1 // SPDX-License-Identifier: GPL-2.0 2 #include <dirent.h> 3 #include <errno.h> 4 #include <stdlib.h> 5 #include <stdio.h> 6 #include <string.h> 7 #include <linux/capability.h> 8 #include <linux/kernel.h> 9 #include <linux/mman.h> 10 #include <linux/string.h> 11 #include <linux/time64.h> 12 #include <sys/types.h> 13 #include <sys/stat.h> 14 #include <sys/param.h> 15 #include <fcntl.h> 16 #include <unistd.h> 17 #include <inttypes.h> 18 #include "annotate.h" 19 #include "build-id.h" 20 #include "cap.h" 21 #include "cpumap.h" 22 #include "debug.h" 23 #include "demangle-cxx.h" 24 #include "demangle-java.h" 25 #include "demangle-ocaml.h" 26 #include "demangle-rust-v0.h" 27 #include "dso.h" 28 #include "util.h" // lsdir() 29 #include "event.h" 30 #include "machine.h" 31 #include "map.h" 32 #include "symbol.h" 33 #include "map_symbol.h" 34 #include "mem-events.h" 35 #include "mem-info.h" 36 #include "symsrc.h" 37 #include "strlist.h" 38 #include "intlist.h" 39 #include "namespaces.h" 40 #include "header.h" 41 #include "path.h" 42 #include <linux/ctype.h> 43 #include <linux/log2.h> 44 #include <linux/zalloc.h> 45 46 #include <elf.h> 47 #include <limits.h> 48 #include <symbol/kallsyms.h> 49 #include <sys/utsname.h> 50 51 static int map_fixup_cb(struct map *map, void *data __maybe_unused) 52 { 53 map__fixup_start(map); 54 map__fixup_end(map); 55 return 0; 56 } 57 58 static int dso__load_kernel_sym(struct dso *dso, struct map *map); 59 static int dso__load_guest_kernel_sym(struct dso *dso, struct map *map); 60 61 int vmlinux_path__nr_entries; 62 char **vmlinux_path; 63 64 struct symbol_conf symbol_conf = { 65 .nanosecs = false, 66 .use_modules = true, 67 .try_vmlinux_path = true, 68 .demangle = true, 69 .demangle_kernel = false, 70 .cumulate_callchain = true, 71 .time_quantum = 100 * NSEC_PER_MSEC, /* 100ms */ 72 .show_hist_headers = true, 73 .symfs = "", 74 .symfs_layout_flat = false, 75 .event_group = true, 76 .inline_name = true, 77 .res_sample = 0, 78 .addr2line_timeout_ms = 5 * 1000, 79 }; 80 81 struct map_list_node { 82 struct list_head node; 83 struct map *map; 84 }; 85 86 static struct map_list_node *map_list_node__new(void) 87 { 88 return malloc(sizeof(struct map_list_node)); 89 } 90 91 static enum dso_binary_type binary_type_symtab[] = { 92 DSO_BINARY_TYPE__KALLSYMS, 93 DSO_BINARY_TYPE__GUEST_KALLSYMS, 94 DSO_BINARY_TYPE__JAVA_JIT, 95 DSO_BINARY_TYPE__DEBUGLINK, 96 DSO_BINARY_TYPE__BUILD_ID_CACHE, 97 DSO_BINARY_TYPE__BUILD_ID_CACHE_DEBUGINFO, 98 DSO_BINARY_TYPE__FEDORA_DEBUGINFO, 99 DSO_BINARY_TYPE__UBUNTU_DEBUGINFO, 100 DSO_BINARY_TYPE__BUILDID_DEBUGINFO, 101 DSO_BINARY_TYPE__GNU_DEBUGDATA, 102 DSO_BINARY_TYPE__SYSTEM_PATH_DSO, 103 DSO_BINARY_TYPE__GUEST_KMODULE, 104 DSO_BINARY_TYPE__GUEST_KMODULE_COMP, 105 DSO_BINARY_TYPE__SYSTEM_PATH_KMODULE, 106 DSO_BINARY_TYPE__SYSTEM_PATH_KMODULE_COMP, 107 DSO_BINARY_TYPE__OPENEMBEDDED_DEBUGINFO, 108 DSO_BINARY_TYPE__MIXEDUP_UBUNTU_DEBUGINFO, 109 DSO_BINARY_TYPE__NOT_FOUND, 110 }; 111 112 #define DSO_BINARY_TYPE__SYMTAB_CNT ARRAY_SIZE(binary_type_symtab) 113 114 static bool symbol_type__filter(char symbol_type) 115 { 116 symbol_type = toupper(symbol_type); 117 return symbol_type == 'T' || symbol_type == 'W' || symbol_type == 'D' || symbol_type == 'B'; 118 } 119 120 static int prefix_underscores_count(const char *str) 121 { 122 const char *tail = str; 123 124 while (*tail == '_') 125 tail++; 126 127 return tail - str; 128 } 129 130 const char * __weak arch__normalize_symbol_name(const char *name) 131 { 132 return name; 133 } 134 135 int __weak arch__compare_symbol_names(const char *namea, const char *nameb) 136 { 137 return strcmp(namea, nameb); 138 } 139 140 int __weak arch__compare_symbol_names_n(const char *namea, const char *nameb, 141 unsigned int n) 142 { 143 return strncmp(namea, nameb, n); 144 } 145 146 int __weak arch__choose_best_symbol(struct symbol *syma, 147 struct symbol *symb __maybe_unused) 148 { 149 /* Avoid "SyS" kernel syscall aliases */ 150 if (strlen(syma->name) >= 3 && !strncmp(syma->name, "SyS", 3)) 151 return SYMBOL_B; 152 if (strlen(syma->name) >= 10 && !strncmp(syma->name, "compat_SyS", 10)) 153 return SYMBOL_B; 154 155 return SYMBOL_A; 156 } 157 158 static int choose_best_symbol(struct symbol *syma, struct symbol *symb) 159 { 160 s64 a; 161 s64 b; 162 size_t na, nb; 163 164 /* Prefer a symbol with non zero length */ 165 a = syma->end - syma->start; 166 b = symb->end - symb->start; 167 if ((b == 0) && (a > 0)) 168 return SYMBOL_A; 169 else if ((a == 0) && (b > 0)) 170 return SYMBOL_B; 171 172 if (symbol__type(syma) != symbol__type(symb)) { 173 if (symbol__type(syma) == STT_NOTYPE) 174 return SYMBOL_B; 175 if (symbol__type(symb) == STT_NOTYPE) 176 return SYMBOL_A; 177 } 178 179 /* Prefer a non weak symbol over a weak one */ 180 a = symbol__binding(syma) == STB_WEAK; 181 b = symbol__binding(symb) == STB_WEAK; 182 if (b && !a) 183 return SYMBOL_A; 184 if (a && !b) 185 return SYMBOL_B; 186 187 /* Prefer a global symbol over a non global one */ 188 a = symbol__binding(syma) == STB_GLOBAL; 189 b = symbol__binding(symb) == STB_GLOBAL; 190 if (a && !b) 191 return SYMBOL_A; 192 if (b && !a) 193 return SYMBOL_B; 194 195 /* Prefer a symbol with less underscores */ 196 a = prefix_underscores_count(syma->name); 197 b = prefix_underscores_count(symb->name); 198 if (b > a) 199 return SYMBOL_A; 200 else if (a > b) 201 return SYMBOL_B; 202 203 /* Choose the symbol with the longest name */ 204 na = strlen(syma->name); 205 nb = strlen(symb->name); 206 if (na > nb) 207 return SYMBOL_A; 208 else if (na < nb) 209 return SYMBOL_B; 210 211 return arch__choose_best_symbol(syma, symb); 212 } 213 214 void symbols__fixup_duplicate(struct rb_root_cached *symbols) 215 { 216 struct rb_node *nd; 217 struct symbol *curr, *next; 218 219 if (symbol_conf.allow_aliases) 220 return; 221 222 nd = rb_first_cached(symbols); 223 224 while (nd) { 225 curr = rb_entry(nd, struct symbol, rb_node); 226 again: 227 nd = rb_next(&curr->rb_node); 228 if (!nd) 229 break; 230 231 next = rb_entry(nd, struct symbol, rb_node); 232 if (curr->start != next->start) 233 continue; 234 235 if (choose_best_symbol(curr, next) == SYMBOL_A) { 236 if (symbol__type(next) == STT_GNU_IFUNC) 237 symbol__set_ifunc_alias(curr, true); 238 rb_erase_cached(&next->rb_node, symbols); 239 symbol__delete(next); 240 goto again; 241 } else { 242 if (symbol__type(curr) == STT_GNU_IFUNC) 243 symbol__set_ifunc_alias(next, true); 244 nd = rb_next(&curr->rb_node); 245 rb_erase_cached(&curr->rb_node, symbols); 246 symbol__delete(curr); 247 } 248 } 249 } 250 251 /* Update zero-sized symbols using the address of the next symbol */ 252 void symbols__fixup_end(struct rb_root_cached *symbols, bool is_kallsyms) 253 { 254 struct rb_node *nd, *prevnd = rb_first_cached(symbols); 255 struct symbol *curr, *prev; 256 257 if (prevnd == NULL) 258 return; 259 260 curr = rb_entry(prevnd, struct symbol, rb_node); 261 262 for (nd = rb_next(prevnd); nd; nd = rb_next(nd)) { 263 prev = curr; 264 curr = rb_entry(nd, struct symbol, rb_node); 265 266 /* 267 * On some architecture kernel text segment start is located at 268 * some low memory address, while modules are located at high 269 * memory addresses (or vice versa). The gap between end of 270 * kernel text segment and beginning of first module's text 271 * segment is very big. Therefore do not fill this gap and do 272 * not assign it to the kernel dso map (kallsyms). 273 * 274 * Also BPF code can be allocated separately from text segments 275 * and modules. So the last entry in a module should not fill 276 * the gap too. 277 * 278 * In kallsyms, it determines module symbols using '[' character 279 * like in: 280 * ffffffffc1937000 T hdmi_driver_init [snd_hda_codec_hdmi] 281 */ 282 if (prev->end == prev->start) { 283 const char *prev_mod; 284 const char *curr_mod; 285 286 if (!is_kallsyms) { 287 prev->end = curr->start; 288 continue; 289 } 290 291 prev_mod = strchr(prev->name, '['); 292 curr_mod = strchr(curr->name, '['); 293 294 /* Last kernel/module symbol mapped to end of page */ 295 if (!prev_mod != !curr_mod) 296 prev->end = roundup(prev->end + 4096, 4096); 297 /* Last symbol in the previous module */ 298 else if (prev_mod && strcmp(prev_mod, curr_mod)) 299 prev->end = roundup(prev->end + 4096, 4096); 300 else 301 prev->end = curr->start; 302 303 pr_debug4("%s sym:%s end:%#" PRIx64 "\n", 304 __func__, prev->name, prev->end); 305 } 306 } 307 308 /* Last entry */ 309 if (curr->end == curr->start) 310 curr->end = roundup(curr->start, 4096) + 4096; 311 } 312 313 struct symbol *symbol__new(u64 start, u64 len, u8 binding, u8 type, const char *name) 314 { 315 size_t namelen = strlen(name) + 1; 316 struct symbol *sym = calloc(1, (symbol_conf.priv_size + 317 sizeof(*sym) + namelen)); 318 if (sym == NULL) 319 return NULL; 320 321 if (symbol_conf.priv_size) { 322 if (symbol_conf.init_annotation) { 323 struct annotation *notes = (void *)sym; 324 annotation__init(notes); 325 } 326 sym = ((void *)sym) + symbol_conf.priv_size; 327 } 328 329 sym->start = start; 330 sym->end = len ? start + len : start; 331 atomic_init(&sym->flags, (type << SYMBOL_FLAG_TYPE_SHIFT) | 332 (binding << SYMBOL_FLAG_BINDING_SHIFT)); 333 sym->namelen = namelen - 1; 334 335 pr_debug4("%s: %s %#" PRIx64 "-%#" PRIx64 "\n", 336 __func__, name, start, sym->end); 337 memcpy(sym->name, name, namelen); 338 339 return sym; 340 } 341 342 void symbol__delete(struct