Loading...
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#include <limits.h>
26
27#ifndef ARRAY_SIZE
28#define ARRAY_SIZE(arr) (sizeof(arr) / sizeof(arr[0]))
29#endif
30
31#define KSYM_NAME_LEN 128
32
33struct sym_entry {
34 unsigned long long addr;
35 unsigned int len;
36 unsigned int start_pos;
37 unsigned char *sym;
38 unsigned int percpu_absolute;
39};
40
41struct addr_range {
42 const char *start_sym, *end_sym;
43 unsigned long long start, end;
44};
45
46static unsigned long long _text;
47static unsigned long long relative_base;
48static struct addr_range text_ranges[] = {
49 { "_stext", "_etext" },
50 { "_sinittext", "_einittext" },
51 { "_stext_l1", "_etext_l1" }, /* Blackfin on-chip L1 inst SRAM */
52 { "_stext_l2", "_etext_l2" }, /* Blackfin on-chip L2 SRAM */
53};
54#define text_range_text (&text_ranges[0])
55#define text_range_inittext (&text_ranges[1])
56
57static struct addr_range percpu_range = {
58 "__per_cpu_start", "__per_cpu_end", -1ULL, 0
59};
60
61static struct sym_entry *table;
62static unsigned int table_size, table_cnt;
63static int all_symbols = 0;
64static int absolute_percpu = 0;
65static char symbol_prefix_char = '\0';
66static unsigned long long kernel_start_addr = 0;
67static int base_relative = 0;
68
69int token_profit[0x10000];
70
71/* the table that holds the result of the compression */
72unsigned char best_table[256][2];
73unsigned char best_table_len[256];
74
75
76static void usage(void)
77{
78 fprintf(stderr, "Usage: kallsyms [--all-symbols] "
79 "[--symbol-prefix=<prefix char>] "
80 "[--page-offset=<CONFIG_PAGE_OFFSET>] "
81 "[--base-relative] < in.map > out.S\n");
82 exit(1);
83}
84
85/*
86 * This ignores the intensely annoying "mapping symbols" found
87 * in ARM ELF files: $a, $t and $d.
88 */
89static inline int is_arm_mapping_symbol(const char *str)
90{
91 return str[0] == '$' && strchr("axtd", str[1])
92 && (str[2] == '\0' || str[2] == '.');
93}
94
95static int check_symbol_range(const char *sym, unsigned long long addr,
96 struct addr_range *ranges, int entries)
97{
98 size_t i;
99 struct addr_range *ar;
100
101 for (i = 0; i < entries; ++i) {
102 ar = &ranges[i];
103
104 if (strcmp(sym, ar->start_sym) == 0) {
105 ar->start = addr;
106 return 0;
107 } else if (strcmp(sym, ar->end_sym) == 0) {
108 ar->end = addr;
109 return 0;
110 }
111 }
112
113 return 1;
114}
115
116static int read_symbol(FILE *in, struct sym_entry *s)
117{
118 char str[500];
119 char *sym, stype;
120 int rc;
121
122 rc = fscanf(in, "%llx %c %499s\n", &s->addr, &stype, str);
123 if (rc != 3) {
124 if (rc != EOF && fgets(str, 500, in) == NULL)
125 fprintf(stderr, "Read error or end of file.\n");
126 return -1;
127 }
128 if (strlen(str) > KSYM_NAME_LEN) {
129 fprintf(stderr, "Symbol %s too long for kallsyms (%zu vs %d).\n"
130 "Please increase KSYM_NAME_LEN both in kernel and kallsyms.c\n",
131 str, strlen(str), KSYM_NAME_LEN);
132 return -1;
133 }
134
135 sym = str;
136 /* skip prefix char */
137 if (symbol_prefix_char && str[0] == symbol_prefix_char)
138 sym++;
139
140 /* Ignore most absolute/undefined (?) symbols. */
141 if (strcmp(sym, "_text") == 0)
142 _text = s->addr;
143 else if (check_symbol_range(sym, s->addr, text_ranges,
144 ARRAY_SIZE(text_ranges)) == 0)
145 /* nothing to do */;
146 else if (toupper(stype) == 'A')
147 {
148 /* Keep these useful absolute symbols */
149 if (strcmp(sym, "__kernel_syscall_via_break") &&
150 strcmp(sym, "__kernel_syscall_via_epc") &&
151 strcmp(sym, "__kernel_sigtramp") &&
152 strcmp(sym, "__gp"))
153 return -1;
154
155 }
156 else if (toupper(stype) == 'U' ||
157 is_arm_mapping_symbol(sym))
158 return -1;
159 /* exclude also MIPS ELF local symbols ($L123 instead of .L123) */
160 else if (str[0] == '$')
161 return -1;
162 /* exclude debugging symbols */
163 else if (stype == 'N')
164 return -1;
165
166 /* include the type field in the symbol name, so that it gets
167 * compressed together */
168 s->len = strlen(str) + 1;
169 s->sym = malloc(s->len + 1);
170 if (!s->sym) {
171 fprintf(stderr, "kallsyms failure: "
172 "unable to allocate required amount of memory\n");
173 exit(EXIT_FAILURE);
174 }
175 strcpy((char *)s->sym + 1, str);
176 s->sym[0] = stype;
177
178 s->percpu_absolute = 0;
179
180 /* Record if we've found __per_cpu_start/end. */
181 check_symbol_range(sym, s->addr, &percpu_range, 1);
182
183 return 0;
184}
185
186static int symbol_in_range(struct sym_entry *s, struct addr_range *ranges,
187 int entries)
188{
189 size_t i;
190 struct addr_range *ar;
191
192 for (i = 0; i < entries; ++i) {
193 ar = &ranges[i];
194
195 if (s->addr >= ar->start && s->addr <= ar->end)
196 return 1;
197 }
198
199 return 0;
200}
201
202static int symbol_valid(struct sym_entry *s)
203{
204 /* Symbols which vary between passes. Passes 1 and 2 must have
205 * identical symbol lists. The kallsyms_* symbols below are only added
206 * after pass 1, they would be included in pass 2 when --all-symbols is
207 * specified so exclude them to get a stable symbol list.
