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 <stdio.h> 22 #include <stdlib.h> 23 #include <string.h> 24 #include <ctype.h> 25 #include <limits.h> 26 27 #ifndef ARRAY_SIZE 28 #define ARRAY_SIZE(arr) (sizeof(arr) / sizeof(arr[0])) 29 #endif 30 31 #define KSYM_NAME_LEN 128 32 33 struct sym_entry { 34 unsigned long long addr; 35 unsigned int len; 36 unsigned int start_pos; 37 unsigned char *sym; 38 unsigned int percpu_absolute; 39 }; 40 41 struct addr_range { 42 const char *start_sym, *end_sym; 43 unsigned long long start, end; 44 }; 45 46 static unsigned long long _text; 47 static unsigned long long relative_base; 48 static struct addr_range text_ranges[] = { 49 { "_stext", "_etext" }, 50 { "_sinittext", "_einittext" }, 51 }; 52 #define text_range_text (&text_ranges[0]) 53 #define text_range_inittext (&text_ranges[1]) 54 55 static struct addr_range percpu_range = { 56 "__per_cpu_start", "__per_cpu_end", -1ULL, 0 57 }; 58 59 static struct sym_entry *table; 60 static unsigned int table_size, table_cnt; 61 static int all_symbols = 0; 62 static int absolute_percpu = 0; 63 static int base_relative = 0; 64 65 int token_profit[0x10000]; 66 67 /* the table that holds the result of the compression */ 68 unsigned char best_table[256][2]; 69 unsigned char best_table_len[256]; 70 71 72 static void usage(void) 73 { 74 fprintf(stderr, "Usage: kallsyms [--all-symbols] " 75 "[--base-relative] < in.map > out.S\n"); 76 exit(1); 77 } 78 79 /* 80 * This ignores the intensely annoying "mapping symbols" found 81 * in ARM ELF files: $a, $t and $d. 82 */ 83 static inline int is_arm_mapping_symbol(const char *str) 84 { 85 return str[0] == '$' && strchr("axtd", str[1]) 86 && (str[2] == '\0' || str[2] == '.'); 87 } 88 89 static int check_symbol_range(const char *sym, unsigned long long addr, 90 struct addr_range *ranges, int entries) 91 { 92 size_t i; 93 struct addr_range *ar; 94 95 for (i = 0; i < entries; ++i) { 96 ar = &ranges[i]; 97 98 if (strcmp(sym, ar->start_sym) == 0) { 99 ar->start = addr; 100 return 0; 101 } else if (strcmp(sym, ar->end_sym) == 0) { 102 ar->end = addr; 103 return 0; 104 } 105 } 106 107 return 1; 108 } 109 110 static int read_symbol(FILE *in, struct sym_entry *s) 111 { 112 char sym[500], stype; 113 int rc; 114 115 rc = fscanf(in, "%llx %c %499s\n", &s->addr, &stype, sym); 116 if (rc != 3) { 117 if (rc != EOF && fgets(sym, 500, in) == NULL) 118 fprintf(stderr, "Read error or end of file.\n"); 119 return -1; 120 } 121 if (strlen(sym) > KSYM_NAME_LEN) { 122 fprintf(stderr, "Symbol %s too long for kallsyms (%zu vs %d).\n" 123 "Please increase KSYM_NAME_LEN both in kernel and kallsyms.c\n", 124 sym, strlen(sym), KSYM_NAME_LEN); 125 return -1; 126 } 127 128 /* Ignore most absolute/undefined (?) symbols. */ 129 if (strcmp(sym, "_text") == 0) 130 _text = s->addr; 131 else if (check_symbol_range(sym, s->addr, text_ranges, 132 ARRAY_SIZE(text_ranges)) == 0) 133 /* nothing to do */; 134 else if (toupper(stype) == 'A') 135 { 136 /* Keep these useful absolute symbols */ 137 if (strcmp(sym, "__kernel_syscall_via_break") && 138 strcmp(sym, "__kernel_syscall_via_epc") && 139 strcmp(sym, "__kernel_sigtramp") && 140 strcmp(sym, "__gp")) 141 return -1; 142 143 } 144 else if (toupper(stype) == 'U' || 145 is_arm_mapping_symbol(sym)) 146 return -1; 147 /* exclude also MIPS ELF local symbols ($L123 instead of .L123) */ 148 else if (sym[0] == '$') 149 return -1; 150 /* exclude debugging symbols */ 151 else if (stype == 'N' || stype == 'n') 152 return -1; 