blob: a9186a98a37d1342cd58446b29804e47f712b168 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/* 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 *
Linus Torvalds1da177e2005-04-16 15:20:36 -070010 * 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>
Ard Biesheuvel2213e9a2016-03-15 14:58:19 -070025#include <limits.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070026
Mike Frysinger17b1f0d2009-06-08 19:12:13 -040027#ifndef ARRAY_SIZE
28#define ARRAY_SIZE(arr) (sizeof(arr) / sizeof(arr[0]))
29#endif
30
Tejun Heo9281ace2007-07-17 04:03:51 -070031#define KSYM_NAME_LEN 128
Linus Torvalds1da177e2005-04-16 15:20:36 -070032
Linus Torvalds1da177e2005-04-16 15:20:36 -070033struct sym_entry {
34 unsigned long long addr;
Paulo Marquesb3dbb4e2005-09-06 15:16:31 -070035 unsigned int len;
Paulo Marquesf2df3f62008-02-06 01:37:33 -080036 unsigned int start_pos;
Linus Torvalds1da177e2005-04-16 15:20:36 -070037 unsigned char *sym;
Ard Biesheuvel8c996942016-03-15 14:58:15 -070038 unsigned int percpu_absolute;
Linus Torvalds1da177e2005-04-16 15:20:36 -070039};
40
Kees Cook78eb7152014-03-17 13:18:27 +103041struct addr_range {
42 const char *start_sym, *end_sym;
Mike Frysinger17b1f0d2009-06-08 19:12:13 -040043 unsigned long long start, end;
44};
45
46static unsigned long long _text;
Ard Biesheuvel2213e9a2016-03-15 14:58:19 -070047static unsigned long long relative_base;
Kees Cook78eb7152014-03-17 13:18:27 +103048static struct addr_range text_ranges[] = {
Mike Frysinger17b1f0d2009-06-08 19:12:13 -040049 { "_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
Rusty Russellc6bda7c2014-03-17 14:05:46 +103057static struct addr_range percpu_range = {
58 "__per_cpu_start", "__per_cpu_end", -1ULL, 0
59};
60
Linus Torvalds1da177e2005-04-16 15:20:36 -070061static struct sym_entry *table;
Paulo Marquesb3dbb4e2005-09-06 15:16:31 -070062static unsigned int table_size, table_cnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -070063static int all_symbols = 0;
Rusty Russellc6bda7c2014-03-17 14:05:46 +103064static int absolute_percpu = 0;
Ard Biesheuvel2213e9a2016-03-15 14:58:19 -070065static int base_relative = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -070066
Paulo Marquesb3dbb4e2005-09-06 15:16:31 -070067int token_profit[0x10000];
Linus Torvalds1da177e2005-04-16 15:20:36 -070068
69/* the table that holds the result of the compression */
Paulo Marquesb3dbb4e2005-09-06 15:16:31 -070070unsigned char best_table[256][2];
Linus Torvalds1da177e2005-04-16 15:20:36 -070071unsigned char best_table_len[256];
72
73
Paulo Marquesb3dbb4e2005-09-06 15:16:31 -070074static void usage(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -070075{
Ming Leif6537f22013-11-02 09:11:33 +103076 fprintf(stderr, "Usage: kallsyms [--all-symbols] "
Ard Biesheuvel2213e9a2016-03-15 14:58:19 -070077 "[--base-relative] < in.map > out.S\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -070078 exit(1);
79}
80
81/*
82 * This ignores the intensely annoying "mapping symbols" found
83 * in ARM ELF files: $a, $t and $d.
