1 /* Generate assembler source containing symbol 1 /* Generate assembler source containing symbol information
2 * 2 *
3 * Copyright 2002 by Kai Germaschewski 3 * Copyright 2002 by Kai Germaschewski
4 * 4 *
5 * This software may be used and distributed a 5 * This software may be used and distributed according to the terms
6 * of the GNU General Public License, incorpor 6 * of the GNU General Public License, incorporated herein by reference.
7 * 7 *
8 * Usage: kallsyms [--all-symbols] [--absolute !! 8 * Usage: kallsyms [--all-symbols] [--absolute-percpu]
>> 9 * [--base-relative] [--lto-clang] in.map > out.S
9 * 10 *
10 * Table compression uses all the unused 11 * Table compression uses all the unused char codes on the symbols and
11 * maps these to the most used substrings (to 12 * maps these to the most used substrings (tokens). For instance, it might
12 * map char code 0xF7 to represent "write_" a 13 * map char code 0xF7 to represent "write_" and then in every symbol where
13 * "write_" appears it can be replaced by 0xF 14 * "write_" appears it can be replaced by 0xF7, saving 5 bytes.
14 * The used codes themselves are also pla 15 * The used codes themselves are also placed in the table so that the
15 * decompresion can work without "special cas 16 * decompresion can work without "special cases".
16 * Applied to kernel symbols, this usuall 17 * Applied to kernel symbols, this usually produces a compression ratio
17 * of about 50%. 18 * of about 50%.
18 * 19 *
19 */ 20 */
20 21
21 #include <errno.h> <<
22 #include <getopt.h> 22 #include <getopt.h>
23 #include <stdbool.h> 23 #include <stdbool.h>
24 #include <stdio.h> 24 #include <stdio.h>
25 #include <stdlib.h> 25 #include <stdlib.h>
26 #include <string.h> 26 #include <string.h>
27 #include <ctype.h> 27 #include <ctype.h>
28 #include <limits.h> 28 #include <limits.h>
29 29
30 #include <xalloc.h> <<
31 <<
32 #define ARRAY_SIZE(arr) (sizeof(arr) / sizeof( 30 #define ARRAY_SIZE(arr) (sizeof(arr) / sizeof(arr[0]))
33 31
>> 32 #define _stringify_1(x) #x
>> 33 #define _stringify(x) _stringify_1(x)
>> 34
34 #define KSYM_NAME_LEN 512 35 #define KSYM_NAME_LEN 512
35 36
>> 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 */
>> 44 #define KSYM_NAME_LEN_BUFFER 2048
>> 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 );
>> 49
36 struct sym_entry { 50 struct sym_entry {
37 unsigned long long addr; 51 unsigned long long addr;
38 unsigned int len; 52 unsigned int len;
39 unsigned int seq; 53 unsigned int seq;
40 bool percpu_absolute; !! 54 unsigned int start_pos;
>> 55 unsigned int percpu_absolute;
41 unsigned char sym[]; 56 unsigned char sym[];
42 }; 57 };
43 58
44 struct addr_range { 59 struct addr_range {
45 const char *start_sym, *end_sym; 60 const char *start_sym, *end_sym;
46 unsigned long long start, end; 61 unsigned long long start, end;
47 }; 62 };
48 63
49 static unsigned long long _text; 64 static unsigned long long _text;
50 static unsigned long long relative_base; 65 static unsigned long long relative_base;
51 static struct addr_range text_ranges[] = { 66 static struct addr_range text_ranges[] = {
52 { "_stext", "_etext" }, 67 { "_stext", "_etext" },
53 { "_sinittext", "_einittext" }, 68 { "_sinittext", "_einittext" },
54 }; 69 };
55 #define text_range_text (&text_ranges[0]) 70 #define text_range_text (&text_ranges[0])
56 #define text_range_inittext (&text_ranges[1]) 71 #define text_range_inittext (&text_ranges[1])
57 72
58 static struct addr_range percpu_range = { 73 static struct addr_range percpu_range = {
59 "__per_cpu_start", "__per_cpu_end", -1 74 "__per_cpu_start", "__per_cpu_end", -1ULL, 0
60 }; 75 };
61 76
62 static struct sym_entry **table; 77 static struct sym_entry **table;
63 static unsigned int table_size, table_cnt; 78 static unsigned int table_size, table_cnt;
64 static int all_symbols; 79 static int all_symbols;
65 static int absolute_percpu; 80 static int absolute_percpu;
>> 81 static int base_relative;
>> 82 static int lto_clang;
66 83
67 static int token_profit[0x10000]; 84 static int token_profit[0x10000];
68 85
69 /* the table that holds the result of the comp 86 /* the table that holds the result of the compression */
70 static unsigned char best_table[256][2]; 87 static unsigned char best_table[256][2];
71 static unsigned char best_table_len[256]; 88 static unsigned char best_table_len[256];
72 89
73 90
74 static void usage(void) 91 static void usage(void)
75 { 92 {
76 fprintf(stderr, "Usage: kallsyms [--al !! 93 fprintf(stderr, "Usage: kallsyms [--all-symbols] [--absolute-percpu] "
>> 94 "[--base-relative] [--lto-clang] in.map > out.S\n");
77 exit(1); 95 exit(1);
78 } 96 }
79 97
80 static char *sym_name(const struct sym_entry * 98 static char *sym_name(const struct sym_entry *s)
81 { 99 {
82 return (char *)s->sym + 1; 100 return (char *)s->sym + 1;
83 } 101 }
84 102
85 static bool is_ignored_symbol(const char *name 103 static bool is_ignored_symbol(const char *name, char type)
86 { 104 {
87 if (type == 'u' || type == 'n') 105 if (type == 'u' || type == 'n')
88 return true; 106 return true;
89 107
90 if (toupper(type) == 'A') { 108 if (toupper(type) == 'A') {
91 /* Keep these useful absolute 109 /* Keep these useful absolute symbols */
