xref: /linux/scripts/kallsyms.c (revision 01b944fe1cd4e21a2a9ed51adbdbafe2d5e905ba)
11da177e4SLinus Torvalds /* Generate assembler source containing symbol information
21da177e4SLinus Torvalds  *
31da177e4SLinus Torvalds  * Copyright 2002       by Kai Germaschewski
41da177e4SLinus Torvalds  *
51da177e4SLinus Torvalds  * This software may be used and distributed according to the terms
61da177e4SLinus Torvalds  * of the GNU General Public License, incorporated herein by reference.
71da177e4SLinus Torvalds  *
81da177e4SLinus Torvalds  * Usage: nm -n vmlinux | scripts/kallsyms [--all-symbols] > symbols.S
91da177e4SLinus Torvalds  *
101da177e4SLinus Torvalds  *      Table compression uses all the unused char codes on the symbols and
111da177e4SLinus Torvalds  *  maps these to the most used substrings (tokens). For instance, it might
121da177e4SLinus Torvalds  *  map char code 0xF7 to represent "write_" and then in every symbol where
131da177e4SLinus Torvalds  *  "write_" appears it can be replaced by 0xF7, saving 5 bytes.
141da177e4SLinus Torvalds  *      The used codes themselves are also placed in the table so that the
151da177e4SLinus Torvalds  *  decompresion can work without "special cases".
161da177e4SLinus Torvalds  *      Applied to kernel symbols, this usually produces a compression ratio
171da177e4SLinus Torvalds  *  of about 50%.
181da177e4SLinus Torvalds  *
191da177e4SLinus Torvalds  */
201da177e4SLinus Torvalds 
211da177e4SLinus Torvalds #include <stdio.h>
221da177e4SLinus Torvalds #include <stdlib.h>
231da177e4SLinus Torvalds #include <string.h>
241da177e4SLinus Torvalds #include <ctype.h>
251da177e4SLinus Torvalds 
2617b1f0deSMike Frysinger #ifndef ARRAY_SIZE
2717b1f0deSMike Frysinger #define ARRAY_SIZE(arr) (sizeof(arr) / sizeof(arr[0]))
2817b1f0deSMike Frysinger #endif
2917b1f0deSMike Frysinger 
309281aceaSTejun Heo #define KSYM_NAME_LEN		128
311da177e4SLinus Torvalds 
321da177e4SLinus Torvalds struct sym_entry {
331da177e4SLinus Torvalds 	unsigned long long addr;
34b3dbb4ecSPaulo Marques 	unsigned int len;
35f2df3f65SPaulo Marques 	unsigned int start_pos;
361da177e4SLinus Torvalds 	unsigned char *sym;
371da177e4SLinus Torvalds };
381da177e4SLinus Torvalds 
3978eb7159SKees Cook struct addr_range {
4078eb7159SKees Cook 	const char *start_sym, *end_sym;
4117b1f0deSMike Frysinger 	unsigned long long start, end;
4217b1f0deSMike Frysinger };
4317b1f0deSMike Frysinger 
4417b1f0deSMike Frysinger static unsigned long long _text;
4578eb7159SKees Cook static struct addr_range text_ranges[] = {
4617b1f0deSMike Frysinger 	{ "_stext",     "_etext"     },
4717b1f0deSMike Frysinger 	{ "_sinittext", "_einittext" },
4817b1f0deSMike Frysinger 	{ "_stext_l1",  "_etext_l1"  },	/* Blackfin on-chip L1 inst SRAM */
4917b1f0deSMike Frysinger 	{ "_stext_l2",  "_etext_l2"  },	/* Blackfin on-chip L2 SRAM */
5017b1f0deSMike Frysinger };
5117b1f0deSMike Frysinger #define text_range_text     (&text_ranges[0])
5217b1f0deSMike Frysinger #define text_range_inittext (&text_ranges[1])
5317b1f0deSMike Frysinger 
54c6bda7c9SRusty Russell static struct addr_range percpu_range = {
55c6bda7c9SRusty Russell 	"__per_cpu_start", "__per_cpu_end", -1ULL, 0
56c6bda7c9SRusty Russell };
57c6bda7c9SRusty Russell 
581da177e4SLinus Torvalds static struct sym_entry *table;
59b3dbb4ecSPaulo Marques static unsigned int table_size, table_cnt;
601da177e4SLinus Torvalds static int all_symbols = 0;
61c6bda7c9SRusty Russell static int absolute_percpu = 0;
6241f11a4fSYoshinori Sato static char symbol_prefix_char = '\0';
63f6537f2fSMing Lei static unsigned long long kernel_start_addr = 0;
641da177e4SLinus Torvalds 
65b3dbb4ecSPaulo Marques int token_profit[0x10000];
661da177e4SLinus Torvalds 
671da177e4SLinus Torvalds /* the table that holds the result of the compression */
68b3dbb4ecSPaulo Marques unsigned char best_table[256][2];
691da177e4SLinus Torvalds unsigned char best_table_len[256];
701da177e4SLinus Torvalds 
711da177e4SLinus Torvalds 
72b3dbb4ecSPaulo Marques static void usage(void)
731da177e4SLinus Torvalds {
74f6537f2fSMing Lei 	fprintf(stderr, "Usage: kallsyms [--all-symbols] "
75f6537f2fSMing Lei 			"[--symbol-prefix=<prefix char>] "
76f6537f2fSMing Lei 			"[--page-offset=<CONFIG_PAGE_OFFSET>] "
77f6537f2fSMing Lei 			"< in.map > out.S\n");
781da177e4SLinus Torvalds 	exit(1);
791da177e4SLinus Torvalds }
801da177e4SLinus Torvalds 
811da177e4SLinus Torvalds /*
821da177e4SLinus Torvalds  * This ignores the intensely annoying "mapping symbols" found
831da177e4SLinus Torvalds  * in ARM ELF files: $a, $t and $d.
