block: ublk: fix doc build warning
[linux-block.git] / scripts / kallsyms.c
CommitLineData
1da177e4
LT
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 *
adc40221
YU
8 * Usage: kallsyms [--all-symbols] [--absolute-percpu]
9 * [--base-relative] in.map > out.S
1da177e4 10 *
1da177e4
LT
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
aa221f2e 22#include <getopt.h>
a41333e0 23#include <stdbool.h>
1da177e4
LT
24#include <stdio.h>
25#include <stdlib.h>
26#include <string.h>
27#include <ctype.h>
2213e9a6 28#include <limits.h>
1da177e4 29
17b1f0de 30#define ARRAY_SIZE(arr) (sizeof(arr) / sizeof(arr[0]))
17b1f0de 31
b471927e
BF
32#define _stringify_1(x) #x
33#define _stringify(x) _stringify_1(x)
34
b8a94bfb 35#define KSYM_NAME_LEN 512
1da177e4 36
6e8c5bbd
MO
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 */
b8a94bfb 44#define KSYM_NAME_LEN_BUFFER 2048
6e8c5bbd
MO
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);
b471927e 49
1da177e4
LT
50struct sym_entry {
51 unsigned long long addr;
b3dbb4ec 52 unsigned int len;
60443c88 53 unsigned int seq;
f2df3f65 54 unsigned int start_pos;
8c996940 55 unsigned int percpu_absolute;
9d82973e 56 unsigned char sym[];
1da177e4
LT
57};
58
78eb7159
KC
59struct addr_range {
60 const char *start_sym, *end_sym;
17b1f0de
MF
61 unsigned long long start, end;
62};
63
64static unsigned long long _text;
2213e9a6 65static unsigned long long relative_base;
78eb7159 66static struct addr_range text_ranges[] = {
17b1f0de
MF
67 { "_stext", "_etext" },
68 { "_sinittext", "_einittext" },
17b1f0de
MF
69};
70#define text_range_text (&text_ranges[0])
71#define text_range_inittext (&text_ranges[1])
72
c6bda7c9
RR
73static struct addr_range percpu_range = {
74 "__per_cpu_start", "__per_cpu_end", -1ULL, 0
75};
76
8d605269 77static struct sym_entry **table;
b3dbb4ec 78static unsigned int table_size, table_cnt;
831362fc
MY
79static int all_symbols;
80static int absolute_percpu;
81static int base_relative;
010a0aad 82static int lto_clang;
1da177e4 83
f43e9daa 84static int token_profit[0x10000];
1da177e4
LT
85
86/* the table that holds the result of the compression */
f43e9daa
MY
87static unsigned char best_table[256][2];
88static unsigned char best_table_len[256];
1da177e4
LT
89
90
b3dbb4ec 91static void usage(void)
1da177e4 92{
8d3a7507 93 fprintf(stderr, "Usage: kallsyms [--all-symbols] [--absolute-percpu] "
010a0aad 94 "[--base-relative] [--lto-clang] in.map > out.S\n");
1da177e4
LT
95 exit(1);
96}
97
29e55ad3
MY
98static char *sym_name(const struct sym_entry *s)
99{
100 return (char *)s->sym + 1;
101}
102
a41333e0
MY
103static bool is_ignored_symbol(const char *name, char type)
104{
516d980f 105 /* Symbol names that exactly match to the following are ignored.*/
a41333e0
MY
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
516d980f 128 /* Symbol names that begin with the following are ignored.*/
a41333e0 129 static const char * const ignored_prefixes[] = {
a41333e0 130 "__efistub_", /* arm64 EFI stub namespace */
6ccf9cb5
KS
131 "__kvm_nvhe_$", /* arm64 local symbols in non-VHE KVM namespace */
132 "__kvm_nvhe_.L", /* arm64 local symbols in non-VHE KVM namespace */
efe6e306
AB