92 if (strcmp(name, "__kernel_sys 110 if (strcmp(name, "__kernel_syscall_via_break") &&
93 strcmp(name, "__kernel_sys 111 strcmp(name, "__kernel_syscall_via_epc") &&
94 strcmp(name, "__kernel_sig 112 strcmp(name, "__kernel_sigtramp") &&
95 strcmp(name, "__gp")) 113 strcmp(name, "__gp"))
96 return true; 114 return true;
97 } 115 }
98 116
99 return false; 117 return false;
100 } 118 }
101 119
102 static void check_symbol_range(const char *sym 120 static void check_symbol_range(const char *sym, unsigned long long addr,
103 struct addr_ran 121 struct addr_range *ranges, int entries)
104 { 122 {
105 size_t i; 123 size_t i;
106 struct addr_range *ar; 124 struct addr_range *ar;
107 125
108 for (i = 0; i < entries; ++i) { 126 for (i = 0; i < entries; ++i) {
109 ar = &ranges[i]; 127 ar = &ranges[i];
110 128
111 if (strcmp(sym, ar->start_sym) 129 if (strcmp(sym, ar->start_sym) == 0) {
112 ar->start = addr; 130 ar->start = addr;
113 return; 131 return;
114 } else if (strcmp(sym, ar->end 132 } else if (strcmp(sym, ar->end_sym) == 0) {
115 ar->end = addr; 133 ar->end = addr;
116 return; 134 return;
117 } 135 }
118 } 136 }
119 } 137 }
120 138
121 static struct sym_entry *read_symbol(FILE *in, !! 139 static struct sym_entry *read_symbol(FILE *in)
122 { 140 {
123 char *name, type, *p; !! 141 char name[KSYM_NAME_LEN_BUFFER+1], type;
124 unsigned long long addr; 142 unsigned long long addr;
125 size_t len; !! 143 unsigned int len;
126 ssize_t readlen; <<
127 struct sym_entry *sym; 144 struct sym_entry *sym;
>> 145 int rc;
128 146
129 errno = 0; !! 147 rc = fscanf(in, "%llx %c %" _stringify(KSYM_NAME_LEN_BUFFER) "s\n", &addr, &type, name);
130 readlen = getline(buf, buf_len, in); !! 148 if (rc != 3) {
131 if (readlen < 0) { !! 149 if (rc != EOF && fgets(name, ARRAY_SIZE(name), in) == NULL)
132 if (errno) { !! 150 fprintf(stderr, "Read error or end of file.\n");
133 perror("read_symbol"); <<
134 exit(EXIT_FAILURE); <<
135 } <<
136 return NULL; 151 return NULL;
137 } 152 }
138 !! 153 if (strlen(name) >= KSYM_NAME_LEN) {
139 if ((*buf)[readlen - 1] == '\n') <<
140 (*buf)[readlen - 1] = 0; <<
141 <<
142 addr = strtoull(*buf, &p, 16); <<
143 <<
144 if (*buf == p || *p++ != ' ' || !isasc <<
145 fprintf(stderr, "line format e <<
146 exit(EXIT_FAILURE); <<
147 } <<
148 <<
149 name = p; <<
150 len = strlen(name); <<
151 <<
152 if (len >= KSYM_NAME_LEN) { <<
153 fprintf(stderr, "Symbol %s too 154 fprintf(stderr, "Symbol %s too long for kallsyms (%zu >= %d).\n"
154 "Please increa 155 "Please increase KSYM_NAME_LEN both in kernel and kallsyms.c\n",
155 name, len, KSYM_NAME_L !! 156 name, strlen(name), KSYM_NAME_LEN);
156 return NULL; 157 return NULL;
157 } 158 }
158 159
159 if (strcmp(name, "_text") == 0) 160 if (strcmp(name, "_text") == 0)
160 _text = addr; 161 _text = addr;
161 162
162 /* Ignore most absolute/undefined (?) 163 /* Ignore most absolute/undefined (?) symbols. */
163 if (is_ignored_symbol(name, type)) 164 if (is_ignored_symbol(name, type))
164 return NULL; 165 return NULL;
165 166
166 check_symbol_range(name, addr, text_ra 167 check_symbol_range(name, addr, text_ranges, ARRAY_SIZE(text_ranges));
167 check_symbol_range(name, addr, &percpu 168 check_symbol_range(name, addr, &percpu_range, 1);
168 169
169 /* include the type field in the symbo 170 /* include the type field in the symbol name, so that it gets
170 * compressed together */ 171 * compressed together */
171 len++; <<
172 172
173 sym = xmalloc(sizeof(*sym) + len + 1); !! 173 len = strlen(name) + 1;
>> 174
>> 175 sym = malloc(sizeof(*sym) + len + 1);
>> 176 if (!sym) {
>> 177 fprintf(stderr, "kallsyms failure: "
>> 178 "unable to allocate required amount of memory\n");
>> 179 exit(EXIT_FAILURE);
>> 180 }
174 sym->addr = addr; 181 sym->addr = addr;
175 sym->len = len; 182 sym->len = len;
176 sym->sym[0] = type; 183 sym->sym[0] = type;
177 strcpy(sym_name(sym), name); 184 strcpy(sym_name(sym), name);
178 sym->percpu_absolute = false; !! 185 sym->percpu_absolute = 0;
179 186
180 return sym; 187 return sym;
181 } 188 }
182 189
183 static int symbol_in_range(const struct sym_en 190 static int symbol_in_range(const struct sym_entry *s,
184 const struct addr_r 191 const struct addr_range *ranges, int entries)
185 { 192 {
186 size_t i; 193 size_t i;
187 const struct addr_range *ar; 194 const struct addr_range *ar;
188 195
189 for (i = 0; i < entries; ++i) { 196 for (i = 0; i < entries; ++i) {
190 ar = &ranges[i]; 197 ar = &ranges[i];
191 198
192 if (s->addr >= ar->start && s- 199 if (s->addr >= ar->start && s->addr <= ar->end)
193 return 1; 200 return 1;
194 } 201 }
195 202
196 return 0; 203 return 0;
197 } 204 }
198 205
199 static bool string_starts_with(const char *s, <<
200 { <<
201 return strncmp(s, prefix, strlen(prefi <<
202 } <<
203 <<
204 static int symbol_valid(const struct sym_entry 206 static int symbol_valid(const struct sym_entry *s)
205 { 207 {
206 const char *name = sym_name(s); 208 const char *name = sym_name(s);
207 209
208 /* if --all-symbols is not specified, 210 /* if --all-symbols is not specified, then symbols outside the text
209 * and inittext sections are discarded 211 * and inittext sections are discarded */
