xref: /openbmc/linux/scripts/kallsyms.c (revision e868d61272caa648214046a096e5a6bfc068dc8c)
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  *
8  * Usage: nm -n vmlinux | scripts/kallsyms [--all-symbols] > symbols.S
9  *
10  * ChangeLog:
11  *
12  * (25/Aug/2004) Paulo Marques <pmarques@grupopie.com>
13  *      Changed the compression method from stem compression to "table lookup"
14  *      compression
15  *
16  *      Table compression uses all the unused char codes on the symbols and
17  *  maps these to the most used substrings (tokens). For instance, it might
18  *  map char code 0xF7 to represent "write_" and then in every symbol where
19  *  "write_" appears it can be replaced by 0xF7, saving 5 bytes.
20  *      The used codes themselves are also placed in the table so that the
21  *  decompresion can work without "special cases".
22  *      Applied to kernel symbols, this usually produces a compression ratio
23  *  of about 50%.
24  *
25  */
26 
27 #define _GNU_SOURCE
28 
29 #include <stdio.h>
30 #include <stdlib.h>
31 #include <string.h>
32 #include <ctype.h>
33 
34 #define KSYM_NAME_LEN		127
35 
36 
37 struct sym_entry {
38 	unsigned long long addr;
39 	unsigned int len;
40 	unsigned char *sym;
41 };
42 
43 
44 static struct sym_entry *table;
45 static unsigned int table_size, table_cnt;
46 static unsigned long long _text, _stext, _etext, _sinittext, _einittext, _sextratext, _eextratext;
47 static int all_symbols = 0;
48 static char symbol_prefix_char = '\0';
49 
50 int token_profit[0x10000];
51 
52 /* the table that holds the result of the compression */
53 unsigned char best_table[256][2];
54 unsigned char best_table_len[256];
55 
56 
57 static void usage(void)
58 {
59 	fprintf(stderr, "Usage: kallsyms [--all-symbols] [--symbol-prefix=<prefix char>] < in.map > out.S\n");
60 	exit(1);
61 }
62 
63 /*
64  * This ignores the intensely annoying "mapping symbols" found
65  * in ARM ELF files: $a, $t and $d.
66  */
67 static inline int is_arm_mapping_symbol(const char *str)
68 {
69 	return str[0] == '$' && strchr("atd", str[1])
70 	       && (str[2] == '\0' || str[2] == '.');
71 }
72 
73 static int read_symbol(FILE *in, struct sym_entry *s)
74 {
75 	char str[500];
76 	char *sym, stype;
77 	int rc;
78 
79 	rc = fscanf(in, "%llx %c %499s\n", &s->addr, &stype, str);
80 	if (rc != 3) {
81 		if (rc != EOF) {
82 			/* skip line */
83 			fgets(str, 500, in);
84 		}
85 		return -1;
86 	}
87 
88 	sym = str;
89 	/* skip prefix char */
90 	if (symbol_prefix_char && str[0] == symbol_prefix_char)
91 		sym++;
92 
93 	/* Ignore most absolute/undefined (?) symbols. */
94 	if (strcmp(sym, "_text") == 0)
95 		_text = s->addr;
96 	else if (strcmp(sym, "_stext") == 0)
97 		_stext = s->addr;
98 	else if (strcmp(sym, "_etext") == 0)
99 		_etext = s->addr;
100 	else if (strcmp(sym, "_sinittext") == 0)
101 		_sinittext = s->addr;
102 	else if (strcmp(sym, "_einittext") == 0)
103 		_einittext = s->addr;
104 	else if (strcmp(sym, "_sextratext") == 0)
105 		_sextratext = s->addr;
106 	else if (strcmp(sym, "_eextratext") == 0)
107 		_eextratext = s->addr;
108 	else if (toupper(stype) == 'A')
109 	{
110 		/* Keep these useful absolute symbols */
111 		if (strcmp(sym, "__kernel_syscall_via_break") &&
112 		    strcmp(sym, "__kernel_syscall_via_epc") &&
113 		    strcmp(sym, "__kernel_sigtramp") &&
114 		    strcmp(sym, "__gp"))
115 			return -1;
116 
117 	}
118 	else if (toupper(stype) == 'U' ||
119 		 is_arm_mapping_symbol(sym))
120 		return -1;
121 	/* exclude also MIPS ELF local symbols ($L123 instead of .L123) */
122 	else if (str[0] == '$')
123 		return -1;
124 
125 	/* include the type field in the symbol name, so that it gets
126 	 * compressed together */
127 	s->len = strlen(str) + 1;
128 	s->sym = malloc(s->len + 1);
129 	if (!s->sym) {
130 		fprintf(stderr, "kallsyms failure: "
131 			"unable to allocate required amount of memory\n");
132 		exit(EXIT_FAILURE);
133 	}
134 	strcpy((char *)s->sym + 1, str);
135 	s->sym[0] = stype;
136 
137 	return 0;
138 }
139 
140 static int symbol_valid(struct sym_entry *s)
141 {
142 	/* Symbols which vary between passes.  Passes 1 and 2 must have
143 	 * identical symbol lists.  The kallsyms_* symbols below are only added
144 	 * after pass 1, they would be included in pass 2 when --all-symbols is
145 	 * specified so exclude them to get a stable symbol list.
