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 <stdbool.h>
22#include <stdio.h>
23#include <stdlib.h>
24#include <string.h>
25#include <ctype.h>
26#include <limits.h>
27
28#define ARRAY_SIZE(arr) (sizeof(arr) / sizeof(arr[0]))
29
30#define KSYM_NAME_LEN 128
31
32struct sym_entry {
33 unsigned long long addr;
34 unsigned int len;
35 unsigned int start_pos;
36 unsigned int percpu_absolute;
37 unsigned char sym[];
38};
39
40struct addr_range {
41 const char *start_sym, *end_sym;
42 unsigned long long start, end;
43};
44
45static unsigned long long _text;
46static unsigned long long relative_base;
47static struct addr_range text_ranges[] = {
48 { "_stext", "_etext" },
49 { "_sinittext", "_einittext" },
50};
51#define text_range_text (&text_ranges[0])
52#define text_range_inittext (&text_ranges[1])
53
54static struct addr_range percpu_range = {
55 "__per_cpu_start", "__per_cpu_end", -1ULL, 0
56};
57
58static struct sym_entry **table;
59static unsigned int table_size, table_cnt;
60static int all_symbols;
61static int absolute_percpu;
62static int base_relative;
63
64static int token_profit[0x10000];
65
66/* the table that holds the result of the compression */
67static unsigned char best_table[256][2];
68static unsigned char best_table_len[256];
69
70
71static void usage(void)
72{
73 fprintf(stderr, "Usage: kallsyms [--all-symbols] "
74 "[--base-relative] < in.map > out.S\n");
75 exit(1);
76}
77
78static char *sym_name(const struct sym_entry *s)
79{
80 return (char *)s->sym + 1;
81}
82
83static bool is_ignored_symbol(const char *name, char type)
84{
85 /* Symbol names that exactly match to the following are ignored.*/
86 static const char * const ignored_symbols[] = {
87 /*
88 * Symbols which vary between passes. Passes 1 and 2 must have
89 * identical symbol lists. The kallsyms_* symbols below are
90 * only added after pass 1, they would be included in pass 2
91 * when --all-symbols is specified so exclude them to get a
92 * stable symbol list.
93 */
94 "kallsyms_addresses",
95 "kallsyms_offsets",
96 "kallsyms_relative_base",
97 "kallsyms_num_syms",
98 "kallsyms_names",
99 "kallsyms_markers",
100 "kallsyms_token_table",
101 "kallsyms_token_index",
102 /* Exclude linker generated symbols which vary between passes */
103 "_SDA_BASE_", /* ppc */
104 "_SDA2_BASE_", /* ppc */
105 NULL
106 };
107
108 /* Symbol names that begin with the following are ignored.*/
109 static const char * const ignored_prefixes[] = {
110 "$", /* local symbols for ARM, MIPS, etc. */
111 ".LASANPC", /* s390 kasan local symbols */
112 "__crc_", /* modversions */
113 "__efistub_", /* arm64 EFI stub namespace */
114 "__kvm_nvhe_", /* arm64 non-VHE KVM namespace */
115 NULL
116 };
117
118 /* Symbol names that end with the following are ignored.*/
119 static const char * const ignored_suffixes[] = {
120 "_from_arm", /* arm */
121 "_from_thumb", /* arm */
122 "_veneer", /* arm */
123 NULL
124 };
125
126 /* Symbol names that contain the following are ignored.*/
127 static const char * const ignored_matches[] = {
128 ".long_branch.", /* ppc stub */
129 ".plt_branch.", /* ppc stub */
130 NULL
131 };
132
133 const char * const *p;
134
135 for (p = ignored_symbols; *p; p++)
136 if (!strcmp(name, *p))
137 return true;
138
139 for (p = ignored_prefixes; *p; p++)
140 if (!strncmp(name, *p, strlen(*p)))
141 return true;
142
143 for (p = ignored_suffixes; *p; p++) {
144 int l = strlen(name) - strlen(*p);
145
146 if (l >= 0 && !strcmp(name + l, *p))
147 return true;
148 }
149
150 for (p = ignored_matches; *p; p++) {
151 if (strstr(name, *p))
152 return true;
153 }
154
155 if (type == 'U' || type == 'u')
156 return true;
157 /* exclude debugging symbols */
158 if (type == 'N' || type == 'n')
159 return true;
160
161 if (toupper(type) == 'A') {
162 /* Keep these useful absolute symbols */
163 if (strcmp(name, "__kernel_syscall_via_break") &&
164 strcmp(name, "__kernel_syscall_via_epc") &&
165 strcmp(name, "__kernel_sigtramp") &&
166 strcmp(name, "__gp"))
167 return true;
168 }
169
170 return false;
171}
172
173static void check_symbol_range(const char *sym, unsigned long long addr,
174 struct addr_range *ranges, int entries)
175{
176 size_t i;
177 struct addr_range *ar;
178
179 for (i = 0; i < entries; ++i) {
180 ar = &ranges[i];
181
182 if (strcmp(sym, ar->start_sym) == 0) {
183 ar->start = addr;
184 return;
185 } else if (strcmp(sym, ar->end_sym) == 0) {
186 ar->end = addr;
187 return;
188 }
189 }
190}
191
192static struct sym_entry *read_symbol(FILE *in)
193{
194 char name[500], type;
195 unsigned long long addr;
196 unsigned int len;
197 struct sym_entry *sym;
198 int rc;
199
200 rc = fscanf(in, "%llx %c %499s\n", &addr, &type, name);
201 if (rc != 3) {
202 if (rc != EOF && fgets(name, 500, in) == NULL)
203 fprintf(stderr, "Read error or end of file.\n");
204 return NULL;
205 }
206 if (strlen(name) >= KSYM_NAME_LEN) {
207 fprintf(stderr, "Symbol %s too long for kallsyms (%zu >= %d).\n"
208 "Please increase KSYM_NAME_LEN both in kernel and kallsyms.c\n",
209 name, strlen(name), KSYM_NAME_LEN);
