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: kallsyms [--all-symbols] [--absolute-percpu]
9 * [--base-relative] 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 <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#define ARRAY_SIZE(arr) (sizeof(arr) / sizeof(arr[0]))
31
32#define _stringify_1(x) #x
33#define _stringify(x) _stringify_1(x)
34
35#define KSYM_NAME_LEN 512
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
50struct sym_entry {
51 unsigned long long addr;
52 unsigned int len;
53 unsigned int seq;
54 unsigned int start_pos;
55 unsigned int percpu_absolute;
56 unsigned char sym[];
57};
58
59struct addr_range {
60 const char *start_sym, *end_sym;
61 unsigned long long start, end;
62};
63
64static unsigned long long _text;
65static unsigned long long relative_base;
66static struct addr_range text_ranges[] = {
67 { "_stext", "_etext" },
68 { "_sinittext", "_einittext" },
69};
70#define text_range_text (&text_ranges[0])
71#define text_range_inittext (&text_ranges[1])
72
73static struct addr_range percpu_range = {
74 "__per_cpu_start", "__per_cpu_end", -1ULL, 0
75};
76
77static struct sym_entry **table;
78static unsigned int table_size, table_cnt;
79static int all_symbols;
80static int absolute_percpu;
81static int base_relative;
82static int lto_clang;
83
84static int token_profit[0x10000];
85
86/* the table that holds the result of the compression */
87static unsigned char best_table[256][2];
88static unsigned char best_table_len[256];
89
90
91static void usage(void)
92{
93 fprintf(stderr, "Usage: kallsyms [--all-symbols] [--absolute-percpu] "
94 "[--base-relative] [--lto-clang] in.map > out.S\n");
95 exit(1);
96}
97
98static char *sym_name(const struct sym_entry *s)
99{
100 return (char *)s->sym + 1;
101}
102
103static bool is_ignored_symbol(const char *name, char type)
104{
105 /* Symbol names that exactly match to the following are ignored.*/
106 static const char * const ignored_symbols[] = {
107 /*
108 * Symbols which vary between passes. Passes 1 and 2 must have
109 * identical symbol lists. The kallsyms_* symbols below are
110 * only added after pass 1, they would be included in pass 2
111 * when --all-symbols is specified so exclude them to get a
112 * stable symbol list.
113 */
114 "kallsyms_addresses",
115 "kallsyms_offsets",
116 "kallsyms_relative_base",
117 "kallsyms_num_syms",
118 "kallsyms_names",
119 "kallsyms_markers",
120 "kallsyms_token_table",
121 "kallsyms_token_index",
122 /* Exclude linker generated symbols which vary between passes */
123 "_SDA_BASE_", /* ppc */
124 "_SDA2_BASE_", /* ppc */
125 NULL
126 };
127
128 /* Symbol names that begin with the following are ignored.*/
129 static const char * const ignored_prefixes[] = {
130 "__efistub_", /* arm64 EFI stub namespace */
131 "__kvm_nvhe_$", /* arm64 local symbols in non-VHE KVM namespace */
132 "__kvm_nvhe_.L", /* arm64 local symbols in non-VHE KVM namespace */
133 "__AArch64ADRPThunk_", /* arm64 lld */
134 "__ARMV5PILongThunk_", /* arm lld */
135 "__ARMV7PILongThunk_",
136 "__ThumbV7PILongThunk_",
137 "__LA25Thunk_", /* mips lld */
138 "__microLA25Thunk_",
139 "__kcfi_typeid_", /* CFI type identifiers */
140 NULL
141 };
142
143 /* Symbol names that end with the following are ignored.*/
144 static const char * const ignored_suffixes[] = {
145 "_from_arm", /* arm */
146 "_from_thumb", /* arm */
147 "_veneer", /* arm */
148 NULL
149 };
150
151 /* Symbol names that contain the following are ignored.*/
152 static const char * const ignored_matches[] = {
153 ".long_branch.", /* ppc stub */
154 ".plt_branch.", /* ppc stub */
155 NULL
156 };
157
158 const char * const *p;
159
160 for (p = ignored_symbols; *p; p++)
161 if (!strcmp(name, *p))
162 return true;
163
164 for (p = ignored_prefixes; *p; p++)
165 if (!strncmp(name, *p, strlen(*p)))
166 return true;
167
168 for (p = ignored_suffixes; *p; p++) {
169 int l = strlen(name) - strlen(*p);
170
171 if (l >= 0 && !strcmp(name + l, *p))
172 return true;
173 }
174
175 for (p = ignored_matches; *p; p++) {
176 if (strstr(name, *p))
177 return true;
178 }
179
180 if (type == 'U' || type == 'u')
181 return true;
182 /* exclude debugging symbols */
183 if (type == 'N' || type == 'n')
184 return true;
185
186 if (toupper(type) == 'A') {
187 /* Keep these useful absolute symbols */
188 if (strcmp(name, "__kernel_syscall_via_break") &&
189 strcmp(name, "__kernel_syscall_via_epc") &&
190 strcmp(name, "__kernel_sigtramp") &&
191 strcmp(name, "__gp"))
192 return true;
193 }
194
195 return false;
196}
197
198static void check_symbol_range(const char *sym, unsigned long long addr,
199 struct addr_range *ranges, int entries)
200{
201 size_t i;
202 struct addr_range *ar;
203
204 for (i = 0; i < entries; ++i) {
205 ar = &ranges[i];
206
207 if (strcmp(sym, ar->start_sym) == 0) {
208 ar->start = addr;
209 return;
210 } else if (strcmp(sym, ar->end_sym) == 0) {
211 ar->end = addr;
212 return;
213 }
214 }
215}
216
217static struct sym_entry *read_symbol(FILE *in)
218{
219 char name[KSYM_NAME_LEN_BUFFER+1], type;
220 unsigned long long addr;
221 unsigned int len;
222 struct sym_entry *sym;
223 int rc;
224
225 rc = fscanf(in, "%llx %c %" _stringify(KSYM_NAME_LEN_BUFFER) "s\n", &addr, &type, name);
