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: kallsyms [--all-symbols] [--absolute-percpu] in.map > out.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 <errno.h> 22 #include <getopt.h> 23 #include <stdbool.h> 24 #include <stdio.h> 25 #include <stdlib.h> 26 #include <string.h> 27 #include <ctype.h> 28 #include <limits.h> 29 30 #include <xalloc.h> 31 32 #define ARRAY_SIZE(arr) (sizeof(arr) / sizeof(arr[0])) 33 34 #define KSYM_NAME_LEN 512 35 36 struct sym_entry { 37 unsigned long long addr; 38 unsigned int len; 39 unsigned int seq; 40 bool percpu_absolute; 41 unsigned char sym[]; 42 }; 43 44 struct addr_range { 45 const char *start_sym, *end_sym; 46 unsigned long long start, end; 47 }; 48 49 static unsigned long long _text; 50 static unsigned long long relative_base; 51 static struct addr_range text_ranges[] = { 52 { "_stext", "_etext" }, 53 { "_sinittext", "_einittext" }, 54 }; 55 #define text_range_text (&text_ranges[0]) 56 #define text_range_inittext (&text_ranges[1]) 57 58 static struct addr_range percpu_range = { 59 "__per_cpu_start", "__per_cpu_end", -1ULL, 0 60 }; 61 62 static struct sym_entry **table; 63 static unsigned int table_size, table_cnt; 64 static int all_symbols; 65 static int absolute_percpu; 66 67 static int token_profit[0x10000]; 68 69 /* the table that holds the result of the compression */ 70 static unsigned char best_table[256][2]; 71 static unsigned char best_table_len[256]; 72 73 74 static void usage(void) 75 { 76 fprintf(stderr, "Usage: kallsyms [--all-symbols] [--absolute-percpu] in.map > out.S\n"); 77 exit(1); 78 } 79 80 static char *sym_name(const struct sym_entry *s) 81 { 82 return (char *)s->sym + 1; 83 } 84 85 static bool is_ignored_symbol(const char *name, char type) 86 { 87 if (type == 'u' || type == 'n') 88 return true; 89 90 if (toupper(type) == 'A') { 91 /* Keep these useful absolute symbols */ 92 if (strcmp(name, "__kernel_syscall_via_break") && 93 strcmp(name, "__kernel_syscall_via_epc") && 94 strcmp(name, "__kernel_sigtramp") && 95 strcmp(name, "__gp")) 96 return true; 97 } 98 99 return false; 100 } 101 102 static void check_symbol_range(const char *sym, unsigned long long addr, 103 struct addr_range *ranges, int entries) 104 { 105 size_t i; 106 struct addr_range *ar; 107 108 for (i = 0; i < entries; ++i) { 109 ar = &ranges[i]; 110 111 if (strcmp(sym, ar->start_sym) == 0) { 112 ar->start = addr; 113 return; 114 } else if (strcmp(sym, ar->end_sym) == 0) { 115 ar->end = addr; 116 return; 117 } 118 } 119 } 120 121 static struct sym_entry *read_symbol(FILE *in, char **buf, size_t *buf_len) 122 { 123 char *name, type, *p; 124 unsigned long long addr; 125 size_t len; 126 ssize_t readlen; 127 struct sym_entry *sym; 128 129 errno = 0; 130 readlen = getline(buf, buf_len, in); 131 if (readlen < 0) { 132 if (errno) { 133 perror("read_symbol"); 134 exit(EXIT_FAILURE); 135 } 136 return NULL; 137 } 138 139 if ((*buf)[readlen - 1] == '\n') 140 (*buf)[readlen - 1] = 0; 141 142 addr = strtoull(*buf, &p, 16); 143 144 if (*buf == p || *p++ != ' ' || !isascii((type = *p++)) || *p++ != ' ') { 145 fprintf(stderr, "line format error\n"); 146 exit(EXIT_FAILURE); 147 } 148 149 name = p; 150 len = strlen(name); 151 152 if (len >= KSYM_NAME_LEN) { 153 fprintf(stderr, "Symbol %s too long for kallsyms (%zu >= %d).