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 | ||
aa221f2e | 22 | #include <getopt.h> |
a41333e0 | 23 | #include <stdbool.h> |
1da177e4 LT |
24 | #include <stdio.h> |
25 | #include <stdlib.h> | |
26 | #include <string.h> | |
27 | #include <ctype.h> | |
2213e9a6 | 28 | #include <limits.h> |
1da177e4 | 29 | |
17b1f0de | 30 | #define ARRAY_SIZE(arr) (sizeof(arr) / sizeof(arr[0])) |
17b1f0de | 31 | |
b471927e BF |
32 | #define _stringify_1(x) #x |
33 | #define _stringify(x) _stringify_1(x) | |
34 | ||
b8a94bfb | 35 | #define KSYM_NAME_LEN 512 |
1da177e4 | 36 | |
6e8c5bbd MO |
37 | /* |
38 | * A substantially bigger size than the current maximum. | |
39 | * | |
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. | |
43 | */ | |
b8a94bfb | 44 | #define KSYM_NAME_LEN_BUFFER 2048 |
6e8c5bbd MO |
45 | _Static_assert( |
46 | KSYM_NAME_LEN_BUFFER == KSYM_NAME_LEN * 4, | |
47 | "Please keep KSYM_NAME_LEN_BUFFER in sync with KSYM_NAME_LEN" | |
48 | ); | |
b471927e | 49 | |
1da177e4 LT |
50 | struct sym_entry { |
51 | unsigned long long addr; | |
b3dbb4ec | 52 | unsigned int len; |
60443c88 | 53 | unsigned int seq; |
f2df3f65 | 54 | unsigned int start_pos; |
8c996940 | 55 | unsigned int percpu_absolute; |
9d82973e | 56 | unsigned char sym[]; |
1da177e4 LT |
57 | }; |
58 | ||
78eb7159 KC |
59 | struct addr_range { |
60 | const char *start_sym, *end_sym; | |
17b1f0de MF |
61 | unsigned long long start, end; |
62 | }; | |
63 | ||
64 | static unsigned long long _text; | |
2213e9a6 | 65 | static unsigned long long relative_base; |
78eb7159 | 66 | static struct addr_range text_ranges[] = { |
17b1f0de MF |
67 | { "_stext", "_etext" }, |
68 | { "_sinittext", "_einittext" }, | |
17b1f0de MF |
69 | }; |
70 | #define text_range_text (&text_ranges[0]) | |
71 | #define text_range_inittext (&text_ranges[1]) | |
72 | ||
c6bda7c9 RR |
73 | static struct addr_range percpu_range = { |
74 | "__per_cpu_start", "__per_cpu_end", -1ULL, 0 | |
75 | }; | |
76 | ||
8d605269 | 77 | static struct sym_entry **table; |
b3dbb4ec | 78 | static unsigned int table_size, table_cnt; |
831362fc MY |
79 | static int all_symbols; |
80 | static int absolute_percpu; | |
81 | static int base_relative; | |
010a0aad | 82 | static int lto_clang; |
1da177e4 | 83 | |
f43e9daa | 84 | static int token_profit[0x10000]; |
1da177e4 LT |
85 | |
86 | /* the table that holds the result of the compression */ | |
f43e9daa MY |
87 | static unsigned char best_table[256][2]; |
88 | static unsigned char best_table_len[256]; | |
1da177e4 LT |
89 | |
90 | ||
b3dbb4ec | 91 | static void usage(void) |
1da177e4 | 92 | { |
8d3a7507 | 93 | fprintf(stderr, "Usage: kallsyms [--all-symbols] [--absolute-percpu] " |
010a0aad | 94 | "[--base-relative] [--lto-clang] in.map > out.S\n"); |
1da177e4 LT |
95 | exit(1); |
96 | } | |
97 | ||
29e55ad3 MY |
98 | static char *sym_name(const struct sym_entry *s) |
99 | { | |
100 | return (char *)s->sym + 1; | |
101 | } | |
102 | ||
a41333e0 MY |
103 | static bool is_ignored_symbol(const char *name, char type) |
104 | { | |
a7b00a18 | 105 | if (type == 'u' || type == 'n') |
887df76d MY |
106 | return true; |
107 | ||
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")) | |
114 | return true; | |
115 | } | |
116 | ||
a41333e0 MY |
117 | return false; |