841da177e4SLinus Torvalds  */
85b3dbb4ecSPaulo Marques static inline int is_arm_mapping_symbol(const char *str)
861da177e4SLinus Torvalds {
876c34f1f5SKyle McMartin 	return str[0] == '$' && strchr("axtd", str[1])
881da177e4SLinus Torvalds 	       && (str[2] == '\0' || str[2] == '.');
891da177e4SLinus Torvalds }
901da177e4SLinus Torvalds 
9178eb7159SKees Cook static int check_symbol_range(const char *sym, unsigned long long addr,
9278eb7159SKees Cook 			      struct addr_range *ranges, int entries)
9317b1f0deSMike Frysinger {
9417b1f0deSMike Frysinger 	size_t i;
9578eb7159SKees Cook 	struct addr_range *ar;
9617b1f0deSMike Frysinger 
9778eb7159SKees Cook 	for (i = 0; i < entries; ++i) {
9878eb7159SKees Cook 		ar = &ranges[i];
9917b1f0deSMike Frysinger 
10078eb7159SKees Cook 		if (strcmp(sym, ar->start_sym) == 0) {
10178eb7159SKees Cook 			ar->start = addr;
10217b1f0deSMike Frysinger 			return 0;
10378eb7159SKees Cook 		} else if (strcmp(sym, ar->end_sym) == 0) {
10478eb7159SKees Cook 			ar->end = addr;
10517b1f0deSMike Frysinger 			return 0;
10617b1f0deSMike Frysinger 		}
10717b1f0deSMike Frysinger 	}
10817b1f0deSMike Frysinger 
10917b1f0deSMike Frysinger 	return 1;
11017b1f0deSMike Frysinger }
11117b1f0deSMike Frysinger 
112b3dbb4ecSPaulo Marques static int read_symbol(FILE *in, struct sym_entry *s)
1131da177e4SLinus Torvalds {
1141da177e4SLinus Torvalds 	char str[500];
115b3dbb4ecSPaulo Marques 	char *sym, stype;
1161da177e4SLinus Torvalds 	int rc;
1171da177e4SLinus Torvalds 
118b3dbb4ecSPaulo Marques 	rc = fscanf(in, "%llx %c %499s\n", &s->addr, &stype, str);
1191da177e4SLinus Torvalds 	if (rc != 3) {
120ef894870SJean Sacren 		if (rc != EOF && fgets(str, 500, in) == NULL)
121ef894870SJean Sacren 			fprintf(stderr, "Read error or end of file.\n");
1221da177e4SLinus Torvalds 		return -1;
1231da177e4SLinus Torvalds 	}
124f3462aa9SAndi Kleen 	if (strlen(str) > KSYM_NAME_LEN) {
1256f62259bSFabio Estevam 		fprintf(stderr, "Symbol %s too long for kallsyms (%zu vs %d).\n"
126f3462aa9SAndi Kleen 				"Please increase KSYM_NAME_LEN both in kernel and kallsyms.c\n",
127f3462aa9SAndi Kleen 			str, strlen(str), KSYM_NAME_LEN);
128f3462aa9SAndi Kleen 		return -1;
129f3462aa9SAndi Kleen 	}
1301da177e4SLinus Torvalds 
13141f11a4fSYoshinori Sato 	sym = str;
13241f11a4fSYoshinori Sato 	/* skip prefix char */
13341f11a4fSYoshinori Sato 	if (symbol_prefix_char && str[0] == symbol_prefix_char)
13441f11a4fSYoshinori Sato 		sym++;
13541f11a4fSYoshinori Sato 
1361da177e4SLinus Torvalds 	/* Ignore most absolute/undefined (?) symbols. */
137fd593d12SEric W. Biederman 	if (strcmp(sym, "_text") == 0)
138fd593d12SEric W. Biederman 		_text = s->addr;
13978eb7159SKees Cook 	else if (check_symbol_range(sym, s->addr, text_ranges,
14078eb7159SKees Cook 				    ARRAY_SIZE(text_ranges)) == 0)
14117b1f0deSMike Frysinger 		/* nothing to do */;
142b3dbb4ecSPaulo Marques 	else if (toupper(stype) == 'A')
1431da177e4SLinus Torvalds 	{
1441da177e4SLinus Torvalds 		/* Keep these useful absolute symbols */
14541f11a4fSYoshinori Sato 		if (strcmp(sym, "__kernel_syscall_via_break") &&
14641f11a4fSYoshinori Sato 		    strcmp(sym, "__kernel_syscall_via_epc") &&
14741f11a4fSYoshinori Sato 		    strcmp(sym, "__kernel_sigtramp") &&
14841f11a4fSYoshinori Sato 		    strcmp(sym, "__gp"))
1491da177e4SLinus Torvalds 			return -1;
1501da177e4SLinus Torvalds 
1511da177e4SLinus Torvalds 	}
152b3dbb4ecSPaulo Marques 	else if (toupper(stype) == 'U' ||
15341f11a4fSYoshinori Sato 		 is_arm_mapping_symbol(sym))
1541da177e4SLinus Torvalds 		return -1;
1556f00df24SRalf Baechle 	/* exclude also MIPS ELF local symbols ($L123 instead of .L123) */
1566f00df24SRalf Baechle 	else if (str[0] == '$')
1576f00df24SRalf Baechle 		return -1;
158aab34ac8SSam Ravnborg 	/* exclude debugging symbols */
159aab34ac8SSam Ravnborg 	else if (stype == 'N')
160aab34ac8SSam Ravnborg 		return -1;
1611da177e4SLinus Torvalds 
1621da177e4SLinus Torvalds 	/* include the type field in the symbol name, so that it gets
1631da177e4SLinus Torvalds 	 * compressed together */
