1 /* Generate assembler source containing symbol information 2 * 3 * Copyright 2002 by Kai Germaschewski 4 * 5 * This software may be used and distributed according to the terms 6 * of the GNU General Public License, incorporated herein by reference. 7 * 8 * Usage: nm -n vmlinux | scripts/kallsyms [--all-symbols] > symbols.S 9 * 10 * Table compression uses all the unused char codes on the symbols and 11 * maps these to the most used substrings (tokens). For instance, it might 12 * map char code 0xF7 to represent "write_" and then in every symbol where 13 * "write_" appears it can be replaced by 0xF7, saving 5 bytes. 14 * The used codes themselves are also placed in the table so that the 15 * decompresion can work without "special cases". 16 * Applied to kernel symbols, this usually produces a compression ratio 17 * of about 50%. 18 * 19 */ 20 21 #include <stdbool.h> 22 #include <stdio.h> 23 #include <stdlib.h> 24 #include <string.h> 25 #include <ctype.h> 26 #include <limits.h> 27 28 #define ARRAY_SIZE(arr) (sizeof(arr) / sizeof(arr[0])) 29 30 #define KSYM_NAME_LEN 128 31 32 struct sym_entry { 33 unsigned long long addr; 34 unsigned int len; 35 unsigned int start_pos; 36 unsigned int percpu_absolute; 37 unsigned char sym[]; 38 }; 39 40 struct addr_range { 41 const char *start_sym, *end_sym; 42 unsigned long long start, end; 43 }; 44 45 static unsigned long long _text; 46 static unsigned long long relative_base; 47 static struct addr_range text_ranges[] = { 48 { "_stext", "_etext" }, 49 { "_sinittext", "_einittext" }, 50 }; 51 #define text_range_text (&text_ranges[0]) 52 #define text_range_inittext (&text_ranges[1]) 53 54 static struct addr_range percpu_range = { 55 "__per_cpu_start", "__per_cpu_end", -1ULL, 0 56 }; 57 58 static struct sym_entry **table; 59 static unsigned int table_size, table_cnt; 60 static int all_symbols; 61 static int absolute_percpu; 62 static int base_relative; 63 64 static int token_profit[0x10000]; 65 66 /* the table that holds the result of the compression */ 67 static unsigned char best_table[256][2]; 68 static unsigned char best_table_len[256]; 69 70 71 static void usage(void) 72 { 73 fprintf(stderr, "Usage: kallsyms [--all-symbols] " 74 "[--base-relative] < in.map > out.S\n"); 75 exit(1); 76 } 77 78 static char *sym_name(const struct sym_entry *s) 79 { 80 return (char *)s->sym + 1; 81 } 82 83 static bool is_ignored_symbol(const char *name, char type) 84 { 85 /* Symbol names that exactly match to the following are ignored.*/ 86 static const char * const ignored_symbols[] = { 87 /* 88 * Symbols which vary between passes. Passes 1 and 2 must have 89 * identical symbol lists. The kallsyms_* symbols below are 90 * only added after pass 1, they would be included in pass 2 91 * when --all-symbols is specified so exclude them to get a 92 * stable symbol list. 93 */ 94 "kallsyms_addresses", 95 "kallsyms_offsets", 96 "kallsyms_relative_base", 97 "kallsyms_num_syms", 98 "kallsyms_names", 99 "kallsyms_markers", 100 "kallsyms_token_table", 101 "kallsyms_token_index", 102 /* Exclude linker generated symbols which vary between passes */ 103 "_SDA_BASE_", /* ppc */ 104 "_SDA2_BASE_", /* ppc */ 105 NULL 106 }; 107 108 /* Symbol names that begin with the following are ignored.