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: kallsyms [--all-symbols] [--absolute-percpu]
9 * [--base-relative] [--lto-clang] in.map > out.S
11 * Table compression uses all the unused char codes on the symbols and
12 * maps these to the most used substrings (tokens). For instance, it might
13 * map char code 0xF7 to represent "write_" and then in every symbol where
14 * "write_" appears it can be replaced by 0xF7, saving 5 bytes.
15 * The used codes themselves are also placed in the table so that the
16 * decompresion can work without "special cases".
17 * Applied to kernel symbols, this usually produces a compression ratio
30 #define ARRAY_SIZE(arr) (sizeof(arr) / sizeof(arr[0]))
32 #define _stringify_1(x) #x
33 #define _stringify(x) _stringify_1(x)
35 #define KSYM_NAME_LEN 512
38 * A substantially bigger size than the current maximum.
40 * It cannot be defined as an expression because it gets stringified
41 * for the fscanf() format string. Therefore, a _Static_assert() is
42 * used instead to maintain the relationship with KSYM_NAME_LEN.
44 #define KSYM_NAME_LEN_BUFFER 2048
46 KSYM_NAME_LEN_BUFFER == KSYM_NAME_LEN * 4,
47 "Please keep KSYM_NAME_LEN_BUFFER in sync with KSYM_NAME_LEN"
51 unsigned long long addr;
54 unsigned int start_pos;
55 unsigned int percpu_absolute;
60 const char *start_sym, *end_sym;
61 unsigned long long start, end;
64 static unsigned long long _text;
65 static unsigned long long relative_base;
66 static struct addr_range text_ranges[] = {
67 { "_stext", "_etext" },
68 { "_sinittext", "_einittext" },
70 #define text_range_text (&text_ranges[0])
71 #define text_range_inittext (&text_ranges[1])
73 static struct addr_range percpu_range = {
74 "__per_cpu_start", "__per_cpu_end", -1ULL, 0
77 static struct sym_entry **table;
78 static unsigned int table_size, table_cnt;
79 static int all_symbols;
80 static int absolute_percpu;
81 static int base_relative;
84 static int token_profit[0x10000];
86 /* the table that holds the result of the compression */
87 static unsigned char best_table[256][2];
88 static unsigned char best_table_len[256];
91 static void usage(void)
93 fprintf(stderr, "Usage: kallsyms [--all-symbols] [--absolute-percpu] "
94 "[--base-relative] [--lto-clang] in.map > out.S\n");
98 static char *sym_name(const struct sym_entry *s)
100 return (char *)s->sym + 1;
103 static bool is_ignored_symbol(const char *name, char type)
105 if (type == 'u' || type == 'n')
108 if (toupper(type) == 'A') {
109 /* Keep these useful absolute symbols */
110 if (strcmp(name, "__kernel_syscall_via_break") &&
111 strcmp(name, "__kernel_syscall_via_epc") &&
112 strcmp(name, "__kernel_sigtramp") &&
113 strcmp(name, "__gp"))
120 static void check_symbol_range(const char *sym, unsigned long long addr,
121 struct addr_range *ranges, int entries)
124 struct addr_range *ar;
126 for (i = 0; i < entries; ++i) {
129 if (strcmp(sym, ar->start_sym) == 0) {
132 } else if (strcmp(sym, ar->end_sym) == 0) {
139 static struct sym_entry *read_symbol(FILE *in)
141 char name[KSYM_NAME_LEN_BUFFER+1], type;
142 unsigned long long addr;
144 struct sym_entry *sym;
147 rc = fscanf(in, "%llx %c %" _stringify(KSYM_NAME_LEN_BUFFER) "s\n", &addr, &type, name);
149 if (rc != EOF && fgets(name, ARRAY_SIZE(name), in) == NULL)
150 fprintf(stderr, "Read error or end of file.\n");
153 if (strlen(name) >= KSYM_NAME_LEN) {
154 fprintf(stderr, "Symbol %s too long for kallsyms (%zu >= %d).\n"
155 "Please increase KSYM_NAME_LEN both in kernel and kallsyms.c\n",
156 name, strlen(name), KSYM_NAME_LEN);
160 if (strcmp(name, "_text") == 0)
163 /* Ignore most absolute/undefined (?) symbols. */
164 if (is_ignored_symbol(name, type))
167 check_symbol_range(name, addr, text_ranges, ARRAY_SIZE(text_ranges));
168 check_symbol_range(name, addr, &percpu_range, 1);
170 /* include the type field in the symbol name, so that it gets
171 * compressed together */
173 len = strlen(name) + 1;
175 sym = malloc(sizeof(*sym) + len + 1);
177 fprintf(stderr, "kallsyms failure: "
178 "unable to allocate required amount of memory\n");
184 strcpy(sym_name(sym), name);
185 sym->percpu_absolute = 0;
190 static int symbol_in_range(const struct sym_entry *s,
191 const struct addr_range *ranges, int entries)
194 const struct addr_range *ar;
196 for (i = 0; i < entries; ++i) {
199 if (s->addr >= ar->start && s->addr <= ar->end)
206 static int symbol_valid(const struct sym_entry *s)
208 const char *name = sym_name(s);
210 /* if --all-symbols is not specified, then symbols outside the text