210 if (!all_symbols) { 212 if (!all_symbols) {
211 /* <<
212 * Symbols starting with __sta <<
213 * section boundaries, and sho <<
214 */ <<
215 if (string_starts_with(name, " <<
216 string_starts_with(name, " <<
217 return 1; <<
218 <<
219 if (symbol_in_range(s, text_ra 213 if (symbol_in_range(s, text_ranges,
220 ARRAY_SIZE 214 ARRAY_SIZE(text_ranges)) == 0)
221 return 0; 215 return 0;
222 /* Corner case. Discard any s 216 /* Corner case. Discard any symbols with the same value as
223 * _etext _einittext; they can 217 * _etext _einittext; they can move between pass 1 and 2 when
224 * the kallsyms data are added 218 * the kallsyms data are added. If these symbols move then
225 * they may get dropped in pas 219 * they may get dropped in pass 2, which breaks the kallsyms
226 * rules. 220 * rules.
227 */ 221 */
228 if ((s->addr == text_range_tex 222 if ((s->addr == text_range_text->end &&
229 strcmp(name, text_range_t 223 strcmp(name, text_range_text->end_sym)) ||
230 (s->addr == text_range_ini 224 (s->addr == text_range_inittext->end &&
231 strcmp(name, text_range_i 225 strcmp(name, text_range_inittext->end_sym)))
232 return 0; 226 return 0;
233 } 227 }
234 228
235 return 1; 229 return 1;
236 } 230 }
237 231
238 /* remove all the invalid symbols from the tab 232 /* remove all the invalid symbols from the table */
239 static void shrink_table(void) 233 static void shrink_table(void)
240 { 234 {
241 unsigned int i, pos; 235 unsigned int i, pos;
242 236
243 pos = 0; 237 pos = 0;
244 for (i = 0; i < table_cnt; i++) { 238 for (i = 0; i < table_cnt; i++) {
245 if (symbol_valid(table[i])) { 239 if (symbol_valid(table[i])) {
246 if (pos != i) 240 if (pos != i)
247 table[pos] = t 241 table[pos] = table[i];
248 pos++; 242 pos++;
249 } else { 243 } else {
250 free(table[i]); 244 free(table[i]);
251 } 245 }
252 } 246 }
253 table_cnt = pos; 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 } 254 }
255 255
256 static void read_map(const char *in) 256 static void read_map(const char *in)
257 { 257 {
258 FILE *fp; 258 FILE *fp;
259 struct sym_entry *sym; 259 struct sym_entry *sym;
260 char *buf = NULL; <<
261 size_t buflen = 0; <<
262 260
263 fp = fopen(in, "r"); 261 fp = fopen(in, "r");
264 if (!fp) { 262 if (!fp) {
265 perror(in); 263 perror(in);
266 exit(1); 264 exit(1);
267 } 265 }
268 266
269 while (!feof(fp)) { 267 while (!feof(fp)) {
270 sym = read_symbol(fp, &buf, &b !! 268 sym = read_symbol(fp);
271 if (!sym) 269 if (!sym)
272 continue; 270 continue;
273 271
274 sym->seq = table_cnt; !! 272 sym->start_pos = table_cnt;
275 273
276 if (table_cnt >= table_size) { 274 if (table_cnt >= table_size) {
277 table_size += 10000; 275 table_size += 10000;
278 table = xrealloc(table !! 276 table = realloc(table, sizeof(*table) * table_size);
>> 277 if (!table) {
>> 278 fprintf(stderr, "out of memory\n");
>> 279 fclose(fp);
>> 280 exit (1);
>> 281 }
279 } 282 }
280 283
281 table[table_cnt++] = sym; 284 table[table_cnt++] = sym;
282 } 285 }
283 286
284 free(buf); <<
285 fclose(fp); 287 fclose(fp);
286 } 288 }
287 289
288 static void output_label(const char *label) 290 static void output_label(const char *label)
289 { 291 {
290 printf(".globl %s\n", label); 292 printf(".globl %s\n", label);
291 printf("\tALGN\n"); 293 printf("\tALGN\n");
292 printf("%s:\n", label); 294 printf("%s:\n", label);
293 } 295 }
294 296
>> 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
295 /* uncompress a compressed symbol. When this f 306 /* uncompress a compressed symbol. When this function is called, the best table
296 * might still be compressed itself, so the fu 307 * might still be compressed itself, so the function needs to be recursive */
297 static int expand_symbol(const unsigned char * 308 static int expand_symbol(const unsigned char *data, int len, char *result)
298 { 309 {
299 int c, rlen, total=0; 310 int c, rlen, total=0;
300 311
301 while (len) { 312 while (len) {
302 c = *data; 313 c = *data;
303 /* if the table holds a single 314 /* if the table holds a single char that is the same as the one
304 * we are looking for, then en 315 * we are looking for, then end the search */
305 if (best_table[c][0]==c && bes 316 if (best_table[c][0]==c && best_table_len[c]==1) {
306 *result++ = c; 317 *result++ = c;
307 total++; 318 total++;
308 } else { 319 } else {
309 /* if not, recurse and 320 /* if not, recurse and expand */
310 rlen = expand_symbol(b 321 rlen = expand_symbol(best_table[c], best_table_len[c], result);
311 total += rlen; 322 total += rlen;
312 result += rlen; 323 result += rlen;
313 } 324 }
314 data++; 325 data++;
315 len--; 326 len--;
316 } 327 }
317 *result=0; 328 *result=0;
318 329
319 return total; 330 return total;
320 } 331 }
321 332
322 static bool symbol_absolute(const struct sym_e !! 333 static int symbol_absolute(const struct sym_entry *s)
323 { 334 {
324 return s->percpu_absolute; 335 return s->percpu_absolute;
325 } 336 }
326 337
>> 338 static void cleanup_symbol_name(char *s)
>> 339 {
>> 340 char *p;
>> 341
>> 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 the first '.' in ".llvm." with '\0' does not
>> 350 * affect the main sorting, but it helps us with subsorting.