226 if (rc != 3) {
227 if (rc != EOF && fgets(name, ARRAY_SIZE(name), in) == NULL)
228 fprintf(stderr, "Read error or end of file.\n");
229 return NULL;
230 }
231 if (strlen(name) >= KSYM_NAME_LEN) {
232 fprintf(stderr, "Symbol %s too long for kallsyms (%zu >= %d).\n"
233 "Please increase KSYM_NAME_LEN both in kernel and kallsyms.c\n",
234 name, strlen(name), KSYM_NAME_LEN);
235 return NULL;
236 }
237
238 if (strcmp(name, "_text") == 0)
239 _text = addr;
240
241 /* Ignore most absolute/undefined (?) symbols. */
242 if (is_ignored_symbol(name, type))
243 return NULL;
244
245 check_symbol_range(name, addr, text_ranges, ARRAY_SIZE(text_ranges));
246 check_symbol_range(name, addr, &percpu_range, 1);
247
248 /* include the type field in the symbol name, so that it gets
249 * compressed together */
250
251 len = strlen(name) + 1;
252
253 sym = malloc(sizeof(*sym) + len + 1);
254 if (!sym) {
255 fprintf(stderr, "kallsyms failure: "
256 "unable to allocate required amount of memory\n");
257 exit(EXIT_FAILURE);
258 }
259 sym->addr = addr;
260 sym->len = len;
261 sym->sym[0] = type;
262 strcpy(sym_name(sym), name);
263 sym->percpu_absolute = 0;
264
265 return sym;
266}
267
268static int symbol_in_range(const struct sym_entry *s,
269 const struct addr_range *ranges, int entries)
270{
271 size_t i;
272 const struct addr_range *ar;
273
274 for (i = 0; i < entries; ++i) {
275 ar = &ranges[i];
276
277 if (s->addr >= ar->start && s->addr <= ar->end)
278 return 1;
279 }
280
281 return 0;
282}
283
284static int symbol_valid(const struct sym_entry *s)
285{
286 const char *name = sym_name(s);
287
288 /* if --all-symbols is not specified, then symbols outside the text
289 * and inittext sections are discarded */
290 if (!all_symbols) {
291 if (symbol_in_range(s, text_ranges,
292 ARRAY_SIZE(text_ranges)) == 0)
293 return 0;
294 /* Corner case. Discard any symbols with the same value as
295 * _etext _einittext; they can move between pass 1 and 2 when
296 * the kallsyms data are added. If these symbols move then
297 * they may get dropped in pass 2, which breaks the kallsyms
298 * rules.
299 */
300 if ((s->addr == text_range_text->end &&
301 strcmp(name, text_range_text->end_sym)) ||
302 (s->addr == text_range_inittext->end &&
303 strcmp(name, text_range_inittext->end_sym)))
304 return 0;
305 }
306
307 return 1;
308}
309
310/* remove all the invalid symbols from the table */
311static void shrink_table(void)
312{
313 unsigned int i, pos;
314
315 pos = 0;
316 for (i = 0; i < table_cnt; i++) {
317 if (symbol_valid(table[i])) {
318 if (pos != i)
319 table[pos] = table[i];
320 pos++;
321 } else {
322 free(table[i]);
323 }
324 }
325 table_cnt = pos;
326
327 /* When valid symbol is not registered, exit to error */
328 if (!table_cnt) {
329 fprintf(stderr, "No valid symbol.\n");
330 exit(1);
331 }
332}
333
334static void read_map(const char *in)
335{
336 FILE *fp;
337 struct sym_entry *sym;
338
339 fp = fopen(in, "r");
340 if (!fp) {
341 perror(in);
342 exit(1);
343 }
344
345 while (!feof(fp)) {
346 sym = read_symbol(fp);
347 if (!sym)
348 continue;
349
350 sym->start_pos = table_cnt;
351
352 if (table_cnt >= table_size) {
353 table_size += 10000;
354 table = realloc(table, sizeof(*table) * table_size);
355 if (!table) {
356 fprintf(stderr, "out of memory\n");
357 fclose(fp);
358 exit (1);
359 }
360 }
361
362 table[table_cnt++] = sym;
363 }
364
365 fclose(fp);
366}
367
368static void output_label(const char *label)
369{
370 printf(".globl %s\n", label);
371 printf("\tALGN\n");
372 printf("%s:\n", label);
373}
374
375/* Provide proper symbols relocatability by their '_text' relativeness. */
376static void output_address(unsigned long long addr)
377{
378 if (_text <= addr)
379 printf("\tPTR\t_text + %#llx\n", addr - _text);
380 else
381 printf("\tPTR\t_text - %#llx\n", _text - addr);
382}
383
384/* uncompress a compressed symbol. When this function is called, the best table
385 * might still be compressed itself, so the function needs to be recursive */
386static int expand_symbol(const unsigned char *data, int len, char *result)
387{
388 int c, rlen, total=0;
389
390 while (len) {
391 c = *data;
392 /* if the table holds a single char that is the same as the one
393 * we are looking for, then end the search */
394 if (best_table[c][0]==c && best_table_len[c]==1) {
395 *result++ = c;
396 total++;
397 } else {
398 /* if not, recurse and expand */
399 rlen = expand_symbol(best_table[c], best_table_len[c], result);
400 total += rlen;
401 result += rlen;
402 }
403 data++;
404 len--;
405 }
406 *result=0;
407
408 return total;
409}
410
411static int symbol_absolute(const struct sym_entry *s)
412{
413 return s->percpu_absolute;
414}
415
416static char * s_name(char *buf)
417{
418 /* Skip the symbol type */
419 return buf + 1;
420}
421
422static void cleanup_symbol_name(char *s)
423{
424 char *p;
425
426 if (!lto_clang)
427 return;
428
429 /*
430 * ASCII[.] = 2e
431 * ASCII[0-9] = 30,39
432 * ASCII[A-Z] = 41,5a
433 * ASCII[_] = 5f
434 * ASCII[a-z] = 61,7a
435 *
436 * As above, replacing '.' with '\0' does not affect the main sorting,
437 * but it helps us with subsorting.