210 return NULL;
211 }
212
213 if (strcmp(name, "_text") == 0)
214 _text = addr;
215
216 /* Ignore most absolute/undefined (?) symbols. */
217 if (is_ignored_symbol(name, type))
218 return NULL;
219
220 check_symbol_range(name, addr, text_ranges, ARRAY_SIZE(text_ranges));
221 check_symbol_range(name, addr, &percpu_range, 1);
222
223 /* include the type field in the symbol name, so that it gets
224 * compressed together */
225
226 len = strlen(name) + 1;
227
228 sym = malloc(sizeof(*sym) + len + 1);
229 if (!sym) {
230 fprintf(stderr, "kallsyms failure: "
231 "unable to allocate required amount of memory\n");
232 exit(EXIT_FAILURE);
233 }
234 sym->addr = addr;
235 sym->len = len;
236 sym->sym[0] = type;
237 strcpy(sym_name(sym), name);
238 sym->percpu_absolute = 0;
239
240 return sym;
241}
242
243static int symbol_in_range(const struct sym_entry *s,
244 const struct addr_range *ranges, int entries)
245{
246 size_t i;
247 const struct addr_range *ar;
248
249 for (i = 0; i < entries; ++i) {
250 ar = &ranges[i];
251
252 if (s->addr >= ar->start && s->addr <= ar->end)
253 return 1;
254 }
255
256 return 0;
257}
258
259static int symbol_valid(const struct sym_entry *s)
260{
261 const char *name = sym_name(s);
262
263 /* if --all-symbols is not specified, then symbols outside the text
264 * and inittext sections are discarded */
265 if (!all_symbols) {
266 if (symbol_in_range(s, text_ranges,
267 ARRAY_SIZE(text_ranges)) == 0)
268 return 0;
269 /* Corner case. Discard any symbols with the same value as
270 * _etext _einittext; they can move between pass 1 and 2 when
271 * the kallsyms data are added. If these symbols move then
272 * they may get dropped in pass 2, which breaks the kallsyms
273 * rules.
274 */
275 if ((s->addr == text_range_text->end &&
276 strcmp(name, text_range_text->end_sym)) ||
277 (s->addr == text_range_inittext->end &&
278 strcmp(name, text_range_inittext->end_sym)))
279 return 0;
280 }
281
282 return 1;
283}
284
285/* remove all the invalid symbols from the table */
286static void shrink_table(void)
287{
288 unsigned int i, pos;
289
290 pos = 0;
291 for (i = 0; i < table_cnt; i++) {
292 if (symbol_valid(table[i])) {
293 if (pos != i)
294 table[pos] = table[i];
295 pos++;
296 } else {
297 free(table[i]);
298 }
299 }
300 table_cnt = pos;
301
302 /* When valid symbol is not registered, exit to error */
303 if (!table_cnt) {
304 fprintf(stderr, "No valid symbol.\n");
305 exit(1);
306 }
307}
308
309static void read_map(FILE *in)
310{
311 struct sym_entry *sym;
312
313 while (!feof(in)) {
314 sym = read_symbol(in);
315 if (!sym)
316 continue;
317
318 sym->start_pos = table_cnt;
319
320 if (table_cnt >= table_size) {
321 table_size += 10000;
322 table = realloc(table, sizeof(*table) * table_size);
323 if (!table) {
324 fprintf(stderr, "out of memory\n");
325 exit (1);
326 }
327 }
328
329 table[table_cnt++] = sym;
330 }
331}
332
333static void output_label(const char *label)
334{
335 printf(".globl %s\n", label);
336 printf("\tALGN\n");
337 printf("%s:\n", label);
338}
339
340/* Provide proper symbols relocatability by their '_text' relativeness. */
341static void output_address(unsigned long long addr)
342{
343 if (_text <= addr)
344 printf("\tPTR\t_text + %#llx\n", addr - _text);
345 else
346 printf("\tPTR\t_text - %#llx\n", _text - addr);
347}
348
349/* uncompress a compressed symbol. When this function is called, the best table
350 * might still be compressed itself, so the function needs to be recursive */
351static int expand_symbol(const unsigned char *data, int len, char *result)
352{
353 int c, rlen, total=0;
354
355 while (len) {
356 c = *data;
357 /* if the table holds a single char that is the same as the one
358 * we are looking for, then end the search */
359 if (best_table[c][0]==c && best_table_len[c]==1) {
360 *result++ = c;
361 total++;
362 } else {
363 /* if not, recurse and expand */
364 rlen = expand_symbol(best_table[c], best_table_len[c], result);
365 total += rlen;
366 result += rlen;
367 }
368 data++;
369 len--;
370 }
371 *result=0;
372
373 return total;
374}
375
376static int symbol_absolute(const struct sym_entry *s)
377{
378 return s->percpu_absolute;
379}
380
381static void write_src(void)
382{
383 unsigned int i, k, off;
384 unsigned int best_idx[256];
385 unsigned int *markers;
386 char buf[KSYM_NAME_LEN];
387
388 printf("#include <asm/bitsperlong.h>\n");
389 printf("#if BITS_PER_LONG == 64\n");
390 printf("#define PTR .quad\n");
391 printf("#define ALGN .balign 8\n");
392 printf("#else\n");
393 printf("#define PTR .long\n");
394 printf("#define ALGN .balign 4\n");
395 printf("#endif\n");
396
397 printf("\t.section .rodata, \"a\"\n");
398
399 if (!base_relative)
400 output_label("kallsyms_addresses");
401 else
402 output_label("kallsyms_offsets");
403
404 for (i = 0; i < table_cnt; i++) {
405 if (base_relative) {
406 /*
407 * Use the offset relative to the lowest value
408 * encountered of all relative symbols, and emit
409 * non-relocatable fixed offsets that will be fixed
410 * up at runtime.