208 */
209 static char *special_symbols[] = {
210 "kallsyms_addresses",
211 "kallsyms_offsets",
212 "kallsyms_relative_base",
213 "kallsyms_num_syms",
214 "kallsyms_names",
215 "kallsyms_markers",
216 "kallsyms_token_table",
217 "kallsyms_token_index",
218
219 /* Exclude linker generated symbols which vary between passes */
220 "_SDA_BASE_", /* ppc */
221 "_SDA2_BASE_", /* ppc */
222 NULL };
223
224 static char *special_suffixes[] = {
225 "_veneer", /* arm */
226 NULL };
227
228 int i;
229 char *sym_name = (char *)s->sym + 1;
230
231
232 if (s->addr < kernel_start_addr)
233 return 0;
234
235 /* skip prefix char */
236 if (symbol_prefix_char && *sym_name == symbol_prefix_char)
237 sym_name++;
238
239
240 /* if --all-symbols is not specified, then symbols outside the text
241 * and inittext sections are discarded */
242 if (!all_symbols) {
243 if (symbol_in_range(s, text_ranges,
244 ARRAY_SIZE(text_ranges)) == 0)
245 return 0;
246 /* Corner case. Discard any symbols with the same value as
247 * _etext _einittext; they can move between pass 1 and 2 when
248 * the kallsyms data are added. If these symbols move then
249 * they may get dropped in pass 2, which breaks the kallsyms
250 * rules.
251 */
252 if ((s->addr == text_range_text->end &&
253 strcmp(sym_name,
254 text_range_text->end_sym)) ||
255 (s->addr == text_range_inittext->end &&
256 strcmp(sym_name,
257 text_range_inittext->end_sym)))
258 return 0;
259 }
260
261 /* Exclude symbols which vary between passes. */
262 for (i = 0; special_symbols[i]; i++)
263 if (strcmp(sym_name, special_symbols[i]) == 0)
264 return 0;
265
266 for (i = 0; special_suffixes[i]; i++) {
267 int l = strlen(sym_name) - strlen(special_suffixes[i]);
268
269 if (l >= 0 && strcmp(sym_name + l, special_suffixes[i]) == 0)
270 return 0;
271 }
272
273 return 1;
274}
275
276static void read_map(FILE *in)
277{
278 while (!feof(in)) {
279 if (table_cnt >= table_size) {
280 table_size += 10000;
281 table = realloc(table, sizeof(*table) * table_size);
282 if (!table) {
283 fprintf(stderr, "out of memory\n");
284 exit (1);
285 }
286 }
287 if (read_symbol(in, &table[table_cnt]) == 0) {
288 table[table_cnt].start_pos = table_cnt;
289 table_cnt++;
290 }
291 }
292}
293
294static void output_label(char *label)
295{
296 if (symbol_prefix_char)
297 printf(".globl %c%s\n", symbol_prefix_char, label);
298 else
299 printf(".globl %s\n", label);
300 printf("\tALGN\n");
301 if (symbol_prefix_char)
302 printf("%c%s:\n", symbol_prefix_char, label);
303 else
304 printf("%s:\n", label);
305}
306
307/* uncompress a compressed symbol. When this function is called, the best table
308 * might still be compressed itself, so the function needs to be recursive */
309static int expand_symbol(unsigned char *data, int len, char *result)
310{
311 int c, rlen, total=0;
312
313 while (len) {
314 c = *data;
315 /* if the table holds a single char that is the same as the one
316 * we are looking for, then end the search */
317 if (best_table[c][0]==c && best_table_len[c]==1) {
318 *result++ = c;
319 total++;
320 } else {
321 /* if not, recurse and expand */
322 rlen = expand_symbol(best_table[c], best_table_len[c], result);
323 total += rlen;
324 result += rlen;
325 }
326 data++;
327 len--;
328 }
329 *result=0;
330
331 return total;
332}
333
334static int symbol_absolute(struct sym_entry *s)
335{
336 return s->percpu_absolute;
337}
338
339static void write_src(void)
340{
341 unsigned int i, k, off;
342 unsigned int best_idx[256];
343 unsigned int *markers;
344 char buf[KSYM_NAME_LEN];
345
346 printf("#include <asm/types.h>\n");
347 printf("#if BITS_PER_LONG == 64\n");
348 printf("#define PTR .quad\n");
349 printf("#define ALGN .align 8\n");
350 printf("#else\n");
351 printf("#define PTR .long\n");
352 printf("#define ALGN .align 4\n");
353 printf("#endif\n");
354
355 printf("\t.section .rodata, \"a\"\n");
356
357 /* Provide proper symbols relocatability by their relativeness
358 * to a fixed anchor point in the runtime image, either '_text'
359 * for absolute address tables, in which case the linker will
360 * emit the final addresses at build time. Otherwise, use the
361 * offset relative to the lowest value encountered of all relative
362 * symbols, and emit non-relocatable fixed offsets that will be fixed
363 * up at runtime.
364 *
365 * The symbol names cannot be used to construct normal symbol
366 * references as the list of symbols contains symbols that are
367 * declared static and are private to their .o files. This prevents
368 * .tmp_kallsyms.o or any other object from referencing them.