438 */
439 p = strchr(s, '.');
440 if (p)
441 *p = '\0';
442}
443
444static int compare_names(const void *a, const void *b)
445{
446 int ret;
447 char sa_namebuf[KSYM_NAME_LEN];
448 char sb_namebuf[KSYM_NAME_LEN];
449 const struct sym_entry *sa = *(const struct sym_entry **)a;
450 const struct sym_entry *sb = *(const struct sym_entry **)b;
451
452 expand_symbol(sa->sym, sa->len, sa_namebuf);
453 expand_symbol(sb->sym, sb->len, sb_namebuf);
454 cleanup_symbol_name(s_name(sa_namebuf));
455 cleanup_symbol_name(s_name(sb_namebuf));
456 ret = strcmp(s_name(sa_namebuf), s_name(sb_namebuf));
457 if (!ret) {
458 if (sa->addr > sb->addr)
459 return 1;
460 else if (sa->addr < sb->addr)
461 return -1;
462
463 /* keep old order */
464 return (int)(sa->seq - sb->seq);
465 }
466
467 return ret;
468}
469
470static void sort_symbols_by_name(void)
471{
472 qsort(table, table_cnt, sizeof(table[0]), compare_names);
473}
474
475static void write_src(void)
476{
477 unsigned int i, k, off;
478 unsigned int best_idx[256];
479 unsigned int *markers;
480 char buf[KSYM_NAME_LEN];
481
482 printf("#include <asm/bitsperlong.h>\n");
483 printf("#if BITS_PER_LONG == 64\n");
484 printf("#define PTR .quad\n");
485 printf("#define ALGN .balign 8\n");
486 printf("#else\n");
487 printf("#define PTR .long\n");
488 printf("#define ALGN .balign 4\n");
489 printf("#endif\n");
490
491 printf("\t.section .rodata, \"a\"\n");
492
493 if (!base_relative)
494 output_label("kallsyms_addresses");
495 else
496 output_label("kallsyms_offsets");
497
498 for (i = 0; i < table_cnt; i++) {
499 if (base_relative) {
500 /*
501 * Use the offset relative to the lowest value
502 * encountered of all relative symbols, and emit
503 * non-relocatable fixed offsets that will be fixed
504 * up at runtime.
505 */
506
507 long long offset;
508 int overflow;
509
510 if (!absolute_percpu) {
511 offset = table[i]->addr - relative_base;
512 overflow = (offset < 0 || offset > UINT_MAX);
513 } else if (symbol_absolute(table[i])) {
514 offset = table[i]->addr;
515 overflow = (offset < 0 || offset > INT_MAX);
516 } else {
517 offset = relative_base - table[i]->addr - 1;
518 overflow = (offset < INT_MIN || offset >= 0);
519 }
520 if (overflow) {
521 fprintf(stderr, "kallsyms failure: "
522 "%s symbol value %#llx out of range in relative mode\n",
523 symbol_absolute(table[i]) ? "absolute" : "relative",
524 table[i]->addr);
525 exit(EXIT_FAILURE);
526 }
527 printf("\t.long\t%#x\n", (int)offset);
528 } else if (!symbol_absolute(table[i])) {
529 output_address(table[i]->addr);
530 } else {
531 printf("\tPTR\t%#llx\n", table[i]->addr);
532 }
533 }
534 printf("\n");
535
536 if (base_relative) {
537 output_label("kallsyms_relative_base");
538 output_address(relative_base);
539 printf("\n");
540 }
541
542 output_label("kallsyms_num_syms");
543 printf("\t.long\t%u\n", table_cnt);
544 printf("\n");
545
546 /* table of offset markers, that give the offset in the compressed stream
547 * every 256 symbols */
548 markers = malloc(sizeof(unsigned int) * ((table_cnt + 255) / 256));
549 if (!markers) {
550 fprintf(stderr, "kallsyms failure: "
551 "unable to allocate required memory\n");
552 exit(EXIT_FAILURE);
553 }
554
555 output_label("kallsyms_names");
556 off = 0;
557 for (i = 0; i < table_cnt; i++) {
558 if ((i & 0xFF) == 0)
559 markers[i >> 8] = off;
560 table[i]->seq = i;
561
562 /* There cannot be any symbol of length zero. */
563 if (table[i]->len == 0) {
564 fprintf(stderr, "kallsyms failure: "
565 "unexpected zero symbol length\n");
566 exit(EXIT_FAILURE);
567 }
568
569 /* Only lengths that fit in up-to-two-byte ULEB128 are supported. */
570 if (table[i]->len > 0x3FFF) {
571 fprintf(stderr, "kallsyms failure: "
572 "unexpected huge symbol length\n");
573 exit(EXIT_FAILURE);
574 }
575
576 /* Encode length with ULEB128. */
577 if (table[i]->len <= 0x7F) {
578 /* Most symbols use a single byte for the length. */
579 printf("\t.byte 0x%02x", table[i]->len);
580 off += table[i]->len + 1;
581 } else {
582 /* "Big" symbols use two bytes. */
583 printf("\t.byte 0x%02x, 0x%02x",
584 (table[i]->len & 0x7F) | 0x80,
585 (table[i]->len >> 7) & 0x7F);
586 off += table[i]->len + 2;
587 }
588 for (k = 0; k < table[i]->len; k++)
589 printf(", 0x%02x", table[i]->sym[k]);
590 printf("\n");
591 }
592 printf("\n");
593
594 output_label("kallsyms_markers");
595 for (i = 0; i < ((table_cnt + 255) >> 8); i++)
596 printf("\t.long\t%u\n", markers[i]);
597 printf("\n");
598
599 free(markers);
600
601 sort_symbols_by_name();
602 output_label("kallsyms_seqs_of_names");
603 for (i = 0; i < table_cnt; i++)
604 printf("\t.byte 0x%02x, 0x%02x, 0x%02x\n",
605 (unsigned char)(table[i]->seq >> 16),
606 (unsigned char)(table[i]->seq >> 8),
607 (unsigned char)(table[i]->seq >> 0));
608 printf("\n");
609
