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 "[--page-offset=<CONFIG_PAGE_OFFSET>] " 80 "[--base-relative] < in.map > out.S\n"); 81 exit(1); 82 } 83 84 /* 85 * This ignores the intensely annoying "mapping symbols" found 86 * in ARM ELF files: $a, $t and $d. 87 */ 88 static inline int is_arm_mapping_symbol(const char *str) 89 { 90 return str[0] == '$' && strchr("axtd", str[1]) 91 && (str[2] == '\0' || str[2] == '.'); 92 } 93 94 static int check_symbol_range(const char *sym, unsigned long long addr, 95 struct addr_range *ranges, int entries) 96 { 97 size_t i; 98 struct addr_range *ar; 99 100 for (i = 0; i < entries; ++i) { 101 ar = &ranges[i]; 102 103 if (strcmp(sym, ar->start_sym) == 0) { 104 ar->start = addr; 105 return 0; 106 } else if (strcmp(sym, ar->end_sym) == 0) { 107 ar->end = addr; 108 return 0; 109 } 110 } 111 112 return 1; 113 } 114 115 static int read_symbol(FILE *in, struct sym_entry *s) 116 { 117 char str[500]; 118 char *sym, stype; 119 int rc; 120 121 rc = fscanf(in, "%llx %c %499s\n", &s->addr, &stype, str); 122 if (rc != 3) { 123 if (rc != EOF && fgets(str, 500, in) == NULL) 124 fprintf(stderr, "Read error or end of file.\n"); 125 return -1; 126 } 127 if (strlen(str) > KSYM_NAME_LEN) { 128 fprintf(stderr, "Symbol %s too long for kallsyms (%zu vs %d).\n" 129 "Please increase KSYM_NAME_LEN both in kernel and kallsyms.c\n", 130 str, strlen(str), KSYM_NAME_LEN); 131 return -1; 132 } 133 134 sym = str; 135 /* skip prefix char */ 136 if (symbol_prefix_char && str[0] == symbol_prefix_char) 137 sym++; 138 139 /* Ignore most absolute/undefined (?) symbols. */ 140 if (strcmp(sym, "_text") == 0) 141 _text = s->addr; 142 else if (check_symbol_range(sym, s->addr, text_ranges, 143 ARRAY_SIZE(text_ranges)) == 0) 144 /* nothing to do */; 145 else if (toupper(stype) == 'A') 146 { 147 /* Keep these useful absolute symbols */ 148 if (strcmp(sym, "__kernel_syscall_via_break") && 149 strcmp(sym, "__kernel_syscall_via_epc") && 150 strcmp(sym, "__kernel_sigtramp") && 151 strcmp(sym, "__gp")) 152 return -1; 153 154 } 155 else if (toupper(stype) == 'U' || 156 is_arm_mapping_symbol(sym)) 157 return -1; 158 /* exclude also MIPS ELF local symbols ($L123 instead of .L123) */ 159 else if (str[0] == '$') 160 return -1; 161 /* exclude debugging symbols */ 162 else if (stype == 'N') 163 return -1; 164 165 /* include the type field in the symbol name, so that it gets 166 * compressed together */ 167 s->len = strlen(str) + 1; 168 s->sym = malloc(s->len + 1); 169 if (!s->sym) { 170 fprintf(stderr, "kallsyms failure: " 171 "unable to allocate required amount of memory\n"); 172 exit(EXIT_FAILURE); 173 } 174 strcpy((char *)s->sym + 1, str); 175 s->sym[0] = stype; 176 177 s->percpu_absolute = 0; 178 179 /* Record if we've found __per_cpu_start/end. */ 180 check_symbol_range(sym, s->addr, &percpu_range, 1); 181 182 return 0; 183 } 184 185 static int symbol_in_range(struct sym_entry *s, struct addr_range *ranges, 186 int entries) 187 { 188 size_t i; 189 struct addr_range *ar; 190 191 for (i = 0; i < entries; ++i) { 192 ar = &ranges[i]; 193 194 if (s->addr >= ar->start && s->addr <= ar->end) 195 return 1; 196 } 197 198 return 0; 199 } 200 201 