84 */
Paulo Marquesb3dbb4e2005-09-06 15:16:31 -070085static inline int is_arm_mapping_symbol(const char *str)
Linus Torvalds1da177e2005-04-16 15:20:36 -070086{
Kyle McMartin6c34f1f2014-09-16 22:37:18 +010087 return str[0] == '$' && strchr("axtd", str[1])
Linus Torvalds1da177e2005-04-16 15:20:36 -070088 && (str[2] == '\0' || str[2] == '.');
89}
90
Kees Cook78eb7152014-03-17 13:18:27 +103091static int check_symbol_range(const char *sym, unsigned long long addr,
92 struct addr_range *ranges, int entries)
Mike Frysinger17b1f0d2009-06-08 19:12:13 -040093{
94 size_t i;
Kees Cook78eb7152014-03-17 13:18:27 +103095 struct addr_range *ar;
Mike Frysinger17b1f0d2009-06-08 19:12:13 -040096
Kees Cook78eb7152014-03-17 13:18:27 +103097 for (i = 0; i < entries; ++i) {
98 ar = &ranges[i];
Mike Frysinger17b1f0d2009-06-08 19:12:13 -040099
Kees Cook78eb7152014-03-17 13:18:27 +1030100 if (strcmp(sym, ar->start_sym) == 0) {
101 ar->start = addr;
Mike Frysinger17b1f0d2009-06-08 19:12:13 -0400102 return 0;
Kees Cook78eb7152014-03-17 13:18:27 +1030103 } else if (strcmp(sym, ar->end_sym) == 0) {
104 ar->end = addr;
Mike Frysinger17b1f0d2009-06-08 19:12:13 -0400105 return 0;
106 }
107 }
108
109 return 1;
110}
111
Paulo Marquesb3dbb4e2005-09-06 15:16:31 -0700112static int read_symbol(FILE *in, struct sym_entry *s)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700113{
Masahiro Yamada534c9f22018-05-09 16:23:47 +0900114 char sym[500], stype;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700115 int rc;
116
Masahiro Yamada534c9f22018-05-09 16:23:47 +0900117 rc = fscanf(in, "%llx %c %499s\n", &s->addr, &stype, sym);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700118 if (rc != 3) {
Masahiro Yamada534c9f22018-05-09 16:23:47 +0900119 if (rc != EOF && fgets(sym, 500, in) == NULL)
Jean Sacrenef894872010-09-10 23:13:33 -0600120 fprintf(stderr, "Read error or end of file.\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700121 return -1;
122 }
Masahiro Yamada534c9f22018-05-09 16:23:47 +0900123 if (strlen(sym) > KSYM_NAME_LEN) {
Fabio Estevam6f622592013-11-08 00:45:01 -0200124 fprintf(stderr, "Symbol %s too long for kallsyms (%zu vs %d).\n"
Masahiro Yamadabb66fc62014-06-10 19:08:13 +0900125 "Please increase KSYM_NAME_LEN both in kernel and kallsyms.c\n",
Masahiro Yamada534c9f22018-05-09 16:23:47 +0900126 sym, strlen(sym), KSYM_NAME_LEN);
Andi Kleenf3462aa2013-10-23 15:07:53 +0200127 return -1;
128 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700129
130 /* Ignore most absolute/undefined (?) symbols. */
Eric W. Biedermanfd593d12006-12-07 02:14:04 +0100131 if (strcmp(sym, "_text") == 0)
132 _text = s->addr;
Kees Cook78eb7152014-03-17 13:18:27 +1030133 else if (check_symbol_range(sym, s->addr, text_ranges,
134 ARRAY_SIZE(text_ranges)) == 0)
Mike Frysinger17b1f0d2009-06-08 19:12:13 -0400135 /* nothing to do */;
Paulo Marquesb3dbb4e2005-09-06 15:16:31 -0700136 else if (toupper(stype) == 'A')
Linus Torvalds1da177e2005-04-16 15:20:36 -0700137 {
138 /* Keep these useful absolute symbols */
Yoshinori Sato41f11a42005-05-01 08:59:06 -0700139 if (strcmp(sym, "__kernel_syscall_via_break") &&
140 strcmp(sym, "__kernel_syscall_via_epc") &&
141 strcmp(sym, "__kernel_sigtramp") &&
142 strcmp(sym, "__gp"))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700143 return -1;
144
145 }
Paulo Marquesb3dbb4e2005-09-06 15:16:31 -0700146 else if (toupper(stype) == 'U' ||
Yoshinori Sato41f11a42005-05-01 08:59:06 -0700147 is_arm_mapping_symbol(sym))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700148 return -1;
Ralf Baechle6f00df22005-09-06 15:16:41 -0700149 /* exclude also MIPS ELF local symbols ($L123 instead of .L123) */
Masahiro Yamada534c9f22018-05-09 16:23:47 +0900150 else if (sym[0] == '$')
Ralf Baechle6f00df22005-09-06 15:16:41 -0700151 return -1;
Sam Ravnborgaab34ac2008-05-19 20:07:58 +0200152 /* exclude debugging symbols */
Guenter Roeck51962a92017-10-13 15:57:58 -0700153 else if (stype == 'N' || stype == 'n')
Sam Ravnborgaab34ac2008-05-19 20:07:58 +0200154 return -1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700155
156 /* include the type field in the symbol name, so that it gets
157 * compressed together */
Masahiro Yamada534c9f22018-05-09 16:23:47 +0900158 s->len = strlen(sym) + 1;
Paulo Marquesb3dbb4e2005-09-06 15:16:31 -0700159 s->sym = malloc(s->len + 1);
Jesper Juhlf1a136e2006-03-25 03:07:46 -0800160 if (!s->sym) {