610 output_label("kallsyms_token_table");
611 off = 0;
612 for (i = 0; i < 256; i++) {
613 best_idx[i] = off;
614 expand_symbol(best_table[i], best_table_len[i], buf);
615 printf("\t.asciz\t\"%s\"\n", buf);
616 off += strlen(buf) + 1;
617 }
618 printf("\n");
619
620 output_label("kallsyms_token_index");
621 for (i = 0; i < 256; i++)
622 printf("\t.short\t%d\n", best_idx[i]);
623 printf("\n");
624}
625
626
627/* table lookup compression functions */
628
629/* count all the possible tokens in a symbol */
630static void learn_symbol(const unsigned char *symbol, int len)
631{
632 int i;
633
634 for (i = 0; i < len - 1; i++)
635 token_profit[ symbol[i] + (symbol[i + 1] << 8) ]++;
636}
637
638/* decrease the count for all the possible tokens in a symbol */
639static void forget_symbol(const unsigned char *symbol, int len)
640{
641 int i;
642
643 for (i = 0; i < len - 1; i++)
644 token_profit[ symbol[i] + (symbol[i + 1] << 8) ]--;
645}
646
647/* do the initial token count */
648static void build_initial_token_table(void)
649{
650 unsigned int i;
651
652 for (i = 0; i < table_cnt; i++)
653 learn_symbol(table[i]->sym, table[i]->len);
654}
655
656static unsigned char *find_token(unsigned char *str, int len,
657 const unsigned char *token)
658{
659 int i;
660
661 for (i = 0; i < len - 1; i++) {
662 if (str[i] == token[0] && str[i+1] == token[1])
663 return &str[i];
664 }
665 return NULL;
666}
667
668/* replace a given token in all the valid symbols. Use the sampled symbols
669 * to update the counts */
670static void compress_symbols(const unsigned char *str, int idx)
671{
672 unsigned int i, len, size;
673 unsigned char *p1, *p2;
674
675 for (i = 0; i < table_cnt; i++) {
676
677 len = table[i]->len;
678 p1 = table[i]->sym;
679
680 /* find the token on the symbol */
681 p2 = find_token(p1, len, str);
682 if (!p2) continue;
683
684 /* decrease the counts for this symbol's tokens */
685 forget_symbol(table[i]->sym, len);
686
687 size = len;
688
689 do {
690 *p2 = idx;
691 p2++;
692 size -= (p2 - p1);
693 memmove(p2, p2 + 1, size);
694 p1 = p2;
695 len--;
696
697 if (size < 2) break;
698
699 /* find the token on the symbol */
700 p2 = find_token(p1, size, str);
701
702 } while (p2);
703
704 table[i]->len = len;
705
706 /* increase the counts for this symbol's new tokens */
707 learn_symbol(table[i]->sym, len);
708 }
709}
710
711/* search the token with the maximum profit */
712static int find_best_token(void)
713{
714 int i, best, bestprofit;
715
716 bestprofit=-10000;
717 best = 0;
718
719 for (i = 0; i < 0x10000; i++) {
720 if (token_profit[i] > bestprofit) {
721 best = i;
722 bestprofit = token_profit[i];
723 }
724 }
725 return best;
726}
727
728/* this is the core of the algorithm: calculate the "best" table */
729static void optimize_result(void)
730{
731 int i, best;
732
733 /* using the '\0' symbol last allows compress_symbols to use standard
734 * fast string functions */
735 for (i = 255; i >= 0; i--) {
736
737 /* if this table slot is empty (it is not used by an actual
738 * original char code */
739 if (!best_table_len[i]) {
740
741 /* find the token with the best profit value */
742 best = find_best_token();
743 if (token_profit[best] == 0)
744 break;
745
746 /* place it in the "best" table */
747 best_table_len[i] = 2;
748 best_table[i][0] = best & 0xFF;
749 best_table[i][1] = (best >> 8) & 0xFF;
750
751 /* replace this token in all the valid symbols */
752 compress_symbols(best_table[i], i);
753 }
754 }
755}
756
757/* start by placing the symbols that are actually used on the table */
758static void insert_real_symbols_in_table(void)
759{
760 unsigned int i, j, c;
761
762 for (i = 0; i < table_cnt; i++) {
763 for (j = 0; j < table[i]->len; j++) {
764 c = table[i]->sym[j];
765 best_table[c][0]=c;
766 best_table_len[c]=1;
767 }
768 }
769}
770
771static void optimize_token_table(void)
772{
773 build_initial_token_table();
774
775 insert_real_symbols_in_table();
776
777 optimize_result();
778}
779
780/* guess for "linker script provide" symbol */
781static int may_be_linker_script_provide_symbol(const struct sym_entry *se)
782{
783 const char *symbol = sym_name(se);
784 int len = se->len - 1;
785
786 if (len < 8)
787 return 0;
788
789 if (symbol[0] != '_' || symbol[1] != '_')
790 return 0;
791
792 /* __start_XXXXX */
793 if (!memcmp(symbol + 2, "start_", 6))
794 return 1;
795
796 /* __stop_XXXXX */
797 if (!memcmp(symbol + 2, "stop_", 5))
798 return 1;
799
800 /* __end_XXXXX */
801 if (!memcmp(symbol + 2, "end_", 4))
802 return 1;
803
804 /* __XXXXX_start */
805 if (!memcmp(symbol + len - 6, "_start", 6))
806 return 1;
807
808 /* __XXXXX_end */
809 if (!memcmp(symbol + len - 4, "_end", 4))
810 return 1;
811
812 return 0;
813}
814
815static int compare_symbols(const void *a, const void *b)
816{
