1 /* Generate assembler source containing symbol information
3 * Copyright 2002 by Kai Germaschewski
5 * This software may be used and distributed according to the terms
6 * of the GNU General Public License, incorporated herein by reference.
8 * Usage: nm -n vmlinux | scripts/kallsyms [--all-symbols] > symbols.S
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
28 #define ARRAY_SIZE(arr) (sizeof(arr) / sizeof(arr[0]))
31 #define KSYM_NAME_LEN 128
34 unsigned long long addr;
36 unsigned int start_pos;
38 unsigned int percpu_absolute;
42 const char *start_sym, *end_sym;
43 unsigned long long start, end;
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 */
54 #define text_range_text (&text_ranges[0])
55 #define text_range_inittext (&text_ranges[1])
57 static struct addr_range percpu_range = {
58 "__per_cpu_start", "__per_cpu_end", -1ULL, 0
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 int base_relative = 0;
67 int token_profit[0x10000];
69 /* the table that holds the result of the compression */
70 unsigned char best_table[256][2];
71 unsigned char best_table_len[256];
74 static void usage(void)
76 fprintf(stderr, "Usage: kallsyms [--all-symbols] "
77 "[--base-relative] < in.map > out.S\n");
82 * This ignores the intensely annoying "mapping symbols" found
83 * in ARM ELF files: $a, $t and $d.
85 static inline int is_arm_mapping_symbol(const char *str)
87 return str[0] == '$' && strchr("axtd", str[1])
88 && (str[2] == '\0' || str[2] == '.');
91 static int check_symbol_range(const char *sym, unsigned long long addr,
92 struct addr_range *ranges, int entries)
95 struct addr_range *ar;
97 for (i = 0; i < entries; ++i) {
100 if (strcmp(sym, ar->start_sym) == 0) {
103 } else if (strcmp(sym, ar->end_sym) == 0) {
112 static int read_symbol(FILE *in, struct sym_entry *s)
114 char sym[500], stype;
117 rc = fscanf(in, "%llx %c %499s\n", &s->addr, &stype, sym);
119 if (rc != EOF && fgets(sym, 500, in) == NULL)
120 fprintf(stderr, "Read error or end of file.\n");
123 if (strlen(sym) >= KSYM_NAME_LEN) {
124 fprintf(stderr, "Symbol %s too long for kallsyms (%zu >= %d).\n"
125 "Please increase KSYM_NAME_LEN both in kernel and kallsyms.c\n",
126 sym, strlen(sym), KSYM_NAME_LEN);
130 /* Ignore most absolute/undefined (?) symbols. */
131 if (strcmp(sym, "_text") == 0)
133 else if (check_symbol_range(sym, s->addr, text_ranges,
134 ARRAY_SIZE(text_ranges)) == 0)
136 else if (toupper(stype) == 'A')
138 /* Keep these useful absolute symbols */
139 if (strcmp(sym, "__kernel_syscall_via_break") &&
140 strcmp(sym, "__kernel_syscall_via_epc") &&
141 strcmp(sym, "__kernel_sigtramp") &&
146 else if (toupper(stype) == 'U' ||
147 is_arm_mapping_symbol(sym))
149 /* exclude also MIPS ELF local symbols ($L123 instead of .L123) */
150 else if (sym[0] == '$')
152 /* exclude debugging symbols */
153 else if (stype == 'N' || stype == 'n')
155 /* exclude s390 kasan local symbols */
156 else if (!strncmp(sym, ".LASANPC", 8))
159 /* include the type field in the symbol name, so that it gets
160 * compressed together */
161 s->len = strlen(sym) + 1;
162 s->sym = malloc(s->len + 1);
164 fprintf(stderr, "kallsyms failure: "
165 "unable to allocate required amount of memory\n");
168 strcpy((char *)s->sym + 1, sym);
171 s->percpu_absolute = 0;
173 /* Record if we've found __per_cpu_start/end. */
174 check_symbol_range(sym, s->addr, &percpu_range, 1);
179 static int symbol_in_range(struct sym_entry *s, struct addr_range *ranges,
183 struct addr_range *ar;
185 for (i = 0; i < entries; ++i) {
188 if (s->addr >= ar->start && s->addr <= ar->end)
195 static int symbol_valid(struct sym_entry *s)
197 /* Symbols which vary between passes. Passes 1 and 2 must have
198 * identical symbol lists. The kallsyms_* symbols below are only added
199 * after pass 1, they would be included in pass 2 when --all-symbols is
200 * specified so exclude them to get a stable symbol list.