133 "__AArch64ADRPThunk_", /* arm64 lld */
134 "__ARMV5PILongThunk_", /* arm lld */
135 "__ARMV7PILongThunk_",
136 "__ThumbV7PILongThunk_",
137 "__LA25Thunk_", /* mips lld */
138 "__microLA25Thunk_",
d0f9562e 139 "__kcfi_typeid_", /* CFI type identifiers */
a41333e0
MY
140 NULL
141 };
142
516d980f 143 /* Symbol names that end with the following are ignored.*/
a41333e0
MY
144 static const char * const ignored_suffixes[] = {
145 "_from_arm", /* arm */
146 "_from_thumb", /* arm */
147 "_veneer", /* arm */
148 NULL
149 };
150
516d980f
MY
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
a41333e0
MY
158 const char * const *p;
159
a41333e0
MY
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
516d980f
MY
175 for (p = ignored_matches; *p; p++) {
176 if (strstr(name, *p))
177 return true;
178 }
179
887df76d
MY
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
a41333e0
MY
195 return false;
196}
197
b6233d0d
MY
198static void check_symbol_range(const char *sym, unsigned long long addr,
199 struct addr_range *ranges, int entries)
17b1f0de
MF
200{
201 size_t i;
78eb7159 202 struct addr_range *ar;
17b1f0de 203
78eb7159
KC
204 for (i = 0; i < entries; ++i) {
205 ar = &ranges[i];
17b1f0de 206
78eb7159
KC
207 if (strcmp(sym, ar->start_sym) == 0) {
208 ar->start = addr;
b6233d0d 209 return;
78eb7159
KC
210 } else if (strcmp(sym, ar->end_sym) == 0) {
211 ar->end = addr;
b6233d0d 212 return;
17b1f0de
MF
213 }
214 }
17b1f0de
MF
215}
216
8d605269 217static struct sym_entry *read_symbol(FILE *in)
1da177e4 218{
b471927e 219 char name[KSYM_NAME_LEN_BUFFER+1], type;
8d605269
MY
220 unsigned long long addr;
221 unsigned int len;
222 struct sym_entry *sym;
1da177e4
LT
223 int rc;
224
b471927e 225 rc = fscanf(in, "%llx %c %" _stringify(KSYM_NAME_LEN_BUFFER) "s\n", &addr, &type, name);
1da177e4 226 if (rc != 3) {
b66c874f 227 if (rc != EOF && fgets(name, ARRAY_SIZE(name), in) == NULL)
ef894870 228 fprintf(stderr, "Read error or end of file.\n");
8d605269 229 return NULL;
1da177e4 230 }
be9f6133 231 if (strlen(name) >= KSYM_NAME_LEN) {
6db2983c 232 fprintf(stderr, "Symbol %s too long for kallsyms (%zu >= %d).\n"
bb66fc67 233 "Please increase KSYM_NAME_LEN both in kernel and kallsyms.c\n",
be9f6133 234 name, strlen(name), KSYM_NAME_LEN);
8d605269 235 return NULL;
f3462aa9 236 }
1da177e4 237
be9f6133 238 if (strcmp(name, "_text") == 0)
8d605269 239 _text = addr;
b6233d0d 240
7883a143
MP
241 /* Ignore most absolute/undefined (?) symbols. */
242 if (is_ignored_symbol(name, type))
243 return NULL;
244
8d605269
MY
245 check_symbol_range(name, addr, text_ranges, ARRAY_SIZE(text_ranges));
246 check_symbol_range(name, addr, &percpu_range, 1);
1da177e4
LT
247
248 /* include the type field in the symbol name, so that it gets
249 * compressed together */
8d605269
MY
250
251 len = strlen(name) + 1;
252
9d1b3895 253 sym = malloc(sizeof(*sym) + len + 1);
8d605269 254 if (!sym) {
f1a136e0
JJ
255 fprintf(stderr, "kallsyms failure: "
256 "unable to allocate required amount of memory\n");
257 exit(EXIT_FAILURE);
258 }
8d605269
MY
259 sym->addr = addr;
260 sym->len = len;
261 sym->sym[0] = type;
9d1b3895 262 strcpy(sym_name(sym), name);
8d605269 263 sym->percpu_absolute = 0;
8c996940 264