817 const struct sym_entry *sa = *(const struct sym_entry **)a;
818 const struct sym_entry *sb = *(const struct sym_entry **)b;
819 int wa, wb;
820
821 /* sort by address first */
822 if (sa->addr > sb->addr)
823 return 1;
824 if (sa->addr < sb->addr)
825 return -1;
826
827 /* sort by "weakness" type */
828 wa = (sa->sym[0] == 'w') || (sa->sym[0] == 'W');
829 wb = (sb->sym[0] == 'w') || (sb->sym[0] == 'W');
830 if (wa != wb)
831 return wa - wb;
832
833 /* sort by "linker script provide" type */
834 wa = may_be_linker_script_provide_symbol(sa);
835 wb = may_be_linker_script_provide_symbol(sb);
836 if (wa != wb)
837 return wa - wb;
838
839 /* sort by the number of prefix underscores */
840 wa = strspn(sym_name(sa), "_");
841 wb = strspn(sym_name(sb), "_");
842 if (wa != wb)
843 return wa - wb;
844
845 /* sort by initial order, so that other symbols are left undisturbed */
846 return sa->start_pos - sb->start_pos;
847}
848
849static void sort_symbols(void)
850{
851 qsort(table, table_cnt, sizeof(table[0]), compare_symbols);
852}
853
854static void make_percpus_absolute(void)
855{
856 unsigned int i;
857
858 for (i = 0; i < table_cnt; i++)
859 if (symbol_in_range(table[i], &percpu_range, 1)) {
860 /*
861 * Keep the 'A' override for percpu symbols to
862 * ensure consistent behavior compared to older
863 * versions of this tool.
864 */
865 table[i]->sym[0] = 'A';
866 table[i]->percpu_absolute = 1;
867 }
868}
869
870/* find the minimum non-absolute symbol address */
871static void record_relative_base(void)
872{
873 unsigned int i;
874
875 for (i = 0; i < table_cnt; i++)
876 if (!symbol_absolute(table[i])) {
877 /*
878 * The table is sorted by address.
879 * Take the first non-absolute symbol value.
880 */
881 relative_base = table[i]->addr;
882 return;
883 }
884}
885
886int main(int argc, char **argv)
887{
888 while (1) {
889 static struct option long_options[] = {
890 {"all-symbols", no_argument, &all_symbols, 1},
891 {"absolute-percpu", no_argument, &absolute_percpu, 1},
892 {"base-relative", no_argument, &base_relative, 1},
893 {"lto-clang", no_argument, <o_clang, 1},
894 {},
895 };
896
897 int c = getopt_long(argc, argv, "", long_options, NULL);
898
899 if (c == -1)
900 break;
901 if (c != 0)
902 usage();
903 }
904
905 if (optind >= argc)
906 usage();
907
908 read_map(argv[optind]);
909 shrink_table();
910 if (absolute_percpu)
911 make_percpus_absolute();
912 sort_symbols();
913 if (base_relative)
914 record_relative_base();
915 optimize_token_table();
916 write_src();
917
918 return 0;
919}
1/* Generate assembler source containing symbol information
2 *
3 * Copyright 2002 by Kai Germaschewski
4 *
5 * This software may be used and distributed according to the terms
6 * of the GNU General Public License, incorporated herein by reference.
7 *
8 * Usage: nm -n vmlinux | scripts/kallsyms [--all-symbols] > symbols.S
9 *
10 * Table compression uses all the unused char codes on the symbols and
11 * maps these to the most used substrings (tokens). For instance, it might
12 * map char code 0xF7 to represent "write_" and then in every symbol where
13 * "write_" appears it can be replaced by 0xF7, saving 5 bytes.
14 * The used codes themselves are also placed in the table so that the
15 * decompresion can work without "special cases".
16 * Applied to kernel symbols, this usually produces a compression ratio
17 * of about 50%.
18 *
19 */
20
21#include <stdio.h>
22#include <stdlib.h>
23#include <string.h>
24#include <ctype.h>
25
26#ifndef ARRAY_SIZE
27#define ARRAY_SIZE(arr) (sizeof(arr) / sizeof(arr[0]))
28#endif
29
30#define KSYM_NAME_LEN 128
31
32struct sym_entry {
33 unsigned long long addr;
34 unsigned int len;
35 unsigned int start_pos;
36 unsigned char *sym;
37};
38
39struct text_range {
40 const char *stext, *etext;
41 unsigned long long start, end;
42};
43
44static unsigned long long _text;
45static struct text_range text_ranges[] = {
46 { "_stext", "_etext" },
47 { "_sinittext", "_einittext" },
48 { "_stext_l1", "_etext_l1" }, /* Blackfin on-chip L1 inst SRAM */
49 { "_stext_l2", "_etext_l2" }, /* Blackfin on-chip L2 SRAM */
50};
51#define text_range_text (&text_ranges[0])
52#define text_range_inittext (&text_ranges[1])
53
54static struct sym_entry *table;
55static unsigned int table_size, table_cnt;
56static int all_symbols = 0;
57static char symbol_prefix_char = '\0';
58
59int token_profit[0x10000];
60
61/* the table that holds the result of the compression */
62unsigned char best_table[256][2];
63unsigned char best_table_len[256];
64
65
66static void usage(void)
67{
68 fprintf(stderr, "Usage: kallsyms [--all-symbols] [--symbol-prefix=<prefix char>] < in.map > out.S\n");
69 exit(1);
70}
71
72/*
73 * This ignores the intensely annoying "mapping symbols" found
74 * in ARM ELF files: $a, $t and $d.