153 154 /* include the type field in the symbol name, so that it gets 155 * compressed together */ 156 s->len = strlen(sym) + 1; 157 s->sym = malloc(s->len + 1); 158 if (!s->sym) { 159 fprintf(stderr, "kallsyms failure: " 160 "unable to allocate required amount of memory\n"); 161 exit(EXIT_FAILURE); 162 } 163 strcpy((char *)s->sym + 1, sym); 164 s->sym[0] = stype; 165 166 s->percpu_absolute = 0; 167 168 /* Record if we've found __per_cpu_start/end. */ 169 check_symbol_range(sym, s->addr, &percpu_range, 1); 170 171 return 0; 172 } 173 174 static int symbol_in_range(struct sym_entry *s, struct addr_range *ranges, 175 int entries) 176 { 177 size_t i; 178 struct addr_range *ar; 179 180 for (i = 0; i < entries; ++i) { 181 ar = &ranges[i]; 182 183 if (s->addr >= ar->start && s->addr <= ar->end) 184 return 1; 185 } 186 187 return 0; 188 } 189 190 static int symbol_valid(struct sym_entry *s) 191 { 192 /* Symbols which vary between passes. Passes 1 and 2 must have 193 * identical symbol lists. The kallsyms_* symbols below are only added 194 * after pass 1, they would be included in pass 2 when --all-symbols is 195 * specified so exclude them to get a stable symbol list. 196 */ 197 static char *special_symbols[] = { 198 "kallsyms_addresses", 199 "kallsyms_offsets", 200 "kallsyms_relative_base", 201 "kallsyms_num_syms", 202 "kallsyms_names", 203 "kallsyms_markers", 204 "kallsyms_token_table", 205 "kallsyms_token_index", 206 207 /* Exclude linker generated symbols which vary between passes */ 208 "_SDA_BASE_", /* ppc */ 209 "_SDA2_BASE_", /* ppc */ 210 NULL }; 211 212 static char *special_prefixes[] = { 213 "__crc_", /* modversions */ 214 "__efistub_", /* arm64 EFI stub namespace */ 215 NULL }; 216 217 static char *special_suffixes[] = { 218 "_veneer", /* arm */ 219 "_from_arm", /* arm */ 220 "_from_thumb", /* arm */ 221 NULL }; 222 223 int i; 224 char *sym_name = (char *)s->sym + 1; 225 226 /* if --all-symbols is not specified, then symbols outside the text 227 * and inittext sections are discarded */ 228 if (!all_symbols) { 229 if (symbol_in_range(s, text_ranges, 230 ARRAY_SIZE(text_ranges)) == 0) 231 return 0; 232 /* Corner case. Discard any symbols with the same value as 233 * _etext _einittext; they can move between pass 1 and 2 when 234 * the kallsyms data are added. If these symbols move then 235 * they may get dropped in pass 2, which breaks the kallsyms 236 * rules. 237 */ 238 if ((s->addr == text_range_text->end && 239 strcmp(sym_name, 240 text_range_text->end_sym)) || 241 (s->addr == text_range_inittext->end && 242 strcmp(sym_name, 243 text_range_inittext->end_sym))) 244 return 0; 245 } 246 247 /* Exclude symbols which vary between passes. */ 248 for (i = 0; special_symbols[i]; i++) 249 if (strcmp(sym_name, special_symbols[i]) == 0) 250 return 0; 251 252 for (i = 0; special_prefixes[i]; i++) { 253 int l = strlen(special_prefixes[i]); 254 255 if (l <= strlen(sym_name) && 256 strncmp(sym_name, special_prefixes[i], l) == 0) 257 return 0; 258 } 259 260 for (i = 0; special_suffixes[i]; i++) { 261 int l = strlen(sym_name) - strlen(special_suffixes[i]); 262 263 if (l >= 0 && strcmp(sym_name + l, special_suffixes[i]) == 0) 264 return 0; 265 } 266 267 return 1; 268 } 269 270 static void read_map(FILE *in) 271 { 272 while (!feof(in)) { 273 if (table_cnt >= table_size) { 274 table_size += 10000; 275 table = realloc(table, sizeof(*table) * table_size); 276 if (!table) { 277 fprintf(stderr, "out of memory\n"); 278 exit (1); 279 } 280 } 281 if (read_symbol(in, &table[table_cnt]) == 0) { 282 table[table_cnt].start_pos = table_cnt; 283 table_cnt++; 284 } 285 } 286 } 287 288 static void output_label(char *label) 289 { 290 printf(".globl %s\n", label); 291 printf("\tALGN\n"); 292 printf("%s:\n", label); 293 } 294 295 /* uncompress a compressed symbol. When this function is called, the best table 296 * might still be compressed itself, so the function needs to be recursive */ 297 static int expand_symbol(unsigned char *data, int len, char *result) 298 { 299 int c, rlen, total=0; 300 301 while (len) { 302 c = *data; 303 /* if the table holds a single char that is the same as the one 304 * we are looking for, then end the search */ 305 if (best_table[c][0]==c && best_table_len[c]==1) { 306 *result++ = c; 307 total++; 308 } else { 309 /* if not, recurse and expand */ 310 rlen = expand_symbol(best_table[c], best_table_len[c], result); 311 total += rlen; 312 result += rlen; 313 } 314 data++; 315 len--; 316 } 317 *result=0; 318 319 return total; 320 } 321 322 static int symbol_absolute(struct sym_entry *s) 323 { 324 return s->percpu_absolute; 325 } 326 327 static void write_src(void) 328 { 329 unsigned int i, k, off; 330 unsigned int best_idx[256]; 331 unsigned int *markers; 332 char buf[KSYM_NAME_LEN]; 333 334 printf("#include <asm/types.h>\n"); 335 printf("#if BITS_PER_LONG == 64\n"); 336 printf("#define PTR .quad\n"); 337 printf("#define ALGN .balign 8\n"); 338 printf("#else\n"); 339 printf("#define PTR .long\n"); 340 printf("#define ALGN .balign 4\n"); 341 printf("#endif\n"); 342 343 printf("\t.section .rodata, \"a\"\n"); 344 345 /* Provide proper symbols relocatability by their relativeness 346 * to a fixed anchor point in the runtime image, either '_text' 347 * for absolute address tables, in which case the linker will 348 * emit the final addresses at build time. Otherwise, use the 349 * offset relative to the lowest value encountered of all relative 350 * symbols, and emit non-relocatable fixed offsets that will be fixed 351 * up at runtime. 352 * 353 * The symbol names cannot be used to construct normal symbol 354 * references as the list of symbols contains symbols that are 355 * declared static and are private to their .o files. This prevents 356 * .tmp_kallsyms.o or any other object from referencing them. 357 */ 358 if (!base_relative) 359 output_label("kallsyms_addresses"); 360 else 361 output_label("kallsyms_offsets"); 362 363 for (i = 0; i < table_cnt; i++) { 364 if (base_relative) { 365 long long offset; 366 int overflow; 367 368 if (!absolute_percpu) { 369 offset = table[i].addr - relative_base; 370 overflow = (offset < 0 || offset > UINT_MAX); 371 } else if (symbol_absolute(&table[i])) { 372 offset = table[i].addr; 373 overflow = (offset < 0 || offset > INT_MAX); 374 } else { 375 offset = relative_base - table[i].addr - 1; 376 overflow = (offset < INT_MIN || offset >= 0); 377 } 378 if (overflow) { 379 fprintf(stderr, "kallsyms failure: " 380 "%s symbol value %#llx out of range in relative mode\n", 381 symbol_absolute(&table[i]) ? "absolute" : "relative", 382 table[i].addr); 383 exit(EXIT_FAILURE); 384 } 385 printf("\t.long\t%#x\n", (int)offset); 386 } else if (!symbol_absolute(&table[i])) { 387 if (_text <= table[i].addr) 388 printf("\tPTR\t_text + %#llx\n", 389 table[i].addr - _text); 390 else 391 printf("\tPTR\t_text - %#llx\n", 392 _text - table[i].addr); 393 } else { 394 printf("\tPTR\t%#llx\n", table[i].addr); 395 } 396 } 397 printf("\n"); 398 399 if (base_relative) { 