\n" 154 "Please increase KSYM_NAME_LEN both in kernel and kallsyms.c\n", 155 name, len, KSYM_NAME_LEN); 156 return NULL; 157 } 158 159 if (strcmp(name, "_text") == 0) 160 _text = addr; 161 162 /* Ignore most absolute/undefined (?) symbols. */ 163 if (is_ignored_symbol(name, type)) 164 return NULL; 165 166 check_symbol_range(name, addr, text_ranges, ARRAY_SIZE(text_ranges)); 167 check_symbol_range(name, addr, &percpu_range, 1); 168 169 /* include the type field in the symbol name, so that it gets 170 * compressed together */ 171 len++; 172 173 sym = xmalloc(sizeof(*sym) + len + 1); 174 sym->addr = addr; 175 sym->len = len; 176 sym->sym[0] = type; 177 strcpy(sym_name(sym), name); 178 sym->percpu_absolute = false; 179 180 return sym; 181 } 182 183 static int symbol_in_range(const struct sym_entry *s, 184 const struct addr_range *ranges, int entries) 185 { 186 size_t i; 187 const struct addr_range *ar; 188 189 for (i = 0; i < entries; ++i) { 190 ar = &ranges[i]; 191 192 if (s->addr >= ar->start && s->addr <= ar->end) 193 return 1; 194 } 195 196 return 0; 197 } 198 199 static bool string_starts_with(const char *s, const char *prefix) 200 { 201 return strncmp(s, prefix, strlen(prefix)) == 0; 202 } 203 204 static int symbol_valid(const struct sym_entry *s) 205 { 206 const char *name = sym_name(s); 207 208 /* if --all-symbols is not specified, then symbols outside the text 209 * and inittext sections are discarded */ 210 if (!all_symbols) { 211 /* 212 * Symbols starting with __start and __stop are used to denote 213 * section boundaries, and should always be included: 214 */ 215 if (string_starts_with(name, "__start_") || 216 string_starts_with(name, "__stop_")) 217 return 1; 218 219 if (symbol_in_range(s, text_ranges, 220 ARRAY_SIZE(text_ranges)) == 0) 221 return 0; 222 /* Corner case. Discard any symbols with the same value as 223 * _etext _einittext; they can move between pass 1 and 2 when 224 * the kallsyms data are added. If these symbols move then 225 * they may get dropped in pass 2, which breaks the kallsyms 226 * rules. 227 */ 228 if ((s->addr == text_range_text->end && 229 strcmp(name, text_range_text->end_sym)) || 230 (s->addr == text_range_inittext->end && 231 strcmp(name, text_range_inittext->end_sym))) 232 return 0; 233 } 234 235 return 1; 236 } 237 238 /* remove all the invalid symbols from the table */ 239 static void shrink_table(void) 240 { 241 unsigned int i, pos; 242 243 pos = 0; 244 for (i = 0; i < table_cnt; i++) { 245 if (symbol_valid(table[i])) { 246 if (pos != i) 247 table[pos] = table[i]; 248 pos++; 249 } else { 250 free(table[i]); 251 } 252 } 253 table_cnt = pos; 254 } 255 256 static void read_map(const char *in) 257 { 258 FILE *fp; 259 struct sym_entry *sym; 260 char *buf = NULL; 261 size_t buflen = 0; 262 263 fp = fopen(in, "r"); 264 if (!fp) { 265 perror(in); 266 exit(1); 267 } 268 269 while (!feof(fp)) { 270 sym = read_symbol(fp, &buf, &buflen); 271 if (!sym) 272 continue; 273 274 sym->seq = table_cnt; 275 276 if (table_cnt >= table_size) { 277 table_size += 10000; 278 table = xrealloc(table, sizeof(*table) * table_size); 279 } 280 281 table[table_cnt++] = sym; 282 } 283 284 free(buf); 285 fclose(fp); 286 } 287 288 static void output_label(const 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(const 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 bool symbol_absolute(const struct sym_entry *s) 323 { 324 return s->percpu_absolute; 325 } 326 327 static int compare_names(const void *a, const void *b) 328 { 329 int ret; 330 const struct sym_entry *sa = *(const struct sym_entry **)a; 331 const struct sym_entry *sb = *(const struct sym_entry **)b; 332 333 ret = strcmp(sym_name(sa), sym_name(sb)); 334 if (!ret) { 335 if (sa->addr > sb->addr) 336 return 1; 337 else if (sa->addr < sb->addr) 338 return -1; 339 340 /* keep old order */ 341 return (int)(sa->seq - sb->seq); 342 } 343 344 return ret; 345 } 346 347 static void sort_symbols_by_name(void) 348 { 349 qsort(table, table_cnt, sizeof(table[0]), compare_names); 350 } 351 352 static void write_src(void) 353 { 354 unsigned int i, k, off; 355 unsigned int best_idx[256]; 356 