411 */
412
413 long long offset;
414 int overflow;
415
416 if (!absolute_percpu) {
417 offset = table[i]->addr - relative_base;
418 overflow = (offset < 0 || offset > UINT_MAX);
419 } else if (symbol_absolute(table[i])) {
420 offset = table[i]->addr;
421 overflow = (offset < 0 || offset > INT_MAX);
422 } else {
423 offset = relative_base - table[i]->addr - 1;
424 overflow = (offset < INT_MIN || offset >= 0);
425 }
426 if (overflow) {
427 fprintf(stderr, "kallsyms failure: "
428 "%s symbol value %#llx out of range in relative mode\n",
429 symbol_absolute(table[i]) ? "absolute" : "relative",
430 table[i]->addr);
431 exit(EXIT_FAILURE);
432 }
433 printf("\t.long\t%#x\n", (int)offset);
434 } else if (!symbol_absolute(table[i])) {
435 output_address(table[i]->addr);
436 } else {
437 printf("\tPTR\t%#llx\n", table[i]->addr);
438 }
439 }
440 printf("\n");
441
442 if (base_relative) {
443 output_label("kallsyms_relative_base");
444 output_address(relative_base);
445 printf("\n");
446 }
447
448 output_label("kallsyms_num_syms");
449 printf("\t.long\t%u\n", table_cnt);
450 printf("\n");
451
452 /* table of offset markers, that give the offset in the compressed stream
453 * every 256 symbols */
454 markers = malloc(sizeof(unsigned int) * ((table_cnt + 255) / 256));
455 if (!markers) {
456 fprintf(stderr, "kallsyms failure: "
457 "unable to allocate required memory\n");
458 exit(EXIT_FAILURE);
459 }
460
461 output_label("kallsyms_names");
462 off = 0;
463 for (i = 0; i < table_cnt; i++) {
464 if ((i & 0xFF) == 0)
465 markers[i >> 8] = off;
466
467 printf("\t.byte 0x%02x", table[i]->len);
468 for (k = 0; k < table[i]->len; k++)
469 printf(", 0x%02x", table[i]->sym[k]);
470 printf("\n");
471
472 off += table[i]->len + 1;
473 }
474 printf("\n");
475
476 output_label("kallsyms_markers");
477 for (i = 0; i < ((table_cnt + 255) >> 8); i++)
478 printf("\t.long\t%u\n", markers[i]);
479 printf("\n");
480
481 free(markers);
482
483 output_label("kallsyms_token_table");
484 off = 0;
485 for (i = 0; i < 256; i++) {
486 best_idx[i] = off;
487 expand_symbol(best_table[i], best_table_len[i], buf);
488 printf("\t.asciz\t\"%s\"\n", buf);
489 off += strlen(buf) + 1;
490 }
491 printf("\n");
492
493 output_label("kallsyms_token_index");
494 for (i = 0; i < 256; i++)
495 printf("\t.short\t%d\n", best_idx[i]);
496 printf("\n");
497}
498
499
500/* table lookup compression functions */
501
502/* count all the possible tokens in a symbol */
503static void learn_symbol(const unsigned char *symbol, int len)
504{
505 int i;
506
507 for (i = 0; i < len - 1; i++)
508 token_profit[ symbol[i] + (symbol[i + 1] << 8) ]++;
509}
510
511/* decrease the count for all the possible tokens in a symbol */
512static void forget_symbol(const unsigned char *symbol, int len)
513{
514 int i;
515
516 for (i = 0; i < len - 1; i++)
517 token_profit[ symbol[i] + (symbol[i + 1] << 8) ]--;
518}
519
520/* do the initial token count */
521static void build_initial_tok_table(void)
522{
523 unsigned int i;
524
525 for (i = 0; i < table_cnt; i++)
526 learn_symbol(table[i]->sym, table[i]->len);
527}
528
529static unsigned char *find_token(unsigned char *str, int len,
530 const unsigned char *token)
531{
532 int i;
533
534 for (i = 0; i < len - 1; i++) {
535 if (str[i] == token[0] && str[i+1] == token[1])
536 return &str[i];
537 }
538 return NULL;
539}
540
541/* replace a given token in all the valid symbols. Use the sampled symbols
542 * to update the counts */
543static void compress_symbols(const unsigned char *str, int idx)
544{
545 unsigned int i, len, size;
546 unsigned char *p1, *p2;
547
548 for (i = 0; i < table_cnt; i++) {
549
550 len = table[i]->len;
551 p1 = table[i]->sym;
552
553 /* find the token on the symbol */
554 p2 = find_token(p1, len, str);
555 if (!p2) continue;
556
557 /* decrease the counts for this symbol's tokens */
558 forget_symbol(table[i]->sym, len);
559
560 size = len;
561
562 do {
563 *p2 = idx;
564 p2++;
565 size -= (p2 - p1);
566 memmove(p2, p2 + 1, size);
567 p1 = p2;
568 len--;
569
570 if (size < 2) break;
571
572 /* find the token on the symbol */
573 p2 = find_token(p1, size, str);
574
575 } while (p2);
576
577 table[i]->len = len;
578
579 /* increase the counts for this symbol's new tokens */
580 learn_symbol(table[i]->sym, len);
581 }
582}
583
584/* search the token with the maximum profit */
585static int find_best_token(void)
586{
587 int i, best, bestprofit;
588
589 bestprofit=-10000;
590 best = 0;
591
592 for (i = 0; i < 0x10000; i++) {
593 if (token_profit[i] > bestprofit) {
594 best = i;
595 bestprofit = token_profit[i];
596 }
597 }
598 return best;
599}
600
601/* this is the core of the algorithm: calculate the "best" table */
602static void optimize_result(void)
603{
604 int i, best;
605
606 /* using the '\0' symbol last allows compress_symbols to use standard
607 * fast string functions */
608 for (i = 255; i >= 0; i--) {