211 * and inittext sections are discarded */
213 if (symbol_in_range(s, text_ranges,
214 ARRAY_SIZE(text_ranges)) == 0)
216 /* Corner case. Discard any symbols with the same value as
217 * _etext _einittext; they can move between pass 1 and 2 when
218 * the kallsyms data are added. If these symbols move then
219 * they may get dropped in pass 2, which breaks the kallsyms
222 if ((s->addr == text_range_text->end &&
223 strcmp(name, text_range_text->end_sym)) ||
224 (s->addr == text_range_inittext->end &&
225 strcmp(name, text_range_inittext->end_sym)))
232 /* remove all the invalid symbols from the table */
233 static void shrink_table(void)
238 for (i = 0; i < table_cnt; i++) {
239 if (symbol_valid(table[i])) {
241 table[pos] = table[i];
249 /* When valid symbol is not registered, exit to error */
251 fprintf(stderr, "No valid symbol.\n");
256 static void read_map(const char *in)
259 struct sym_entry *sym;
268 sym = read_symbol(fp);
272 sym->start_pos = table_cnt;
274 if (table_cnt >= table_size) {
276 table = realloc(table, sizeof(*table) * table_size);
278 fprintf(stderr, "out of memory\n");
284 table[table_cnt++] = sym;
290 static void output_label(const char *label)
292 printf(".globl %s\n", label);
294 printf("%s:\n", label);
297 /* Provide proper symbols relocatability by their '_text' relativeness. */
298 static void output_address(unsigned long long addr)
301 printf("\tPTR\t_text + %#llx\n", addr - _text);
303 printf("\tPTR\t_text - %#llx\n", _text - addr);
306 /* uncompress a compressed symbol. When this function is called, the best table
307 * might still be compressed itself, so the function needs to be recursive */
308 static int expand_symbol(const unsigned char *data, int len, char *result)
310 int c, rlen, total=0;
314 /* if the table holds a single char that is the same as the one
315 * we are looking for, then end the search */
316 if (best_table[c][0]==c && best_table_len[c]==1) {
320 /* if not, recurse and expand */
321 rlen = expand_symbol(best_table[c], best_table_len[c], result);
333 static int symbol_absolute(const struct sym_entry *s)
335 return s->percpu_absolute;
338 static void cleanup_symbol_name(char *s)
349 * As above, replacing '.' with '\0' does not affect the main sorting,
350 * but it helps us with subsorting.
357 static int compare_names(const void *a, const void *b)
360 const struct sym_entry *sa = *(const struct sym_entry **)a;
361 const struct sym_entry *sb = *(const struct sym_entry **)b;
363 ret = strcmp(sym_name(sa), sym_name(sb));
365 if (sa->addr > sb->addr)
367 else if (sa->addr < sb->addr)
371 return (int)(sa->seq - sb->seq);
377 static void sort_symbols_by_name(void)
379 qsort(table, table_cnt, sizeof(table[0]), compare_names);
382 static void write_src(void)
384 unsigned int i, k, off;
385 unsigned int best_idx[256];
386 unsigned int *markers;
387 char buf[KSYM_NAME_LEN];
389 printf("#include <asm/bitsperlong.h>\n");
390 printf("#if BITS_PER_LONG == 64\n");
391 printf("#define PTR .quad\n");
392 printf("#define ALGN .balign 8\n");
394 printf("#define PTR .long\n");
395 printf("#define ALGN .balign 4\n");
398 printf("\t.section .rodata, \"a\"\n");
400 output_label("kallsyms_num_syms");
401 printf("\t.long\t%u\n", table_cnt);
404 /* table of offset markers, that give the offset in the compressed stream
405 * every 256 symbols */
406 markers = malloc(sizeof(unsigned int) * ((table_cnt + 255) / 256));
408 fprintf(stderr, "kallsyms failure: "
409 "unable to allocate required memory\n");
413 output_label("kallsyms_names");
415 for (i = 0; i < table_cnt; i++) {
417 markers[i >> 8] = off;
420 /* There cannot be any symbol of length zero. */
421 if (table[i]->len == 0) {
422 fprintf(stderr, "kallsyms failure: "
423 "unexpected zero symbol length\n");
427 /* Only lengths that fit in up-to-two-byte ULEB128 are supported. */
428 if (table[i]->len > 0x3FFF) {
429 fprintf(stderr, "kallsyms failure: "
430 "unexpected huge symbol length\n");
434 /* Encode length with ULEB128. */
435 if (table[i]->len <= 0x7F) {
436 /* Most symbols use a single byte for the length. */
437 printf("\t.byte 0x%02x", table[i]->len);
438 off += table[i]->len + 1;
440 /* "Big" symbols use two bytes. */
441 printf("\t.byte 0x%02x, 0x%02x",
442 (table[i]->len & 0x7F) | 0x80,
443 (table[i]->len >> 7) & 0x7F);
444 off += table[i]->len + 2;
446 for (k = 0; k < table[i]->len; k++)
447 printf(", 0x%02x", table[i]->sym[k]);
453 * Now that we wrote out the compressed symbol names, restore the
454 * original names, which are needed in some of the later steps.
