tccgen.c: In parse_btype, handle type qualifiers applied to arrays.
[tinycc.git] / tccelf.c
blobb90cf0ff80df00b23aecfb7ead6c6ffb95081e93
1 /*
2 * ELF file handling for TCC
3 *
4 * Copyright (c) 2001-2004 Fabrice Bellard
5 *
6 * This library is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU Lesser General Public
8 * License as published by the Free Software Foundation; either
9 * version 2 of the License, or (at your option) any later version.
10 *
11 * This library is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 * Lesser General Public License for more details.
15 *
16 * You should have received a copy of the GNU Lesser General Public
17 * License along with this library; if not, write to the Free Software
18 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
19 */
21 #include "tcc.h"
23 /* Define this to get some debug output during relocation processing. */
24 #undef DEBUG_RELOC
26 /* XXX: avoid static variable */
27 static int new_undef_sym = 0; /* Is there a new undefined sym since last new_undef_sym() */
29 ST_FUNC int put_elf_str(Section *s, const char *sym)
30 {
31 int offset, len;
32 char *ptr;
34 len = strlen(sym) + 1;
35 offset = s->data_offset;
36 ptr = section_ptr_add(s, len);
37 memcpy(ptr, sym, len);
38 return offset;
39 }
41 /* elf symbol hashing function */
42 static unsigned long elf_hash(const unsigned char *name)
43 {
44 unsigned long h = 0, g;
46 while (*name) {
47 h = (h << 4) + *name++;
48 g = h & 0xf0000000;
49 if (g)
50 h ^= g >> 24;
51 h &= ~g;
52 }
53 return h;
54 }
56 /* rebuild hash table of section s */
57 /* NOTE: we do factorize the hash table code to go faster */
58 static void rebuild_hash(Section *s, unsigned int nb_buckets)
59 {
60 ElfW(Sym) *sym;
61 int *ptr, *hash, nb_syms, sym_index, h;
62 unsigned char *strtab;
64 strtab = s->link->data;
65 nb_syms = s->data_offset / sizeof(ElfW(Sym));
67 s->hash->data_offset = 0;
68 ptr = section_ptr_add(s->hash, (2 + nb_buckets + nb_syms) * sizeof(int));
69 ptr[0] = nb_buckets;
70 ptr[1] = nb_syms;
71 ptr += 2;
72 hash = ptr;
73 memset(hash, 0, (nb_buckets + 1) * sizeof(int));
74 ptr += nb_buckets + 1;
76 sym = (ElfW(Sym) *)s->data + 1;
77 for(sym_index = 1; sym_index < nb_syms; sym_index++) {
78 if (ELFW(ST_BIND)(sym->st_info) != STB_LOCAL) {
79 h = elf_hash(strtab + sym->st_name) % nb_buckets;
80 *ptr = hash[h];
81 hash[h] = sym_index;
82 } else {
83 *ptr = 0;
84 }
85 ptr++;
86 sym++;
87 }
88 }
90 /* return the symbol number */
91 ST_FUNC int put_elf_sym(Section *s, addr_t value, unsigned long size,
92 int info, int other, int shndx, const char *name)
93 {
94 int name_offset, sym_index;
95 int nbuckets, h;
96 ElfW(Sym) *sym;
97 Section *hs;
99 sym = section_ptr_add(s, sizeof(ElfW(Sym)));
100 if (name)
101 name_offset = put_elf_str(s->link, name);
102 else
103 name_offset = 0;
104 /* XXX: endianness */
105 sym->st_name = name_offset;
106 sym->st_value = value;
107 sym->st_size = size;
108 sym->st_info = info;
109 sym->st_other = other;
110 sym->st_shndx = shndx;
111 sym_index = sym - (ElfW(Sym) *)s->data;
112 hs = s->hash;
113 if (hs) {
114 int *ptr, *base;
115 ptr = section_ptr_add(hs, sizeof(int));
116 base = (int *)hs->data;
117 /* only add global or weak symbols */
118 if (ELFW(ST_BIND)(info) != STB_LOCAL) {
119 /* add another hashing entry */
120 nbuckets = base[0];
121 h = elf_hash((unsigned char *) name) % nbuckets;
122 *ptr = base[2 + h];
123 base[2 + h] = sym_index;
124 base[1]++;
125 /* we resize the hash table */
126 hs->nb_hashed_syms++;
127 if (hs->nb_hashed_syms > 2 * nbuckets) {
128 rebuild_hash(s, 2 * nbuckets);
130 } else {
131 *ptr = 0;
132 base[1]++;
135 return sym_index;
138 /* find global ELF symbol 'name' and return its index. Return 0 if not
139 found. */
140 ST_FUNC int find_elf_sym(Section *s, const char *name)
142 ElfW(Sym) *sym;
143 Section *hs;
144 int nbuckets, sym_index, h;
145 const char *name1;
147 hs = s->hash;
148 if (!hs)
149 return 0;
150 nbuckets = ((int *)hs->data)[0];
151 h = elf_hash((unsigned char *) name) % nbuckets;
152 sym_index = ((int *)hs->data)[2 + h];
153 while (sym_index != 0) {
154 sym = &((ElfW(Sym) *)s->data)[sym_index];
155 name1 = (char *) s->link->data + sym->st_name;
156 if (!strcmp(name, name1))
157 return sym_index;
158 sym_index = ((int *)hs->data)[2 + nbuckets + sym_index];
160 return 0;
163 /* return elf symbol value, signal error if 'err' is nonzero */
164 ST_FUNC addr_t get_elf_sym_addr(TCCState *s, const char *name, int err)
166 int sym_index;
167 ElfW(Sym) *sym;
169 sym_index = find_elf_sym(s->symtab, name);
170 sym = &((ElfW(Sym) *)s->symtab->data)[sym_index];
171 if (!sym_index || sym->st_shndx == SHN_UNDEF) {
172 if (err)
173 tcc_error("%s not defined", name);
174 return 0;
176 return sym->st_value;
179 /* return elf symbol value */
180 LIBTCCAPI void *tcc_get_symbol(TCCState *s, const char *name)
182 return (void*)(uintptr_t)get_elf_sym_addr(s, name, 0);
185 #if defined TCC_IS_NATIVE || defined TCC_TARGET_PE
186 /* return elf symbol value or error */
187 ST_FUNC void* tcc_get_symbol_err(TCCState *s, const char *name)
189 return (void*)(uintptr_t)get_elf_sym_addr(s, name, 1);
191 #endif
193 /* add an elf symbol : check if it is already defined and patch
194 it. Return symbol index. NOTE that sh_num can be SHN_UNDEF. */
195 ST_FUNC int add_elf_sym(Section *s, addr_t value, unsigned long size,
196 int info, int other, int sh_num, const char *name)
198 ElfW(Sym) *esym;
199 int sym_bind, sym_index, sym_type, esym_bind;
200 unsigned char sym_vis, esym_vis, new_vis;
202 sym_bind = ELFW(ST_BIND)(info);
203 sym_type = ELFW(ST_TYPE)(info);
204 sym_vis = ELFW(ST_VISIBILITY)(other);
206 if (sym_bind != STB_LOCAL) {
207 /* we search global or weak symbols */
208 sym_index = find_elf_sym(s, name);
209 if (!sym_index)
210 goto do_def;
211 esym = &((ElfW(Sym) *)s->data)[sym_index];
212 if (esym->st_shndx != SHN_UNDEF) {
213 esym_bind = ELFW(ST_BIND)(esym->st_info);
214 /* propagate the most constraining visibility */
215 /* STV_DEFAULT(0)<STV_PROTECTED(3)<STV_HIDDEN(2)<STV_INTERNAL(1) */
216 esym_vis = ELFW(ST_VISIBILITY)(esym->st_other);
217 if (esym_vis == STV_DEFAULT) {
218 new_vis = sym_vis;
219 } else if (sym_vis == STV_DEFAULT) {
220 new_vis = esym_vis;
221 } else {
222 new_vis = (esym_vis < sym_vis) ? esym_vis : sym_vis;
224 esym->st_other = (esym->st_other & ~ELFW(ST_VISIBILITY)(-1))
225 | new_vis;
226 other = esym->st_other; /* in case we have to patch esym */
227 if (sh_num == SHN_UNDEF) {
228 /* ignore adding of undefined symbol if the
229 corresponding symbol is already defined */
230 } else if (sym_bind == STB_GLOBAL && esym_bind == STB_WEAK) {
231 /* global overrides weak, so patch */
232 goto do_patch;
233 } else if (sym_bind == STB_WEAK && esym_bind == STB_GLOBAL) {
234 /* weak is ignored if already global */
235 } else if (sym_bind == STB_WEAK && esym_bind == STB_WEAK) {
236 /* keep first-found weak definition, ignore subsequents */
237 } else if (sym_vis == STV_HIDDEN || sym_vis == STV_INTERNAL) {
238 /* ignore hidden symbols after */
239 } else if (esym->st_shndx == SHN_COMMON
240 && (sh_num < SHN_LORESERVE || sh_num == SHN_COMMON)) {
241 /* gr: Happens with 'tcc ... -static tcctest.c' on e.g. Ubuntu 6.01
242 No idea if this is the correct solution ... */
243 goto do_patch;
244 } else if (s == tcc_state->dynsymtab_section) {
245 /* we accept that two DLL define the same symbol */
246 } else {
247 #if 0
248 printf("new_bind=%x new_shndx=%x new_vis=%x old_bind=%x old_shndx=%x old_vis=%x\n",
249 sym_bind, sh_num, new_vis, esym_bind, esym->st_shndx, esym_vis);
250 #endif
251 tcc_error_noabort("'%s' defined twice... may be -fcommon is needed?", name);
253 } else {
254 do_patch:
255 esym->st_info = ELFW(ST_INFO)(sym_bind, sym_type);
256 esym->st_shndx = sh_num;
257 new_undef_sym = 1;