118 | } | |
119 | ||
b6233d0d MY |
120 | static void check_symbol_range(const char *sym, unsigned long long addr, |
121 | struct addr_range *ranges, int entries) | |
17b1f0de MF |
122 | { |
123 | size_t i; | |
78eb7159 | 124 | struct addr_range *ar; |
17b1f0de | 125 | |
78eb7159 KC |
126 | for (i = 0; i < entries; ++i) { |
127 | ar = &ranges[i]; | |
17b1f0de | 128 | |
78eb7159 KC |
129 | if (strcmp(sym, ar->start_sym) == 0) { |
130 | ar->start = addr; | |
b6233d0d | 131 | return; |
78eb7159 KC |
132 | } else if (strcmp(sym, ar->end_sym) == 0) { |
133 | ar->end = addr; | |
b6233d0d | 134 | return; |
17b1f0de MF |
135 | } |
136 | } | |
17b1f0de MF |
137 | } |
138 | ||
8d605269 | 139 | static struct sym_entry *read_symbol(FILE *in) |
1da177e4 | 140 | { |
b471927e | 141 | char name[KSYM_NAME_LEN_BUFFER+1], type; |
8d605269 MY |
142 | unsigned long long addr; |
143 | unsigned int len; | |
144 | struct sym_entry *sym; | |
1da177e4 LT |
145 | int rc; |
146 | ||
b471927e | 147 | rc = fscanf(in, "%llx %c %" _stringify(KSYM_NAME_LEN_BUFFER) "s\n", &addr, &type, name); |
1da177e4 | 148 | if (rc != 3) { |
b66c874f | 149 | if (rc != EOF && fgets(name, ARRAY_SIZE(name), in) == NULL) |
ef894870 | 150 | fprintf(stderr, "Read error or end of file.\n"); |
8d605269 | 151 | return NULL; |
1da177e4 | 152 | } |
be9f6133 | 153 | if (strlen(name) >= KSYM_NAME_LEN) { |
6db2983c | 154 | fprintf(stderr, "Symbol %s too long for kallsyms (%zu >= %d).\n" |
bb66fc67 | 155 | "Please increase KSYM_NAME_LEN both in kernel and kallsyms.c\n", |
be9f6133 | 156 | name, strlen(name), KSYM_NAME_LEN); |
8d605269 | 157 | return NULL; |
f3462aa9 | 158 | } |
1da177e4 | 159 | |
be9f6133 | 160 | if (strcmp(name, "_text") == 0) |
8d605269 | 161 | _text = addr; |
b6233d0d | 162 | |
7883a143 MP |
163 | /* Ignore most absolute/undefined (?) symbols. */ |
164 | if (is_ignored_symbol(name, type)) | |
165 | return NULL; | |
166 | ||
8d605269 MY |
167 | check_symbol_range(name, addr, text_ranges, ARRAY_SIZE(text_ranges)); |
168 | check_symbol_range(name, addr, &percpu_range, 1); | |
1da177e4 LT |
169 | |
170 | /* include the type field in the symbol name, so that it gets | |
171 | * compressed together */ | |
8d605269 MY |
172 | |
173 | len = strlen(name) + 1; | |
174 | ||
9d1b3895 | 175 | sym = malloc(sizeof(*sym) + len + 1); |
8d605269 | 176 | if (!sym) { |
f1a136e0 JJ |
177 | fprintf(stderr, "kallsyms failure: " |
178 | "unable to allocate required amount of memory\n"); | |
179 | exit(EXIT_FAILURE); | |
180 | } | |
8d605269 MY |
181 | sym->addr = addr; |
182 | sym->len = len; | |
183 | sym->sym[0] = type; | |
9d1b3895 | 184 | strcpy(sym_name(sym), name); |
8d605269 | 185 | sym->percpu_absolute = 0; |
8c996940 | 186 | |
8d605269 | 187 | return sym; |
1da177e4 LT |
188 | } |
189 | ||
4bfe2b78 MY |
190 | static int symbol_in_range(const struct sym_entry *s, |
191 | const struct addr_range *ranges, int entries) | |
17b1f0de MF |
192 | { |
193 | size_t i; | |
4bfe2b78 | 194 | const struct addr_range *ar; |
17b1f0de | 195 | |
78eb7159 KC |
196 | for (i = 0; i < entries; ++i) { |
197 | ar = &ranges[i]; | |
17b1f0de | 198 | |
78eb7159 | 199 | if (s->addr >= ar->start && s->addr <= ar->end) |
ac6ca5c8 | 200 | return 1; |
17b1f0de MF |
201 | } |
202 | ||
ac6ca5c8 | 203 | return 0; |