>> 351 */
>> 352 p = strstr(s, ".llvm.");
>> 353 if (p)
>> 354 *p = '\0';
>> 355 }
>> 356
327 static int compare_names(const void *a, const 357 static int compare_names(const void *a, const void *b)
328 { 358 {
329 int ret; 359 int ret;
330 const struct sym_entry *sa = *(const s 360 const struct sym_entry *sa = *(const struct sym_entry **)a;
331 const struct sym_entry *sb = *(const s 361 const struct sym_entry *sb = *(const struct sym_entry **)b;
332 362
333 ret = strcmp(sym_name(sa), sym_name(sb 363 ret = strcmp(sym_name(sa), sym_name(sb));
334 if (!ret) { 364 if (!ret) {
335 if (sa->addr > sb->addr) 365 if (sa->addr > sb->addr)
336 return 1; 366 return 1;
337 else if (sa->addr < sb->addr) 367 else if (sa->addr < sb->addr)
338 return -1; 368 return -1;
339 369
340 /* keep old order */ 370 /* keep old order */
341 return (int)(sa->seq - sb->seq 371 return (int)(sa->seq - sb->seq);
342 } 372 }
343 373
344 return ret; 374 return ret;
345 } 375 }
346 376
347 static void sort_symbols_by_name(void) 377 static void sort_symbols_by_name(void)
348 { 378 {
349 qsort(table, table_cnt, sizeof(table[0 379 qsort(table, table_cnt, sizeof(table[0]), compare_names);
350 } 380 }
351 381
352 static void write_src(void) 382 static void write_src(void)
353 { 383 {
354 unsigned int i, k, off; 384 unsigned int i, k, off;
355 unsigned int best_idx[256]; 385 unsigned int best_idx[256];
356 unsigned int *markers, markers_cnt; !! 386 unsigned int *markers;
357 char buf[KSYM_NAME_LEN]; 387 char buf[KSYM_NAME_LEN];
358 388
359 printf("#include <asm/bitsperlong.h>\n 389 printf("#include <asm/bitsperlong.h>\n");
360 printf("#if BITS_PER_LONG == 64\n"); 390 printf("#if BITS_PER_LONG == 64\n");
361 printf("#define PTR .quad\n"); 391 printf("#define PTR .quad\n");
362 printf("#define ALGN .balign 8\n"); 392 printf("#define ALGN .balign 8\n");
363 printf("#else\n"); 393 printf("#else\n");
364 printf("#define PTR .long\n"); 394 printf("#define PTR .long\n");
365 printf("#define ALGN .balign 4\n"); 395 printf("#define ALGN .balign 4\n");
366 printf("#endif\n"); 396 printf("#endif\n");
367 397
368 printf("\t.section .rodata, \"a\"\n"); 398 printf("\t.section .rodata, \"a\"\n");
369 399
370 output_label("kallsyms_num_syms"); 400 output_label("kallsyms_num_syms");
371 printf("\t.long\t%u\n", table_cnt); 401 printf("\t.long\t%u\n", table_cnt);
372 printf("\n"); 402 printf("\n");
373 403
374 /* table of offset markers, that give 404 /* table of offset markers, that give the offset in the compressed stream
375 * every 256 symbols */ 405 * every 256 symbols */
376 markers_cnt = (table_cnt + 255) / 256; !! 406 markers = malloc(sizeof(unsigned int) * ((table_cnt + 255) / 256));
377 markers = xmalloc(sizeof(*markers) * m !! 407 if (!markers) {
>> 408 fprintf(stderr, "kallsyms failure: "
>> 409 "unable to allocate required memory\n");
>> 410 exit(EXIT_FAILURE);
>> 411 }
378 412
379 output_label("kallsyms_names"); 413 output_label("kallsyms_names");
380 off = 0; 414 off = 0;
381 for (i = 0; i < table_cnt; i++) { 415 for (i = 0; i < table_cnt; i++) {
382 if ((i & 0xFF) == 0) 416 if ((i & 0xFF) == 0)
383 markers[i >> 8] = off; 417 markers[i >> 8] = off;
384 table[i]->seq = i; 418 table[i]->seq = i;
385 419
386 /* There cannot be any symbol 420 /* There cannot be any symbol of length zero. */
387 if (table[i]->len == 0) { 421 if (table[i]->len == 0) {
388 fprintf(stderr, "kalls 422 fprintf(stderr, "kallsyms failure: "
389 "unexpected ze 423 "unexpected zero symbol length\n");
390 exit(EXIT_FAILURE); 424 exit(EXIT_FAILURE);