1641da177e4SLinus Torvalds 	s->len = strlen(str) + 1;
165b3dbb4ecSPaulo Marques 	s->sym = malloc(s->len + 1);
166f1a136e0SJesper Juhl 	if (!s->sym) {
167f1a136e0SJesper Juhl 		fprintf(stderr, "kallsyms failure: "
168f1a136e0SJesper Juhl 			"unable to allocate required amount of memory\n");
169f1a136e0SJesper Juhl 		exit(EXIT_FAILURE);
170f1a136e0SJesper Juhl 	}
171b3dbb4ecSPaulo Marques 	strcpy((char *)s->sym + 1, str);
172b3dbb4ecSPaulo Marques 	s->sym[0] = stype;
1731da177e4SLinus Torvalds 
174c6bda7c9SRusty Russell 	/* Record if we've found __per_cpu_start/end. */
175c6bda7c9SRusty Russell 	check_symbol_range(sym, s->addr, &percpu_range, 1);
176c6bda7c9SRusty Russell 
1771da177e4SLinus Torvalds 	return 0;
1781da177e4SLinus Torvalds }
1791da177e4SLinus Torvalds 
18078eb7159SKees Cook static int symbol_in_range(struct sym_entry *s, struct addr_range *ranges,
18178eb7159SKees Cook 			   int entries)
18217b1f0deSMike Frysinger {
18317b1f0deSMike Frysinger 	size_t i;
18478eb7159SKees Cook 	struct addr_range *ar;
18517b1f0deSMike Frysinger 
18678eb7159SKees Cook 	for (i = 0; i < entries; ++i) {
18778eb7159SKees Cook 		ar = &ranges[i];
18817b1f0deSMike Frysinger 
18978eb7159SKees Cook 		if (s->addr >= ar->start && s->addr <= ar->end)
190ac6ca5c8SMike Frysinger 			return 1;
19117b1f0deSMike Frysinger 	}
19217b1f0deSMike Frysinger 
193ac6ca5c8SMike Frysinger 	return 0;
19417b1f0deSMike Frysinger }
19517b1f0deSMike Frysinger 
196b3dbb4ecSPaulo Marques static int symbol_valid(struct sym_entry *s)
1971da177e4SLinus Torvalds {
1981da177e4SLinus Torvalds 	/* Symbols which vary between passes.  Passes 1 and 2 must have
1992ea03891SSam Ravnborg 	 * identical symbol lists.  The kallsyms_* symbols below are only added
2002ea03891SSam Ravnborg 	 * after pass 1, they would be included in pass 2 when --all-symbols is
2012ea03891SSam Ravnborg 	 * specified so exclude them to get a stable symbol list.
2021da177e4SLinus Torvalds 	 */
2031da177e4SLinus Torvalds 	static char *special_symbols[] = {
2042ea03891SSam Ravnborg 		"kallsyms_addresses",
2052ea03891SSam Ravnborg 		"kallsyms_num_syms",
2062ea03891SSam Ravnborg 		"kallsyms_names",
2072ea03891SSam Ravnborg 		"kallsyms_markers",
2082ea03891SSam Ravnborg 		"kallsyms_token_table",
2092ea03891SSam Ravnborg 		"kallsyms_token_index",
2102ea03891SSam Ravnborg 
2111da177e4SLinus Torvalds 	/* Exclude linker generated symbols which vary between passes */
2121da177e4SLinus Torvalds 		"_SDA_BASE_",		/* ppc */
2131da177e4SLinus Torvalds 		"_SDA2_BASE_",		/* ppc */
2141da177e4SLinus Torvalds 		NULL };
215bd8b22d2SArd Biesheuvel 
216bd8b22d2SArd Biesheuvel 	static char *special_suffixes[] = {
217bd8b22d2SArd Biesheuvel 		"_veneer",		/* arm */
218bd8b22d2SArd Biesheuvel 		NULL };
219bd8b22d2SArd Biesheuvel 
2201da177e4SLinus Torvalds 	int i;
221bd8b22d2SArd Biesheuvel 	char *sym_name = (char *)s->sym + 1;
222bd8b22d2SArd Biesheuvel 
22341f11a4fSYoshinori Sato 
224f6537f2fSMing Lei 	if (s->addr < kernel_start_addr)
225f6537f2fSMing Lei 		return 0;
226f6537f2fSMing Lei 
22741f11a4fSYoshinori Sato 	/* skip prefix char */
228bd8b22d2SArd Biesheuvel 	if (symbol_prefix_char && *sym_name == symbol_prefix_char)
229bd8b22d2SArd Biesheuvel 		sym_name++;
230bd8b22d2SArd Biesheuvel 
2311da177e4SLinus Torvalds 
2321da177e4SLinus Torvalds 	/* if --all-symbols is not specified, then symbols outside the text
2331da177e4SLinus Torvalds 	 * and inittext sections are discarded */
2341da177e4SLinus Torvalds 	if (!all_symbols) {
23578eb7159SKees Cook 		if (symbol_in_range(s, text_ranges,
23678eb7159SKees Cook 				    ARRAY_SIZE(text_ranges)) == 0)
2371da177e4SLinus Torvalds 			return 0;
2381da177e4SLinus Torvalds 		/* Corner case.  Discard any symbols with the same value as
239a3b81113SRobin Getz 		 * _etext _einittext; they can move between pass 1 and 2 when
240a3b81113SRobin Getz 		 * the kallsyms data are added.  If these symbols move then
241a3b81113SRobin Getz 		 * they may get dropped in pass 2, which breaks the kallsyms
242a3b81113SRobin Getz 		 * rules.
