Commit | Line | Data |
---|---|---|
1da177e4 LT |
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 | * | |
adc40221 | 8 | * Usage: kallsyms [--all-symbols] [--absolute-percpu] |
79549da6 | 9 | * [--base-relative] [--lto-clang] in.map > out.S |
1da177e4 | 10 | * |
1da177e4 LT |
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 | |
18 | * of about 50%. | |
19 | * | |
20 | */ | |
21 | ||
1c975da5 | 22 | #include <errno.h> |
aa221f2e | 23 | #include <getopt.h> |
a41333e0 | 24 | #include <stdbool.h> |
1da177e4 LT |
25 | #include <stdio.h> |
26 | #include <stdlib.h> | |
27 | #include <string.h> | |
28 | #include <ctype.h> | |
2213e9a6 | 29 | #include <limits.h> |
1da177e4 | 30 | |
17b1f0de | 31 | #define ARRAY_SIZE(arr) (sizeof(arr) / sizeof(arr[0])) |
17b1f0de | 32 | |
b8a94bfb | 33 | #define KSYM_NAME_LEN 512 |
1da177e4 | 34 | |
1da177e4 LT |
35 | struct sym_entry { |
36 | unsigned long long addr; | |
b3dbb4ec | 37 | unsigned int len; |
60443c88 | 38 | unsigned int seq; |
f2df3f65 | 39 | unsigned int start_pos; |
8c996940 | 40 | unsigned int percpu_absolute; |
9d82973e | 41 | unsigned char sym[]; |
1da177e4 LT |
42 | }; |
43 | ||
78eb7159 KC |
44 | struct addr_range { |
45 | const char *start_sym, *end_sym; | |
17b1f0de MF |
46 | unsigned long long start, end; |
47 | }; | |
48 | ||
49 | static unsigned long long _text; | |
2213e9a6 | 50 | static unsigned long long relative_base; |
78eb7159 | 51 | static struct addr_range text_ranges[] = { |
17b1f0de MF |
52 | { "_stext", "_etext" }, |
53 | { "_sinittext", "_einittext" }, | |
17b1f0de MF |
54 | }; |
55 | #define text_range_text (&text_ranges[0]) | |
56 | #define text_range_inittext (&text_ranges[1]) | |
57 | ||
c6bda7c9 RR |
58 | static struct addr_range percpu_range = { |
59 | "__per_cpu_start", "__per_cpu_end", -1ULL, 0 | |
60 | }; | |
61 | ||
8d605269 | 62 | static struct sym_entry **table; |
b3dbb4ec | 63 | static unsigned int table_size, table_cnt; |
831362fc MY |
64 | static int all_symbols; |
65 | static int absolute_percpu; | |
66 | static int base_relative; | |
010a0aad | 67 | static int lto_clang; |
1da177e4 | 68 | |
f43e9daa | 69 | static int token_profit[0x10000]; |
1da177e4 LT |
70 | |
71 | /* the table that holds the result of the compression */ | |
f43e9daa MY |
72 | static unsigned char best_table[256][2]; |
73 | static unsigned char best_table_len[256]; | |
1da177e4 LT |
74 | |
75 | ||
b3dbb4ec | 76 | static void usage(void) |
1da177e4 | 77 | { |
8d3a7507 | 78 | fprintf(stderr, "Usage: kallsyms [--all-symbols] [--absolute-percpu] " |
010a0aad | 79 | "[--base-relative] [--lto-clang] in.map > out.S\n"); |
1da177e4 LT |
80 | exit(1); |
81 | } | |
82 | ||
29e55ad3 MY |
83 | static char *sym_name(const struct sym_entry *s) |
84 | { | |
85 | return (char *)s->sym + 1; | |
86 | } | |
87 | ||
a41333e0 MY |
88 | static bool is_ignored_symbol(const char *name, char type) |
89 | { | |
a7b00a18 | 90 | if (type == 'u' || type == 'n') |
887df76d MY |
91 | return true; |
92 | ||
93 | if (toupper(type) == 'A') { | |
94 | /* Keep these useful absolute symbols */ | |
95 | if (strcmp(name, "__kernel_syscall_via_break") && | |
96 | strcmp(name, "__kernel_syscall_via_epc") && | |
97 | strcmp(name, "__kernel_sigtramp") && | |
98 | strcmp(name, "__gp")) | |
99 | return true; | |
100 | } | |
101 | ||
a41333e0 MY |
102 | return false; |
103 | } | |
104 | ||
b6233d0d MY |
105 | static void check_symbol_range(const char *sym, unsigned long long addr, |
106 | struct addr_range *ranges, int entries) | |
17b1f0de MF |
107 | { |
108 | size_t i; | |
78eb7159 | 109 | struct addr_range *ar; |
17b1f0de | 110 | |
78eb7159 KC |
111 | for (i = 0; i < entries; ++i) { |