17b1f0de MF |
204 | } |
205 | ||
4bfe2b78 | 206 | static int symbol_valid(const struct sym_entry *s) |
1da177e4 | 207 | { |
29e55ad3 | 208 | const char *name = sym_name(s); |
bd8b22d2 | 209 | |
1da177e4 LT |
210 | /* if --all-symbols is not specified, then symbols outside the text |
211 | * and inittext sections are discarded */ | |
212 | if (!all_symbols) { | |
78eb7159 KC |
213 | if (symbol_in_range(s, text_ranges, |
214 | ARRAY_SIZE(text_ranges)) == 0) | |
1da177e4 LT |
215 | return 0; |
216 | /* Corner case. Discard any symbols with the same value as | |
a3b81113 RG |
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 | |
220 | * rules. | |
1da177e4 | 221 | */ |
17b1f0de | 222 | if ((s->addr == text_range_text->end && |
29e55ad3 | 223 | strcmp(name, text_range_text->end_sym)) || |
17b1f0de | 224 | (s->addr == text_range_inittext->end && |
29e55ad3 | 225 | strcmp(name, text_range_inittext->end_sym))) |
1da177e4 LT |
226 | return 0; |
227 | } | |
228 | ||
1da177e4 LT |
229 | return 1; |
230 | } | |
231 | ||
5e5c4fa7 MY |
232 | /* remove all the invalid symbols from the table */ |
233 | static void shrink_table(void) | |
234 | { | |
235 | unsigned int i, pos; | |
236 | ||
237 | pos = 0; | |
238 | for (i = 0; i < table_cnt; i++) { | |
8d605269 | 239 | if (symbol_valid(table[i])) { |
5e5c4fa7 MY |
240 | if (pos != i) |
241 | table[pos] = table[i]; | |
242 | pos++; | |
243 | } else { | |
8d605269 | 244 | free(table[i]); |
5e5c4fa7 MY |
245 | } |
246 | } | |
247 | table_cnt = pos; | |
248 | ||
249 | /* When valid symbol is not registered, exit to error */ | |
250 | if (!table_cnt) { | |
251 | fprintf(stderr, "No valid symbol.\n"); | |
252 | exit(1); | |
253 | } | |
254 | } | |
255 | ||
aa221f2e | 256 | static void read_map(const char *in) |
1da177e4 | 257 | { |
aa221f2e | 258 | FILE *fp; |
8d605269 MY |
259 | struct sym_entry *sym; |
260 | ||
aa221f2e MY |
261 | fp = fopen(in, "r"); |
262 | if (!fp) { | |
263 | perror(in); | |
264 | exit(1); | |
265 | } | |
266 | ||
267 | while (!feof(fp)) { | |
268 | sym = read_symbol(fp); | |
8d605269 MY |
269 | if (!sym) |
270 | continue; | |
271 | ||
272 | sym->start_pos = table_cnt; | |
273 | ||
b3dbb4ec PM |
274 | if (table_cnt >= table_size) { |
275 | table_size += 10000; | |
276 | table = realloc(table, sizeof(*table) * table_size); | |
1da177e4 LT |
277 | if (!table) { |
278 | fprintf(stderr, "out of memory\n"); | |
aa221f2e | 279 | fclose(fp); |
1da177e4 LT |
280 | exit (1); |
281 | } | |
282 | } | |
8d605269 MY |
283 | |
284 | table[table_cnt++] = sym; | |
1da177e4 | 285 | } |
aa221f2e MY |
286 | |
287 | fclose(fp); | |
1da177e4 LT |
288 | } |
289 | ||
4bfe2b78 | 290 | static void output_label(const char *label) |
1da177e4 | 291 | { |
534c9f2e | 292 | printf(".globl %s\n", label); |
1da177e4 | 293 | printf("\tALGN\n"); |
534c9f2e | 294 | printf("%s:\n", label); |
1da177e4 LT |
295 | } |
296 | ||
fd2ab2f6 MY |
297 | /* Provide proper symbols relocatability by their '_text' relativeness. */ |
298 | static void output_address(unsigned long long addr) | |
299 | { | |
300 | if (_text <= addr) | |
301 | printf("\tPTR\t_text + %#llx\n", addr - _text); | |
302 | else | |
303 | printf("\tPTR\t_text - %#llx\n", _text - addr); | |
304 | } | |
305 | ||
1da177e4 LT |