8d605269 265 return sym;
1da177e4
LT
266}
267
4bfe2b78
MY
268static int symbol_in_range(const struct sym_entry *s,
269 const struct addr_range *ranges, int entries)
17b1f0de
MF
270{
271 size_t i;
4bfe2b78 272 const struct addr_range *ar;
17b1f0de 273
78eb7159
KC
274 for (i = 0; i < entries; ++i) {
275 ar = &ranges[i];
17b1f0de 276
78eb7159 277 if (s->addr >= ar->start && s->addr <= ar->end)
ac6ca5c8 278 return 1;
17b1f0de
MF
279 }
280
ac6ca5c8 281 return 0;
17b1f0de
MF
282}
283
4bfe2b78 284static int symbol_valid(const struct sym_entry *s)
1da177e4 285{
29e55ad3 286 const char *name = sym_name(s);
bd8b22d2 287
1da177e4
LT
288 /* if --all-symbols is not specified, then symbols outside the text
289 * and inittext sections are discarded */
290 if (!all_symbols) {
78eb7159
KC
291 if (symbol_in_range(s, text_ranges,
292 ARRAY_SIZE(text_ranges)) == 0)
1da177e4
LT
293 return 0;
294 /* Corner case. Discard any symbols with the same value as
a3b81113
RG
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.
1da177e4 299 */
17b1f0de 300 if ((s->addr == text_range_text->end &&
29e55ad3 301 strcmp(name, text_range_text->end_sym)) ||
17b1f0de 302 (s->addr == text_range_inittext->end &&
29e55ad3 303 strcmp(name, text_range_inittext->end_sym)))
1da177e4
LT
304 return 0;
305 }
306
1da177e4
LT
307 return 1;
308}
309
5e5c4fa7
MY
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++) {
8d605269 317 if (symbol_valid(table[i])) {
5e5c4fa7
MY
318 if (pos != i)
319 table[pos] = table[i];
320 pos++;
321 } else {
8d605269 322 free(table[i]);
5e5c4fa7
MY
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
aa221f2e 334static void read_map(const char *in)
1da177e4 335{
aa221f2e 336 FILE *fp;
8d605269
MY
337 struct sym_entry *sym;
338
aa221f2e
MY
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);
8d605269
MY
347 if (!sym)
348 continue;
349
350 sym->start_pos = table_cnt;
351
b3dbb4ec
PM
352 if (table_cnt >= table_size) {
353 table_size += 10000;
354 table = realloc(table, sizeof(*table) * table_size);
1da177e4
LT
355 if (!table) {
356 fprintf(stderr, "out of memory\n");
aa221f2e 357 fclose(fp);
1da177e4
LT
358 exit (1);
359 }
360 }
8d605269
MY
361
362 table[table_cnt++] = sym;
1da177e4 363 }
aa221f2e
MY
364
365 fclose(fp);
1da177e4
LT
366}
367
4bfe2b78 368static void output_label(const char *label)
1da177e4 369{
534c9f2e 370 printf(".globl %s\n", label);
1da177e4 371 printf("\tALGN\n");
534c9f2e 372 printf("%s:\n", label);
1da177e4
LT
373}
374
fd2ab2f6
MY
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
1da177e4
LT
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 */
4bfe2b78 386static int expand_symbol(const unsigned char *data, int len, char *result)
1da177e4
LT
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
4bfe2b78 411static int symbol_absolute(const struct sym_entry *s)
78eb7159 412{
8c996940 413 return s->percpu_absolute;
78eb7159
KC
414}
415
010a0aad
ZL
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
60443c88
ZL
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);
010a0aad
ZL
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));
60443c88
ZL
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
