ITRACE_DEFAULT_LAST_BRANCH_SZ 64 931 #define PERF_ITRACE_MAX_LAST_BRANCH_SZ 1024 932 933 void itrace_synth_opts__set_default(struct itrace_synth_opts *synth_opts) 934 { 935 synth_opts->instructions = true; 936 synth_opts->branches = true; 937 synth_opts->transactions = true; 938 synth_opts->errors = true; 939 synth_opts->period_type = PERF_ITRACE_DEFAULT_PERIOD_TYPE; 940 synth_opts->period = PERF_ITRACE_DEFAULT_PERIOD; 941 synth_opts->callchain_sz = PERF_ITRACE_DEFAULT_CALLCHAIN_SZ; 942 synth_opts->last_branch_sz = PERF_ITRACE_DEFAULT_LAST_BRANCH_SZ; 943 synth_opts->initial_skip = 0; 944 } 945 946 /* 947 * Please check tools/perf/Documentation/perf-script.txt for information 948 * about the options parsed here, which is introduced after this cset, 949 * when support in 'perf script' for these options is introduced. 950 */ 951 int itrace_parse_synth_opts(const struct option *opt, const char *str, 952 int unset) 953 { 954 struct itrace_synth_opts *synth_opts = opt->value; 955 const char *p; 956 char *endptr; 957 bool period_type_set = false; 958 bool period_set = false; 959 960 synth_opts->set = true; 961 962 if (unset) { 963 synth_opts->dont_decode = true; 964 return 0; 965 } 966 967 if (!str) { 968 itrace_synth_opts__set_default(synth_opts); 969 return 0; 970 } 971 972 for (p = str; *p;) { 973 switch (*p++) { 974 case 'i': 975 synth_opts->instructions = true; 976 while (*p == ' ' || *p == ',') 977 p += 1; 978 if (isdigit(*p)) { 979 synth_opts->period = strtoull(p, &endptr, 10); 980 period_set = true; 981 p = endptr; 982 while (*p == ' ' || *p == ',') 983 p += 1; 984 switch (*p++) { 985 case 'i': 986 synth_opts->period_type = 987 PERF_ITRACE_PERIOD_INSTRUCTIONS; 988 period_type_set = true; 989 break; 990 case 't': 991 synth_opts->period_type = 992 PERF_ITRACE_PERIOD_TICKS; 993 period_type_set = true; 994 break; 995 case 'm': 996 synth_opts->period *= 1000; 997 /* Fall through */ 998 case 'u': 999 synth_opts->period *= 1000; 1000 /* Fall through */ 1001 case 'n': 1002 if (*p++ != 's') 1003 goto out_err; 1004 synth_opts->period_type = 1005 PERF_ITRACE_PERIOD_NANOSECS; 1006 period_type_set = true; 1007 break; 1008 case '\0': 1009 goto out; 1010 default: 1011 goto out_err; 1012 } 1013 } 1014 break; 1015 case 'b': 1016 synth_opts->branches = true; 1017 break; 1018 case 'x': 1019 synth_opts->transactions = true; 1020 break; 1021 case 'e': 1022 synth_opts->errors = true; 1023 break; 1024 case 'd': 1025 synth_opts->log = true; 1026 break; 1027 case 'c': 1028 synth_opts->branches = true; 1029 synth_opts->calls = true; 1030 break; 1031 case 'r': 1032 synth_opts->branches = true; 1033 synth_opts->returns = true; 1034 break; 1035 case 'g': 1036 synth_opts->callchain = true; 1037 synth_opts->callchain_sz = 1038 PERF_ITRACE_DEFAULT_CALLCHAIN_SZ; 1039 while (*p == ' ' || *p == ',') 1040 p += 1; 1041 if (isdigit(*p)) { 1042 unsigned int val; 1043 1044 val = strtoul(p, &endptr, 10); 1045 p = endptr; 1046 if (!val || val > PERF_ITRACE_MAX_CALLCHAIN_SZ) 1047 goto