*/ 109 static const char * const ignored_prefixes[] = { 110 "$", /* local symbols for ARM, MIPS, etc. */ 111 ".L", /* local labels, .LBB,.Ltmpxxx,.L__unnamed_xx,.LASANPC, etc. */ 112 "__crc_", /* modversions */ 113 "__efistub_", /* arm64 EFI stub namespace */ 114 "__kvm_nvhe_", /* arm64 non-VHE KVM namespace */ 115 "__AArch64ADRPThunk_", /* arm64 lld */ 116 "__ARMV5PILongThunk_", /* arm lld */ 117 "__ARMV7PILongThunk_", 118 "__ThumbV7PILongThunk_", 119 "__LA25Thunk_", /* mips lld */ 120 "__microLA25Thunk_", 121 NULL 122 }; 123 124 /* Symbol names that end with the following are ignored.*/ 125 static const char * const ignored_suffixes[] = { 126 "_from_arm", /* arm */ 127 "_from_thumb", /* arm */ 128 "_veneer", /* arm */ 129 NULL 130 }; 131 132 /* Symbol names that contain the following are ignored.*/ 133 static const char * const ignored_matches[] = { 134 ".long_branch.", /* ppc stub */ 135 ".plt_branch.", /* ppc stub */ 136 NULL 137 }; 138 139 const char * const *p; 140 141 for (p = ignored_symbols; *p; p++) 142 if (!strcmp(name, *p)) 143 return true; 144 145 for (p = ignored_prefixes; *p; p++) 146 if (!strncmp(name, *p, strlen(*p))) 147 return true; 148 149 for (p = ignored_suffixes; *p; p++) { 150 int l = strlen(name) - strlen(*p); 151 152 if (l >= 0 && !strcmp(name + l, *p)) 153 return true; 154 } 155 156 for (p = ignored_matches; *p; p++) { 157 if (strstr(name, *p)) 158 return true; 159 } 160 161 if (type == 'U' || type == 'u') 162 return true; 163 /* exclude debugging symbols */ 164 if (type == 'N' || type == 'n') 165 return true; 166 167 if (toupper(type) == 'A') { 168 /* Keep these useful absolute symbols */ 169 if (strcmp(name, "__kernel_syscall_via_break") && 170 strcmp(name, "__kernel_syscall_via_epc") && 171 strcmp(name, "__kernel_sigtramp") && 172 strcmp(name, "__gp")) 173 return true; 174 } 175 176 return false; 177 } 178 179 static void check_symbol_range(const char *sym, unsigned long long addr, 180 struct addr_range *ranges, int entries) 181 { 182 size_t i; 183 struct addr_range *ar; 184 185 for (i = 0; i < entries; ++i) { 186 ar = &ranges[i]; 187 188 if (strcmp(sym, ar->start_sym) == 0) { 189 ar->start = addr; 190 return; 191 } else if (strcmp(sym, ar->end_sym) == 0) { 192 ar->end = addr; 193 return; 194 } 195 } 196 } 197 198 static struct sym_entry *read_symbol(FILE *in) 199 { 200 char name[500], type; 201 unsigned long long addr; 202 unsigned int len; 203 struct sym_entry *sym; 204 int rc; 205 206 rc = fscanf(in, "%llx %c %499s\n", &addr, &type, name); 207 if (rc != 3) { 208 if (rc != EOF && fgets(name, 500, in) == NULL) 209 fprintf(stderr, "Read error or end of file.\n"); 210 return NULL; 211 } 212 if (strlen(name) >= KSYM_NAME_LEN) { 213 fprintf(stderr, "Symbol %s too long for kallsyms (%zu >= %d).\n" 214 "Please increase KSYM_NAME_LEN both in kernel and kallsyms.c\n", 215 name, strlen(name), KSYM_NAME_LEN); 216 return NULL; 217 } 218 219 if (strcmp(name, "_text") == 0) 220 _text = addr; 221 222 /* Ignore most absolute/undefined (?) symbols. */ 223 if (is_ignored_symbol(name, type)) 224 return NULL; 225 226 check_symbol_range(name, addr, text_ranges, ARRAY_SIZE(text_ranges)); 227 check_symbol_range(name, addr, &percpu_range, 1); 228 229 /* include the type field in the symbol name, so that it gets 230 * compressed together */ 231 232 len = strlen(name) + 1; 233 234 sym = malloc(sizeof(*sym) + len + 1); 235 if (!sym) { 236 fprintf(stderr, "kallsyms failure: " 237 "unable to allocate required amount of memory\n"); 238 exit(EXIT_FAILURE); 239 } 240 sym->addr = addr; 