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 */ | |
4bfe2b78 | 308 | static int expand_symbol(const unsigned char *data, int len, char *result) |
1da177e4 LT |
309 | { |
310 | int c, rlen, total=0; | |
311 | ||
312 | while (len) { | |
313 | c = *data; | |
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) { | |
317 | *result++ = c; | |
318 | total++; | |
319 | } else { | |
320 | /* if not, recurse and expand */ | |
321 | rlen = expand_symbol(best_table[c], best_table_len[c], result); | |
322 | total += rlen; | |
323 | result += rlen; | |
324 | } | |
325 | data++; | |
326 | len--; | |
327 | } | |
328 | *result=0; | |
329 | ||
330 | return total; | |
331 | } | |
332 | ||
4bfe2b78 | 333 | static int symbol_absolute(const struct sym_entry *s) |
78eb7159 | 334 | { |
8c996940 | 335 | return s->percpu_absolute; |
78eb7159 KC |
336 | } |
337 | ||
010a0aad ZL |
338 | static void cleanup_symbol_name(char *s) |
339 | { | |
340 | char *p; | |
341 | ||
010a0aad ZL |
342 | /* |
343 | * ASCII[.] = 2e | |
344 | * ASCII[0-9] = 30,39 | |
345 | * ASCII[A-Z] = 41,5a | |
346 | * ASCII[_] = 5f | |
347 | * ASCII[a-z] = 61,7a | |
348 | * | |
349 | * As above, replacing '.' with '\0' does not affect the main sorting, | |
350 | * but it helps us with subsorting. | |
351 | */ | |
352 | p = strchr(s, '.'); | |
353 | if (p) | |
354 | *p = '\0'; | |
355 | } | |
356 | ||
60443c88 ZL |
357 | static int compare_names(const void *a, const void *b) |
358 | { | |
359 | int ret; | |
60443c88 ZL |
360 | const struct sym_entry *sa = *(const struct sym_entry **)a; |
361 | const struct sym_entry *sb = *(const struct sym_entry **)b; | |
362 | ||
dd1553b8 | 363 | ret = strcmp(sym_name(sa), sym_name(sb)); |
60443c88 ZL |
364 | if (!ret) { |
365 | if (sa->addr > sb->addr) | |
366 | return 1; | |
367 | else if (sa->addr < sb->addr) | |
368 | return -1; | |
369 | ||
370 | /* keep old order */ | |
371 | return (int)(sa->seq - sb->seq); | |
372 | } | |
373 | ||
374 | return ret; | |
375 | } | |
376 | ||
377 | static void sort_symbols_by_name(void) | |
378 | { | |
379 | qsort(table, table_cnt, sizeof(table[0]), compare_names); | |
380 | } | |
381 | ||
b3dbb4ec | 382 | static void write_src(void) |
1da177e4 | 383 | { |
b3dbb4ec | 384 | unsigned int i, k, off; |
1da177e4 LT |
385 | unsigned int best_idx[256]; |
386 | unsigned int *markers; | |
9281acea | 387 | char buf[KSYM_NAME_LEN]; |
1da177e4 | 388 | |
500193ec | 389 | printf("#include <asm/bitsperlong.h>\n"); |
1da177e4 LT |
390 | printf("#if BITS_PER_LONG == 64\n"); |
391 | printf("#define PTR .quad\n"); | |
72d3ebb9 | 392 | printf("#define ALGN .balign 8\n"); |
1da177e4 LT |
393 | printf("#else\n"); |
394 | printf("#define PTR .long\n"); | |
72d3ebb9 | 395 | printf("#define ALGN .balign 4\n"); |
1da177e4 LT |
396 | printf("#endif\n"); |
397 | ||
aad09470 | 398 | printf("\t.section .rodata, \"a\"\n"); |
1da177e4 | 399 | |
1da177e4 | 400 | output_label("kallsyms_num_syms"); |
80ffbaa5 | 401 | printf("\t.long\t%u\n", table_cnt); |
1da177e4 LT |
402 | printf("\n"); |
403 | ||
404 | /* table of offset markers, that give the offset in the compressed stream | |
405 | * every 256 symbols */ | |
f1a136e0 JJ |
406 | markers = malloc(sizeof(unsigned int) * ((table_cnt + 255) / 256)); |
407 | if (!markers) { | |
408 | fprintf(stderr, "kallsyms failure: " | |
