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 /* uncompress a compressed symbol. When this function is called, the best table 314 * might still be compressed itself, so the function needs to be recursive */ 315 static int expand_symbol(const unsigned char *data, int len, char *result) 316 { 317 int c, rlen, total=0; 318 319 while (len) { 320 c = *data; 321 /* if the table holds a single char that is the same as the one 322 * we are looking for, then end the search */ 323 if (best_table[c][0]==c && best_table_len[c]==1) { 324 *result++ = c; 325 total++; 326 } else { 327 /* if not, recurse and expand */ 328 rlen = expand_symbol(best_table[c], best_table_len[c], result); 329 total += rlen; 330 result += rlen; 331 } 332 data++; 333 len--; 334 } 335 *result=0; 336 337 return total; 338 } 339 340 static int symbol_absolute(const struct sym_entry *s) 341 { 342 return s->percpu_absolute; 343 } 344 345 static void write_src(void) 346 { 347 unsigned int i, k, off; 348 unsigned int best_idx[256]; 349 unsigned int *markers; 350 char buf[KSYM_NAME_LEN]; 351 352 printf("#include <asm/bitsperlong.h>\n"); 353 printf("#if BITS_PER_LONG == 64\n"); 354 printf("#define PTR .quad\n"); 355 printf("#define ALGN .balign 8\n"); 356 printf("#else\n"); 357 printf("#define PTR .long\n"); 358 printf("#define ALGN .balign 4\n"); 359 printf("#endif\n"); 360 361 printf("\t.section .rodata, \"a\"\n"); 362 363 /* Provide proper symbols relocatability by their relativeness 364 * to a fixed anchor point in the runtime image, either '_text' 365 * for absolute address tables, in which case the linker will 366 * emit the final addresses at build time. Otherwise, use the 367 * offset relative to the lowest value encountered of all relative 368 * symbols, and emit non-relocatable fixed offsets that will be fixed 369 * up at runtime. 370 * 371 * The symbol names cannot be used to construct normal symbol 372 * references as the list of symbols contains symbols that are 373 * declared static and are private to their .o files. This prevents 374 * .tmp_kallsyms.o or any other object from referencing them. 375 */ 376 if (!base_relative) 377 output_label("kallsyms_addresses"); 378 else 379 output_label("kallsyms_offsets"); 380 381 for (i = 0; i < table_cnt; i++) { 382 if (base_relative) { 383 long long offset; 384 int overflow; 385 386 if (!absolute_percpu) { 387 offset = table[i].addr - relative_base; 388 overflow = (offset < 0 || offset > UINT_MAX); 389 } else if (symbol_absolute(&table[i])) { 390 offset = table[i].addr; 391 overflow = (offset < 0 || offset > INT_MAX); 392 } else { 393 offset = relative_base - table[i].addr - 1; 394 overflow = (offset < INT_MIN || offset >= 0); 395 } 396 if (overflow) { 397 fprintf(stderr, "kallsyms failure: " 398 "%s symbol value %#llx out of range in relative mode\n", 399 symbol_absolute(&table[i]) ? "absolute" : "relative", 400 table[i].addr); 401 exit(EXIT_FAILURE); 402 } 403 printf("\t.long\t%#x\n", (int)offset); 404 } else if (!symbol_absolute(&table[i])) { 405 if (_text <= table[i].addr) 406 printf("\tPTR\t_text + %#llx\n", 407 table[i].addr - _text); 408 else 409 printf("\tPTR\t_text - %#llx\n", 410 _text - table[i].addr); 411 } else { 412 printf("\tPTR\t%#llx\n", table[i].addr); 413 } 414 } 415 printf("\n"); 416 417 if (base_relative) { 418 output_label("kallsyms_relative_base"); 419 printf("\tPTR\t_text - %#llx\n", _text - relative_base); 420 printf("\n"); 421 } 422 423 output_label("kallsyms_num_syms"); 424 printf("\t.long\t%u\n", table_cnt); 425 printf("\n"); 426 427 /* table of offset markers, that give the offset in the compressed stream 428 * every 256 symbols */ 429 markers = malloc(sizeof(unsigned int) * ((table_cnt + 255) / 256)); 430 if (!markers) { 431 fprintf(stderr, "kallsyms failure: " 432 "unable to allocate required memory\n"); 433 exit(EXIT_FAILURE); 434 } 435 436 