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 static const char * const ignored_symbols[] = { 86 /* 87 * Symbols which vary between passes. Passes 1 and 2 must have 88 * identical symbol lists. The kallsyms_* symbols below are 89 * only added after pass 1, they would be included in pass 2 90 * when --all-symbols is specified so exclude them to get a 91 * stable symbol list. 92 */ 93 "kallsyms_addresses", 94 "kallsyms_offsets", 95 "kallsyms_relative_base", 96 "kallsyms_num_syms", 97 "kallsyms_names", 98 "kallsyms_markers", 99 "kallsyms_token_table", 100 "kallsyms_token_index", 101 /* Exclude linker generated symbols which vary between passes */ 102 "_SDA_BASE_", /* ppc */ 103 "_SDA2_BASE_", /* ppc */ 104 NULL 105 }; 106 107 static const char * const ignored_prefixes[] = { 108 "$", /* local symbols for ARM, MIPS, etc. */ 109 ".LASANPC", /* s390 kasan local symbols */ 110 "__crc_", /* modversions */ 111 "__efistub_", /* arm64 EFI stub namespace */ 112 "__kvm_nvhe_", /* arm64 non-VHE KVM namespace */ 113 NULL 114 }; 115 116 static const char * const ignored_suffixes[] = { 117 "_from_arm", /* arm */ 118 "_from_thumb", /* arm */ 119 "_veneer", /* arm */ 120 NULL 121 }; 122 123 const char * const *p; 124 125 /* Exclude symbols which vary between passes. */ 126 for (p = ignored_symbols; *p; p++) 127 if (!strcmp(name, *p)) 128 return true; 129 130 for (p = ignored_prefixes; *p; p++) 131 if (!strncmp(name, *p, strlen(*p))) 132 return true; 133 134 for (p = ignored_suffixes; *p; p++) { 135 int l = strlen(name) - strlen(*p); 136 137 if (l >= 0 && !strcmp(name + l, *p)) 138 return true; 139 } 140 141 if (type == 'U' || type == 'u') 142 return true; 143 /* exclude debugging symbols */ 144 if (type == 'N' || type == 'n') 145 return true; 146 147 if (toupper(type) == 'A') { 148 /* Keep these useful absolute symbols */ 149 if (strcmp(name, "__kernel_syscall_via_break") && 150 strcmp(name, "__kernel_syscall_via_epc") && 151 strcmp(name, "__kernel_sigtramp") && 152 strcmp(name, "__gp")) 153 return true; 154 } 155 156 return false; 157 } 158 159 static void check_symbol_range(const char *sym, unsigned long long addr, 160 struct addr_range *ranges, int entries) 161 { 162 size_t i; 163 struct addr_range *ar; 164 165 for (i = 0; i < entries; ++i) { 166 ar = &ranges[i]; 167 168 if (strcmp(sym, ar->start_sym) == 0) { 169 ar->start = addr; 170 return; 171 } else if (strcmp(sym, ar->end_sym) == 0) { 172 ar->end = addr; 173 return; 174 } 175 } 176 } 177 178 static struct sym_entry *read_symbol(FILE *in) 179 { 180 char name[500], type; 181 unsigned long long addr; 182 unsigned int len; 183 struct sym_entry *sym; 184 int rc; 185 186 rc = fscanf(in, "%llx %c %499s\n", &addr, &type, name); 187 if (rc != 3) { 188 if (rc != EOF && fgets(name, 500, in) == NULL) 189 fprintf(stderr, "Read error or end of file.\n"); 190 return NULL; 191 } 192 if (strlen(name) >= KSYM_NAME_LEN) { 193 fprintf(stderr, "Symbol %s too long for kallsyms (%zu >= %d).