391 } 425 }
392 426
393 /* Only lengths that fit in up 427 /* Only lengths that fit in up-to-two-byte ULEB128 are supported. */
394 if (table[i]->len > 0x3FFF) { 428 if (table[i]->len > 0x3FFF) {
395 fprintf(stderr, "kalls 429 fprintf(stderr, "kallsyms failure: "
396 "unexpected hu 430 "unexpected huge symbol length\n");
397 exit(EXIT_FAILURE); 431 exit(EXIT_FAILURE);
398 } 432 }
399 433
400 /* Encode length with ULEB128. 434 /* Encode length with ULEB128. */
401 if (table[i]->len <= 0x7F) { 435 if (table[i]->len <= 0x7F) {
402 /* Most symbols use a 436 /* Most symbols use a single byte for the length. */
403 printf("\t.byte 0x%02x 437 printf("\t.byte 0x%02x", table[i]->len);
404 off += table[i]->len + 438 off += table[i]->len + 1;
405 } else { 439 } else {
406 /* "Big" symbols use t 440 /* "Big" symbols use two bytes. */
407 printf("\t.byte 0x%02x 441 printf("\t.byte 0x%02x, 0x%02x",
408 (table[i]->len 442 (table[i]->len & 0x7F) | 0x80,
409 (table[i]->len 443 (table[i]->len >> 7) & 0x7F);
410 off += table[i]->len + 444 off += table[i]->len + 2;
411 } 445 }
412 for (k = 0; k < table[i]->len; 446 for (k = 0; k < table[i]->len; k++)
413 printf(", 0x%02x", tab 447 printf(", 0x%02x", table[i]->sym[k]);
>> 448 printf("\n");
>> 449 }
>> 450 printf("\n");
414 451
415 /* !! 452 /*
416 * Now that we wrote out the c !! 453 * Now that we wrote out the compressed symbol names, restore the
417 * original name and print it !! 454 * original names, which are needed in some of the later steps.
418 */ !! 455 */
>> 456 for (i = 0; i < table_cnt; i++) {
419 expand_symbol(table[i]->sym, t 457 expand_symbol(table[i]->sym, table[i]->len, buf);
420 strcpy((char *)table[i]->sym, 458 strcpy((char *)table[i]->sym, buf);
421 printf("\t/* %s */\n", table[i <<
422 } 459 }
423 printf("\n"); <<
424 460
425 output_label("kallsyms_markers"); 461 output_label("kallsyms_markers");
426 for (i = 0; i < markers_cnt; i++) !! 462 for (i = 0; i < ((table_cnt + 255) >> 8); i++)
427 printf("\t.long\t%u\n", marker 463 printf("\t.long\t%u\n", markers[i]);
428 printf("\n"); 464 printf("\n");
429 465
430 free(markers); 466 free(markers);
431 467
432 output_label("kallsyms_token_table"); 468 output_label("kallsyms_token_table");
433 off = 0; 469 off = 0;
434 for (i = 0; i < 256; i++) { 470 for (i = 0; i < 256; i++) {
435 best_idx[i] = off; 471 best_idx[i] = off;
436 expand_symbol(best_table[i], b 472 expand_symbol(best_table[i], best_table_len[i], buf);
437 printf("\t.asciz\t\"%s\"\n", b 473 printf("\t.asciz\t\"%s\"\n", buf);
438 off += strlen(buf) + 1; 474 off += strlen(buf) + 1;
439 } 475 }
440 printf("\n"); 476 printf("\n");
441 477
442 output_label("kallsyms_token_index"); 478 output_label("kallsyms_token_index");
443 for (i = 0; i < 256; i++) 479 for (i = 0; i < 256; i++)
444 printf("\t.short\t%d\n", best_ 480 printf("\t.short\t%d\n", best_idx[i]);
445 printf("\n"); 481 printf("\n");
446 482
447 output_label("kallsyms_offsets"); !! 483 if (!base_relative)
>> 484 output_label("kallsyms_addresses");
>> 485 else
>> 486 output_label("kallsyms_offsets");
448 487
449 for (i = 0; i < table_cnt; i++) { 488 for (i = 0; i < table_cnt; i++) {
450 /* !! 489 if (base_relative) {
451 * Use the offset relative to !! 490 /*
452 * encountered of all relative !! 491 * Use the offset relative to the lowest value
453 * non-relocatable fixed offse !! 492 * encountered of all relative symbols, and emit
454 * up at runtime. !! 493 * non-relocatable fixed offsets that will be fixed
455 */ !! 494 * up at runtime.