symbol *sym) 343 { 344 if (symbol_conf.priv_size) { 345 if (symbol_conf.init_annotation) { 346 struct annotation *notes = symbol__annotation(sym); 347 348 annotation__exit(notes); 349 } 350 } 351 free(((void *)sym) - symbol_conf.priv_size); 352 } 353 354 void symbol__set_ignore(struct symbol *sym, bool ignore) 355 { 356 if (ignore) 357 atomic_fetch_or(&sym->flags, SYMBOL_FLAG_IGNORE); 358 else 359 atomic_fetch_and(&sym->flags, ~SYMBOL_FLAG_IGNORE); 360 } 361 362 void symbol__set_annotate2(struct symbol *sym, bool annotate2) 363 { 364 if (annotate2) 365 atomic_fetch_or(&sym->flags, SYMBOL_FLAG_ANNOTATE2); 366 else 367 atomic_fetch_and(&sym->flags, ~SYMBOL_FLAG_ANNOTATE2); 368 } 369 370 void symbol__set_inlined(struct symbol *sym, bool inlined) 371 { 372 if (inlined) 373 atomic_fetch_or(&sym->flags, SYMBOL_FLAG_INLINED); 374 else 375 atomic_fetch_and(&sym->flags, ~SYMBOL_FLAG_INLINED); 376 } 377 378 void symbol__set_ifunc_alias(struct symbol *sym, bool ifunc_alias) 379 { 380 if (ifunc_alias) 381 atomic_fetch_or(&sym->flags, SYMBOL_FLAG_IFUNC_ALIAS); 382 else 383 atomic_fetch_and(&sym->flags, ~SYMBOL_FLAG_IFUNC_ALIAS); 384 } 385 386 static void symbol__set_idle(struct symbol *sym, bool idle) 387 { 388 uint16_t old_flags = atomic_load_explicit(&sym->flags, memory_order_relaxed); 389 uint16_t new_flags; 390 uint16_t idle_val = idle ? SYMBOL_IDLE__IDLE : SYMBOL_IDLE__NOT_IDLE; 391 392 do { 393 new_flags = old_flags & ~SYMBOL_FLAG_IDLE_MASK; 394 new_flags |= (idle_val << SYMBOL_FLAG_IDLE_SHIFT); 395 } while (!atomic_compare_exchange_weak(&sym->flags, &old_flags, new_flags)); 396 } 397 void symbols__delete(struct rb_root_cached *symbols) 398 { 399 struct symbol *pos; 400 struct rb_node *next = rb_first_cached(symbols); 401 402 while (next) { 403 pos = rb_entry(next, struct symbol, rb_node); 404 next = rb_next(&pos->rb_node); 405 rb_erase_cached(&pos->rb_node, symbols); 406 symbol__delete(pos); 407 } 408 } 409 410 void __symbols__insert(struct rb_root_cached *symbols, struct symbol *sym) 411 { 412 struct rb_node **p = &symbols->rb_root.rb_node; 413 struct rb_node *parent = NULL; 414 const u64 ip = sym->start; 415 struct symbol *s; 416 bool leftmost = true; 417 418 while (*p != NULL) { 419 parent = *p; 420 s = rb_entry(parent, struct symbol, rb_node); 421 if (ip < s->start) 422 p = &(*p)->rb_left; 423 else { 424 p = &(*p)->rb_right; 425 leftmost = false; 426 } 427 } 428 rb_link_node(&sym->rb_node, parent, p); 429 rb_insert_color_cached(&sym->rb_node, symbols, leftmost); 430 } 431 432 void symbols__insert(struct rb_root_cached *symbols, struct symbol *sym) 433 { 434 __symbols__insert(symbols, sym); 435 } 436 437 static struct symbol *symbols__find(struct rb_root_cached *symbols, u64 ip) 438 { 439 struct rb_node *n; 440 441 if (symbols == NULL) 442 return NULL; 443 444 n = symbols->rb_root.rb_node; 445 446 while (n) { 447 struct symbol *s = rb_entry(n, struct symbol, rb_node); 448 449 if (ip < s->start) 450 n = n->rb_left; 451 else if (ip > s->end || (ip == s->end && ip != s->start)) 452 n = n->rb_right; 453 else 454 return s; 455 } 456 457 return NULL; 458 } 459 460 static struct symbol *symbols__first(struct rb_root_cached *symbols) 461 { 462 struct rb_node *n = rb_first_cached(symbols); 463 464 if (n) 465 return rb_entry(n, struct symbol, rb_node); 466 467 return NULL; 468 } 469 470 static struct symbol *symbols__last(struct rb_root_cached *symbols) 471 { 472 struct rb_node *n = rb_last(&symbols->rb_root); 473 474 if (n) 475 return rb_entry(n, struct symbol, rb_node); 476 477 return NULL; 478 } 479 480 static struct symbol *symbols__next(struct symbol *sym) 481 { 482 struct rb_node *n = rb_next(&sym->rb_node); 483 484 if (n) 485 return rb_entry(n, struct symbol, rb_node); 486 487 return NULL; 488 } 489 490 static int symbols__sort_name_cmp(const void *vlhs, const void *vrhs) 491 { 492 const struct symbol *lhs = *((const struct symbol **)vlhs); 493 const struct symbol *rhs = *((const struct symbol **)vrhs); 494 495 return strcmp(lhs->name, rhs->name); 496 } 497 498 static struct symbol **symbols__sort_by_name(struct rb_root_cached *source, size_t *len) 499 { 500 struct rb_node *nd; 501 struct symbol **result; 502 size_t i = 0, size = 0; 503 504 for (nd = rb_first_cached(source); nd; nd = rb_next(nd)) 505 size++; 506 507 result = malloc(sizeof(*result) * size); 508 if (!result) 509 return NULL; 510 511 for (nd = rb_first_cached(source); nd; nd = rb_next(nd)) { 512 struct symbol *pos = rb_entry(nd, struct symbol, rb_node); 513 514 result[i++] = pos; 515 } 516 qsort(result, size, sizeof(*result), symbols__sort_name_cmp); 517 *len = size; 518 return result; 519 } 520 521 int symbol__match_symbol_name(const char *name, const char *str, 522 enum symbol_tag_include includes) 523 { 524 const char *versioning; 525 526 if (includes == SYMBOL_TAG_INCLUDE__DEFAULT_ONLY && 527 (versioning = strstr(name, "@@"))) { 528 int len = strlen(str); 529 530 if (len < versioning - name) 531 len = versioning - name; 532 533 return arch__compare_symbol_names_n(name, str, len); 534 } else 535 return arch__compare_symbol_names(name, str); 536 } 537 538 static struct symbol *symbols__find_by_name(struct symbol *symbols[], 539 size_t symbols_len, 540 const char *name, 541 enum symbol_tag_include includes, 542 size_t *found_idx) 543 { 544 size_t i, lower = 0, upper = symbols_len; 545 struct symbol *s = NULL; 546 547 if (found_idx) 548 *found_idx = SIZE_MAX; 549 550 if (!symbols_len) 551 return NULL; 552 553 while (lower < upper) { 554 int cmp; 555 556 i = (lower + upper) / 2; 557 cmp = symbol__match_symbol_name(symbols[i]->name, name, includes); 558 559 if (cmp > 0) 560 upper = i; 561 else if (cmp < 0) 562 lower = i + 1; 563 else { 564 if (found_idx) 565 *found_idx = i; 566 s = symbols[i]; 567 break; 568 } 569 } 570 if (s && includes != SYMBOL_TAG_INCLUDE__DEFAULT_ONLY) { 571 /* return first symbol that has same name (if any) */ 572 for (; i > 0; i--) { 573 struct symbol *tmp = symbols[i - 1]; 574 575 if (!arch__compare_symbol_names(tmp->name, s->name)) { 576 if (found_idx) 577 *found_idx = i - 1; 578 s = tmp; 579 } else 580 break; 581 } 582 } 583 assert(!found_idx || !s || s == symbols[*found_idx]); 584 return s; 585 } 586 587 void dso__reset_find_symbol_cache(struct dso *dso) 588 { 589 dso__set_last_find_result_addr(dso, 0); 590 dso__set_last_find_result_symbol(dso, NULL); 591 } 592 593 void dso__insert_symbol(struct dso *dso, struct symbol *sym) 594 { 595 __symbols__insert(dso__symbols(dso), sym); 596 597 /* update the symbol cache if necessary */ 598 if (dso__last_find_result_addr(dso) >= sym->start && 599 (dso__last_find_result_addr(dso) < sym->end || 600 sym->start == sym->end)) { 601 dso__set_last_find_result_symbol(dso, sym); 602 } 603 } 604 605 void dso__delete_symbol(struct dso *dso, struct symbol *sym) 606 { 607 rb_erase_cached(&sym->rb_node, dso__symbols(dso)); 608 symbol__delete(sym); 609 dso__reset_find_symbol_cache(dso); 610 } 611 612 struct symbol *dso__find_symbol(struct dso *dso, u64 addr) 613 { 614 if (dso__last_find_result_addr(dso) != addr || dso__last_find_result_symbol(dso) == NULL) { 615 dso__set_last_find_result_addr(dso, addr); 616 dso__set_last_find_result_symbol(dso, symbols__find(dso__symbols(dso), addr)); 617 } 618 619 return dso__last_find_result_symbol(dso); 620 } 621 622 struct symbol *dso__find_symbol_nocache(struct dso *dso, u64 addr) 623 { 624 return symbols__find(dso__symbols(dso), addr); 625 } 626 627 struct symbol *dso__first_symbol(struct dso *dso) 628 { 629 return symbols__first(dso__symbols(dso)); 630 } 631 632 struct symbol *dso__last_symbol(struct dso *dso) 633 { 634 return symbols__last(dso__symbols(dso)); 635 } 636 637 struct symbol *dso__next_symbol(struct symbol *sym) 638 { 639 return symbols__next(sym); 640 } 641 642 struct symbol *dso__next_symbol_by_name(struct dso *dso, size_t *idx) 643 { 644 if (*idx + 1 >= dso__symbol_names_len(dso)) 645 return NULL; 646 647 ++*idx; 648 return dso__symbol_names(dso)[*idx]; 649 } 650 651 /* 652 * Returns first symbol that matched with @name. 653 */ 654 struct symbol *dso__find_symbol_by_name(struct dso *dso, const char *name, size_t *idx) 655 { 656 struct symbol *s = symbols__find_by_name(dso__symbol_names(dso), 657 dso__symbol_names_len(dso), 658 name, SYMBOL_TAG_INCLUDE__NONE, idx); 659 if (!s) { 660 s = symbols__find_by_name(dso__symbol_names(dso), dso__symbol_names_len(dso), 661 name, SYMBOL_TAG_INCLUDE__DEFAULT_ONLY, idx); 662 } 663 return s; 664 } 665 666 void dso__sort_by_name(struct dso *dso) 667 { 668 mutex_lock(dso__lock(dso)); 669 if (!dso__sorted_by_name(dso)) { 670 size_t len = 0; 671 672 dso__set_symbol_names(dso, symbols__sort_by_name(dso__symbols(dso), &len)); 673 if (dso__symbol_names(dso)) { 674 dso__set_symbol_names_len(dso, len); 675 dso__set_sorted_by_name(dso); 676 } 677 } 678 mutex_unlock(dso__lock(dso)); 679 } 680 681 /* 682 * While we find nice hex chars, build a long_val. 683 * Return number of chars processed. 