258 esym->st_value = value;
259 esym->st_size = size;
260 esym->st_other = other;
262 } else {
263 do_def:
264 sym_index = put_elf_sym(s, value, size,
265 ELFW(ST_INFO)(sym_bind, sym_type), other,
266 sh_num, name);
268 return sym_index;
271 /* put relocation */
272 ST_FUNC void put_elf_reloca(Section *symtab, Section *s, unsigned long offset,
273 int type, int symbol, addr_t addend)
275 char buf[256];
276 Section *sr;
277 ElfW_Rel *rel;
279 sr = s->reloc;
280 if (!sr) {
281 /* if no relocation section, create it */
282 snprintf(buf, sizeof(buf), REL_SECTION_FMT, s->name);
283 /* if the symtab is allocated, then we consider the relocation
284 are also */
285 sr = new_section(tcc_state, buf, SHT_RELX, symtab->sh_flags);
286 sr->sh_entsize = sizeof(ElfW_Rel);
287 sr->link = symtab;
288 sr->sh_info = s->sh_num;
289 s->reloc = sr;
291 rel = section_ptr_add(sr, sizeof(ElfW_Rel));
292 rel->r_offset = offset;
293 rel->r_info = ELFW(R_INFO)(symbol, type);
294 #if defined(TCC_TARGET_ARM64) || defined(TCC_TARGET_X86_64)
295 rel->r_addend = addend;
296 #else
297 if (addend)
298 tcc_error("non-zero addend on REL architecture");
299 #endif
302 ST_FUNC void put_elf_reloc(Section *symtab, Section *s, unsigned long offset,
303 int type, int symbol)
305 put_elf_reloca(symtab, s, offset, type, symbol, 0);
308 /* put stab debug information */
310 ST_FUNC void put_stabs(const char *str, int type, int other, int desc,
311 unsigned long value)
313 Stab_Sym *sym;
315 sym = section_ptr_add(stab_section, sizeof(Stab_Sym));
316 if (str) {
317 sym->n_strx = put_elf_str(stabstr_section, str);
318 } else {
319 sym->n_strx = 0;
321 sym->n_type = type;
322 sym->n_other = other;
323 sym->n_desc = desc;
324 sym->n_value = value;
327 ST_FUNC void put_stabs_r(const char *str, int type, int other, int desc,
328 unsigned long value, Section *sec, int sym_index)
330 put_stabs(str, type, other, desc, value);
331 put_elf_reloc(symtab_section, stab_section,
332 stab_section->data_offset - sizeof(unsigned int),
333 R_DATA_32, sym_index);
336 ST_FUNC void put_stabn(int type, int other, int desc, int value)
338 put_stabs(NULL, type, other, desc, value);
341 ST_FUNC void put_stabd(int type, int other, int desc)
343 put_stabs(NULL, type, other, desc, 0);
346 /* Browse each elem of type <type> in section <sec> starting at elem <startoff>
347 using variable <elem> */
348 #define for_each_elem(sec, startoff, elem, type) \
349 for (elem = (type *) sec->data + startoff; \
350 elem < (type *) (sec->data + sec->data_offset); elem++)
352 /* In an ELF file symbol table, the local symbols must appear below
353 the global and weak ones. Since TCC cannot sort it while generating
354 the code, we must do it after. All the relocation tables are also
355 modified to take into account the symbol table sorting */
356 static void sort_syms(TCCState *s1, Section *s)
358 int *old_to_new_syms;
359 ElfW(Sym) *new_syms;
360 int nb_syms, i;
361 ElfW(Sym) *p, *q;
362 ElfW_Rel *rel;
363 Section *sr;
364 int type, sym_index;
366 nb_syms = s->data_offset / sizeof(ElfW(Sym));
367 new_syms = tcc_malloc(nb_syms * sizeof(ElfW(Sym)));
368 old_to_new_syms = tcc_malloc(nb_syms * sizeof(int));
370 /* first pass for local symbols */
371 p = (ElfW(Sym) *)s->data;
372 q = new_syms;
373 for(i = 0; i < nb_syms; i++) {
374 if (ELFW(ST_BIND)(p->st_info) == STB_LOCAL) {
375 old_to_new_syms[i] = q - new_syms;
376 *q++ = *p;
378 p++;
380 /* save the number of local symbols in section header */
381 s->sh_info = q - new_syms;
383 /* then second pass for non local symbols */
384 p = (ElfW(Sym) *)s->data;
385 for(i = 0; i < nb_syms; i++) {
386 if (ELFW(ST_BIND)(p->st_info) != STB_LOCAL) {
387 old_to_new_syms[i] = q - new_syms;
388 *q++ = *p;
390 p++;
393 /* we copy the new symbols to the old */
394 memcpy(s->data, new_syms, nb_syms * sizeof(ElfW(Sym)));
395 tcc_free(new_syms);
397 /* now we modify all the relocations */
398 for(i = 1; i < s1->nb_sections; i++) {
399 sr = s1->sections[i];
400 if (sr->sh_type == SHT_RELX && sr->link == s) {
401 for_each_elem(sr, 0, rel, ElfW_Rel) {
402 sym_index = ELFW(R_SYM)(rel->r_info);
403 type = ELFW(R_TYPE)(rel->r_info);
404 sym_index = old_to_new_syms[sym_index];
405 rel->r_info = ELFW(R_INFO)(sym_index, type);
410 tcc_free(old_to_new_syms);
413 /* relocate common symbols in the .bss section */
414 ST_FUNC void relocate_common_syms(void)
416 ElfW(Sym) *sym;
417 unsigned long offset, align;
419 for_each_elem(symtab_section, 1, sym, ElfW(Sym)) {
420 if (sym->st_shndx == SHN_COMMON) {
421 /* align symbol */
422 align = sym->st_value;
423 offset = bss_section->data_offset;
424 offset = (offset + align - 1) & -align;
425 sym->st_value = offset;
426 sym->st_shndx = bss_section->sh_num;
427 offset += sym->st_size;
428 bss_section->data_offset = offset;
433 /* relocate symbol table, resolve undefined symbols if do_resolve is
434 true and output error if undefined symbol. */
435 ST_FUNC void relocate_syms(TCCState *s1, int do_resolve)
437 ElfW(Sym) *sym, *esym;
438 int sym_bind, sh_num, sym_index;
439 const char *name;
441 for_each_elem(symtab_section, 1, sym, ElfW(Sym)) {
442 sh_num = sym->st_shndx;
443 if (sh_num == SHN_UNDEF) {
444 name = (char *) strtab_section->data + sym->st_name;
445 /* Use ld.so to resolve symbol for us (for tcc -run) */
446 if (do_resolve) {
447 #if defined TCC_IS_NATIVE && !defined _WIN32
448 void *addr;
449 name = (char *) symtab_section->link->data + sym->st_name;
450 addr = resolve_sym(s1, name);
451 if (addr) {
452 sym->st_value = (addr_t)addr;
453 #ifdef DEBUG_RELOC
454 printf ("relocate_sym: %s -> 0x%lx\n", name, sym->st_value);
455 #endif
456 goto found;
458 #endif
459 } else if (s1->dynsym) {
460 /* if dynamic symbol exist, then use it */
461 sym_index = find_elf_sym(s1->dynsym, name);
462 if (sym_index) {
463 esym = &((ElfW(Sym) *)s1->dynsym->data)[sym_index];
464 sym->st_value = esym->st_value;
465 goto found;
468 /* XXX: _fp_hw seems to be part of the ABI, so we ignore
469 it */
470 if (!strcmp(name, "_fp_hw"))
471 goto found;
472 /* only weak symbols are accepted to be undefined. Their
473 value is zero */
474 sym_bind = ELFW(ST_BIND)(sym->st_info);
475 if (sym_bind == STB_WEAK) {
476 sym->st_value = 0;
477 } else {
478 tcc_error_noabort("undefined symbol '%s'", name);
480 } else if (sh_num < SHN_LORESERVE) {
481 /* add section base */
482 sym->st_value += s1->sections[sym->st_shndx]->sh_addr;
484 found: ;
488 /* relocate a given section (CPU dependent) by applying the relocations
489 in the associated relocation section */
490 ST_FUNC void relocate_section(TCCState *s1, Section *s)
492 Section *sr = s->reloc;
493 ElfW_Rel *rel;
494 ElfW(Sym) *sym;
495 int type, sym_index;
496 unsigned char *ptr;
497 addr_t val, addr;
498 #if defined TCC_TARGET_I386 || defined TCC_TARGET_X86_64
499 ElfW_Rel *qrel = (ElfW_Rel *) sr->data; /* ptr to next reloc entry reused */
500 int esym_index;
501 #endif
503 for_each_elem(sr, 0, rel, ElfW_Rel) {
504 ptr = s->data + rel->r_offset;
506 sym_index = ELFW(R_SYM)(rel->r_info);
507 sym = &((ElfW(Sym) *)symtab_section->data)[sym_index];
508 val = sym->st_value;
509 #if defined(TCC_TARGET_ARM64) || defined(TCC_TARGET_X86_64)
510 val += rel->r_addend;
511 #endif
512 type = ELFW(R_TYPE)(rel->r_info);
513 addr = s->sh_addr + rel->r_offset;
515 /* CPU specific */
516 switch(type) {
517 #if defined(TCC_TARGET_I386)
518 case R_386_32:
519 if (s1->output_type == TCC_OUTPUT_DLL) {
520 esym_index = s1->symtab_to_dynsym[sym_index];
521 qrel->r_offset = rel->r_offset;
522 if (esym_index) {
523 qrel->r_info = ELFW(R_INFO)(esym_index, R_386_32);
524 qrel++;
525 break;
526 } else {
527 qrel->r_info = ELFW(R_INFO)(0, R_386_RELATIVE);
528 qrel++;
531 write32le(ptr, read32le(ptr) + val);
532 break;
533 case R_386_PC32:
534 if (s1->output_type == TCC_OUTPUT_DLL) {
