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