146 	 */
147 	static char *special_symbols[] = {
148 		"kallsyms_addresses",
149 		"kallsyms_num_syms",
150 		"kallsyms_names",
151 		"kallsyms_markers",
152 		"kallsyms_token_table",
153 		"kallsyms_token_index",
154 
155 	/* Exclude linker generated symbols which vary between passes */
156 		"_SDA_BASE_",		/* ppc */
157 		"_SDA2_BASE_",		/* ppc */
158 		NULL };
159 	int i;
160 	int offset = 1;
161 
162 	/* skip prefix char */
163 	if (symbol_prefix_char && *(s->sym + 1) == symbol_prefix_char)
164 		offset++;
165 
166 	/* if --all-symbols is not specified, then symbols outside the text
167 	 * and inittext sections are discarded */
168 	if (!all_symbols) {
169 		if ((s->addr < _stext || s->addr > _etext)
170 		    && (s->addr < _sinittext || s->addr > _einittext)
171 		    && (s->addr < _sextratext || s->addr > _eextratext))
172 			return 0;
173 		/* Corner case.  Discard any symbols with the same value as
174 		 * _etext _einittext or _eextratext; they can move between pass
175 		 * 1 and 2 when the kallsyms data are added.  If these symbols
176 		 * move then they may get dropped in pass 2, which breaks the
177 		 * kallsyms rules.
178 		 */
179 		if ((s->addr == _etext && strcmp((char*)s->sym + offset, "_etext")) ||
180 		    (s->addr == _einittext && strcmp((char*)s->sym + offset, "_einittext")) ||
181 		    (s->addr == _eextratext && strcmp((char*)s->sym + offset, "_eextratext")))
182 			return 0;
183 	}
184 
185 	/* Exclude symbols which vary between passes. */
186 	if (strstr((char *)s->sym + offset, "_compiled."))
187 		return 0;
188 
189 	for (i = 0; special_symbols[i]; i++)
190 		if( strcmp((char *)s->sym + offset, special_symbols[i]) == 0 )
191 			return 0;
192 
193 	return 1;
194 }
195 
196 static void read_map(FILE *in)
197 {
198 	while (!feof(in)) {
199 		if (table_cnt >= table_size) {
200 			table_size += 10000;
201 			table = realloc(table, sizeof(*table) * table_size);
202 			if (!table) {
203 				fprintf(stderr, "out of memory\n");
204 				exit (1);
205 			}
206 		}
207 		if (read_symbol(in, &table[table_cnt]) == 0)
208 			table_cnt++;
209 	}
210 }
211 
212 static void output_label(char *label)
213 {
214 	if (symbol_prefix_char)
215 		printf(".globl %c%s\n", symbol_prefix_char, label);
216 	else
217 		printf(".globl %s\n", label);
218 	printf("\tALGN\n");
219 	if (symbol_prefix_char)
220 		printf("%c%s:\n", symbol_prefix_char, label);
221 	else
222 		printf("%s:\n", label);
223 }
224 
225 /* uncompress a compressed symbol. When this function is called, the best table
226  * might still be compressed itself, so the function needs to be recursive */
227 static int expand_symbol(unsigned char *data, int len, char *result)
228 {
229 	int c, rlen, total=0;
230 
231 	while (len) {
232 		c = *data;
233 		/* if the table holds a single char that is the same as the one
234 		 * we are looking for, then end the search */
235 		if (best_table[c][0]==c && best_table_len[c]==1) {
236 			*result++ = c;
237 			total++;
238 		} else {
239 			/* if not, recurse and expand */
240 			rlen = expand_symbol(best_table[c], best_table_len[c], result);
241 			total += rlen;
242 			result += rlen;
243 		}
244 		data++;
245 		len--;
246 	}
247 	*result=0;
248 
249 	return total;
250 }
251 
252 static void write_src(void)
253 {
254 	unsigned int i, k, off;
255 	unsigned int best_idx[256];
256 	unsigned int *markers;
257 	char buf[KSYM_NAME_LEN+1];
258 
259 	printf("#include <asm/types.h>\n");
260 	printf("#if BITS_PER_LONG == 64\n");
261 	printf("#define PTR .quad\n");
262 	printf("#define ALGN .align 8\n");
263 	printf("#else\n");
264 	printf("#define PTR .long\n");
265 	printf("#define ALGN .align 4\n");
266 	printf("#endif\n");
267 
268 	printf("\t.section .rodata, \"a\"\n");
269 
270 	/* Provide proper symbols relocatability by their '_text'
271 	 * relativeness.  The symbol names cannot be used to construct
272 	 * normal symbol references as the list of symbols contains
273 	 * symbols that are declared static and are private to their
274 	 * .o files.  This prevents .tmp_kallsyms.o or any other
275 	 * object from referencing them.