609
610 /* if this table slot is empty (it is not used by an actual
611 * original char code */
612 if (!best_table_len[i]) {
613
614 /* find the token with the best profit value */
615 best = find_best_token();
616 if (token_profit[best] == 0)
617 break;
618
619 /* place it in the "best" table */
620 best_table_len[i] = 2;
621 best_table[i][0] = best & 0xFF;
622 best_table[i][1] = (best >> 8) & 0xFF;
623
624 /* replace this token in all the valid symbols */
625 compress_symbols(best_table[i], i);
626 }
627 }
628}
629
630/* start by placing the symbols that are actually used on the table */
631static void insert_real_symbols_in_table(void)
632{
633 unsigned int i, j, c;
634
635 for (i = 0; i < table_cnt; i++) {
636 for (j = 0; j < table[i]->len; j++) {
637 c = table[i]->sym[j];
638 best_table[c][0]=c;
639 best_table_len[c]=1;
640 }
641 }
642}
643
644static void optimize_token_table(void)
645{
646 build_initial_tok_table();
647
648 insert_real_symbols_in_table();
649
650 optimize_result();
651}
652
653/* guess for "linker script provide" symbol */
654static int may_be_linker_script_provide_symbol(const struct sym_entry *se)
655{
656 const char *symbol = sym_name(se);
657 int len = se->len - 1;
658
659 if (len < 8)
660 return 0;
661
662 if (symbol[0] != '_' || symbol[1] != '_')
663 return 0;
664
665 /* __start_XXXXX */
666 if (!memcmp(symbol + 2, "start_", 6))
667 return 1;
668
669 /* __stop_XXXXX */
670 if (!memcmp(symbol + 2, "stop_", 5))
671 return 1;
672
673 /* __end_XXXXX */
674 if (!memcmp(symbol + 2, "end_", 4))
675 return 1;
676
677 /* __XXXXX_start */
678 if (!memcmp(symbol + len - 6, "_start", 6))
679 return 1;
680
681 /* __XXXXX_end */
682 if (!memcmp(symbol + len - 4, "_end", 4))
683 return 1;
684
685 return 0;
686}
687
688static int compare_symbols(const void *a, const void *b)
689{
690 const struct sym_entry *sa = *(const struct sym_entry **)a;
691 const struct sym_entry *sb = *(const struct sym_entry **)b;
692 int wa, wb;
693
694 /* sort by address first */
695 if (sa->addr > sb->addr)
696 return 1;
697 if (sa->addr < sb->addr)
698 return -1;
699
700 /* sort by "weakness" type */
701 wa = (sa->sym[0] == 'w') || (sa->sym[0] == 'W');
702 wb = (sb->sym[0] == 'w') || (sb->sym[0] == 'W');
703 if (wa != wb)
704 return wa - wb;
705
706 /* sort by "linker script provide" type */
707 wa = may_be_linker_script_provide_symbol(sa);
708 wb = may_be_linker_script_provide_symbol(sb);
709 if (wa != wb)
710 return wa - wb;
711
712 /* sort by the number of prefix underscores */
713 wa = strspn(sym_name(sa), "_");
714 wb = strspn(sym_name(sb), "_");
715 if (wa != wb)
716 return wa - wb;
717
718 /* sort by initial order, so that other symbols are left undisturbed */
719 return sa->start_pos - sb->start_pos;
720}
721
722static void sort_symbols(void)
723{
724 qsort(table, table_cnt, sizeof(table[0]), compare_symbols);
725}
726
727static void make_percpus_absolute(void)
728{
729 unsigned int i;
730
731 for (i = 0; i < table_cnt; i++)
732 if (symbol_in_range(table[i], &percpu_range, 1)) {
733 /*
734 * Keep the 'A' override for percpu symbols to
735 * ensure consistent behavior compared to older
736 * versions of this tool.
737 */
738 table[i]->sym[0] = 'A';
739 table[i]->percpu_absolute = 1;
740 }
741}
742
743/* find the minimum non-absolute symbol address */
744static void record_relative_base(void)
745{
746 unsigned int i;
747
748 for (i = 0; i < table_cnt; i++)
749 if (!symbol_absolute(table[i])) {
750 /*
751 * The table is sorted by address.
752 * Take the first non-absolute symbol value.
753 */
754 relative_base = table[i]->addr;
755 return;
756 }
757}
758
759int main(int argc, char **argv)
760{
761 if (argc >= 2) {
762 int i;
763 for (i = 1; i < argc; i++) {
764 if(strcmp(argv[i], "--all-symbols") == 0)
765 all_symbols = 1;
766 else if (strcmp(argv[i], "--absolute-percpu") == 0)
767 absolute_percpu = 1;
768 else if (strcmp(argv[i], "--base-relative") == 0)
769 base_relative = 1;
770 else
771 usage();
772 }
773 } else if (argc != 1)
774 usage();
775
776 read_map(stdin);
777 shrink_table();
778 if (absolute_percpu)
779 make_percpus_absolute();
780 sort_symbols();
781 if (base_relative)
782 record_relative_base();
783 optimize_token_table();
784 write_src();
785
786 return 0;
787}
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]
9 * [--base-relative] [--lto-clang] in.map > out.S
10 *
11 * Table compression uses all the unused char codes on the symbols and
12 * maps these to the most used substrings (tokens). For instance, it might
13 * map char code 0xF7 to represent "write_" and then in every symbol where
14 * "write_" appears it can be replaced by 0xF7, saving 5 bytes.
15 * The used codes themselves are also placed in the table so that the
16 * decompresion can work without "special cases".
17 * Applied to kernel symbols, this usually produces a compression ratio
18 * of about 50%.