535 /* DLL relocation */
536 esym_index = s1->symtab_to_dynsym[sym_index];
537 if (esym_index) {
538 qrel->r_offset = rel->r_offset;
539 qrel->r_info = ELFW(R_INFO)(esym_index, R_386_PC32);
540 qrel++;
541 break;
544 write32le(ptr, read32le(ptr) + val - addr);
545 break;
546 case R_386_PLT32:
547 write32le(ptr, read32le(ptr) + val - addr);
548 break;
549 case R_386_GLOB_DAT:
550 case R_386_JMP_SLOT:
551 write32le(ptr, val);
552 break;
553 case R_386_GOTPC:
554 write32le(ptr, read32le(ptr) + s1->got->sh_addr - addr);
555 break;
556 case R_386_GOTOFF:
557 write32le(ptr, read32le(ptr) + val - s1->got->sh_addr);
558 break;
559 case R_386_GOT32:
560 case R_386_GOT32X:
561 /* we load the got offset */
562 write32le(ptr, read32le(ptr) + s1->sym_attrs[sym_index].got_offset);
563 break;
564 case R_386_16:
565 if (s1->output_format != TCC_OUTPUT_FORMAT_BINARY) {
566 output_file:
567 tcc_error("can only produce 16-bit binary files");
569 write16le(ptr, read16le(ptr) + val);
570 break;
571 case R_386_PC16:
572 if (s1->output_format != TCC_OUTPUT_FORMAT_BINARY)
573 goto output_file;
574 write16le(ptr, read16le(ptr) + val - addr);
575 break;
576 default:
577 fprintf(stderr,"FIXME: handle reloc type %d at %x [%p] to %x\n",
578 type, (unsigned)addr, ptr, (unsigned)val);
579 break;
580 #elif defined(TCC_TARGET_ARM)
581 case R_ARM_PC24:
582 case R_ARM_CALL:
583 case R_ARM_JUMP24:
584 case R_ARM_PLT32:
586 int x, is_thumb, is_call, h, blx_avail, is_bl, th_ko;
587 x = (*(int *) ptr) & 0xffffff;
588 if (sym->st_shndx == SHN_UNDEF)
589 val = s1->plt->sh_addr;
590 #ifdef DEBUG_RELOC
591 printf ("reloc %d: x=0x%x val=0x%x ", type, x, val);
592 #endif
593 (*(int *)ptr) &= 0xff000000;
594 if (x & 0x800000)
595 x -= 0x1000000;
596 x <<= 2;
597 blx_avail = (TCC_ARM_VERSION >= 5);
598 is_thumb = val & 1;
599 is_bl = (*(unsigned *) ptr) >> 24 == 0xeb;
600 is_call = (type == R_ARM_CALL || (type == R_ARM_PC24 && is_bl));
601 x += val - addr;
602 #ifdef DEBUG_RELOC
603 printf (" newx=0x%x name=%s\n", x,
604 (char *) symtab_section->link->data + sym->st_name);
605 #endif
606 h = x & 2;
607 th_ko = (x & 3) && (!blx_avail || !is_call);
608 if (th_ko || x >= 0x2000000 || x < -0x2000000)
609 tcc_error("can't relocate value at %x,%d",addr, type);
610 x >>= 2;
611 x &= 0xffffff;
612 /* Only reached if blx is avail and it is a call */
613 if (is_thumb) {
614 x |= h << 24;
615 (*(int *)ptr) = 0xfa << 24; /* bl -> blx */
617 (*(int *) ptr) |= x;
619 break;
620 /* Since these relocations only concern Thumb-2 and blx instruction was
621 introduced before Thumb-2, we can assume blx is available and not
622 guard its use */
623 case R_ARM_THM_PC22:
624 case R_ARM_THM_JUMP24:
626 int x, hi, lo, s, j1, j2, i1, i2, imm10, imm11;
627 int to_thumb, is_call, to_plt, blx_bit = 1 << 12;
628 Section *plt;
630 /* weak reference */
631 if (sym->st_shndx == SHN_UNDEF &&
632 ELFW(ST_BIND)(sym->st_info) == STB_WEAK)
633 break;
635 /* Get initial offset */
636 hi = (*(uint16_t *)ptr);
637 lo = (*(uint16_t *)(ptr+2));
638 s = (hi >> 10) & 1;
639 j1 = (lo >> 13) & 1;
640 j2 = (lo >> 11) & 1;
641 i1 = (j1 ^ s) ^ 1;
642 i2 = (j2 ^ s) ^ 1;
643 imm10 = hi & 0x3ff;
644 imm11 = lo & 0x7ff;
645 x = (s << 24) | (i1 << 23) | (i2 << 22) |
646 (imm10 << 12) | (imm11 << 1);
647 if (x & 0x01000000)
648 x -= 0x02000000;
650 /* Relocation infos */
651 to_thumb = val & 1;
652 plt = s1->plt;
653 to_plt = (val >= plt->sh_addr) &&
654 (val < plt->sh_addr + plt->data_offset);
655 is_call = (type == R_ARM_THM_PC22);
657 /* Compute final offset */
658 if (to_plt && !is_call) /* Point to 1st instr of Thumb stub */
659 x -= 4;
660 x += val - addr;
661 if (!to_thumb && is_call) {
662 blx_bit = 0; /* bl -> blx */
663 x = (x + 3) & -4; /* Compute offset from aligned PC */
666 /* Check that relocation is possible
667 * offset must not be out of range
668 * if target is to be entered in arm mode:
669 - bit 1 must not set
670 - instruction must be a call (bl) or a jump to PLT */
671 if (!to_thumb || x >= 0x1000000 || x < -0x1000000)
672 if (to_thumb || (val & 2) || (!is_call && !to_plt))
673 tcc_error("can't relocate value at %x,%d",addr, type);
675 /* Compute and store final offset */
676 s = (x >> 24) & 1;
677 i1 = (x >> 23) & 1;
678 i2 = (x >> 22) & 1;
679 j1 = s ^ (i1 ^ 1);
680 j2 = s ^ (i2 ^ 1);
681 imm10 = (x >> 12) & 0x3ff;
682 imm11 = (x >> 1) & 0x7ff;
683 (*(uint16_t *)ptr) = (uint16_t) ((hi & 0xf800) |
684 (s << 10) | imm10);
685 (*(uint16_t *)(ptr+2)) = (uint16_t) ((lo & 0xc000) |
686 (j1 << 13) | blx_bit | (j2 << 11) |
687 imm11);
689 break;
690 case R_ARM_MOVT_ABS:
691 case R_ARM_MOVW_ABS_NC:
693 int x, imm4, imm12;
694 if (type == R_ARM_MOVT_ABS)
695 val >>= 16;
696 imm12 = val & 0xfff;
697 imm4 = (val >> 12) & 0xf;
698 x = (imm4 << 16) | imm12;
699 if (type == R_ARM_THM_MOVT_ABS)
700 *(int *)ptr |= x;
701 else
702 *(int *)ptr += x;
704 break;
705 case R_ARM_THM_MOVT_ABS:
706 case R_ARM_THM_MOVW_ABS_NC:
708 int x, i, imm4, imm3, imm8;
709 if (type == R_ARM_THM_MOVT_ABS)
710 val >>= 16;
711 imm8 = val & 0xff;
712 imm3 = (val >> 8) & 0x7;
713 i = (val >> 11) & 1;
714 imm4 = (val >> 12) & 0xf;
715 x = (imm3 << 28) | (imm8 << 16) | (i << 10) | imm4;
716 if (type == R_ARM_THM_MOVT_ABS)
717 *(int *)ptr |= x;
718 else
719 *(int *)ptr += x;
721 break;
722 case R_ARM_PREL31:
724 int x;
725 x = (*(int *)ptr) & 0x7fffffff;
726 (*(int *)ptr) &= 0x80000000;
727 x = (x * 2) / 2;
728 x += val - addr;
729 if((x^(x>>1))&0x40000000)
730 tcc_error("can't relocate value at %x,%d",addr, type);
731 (*(int *)ptr) |= x & 0x7fffffff;
733 case R_ARM_ABS32:
734 *(int *)ptr += val;
735 break;
736 case R_ARM_REL32:
737 *(int *)ptr += val - addr;
738 break;
739 case R_ARM_GOTPC:
740 *(int *)ptr += s1->got->sh_addr - addr;
741 break;
742 case R_ARM_GOTOFF:
743 *(int *)ptr += val - s1->got->sh_addr;
744 break;
745 case R_ARM_GOT32:
746 /* we load the got offset */
747 *(int *)ptr += s1->sym_attrs[sym_index].got_offset;
748 break;
749 case R_ARM_COPY:
750 break;
751 case R_ARM_V4BX:
752 /* trade Thumb support for ARMv4 support */
753 if ((0x0ffffff0 & *(int*)ptr) == 0x012FFF10)
754 *(int*)ptr ^= 0xE12FFF10 ^ 0xE1A0F000; /* BX Rm -> MOV PC, Rm */
755 break;
756 case R_ARM_GLOB_DAT:
757 case R_ARM_JUMP_SLOT:
758 *(addr_t *)ptr = val;
759 break;
760 case R_ARM_NONE:
761 /* Nothing to do. Normally used to indicate a dependency
762 on a certain symbol (like for exception handling under EABI). */
763 break;
764 default:
765 fprintf(stderr,"FIXME: handle reloc type %x at %x [%p] to %x\n",
766 type, (unsigned)addr, ptr, (unsigned)val);
767 break;
768 #elif defined(TCC_TARGET_ARM64)
769 case R_AARCH64_ABS64:
770 write64le(ptr, val);
771 break;
772 case R_AARCH64_ABS32:
773 write32le(ptr, val);
774 break;
775 case R_AARCH64_MOVW_UABS_G0_NC:
776 write32le(ptr, ((read32le(ptr) & 0xffe0001f) |
777 (val & 0xffff) << 5));
778 break;
779 case R_AARCH64_MOVW_UABS_G1_NC:
780 write32le(ptr, ((read32le(ptr) & 0xffe0001f) |
781 (val >> 16 & 0xffff) << 5));
782 break;
783 case R_AARCH64_MOVW_UABS_G2_NC:
784 write32le(ptr, ((read32le(ptr) & 0xffe0001f) |
785 (val >> 32 & 0xffff) << 5));
786 break;
787 case R_AARCH64_MOVW_UABS_G3:
788 write32le(ptr, ((read32le(ptr) & 0xffe0001f) |
789 (val >> 48 & 0xffff) << 5));
790 break;
791 case R_AARCH64_ADR_PREL_PG_HI21: {
792 uint64_t off = (val >> 12) - (addr >> 12);
793 if ((off + ((uint64_t)1 << 20)) >> 21)
794 tcc_error("R_AARCH64_ADR_PREL_PG_HI21 relocation failed");
795 write32le(ptr, ((read32le(ptr) & 0x9f00001f) |
796 (off & 0x1ffffc) << 3 | (off & 3) << 29));
797 break;
799 case R_AARCH64_ADD_ABS_LO12_NC:
800 write32le(ptr, ((read32le(ptr) & 0xffc003ff) |
801 (val & 0xfff) << 10));
802 break;
803 case R_AARCH64_JUMP26:
804 case R_AARCH64_CALL26:
805 /* This check must match the one in build_got_entries, testing
806 if we really need a PLT slot. */