409 | "unable to allocate required memory\n"); | |
410 | exit(EXIT_FAILURE); | |
411 | } | |
1da177e4 LT |
412 | |
413 | output_label("kallsyms_names"); | |
1da177e4 | 414 | off = 0; |
b3dbb4ec PM |
415 | for (i = 0; i < table_cnt; i++) { |
416 | if ((i & 0xFF) == 0) | |
417 | markers[i >> 8] = off; | |
60443c88 | 418 | table[i]->seq = i; |
1da177e4 | 419 | |
73bbb944 MO |
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"); | |
424 | exit(EXIT_FAILURE); | |
425 | } | |
426 | ||
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"); | |
431 | exit(EXIT_FAILURE); | |
432 | } | |
433 | ||
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; | |
439 | } else { | |
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; | |
445 | } | |
8d605269 MY |
446 | for (k = 0; k < table[i]->len; k++) |
447 | printf(", 0x%02x", table[i]->sym[k]); | |
1da177e4 | 448 | printf("\n"); |
1da177e4 LT |
449 | } |
450 | printf("\n"); | |
451 | ||
dd1553b8 MY |
452 | /* |
453 | * Now that we wrote out the compressed symbol names, restore the | |
454 | * original names, which are needed in some of the later steps. | |
455 | */ | |
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); | |
459 | } | |
460 | ||
1da177e4 | 461 | output_label("kallsyms_markers"); |
b3dbb4ec | 462 | for (i = 0; i < ((table_cnt + 255) >> 8); i++) |
80ffbaa5 | 463 | printf("\t.long\t%u\n", markers[i]); |
1da177e4 LT |
464 | printf("\n"); |
465 | ||
466 | free(markers); | |
467 | ||
468 | output_label("kallsyms_token_table"); | |
469 | off = 0; | |
470 | for (i = 0; i < 256; i++) { | |
471 | best_idx[i] = off; | |
b3dbb4ec | 472 | expand_symbol(best_table[i], best_table_len[i], buf); |
1da177e4 LT |
473 | printf("\t.asciz\t\"%s\"\n", buf); |
474 | off += strlen(buf) + 1; | |
475 | } | |
476 | printf("\n"); | |
477 | ||
478 | output_label("kallsyms_token_index"); | |
479 | for (i = 0; i < 256; i++) | |
480 | printf("\t.short\t%d\n", best_idx[i]); | |
481 | printf("\n"); | |
404bad70 MY |
482 | |
483 | if (!base_relative) | |
484 | output_label("kallsyms_addresses"); | |
485 | else | |
486 | output_label("kallsyms_offsets"); | |
487 | ||
488 | for (i = 0; i < table_cnt; i++) { | |
489 | if (base_relative) { | |
490 | /* | |
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 | |
494 | * up at runtime. | |
495 | */ | |
496 | ||
497 | long long offset; | |
498 | int overflow; | |
499 | ||
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); | |
506 | } else { | |
507 | offset = relative_base - table[i]->addr - 1; | |
508 | overflow = (offset < INT_MIN || offset >= 0); | |
509 | } | |
510 | if (overflow) { | |
511 | fprintf(stderr, "kallsyms failure: " | |
512 | "%s symbol value %#llx out of range in relative mode\n", | |
513 | symbol_absolute(table[i]) ? "absolute" : "relative", | |
514 | table[i]->addr); | |
515 | exit(EXIT_FAILURE); | |
516 | } | |
dd1553b8 | 517 | printf("\t.long\t%#x /* %s */\n", (int)offset, table[i]->sym); |
404bad70 MY |
518 | } else if (!symbol_absolute(table[i])) { |
519 | output_address(table[i]->addr); | |
520 | } else { | |
521 | printf("\tPTR\t%#llx\n", table[i]->addr); | |
522 | } | |
523 | } | |
524 | printf("\n"); | |
525 | ||
526 | if (base_relative) { | |
527 | output_label("kallsyms_relative_base"); | |