684 */ 685 static int hex2u64(const char *ptr, u64 *long_val) 686 { 687 char *p; 688 689 *long_val = strtoull(ptr, &p, 16); 690 691 return p - ptr; 692 } 693 694 695 int modules__parse(const char *filename, void *arg, 696 int (*process_module)(void *arg, const char *name, 697 u64 start, u64 size)) 698 { 699 char *line = NULL; 700 size_t n; 701 FILE *file; 702 int err = 0; 703 704 file = fopen(filename, "r"); 705 if (file == NULL) 706 return -1; 707 708 while (1) { 709 char name[PATH_MAX]; 710 u64 start, size; 711 char *sep, *endptr; 712 ssize_t line_len; 713 714 line_len = getline(&line, &n, file); 715 if (line_len < 0) { 716 if (feof(file)) 717 break; 718 err = -1; 719 goto out; 720 } 721 722 if (!line) { 723 err = -1; 724 goto out; 725 } 726 727 line[--line_len] = '\0'; /* \n */ 728 729 sep = strrchr(line, 'x'); 730 if (sep == NULL) 731 continue; 732 733 hex2u64(sep + 1, &start); 734 735 sep = strchr(line, ' '); 736 if (sep == NULL) 737 continue; 738 739 *sep = '\0'; 740 741 scnprintf(name, sizeof(name), "[%s]", line); 742 743 size = strtoul(sep + 1, &endptr, 0); 744 if (*endptr != ' ' && *endptr != '\t') 745 continue; 746 747 err = process_module(arg, name, start, size); 748 if (err) 749 break; 750 } 751 out: 752 free(line); 753 fclose(file); 754 return err; 755 } 756 757 758 /* 759 * These are symbols in the kernel image, so make sure that 760 * sym is from a kernel DSO. 761 */ 762 static int sym_name_cmp(const void *a, const void *b) 763 { 764 const char *name = a; 765 const char *const *sym = b; 766 767 return strcmp(name, *sym); 768 } 769 770 static bool match_x86_idle_routine(const char *name, const char *base) 771 { 772 if (strstarts(name, base)) { 773 size_t len = strlen(base); 774 775 if (name[len] == '\0' || name[len] == '.') 776 return true; 777 } 778 return false; 779 } 780 781 bool symbol__is_idle(struct symbol *sym, const struct dso *dso, struct perf_env *env) 782 { 783 static const char * const idle_symbols[] = { 784 "acpi_idle_do_entry", 785 "acpi_processor_ffh_cstate_enter", 786 "arch_cpu_idle", 787 "cpu_idle", 788 "cpu_startup_entry", 789 "default_idle", 790 "enter_idle", 791 "exit_idle", 792 "idle_cpu", 793 "native_safe_halt", 794 "poll_idle", 795 "pseries_dedicated_idle_sleep", 796 }; 797 const char *name = sym->name; 798 uint16_t e_machine; 799 800 { 801 uint16_t flags = atomic_load_explicit(&sym->flags, memory_order_relaxed); 802 uint16_t idle_val = (flags & SYMBOL_FLAG_IDLE_MASK) >> SYMBOL_FLAG_IDLE_SHIFT; 803 804 if (idle_val != SYMBOL_IDLE__UNKNOWN) 805 return idle_val == SYMBOL_IDLE__IDLE; 806 } 807 808 if (!dso || dso__kernel(dso) == DSO_SPACE__USER) { 809 symbol__set_idle(sym, /*idle=*/false); 810 return false; 811 } 812 813 /* 814 * ppc64 uses function descriptors and appends a '.' to the 815 * start of every instruction address. Remove it. 816 */ 817 if (name[0] == '.') 818 name++; 819 820 if (bsearch(name, idle_symbols, ARRAY_SIZE(idle_symbols), 821 sizeof(idle_symbols[0]), sym_name_cmp)) { 822 symbol__set_idle(sym, /*idle=*/true); 823 return true; 824 } 825 826 e_machine = (env && env->arch) ? perf_env__e_machine(env, NULL) : EM_NONE; 827 if (e_machine == EM_NONE && dso) 828 e_machine = dso__e_machine((struct dso *)dso, NULL, NULL); 829 if (e_machine == EM_NONE && env) 830 e_machine = perf_env__e_machine(env, NULL); 831 832 if (e_machine == EM_386 || e_machine == EM_X86_64) { 833 if (match_x86_idle_routine(name, "intel_idle") || 834 match_x86_idle_routine(name, "intel_idle_irq") || 835 match_x86_idle_routine(name, "intel_idle_ibrs") || 836 match_x86_idle_routine(name, "mwait_idle") || 837 match_x86_idle_routine(name, "mwait_idle_with_hints")) { 838 symbol__set_idle(sym, /*idle=*/true); 839 return true; 840 } 841 } 842 843 if (e_machine == EM_PPC64 && !strcmp(name, "ppc64_runlatch_off")) { 844 symbol__set_idle(sym, /*idle=*/true); 845 return true; 846 } 847 848 if (e_machine == EM_S390 && strstarts(name, "psw_idle")) { 849 int major = 0, minor = 0; 850 const char *release = env ? perf_env__os_release(env) : NULL; 851 852 /* 853 * If we can't determine the release (e.g. unpopulated guest traces), 854 * default to idle. 855 */ 856 if (!release) { 857 symbol__set_idle(sym, /*idle=*/true); 858 return true; 859 } 860 861 /* Before v6.10, s390 used psw_idle. */ 862 if (sscanf(release, "%d.%d", &major, &minor) == 2 && 863 (major < 6 || (major == 6 && minor < 10))) { 864 symbol__set_idle(sym, /*idle=*/true); 865 return true; 866 } 867 } 868 869 symbol__set_idle(sym, /*idle=*/false); 870 return false; 871 } 872 873 static int map__process_kallsym_symbol(void *arg, const char *name, 874 char type, u64 start) 875 { 876 struct symbol *sym; 877 struct dso *dso = arg; 878 struct rb_root_cached *root = dso__symbols(dso); 879 880 if (!symbol_type__filter(type)) 881 return 0; 882 883 /* Ignore mapping symbols in kallsyms */ 884 if (is_ignored_kernel_symbol(name)) 885 return 0; 886 887 /* 888 * module symbols are not sorted so we add all 889 * symbols, setting length to 0, and rely on 890 * symbols__fixup_end() to fix it up. 891 */ 892 sym = symbol__new(start, 0, kallsyms2elf_binding(type), kallsyms2elf_type(type), name); 893 if (sym == NULL) 894 return -ENOMEM; 895 /* 896 * We will pass the symbols to the filter later, in 897 * map__split_kallsyms, when we have split the maps per module 898 */ 899 __symbols__insert(root, sym); 900 901 return 0; 902 } 903 904 /* 905 * Loads the function entries in /proc/kallsyms into kernel_map->dso, 906 * so that we can in the next step set the symbol ->end address and then 907 * call kernel_maps__split_kallsyms. 908 */ 909 static int dso__load_all_kallsyms(struct dso *dso, const char *filename) 910 { 911 return kallsyms__parse(filename, dso, map__process_kallsym_symbol); 912 } 913 914 static int maps__split_kallsyms_for_kcore(struct maps *kmaps, struct dso *dso) 915 { 916 struct symbol *pos; 917 int count = 0; 918 struct rb_root_cached *root = dso__symbols(dso); 919 struct rb_root_cached old_root = *root; 920 struct rb_node *next = rb_first_cached(root); 921 922 if (!kmaps) 923 return -1; 924 925 *root = RB_ROOT_CACHED; 926 927 while (next) { 928 struct map *curr_map; 929 struct dso *curr_map_dso; 930 char *module; 931 932 pos = rb_entry(next, struct symbol, rb_node); 933 next = rb_next(&pos->rb_node); 934 935 rb_erase_cached(&pos->rb_node, &old_root); 936 RB_CLEAR_NODE(&pos->rb_node); 937 module = strchr(pos->name, '\t'); 938 if (module) 939 *module = '\0'; 940 941 curr_map = maps__find(kmaps, pos->start); 942 943 if (!curr_map) { 944 symbol__delete(pos); 945 continue; 946 } 947 curr_map_dso = map__dso(curr_map); 948 pos->start -= map__start(curr_map) - map__pgoff(curr_map); 949 if (pos->end > map__end(curr_map)) 950 pos->end = map__end(curr_map); 951 if (pos->end) 952 pos->end -= map__start(curr_map) - map__pgoff(curr_map); 953 symbols__insert(dso__symbols(curr_map_dso), pos); 954 ++count; 955 map__put(curr_map); 956 } 957 958 /* Symbols have been adjusted */ 959 dso__set_adjust_symbols(dso, true); 960 961 return count; 962 } 963 964 static uint16_t machine_or_dso_e_machine(struct machine *machine, struct dso *dso) 965 { 966 uint16_t e_machine = EM_NONE; 967 /* DSO should be most accurate */ 968 if (dso) 969 e_machine = dso__e_machine(dso, machine, /*e_flags=*/NULL); 970 971 if (e_machine != EM_NONE) 972 return e_machine; 973 974 /* Check the global environment next. */ 975 if (machine && machine->env && machine->env->e_machine != EM_NONE) 976 return machine->env->e_machine; 977 978 return perf_env__e_machine(machine ? machine->env : NULL, /*e_flags=*/NULL); 979 } 980 981 /* 982 * Split the symbols into maps, making sure there are no overlaps, i.e. the 983 * kernel range is broken in several maps, named [kernel].N, as we don't have 984 * the original ELF section names vmlinux have. 985 */ 986 static int maps__split_kallsyms(struct maps *kmaps, struct dso *dso, u64 delta, 987 struct map *initial_map) 988 { 989 struct machine *machine; 990 struct map *curr_map = map__get(initial_map); 991 struct symbol *pos; 992 int count = 0, moved = 0; 993 struct rb_root_cached *root = dso__symbols(dso); 994 struct rb_node *next = rb_first_cached(root); 995 int kernel_range = 0; 996 uint16_t e_machine = EM_NONE; 997 998 if (!kmaps) 999 return -1; 1000 1001 machine = maps__machine(kmaps); 1002 e_machine = machine_or_dso_e_machine(machine, dso); 1003 1004 while (next) { 1005 char *module; 1006 1007 pos = rb_entry(next, struct symbol, rb_node); 1008 next = rb_next(&pos->rb_node); 1009 1010 module = strchr(pos->name, '\t'); 1011 if (module) { 1012 struct dso *curr_map_dso; 1013 1014 if (!symbol_conf.use_modules) 1015 goto discard_symbol; 1016 1017 *module++ = '\0'; 1018 curr_map_dso = map__dso(curr_map); 1019 if (strcmp(dso__short_name(curr_map_dso), module)) { 1020 if (!RC_CHK_EQUAL(curr_map, initial_map) && 1021 dso__kernel(dso) == DSO_SPACE__KERNEL_GUEST && 1022 machine__is_default_guest(machine)) { 1023 /* 1024 * We assume all symbols of a module are 1025 * continuous in * kallsyms, so curr_map 1026 * points to a module and all its 1027 * symbols are in its kmap. Mark it as 1028 * loaded. 1029 */ 1030 dso__set_loaded(curr_map_dso); 1031 } 1032 1033 map__zput(curr_map); 1034 curr_map = maps__find_by_name(kmaps, module); 1035 if (curr_map == NULL) { 1036 pr_debug("%s/proc/{kallsyms,modules} " 1037 "inconsistency while looking " 1038 "for \"%s\" module!\n", 1039 machine->root_dir, module); 1040 curr_map = map__get(initial_map); 1041 goto discard_symbol; 1042 } 1043 curr_map_dso = map__dso(curr_map); 1044 if (dso__loaded(curr_map_dso) && 1045 !machine__is_default_guest(machine)) 1046 goto discard_symbol; 1047 } 1048 /* 1049 * So that we look just like we get from .ko files, 1050 * i.e. not prelinked, relative to initial_map->start. 