b3dbb4ec 475static void write_src(void)
1da177e4 476{
b3dbb4ec 477 unsigned int i, k, off;
1da177e4
LT
478 unsigned int best_idx[256];
479 unsigned int *markers;
9281acea 480 char buf[KSYM_NAME_LEN];
1da177e4 481
500193ec 482 printf("#include <asm/bitsperlong.h>\n");
1da177e4
LT
483 printf("#if BITS_PER_LONG == 64\n");
484 printf("#define PTR .quad\n");
72d3ebb9 485 printf("#define ALGN .balign 8\n");
1da177e4
LT
486 printf("#else\n");
487 printf("#define PTR .long\n");
72d3ebb9 488 printf("#define ALGN .balign 4\n");
1da177e4
LT
489 printf("#endif\n");
490
aad09470 491 printf("\t.section .rodata, \"a\"\n");
1da177e4 492
2213e9a6
AB
493 if (!base_relative)
494 output_label("kallsyms_addresses");
495 else
496 output_label("kallsyms_offsets");
497
b3dbb4ec 498 for (i = 0; i < table_cnt; i++) {
2213e9a6 499 if (base_relative) {
fd2ab2f6
MY
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
2213e9a6
AB
507 long long offset;
508 int overflow;
509
510 if (!absolute_percpu) {
8d605269 511 offset = table[i]->addr - relative_base;
2213e9a6 512 overflow = (offset < 0 || offset > UINT_MAX);
8d605269
MY
513 } else if (symbol_absolute(table[i])) {
514 offset = table[i]->addr;
2213e9a6
AB
515 overflow = (offset < 0 || offset > INT_MAX);
516 } else {
8d605269 517 offset = relative_base - table[i]->addr - 1;
2213e9a6
AB
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",
8d605269
MY
523 symbol_absolute(table[i]) ? "absolute" : "relative",
524 table[i]->addr);
2213e9a6
AB
525 exit(EXIT_FAILURE);
526 }
527 printf("\t.long\t%#x\n", (int)offset);
8d605269
MY
528 } else if (!symbol_absolute(table[i])) {
529 output_address(table[i]->addr);
fd593d12 530 } else {
8d605269 531 printf("\tPTR\t%#llx\n", table[i]->addr);
fd593d12 532 }
1da177e4
LT
533 }
534 printf("\n");
535
2213e9a6
AB
536 if (base_relative) {
537 output_label("kallsyms_relative_base");
fd2ab2f6 538 output_address(relative_base);
2213e9a6
AB
539 printf("\n");
540 }
541
1da177e4 542 output_label("kallsyms_num_syms");
80ffbaa5 543 printf("\t.long\t%u\n", table_cnt);
1da177e4
LT
544 printf("\n");
545
546 /* table of offset markers, that give the offset in the compressed stream
547 * every 256 symbols */
f1a136e0
JJ
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 }
1da177e4
LT
554
555 output_label("kallsyms_names");
1da177e4 556 off = 0;
b3dbb4ec
PM
557 for (i = 0; i < table_cnt; i++) {
558 if ((i & 0xFF) == 0)
559 markers[i >> 8] = off;
60443c88 560 table[i]->seq = i;
1da177e4 561
73bbb944
MO
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 }
8d605269
MY
588 for (k = 0; k < table[i]->len; k++)
589 printf(", 0x%02x", table[i]->sym[k]);
1da177e4 590 printf("\n");
1da177e4
LT
591 }
592 printf("\n");
593
594 output_label("kallsyms_markers");
b3dbb4ec 595 for (i = 0; i < ((table_cnt + 255) >> 8); i++)
80ffbaa5 596 printf("\t.long\t%u\n", markers[i]);
1da177e4
LT
597 printf("\n");
598
599 free(markers);
600
60443c88
ZL
601 sort_symbols_by_name();
602 output_label("kallsyms_seqs_of_names");
603 for (i = 0; i < table_cnt; i++)
19bd8981
ZL
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));
60443c88
ZL
608 printf("\n");
609
1da177e4
LT
610 output_label("kallsyms_token_table");
611 off = 0;