112 | ar = &ranges[i]; | |
17b1f0de | 113 | |
78eb7159 KC |
114 | if (strcmp(sym, ar->start_sym) == 0) { |
115 | ar->start = addr; | |
b6233d0d | 116 | return; |
78eb7159 KC |
117 | } else if (strcmp(sym, ar->end_sym) == 0) { |
118 | ar->end = addr; | |
b6233d0d | 119 | return; |
17b1f0de MF |
120 | } |
121 | } | |
17b1f0de MF |
122 | } |
123 | ||
1c975da5 | 124 | static struct sym_entry *read_symbol(FILE *in, char **buf, size_t *buf_len) |
1da177e4 | 125 | { |
1c975da5 | 126 | char *name, type, *p; |
8d605269 | 127 | unsigned long long addr; |
1c975da5 MY |
128 | size_t len; |
129 | ssize_t readlen; | |
8d605269 | 130 | struct sym_entry *sym; |
1da177e4 | 131 | |
23835308 | 132 | errno = 0; |
1c975da5 MY |
133 | readlen = getline(buf, buf_len, in); |
134 | if (readlen < 0) { | |
135 | if (errno) { | |
136 | perror("read_symbol"); | |
137 | exit(EXIT_FAILURE); | |
138 | } | |
8d605269 | 139 | return NULL; |
1da177e4 | 140 | } |
1c975da5 MY |
141 | |
142 | if ((*buf)[readlen - 1] == '\n') | |
143 | (*buf)[readlen - 1] = 0; | |
144 | ||
145 | addr = strtoull(*buf, &p, 16); | |
146 | ||
147 | if (*buf == p || *p++ != ' ' || !isascii((type = *p++)) || *p++ != ' ') { | |
148 | fprintf(stderr, "line format error\n"); | |
149 | exit(EXIT_FAILURE); | |
150 | } | |
151 | ||
152 | name = p; | |
153 | len = strlen(name); | |
154 | ||
155 | if (len >= KSYM_NAME_LEN) { | |
6db2983c | 156 | fprintf(stderr, "Symbol %s too long for kallsyms (%zu >= %d).\n" |
bb66fc67 | 157 | "Please increase KSYM_NAME_LEN both in kernel and kallsyms.c\n", |
1c975da5 | 158 | name, len, KSYM_NAME_LEN); |
8d605269 | 159 | return NULL; |
f3462aa9 | 160 | } |
1da177e4 | 161 | |
be9f6133 | 162 | if (strcmp(name, "_text") == 0) |
8d605269 | 163 | _text = addr; |
b6233d0d | 164 | |
7883a143 MP |
165 | /* Ignore most absolute/undefined (?) symbols. */ |
166 | if (is_ignored_symbol(name, type)) | |
167 | return NULL; | |
168 | ||
8d605269 MY |
169 | check_symbol_range(name, addr, text_ranges, ARRAY_SIZE(text_ranges)); |
170 | check_symbol_range(name, addr, &percpu_range, 1); | |
1da177e4 LT |
171 | |
172 | /* include the type field in the symbol name, so that it gets | |
173 | * compressed together */ | |
1c975da5 | 174 | len++; |
8d605269 | 175 | |
9d1b3895 | 176 | sym = malloc(sizeof(*sym) + len + 1); |
8d605269 | 177 | if (!sym) { |
f1a136e0 JJ |
178 | fprintf(stderr, "kallsyms failure: " |
179 | "unable to allocate required amount of memory\n"); | |
180 | exit(EXIT_FAILURE); | |
181 | } | |
8d605269 MY |
182 | sym->addr = addr; |
183 | sym->len = len; | |
184 | sym->sym[0] = type; | |
9d1b3895 | 185 | strcpy(sym_name(sym), name); |
8d605269 | 186 | sym->percpu_absolute = 0; |
8c996940 | 187 | |
8d605269 | 188 | return sym; |
1da177e4 LT |
189 | } |
190 | ||
4bfe2b78 MY |
191 | static int symbol_in_range(const struct sym_entry *s, |
192 | const struct addr_range *ranges, int entries) | |
17b1f0de MF |
193 | { |
194 | size_t i; | |
4bfe2b78 | 195 | const struct addr_range *ar; |
17b1f0de | 196 | |
78eb7159 KC |
197 | for (i = 0; i < entries; ++i) { |
198 | ar = &ranges[i]; | |
17b1f0de | 199 | |
78eb7159 | 200 | if (s->addr >= ar->start && s->addr <= ar->end) |
ac6ca5c8 | 201 | return 1; |
17b1f0de MF |
202 | } |
203 | ||
ac6ca5c8 | 204 | return 0; |
17b1f0de MF |
205 | } |
206 | ||
a7f13d0f KO |
207 | static bool string_starts_with(const char *s, const char *prefix) |
208 | { | |
209 | return strncmp(s, prefix, strlen(prefix)) == 0; | |
210 | } | |
211 | ||
4bfe2b78 | 212 | static int symbol_valid(const struct sym_entry *s) |
1da177e4 | 213 | { |
29e55ad3 | 214 | const char *name = sym_name(s); |
bd8b22d2 | 215 | |
1da177e4 LT |