276 	 */
277 	output_label("kallsyms_addresses");
278 	for (i = 0; i < table_cnt; i++) {
279 		if (toupper(table[i].sym[0]) != 'A') {
280 			if (_text <= table[i].addr)
281 				printf("\tPTR\t_text + %#llx\n",
282 					table[i].addr - _text);
283 			else
284 				printf("\tPTR\t_text - %#llx\n",
285 					_text - table[i].addr);
286 		} else {
287 			printf("\tPTR\t%#llx\n", table[i].addr);
288 		}
289 	}
290 	printf("\n");
291 
292 	output_label("kallsyms_num_syms");
293 	printf("\tPTR\t%d\n", table_cnt);
294 	printf("\n");
295 
296 	/* table of offset markers, that give the offset in the compressed stream
297 	 * every 256 symbols */
298 	markers = malloc(sizeof(unsigned int) * ((table_cnt + 255) / 256));
299 	if (!markers) {
300 		fprintf(stderr, "kallsyms failure: "
301 			"unable to allocate required memory\n");
302 		exit(EXIT_FAILURE);
303 	}
304 
305 	output_label("kallsyms_names");
306 	off = 0;
307 	for (i = 0; i < table_cnt; i++) {
308 		if ((i & 0xFF) == 0)
309 			markers[i >> 8] = off;
310 
311 		printf("\t.byte 0x%02x", table[i].len);
312 		for (k = 0; k < table[i].len; k++)
313 			printf(", 0x%02x", table[i].sym[k]);
314 		printf("\n");
315 
316 		off += table[i].len + 1;
317 	}
318 	printf("\n");
319 
320 	output_label("kallsyms_markers");
321 	for (i = 0; i < ((table_cnt + 255) >> 8); i++)
322 		printf("\tPTR\t%d\n", markers[i]);
323 	printf("\n");
324 
325 	free(markers);
326 
327 	output_label("kallsyms_token_table");
328 	off = 0;
329 	for (i = 0; i < 256; i++) {
330 		best_idx[i] = off;
331 		expand_symbol(best_table[i], best_table_len[i], buf);
332 		printf("\t.asciz\t\"%s\"\n", buf);
333 		off += strlen(buf) + 1;
334 	}
335 	printf("\n");
336 
337 	output_label("kallsyms_token_index");
338 	for (i = 0; i < 256; i++)
339 		printf("\t.short\t%d\n", best_idx[i]);
340 	printf("\n");
341 }
342 
343 
344 /* table lookup compression functions */
345 
346 /* count all the possible tokens in a symbol */
347 static void learn_symbol(unsigned char *symbol, int len)
348 {
349 	int i;
350 
351 	for (i = 0; i < len - 1; i++)
352 		token_profit[ symbol[i] + (symbol[i + 1] << 8) ]++;
353 }
354 
355 /* decrease the count for all the possible tokens in a symbol */
356 static void forget_symbol(unsigned char *symbol, int len)
357 {
358 	int i;
359 
360 	for (i = 0; i < len - 1; i++)
361 		token_profit[ symbol[i] + (symbol[i + 1] << 8) ]--;
362 }
363 
364 /* remove all the invalid symbols from the table and do the initial token count */
365 static void build_initial_tok_table(void)
366 {
367 	unsigned int i, pos;
368 
369 	pos = 0;
370 	for (i = 0; i < table_cnt; i++) {
371 		if ( symbol_valid(&table[i]) ) {
372 			if (pos != i)
373 				table[pos] = table[i];
374 			learn_symbol(table[pos].sym, table[pos].len);
375 			pos++;
376 		}
377 	}
378 	table_cnt = pos;
379 }
380 
381 /* replace a given token in all the valid symbols. Use the sampled symbols
382  * to update the counts */
383 static void compress_symbols(unsigned char *str, int idx)
384 {
385 	unsigned int i, len, size;
386 	unsigned char *p1, *p2;
387 
388 	for (i = 0; i < table_cnt; i++) {
389 
390 		len = table[i].len;
391 		p1 = table[i].sym;
392 
393 		/* find the token on the symbol */
