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 *      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>
25
26#ifndef ARRAY_SIZE
27#define ARRAY_SIZE(arr) (sizeof(arr) / sizeof(arr[0]))
28#endif
29
30#define KSYM_NAME_LEN		128
31
32struct sym_entry {
33	unsigned long long addr;
34	unsigned int len;
35	unsigned int start_pos;
36	unsigned char *sym;
37};
38
39struct text_range {
40	const char *stext, *etext;
41	unsigned long long start, end;
42};
43
44static unsigned long long _text;
45static struct text_range text_ranges[] = {
46	{ "_stext",     "_etext"     },
47	{ "_sinittext", "_einittext" },
48	{ "_stext_l1",  "_etext_l1"  },	/* Blackfin on-chip L1 inst SRAM */
49	{ "_stext_l2",  "_etext_l2"  },	/* Blackfin on-chip L2 SRAM */
50};
51#define text_range_text     (&text_ranges[0])
52#define text_range_inittext (&text_ranges[1])
53
54static struct sym_entry *table;
55static unsigned int table_size, table_cnt;
56static int all_symbols = 0;
57static char symbol_prefix_char = '\0';
58
59int token_profit[0x10000];
60
61/* the table that holds the result of the compression */
62unsigned char best_table[256][2];
63unsigned char best_table_len[256];
64
65
66static void usage(void)
67{
68	fprintf(stderr, "Usage: kallsyms [--all-symbols] [--symbol-prefix=<prefix char>] < in.map > out.S\n");
69	exit(1);
70}
71
72/*
73 * This ignores the intensely annoying "mapping symbols" found
74 * in ARM ELF files: $a, $t and $d.
75 */
76static inline int is_arm_mapping_symbol(const char *str)
77{
78	return str[0] == '$' && strchr("atd", str[1])
79	       && (str[2] == '\0' || str[2] == '.');
80}
81
82static int read_symbol_tr(const char *sym, unsigned long long addr)
83{
84	size_t i;
85	struct text_range *tr;
86
87	for (i = 0; i < ARRAY_SIZE(text_ranges); ++i) {
88		tr = &text_ranges[i];
89
90		if (strcmp(sym, tr->stext) == 0) {
91			tr->start = addr;
92			return 0;
93		} else if (strcmp(sym, tr->etext) == 0) {
94			tr->end = addr;
95			return 0;
96		}
97	}
98
99	return 1;
100}
101
102static int read_symbol(FILE *in, struct sym_entry *s)
103{
104	char str[500];
105	char *sym, stype;
106	int rc;
107
108	rc = fscanf(in, "%llx %c %499s\n", &s->addr, &stype, str);
109	if (rc != 3) {
110		if (rc != EOF && fgets(str, 500, in) == NULL)
111			fprintf(stderr, "Read error or end of file.\n");
112		return -1;
113	}
114
115	sym = str;
116	/* skip prefix char */
117	if (symbol_prefix_char && str[0] == symbol_prefix_char)
118		sym++;
119
120	/* Ignore most absolute/undefined (?) symbols. */
121	if (strcmp(sym, "_text") == 0)
122		_text = s->addr;
123	else if (read_symbol_tr(sym, s->addr) == 0)
124		/* nothing to do */;
125	else if (toupper(stype) == 'A')
126	{
127		/* Keep these useful absolute symbols */
128		if (strcmp(sym, "__kernel_syscall_via_break") &&
129		    strcmp(sym, "__kernel_syscall_via_epc") &&
130		    strcmp(sym, "__kernel_sigtramp") &&
131		    strcmp(sym, "__gp"))
132			return -1;
133
134	}
135	else if (toupper(stype) == 'U' ||
136		 is_arm_mapping_symbol(sym))
137		return -1;
138	/* exclude also MIPS ELF local symbols ($L123 instead of .L123) */
139	else if (str[0] == '$')
140		return -1;
141	/* exclude debugging symbols */
142	else if (stype == 'N')
143		return -1;
144
145	/* include the type field in the symbol name, so that it gets
146	 * compressed together */