216 | /* if --all-symbols is not specified, then symbols outside the text |
217 | * and inittext sections are discarded */ | |
218 | if (!all_symbols) { | |
a7f13d0f KO |
219 | /* |
220 | * Symbols starting with __start and __stop are used to denote | |
221 | * section boundaries, and should always be included: | |
222 | */ | |
223 | if (string_starts_with(name, "__start_") || | |
224 | string_starts_with(name, "__stop_")) | |
225 | return 1; | |
226 | ||
78eb7159 KC |
227 | if (symbol_in_range(s, text_ranges, |
228 | ARRAY_SIZE(text_ranges)) == 0) | |
1da177e4 LT |
229 | return 0; |
230 | /* Corner case. Discard any symbols with the same value as | |
a3b81113 RG |
231 | * _etext _einittext; they can move between pass 1 and 2 when |
232 | * the kallsyms data are added. If these symbols move then | |
233 | * they may get dropped in pass 2, which breaks the kallsyms | |
234 | * rules. | |
1da177e4 | 235 | */ |
17b1f0de | 236 | if ((s->addr == text_range_text->end && |
29e55ad3 | 237 | strcmp(name, text_range_text->end_sym)) || |
17b1f0de | 238 | (s->addr == text_range_inittext->end && |
29e55ad3 | 239 | strcmp(name, text_range_inittext->end_sym))) |
1da177e4 LT |
240 | return 0; |
241 | } | |
242 | ||
1da177e4 LT |
243 | return 1; |
244 | } | |
245 | ||
5e5c4fa7 MY |
246 | /* remove all the invalid symbols from the table */ |
247 | static void shrink_table(void) | |
248 | { | |
249 | unsigned int i, pos; | |
250 | ||
251 | pos = 0; | |
252 | for (i = 0; i < table_cnt; i++) { | |
8d605269 | 253 | if (symbol_valid(table[i])) { |
5e5c4fa7 MY |
254 | if (pos != i) |
255 | table[pos] = table[i]; | |
256 | pos++; | |
257 | } else { | |
8d605269 | 258 | free(table[i]); |
5e5c4fa7 MY |
259 | } |
260 | } | |
261 | table_cnt = pos; | |
262 | ||
263 | /* When valid symbol is not registered, exit to error */ | |
264 | if (!table_cnt) { | |
265 | fprintf(stderr, "No valid symbol.\n"); | |
266 | exit(1); | |
267 | } | |
268 | } | |
269 | ||
aa221f2e | 270 | static void read_map(const char *in) |
1da177e4 | 271 | { |
aa221f2e | 272 | FILE *fp; |
8d605269 | 273 | struct sym_entry *sym; |
1c975da5 MY |
274 | char *buf = NULL; |
275 | size_t buflen = 0; | |
8d605269 | 276 | |
aa221f2e MY |
277 | fp = fopen(in, "r"); |
278 | if (!fp) { | |
279 | perror(in); | |
280 | exit(1); | |
281 | } | |
282 | ||
283 | while (!feof(fp)) { | |
1c975da5 | 284 | sym = read_symbol(fp, &buf, &buflen); |
8d605269 MY |
285 | if (!sym) |
286 | continue; | |
287 | ||
288 | sym->start_pos = table_cnt; | |
289 | ||
b3dbb4ec PM |
290 | if (table_cnt >= table_size) { |
291 | table_size += 10000; | |
292 | table = realloc(table, sizeof(*table) * table_size); | |
1da177e4 LT |
293 | if (!table) { |
294 | fprintf(stderr, "out of memory\n"); | |
aa221f2e | 295 | fclose(fp); |
1da177e4 LT |
296 | exit (1); |
297 | } | |
298 | } | |
8d605269 MY |
299 | |
300 | table[table_cnt++] = sym; | |
1da177e4 | 301 | } |
aa221f2e | 302 | |
1c975da5 | 303 | free(buf); |
aa221f2e | 304 | fclose(fp); |
1da177e4 LT |
305 | } |
306 | ||
4bfe2b78 | 307 | static void output_label(const char *label) |
1da177e4 | 308 | { |
534c9f2e | 309 | printf(".globl %s\n", label); |
1da177e4 | 310 | printf("\tALGN\n"); |
534c9f2e | 311 | printf("%s:\n", label); |
1da177e4 LT |
312 | } |
313 | ||
fd2ab2f6 MY |
314 | /* Provide proper symbols relocatability by their '_text' relativeness. */ |
315 | static void output_address(unsigned long long addr) | |
316 | { | |
317 | if (_text <= addr) | |
318 | printf("\tPTR\t_text + %#llx\n", addr - _text); | |
319 | else | |
320 | printf("\tPTR\t_text - %#llx\n", _text - addr); | |
321 | } | |
322 | ||
1da177e4 LT |
323 | /* uncompress a compressed symbol. When this function is called, the best table |