2431da177e4SLinus Torvalds 		 */
24417b1f0deSMike Frysinger 		if ((s->addr == text_range_text->end &&
245bd8b22d2SArd Biesheuvel 				strcmp(sym_name,
24678eb7159SKees Cook 				       text_range_text->end_sym)) ||
24717b1f0deSMike Frysinger 		    (s->addr == text_range_inittext->end &&
248bd8b22d2SArd Biesheuvel 				strcmp(sym_name,
24978eb7159SKees Cook 				       text_range_inittext->end_sym)))
2501da177e4SLinus Torvalds 			return 0;
2511da177e4SLinus Torvalds 	}
2521da177e4SLinus Torvalds 
2531da177e4SLinus Torvalds 	/* Exclude symbols which vary between passes. */
254bd8b22d2SArd Biesheuvel 	for (i = 0; special_symbols[i]; i++)
255bd8b22d2SArd Biesheuvel 		if (strcmp(sym_name, special_symbols[i]) == 0)
2561da177e4SLinus Torvalds 			return 0;
2571da177e4SLinus Torvalds 
258bd8b22d2SArd Biesheuvel 	for (i = 0; special_suffixes[i]; i++) {
259bd8b22d2SArd Biesheuvel 		int l = strlen(sym_name) - strlen(special_suffixes[i]);
260bd8b22d2SArd Biesheuvel 
261bd8b22d2SArd Biesheuvel 		if (l >= 0 && strcmp(sym_name + l, special_suffixes[i]) == 0)
2621da177e4SLinus Torvalds 			return 0;
263bd8b22d2SArd Biesheuvel 	}
2641da177e4SLinus Torvalds 
2651da177e4SLinus Torvalds 	return 1;
2661da177e4SLinus Torvalds }
2671da177e4SLinus Torvalds 
268b3dbb4ecSPaulo Marques static void read_map(FILE *in)
2691da177e4SLinus Torvalds {
2701da177e4SLinus Torvalds 	while (!feof(in)) {
271b3dbb4ecSPaulo Marques 		if (table_cnt >= table_size) {
272b3dbb4ecSPaulo Marques 			table_size += 10000;
273b3dbb4ecSPaulo Marques 			table = realloc(table, sizeof(*table) * table_size);
2741da177e4SLinus Torvalds 			if (!table) {
2751da177e4SLinus Torvalds 				fprintf(stderr, "out of memory\n");
2761da177e4SLinus Torvalds 				exit (1);
2771da177e4SLinus Torvalds 			}
2781da177e4SLinus Torvalds 		}
279f2df3f65SPaulo Marques 		if (read_symbol(in, &table[table_cnt]) == 0) {
280f2df3f65SPaulo Marques 			table[table_cnt].start_pos = table_cnt;
281b3dbb4ecSPaulo Marques 			table_cnt++;
2821da177e4SLinus Torvalds 		}
2831da177e4SLinus Torvalds 	}
284f2df3f65SPaulo Marques }
2851da177e4SLinus Torvalds 
2861da177e4SLinus Torvalds static void output_label(char *label)
2871da177e4SLinus Torvalds {
28841f11a4fSYoshinori Sato 	if (symbol_prefix_char)
28941f11a4fSYoshinori Sato 		printf(".globl %c%s\n", symbol_prefix_char, label);
29041f11a4fSYoshinori Sato 	else
2911da177e4SLinus Torvalds 		printf(".globl %s\n", label);
2921da177e4SLinus Torvalds 	printf("\tALGN\n");
29341f11a4fSYoshinori Sato 	if (symbol_prefix_char)
29441f11a4fSYoshinori Sato 		printf("%c%s:\n", symbol_prefix_char, label);
29541f11a4fSYoshinori Sato 	else
2961da177e4SLinus Torvalds 		printf("%s:\n", label);
2971da177e4SLinus Torvalds }
2981da177e4SLinus Torvalds 
2991da177e4SLinus Torvalds /* uncompress a compressed symbol. When this function is called, the best table
3001da177e4SLinus Torvalds  * might still be compressed itself, so the function needs to be recursive */
3011da177e4SLinus Torvalds static int expand_symbol(unsigned char *data, int len, char *result)
3021da177e4SLinus Torvalds {
3031da177e4SLinus Torvalds 	int c, rlen, total=0;
3041da177e4SLinus Torvalds 
3051da177e4SLinus Torvalds 	while (len) {
3061da177e4SLinus Torvalds 		c = *data;
3071da177e4SLinus Torvalds 		/* if the table holds a single char that is the same as the one
3081da177e4SLinus Torvalds 		 * we are looking for, then end the search */
3091da177e4SLinus Torvalds 		if (best_table[c][0]==c && best_table_len[c]==1) {
3101da177e4SLinus Torvalds 			*result++ = c;
3111da177e4SLinus Torvalds 			total++;
3121da177e4SLinus Torvalds 		} else {
3131da177e4SLinus Torvalds 			/* if not, recurse and expand */
3141da177e4SLinus Torvalds 			rlen = expand_symbol(best_table[c], best_table_len[c], result);
3151da177e4SLinus Torvalds 			total += rlen;
3161da177e4SLinus Torvalds 			result += rlen;
3171da177e4SLinus Torvalds 		}
3181da177e4SLinus Torvalds 		data++;
3191da177e4SLinus Torvalds 		len--;
3201da177e4SLinus Torvalds 	}
3211da177e4SLinus Torvalds 	*result=0;
3221da177e4SLinus Torvalds 
3231da177e4SLinus Torvalds 	return total;
3241da177e4SLinus Torvalds }
3251da177e4SLinus Torvalds 
32678eb7159SKees Cook static int symbol_absolute(struct sym_entry *s)
32778eb7159SKees Cook {
32878eb7159SKees Cook 	return toupper(s->sym[0]) == 'A';
32978eb7159SKees Cook }
33078eb7159SKees Cook 
331b3dbb4ecSPaulo Marques static void write_src(void)
3321da177e4SLinus Torvalds {
333b3dbb4ecSPaulo Marques 	unsigned int i, k, off;
3341da177e4SLinus Torvalds 	unsigned int best_idx[256];
3351da177e4SLinus Torvalds 	unsigned int *markers;
3369281aceaSTejun Heo 	char buf[KSYM_NAME_LEN];
3371da177e4SLinus Torvalds 
3381da177e4SLinus Torvalds 	printf("#include <asm/types.h>\n");
3391da177e4SLinus Torvalds 	printf("#if BITS_PER_LONG == 64\n");
3401da177e4SLinus Torvalds 	printf("#define PTR .quad\n");
3411da177e4SLinus Torvalds 	printf("#define ALGN .align 8\n");
3421da177e4SLinus Torvalds 	printf("#else\n");
3431da177e4SLinus Torvalds 	printf("#define PTR .long\n");
3441da177e4SLinus Torvalds 	printf("#define ALGN .align 4\n");
3451da177e4SLinus Torvalds 	printf("#endif\n");
3461da177e4SLinus Torvalds 
347aad09470SJan Beulich 	printf("\t.section .rodata, \"a\"\n");
3481da177e4SLinus Torvalds 
349fd593d12SEric W. Biederman 	/* Provide proper symbols relocatability by their '_text'
350fd593d12SEric W. Biederman 	 * relativeness.  The symbol names cannot be used to construct
351fd593d12SEric W. Biederman 	 * normal symbol references as the list of symbols contains
352fd593d12SEric W. Biederman 	 * symbols that are declared static and are private to their
353fd593d12SEric W. Biederman 	 * .o files.  This prevents .tmp_kallsyms.o or any other
354fd593d12SEric W. Biederman 	 * object from referencing them.