unsigned int *markers, markers_cnt; 357 char buf[KSYM_NAME_LEN]; 358 359 printf("#include <asm/bitsperlong.h>\n"); 360 printf("#if BITS_PER_LONG == 64\n"); 361 printf("#define PTR .quad\n"); 362 printf("#define ALGN .balign 8\n"); 363 printf("#else\n"); 364 printf("#define PTR .long\n"); 365 printf("#define ALGN .balign 4\n"); 366 printf("#endif\n"); 367 368 printf("\t.section .rodata, \"a\"\n"); 369 370 output_label("kallsyms_num_syms"); 371 printf("\t.long\t%u\n", table_cnt); 372 printf("\n"); 373 374 /* table of offset markers, that give the offset in the compressed stream 375 * every 256 symbols */ 376 markers_cnt = (table_cnt + 255) / 256; 377 markers = xmalloc(sizeof(*markers) * markers_cnt); 378 379 output_label("kallsyms_names"); 380 off = 0; 381 for (i = 0; i < table_cnt; i++) { 382 if ((i & 0xFF) == 0) 383 markers[i >> 8] = off; 384 table[i]->seq = i; 385 386 /* There cannot be any symbol of length zero. */ 387 if (table[i]->len == 0) { 388 fprintf(stderr, "kallsyms failure: " 389 "unexpected zero symbol length\n"); 390 exit(EXIT_FAILURE); 391 } 392 393 /* Only lengths that fit in up-to-two-byte ULEB128 are supported. */ 394 if (table[i]->len > 0x3FFF) { 395 fprintf(stderr, "kallsyms failure: " 396 "unexpected huge symbol length\n"); 397 exit(EXIT_FAILURE); 398 } 399 400 /* Encode length with ULEB128. */ 401 if (table[i]->len <= 0x7F) { 402 /* Most symbols use a single byte for the length. */ 403 printf("\t.byte 0x%02x", table[i]->len); 404 off += table[i]->len + 1; 405 } else { 406 /* "Big" symbols use two bytes. */ 407 printf("\t.byte 0x%02x, 0x%02x", 408 (table[i]->len & 0x7F) | 0x80, 409 (table[i]->len >> 7) & 0x7F); 410 off += table[i]->len + 2; 411 } 412 for (k = 0; k < table[i]->len; k++) 413 printf(", 0x%02x", table[i]->sym[k]); 414 415 /* 416 * Now that we wrote out the compressed symbol name, restore the 417 * original name and print it in the comment. 418 */ 419 expand_symbol(table[i]->sym, table[i]->len, buf); 420 strcpy((char *)table[i]->sym, buf); 421 printf("\t/* %s */\n", table[i]->sym); 422 } 423 printf("\n"); 424 425 output_label("kallsyms_markers"); 426 for (i = 0; i < markers_cnt; i++) 427 printf("\t.long\t%u\n", markers[i]); 428 printf("\n"); 429 430 free(markers); 431 432 output_label("kallsyms_token_table"); 433 off = 0; 434 for (i = 0; i < 256; i++) { 435 best_idx[i] = off; 436 expand_symbol(best_table[i], best_table_len[i], buf); 437 printf("\t.asciz\t\"%s\"\n", buf); 438 off += strlen(buf) + 1; 439 } 440 printf("\n"); 441 442 output_label("kallsyms_token_index"); 443 for (i = 0; i < 256; i++) 444 printf("\t.short\t%d\n", best_idx[i]); 445 printf("\n"); 446 447 output_label("kallsyms_offsets"); 448 449 for (i = 0; i < table_cnt; i++) { 450 /* 451 * Use the offset relative to the lowest value 452 * encountered of all relative symbols, and emit 453 * non-relocatable fixed offsets that will be fixed 454 * up at runtime. 455 */ 456 457 long long offset; 458 bool overflow; 459 460 if (!absolute_percpu) { 461 offset = table[i]->addr - relative_base; 462 overflow = offset < 0 || offset > UINT_MAX; 463 } else if (symbol_absolute(table[i])) { 464 offset = table[i]->addr; 465 overflow = offset < 0 || offset > INT_MAX; 466 } else { 467 offset = relative_base - table[i]->addr - 1; 468 overflow = offset < INT_MIN || offset >= 0; 469 } 470 if (overflow) { 471 fprintf(stderr, "kallsyms failure: " 472 "%s symbol value %#llx out of range in relative mode\n", 473 symbol_absolute(table[i]) ? "absolute" : "relative", 474 table[i]->addr); 475 exit(EXIT_FAILURE); 476 } 477 printf("\t.long\t%#x\t/* %s */\n", (int)offset, table[i]->sym); 478 } 479 printf("\n"); 480 481 output_label("kallsyms_relative_base"); 482 /* Provide proper