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 { "_stext_l1", "_etext_l1" }, /* Blackfin on-chip L1 inst SRAM */ 52 { "_stext_l2", "_etext_l2" }, /* Blackfin on-chip L2 SRAM */ 53 }; 54 #define text_range_text (&text_ranges[0]) 55 #define text_range_inittext (&text_ranges[1]) 56 57 static struct addr_range percpu_range = { 58 "__per_cpu_start", "__per_cpu_end", -1ULL, 0 59 }; 60 61 static struct sym_entry *table; 62 static unsigned int table_size, table_cnt; 63 static int all_symbols = 0; 64 static int absolute_percpu = 0; 65 static char symbol_prefix_char = '\0'; 66 static int base_relative = 0; 67 68 int token_profit[0x10000]; 69 70 /* the table that holds the result of the compression */ 71 unsigned char best_table[256][2]; 72 unsigned char best_table_len[256]; 73 74 75 static void usage(void) 76 { 77 fprintf(stderr, "Usage: kallsyms [--all-symbols] " 78 "[--symbol-prefix=<prefix char>] " 79 "[--base-relative] < in.map > out.S\n"); 80 exit(1); 81 } 82 83 /* 84 * This ignores the intensely annoying "mapping symbols" found 85 * in ARM ELF files: $a, $t and $d. 86 */ 87 static inline int is_arm_mapping_symbol(const char *str) 88 { 89 return str[0] == '$' && strchr("axtd", str[1]) 90 && (str[2] == '\0' || str[2] == '.'); 91 } 92 93 static int check_symbol_range(const char *sym, unsigned long long addr, 94 struct addr_range *ranges, int entries) 95 { 96 size_t i; 97 struct addr_range *ar; 98 99 for (i = 0; i < entries; ++i) { 100 ar = &ranges[i]; 101 102 if (strcmp(sym, ar->start_sym) == 0) { 103 ar->start = addr; 104 return 0; 105 } else if (strcmp(sym, ar->end_sym) == 0) { 106 ar->end = addr; 107 return 0; 108 } 109 } 110 111 return 1; 112 } 113 114 static int read_symbol(FILE *in, struct sym_entry *s) 115 { 116 char str[500]; 117 char *sym, stype; 118 int rc; 119 120 rc = fscanf(in, "%llx %c %499s\n", &s->addr, &stype, str); 121 if (rc != 3) { 122 if (rc != EOF && fgets(str, 500, in) == NULL) 123 fprintf(stderr, "Read error or end of file.\n"); 124 return -1; 125 } 126 if (strlen(str) > KSYM_NAME_LEN) { 127 fprintf(stderr, "Symbol %s too long for kallsyms (%zu vs %d).\n" 128 "Please increase KSYM_NAME_LEN both in kernel and kallsyms.c\n", 129 str, strlen(str), KSYM_NAME_LEN); 130 return -1; 131 } 132 133 sym = str; 134 /* skip prefix char */ 135 if (symbol_prefix_char && str[0] == symbol_prefix_char) 136 sym++; 137 138 /* Ignore most absolute/undefined (?) symbols. */ 139 if (strcmp(sym, "_text") == 0) 140 _text = s->addr; 141 else if (check_symbol_range(sym, s->addr, text_ranges, 142 ARRAY_SIZE(text_ranges)) == 0) 143 /* nothing to do */; 144 else if (toupper(stype) == 'A') 145 { 146 /* Keep these useful absolute symbols */ 147 if (strcmp(sym, "__kernel_syscall_via_break") && 148 strcmp(sym, "__kernel_syscall_via_epc") && 149 strcmp(sym, "__kernel_sigtramp") && 150 strcmp(sym, "__gp")) 151 return -1; 152 153 } 154 else if (toupper(stype) == 'U' || 155 is_arm_mapping_symbol(sym)) 156 return -1; 157 /* exclude also MIPS ELF local symbols ($L123 instead of .L123) */ 158 else if (str[0] == '$') 159 return -1; 160 /* exclude debugging symbols */ 161 else if (stype == 'N') 162 return -1; 163 164 /* include the type field in the symbol name, so that it gets 165 * compressed together */ 166 s->len = strlen(str) + 1; 167 s->sym = malloc(s->len + 1); 168 if (!s->sym) { 169 fprintf(stderr, "kallsyms failure: " 170 "unable to allocate required amount of memory\n"); 171 exit(EXIT_FAILURE); 172 } 173 strcpy((char *)s->sym + 1, str); 174 s->sym[0] = stype; 175 176 s->percpu_absolute = 0; 177 178 /* Record if we've found __per_cpu_start/end. */ 179 check_symbol_range(sym, s->addr, &percpu_range, 1); 180 181 return 0; 182 } 183 184 static int symbol_in_range(struct sym_entry *s, struct addr_range *ranges, 185 int entries) 186 { 187 size_t i; 188 struct addr_range *ar; 189 190 for (i = 0; i < entries; ++i) { 191 ar = &ranges[i]; 192 193 if (s->addr >= ar->start && s->addr <= ar->end) 194 return 1; 195 } 196 197 return 0; 198 } 199 200 static int symbol_valid(struct sym_entry *s) 201 { 202 /* Symbols which vary between passes. Passes 1 and 2 must have 203 * identical symbol lists. The kallsyms_* symbols below are only added 204 * after pass 1, they would be included in pass 2 when --all-symbols is 205 * specified so exclude them to get a stable symbol list. 206 */ 207 static char *special_symbols[] = { 208 "kallsyms_addresses", 209 "kallsyms_offsets", 210 "kallsyms_relative_base", 211 "kallsyms_num_syms", 212 "kallsyms_names", 213 "kallsyms_markers", 214 "kallsyms_token_table", 215 "kallsyms_token_index", 216 217 /* Exclude linker generated symbols which vary between passes */ 218 "_SDA_BASE_", /* ppc */ 219 "_SDA2_BASE_", /* ppc */ 220 NULL }; 221 222 static char *special_suffixes[] = { 223 "_veneer", /* arm */ 224 "_from_arm", /* arm */ 225 "_from_thumb", /* arm */ 226 NULL }; 227 228 int i; 229 char *sym_name = (char *)s->sym + 1; 230 231 /* skip prefix char */ 232 if (symbol_prefix_char && *sym_name == symbol_prefix_char) 233 sym_name++; 234 235 236 /* if --all-symbols is not specified, then symbols outside the text 237 * and inittext sections are discarded */ 238 if (!all_symbols) { 239 if (symbol_in_range(s, text_ranges, 240 ARRAY_SIZE(text_ranges)) == 0) 241 return 0; 242 /* Corner case. Discard any symbols with the same value as 243 * _etext _einittext; they can move between pass 1 and 2 when 244 * the kallsyms data are added. If these symbols move then 245 * they may get dropped in pass 2, which breaks the kallsyms 246 * rules. 247 */ 248 if ((s->addr == text_range_text->end && 249 strcmp(sym_name, 250 text_range_text->end_sym)) || 251 (s->addr == text_range_inittext->end && 252 strcmp(sym_name, 253 text_range_inittext->end_sym))) 254 return 0; 255 } 256 257 /* Exclude symbols which vary between passes. */ 258 for (i = 0; special_symbols[i]; i++) 259 if (strcmp(sym_name, special_symbols[i]) == 0) 260 return 0; 261 262 for (i = 0; special_suffixes[i]; i++) { 263 int l = strlen(sym_name) - strlen(special_suffixes[i]); 264 265 if (l >= 0 && strcmp(sym_name + l, special_suffixes[i]) == 0) 266 return 0; 267 } 268 269 return 1; 270 } 271 272 static void read_map(FILE *in) 273 { 274 while (!feof(in)) { 275 if (table_cnt >= table_size) { 276 table_size += 10000; 277 table = realloc(table, sizeof(*table) * table_size); 278 if (!table) { 279 fprintf(stderr, "out of memory\n"); 280 exit (1); 281 } 282 } 283 if (read_symbol(in, &table[table_cnt]) == 0) { 284 table[table_cnt].start_pos = table_cnt; 285 table_cnt++; 286 } 287 } 288 } 289 290 static void output_label(char *label) 291 { 292 if (symbol_prefix_char) 293 printf(".globl %c%s\n", symbol_prefix_char, label); 294 else 295 printf(".globl %s\n", label); 296 printf("\tALGN\n"); 297 if (symbol_prefix_char) 298 printf("%c%s:\n", symbol_prefix_char, label); 299 else 300 printf("%s:\n", label); 301 } 302 303 /* uncompress a compressed symbol. When this function is called, the best table 304 * might still be compressed itself, so the function needs to be recursive */ 305 static int expand_symbol(unsigned char *data, int len, char *result) 306 { 307 int c, rlen, total=0; 308 309 while (len) { 310 c = *data; 311 /* if the table holds a single char that is the same as the one 312 * we are looking for, then end the search */ 313 if (best_table[c][0]==c && best_table_len[c]==1) { 314 *result++ = c; 315 total++; 316 } else { 317 /* if not, recurse and expand */ 318 rlen = expand_symbol(best_table[c], best_table_len[c], result); 319 total += rlen; 320 result += rlen; 321 } 322 data++; 323 len--; 324 } 