161 fprintf(stderr, "kallsyms failure: "
162 "unable to allocate required amount of memory\n");
163 exit(EXIT_FAILURE);
164 }
Masahiro Yamada534c9f22018-05-09 16:23:47 +0900165 strcpy((char *)s->sym + 1, sym);
Paulo Marquesb3dbb4e2005-09-06 15:16:31 -0700166 s->sym[0] = stype;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700167
Ard Biesheuvel8c996942016-03-15 14:58:15 -0700168 s->percpu_absolute = 0;
169
Rusty Russellc6bda7c2014-03-17 14:05:46 +1030170 /* Record if we've found __per_cpu_start/end. */
171 check_symbol_range(sym, s->addr, &percpu_range, 1);
172
Linus Torvalds1da177e2005-04-16 15:20:36 -0700173 return 0;
174}
175
Kees Cook78eb7152014-03-17 13:18:27 +1030176static int symbol_in_range(struct sym_entry *s, struct addr_range *ranges,
177 int entries)
Mike Frysinger17b1f0d2009-06-08 19:12:13 -0400178{
179 size_t i;
Kees Cook78eb7152014-03-17 13:18:27 +1030180 struct addr_range *ar;
Mike Frysinger17b1f0d2009-06-08 19:12:13 -0400181
Kees Cook78eb7152014-03-17 13:18:27 +1030182 for (i = 0; i < entries; ++i) {
183 ar = &ranges[i];
Mike Frysinger17b1f0d2009-06-08 19:12:13 -0400184
Kees Cook78eb7152014-03-17 13:18:27 +1030185 if (s->addr >= ar->start && s->addr <= ar->end)
Mike Frysingerac6ca5c2009-06-15 07:52:48 -0400186 return 1;
Mike Frysinger17b1f0d2009-06-08 19:12:13 -0400187 }
188
Mike Frysingerac6ca5c2009-06-15 07:52:48 -0400189 return 0;
Mike Frysinger17b1f0d2009-06-08 19:12:13 -0400190}
191
Paulo Marquesb3dbb4e2005-09-06 15:16:31 -0700192static int symbol_valid(struct sym_entry *s)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700193{
194 /* Symbols which vary between passes. Passes 1 and 2 must have
Sam Ravnborg2ea03892009-01-14 21:38:20 +0100195 * identical symbol lists. The kallsyms_* symbols below are only added
196 * after pass 1, they would be included in pass 2 when --all-symbols is
197 * specified so exclude them to get a stable symbol list.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700198 */
199 static char *special_symbols[] = {
Sam Ravnborg2ea03892009-01-14 21:38:20 +0100200 "kallsyms_addresses",
Ard Biesheuvel2213e9a2016-03-15 14:58:19 -0700201 "kallsyms_offsets",
202 "kallsyms_relative_base",
Sam Ravnborg2ea03892009-01-14 21:38:20 +0100203 "kallsyms_num_syms",
204 "kallsyms_names",
205 "kallsyms_markers",
206 "kallsyms_token_table",
207 "kallsyms_token_index",
208
Linus Torvalds1da177e2005-04-16 15:20:36 -0700209 /* Exclude linker generated symbols which vary between passes */
210 "_SDA_BASE_", /* ppc */
211 "_SDA2_BASE_", /* ppc */
212 NULL };
Ard Biesheuvelbd8b22d2015-03-30 15:20:31 +0200213
Ard Biesheuvel56067812017-02-03 09:54:05 +0000214 static char *special_prefixes[] = {
215 "__crc_", /* modversions */
Ard Biesheuvel1212f7a2018-03-01 17:19:01 +0000216 "__efistub_", /* arm64 EFI stub namespace */
Ard Biesheuvel56067812017-02-03 09:54:05 +0000217 NULL };
218
Ard Biesheuvelbd8b22d2015-03-30 15:20:31 +0200219 static char *special_suffixes[] = {
Ard Biesheuvelbd8b22d2015-03-30 15:20:31 +0200220 "_veneer", /* arm */
Ard Biesheuvelb9b74be2016-04-01 14:33:48 +0100221 "_from_arm", /* arm */
222 "_from_thumb", /* arm */
Ard Biesheuvelbd8b22d2015-03-30 15:20:31 +0200223 NULL };
224
Linus Torvalds1da177e2005-04-16 15:20:36 -0700225 int i;
Ard Biesheuvelbd8b22d2015-03-30 15:20:31 +0200226 char *sym_name = (char *)s->sym + 1;
227
Linus Torvalds1da177e2005-04-16 15:20:36 -0700228 /* if --all-symbols is not specified, then symbols outside the text
229 * and inittext sections are discarded */
230 if (!all_symbols) {
Kees Cook78eb7152014-03-17 13:18:27 +1030231 if (symbol_in_range(s, text_ranges,
232 ARRAY_SIZE(text_ranges)) == 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700233 return 0;
234 /* Corner case. Discard any symbols with the same value as
Robin Getza3b81112008-02-06 01:36:26 -0800235 * _etext _einittext; they can move between pass 1 and 2 when
236 * the kallsyms data are added. If these symbols move then
237 * they may get dropped in pass 2, which breaks the kallsyms
238 * rules.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700239 */
Mike Frysinger17b1f0d2009-06-08 19:12:13 -0400240 if ((s->addr == text_range_text->end &&
Ard Biesheuvelbd8b22d2015-03-30 15:20:31 +0200241 strcmp(sym_name,
Kees Cook78eb7152014-03-17 13:18:27 +1030242 text_range_text->end_sym)) ||
Mike Frysinger17b1f0d2009-06-08 19:12:13 -0400243 (s->addr == text_range_inittext->end &&