\n" 194 "Please increase KSYM_NAME_LEN both in kernel and kallsyms.c\n", 195 name, strlen(name), KSYM_NAME_LEN); 196 return NULL; 197 } 198 199 if (strcmp(name, "_text") == 0) 200 _text = addr; 201 202 /* Ignore most absolute/undefined (?) symbols. */ 203 if (is_ignored_symbol(name, type)) 204 return NULL; 205 206 check_symbol_range(name, addr, text_ranges, ARRAY_SIZE(text_ranges)); 207 check_symbol_range(name, addr, &percpu_range, 1); 208 209 /* include the type field in the symbol name, so that it gets 210 * compressed together */ 211 212 len = strlen(name) + 1; 213 214 sym = malloc(sizeof(*sym) + len + 1); 215 if (!sym) { 216 fprintf(stderr, "kallsyms failure: " 217 "unable to allocate required amount of memory\n"); 218 exit(EXIT_FAILURE); 219 } 220 sym->addr = addr; 221 sym->len = len; 222 sym->sym[0] = type; 223 strcpy(sym_name(sym), name); 224 sym->percpu_absolute = 0; 225 226 return sym; 227 } 228 229 static int symbol_in_range(const struct sym_entry *s, 230 const struct addr_range *ranges, int entries) 231 { 232 size_t i; 233 const struct addr_range *ar; 234 235 for (i = 0; i < entries; ++i) { 236 ar = &ranges[i]; 237 238 if (s->addr >= ar->start && s->addr <= ar->end) 239 return 1; 240 } 241 242 return 0; 243 } 244 245 static int symbol_valid(const struct sym_entry *s) 246 { 247 const char *name = sym_name(s); 248 249 /* if --all-symbols is not specified, then symbols outside the text 250 * and inittext sections are discarded */ 251 if (!all_symbols) { 252 if (symbol_in_range(s, text_ranges, 253 ARRAY_SIZE(text_ranges)) == 0) 254 return 0; 255 /* Corner case. Discard any symbols with the same value as 256 * _etext _einittext; they can move between pass 1 and 2 when 257 * the kallsyms data are added. If these symbols move then 258 * they may get dropped in pass 2, which breaks the kallsyms 259 * rules. 260 */ 261 if ((s->addr == text_range_text->end && 262 strcmp(name, text_range_text->end_sym)) || 263 (s->addr == text_range_inittext->end && 264 strcmp(name, text_range_inittext->end_sym))) 265 return 0; 266 } 267 268 return 1; 269 } 270 271 /* remove all the invalid symbols from the table */ 272 static void shrink_table(void) 273 { 274 unsigned int i, pos; 275 276 pos = 0; 277 for (i = 0; i < table_cnt; i++) { 278 if (symbol_valid(table[i])) { 279 if (pos != i) 280 table[pos] = table[i]; 281 pos++; 282 } else { 283 free(table[i]); 284 } 285 } 286 table_cnt = pos; 287 288 /* When valid symbol is not registered, exit to error */ 289 if (!table_cnt) { 290 fprintf(stderr, "No valid symbol.\n"); 291 exit(1); 292 } 293 } 294 295 static void read_map(FILE *in) 296 { 297 struct sym_entry *sym; 298 299 while (!feof(in)) { 300 sym = read_symbol(in); 301 if (!sym) 302 continue; 303 304 sym->start_pos = table_cnt; 305 306 if (table_cnt >= table_size) { 307 table_size += 10000; 308 table = realloc(table, sizeof(*table) * table_size); 309 if (!table) { 310 fprintf(stderr, "out of memory\n"); 311 exit (1); 312 } 313 } 314 315 table[table_cnt++] = sym; 316 } 317 } 318 319 static void output_label(const char *label) 320 { 321 printf(".globl %s\n", label); 322 printf("\tALGN\n"); 323 printf("%s:\n", label); 324 } 325 326 /* Provide proper symbols relocatability by their '_text' relativeness. */ 327 static void output_address(unsigned long long addr) 328 { 329 if (_text <= addr) 330 printf("\tPTR\t_text + %#llx\n", addr - _text); 331 else 332 printf("\tPTR\t_text - %#llx\n", _text - addr); 333 } 334 335 /* uncompress a compressed symbol. When this function is called, the best table 336 * might still be compressed itself, so the function needs to be recursive */ 337 static int expand_symbol(const unsigned char *data, int len, char *result) 338 { 339 int c, rlen, total=0; 340 341 while (len) { 342 c = *data; 343 /* if the table holds a single char that is the same as the one 344 * we are looking for, then end the search */ 345 if (best_table[c][0]==c && best_table_len[c]==1) { 346 *result++ = c; 347 total++; 