355fd593d12SEric W. Biederman 	 */
3561da177e4SLinus Torvalds 	output_label("kallsyms_addresses");
357b3dbb4ecSPaulo Marques 	for (i = 0; i < table_cnt; i++) {
35878eb7159SKees Cook 		if (!symbol_absolute(&table[i])) {
3592c22d8baSVivek Goyal 			if (_text <= table[i].addr)
360fd593d12SEric W. Biederman 				printf("\tPTR\t_text + %#llx\n",
361fd593d12SEric W. Biederman 					table[i].addr - _text);
3622c22d8baSVivek Goyal 			else
3632930ffc7SAndrew Morton 				printf("\tPTR\t_text - %#llx\n",
3642930ffc7SAndrew Morton 					_text - table[i].addr);
365fd593d12SEric W. Biederman 		} else {
3661da177e4SLinus Torvalds 			printf("\tPTR\t%#llx\n", table[i].addr);
3671da177e4SLinus Torvalds 		}
368fd593d12SEric W. Biederman 	}
3691da177e4SLinus Torvalds 	printf("\n");
3701da177e4SLinus Torvalds 
3711da177e4SLinus Torvalds 	output_label("kallsyms_num_syms");
372b3dbb4ecSPaulo Marques 	printf("\tPTR\t%d\n", table_cnt);
3731da177e4SLinus Torvalds 	printf("\n");
3741da177e4SLinus Torvalds 
3751da177e4SLinus Torvalds 	/* table of offset markers, that give the offset in the compressed stream
3761da177e4SLinus Torvalds 	 * every 256 symbols */
377f1a136e0SJesper Juhl 	markers = malloc(sizeof(unsigned int) * ((table_cnt + 255) / 256));
378f1a136e0SJesper Juhl 	if (!markers) {
379f1a136e0SJesper Juhl 		fprintf(stderr, "kallsyms failure: "
380f1a136e0SJesper Juhl 			"unable to allocate required memory\n");
381f1a136e0SJesper Juhl 		exit(EXIT_FAILURE);
382f1a136e0SJesper Juhl 	}
3831da177e4SLinus Torvalds 
3841da177e4SLinus Torvalds 	output_label("kallsyms_names");
3851da177e4SLinus Torvalds 	off = 0;
386b3dbb4ecSPaulo Marques 	for (i = 0; i < table_cnt; i++) {
387b3dbb4ecSPaulo Marques 		if ((i & 0xFF) == 0)
388b3dbb4ecSPaulo Marques 			markers[i >> 8] = off;
3891da177e4SLinus Torvalds 
3901da177e4SLinus Torvalds 		printf("\t.byte 0x%02x", table[i].len);
3911da177e4SLinus Torvalds 		for (k = 0; k < table[i].len; k++)
3921da177e4SLinus Torvalds 			printf(", 0x%02x", table[i].sym[k]);
3931da177e4SLinus Torvalds 		printf("\n");
3941da177e4SLinus Torvalds 
3951da177e4SLinus Torvalds 		off += table[i].len + 1;
3961da177e4SLinus Torvalds 	}
3971da177e4SLinus Torvalds 	printf("\n");
3981da177e4SLinus Torvalds 
3991da177e4SLinus Torvalds 	output_label("kallsyms_markers");
400b3dbb4ecSPaulo Marques 	for (i = 0; i < ((table_cnt + 255) >> 8); i++)
4011da177e4SLinus Torvalds 		printf("\tPTR\t%d\n", markers[i]);
4021da177e4SLinus Torvalds 	printf("\n");
4031da177e4SLinus Torvalds 
4041da177e4SLinus Torvalds 	free(markers);
4051da177e4SLinus Torvalds 
4061da177e4SLinus Torvalds 	output_label("kallsyms_token_table");
4071da177e4SLinus Torvalds 	off = 0;
4081da177e4SLinus Torvalds 	for (i = 0; i < 256; i++) {
4091da177e4SLinus Torvalds 		best_idx[i] = off;
4101da177e4SLinus Torvalds 		expand_symbol(best_table[i], best_table_len[i], buf);
4111da177e4SLinus Torvalds 		printf("\t.asciz\t\"%s\"\n", buf);
4121da177e4SLinus Torvalds 		off += strlen(buf) + 1;
4131da177e4SLinus Torvalds 	}
4141da177e4SLinus Torvalds 	printf("\n");
4151da177e4SLinus Torvalds 
4161da177e4SLinus Torvalds 	output_label("kallsyms_token_index");
4171da177e4SLinus Torvalds 	for (i = 0; i < 256; i++)
4181da177e4SLinus Torvalds 		printf("\t.short\t%d\n", best_idx[i]);
4191da177e4SLinus Torvalds 	printf("\n");
4201da177e4SLinus Torvalds }
4211da177e4SLinus Torvalds 
4221da177e4SLinus Torvalds 
4231da177e4SLinus Torvalds /* table lookup compression functions */
4241da177e4SLinus Torvalds 
4251da177e4SLinus Torvalds /* count all the possible tokens in a symbol */
4261da177e4SLinus Torvalds static void learn_symbol(unsigned char *symbol, int len)
4271da177e4SLinus Torvalds {
4281da177e4SLinus Torvalds 	int i;
4291da177e4SLinus Torvalds 
4301da177e4SLinus Torvalds 	for (i = 0; i < len - 1; i++)
431b3dbb4ecSPaulo Marques 		token_profit[ symbol[i] + (symbol[i + 1] << 8) ]++;
4321da177e4SLinus Torvalds }
4331da177e4SLinus Torvalds 
4341da177e4SLinus Torvalds /* decrease the count for all the possible tokens in a symbol */
4351da177e4SLinus Torvalds static void forget_symbol(unsigned char *symbol, int len)
4361da177e4SLinus Torvalds {
4371da177e4SLinus Torvalds 	int i;
4381da177e4SLinus Torvalds 
4391da177e4SLinus Torvalds 	for (i = 0; i < len - 1; i++)
440b3dbb4ecSPaulo Marques 		token_profit[ symbol[i] + (symbol[i + 1] << 8) ]--;
4411da177e4SLinus Torvalds }
4421da177e4SLinus Torvalds 