202 static char *special_symbols[] = {
203 "kallsyms_addresses",
205 "kallsyms_relative_base",
209 "kallsyms_token_table",
210 "kallsyms_token_index",
212 /* Exclude linker generated symbols which vary between passes */
213 "_SDA_BASE_", /* ppc */
214 "_SDA2_BASE_", /* ppc */
217 static char *special_prefixes[] = {
218 "__crc_", /* modversions */
219 "__efistub_", /* arm64 EFI stub namespace */
222 static char *special_suffixes[] = {
224 "_from_arm", /* arm */
225 "_from_thumb", /* arm */
229 char *sym_name = (char *)s->sym + 1;
231 /* if --all-symbols is not specified, then symbols outside the text
232 * and inittext sections are discarded */
234 if (symbol_in_range(s, text_ranges,
235 ARRAY_SIZE(text_ranges)) == 0)
237 /* Corner case. Discard any symbols with the same value as
238 * _etext _einittext; they can move between pass 1 and 2 when
239 * the kallsyms data are added. If these symbols move then
240 * they may get dropped in pass 2, which breaks the kallsyms
243 if ((s->addr == text_range_text->end &&
245 text_range_text->end_sym)) ||
246 (s->addr == text_range_inittext->end &&
248 text_range_inittext->end_sym)))
252 /* Exclude symbols which vary between passes. */
253 for (i = 0; special_symbols[i]; i++)
254 if (strcmp(sym_name, special_symbols[i]) == 0)
257 for (i = 0; special_prefixes[i]; i++) {
258 int l = strlen(special_prefixes[i]);
260 if (l <= strlen(sym_name) &&
261 strncmp(sym_name, special_prefixes[i], l) == 0)
265 for (i = 0; special_suffixes[i]; i++) {
266 int l = strlen(sym_name) - strlen(special_suffixes[i]);
268 if (l >= 0 && strcmp(sym_name + l, special_suffixes[i]) == 0)
275 static void read_map(FILE *in)
278 if (table_cnt >= table_size) {
280 table = realloc(table, sizeof(*table) * table_size);
282 fprintf(stderr, "out of memory\n");
286 if (read_symbol(in, &table[table_cnt]) == 0) {
287 table[table_cnt].start_pos = table_cnt;
293 static void output_label(char *label)
295 printf(".globl %s\n", label);
297 printf("%s:\n", label);
300 /* uncompress a compressed symbol. When this function is called, the best table
301 * might still be compressed itself, so the function needs to be recursive */
302 static int expand_symbol(unsigned char *data, int len, char *result)
304 int c, rlen, total=0;
308 /* if the table holds a single char that is the same as the one
309 * we are looking for, then end the search */
310 if (best_table[c][0]==c && best_table_len[c]==1) {
314 /* if not, recurse and expand */
315 rlen = expand_symbol(best_table[c], best_table_len[c], result);
327 static int symbol_absolute(struct sym_entry *s)
329 return s->percpu_absolute;
332 static void write_src(void)
334 unsigned int i, k, off;
335 unsigned int best_idx[256];
336 unsigned int *markers;
337 char buf[KSYM_NAME_LEN];
339 printf("#include <asm/types.h>\n");
340 printf("#if BITS_PER_LONG == 64\n");
341 printf("#define PTR .quad\n");
342 printf("#define ALGN .align 8\n");
344 printf("#define PTR .long\n");
345 printf("#define ALGN .align 4\n");
348 printf("\t.section .rodata, \"a\"\n");
350 /* Provide proper symbols relocatability by their relativeness
351 * to a fixed anchor point in the runtime image, either '_text'
352 * for absolute address tables, in which case the linker will
353 * emit the final addresses at build time. Otherwise, use the
354 * offset relative to the lowest value encountered of all relative
355 * symbols, and emit non-relocatable fixed offsets that will be fixed
358 * The symbol names cannot be used to construct normal symbol
359 * references as the list of symbols contains symbols that are
360 * declared static and are private to their .o files. This prevents
361 * .tmp_kallsyms.o or any other object from referencing them.