147	s->len = strlen(str) + 1;
148	s->sym = malloc(s->len + 1);
149	if (!s->sym) {
150		fprintf(stderr, "kallsyms failure: "
151			"unable to allocate required amount of memory\n");
152		exit(EXIT_FAILURE);
153	}
154	strcpy((char *)s->sym + 1, str);
155	s->sym[0] = stype;
156
157	return 0;
158}
159
160static int symbol_valid_tr(struct sym_entry *s)
161{
162	size_t i;
163	struct text_range *tr;
164
165	for (i = 0; i < ARRAY_SIZE(text_ranges); ++i) {
166		tr = &text_ranges[i];
167
168		if (s->addr >= tr->start && s->addr <= tr->end)
169			return 1;
170	}
171
172	return 0;
173}
174
175static int symbol_valid(struct sym_entry *s)
176{
177	/* Symbols which vary between passes.  Passes 1 and 2 must have
178	 * identical symbol lists.  The kallsyms_* symbols below are only added
179	 * after pass 1, they would be included in pass 2 when --all-symbols is
180	 * specified so exclude them to get a stable symbol list.
181	 */
182	static char *special_symbols[] = {
183		"kallsyms_addresses",
184		"kallsyms_num_syms",
185		"kallsyms_names",
186		"kallsyms_markers",
187		"kallsyms_token_table",
188		"kallsyms_token_index",
189
190	/* Exclude linker generated symbols which vary between passes */
191		"_SDA_BASE_",		/* ppc */
192		"_SDA2_BASE_",		/* ppc */
193		NULL };
194	int i;
195	int offset = 1;
196
197	/* skip prefix char */
198	if (symbol_prefix_char && *(s->sym + 1) == symbol_prefix_char)
199		offset++;
200
201	/* if --all-symbols is not specified, then symbols outside the text
202	 * and inittext sections are discarded */
203	if (!all_symbols) {
204		if (symbol_valid_tr(s) == 0)
205			return 0;
206		/* Corner case.  Discard any symbols with the same value as
207		 * _etext _einittext; they can move between pass 1 and 2 when
208		 * the kallsyms data are added.  If these symbols move then
209		 * they may get dropped in pass 2, which breaks the kallsyms
210		 * rules.
211		 */
212		if ((s->addr == text_range_text->end &&
213				strcmp((char *)s->sym + offset, text_range_text->etext)) ||
214		    (s->addr == text_range_inittext->end &&
215				strcmp((char *)s->sym + offset, text_range_inittext->etext)))
216			return 0;
217	}
218
219	/* Exclude symbols which vary between passes. */
220	if (strstr((char *)s->sym + offset, "_compiled."))
221		return 0;
222
223	for (i = 0; special_symbols[i]; i++)
224		if( strcmp((char *)s->sym + offset, special_symbols[i]) == 0 )
225			return 0;
226
227	return 1;
228}
229
230static void read_map(FILE *in)
231{
232	while (!feof(in)) {
233		if (table_cnt >= table_size) {
234			table_size += 10000;
235			table = realloc(table, sizeof(*table) * table_size);
236			if (!table) {
237				fprintf(stderr, "out of memory\n");
238				exit (1);
239			}
240		}
241		if (read_symbol(in, &table[table_cnt]) == 0) {
242			table[table_cnt].start_pos = table_cnt;
243			table_cnt++;
244		}
245	}
246}
247
248static void output_label(char *label)
249{
250	if (symbol_prefix_char)
251		printf(".globl %c%s\n", symbol_prefix_char, label);
252	else
253		printf(".globl %s\n", label);
254	printf("\tALGN\n");
255	if (symbol_prefix_char)
256		printf("%c%s:\n", symbol_prefix_char, label);
257	else
258		printf("%s:\n", label);
259}
260
261/* uncompress a compressed symbol. When this function is called, the best table
262 * might still be compressed itself, so the function needs to be recursive */