807 if (sym->st_shndx == SHN_UNDEF)
808 /* We've put the PLT slot offset into r_addend when generating
809 it, and that's what we must use as relocation value (adjusted
810 by section offset of course). */
811 val = s1->plt->sh_addr + rel->r_addend;
812 #ifdef DEBUG_RELOC
813 printf ("reloc %d @ 0x%lx: val=0x%lx name=%s\n", type, addr, val,
814 (char *) symtab_section->link->data + sym->st_name);
815 #endif
816 if (((val - addr) + ((uint64_t)1 << 27)) & ~(uint64_t)0xffffffc)
818 tcc_error("R_AARCH64_(JUMP|CALL)26 relocation failed (val=%lx, addr=%lx)", addr, val);
820 write32le(ptr, (0x14000000 |
821 (uint32_t)(type == R_AARCH64_CALL26) << 31 |
822 ((val - addr) >> 2 & 0x3ffffff)));
823 break;
824 case R_AARCH64_ADR_GOT_PAGE: {
825 uint64_t off =
826 (((s1->got->sh_addr +
827 s1->sym_attrs[sym_index].got_offset) >> 12) - (addr >> 12));
828 if ((off + ((uint64_t)1 << 20)) >> 21)
829 tcc_error("R_AARCH64_ADR_GOT_PAGE relocation failed");
830 write32le(ptr, ((read32le(ptr) & 0x9f00001f) |
831 (off & 0x1ffffc) << 3 | (off & 3) << 29));
832 break;
834 case R_AARCH64_LD64_GOT_LO12_NC:
835 write32le(ptr,
836 ((read32le(ptr) & 0xfff803ff) |
837 ((s1->got->sh_addr +
838 s1->sym_attrs[sym_index].got_offset) & 0xff8) << 7));
839 break;
840 case R_AARCH64_COPY:
841 break;
842 case R_AARCH64_GLOB_DAT:
843 case R_AARCH64_JUMP_SLOT:
844 /* They don't need addend */
845 #ifdef DEBUG_RELOC
846 printf ("reloc %d @ 0x%lx: val=0x%lx name=%s\n", type, addr,
847 val - rel->r_addend,
848 (char *) symtab_section->link->data + sym->st_name);
849 #endif
850 write64le(ptr, val - rel->r_addend);
851 break;
852 default:
853 fprintf(stderr, "FIXME: handle reloc type %x at %x [%p] to %x\n",
854 type, (unsigned)addr, ptr, (unsigned)val);
855 break;
856 #elif defined(TCC_TARGET_C67)
857 case R_C60_32:
858 *(int *)ptr += val;
859 break;
860 case R_C60LO16:
862 uint32_t orig;
864 /* put the low 16 bits of the absolute address
865 add to what is already there */
867 orig = ((*(int *)(ptr )) >> 7) & 0xffff;
868 orig |= (((*(int *)(ptr+4)) >> 7) & 0xffff) << 16;
870 /* patch both at once - assumes always in pairs Low - High */
872 *(int *) ptr = (*(int *) ptr & (~(0xffff << 7)) ) | (((val+orig) & 0xffff) << 7);
873 *(int *)(ptr+4) = (*(int *)(ptr+4) & (~(0xffff << 7)) ) | ((((val+orig)>>16) & 0xffff) << 7);
875 break;
876 case R_C60HI16:
877 break;
878 default:
879 fprintf(stderr,"FIXME: handle reloc type %x at %x [%p] to %x\n",
880 type, (unsigned)addr, ptr, (unsigned)val);
881 break;
882 #elif defined(TCC_TARGET_X86_64)
883 case R_X86_64_64:
884 if (s1->output_type == TCC_OUTPUT_DLL) {
885 esym_index = s1->symtab_to_dynsym[sym_index];
886 qrel->r_offset = rel->r_offset;
887 if (esym_index) {
888 qrel->r_info = ELFW(R_INFO)(esym_index, R_X86_64_64);
889 qrel->r_addend = rel->r_addend;
890 qrel++;
891 break;
892 } else {
893 qrel->r_info = ELFW(R_INFO)(0, R_X86_64_RELATIVE);
894 qrel->r_addend = read64le(ptr) + val;
895 qrel++;
898 write64le(ptr, read64le(ptr) + val);
899 break;
900 case R_X86_64_32:
901 case R_X86_64_32S:
902 if (s1->output_type == TCC_OUTPUT_DLL) {
903 /* XXX: this logic may depend on TCC's codegen
904 now TCC uses R_X86_64_32 even for a 64bit pointer */
905 qrel->r_info = ELFW(R_INFO)(0, R_X86_64_RELATIVE);
906 /* Use sign extension! */
907 qrel->r_addend = (int)read32le(ptr) + val;
908 qrel++;
910 write32le(ptr, read32le(ptr) + val);
911 break;
913 case R_X86_64_PC32:
914 if (s1->output_type == TCC_OUTPUT_DLL) {
915 /* DLL relocation */
916 esym_index = s1->symtab_to_dynsym[sym_index];
917 if (esym_index) {
918 qrel->r_offset = rel->r_offset;
919 qrel->r_info = ELFW(R_INFO)(esym_index, R_X86_64_PC32);
920 /* Use sign extension! */
921 qrel->r_addend = (int)read32le(ptr);
922 qrel++;
923 break;
926 goto plt32pc32;
928 case R_X86_64_PLT32:
929 /* We've put the PLT slot offset into r_addend when generating
930 it, and that's what we must use as relocation value (adjusted
931 by section offset of course). */
932 val = s1->plt->sh_addr + rel->r_addend;
933 /* fallthrough. */
935 plt32pc32:
937 long long diff;
938 diff = (long long)val - addr;
939 if (diff < -2147483648LL || diff > 2147483647LL) {
940 tcc_error("internal error: relocation failed");
942 write32le(ptr, read32le(ptr) + diff);
944 break;
945 case R_X86_64_GLOB_DAT:
946 case R_X86_64_JUMP_SLOT:
947 /* They don't need addend */
948 write64le(ptr, val - rel->r_addend);
949 break;
950 case R_X86_64_GOTPCREL:
951 case R_X86_64_GOTPCRELX:
952 case R_X86_64_REX_GOTPCRELX:
953 write32le(ptr, read32le(ptr) +
954 (s1->got->sh_addr - addr +
955 s1->sym_attrs[sym_index].got_offset - 4));
956 break;
957 case R_X86_64_GOTTPOFF:
958 write32le(ptr, read32le(ptr) + val - s1->got->sh_addr);
959 break;
960 case R_X86_64_GOT32:
961 /* we load the got offset */
962 write32le(ptr, read32le(ptr) + s1->sym_attrs[sym_index].got_offset);
963 break;
964 #else
965 #error unsupported processor
966 #endif
969 /* if the relocation is allocated, we change its symbol table */
970 if (sr->sh_flags & SHF_ALLOC)
971 sr->link = s1->dynsym;
974 /* relocate relocation table in 'sr' */
975 static void relocate_rel(TCCState *s1, Section *sr)
977 Section *s;
978 ElfW_Rel *rel;
980 s = s1->sections[sr->sh_info];
981 for_each_elem(sr, 0, rel, ElfW_Rel)
982 rel->r_offset += s->sh_addr;
985 /* count the number of dynamic relocations so that we can reserve
986 their space */
987 static int prepare_dynamic_rel(TCCState *s1, Section *sr)
989 ElfW_Rel *rel;
990 int sym_index, esym_index, type, count;
992 count = 0;
993 for_each_elem(sr, 0, rel, ElfW_Rel) {
994 sym_index = ELFW(R_SYM)(rel->r_info);
995 type = ELFW(R_TYPE)(rel->r_info);
996 switch(type) {
997 #if defined(TCC_TARGET_I386)
998 case R_386_32:
999 #elif defined(TCC_TARGET_X86_64)
1000 case R_X86_64_32:
1001 case R_X86_64_32S:
1002 case R_X86_64_64:
1003 #endif
1004 count++;
1005 break;
1006 #if defined(TCC_TARGET_I386)
1007 case R_386_PC32:
1008 #elif defined(TCC_TARGET_X86_64)
1009 case R_X86_64_PC32:
1010 #endif
1011 esym_index = s1->symtab_to_dynsym[sym_index];
1012 if (esym_index)
1013 count++;
1014 break;
1015 default:
1016 break;
1019 if (count) {
1020 /* allocate the section */
1021 sr->sh_flags |= SHF_ALLOC;
1022 sr->sh_size = count * sizeof(ElfW_Rel);
1024 return count;
1027 static struct sym_attr *alloc_sym_attr(TCCState *s1, int index)
1029 int n;
1030 struct sym_attr *tab;
1032 if (index >= s1->nb_sym_attrs) {
1033 /* find immediately bigger power of 2 and reallocate array */
1034 n = 1;
1035 while (index >= n)
1036 n *= 2;
1037 tab = tcc_realloc(s1->sym_attrs, n * sizeof(*s1->sym_attrs));
1038 s1->sym_attrs = tab;
1039 memset(s1->sym_attrs + s1->nb_sym_attrs, 0,
1040 (n - s1->nb_sym_attrs) * sizeof(*s1->sym_attrs));
1041 s1->nb_sym_attrs = n;
1043 return &s1->sym_attrs[index];
1046 static void build_got(TCCState *s1)
1048 unsigned char *ptr;
1050 /* if no got, then create it */
1051 s1->got = new_section(s1, ".got", SHT_PROGBITS, SHF_ALLOC | SHF_WRITE);
1052 s1->got->sh_entsize = 4;
1053 add_elf_sym(symtab_section, 0, 4, ELFW(ST_INFO)(STB_GLOBAL, STT_OBJECT),
1054 0, s1->got->sh_num, "_GLOBAL_OFFSET_TABLE_");
1055 ptr = section_ptr_add(s1->got, 3 * PTR_SIZE);
1056 #if PTR_SIZE == 4
1057 /* keep space for _DYNAMIC pointer, if present */
1058 write32le(ptr, 0);
1059 /* two dummy got entries */
1060 write32le(ptr + 4, 0);
1061 write32le(ptr + 8, 0);
1062 #else
1063 /* keep space for _DYNAMIC pointer, if present */
1064 write32le(ptr, 0);
1065 write32le(ptr + 4, 0);
1066 /* two dummy got entries */
1067 write32le(ptr + 8, 0);
1068 write32le(ptr + 12, 0);
1069 write32le(ptr + 16, 0);
1070 write32le(ptr + 20, 0);
1071 #endif
1074 /* put a got or plt entry corresponding to a symbol in symtab_section. 'size'
1075 and 'info' can be modifed if more precise info comes from the DLL.