369 */
370 if (!base_relative)
371 output_label("kallsyms_addresses");
372 else
373 output_label("kallsyms_offsets");
374
375 for (i = 0; i < table_cnt; i++) {
376 if (base_relative) {
377 long long offset;
378 int overflow;
379
380 if (!absolute_percpu) {
381 offset = table[i].addr - relative_base;
382 overflow = (offset < 0 || offset > UINT_MAX);
383 } else if (symbol_absolute(&table[i])) {
384 offset = table[i].addr;
385 overflow = (offset < 0 || offset > INT_MAX);
386 } else {
387 offset = relative_base - table[i].addr - 1;
388 overflow = (offset < INT_MIN || offset >= 0);
389 }
390 if (overflow) {
391 fprintf(stderr, "kallsyms failure: "
392 "%s symbol value %#llx out of range in relative mode\n",
393 symbol_absolute(&table[i]) ? "absolute" : "relative",
394 table[i].addr);
395 exit(EXIT_FAILURE);
396 }
397 printf("\t.long\t%#x\n", (int)offset);
398 } else if (!symbol_absolute(&table[i])) {
399 if (_text <= table[i].addr)
400 printf("\tPTR\t_text + %#llx\n",
401 table[i].addr - _text);
402 else
403 printf("\tPTR\t_text - %#llx\n",
404 _text - table[i].addr);
405 } else {
406 printf("\tPTR\t%#llx\n", table[i].addr);
407 }
408 }
409 printf("\n");
410
411 if (base_relative) {
412 output_label("kallsyms_relative_base");
413 printf("\tPTR\t_text - %#llx\n", _text - relative_base);
414 printf("\n");
415 }
416
417 output_label("kallsyms_num_syms");
418 printf("\tPTR\t%d\n", table_cnt);
419 printf("\n");
420
421 /* table of offset markers, that give the offset in the compressed stream
422 * every 256 symbols */
423 markers = malloc(sizeof(unsigned int) * ((table_cnt + 255) / 256));
424 if (!markers) {
425 fprintf(stderr, "kallsyms failure: "
426 "unable to allocate required memory\n");
427 exit(EXIT_FAILURE);
428 }
429
430 output_label("kallsyms_names");
431 off = 0;
432 for (i = 0; i < table_cnt; i++) {
433 if ((i & 0xFF) == 0)
434 markers[i >> 8] = off;
435
436 printf("\t.byte 0x%02x", table[i].len);
437 for (k = 0; k < table[i].len; k++)
438 printf(", 0x%02x", table[i].sym[k]);
439 printf("\n");
440
441 off += table[i].len + 1;
442 }
443 printf("\n");
444
445 output_label("kallsyms_markers");
446 for (i = 0; i < ((table_cnt + 255) >> 8); i++)
447 printf("\tPTR\t%d\n", markers[i]);
448 printf("\n");
449
450 free(markers);
451
452 output_label("kallsyms_token_table");
453 off = 0;
454 for (i = 0; i < 256; i++) {
455 best_idx[i] = off;
456 expand_symbol(best_table[i], best_table_len[i], buf);
457 printf("\t.asciz\t\"%s\"\n", buf);
458 off += strlen(buf) + 1;
459 }
460 printf("\n");
461
462 output_label("kallsyms_token_index");
463 for (i = 0; i < 256; i++)
464 printf("\t.short\t%d\n", best_idx[i]);
465 printf("\n");
466}
467
468
469/* table lookup compression functions */
470
471/* count all the possible tokens in a symbol */
472static void learn_symbol(unsigned char *symbol, int len)
473{
474 int i;
475
476 for (i = 0; i < len - 1; i++)
477 token_profit[ symbol[i] + (symbol[i + 1] << 8) ]++;
478}
479
480/* decrease the count for all the possible tokens in a symbol */
481static void forget_symbol(unsigned char *symbol, int len)
482{
483 int i;
484
485 for (i = 0; i < len - 1; i++)
486 token_profit[ symbol[i] + (symbol[i + 1] << 8) ]--;
487}
488
489/* remove all the invalid symbols from the table and do the initial token count */
490static void build_initial_tok_table(void)
491{
492 unsigned int i, pos;
493
494 pos = 0;
495 for (i = 0; i < table_cnt; i++) {
496 if ( symbol_valid(&table[i]) ) {
497 if (pos != i)
498 table[pos] = table[i];
499 learn_symbol(table[pos].sym, table[pos].len);
500 pos++;
501 }
502 }
503 table_cnt = pos;
504}
505
506static void *find_token(unsigned char *str, int len, unsigned char *token)
507{
508 int i;
509
510 for (i = 0; i < len - 1; i++) {
511 if (str[i] == token[0] && str[i+1] == token[1])
512 return &str[i];
513 }
514 return NULL;
515}
516
517/* replace a given token in all the valid symbols. Use the sampled symbols
518 * to update the counts */
519static void compress_symbols(unsigned char *str, int idx)
520{
521 unsigned int i, len, size;
522 unsigned char *p1, *p2;
523
524 for (i = 0; i < table_cnt; i++) {
525
526 len = table[i].len;
527 p1 = table[i].sym;
528
529 /* find the token on the symbol */
530 p2 = find_token(p1, len, str);
531 if (!p2) continue;
532
533 /* decrease the counts for this symbol's tokens */
534 forget_symbol(table[i].sym, len);
535
536 size = len;
537
538 do {
539 *p2 = idx;
540 p2++;
541 size -= (p2 - p1);
542 memmove(p2, p2 + 1, size);
543 p1 = p2;
544 len--;