325 *result=0; 326 327 return total; 328 } 329 330 static int symbol_absolute(struct sym_entry *s) 331 { 332 return s->percpu_absolute; 333 } 334 335 static void write_src(void) 336 { 337 unsigned int i, k, off; 338 unsigned int best_idx[256]; 339 unsigned int *markers; 340 char buf[KSYM_NAME_LEN]; 341 342 printf("#include <asm/types.h>\n"); 343 printf("#if BITS_PER_LONG == 64\n"); 344 printf("#define PTR .quad\n"); 345 printf("#define ALGN .align 8\n"); 346 printf("#else\n"); 347 printf("#define PTR .long\n"); 348 printf("#define ALGN .align 4\n"); 349 printf("#endif\n"); 350 351 printf("\t.section .rodata, \"a\"\n"); 352 353 /* Provide proper symbols relocatability by their relativeness 354 * to a fixed anchor point in the runtime image, either '_text' 355 * for absolute address tables, in which case the linker will 356 * emit the final addresses at build time. Otherwise, use the 357 * offset relative to the lowest value encountered of all relative 358 * symbols, and emit non-relocatable fixed offsets that will be fixed 359 * up at runtime. 360 * 361 * The symbol names cannot be used to construct normal symbol 362 * references as the list of symbols contains symbols that are 363 * declared static and are private to their .o files. This prevents 364 * .tmp_kallsyms.o or any other object from referencing them. 365 */ 366 if (!base_relative) 367 output_label("kallsyms_addresses"); 368 else 369 output_label("kallsyms_offsets"); 370 371 for (i = 0; i < table_cnt; i++) { 372 if (base_relative) { 373 long long offset; 374 int overflow; 375 376 if (!absolute_percpu) { 377 offset = table[i].addr - relative_base; 378 overflow = (offset < 0 || offset > UINT_MAX); 379 } else if (symbol_absolute(&table[i])) { 380 offset = table[i].addr; 381 overflow = (offset < 0 || offset > INT_MAX); 382 } else { 383 offset = relative_base - table[i].addr - 1; 384 overflow = (offset < INT_MIN || offset >= 0); 385 } 386 if (overflow) { 387 fprintf(stderr, "kallsyms failure: " 388 "%s symbol value %#llx out of range in relative mode\n", 389 symbol_absolute(&table[i]) ? "absolute" : "relative", 390 table[i].addr); 391 exit(EXIT_FAILURE); 392 } 393 printf("\t.long\t%#x\n", (int)offset); 394 } else if (!symbol_absolute(&table[i])) { 395 if (_text <= table[i].addr) 396 printf("\tPTR\t_text + %#llx\n", 397 table[i].addr - _text); 398 else 399 printf("\tPTR\t_text - %#llx\n", 400 _text - table[i].addr); 401 } else { 402 printf("\tPTR\t%#llx\n", table[i].addr); 403 } 404 } 405 printf("\n"); 406 407 if (base_relative) { 408 output_label("kallsyms_relative_base"); 409 printf("\tPTR\t_text - %#llx\n", _text - relative_base); 410 printf("\n"); 411 } 412 413 output_label("kallsyms_num_syms"); 414 printf("\tPTR\t%d\n", table_cnt); 415 printf("\n"); 416 417 /* table of offset markers, that give the offset in the compressed stream 418 * every 256 symbols */ 419 markers = malloc(sizeof(unsigned int) * ((table_cnt + 255) / 256)); 420 if (!markers) { 421 fprintf(stderr, "kallsyms failure: " 422 "unable to allocate required memory\n"); 423 exit(EXIT_FAILURE); 424 } 425 426 output_label("kallsyms_names"); 427 off = 0; 428 for (i = 0; i < table_cnt; i++) { 429 if ((i & 0xFF) == 0) 430 markers[i >> 8] = off; 431 432 printf("\t.byte 0x%02x", table[i].len); 433 for (k = 0; k < table[i].len; k++) 434 printf(", 0x%02x", table[i].sym[k]); 435 printf("\n"); 436 437 off += table[i].len + 1; 438 } 439 printf("\n"); 440 441 output_label("kallsyms_markers"); 442 for (i = 0; i < ((table_cnt + 255) >> 8); i++) 443 printf("\tPTR\t%d\n", markers[i]); 444 printf("\n"); 445 446 free(markers); 447 448 output_label("kallsyms_token_table"); 449 off = 0; 450 for (i = 0; i < 256; i++) { 451 best_idx[i] = off; 452 expand_symbol(best_table[i], best_table_len[i], buf); 453 printf("\t.asciz\t\"%s\"\n", buf); 