394 		p2 = memmem(p1, len, str, 2);
395 		if (!p2) continue;
396 
397 		/* decrease the counts for this symbol's tokens */
398 		forget_symbol(table[i].sym, len);
399 
400 		size = len;
401 
402 		do {
403 			*p2 = idx;
404 			p2++;
405 			size -= (p2 - p1);
406 			memmove(p2, p2 + 1, size);
407 			p1 = p2;
408 			len--;
409 
410 			if (size < 2) break;
411 
412 			/* find the token on the symbol */
413 			p2 = memmem(p1, size, str, 2);
414 
415 		} while (p2);
416 
417 		table[i].len = len;
418 
419 		/* increase the counts for this symbol's new tokens */
420 		learn_symbol(table[i].sym, len);
421 	}
422 }
423 
424 /* search the token with the maximum profit */
425 static int find_best_token(void)
426 {
427 	int i, best, bestprofit;
428 
429 	bestprofit=-10000;
430 	best = 0;
431 
432 	for (i = 0; i < 0x10000; i++) {
433 		if (token_profit[i] > bestprofit) {
434 			best = i;
435 			bestprofit = token_profit[i];
436 		}
437 	}
438 	return best;
439 }
440 
441 /* this is the core of the algorithm: calculate the "best" table */
442 static void optimize_result(void)
443 {
444 	int i, best;
445 
446 	/* using the '\0' symbol last allows compress_symbols to use standard
447 	 * fast string functions */
448 	for (i = 255; i >= 0; i--) {
449 
450 		/* if this table slot is empty (it is not used by an actual
451 		 * original char code */
452 		if (!best_table_len[i]) {
453 
454 			/* find the token with the breates profit value */
455 			best = find_best_token();
456 
457 			/* place it in the "best" table */
458 			best_table_len[i] = 2;
459 			best_table[i][0] = best & 0xFF;
460 			best_table[i][1] = (best >> 8) & 0xFF;
461 
462 			/* replace this token in all the valid symbols */
463 			compress_symbols(best_table[i], i);
464 		}
465 	}
466 }
467 
468 /* start by placing the symbols that are actually used on the table */
469 static void insert_real_symbols_in_table(void)
470 {
471 	unsigned int i, j, c;
472 
473 	memset(best_table, 0, sizeof(best_table));
474 	memset(best_table_len, 0, sizeof(best_table_len));
475 
476 	for (i = 0; i < table_cnt; i++) {
477 		for (j = 0; j < table[i].len; j++) {
478 			c = table[i].sym[j];
479 			best_table[c][0]=c;
480 			best_table_len[c]=1;
481 		}
482 	}
483 }
484 
485 static void optimize_token_table(void)
486 {
487 	build_initial_tok_table();
488 
489 	insert_real_symbols_in_table();
490 
491 	/* When valid symbol is not registered, exit to error */
492 	if (!table_cnt) {
493 		fprintf(stderr, "No valid symbol.\n");
494 		exit(1);
495 	}
496 
497 	optimize_result();
498 }
499 
500 
501 int main(int argc, char **argv)
502 {
503 	if (argc >= 2) {
504 		int i;
505 		for (i = 1; i < argc; i++) {
506 			if(strcmp(argv[i], "--all-symbols") == 0)
507 				all_symbols = 1;
508 			else if (strncmp(argv[i], "--symbol-prefix=", 16) == 0) {
509 				char *p = &argv[i][16];
510 				/* skip quote */
511 				if ((*p == '"' && *(p+2) == '"') || (*p == '\'' && *(p+2) == '\''))
512 					p++;
513 				symbol_prefix_char = *p;
514 			} else
515 				usage();
516 		}
517 	} else if (argc != 1)
518 		usage();
519 
520 	read_map(stdin);
521 	optimize_token_table();
522 	write_src();
523 
524 	return 0;
525 }
526