1051 */ 1052 pos->start = map__map_ip(curr_map, pos->start); 1053 pos->end = map__map_ip(curr_map, pos->end); 1054 } else if (e_machine == EM_X86_64 && is_entry_trampoline(pos->name)) { 1055 /* 1056 * These symbols are not needed anymore since the 1057 * trampoline maps refer to the text section and it's 1058 * symbols instead. Avoid having to deal with 1059 * relocations, and the assumption that the first symbol 1060 * is the start of kernel text, by simply removing the 1061 * symbols at this point. 1062 */ 1063 goto discard_symbol; 1064 } else if (!RC_CHK_EQUAL(curr_map, initial_map)) { 1065 char dso_name[PATH_MAX]; 1066 struct dso *ndso; 1067 1068 if (delta) { 1069 /* Kernel was relocated at boot time */ 1070 pos->start -= delta; 1071 pos->end -= delta; 1072 } 1073 1074 if (map__start(initial_map) <= (pos->start + delta) && 1075 (pos->start + delta) < map__end(initial_map)) { 1076 map__zput(curr_map); 1077 curr_map = map__get(initial_map); 1078 goto add_symbol; 1079 } 1080 1081 if (dso__kernel(dso) == DSO_SPACE__KERNEL_GUEST) 1082 snprintf(dso_name, sizeof(dso_name), 1083 "[guest.kernel].%d", 1084 kernel_range); 1085 else 1086 snprintf(dso_name, sizeof(dso_name), 1087 "[kernel].%d", 1088 kernel_range); 1089 1090 ndso = dso__new(dso_name); 1091 map__zput(curr_map); 1092 if (ndso == NULL) 1093 return -1; 1094 1095 dso__set_kernel(ndso, dso__kernel(dso)); 1096 dso__set_loaded(ndso); 1097 1098 curr_map = map__new2(pos->start, ndso); 1099 if (curr_map == NULL) { 1100 dso__put(ndso); 1101 return -1; 1102 } 1103 1104 map__set_mapping_type(curr_map, MAPPING_TYPE__IDENTITY); 1105 if (maps__insert(kmaps, curr_map)) { 1106 map__zput(curr_map); 1107 dso__put(ndso); 1108 return -1; 1109 } 1110 dso__put(ndso); 1111 ++kernel_range; 1112 } else if (delta) { 1113 /* Kernel was relocated at boot time */ 1114 pos->start -= delta; 1115 pos->end -= delta; 1116 } 1117 add_symbol: 1118 if (!RC_CHK_EQUAL(curr_map, initial_map)) { 1119 struct dso *curr_map_dso = map__dso(curr_map); 1120 1121 rb_erase_cached(&pos->rb_node, root); 1122 symbols__insert(dso__symbols(curr_map_dso), pos); 1123 ++moved; 1124 } else 1125 ++count; 1126 1127 continue; 1128 discard_symbol: 1129 rb_erase_cached(&pos->rb_node, root); 1130 symbol__delete(pos); 1131 } 1132 1133 if (!RC_CHK_EQUAL(curr_map, initial_map) && 1134 dso__kernel(dso) == DSO_SPACE__KERNEL_GUEST && 1135 machine__is_default_guest(maps__machine(kmaps))) { 1136 dso__set_loaded(map__dso(curr_map)); 1137 } 1138 map__put(curr_map); 1139 return count + moved; 1140 } 1141 1142 bool symbol__restricted_filename(const char *filename, 1143 const char *restricted_filename) 1144 { 1145 bool restricted = false; 1146 1147 if (symbol_conf.kptr_restrict) { 1148 char *r = realpath(filename, NULL); 1149 1150 if (r != NULL) { 1151 restricted = strcmp(r, restricted_filename) == 0; 1152 free(r); 1153 return restricted; 1154 } 1155 } 1156 1157 return restricted; 1158 } 1159 1160 struct module_info { 1161 struct rb_node rb_node; 1162 char *name; 1163 u64 start; 1164 }; 1165 1166 static void add_module(struct module_info *mi, struct rb_root *modules) 1167 { 1168 struct rb_node **p = &modules->rb_node; 1169 struct rb_node *parent = NULL; 1170 struct module_info *m; 1171 1172 while (*p != NULL) { 1173 parent = *p; 1174 m = rb_entry(parent, struct module_info, rb_node); 1175 if (strcmp(mi->name, m->name) < 0) 1176 p = &(*p)->rb_left; 1177 else 1178 p = &(*p)->rb_right; 1179 } 1180 rb_link_node(&mi->rb_node, parent, p); 1181 rb_insert_color(&mi->rb_node, modules); 1182 } 1183 1184 static void delete_modules(struct rb_root *modules) 1185 { 1186 struct module_info *mi; 1187 struct rb_node *next = rb_first(modules); 1188 1189 while (next) { 1190 mi = rb_entry(next, struct module_info, rb_node); 1191 next = rb_next(&mi->rb_node); 1192 rb_erase(&mi->rb_node, modules); 1193 zfree(&mi->name); 1194 free(mi); 1195 } 1196 } 1197 1198 static struct module_info *find_module(const char *name, 1199 struct rb_root *modules) 1200 { 1201 struct rb_node *n = modules->rb_node; 1202 1203 while (n) { 1204 struct module_info *m; 1205 int cmp; 1206 1207 m = rb_entry(n, struct module_info, rb_node); 1208 cmp = strcmp(name, m->name); 1209 if (cmp < 0) 1210 n = n->rb_left; 1211 else if (cmp > 0) 1212 n = n->rb_right; 1213 else 1214 return m; 1215 } 1216 1217 return NULL; 1218 } 1219 1220 static int __read_proc_modules(void *arg, const char *name, u64 start, 1221 u64 size __maybe_unused) 1222 { 1223 struct rb_root *modules = arg; 1224 struct module_info *mi; 1225 1226 mi = zalloc(sizeof(struct module_info)); 1227 if (!mi) 1228 return -ENOMEM; 1229 1230 mi->name = strdup(name); 1231 mi->start = start; 1232 1233 if (!mi->name) { 1234 free(mi); 1235 return -ENOMEM; 1236 } 1237 1238 add_module(mi, modules); 1239 1240 return 0; 1241 } 1242 1243 static int read_proc_modules(const char *filename, struct rb_root *modules) 1244 { 1245 if (symbol__restricted_filename(filename, "/proc/modules")) 1246 return -1; 1247 1248 if (modules__parse(filename, modules, __read_proc_modules)) { 1249 delete_modules(modules); 1250 return -1; 1251 } 1252 1253 return 0; 1254 } 1255 1256 int compare_proc_modules(const char *from, const char *to) 1257 { 1258 struct rb_root from_modules = RB_ROOT; 1259 struct rb_root to_modules = RB_ROOT; 1260 struct rb_node *from_node, *to_node; 1261 struct module_info *from_m, *to_m; 1262 int ret = -1; 1263 1264 if (read_proc_modules(from, &from_modules)) 1265 return -1; 1266 1267 if (read_proc_modules(to, &to_modules)) 1268 goto out_delete_from; 1269 1270 from_node = rb_first(&from_modules); 1271 to_node = rb_first(&to_modules); 1272 while (from_node) { 1273 if (!to_node) 1274 break; 1275 1276 from_m = rb_entry(from_node, struct module_info, rb_node); 1277 to_m = rb_entry(to_node, struct module_info, rb_node); 1278 1279 if (from_m->start != to_m->start || 1280 strcmp(from_m->name, to_m->name)) 1281 break; 1282 1283 from_node = rb_next(from_node); 1284 to_node = rb_next(to_node); 1285 } 1286 1287 if (!from_node && !to_node) 1288 ret = 0; 1289 1290 delete_modules(&to_modules); 1291 out_delete_from: 1292 delete_modules(&from_modules); 1293 1294 return ret; 1295 } 1296 1297 static int do_validate_kcore_modules_cb(struct map *old_map, void *data) 1298 { 1299 struct rb_root *modules = data; 1300 struct module_info *mi; 1301 struct dso *dso; 1302 1303 if (!__map__is_kmodule(old_map)) 1304 return 0; 1305 1306 dso = map__dso(old_map); 1307 /* Module must be in memory at the same address */ 1308 mi = find_module(dso__short_name(dso), modules); 1309 if (!mi || mi->start != map__start(old_map)) 1310 return -EINVAL; 1311 1312 return 0; 1313 } 1314 1315 static int do_validate_kcore_modules(const char *filename, struct maps *kmaps) 1316 { 1317 struct rb_root modules = RB_ROOT; 1318 int err; 1319 1320 err = read_proc_modules(filename, &modules); 1321 if (err) 1322 return err; 1323 1324 err = maps__for_each_map(kmaps, do_validate_kcore_modules_cb, &modules); 1325 1326 delete_modules(&modules); 1327 return err; 1328 } 1329 1330 /* 1331 * If kallsyms is referenced by name then we look for filename in the same 1332 * directory. 1333 */ 1334 static bool filename_from_kallsyms_filename(char *filename, 1335 const char *base_name, 1336 const char *kallsyms_filename) 1337 { 1338 char *name; 1339 1340 strcpy(filename, kallsyms_filename); 1341 name = strrchr(filename, '/'); 1342 if (!name) 1343 return false; 1344 1345 name += 1; 1346 1347 if (!strcmp(name, "kallsyms")) { 1348 strcpy(name, base_name); 1349 return true; 1350 } 1351 1352 return false; 1353 } 1354 1355 static int validate_kcore_modules(const char *kallsyms_filename, 1356 struct map *map) 1357 { 1358 struct maps *kmaps = map__kmaps(map); 1359 char modules_filename[PATH_MAX]; 1360 1361 if (!kmaps) 1362 return -EINVAL; 1363 1364 if (!filename_from_kallsyms_filename(modules_filename, "modules", 1365 kallsyms_filename)) 1366 return -EINVAL; 1367 1368 if (do_validate_kcore_modules(modules_filename, kmaps)) 1369 return -EINVAL; 1370 1371 return 0; 1372 } 1373 1374 static int validate_kcore_addresses(const char *kallsyms_filename, 1375 struct map *map) 1376 { 1377 struct kmap *kmap = map__kmap(map); 1378 1379 if (!kmap) 1380 return -EINVAL; 1381 1382 if (kmap->ref_reloc_sym && kmap->ref_reloc_sym->name) { 1383 u64 start; 1384 1385 if (kallsyms__get_function_start(kallsyms_filename, 1386 kmap->ref_reloc_sym->name, &start)) 1387 return -ENOENT; 1388 if (start != kmap->ref_reloc_sym->addr) 1389 return -EINVAL; 1390 } 1391 1392 return validate_kcore_modules(kallsyms_filename, map); 1393 } 1394 1395 struct kcore_mapfn_data { 1396 struct dso *dso; 1397 struct list_head maps; 1398 }; 1399 1400 static int kcore_mapfn(u64 start, u64 len, u64 pgoff, void *data) 1401 { 1402 struct kcore_mapfn_data *md = data; 1403 struct map_list_node *list_node = map_list_node__new(); 1404 1405 if (!list_node) 1406 return -ENOMEM; 1407 1408 list_node->map = map__new2(start, md->dso); 1409 if (!list_node->map) { 1410 free(list_node); 1411 return -ENOMEM; 1412 } 1413 1414 map__set_end(list_node->map, map__start(list_node->map) + len); 1415 map__set_pgoff(list_node->map, pgoff); 1416 1417 list_add(&list_node->node, &md->maps); 1418 1419 return 0; 1420 } 1421 1422 static bool remove_old_maps(struct map *map, void *data) 1423 { 1424 const struct map *map_to_save = data; 1425 1426 /* 1427 * We need to preserve eBPF maps even if they are covered by kcore, 1428 * because we need to access eBPF dso for source data. 1429 */ 1430 return !RC_CHK_EQUAL(map, map_to_save) && !