output_label("kallsyms_names"); 437 off = 0; 438 for (i = 0; i < table_cnt; i++) { 439 if ((i & 0xFF) == 0) 440 markers[i >> 8] = off; 441 442 printf("\t.byte 0x%02x", table[i].len); 443 for (k = 0; k < table[i].len; k++) 444 printf(", 0x%02x", table[i].sym[k]); 445 printf("\n"); 446 447 off += table[i].len + 1; 448 } 449 printf("\n"); 450 451 output_label("kallsyms_markers"); 452 for (i = 0; i < ((table_cnt + 255) >> 8); i++) 453 printf("\t.long\t%u\n", markers[i]); 454 printf("\n"); 455 456 free(markers); 457 458 output_label("kallsyms_token_table"); 459 off = 0; 460 for (i = 0; i < 256; i++) { 461 best_idx[i] = off; 462 expand_symbol(best_table[i], best_table_len[i], buf); 463 printf("\t.asciz\t\"%s\"\n", buf); 464 off += strlen(buf) + 1; 465 } 466 printf("\n"); 467 468 output_label("kallsyms_token_index"); 469 for (i = 0; i < 256; i++) 470 printf("\t.short\t%d\n", best_idx[i]); 471 printf("\n"); 472 } 473 474 475 /* table lookup compression functions */ 476 477 /* count all the possible tokens in a symbol */ 478 static void learn_symbol(const unsigned char *symbol, int len) 479 { 480 int i; 481 482 for (i = 0; i < len - 1; i++) 483 token_profit[ symbol[i] + (symbol[i + 1] << 8) ]++; 484 } 485 486 /* decrease the count for all the possible tokens in a symbol */ 487 static void forget_symbol(const unsigned char *symbol, int len) 488 { 489 int i; 490 491 for (i = 0; i < len - 1; i++) 492 token_profit[ symbol[i] + (symbol[i + 1] << 8) ]--; 493 } 494 495 /* do the initial token count */ 496 static void build_initial_tok_table(void) 497 { 498 unsigned int i; 499 500 for (i = 0; i < table_cnt; i++) 501 learn_symbol(table[i].sym, table[i].len); 502 } 503 504 static unsigned char *find_token(unsigned char *str, int len, 505 const unsigned char *token) 506 { 507 int i; 508 509 for (i = 0; i < len - 1; i++) { 510 if (str[i] == token[0] && str[i+1] == token[1]) 511 return &str[i]; 512 } 513 return NULL; 514 } 515 516 /* replace a given token in all the valid symbols. Use the sampled symbols 517 * to update the counts */ 518 static void compress_symbols(const unsigned char *str, int idx) 519 { 520 unsigned int i, len, size; 521 unsigned char *p1, *p2; 522 523 for (i = 0; i < table_cnt; i++) { 524 525 len = table[i].len; 526 p1 = table[i].sym; 527 528 /* find the token on the symbol */ 529 p2 = find_token(p1, len, str); 530 if (!p2) continue; 531 532 /* decrease the counts for this symbol's tokens */ 533 forget_symbol(table[i].sym, len); 534 535 size = len; 536 537 do { 538 *p2 = idx; 539 p2++; 540 size -= (p2 - p1); 541 memmove(p2, p2 + 1, size); 542 p1 = p2; 543 len--; 544 545 if (size < 2) break; 546 547 /* find the token on the symbol */ 548 p2 = find_token(p1, size, str); 549 550 } while (p2); 551 552 table[i].len = len; 553 554 /* increase the counts for this symbol's new tokens */ 555 learn_symbol(table[i].sym, len); 556 } 557 } 558 559 /* search the token with the maximum profit */ 560 static int find_best_token(void) 561 { 562 int i, best, bestprofit; 563 564 bestprofit=-10000; 565 best = 0; 566 567 for (i = 0; i < 0x10000; i++) { 568 if (token_profit[i] > bestprofit) { 569 best = i; 570 bestprofit = token_profit[i]; 571 } 572 } 573 return best; 574 } 575 576 /* this is the core of the algorithm: calculate the "best" table */ 577 static void optimize_result(void) 578 { 579 int i, best; 580 581 /* using the '\0' symbol last allows compress_symbols to use standard 582 * fast string functions */ 583 for (i = 255; i >= 0; i--) { 584 585 /* if this table slot is empty (it is not used by an actual 586 * original char code */ 587 if (!best_table_len[i]) { 588 589 /* find the token with the best profit value */ 590 best = find_best_token(); 591 if (token_profit[best] == 0) 592 break; 593 594 /* place it in the "best" table */ 595 best_table_len[i] = 2; 596 best_table[i][0] = best & 0xFF; 597 best_table[i][1] = (best >> 8) & 0xFF; 598 599 /* replace this token in all the valid symbols */ 600 compress_symbols(best_table[i], i); 601 } 602 } 603 } 604 605 /* start by placing the symbols that are actually used on the table */ 606 static void insert_real_symbols_in_table(void) 607 { 608 unsigned int i, j, c; 609 610 for (i = 0; i < table_cnt; i++) { 611 for (j = 0; j < table[i].len; j++) { 612 c = table[i].sym[j]; 613 best_table[c][0]=c; 614 best_table_len[c]=1; 615 } 616 } 617 } 618 619 static void optimize_token_table(void) 620 { 621 build_initial_tok_table(); 622 623 insert_real_symbols_in_table(); 624 625 optimize_result(); 626 } 627 628 /* guess for "linker script provide" symbol */ 629 static int may_be_linker_script_provide_symbol(const struct sym_entry *se) 630 { 631 const char *symbol = sym_name(se); 632 int len = se->len - 1; 633 634 if (len < 8) 635 return 0; 636 637 if (symbol[0] != '_' || symbol[1] != '_') 638 return 0; 639 640 /* __start_XXXXX */ 641 if (!memcmp(symbol + 2, "start_", 6)) 642 return 1; 643 644 /* __stop_XXXXX */ 645 if (!memcmp(symbol + 2, "stop_", 5)) 646 return 1; 647 648 /* __end_XXXXX */ 649 if (!memcmp(symbol + 2, "end_", 4)) 650 return 1; 651 652 /* __XXXXX_start */ 653 if (!memcmp(symbol + len - 6, "_start", 6)) 654 return 1; 655 656 /* __XXXXX_end */ 657 if (!memcmp(symbol + len - 4, "_end", 4)) 658 return 1; 659 660 return 0; 661 } 662 663 static int compare_symbols(const void *a, const void *b) 664 { 665 const struct sym_entry *sa; 666 const struct sym_entry *sb; 667 int wa, wb; 668 669 sa = a; 670 sb = b; 671 672 /* sort by address first */ 673 if (sa->addr > sb->addr) 674 return 1; 675 if (sa->addr < sb->addr) 676 return -1; 677 678 /* sort by "weakness" type */ 679 wa = (sa->sym[0] == 'w') || (sa->sym[0] == 'W'); 680 wb = (sb->sym[0] == 'w') || (sb->sym[0] == 'W'); 681 if (wa != wb) 682 return wa - wb; 683 684 /* sort by "linker script provide" type */ 685 wa = may_be_linker_script_provide_symbol(sa); 686 wb = may_be_linker_script_provide_symbol(sb); 687 if (wa != wb) 688 return wa - wb; 689 690 /* sort by the number of prefix underscores */ 691 wa = strspn(sym_name(sa), "_"); 692 wb = strspn(sym_name(sb), "_"); 693 if (wa != wb) 694 return wa - wb; 695 696 /* sort by initial order, so that other symbols are left undisturbed */ 697 return sa->start_pos - sb->start_pos; 698 } 699 700 static void sort_symbols(void) 701 { 702 qsort(table, table_cnt, sizeof(struct sym_entry), compare_symbols); 703 } 704 705 static void make_percpus_absolute(void) 706 { 707 unsigned int i; 708 709 for (i = 0; i < table_cnt; i++) 710 if (symbol_in_range(&table[i], &percpu_range, 1)) { 711 /* 712 * Keep the 'A' override for percpu symbols to 713 * ensure consistent behavior compared to older 714 * versions of this tool. 715 */ 716 table[i].sym[0] = 'A'; 717 table[i].percpu_absolute = 1; 718 } 719 } 720 721 /* find the minimum non-absolute symbol address */ 722 static void record_relative_base(void) 723 { 724 unsigned int i; 725 726 for (i = 0; i < table_cnt; i++) 727 if (!symbol_absolute(&table[i])) { 728 /* 729 * The table is sorted by address. 730 * Take the first non-absolute symbol value. 731 */ 732 relative_base = table[i].addr; 733 return; 734 } 735 } 736 737 int main(int argc, char **argv) 738 { 739 if (argc >= 2) { 740 int i; 741 for (i = 1; i < argc; i++) { 742 if(strcmp(argv[i], "--all-symbols") == 0) 743 all_symbols = 1; 744 else if (strcmp(argv[i], "--absolute-percpu") == 0) 745 absolute_percpu = 1; 746 else if (strcmp(argv[i], "--base-relative") == 0) 747 base_relative = 1; 748 else 749 usage(); 750 } 751 } else if (argc != 1) 752 usage(); 753 754 read_map(stdin); 755 shrink_table(); 756 if (absolute_percpu) 757 make_percpus_absolute(); 758 sort_symbols(); 759 if (base_relative) 760 record_relative_base(); 761 optimize_token_table(); 762 write_src(); 763 764 return 0; 765 } 766