1076 Returns offset of GOT or PLT slot. */
1077 static unsigned long put_got_entry(TCCState *s1,
1078 int reloc_type, unsigned long size, int info,
1079 int sym_index)
1081 int index, need_plt_entry;
1082 const char *name;
1083 ElfW(Sym) *sym;
1084 unsigned long offset;
1085 int *ptr;
1086 struct sym_attr *symattr;
1088 if (!s1->got)
1089 build_got(s1);
1091 need_plt_entry =
1092 #ifdef TCC_TARGET_X86_64
1093 (reloc_type == R_X86_64_JUMP_SLOT);
1094 #elif defined(TCC_TARGET_I386)
1095 (reloc_type == R_386_JMP_SLOT);
1096 #elif defined(TCC_TARGET_ARM)
1097 (reloc_type == R_ARM_JUMP_SLOT);
1098 #elif defined(TCC_TARGET_ARM64)
1099 (reloc_type == R_AARCH64_JUMP_SLOT);
1100 #else
1101 0;
1102 #endif
1104 if (need_plt_entry && !s1->plt) {
1105 /* add PLT */
1106 s1->plt = new_section(s1, ".plt", SHT_PROGBITS,
1107 SHF_ALLOC | SHF_EXECINSTR);
1108 s1->plt->sh_entsize = 4;
1111 /* If a got/plt entry already exists for that symbol, no need to add one */
1112 if (sym_index < s1->nb_sym_attrs) {
1113 if (need_plt_entry && s1->sym_attrs[sym_index].plt_offset)
1114 return s1->sym_attrs[sym_index].plt_offset;
1115 else if (!need_plt_entry && s1->sym_attrs[sym_index].got_offset)
1116 return s1->sym_attrs[sym_index].got_offset;
1119 symattr = alloc_sym_attr(s1, sym_index);
1121 /* Only store the GOT offset if it's not generated for the PLT entry. */
1122 if (!need_plt_entry)
1123 symattr->got_offset = s1->got->data_offset;
1125 sym = &((ElfW(Sym) *)symtab_section->data)[sym_index];
1126 name = (char *) symtab_section->link->data + sym->st_name;
1127 offset = sym->st_value;
1128 #if defined(TCC_TARGET_I386) || defined(TCC_TARGET_X86_64)
1129 if (need_plt_entry) {
1130 Section *plt;
1131 uint8_t *p;
1132 int modrm;
1133 unsigned long relofs;
1135 #if defined(TCC_OUTPUT_DLL_WITH_PLT)
1136 modrm = 0x25;
1137 #else
1138 /* if we build a DLL, we add a %ebx offset */
1139 if (s1->output_type == TCC_OUTPUT_DLL)
1140 modrm = 0xa3;
1141 else
1142 modrm = 0x25;
1143 #endif
1145 /* add a PLT entry */
1146 plt = s1->plt;
1147 if (plt->data_offset == 0) {
1148 /* first plt entry */
1149 p = section_ptr_add(plt, 16);
1150 p[0] = 0xff; /* pushl got + PTR_SIZE */
1151 p[1] = modrm + 0x10;
1152 write32le(p + 2, PTR_SIZE);
1153 p[6] = 0xff; /* jmp *(got + PTR_SIZE * 2) */
1154 p[7] = modrm;
1155 write32le(p + 8, PTR_SIZE * 2);
1158 /* The PLT slot refers to the relocation entry it needs
1159 via offset. The reloc entry is created below, so its
1160 offset is the current data_offset. */
1161 relofs = s1->got->reloc ? s1->got->reloc->data_offset : 0;
1162 symattr->plt_offset = plt->data_offset;
1163 p = section_ptr_add(plt, 16);
1164 p[0] = 0xff; /* jmp *(got + x) */
1165 p[1] = modrm;
1166 write32le(p + 2, s1->got->data_offset);
1167 p[6] = 0x68; /* push $xxx */
1168 #ifdef TCC_TARGET_X86_64
1169 /* On x86-64, the relocation is referred to by _index_. */
1170 write32le(p + 7, relofs / sizeof (ElfW_Rel));
1171 #else
1172 write32le(p + 7, relofs);
1173 #endif
1174 p[11] = 0xe9; /* jmp plt_start */
1175 write32le(p + 12, -(plt->data_offset));
1177 /* If this was an UNDEF symbol set the offset in the
1178 dynsymtab to the PLT slot, so that PC32 relocs to it
1179 can be resolved. */
1180 if (sym->st_shndx == SHN_UNDEF)
1181 offset = plt->data_offset - 16;
1183 #elif defined(TCC_TARGET_ARM)
1184 if (need_plt_entry) {
1185 Section *plt;
1186 uint8_t *p;
1188 /* if we build a DLL, we add a %ebx offset */
1189 if (s1->output_type == TCC_OUTPUT_DLL)
1190 tcc_error("DLLs unimplemented!");
1192 /* add a PLT entry */
1193 plt = s1->plt;
1194 if (plt->data_offset == 0) {
1195 /* first plt entry */
1196 p = section_ptr_add(plt, 16);
1197 write32le(p, 0xe52de004); /* push {lr} */
1198 write32le(p+4, 0xe59fe010); /* ldr lr, [pc, #16] */
1199 write32le(p+8, 0xe08fe00e); /* add lr, pc, lr */
1200 write32le(p+12, 0xe5bef008); /* ldr pc, [lr, #8]! */
1203 symattr->plt_offset = plt->data_offset;
1204 if (symattr->plt_thumb_stub) {
1205 p = section_ptr_add(plt, 20);
1206 write32le(p, 0x4778); /* bx pc */
1207 write32le(p+2, 0x46c0); /* nop */
1208 p += 4;
1209 } else
1210 p = section_ptr_add(plt, 16);
1211 write32le(p, 0xe59fc004); /* ldr ip, [pc, #4] ; GOT entry offset */
1212 write32le(p+4, 0xe08fc00c); /* add ip, pc, ip ; addr of GOT entry */
1213 write32le(p+8, 0xe59cf000); /* ldr pc, [ip] ; jump to GOT entry */
1214 write32le(p+12, s1->got->data_offset); /* GOT entry off once patched */
1216 /* the symbol is modified so that it will be relocated to
1217 the PLT */
1218 if (sym->st_shndx == SHN_UNDEF)
1219 offset = plt->data_offset - 16;
1221 #elif defined(TCC_TARGET_ARM64)
1222 if (need_plt_entry) {
1223 Section *plt;
1224 uint8_t *p;
1226 if (s1->output_type == TCC_OUTPUT_DLL)
1227 tcc_error("DLLs unimplemented!");
1229 plt = s1->plt;
1230 if (plt->data_offset == 0)
1231 section_ptr_add(plt, 32);
1232 symattr->plt_offset = plt->data_offset;
1233 p = section_ptr_add(plt, 16);
1234 write32le(p, s1->got->data_offset);
1235 write32le(p + 4, (uint64_t)s1->got->data_offset >> 32);
1237 if (sym->st_shndx == SHN_UNDEF)
1238 offset = plt->data_offset - 16;
1240 #elif defined(TCC_TARGET_C67)
1241 if (s1->dynsym) {
1242 tcc_error("C67 got not implemented");
1244 #else
1245 #error unsupported CPU
1246 #endif
1247 if (s1->dynsym) {
1248 /* XXX This might generate multiple syms for name. */
1249 index = put_elf_sym(s1->dynsym, offset,
1250 size, info, 0, sym->st_shndx, name);
1251 /* Create the relocation (it's against the GOT for PLT
1252 and GOT relocs). */
1253 put_elf_reloc(s1->dynsym, s1->got,
1254 s1->got->data_offset,
1255 reloc_type, index);
1256 } else {
1257 /* Without .dynsym (i.e. static link or memory output) we
1258 still need relocs against the generated got, so as to fill
1259 the entries with the symbol values (determined later). */
1260 put_elf_reloc(symtab_section, s1->got,
1261 s1->got->data_offset,
1262 reloc_type, sym_index);
1264 /* And now create the GOT slot itself. */
1265 ptr = section_ptr_add(s1->got, PTR_SIZE);
1266 *ptr = 0;
1267 if (need_plt_entry)
1268 return symattr->plt_offset;
1269 else
1270 return symattr->got_offset;
1273 /* build GOT and PLT entries */
1274 ST_FUNC void build_got_entries(TCCState *s1)
1276 Section *s;
1277 ElfW_Rel *rel;
1278 ElfW(Sym) *sym;
1279 int i, type, reloc_type, sym_index;
1281 for(i = 1; i < s1->nb_sections; i++) {
1282 s = s1->sections[i];
1283 if (s->sh_type != SHT_RELX)
1284 continue;
1285 /* no need to handle got relocations */
1286 if (s->link != symtab_section)
1287 continue;
1288 for_each_elem(s, 0, rel, ElfW_Rel) {
1289 type = ELFW(R_TYPE)(rel->r_info);
1290 switch(type) {
1291 #if defined(TCC_TARGET_I386)
1292 case R_386_GOT32:
1293 case R_386_GOT32X:
1294 case R_386_GOTOFF:
1295 case R_386_GOTPC:
1296 case R_386_PLT32:
1297 if (!s1->got)
1298 build_got(s1);
1299 if (type == R_386_GOT32 || type == R_386_GOT32X ||
1300 type == R_386_PLT32) {
1301 sym_index = ELFW(R_SYM)(rel->r_info);
1302 sym = &((ElfW(Sym) *)symtab_section->data)[sym_index];
1303 /* look at the symbol got offset. If none, then add one */
1304 if (type == R_386_GOT32 || type == R_386_GOT32X)
1305 reloc_type = R_386_GLOB_DAT;
1306 else
1307 reloc_type = R_386_JMP_SLOT;
1308 put_got_entry(s1, reloc_type, sym->st_size, sym->st_info,
1309 sym_index);
1311 break;
1312 #elif defined(TCC_TARGET_ARM)
1313 case R_ARM_PC24:
1314 case R_ARM_CALL:
1315 case R_ARM_JUMP24:
1316 case R_ARM_GOT32:
1317 case R_ARM_GOTOFF:
1318 case R_ARM_GOTPC:
1319 case R_ARM_PLT32:
1320 if (!s1->got)
1321 build_got(s1);
1322 sym_index = ELFW(R_SYM)(rel->r_info);
1323 sym = &((ElfW(Sym) *)symtab_section->data)[sym_index];
1324 if (type != R_ARM_GOTOFF && type != R_ARM_GOTPC
1325 && sym->st_shndx == SHN_UNDEF) {
1326 unsigned long ofs;
1327 /* look at the symbol got offset. If none, then add one */
1328 if (type == R_ARM_GOT32)
1329 reloc_type = R_ARM_GLOB_DAT;
1330 else
1331 reloc_type = R_ARM_JUMP_SLOT;
1332 ofs = put_got_entry(s1, reloc_type, sym->st_size,
1333 sym->st_info, sym_index);
1334 #ifdef DEBUG_RELOC
1335 printf ("maybegot: %s, %d, %d --> ofs=0x%x\n",
1336 (char *) symtab_section->link->data + sym->st_name,
1337 type, sym->st_shndx, ofs);
1338 #endif
1339 if (type != R_ARM_GOT32) {
1340 addr_t *ptr = (addr_t*)(s1->sections[s->sh_info]->data
1341 + rel->r_offset);
1342 /* x must be signed! */
1343 int x = *ptr & 0xffffff;
1344 x = (x << 8) >> 8;
1345 x <<= 2;
1346 x += ofs;
1347 x >>= 2;
1348 #ifdef DEBUG_RELOC
1349 printf ("insn=0x%x --> 0x%x (x==0x%x)\n", *ptr,
1350 (*ptr & 0xff000000) | x, x);
1351 #endif
1352 *ptr = (*ptr & 0xff000000) | x;
1355 break;
1356 case R_ARM_THM_JUMP24:
1357 sym_index = ELFW(R_SYM)(rel->r_info);
1358 sym = &((ElfW(Sym) *)symtab_section->data)[sym_index];
1359 /* We are relocating a jump from thumb code to arm code */
1360 if (sym->st_shndx != SHN_UNDEF && !(sym->st_value & 1)) {
1361 int index;
1362 uint8_t *p;
1363 char *name, buf[1024];
1364 Section *text_section;
1366 name = (char *) symtab_section->link->data + sym->st_name;
1367 text_section = s1->sections[sym->st_shndx];
1368 /* Modify reloc to target a thumb stub to switch to ARM */