__map__is_bpf_prog(map); 1431 } 1432 1433 static int dso__load_kcore(struct dso *dso, struct map *map, 1434 const char *kallsyms_filename) 1435 { 1436 struct maps *kmaps = map__kmaps(map); 1437 struct kcore_mapfn_data md; 1438 struct map *map_ref, *replacement_map = NULL; 1439 struct machine *machine; 1440 bool is_64_bit; 1441 int err, fd; 1442 char kcore_filename[PATH_MAX]; 1443 u64 stext; 1444 1445 if (!kmaps) 1446 return -EINVAL; 1447 1448 machine = maps__machine(kmaps); 1449 1450 /* This function requires that the map is the kernel map */ 1451 if (!__map__is_kernel(map)) 1452 return -EINVAL; 1453 1454 if (!filename_from_kallsyms_filename(kcore_filename, "kcore", 1455 kallsyms_filename)) 1456 return -EINVAL; 1457 1458 /* Modules and kernel must be present at their original addresses */ 1459 if (validate_kcore_addresses(kallsyms_filename, map)) 1460 return -EINVAL; 1461 1462 md.dso = dso; 1463 INIT_LIST_HEAD(&md.maps); 1464 1465 fd = open(kcore_filename, O_RDONLY); 1466 if (fd < 0) { 1467 pr_debug("Failed to open %s. Note /proc/kcore requires CAP_SYS_RAWIO capability to access.\n", 1468 kcore_filename); 1469 return -EINVAL; 1470 } 1471 1472 /* Read new maps into temporary lists */ 1473 err = file__read_maps(fd, map__prot(map) & PROT_EXEC, kcore_mapfn, &md, 1474 &is_64_bit); 1475 if (err) 1476 goto out_err; 1477 dso__set_is_64_bit(dso, is_64_bit); 1478 1479 if (list_empty(&md.maps)) { 1480 err = -EINVAL; 1481 goto out_err; 1482 } 1483 1484 /* Remove old maps */ 1485 maps__remove_maps(kmaps, remove_old_maps, map); 1486 machine->trampolines_mapped = false; 1487 1488 /* Find the kernel map using the '_stext' symbol */ 1489 if (!kallsyms__get_function_start(kallsyms_filename, "_stext", &stext)) { 1490 u64 replacement_size = 0; 1491 struct map_list_node *new_node; 1492 1493 list_for_each_entry(new_node, &md.maps, node) { 1494 struct map *new_map = new_node->map; 1495 u64 new_size = map__size(new_map); 1496 1497 if (!(stext >= map__start(new_map) && stext < map__end(new_map))) 1498 continue; 1499 1500 /* 1501 * On some architectures, ARM64 for example, the kernel 1502 * text can get allocated inside of the vmalloc segment. 1503 * Select the smallest matching segment, in case stext 1504 * falls within more than one in the list. 1505 */ 1506 if (!replacement_map || new_size < replacement_size) { 1507 replacement_map = new_map; 1508 replacement_size = new_size; 1509 } 1510 } 1511 } 1512 1513 if (!replacement_map) 1514 replacement_map = list_entry(md.maps.next, struct map_list_node, node)->map; 1515 1516 /* 1517 * Update addresses of vmlinux map. Re-insert it to ensure maps are 1518 * correctly ordered. Do this before using maps__merge_in() for the 1519 * remaining maps so vmlinux gets split if necessary. 1520 */ 1521 map_ref = map__get(map); 1522 maps__remove(kmaps, map_ref); 1523 1524 map__set_start(map_ref, map__start(replacement_map)); 1525 map__set_end(map_ref, map__end(replacement_map)); 1526 map__set_pgoff(map_ref, map__pgoff(replacement_map)); 1527 map__set_mapping_type(map_ref, map__mapping_type(replacement_map)); 1528 1529 err = maps__insert(kmaps, map_ref); 1530 map__put(map_ref); 1531 if (err) 1532 goto out_err; 1533 1534 /* Add new maps */ 1535 while (!list_empty(&md.maps)) { 1536 struct map_list_node *new_node = list_entry(md.maps.next, struct map_list_node, node); 1537 struct map *new_map = new_node->map; 1538 1539 list_del_init(&new_node->node); 1540 1541 /* skip if replacement_map, already inserted above */ 1542 if (!RC_CHK_EQUAL(new_map, replacement_map)) { 1543 /* 1544 * Merge kcore map into existing maps, 1545 * and ensure that current maps (eBPF) 1546 * stay intact. 1547 */ 1548 if (maps__merge_in(kmaps, new_map)) { 1549 err = -EINVAL; 1550 goto out_err; 1551 } 1552 } 1553 map__zput(new_node->map); 1554 free(new_node); 1555 } 1556 1557 if (machine_or_dso_e_machine(machine, dso) == EM_X86_64) { 1558 u64 addr; 1559 1560 /* 1561 * If one of the corresponding symbols is there, assume the 1562 * entry trampoline maps are too. 1563 */ 1564 if (!kallsyms__get_function_start(kallsyms_filename, 1565 ENTRY_TRAMPOLINE_NAME, 1566 &addr)) 1567 machine->trampolines_mapped = true; 1568 } 1569 1570 /* 1571 * Set the data type and long name so that kcore can be read via 1572 * dso__data_read_addr(). 1573 */ 1574 if (dso__kernel(dso) == DSO_SPACE__KERNEL_GUEST) 1575 dso__set_binary_type(dso, DSO_BINARY_TYPE__GUEST_KCORE); 1576 else 1577 dso__set_binary_type(dso, DSO_BINARY_TYPE__KCORE); 1578 dso__set_long_name(dso, strdup(kcore_filename), true); 1579 1580 close(fd); 1581 1582 if (map__prot(map) & PROT_EXEC) 1583 pr_debug("Using %s for kernel object code\n", kcore_filename); 1584 else 1585 pr_debug("Using %s for kernel data\n", kcore_filename); 1586 1587 return 0; 1588 1589 out_err: 1590 while (!list_empty(&md.maps)) { 1591 struct map_list_node *list_node; 1592 1593 list_node = list_entry(md.maps.next, struct map_list_node, node); 1594 list_del_init(&list_node->node); 1595 map__zput(list_node->map); 1596 free(list_node); 1597 } 1598 close(fd); 1599 return err; 1600 } 1601 1602 /* 1603 * If the kernel is relocated at boot time, kallsyms won't match. Compute the 1604 * delta based on the relocation reference symbol. 1605 */ 1606 static int kallsyms__delta(struct kmap *kmap, const char *filename, u64 *delta) 1607 { 1608 u64 addr; 1609 1610 if (!kmap->ref_reloc_sym || !kmap->ref_reloc_sym->name) 1611 return 0; 1612 1613 if (kallsyms__get_function_start(filename, kmap->ref_reloc_sym->name, &addr)) 1614 return -1; 1615 1616 *delta = addr - kmap->ref_reloc_sym->addr; 1617 return 0; 1618 } 1619 1620 int __dso__load_kallsyms(struct dso *dso, const char *filename, 1621 struct map *map, bool no_kcore) 1622 { 1623 struct kmap *kmap = map__kmap(map); 1624 u64 delta = 0; 1625 1626 if (symbol__restricted_filename(filename, "/proc/kallsyms")) 1627 return -1; 1628 1629 if (!kmap || !kmap->kmaps) 1630 return -1; 1631 1632 if (dso__load_all_kallsyms(dso, filename) < 0) 1633 return -1; 1634 1635 if (kallsyms__delta(kmap, filename, &delta)) 1636 return -1; 1637 1638 symbols__fixup_end(dso__symbols(dso), true); 1639 symbols__fixup_duplicate(dso__symbols(dso)); 1640 1641 if (dso__kernel(dso) == DSO_SPACE__KERNEL_GUEST) 1642 dso__set_symtab_type(dso, DSO_BINARY_TYPE__GUEST_KALLSYMS); 1643 else 1644 dso__set_symtab_type(dso, DSO_BINARY_TYPE__KALLSYMS); 1645 1646 if (!no_kcore && !dso__load_kcore(dso, map, filename)) 1647 return maps__split_kallsyms_for_kcore(kmap->kmaps, dso); 1648 else 1649 return maps__split_kallsyms(kmap->kmaps, dso, delta, map); 1650 } 1651 1652 int dso__load_kallsyms(struct dso *dso, const char *filename, 1653 struct map *map) 1654 { 1655 return __dso__load_kallsyms(dso, filename, map, false); 1656 } 1657 1658 static int dso__load_perf_map(const char *map_path, struct dso *dso) 1659 { 1660 char *line = NULL; 1661 size_t n; 1662 FILE *file; 1663 int nr_syms = 0; 1664 1665 file = fopen(map_path, "r"); 1666 if (file == NULL) 1667 goto out_failure; 1668 1669 while (!feof(file)) { 1670 u64 start, size; 1671 struct symbol *sym; 1672 int line_len, len; 1673 1674 line_len = getline(&line, &n, file); 1675 if (line_len < 0) 1676 break; 1677 1678 if (!line) 1679 goto out_failure; 1680 1681 line[--line_len] = '\0'; /* \n */ 1682 1683 len = hex2u64(line, &start); 1684 1685 len++; 1686 if (len + 2 >= line_len) 1687 continue; 1688 1689 len += hex2u64(line + len, &size); 1690 1691 len++; 1692 if (len + 2 >= line_len) 1693 continue; 1694 1695 sym = symbol__new(start, size, STB_GLOBAL, STT_FUNC, line + len); 1696 1697 if (sym == NULL) 1698 goto out_delete_line; 1699 1700 symbols__insert(dso__symbols(dso), sym); 1701 nr_syms++; 1702 } 1703 1704 free(line); 1705 fclose(file); 1706 1707 return nr_syms; 1708 1709 out_delete_line: 1710 free(line); 1711 out_failure: 1712 return -1; 1713 } 1714 1715 static bool dso__is_compatible_symtab_type(struct dso *dso, bool kmod, 1716 enum dso_binary_type type) 1717 { 1718 switch (type) { 1719 case DSO_BINARY_TYPE__JAVA_JIT: 1720 case DSO_BINARY_TYPE__DEBUGLINK: 1721 case DSO_BINARY_TYPE__SYSTEM_PATH_DSO: 1722 case DSO_BINARY_TYPE__FEDORA_DEBUGINFO: 1723 case DSO_BINARY_TYPE__UBUNTU_DEBUGINFO: 1724 case DSO_BINARY_TYPE__MIXEDUP_UBUNTU_DEBUGINFO: 1725 case DSO_BINARY_TYPE__BUILDID_DEBUGINFO: 1726 case DSO_BINARY_TYPE__OPENEMBEDDED_DEBUGINFO: 1727 case DSO_BINARY_TYPE__GNU_DEBUGDATA: 1728 return !kmod && dso__kernel(dso) == DSO_SPACE__USER; 1729 1730 case DSO_BINARY_TYPE__KALLSYMS: 1731 case DSO_BINARY_TYPE__VMLINUX: 1732 case DSO_BINARY_TYPE__KCORE: 1733 return dso__kernel(dso) == DSO_SPACE__KERNEL; 1734 1735 case DSO_BINARY_TYPE__GUEST_KALLSYMS: 1736 case DSO_BINARY_TYPE__GUEST_VMLINUX: 1737 case DSO_BINARY_TYPE__GUEST_KCORE: 1738 return dso__kernel(dso) == DSO_SPACE__KERNEL_GUEST; 1739 1740 case DSO_BINARY_TYPE__GUEST_KMODULE: 1741 case DSO_BINARY_TYPE__GUEST_KMODULE_COMP: 1742 case DSO_BINARY_TYPE__SYSTEM_PATH_KMODULE: 1743 case DSO_BINARY_TYPE__SYSTEM_PATH_KMODULE_COMP: 1744 /* 1745 * kernel modules know their symtab type - it's set when 1746 * creating a module dso in machine__addnew_module_map(). 