static int symbol_valid(struct sym_entry *s) 202 { 203 /* Symbols which vary between passes. Passes 1 and 2 must have 204 * identical symbol lists. The kallsyms_* symbols below are only added 205 * after pass 1, they would be included in pass 2 when --all-symbols is 206 * specified so exclude them to get a stable symbol list. 207 */ 208 static char *special_symbols[] = { 209 "kallsyms_addresses", 210 "kallsyms_offsets", 211 "kallsyms_relative_base", 212 "kallsyms_num_syms", 213 "kallsyms_names", 214 "kallsyms_markers", 215 "kallsyms_token_table", 216 "kallsyms_token_index", 217 218 /* Exclude linker generated symbols which vary between passes */ 219 "_SDA_BASE_", /* ppc */ 220 "_SDA2_BASE_", /* ppc */ 221 NULL }; 222 223 static char *special_suffixes[] = { 224 "_veneer", /* arm */ 225 "_from_arm", /* arm */ 226 "_from_thumb", /* arm */ 227 NULL }; 228 229 int i; 230 char *sym_name = (char *)s->sym + 1; 231 232 /* skip prefix char */ 233 if (symbol_prefix_char && *sym_name == symbol_prefix_char) 234 sym_name++; 235 236 237 /* if --all-symbols is not specified, then symbols outside the text 238 * and inittext sections are discarded */ 239 if (!all_symbols) { 240 if (symbol_in_range(s, text_ranges, 241 ARRAY_SIZE(text_ranges)) == 0) 242 return 0; 243 /* Corner case. Discard any symbols with the same value as 244 * _etext _einittext; they can move between pass 1 and 2 when 245 * the kallsyms data are added. If these symbols move then 246 * they may get dropped in pass 2, which breaks the kallsyms 247 * rules. 248 */ 249 if ((s->addr == text_range_text->end && 250 strcmp(sym_name, 251 text_range_text->end_sym)) || 252 (s->addr == text_range_inittext->end && 253 strcmp(sym_name, 254 text_range_inittext->end_sym))) 255 return 0; 256 } 257 258 /* Exclude symbols which vary between passes. */ 259 for (i = 0; special_symbols[i]; i++) 260 if (strcmp(sym_name, special_symbols[i]) == 0) 261 return 0; 262 263 for (i = 0; special_suffixes[i]; i++) { 264 int l = strlen(sym_name) - strlen(special_suffixes[i]); 265 266 if (l >= 0 && strcmp(sym_name + l, special_suffixes[i]) == 0) 267 return 0; 268 } 269 270 return 1; 271 } 272 273 static void read_map(FILE *in) 274 { 275 while (!feof(in)) { 276 if (table_cnt >= table_size) { 277 table_size += 10000; 278 table = realloc(table, sizeof(*table) * table_size); 279 if (!table) { 280 fprintf(stderr, "out of memory\n"); 281 exit (1); 282 } 283 } 284 if (read_symbol(in, &table[table_cnt]) == 0) { 285 table[table_cnt].start_pos = table_cnt; 286 table_cnt++; 287 } 288 } 289 } 290 291 static void output_label(char *label) 292 { 293 if (symbol_prefix_char) 294 printf(".globl %c%s\n", symbol_prefix_char, label); 295 else 296 printf(".globl %s\n", label); 297 printf("\tALGN\n"); 298 if (symbol_prefix_char) 299 printf("%c%s:\n", symbol_prefix_char, label); 300 else 301 printf("%s:\n", label); 302 } 303 304 /* uncompress a compressed symbol. When this function is called, the best table 305 * might still be compressed itself, so the function needs to be recursive */ 306 static int expand_symbol(unsigned char *data, int len, char *result) 307 { 308 int c, rlen, total=0; 309 310 while (len) { 311 c = *data; 312 /* if the table holds a single char that is the same as the one 313 * we are looking for, then end the search */ 314 if (best_table[c][0]==c && best_table_len[c]==1) { 315 *result++ = c; 316 total++; 317 } else { 318 /* if not, recurse and expand */ 319 rlen = expand_symbol(best_table[c], best_table_len[c], result); 320 total += rlen; 321 result += rlen; 322 } 323 data++; 324 len--; 325 } 326 *result=0; 327 328 return total; 329 } 330 331 static int symbol_absolute(struct sym_entry *s) 332 { 333 return s->percpu_absolute; 334 } 335 336 static void write_src(void) 337 { 338 unsigned int i, k, off; 339 unsigned int best_idx[256]; 340 unsigned int *markers; 341 char buf[KSYM_NAME_LEN]; 342 343 printf("#include <asm/types.h>\n"); 344 printf("#if BITS_PER_LONG == 64\n"); 345 printf("#define PTR .quad\n"); 346 printf("#define ALGN .align 8\n"); 347 printf("#else\n"); 348 printf("#define PTR .long\n"); 349 printf("#define ALGN .align 4\n"); 350 printf("#endif\n"); 351 352 printf("\t.section .rodata, \"a\"\n"); 353 354 /* Provide proper symbols relocatability by their relativeness 355 * to a fixed anchor point in the runtime image, either '_text' 356 * for absolute address tables, in which case the linker will 357 * emit the final addresses at build time. Otherwise, use the 358 * offset relative to the lowest value encountered of all relative 359 * symbols, and emit non-relocatable fixed offsets that will be fixed 360 * up at runtime. 361 * 362 * The symbol names cannot be used to construct normal symbol 363 * references as the list of symbols contains symbols that are 364 * declared static and are private to their .o files. This prevents 365 * .tmp_kallsyms.o or any other object from referencing them. 366 */ 367 if (!base_relative) 368 output_label("kallsyms_addresses"); 369 else 370 output_label("kallsyms_offsets"); 371 372 for (i = 0; i < table_cnt; i++) { 373 if (base_relative) { 374 long long offset; 375 int overflow; 376 377 if (!absolute_percpu) { 378 offset = table[i].addr - relative_base; 379 overflow = (offset < 0 || offset > UINT_MAX); 380 } else if (symbol_absolute(&table[i])) { 381 offset = table[i].addr; 382 overflow = (offset < 0 || offset > INT_MAX); 383 } else { 384 offset = relative_base - table[i].addr - 1; 385 overflow = (offset < INT_MIN || offset >= 0); 386 } 387 if (overflow) { 388 fprintf(stderr, "kallsyms failure: " 389 "%s symbol value %#llx out of range in relative mode\n", 390 symbol_absolute(&table[i]) ? "absolute" : "relative", 391 table[i].addr); 392 exit(EXIT_FAILURE); 393 } 394 printf("\t.long\t%#x\n", (int)offset); 395 } else if (!symbol_absolute(&table[i])) { 396 if (_text <= table[i].addr) 397 printf("\tPTR\t_text + %#llx\n", 398 table[i].addr - _text); 399 else 400 printf("\tPTR\t_text - %#llx\n", 401 _text - table[i].addr); 402 } else { 403 printf("\tPTR\t%#llx\n", table[i].addr); 404 } 405 } 406 printf("\n"); 407 408 if (base_relative) { 409 output_label("kallsyms_relative_base"); 410 printf("\tPTR\t_text - %#llx\n", _text - relative_base); 411 printf("\n"); 412 } 413 414 output_label("kallsyms_num_syms"); 415 printf("\tPTR\t%d\n", table_cnt); 416 printf("\n"); 417 418 /* table of offset