454 off += strlen(buf) + 1; 455 } 456 printf("\n"); 457 458 output_label("kallsyms_token_index"); 459 for (i = 0; i < 256; i++) 460 printf("\t.short\t%d\n", best_idx[i]); 461 printf("\n"); 462 } 463 464 465 /* table lookup compression functions */ 466 467 /* count all the possible tokens in a symbol */ 468 static void learn_symbol(unsigned char *symbol, int len) 469 { 470 int i; 471 472 for (i = 0; i < len - 1; i++) 473 token_profit[ symbol[i] + (symbol[i + 1] << 8) ]++; 474 } 475 476 /* decrease the count for all the possible tokens in a symbol */ 477 static void forget_symbol(unsigned char *symbol, int len) 478 { 479 int i; 480 481 for (i = 0; i < len - 1; i++) 482 token_profit[ symbol[i] + (symbol[i + 1] << 8) ]--; 483 } 484 485 /* remove all the invalid symbols from the table and do the initial token count */ 486 static void build_initial_tok_table(void) 487 { 488 unsigned int i, pos; 489 490 pos = 0; 491 for (i = 0; i < table_cnt; i++) { 492 if ( symbol_valid(&table[i]) ) { 493 if (pos != i) 494 table[pos] = table[i]; 495 learn_symbol(table[pos].sym, table[pos].len); 496 pos++; 497 } 498 } 499 table_cnt = pos; 500 } 501 502 static void *find_token(unsigned char *str, int len, unsigned char *token) 503 { 504 int i; 505 506 for (i = 0; i < len - 1; i++) { 507 if (str[i] == token[0] && str[i+1] == token[1]) 508 return &str[i]; 509 } 510 return NULL; 511 } 512 513 /* replace a given token in all the valid symbols. Use the sampled symbols 514 * to update the counts */ 515 static void compress_symbols(unsigned char *str, int idx) 516 { 517 unsigned int i, len, size; 518 unsigned char *p1, *p2; 519 520 for (i = 0; i < table_cnt; i++) { 521 522 len = table[i].len; 523 p1 = table[i].sym; 524 525 /* find the token on the symbol */ 526 p2 = find_token(p1, len, str); 527 if (!p2) continue; 528 529 /* decrease the counts for this symbol's tokens */ 530 forget_symbol(table[i].sym, len); 531 532 size = len; 533 534 do { 535 *p2 = idx; 536 p2++; 537 size -= (p2 - p1); 538 memmove(p2, p2 + 1, size); 539 p1 = p2; 540 len--; 541 542 if (size < 2) break; 543 544 /* find the token on the symbol */ 545 p2 = find_token(p1, size, str); 546 547 } while (p2); 548 549 table[i].len = len; 550 551 /* increase the counts for this symbol's new tokens */ 552 learn_symbol(table[i].sym, len); 553 } 554 } 555 556 /* search the token with the maximum profit */ 557 static int find_best_token(void) 558 { 559 int i, best, bestprofit; 560 561 bestprofit=-10000; 562 best = 0; 563 564 for (i = 0; i < 0x10000; i++) { 565 if (token_profit[i] > bestprofit) { 566 best = i; 567 bestprofit = token_profit[i]; 568 } 569 } 570 return best; 571 } 572 573 /* this is the core of the algorithm: calculate the "best" table */ 574 static void optimize_result(void) 575 { 576 int i, best; 577 578 /* using the '\0' symbol last allows compress_symbols to use standard 579 * fast string functions */ 580 for (i = 255; i >= 0; i--) { 581 582 /* if this table slot is empty (it is not used by an actual 583 * original char code */ 584 if (!best_table_len[i]) { 585 586 /* find the token with the breates profit value */ 587 best = find_best_token(); 588 if (token_profit[best] == 0) 589 break; 590 591 /* place it in the "best" table */ 592 best_table_len[i] = 2; 593 best_table[i][0] = best & 0xFF; 594 best_table[i][1] = (best >> 8) & 0xFF; 595 596 /* replace this token in all the valid symbols */ 597 compress_symbols(best_table[i], i); 598 } 599 } 600 } 601 602 /* start by placing the symbols that are actually used on the table */ 603 static void insert_real_symbols_in_table(void) 604 { 605 unsigned int i, j, c; 606 607 memset(best_table, 0, sizeof(best_table)); 608 memset(best_table_len, 0, sizeof(best_table_len)); 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 /* When valid symbol is not registered, exit to error */ 626 if (!table_cnt) { 627 fprintf(stderr, "No valid symbol.