348 } else { 349 /* if not, recurse and expand */ 350 rlen = expand_symbol(best_table[c], best_table_len[c], result); 351 total += rlen; 352 result += rlen; 353 } 354 data++; 355 len--; 356 } 357 *result=0; 358 359 return total; 360 } 361 362 static int symbol_absolute(const struct sym_entry *s) 363 { 364 return s->percpu_absolute; 365 } 366 367 static void write_src(void) 368 { 369 unsigned int i, k, off; 370 unsigned int best_idx[256]; 371 unsigned int *markers; 372 char buf[KSYM_NAME_LEN]; 373 374 printf("#include <asm/bitsperlong.h>\n"); 375 printf("#if BITS_PER_LONG == 64\n"); 376 printf("#define PTR .quad\n"); 377 printf("#define ALGN .balign 8\n"); 378 printf("#else\n"); 379 printf("#define PTR .long\n"); 380 printf("#define ALGN .balign 4\n"); 381 printf("#endif\n"); 382 383 printf("\t.section .rodata, \"a\"\n"); 384 385 if (!base_relative) 386 output_label("kallsyms_addresses"); 387 else 388 output_label("kallsyms_offsets"); 389 390 for (i = 0; i < table_cnt; i++) { 391 if (base_relative) { 392 /* 393 * Use the offset relative to the lowest value 394 * encountered of all relative symbols, and emit 395 * non-relocatable fixed offsets that will be fixed 396 * up at runtime. 397 */ 398 399 long long offset; 400 int overflow; 401 402 if (!absolute_percpu) { 403 offset = table[i]->addr - relative_base; 404 overflow = (offset < 0 || offset > UINT_MAX); 405 } else if (symbol_absolute(table[i])) { 406 offset = table[i]->addr; 407 overflow = (offset < 0 || offset > INT_MAX); 408 } else { 409 offset = relative_base - table[i]->addr - 1; 410 overflow = (offset < INT_MIN || offset >= 0); 411 } 412 if (overflow) { 413 fprintf(stderr, "kallsyms failure: " 414 "%s symbol value %#llx out of range in relative mode\n", 415 symbol_absolute(table[i]) ? "absolute" : "relative", 416 table[i]->addr); 417 exit(EXIT_FAILURE); 418 } 419 printf("\t.long\t%#x\n", (int)offset); 420 } else if (!symbol_absolute(table[i])) { 421 output_address(table[i]->addr); 422 } else { 423 printf("\tPTR\t%#llx\n", table[i]->addr); 424 } 425 } 426 printf("\n"); 427 428 if (base_relative) { 429 output_label("kallsyms_relative_base"); 430 output_address(relative_base); 431 printf("\n"); 432 } 433 434 output_label("kallsyms_num_syms"); 435 printf("\t.long\t%u\n", table_cnt); 436 printf("\n"); 437 438 /* table of offset markers, that give the offset in the compressed stream 439 * every 256 symbols */ 440 markers = malloc(sizeof(unsigned int) * ((table_cnt + 255) / 256)); 441 if (!markers) { 442 fprintf(stderr, "kallsyms failure: " 443 "unable to allocate required memory\n"); 444 exit(EXIT_FAILURE); 445 } 446 447 output_label("kallsyms_names"); 448 off = 0; 449 for (i = 0; i < table_cnt; i++) { 450 if ((i & 0xFF) == 0) 451 markers[i >> 8] = off; 452 453 printf("\t.byte 0x%02x", table[i]->len); 454 for (k = 0; k < table[i]->len; k++) 455 printf(", 0x%02x", table[i]->sym[k]); 456 printf("\n"); 457 458 off += table[i]->len + 1; 459 } 460 printf("\n"); 461 462 output_label("kallsyms_markers"); 463 for (i = 0; i < ((table_cnt + 255) >> 8); i++) 464 printf("\t.long\t%u\n", markers[i]); 465 printf("\n"); 466 467 free(markers); 468 469 output_label("kallsyms_token_table"); 470 off = 0; 471 for (i = 0; i < 256; i++) { 472 best_idx[i] = off; 473 expand_symbol(best_table[i], best_table_len[i], buf); 474 printf("\t.asciz\t\"%s\"\n", buf); 475 off += strlen(buf) + 1; 476 } 477 printf("\n"); 478 479 output_label("kallsyms_token_index"); 480 for (i = 0; i < 