263static int expand_symbol(unsigned char *data, int len, char *result)
264{
265	int c, rlen, total=0;
266
267	while (len) {
268		c = *data;
269		/* if the table holds a single char that is the same as the one
270		 * we are looking for, then end the search */
271		if (best_table[c][0]==c && best_table_len[c]==1) {
272			*result++ = c;
273			total++;
274		} else {
275			/* if not, recurse and expand */
276			rlen = expand_symbol(best_table[c], best_table_len[c], result);
277			total += rlen;
278			result += rlen;
279		}
280		data++;
281		len--;
282	}
283	*result=0;
284
285	return total;
286}
287
288static void write_src(void)
289{
290	unsigned int i, k, off;
291	unsigned int best_idx[256];
292	unsigned int *markers;
293	char buf[KSYM_NAME_LEN];
294
295	printf("#include <asm/types.h>\n");
296	printf("#if BITS_PER_LONG == 64\n");
297	printf("#define PTR .quad\n");
298	printf("#define ALGN .align 8\n");
299	printf("#else\n");
300	printf("#define PTR .long\n");
301	printf("#define ALGN .align 4\n");
302	printf("#endif\n");
303
304	printf("\t.section .rodata, \"a\"\n");
305
306	/* Provide proper symbols relocatability by their '_text'
307	 * relativeness.  The symbol names cannot be used to construct
308	 * normal symbol references as the list of symbols contains
309	 * symbols that are declared static and are private to their
310	 * .o files.  This prevents .tmp_kallsyms.o or any other
311	 * object from referencing them.
312	 */
313	output_label("kallsyms_addresses");
314	for (i = 0; i < table_cnt; i++) {
315		if (toupper(table[i].sym[0]) != 'A') {
316			if (_text <= table[i].addr)
317				printf("\tPTR\t_text + %#llx\n",
318					table[i].addr - _text);
319			else
320				printf("\tPTR\t_text - %#llx\n",
321					_text - table[i].addr);
322		} else {
323			printf("\tPTR\t%#llx\n", table[i].addr);
324		}
325	}
326	printf("\n");
327
328	output_label("kallsyms_num_syms");
329	printf("\tPTR\t%d\n", table_cnt);
330	printf("\n");
331
332	/* table of offset markers, that give the offset in the compressed stream
333	 * every 256 symbols */
334	markers = malloc(sizeof(unsigned int) * ((table_cnt + 255) / 256));
335	if (!markers) {
336		fprintf(stderr, "kallsyms failure: "
337			"unable to allocate required memory\n");
338		exit(EXIT_FAILURE);
339	}
340
341	output_label("kallsyms_names");
342	off = 0;
343	for (i = 0; i < table_cnt; i++) {
344		if ((i & 0xFF) == 0)
345			markers[i >> 8] = off;
346
347		printf("\t.byte 0x%02x", table[i].len);
348		for (k = 0; k < table[i].len; k++)
349			printf(", 0x%02x", table[i].sym[k]);
350		printf("\n");
351
352		off += table[i].len + 1;
353	}
354	printf("\n");
355
356	output_label("kallsyms_markers");
357	for (i = 0; i < ((table_cnt + 255) >> 8); i++)
358		printf("\tPTR\t%d\n", markers[i]);
359	printf("\n");
360
361	free(markers);
362
363	output_label("kallsyms_token_table");
364	off = 0;
365	for (i = 0; i < 256; i++) {
366		best_idx[i] = off;
367		expand_symbol(best_table[i], best_table_len[i], buf);
368		printf("\t.asciz\t\"%s\"\n", buf);
369		off += strlen(buf) + 1;
370	}
371	printf("\n");
372
373	output_label("kallsyms_token_index");
374	for (i = 0; i < 256; i++)
375		printf("\t.short\t%d\n", best_idx[i]);
376	printf("\n");
377}
378
379
380/* table lookup compression functions */
381
382/* count all the possible tokens in a symbol */