Ard Biesheuvelbd8b22d2015-03-30 15:20:31 +0200244 strcmp(sym_name,
Kees Cook78eb7152014-03-17 13:18:27 +1030245 text_range_inittext->end_sym)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700246 return 0;
247 }
248
249 /* Exclude symbols which vary between passes. */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700250 for (i = 0; special_symbols[i]; i++)
Ard Biesheuvelbd8b22d2015-03-30 15:20:31 +0200251 if (strcmp(sym_name, special_symbols[i]) == 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700252 return 0;
253
Ard Biesheuvel56067812017-02-03 09:54:05 +0000254 for (i = 0; special_prefixes[i]; i++) {
255 int l = strlen(special_prefixes[i]);
256
257 if (l <= strlen(sym_name) &&
258 strncmp(sym_name, special_prefixes[i], l) == 0)
259 return 0;
260 }
261
Ard Biesheuvelbd8b22d2015-03-30 15:20:31 +0200262 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
Linus Torvalds1da177e2005-04-16 15:20:36 -0700269 return 1;
270}
271
Paulo Marquesb3dbb4e2005-09-06 15:16:31 -0700272static void read_map(FILE *in)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700273{
274 while (!feof(in)) {
Paulo Marquesb3dbb4e2005-09-06 15:16:31 -0700275 if (table_cnt >= table_size) {
276 table_size += 10000;
277 table = realloc(table, sizeof(*table) * table_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700278 if (!table) {
279 fprintf(stderr, "out of memory\n");
280 exit (1);
281 }
282 }
Paulo Marquesf2df3f62008-02-06 01:37:33 -0800283 if (read_symbol(in, &table[table_cnt]) == 0) {
284 table[table_cnt].start_pos = table_cnt;
Paulo Marquesb3dbb4e2005-09-06 15:16:31 -0700285 table_cnt++;
Paulo Marquesf2df3f62008-02-06 01:37:33 -0800286 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700287 }
288}
289
290static void output_label(char *label)
291{
Masahiro Yamada534c9f22018-05-09 16:23:47 +0900292 printf(".globl %s\n", label);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700293 printf("\tALGN\n");
Masahiro Yamada534c9f22018-05-09 16:23:47 +0900294 printf("%s:\n", label);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700295}
296
297/* uncompress a compressed symbol. When this function is called, the best table
298 * might still be compressed itself, so the function needs to be recursive */
299static int expand_symbol(unsigned char *data, int len, char *result)
300{
301 int c, rlen, total=0;
302
303 while (len) {
304 c = *data;
305 /* if the table holds a single char that is the same as the one
306 * we are looking for, then end the search */
307 if (best_table[c][0]==c && best_table_len[c]==1) {
308 *result++ = c;
309 total++;
310 } else {
311 /* if not, recurse and expand */
312 rlen = expand_symbol(best_table[c], best_table_len[c], result);
313 total += rlen;
314 result += rlen;
315 }
316 data++;
317 len--;
318 }
319 *result=0;
320
321 return total;
322}
323
Kees Cook78eb7152014-03-17 13:18:27 +1030324static int symbol_absolute(struct sym_entry *s)
325{
Ard Biesheuvel8c996942016-03-15 14:58:15 -0700326 return s->percpu_absolute;
Kees Cook78eb7152014-03-17 13:18:27 +1030327}
328
Paulo Marquesb3dbb4e2005-09-06 15:16:31 -0700329static void write_src(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700330{
Paulo Marquesb3dbb4e2005-09-06 15:16:31 -0700331 unsigned int i, k, off;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700332 unsigned int best_idx[256];
333 unsigned int *markers;
Tejun Heo9281ace2007-07-17 04:03:51 -0700334 char buf[KSYM_NAME_LEN];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700335
336 printf("#include <asm/types.h>\n");
337 printf("#if BITS_PER_LONG == 64\n");
338 printf("#define PTR .quad\n");
339 printf("#define ALGN .align 8\n");
340 printf("#else\n");
341 printf("#define PTR .long\n");
342 printf("#define ALGN .align 4\n");
343 printf("#endif\n");
344
Jan Beulichaad09472006-12-08 02:35:57 -0800345 printf("\t.section .rodata, \"a\"\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700346
Ard Biesheuvel2213e9a2016-03-15 14:58:19 -0700347 /* Provide proper symbols relocatability by their relativeness
348 * to a fixed anchor point in the runtime image, either '_text'
349 * for absolute address tables, in which case the linker will
350 * emit the final addresses at build time. Otherwise, use the
351 * offset relative to the lowest value encountered of all relative
352 * symbols, and emit non-relocatable fixed offsets that will be fixed
353 * up at runtime.