\n"); 628 exit(1); 629 } 630 631 optimize_result(); 632 } 633 634 /* guess for "linker script provide" symbol */ 635 static int may_be_linker_script_provide_symbol(const struct sym_entry *se) 636 { 637 const char *symbol = (char *)se->sym + 1; 638 int len = se->len - 1; 639 640 if (len < 8) 641 return 0; 642 643 if (symbol[0] != '_' || symbol[1] != '_') 644 return 0; 645 646 /* __start_XXXXX */ 647 if (!memcmp(symbol + 2, "start_", 6)) 648 return 1; 649 650 /* __stop_XXXXX */ 651 if (!memcmp(symbol + 2, "stop_", 5)) 652 return 1; 653 654 /* __end_XXXXX */ 655 if (!memcmp(symbol + 2, "end_", 4)) 656 return 1; 657 658 /* __XXXXX_start */ 659 if (!memcmp(symbol + len - 6, "_start", 6)) 660 return 1; 661 662 /* __XXXXX_end */ 663 if (!memcmp(symbol + len - 4, "_end", 4)) 664 return 1; 665 666 return 0; 667 } 668 669 static int prefix_underscores_count(const char *str) 670 { 671 const char *tail = str; 672 673 while (*tail == '_') 674 tail++; 675 676 return tail - str; 677 } 678 679 static int compare_symbols(const void *a, const void *b) 680 { 681 const struct sym_entry *sa; 682 const struct sym_entry *sb; 683 int wa, wb; 684 685 sa = a; 686 sb = b; 687 688 /* sort by address first */ 689 if (sa->addr > sb->addr) 690 return 1; 691 if (sa->addr < sb->addr) 692 return -1; 693 694 /* sort by "weakness" type */ 695 wa = (sa->sym[0] == 'w') || (sa->sym[0] == 'W'); 696 wb = (sb->sym[0] == 'w') || (sb->sym[0] == 'W'); 697 if (wa != wb) 698 return wa - wb; 699 700 /* sort by "linker script provide" type */ 701 wa = may_be_linker_script_provide_symbol(sa); 702 wb = may_be_linker_script_provide_symbol(sb); 703 if (wa != wb) 704 return wa - wb; 705 706 /* sort by the number of prefix underscores */ 707 wa = prefix_underscores_count((const char *)sa->sym + 1); 708 wb = prefix_underscores_count((const char *)sb->sym + 1); 709 if (wa != wb) 710 return wa - wb; 711 712 /* sort by initial order, so that other symbols are left undisturbed */ 713 return sa->start_pos - sb->start_pos; 714 } 715 716 static void sort_symbols(void) 717 { 718 qsort(table, table_cnt, sizeof(struct sym_entry), compare_symbols); 719 } 720 721 static void make_percpus_absolute(void) 722 { 723 unsigned int i; 724 725 for (i = 0; i < table_cnt; i++) 726 if (symbol_in_range(&table[i], &percpu_range, 1)) { 727 /* 728 * Keep the 'A' override for percpu symbols to 729 * ensure consistent behavior compared to older 730 * versions of this tool. 731 */ 732 table[i].sym[0] = 'A'; 733 table[i].percpu_absolute = 1; 734 } 735 } 736 737 /* find the minimum non-absolute symbol address */ 738 static void record_relative_base(void) 739 { 740 unsigned int i; 741 742 relative_base = -1ULL; 743 for (i = 0; i < table_cnt; i++) 744 if (!symbol_absolute(&table[i]) && 745 table[i].addr < relative_base) 746 relative_base = table[i].addr; 747 } 748 749 int main(int argc, char **argv) 750 { 751 if (argc >= 2) { 752 int i; 753 for (i = 1; i < argc; i++) { 754 if(strcmp(argv[i], "--all-symbols") == 0) 755 all_symbols = 1; 756 else if (strcmp(argv[i], "--absolute-percpu") == 0) 757 absolute_percpu = 1; 758 else if (strncmp(argv[i], "--symbol-prefix=", 16) == 0) { 759 char *p = &argv[i][16]; 760 /* skip quote */ 761 if ((*p == '"' && *(p+2) == '"') || (*p == '\'' && *(p+2) == '\'')) 762 p++; 763 symbol_prefix_char = *p; 764 } else if (strcmp(argv[i], "--base-relative") == 0) 765 base_relative = 1; 766 else 767 usage(); 768 } 769 } else if (argc != 1) 770 usage(); 771 772 read_map(stdin); 773 if (absolute_percpu) 774 make_percpus_absolute(); 775 if (base_relative) 776 record_relative_base(); 777 sort_symbols(); 778 optimize_token_table(); 779 write_src(); 780 781 return 0; 782 } 783