545
546 if (size < 2) break;
547
548 /* find the token on the symbol */
549 p2 = find_token(p1, size, str);
550
551 } while (p2);
552
553 table[i].len = len;
554
555 /* increase the counts for this symbol's new tokens */
556 learn_symbol(table[i].sym, len);
557 }
558}
559
560/* search the token with the maximum profit */
561static int find_best_token(void)
562{
563 int i, best, bestprofit;
564
565 bestprofit=-10000;
566 best = 0;
567
568 for (i = 0; i < 0x10000; i++) {
569 if (token_profit[i] > bestprofit) {
570 best = i;
571 bestprofit = token_profit[i];
572 }
573 }
574 return best;
575}
576
577/* this is the core of the algorithm: calculate the "best" table */
578static void optimize_result(void)
579{
580 int i, best;
581
582 /* using the '\0' symbol last allows compress_symbols to use standard
583 * fast string functions */
584 for (i = 255; i >= 0; i--) {
585
586 /* if this table slot is empty (it is not used by an actual
587 * original char code */
588 if (!best_table_len[i]) {
589
590 /* find the token with the breates profit value */
591 best = find_best_token();
592 if (token_profit[best] == 0)
593 break;
594
595 /* place it in the "best" table */
596 best_table_len[i] = 2;
597 best_table[i][0] = best & 0xFF;
598 best_table[i][1] = (best >> 8) & 0xFF;
599
600 /* replace this token in all the valid symbols */
601 compress_symbols(best_table[i], i);
602 }
603 }
604}
605
606/* start by placing the symbols that are actually used on the table */
607static void insert_real_symbols_in_table(void)
608{
609 unsigned int i, j, c;
610
611 memset(best_table, 0, sizeof(best_table));
612 memset(best_table_len, 0, sizeof(best_table_len));
613
614 for (i = 0; i < table_cnt; i++) {
615 for (j = 0; j < table[i].len; j++) {
616 c = table[i].sym[j];
617 best_table[c][0]=c;
618 best_table_len[c]=1;
619 }
620 }
621}
622
623static void optimize_token_table(void)
624{
625 build_initial_tok_table();
626
627 insert_real_symbols_in_table();
628
629 /* When valid symbol is not registered, exit to error */
630 if (!table_cnt) {
631 fprintf(stderr, "No valid symbol.\n");
632 exit(1);
633 }
634
635 optimize_result();
636}
637
638/* guess for "linker script provide" symbol */
639static int may_be_linker_script_provide_symbol(const struct sym_entry *se)
640{
641 const char *symbol = (char *)se->sym + 1;
642 int len = se->len - 1;
643
644 if (len < 8)
645 return 0;
646
647 if (symbol[0] != '_' || symbol[1] != '_')
648 return 0;
649
650 /* __start_XXXXX */
651 if (!memcmp(symbol + 2, "start_", 6))
652 return 1;
653
654 /* __stop_XXXXX */
655 if (!memcmp(symbol + 2, "stop_", 5))
656 return 1;
657
658 /* __end_XXXXX */
659 if (!memcmp(symbol + 2, "end_", 4))
660 return 1;
661
662 /* __XXXXX_start */
663 if (!memcmp(symbol + len - 6, "_start", 6))
664 return 1;
665
666 /* __XXXXX_end */
667 if (!memcmp(symbol + len - 4, "_end", 4))
668 return 1;
669
670 return 0;
671}
672
673static int prefix_underscores_count(const char *str)
674{
675 const char *tail = str;
676
677 while (*tail == '_')
678 tail++;
679
680 return tail - str;
681}
682
683static int compare_symbols(const void *a, const void *b)
684{
685 const struct sym_entry *sa;
686 const struct sym_entry *sb;
687 int wa, wb;
688
689 sa = a;
690 sb = b;
691
692 /* sort by address first */
693 if (sa->addr > sb->addr)
694 return 1;
695 if (sa->addr < sb->addr)
696 return -1;
697
698 /* sort by "weakness" type */
699 wa = (sa->sym[0] == 'w') || (sa->sym[0] == 'W');
700 wb = (sb->sym[0] == 'w') || (sb->sym[0] == 'W');
701 if (wa != wb)
702 return wa - wb;
703
704 /* sort by "linker script provide" type */
705 wa = may_be_linker_script_provide_symbol(sa);
706 wb = may_be_linker_script_provide_symbol(sb);
707 if (wa != wb)
708 return wa - wb;
709
710 /* sort by the number of prefix underscores */
711 wa = prefix_underscores_count((const char *)sa->sym + 1);
712 wb = prefix_underscores_count((const char *)sb->sym + 1);
713 if (wa != wb)
714 return wa - wb;
715
716 /* sort by initial order, so that other symbols are left undisturbed */
717 return sa->start_pos - sb->start_pos;
718}
719
720static void sort_symbols(void)
721{
722 qsort(table, table_cnt, sizeof(struct sym_entry), compare_symbols);
723}
724
725static void make_percpus_absolute(void)
726{
727 unsigned int i;
728
729 for (i = 0; i < table_cnt; i++)
730 if (symbol_in_range(&table[i], &percpu_range, 1)) {
731 /*
732 * Keep the 'A' override for percpu symbols to
733 * ensure consistent behavior compared to older
734 * versions of this tool.