19 *
20 */
21
22#include <errno.h>
23#include <getopt.h>
24#include <stdbool.h>
25#include <stdio.h>
26#include <stdlib.h>
27#include <string.h>
28#include <ctype.h>
29#include <limits.h>
30
31#define ARRAY_SIZE(arr) (sizeof(arr) / sizeof(arr[0]))
32
33#define KSYM_NAME_LEN 512
34
35struct sym_entry {
36 unsigned long long addr;
37 unsigned int len;
38 unsigned int seq;
39 unsigned int start_pos;
40 unsigned int 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;
66static int base_relative;
67static int lto_clang;
68
69static int token_profit[0x10000];
70
71/* the table that holds the result of the compression */
72static unsigned char best_table[256][2];
73static unsigned char best_table_len[256];
74
75
76static void usage(void)
77{
78 fprintf(stderr, "Usage: kallsyms [--all-symbols] [--absolute-percpu] "
79 "[--base-relative] [--lto-clang] in.map > out.S\n");
80 exit(1);
81}
82
83static char *sym_name(const struct sym_entry *s)
84{
85 return (char *)s->sym + 1;
86}
87
88static bool is_ignored_symbol(const char *name, char type)
89{
90 if (type == 'u' || type == 'n')
91 return true;
92
93 if (toupper(type) == 'A') {
94 /* Keep these useful absolute symbols */
95 if (strcmp(name, "__kernel_syscall_via_break") &&
96 strcmp(name, "__kernel_syscall_via_epc") &&
97 strcmp(name, "__kernel_sigtramp") &&
98 strcmp(name, "__gp"))
99 return true;
100 }
101
102 return false;
103}
104
105static void check_symbol_range(const char *sym, unsigned long long addr,
106 struct addr_range *ranges, int entries)
107{
108 size_t i;
109 struct addr_range *ar;
110
111 for (i = 0; i < entries; ++i) {
112 ar = &ranges[i];
113
114 if (strcmp(sym, ar->start_sym) == 0) {
115 ar->start = addr;
116 return;
117 } else if (strcmp(sym, ar->end_sym) == 0) {
118 ar->end = addr;
119 return;
120 }
121 }
122}
123
124static struct sym_entry *read_symbol(FILE *in, char **buf, size_t *buf_len)
125{
126 char *name, type, *p;
127 unsigned long long addr;
128 size_t len;
129 ssize_t readlen;
130 struct sym_entry *sym;
131
132 errno = 0;
133 readlen = getline(buf, buf_len, in);
134 if (readlen < 0) {
135 if (errno) {
136 perror("read_symbol");
137 exit(EXIT_FAILURE);
138 }
139 return NULL;
140 }
141
142 if ((*buf)[readlen - 1] == '\n')
143 (*buf)[readlen - 1] = 0;
144
145 addr = strtoull(*buf, &p, 16);
146
147 if (*buf == p || *p++ != ' ' || !isascii((type = *p++)) || *p++ != ' ') {
148 fprintf(stderr, "line format error\n");
149 exit(EXIT_FAILURE);
150 }
151
152 name = p;
153 len = strlen(name);
154
155 if (len >= KSYM_NAME_LEN) {
156 fprintf(stderr, "Symbol %s too long for kallsyms (%zu >= %d).\n"
157 "Please increase KSYM_NAME_LEN both in kernel and kallsyms.c\n",
158 name, len, KSYM_NAME_LEN);
159 return NULL;
160 }
161
162 if (strcmp(name, "_text") == 0)
163 _text = addr;
164
165 /* Ignore most absolute/undefined (?) symbols. */
166 if (is_ignored_symbol(name, type))
167 return NULL;
168
169 check_symbol_range(name, addr, text_ranges, ARRAY_SIZE(text_ranges));
170 check_symbol_range(name, addr, &percpu_range, 1);
171
172 /* include the type field in the symbol name, so that it gets
173 * compressed together */
174 len++;
175
176 sym = malloc(sizeof(*sym) + len + 1);
177 if (!sym) {
178 fprintf(stderr, "kallsyms failure: "
179 "unable to allocate required amount of memory\n");
180 exit(EXIT_FAILURE);
181 }
182 sym->addr = addr;
183 sym->len = len;
184 sym->sym[0] = type;
185 strcpy(sym_name(sym), name);
186 sym->percpu_absolute = 0;
187
188 return sym;
189}
190
191static int symbol_in_range(const struct sym_entry *s,
192 const struct addr_range *ranges, int entries)
193{
194 size_t i;
195 const struct addr_range *ar;
196
197 for (i = 0; i < entries; ++i) {
198 ar = &ranges[i];
199
200 if (s->addr >= ar->start && s->addr <= ar->end)
201 return 1;
202 }
203
204 return 0;
205}
206
207static int symbol_valid(const struct sym_entry *s)
208{
209 const char *name = sym_name(s);
210
211 /* if --all-symbols is not specified, then symbols outside the text
212 * and inittext sections are discarded */
213 if (!all_symbols) {
214 if (symbol_in_range(s, text_ranges,
215 ARRAY_SIZE(text_ranges)) == 0)
216 return 0;
217 /* Corner case. Discard any symbols with the same value as
218 * _etext _einittext; they can move between pass 1 and 2 when
219 * the kallsyms data are added. If these symbols move then
220 * they may get dropped in pass 2, which breaks the kallsyms
221 * rules.