456 for (i = 0; i < table_cnt; i++) {
457 expand_symbol(table[i]->sym, table[i]->len, buf);
458 strcpy((char *)table[i]->sym, buf);
461 output_label("kallsyms_markers");
462 for (i = 0; i < ((table_cnt + 255) >> 8); i++)
463 printf("\t.long\t%u\n", markers[i]);
468 output_label("kallsyms_token_table");
470 for (i = 0; i < 256; i++) {
472 expand_symbol(best_table[i], best_table_len[i], buf);
473 printf("\t.asciz\t\"%s\"\n", buf);
474 off += strlen(buf) + 1;
478 output_label("kallsyms_token_index");
479 for (i = 0; i < 256; i++)
480 printf("\t.short\t%d\n", best_idx[i]);
484 output_label("kallsyms_addresses");
486 output_label("kallsyms_offsets");
488 for (i = 0; i < table_cnt; i++) {
491 * Use the offset relative to the lowest value
492 * encountered of all relative symbols, and emit
493 * non-relocatable fixed offsets that will be fixed
500 if (!absolute_percpu) {
501 offset = table[i]->addr - relative_base;
502 overflow = (offset < 0 || offset > UINT_MAX);
503 } else if (symbol_absolute(table[i])) {
504 offset = table[i]->addr;
505 overflow = (offset < 0 || offset > INT_MAX);
507 offset = relative_base - table[i]->addr - 1;
508 overflow = (offset < INT_MIN || offset >= 0);
511 fprintf(stderr, "kallsyms failure: "
512 "%s symbol value %#llx out of range in relative mode\n",
513 symbol_absolute(table[i]) ? "absolute" : "relative",
517 printf("\t.long\t%#x /* %s */\n", (int)offset, table[i]->sym);
518 } else if (!symbol_absolute(table[i])) {
519 output_address(table[i]->addr);
521 printf("\tPTR\t%#llx\n", table[i]->addr);
527 output_label("kallsyms_relative_base");
528 output_address(relative_base);
533 for (i = 0; i < table_cnt; i++)
534 cleanup_symbol_name((char *)table[i]->sym);
536 sort_symbols_by_name();
537 output_label("kallsyms_seqs_of_names");
538 for (i = 0; i < table_cnt; i++)
539 printf("\t.byte 0x%02x, 0x%02x, 0x%02x\n",
540 (unsigned char)(table[i]->seq >> 16),
541 (unsigned char)(table[i]->seq >> 8),
542 (unsigned char)(table[i]->seq >> 0));
547 /* table lookup compression functions */
549 /* count all the possible tokens in a symbol */
550 static void learn_symbol(const unsigned char *symbol, int len)
554 for (i = 0; i < len - 1; i++)
555 token_profit[ symbol[i] + (symbol[i + 1] << 8) ]++;
558 /* decrease the count for all the possible tokens in a symbol */
559 static void forget_symbol(const unsigned char *symbol, int len)
563 for (i = 0; i < len - 1; i++)
564 token_profit[ symbol[i] + (symbol[i + 1] << 8) ]--;
567 /* do the initial token count */
568 static void build_initial_token_table(void)
572 for (i = 0; i < table_cnt; i++)
573 learn_symbol(table[i]->sym, table[i]->len);
576 static unsigned char *find_token(unsigned char *str, int len,
577 const unsigned char *token)
581 for (i = 0; i < len - 1; i++) {
582 if (str[i] == token[0] && str[i+1] == token[1])
588 /* replace a given token in all the valid symbols. Use the sampled symbols
589 * to update the counts */
590 static void compress_symbols(const unsigned char *str, int idx)
592 unsigned int i, len, size;
593 unsigned char *p1, *p2;
595 for (i = 0; i < table_cnt; i++) {
600 /* find the token on the symbol */
601 p2 = find_token(p1, len, str);
604 /* decrease the counts for this symbol's tokens */
605 forget_symbol(table[i]->sym, len);
613 memmove(p2, p2 + 1, size);
619 /* find the token on the symbol */
620 p2 = find_token(p1, size, str);
626 /* increase the counts for this symbol's new tokens */
627 learn_symbol(table[i]->sym, len);
631 /* search the token with the maximum profit */
632 static int find_best_token(void)
634 int i, best, bestprofit;
639 for (i = 0; i < 0x10000; i++) {