443b3dbb4ecSPaulo Marques /* remove all the invalid symbols from the table and do the initial token count */
4441da177e4SLinus Torvalds static void build_initial_tok_table(void)
4451da177e4SLinus Torvalds {
446b3dbb4ecSPaulo Marques 	unsigned int i, pos;
4471da177e4SLinus Torvalds 
448b3dbb4ecSPaulo Marques 	pos = 0;
449b3dbb4ecSPaulo Marques 	for (i = 0; i < table_cnt; i++) {
4501da177e4SLinus Torvalds 		if ( symbol_valid(&table[i]) ) {
451b3dbb4ecSPaulo Marques 			if (pos != i)
452b3dbb4ecSPaulo Marques 				table[pos] = table[i];
453b3dbb4ecSPaulo Marques 			learn_symbol(table[pos].sym, table[pos].len);
454b3dbb4ecSPaulo Marques 			pos++;
4551da177e4SLinus Torvalds 		}
4561da177e4SLinus Torvalds 	}
457b3dbb4ecSPaulo Marques 	table_cnt = pos;
4581da177e4SLinus Torvalds }
4591da177e4SLinus Torvalds 
4607c5d249aSPaulo Marques static void *find_token(unsigned char *str, int len, unsigned char *token)
4617c5d249aSPaulo Marques {
4627c5d249aSPaulo Marques 	int i;
4637c5d249aSPaulo Marques 
4647c5d249aSPaulo Marques 	for (i = 0; i < len - 1; i++) {
4657c5d249aSPaulo Marques 		if (str[i] == token[0] && str[i+1] == token[1])
4667c5d249aSPaulo Marques 			return &str[i];
4677c5d249aSPaulo Marques 	}
4687c5d249aSPaulo Marques 	return NULL;
4697c5d249aSPaulo Marques }
4707c5d249aSPaulo Marques 
4711da177e4SLinus Torvalds /* replace a given token in all the valid symbols. Use the sampled symbols
4721da177e4SLinus Torvalds  * to update the counts */
473b3dbb4ecSPaulo Marques static void compress_symbols(unsigned char *str, int idx)
4741da177e4SLinus Torvalds {
475b3dbb4ecSPaulo Marques 	unsigned int i, len, size;
476b3dbb4ecSPaulo Marques 	unsigned char *p1, *p2;
4771da177e4SLinus Torvalds 
478b3dbb4ecSPaulo Marques 	for (i = 0; i < table_cnt; i++) {
4791da177e4SLinus Torvalds 
4801da177e4SLinus Torvalds 		len = table[i].len;
481b3dbb4ecSPaulo Marques 		p1 = table[i].sym;
482b3dbb4ecSPaulo Marques 
483b3dbb4ecSPaulo Marques 		/* find the token on the symbol */
4847c5d249aSPaulo Marques 		p2 = find_token(p1, len, str);
485b3dbb4ecSPaulo Marques 		if (!p2) continue;
486b3dbb4ecSPaulo Marques 
487b3dbb4ecSPaulo Marques 		/* decrease the counts for this symbol's tokens */
488b3dbb4ecSPaulo Marques 		forget_symbol(table[i].sym, len);
489b3dbb4ecSPaulo Marques 
490b3dbb4ecSPaulo Marques 		size = len;
4911da177e4SLinus Torvalds 
4921da177e4SLinus Torvalds 		do {
493b3dbb4ecSPaulo Marques 			*p2 = idx;
494b3dbb4ecSPaulo Marques 			p2++;
495b3dbb4ecSPaulo Marques 			size -= (p2 - p1);
496b3dbb4ecSPaulo Marques 			memmove(p2, p2 + 1, size);
497b3dbb4ecSPaulo Marques 			p1 = p2;
498b3dbb4ecSPaulo Marques 			len--;
499b3dbb4ecSPaulo Marques 
500b3dbb4ecSPaulo Marques 			if (size < 2) break;
501b3dbb4ecSPaulo Marques 
5021da177e4SLinus Torvalds 			/* find the token on the symbol */
5037c5d249aSPaulo Marques 			p2 = find_token(p1, size, str);
5041da177e4SLinus Torvalds 
505b3dbb4ecSPaulo Marques 		} while (p2);
5061da177e4SLinus Torvalds 
5071da177e4SLinus Torvalds 		table[i].len = len;
508b3dbb4ecSPaulo Marques 
509b3dbb4ecSPaulo Marques 		/* increase the counts for this symbol's new tokens */
5101da177e4SLinus Torvalds 		learn_symbol(table[i].sym, len);
5111da177e4SLinus Torvalds 	}
5121da177e4SLinus Torvalds }
5131da177e4SLinus Torvalds 
5141da177e4SLinus Torvalds /* search the token with the maximum profit */
515b3dbb4ecSPaulo Marques static int find_best_token(void)
5161da177e4SLinus Torvalds {
517b3dbb4ecSPaulo Marques 	int i, best, bestprofit;
5181da177e4SLinus Torvalds 
5191da177e4SLinus Torvalds 	bestprofit=-10000;
520b3dbb4ecSPaulo Marques 	best = 0;
5211da177e4SLinus Torvalds 
522b3dbb4ecSPaulo Marques 	for (i = 0; i < 0x10000; i++) {
523b3dbb4ecSPaulo Marques 		if (token_profit[i] > bestprofit) {
524b3dbb4ecSPaulo Marques 			best = i;
525b3dbb4ecSPaulo Marques 			bestprofit = token_profit[i];
5261da177e4SLinus Torvalds 		}
5271da177e4SLinus Torvalds 	}
5281da177e4SLinus Torvalds 	return best;
5291da177e4SLinus Torvalds }
5301da177e4SLinus Torvalds 
5311da177e4SLinus Torvalds /* this is the core of the algorithm: calculate the "best" table */
5321da177e4SLinus Torvalds static void optimize_result(void)
5331da177e4SLinus Torvalds {
534b3dbb4ecSPaulo Marques 	int i, best;
5351da177e4SLinus Torvalds 
5361da177e4SLinus Torvalds 	/* using the '\0' symbol last allows compress_symbols to use standard
5371da177e4SLinus Torvalds 	 * fast string functions */