354 *
355 * The symbol names cannot be used to construct normal symbol
356 * references as the list of symbols contains symbols that are
357 * declared static and are private to their .o files. This prevents
358 * .tmp_kallsyms.o or any other object from referencing them.
Eric W. Biedermanfd593d12006-12-07 02:14:04 +0100359 */
Ard Biesheuvel2213e9a2016-03-15 14:58:19 -0700360 if (!base_relative)
361 output_label("kallsyms_addresses");
362 else
363 output_label("kallsyms_offsets");
364
Paulo Marquesb3dbb4e2005-09-06 15:16:31 -0700365 for (i = 0; i < table_cnt; i++) {
Ard Biesheuvel2213e9a2016-03-15 14:58:19 -0700366 if (base_relative) {
367 long long offset;
368 int overflow;
369
370 if (!absolute_percpu) {
371 offset = table[i].addr - relative_base;
372 overflow = (offset < 0 || offset > UINT_MAX);
373 } else if (symbol_absolute(&table[i])) {
374 offset = table[i].addr;
375 overflow = (offset < 0 || offset > INT_MAX);
376 } else {
377 offset = relative_base - table[i].addr - 1;
378 overflow = (offset < INT_MIN || offset >= 0);
379 }
380 if (overflow) {
381 fprintf(stderr, "kallsyms failure: "
382 "%s symbol value %#llx out of range in relative mode\n",
383 symbol_absolute(&table[i]) ? "absolute" : "relative",
384 table[i].addr);
385 exit(EXIT_FAILURE);
386 }
387 printf("\t.long\t%#x\n", (int)offset);
388 } else if (!symbol_absolute(&table[i])) {
Vivek Goyal2c22d8b2006-12-07 02:14:10 +0100389 if (_text <= table[i].addr)
390 printf("\tPTR\t_text + %#llx\n",
391 table[i].addr - _text);
392 else
Andrew Morton2930ffc2014-03-10 15:49:48 -0700393 printf("\tPTR\t_text - %#llx\n",
394 _text - table[i].addr);
Eric W. Biedermanfd593d12006-12-07 02:14:04 +0100395 } else {
396 printf("\tPTR\t%#llx\n", table[i].addr);
397 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700398 }
399 printf("\n");
400
Ard Biesheuvel2213e9a2016-03-15 14:58:19 -0700401 if (base_relative) {
402 output_label("kallsyms_relative_base");
403 printf("\tPTR\t_text - %#llx\n", _text - relative_base);
404 printf("\n");
405 }
406
Linus Torvalds1da177e2005-04-16 15:20:36 -0700407 output_label("kallsyms_num_syms");
nixiaomingac5db1f2018-05-24 11:16:12 +0800408 printf("\tPTR\t%u\n", table_cnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700409 printf("\n");
410
411 /* table of offset markers, that give the offset in the compressed stream
412 * every 256 symbols */
Jesper Juhlf1a136e2006-03-25 03:07:46 -0800413 markers = malloc(sizeof(unsigned int) * ((table_cnt + 255) / 256));
414 if (!markers) {
415 fprintf(stderr, "kallsyms failure: "
416 "unable to allocate required memory\n");
417 exit(EXIT_FAILURE);
418 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700419
420 output_label("kallsyms_names");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700421 off = 0;
Paulo Marquesb3dbb4e2005-09-06 15:16:31 -0700422 for (i = 0; i < table_cnt; i++) {
423 if ((i & 0xFF) == 0)
424 markers[i >> 8] = off;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700425
426 printf("\t.byte 0x%02x", table[i].len);
427 for (k = 0; k < table[i].len; k++)
428 printf(", 0x%02x", table[i].sym[k]);
429 printf("\n");
430
431 off += table[i].len + 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700432 }
433 printf("\n");
434
435 output_label("kallsyms_markers");
Paulo Marquesb3dbb4e2005-09-06 15:16:31 -0700436 for (i = 0; i < ((table_cnt + 255) >> 8); i++)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700437 printf("\tPTR\t%d\n", markers[i]);
438 printf("\n");
439
440 free(markers);
441
442 output_label("kallsyms_token_table");
443 off = 0;
444 for (i = 0; i < 256; i++) {
445 best_idx[i] = off;
Paulo Marquesb3dbb4e2005-09-06 15:16:31 -0700446 expand_symbol(best_table[i], best_table_len[i], buf);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700447 printf("\t.asciz\t\"%s\"\n", buf);
448 off += strlen(buf) + 1;
449 }
450 printf("\n");
451
452 output_label("kallsyms_token_index");
453 for (i = 0; i < 256; i++)