400 output_label("kallsyms_relative_base"); 401 printf("\tPTR\t_text - %#llx\n", _text - relative_base); 402 printf("\n"); 403 } 404 405 output_label("kallsyms_num_syms"); 406 printf("\t.long\t%u\n", table_cnt); 407 printf("\n"); 408 409 /* table of offset markers, that give the offset in the compressed stream 410 * every 256 symbols */ 411 markers = malloc(sizeof(unsigned int) * ((table_cnt + 255) / 256)); 412 if (!markers) { 413 fprintf(stderr, "kallsyms failure: " 414 "unable to allocate required memory\n"); 415 exit(EXIT_FAILURE); 416 } 417 418 output_label("kallsyms_names"); 419 off = 0; 420 for (i = 0; i < table_cnt; i++) { 421 if ((i & 0xFF) == 0) 422 markers[i >> 8] = off; 423 424 printf("\t.byte 0x%02x", table[i].len); 425 for (k = 0; k < table[i].len; k++) 426 printf(", 0x%02x", table[i].sym[k]); 427 printf("\n"); 428 429 off += table[i].len + 1; 430 } 431 printf("\n"); 432 433 output_label("kallsyms_markers"); 434 for (i = 0; i < ((table_cnt + 255) >> 8); i++) 435 printf("\t.long\t%u\n", markers[i]); 436 printf("\n"); 437 438 free(markers); 439 440 output_label("kallsyms_token_table"); 441 off = 0; 442 for (i = 0; i < 256; i++) { 443 best_idx[i] = off; 444 expand_symbol(best_table[i], best_table_len[i], buf); 445 printf("\t.asciz\t\"%s\"\n", buf); 446 off += strlen(buf) + 1; 447 } 448 printf("\n"); 449 450 output_label("kallsyms_token_index"); 451 for (i = 0; i < 256; i++) 452 printf("\t.short\t%d\n", best_idx[i]); 453 printf("\n"); 454 } 455 456 457 /* table lookup compression functions */ 458 459 /* count all the possible tokens in a symbol */ 460 static void learn_symbol(unsigned char *symbol, int len) 461 { 462 int i; 463 464 for (i = 0; i < len - 1; i++) 465 token_profit[ symbol[i] + (symbol[i + 1] << 8) ]++; 466 } 467 468 /* decrease the count for all the possible tokens in a symbol */ 469 static void forget_symbol(unsigned char *symbol, int len) 470 { 471 int i; 472 473 for (i = 0; i < len - 1; i++) 474 token_profit[ symbol[i] + (symbol[i + 1] << 8) ]--; 475 } 476 477 /* remove all the invalid symbols from the table and do the initial token count */ 478 static void build_initial_tok_table(void) 479 { 480 unsigned int i, pos; 481 482 pos = 0; 483 for (i = 0; i < table_cnt; i++) { 484 if ( symbol_valid(&table[i]) ) { 485 if (pos != i) 486 table[pos] = table[i]; 487 learn_symbol(table[pos].sym, table[pos].len); 488 pos++; 489 } 490 } 491 table_cnt = pos; 492 } 493 494 static void *find_token(unsigned char *str, int len, unsigned char *token) 495 { 496 int i; 497 498 for (i = 0; i < len - 1; i++) { 499 if (str[i] == token[0] && str[i+1] == token[1]) 500 return &str[i]; 501 } 502 return NULL; 503 } 504 505 /* replace a given token in all the valid symbols. Use the sampled symbols 506 * to update the counts */ 507 static void compress_symbols(unsigned char *str, int idx) 508 { 509 unsigned int i, len, size; 510 unsigned char *p1, *p2; 511 512 for (i = 0; i < table_cnt; i++) { 513 514 len = table[i].len; 515 p1 = table[i].sym; 516 517 /* find the token on the symbol */ 518 p2 = find_token(p1, len, str); 519 if (!p2) continue; 520 521 /* decrease the counts for this symbol's tokens */ 522 forget_symbol(table[i].sym, len); 523 524 size = len; 525 526 do { 527 *p2 = idx; 528 p2++; 529 size -= (p2 - p1); 530 memmove(p2, p2 + 1, size); 531 p1 = p2; 532 len--; 533 534 if (size < 2) break; 535 536 /* find the token on the symbol */ 537 p2 = find_token(p1, size, str); 538 539 } while (p2); 540 541 table[i].len = len; 542 543 /* increase