364 output_label("kallsyms_addresses");
366 output_label("kallsyms_offsets");
368 for (i = 0; i < table_cnt; i++) {
373 if (!absolute_percpu) {
374 offset = table[i].addr - relative_base;
375 overflow = (offset < 0 || offset > UINT_MAX);
376 } else if (symbol_absolute(&table[i])) {
377 offset = table[i].addr;
378 overflow = (offset < 0 || offset > INT_MAX);
380 offset = relative_base - table[i].addr - 1;
381 overflow = (offset < INT_MIN || offset >= 0);
384 fprintf(stderr, "kallsyms failure: "
385 "%s symbol value %#llx out of range in relative mode\n",
386 symbol_absolute(&table[i]) ? "absolute" : "relative",
390 printf("\t.long\t%#x\n", (int)offset);
391 } else if (!symbol_absolute(&table[i])) {
392 if (_text <= table[i].addr)
393 printf("\tPTR\t_text + %#llx\n",
394 table[i].addr - _text);
396 printf("\tPTR\t_text - %#llx\n",
397 _text - table[i].addr);
399 printf("\tPTR\t%#llx\n", table[i].addr);
405 output_label("kallsyms_relative_base");
406 printf("\tPTR\t_text - %#llx\n", _text - relative_base);
410 output_label("kallsyms_num_syms");
411 printf("\tPTR\t%u\n", table_cnt);
414 /* table of offset markers, that give the offset in the compressed stream
415 * every 256 symbols */
416 markers = malloc(sizeof(unsigned int) * ((table_cnt + 255) / 256));
418 fprintf(stderr, "kallsyms failure: "
419 "unable to allocate required memory\n");
423 output_label("kallsyms_names");
425 for (i = 0; i < table_cnt; i++) {
427 markers[i >> 8] = off;
429 printf("\t.byte 0x%02x", table[i].len);
430 for (k = 0; k < table[i].len; k++)
431 printf(", 0x%02x", table[i].sym[k]);
434 off += table[i].len + 1;
438 output_label("kallsyms_markers");
439 for (i = 0; i < ((table_cnt + 255) >> 8); i++)
440 printf("\tPTR\t%d\n", markers[i]);
445 output_label("kallsyms_token_table");
447 for (i = 0; i < 256; i++) {
449 expand_symbol(best_table[i], best_table_len[i], buf);
450 printf("\t.asciz\t\"%s\"\n", buf);
451 off += strlen(buf) + 1;
455 output_label("kallsyms_token_index");
456 for (i = 0; i < 256; i++)
457 printf("\t.short\t%d\n", best_idx[i]);
462 /* table lookup compression functions */
464 /* count all the possible tokens in a symbol */
465 static void learn_symbol(unsigned char *symbol, int len)
469 for (i = 0; i < len - 1; i++)
470 token_profit[ symbol[i] + (symbol[i + 1] << 8) ]++;
473 /* decrease the count for all the possible tokens in a symbol */
474 static void forget_symbol(unsigned char *symbol, int len)
478 for (i = 0; i < len - 1; i++)
479 token_profit[ symbol[i] + (symbol[i + 1] << 8) ]--;
482 /* remove all the invalid symbols from the table and do the initial token count */
483 static void build_initial_tok_table(void)
488 for (i = 0; i < table_cnt; i++) {
489 if ( symbol_valid(&table[i]) ) {
491 table[pos] = table[i];
492 learn_symbol(table[pos].sym, table[pos].len);
499 static void *find_token(unsigned char *str, int len, unsigned char *token)
503 for (i = 0; i < len - 1; i++) {
504 if (str[i] == token[0] && str[i+1] == token[1])
510 /* replace a given token in all the valid symbols. Use the sampled symbols
511 * to update the counts */
512 static void compress_symbols(unsigned char *str, int idx)
514 unsigned int i, len, size;
515 unsigned char *p1, *p2;
517 for (i = 0; i < table_cnt; i++) {
522 /* find the token on the symbol */
523 p2 = find_token(p1, len, str);
526 /* decrease the counts for this symbol's tokens */
527 forget_symbol(table[i].sym, len);
535 memmove(p2, p2 + 1, size);
541 /* find the token on the symbol */
542 p2 = find_token(p1, size, str);
548 /* increase the counts for this symbol's new tokens */
549 learn_symbol(table[i].sym, len);
553 /* search the token with the maximum profit */
554 static int find_best_token(void)
556 int i, best, bestprofit;