5381da177e4SLinus Torvalds 	for (i = 255; i >= 0; i--) {
5391da177e4SLinus Torvalds 
5401da177e4SLinus Torvalds 		/* if this table slot is empty (it is not used by an actual
5411da177e4SLinus Torvalds 		 * original char code */
5421da177e4SLinus Torvalds 		if (!best_table_len[i]) {
5431da177e4SLinus Torvalds 
5441da177e4SLinus Torvalds 			/* find the token with the breates profit value */
5451da177e4SLinus Torvalds 			best = find_best_token();
546e0a04b11SXiaochen Wang 			if (token_profit[best] == 0)
547e0a04b11SXiaochen Wang 				break;
5481da177e4SLinus Torvalds 
5491da177e4SLinus Torvalds 			/* place it in the "best" table */
550b3dbb4ecSPaulo Marques 			best_table_len[i] = 2;
551b3dbb4ecSPaulo Marques 			best_table[i][0] = best & 0xFF;
552b3dbb4ecSPaulo Marques 			best_table[i][1] = (best >> 8) & 0xFF;
5531da177e4SLinus Torvalds 
5541da177e4SLinus Torvalds 			/* replace this token in all the valid symbols */
555b3dbb4ecSPaulo Marques 			compress_symbols(best_table[i], i);
5561da177e4SLinus Torvalds 		}
5571da177e4SLinus Torvalds 	}
5581da177e4SLinus Torvalds }
5591da177e4SLinus Torvalds 
5601da177e4SLinus Torvalds /* start by placing the symbols that are actually used on the table */
5611da177e4SLinus Torvalds static void insert_real_symbols_in_table(void)
5621da177e4SLinus Torvalds {
563b3dbb4ecSPaulo Marques 	unsigned int i, j, c;
5641da177e4SLinus Torvalds 
5651da177e4SLinus Torvalds 	memset(best_table, 0, sizeof(best_table));
5661da177e4SLinus Torvalds 	memset(best_table_len, 0, sizeof(best_table_len));
5671da177e4SLinus Torvalds 
568b3dbb4ecSPaulo Marques 	for (i = 0; i < table_cnt; i++) {
5691da177e4SLinus Torvalds 		for (j = 0; j < table[i].len; j++) {
5701da177e4SLinus Torvalds 			c = table[i].sym[j];
5711da177e4SLinus Torvalds 			best_table[c][0]=c;
5721da177e4SLinus Torvalds 			best_table_len[c]=1;
5731da177e4SLinus Torvalds 		}
5741da177e4SLinus Torvalds 	}
5751da177e4SLinus Torvalds }
5761da177e4SLinus Torvalds 
5771da177e4SLinus Torvalds static void optimize_token_table(void)
5781da177e4SLinus Torvalds {
5791da177e4SLinus Torvalds 	build_initial_tok_table();
5801da177e4SLinus Torvalds 
5811da177e4SLinus Torvalds 	insert_real_symbols_in_table();
5821da177e4SLinus Torvalds 
58341f11a4fSYoshinori Sato 	/* When valid symbol is not registered, exit to error */
584b3dbb4ecSPaulo Marques 	if (!table_cnt) {
58541f11a4fSYoshinori Sato 		fprintf(stderr, "No valid symbol.\n");
58641f11a4fSYoshinori Sato 		exit(1);
58741f11a4fSYoshinori Sato 	}
58841f11a4fSYoshinori Sato 
5891da177e4SLinus Torvalds 	optimize_result();
5901da177e4SLinus Torvalds }
5911da177e4SLinus Torvalds 
592b478b782SLai Jiangshan /* guess for "linker script provide" symbol */
593b478b782SLai Jiangshan static int may_be_linker_script_provide_symbol(const struct sym_entry *se)
594b478b782SLai Jiangshan {
595b478b782SLai Jiangshan 	const char *symbol = (char *)se->sym + 1;
596b478b782SLai Jiangshan 	int len = se->len - 1;
597b478b782SLai Jiangshan 
598b478b782SLai Jiangshan 	if (len < 8)
599b478b782SLai Jiangshan 		return 0;
600b478b782SLai Jiangshan 
601b478b782SLai Jiangshan 	if (symbol[0] != '_' || symbol[1] != '_')
602b478b782SLai Jiangshan 		return 0;
603b478b782SLai Jiangshan 
604b478b782SLai Jiangshan 	/* __start_XXXXX */
605b478b782SLai Jiangshan 	if (!memcmp(symbol + 2, "start_", 6))
606b478b782SLai Jiangshan 		return 1;
607b478b782SLai Jiangshan 
608b478b782SLai Jiangshan 	/* __stop_XXXXX */
609b478b782SLai Jiangshan 	if (!memcmp(symbol + 2, "stop_", 5))
610b478b782SLai Jiangshan 		return 1;
611b478b782SLai Jiangshan 
612b478b782SLai Jiangshan 	/* __end_XXXXX */
613b478b782SLai Jiangshan 	if (!memcmp(symbol + 2, "end_", 4))
614b478b782SLai Jiangshan 		return 1;
615b478b782SLai Jiangshan 
616b478b782SLai Jiangshan 	/* __XXXXX_start */
617b478b782SLai Jiangshan 	if (!memcmp(symbol + len - 6, "_start", 6))
618b478b782SLai Jiangshan 		return 1;
619b478b782SLai Jiangshan 
620b478b782SLai Jiangshan 	/* __XXXXX_end */
621b478b782SLai Jiangshan 	if (!memcmp(symbol + len - 4, "_end", 4))
622b478b782SLai Jiangshan 		return 1;
623b478b782SLai Jiangshan 
624b478b782SLai Jiangshan 	return 0;
625b478b782SLai Jiangshan }
626b478b782SLai Jiangshan 
627b478b782SLai Jiangshan static int prefix_underscores_count(const char *str)
628b478b782SLai Jiangshan {
629b478b782SLai Jiangshan 	const char *tail = str;
630b478b782SLai Jiangshan 