the counts for this symbol's new tokens */ 544 learn_symbol(table[i].sym, len); 545 } 546 } 547 548 /* search the token with the maximum profit */ 549 static int find_best_token(void) 550 { 551 int i, best, bestprofit; 552 553 bestprofit=-10000; 554 best = 0; 555 556 for (i = 0; i < 0x10000; i++) { 557 if (token_profit[i] > bestprofit) { 558 best = i; 559 bestprofit = token_profit[i]; 560 } 561 } 562 return best; 563 } 564 565 /* this is the core of the algorithm: calculate the "best" table */ 566 static void optimize_result(void) 567 { 568 int i, best; 569 570 /* using the '\0' symbol last allows compress_symbols to use standard 571 * fast string functions */ 572 for (i = 255; i >= 0; i--) { 573 574 /* if this table slot is empty (it is not used by an actual 575 * original char code */ 576 if (!best_table_len[i]) { 577 578 /* find the token with the best profit value */ 579 best = find_best_token(); 580 if (token_profit[best] == 0) 581 break; 582 583 /* place it in the "best" table */ 584 best_table_len[i] = 2; 585 best_table[i][0] = best & 0xFF; 586 best_table[i][1] = (best >> 8) & 0xFF; 587 588 /* replace this token in all the valid symbols */ 589 compress_symbols(best_table[i], i); 590 } 591 } 592 } 593 594 /* start by placing the symbols that are actually used on the table */ 595 static void insert_real_symbols_in_table(void) 596 { 597 unsigned int i, j, c; 598 599 memset(best_table, 0, sizeof(best_table)); 600 memset(best_table_len, 0, sizeof(best_table_len)); 601 602 for (i = 0; i < table_cnt; i++) { 603 for (j = 0; j < table[i].len; j++) { 604 c = table[i].sym[j]; 605 best_table[c][0]=c; 606 best_table_len[c]=1; 607 } 608 } 609 } 610 611 static void optimize_token_table(void) 612 { 613 build_initial_tok_table(); 614 615 insert_real_symbols_in_table(); 616 617 /* When valid symbol is not registered, exit to error */ 618 if (!table_cnt) { 619 fprintf(stderr, "No valid symbol.\n"); 620 exit(1); 621 } 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 = (char *)se->sym + 1; 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 prefix_underscores_count(const char *str) 662 { 663 const char *tail = str; 664 665 while (*tail == '_') 666 tail++; 667 668 return tail - str; 669 } 670 671 static int compare_symbols(const void *a, const void *b) 672 { 673 const struct sym_entry *sa; 674 const struct sym_entry *sb; 675 int wa, wb; 676 677 sa = a; 678 sb = b; 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 = prefix_underscores_count((const char *)sa->sym + 1); 700 wb = prefix_underscores_count((const char *)sb->sym + 1); 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(struct sym_entry), 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 relative_base = -1ULL; 735 for (i = 0; i < table_cnt; i++) 736 if (!symbol_absolute(&table[i]) && 737 table[i].addr < relative_base) 738 relative_base = table[i].addr; 739 } 740 741 int main(int argc, char **argv) 742 { 743 if (argc >= 2) { 744 int i; 745 for (i = 1; i < argc; i++) { 746 if(strcmp(argv[i], "--all-symbols") == 0) 747 all_symbols = 1; 748 else if (strcmp(argv[i], "--absolute-percpu") == 0) 749 absolute_percpu = 1; 750 else if (strcmp(argv[i], "--base-relative") == 0) 751 base_relative = 1; 752 else 753 usage(); 754 } 755 } else if (argc != 1) 756 usage(); 757 758 read_map(stdin); 759 if (absolute_percpu) 760 make_percpus_absolute(); 761 if (base_relative) 762 record_relative_base(); 763 sort_symbols(); 764 optimize_token_table(); 765 write_src(); 766 767 return 0; 768 } 769