241 sym->len = len; 242 sym->sym[0] = type; 243 strcpy(sym_name(sym), name); 244 sym->percpu_absolute = 0; 245 246 return sym; 247 } 248 249 static int symbol_in_range(const struct sym_entry *s, 250 const struct addr_range *ranges, int entries) 251 { 252 size_t i; 253 const struct addr_range *ar; 254 255 for (i = 0; i < entries; ++i) { 256 ar = &ranges[i]; 257 258 if (s->addr >= ar->start && s->addr <= ar->end) 259 return 1; 260 } 261 262 return 0; 263 } 264 265 static int symbol_valid(const struct sym_entry *s) 266 { 267 const char *name = sym_name(s); 268 269 /* if --all-symbols is not specified, then symbols outside the text 270 * and inittext sections are discarded */ 271 if (!all_symbols) { 272 if (symbol_in_range(s, text_ranges, 273 ARRAY_SIZE(text_ranges)) == 0) 274 return 0; 275 /* Corner case. Discard any symbols with the same value as 276 * _etext _einittext; they can move between pass 1 and 2 when 277 * the kallsyms data are added. If these symbols move then 278 * they may get dropped in pass 2, which breaks the kallsyms 279 * rules. 280 */ 281 if ((s->addr == text_range_text->end && 282 strcmp(name, text_range_text->end_sym)) || 283 (s->addr == text_range_inittext->end && 284 strcmp(name, text_range_inittext->end_sym))) 285 return 0; 286 } 287 288 return 1; 289 } 290 291 /* remove all the invalid symbols from the table */ 292 static void shrink_table(void) 293 { 294 unsigned int i, pos; 295 296 pos = 0; 297 for (i = 0; i < table_cnt; i++) { 298 if (symbol_valid(table[i])) { 299 if (pos != i) 300 table[pos] = table[i]; 301 pos++; 302 } else { 303 free(table[i]); 304 } 305 } 306 table_cnt = pos; 307 308 /* When valid symbol is not registered, exit to error */ 309 if (!table_cnt) { 310 fprintf(stderr, "No valid symbol.\n"); 311 exit(1); 312 } 313 } 314 315 static void read_map(FILE *in) 316 { 317 struct sym_entry *sym; 318 319 while (!feof(in)) { 320 sym = read_symbol(in); 321 if (!sym) 322 continue; 323 324 sym->start_pos = table_cnt; 325 326 if (table_cnt >= table_size) { 327 table_size += 10000; 328 table = realloc(table, sizeof(*table) * table_size); 329 if (!table) { 330 fprintf(stderr, "out of memory\n"); 331 exit (1); 332 } 333 } 334 335 table[table_cnt++] = sym; 336 } 337 } 338 339 static void output_label(const char *label) 340 { 341 printf(".globl %s\n", label); 342 printf("\tALGN\n"); 343 printf("%s:\n", label); 344 } 345 346 /* Provide proper symbols relocatability by their '_text' relativeness. */ 347 static void output_address(unsigned long long addr) 348 { 349 if (_text <= addr) 350 printf("\tPTR\t_text + %#llx\n", addr - _text); 351 else 352 printf("\tPTR\t_text - %#llx\n", _text - addr); 353 } 354 355 /* uncompress a compressed symbol. When this function is called, the best table 356 * might still be compressed itself, so the function needs to be recursive */ 357 static int expand_symbol(const unsigned char *data, int len, char *result) 358 { 359 int c, rlen, total=0; 360 361 while (len) { 362 c = *data; 363 /* if the table holds a single char that is the same as the one 364 * we are looking for, then end the search */ 365 if (best_table[c][0]==c && best_table_len[c]==1) { 366 *result++ = c; 367 total++; 368 } else { 369 /* if not, recurse and expand */ 370 rlen = expand_symbol(best_table[c], best_table_len[c], result); 371 total += rlen; 372 result += rlen; 373 } 374 data++; 375 len--; 376 } 377 *result=0; 378 