out_err; 1048 synth_opts->callchain_sz = val; 1049 } 1050 break; 1051 case 'l': 1052 synth_opts->last_branch = true; 1053 synth_opts->last_branch_sz = 1054 PERF_ITRACE_DEFAULT_LAST_BRANCH_SZ; 1055 while (*p == ' ' || *p == ',') 1056 p += 1; 1057 if (isdigit(*p)) { 1058 unsigned int val; 1059 1060 val = strtoul(p, &endptr, 10); 1061 p = endptr; 1062 if (!val || 1063 val > PERF_ITRACE_MAX_LAST_BRANCH_SZ) 1064 goto out_err; 1065 synth_opts->last_branch_sz = val; 1066 } 1067 break; 1068 case 's': 1069 synth_opts->initial_skip = strtoul(p, &endptr, 10); 1070 if (p == endptr) 1071 goto out_err; 1072 p = endptr; 1073 break; 1074 case ' ': 1075 case ',': 1076 break; 1077 default: 1078 goto out_err; 1079 } 1080 } 1081 out: 1082 if (synth_opts->instructions) { 1083 if (!period_type_set) 1084 synth_opts->period_type = 1085 PERF_ITRACE_DEFAULT_PERIOD_TYPE; 1086 if (!period_set) 1087 synth_opts->period = PERF_ITRACE_DEFAULT_PERIOD; 1088 } 1089 1090 return 0; 1091 1092 out_err: 1093 pr_err("Bad Instruction Tracing options '%s'\n", str); 1094 return -EINVAL; 1095 } 1096 1097 static const char * const auxtrace_error_type_name[] = { 1098 [PERF_AUXTRACE_ERROR_ITRACE] = "instruction trace", 1099 }; 1100 1101 static const char *auxtrace_error_name(int type) 1102 { 1103 const char *error_type_name = NULL; 1104 1105 if (type < PERF_AUXTRACE_ERROR_MAX) 1106 error_type_name = auxtrace_error_type_name[type]; 1107 if (!error_type_name) 1108 error_type_name = "unknown AUX"; 1109 return error_type_name; 1110 } 1111 1112 size_t perf_event__fprintf_auxtrace_error(union perf_event *event, FILE *fp) 1113 { 1114 struct auxtrace_error_event *e = &event->auxtrace_error; 1115 int ret; 1116 1117 ret = fprintf(fp, " %s error type %u", 1118 auxtrace_error_name(e->type), e->type); 1119 ret += fprintf(fp, " cpu %d pid %d tid %d ip %#"PRIx64" code %u: %s\n", 1120 e->cpu, e->pid, e->tid, e->ip, e->code, e->msg); 1121 return ret; 1122 } 1123 1124 void perf_session__auxtrace_error_inc(struct perf_session *session, 1125 union perf_event *event) 1126 { 1127 struct auxtrace_error_event *e = &event->auxtrace_error; 1128 1129 if (e->type < PERF_AUXTRACE_ERROR_MAX) 1130 session->evlist->stats.nr_auxtrace_errors[e->type] += 1; 1131 } 1132 1133 void events_stats__auxtrace_error_warn(const struct events_stats *stats) 1134 { 1135 int i; 1136 1137 for (i = 0; i < PERF_AUXTRACE_ERROR_MAX; i++) { 1138 if (!stats->nr_auxtrace_errors[i]) 1139 continue; 1140 ui__warning("%u %s errors\n", 1141 stats->nr_auxtrace_errors[i], 1142 auxtrace_error_name(i)); 1143 } 1144 } 1145 1146 int perf_event__process_auxtrace_error(struct perf_tool *tool __maybe_unused, 1147 union perf_event *event, 1148 struct perf_session *session) 1149 { 1150 if (auxtrace__dont_decode(session)) 1151 return 0; 1152 1153 perf_event__fprintf_auxtrace_error(event, stdout); 1154 return 0; 1155 } 1156 1157 static int __auxtrace_mmap__read(struct auxtrace_mmap *mm, 1158 struct auxtrace_record *itr, 1159 struct perf_tool *tool, process_auxtrace_t