256; i++) 481 printf("\t.short\t%d\n", best_idx[i]); 482 printf("\n"); 483 } 484 485 486 /* table lookup compression functions */ 487 488 /* count all the possible tokens in a symbol */ 489 static void learn_symbol(const unsigned char *symbol, int len) 490 { 491 int i; 492 493 for (i = 0; i < len - 1; i++) 494 token_profit[ symbol[i] + (symbol[i + 1] << 8) ]++; 495 } 496 497 /* decrease the count for all the possible tokens in a symbol */ 498 static void forget_symbol(const unsigned char *symbol, int len) 499 { 500 int i; 501 502 for (i = 0; i < len - 1; i++) 503 token_profit[ symbol[i] + (symbol[i + 1] << 8) ]--; 504 } 505 506 /* do the initial token count */ 507 static void build_initial_tok_table(void) 508 { 509 unsigned int i; 510 511 for (i = 0; i < table_cnt; i++) 512 learn_symbol(table[i]->sym, table[i]->len); 513 } 514 515 static unsigned char *find_token(unsigned char *str, int len, 516 const unsigned char *token) 517 { 518 int i; 519 520 for (i = 0; i < len - 1; i++) { 521 if (str[i] == token[0] && str[i+1] == token[1]) 522 return &str[i]; 523 } 524 return NULL; 525 } 526 527 /* replace a given token in all the valid symbols. Use the sampled symbols 528 * to update the counts */ 529 static void compress_symbols(const unsigned char *str, int idx) 530 { 531 unsigned int i, len, size; 532 unsigned char *p1, *p2; 533 534 for (i = 0; i < table_cnt; i++) { 535 536 len = table[i]->len; 537 p1 = table[i]->sym; 538 539 /* find the token on the symbol */ 540 p2 = find_token(p1, len, str); 541 if (!p2) continue; 542 543 /* decrease the counts for this symbol's tokens */ 544 forget_symbol(table[i]->sym, len); 545 546 size = len; 547 548 do { 549 *p2 = idx; 550 p2++; 551 size -= (p2 - p1); 552 memmove(p2, p2 + 1, size); 553 p1 = p2; 554 len--; 555 556 if (size < 2) break; 557 558 /* find the token on the symbol */ 559 p2 = find_token(p1, size, str); 560 561 } while (p2); 562 563 table[i]->len = len; 564 565 /* increase the counts for this symbol's new tokens */ 566 learn_symbol(table[i]->sym, len); 567 } 568 } 569 570 /* search the token with the maximum profit */ 571 static int find_best_token(void) 572 { 573 int i, best, bestprofit; 574 575 bestprofit=-10000; 576 best = 0; 577 578 for (i = 0; i < 0x10000; i++) { 579 if (token_profit[i] > bestprofit) { 580 best = i; 581 bestprofit = token_profit[i]; 582 } 583 } 584 return best; 585 } 586 587 /* this is the core of the algorithm: calculate the "best" table */ 588 static void optimize_result(void) 589 { 590 int i, best; 591 592 /* using the '\0' symbol last allows compress_symbols to use standard 593 * fast string functions */ 594 for (i = 255; i >= 0; i--) { 595 596 /* if this table slot is empty (it is not used by an actual 597 * original char code */ 598 if (!best_table_len[i]) { 599 600 /* find the token with the best profit value */ 601 best = find_best_token(); 602 if (token_profit[best] == 0) 603 break; 604 605 /* place it in the "best" table */ 606 best_table_len[i] = 2; 607 best_table[i][0] = best & 0xFF; 608 best_table[i][1] = (best >> 8) & 0xFF; 609 610 /* replace this token in all the valid symbols */ 611 compress_symbols(best_table[i], i); 612 } 613 } 614 } 615 616 /* start by placing the symbols that are actually used on the table */ 617 static void insert_real_symbols_in_table(void) 618 { 619 unsigned int i, j, c; 620 621 for (i = 0; i < table_cnt; i++) { 622 for (j = 0; j < table[i]->len; j++) { 623 c = table[i]->sym[j]; 624 best_table[c][0]=c; 