324 | * might still be compressed itself, so the function needs to be recursive */ | |
4bfe2b78 | 325 | static int expand_symbol(const unsigned char *data, int len, char *result) |
1da177e4 LT |
326 | { |
327 | int c, rlen, total=0; | |
328 | ||
329 | while (len) { | |
330 | c = *data; | |
331 | /* if the table holds a single char that is the same as the one | |
332 | * we are looking for, then end the search */ | |
333 | if (best_table[c][0]==c && best_table_len[c]==1) { | |
334 | *result++ = c; | |
335 | total++; | |
336 | } else { | |
337 | /* if not, recurse and expand */ | |
338 | rlen = expand_symbol(best_table[c], best_table_len[c], result); | |
339 | total += rlen; | |
340 | result += rlen; | |
341 | } | |
342 | data++; | |
343 | len--; | |
344 | } | |
345 | *result=0; | |
346 | ||
347 | return total; | |
348 | } | |
349 | ||
4bfe2b78 | 350 | static int symbol_absolute(const struct sym_entry *s) |
78eb7159 | 351 | { |
8c996940 | 352 | return s->percpu_absolute; |
78eb7159 KC |
353 | } |
354 | ||
010a0aad ZL |
355 | static void cleanup_symbol_name(char *s) |
356 | { | |
357 | char *p; | |
358 | ||
010a0aad ZL |
359 | /* |
360 | * ASCII[.] = 2e | |
361 | * ASCII[0-9] = 30,39 | |
362 | * ASCII[A-Z] = 41,5a | |
363 | * ASCII[_] = 5f | |
364 | * ASCII[a-z] = 61,7a | |
365 | * | |
8cc32a9b YS |
366 | * As above, replacing the first '.' in ".llvm." with '\0' does not |
367 | * affect the main sorting, but it helps us with subsorting. | |
010a0aad | 368 | */ |
8cc32a9b | 369 | p = strstr(s, ".llvm."); |
010a0aad ZL |
370 | if (p) |
371 | *p = '\0'; | |
372 | } | |
373 | ||
60443c88 ZL |
374 | static int compare_names(const void *a, const void *b) |
375 | { | |
376 | int ret; | |
60443c88 ZL |
377 | const struct sym_entry *sa = *(const struct sym_entry **)a; |
378 | const struct sym_entry *sb = *(const struct sym_entry **)b; | |
379 | ||
dd1553b8 | 380 | ret = strcmp(sym_name(sa), sym_name(sb)); |
60443c88 ZL |
381 | if (!ret) { |
382 | if (sa->addr > sb->addr) | |
383 | return 1; | |
384 | else if (sa->addr < sb->addr) | |
385 | return -1; | |
386 | ||
387 | /* keep old order */ | |
388 | return (int)(sa->seq - sb->seq); | |
389 | } | |
390 | ||
391 | return ret; | |
392 | } | |
393 | ||
394 | static void sort_symbols_by_name(void) | |
395 | { | |
396 | qsort(table, table_cnt, sizeof(table[0]), compare_names); | |
397 | } | |
398 | ||
b3dbb4ec | 399 | static void write_src(void) |
1da177e4 | 400 | { |
b3dbb4ec | 401 | unsigned int i, k, off; |
1da177e4 LT |
402 | unsigned int best_idx[256]; |
403 | unsigned int *markers; | |
9281acea | 404 | char buf[KSYM_NAME_LEN]; |
1da177e4 | 405 | |
500193ec | 406 | printf("#include <asm/bitsperlong.h>\n"); |
1da177e4 LT |
407 | printf("#if BITS_PER_LONG == 64\n"); |
408 | printf("#define PTR .quad\n"); | |
72d3ebb9 | 409 | printf("#define ALGN .balign 8\n"); |
1da177e4 LT |
410 | printf("#else\n"); |
411 | printf("#define PTR .long\n"); | |
72d3ebb9 | 412 | printf("#define ALGN .balign 4\n"); |
1da177e4 LT |
413 | printf("#endif\n"); |
414 | ||
aad09470 | 415 | printf("\t.section .rodata, \"a\"\n"); |
1da177e4 | 416 | |
1da177e4 | 417 | output_label("kallsyms_num_syms"); |
80ffbaa5 | 418 | printf("\t.long\t%u\n", table_cnt); |
1da177e4 LT |
419 | printf("\n"); |
420 | ||
421 | /* table of offset markers, that give the offset in the compressed stream | |
422 | * every 256 symbols */ | |
f1a136e0 JJ |
423 | markers = malloc(sizeof(unsigned int) * ((table_cnt + 255) / 256)); |
424 | if (!markers) { | |
425 | fprintf(stderr, "kallsyms failure: " | |
426 | "unable to allocate required memory\n"); | |
427 | exit(EXIT_FAILURE); | |
428 | } | |