markers, that give the offset in the compressed stream 419 * every 256 symbols */ 420 markers = malloc(sizeof(unsigned int) * ((table_cnt + 255) / 256)); 421 if (!markers) { 422 fprintf(stderr, "kallsyms failure: " 423 "unable to allocate required memory\n"); 424 exit(EXIT_FAILURE); 425 } 426 427 output_label("kallsyms_names"); 428 off = 0; 429 for (i = 0; i < table_cnt; i++) { 430 if ((i & 0xFF) == 0) 431 markers[i >> 8] = off; 432 433 printf("\t.byte 0x%02x", table[i].len); 434 for (k = 0; k < table[i].len; k++) 435 printf(", 0x%02x", table[i].sym[k]); 436 printf("\n"); 437 438 off += table[i].len + 1; 439 } 440 printf("\n"); 441 442 output_label("kallsyms_markers"); 443 for (i = 0; i < ((table_cnt + 255) >> 8); i++) 444 printf("\tPTR\t%d\n", markers[i]); 445 printf("\n"); 446 447 free(markers); 448 449 output_label("kallsyms_token_table"); 450 off = 0; 451 for (i = 0; i < 256; i++) { 452 best_idx[i] = off; 453 expand_symbol(best_table[i], best_table_len[i], buf); 454 printf("\t.asciz\t\"%s\"\n", buf); 455 off += strlen(buf) + 1; 456 } 457 printf("\n"); 458 459 output_label("kallsyms_token_index"); 460 for (i = 0; i < 256; i++) 461 printf("\t.short\t%d\n", best_idx[i]); 462 printf("\n"); 463 } 464 465 466 /* table lookup compression functions */ 467 468 /* count all the possible tokens in a symbol */ 469 static void learn_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 /* decrease the count for all the possible tokens in a symbol */ 478 static void forget_symbol(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 /* remove all the invalid symbols from the table and do the initial token count */ 487 static void build_initial_tok_table(void) 488 { 489 unsigned int i, pos; 490 491 pos = 0; 492 for (i = 0; i < table_cnt; i++) { 493 if ( symbol_valid(&table[i]) ) { 494 if (pos != i) 495 table[pos] = table[i]; 496 learn_symbol(table[pos].sym, table[pos].len); 497 pos++; 498 } 499 } 500 table_cnt = pos; 501 } 502 503 static void *find_token(unsigned char *str, int len, 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(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 breates 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 memset(best_table, 0, sizeof(best_table)); 609 memset(best_table_len, 0, sizeof(best_table_len)); 610 611 for (i = 0; i < table_cnt; i++) { 612 for (j = 0; j < table[i].len; j++) { 613 c = table[i].sym[j]; 614 best_table[c][0]=c; 615 best_table_len[c]=1; 616 } 617 } 618 } 619 620 static void optimize_token_table(void) 621 { 622 build_initial_tok_table(); 623 624 insert_real_symbols_in_table(); 625 626 /* When valid symbol is not registered, exit to error */ 627 if (!table_cnt) { 628 fprintf(stderr, "No valid symbol.\n"); 629 exit(1); 630 } 631 632 optimize_result(); 633 } 634 635 /* guess for "linker script provide" symbol */ 636 static int may_be_linker_script_provide_symbol(const struct sym_entry *se) 637 { 638 const char *symbol = (char *)se->sym + 1; 639 int len = se->len - 1; 640 641 if (len < 8) 642 return 0; 643 644 if (symbol[0] != '_' || symbol[1] != '_') 645 return 0; 646 647 /* __start_XXXXX */ 648 if (!memcmp(symbol + 2, "start_", 6)) 649 return 1; 650 651 /* __stop_XXXXX */ 652 if (!memcmp(symbol + 2, "stop_", 5)) 653 return 1; 654 655 /* __end_XXXXX */ 656 if (!memcmp(symbol + 2, "end_", 4)) 657 return 1; 658 659 /* __XXXXX_start */ 660 if (!memcmp(symbol + len - 6, "_start", 6)) 661 return 1; 662 663 /* __XXXXX_end */ 664 if (!memcmp(symbol + len - 4, "_end", 4)) 665 return 1; 666 667 return 0; 668 } 669 670 static int prefix_underscores_count(const char *str) 671 { 672 const char *tail = str; 673 674 while (*tail == '_') 675 tail++; 676 677 return tail - str; 678 } 679 680 static int compare_symbols(const void *a, const void *b) 681 { 682 const struct sym_entry *sa; 683 const struct sym_entry *sb; 684 int wa, wb; 685 686 sa = a; 687 sb = b; 688 689 /* sort by address first */ 690 if (sa->addr > sb->addr) 691 return 1; 692 if (sa->addr < sb->addr) 693 return -1; 694 695 /* sort by "weakness" type */ 696 wa = (sa->sym[0] == 'w') || (sa->sym[0] == 'W'); 697 wb = (sb->sym[0] == 'w') || (sb->sym[0] == 'W'); 698 if (wa != wb) 699 return wa - wb; 700 701 /* sort by "linker script provide" type */ 702 wa = may_be_linker_script_provide_symbol(sa); 703 wb = may_be_linker_script_provide_symbol(sb); 704 if (wa != wb) 705 return wa - wb; 706 707 /* sort by the number of prefix underscores */ 708 wa = prefix_underscores_count((const char *)sa->sym + 1); 709 wb = prefix_underscores_count((const char *)sb->sym + 1); 710 if (wa != wb) 711 return wa - wb; 712 713 /* sort by initial order, so that other symbols are left undisturbed */ 714 return sa->start_pos - sb->start_pos; 715 } 716 717 static void sort_symbols(void) 718 { 719 qsort(table, table_cnt, sizeof(struct sym_entry), compare_symbols); 720 } 721 722 static void make_percpus_absolute(void) 723 { 724 unsigned int i; 725 726 for (i = 0; i < table_cnt; i++) 727 if (symbol_in_range(&table[i], &percpu_range, 1)) { 728 /* 729 * Keep the 'A' override for percpu symbols to 730 * ensure consistent behavior compared to older 731 * versions of this tool. 732 */ 733 table[i].sym[0] = 'A'; 734 table[i].percpu_absolute = 1; 735 } 736 } 737 738 /* find the minimum non-absolute symbol address */ 739 static void record_relative_base(void) 740 { 741 unsigned int i; 742 743 relative_base = -1ULL; 744 for (i = 0; i < table_cnt; i++) 745 if (!symbol_absolute(&table[i]) && 746 table[i].addr < relative_base) 747 relative_base = table[i].addr; 748 } 749 750 int main(int argc, char **argv) 751 { 752 if (argc >= 2) { 753 int i; 754 for (i = 1; i < argc; i++) { 755 if(strcmp(argv[i], "--all-symbols") == 0) 756 all_symbols = 1; 757 else if (strcmp(argv[i], "--absolute-percpu") == 0) 758 absolute_percpu = 1; 759 else if (strncmp(argv[i], "--symbol-prefix=", 16) == 0) { 760 char *p = &argv[i][16]; 761 /* skip quote */ 762 if ((*p == '"' && *(p+2) == '"') || (*p == '\'' && *(p+2) == '\'')) 763 p++; 764 symbol_prefix_char = *p; 765 } else if (strcmp(argv[i], "--base-relative") == 0) 766 base_relative = 1; 767 else 768 usage(); 769 } 770 } else if (argc != 1) 771 usage(); 772 773 read_map(stdin); 774 if (absolute_percpu) 775 make_percpus_absolute(); 776 if (base_relative) 777 record_relative_base(); 778 sort_symbols(); 779 optimize_token_table(); 780 write_src(); 781 782 return 0; 783 } 784