symbols relocatability by their '_text' relativeness. */ 483 if (_text <= relative_base) 484 printf("\tPTR\t_text + %#llx\n", relative_base - _text); 485 else 486 printf("\tPTR\t_text - %#llx\n", _text - relative_base); 487 printf("\n"); 488 489 sort_symbols_by_name(); 490 output_label("kallsyms_seqs_of_names"); 491 for (i = 0; i < table_cnt; i++) 492 printf("\t.byte 0x%02x, 0x%02x, 0x%02x\t/* %s */\n", 493 (unsigned char)(table[i]->seq >> 16), 494 (unsigned char)(table[i]->seq >> 8), 495 (unsigned char)(table[i]->seq >> 0), 496 table[i]->sym); 497 printf("\n"); 498 } 499 500 501 /* table lookup compression functions */ 502 503 /* count all the possible tokens in a symbol */ 504 static void learn_symbol(const unsigned char *symbol, int len) 505 { 506 int i; 507 508 for (i = 0; i < len - 1; i++) 509 token_profit[ symbol[i] + (symbol[i + 1] << 8) ]++; 510 } 511 512 /* decrease the count for all the possible tokens in a symbol */ 513 static void forget_symbol(const unsigned char *symbol, int len) 514 { 515 int i; 516 517 for (i = 0; i < len - 1; i++) 518 token_profit[ symbol[i] + (symbol[i + 1] << 8) ]--; 519 } 520 521 /* do the initial token count */ 522 static void build_initial_token_table(void) 523 { 524 unsigned int i; 525 526 for (i = 0; i < table_cnt; i++) 527 learn_symbol(table[i]->sym, table[i]->len); 528 } 529 530 static unsigned char *find_token(unsigned char *str, int len, 531 const unsigned char *token) 532 { 533 int i; 534 535 for (i = 0; i < len - 1; i++) { 536 if (str[i] == token[0] && str[i+1] == token[1]) 537 return &str[i]; 538 } 539 return NULL; 540 } 541 542 /* replace a given token in all the valid symbols. Use the sampled symbols 543 * to update the counts */ 544 static void compress_symbols(const unsigned char *str, int idx) 545 { 546 unsigned int i, len, size; 547 unsigned char *p1, *p2; 548 549 for (i = 0; i < table_cnt; i++) { 550 551 len = table[i]->len; 552 p1 = table[i]->sym; 553 554 /* find the token on the symbol */ 555 p2 = find_token(p1, len, str); 556 if (!p2) continue; 557 558 /* decrease the counts for this symbol's tokens */ 559 forget_symbol(table[i]->sym, len); 560 561 size = len; 562 563 do { 564 *p2 = idx; 565 p2++; 566 size -= (p2 - p1); 567 memmove(p2, p2 + 1, size); 568 p1 = p2; 569 len--; 570 571 if (size < 2) break; 572 573 /* find the token on the symbol */ 574 p2 = find_token(p1, size, str); 575 576 } while (p2); 577 578 table[i]->len = len; 579 580 /* increase the counts for this symbol's new tokens */ 581 learn_symbol(table[i]->sym, len); 582 } 583 } 584 585 /* search the token with the maximum profit */ 586 static int find_best_token(void) 587 { 588 int i, best, bestprofit; 589 590 bestprofit=-10000; 591 best = 0; 592 593 for (i = 0; i < 0x10000; i++) { 594 if (token_profit[i] > bestprofit) { 595 best = i; 596 bestprofit = token_profit[i]; 597 } 598 } 599 return best; 600 } 601 602 /* this is the core of the algorithm: calculate the "best" table */ 603 static void optimize_result(void) 604 { 605 int i, best; 606 607 /* using the '\0' symbol last allows compress_symbols to use standard 608 * fast string functions */ 609 for (i = 255; i >= 0; i--) { 610 611 /* if this table slot is empty (it is not used by an actual 612 * original char code */ 613 if (!best_table_len[i]) { 614 615 /* find the token with the best profit value */ 616 best = find_best_token(); 617 if (token_profit[best] == 0) 618 break; 619 620 /* place it in the "best" table */ 621 best_table_len[i] = 2; 622 best_table[i][0] = best & 0xFF; 623 best_table[i][1] = (best >> 8) & 0xFF; 624 625 /* replace this token in all the valid symbols */ 626 compress_symbols(best_table[i], i); 627 } 628 } 629 } 630 631 /* start by placing the symbols