379 return total; 380 } 381 382 static int symbol_absolute(const struct sym_entry *s) 383 { 384 return s->percpu_absolute; 385 } 386 387 static void write_src(void) 388 { 389 unsigned int i, k, off; 390 unsigned int best_idx[256]; 391 unsigned int *markers; 392 char buf[KSYM_NAME_LEN]; 393 394 printf("#include <asm/bitsperlong.h>\n"); 395 printf("#if BITS_PER_LONG == 64\n"); 396 printf("#define PTR .quad\n"); 397 printf("#define ALGN .balign 8\n"); 398 printf("#else\n"); 399 printf("#define PTR .long\n"); 400 printf("#define ALGN .balign 4\n"); 401 printf("#endif\n"); 402 403 printf("\t.section .rodata, \"a\"\n"); 404 405 if (!base_relative) 406 output_label("kallsyms_addresses"); 407 else 408 output_label("kallsyms_offsets"); 409 410 for (i = 0; i < table_cnt; i++) { 411 if (base_relative) { 412 /* 413 * Use the offset relative to the lowest value 414 * encountered of all relative symbols, and emit 415 * non-relocatable fixed offsets that will be fixed 416 * up at runtime. 417 */ 418 419 long long offset; 420 int overflow; 421 422 if (!absolute_percpu) { 423 offset = table[i]->addr - relative_base; 424 overflow = (offset < 0 || offset > UINT_MAX); 425 } else if (symbol_absolute(table[i])) { 426 offset = table[i]->addr; 427 overflow = (offset < 0 || offset > INT_MAX); 428 } else { 429 offset = relative_base - table[i]->addr - 1; 430 overflow = (offset < INT_MIN || offset >= 0); 431 } 432 if (overflow) { 433 fprintf(stderr, "kallsyms failure: " 434 "%s symbol value %#llx out of range in relative mode\n", 435 symbol_absolute(table[i]) ? "absolute" : "relative", 436 table[i]->addr); 437 exit(EXIT_FAILURE); 438 } 439 printf("\t.long\t%#x\n", (int)offset); 440 } else if (!symbol_absolute(table[i])) { 441 output_address(table[i]->addr); 442 } else { 443 printf("\tPTR\t%#llx\n", table[i]->addr); 444 } 445 } 446 printf("\n"); 447 448 if (base_relative) { 449 output_label("kallsyms_relative_base"); 450 output_address(relative_base); 451 printf("\n"); 452 } 453 454 output_label("kallsyms_num_syms"); 455 printf("\t.long\t%u\n", table_cnt); 456 printf("\n"); 457 458 /* table of offset markers, that give the offset in the compressed stream 459 * every 256 symbols */ 460 markers = malloc(sizeof(unsigned int) * ((table_cnt + 255) / 256)); 461 if (!markers) { 462 fprintf(stderr, "kallsyms failure: " 463 "unable to allocate required memory\n"); 464 exit(EXIT_FAILURE); 465 } 466 467 output_label("kallsyms_names"); 468 off = 0; 469 for (i = 0; i < table_cnt; i++) { 470 if ((i & 0xFF) == 0) 471 markers[i >> 8] = off; 472 473 printf("\t.byte 0x%02x", table[i]->len); 474 for (k = 0; k < table[i]->len; k++) 475 printf(", 0x%02x", table[i]->sym[k]); 476 printf("\n"); 477 478 off += table[i]->len + 1; 479 } 480 printf("\n"); 481 482 output_label("kallsyms_markers"); 483 for (i = 0; i < ((table_cnt + 255) >> 8); i++) 484 printf("\t.long\t%u\n", markers[i]); 485 printf("\n"); 486 487 free(markers); 488 489 output_label("kallsyms_token_table"); 490 off = 0; 491 for (i = 0; i < 256; i++) { 492 best_idx[i] = off; 493 expand_symbol(best_table[i], best_table_len[i], buf); 494 printf("\t.asciz\t\"%s\"\n", buf); 495 off += strlen(buf) + 1; 496 } 497 printf("\n"); 498 499 output_label("kallsyms_token_index"); 500 for (i = 0; i < 256; i++) 501 printf("\t.short\t%d\n", best_idx[i]); 502 printf("\n"); 503 } 504 505 506 /* table lookup compression functions */ 507 508 /* count all the possible tokens in a symbol */ 509 static void learn_symbol(const unsigned char *symbol, int len) 510 { 511 int i; 512 513 for (i = 0; i < len - 1; i++) 514 token_profit[ symbol[i] + (symbol[i + 1] << 8) ]++; 515 } 516 517 /* decrease the count for all the possible tokens in a symbol */ 518 static void forget_symbol(const unsigned char *symbol, int len) 519 { 520 int i; 521 522 for (i = 0; i < len - 1; i++) 523 token_profit[ symbol[i] + (symbol[i + 1] << 8) ]--; 524 } 525 526 /* do the initial token count */ 527 static void build_initial_tok_table(void) 528 { 529 unsigned int i; 530 531 for (i = 0; i < table_cnt; i++) 532 learn_symbol(table[i]->sym, table[i]->len); 533 } 534 535 static unsigned char *find_token(unsigned char *str, int len, 536 const unsigned char *token) 537 { 538 int i; 539 540 for (i = 0; i < len - 1; i++) { 541 if (str[i] == token[0] && str[i+1] == token[1]) 542 return &str[i]; 543 } 544 return NULL; 545 } 546 547 /* replace a given token in all the valid symbols. Use the sampled symbols 548 * to update the counts */ 549 static void compress_symbols(const unsigned char *str, int idx) 550 { 551 unsigned int i, len, size; 552 unsigned char *p1, *p2; 553 554 for (i = 0; i < table_cnt; i++) { 555 556 len = table[i]->len; 557 p1 = table[i]->sym; 558 559 /* find the token on the symbol */ 560 p2 = find_token(p1, len, str); 561 if (!p2) continue; 562 563 /* decrease the counts for this symbol's tokens */ 564 forget_symbol(table[i]->sym, len); 565 566 size = len; 567 568 do { 569 *p2 = idx; 570 p2++; 571 size -= (p2 - p1); 572 memmove(p2, p2 + 1, size); 573 p1 = p2; 574 len--; 575 576 if (size < 2) break; 577 578 /* find the token on the symbol */ 579 p2 = find_token(p1, size, str); 580 581 } while (p2); 582 583 table[i]->len = len; 584 585 /* increase the counts for this symbol's new tokens */ 586 learn_symbol(table[i]->sym, len); 587 } 588 } 589 590 /* search the token with the maximum profit */ 591 static int find_best_token(void) 592 { 593 int i, best, bestprofit; 594 595 bestprofit=-10000; 596 best = 0; 597 598 for (i = 0; i < 0x10000; i++) { 599 if (token_profit[i] > bestprofit) { 600 best = i; 601 bestprofit = token_profit[i]; 602 } 603 } 604 return best; 605 } 606 607 /* this is the core of the algorithm: calculate the "best" table */ 608 static void optimize_result(void) 609 { 610 int i, best; 611 612 /* using the '\0' symbol last allows compress_symbols to use standard 613 * fast string functions */ 614 for (i = 255; i >= 0; i--) { 615 616 /* if this table slot is empty (it is not used by an actual 617 * original char code */ 618 if (!best_table_len[i]) { 619 620 /* find the token with the best profit value */ 621 best = find_best_token(); 622 if (token_profit[best] == 0) 623 break; 624 625 /* place it in the "best" table */ 626 best_table_len[i] = 2; 627 best_table[i][0] = best & 0xFF; 628 best_table[i][1] = (best >> 8) & 0xFF; 629 630 /* replace this token in all the valid symbols */ 631 compress_symbols(best_table[i], i); 632 } 633 } 634 } 635 636 /* start by placing the symbols that are actually used on the table */ 637 static void insert_real_symbols_in_table(void) 638 { 639 unsigned int i, j, c; 640 641 for (i = 0; i < table_cnt; i++) { 642 for (j = 0; j < table[i]->len; j++) { 643 c = table[i]->sym[j]; 644 best_table[c][0]=c; 645 best_table_len[c]=1; 646 } 647 } 648 } 649 650 static void