528 | output_address(relative_base); | |
529 | printf("\n"); | |
530 | } | |
531 | ||
dd1553b8 MY |
532 | if (lto_clang) |
533 | for (i = 0; i < table_cnt; i++) | |
534 | cleanup_symbol_name((char *)table[i]->sym); | |
535 | ||
404bad70 MY |
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)); | |
543 | printf("\n"); | |
1da177e4 LT |
544 | } |
545 | ||
546 | ||
547 | /* table lookup compression functions */ | |
548 | ||
1da177e4 | 549 | /* count all the possible tokens in a symbol */ |
4bfe2b78 | 550 | static void learn_symbol(const unsigned char *symbol, int len) |
1da177e4 LT |
551 | { |
552 | int i; | |
553 | ||
554 | for (i = 0; i < len - 1; i++) | |
b3dbb4ec | 555 | token_profit[ symbol[i] + (symbol[i + 1] << 8) ]++; |
1da177e4 LT |
556 | } |
557 | ||
558 | /* decrease the count for all the possible tokens in a symbol */ | |
4bfe2b78 | 559 | static void forget_symbol(const unsigned char *symbol, int len) |
1da177e4 LT |
560 | { |
561 | int i; | |
562 | ||
563 | for (i = 0; i < len - 1; i++) | |
b3dbb4ec | 564 | token_profit[ symbol[i] + (symbol[i + 1] << 8) ]--; |
1da177e4 LT |
565 | } |
566 | ||
5e5c4fa7 | 567 | /* do the initial token count */ |
fcdf7197 | 568 | static void build_initial_token_table(void) |
1da177e4 | 569 | { |
5e5c4fa7 | 570 | unsigned int i; |
1da177e4 | 571 | |
5e5c4fa7 | 572 | for (i = 0; i < table_cnt; i++) |
8d605269 | 573 | learn_symbol(table[i]->sym, table[i]->len); |
1da177e4 LT |
574 | } |
575 | ||
2558c138 | 576 | static unsigned char *find_token(unsigned char *str, int len, |
4bfe2b78 | 577 | const unsigned char *token) |
7c5d249a PM |
578 | { |
579 | int i; | |
580 | ||
581 | for (i = 0; i < len - 1; i++) { | |
582 | if (str[i] == token[0] && str[i+1] == token[1]) | |
583 | return &str[i]; | |
584 | } | |
585 | return NULL; | |
586 | } | |
587 | ||
1da177e4 LT |
588 | /* replace a given token in all the valid symbols. Use the sampled symbols |
589 | * to update the counts */ | |
4bfe2b78 | 590 | static void compress_symbols(const unsigned char *str, int idx) |
1da177e4 | 591 | { |
b3dbb4ec PM |
592 | unsigned int i, len, size; |
593 | unsigned char *p1, *p2; | |
1da177e4 | 594 | |
b3dbb4ec | 595 | for (i = 0; i < table_cnt; i++) { |
1da177e4 | 596 | |
8d605269 MY |
597 | len = table[i]->len; |
598 | p1 = table[i]->sym; | |
b3dbb4ec PM |
599 | |
600 | /* find the token on the symbol */ | |
7c5d249a | 601 | p2 = find_token(p1, len, str); |
b3dbb4ec PM |
602 | if (!p2) continue; |
603 | ||
604 | /* decrease the counts for this symbol's tokens */ | |
8d605269 | 605 | forget_symbol(table[i]->sym, len); |
b3dbb4ec PM |
606 | |
607 | size = len; | |
1da177e4 LT |
608 | |
609 | do { | |
b3dbb4ec PM |
610 | *p2 = idx; |
611 | p2++; | |
612 | size -= (p2 - p1); | |
613 | memmove(p2, p2 + 1, size); | |
614 | p1 = p2; | |
615 | len--; | |
616 | ||
617 | if (size < 2) break; | |
618 | ||
1da177e4 | 619 | /* find the token on the symbol */ |
7c5d249a | 620 | p2 = find_token(p1, size, str); |
1da177e4 | 621 | |
b3dbb4ec | 622 | } while (p2); |
1da177e4 | 623 | |
8d605269 | 624 | table[i]->len = len; |
1da177e4 | 625 | |
b3dbb4ec | 626 | /* increase the counts for this symbol's new tokens */ |
8d605269 | 627 | learn_symbol(table[i]->sym, len); |
1da177e4 LT |
628 | } |
629 | } | |
630 | ||