1da177e4 LT |
429 | |
430 | output_label("kallsyms_names"); | |
1da177e4 | 431 | off = 0; |
b3dbb4ec PM |
432 | for (i = 0; i < table_cnt; i++) { |
433 | if ((i & 0xFF) == 0) | |
434 | markers[i >> 8] = off; | |
60443c88 | 435 | table[i]->seq = i; |
1da177e4 | 436 | |
73bbb944 MO |
437 | /* There cannot be any symbol of length zero. */ |
438 | if (table[i]->len == 0) { | |
439 | fprintf(stderr, "kallsyms failure: " | |
440 | "unexpected zero symbol length\n"); | |
441 | exit(EXIT_FAILURE); | |
442 | } | |
443 | ||
444 | /* Only lengths that fit in up-to-two-byte ULEB128 are supported. */ | |
445 | if (table[i]->len > 0x3FFF) { | |
446 | fprintf(stderr, "kallsyms failure: " | |
447 | "unexpected huge symbol length\n"); | |
448 | exit(EXIT_FAILURE); | |
449 | } | |
450 | ||
451 | /* Encode length with ULEB128. */ | |
452 | if (table[i]->len <= 0x7F) { | |
453 | /* Most symbols use a single byte for the length. */ | |
454 | printf("\t.byte 0x%02x", table[i]->len); | |
455 | off += table[i]->len + 1; | |
456 | } else { | |
457 | /* "Big" symbols use two bytes. */ | |
458 | printf("\t.byte 0x%02x, 0x%02x", | |
459 | (table[i]->len & 0x7F) | 0x80, | |
460 | (table[i]->len >> 7) & 0x7F); | |
461 | off += table[i]->len + 2; | |
462 | } | |
8d605269 MY |
463 | for (k = 0; k < table[i]->len; k++) |
464 | printf(", 0x%02x", table[i]->sym[k]); | |
1da177e4 | 465 | printf("\n"); |
1da177e4 LT |
466 | } |
467 | printf("\n"); | |
468 | ||
dd1553b8 MY |
469 | /* |
470 | * Now that we wrote out the compressed symbol names, restore the | |
471 | * original names, which are needed in some of the later steps. | |
472 | */ | |
473 | for (i = 0; i < table_cnt; i++) { | |
474 | expand_symbol(table[i]->sym, table[i]->len, buf); | |
475 | strcpy((char *)table[i]->sym, buf); | |
476 | } | |
477 | ||
1da177e4 | 478 | output_label("kallsyms_markers"); |
b3dbb4ec | 479 | for (i = 0; i < ((table_cnt + 255) >> 8); i++) |
80ffbaa5 | 480 | printf("\t.long\t%u\n", markers[i]); |
1da177e4 LT |
481 | printf("\n"); |
482 | ||
483 | free(markers); | |
484 | ||
485 | output_label("kallsyms_token_table"); | |
486 | off = 0; | |
487 | for (i = 0; i < 256; i++) { | |
488 | best_idx[i] = off; | |
b3dbb4ec | 489 | expand_symbol(best_table[i], best_table_len[i], buf); |
1da177e4 LT |
490 | printf("\t.asciz\t\"%s\"\n", buf); |
491 | off += strlen(buf) + 1; | |
492 | } | |
493 | printf("\n"); | |
494 | ||
495 | output_label("kallsyms_token_index"); | |
496 | for (i = 0; i < 256; i++) | |
497 | printf("\t.short\t%d\n", best_idx[i]); | |
498 | printf("\n"); | |
404bad70 MY |
499 | |
500 | if (!base_relative) | |
501 | output_label("kallsyms_addresses"); | |
502 | else | |
503 | output_label("kallsyms_offsets"); | |
504 | ||
505 | for (i = 0; i < table_cnt; i++) { | |
506 | if (base_relative) { | |
507 | /* | |
508 | * Use the offset relative to the lowest value | |
509 | * encountered of all relative symbols, and emit | |
510 | * non-relocatable fixed offsets that will be fixed | |
511 | * up at runtime. | |
512 | */ | |
513 | ||
514 | long long offset; | |
515 | int overflow; | |
516 | ||
517 | if (!absolute_percpu) { | |
518 | offset = table[i]->addr - relative_base; | |
519 | overflow = (offset < 0 || offset > UINT_MAX); | |
520 | } else if (symbol_absolute(table[i])) { | |
521 | offset = table[i]->addr; | |
522 | overflow = (offset < 0 || offset > INT_MAX); | |
523 | } else { | |
524 | offset = relative_base - table[i]->addr - 1; | |
525 | overflow = (offset < INT_MIN || offset >= 0); | |
526 | } | |
527 | if (overflow) { | |
528 | fprintf(stderr, "kallsyms failure: " | |
529 | "%s symbol value %#llx out of range in relative mode\n", | |