>> 495 */
456 496
457 long long offset; !! 497 long long offset;
458 bool overflow; !! 498 int overflow;
459 499
460 if (!absolute_percpu) { !! 500 if (!absolute_percpu) {
461 offset = table[i]->add !! 501 offset = table[i]->addr - relative_base;
462 overflow = offset < 0 !! 502 overflow = (offset < 0 || offset > UINT_MAX);
463 } else if (symbol_absolute(tab !! 503 } else if (symbol_absolute(table[i])) {
464 offset = table[i]->add !! 504 offset = table[i]->addr;
465 overflow = offset < 0 !! 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 }
>> 517 printf("\t.long\t%#x /* %s */\n", (int)offset, table[i]->sym);
>> 518 } else if (!symbol_absolute(table[i])) {
>> 519 output_address(table[i]->addr);
466 } else { 520 } else {
467 offset = relative_base !! 521 printf("\tPTR\t%#llx\n", table[i]->addr);
468 overflow = offset < IN <<
469 } 522 }
470 if (overflow) { <<
471 fprintf(stderr, "kalls <<
472 "%s symbol val <<
473 symbol_absolut <<
474 table[i]->addr <<
475 exit(EXIT_FAILURE); <<
476 } <<
477 printf("\t.long\t%#x\t/* %s */ <<
478 } 523 }
479 printf("\n"); 524 printf("\n");
480 525
481 output_label("kallsyms_relative_base") !! 526 if (base_relative) {
482 /* Provide proper symbols relocatabili !! 527 output_label("kallsyms_relative_base");
483 if (_text <= relative_base) !! 528 output_address(relative_base);
484 printf("\tPTR\t_text + %#llx\n !! 529 printf("\n");
485 else !! 530 }
486 printf("\tPTR\t_text - %#llx\n !! 531
487 printf("\n"); !! 532 if (lto_clang)
>> 533 for (i = 0; i < table_cnt; i++)
>> 534 cleanup_symbol_name((char *)table[i]->sym);
488 535
489 sort_symbols_by_name(); 536 sort_symbols_by_name();
490 output_label("kallsyms_seqs_of_names") 537 output_label("kallsyms_seqs_of_names");
491 for (i = 0; i < table_cnt; i++) 538 for (i = 0; i < table_cnt; i++)
492 printf("\t.byte 0x%02x, 0x%02x !! 539 printf("\t.byte 0x%02x, 0x%02x, 0x%02x\n",
493 (unsigned char)(table[ 540 (unsigned char)(table[i]->seq >> 16),
494 (unsigned char)(table[ 541 (unsigned char)(table[i]->seq >> 8),
495 (unsigned char)(table[ !! 542 (unsigned char)(table[i]->seq >> 0));
496 table[i]->sym); <<
497 printf("\n"); 543 printf("\n");
498 } 544 }
499 545
500 546
501 /* table lookup compression functions */ 547 /* table lookup compression functions */
502 548
503 /* count all the possible tokens in a symbol * 549 /* count all the possible tokens in a symbol */
504 static void learn_symbol(const unsigned char * 550 static void learn_symbol(const unsigned char *symbol, int len)
505 { 551 {
506 int i; 552 int i;
507 553
508 for (i = 0; i < len - 1; i++) 554 for (i = 0; i < len - 1; i++)
509 token_profit[ symbol[i] + (sym 555 token_profit[ symbol[i] + (symbol[i + 1] << 8) ]++;
510 } 556 }
511 557
512 /* decrease the count for all the possible tok 558 /* decrease the count for all the possible tokens in a symbol */
513 static void forget_symbol(const unsigned char 559 static void forget_symbol(const unsigned char *symbol, int len)
514 { 560 {
515 int i; 561 int i;
516 562
517 for (i = 0; i < len - 1; i++) 563 for (i = 0; i < len - 1; i++)
518 token_profit[ symbol[i] + (sym 564 token_profit[ symbol[i] + (symbol[i + 1] << 8) ]--;
519 } 565 }
520 566
521 /* do the initial token count */ 567 /* do the initial token count */
522 static void build_initial_token_table(void) 568 static void build_initial_token_table(void)
523 { 569 {
524 unsigned int i; 570 unsigned int i;
525 571
526 for (i = 0; i < table_cnt; i++) 572 for (i = 0; i < table_cnt; i++)
527 learn_symbol(table[i]->sym, ta 573 learn_symbol(table[i]->sym, table[i]->len);
528 } 574 }
529 575
530 static unsigned char *find_token(unsigned char 576 static unsigned char *find_token(unsigned char *str, int len,
531 const unsigne 577 const unsigned char *token)
532 { 578 {
533 int i; 579 int i;
534 580
535 for (i = 0; i < len - 1; i++) { 581 for (i = 0; i < len - 1; i++) {
536 if (str[i] == token[0] && str[ 582 if (str[i] == token[0] && str[i+1] == token[1])
537 return &str[i]; 583 return &str[i];
538 } 584 }
539 return NULL; 585 return NULL;
540 } 586 }
541 587
542 /* replace a given token in all the valid symb 588 /* replace a given token in all the valid symbols. Use the sampled symbols
543 * to update the counts */ 589 * to update the counts */
544 static void compress_symbols(const unsigned ch 590 static void compress_symbols(const unsigned char *str, int idx)
545 { 591 {
546 unsigned int i, len, size; 592 unsigned int i, len, size;
547 unsigned char *p1, *p2; 593 unsigned char *p1, *p2;
548 594
549 for (i = 0; i < table_cnt; i++) { 595 for (i = 0; i < table_cnt; i++) {
550 596
551 len = table[i]->len; 597 len = table[i]->len;
552 p1 = table[i]->sym; 598 p1 = table[i]->sym;
553 599
554 /* find the token on the symbo 600 /* find the token on the symbol */
555 p2 = find_token(p1, len, str); 601 p2 = find_token(p1, len, str);
556 if (!p2) continue; 602 if (!p2) continue;
557 603
558 /* decrease the counts for thi 604 /* decrease the counts for this symbol's tokens */
559 forget_symbol(table[i]->sym, l 605 forget_symbol(table[i]->sym, len);