1747 */ 1748 return kmod && dso__symtab_type(dso) == type; 1749 1750 case DSO_BINARY_TYPE__BUILD_ID_CACHE: 1751 case DSO_BINARY_TYPE__BUILD_ID_CACHE_DEBUGINFO: 1752 return true; 1753 1754 case DSO_BINARY_TYPE__BPF_PROG_INFO: 1755 case DSO_BINARY_TYPE__BPF_IMAGE: 1756 case DSO_BINARY_TYPE__OOL: 1757 case DSO_BINARY_TYPE__NOT_FOUND: 1758 default: 1759 return false; 1760 } 1761 } 1762 1763 /* Checks for the existence of the perf-<pid>.map file in two different 1764 * locations. First, if the process is a separate mount namespace, check in 1765 * that namespace using the pid of the innermost pid namespace. If's not in a 1766 * namespace, or the file can't be found there, try in the mount namespace of 1767 * the tracing process using our view of its pid. 1768 */ 1769 static int dso__find_perf_map(char *filebuf, size_t bufsz, 1770 struct nsinfo **nsip) 1771 { 1772 struct nscookie nsc; 1773 struct nsinfo *nsi; 1774 struct nsinfo *nnsi; 1775 int rc = -1; 1776 1777 nsi = *nsip; 1778 1779 if (nsinfo__need_setns(nsi)) { 1780 snprintf(filebuf, bufsz, "/tmp/perf-%d.map", nsinfo__nstgid(nsi)); 1781 nsinfo__mountns_enter(nsi, &nsc); 1782 rc = access(filebuf, R_OK); 1783 nsinfo__mountns_exit(&nsc); 1784 if (rc == 0) 1785 return rc; 1786 } 1787 1788 nnsi = nsinfo__copy(nsi); 1789 if (nnsi) { 1790 nsinfo__put(nsi); 1791 1792 nsinfo__clear_need_setns(nnsi); 1793 snprintf(filebuf, bufsz, "/tmp/perf-%d.map", nsinfo__tgid(nnsi)); 1794 *nsip = nnsi; 1795 rc = 0; 1796 } 1797 1798 return rc; 1799 } 1800 1801 int dso__load(struct dso *dso, struct map *map) 1802 { 1803 char *name; 1804 int ret = -1; 1805 u_int i; 1806 struct machine *machine = NULL; 1807 char *root_dir = (char *) ""; 1808 int ss_pos = 0; 1809 struct symsrc ss_[2]; 1810 struct symsrc *syms_ss = NULL, *runtime_ss = NULL; 1811 bool kmod; 1812 bool perfmap; 1813 struct nscookie nsc; 1814 char newmapname[PATH_MAX]; 1815 const char *map_path = dso__long_name(dso); 1816 1817 mutex_lock(dso__lock(dso)); 1818 perfmap = is_perf_pid_map_name(map_path); 1819 1820 if (perfmap) { 1821 if (dso__nsinfo(dso) && 1822 (dso__find_perf_map(newmapname, sizeof(newmapname), 1823 dso__nsinfo_ptr(dso)) == 0)) { 1824 map_path = newmapname; 1825 } 1826 } 1827 1828 nsinfo__mountns_enter(dso__nsinfo(dso), &nsc); 1829 1830 /* check again under the dso->lock */ 1831 if (dso__loaded(dso)) { 1832 ret = 1; 1833 goto out; 1834 } 1835 1836 kmod = dso__is_kmod(dso); 1837 1838 if (dso__kernel(dso) && !kmod) { 1839 if (dso__kernel(dso) == DSO_SPACE__KERNEL) 1840 ret = dso__load_kernel_sym(dso, map); 1841 else if (dso__kernel(dso) == DSO_SPACE__KERNEL_GUEST) 1842 ret = dso__load_guest_kernel_sym(dso, map); 1843 1844 machine = maps__machine(map__kmaps(map)); 1845 if (machine && machine_or_dso_e_machine(machine, dso) == EM_X86_64) 1846 machine__map_x86_64_entry_trampolines(machine, dso); 1847 goto out; 1848 } 1849 1850 dso__set_adjust_symbols(dso, false); 1851 1852 if (perfmap) { 1853 ret = dso__load_perf_map(map_path, dso); 1854 dso__set_symtab_type(dso, ret > 0 1855 ? DSO_BINARY_TYPE__JAVA_JIT 1856 : DSO_BINARY_TYPE__NOT_FOUND); 1857 goto out; 1858 } 1859 1860 if (machine) 1861 root_dir = machine->root_dir; 1862 1863 name = malloc(PATH_MAX); 1864 if (!name) 1865 goto out; 1866 1867 /* 1868 * Read the build id if possible. This is required for 1869 * DSO_BINARY_TYPE__BUILDID_DEBUGINFO to work. 1870 */ 1871 if (!dso__has_build_id(dso)) { 1872 struct build_id bid = { .size = 0, }; 1873 1874 __symbol__join_symfs(name, PATH_MAX, dso__long_name(dso)); 1875 if (filename__read_build_id(name, &bid) > 0) 1876 dso__set_build_id(dso, &bid); 1877 } 1878 1879 /* 1880 * Iterate over candidate debug images. 1881 * Keep track of "interesting" ones (those which have a symtab, dynsym, 1882 * and/or opd section) for processing. 1883 */ 1884 for (i = 0; i < DSO_BINARY_TYPE__SYMTAB_CNT; i++) { 1885 struct symsrc *ss = &ss_[ss_pos]; 1886 bool next_slot = false; 1887 bool is_reg; 1888 bool nsexit; 1889 int bfdrc = -1; 1890 int sirc = -1; 1891 1892 enum dso_binary_type symtab_type = binary_type_symtab[i]; 1893 1894 nsexit = (symtab_type == DSO_BINARY_TYPE__BUILD_ID_CACHE || 1895 symtab_type == DSO_BINARY_TYPE__BUILD_ID_CACHE_DEBUGINFO); 1896 1897 if (!dso__is_compatible_symtab_type(dso, kmod, symtab_type)) 1898 continue; 1899 1900 if (dso__read_binary_type_filename(dso, symtab_type, 1901 root_dir, name, PATH_MAX)) 1902 continue; 1903 1904 if (nsexit) 1905 nsinfo__mountns_exit(&nsc); 1906 1907 is_reg = is_regular_file(name); 1908 if (!is_reg && errno == ENOENT && dso__nsinfo(dso)) { 1909 char *new_name = dso__filename_with_chroot(dso, name); 1910 if (new_name) { 1911 is_reg = is_regular_file(new_name); 1912 strlcpy(name, new_name, PATH_MAX); 1913 free(new_name); 1914 } 1915 } 1916 1917 #ifdef HAVE_LIBBFD_SUPPORT 1918 if (is_reg) 1919 bfdrc = dso__load_bfd_symbols(dso, name); 1920 #endif 1921 if (is_reg && bfdrc < 0) 1922 sirc = symsrc__init(ss, dso, name, symtab_type); 1923 1924 if (nsexit) 1925 nsinfo__mountns_enter(dso__nsinfo(dso), &nsc); 1926 1927 if (bfdrc == 0) { 1928 ret = 0; 1929 break; 1930 } 1931 1932 if (!is_reg || sirc < 0) 1933 continue; 1934 1935 if (!syms_ss && symsrc__has_symtab(ss)) { 1936 syms_ss = ss; 1937 next_slot = true; 1938 if (!dso__symsrc_filename(dso)) 1939 dso__set_symsrc_filename(dso, strdup(name)); 1940 } 1941 1942 if (!runtime_ss && symsrc__possibly_runtime(ss)) { 1943 runtime_ss = ss; 1944 next_slot = true; 1945 } 1946 1947 if (next_slot) { 1948 ss_pos++; 1949 1950 if (dso__binary_type(dso) == DSO_BINARY_TYPE__NOT_FOUND) 1951 dso__set_binary_type(dso, symtab_type); 1952 1953 if (syms_ss && runtime_ss) 1954 break; 1955 } else { 1956 symsrc__destroy(ss); 1957 } 1958 1959 } 1960 1961 if (!runtime_ss && !syms_ss) 1962 goto out_free; 1963 1964 if (runtime_ss && !syms_ss) { 1965 syms_ss = runtime_ss; 1966 } 1967 1968 /* We'll have to hope for the best */ 1969 if (!runtime_ss && syms_ss) 1970 runtime_ss = syms_ss; 1971 1972 if (syms_ss) 1973 ret = dso__load_sym(dso, map, syms_ss, runtime_ss, kmod); 1974 else 1975 ret = -1; 1976 1977 if (ret > 0) { 1978 int nr_plt; 1979 1980 nr_plt = dso__synthesize_plt_symbols(dso, runtime_ss); 1981 if (nr_plt > 0) 1982 ret += nr_plt; 1983 } 1984 1985 for (; ss_pos > 0; ss_pos--) 1986 symsrc__destroy(&ss_[ss_pos - 1]); 1987 out_free: 1988 free(name); 1989 if (ret < 0 && strstr(dso__name(dso), " (deleted)") != NULL) 1990 ret = 0; 1991 out: 1992 dso__set_loaded(dso); 1993 mutex_unlock(dso__lock(dso)); 1994 nsinfo__mountns_exit(&nsc); 1995 1996 return ret; 1997 } 1998 1999 /* 2000 * Always takes ownership of vmlinux when vmlinux_allocated == true, even if 2001 * it returns an error. 2002 */ 2003 int dso__load_vmlinux(struct dso *dso, struct map *map, 2004 const char *vmlinux, bool vmlinux_allocated) 2005 { 2006 int err = -1; 2007 struct symsrc ss; 2008 char symfs_vmlinux[PATH_MAX]; 2009 enum dso_binary_type symtab_type; 2010 2011 if (vmlinux[0] == '/') 2012 snprintf(symfs_vmlinux, sizeof(symfs_vmlinux), "%s", vmlinux); 2013 else 2014 symbol__join_symfs(symfs_vmlinux, vmlinux); 2015 2016 if (dso__kernel(dso) == DSO_SPACE__KERNEL_GUEST) 2017 symtab_type = DSO_BINARY_TYPE__GUEST_VMLINUX; 2018 else 2019 symtab_type = DSO_BINARY_TYPE__VMLINUX; 2020 2021 if (symsrc__init(&ss, dso, symfs_vmlinux, symtab_type)) { 2022 if (vmlinux_allocated) 2023 free((char *) vmlinux); 2024 return -1; 2025 } 2026 2027 /* 2028 * dso__load_sym() may copy 'dso' which will result in the copies having 2029 * an incorrect long name unless we set it here first. 2030 */ 2031 dso__set_long_name(dso, vmlinux, vmlinux_allocated); 2032 if (dso__kernel(dso) == DSO_SPACE__KERNEL_GUEST) 2033 dso__set_binary_type(dso, DSO_BINARY_TYPE__GUEST_VMLINUX); 2034 else 2035 dso__set_binary_type(dso, DSO_BINARY_TYPE__VMLINUX); 2036 2037 err = dso__load_sym(dso, map, &ss, &ss, 0); 2038 symsrc__destroy(&ss); 2039 2040 if (err > 0) { 2041 dso__set_loaded(dso); 2042 pr_debug("Using %s for symbols\n", symfs_vmlinux); 2043 } 2044 2045 return err; 2046 } 2047 2048 int dso__load_vmlinux_path(struct dso *dso, struct map *map) 2049 { 2050 int i, err = 0; 2051 char *filename = NULL; 2052 2053 pr_debug("Looking at the vmlinux_path (%d entries long)\n", 2054 vmlinux_path__nr_entries + 1); 2055 2056 for (i = 0; i < vmlinux_path__nr_entries; ++i) { 2057 err = dso__load_vmlinux(dso, map, vmlinux_path[i], false); 2058 if (err > 0) 2059 goto out; 2060 } 2061 2062 if (!symbol_conf.ignore_vmlinux_buildid) 2063 filename = dso__build_id_filename(dso, NULL, 0, false); 2064 if (filename != NULL) { 2065 err = dso__load_vmlinux(dso, map, filename, true); 2066 if (err > 0) 2067 goto out; 2068 } 2069 out: 2070 return err; 2071 } 2072 2073 static bool visible_dir_filter(const char *name, struct dirent *d) 2074 { 2075 if (d->d_type != DT_DIR) 2076 return false; 2077 return lsdir_no_dot_filter(name, d); 2078 } 2079 2080 static int find_matching_kcore(struct map *map, char *dir, size_t dir_sz) 2081 { 2082 char kallsyms_filename[PATH_MAX]; 2083 int ret = -1; 2084 struct strlist *dirs; 2085 struct str_node *nd; 2086 2087 dirs = lsdir(dir, visible_dir_filter); 2088 if (!dirs) 2089 return -1; 2090 2091 strlist__for_each_entry(nd, dirs) { 2092 scnprintf(kallsyms_filename, sizeof(kallsyms_filename), 2093 "%s/%s/kallsyms", dir, nd->s); 2094 if (!validate_kcore_addresses(kallsyms_filename, map)) { 2095 strlcpy(dir, kallsyms_filename, dir_sz); 2096 ret = 0; 2097 break; 2098 } 2099 } 2100 2101 strlist__delete(dirs); 2102 2103 return ret; 2104 } 2105 2106 /* 2107 * Use open(O_RDONLY) to check readability directly instead of access(R_OK) 2108 * since access(R_OK) only checks with real UID/GID but open() use effective 2109 * UID/GID and actual capabilities (e.g. /proc/kcore requires CAP_SYS_RAWIO). 