530 | symbol_absolute(table[i]) ? "absolute" : "relative", | |
531 | table[i]->addr); | |
532 | exit(EXIT_FAILURE); | |
533 | } | |
dd1553b8 | 534 | printf("\t.long\t%#x /* %s */\n", (int)offset, table[i]->sym); |
404bad70 MY |
535 | } else if (!symbol_absolute(table[i])) { |
536 | output_address(table[i]->addr); | |
537 | } else { | |
538 | printf("\tPTR\t%#llx\n", table[i]->addr); | |
539 | } | |
540 | } | |
541 | printf("\n"); | |
542 | ||
543 | if (base_relative) { | |
544 | output_label("kallsyms_relative_base"); | |
545 | output_address(relative_base); | |
546 | printf("\n"); | |
547 | } | |
548 | ||
dd1553b8 MY |
549 | if (lto_clang) |
550 | for (i = 0; i < table_cnt; i++) | |
551 | cleanup_symbol_name((char *)table[i]->sym); | |
552 | ||
404bad70 MY |
553 | sort_symbols_by_name(); |
554 | output_label("kallsyms_seqs_of_names"); | |
555 | for (i = 0; i < table_cnt; i++) | |
556 | printf("\t.byte 0x%02x, 0x%02x, 0x%02x\n", | |
557 | (unsigned char)(table[i]->seq >> 16), | |
558 | (unsigned char)(table[i]->seq >> 8), | |
559 | (unsigned char)(table[i]->seq >> 0)); | |
560 | printf("\n"); | |
1da177e4 LT |
561 | } |
562 | ||
563 | ||
564 | /* table lookup compression functions */ | |
565 | ||
1da177e4 | 566 | /* count all the possible tokens in a symbol */ |
4bfe2b78 | 567 | static void learn_symbol(const unsigned char *symbol, int len) |
1da177e4 LT |
568 | { |
569 | int i; | |
570 | ||
571 | for (i = 0; i < len - 1; i++) | |
b3dbb4ec | 572 | token_profit[ symbol[i] + (symbol[i + 1] << 8) ]++; |
1da177e4 LT |
573 | } |
574 | ||
575 | /* decrease the count for all the possible tokens in a symbol */ | |
4bfe2b78 | 576 | static void forget_symbol(const unsigned char *symbol, int len) |
1da177e4 LT |
577 | { |
578 | int i; | |
579 | ||
580 | for (i = 0; i < len - 1; i++) | |
b3dbb4ec | 581 | token_profit[ symbol[i] + (symbol[i + 1] << 8) ]--; |
1da177e4 LT |
582 | } |
583 | ||
5e5c4fa7 | 584 | /* do the initial token count */ |
fcdf7197 | 585 | static void build_initial_token_table(void) |
1da177e4 | 586 | { |
5e5c4fa7 | 587 | unsigned int i; |
1da177e4 | 588 | |
5e5c4fa7 | 589 | for (i = 0; i < table_cnt; i++) |
8d605269 | 590 | learn_symbol(table[i]->sym, table[i]->len); |
1da177e4 LT |
591 | } |
592 | ||
2558c138 | 593 | static unsigned char *find_token(unsigned char *str, int len, |
4bfe2b78 | 594 | const unsigned char *token) |
7c5d249a PM |
595 | { |
596 | int i; | |
597 | ||
598 | for (i = 0; i < len - 1; i++) { | |
599 | if (str[i] == token[0] && str[i+1] == token[1]) | |
600 | return &str[i]; | |
601 | } | |
602 | return NULL; | |
603 | } | |
604 | ||
1da177e4 LT |
605 | /* replace a given token in all the valid symbols. Use the sampled symbols |
606 | * to update the counts */ | |
4bfe2b78 | 607 | static void compress_symbols(const unsigned char *str, int idx) |
1da177e4 | 608 | { |
b3dbb4ec PM |
609 | unsigned int i, len, size; |
610 | unsigned char *p1, *p2; | |
1da177e4 | 611 | |
b3dbb4ec | 612 | for (i = 0; i < table_cnt; i++) { |
1da177e4 | 613 | |
8d605269 MY |
614 | len = table[i]->len; |
615 | p1 = table[i]->sym; | |
b3dbb4ec PM |
616 | |
617 | /* find the token on the symbol */ | |
7c5d249a | 618 | p2 = find_token(p1, len, str); |
b3dbb4ec PM |
619 | if (!p2) continue; |
620 | ||
621 | /* decrease the counts for this symbol's tokens */ | |
8d605269 | 622 | forget_symbol(table[i]->sym, len); |
b3dbb4ec PM |
623 | |
624 | size = len; | |
1da177e4 LT |
625 | |
626 | do { | |
b3dbb4ec PM |
627 | *p2 = idx; |
628 | p2++; | |
629 | size -= (p2 - p1); | |
630 | memmove(p2, p2 + 1, size); | |
631 | p1 = p2; | |
632 | len--; | |
633 | ||
634 | if (size < 2) break; | |
635 | ||