1369 snprintf(buf, sizeof(buf), "%s_from_thumb", name);
1370 index = put_elf_sym(symtab_section,
1371 text_section->data_offset + 1,
1372 sym->st_size, sym->st_info, 0,
1373 sym->st_shndx, buf);
1374 rel->r_info = ELFW(R_INFO)(index, type);
1375 /* Create a thumb stub fonction to switch to ARM mode */
1376 put_elf_reloc(symtab_section, text_section,
1377 text_section->data_offset + 4, R_ARM_JUMP24,
1378 sym_index);
1379 p = section_ptr_add(text_section, 8);
1380 write32le(p, 0x4778); /* bx pc */
1381 write32le(p+2, 0x46c0); /* nop */
1382 write32le(p+4, 0xeafffffe); /* b $sym */
1384 #elif defined(TCC_TARGET_ARM64)
1385 //xx Other cases may be required here:
1386 case R_AARCH64_ADR_GOT_PAGE:
1387 case R_AARCH64_LD64_GOT_LO12_NC:
1388 if (!s1->got)
1389 build_got(s1);
1390 sym_index = ELFW(R_SYM)(rel->r_info);
1391 sym = &((ElfW(Sym) *)symtab_section->data)[sym_index];
1392 reloc_type = R_AARCH64_GLOB_DAT;
1393 put_got_entry(s1, reloc_type, sym->st_size, sym->st_info,
1394 sym_index);
1395 break;
1397 case R_AARCH64_JUMP26:
1398 case R_AARCH64_CALL26:
1399 if (!s1->got)
1400 build_got(s1);
1401 sym_index = ELFW(R_SYM)(rel->r_info);
1402 sym = &((ElfW(Sym) *)symtab_section->data)[sym_index];
1403 if (sym->st_shndx == SHN_UNDEF) {
1404 unsigned long ofs;
1405 reloc_type = R_AARCH64_JUMP_SLOT;
1406 ofs = put_got_entry(s1, reloc_type, sym->st_size,
1407 sym->st_info, sym_index);
1408 /* We store the place of the generated PLT slot
1409 in our addend. */
1410 rel->r_addend += ofs;
1412 break;
1413 #elif defined(TCC_TARGET_C67)
1414 case R_C60_GOT32:
1415 case R_C60_GOTOFF:
1416 case R_C60_GOTPC:
1417 case R_C60_PLT32:
1418 if (!s1->got)
1419 build_got(s1);
1420 if (type == R_C60_GOT32 || type == R_C60_PLT32) {
1421 sym_index = ELFW(R_SYM)(rel->r_info);
1422 sym = &((ElfW(Sym) *)symtab_section->data)[sym_index];
1423 /* look at the symbol got offset. If none, then add one */
1424 if (type == R_C60_GOT32)
1425 reloc_type = R_C60_GLOB_DAT;
1426 else
1427 reloc_type = R_C60_JMP_SLOT;
1428 put_got_entry(s1, reloc_type, sym->st_size, sym->st_info,
1429 sym_index);
1431 break;
1432 #elif defined(TCC_TARGET_X86_64)
1433 case R_X86_64_GOT32:
1434 case R_X86_64_GOTTPOFF:
1435 case R_X86_64_GOTPCREL:
1436 case R_X86_64_GOTPCRELX:
1437 case R_X86_64_REX_GOTPCRELX:
1438 case R_X86_64_PLT32:
1439 sym_index = ELFW(R_SYM)(rel->r_info);
1440 sym = &((ElfW(Sym) *)symtab_section->data)[sym_index];
1441 if (type == R_X86_64_PLT32 &&
1442 ELFW(ST_VISIBILITY)(sym->st_other) != STV_DEFAULT)
1444 rel->r_info = ELFW(R_INFO)(sym_index, R_X86_64_PC32);
1445 break;
1448 if (!s1->got) {
1449 build_got(s1);
1450 sym = &((ElfW(Sym) *)symtab_section->data)[sym_index];
1452 if (type == R_X86_64_GOT32 || type == R_X86_64_GOTPCREL ||
1453 type == R_X86_64_GOTPCRELX ||
1454 type == R_X86_64_REX_GOTPCRELX ||
1455 type == R_X86_64_PLT32) {
1456 unsigned long ofs;
1457 /* look at the symbol got offset. If none, then add one */
1458 if (type == R_X86_64_PLT32)
1459 reloc_type = R_X86_64_JUMP_SLOT;
1460 else
1461 reloc_type = R_X86_64_GLOB_DAT;
1462 ofs = put_got_entry(s1, reloc_type, sym->st_size,
1463 sym->st_info, sym_index);
1464 if (type == R_X86_64_PLT32)
1465 /* We store the place of the generated PLT slot
1466 in our addend. */
1467 rel->r_addend += ofs;
1469 break;
1470 #else
1471 #error unsupported CPU
1472 #endif
1473 default:
1474 break;
1480 ST_FUNC Section *new_symtab(TCCState *s1,
1481 const char *symtab_name, int sh_type, int sh_flags,
1482 const char *strtab_name,
1483 const char *hash_name, int hash_sh_flags)
1485 Section *symtab, *strtab, *hash;
1486 int *ptr, nb_buckets;
1488 symtab = new_section(s1, symtab_name, sh_type, sh_flags);
1489 symtab->sh_entsize = sizeof(ElfW(Sym));
1490 strtab = new_section(s1, strtab_name, SHT_STRTAB, sh_flags);
1491 put_elf_str(strtab, "");
1492 symtab->link = strtab;
1493 put_elf_sym(symtab, 0, 0, 0, 0, 0, NULL);
1495 nb_buckets = 1;
1497 hash = new_section(s1, hash_name, SHT_HASH, hash_sh_flags);
1498 hash->sh_entsize = sizeof(int);
1499 symtab->hash = hash;
1500 hash->link = symtab;
1502 ptr = section_ptr_add(hash, (2 + nb_buckets + 1) * sizeof(int));
1503 ptr[0] = nb_buckets;
1504 ptr[1] = 1;
1505 memset(ptr + 2, 0, (nb_buckets + 1) * sizeof(int));
1506 return symtab;
1509 /* put dynamic tag */
1510 static void put_dt(Section *dynamic, int dt, addr_t val)
1512 ElfW(Dyn) *dyn;
1513 dyn = section_ptr_add(dynamic, sizeof(ElfW(Dyn)));
1514 dyn->d_tag = dt;
1515 dyn->d_un.d_val = val;
1518 static void add_init_array_defines(TCCState *s1, const char *section_name)
1520 Section *s;
1521 long end_offset;
1522 char sym_start[1024];
1523 char sym_end[1024];
1525 snprintf(sym_start, sizeof(sym_start), "__%s_start", section_name + 1);
1526 snprintf(sym_end, sizeof(sym_end), "__%s_end", section_name + 1);
1528 s = find_section(s1, section_name);
1529 if (!s) {
1530 end_offset = 0;
1531 s = data_section;
1532 } else {
1533 end_offset = s->data_offset;
1536 add_elf_sym(symtab_section,
1537 0, 0,
1538 ELFW(ST_INFO)(STB_GLOBAL, STT_NOTYPE), 0,
1539 s->sh_num, sym_start);
1540 add_elf_sym(symtab_section,
1541 end_offset, 0,
1542 ELFW(ST_INFO)(STB_GLOBAL, STT_NOTYPE), 0,
1543 s->sh_num, sym_end);
1546 static int tcc_add_support(TCCState *s1, const char *filename)
1548 char buf[1024];
1549 snprintf(buf, sizeof(buf), "%s/%s/%s", s1->tcc_lib_path,
1550 /* an cpu specific path inside tcc_lib_path, mainly for keeping libtcc1.a */
1551 #ifdef TCC_TARGET_I386
1552 "i386"
1553 #endif
1554 #ifdef TCC_TARGET_X86_64
1555 "x86-64"
1556 #endif
1557 #ifdef TCC_TARGET_ARM
1558 "arm"
1559 #endif
1560 #ifdef TCC_TARGET_ARM64
1561 "arm64"
1562 #endif
1563 #ifdef TCC_TARGET_C67
1564 "C67"
1565 #endif
1566 ,filename);
1568 return tcc_add_file(s1, buf, TCC_FILETYPE_BINARY);
1571 ST_FUNC void tcc_add_bcheck(TCCState *s1)
1573 #ifdef CONFIG_TCC_BCHECK
1574 addr_t *ptr;
1576 if (0 == s1->do_bounds_check)
1577 return;
1579 /* XXX: add an object file to do that */
1580 ptr = section_ptr_add(bounds_section, sizeof(*ptr));
1581 *ptr = 0;
1582 add_elf_sym(symtab_section, 0, 0,
1583 ELFW(ST_INFO)(STB_GLOBAL, STT_NOTYPE), 0,
1584 bounds_section->sh_num, "__bounds_start");
1585 if (s1->output_type != TCC_OUTPUT_MEMORY) {
1586 /* add 'call __bound_init()' in .init section */
1588 /* XXX not called on MSYS, reason is unknown. For this
1589 case a call to __bound_init is performed in bcheck.c
1590 when __bound_ptr_add, __bound_new_region,
1591 __bound_delete_region called */
1593 int sym_index = find_elf_sym(symtab_section, "__bound_init");
1594 if (sym_index) {
1595 Section *init_section = find_section(s1, ".init");
1596 unsigned char *pinit = section_ptr_add(init_section, 5);
1597 pinit[0] = 0xe8;
1598 write32le(pinit + 1, -4);
1599 put_elf_reloc(symtab_section, init_section,
1600 init_section->data_offset - 4, R_386_PC32, sym_index);
1602 else
1603 tcc_warning("__bound_init not defined");
1605 #endif
1608 /* add tcc runtime libraries */
1609 ST_FUNC void tcc_add_runtime(TCCState *s1)
1611 tcc_add_pragma_libs(s1);
1613 /* add libc */
1614 if (!s1->nostdlib) {
1615 tcc_add_library(s1, "c");
1616 #ifdef CONFIG_USE_LIBGCC
1617 if (!s1->static_link) {
1618 tcc_add_file(s1, TCC_LIBGCC, TCC_FILETYPE_BINARY);
1620 #endif
1621 tcc_add_support(s1, "libtcc1.a");
1624 /* tcc_add_bcheck tries to relocate a call to __bound_init in _init so
1625 libtcc1.a must be loaded before for __bound_init to be defined and
1626 crtn.o must be loaded after to not finalize _init too early. */
1627 tcc_add_bcheck(s1);
1629 if (!s1->nostdlib) {
1630 /* add crt end if not memory output */
1631 if (s1->output_type != TCC_OUTPUT_MEMORY)
1632 tcc_add_crt(s1, "crtn.o");
1636 /* add various standard linker symbols (must be done after the
1637 sections are filled (for example after allocating common
1638 symbols)) */
1639 ST_FUNC void tcc_add_linker_symbols(TCCState *s1)
1641 char buf[1024];
1642 int i;
1643 Section *s;
1645 add_elf_sym(symtab_section,
1646 text_section->data_offset, 0,
1647 ELFW(ST_INFO)(STB_GLOBAL, STT_NOTYPE), 0,
1648 text_section->sh_num, "_etext");
1649 add_elf_sym(symtab_section,
1650 data_section->data_offset, 0,
1651 ELFW(ST_INFO)(STB_GLOBAL, STT_NOTYPE), 0,
1652 data_section->sh_num, "_edata");
1653 add_elf_sym(symtab_section,
1654 bss_section->data_offset, 0,
1655 ELFW(ST_INFO)(STB_GLOBAL, STT_NOTYPE), 0,
1656 bss_section->sh_num, "_end");
1657 /* horrible new standard ldscript defines */
1658 add_init_array_defines(s1, ".preinit_array");
1659 add_init_array_defines(s1, ".init_array");
1660 add_init_array_defines(s1, ".fini_array");
1662 /* add start and stop symbols for sections whose name can be
1663 expressed in C */
1664 for(i = 1; i < s1->nb_sections; i++) {
1665 s = s1->sections[i];
1666 if (s->sh_type == SHT_PROGBITS &&
1667 (s->sh_flags & SHF_ALLOC)) {
1668 const char *p;
1669 int ch;
1671 /* check if section name can be expressed in C */
1672 p = s->name;
1673 for(;;) {
1674 ch = *p;
1675 if (!ch)
1676 break;
1677 if (!isid(ch) && !isnum(ch))
1678 goto next_sec;
1679 p++;
1681 snprintf(buf, sizeof(buf), "__start_%s", s->name);
1682 add_elf_sym(symtab_section,
1683 0, 0,
1684 ELFW(ST_INFO)(STB_GLOBAL, STT_NOTYPE), 0,