631a9ece53cSPaul Mundt 	while (*tail == '_')
632b478b782SLai Jiangshan 		tail++;
633b478b782SLai Jiangshan 
634b478b782SLai Jiangshan 	return tail - str;
635b478b782SLai Jiangshan }
636b478b782SLai Jiangshan 
637f2df3f65SPaulo Marques static int compare_symbols(const void *a, const void *b)
638f2df3f65SPaulo Marques {
639f2df3f65SPaulo Marques 	const struct sym_entry *sa;
640f2df3f65SPaulo Marques 	const struct sym_entry *sb;
641f2df3f65SPaulo Marques 	int wa, wb;
642f2df3f65SPaulo Marques 
643f2df3f65SPaulo Marques 	sa = a;
644f2df3f65SPaulo Marques 	sb = b;
645f2df3f65SPaulo Marques 
646f2df3f65SPaulo Marques 	/* sort by address first */
647f2df3f65SPaulo Marques 	if (sa->addr > sb->addr)
648f2df3f65SPaulo Marques 		return 1;
649f2df3f65SPaulo Marques 	if (sa->addr < sb->addr)
650f2df3f65SPaulo Marques 		return -1;
651f2df3f65SPaulo Marques 
652f2df3f65SPaulo Marques 	/* sort by "weakness" type */
653f2df3f65SPaulo Marques 	wa = (sa->sym[0] == 'w') || (sa->sym[0] == 'W');
654f2df3f65SPaulo Marques 	wb = (sb->sym[0] == 'w') || (sb->sym[0] == 'W');
655f2df3f65SPaulo Marques 	if (wa != wb)
656f2df3f65SPaulo Marques 		return wa - wb;
657f2df3f65SPaulo Marques 
658b478b782SLai Jiangshan 	/* sort by "linker script provide" type */
659b478b782SLai Jiangshan 	wa = may_be_linker_script_provide_symbol(sa);
660b478b782SLai Jiangshan 	wb = may_be_linker_script_provide_symbol(sb);
661b478b782SLai Jiangshan 	if (wa != wb)
662b478b782SLai Jiangshan 		return wa - wb;
663b478b782SLai Jiangshan 
664b478b782SLai Jiangshan 	/* sort by the number of prefix underscores */
665b478b782SLai Jiangshan 	wa = prefix_underscores_count((const char *)sa->sym + 1);
666b478b782SLai Jiangshan 	wb = prefix_underscores_count((const char *)sb->sym + 1);
667b478b782SLai Jiangshan 	if (wa != wb)
668b478b782SLai Jiangshan 		return wa - wb;
669b478b782SLai Jiangshan 
670f2df3f65SPaulo Marques 	/* sort by initial order, so that other symbols are left undisturbed */
671f2df3f65SPaulo Marques 	return sa->start_pos - sb->start_pos;
672f2df3f65SPaulo Marques }
673f2df3f65SPaulo Marques 
674f2df3f65SPaulo Marques static void sort_symbols(void)
675f2df3f65SPaulo Marques {
676f2df3f65SPaulo Marques 	qsort(table, table_cnt, sizeof(struct sym_entry), compare_symbols);
677f2df3f65SPaulo Marques }
6781da177e4SLinus Torvalds 
679c6bda7c9SRusty Russell static void make_percpus_absolute(void)
680c6bda7c9SRusty Russell {
681c6bda7c9SRusty Russell 	unsigned int i;
682c6bda7c9SRusty Russell 
683c6bda7c9SRusty Russell 	for (i = 0; i < table_cnt; i++)
684c6bda7c9SRusty Russell 		if (symbol_in_range(&table[i], &percpu_range, 1))
685c6bda7c9SRusty Russell 			table[i].sym[0] = 'A';
686c6bda7c9SRusty Russell }
687c6bda7c9SRusty Russell 
688b3dbb4ecSPaulo Marques int main(int argc, char **argv)
6891da177e4SLinus Torvalds {
69041f11a4fSYoshinori Sato 	if (argc >= 2) {
69141f11a4fSYoshinori Sato 		int i;
69241f11a4fSYoshinori Sato 		for (i = 1; i < argc; i++) {
69341f11a4fSYoshinori Sato 			if(strcmp(argv[i], "--all-symbols") == 0)
6941da177e4SLinus Torvalds 				all_symbols = 1;
695c6bda7c9SRusty Russell 			else if (strcmp(argv[i], "--absolute-percpu") == 0)
696c6bda7c9SRusty Russell 				absolute_percpu = 1;
69741f11a4fSYoshinori Sato 			else if (strncmp(argv[i], "--symbol-prefix=", 16) == 0) {
69841f11a4fSYoshinori Sato 				char *p = &argv[i][16];
69941f11a4fSYoshinori Sato 				/* skip quote */
70041f11a4fSYoshinori Sato 				if ((*p == '"' && *(p+2) == '"') || (*p == '\'' && *(p+2) == '\''))
70141f11a4fSYoshinori Sato 					p++;
70241f11a4fSYoshinori Sato 				symbol_prefix_char = *p;
703f6537f2fSMing Lei 			} else if (strncmp(argv[i], "--page-offset=", 14) == 0) {
704f6537f2fSMing Lei 				const char *p = &argv[i][14];
705f6537f2fSMing Lei 				kernel_start_addr = strtoull(p, NULL, 16);
70641f11a4fSYoshinori Sato 			} else
70741f11a4fSYoshinori Sato 				usage();
70841f11a4fSYoshinori Sato 		}
70941f11a4fSYoshinori Sato 	} else if (argc != 1)
7101da177e4SLinus Torvalds 		usage();
7111da177e4SLinus Torvalds 
7121da177e4SLinus Torvalds 	read_map(stdin);
713c6bda7c9SRusty Russell 	if (absolute_percpu)
714c6bda7c9SRusty Russell 		make_percpus_absolute();
715f2df3f65SPaulo Marques 	sort_symbols();
7161da177e4SLinus Torvalds 	optimize_token_table();
7171da177e4SLinus Torvalds 	write_src();
7181da177e4SLinus Torvalds 
7191da177e4SLinus Torvalds 	return 0;
7201da177e4SLinus Torvalds }
721