631 | /* search the token with the maximum profit */ | |
b3dbb4ec | 632 | static int find_best_token(void) |
1da177e4 | 633 | { |
b3dbb4ec | 634 | int i, best, bestprofit; |
1da177e4 LT |
635 | |
636 | bestprofit=-10000; | |
b3dbb4ec | 637 | best = 0; |
1da177e4 | 638 | |
b3dbb4ec PM |
639 | for (i = 0; i < 0x10000; i++) { |
640 | if (token_profit[i] > bestprofit) { | |
641 | best = i; | |
642 | bestprofit = token_profit[i]; | |
1da177e4 | 643 | } |
1da177e4 | 644 | } |
1da177e4 LT |
645 | return best; |
646 | } | |
647 | ||
648 | /* this is the core of the algorithm: calculate the "best" table */ | |
649 | static void optimize_result(void) | |
650 | { | |
b3dbb4ec | 651 | int i, best; |
1da177e4 LT |
652 | |
653 | /* using the '\0' symbol last allows compress_symbols to use standard | |
654 | * fast string functions */ | |
655 | for (i = 255; i >= 0; i--) { | |
656 | ||
657 | /* if this table slot is empty (it is not used by an actual | |
658 | * original char code */ | |
659 | if (!best_table_len[i]) { | |
660 | ||
cbf7a90e | 661 | /* find the token with the best profit value */ |
1da177e4 | 662 | best = find_best_token(); |
e0a04b11 XW |
663 | if (token_profit[best] == 0) |
664 | break; | |
1da177e4 LT |
665 | |
666 | /* place it in the "best" table */ | |
b3dbb4ec PM |
667 | best_table_len[i] = 2; |
668 | best_table[i][0] = best & 0xFF; | |
669 | best_table[i][1] = (best >> 8) & 0xFF; | |
1da177e4 LT |
670 | |
671 | /* replace this token in all the valid symbols */ | |
b3dbb4ec | 672 | compress_symbols(best_table[i], i); |
1da177e4 LT |
673 | } |
674 | } | |
675 | } | |
676 | ||
677 | /* start by placing the symbols that are actually used on the table */ | |
678 | static void insert_real_symbols_in_table(void) | |
679 | { | |
b3dbb4ec | 680 | unsigned int i, j, c; |
1da177e4 | 681 | |
b3dbb4ec | 682 | for (i = 0; i < table_cnt; i++) { |
8d605269 MY |
683 | for (j = 0; j < table[i]->len; j++) { |
684 | c = table[i]->sym[j]; | |
b3dbb4ec PM |
685 | best_table[c][0]=c; |
686 | best_table_len[c]=1; | |
1da177e4 LT |
687 | } |
688 | } | |
689 | } | |
690 | ||
691 | static void optimize_token_table(void) | |
692 | { | |
fcdf7197 | 693 | build_initial_token_table(); |
1da177e4 LT |
694 | |
695 | insert_real_symbols_in_table(); | |
696 | ||
697 | optimize_result(); | |
698 | } | |
699 | ||
b478b782 LJ |
700 | /* guess for "linker script provide" symbol */ |
701 | static int may_be_linker_script_provide_symbol(const struct sym_entry *se) | |
702 | { | |
29e55ad3 | 703 | const char *symbol = sym_name(se); |
b478b782 LJ |
704 | int len = se->len - 1; |
705 | ||
706 | if (len < 8) | |
707 | return 0; | |
708 | ||
709 | if (symbol[0] != '_' || symbol[1] != '_') | |
710 | return 0; | |
711 | ||
712 | /* __start_XXXXX */ | |
713 | if (!memcmp(symbol + 2, "start_", 6)) | |
714 | return 1; | |
715 | ||
716 | /* __stop_XXXXX */ | |
717 | if (!memcmp(symbol + 2, "stop_", 5)) | |
718 | return 1; | |
719 | ||
720 | /* __end_XXXXX */ | |
721 | if (!memcmp(symbol + 2, "end_", 4)) | |
722 | return 1; | |
723 | ||
724 | /* __XXXXX_start */ | |
725 | if (!memcmp(symbol + len - 6, "_start", 6)) | |
726 | return 1; | |
727 | ||
728 | /* __XXXXX_end */ | |
729 | if (!memcmp(symbol + len - 4, "_end", 4)) | |
730 | return 1; | |
731 | ||
732 | return 0; | |