75 */
76static inline int is_arm_mapping_symbol(const char *str)
77{
78 return str[0] == '$' && strchr("atd", str[1])
79 && (str[2] == '\0' || str[2] == '.');
80}
81
82static int read_symbol_tr(const char *sym, unsigned long long addr)
83{
84 size_t i;
85 struct text_range *tr;
86
87 for (i = 0; i < ARRAY_SIZE(text_ranges); ++i) {
88 tr = &text_ranges[i];
89
90 if (strcmp(sym, tr->stext) == 0) {
91 tr->start = addr;
92 return 0;
93 } else if (strcmp(sym, tr->etext) == 0) {
94 tr->end = addr;
95 return 0;
96 }
97 }
98
99 return 1;
100}
101
102static int read_symbol(FILE *in, struct sym_entry *s)
103{
104 char str[500];
105 char *sym, stype;
106 int rc;
107
108 rc = fscanf(in, "%llx %c %499s\n", &s->addr, &stype, str);
109 if (rc != 3) {
110 if (rc != EOF && fgets(str, 500, in) == NULL)
111 fprintf(stderr, "Read error or end of file.\n");
112 return -1;
113 }
114
115 sym = str;
116 /* skip prefix char */
117 if (symbol_prefix_char && str[0] == symbol_prefix_char)
118 sym++;
119
120 /* Ignore most absolute/undefined (?) symbols. */
121 if (strcmp(sym, "_text") == 0)
122 _text = s->addr;
123 else if (read_symbol_tr(sym, s->addr) == 0)
124 /* nothing to do */;
125 else if (toupper(stype) == 'A')
126 {
127 /* Keep these useful absolute symbols */
128 if (strcmp(sym, "__kernel_syscall_via_break") &&
129 strcmp(sym, "__kernel_syscall_via_epc") &&
130 strcmp(sym, "__kernel_sigtramp") &&
131 strcmp(sym, "__gp"))
132 return -1;
133
134 }
135 else if (toupper(stype) == 'U' ||
136 is_arm_mapping_symbol(sym))
137 return -1;
138 /* exclude also MIPS ELF local symbols ($L123 instead of .L123) */
139 else if (str[0] == '$')
140 return -1;
141 /* exclude debugging symbols */
142 else if (stype == 'N')
143 return -1;
144
145 /* include the type field in the symbol name, so that it gets
146 * compressed together */
147 s->len = strlen(str) + 1;
148 s->sym = malloc(s->len + 1);
149 if (!s->sym) {
150 fprintf(stderr, "kallsyms failure: "
151 "unable to allocate required amount of memory\n");
152 exit(EXIT_FAILURE);
153 }
154 strcpy((char *)s->sym + 1, str);
155 s->sym[0] = stype;
156
157 return 0;
158}
159
160static int symbol_valid_tr(struct sym_entry *s)
161{
162 size_t i;
163 struct text_range *tr;
164
165 for (i = 0; i < ARRAY_SIZE(text_ranges); ++i) {
166 tr = &text_ranges[i];
167
168 if (s->addr >= tr->start && s->addr <= tr->end)
169 return 1;
170 }
171
172 return 0;
173}
174
175static int symbol_valid(struct sym_entry *s)
176{
177 /* Symbols which vary between passes. Passes 1 and 2 must have
178 * identical symbol lists. The kallsyms_* symbols below are only added
179 * after pass 1, they would be included in pass 2 when --all-symbols is
180 * specified so exclude them to get a stable symbol list.
181 */
182 static char *special_symbols[] = {
183 "kallsyms_addresses",
184 "kallsyms_num_syms",
185 "kallsyms_names",
186 "kallsyms_markers",
187 "kallsyms_token_table",
188 "kallsyms_token_index",
189
190 /* Exclude linker generated symbols which vary between passes */
191 "_SDA_BASE_", /* ppc */
192 "_SDA2_BASE_", /* ppc */
193 NULL };
194 int i;
195 int offset = 1;
196
197 /* skip prefix char */
198 if (symbol_prefix_char && *(s->sym + 1) == symbol_prefix_char)
199 offset++;
200
201 /* if --all-symbols is not specified, then symbols outside the text
202 * and inittext sections are discarded */
203 if (!all_symbols) {
204 if (symbol_valid_tr(s) == 0)
205 return 0;
206 /* Corner case. Discard any symbols with the same value as
207 * _etext _einittext; they can move between pass 1 and 2 when
208 * the kallsyms data are added. If these symbols move then
209 * they may get dropped in pass 2, which breaks the kallsyms
210 * rules.
211 */
212 if ((s->addr == text_range_text->end &&
213 strcmp((char *)s->sym + offset, text_range_text->etext)) ||
214 (s->addr == text_range_inittext->end &&
215 strcmp((char *)s->sym + offset, text_range_inittext->etext)))
216 return 0;
217 }
218
219 /* Exclude symbols which vary between passes. */
220 if (strstr((char *)s->sym + offset, "_compiled."))