383static void learn_symbol(unsigned char *symbol, int len)
384{
385	int i;
386
387	for (i = 0; i < len - 1; i++)
388		token_profit[ symbol[i] + (symbol[i + 1] << 8) ]++;
389}
390
391/* decrease the count for all the possible tokens in a symbol */
392static void forget_symbol(unsigned char *symbol, int len)
393{
394	int i;
395
396	for (i = 0; i < len - 1; i++)
397		token_profit[ symbol[i] + (symbol[i + 1] << 8) ]--;
398}
399
400/* remove all the invalid symbols from the table and do the initial token count */
401static void build_initial_tok_table(void)
402{
403	unsigned int i, pos;
404
405	pos = 0;
406	for (i = 0; i < table_cnt; i++) {
407		if ( symbol_valid(&table[i]) ) {
408			if (pos != i)
409				table[pos] = table[i];
410			learn_symbol(table[pos].sym, table[pos].len);
411			pos++;
412		}
413	}
414	table_cnt = pos;
415}
416
417static void *find_token(unsigned char *str, int len, unsigned char *token)
418{
419	int i;
420
421	for (i = 0; i < len - 1; i++) {
422		if (str[i] == token[0] && str[i+1] == token[1])
423			return &str[i];
424	}
425	return NULL;
426}
427
428/* replace a given token in all the valid symbols. Use the sampled symbols
429 * to update the counts */
430static void compress_symbols(unsigned char *str, int idx)
431{
432	unsigned int i, len, size;
433	unsigned char *p1, *p2;
434
435	for (i = 0; i < table_cnt; i++) {
436
437		len = table[i].len;
438		p1 = table[i].sym;
439
440		/* find the token on the symbol */
441		p2 = find_token(p1, len, str);
442		if (!p2) continue;
443
444		/* decrease the counts for this symbol's tokens */
445		forget_symbol(table[i].sym, len);
446
447		size = len;
448
449		do {
450			*p2 = idx;
451			p2++;
452			size -= (p2 - p1);
453			memmove(p2, p2 + 1, size);
454			p1 = p2;
455			len--;
456
457			if (size < 2) break;
458
459			/* find the token on the symbol */
460			p2 = find_token(p1, size, str);
461
462		} while (p2);
463
464		table[i].len = len;
465
466		/* increase the counts for this symbol's new tokens */
467		learn_symbol(table[i].sym, len);
468	}
469}
470
471/* search the token with the maximum profit */
472static int find_best_token(void)
473{
474	int i, best, bestprofit;
475
476	bestprofit=-10000;
477	best = 0;
478
479	for (i = 0; i < 0x10000; i++) {
480		if (token_profit[i] > bestprofit) {
481			best = i;
482			bestprofit = token_profit[i];
483		}
484	}
485	return best;
486}
487
488/* this is the core of the algorithm: calculate the "best" table */
489static void optimize_result(void)
490{
491	int i, best;
492
493	/* using the '\0' symbol last allows compress_symbols to use standard
494	 * fast string functions */
495	for (i = 255; i >= 0; i--) {
496
497		/* if this table slot is empty (it is not used by an actual
498		 * original char code */
499		if (!best_table_len[i]) {
500
501			/* find the token with the breates profit value */
502			best = find_best_token();
503			if (token_profit[best] == 0)
504				break;
505
506			/* place it in the "best" table */
507			best_table_len[i] = 2;
508			best_table[i][0] = best & 0xFF;
509			best_table[i][1] = (best >> 8) & 0xFF;
510
511			/* replace this token in all the valid symbols */
512			compress_symbols(best_table[i], i);
513		}
514	}
515}
516
517/* start by placing the symbols that are actually used on the table */
518static void insert_real_symbols_in_table(void)