454 printf("\t.short\t%d\n", best_idx[i]);
455 printf("\n");
456}
457
458
459/* table lookup compression functions */
460
Linus Torvalds1da177e2005-04-16 15:20:36 -0700461/* count all the possible tokens in a symbol */
462static void learn_symbol(unsigned char *symbol, int len)
463{
464 int i;
465
466 for (i = 0; i < len - 1; i++)
Paulo Marquesb3dbb4e2005-09-06 15:16:31 -0700467 token_profit[ symbol[i] + (symbol[i + 1] << 8) ]++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700468}
469
470/* decrease the count for all the possible tokens in a symbol */
471static void forget_symbol(unsigned char *symbol, int len)
472{
473 int i;
474
475 for (i = 0; i < len - 1; i++)
Paulo Marquesb3dbb4e2005-09-06 15:16:31 -0700476 token_profit[ symbol[i] + (symbol[i + 1] << 8) ]--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700477}
478
Paulo Marquesb3dbb4e2005-09-06 15:16:31 -0700479/* remove all the invalid symbols from the table and do the initial token count */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700480static void build_initial_tok_table(void)
481{
Paulo Marquesb3dbb4e2005-09-06 15:16:31 -0700482 unsigned int i, pos;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700483
Paulo Marquesb3dbb4e2005-09-06 15:16:31 -0700484 pos = 0;
485 for (i = 0; i < table_cnt; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700486 if ( symbol_valid(&table[i]) ) {
Paulo Marquesb3dbb4e2005-09-06 15:16:31 -0700487 if (pos != i)
488 table[pos] = table[i];
489 learn_symbol(table[pos].sym, table[pos].len);
490 pos++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700491 }
492 }
Paulo Marquesb3dbb4e2005-09-06 15:16:31 -0700493 table_cnt = pos;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700494}
495
Paulo Marques7c5d2492007-06-20 18:09:00 +0100496static void *find_token(unsigned char *str, int len, unsigned char *token)
497{
498 int i;
499
500 for (i = 0; i < len - 1; i++) {
501 if (str[i] == token[0] && str[i+1] == token[1])
502 return &str[i];
503 }
504 return NULL;
505}
506
Linus Torvalds1da177e2005-04-16 15:20:36 -0700507/* replace a given token in all the valid symbols. Use the sampled symbols
508 * to update the counts */
Paulo Marquesb3dbb4e2005-09-06 15:16:31 -0700509static void compress_symbols(unsigned char *str, int idx)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700510{
Paulo Marquesb3dbb4e2005-09-06 15:16:31 -0700511 unsigned int i, len, size;
512 unsigned char *p1, *p2;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700513
Paulo Marquesb3dbb4e2005-09-06 15:16:31 -0700514 for (i = 0; i < table_cnt; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700515
516 len = table[i].len;
Paulo Marquesb3dbb4e2005-09-06 15:16:31 -0700517 p1 = table[i].sym;
518
519 /* find the token on the symbol */
Paulo Marques7c5d2492007-06-20 18:09:00 +0100520 p2 = find_token(p1, len, str);
Paulo Marquesb3dbb4e2005-09-06 15:16:31 -0700521 if (!p2) continue;
522
523 /* decrease the counts for this symbol's tokens */
524 forget_symbol(table[i].sym, len);
525
526 size = len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700527
528 do {
Paulo Marquesb3dbb4e2005-09-06 15:16:31 -0700529 *p2 = idx;
530 p2++;
531 size -= (p2 - p1);
532 memmove(p2, p2 + 1, size);
533 p1 = p2;
534 len--;
535
536 if (size < 2) break;
537
Linus Torvalds1da177e2005-04-16 15:20:36 -0700538 /* find the token on the symbol */
Paulo Marques7c5d2492007-06-20 18:09:00 +0100539 p2 = find_token(p1, size, str);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700540
Paulo Marquesb3dbb4e2005-09-06 15:16:31 -0700541 } while (p2);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700542
Paulo Marquesb3dbb4e2005-09-06 15:16:31 -0700543 table[i].len = len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700544
Paulo Marquesb3dbb4e2005-09-06 15:16:31 -0700545 /* increase the counts for this symbol's new tokens */
546 learn_symbol(table[i].sym, len);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700547 }