612 for (i = 0; i < 256; i++) {
613 best_idx[i] = off;
b3dbb4ec 614 expand_symbol(best_table[i], best_table_len[i], buf);
1da177e4
LT
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
1da177e4 629/* count all the possible tokens in a symbol */
4bfe2b78 630static void learn_symbol(const unsigned char *symbol, int len)
1da177e4
LT
631{
632 int i;
633
634 for (i = 0; i < len - 1; i++)
b3dbb4ec 635 token_profit[ symbol[i] + (symbol[i + 1] << 8) ]++;
1da177e4
LT
636}
637
638/* decrease the count for all the possible tokens in a symbol */
4bfe2b78 639static void forget_symbol(const unsigned char *symbol, int len)
1da177e4
LT
640{
641 int i;
642
643 for (i = 0; i < len - 1; i++)
b3dbb4ec 644 token_profit[ symbol[i] + (symbol[i + 1] << 8) ]--;
1da177e4
LT
645}
646
5e5c4fa7 647/* do the initial token count */
fcdf7197 648static void build_initial_token_table(void)
1da177e4 649{
5e5c4fa7 650 unsigned int i;
1da177e4 651
5e5c4fa7 652 for (i = 0; i < table_cnt; i++)
8d605269 653 learn_symbol(table[i]->sym, table[i]->len);
1da177e4
LT
654}
655
2558c138 656static unsigned char *find_token(unsigned char *str, int len,
4bfe2b78 657 const unsigned char *token)
7c5d249a
PM
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
1da177e4
LT
668/* replace a given token in all the valid symbols. Use the sampled symbols
669 * to update the counts */
4bfe2b78 670static void compress_symbols(const unsigned char *str, int idx)
1da177e4 671{
b3dbb4ec
PM
672 unsigned int i, len, size;
673 unsigned char *p1, *p2;
1da177e4 674
b3dbb4ec 675 for (i = 0; i < table_cnt; i++) {
1da177e4 676
8d605269
MY
677 len = table[i]->len;
678 p1 = table[i]->sym;
b3dbb4ec
PM
679
680 /* find the token on the symbol */
7c5d249a 681 p2 = find_token(p1, len, str);
b3dbb4ec
PM
682 if (!p2) continue;
683
684 /* decrease the counts for this symbol's tokens */
8d605269 685 forget_symbol(table[i]->sym, len);
b3dbb4ec
PM
686
687 size = len;
1da177e4
LT
688
689 do {
b3dbb4ec
PM
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
1da177e4 699 /* find the token on the symbol */
7c5d249a 700 p2 = find_token(p1, size, str);
1da177e4 701
b3dbb4ec 702 } while (p2);
1da177e4 703
8d605269 704 table[i]->len = len;
1da177e4 705
b3dbb4ec 706 /* increase the counts for this symbol's new tokens */
8d605269 707 learn_symbol(table[i]->sym, len);
1da177e4
LT
708 }
709}
710
711/* search the token with the maximum profit */
b3dbb4ec 712static int find_best_token(void)
1da177e4 713{
b3dbb4ec 714 int i, best, bestprofit;
1da177e4
LT
715
716 bestprofit=-10000;
b3dbb4ec 717 best = 0;
1da177e4 718
b3dbb4ec
PM
719 for (i = 0; i < 0x10000; i++) {
720 if (token_profit[i] > bestprofit) {
721 best = i;
722 bestprofit = token_profit[i];
1da177e4 723 }
1da177e4 724 }
1da177e4
LT
725 return best;
726}
727
728/* this is the core of the algorithm: calculate the "best" table */
729static void optimize_result(void)
730{
b3dbb4ec 731 int i, best;
1da177e4
LT
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
cbf7a90e 741 /* find the token with the best profit value */
1da177e4 742 best = find_best_token();
e0a04b11
XW
743 if (token_profit[best] == 0)
744 break;
1da177e4
LT
745
746 /* place it in the "best" table */