560 606
561 size = len; 607 size = len;
562 608
563 do { 609 do {
564 *p2 = idx; 610 *p2 = idx;
565 p2++; 611 p2++;
566 size -= (p2 - p1); 612 size -= (p2 - p1);
567 memmove(p2, p2 + 1, si 613 memmove(p2, p2 + 1, size);
568 p1 = p2; 614 p1 = p2;
569 len--; 615 len--;
570 616
571 if (size < 2) break; 617 if (size < 2) break;
572 618
573 /* find the token on t 619 /* find the token on the symbol */
574 p2 = find_token(p1, si 620 p2 = find_token(p1, size, str);
575 621
576 } while (p2); 622 } while (p2);
577 623
578 table[i]->len = len; 624 table[i]->len = len;
579 625
580 /* increase the counts for thi 626 /* increase the counts for this symbol's new tokens */
581 learn_symbol(table[i]->sym, le 627 learn_symbol(table[i]->sym, len);
582 } 628 }
583 } 629 }
584 630
585 /* search the token with the maximum profit */ 631 /* search the token with the maximum profit */
586 static int find_best_token(void) 632 static int find_best_token(void)
587 { 633 {
588 int i, best, bestprofit; 634 int i, best, bestprofit;
589 635
590 bestprofit=-10000; 636 bestprofit=-10000;
591 best = 0; 637 best = 0;
592 638
593 for (i = 0; i < 0x10000; i++) { 639 for (i = 0; i < 0x10000; i++) {
594 if (token_profit[i] > bestprof 640 if (token_profit[i] > bestprofit) {
595 best = i; 641 best = i;
596 bestprofit = token_pro 642 bestprofit = token_profit[i];
597 } 643 }
598 } 644 }
599 return best; 645 return best;
600 } 646 }
601 647
602 /* this is the core of the algorithm: calculat 648 /* this is the core of the algorithm: calculate the "best" table */
603 static void optimize_result(void) 649 static void optimize_result(void)
604 { 650 {
605 int i, best; 651 int i, best;
606 652
607 /* using the '\0' symbol last allows c 653 /* using the '\0' symbol last allows compress_symbols to use standard
608 * fast string functions */ 654 * fast string functions */
609 for (i = 255; i >= 0; i--) { 655 for (i = 255; i >= 0; i--) {
610 656
611 /* if this table slot is empty 657 /* if this table slot is empty (it is not used by an actual
612 * original char code */ 658 * original char code */
613 if (!best_table_len[i]) { 659 if (!best_table_len[i]) {
614 660
615 /* find the token with 661 /* find the token with the best profit value */
616 best = find_best_token 662 best = find_best_token();
617 if (token_profit[best] 663 if (token_profit[best] == 0)
618 break; 664 break;
619 665
620 /* place it in the "be 666 /* place it in the "best" table */
621 best_table_len[i] = 2; 667 best_table_len[i] = 2;
622 best_table[i][0] = bes 668 best_table[i][0] = best & 0xFF;
623 best_table[i][1] = (be 669 best_table[i][1] = (best >> 8) & 0xFF;
624 670
625 /* replace this token 671 /* replace this token in all the valid symbols */
626 compress_symbols(best_ 672 compress_symbols(best_table[i], i);
627 } 673 }
628 } 674 }
629 } 675 }
630 676
631 /* start by placing the symbols that are actua 677 /* start by placing the symbols that are actually used on the table */
632 static void insert_real_symbols_in_table(void) 678 static void insert_real_symbols_in_table(void)
633 { 679 {
634 unsigned int i, j, c; 680 unsigned int i, j, c;
635 681
636 for (i = 0; i < table_cnt; i++) { 682 for (i = 0; i < table_cnt; i++) {
637 for (j = 0; j < table[i]->len; 683 for (j = 0; j < table[i]->len; j++) {
638 c = table[i]->sym[j]; 684 c = table[i]->sym[j];
639 best_table[c][0]=c; 685 best_table[c][0]=c;
640 best_table_len[c]=1; 686 best_table_len[c]=1;
641 } 687 }
642 } 688 }
643 } 689 }
644 690
645 static void optimize_token_table(void) 691 static void optimize_token_table(void)
646 { 692 {
647 build_initial_token_table(); 693 build_initial_token_table();
648 694
649 insert_real_symbols_in_table(); 695 insert_real_symbols_in_table();
650 696
651 optimize_result(); 697 optimize_result();
652 } 698 }
653 699
654 /* guess for "linker script provide" symbol */ 700 /* guess for "linker script provide" symbol */
655 static int may_be_linker_script_provide_symbol 701 static int may_be_linker_script_provide_symbol(const struct sym_entry *se)
656 { 702 {
657 const char *symbol = sym_name(se); 703 const char *symbol = sym_name(se);
658 int len = se->len - 1; 704 int len = se->len - 1;
659 705
660 if (len < 8) 706 if (len < 8)
661 return 0; 707 return 0;
662 708
663 if (symbol[0] != '_' || symbol[1] != ' 709 if (symbol[0] != '_' || symbol[1] != '_')
664 return 0; 710 return 0;
665 711
666 /* __start_XXXXX */ 712 /* __start_XXXXX */
667 if (!memcmp(symbol + 2, "start_", 6)) 713 if (!memcmp(symbol + 2, "start_", 6))
668 return 1; 714 return 1;
669 715
670 /* __stop_XXXXX */ 716 /* __stop_XXXXX */
671 if (!memcmp(symbol + 2, "stop_", 5)) 717 if (!memcmp(symbol + 2, "stop_", 5))
672 return 1; 718 return 1;
673 719
674 /* __end_XXXXX */ 720 /* __end_XXXXX */
675 if (!memcmp(symbol + 2, "end_", 4)) 721 if (!memcmp(symbol + 2, "end_", 4))
676 return 1; 722 return 1;
677 723
678 /* __XXXXX_start */ 724 /* __XXXXX_start */
679 if (!memcmp(symbol + len - 6, "_start" 725 if (!memcmp(symbol + len - 6, "_start", 6))
680 return 1; 726 return 1;