548}
549
550/* search the token with the maximum profit */
Paulo Marquesb3dbb4e2005-09-06 15:16:31 -0700551static int find_best_token(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700552{
Paulo Marquesb3dbb4e2005-09-06 15:16:31 -0700553 int i, best, bestprofit;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700554
555 bestprofit=-10000;
Paulo Marquesb3dbb4e2005-09-06 15:16:31 -0700556 best = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700557
Paulo Marquesb3dbb4e2005-09-06 15:16:31 -0700558 for (i = 0; i < 0x10000; i++) {
559 if (token_profit[i] > bestprofit) {
560 best = i;
561 bestprofit = token_profit[i];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700562 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700563 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700564 return best;
565}
566
567/* this is the core of the algorithm: calculate the "best" table */
568static void optimize_result(void)
569{
Paulo Marquesb3dbb4e2005-09-06 15:16:31 -0700570 int i, best;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700571
572 /* using the '\0' symbol last allows compress_symbols to use standard
573 * fast string functions */
574 for (i = 255; i >= 0; i--) {
575
576 /* if this table slot is empty (it is not used by an actual
577 * original char code */
578 if (!best_table_len[i]) {
579
Cao jincbf7a902018-02-27 16:16:19 +0800580 /* find the token with the best profit value */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700581 best = find_best_token();
Xiaochen Wange0a04b12011-05-01 11:41:41 +0800582 if (token_profit[best] == 0)
583 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700584
585 /* place it in the "best" table */
Paulo Marquesb3dbb4e2005-09-06 15:16:31 -0700586 best_table_len[i] = 2;
587 best_table[i][0] = best & 0xFF;
588 best_table[i][1] = (best >> 8) & 0xFF;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700589
590 /* replace this token in all the valid symbols */
Paulo Marquesb3dbb4e2005-09-06 15:16:31 -0700591 compress_symbols(best_table[i], i);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700592 }
593 }
594}
595
596/* start by placing the symbols that are actually used on the table */
597static void insert_real_symbols_in_table(void)
598{
Paulo Marquesb3dbb4e2005-09-06 15:16:31 -0700599 unsigned int i, j, c;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700600
601 memset(best_table, 0, sizeof(best_table));
602 memset(best_table_len, 0, sizeof(best_table_len));
603
Paulo Marquesb3dbb4e2005-09-06 15:16:31 -0700604 for (i = 0; i < table_cnt; i++) {
605 for (j = 0; j < table[i].len; j++) {
606 c = table[i].sym[j];
607 best_table[c][0]=c;
608 best_table_len[c]=1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700609 }
610 }
611}
612
613static void optimize_token_table(void)
614{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700615 build_initial_tok_table();
616
617 insert_real_symbols_in_table();
618
Yoshinori Sato41f11a42005-05-01 08:59:06 -0700619 /* When valid symbol is not registered, exit to error */
Paulo Marquesb3dbb4e2005-09-06 15:16:31 -0700620 if (!table_cnt) {
Yoshinori Sato41f11a42005-05-01 08:59:06 -0700621 fprintf(stderr, "No valid symbol.\n");
622 exit(1);
623 }
624
Linus Torvalds1da177e2005-04-16 15:20:36 -0700625 optimize_result();
626}
627
Lai Jiangshanb478b7822009-03-13 15:10:26 +0800628/* guess for "linker script provide" symbol */
629static int may_be_linker_script_provide_symbol(const struct sym_entry *se)
630{
631 const char *symbol = (char *)se->sym + 1;
632 int len = se->len - 1;
633
634 if (len < 8)
635 return 0;
636
637 if (symbol[0] != '_' || symbol[1] != '_')
638 return 0;
639
640 /* __start_XXXXX */
641 if (!memcmp(symbol + 2, "start_", 6))
642 return 1;
643
644 /* __stop_XXXXX */
645 if (!memcmp(symbol + 2, "stop_", 5))
646 return 1;
647
648 /* __end_XXXXX */
649 if (!memcmp(symbol + 2, "end_", 4))
650 return 1;