640 if (token_profit[i] > bestprofit) {
642 bestprofit = token_profit[i];
648 /* this is the core of the algorithm: calculate the "best" table */
649 static void optimize_result(void)
653 /* using the '\0' symbol last allows compress_symbols to use standard
654 * fast string functions */
655 for (i = 255; i >= 0; i--) {
657 /* if this table slot is empty (it is not used by an actual
658 * original char code */
659 if (!best_table_len[i]) {
661 /* find the token with the best profit value */
662 best = find_best_token();
663 if (token_profit[best] == 0)
666 /* place it in the "best" table */
667 best_table_len[i] = 2;
668 best_table[i][0] = best & 0xFF;
669 best_table[i][1] = (best >> 8) & 0xFF;
671 /* replace this token in all the valid symbols */
672 compress_symbols(best_table[i], i);
677 /* start by placing the symbols that are actually used on the table */
678 static void insert_real_symbols_in_table(void)
680 unsigned int i, j, c;
682 for (i = 0; i < table_cnt; i++) {
683 for (j = 0; j < table[i]->len; j++) {
684 c = table[i]->sym[j];
691 static void optimize_token_table(void)
693 build_initial_token_table();
695 insert_real_symbols_in_table();
700 /* guess for "linker script provide" symbol */
701 static int may_be_linker_script_provide_symbol(const struct sym_entry *se)
703 const char *symbol = sym_name(se);
704 int len = se->len - 1;
709 if (symbol[0] != '_' || symbol[1] != '_')
713 if (!memcmp(symbol + 2, "start_", 6))
717 if (!memcmp(symbol + 2, "stop_", 5))
721 if (!memcmp(symbol + 2, "end_", 4))
725 if (!memcmp(symbol + len - 6, "_start", 6))
729 if (!memcmp(symbol + len - 4, "_end", 4))
735 static int compare_symbols(const void *a, const void *b)
737 const struct sym_entry *sa = *(const struct sym_entry **)a;
738 const struct sym_entry *sb = *(const struct sym_entry **)b;
741 /* sort by address first */
742 if (sa->addr > sb->addr)
744 if (sa->addr < sb->addr)
747 /* sort by "weakness" type */
748 wa = (sa->sym[0] == 'w') || (sa->sym[0] == 'W');
749 wb = (sb->sym[0] == 'w') || (sb->sym[0] == 'W');
753 /* sort by "linker script provide" type */
754 wa = may_be_linker_script_provide_symbol(sa);
755 wb = may_be_linker_script_provide_symbol(sb);
759 /* sort by the number of prefix underscores */
760 wa = strspn(sym_name(sa), "_");
761 wb = strspn(sym_name(sb), "_");
765 /* sort by initial order, so that other symbols are left undisturbed */
766 return sa->start_pos - sb->start_pos;
769 static void sort_symbols(void)
771 qsort(table, table_cnt, sizeof(table[0]), compare_symbols);
774 static void make_percpus_absolute(void)
778 for (i = 0; i < table_cnt; i++)
779 if (symbol_in_range(table[i], &percpu_range, 1)) {
781 * Keep the 'A' override for percpu symbols to
782 * ensure consistent behavior compared to older
783 * versions of this tool.
785 table[i]->sym[0] = 'A';
786 table[i]->percpu_absolute = 1;
790 /* find the minimum non-absolute symbol address */
791 static void record_relative_base(void)
795 for (i = 0; i < table_cnt; i++)
796 if (!symbol_absolute(table[i])) {
798 * The table is sorted by address.
799 * Take the first non-absolute symbol value.
801 relative_base = table[i]->addr;
806 int main(int argc, char **argv)
809 static struct option long_options[] = {
810 {"all-symbols", no_argument, &all_symbols, 1},
811 {"absolute-percpu", no_argument, &absolute_percpu, 1},
812 {"base-relative", no_argument, &base_relative, 1},
813 {"lto-clang", no_argument, <o_clang, 1},
817 int c = getopt_long(argc, argv, "", long_options, NULL);
828 read_map(argv[optind]);
831 make_percpus_absolute();
834 record_relative_base();
835 optimize_token_table();