735 */
736 table[i].sym[0] = 'A';
737 table[i].percpu_absolute = 1;
738 }
739}
740
741/* find the minimum non-absolute symbol address */
742static void record_relative_base(void)
743{
744 unsigned int i;
745
746 relative_base = -1ULL;
747 for (i = 0; i < table_cnt; i++)
748 if (!symbol_absolute(&table[i]) &&
749 table[i].addr < relative_base)
750 relative_base = table[i].addr;
751}
752
753int main(int argc, char **argv)
754{
755 if (argc >= 2) {
756 int i;
757 for (i = 1; i < argc; i++) {
758 if(strcmp(argv[i], "--all-symbols") == 0)
759 all_symbols = 1;
760 else if (strcmp(argv[i], "--absolute-percpu") == 0)
761 absolute_percpu = 1;
762 else if (strncmp(argv[i], "--symbol-prefix=", 16) == 0) {
763 char *p = &argv[i][16];
764 /* skip quote */
765 if ((*p == '"' && *(p+2) == '"') || (*p == '\'' && *(p+2) == '\''))
766 p++;
767 symbol_prefix_char = *p;
768 } else if (strncmp(argv[i], "--page-offset=", 14) == 0) {
769 const char *p = &argv[i][14];
770 kernel_start_addr = strtoull(p, NULL, 16);
771 } else if (strcmp(argv[i], "--base-relative") == 0)
772 base_relative = 1;
773 else
774 usage();
775 }
776 } else if (argc != 1)
777 usage();
778
779 read_map(stdin);
780 if (absolute_percpu)
781 make_percpus_absolute();
782 if (base_relative)
783 record_relative_base();
784 sort_symbols();
785 optimize_token_table();
786 write_src();
787
788 return 0;
789}
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: kallsyms [--all-symbols] [--absolute-percpu] in.map > out.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 <errno.h>
22#include <getopt.h>
23#include <stdbool.h>
24#include <stdio.h>
25#include <stdlib.h>
26#include <string.h>
27#include <ctype.h>
28#include <limits.h>
29
30#include <xalloc.h>
31
32#define ARRAY_SIZE(arr) (sizeof(arr) / sizeof(arr[0]))
33
34#define KSYM_NAME_LEN 512
35
36struct sym_entry {
37 unsigned long long addr;
38 unsigned int len;
39 unsigned int seq;
40 bool percpu_absolute;
41 unsigned char sym[];
42};
43
44struct addr_range {
45 const char *start_sym, *end_sym;
46 unsigned long long start, end;
47};
48
49static unsigned long long _text;
50static unsigned long long relative_base;
51static struct addr_range text_ranges[] = {
52 { "_stext", "_etext" },
53 { "_sinittext", "_einittext" },
54};
55#define text_range_text (&text_ranges[0])
56#define text_range_inittext (&text_ranges[1])
57
58static struct addr_range percpu_range = {
59 "__per_cpu_start", "__per_cpu_end", -1ULL, 0
60};
61
62static struct sym_entry **table;
63static unsigned int table_size, table_cnt;
64static int all_symbols;
65static int absolute_percpu;
66
67static int token_profit[0x10000];
68
69/* the table that holds the result of the compression */
70static unsigned char best_table[256][2];
71static unsigned char best_table_len[256];
72
73
74static void usage(void)
75{
76 fprintf(stderr, "Usage: kallsyms [--all-symbols] [--absolute-percpu] in.map > out.S\n");
77 exit(1);
78}
79
80static char *sym_name(const struct sym_entry *s)
81{
82 return (char *)s->sym + 1;
83}
84
85static bool is_ignored_symbol(const char *name, char type)
86{
87 if (type == 'u' || type == 'n')
88 return true;
89
90 if (toupper(type) == 'A') {
91 /* Keep these useful absolute symbols */
92 if (strcmp(name, "__kernel_syscall_via_break") &&
93 strcmp(name, "__kernel_syscall_via_epc") &&
94 strcmp(name, "__kernel_sigtramp") &&
95 strcmp(name, "__gp"))
96 return true;
97 }
98
99 return false;
100}
101
102static void check_symbol_range(const char *sym, unsigned long long addr,
103 struct addr_range *ranges, int entries)
104{
105 size_t i;
106 struct addr_range *ar;
107
108 for (i = 0; i < entries; ++i) {
109 ar = &ranges[i];
110
111 if (strcmp(sym, ar->start_sym) == 0) {
112 ar->start = addr;
113 return;
114 } else if (strcmp(sym, ar->end_sym) == 0) {
115 ar->end = addr;
116 return;
117 }
118 }
119}
120
121static struct sym_entry *read_symbol(FILE *in, char **buf, size_t *buf_len)
122{
123 char *name, type, *p;
124 unsigned long long addr;
125 size_t len;
126 ssize_t readlen;
127 struct sym_entry *sym;
128
129 errno = 0;
130 readlen = getline(buf, buf_len, in);
131 if (readlen < 0) {
132 if (errno) {
133 perror("read_symbol");
134 exit(EXIT_FAILURE);
135 }
136 return NULL;
137 }
138
139 if ((*buf)[readlen - 1] == '\n')
140 (*buf)[readlen - 1] = 0;
141
142 addr = strtoull(*buf, &p, 16);
143
144 if (*buf == p || *p++ != ' ' || !isascii((type = *p++)) || *p++ != ' ') {
145 fprintf(stderr, "line format error\n");
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 long for kallsyms (%zu >= %d).\n"
154 "Please increase KSYM_NAME_LEN both in kernel and kallsyms.c\n",
155 name, len, KSYM_NAME_LEN);
156 return NULL;
157 }
158
159 if (strcmp(name, "_text") == 0)
160 _text = addr;
161
162 /* Ignore most absolute/undefined (?) symbols. */
163 if (is_ignored_symbol(name, type))
164 return NULL;
165
166 check_symbol_range(name, addr, text_ranges, ARRAY_SIZE(text_ranges));
167 check_symbol_range(name, addr, &percpu_range, 1);
168
169 /* include the type field in the symbol name, so that it gets
170 * compressed together */
171 len++;
172
173 sym = xmalloc(sizeof(*sym) + len + 1);
174 sym->addr = addr;
175 sym->len = len;
176 sym->sym[0] = type;
177 strcpy(sym_name(sym), name);
178 sym->percpu_absolute = false;
179
180 return sym;
181}
182
183static int symbol_in_range(const struct sym_entry *s,
184 const struct addr_range *ranges, int entries)
185{
186 size_t i;
187 const struct addr_range *ar;
188
189 for (i = 0; i < entries; ++i) {
190 ar = &ranges[i];
191
192 if (s->addr >= ar->start && s->addr <= ar->end)
193 return 1;
194 }
195
196 return 0;
197}
198
199static bool string_starts_with(const char *s, const char *prefix)
200{
201 return strncmp(s, prefix, strlen(prefix)) == 0;
202}
203
204static int symbol_valid(const struct sym_entry *s)
205{
206 const char *name = sym_name(s);
207
208 /* if --all-symbols is not specified, then symbols outside the text
209 * and inittext sections are discarded */
210 if (!all_symbols) {
211 /*
212 * Symbols starting with __start and __stop are used to denote
213 * section boundaries, and should always be included:
214 */
215 if (string_starts_with(name, "__start_") ||
216 string_starts_with(name, "__stop_"))
217 return 1;
218
219 if (symbol_in_range(s, text_ranges,
220 ARRAY_SIZE(text_ranges)) == 0)
221 return 0;
222 /* Corner case. Discard any symbols with the same value as
223 * _etext _einittext; they can move between pass 1 and 2 when
224 * the kallsyms data are added. If these symbols move then
225 * they may get dropped in pass 2, which breaks the kallsyms
226 * rules.