1685 s->sh_num, buf);
1686 snprintf(buf, sizeof(buf), "__stop_%s", s->name);
1687 add_elf_sym(symtab_section,
1688 s->data_offset, 0,
1689 ELFW(ST_INFO)(STB_GLOBAL, STT_NOTYPE), 0,
1690 s->sh_num, buf);
1692 next_sec: ;
1696 static void tcc_output_binary(TCCState *s1, FILE *f,
1697 const int *sec_order)
1699 Section *s;
1700 int i, offset, size;
1702 offset = 0;
1703 for(i=1;i<s1->nb_sections;i++) {
1704 s = s1->sections[sec_order[i]];
1705 if (s->sh_type != SHT_NOBITS &&
1706 (s->sh_flags & SHF_ALLOC)) {
1707 while (offset < s->sh_offset) {
1708 fputc(0, f);
1709 offset++;
1711 size = s->sh_size;
1712 fwrite(s->data, 1, size, f);
1713 offset += size;
1718 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
1719 #define HAVE_PHDR 1
1720 #define EXTRA_RELITEMS 14
1722 /* move the relocation value from .dynsym to .got */
1723 void patch_dynsym_undef(TCCState *s1, Section *s)
1725 uint32_t *gotd = (void *)s1->got->data;
1726 ElfW(Sym) *sym;
1728 gotd += 3; /* dummy entries in .got */
1729 /* relocate symbols in .dynsym */
1730 for_each_elem(s, 1, sym, ElfW(Sym)) {
1731 if (sym->st_shndx == SHN_UNDEF) {
1732 *gotd++ = sym->st_value + 6; /* XXX 6 is magic ? */
1733 sym->st_value = 0;
1737 #else
1738 #define HAVE_PHDR 1
1739 #define EXTRA_RELITEMS 9
1741 /* zero plt offsets of weak symbols in .dynsym */
1742 void patch_dynsym_undef(TCCState *s1, Section *s)
1744 ElfW(Sym) *sym;
1746 for_each_elem(s, 1, sym, ElfW(Sym))
1747 if (sym->st_shndx == SHN_UNDEF && ELFW(ST_BIND)(sym->st_info) == STB_WEAK)
1748 sym->st_value = 0;
1750 #endif
1752 ST_FUNC void fill_got_entry(TCCState *s1, ElfW_Rel *rel)
1754 int sym_index = ELFW(R_SYM) (rel->r_info);
1755 ElfW(Sym) *sym = &((ElfW(Sym) *) symtab_section->data)[sym_index];
1756 unsigned long offset;
1758 if (sym_index >= s1->nb_sym_attrs)
1759 return;
1760 offset = s1->sym_attrs[sym_index].got_offset;
1761 section_reserve(s1->got, offset + PTR_SIZE);
1762 #ifdef TCC_TARGET_X86_64
1763 /* only works for x86-64 */
1764 write32le(s1->got->data + offset + 4, sym->st_value >> 32);
1765 #endif
1766 write32le(s1->got->data + offset, sym->st_value & 0xffffffff);
1769 /* Perform relocation to GOT or PLT entries */
1770 ST_FUNC void fill_got(TCCState *s1)
1772 Section *s;
1773 ElfW_Rel *rel;
1774 int i;
1776 for(i = 1; i < s1->nb_sections; i++) {
1777 s = s1->sections[i];
1778 if (s->sh_type != SHT_RELX)
1779 continue;
1780 /* no need to handle got relocations */
1781 if (s->link != symtab_section)
1782 continue;
1783 for_each_elem(s, 0, rel, ElfW_Rel) {
1784 switch (ELFW(R_TYPE) (rel->r_info)) {
1785 case R_X86_64_GOT32:
1786 case R_X86_64_GOTPCREL:
1787 case R_X86_64_GOTPCRELX:
1788 case R_X86_64_REX_GOTPCRELX:
1789 case R_X86_64_PLT32:
1790 fill_got_entry(s1, rel);
1791 break;
1797 /* Bind symbols of executable: resolve undefined symbols from exported symbols
1798 in shared libraries and export non local defined symbols to shared libraries
1799 if -rdynamic switch was given on command line */
1800 static void bind_exe_dynsyms(TCCState *s1)
1802 const char *name;
1803 int sym_index, index;
1804 ElfW(Sym) *sym, *esym;
1805 int type;
1807 /* Resolve undefined symbols from dynamic symbols. When there is a match:
1808 - if STT_FUNC or STT_GNU_IFUNC symbol -> add it in PLT
1809 - if STT_OBJECT symbol -> add it in .bss section with suitable reloc */
1810 for_each_elem(symtab_section, 1, sym, ElfW(Sym)) {
1811 if (sym->st_shndx == SHN_UNDEF) {
1812 name = (char *) symtab_section->link->data + sym->st_name;
1813 sym_index = find_elf_sym(s1->dynsymtab_section, name);
1814 if (sym_index) {
1815 esym = &((ElfW(Sym) *)s1->dynsymtab_section->data)[sym_index];
1816 type = ELFW(ST_TYPE)(esym->st_info);
1817 if ((type == STT_FUNC) || (type == STT_GNU_IFUNC)) {
1818 /* Indirect functions shall have STT_FUNC type in executable
1819 * dynsym section. Indeed, a dlsym call following a lazy
1820 * resolution would pick the symbol value from the
1821 * executable dynsym entry which would contain the address
1822 * of the function wanted by the caller of dlsym instead of
1823 * the address of the function that would return that
1824 * address */
1825 put_got_entry(s1, R_JMP_SLOT, esym->st_size,
1826 ELFW(ST_INFO)(STB_GLOBAL,STT_FUNC),
1827 sym - (ElfW(Sym) *)symtab_section->data);
1828 } else if (type == STT_OBJECT) {
1829 unsigned long offset;
1830 ElfW(Sym) *dynsym;
1831 offset = bss_section->data_offset;
1832 /* XXX: which alignment ? */
1833 offset = (offset + 16 - 1) & -16;
1834 index = put_elf_sym(s1->dynsym, offset, esym->st_size,
1835 esym->st_info, 0, bss_section->sh_num,
1836 name);
1837 /* Ensure R_COPY works for weak symbol aliases */
1838 if (ELFW(ST_BIND)(esym->st_info) == STB_WEAK) {
1839 for_each_elem(s1->dynsymtab_section, 1, dynsym, ElfW(Sym)) {
1840 if ((dynsym->st_value == esym->st_value)
1841 && (ELFW(ST_BIND)(dynsym->st_info) == STB_GLOBAL)) {
1842 char *dynname = (char *) s1->dynsymtab_section->link->data
1843 + dynsym->st_name;
1844 put_elf_sym(s1->dynsym, offset, dynsym->st_size,
1845 dynsym->st_info, 0,
1846 bss_section->sh_num, dynname);
1847 break;
1851 put_elf_reloc(s1->dynsym, bss_section,
1852 offset, R_COPY, index);
1853 offset += esym->st_size;
1854 bss_section->data_offset = offset;
1856 } else {
1857 /* STB_WEAK undefined symbols are accepted */
1858 /* XXX: _fp_hw seems to be part of the ABI, so we ignore it */
1859 if (ELFW(ST_BIND)(sym->st_info) == STB_WEAK ||
1860 !strcmp(name, "_fp_hw")) {
1861 } else {
1862 tcc_error_noabort("undefined symbol '%s'", name);
1865 } else if (s1->rdynamic && ELFW(ST_BIND)(sym->st_info) != STB_LOCAL) {
1866 /* if -rdynamic option, then export all non local symbols */
1867 name = (char *) symtab_section->link->data + sym->st_name;
1868 put_elf_sym(s1->dynsym, sym->st_value, sym->st_size, sym->st_info,
1869 0, sym->st_shndx, name);
1874 /* Bind symbols of libraries: export non local symbols of executable that
1875 resolve undefined symbols of shared libraries */
1876 static void bind_libs_dynsyms(TCCState *s1)
1878 const char *name;
1879 int sym_index;
1880 ElfW(Sym) *sym, *esym;
1882 /* now look at unresolved dynamic symbols and export
1883 corresponding symbol */
1884 for_each_elem(s1->dynsymtab_section, 1, esym, ElfW(Sym)) {
1885 name = (char *) s1->dynsymtab_section->link->data + esym->st_name;
1886 sym_index = find_elf_sym(symtab_section, name);
1887 if (sym_index) {
1888 /* XXX: avoid adding a symbol if already present because of
1889 -rdynamic ? */
1890 sym = &((ElfW(Sym) *)symtab_section->data)[sym_index];
1891 if (sym->st_shndx != SHN_UNDEF)
1892 put_elf_sym(s1->dynsym, sym->st_value, sym->st_size,
1893 sym->st_info, 0, sym->st_shndx, name);
1894 } else if (esym->st_shndx == SHN_UNDEF) {
1895 /* weak symbols can stay undefined */
1896 if (ELFW(ST_BIND)(esym->st_info) != STB_WEAK)
1897 tcc_warning("undefined dynamic symbol '%s'", name);
1902 /* Export all non local symbols (for shared libraries) */
1903 static void export_global_syms(TCCState *s1)
1905 int nb_syms, dynindex, index;
1906 const char *name;
1907 ElfW(Sym) *sym;
1909 nb_syms = symtab_section->data_offset / sizeof(ElfW(Sym));
1910 s1->symtab_to_dynsym = tcc_mallocz(sizeof(int) * nb_syms);
1911 for_each_elem(symtab_section, 1, sym, ElfW(Sym)) {
1912 if (ELFW(ST_BIND)(sym->st_info) != STB_LOCAL) {
1913 name = (char *) symtab_section->link->data + sym->st_name;
1914 dynindex = put_elf_sym(s1->dynsym, sym->st_value, sym->st_size,
1915 sym->st_info, 0, sym->st_shndx, name);
1916 index = sym - (ElfW(Sym) *) symtab_section->data;
1917 s1->symtab_to_dynsym[index] = dynindex;
1922 /* relocate the PLT: compute addresses and offsets in the PLT now that final
1923 address for PLT and GOT are known (see fill_program_header) */
1924 ST_FUNC void relocate_plt(TCCState *s1)
1926 uint8_t *p, *p_end;
1928 if (!s1->plt)
1929 return;
1931 p = s1->plt->data;
1932 p_end = p + s1->plt->data_offset;
1933 if (p < p_end) {
1934 #if defined(TCC_TARGET_I386)
1935 write32le(p + 2, read32le(p + 2) + s1->got->sh_addr);
1936 write32le(p + 8, read32le(p + 8) + s1->got->sh_addr);
1937 p += 16;
1938 while (p < p_end) {
1939 write32le(p + 2, read32le(p + 2) + s1->got->sh_addr);
1940 p += 16;
1942 #elif defined(TCC_TARGET_X86_64)
1943 int x = s1->got->sh_addr - s1->plt->sh_addr - 6;
1944 write32le(p + 2, read32le(p + 2) + x);
1945 write32le(p + 8, read32le(p + 8) + x - 6);
1946 p += 16;
1947 while (p < p_end) {
1948 write32le(p + 2, read32le(p + 2) + x + s1->plt->data - p);
1949 p += 16;
1951 #elif defined(TCC_TARGET_ARM)
1952 int x;
1953 x=s1->got->sh_addr - s1->plt->sh_addr - 12;
1954 p += 16;
1955 while (p < p_end) {
1956 if (read32le(p) == 0x46c04778) /* PLT Thumb stub present */
1957 p += 4;
1958 write32le(p + 12, x + read32le(p + 12) + s1->plt->data - p);
1959 p += 16;
1961 #elif defined(TCC_TARGET_ARM64)
1962 uint64_t plt = s1->plt->sh_addr;
1963 uint64_t got = s1->got->sh_addr;
1964 uint64_t off = (got >> 12) - (plt >> 12);
1965 if ((off + ((uint32_t)1 << 20)) >> 21)
1966 tcc_error("Failed relocating PLT (off=0x%lx, got=0x%lx, plt=0x%lx)", off, got, plt);
1967 write32le(p, 0xa9bf7bf0); // stp x16,x30,[sp,#-16]!