221 return 0;
222
223 for (i = 0; special_symbols[i]; i++)
224 if( strcmp((char *)s->sym + offset, special_symbols[i]) == 0 )
225 return 0;
226
227 return 1;
228}
229
230static void read_map(FILE *in)
231{
232 while (!feof(in)) {
233 if (table_cnt >= table_size) {
234 table_size += 10000;
235 table = realloc(table, sizeof(*table) * table_size);
236 if (!table) {
237 fprintf(stderr, "out of memory\n");
238 exit (1);
239 }
240 }
241 if (read_symbol(in, &table[table_cnt]) == 0) {
242 table[table_cnt].start_pos = table_cnt;
243 table_cnt++;
244 }
245 }
246}
247
248static void output_label(char *label)
249{
250 if (symbol_prefix_char)
251 printf(".globl %c%s\n", symbol_prefix_char, label);
252 else
253 printf(".globl %s\n", label);
254 printf("\tALGN\n");
255 if (symbol_prefix_char)
256 printf("%c%s:\n", symbol_prefix_char, label);
257 else
258 printf("%s:\n", label);
259}
260
261/* uncompress a compressed symbol. When this function is called, the best table
262 * might still be compressed itself, so the function needs to be recursive */
263static int expand_symbol(unsigned char *data, int len, char *result)
264{
265 int c, rlen, total=0;
266
267 while (len) {
268 c = *data;
269 /* if the table holds a single char that is the same as the one
270 * we are looking for, then end the search */
271 if (best_table[c][0]==c && best_table_len[c]==1) {
272 *result++ = c;
273 total++;
274 } else {
275 /* if not, recurse and expand */
276 rlen = expand_symbol(best_table[c], best_table_len[c], result);
277 total += rlen;
278 result += rlen;
279 }
280 data++;
281 len--;
282 }
283 *result=0;
284
285 return total;
286}
287
288static void write_src(void)
289{
290 unsigned int i, k, off;
291 unsigned int best_idx[256];
292 unsigned int *markers;
293 char buf[KSYM_NAME_LEN];
294
295 printf("#include <asm/types.h>\n");
296 printf("#if BITS_PER_LONG == 64\n");
297 printf("#define PTR .quad\n");
298 printf("#define ALGN .align 8\n");
299 printf("#else\n");
300 printf("#define PTR .long\n");
301 printf("#define ALGN .align 4\n");
302 printf("#endif\n");
303
304 printf("\t.section .rodata, \"a\"\n");
305
306 /* Provide proper symbols relocatability by their '_text'
307 * relativeness. The symbol names cannot be used to construct
308 * normal symbol references as the list of symbols contains
309 * symbols that are declared static and are private to their
310 * .o files. This prevents .tmp_kallsyms.o or any other
311 * object from referencing them.
312 */
313 output_label("kallsyms_addresses");
314 for (i = 0; i < table_cnt; i++) {
315 if (toupper(table[i].sym[0]) != 'A') {
316 if (_text <= table[i].addr)
317 printf("\tPTR\t_text + %#llx\n",
318 table[i].addr - _text);
319 else
320 printf("\tPTR\t_text - %#llx\n",
321 _text - table[i].addr);
322 } else {
323 printf("\tPTR\t%#llx\n", table[i].addr);
324 }
325 }
326 printf("\n");
327
328 output_label("kallsyms_num_syms");
329 printf("\tPTR\t%d\n", table_cnt);
330 printf("\n");
331
332 /* table of offset markers, that give the offset in the compressed stream
333 * every 256 symbols */
334 markers = malloc(sizeof(unsigned int) * ((table_cnt + 255) / 256));
335 if (!markers) {
336 fprintf(stderr, "kallsyms failure: "
337 "unable to allocate required memory\n");
338 exit(EXIT_FAILURE);
339 }
340
341 output_label("kallsyms_names");
342 off = 0;
343 for (i = 0; i < table_cnt; i++) {
344 if ((i & 0xFF) == 0)
345 markers[i >> 8] = off;
346
347 printf("\t.byte 0x%02x", table[i].len);
348 for (k = 0; k < table[i].len; k++)
349 printf(", 0x%02x", table[i].sym[k]);
350 printf("\n");
351
352 off += table[i].len + 1;
353 }
354 printf("\n");
355
356 output_label("kallsyms_markers");
357 for (i = 0; i < ((table_cnt + 255) >> 8); i++)
358 printf("\tPTR\t%d\n", markers[i]);
359 printf("\n");
360
361 free(markers);
362
363 output_label("kallsyms_token_table");
364 off = 0;
365 for (i = 0; i < 256; i++) {
366 best_idx[i] = off;
367 expand_symbol(best_table[i], best_table_len[i], buf);
368 printf("\t.asciz\t\"%s\"\n", buf);
369 off += strlen(buf) + 1;
370 }
371 printf("\n");
372
373 output_label("kallsyms_token_index");
374 for (i = 0; i < 256; i++)
375 printf("\t.short\t%d\n", best_idx[i]);
376 printf("\n");
377}
378
379
380/* table lookup compression functions */
381
382/* count all the possible tokens in a symbol */
383static void learn_symbol(unsigned char *symbol, int len)
384{
385 int i;
386
387 for (i = 0; i < len - 1; i++)
388 token_profit[ symbol[i] + (symbol[i + 1] << 8) ]++;
389}
390
391/* decrease the count for all the possible tokens in a symbol */
392static void forget_symbol(unsigned char *symbol, int len)
393{
394 int i;
395
396 for (i = 0; i < len - 1; i++)
397 token_profit[ symbol[i] + (symbol[i + 1] << 8) ]--;
398}
399
400/* remove all the invalid symbols from the table and do the initial token count */
401static void build_initial_tok_table(void)
402{
403 unsigned int i, pos;
404
405 pos = 0;
406 for (i = 0; i < table_cnt; i++) {
407 if ( symbol_valid(&table[i]) ) {
408 if (pos != i)
409 table[pos] = table[i];
410 learn_symbol(table[pos].sym, table[pos].len);
411 pos++;
412 }
413 }
414 table_cnt = pos;
415}
416
417static void *find_token(unsigned char *str, int len, unsigned char *token)
418{
419 int i;
420
421 for (i = 0; i < len - 1; i++) {
422 if (str[i] == token[0] && str[i+1] == token[1])
423 return &str[i];
424 }
425 return NULL;
426}
427