227 */
228 if ((s->addr == text_range_text->end &&
229 strcmp(name, text_range_text->end_sym)) ||
230 (s->addr == text_range_inittext->end &&
231 strcmp(name, text_range_inittext->end_sym)))
232 return 0;
233 }
234
235 return 1;
236}
237
238/* remove all the invalid symbols from the table */
239static void shrink_table(void)
240{
241 unsigned int i, pos;
242
243 pos = 0;
244 for (i = 0; i < table_cnt; i++) {
245 if (symbol_valid(table[i])) {
246 if (pos != i)
247 table[pos] = table[i];
248 pos++;
249 } else {
250 free(table[i]);
251 }
252 }
253 table_cnt = pos;
254}
255
256static void read_map(const char *in)
257{
258 FILE *fp;
259 struct sym_entry *sym;
260 char *buf = NULL;
261 size_t buflen = 0;
262
263 fp = fopen(in, "r");
264 if (!fp) {
265 perror(in);
266 exit(1);
267 }
268
269 while (!feof(fp)) {
270 sym = read_symbol(fp, &buf, &buflen);
271 if (!sym)
272 continue;
273
274 sym->seq = table_cnt;
275
276 if (table_cnt >= table_size) {
277 table_size += 10000;
278 table = xrealloc(table, sizeof(*table) * table_size);
279 }
280
281 table[table_cnt++] = sym;
282 }
283
284 free(buf);
285 fclose(fp);
286}
287
288static void output_label(const char *label)
289{
290 printf(".globl %s\n", label);
291 printf("\tALGN\n");
292 printf("%s:\n", label);
293}
294
295/* uncompress a compressed symbol. When this function is called, the best table
296 * might still be compressed itself, so the function needs to be recursive */
297static int expand_symbol(const unsigned char *data, int len, char *result)
298{
299 int c, rlen, total=0;
300
301 while (len) {
302 c = *data;
303 /* if the table holds a single char that is the same as the one
304 * we are looking for, then end the search */
305 if (best_table[c][0]==c && best_table_len[c]==1) {
306 *result++ = c;
307 total++;
308 } else {
309 /* if not, recurse and expand */
310 rlen = expand_symbol(best_table[c], best_table_len[c], result);
311 total += rlen;
312 result += rlen;
313 }
314 data++;
315 len--;
316 }
317 *result=0;
318
319 return total;
320}
321
322static bool symbol_absolute(const struct sym_entry *s)
323{
324 return s->percpu_absolute;
325}
326
327static int compare_names(const void *a, const void *b)
328{
329 int ret;
330 const struct sym_entry *sa = *(const struct sym_entry **)a;
331 const struct sym_entry *sb = *(const struct sym_entry **)b;
332
333 ret = strcmp(sym_name(sa), sym_name(sb));
334 if (!ret) {
335 if (sa->addr > sb->addr)
336 return 1;
337 else if (sa->addr < sb->addr)
338 return -1;
339
340 /* keep old order */
341 return (int)(sa->seq - sb->seq);
342 }
343
344 return ret;
345}
346
347static void sort_symbols_by_name(void)
348{
349 qsort(table, table_cnt, sizeof(table[0]), compare_names);
350}
351
352static void write_src(void)
353{
354 unsigned int i, k, off;
355 unsigned int best_idx[256];
356 unsigned int *markers, markers_cnt;
357 char buf[KSYM_NAME_LEN];
358
359 printf("#include <asm/bitsperlong.h>\n");
360 printf("#if BITS_PER_LONG == 64\n");
361 printf("#define PTR .quad\n");
362 printf("#define ALGN .balign 8\n");
363 printf("#else\n");
364 printf("#define PTR .long\n");
365 printf("#define ALGN .balign 4\n");
366 printf("#endif\n");
367
368 printf("\t.section .rodata, \"a\"\n");
369
370 output_label("kallsyms_num_syms");
371 printf("\t.long\t%u\n", table_cnt);
372 printf("\n");
373
374 /* table of offset markers, that give the offset in the compressed stream
375 * every 256 symbols */
376 markers_cnt = (table_cnt + 255) / 256;
377 markers = xmalloc(sizeof(*markers) * markers_cnt);
378
379 output_label("kallsyms_names");
380 off = 0;
381 for (i = 0; i < table_cnt; i++) {
382 if ((i & 0xFF) == 0)
383 markers[i >> 8] = off;
384 table[i]->seq = i;
385
386 /* There cannot be any symbol of length zero. */
387 if (table[i]->len == 0) {
388 fprintf(stderr, "kallsyms failure: "
389 "unexpected zero symbol length\n");
390 exit(EXIT_FAILURE);
391 }
392
393 /* Only lengths that fit in up-to-two-byte ULEB128 are supported. */
394 if (table[i]->len > 0x3FFF) {
395 fprintf(stderr, "kallsyms failure: "
396 "unexpected huge symbol length\n");
397 exit(EXIT_FAILURE);
398 }
399
400 /* Encode length with ULEB128. */
401 if (table[i]->len <= 0x7F) {
402 /* Most symbols use a single byte for the length. */
403 printf("\t.byte 0x%02x", table[i]->len);
404 off += table[i]->len + 1;
405 } else {
406 /* "Big" symbols use two bytes. */
407 printf("\t.byte 0x%02x, 0x%02x",
408 (table[i]->len & 0x7F) | 0x80,
409 (table[i]->len >> 7) & 0x7F);
410 off += table[i]->len + 2;
411 }
412 for (k = 0; k < table[i]->len; k++)
413 printf(", 0x%02x", table[i]->sym[k]);
414
415 /*
416 * Now that we wrote out the compressed symbol name, restore the
417 * original name and print it in the comment.