519{
520	unsigned int i, j, c;
521
522	memset(best_table, 0, sizeof(best_table));
523	memset(best_table_len, 0, sizeof(best_table_len));
524
525	for (i = 0; i < table_cnt; i++) {
526		for (j = 0; j < table[i].len; j++) {
527			c = table[i].sym[j];
528			best_table[c][0]=c;
529			best_table_len[c]=1;
530		}
531	}
532}
533
534static void optimize_token_table(void)
535{
536	build_initial_tok_table();
537
538	insert_real_symbols_in_table();
539
540	/* When valid symbol is not registered, exit to error */
541	if (!table_cnt) {
542		fprintf(stderr, "No valid symbol.\n");
543		exit(1);
544	}
545
546	optimize_result();
547}
548
549/* guess for "linker script provide" symbol */
550static int may_be_linker_script_provide_symbol(const struct sym_entry *se)
551{
552	const char *symbol = (char *)se->sym + 1;
553	int len = se->len - 1;
554
555	if (len < 8)
556		return 0;
557
558	if (symbol[0] != '_' || symbol[1] != '_')
559		return 0;
560
561	/* __start_XXXXX */
562	if (!memcmp(symbol + 2, "start_", 6))
563		return 1;
564
565	/* __stop_XXXXX */
566	if (!memcmp(symbol + 2, "stop_", 5))
567		return 1;
568
569	/* __end_XXXXX */
570	if (!memcmp(symbol + 2, "end_", 4))
571		return 1;
572
573	/* __XXXXX_start */
574	if (!memcmp(symbol + len - 6, "_start", 6))
575		return 1;
576
577	/* __XXXXX_end */
578	if (!memcmp(symbol + len - 4, "_end", 4))
579		return 1;
580
581	return 0;
582}
583
584static int prefix_underscores_count(const char *str)
585{
586	const char *tail = str;
587
588	while (*tail == '_')
589		tail++;
590
591	return tail - str;
592}
593
594static int compare_symbols(const void *a, const void *b)
595{
596	const struct sym_entry *sa;
597	const struct sym_entry *sb;
598	int wa, wb;
599
600	sa = a;
601	sb = b;
602
603	/* sort by address first */
604	if (sa->addr > sb->addr)
605		return 1;
606	if (sa->addr < sb->addr)
607		return -1;
608
609	/* sort by "weakness" type */
610	wa = (sa->sym[0] == 'w') || (sa->sym[0] == 'W');
611	wb = (sb->sym[0] == 'w') || (sb->sym[0] == 'W');
612	if (wa != wb)
613		return wa - wb;
614
615	/* sort by "linker script provide" type */
616	wa = may_be_linker_script_provide_symbol(sa);
617	wb = may_be_linker_script_provide_symbol(sb);
618	if (wa != wb)
619		return wa - wb;
620
621	/* sort by the number of prefix underscores */
622	wa = prefix_underscores_count((const char *)sa->sym + 1);
623	wb = prefix_underscores_count((const char *)sb->sym + 1);
624	if (wa != wb)
625		return wa - wb;
626
627	/* sort by initial order, so that other symbols are left undisturbed */
628	return sa->start_pos - sb->start_pos;
629}
630
631static void sort_symbols(void)
632{
633	qsort(table, table_cnt, sizeof(struct sym_entry), compare_symbols);
634}
635
636int main(int argc, char **argv)
637{
638	if (argc >= 2) {
639		int i;
640		for (i = 1; i < argc; i++) {
641			if(strcmp(argv[i], "--all-symbols") == 0)
642				all_symbols = 1;
643			else if (strncmp(argv[i], "--symbol-prefix=", 16) == 0) {
644				char *p = &argv[i][16];
645				/* skip quote */
646				if ((*p == '"' && *(p+2) == '"') || (*p == '\'' && *(p+2) == '\''))
647					p++;
648				symbol_prefix_char = *p;
649			} else
650				usage();
651		}
652	} else if (argc != 1)
653		usage();
654
655	read_map(stdin);
656	sort_symbols();
657	optimize_token_table();
658	write_src();
659
660	return 0;
661}
662