681 727
682 /* __XXXXX_end */ 728 /* __XXXXX_end */
683 if (!memcmp(symbol + len - 4, "_end", 729 if (!memcmp(symbol + len - 4, "_end", 4))
684 return 1; 730 return 1;
685 731
686 return 0; 732 return 0;
687 } 733 }
688 734
689 static int compare_symbols(const void *a, cons 735 static int compare_symbols(const void *a, const void *b)
690 { 736 {
691 const struct sym_entry *sa = *(const s 737 const struct sym_entry *sa = *(const struct sym_entry **)a;
692 const struct sym_entry *sb = *(const s 738 const struct sym_entry *sb = *(const struct sym_entry **)b;
693 int wa, wb; 739 int wa, wb;
694 740
695 /* sort by address first */ 741 /* sort by address first */
696 if (sa->addr > sb->addr) 742 if (sa->addr > sb->addr)
697 return 1; 743 return 1;
698 if (sa->addr < sb->addr) 744 if (sa->addr < sb->addr)
699 return -1; 745 return -1;
700 746
701 /* sort by "weakness" type */ 747 /* sort by "weakness" type */
702 wa = (sa->sym[0] == 'w') || (sa->sym[0 748 wa = (sa->sym[0] == 'w') || (sa->sym[0] == 'W');
703 wb = (sb->sym[0] == 'w') || (sb->sym[0 749 wb = (sb->sym[0] == 'w') || (sb->sym[0] == 'W');
704 if (wa != wb) 750 if (wa != wb)
705 return wa - wb; 751 return wa - wb;
706 752
707 /* sort by "linker script provide" typ 753 /* sort by "linker script provide" type */
708 wa = may_be_linker_script_provide_symb 754 wa = may_be_linker_script_provide_symbol(sa);
709 wb = may_be_linker_script_provide_symb 755 wb = may_be_linker_script_provide_symbol(sb);
710 if (wa != wb) 756 if (wa != wb)
711 return wa - wb; 757 return wa - wb;
712 758
713 /* sort by the number of prefix unders 759 /* sort by the number of prefix underscores */
714 wa = strspn(sym_name(sa), "_"); 760 wa = strspn(sym_name(sa), "_");
715 wb = strspn(sym_name(sb), "_"); 761 wb = strspn(sym_name(sb), "_");
716 if (wa != wb) 762 if (wa != wb)
717 return wa - wb; 763 return wa - wb;
718 764
719 /* sort by initial order, so that othe 765 /* sort by initial order, so that other symbols are left undisturbed */
720 return sa->seq - sb->seq; !! 766 return sa->start_pos - sb->start_pos;
721 } 767 }
722 768
723 static void sort_symbols(void) 769 static void sort_symbols(void)
724 { 770 {
725 qsort(table, table_cnt, sizeof(table[0 771 qsort(table, table_cnt, sizeof(table[0]), compare_symbols);
726 } 772 }
727 773
728 static void make_percpus_absolute(void) 774 static void make_percpus_absolute(void)
729 { 775 {
730 unsigned int i; 776 unsigned int i;
731 777
732 for (i = 0; i < table_cnt; i++) 778 for (i = 0; i < table_cnt; i++)
733 if (symbol_in_range(table[i], 779 if (symbol_in_range(table[i], &percpu_range, 1)) {
734 /* 780 /*
735 * Keep the 'A' overri 781 * Keep the 'A' override for percpu symbols to
736 * ensure consistent b 782 * ensure consistent behavior compared to older
737 * versions of this to 783 * versions of this tool.
738 */ 784 */
739 table[i]->sym[0] = 'A' 785 table[i]->sym[0] = 'A';
740 table[i]->percpu_absol !! 786 table[i]->percpu_absolute = 1;
741 } 787 }
742 } 788 }
743 789
744 /* find the minimum non-absolute symbol addres 790 /* find the minimum non-absolute symbol address */
745 static void record_relative_base(void) 791 static void record_relative_base(void)
746 { 792 {
747 unsigned int i; 793 unsigned int i;
748 794
749 for (i = 0; i < table_cnt; i++) 795 for (i = 0; i < table_cnt; i++)
750 if (!symbol_absolute(table[i]) 796 if (!symbol_absolute(table[i])) {
751 /* 797 /*
752 * The table is sorted 798 * The table is sorted by address.
753 * Take the first non- 799 * Take the first non-absolute symbol value.
754 */ 800 */
755 relative_base = table[ 801 relative_base = table[i]->addr;
756 return; 802 return;
757 } 803 }
758 } 804 }
759 805
760 int main(int argc, char **argv) 806 int main(int argc, char **argv)
761 { 807 {
762 while (1) { 808 while (1) {
763 static const struct option lon !! 809 static struct option long_options[] = {
764 {"all-symbols", no 810 {"all-symbols", no_argument, &all_symbols, 1},
765 {"absolute-percpu", no 811 {"absolute-percpu", no_argument, &absolute_percpu, 1},
>> 812 {"base-relative", no_argument, &base_relative, 1},
>> 813 {"lto-clang", no_argument, <o_clang, 1},
766 {}, 814 {},
767 }; 815 };
768 816
769 int c = getopt_long(argc, argv 817 int c = getopt_long(argc, argv, "", long_options, NULL);
770 818
771 if (c == -1) 819 if (c == -1)
772 break; 820 break;
773 if (c != 0) 821 if (c != 0)
774 usage(); 822 usage();
775 } 823 }
776 824
777 if (optind >= argc) 825 if (optind >= argc)
778 usage(); 826 usage();
779 827
780 read_map(argv[optind]); 828 read_map(argv[optind]);
781 shrink_table(); 829 shrink_table();
782 if (absolute_percpu) 830 if (absolute_percpu)
783 make_percpus_absolute(); 831 make_percpus_absolute();
784 sort_symbols(); 832 sort_symbols();
785 record_relative_base(); !! 833 if (base_relative)
>> 834 record_relative_base();
786 optimize_token_table(); 835 optimize_token_table();
787 write_src(); 836 write_src();
788 837
789 return 0; 838 return 0;
790 } 839 }
791 840
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.