561 for (i = 0; i < 0x10000; i++) {
562 if (token_profit[i] > bestprofit) {
564 bestprofit = token_profit[i];
570 /* this is the core of the algorithm: calculate the "best" table */
571 static void optimize_result(void)
575 /* using the '\0' symbol last allows compress_symbols to use standard
576 * fast string functions */
577 for (i = 255; i >= 0; i--) {
579 /* if this table slot is empty (it is not used by an actual
580 * original char code */
581 if (!best_table_len[i]) {
583 /* find the token with the best profit value */
584 best = find_best_token();
585 if (token_profit[best] == 0)
588 /* place it in the "best" table */
589 best_table_len[i] = 2;
590 best_table[i][0] = best & 0xFF;
591 best_table[i][1] = (best >> 8) & 0xFF;
593 /* replace this token in all the valid symbols */
594 compress_symbols(best_table[i], i);
599 /* start by placing the symbols that are actually used on the table */
600 static void insert_real_symbols_in_table(void)
602 unsigned int i, j, c;
604 memset(best_table, 0, sizeof(best_table));
605 memset(best_table_len, 0, sizeof(best_table_len));
607 for (i = 0; i < table_cnt; i++) {
608 for (j = 0; j < table[i].len; j++) {
616 static void optimize_token_table(void)
618 build_initial_tok_table();
620 insert_real_symbols_in_table();
622 /* When valid symbol is not registered, exit to error */
624 fprintf(stderr, "No valid symbol.\n");
631 /* guess for "linker script provide" symbol */
632 static int may_be_linker_script_provide_symbol(const struct sym_entry *se)
634 const char *symbol = (char *)se->sym + 1;
635 int len = se->len - 1;
640 if (symbol[0] != '_' || symbol[1] != '_')
644 if (!memcmp(symbol + 2, "start_", 6))
648 if (!memcmp(symbol + 2, "stop_", 5))
652 if (!memcmp(symbol + 2, "end_", 4))
656 if (!memcmp(symbol + len - 6, "_start", 6))
660 if (!memcmp(symbol + len - 4, "_end", 4))
666 static int prefix_underscores_count(const char *str)
668 const char *tail = str;
676 static int compare_symbols(const void *a, const void *b)
678 const struct sym_entry *sa;
679 const struct sym_entry *sb;
685 /* sort by address first */
686 if (sa->addr > sb->addr)
688 if (sa->addr < sb->addr)
691 /* sort by "weakness" type */
692 wa = (sa->sym[0] == 'w') || (sa->sym[0] == 'W');
693 wb = (sb->sym[0] == 'w') || (sb->sym[0] == 'W');
697 /* sort by "linker script provide" type */
698 wa = may_be_linker_script_provide_symbol(sa);
699 wb = may_be_linker_script_provide_symbol(sb);
703 /* sort by the number of prefix underscores */
704 wa = prefix_underscores_count((const char *)sa->sym + 1);
705 wb = prefix_underscores_count((const char *)sb->sym + 1);
709 /* sort by initial order, so that other symbols are left undisturbed */
710 return sa->start_pos - sb->start_pos;
713 static void sort_symbols(void)
715 qsort(table, table_cnt, sizeof(struct sym_entry), compare_symbols);
718 static void make_percpus_absolute(void)
722 for (i = 0; i < table_cnt; i++)
723 if (symbol_in_range(&table[i], &percpu_range, 1)) {
725 * Keep the 'A' override for percpu symbols to
726 * ensure consistent behavior compared to older
727 * versions of this tool.
729 table[i].sym[0] = 'A';
730 table[i].percpu_absolute = 1;
734 /* find the minimum non-absolute symbol address */
735 static void record_relative_base(void)
739 relative_base = -1ULL;
740 for (i = 0; i < table_cnt; i++)
741 if (!symbol_absolute(&table[i]) &&
742 table[i].addr < relative_base)
743 relative_base = table[i].addr;
746 int main(int argc, char **argv)
750 for (i = 1; i < argc; i++) {
751 if(strcmp(argv[i], "--all-symbols") == 0)
753 else if (strcmp(argv[i], "--absolute-percpu") == 0)
755 else if (strcmp(argv[i], "--base-relative") == 0)
760 } else if (argc != 1)
765 make_percpus_absolute();
767 record_relative_base();
769 optimize_token_table();