222 */
223 if ((s->addr == text_range_text->end &&
224 strcmp(name, text_range_text->end_sym)) ||
225 (s->addr == text_range_inittext->end &&
226 strcmp(name, text_range_inittext->end_sym)))
227 return 0;
228 }
229
230 return 1;
231}
232
233/* remove all the invalid symbols from the table */
234static void shrink_table(void)
235{
236 unsigned int i, pos;
237
238 pos = 0;
239 for (i = 0; i < table_cnt; i++) {
240 if (symbol_valid(table[i])) {
241 if (pos != i)
242 table[pos] = table[i];
243 pos++;
244 } else {
245 free(table[i]);
246 }
247 }
248 table_cnt = pos;
249
250 /* When valid symbol is not registered, exit to error */
251 if (!table_cnt) {
252 fprintf(stderr, "No valid symbol.\n");
253 exit(1);
254 }
255}
256
257static void read_map(const char *in)
258{
259 FILE *fp;
260 struct sym_entry *sym;
261 char *buf = NULL;
262 size_t buflen = 0;
263
264 fp = fopen(in, "r");
265 if (!fp) {
266 perror(in);
267 exit(1);
268 }
269
270 while (!feof(fp)) {
271 sym = read_symbol(fp, &buf, &buflen);
272 if (!sym)
273 continue;
274
275 sym->start_pos = table_cnt;
276
277 if (table_cnt >= table_size) {
278 table_size += 10000;
279 table = realloc(table, sizeof(*table) * table_size);
280 if (!table) {
281 fprintf(stderr, "out of memory\n");
282 fclose(fp);
283 exit (1);
284 }
285 }
286
287 table[table_cnt++] = sym;
288 }
289
290 free(buf);
291 fclose(fp);
292}
293
294static void output_label(const char *label)
295{
296 printf(".globl %s\n", label);
297 printf("\tALGN\n");
298 printf("%s:\n", label);
299}
300
301/* Provide proper symbols relocatability by their '_text' relativeness. */
302static void output_address(unsigned long long addr)
303{
304 if (_text <= addr)
305 printf("\tPTR\t_text + %#llx\n", addr - _text);
306 else
307 printf("\tPTR\t_text - %#llx\n", _text - addr);
308}
309
310/* uncompress a compressed symbol. When this function is called, the best table
311 * might still be compressed itself, so the function needs to be recursive */
312static int expand_symbol(const unsigned char *data, int len, char *result)
313{
314 int c, rlen, total=0;
315
316 while (len) {
317 c = *data;
318 /* if the table holds a single char that is the same as the one
319 * we are looking for, then end the search */
320 if (best_table[c][0]==c && best_table_len[c]==1) {
321 *result++ = c;
322 total++;
323 } else {
324 /* if not, recurse and expand */
325 rlen = expand_symbol(best_table[c], best_table_len[c], result);
326 total += rlen;
327 result += rlen;
328 }
329 data++;
330 len--;
331 }
332 *result=0;
333
334 return total;
335}
336
337static int symbol_absolute(const struct sym_entry *s)
338{
339 return s->percpu_absolute;
340}
341
342static void cleanup_symbol_name(char *s)
343{
344 char *p;
345
346 /*
347 * ASCII[.] = 2e
348 * ASCII[0-9] = 30,39
349 * ASCII[A-Z] = 41,5a
350 * ASCII[_] = 5f
351 * ASCII[a-z] = 61,7a
352 *
353 * As above, replacing the first '.' in ".llvm." with '\0' does not
354 * affect the main sorting, but it helps us with subsorting.
355 */
356 p = strstr(s, ".llvm.");
357 if (p)
358 *p = '\0';
359}
360
361static int compare_names(const void *a, const void *b)
362{
363 int ret;
364 const struct sym_entry *sa = *(const struct sym_entry **)a;
365 const struct sym_entry *sb = *(const struct sym_entry **)b;
366
367 ret = strcmp(sym_name(sa), sym_name(sb));
368 if (!ret) {
369 if (sa->addr > sb->addr)
370 return 1;
371 else if (sa->addr < sb->addr)
372 return -1;
373
374 /* keep old order */
375 return (int)(sa->seq - sb->seq);
376 }
377
378 return ret;
379}
380
381static void sort_symbols_by_name(void)
382{
383 qsort(table, table_cnt, sizeof(table[0]), compare_names);
384}
385
386static void write_src(void)
387{
388 unsigned int i, k, off;
389 unsigned int best_idx[256];
390 unsigned int *markers;
391 char buf[KSYM_NAME_LEN];
392
393 printf("#include <asm/bitsperlong.h>\n");
394 printf("#if BITS_PER_LONG == 64\n");
395 printf("#define PTR .quad\n");
396 printf("#define ALGN .balign 8\n");
397 printf("#else\n");
398 printf("#define PTR .long\n");
399 printf("#define ALGN .balign 4\n");
400 printf("#endif\n");
401
402 printf("\t.section .rodata, \"a\"\n");
403
404 output_label("kallsyms_num_syms");
405 printf("\t.long\t%u\n", table_cnt);
406 printf("\n");
407
408 /* table of offset markers, that give the offset in the compressed stream
409 * every 256 symbols */
410 markers = malloc(sizeof(unsigned int) * ((table_cnt + 255) / 256));
411 if (!markers) {
412 fprintf(stderr, "kallsyms failure: "
413 "unable to allocate required memory\n");
414 exit(EXIT_FAILURE);
415 }
416
417 output_label("kallsyms_names");
418 off = 0;
419 for (i = 0; i < table_cnt; i++) {
420 if ((i & 0xFF) == 0)
421 markers[i >> 8] = off;
422 table[i]->seq = i;
423
424 /* There cannot be any symbol of length zero. */
425 if (table[i]->len == 0) {
426 fprintf(stderr, "kallsyms failure: "
427 "unexpected zero symbol length\n");
428 exit(EXIT_FAILURE);
429 }
430
431 /* Only lengths that fit in up-to-two-byte ULEB128 are supported. */
432 if (table[i]->len > 0x3FFF) {
433 fprintf(stderr, "kallsyms failure: "
434 "unexpected huge symbol length\n");
435 exit(EXIT_FAILURE);
436 }
437
438 /* Encode length with ULEB128. */
439 if (table[i]->len <= 0x7F) {
440 /* Most symbols use a single byte for the length. */
441 printf("\t.byte 0x%02x", table[i]->len);
442 off += table[i]->len + 1;
443 } else {
444 /* "Big" symbols use two bytes. */
445 printf("\t.byte 0x%02x, 0x%02x",
446 (table[i]->len & 0x7F) | 0x80,
447 (table[i]->len >> 7) & 0x7F);
448 off += table[i]->len + 2;
449 }
450 for (k = 0; k < table[i]->len; k++)
451 printf(", 0x%02x", table[i]->sym[k]);
452 printf("\n");
453 }
454 printf("\n");
455
456 /*
457 * Now that we wrote out the compressed symbol names, restore the
458 * original names, which are needed in some of the later steps.