733 | } | |
734 | ||
f2df3f65 PM |
735 | static int compare_symbols(const void *a, const void *b) |
736 | { | |
8d605269 MY |
737 | const struct sym_entry *sa = *(const struct sym_entry **)a; |
738 | const struct sym_entry *sb = *(const struct sym_entry **)b; | |
f2df3f65 PM |
739 | int wa, wb; |
740 | ||
f2df3f65 PM |
741 | /* sort by address first */ |
742 | if (sa->addr > sb->addr) | |
743 | return 1; | |
744 | if (sa->addr < sb->addr) | |
745 | return -1; | |
746 | ||
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'); | |
750 | if (wa != wb) | |
751 | return wa - wb; | |
752 | ||
b478b782 LJ |
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); | |
756 | if (wa != wb) | |
757 | return wa - wb; | |
758 | ||
759 | /* sort by the number of prefix underscores */ | |
aa915245 MY |
760 | wa = strspn(sym_name(sa), "_"); |
761 | wb = strspn(sym_name(sb), "_"); | |
b478b782 LJ |
762 | if (wa != wb) |
763 | return wa - wb; | |
764 | ||
f2df3f65 PM |
765 | /* sort by initial order, so that other symbols are left undisturbed */ |
766 | return sa->start_pos - sb->start_pos; | |
767 | } | |
768 | ||
769 | static void sort_symbols(void) | |
770 | { | |
8d605269 | 771 | qsort(table, table_cnt, sizeof(table[0]), compare_symbols); |
f2df3f65 | 772 | } |
1da177e4 | 773 | |
c6bda7c9 RR |
774 | static void make_percpus_absolute(void) |
775 | { | |
776 | unsigned int i; | |
777 | ||
778 | for (i = 0; i < table_cnt; i++) | |
8d605269 | 779 | if (symbol_in_range(table[i], &percpu_range, 1)) { |
8c996940 AB |
780 | /* |
781 | * Keep the 'A' override for percpu symbols to | |
782 | * ensure consistent behavior compared to older | |
783 | * versions of this tool. | |
784 | */ | |
8d605269 MY |
785 | table[i]->sym[0] = 'A'; |
786 | table[i]->percpu_absolute = 1; | |
8c996940 | 787 | } |
c6bda7c9 RR |
788 | } |
789 | ||
2213e9a6 AB |
790 | /* find the minimum non-absolute symbol address */ |
791 | static void record_relative_base(void) | |
792 | { | |
793 | unsigned int i; | |
794 | ||
2213e9a6 | 795 | for (i = 0; i < table_cnt; i++) |
8d605269 | 796 | if (!symbol_absolute(table[i])) { |
f34ea029 MY |
797 | /* |
798 | * The table is sorted by address. | |
799 | * Take the first non-absolute symbol value. | |
800 | */ | |
8d605269 | 801 | relative_base = table[i]->addr; |
f34ea029 MY |
802 | return; |
803 | } | |
2213e9a6 AB |
804 | } |
805 | ||
b3dbb4ec | 806 | int main(int argc, char **argv) |
1da177e4 | 807 | { |
aa221f2e MY |
808 | while (1) { |
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}, | |
010a0aad | 813 | {"lto-clang", no_argument, <o_clang, 1}, |
aa221f2e MY |
814 | {}, |
815 | }; | |
816 | ||
817 | int c = getopt_long(argc, argv, "", long_options, NULL); | |
818 | ||
819 | if (c == -1) | |
820 | break; | |
821 | if (c != 0) | |
822 | usage(); | |
823 | } | |
824 | ||
825 | if (optind >= argc) | |
1da177e4 LT |
826 | usage(); |
827 | ||
aa221f2e | 828 | read_map(argv[optind]); |
5e5c4fa7 | 829 | shrink_table(); |
c6bda7c9 RR |
830 | if (absolute_percpu) |
831 | make_percpus_absolute(); | |
f34ea029 | 832 | sort_symbols(); |
2213e9a6 AB |
833 | if (base_relative) |
834 | record_relative_base(); | |
2ea03891 | 835 | optimize_token_table(); |
1da177e4 LT |
836 | write_src(); |
837 | ||
838 | return 0; | |
839 | } |