fn, 1160 bool snapshot, size_t snapshot_size) 1161 { 1162 u64 head, old = mm->prev, offset, ref; 1163 unsigned char *data = mm->base; 1164 size_t size, head_off, old_off, len1, len2, padding; 1165 union perf_event ev; 1166 void *data1, *data2; 1167 1168 if (snapshot) { 1169 head = auxtrace_mmap__read_snapshot_head(mm); 1170 if (auxtrace_record__find_snapshot(itr, mm->idx, mm, data, 1171 &head, &old)) 1172 return -1; 1173 } else { 1174 head = auxtrace_mmap__read_head(mm); 1175 } 1176 1177 if (old == head) 1178 return 0; 1179 1180 pr_debug3("auxtrace idx %d old %#"PRIx64" head %#"PRIx64" diff %#"PRIx64"\n", 1181 mm->idx, old, head, head - old); 1182 1183 if (mm->mask) { 1184 head_off = head & mm->mask; 1185 old_off = old & mm->mask; 1186 } else { 1187 head_off = head % mm->len; 1188 old_off = old % mm->len; 1189 } 1190 1191 if (head_off > old_off) 1192 size = head_off - old_off; 1193 else 1194 size = mm->len - (old_off - head_off); 1195 1196 if (snapshot && size > snapshot_size) 1197 size = snapshot_size; 1198 1199 ref = auxtrace_record__reference(itr); 1200 1201 if (head > old || size <= head || mm->mask) { 1202 offset = head - size; 1203 } else { 1204 /* 1205 * When the buffer size is not a power of 2, 'head' wraps at the 1206 * highest multiple of the buffer size, so we have to subtract 1207 * the remainder here. 1208 */ 1209 u64 rem = (0ULL - mm->len) % mm->len; 1210 1211 offset = head - size - rem; 1212 } 1213 1214 if (size > head_off) { 1215 len1 = size - head_off; 1216 data1 = &data[mm->len - len1]; 1217 len2 = head_off; 1218 data2 = &data[0]; 1219 } else { 1220 len1 = size; 1221 data1 = &data[head_off - len1]; 1222 len2 = 0; 1223 data2 = NULL; 1224 } 1225 1226 if (itr->alignment) { 1227 unsigned int unwanted = len1 % itr->alignment; 1228 1229 len1 -= unwanted; 1230 size -= unwanted; 1231 } 1232 1233 /* padding must be written by fn() e.g. record__process_auxtrace() */ 1234 padding = size & 7; 1235 if (padding) 1236 padding = 8 - padding; 1237 1238 memset(&ev, 0, sizeof(ev)); 1239 ev.auxtrace.header.type = PERF_RECORD_AUXTRACE; 1240 ev.auxtrace.header.size = sizeof(ev.auxtrace); 1241 ev.auxtrace.size = size + padding; 1242 ev.auxtrace.offset = offset; 1243 ev.auxtrace.reference = ref; 1244 ev.auxtrace.idx = mm->idx; 1245 ev.auxtrace.tid = mm->tid; 1246 ev.auxtrace.cpu = mm->cpu; 1247 1248 if (fn(tool, &ev, data1, len1, data2, len2)) 1249 return -1; 1250 1251 mm->prev = head; 1252 1253 if (!snapshot) { 1254 auxtrace_mmap__write_tail(mm, head); 1255 if (itr->read_finish) { 1256 int err; 1257 1258 err = itr->read_finish(itr, mm->idx); 1259 if (err < 0) 1260 return err; 1261 } 1262 } 1263 1264 return 1; 1265 } 1266 1267 int auxtrace_mmap__read(struct auxtrace_mmap *mm, struct auxtrace_record *itr, 1268 struct perf_tool *tool, process_auxtrace_t fn) 1269 { 1270 return __auxtrace_mmap__read(mm, itr, tool, fn, false, 0); 1271 } 1272 1273 int auxtrace_mmap__read_snapshot(struct auxtrace_mmap *mm, 1274 struct auxtrace_record *itr, 1275 struct