418 */
419 expand_symbol(table[i]->sym, table[i]->len, buf);
420 strcpy((char *)table[i]->sym, buf);
421 printf("\t/* %s */\n", table[i]->sym);
422 }
423 printf("\n");
424
425 output_label("kallsyms_markers");
426 for (i = 0; i < markers_cnt; i++)
427 printf("\t.long\t%u\n", markers[i]);
428 printf("\n");
429
430 free(markers);
431
432 output_label("kallsyms_token_table");
433 off = 0;
434 for (i = 0; i < 256; i++) {
435 best_idx[i] = off;
436 expand_symbol(best_table[i], best_table_len[i], buf);
437 printf("\t.asciz\t\"%s\"\n", buf);
438 off += strlen(buf) + 1;
439 }
440 printf("\n");
441
442 output_label("kallsyms_token_index");
443 for (i = 0; i < 256; i++)
444 printf("\t.short\t%d\n", best_idx[i]);
445 printf("\n");
446
447 output_label("kallsyms_offsets");
448
449 for (i = 0; i < table_cnt; i++) {
450 /*
451 * Use the offset relative to the lowest value
452 * encountered of all relative symbols, and emit
453 * non-relocatable fixed offsets that will be fixed
454 * up at runtime.
455 */
456
457 long long offset;
458 bool overflow;
459
460 if (!absolute_percpu) {
461 offset = table[i]->addr - relative_base;
462 overflow = offset < 0 || offset > UINT_MAX;
463 } else if (symbol_absolute(table[i])) {
464 offset = table[i]->addr;
465 overflow = offset < 0 || offset > INT_MAX;
466 } else {
467 offset = relative_base - table[i]->addr - 1;
468 overflow = offset < INT_MIN || offset >= 0;
469 }
470 if (overflow) {
471 fprintf(stderr, "kallsyms failure: "
472 "%s symbol value %#llx out of range in relative mode\n",
473 symbol_absolute(table[i]) ? "absolute" : "relative",
474 table[i]->addr);
475 exit(EXIT_FAILURE);
476 }
477 printf("\t.long\t%#x\t/* %s */\n", (int)offset, table[i]->sym);
478 }
479 printf("\n");
480
481 output_label("kallsyms_relative_base");
482 /* Provide proper symbols relocatability by their '_text' relativeness. */
483 if (_text <= relative_base)
484 printf("\tPTR\t_text + %#llx\n", relative_base - _text);
485 else
486 printf("\tPTR\t_text - %#llx\n", _text - relative_base);
487 printf("\n");
488
489 sort_symbols_by_name();
490 output_label("kallsyms_seqs_of_names");
491 for (i = 0; i < table_cnt; i++)
492 printf("\t.byte 0x%02x, 0x%02x, 0x%02x\t/* %s */\n",
493 (unsigned char)(table[i]->seq >> 16),
494 (unsigned char)(table[i]->seq >> 8),
495 (unsigned char)(table[i]->seq >> 0),
496 table[i]->sym);
497 printf("\n");
498}
499
500
501/* table lookup compression functions */
502
503/* count all the possible tokens in a symbol */
504static void learn_symbol(const unsigned char *symbol, int len)
505{
506 int i;
507
508 for (i = 0; i < len - 1; i++)
509 token_profit[ symbol[i] + (symbol[i + 1] << 8) ]++;
510}
511
512/* decrease the count for all the possible tokens in a symbol */
513static void forget_symbol(const unsigned char *symbol, int len)
514{
515 int i;
516
517 for (i = 0; i < len - 1; i++)
518 token_profit[ symbol[i] + (symbol[i + 1] << 8) ]--;
519}
520
521/* do the initial token count */
522static void build_initial_token_table(void)
523{
524 unsigned int i;
525
526 for (i = 0; i < table_cnt; i++)
527 learn_symbol(table[i]->sym, table[i]->len);
528}
529
530static unsigned char *find_token(unsigned char *str, int len,
531 const unsigned char *token)
532{
533 int i;
534
535 for (i = 0; i < len - 1; i++) {
536 if (str[i] == token[0] && str[i+1] == token[1])
537 return &str[i];
538 }
539 return NULL;
540}
541
542/* replace a given token in all the valid symbols. Use the sampled symbols
543 * to update the counts */
544static void compress_symbols(const unsigned char *str, int idx)
545{
546 unsigned int i, len, size;
547 unsigned char *p1, *p2;
548
549 for (i = 0; i < table_cnt; i++) {
550
551 len = table[i]->len;
552 p1 = table[i]->sym;
553
554 /* find the token on the symbol */
555 p2 = find_token(p1, len, str);
556 if (!p2) continue;
557
558 /* decrease the counts for this symbol's tokens */
559 forget_symbol(table[i]->sym, len);
560
561 size = len;
562
563 do {
564 *p2 = idx;
565 p2++;
566 size -= (p2 - p1);
567 memmove(p2, p2 + 1, size);
568 p1 = p2;
569 len--;
570
571 if (size < 2) break;
572
573 /* find the token on the symbol */
574 p2 = find_token(p1, size, str);
575
576 } while (p2);
577
578 table[i]->len = len;
579
580 /* increase the counts for this symbol's new tokens */
581 learn_symbol(table[i]->sym, len);
582 }
583}
584
585/* search the token with the maximum profit */
586static int find_best_token(void)
587{
588 int i, best, bestprofit;
589
590 bestprofit=-10000;
591 best = 0;
592
593 for (i = 0; i < 0x10000; i++) {
594 if (token_profit[i] > bestprofit) {
595 best = i;