b3dbb4ec
PM
747 best_table_len[i] = 2;
748 best_table[i][0] = best & 0xFF;
749 best_table[i][1] = (best >> 8) & 0xFF;
1da177e4
LT
750
751 /* replace this token in all the valid symbols */
b3dbb4ec 752 compress_symbols(best_table[i], i);
1da177e4
LT
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{
b3dbb4ec 760 unsigned int i, j, c;
1da177e4 761
b3dbb4ec 762 for (i = 0; i < table_cnt; i++) {
8d605269
MY
763 for (j = 0; j < table[i]->len; j++) {
764 c = table[i]->sym[j];
b3dbb4ec
PM
765 best_table[c][0]=c;
766 best_table_len[c]=1;
1da177e4
LT
767 }
768 }
769}
770
771static void optimize_token_table(void)
772{
fcdf7197 773 build_initial_token_table();
1da177e4
LT
774
775 insert_real_symbols_in_table();
776
777 optimize_result();
778}
779
b478b782
LJ
780/* guess for "linker script provide" symbol */
781static int may_be_linker_script_provide_symbol(const struct sym_entry *se)
782{
29e55ad3 783 const char *symbol = sym_name(se);
b478b782
LJ
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
f2df3f65
PM
815static int compare_symbols(const void *a, const void *b)
816{
8d605269
MY
817 const struct sym_entry *sa = *(const struct sym_entry **)a;
818 const struct sym_entry *sb = *(const struct sym_entry **)b;
f2df3f65
PM
819 int wa, wb;
820
f2df3f65
PM
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
b478b782
LJ
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 */
aa915245
MY
840 wa = strspn(sym_name(sa), "_");
841 wb = strspn(sym_name(sb), "_");
b478b782
LJ
842 if (wa != wb)
843 return wa - wb;
844
f2df3f65
PM
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{
8d605269 851 qsort(table, table_cnt, sizeof(table[0]), compare_symbols);
f2df3f65 852}
1da177e4 853
c6bda7c9
RR
854static void make_percpus_absolute(void)
855{
856 unsigned int i;
857
858 for (i = 0; i < table_cnt; i++)
8d605269 859 if (symbol_in_range(table[i], &percpu_range, 1)) {
8c996940
AB
860 /*
861 * Keep the 'A' override for percpu symbols to
862 * ensure consistent behavior compared to older
863 * versions of this tool.
864 */
8d605269
MY
865 table[i]->sym[0] = 'A';
866 table[i]->percpu_absolute = 1;
8c996940 867 }
c6bda7c9
RR
868}
869
2213e9a6
AB
870/* find the minimum non-absolute symbol address */
871static void record_relative_base(void)
872{
873 unsigned int i;
874
2213e9a6 875 for (i = 0; i < table_cnt; i++)
8d605269 876 if (!symbol_absolute(table[i])) {
f34ea029
MY
877 /*
878 * The table is sorted by address.
879 * Take the first non-absolute symbol value.
880 */
8d605269 881 relative_base = table[i]->addr;
f34ea029
MY
882 return;
883 }
2213e9a6
AB
884}
885
b3dbb4ec 886int main(int argc, char **argv)
1da177e4 887{
aa221f2e
MY
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},
010a0aad 893 {"lto-clang", no_argument, &lto_clang, 1},
aa221f2e
MY
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)
1da177e4
LT
906 usage();
907
aa221f2e 908 read_map(argv[optind]);
5e5c4fa7 909 shrink_table();
c6bda7c9
RR
910 if (absolute_percpu)
911 make_percpus_absolute();
f34ea029 912 sort_symbols();
2213e9a6
AB
913 if (base_relative)
914 record_relative_base();
2ea03891 915 optimize_token_table();
1da177e4
LT
916 write_src();
917
918 return 0;
919}