optimize_token_table(void) 651 { 652 build_initial_tok_table(); 653 654 insert_real_symbols_in_table(); 655 656 optimize_result(); 657 } 658 659 /* guess for "linker script provide" symbol */ 660 static int may_be_linker_script_provide_symbol(const struct sym_entry *se) 661 { 662 const char *symbol = sym_name(se); 663 int len = se->len - 1; 664 665 if (len < 8) 666 return 0; 667 668 if (symbol[0] != '_' || symbol[1] != '_') 669 return 0; 670 671 /* __start_XXXXX */ 672 if (!memcmp(symbol + 2, "start_", 6)) 673 return 1; 674 675 /* __stop_XXXXX */ 676 if (!memcmp(symbol + 2, "stop_", 5)) 677 return 1; 678 679 /* __end_XXXXX */ 680 if (!memcmp(symbol + 2, "end_", 4)) 681 return 1; 682 683 /* __XXXXX_start */ 684 if (!memcmp(symbol + len - 6, "_start", 6)) 685 return 1; 686 687 /* __XXXXX_end */ 688 if (!memcmp(symbol + len - 4, "_end", 4)) 689 return 1; 690 691 return 0; 692 } 693 694 static int compare_symbols(const void *a, const void *b) 695 { 696 const struct sym_entry *sa = *(const struct sym_entry **)a; 697 const struct sym_entry *sb = *(const struct sym_entry **)b; 698 int wa, wb; 699 700 /* sort by address first */ 701 if (sa->addr > sb->addr) 702 return 1; 703 if (sa->addr < sb->addr) 704 return -1; 705 706 /* sort by "weakness" type */ 707 wa = (sa->sym[0] == 'w') || (sa->sym[0] == 'W'); 708 wb = (sb->sym[0] == 'w') || (sb->sym[0] == 'W'); 709 if (wa != wb) 710 return wa - wb; 711 712 /* sort by "linker script provide" type */ 713 wa = may_be_linker_script_provide_symbol(sa); 714 wb = may_be_linker_script_provide_symbol(sb); 715 if (wa != wb) 716 return wa - wb; 717 718 /* sort by the number of prefix underscores */ 719 wa = strspn(sym_name(sa), "_"); 720 wb = strspn(sym_name(sb), "_"); 721 if (wa != wb) 722 return wa - wb; 723 724 /* sort by initial order, so that other symbols are left undisturbed */ 725 return sa->start_pos - sb->start_pos; 726 } 727 728 static void sort_symbols(void) 729 { 730 qsort(table, table_cnt, sizeof(table[0]), compare_symbols); 731 } 732 733 static void make_percpus_absolute(void) 734 { 735 unsigned int i; 736 737 for (i = 0; i < table_cnt; i++) 738 if (symbol_in_range(table[i], &percpu_range, 1)) { 739 /* 740 * Keep the 'A' override for percpu symbols to 741 * ensure consistent behavior compared to older 742 * versions of this tool. 743 */ 744 table[i]->sym[0] = 'A'; 745 table[i]->percpu_absolute = 1; 746 } 747 } 748 749 /* find the minimum non-absolute symbol address */ 750 static void record_relative_base(void) 751 { 752 unsigned int i; 753 754 for (i = 0; i < table_cnt; i++) 755 if (!symbol_absolute(table[i])) { 756 /* 757 * The table is sorted by address. 758 * Take the first non-absolute symbol value. 759 */ 760 relative_base = table[i]->addr; 761 return; 762 } 763 } 764 765 int main(int argc, char **argv) 766 { 767 if (argc >= 2) { 768 int i; 769 for (i = 1; i < argc; i++) { 770 if(strcmp(argv[i], "--all-symbols") == 0) 771 all_symbols = 1; 772 else if (strcmp(argv[i], "--absolute-percpu") == 0) 773 absolute_percpu = 1; 774 else if (strcmp(argv[i], "--base-relative") == 0) 775 base_relative = 1; 776 else 777 usage(); 778 } 779 } else if (argc != 1) 780 usage(); 781 782 read_map(stdin); 783 shrink_table(); 784 if (absolute_percpu) 785 make_percpus_absolute(); 786 sort_symbols(); 787 if (base_relative) 788 record_relative_base(); 789 optimize_token_table(); 790 write_src(); 791 792 return 0; 793 } 794