that are actually used on the table */ 632 static void insert_real_symbols_in_table(void) 633 { 634 unsigned int i, j, c; 635 636 for (i = 0; i < table_cnt; i++) { 637 for (j = 0; j < table[i]->len; j++) { 638 c = table[i]->sym[j]; 639 best_table[c][0]=c; 640 best_table_len[c]=1; 641 } 642 } 643 } 644 645 static void optimize_token_table(void) 646 { 647 build_initial_token_table(); 648 649 insert_real_symbols_in_table(); 650 651 optimize_result(); 652 } 653 654 /* guess for "linker script provide" symbol */ 655 static int may_be_linker_script_provide_symbol(const struct sym_entry *se) 656 { 657 const char *symbol = sym_name(se); 658 int len = se->len - 1; 659 660 if (len < 8) 661 return 0; 662 663 if (symbol[0] != '_' || symbol[1] != '_') 664 return 0; 665 666 /* __start_XXXXX */ 667 if (!memcmp(symbol + 2, "start_", 6)) 668 return 1; 669 670 /* __stop_XXXXX */ 671 if (!memcmp(symbol + 2, "stop_", 5)) 672 return 1; 673 674 /* __end_XXXXX */ 675 if (!memcmp(symbol + 2, "end_", 4)) 676 return 1; 677 678 /* __XXXXX_start */ 679 if (!memcmp(symbol + len - 6, "_start", 6)) 680 return 1; 681 682 /* __XXXXX_end */ 683 if (!memcmp(symbol + len - 4, "_end", 4)) 684 return 1; 685 686 return 0; 687 } 688 689 static int compare_symbols(const void *a, const void *b) 690 { 691 const struct sym_entry *sa = *(const struct sym_entry **)a; 692 const struct sym_entry *sb = *(const struct sym_entry **)b; 693 int wa, wb; 694 695 /* sort by address first */ 696 if (sa->addr > sb->addr) 697 return 1; 698 if (sa->addr < sb->addr) 699 return -1; 700 701 /* sort by "weakness" type */ 702 wa = (sa->sym[0] == 'w') || (sa->sym[0] == 'W'); 703 wb = (sb->sym[0] == 'w') || (sb->sym[0] == 'W'); 704 if (wa != wb) 705 return wa - wb; 706 707 /* sort by "linker script provide" type */ 708 wa = may_be_linker_script_provide_symbol(sa); 709 wb = may_be_linker_script_provide_symbol(sb); 710 if (wa != wb) 711 return wa - wb; 712 713 /* sort by the number of prefix underscores */ 714 wa = strspn(sym_name(sa), "_"); 715 wb = strspn(sym_name(sb), "_"); 716 if (wa != wb) 717 return wa - wb; 718 719 /* sort by initial order, so that other symbols are left undisturbed */ 720 return sa->seq - sb->seq; 721 } 722 723 static void sort_symbols(void) 724 { 725 qsort(table, table_cnt, sizeof(table[0]), compare_symbols); 726 } 727 728 static void make_percpus_absolute(void) 729 { 730 unsigned int i; 731 732 for (i = 0; i < table_cnt; i++) 733 if (symbol_in_range(table[i], &percpu_range, 1)) { 734 /* 735 * Keep the 'A' override for percpu symbols to 736 * ensure consistent behavior compared to older 737 * versions of this tool. 738 */ 739 table[i]->sym[0] = 'A'; 740 table[i]->percpu_absolute = true; 741 } 742 } 743 744 /* find the minimum non-absolute symbol address */ 745 static void record_relative_base(void) 746 { 747 unsigned int i; 748 749 for (i = 0; i < table_cnt; i++) 750 if (!symbol_absolute(table[i])) { 751 /* 752 * The table is sorted by address. 753 * Take the first non-absolute symbol value. 754 */ 755 relative_base = table[i]->addr; 756 return; 757 } 758 } 759 760 int main(int argc, char **argv) 761 { 762 while (1) { 763 static const struct option long_options[] = { 764 {"all-symbols", no_argument, &all_symbols, 1}, 765 {"absolute-percpu", no_argument, &absolute_percpu, 1}, 766 {}, 767 }; 768 769 int c = getopt_long(argc, argv, "", long_options, NULL); 770 771 if (c == -1) 772 break; 773 if (c != 0) 774 usage(); 775 } 776 777 if (optind >= argc) 778 usage(); 779 780 read_map(argv[optind]); 781 shrink_table(); 782 if (absolute_percpu) 783 make_percpus_absolute(); 784 sort_symbols(); 785 record_relative_base(); 786 optimize_token_table(); 787 write_src(); 788 789 return 0; 790 } 791