perf_tool *tool, process_auxtrace_t fn, 1276 size_t snapshot_size) 1277 { 1278 return __auxtrace_mmap__read(mm, itr, tool, fn, true, snapshot_size); 1279 } 1280 1281 /** 1282 * struct auxtrace_cache - hash table to implement a cache 1283 * @hashtable: the hashtable 1284 * @sz: hashtable size (number of hlists) 1285 * @entry_size: size of an entry 1286 * @limit: limit the number of entries to this maximum, when reached the cache 1287 * is dropped and caching begins again with an empty cache 1288 * @cnt: current number of entries 1289 * @bits: hashtable size (@sz = 2^@bits) 1290 */ 1291 struct auxtrace_cache { 1292 struct hlist_head *hashtable; 1293 size_t sz; 1294 size_t entry_size; 1295 size_t limit; 1296 size_t cnt; 1297 unsigned int bits; 1298 }; 1299 1300 struct auxtrace_cache *auxtrace_cache__new(unsigned int bits, size_t entry_size, 1301 unsigned int limit_percent) 1302 { 1303 struct auxtrace_cache *c; 1304 struct hlist_head *ht; 1305 size_t sz, i; 1306 1307 c = zalloc(sizeof(struct auxtrace_cache)); 1308 if (!c) 1309 return NULL; 1310 1311 sz = 1UL << bits; 1312 1313 ht = calloc(sz, sizeof(struct hlist_head)); 1314 if (!ht) 1315 goto out_free; 1316 1317 for (i = 0; i < sz; i++) 1318 INIT_HLIST_HEAD(&ht[i]); 1319 1320 c->hashtable = ht; 1321 c->sz = sz; 1322 c->entry_size = entry_size; 1323 c->limit = (c->sz * limit_percent) / 100; 1324 c->bits = bits; 1325 1326 return c; 1327 1328 out_free: 1329 free(c); 1330 return NULL; 1331 } 1332 1333 static void auxtrace_cache__drop(struct auxtrace_cache *c) 1334 { 1335 struct auxtrace_cache_entry *entry; 1336 struct hlist_node *tmp; 1337 size_t i; 1338 1339 if (!c) 1340 return; 1341 1342 for (i = 0; i < c->sz; i++) { 1343 hlist_for_each_entry_safe(entry, tmp, &c->hashtable[i], hash) { 1344 hlist_del(&entry->hash); 1345 auxtrace_cache__free_entry(c, entry); 1346 } 1347 } 1348 1349 c->cnt = 0; 1350 } 1351 1352 void auxtrace_cache__free(struct auxtrace_cache *c) 1353 { 1354 if (!c) 1355 return; 1356 1357 auxtrace_cache__drop(c); 1358 free(c->hashtable); 1359 free(c); 1360 } 1361 1362 void *auxtrace_cache__alloc_entry(struct auxtrace_cache *c) 1363 { 1364 return malloc(c->entry_size); 1365 } 1366 1367 void auxtrace_cache__free_entry(struct auxtrace_cache *c __maybe_unused, 1368 void *entry) 1369 { 1370 free(entry); 1371 } 1372 1373 int auxtrace_cache__add(struct auxtrace_cache *c, u32 key, 1374 struct auxtrace_cache_entry *entry) 1375 { 1376 if (c->limit && ++c->cnt > c->limit) 1377 auxtrace_cache__drop(c); 1378 1379 entry->key = key; 1380 hlist_add_head(&entry->hash, &c->hashtable[hash_32(key, c->bits)]); 1381 1382 return 0; 1383 } 1384 1385 void *auxtrace_cache__lookup(struct auxtrace_cache *c, u32 key) 1386 { 1387 struct auxtrace_cache_entry *entry; 1388 struct hlist_head *hlist; 1389 1390 if (!c) 1391 return NULL; 1392 1393 hlist = &c->hashtable[hash_32(key, c->bits)]; 1394 hlist_for_each_entry(entry, hlist, hash) { 1395 if (entry->key == key) 1396 return entry; 1397 } 1398 1399 return NULL; 1400 } 1401