459 */
460 for (i = 0; i < table_cnt; i++) {
461 expand_symbol(table[i]->sym, table[i]->len, buf);
462 strcpy((char *)table[i]->sym, buf);
463 }
464
465 output_label("kallsyms_markers");
466 for (i = 0; i < ((table_cnt + 255) >> 8); i++)
467 printf("\t.long\t%u\n", markers[i]);
468 printf("\n");
469
470 free(markers);
471
472 output_label("kallsyms_token_table");
473 off = 0;
474 for (i = 0; i < 256; i++) {
475 best_idx[i] = off;
476 expand_symbol(best_table[i], best_table_len[i], buf);
477 printf("\t.asciz\t\"%s\"\n", buf);
478 off += strlen(buf) + 1;
479 }
480 printf("\n");
481
482 output_label("kallsyms_token_index");
483 for (i = 0; i < 256; i++)
484 printf("\t.short\t%d\n", best_idx[i]);
485 printf("\n");
486
487 if (!base_relative)
488 output_label("kallsyms_addresses");
489 else
490 output_label("kallsyms_offsets");
491
492 for (i = 0; i < table_cnt; i++) {
493 if (base_relative) {
494 /*
495 * Use the offset relative to the lowest value
496 * encountered of all relative symbols, and emit
497 * non-relocatable fixed offsets that will be fixed
498 * up at runtime.
499 */
500
501 long long offset;
502 int overflow;
503
504 if (!absolute_percpu) {
505 offset = table[i]->addr - relative_base;
506 overflow = (offset < 0 || offset > UINT_MAX);
507 } else if (symbol_absolute(table[i])) {
508 offset = table[i]->addr;
509 overflow = (offset < 0 || offset > INT_MAX);
510 } else {
511 offset = relative_base - table[i]->addr - 1;
512 overflow = (offset < INT_MIN || offset >= 0);
513 }
514 if (overflow) {
515 fprintf(stderr, "kallsyms failure: "
516 "%s symbol value %#llx out of range in relative mode\n",
517 symbol_absolute(table[i]) ? "absolute" : "relative",
518 table[i]->addr);
519 exit(EXIT_FAILURE);
520 }
521 printf("\t.long\t%#x /* %s */\n", (int)offset, table[i]->sym);
522 } else if (!symbol_absolute(table[i])) {
523 output_address(table[i]->addr);
524 } else {
525 printf("\tPTR\t%#llx\n", table[i]->addr);
526 }
527 }
528 printf("\n");
529
530 if (base_relative) {
531 output_label("kallsyms_relative_base");
532 output_address(relative_base);
533 printf("\n");
534 }
535
536 if (lto_clang)
537 for (i = 0; i < table_cnt; i++)
538 cleanup_symbol_name((char *)table[i]->sym);
539
540 sort_symbols_by_name();
541 output_label("kallsyms_seqs_of_names");
542 for (i = 0; i < table_cnt; i++)
543 printf("\t.byte 0x%02x, 0x%02x, 0x%02x\n",
544 (unsigned char)(table[i]->seq >> 16),
545 (unsigned char)(table[i]->seq >> 8),
546 (unsigned char)(table[i]->seq >> 0));
547 printf("\n");
548}
549
550
551/* table lookup compression functions */
552
553/* count all the possible tokens in a symbol */
554static void learn_symbol(const unsigned char *symbol, int len)
555{
556 int i;
557
558 for (i = 0; i < len - 1; i++)
559 token_profit[ symbol[i] + (symbol[i + 1] << 8) ]++;
560}
561
562/* decrease the count for all the possible tokens in a symbol */
563static void forget_symbol(const unsigned char *symbol, int len)
564{
565 int i;
566
567 for (i = 0; i < len - 1; i++)
568 token_profit[ symbol[i] + (symbol[i + 1] << 8) ]--;
569}
570
571/* do the initial token count */
572static void build_initial_token_table(void)
573{
574 unsigned int i;
575
576 for (i = 0; i < table_cnt; i++)
577 learn_symbol(table[i]->sym, table[i]->len);
578}
579
580static unsigned char *find_token(unsigned char *str, int len,
581 const unsigned char *token)
582{
583 int i;
584
585 for (i = 0; i < len - 1; i++) {
586 if (str[i] == token[0] && str[i+1] == token[1])
587 return &str[i];
588 }
589 return NULL;
590}
591
592/* replace a given token in all the valid symbols. Use the sampled symbols
593 * to update the counts */
594static void compress_symbols(const unsigned char *str, int idx)
595{
596 unsigned int i, len, size;
597 unsigned char *p1, *p2;
598
599 for (i = 0; i < table_cnt; i++) {
600
601 len = table[i]->len;
602 p1 = table[i]->sym;
603
604 /* find the token on the symbol */
605 p2 = find_token(p1, len, str);
606 if (!p2) continue;
607
608 /* decrease the counts for this symbol's tokens */
609 forget_symbol(table[i]->sym, len);
610
611 size = len;
612
613 do {
614 *p2 = idx;
615 p2++;
616 size -= (p2 - p1);
617 memmove(p2, p2 + 1, size);
618 p1 = p2;
619 len--;
620
621 if (size < 2) break;
622
623 /* find the token on the symbol */
624 p2 = find_token(p1, size, str);
625
626 } while (p2);
627
628 table[i]->len = len;
629
630 /* increase the counts for this symbol's new tokens */
631 learn_symbol(table[i]->sym, len);
632 }
633}
634
635/* search the token with the maximum profit */
636static int find_best_token(void)
637{
638 int i, best, bestprofit;
639
640 bestprofit=-10000;
641 best = 0;
642
643 for (i = 0; i < 0x10000; i++) {
644 if (token_profit[i] > bestprofit) {
645 best = i;