651
652 /* __XXXXX_start */
653 if (!memcmp(symbol + len - 6, "_start", 6))
654 return 1;
655
656 /* __XXXXX_end */
657 if (!memcmp(symbol + len - 4, "_end", 4))
658 return 1;
659
660 return 0;
661}
662
663static int prefix_underscores_count(const char *str)
664{
665 const char *tail = str;
666
Paul Mundta9ece532009-09-22 16:44:12 -0700667 while (*tail == '_')
Lai Jiangshanb478b7822009-03-13 15:10:26 +0800668 tail++;
669
670 return tail - str;
671}
672
Paulo Marquesf2df3f62008-02-06 01:37:33 -0800673static int compare_symbols(const void *a, const void *b)
674{
675 const struct sym_entry *sa;
676 const struct sym_entry *sb;
677 int wa, wb;
678
679 sa = a;
680 sb = b;
681
682 /* sort by address first */
683 if (sa->addr > sb->addr)
684 return 1;
685 if (sa->addr < sb->addr)
686 return -1;
687
688 /* sort by "weakness" type */
689 wa = (sa->sym[0] == 'w') || (sa->sym[0] == 'W');
690 wb = (sb->sym[0] == 'w') || (sb->sym[0] == 'W');
691 if (wa != wb)
692 return wa - wb;
693
Lai Jiangshanb478b7822009-03-13 15:10:26 +0800694 /* sort by "linker script provide" type */
695 wa = may_be_linker_script_provide_symbol(sa);
696 wb = may_be_linker_script_provide_symbol(sb);
697 if (wa != wb)
698 return wa - wb;
699
700 /* sort by the number of prefix underscores */
701 wa = prefix_underscores_count((const char *)sa->sym + 1);
702 wb = prefix_underscores_count((const char *)sb->sym + 1);
703 if (wa != wb)
704 return wa - wb;
705
Paulo Marquesf2df3f62008-02-06 01:37:33 -0800706 /* sort by initial order, so that other symbols are left undisturbed */
707 return sa->start_pos - sb->start_pos;
708}
709
710static void sort_symbols(void)
711{
712 qsort(table, table_cnt, sizeof(struct sym_entry), compare_symbols);
713}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700714
Rusty Russellc6bda7c2014-03-17 14:05:46 +1030715static void make_percpus_absolute(void)
716{
717 unsigned int i;
718
719 for (i = 0; i < table_cnt; i++)
Ard Biesheuvel8c996942016-03-15 14:58:15 -0700720 if (symbol_in_range(&table[i], &percpu_range, 1)) {
721 /*
722 * Keep the 'A' override for percpu symbols to
723 * ensure consistent behavior compared to older
724 * versions of this tool.
725 */
Rusty Russellc6bda7c2014-03-17 14:05:46 +1030726 table[i].sym[0] = 'A';
Ard Biesheuvel8c996942016-03-15 14:58:15 -0700727 table[i].percpu_absolute = 1;
728 }
Rusty Russellc6bda7c2014-03-17 14:05:46 +1030729}
730
Ard Biesheuvel2213e9a2016-03-15 14:58:19 -0700731/* find the minimum non-absolute symbol address */
732static void record_relative_base(void)
733{
734 unsigned int i;
735
736 relative_base = -1ULL;
737 for (i = 0; i < table_cnt; i++)
738 if (!symbol_absolute(&table[i]) &&
739 table[i].addr < relative_base)
740 relative_base = table[i].addr;
741}
742
Paulo Marquesb3dbb4e2005-09-06 15:16:31 -0700743int main(int argc, char **argv)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700744{
Yoshinori Sato41f11a42005-05-01 08:59:06 -0700745 if (argc >= 2) {
746 int i;
747 for (i = 1; i < argc; i++) {
748 if(strcmp(argv[i], "--all-symbols") == 0)
749 all_symbols = 1;
Rusty Russellc6bda7c2014-03-17 14:05:46 +1030750 else if (strcmp(argv[i], "--absolute-percpu") == 0)
751 absolute_percpu = 1;
Masahiro Yamada534c9f22018-05-09 16:23:47 +0900752 else if (strcmp(argv[i], "--base-relative") == 0)
Ard Biesheuvel2213e9a2016-03-15 14:58:19 -0700753 base_relative = 1;
754 else
Yoshinori Sato41f11a42005-05-01 08:59:06 -0700755 usage();
756 }
757 } else if (argc != 1)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700758 usage();
759
760 read_map(stdin);
Rusty Russellc6bda7c2014-03-17 14:05:46 +1030761 if (absolute_percpu)
762 make_percpus_absolute();
Ard Biesheuvel2213e9a2016-03-15 14:58:19 -0700763 if (base_relative)
764 record_relative_base();
Sam Ravnborg2ea03892009-01-14 21:38:20 +0100765 sort_symbols();
766 optimize_token_table();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700767 write_src();
768
769 return 0;
770}