1da177e4 | 636 | /* find the token on the symbol */ |
7c5d249a | 637 | p2 = find_token(p1, size, str); |
1da177e4 | 638 | |
b3dbb4ec | 639 | } while (p2); |
1da177e4 | 640 | |
8d605269 | 641 | table[i]->len = len; |
1da177e4 | 642 | |
b3dbb4ec | 643 | /* increase the counts for this symbol's new tokens */ |
8d605269 | 644 | learn_symbol(table[i]->sym, len); |
1da177e4 LT |
645 | } |
646 | } | |
647 | ||
648 | /* search the token with the maximum profit */ | |
b3dbb4ec | 649 | static int find_best_token(void) |
1da177e4 | 650 | { |
b3dbb4ec | 651 | int i, best, bestprofit; |
1da177e4 LT |
652 | |
653 | bestprofit=-10000; | |
b3dbb4ec | 654 | best = 0; |
1da177e4 | 655 | |
b3dbb4ec PM |
656 | for (i = 0; i < 0x10000; i++) { |
657 | if (token_profit[i] > bestprofit) { | |
658 | best = i; | |
659 | bestprofit = token_profit[i]; | |
1da177e4 | 660 | } |
1da177e4 | 661 | } |
1da177e4 LT |
662 | return best; |
663 | } | |
664 | ||
665 | /* this is the core of the algorithm: calculate the "best" table */ | |
666 | static void optimize_result(void) | |
667 | { | |
b3dbb4ec | 668 | int i, best; |
1da177e4 LT |
669 | |
670 | /* using the '\0' symbol last allows compress_symbols to use standard | |
671 | * fast string functions */ | |
672 | for (i = 255; i >= 0; i--) { | |
673 | ||
674 | /* if this table slot is empty (it is not used by an actual | |
675 | * original char code */ | |
676 | if (!best_table_len[i]) { | |
677 | ||
cbf7a90e | 678 | /* find the token with the best profit value */ |
1da177e4 | 679 | best = find_best_token(); |
e0a04b11 XW |
680 | if (token_profit[best] == 0) |
681 | break; | |
1da177e4 LT |
682 | |
683 | /* place it in the "best" table */ | |
b3dbb4ec PM |
684 | best_table_len[i] = 2; |
685 | best_table[i][0] = best & 0xFF; | |
686 | best_table[i][1] = (best >> 8) & 0xFF; | |
1da177e4 LT |
687 | |
688 | /* replace this token in all the valid symbols */ | |
b3dbb4ec | 689 | compress_symbols(best_table[i], i); |
1da177e4 LT |
690 | } |
691 | } | |
692 | } | |
693 | ||
694 | /* start by placing the symbols that are actually used on the table */ | |
695 | static void insert_real_symbols_in_table(void) | |
696 | { | |
b3dbb4ec | 697 | unsigned int i, j, c; |
1da177e4 | 698 | |
b3dbb4ec | 699 | for (i = 0; i < table_cnt; i++) { |
8d605269 MY |
700 | for (j = 0; j < table[i]->len; j++) { |
701 | c = table[i]->sym[j]; | |
b3dbb4ec PM |
702 | best_table[c][0]=c; |
703 | best_table_len[c]=1; | |
1da177e4 LT |
704 | } |
705 | } | |
706 | } | |
707 | ||
708 | static void optimize_token_table(void) | |
709 | { | |
fcdf7197 | 710 | build_initial_token_table(); |
1da177e4 LT |
711 | |
712 | insert_real_symbols_in_table(); | |
713 | ||
714 | optimize_result(); | |
715 | } | |
716 | ||
b478b782 LJ |
717 | /* guess for "linker script provide" symbol */ |
718 | static int may_be_linker_script_provide_symbol(const struct sym_entry *se) | |
719 | { | |
29e55ad3 | 720 | const char *symbol = sym_name(se); |
b478b782 LJ |
721 | int len = se->len - 1; |
722 | ||
723 | if (len < 8) | |
724 | return 0; | |
725 | ||
726 | if (symbol[0] != '_' || symbol[1] != '_') | |
727 | return 0; | |
728 | ||
729 | /* __start_XXXXX */ | |
730 | if (!memcmp(symbol + 2, "start_", 6)) | |
731 | return 1; | |
732 | ||
733 | /* __stop_XXXXX */ | |
734 | if (!memcmp(symbol + 2, "stop_", 5)) | |
735 | return 1; | |
736 | ||
737 | /* __end_XXXXX */ | |
738 | if (!memcmp(symbol + 2, "end_", 4)) | |
739 | return 1; | |
740 | ||
741 | /* __XXXXX_start */ | |
742 | if (!memcmp(symbol + len - 6, "_start", 6)) | |
743 | return 1; | |
744 | ||
745 | /* __XXXXX_end */ | |
746 | if (!memcmp(symbol + len - 4, "_end", 4)) | |
747 | return 1; | |
748 | ||
749 | return 0; | |