646 bestprofit = token_profit[i];
647 }
648 }
649 return best;
650}
651
652/* this is the core of the algorithm: calculate the "best" table */
653static void optimize_result(void)
654{
655 int i, best;
656
657 /* using the '\0' symbol last allows compress_symbols to use standard
658 * fast string functions */
659 for (i = 255; i >= 0; i--) {
660
661 /* if this table slot is empty (it is not used by an actual
662 * original char code */
663 if (!best_table_len[i]) {
664
665 /* find the token with the best profit value */
666 best = find_best_token();
667 if (token_profit[best] == 0)
668 break;
669
670 /* place it in the "best" table */
671 best_table_len[i] = 2;
672 best_table[i][0] = best & 0xFF;
673 best_table[i][1] = (best >> 8) & 0xFF;
674
675 /* replace this token in all the valid symbols */
676 compress_symbols(best_table[i], i);
677 }
678 }
679}
680
681/* start by placing the symbols that are actually used on the table */
682static void insert_real_symbols_in_table(void)
683{
684 unsigned int i, j, c;
685
686 for (i = 0; i < table_cnt; i++) {
687 for (j = 0; j < table[i]->len; j++) {
688 c = table[i]->sym[j];
689 best_table[c][0]=c;
690 best_table_len[c]=1;
691 }
692 }
693}
694
695static void optimize_token_table(void)
696{
697 build_initial_token_table();
698
699 insert_real_symbols_in_table();
700
701 optimize_result();
702}
703
704/* guess for "linker script provide" symbol */
705static int may_be_linker_script_provide_symbol(const struct sym_entry *se)
706{
707 const char *symbol = sym_name(se);
708 int len = se->len - 1;
709
710 if (len < 8)
711 return 0;
712
713 if (symbol[0] != '_' || symbol[1] != '_')
714 return 0;
715
716 /* __start_XXXXX */
717 if (!memcmp(symbol + 2, "start_", 6))
718 return 1;
719
720 /* __stop_XXXXX */
721 if (!memcmp(symbol + 2, "stop_", 5))
722 return 1;
723
724 /* __end_XXXXX */
725 if (!memcmp(symbol + 2, "end_", 4))
726 return 1;
727
728 /* __XXXXX_start */
729 if (!memcmp(symbol + len - 6, "_start", 6))
730 return 1;
731
732 /* __XXXXX_end */
733 if (!memcmp(symbol + len - 4, "_end", 4))
734 return 1;
735
736 return 0;
737}
738
739static int compare_symbols(const void *a, const void *b)
740{
741 const struct sym_entry *sa = *(const struct sym_entry **)a;
742 const struct sym_entry *sb = *(const struct sym_entry **)b;
743 int wa, wb;
744
745 /* sort by address first */
746 if (sa->addr > sb->addr)
747 return 1;
748 if (sa->addr < sb->addr)
749 return -1;
750
751 /* sort by "weakness" type */
752 wa = (sa->sym[0] == 'w') || (sa->sym[0] == 'W');
753 wb = (sb->sym[0] == 'w') || (sb->sym[0] == 'W');
754 if (wa != wb)
755 return wa - wb;
756
757 /* sort by "linker script provide" type */
758 wa = may_be_linker_script_provide_symbol(sa);
759 wb = may_be_linker_script_provide_symbol(sb);
760 if (wa != wb)
761 return wa - wb;
762
763 /* sort by the number of prefix underscores */
764 wa = strspn(sym_name(sa), "_");
765 wb = strspn(sym_name(sb), "_");
766 if (wa != wb)
767 return wa - wb;
768
769 /* sort by initial order, so that other symbols are left undisturbed */
770 return sa->start_pos - sb->start_pos;
771}
772
773static void sort_symbols(void)
774{
775 qsort(table, table_cnt, sizeof(table[0]), compare_symbols);
776}
777
778static void make_percpus_absolute(void)
779{
780 unsigned int i;
781
782 for (i = 0; i < table_cnt; i++)
783 if (symbol_in_range(table[i], &percpu_range, 1)) {
784 /*
785 * Keep the 'A' override for percpu symbols to
786 * ensure consistent behavior compared to older
787 * versions of this tool.
788 */
789 table[i]->sym[0] = 'A';
790 table[i]->percpu_absolute = 1;
791 }
792}
793
794/* find the minimum non-absolute symbol address */
795static void record_relative_base(void)
796{
797 unsigned int i;
798
799 for (i = 0; i < table_cnt; i++)
800 if (!symbol_absolute(table[i])) {
801 /*
802 * The table is sorted by address.
803 * Take the first non-absolute symbol value.
804 */
805 relative_base = table[i]->addr;
806 return;
807 }
808}
809
810int main(int argc, char **argv)
811{
812 while (1) {
813 static const struct option long_options[] = {
814 {"all-symbols", no_argument, &all_symbols, 1},
815 {"absolute-percpu", no_argument, &absolute_percpu, 1},
816 {"base-relative", no_argument, &base_relative, 1},
817 {"lto-clang", no_argument, <o_clang, 1},
818 {},
819 };
820
821 int c = getopt_long(argc, argv, "", long_options, NULL);
822
823 if (c == -1)
824 break;
825 if (c != 0)
826 usage();
827 }
828
829 if (optind >= argc)
830 usage();
831
832 read_map(argv[optind]);
833 shrink_table();
834 if (absolute_percpu)
835 make_percpus_absolute();
836 sort_symbols();
837 if (base_relative)
838 record_relative_base();
839 optimize_token_table();
840 write_src();
841
842 return 0;
843}