625 best_table_len[c]=1; 626 } 627 } 628 } 629 630 static void optimize_token_table(void) 631 { 632 build_initial_tok_table(); 633 634 insert_real_symbols_in_table(); 635 636 optimize_result(); 637 } 638 639 /* guess for "linker script provide" symbol */ 640 static int may_be_linker_script_provide_symbol(const struct sym_entry *se) 641 { 642 const char *symbol = sym_name(se); 643 int len = se->len - 1; 644 645 if (len < 8) 646 return 0; 647 648 if (symbol[0] != '_' || symbol[1] != '_') 649 return 0; 650 651 /* __start_XXXXX */ 652 if (!memcmp(symbol + 2, "start_", 6)) 653 return 1; 654 655 /* __stop_XXXXX */ 656 if (!memcmp(symbol + 2, "stop_", 5)) 657 return 1; 658 659 /* __end_XXXXX */ 660 if (!memcmp(symbol + 2, "end_", 4)) 661 return 1; 662 663 /* __XXXXX_start */ 664 if (!memcmp(symbol + len - 6, "_start", 6)) 665 return 1; 666 667 /* __XXXXX_end */ 668 if (!memcmp(symbol + len - 4, "_end", 4)) 669 return 1; 670 671 return 0; 672 } 673 674 static int compare_symbols(const void *a, const void *b) 675 { 676 const struct sym_entry *sa = *(const struct sym_entry **)a; 677 const struct sym_entry *sb = *(const struct sym_entry **)b; 678 int wa, wb; 679 680 /* sort by address first */ 681 if (sa->addr > sb->addr) 682 return 1; 683 if (sa->addr < sb->addr) 684 return -1; 685 686 /* sort by "weakness" type */ 687 wa = (sa->sym[0] == 'w') || (sa->sym[0] == 'W'); 688 wb = (sb->sym[0] == 'w') || (sb->sym[0] == 'W'); 689 if (wa != wb) 690 return wa - wb; 691 692 /* sort by "linker script provide" type */ 693 wa = may_be_linker_script_provide_symbol(sa); 694 wb = may_be_linker_script_provide_symbol(sb); 695 if (wa != wb) 696 return wa - wb; 697 698 /* sort by the number of prefix underscores */ 699 wa = strspn(sym_name(sa), "_"); 700 wb = strspn(sym_name(sb), "_"); 701 if (wa != wb) 702 return wa - wb; 703 704 /* sort by initial order, so that other symbols are left undisturbed */ 705 return sa->start_pos - sb->start_pos; 706 } 707 708 static void sort_symbols(void) 709 { 710 qsort(table, table_cnt, sizeof(table[0]), compare_symbols); 711 } 712 713 static void make_percpus_absolute(void) 714 { 715 unsigned int i; 716 717 for (i = 0; i < table_cnt; i++) 718 if (symbol_in_range(table[i], &percpu_range, 1)) { 719 /* 720 * Keep the 'A' override for percpu symbols to 721 * ensure consistent behavior compared to older 722 * versions of this tool. 723 */ 724 table[i]->sym[0] = 'A'; 725 table[i]->percpu_absolute = 1; 726 } 727 } 728 729 /* find the minimum non-absolute symbol address */ 730 static void record_relative_base(void) 731 { 732 unsigned int i; 733 734 for (i = 0; i < table_cnt; i++) 735 if (!symbol_absolute(table[i])) { 736 /* 737 * The table is sorted by address. 738 * Take the first non-absolute symbol value. 739 */ 740 relative_base = table[i]->addr; 741 return; 742 } 743 } 744 745 int main(int argc, char **argv) 746 { 747 if (argc >= 2) { 748 int i; 749 for (i = 1; i < argc; i++) { 750 if(strcmp(argv[i], "--all-symbols") == 0) 751 all_symbols = 1; 752 else if (strcmp(argv[i], "--absolute-percpu") == 0) 753 absolute_percpu = 1; 754 else if (strcmp(argv[i], "--base-relative") == 0) 755 base_relative = 1; 756 else 757 usage(); 758 } 759 } else if (argc != 1) 760 usage(); 761 762 read_map(stdin); 763 shrink_table(); 764 if (absolute_percpu) 765 make_percpus_absolute(); 766 sort_symbols(); 767 if (base_relative) 768 record_relative_base(); 769 optimize_token_table(); 770 write_src(); 771 772 return 0; 773 } 774