2110 */ 2111 static bool filename__readable(const char *file) 2112 { 2113 int fd = open(file, O_RDONLY); 2114 if (fd < 0) 2115 return false; 2116 close(fd); 2117 return true; 2118 } 2119 2120 static char *dso__find_kallsyms(struct dso *dso, struct map *map) 2121 { 2122 struct build_id bid = { .size = 0, }; 2123 char sbuild_id[SBUILD_ID_SIZE]; 2124 bool is_host = false; 2125 char path[PATH_MAX]; 2126 struct maps *kmaps = map__kmaps(map); 2127 2128 if (!dso__has_build_id(dso)) { 2129 /* 2130 * Last resort, if we don't have a build-id and couldn't find 2131 * any vmlinux file, try the running kernel kallsyms table. 2132 */ 2133 goto proc_kallsyms; 2134 } 2135 2136 if (sysfs__read_build_id("/sys/kernel/notes", &bid) == 0) 2137 is_host = dso__build_id_equal(dso, &bid); 2138 2139 /* Try a fast path for /proc/kallsyms if possible */ 2140 if (is_host) { 2141 /* 2142 * Do not check the build-id cache, unless we know we cannot use 2143 * /proc/kcore or module maps don't match to /proc/kallsyms. 2144 * To check readability of /proc/kcore, do not use access(R_OK) 2145 * since /proc/kcore requires CAP_SYS_RAWIO to read and access 2146 * can't check it. 2147 */ 2148 if (filename__readable("/proc/kcore") && 2149 !validate_kcore_addresses("/proc/kallsyms", map)) 2150 goto proc_kallsyms; 2151 } 2152 2153 build_id__snprintf(dso__bid(dso), sbuild_id, sizeof(sbuild_id)); 2154 2155 /* Find kallsyms in build-id cache with kcore */ 2156 scnprintf(path, sizeof(path), "%s/%s/%s", 2157 buildid_dir, DSO__NAME_KCORE, sbuild_id); 2158 2159 if (!find_matching_kcore(map, path, sizeof(path))) 2160 return strdup(path); 2161 2162 /* Use current /proc/kallsyms if possible */ 2163 proc_kallsyms: 2164 if (kmaps) { 2165 struct machine *machine = maps__machine(kmaps); 2166 2167 scnprintf(path, sizeof(path), "%s/proc/kallsyms", machine->root_dir); 2168 return strdup(path); 2169 } else if (is_host) { 2170 return strdup("/proc/kallsyms"); 2171 } 2172 2173 /* Finally, find a cache of kallsyms */ 2174 if (!build_id_cache__kallsyms_path(sbuild_id, path, sizeof(path))) { 2175 pr_err("No kallsyms or vmlinux with build-id %s was found\n", 2176 sbuild_id); 2177 return NULL; 2178 } 2179 2180 return strdup(path); 2181 } 2182 2183 static int dso__load_kernel_sym(struct dso *dso, struct map *map) 2184 { 2185 int err; 2186 const char *kallsyms_filename = NULL; 2187 char *kallsyms_allocated_filename = NULL; 2188 char *filename = NULL; 2189 2190 /* 2191 * Step 1: if the user specified a kallsyms or vmlinux filename, use 2192 * it and only it, reporting errors to the user if it cannot be used. 2193 * 2194 * For instance, try to analyse an ARM perf.data file _without_ a 2195 * build-id, or if the user specifies the wrong path to the right 2196 * vmlinux file, obviously we can't fallback to another vmlinux (a 2197 * x86_86 one, on the machine where analysis is being performed, say), 2198 * or worse, /proc/kallsyms. 2199 * 2200 * If the specified file _has_ a build-id and there is a build-id 2201 * section in the perf.data file, we will still do the expected 2202 * validation in dso__load_vmlinux and will bail out if they don't 2203 * match. 2204 */ 2205 if (symbol_conf.kallsyms_name != NULL) { 2206 kallsyms_filename = symbol_conf.kallsyms_name; 2207 goto do_kallsyms; 2208 } 2209 2210 if (!symbol_conf.ignore_vmlinux && symbol_conf.vmlinux_name != NULL) { 2211 return dso__load_vmlinux(dso, map, symbol_conf.vmlinux_name, false); 2212 } 2213 2214 /* 2215 * Before checking on common vmlinux locations, check if it's 2216 * stored as standard build id binary (not kallsyms) under 2217 * .debug cache. 2218 */ 2219 if (!symbol_conf.ignore_vmlinux_buildid) 2220 filename = __dso__build_id_filename(dso, NULL, 0, false, false); 2221 if (filename != NULL) { 2222 err = dso__load_vmlinux(dso, map, filename, true); 2223 if (err > 0) 2224 return err; 2225 } 2226 2227 if (!symbol_conf.ignore_vmlinux && vmlinux_path != NULL) { 2228 err = dso__load_vmlinux_path(dso, map); 2229 if (err > 0) 2230 return err; 2231 } 2232 2233 /* do not try local files if a symfs was given */ 2234 if (symbol_conf.symfs[0] != 0) 2235 return -1; 2236 2237 kallsyms_allocated_filename = dso__find_kallsyms(dso, map); 2238 if (!kallsyms_allocated_filename) 2239 return -1; 2240 2241 kallsyms_filename = kallsyms_allocated_filename; 2242 2243 do_kallsyms: 2244 err = dso__load_kallsyms(dso, kallsyms_filename, map); 2245 if (err > 0) 2246 pr_debug("Using %s for symbols\n", kallsyms_filename); 2247 free(kallsyms_allocated_filename); 2248 2249 if (err > 0 && !dso__is_kcore(dso)) { 2250 struct maps *kmaps = map__kmaps(map); 2251 2252 dso__set_binary_type(dso, DSO_BINARY_TYPE__KALLSYMS); 2253 dso__set_long_name(dso, DSO__NAME_KALLSYMS, false); 2254 maps__mutate_mapping(kmaps, map, map_fixup_cb, NULL); 2255 } 2256 2257 return err; 2258 } 2259 2260 static int dso__load_guest_kernel_sym(struct dso *dso, struct map *map) 2261 { 2262 int err; 2263 const char *kallsyms_filename; 2264 struct machine *machine = maps__machine(map__kmaps(map)); 2265 char path[PATH_MAX]; 2266 2267 if (machine->kallsyms_filename) { 2268 kallsyms_filename = machine->kallsyms_filename; 2269 } else if (machine__is_default_guest(machine)) { 2270 /* 2271 * if the user specified a vmlinux filename, use it and only 2272 * it, reporting errors to the user if it cannot be used. 2273 * Or use file guest_kallsyms inputted by user on commandline 2274 */ 2275 if (symbol_conf.default_guest_vmlinux_name != NULL) { 2276 err = dso__load_vmlinux(dso, map, 2277 symbol_conf.default_guest_vmlinux_name, 2278 false); 2279 return err; 2280 } 2281 2282 kallsyms_filename = symbol_conf.default_guest_kallsyms; 2283 if (!kallsyms_filename) 2284 return -1; 2285 } else { 2286 snprintf(path, sizeof(path), "%s/proc/kallsyms", machine->root_dir); 2287 kallsyms_filename = path; 2288 } 2289 2290 err = dso__load_kallsyms(dso, kallsyms_filename, map); 2291 if (err > 0) 2292 pr_debug("Using %s for symbols\n", kallsyms_filename); 2293 if (err > 0 && !dso__is_kcore(dso)) { 2294 struct maps *kmaps = map__kmaps(map); 2295 2296 dso__set_binary_type(dso, DSO_BINARY_TYPE__GUEST_KALLSYMS); 2297 dso__set_long_name(dso, machine->mmap_name, false); 2298 maps__mutate_mapping(kmaps, map, map_fixup_cb, NULL); 2299 } 2300 2301 return err; 2302 } 2303 2304 static void vmlinux_path__exit(void) 2305 { 2306 while (--vmlinux_path__nr_entries >= 0) 2307 zfree(&vmlinux_path[vmlinux_path__nr_entries]); 2308 vmlinux_path__nr_entries = 0; 2309 2310 zfree(&vmlinux_path); 2311 } 2312 2313 static const char * const vmlinux_paths[] = { 2314 "vmlinux", 2315 "/boot/vmlinux" 2316 }; 2317 2318 static const char * const vmlinux_paths_upd[] = { 2319 "/boot/vmlinux-%s", 2320 "/usr/lib/debug/boot/vmlinux-%s", 2321 "/lib/modules/%s/build/vmlinux", 2322 "/usr/lib/debug/lib/modules/%s/vmlinux", 2323 "/usr/lib/debug/boot/vmlinux-%s.debug" 2324 }; 2325 2326 static int vmlinux_path__add(const char *new_entry) 2327 { 2328 vmlinux_path[vmlinux_path__nr_entries] = strdup(new_entry); 2329 if (vmlinux_path[vmlinux_path__nr_entries] == NULL) 2330 return -1; 2331 ++vmlinux_path__nr_entries; 2332 2333 return 0; 2334 } 2335 2336 static int vmlinux_path__init(struct perf_env *env) 2337 { 2338 struct utsname uts; 2339 char bf[PATH_MAX]; 2340 const char *kernel_version; 2341 unsigned int i; 2342 2343 vmlinux_path = malloc(sizeof(char *) * (ARRAY_SIZE(vmlinux_paths) + 2344 ARRAY_SIZE(vmlinux_paths_upd))); 2345 if (vmlinux_path == NULL) 2346 return -1; 2347 2348 for (i = 0; i < ARRAY_SIZE(vmlinux_paths); i++) 2349 if (vmlinux_path__add(vmlinux_paths[i]) < 0) 2350 goto out_fail; 2351 2352 /* only try kernel version if no symfs was given */ 2353 if (symbol_conf.symfs[0] != 0) 2354 return 0; 2355 2356 if (env) { 2357 kernel_version = perf_env__os_release(env); 2358 } else { 2359 if (uname(&uts) < 0) 2360 goto out_fail; 2361 2362 kernel_version = uts.release; 2363 } 2364 2365 for (i = 0; i < ARRAY_SIZE(vmlinux_paths_upd); i++) { 2366 snprintf(bf, sizeof(bf), vmlinux_paths_upd[i], kernel_version); 2367 if (vmlinux_path__add(bf) < 0) 2368 goto out_fail; 2369 } 2370 2371 return 0; 2372 2373 out_fail: 2374 vmlinux_path__exit(); 2375 return -1; 2376 } 2377 2378 int setup_list(struct strlist **list, const char *list_str, 2379 const char *list_name) 2380 { 2381 if (list_str == NULL) 2382 return 0; 2383 2384 *list = strlist__new(list_str, NULL); 2385 if (!*list) { 2386 pr_err("problems parsing %s list\n", list_name); 2387 return -1; 2388 } 2389 2390 symbol_conf.has_filter = true; 2391 return 0; 2392 } 2393 2394 int setup_intlist(struct intlist **list, const char *list_str, 2395 const char *list_name) 2396 { 2397 if (list_str == NULL) 2398 return 0; 2399 2400 *list = intlist__new(list_str); 2401 if (!*list) { 2402 pr_err("problems parsing %s list\n", list_name); 2403 return -1; 2404 } 2405 return 0; 2406 } 2407 2408 static int setup_addrlist(struct intlist **addr_list, struct strlist *sym_list) 2409 { 2410 struct str_node *pos, *tmp; 2411 unsigned long val; 2412 char *sep; 2413 const char *end; 2414 int i = 0, err; 2415 2416 *addr_list = intlist__new(NULL); 2417 if (!