750 | } | |
751 | ||
f2df3f65 PM |
752 | static int compare_symbols(const void *a, const void *b) |
753 | { | |
8d605269 MY |
754 | const struct sym_entry *sa = *(const struct sym_entry **)a; |
755 | const struct sym_entry *sb = *(const struct sym_entry **)b; | |
f2df3f65 PM |
756 | int wa, wb; |
757 | ||
f2df3f65 PM |
758 | /* sort by address first */ |
759 | if (sa->addr > sb->addr) | |
760 | return 1; | |
761 | if (sa->addr < sb->addr) | |
762 | return -1; | |
763 | ||
764 | /* sort by "weakness" type */ | |
765 | wa = (sa->sym[0] == 'w') || (sa->sym[0] == 'W'); | |
766 | wb = (sb->sym[0] == 'w') || (sb->sym[0] == 'W'); | |
767 | if (wa != wb) | |
768 | return wa - wb; | |
769 | ||
b478b782 LJ |
770 | /* sort by "linker script provide" type */ |
771 | wa = may_be_linker_script_provide_symbol(sa); | |
772 | wb = may_be_linker_script_provide_symbol(sb); | |
773 | if (wa != wb) | |
774 | return wa - wb; | |
775 | ||
776 | /* sort by the number of prefix underscores */ | |
aa915245 MY |
777 | wa = strspn(sym_name(sa), "_"); |
778 | wb = strspn(sym_name(sb), "_"); | |
b478b782 LJ |
779 | if (wa != wb) |
780 | return wa - wb; | |
781 | ||
f2df3f65 PM |
782 | /* sort by initial order, so that other symbols are left undisturbed */ |
783 | return sa->start_pos - sb->start_pos; | |
784 | } | |
785 | ||
786 | static void sort_symbols(void) | |
787 | { | |
8d605269 | 788 | qsort(table, table_cnt, sizeof(table[0]), compare_symbols); |
f2df3f65 | 789 | } |
1da177e4 | 790 | |
c6bda7c9 RR |
791 | static void make_percpus_absolute(void) |
792 | { | |
793 | unsigned int i; | |
794 | ||
795 | for (i = 0; i < table_cnt; i++) | |
8d605269 | 796 | if (symbol_in_range(table[i], &percpu_range, 1)) { |
8c996940 AB |
797 | /* |
798 | * Keep the 'A' override for percpu symbols to | |
799 | * ensure consistent behavior compared to older | |
800 | * versions of this tool. | |
801 | */ | |
8d605269 MY |
802 | table[i]->sym[0] = 'A'; |
803 | table[i]->percpu_absolute = 1; | |
8c996940 | 804 | } |
c6bda7c9 RR |
805 | } |
806 | ||
2213e9a6 AB |
807 | /* find the minimum non-absolute symbol address */ |
808 | static void record_relative_base(void) | |
809 | { | |
810 | unsigned int i; | |
811 | ||
2213e9a6 | 812 | for (i = 0; i < table_cnt; i++) |
8d605269 | 813 | if (!symbol_absolute(table[i])) { |
f34ea029 MY |
814 | /* |
815 | * The table is sorted by address. | |
816 | * Take the first non-absolute symbol value. | |
817 | */ | |
8d605269 | 818 | relative_base = table[i]->addr; |
f34ea029 MY |
819 | return; |
820 | } | |
2213e9a6 AB |
821 | } |
822 | ||
b3dbb4ec | 823 | int main(int argc, char **argv) |
1da177e4 | 824 | { |
aa221f2e | 825 | while (1) { |
92e74fb6 | 826 | static const struct option long_options[] = { |
aa221f2e MY |
827 | {"all-symbols", no_argument, &all_symbols, 1}, |
828 | {"absolute-percpu", no_argument, &absolute_percpu, 1}, | |
829 | {"base-relative", no_argument, &base_relative, 1}, | |
010a0aad | 830 | {"lto-clang", no_argument, <o_clang, 1}, |
aa221f2e MY |
831 | {}, |
832 | }; | |
833 | ||
834 | int c = getopt_long(argc, argv, "", long_options, NULL); | |
835 | ||
836 | if (c == -1) | |
837 | break; | |
838 | if (c != 0) | |
839 | usage(); | |
840 | } | |
841 | ||
842 | if (optind >= argc) | |
1da177e4 LT |
843 | usage(); |
844 | ||
aa221f2e | 845 | read_map(argv[optind]); |
5e5c4fa7 | 846 | shrink_table(); |
c6bda7c9 RR |
847 | if (absolute_percpu) |
848 | make_percpus_absolute(); | |
f34ea029 | 849 | sort_symbols(); |
2213e9a6 AB |
850 | if (base_relative) |
851 | record_relative_base(); | |
2ea03891 | 852 | optimize_token_table(); |
1da177e4 LT |
853 | write_src(); |
854 | ||
855 | return 0; | |
856 | } |