596 bestprofit = token_profit[i];
597 }
598 }
599 return best;
600}
601
602/* this is the core of the algorithm: calculate the "best" table */
603static void optimize_result(void)
604{
605 int i, best;
606
607 /* using the '\0' symbol last allows compress_symbols to use standard
608 * fast string functions */
609 for (i = 255; i >= 0; i--) {
610
611 /* if this table slot is empty (it is not used by an actual
612 * original char code */
613 if (!best_table_len[i]) {
614
615 /* find the token with the best profit value */
616 best = find_best_token();
617 if (token_profit[best] == 0)
618 break;
619
620 /* place it in the "best" table */
621 best_table_len[i] = 2;
622 best_table[i][0] = best & 0xFF;
623 best_table[i][1] = (best >> 8) & 0xFF;
624
625 /* replace this token in all the valid symbols */
626 compress_symbols(best_table[i], i);
627 }
628 }
629}
630
631/* start by placing the symbols that are actually used on the table */
632static void insert_real_symbols_in_table(void)
633{
634 unsigned int i, j, c;
635
636 for (i = 0; i < table_cnt; i++) {
637 for (j = 0; j < table[i]->len; j++) {
638 c = table[i]->sym[j];
639 best_table[c][0]=c;
640 best_table_len[c]=1;
641 }
642 }
643}
644
645static void optimize_token_table(void)
646{
647 build_initial_token_table();
648
649 insert_real_symbols_in_table();
650
651 optimize_result();
652}
653
654/* guess for "linker script provide" symbol */
655static int may_be_linker_script_provide_symbol(const struct sym_entry *se)
656{
657 const char *symbol = sym_name(se);
658 int len = se->len - 1;
659
660 if (len < 8)
661 return 0;
662
663 if (symbol[0] != '_' || symbol[1] != '_')
664 return 0;
665
666 /* __start_XXXXX */
667 if (!memcmp(symbol + 2, "start_", 6))
668 return 1;
669
670 /* __stop_XXXXX */
671 if (!memcmp(symbol + 2, "stop_", 5))
672 return 1;
673
674 /* __end_XXXXX */
675 if (!memcmp(symbol + 2, "end_", 4))
676 return 1;
677
678 /* __XXXXX_start */
679 if (!memcmp(symbol + len - 6, "_start", 6))
680 return 1;
681
682 /* __XXXXX_end */
683 if (!memcmp(symbol + len - 4, "_end", 4))
684 return 1;
685
686 return 0;
687}
688
689static int compare_symbols(const void *a, const void *b)
690{
691 const struct sym_entry *sa = *(const struct sym_entry **)a;
692 const struct sym_entry *sb = *(const struct sym_entry **)b;
693 int wa, wb;
694
695 /* sort by address first */
696 if (sa->addr > sb->addr)
697 return 1;
698 if (sa->addr < sb->addr)
699 return -1;
700
701 /* sort by "weakness" type */
702 wa = (sa->sym[0] == 'w') || (sa->sym[0] == 'W');
703 wb = (sb->sym[0] == 'w') || (sb->sym[0] == 'W');
704 if (wa != wb)
705 return wa - wb;
706
707 /* sort by "linker script provide" type */
708 wa = may_be_linker_script_provide_symbol(sa);
709 wb = may_be_linker_script_provide_symbol(sb);
710 if (wa != wb)
711 return wa - wb;
712
713 /* sort by the number of prefix underscores */
714 wa = strspn(sym_name(sa), "_");
715 wb = strspn(sym_name(sb), "_");
716 if (wa != wb)
717 return wa - wb;
718
719 /* sort by initial order, so that other symbols are left undisturbed */
720 return sa->seq - sb->seq;
721}
722
723static void sort_symbols(void)
724{
725 qsort(table, table_cnt, sizeof(table[0]), compare_symbols);
726}
727
728static void make_percpus_absolute(void)
729{
730 unsigned int i;
731
732 for (i = 0; i < table_cnt; i++)
733 if (symbol_in_range(table[i], &percpu_range, 1)) {
734 /*
735 * Keep the 'A' override for percpu symbols to
736 * ensure consistent behavior compared to older
737 * versions of this tool.
738 */
739 table[i]->sym[0] = 'A';
740 table[i]->percpu_absolute = true;
741 }
742}
743
744/* find the minimum non-absolute symbol address */
745static void record_relative_base(void)
746{
747 unsigned int i;
748
749 for (i = 0; i < table_cnt; i++)
750 if (!symbol_absolute(table[i])) {
751 /*
752 * The table is sorted by address.
753 * Take the first non-absolute symbol value.
754 */
755 relative_base = table[i]->addr;
756 return;
757 }
758}
759
760int main(int argc, char **argv)
761{
762 while (1) {
763 static const struct option long_options[] = {
764 {"all-symbols", no_argument, &all_symbols, 1},
765 {"absolute-percpu", no_argument, &absolute_percpu, 1},
766 {},
767 };
768
769 int c = getopt_long(argc, argv, "", long_options, NULL);
770
771 if (c == -1)
772 break;
773 if (c != 0)
774 usage();
775 }
776
777 if (optind >= argc)
778 usage();
779
780 read_map(argv[optind]);
781 shrink_table();
782 if (absolute_percpu)
783 make_percpus_absolute();
784 sort_symbols();
785 record_relative_base();
786 optimize_token_table();
787 write_src();
788
789 return 0;
790}