*addr_list) 2418 return -1; 2419 2420 strlist__for_each_entry_safe(pos, tmp, sym_list) { 2421 errno = 0; 2422 val = strtoul(pos->s, &sep, 16); 2423 if (errno || (sep == pos->s)) 2424 continue; 2425 2426 if (*sep != '\0') { 2427 end = pos->s + strlen(pos->s) - 1; 2428 while (end >= sep && isspace(*end)) 2429 end--; 2430 2431 if (end >= sep) 2432 continue; 2433 } 2434 2435 err = intlist__add(*addr_list, val); 2436 if (err) 2437 break; 2438 2439 strlist__remove(sym_list, pos); 2440 i++; 2441 } 2442 2443 if (i == 0) { 2444 intlist__delete(*addr_list); 2445 *addr_list = NULL; 2446 } 2447 2448 return 0; 2449 } 2450 2451 static bool symbol__read_kptr_restrict(void) 2452 { 2453 bool value = false; 2454 FILE *fp = fopen("/proc/sys/kernel/kptr_restrict", "r"); 2455 bool used_root; 2456 bool cap_syslog = perf_cap__capable(CAP_SYSLOG, &used_root); 2457 2458 if (fp != NULL) { 2459 char line[8]; 2460 2461 if (fgets(line, sizeof(line), fp) != NULL) 2462 value = cap_syslog ? (atoi(line) >= 2) : (atoi(line) != 0); 2463 2464 fclose(fp); 2465 } 2466 2467 /* Per kernel/kallsyms.c: 2468 * we also restrict when perf_event_paranoid > 1 w/o CAP_SYSLOG 2469 */ 2470 if (perf_event_paranoid() > 1 && !cap_syslog) 2471 value = true; 2472 2473 return value; 2474 } 2475 2476 int symbol__annotation_init(void) 2477 { 2478 if (symbol_conf.init_annotation) 2479 return 0; 2480 2481 if (symbol_conf.initialized) { 2482 pr_err("Annotation needs to be init before symbol__init()\n"); 2483 return -1; 2484 } 2485 2486 symbol_conf.priv_size += sizeof(struct annotation); 2487 symbol_conf.init_annotation = true; 2488 return 0; 2489 } 2490 2491 static int setup_parallelism_bitmap(void) 2492 { 2493 struct perf_cpu_map *map; 2494 struct perf_cpu cpu; 2495 unsigned int i; 2496 int err = -1; 2497 2498 if (symbol_conf.parallelism_list_str == NULL) 2499 return 0; 2500 2501 map = perf_cpu_map__new(symbol_conf.parallelism_list_str); 2502 if (map == NULL) { 2503 pr_err("failed to parse parallelism filter list\n"); 2504 return -1; 2505 } 2506 2507 bitmap_fill(symbol_conf.parallelism_filter, MAX_NR_CPUS + 1); 2508 perf_cpu_map__for_each_cpu(cpu, i, map) { 2509 if (cpu.cpu <= 0 || cpu.cpu > MAX_NR_CPUS) { 2510 pr_err("Requested parallelism level %d is invalid.\n", cpu.cpu); 2511 goto out_delete_map; 2512 } 2513 __clear_bit(cpu.cpu, symbol_conf.parallelism_filter); 2514 } 2515 2516 err = 0; 2517 out_delete_map: 2518 perf_cpu_map__put(map); 2519 return err; 2520 } 2521 2522 int symbol__init(struct perf_env *env) 2523 { 2524 const char *symfs; 2525 2526 if (symbol_conf.initialized) 2527 return 0; 2528 2529 symbol_conf.priv_size = PERF_ALIGN(symbol_conf.priv_size, sizeof(u64)); 2530 2531 symbol__elf_init(); 2532 2533 if (symbol_conf.try_vmlinux_path && vmlinux_path__init(env) < 0) 2534 return -1; 2535 2536 if (symbol_conf.field_sep && *symbol_conf.field_sep == '.') { 2537 pr_err("'.' is the only non valid --field-separator argument\n"); 2538 return -1; 2539 } 2540 2541 if (setup_parallelism_bitmap()) 2542 return -1; 2543 2544 if (setup_list(&symbol_conf.dso_list, 2545 symbol_conf.dso_list_str, "dso") < 0) 2546 return -1; 2547 2548 if (setup_list(&symbol_conf.comm_list, 2549 symbol_conf.comm_list_str, "comm") < 0) 2550 goto out_free_dso_list; 2551 2552 if (setup_intlist(&symbol_conf.pid_list, 2553 symbol_conf.pid_list_str, "pid") < 0) 2554 goto out_free_comm_list; 2555 2556 if (setup_intlist(&symbol_conf.tid_list, 2557 symbol_conf.tid_list_str, "tid") < 0) 2558 goto out_free_pid_list; 2559 2560 if (setup_list(&symbol_conf.sym_list, 2561 symbol_conf.sym_list_str, "symbol") < 0) 2562 goto out_free_tid_list; 2563 2564 if (symbol_conf.sym_list && 2565 setup_addrlist(&symbol_conf.addr_list, symbol_conf.sym_list) < 0) 2566 goto out_free_sym_list; 2567 2568 if (setup_list(&symbol_conf.bt_stop_list, 2569 symbol_conf.bt_stop_list_str, "symbol") < 0) 2570 goto out_free_sym_list; 2571 2572 /* 2573 * A path to symbols of "/" is identical to "" 2574 * reset here for simplicity. 2575 */ 2576 symfs = realpath(symbol_conf.symfs, NULL); 2577 if (symfs == NULL) 2578 symfs = symbol_conf.symfs; 2579 if (strcmp(symfs, "/") == 0) 2580 symbol_conf.symfs = ""; 2581 if (symfs != symbol_conf.symfs) 2582 free((void *)symfs); 2583 2584 symbol_conf.kptr_restrict = symbol__read_kptr_restrict(); 2585 2586 symbol_conf.initialized = true; 2587 return 0; 2588 2589 out_free_sym_list: 2590 strlist__delete(symbol_conf.sym_list); 2591 intlist__delete(symbol_conf.addr_list); 2592 out_free_tid_list: 2593 intlist__delete(symbol_conf.tid_list); 2594 out_free_pid_list: 2595 intlist__delete(symbol_conf.pid_list); 2596 out_free_comm_list: 2597 strlist__delete(symbol_conf.comm_list); 2598 out_free_dso_list: 2599 strlist__delete(symbol_conf.dso_list); 2600 return -1; 2601 } 2602 2603 void symbol__exit(void) 2604 { 2605 if (!symbol_conf.initialized) 2606 return; 2607 2608 strlist__delete(symbol_conf.bt_stop_list); 2609 strlist__delete(symbol_conf.sym_list); 2610 strlist__delete(symbol_conf.dso_list); 2611 strlist__delete(symbol_conf.comm_list); 2612 intlist__delete(symbol_conf.tid_list); 2613 intlist__delete(symbol_conf.pid_list); 2614 intlist__delete(symbol_conf.addr_list); 2615 vmlinux_path__exit(); 2616 symbol_conf.sym_list = symbol_conf.dso_list = symbol_conf.comm_list = NULL; 2617 symbol_conf.bt_stop_list = NULL; 2618 symbol_conf.initialized = false; 2619 } 2620 2621 int symbol__config_symfs(const struct option *opt __maybe_unused, 2622 const char *dir, int unset __maybe_unused) 2623 { 2624 char *bf = NULL; 2625 const char *layout_str; 2626 char *dir_copy; 2627 int ret; 2628 2629 layout_str = strrchr(dir, ','); 2630 if (layout_str) { 2631 size_t dir_len = layout_str - dir; 2632 2633 dir_copy = strndup(dir, dir_len); 2634 if (dir_copy == NULL) 2635 return -ENOMEM; 2636 2637 symbol_conf.symfs = dir_copy; 2638 2639 layout_str++; 2640 if (!strcmp(layout_str, "flat")) 2641 symbol_conf.symfs_layout_flat = true; 2642 else if (!strcmp(layout_str, "hierarchy")) 2643 symbol_conf.symfs_layout_flat = false; 2644 else { 2645 pr_err("Invalid layout: '%s', use 'hierarchy' or 'flat'\n", 2646 layout_str); 2647 free(dir_copy); 2648 return -EINVAL; 2649 } 2650 } else { 2651 symbol_conf.symfs = strdup(dir); 2652 if (symbol_conf.symfs == NULL) 2653 return -ENOMEM; 2654 symbol_conf.symfs_layout_flat = false; 2655 } 2656 2657 /* skip the locally configured cache if a symfs is given, and 2658 * config buildid dir to symfs/.debug 2659 */ 2660 ret = asprintf(&bf, "%s/%s", symbol_conf.symfs, ".debug"); 2661 if (ret < 0) 2662 return -ENOMEM; 2663 2664 set_buildid_dir(bf); 2665 2666 free(bf); 2667 return 0; 2668 } 2669 2670 /* 2671 * Checks that user supplied symbol kernel files are accessible because 2672 * the default mechanism for accessing elf files fails silently. i.e. if 2673 * debug syms for a build ID aren't found perf carries on normally. When 2674 * they are user supplied we should assume that the user doesn't want to 2675 * silently fail. 2676 */ 2677 int symbol__validate_sym_arguments(void) 2678 { 2679 if (symbol_conf.vmlinux_name && 2680 access(symbol_conf.vmlinux_name, R_OK)) { 2681 pr_err("Invalid file: %s\n", symbol_conf.vmlinux_name); 2682 return -EINVAL; 2683 } 2684 if (symbol_conf.kallsyms_name && 2685 access(symbol_conf.kallsyms_name, R_OK)) { 2686 pr_err("Invalid file: %s\n", symbol_conf.kallsyms_name); 2687 return -EINVAL; 2688 } 2689 return 0; 2690 } 2691 2692 static bool want_demangle(bool is_kernel_sym) 2693 { 2694 return is_kernel_sym ? symbol_conf.demangle_kernel : symbol_conf.demangle; 2695 } 2696 2697 /* 2698 * Demangle C++ function signature, typically replaced by demangle-cxx.cpp 2699 * version. 2700 */ 2701 #ifndef HAVE_CXA_DEMANGLE_SUPPORT 2702 char *cxx_demangle_sym(const char *str __maybe_unused, bool params __maybe_unused, 2703 bool modifiers __maybe_unused) 2704 { 2705 #ifdef HAVE_LIBBFD_SUPPORT 2706 int flags = (params ? DMGL_PARAMS : 0) | (modifiers ? DMGL_ANSI : 0); 2707 2708 return bfd_demangle(NULL, str, flags); 2709 #elif defined(HAVE_CPLUS_DEMANGLE_SUPPORT) 2710 int flags = (params ? DMGL_PARAMS : 0) | (modifiers ? DMGL_ANSI : 0); 2711 2712 return cplus_demangle(str, flags); 2713 #else 2714 return NULL; 2715 #endif 2716 } 2717 #endif /* !HAVE_CXA_DEMANGLE_SUPPORT */ 2718 2719 char *dso__demangle_sym(struct dso *dso, int kmodule, const char *elf_name) 2720 { 2721 struct demangle rust_demangle = { 2722 .style = DemangleStyleUnknown, 2723 }; 2724 char *demangled = NULL; 2725 2726 /* 2727 * We need to figure out if the object was created from C++ sources 2728 * DWARF DW_compile_unit has this, but we don't always have access 2729 * to it... 2730 */ 2731 if (!want_demangle((dso && dso__kernel(dso)) || kmodule)) 2732 return demangled; 2733 2734 rust_demangle_demangle(elf_name, &rust_demangle); 2735 if (rust_demangle_is_known(&rust_demangle)) { 2736 /* A rust mangled name. */ 2737 if (rust_demangle.mangled_len == 0) 2738 return demangled; 2739 2740 for (size_t buf_len = roundup_pow_of_two(rust_demangle.mangled_len * 2); 2741 buf_len < 1024 * 1024; buf_len += 32) { 2742 char *tmp = realloc(demangled, buf_len); 2743 2744 if (!tmp) { 2745 /* Failure to grow output buffer, return what is there. */ 2746 return demangled; 2747 } 2748 demangled = tmp; 2749 if (rust_demangle_display_demangle(&rust_demangle, demangled, buf_len, 2750 /*alternate=*/true) == OverflowOk) 2751 return demangled; 2752 } 2753 /* Buffer exceeded sensible bounds, return what is there. */ 2754 return demangled; 2755 } 2756 2757 demangled = cxx_demangle_sym(elf_name, verbose > 0, verbose > 0); 2758 if (demangled) 2759 return demangled; 2760 2761 demangled = ocaml_demangle_sym(elf_name); 2762 if (demangled) 2763 return demangled; 2764 2765 return java_demangle_sym(elf_name, JAVA_DEMANGLE_NORET); 2766 } 2767