428/* replace a given token in all the valid symbols. Use the sampled symbols
429 * to update the counts */
430static void compress_symbols(unsigned char *str, int idx)
431{
432 unsigned int i, len, size;
433 unsigned char *p1, *p2;
434
435 for (i = 0; i < table_cnt; i++) {
436
437 len = table[i].len;
438 p1 = table[i].sym;
439
440 /* find the token on the symbol */
441 p2 = find_token(p1, len, str);
442 if (!p2) continue;
443
444 /* decrease the counts for this symbol's tokens */
445 forget_symbol(table[i].sym, len);
446
447 size = len;
448
449 do {
450 *p2 = idx;
451 p2++;
452 size -= (p2 - p1);
453 memmove(p2, p2 + 1, size);
454 p1 = p2;
455 len--;
456
457 if (size < 2) break;
458
459 /* find the token on the symbol */
460 p2 = find_token(p1, size, str);
461
462 } while (p2);
463
464 table[i].len = len;
465
466 /* increase the counts for this symbol's new tokens */
467 learn_symbol(table[i].sym, len);
468 }
469}
470
471/* search the token with the maximum profit */
472static int find_best_token(void)
473{
474 int i, best, bestprofit;
475
476 bestprofit=-10000;
477 best = 0;
478
479 for (i = 0; i < 0x10000; i++) {
480 if (token_profit[i] > bestprofit) {
481 best = i;
482 bestprofit = token_profit[i];
483 }
484 }
485 return best;
486}
487
488/* this is the core of the algorithm: calculate the "best" table */
489static void optimize_result(void)
490{
491 int i, best;
492
493 /* using the '\0' symbol last allows compress_symbols to use standard
494 * fast string functions */
495 for (i = 255; i >= 0; i--) {
496
497 /* if this table slot is empty (it is not used by an actual
498 * original char code */
499 if (!best_table_len[i]) {
500
501 /* find the token with the breates profit value */
502 best = find_best_token();
503 if (token_profit[best] == 0)
504 break;
505
506 /* place it in the "best" table */
507 best_table_len[i] = 2;
508 best_table[i][0] = best & 0xFF;
509 best_table[i][1] = (best >> 8) & 0xFF;
510
511 /* replace this token in all the valid symbols */
512 compress_symbols(best_table[i], i);
513 }
514 }
515}
516
517/* start by placing the symbols that are actually used on the table */
518static void insert_real_symbols_in_table(void)
519{
520 unsigned int i, j, c;
521
522 memset(best_table, 0, sizeof(best_table));
523 memset(best_table_len, 0, sizeof(best_table_len));
524
525 for (i = 0; i < table_cnt; i++) {
526 for (j = 0; j < table[i].len; j++) {
527 c = table[i].sym[j];
528 best_table[c][0]=c;
529 best_table_len[c]=1;
530 }
531 }
532}
533
534static void optimize_token_table(void)
535{
536 build_initial_tok_table();
537
538 insert_real_symbols_in_table();
539
540 /* When valid symbol is not registered, exit to error */
541 if (!table_cnt) {
542 fprintf(stderr, "No valid symbol.\n");
543 exit(1);
544 }
545
546 optimize_result();
547}
548
549/* guess for "linker script provide" symbol */
550static int may_be_linker_script_provide_symbol(const struct sym_entry *se)
551{
552 const char *symbol = (char *)se->sym + 1;
553 int len = se->len - 1;
554
555 if (len < 8)
556 return 0;
557
558 if (symbol[0] != '_' || symbol[1] != '_')
559 return 0;
560
561 /* __start_XXXXX */
562 if (!memcmp(symbol + 2, "start_", 6))
563 return 1;
564
565 /* __stop_XXXXX */
566 if (!memcmp(symbol + 2, "stop_", 5))
567 return 1;
568
569 /* __end_XXXXX */
570 if (!memcmp(symbol + 2, "end_", 4))
571 return 1;
572
573 /* __XXXXX_start */
574 if (!memcmp(symbol + len - 6, "_start", 6))
575 return 1;
576
577 /* __XXXXX_end */
578 if (!memcmp(symbol + len - 4, "_end", 4))
579 return 1;
580
581 return 0;
582}
583
584static int prefix_underscores_count(const char *str)
585{
586 const char *tail = str;
587
588 while (*tail == '_')
589 tail++;
590
591 return tail - str;
592}
593
594static int compare_symbols(const void *a, const void *b)
595{
596 const struct sym_entry *sa;
597 const struct sym_entry *sb;
598 int wa, wb;
599
600 sa = a;
601 sb = b;
602
603 /* sort by address first */
604 if (sa->addr > sb->addr)
605 return 1;
606 if (sa->addr < sb->addr)
607 return -1;
608
609 /* sort by "weakness" type */
610 wa = (sa->sym[0] == 'w') || (sa->sym[0] == 'W');
611 wb = (sb->sym[0] == 'w') || (sb->sym[0] == 'W');
612 if (wa != wb)
613 return wa - wb;
614
615 /* sort by "linker script provide" type */
616 wa = may_be_linker_script_provide_symbol(sa);
617 wb = may_be_linker_script_provide_symbol(sb);
618 if (wa != wb)
619 return wa - wb;
620
621 /* sort by the number of prefix underscores */
622 wa = prefix_underscores_count((const char *)sa->sym + 1);
623 wb = prefix_underscores_count((const char *)sb->sym + 1);
624 if (wa != wb)
625 return wa - wb;
626
627 /* sort by initial order, so that other symbols are left undisturbed */
628 return sa->start_pos - sb->start_pos;
629}
630
631static void sort_symbols(void)
632{
633 qsort(table, table_cnt, sizeof(struct sym_entry), compare_symbols);
634}
635
636int main(int argc, char **argv)
637{
638 if (argc >= 2) {
639 int i;
640 for (i = 1; i < argc; i++) {
641 if(strcmp(argv[i], "--all-symbols") == 0)
642 all_symbols = 1;
643 else if (strncmp(argv[i], "--symbol-prefix=", 16) == 0) {
644 char *p = &argv[i][16];
645 /* skip quote */
646 if ((*p == '"' && *(p+2) == '"') || (*p == '\'' && *(p+2) == '\''))
647 p++;
648 symbol_prefix_char = *p;
649 } else
650 usage();
651 }
652 } else if (argc != 1)
653 usage();
654
655 read_map(stdin);
656 sort_symbols();
657 optimize_token_table();
658 write_src();
659
660 return 0;
661}