1968 write32le(p + 4, (0x90000010 | // adrp x16,...
1969 (off & 0x1ffffc) << 3 | (off & 3) << 29));
1970 write32le(p + 8, (0xf9400211 | // ldr x17,[x16,#...]
1971 (got & 0xff8) << 7));
1972 write32le(p + 12, (0x91000210 | // add x16,x16,#...
1973 (got & 0xfff) << 10));
1974 write32le(p + 16, 0xd61f0220); // br x17
1975 write32le(p + 20, 0xd503201f); // nop
1976 write32le(p + 24, 0xd503201f); // nop
1977 write32le(p + 28, 0xd503201f); // nop
1978 p += 32;
1979 while (p < p_end) {
1980 uint64_t pc = plt + (p - s1->plt->data);
1981 uint64_t addr = got + read64le(p);
1982 uint64_t off = (addr >> 12) - (pc >> 12);
1983 if ((off + ((uint32_t)1 << 20)) >> 21)
1984 tcc_error("Failed relocating PLT (off=0x%lx, addr=0x%lx, pc=0x%lx)", off, addr, pc);
1985 write32le(p, (0x90000010 | // adrp x16,...
1986 (off & 0x1ffffc) << 3 | (off & 3) << 29));
1987 write32le(p + 4, (0xf9400211 | // ldr x17,[x16,#...]
1988 (addr & 0xff8) << 7));
1989 write32le(p + 8, (0x91000210 | // add x16,x16,#...
1990 (addr & 0xfff) << 10));
1991 write32le(p + 12, 0xd61f0220); // br x17
1992 p += 16;
1994 #elif defined(TCC_TARGET_C67)
1995 /* XXX: TODO */
1996 #else
1997 #error unsupported CPU
1998 #endif
2002 /* Allocate strings for section names and decide if an unallocated section
2003 should be output.
2005 NOTE: the strsec section comes last, so its size is also correct ! */
2006 static void alloc_sec_names(TCCState *s1, int file_type, Section *strsec)
2008 int i;
2009 Section *s;
2011 /* Allocate strings for section names */
2012 for(i = 1; i < s1->nb_sections; i++) {
2013 s = s1->sections[i];
2014 s->sh_name = put_elf_str(strsec, s->name);
2015 /* when generating a DLL, we include relocations but we may
2016 patch them */
2017 if (file_type == TCC_OUTPUT_DLL &&
2018 s->sh_type == SHT_RELX &&
2019 !(s->sh_flags & SHF_ALLOC)) {
2020 /* gr: avoid bogus relocs for empty (debug) sections */
2021 if (s1->sections[s->sh_info]->sh_flags & SHF_ALLOC)
2022 prepare_dynamic_rel(s1, s);
2023 else if (s1->do_debug)
2024 s->sh_size = s->data_offset;
2025 } else if (s1->do_debug ||
2026 file_type == TCC_OUTPUT_OBJ ||
2027 file_type == TCC_OUTPUT_EXE ||
2028 (s->sh_flags & SHF_ALLOC) ||
2029 i == (s1->nb_sections - 1)) {
2030 /* we output all sections if debug or object file */
2031 s->sh_size = s->data_offset;
2036 /* Info to be copied in dynamic section */
2037 struct dyn_inf {
2038 Section *dynamic;
2039 Section *dynstr;
2040 unsigned long dyn_rel_off;
2041 addr_t rel_addr;
2042 addr_t rel_size;
2043 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
2044 addr_t bss_addr;
2045 addr_t bss_size;
2046 #endif
2047 };
2049 /* Assign sections to segments and decide how are sections laid out when loaded
2050 in memory. This function also fills corresponding program headers. */
2051 static int layout_sections(TCCState *s1, ElfW(Phdr) *phdr, int phnum,
2052 Section *interp, Section* strsec,
2053 struct dyn_inf *dyninf, int *sec_order)
2055 int i, j, k, file_type, sh_order_index, file_offset;
2056 unsigned long s_align;
2057 long long tmp;
2058 addr_t addr;
2059 ElfW(Phdr) *ph;
2060 Section *s;
2062 file_type = s1->output_type;
2063 sh_order_index = 1;
2064 file_offset = 0;
2065 if (s1->output_format == TCC_OUTPUT_FORMAT_ELF)
2066 file_offset = sizeof(ElfW(Ehdr)) + phnum * sizeof(ElfW(Phdr));
2067 s_align = ELF_PAGE_SIZE;
2068 if (s1->section_align)
2069 s_align = s1->section_align;
2071 if (phnum > 0) {
2072 if (s1->has_text_addr) {
2073 int a_offset, p_offset;
2074 addr = s1->text_addr;
2075 /* we ensure that (addr % ELF_PAGE_SIZE) == file_offset %
2076 ELF_PAGE_SIZE */
2077 a_offset = (int) (addr & (s_align - 1));
2078 p_offset = file_offset & (s_align - 1);
2079 if (a_offset < p_offset)
2080 a_offset += s_align;
2081 file_offset += (a_offset - p_offset);
2082 } else {
2083 if (file_type == TCC_OUTPUT_DLL)
2084 addr = 0;
2085 else
2086 addr = ELF_START_ADDR;
2087 /* compute address after headers */
2088 addr += (file_offset & (s_align - 1));
2091 ph = &phdr[0];
2092 /* Leave one program headers for the program interpreter and one for
2093 the program header table itself if needed. These are done later as
2094 they require section layout to be done first. */
2095 if (interp)
2096 ph += 1 + HAVE_PHDR;
2098 /* dynamic relocation table information, for .dynamic section */
2099 dyninf->rel_addr = dyninf->rel_size = 0;
2100 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
2101 dyninf->bss_addr = dyninf->bss_size = 0;
2102 #endif
2104 for(j = 0; j < 2; j++) {
2105 ph->p_type = PT_LOAD;
2106 if (j == 0)
2107 ph->p_flags = PF_R | PF_X;
2108 else
2109 ph->p_flags = PF_R | PF_W;
2110 ph->p_align = s_align;
2112 /* Decide the layout of sections loaded in memory. This must
2113 be done before program headers are filled since they contain
2114 info about the layout. We do the following ordering: interp,
2115 symbol tables, relocations, progbits, nobits */
2116 /* XXX: do faster and simpler sorting */
2117 for(k = 0; k < 5; k++) {
2118 for(i = 1; i < s1->nb_sections; i++) {
2119 s = s1->sections[i];
2120 /* compute if section should be included */
2121 if (j == 0) {
2122 if ((s->sh_flags & (SHF_ALLOC | SHF_WRITE)) !=
2123 SHF_ALLOC)
2124 continue;
2125 } else {
2126 if ((s->sh_flags & (SHF_ALLOC | SHF_WRITE)) !=
2127 (SHF_ALLOC | SHF_WRITE))
2128 continue;
2130 if (s == interp) {
2131 if (k != 0)
2132 continue;
2133 } else if (s->sh_type == SHT_DYNSYM ||
2134 s->sh_type == SHT_STRTAB ||
2135 s->sh_type == SHT_HASH) {
2136 if (k != 1)
2137 continue;
2138 } else if (s->sh_type == SHT_RELX) {
2139 if (k != 2)
2140 continue;
2141 } else if (s->sh_type == SHT_NOBITS) {
2142 if (k != 4)
2143 continue;
2144 } else {
2145 if (k != 3)
2146 continue;
2148 sec_order[sh_order_index++] = i;
2150 /* section matches: we align it and add its size */
2151 tmp = addr;
2152 addr = (addr + s->sh_addralign - 1) &
2153 ~(s->sh_addralign - 1);
2154 file_offset += (int) ( addr - tmp );
2155 s->sh_offset = file_offset;
2156 s->sh_addr = addr;
2158 /* update program header infos */
2159 if (ph->p_offset == 0) {
2160 ph->p_offset = file_offset;
2161 ph->p_vaddr = addr;
2162 ph->p_paddr = ph->p_vaddr;
2164 /* update dynamic relocation infos */
2165 if (s->sh_type == SHT_RELX) {
2166 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
2167 if (!strcmp(strsec->data + s->sh_name, ".rel.got")) {
2168 dyninf->rel_addr = addr;
2169 dyninf->rel_size += s->sh_size; /* XXX only first rel. */
2171 if (!strcmp(strsec->data + s->sh_name, ".rel.bss")) {
2172 dyninf->bss_addr = addr;
2173 dyninf->bss_size = s->sh_size; /* XXX only first rel. */
2175 #else
2176 if (dyninf->rel_size == 0)
2177 dyninf->rel_addr = addr;
2178 dyninf->rel_size += s->sh_size;
2179 #endif
2181 addr += s->sh_size;
2182 if (s->sh_type != SHT_NOBITS)
2183 file_offset += s->sh_size;
2186 if (j == 0) {
2187 /* Make the first PT_LOAD segment include the program
2188 headers itself (and the ELF header as well), it'll
2189 come out with same memory use but will make various
2190 tools like binutils strip work better. */
2191 ph->p_offset &= ~(ph->p_align - 1);
2192 ph->p_vaddr &= ~(ph->p_align - 1);
2193 ph->p_paddr &= ~(ph->p_align - 1);
2195 ph->p_filesz = file_offset - ph->p_offset;
2196 ph->p_memsz = addr - ph->p_vaddr;
2197 ph++;
2198 if (j == 0) {
2199 if (s1->output_format == TCC_OUTPUT_FORMAT_ELF) {
2200 /* if in the middle of a page, we duplicate the page in
2201 memory so that one copy is RX and the other is RW */
2202 if ((addr & (s_align - 1)) != 0)
2203 addr += s_align;
2204 } else {
2205 addr = (addr + s_align - 1) & ~(s_align - 1);
2206 file_offset = (file_offset + s_align - 1) & ~(s_align - 1);
2212 /* all other sections come after */
2213 for(i = 1; i < s1->nb_sections; i++) {
2214 s = s1->sections[i];
2215 if (phnum > 0 && (s->sh_flags & SHF_ALLOC))
2216 continue;
2217 sec_order[sh_order_index++] = i;
2219 file_offset = (file_offset + s->sh_addralign - 1) &
2220 ~(s->sh_addralign - 1);
2221 s->sh_offset = file_offset;
2222 if (s->sh_type != SHT_NOBITS)
2223 file_offset += s->sh_size;
2226 return file_offset;
2229 static void fill_unloadable_phdr(ElfW(Phdr) *phdr, int phnum, Section *interp,
2230 Section *dynamic)
2232 ElfW(Phdr) *ph;
2234 /* if interpreter, then add corresponding program header */
2235 if (interp) {
2236 ph = &phdr[0];
2238 if (HAVE_PHDR)
2240 int len = phnum * sizeof(ElfW(Phdr));
2242 ph->p_type = PT_PHDR;
2243 ph->p_offset = sizeof(ElfW(Ehdr));
2244 ph->p_vaddr = interp->sh_addr - len;
2245 ph->p_paddr = ph->p_vaddr;
2246 ph->p_filesz = ph->p_memsz = len;
2247 ph->p_flags = PF_R | PF_X;
2248 ph->p_align = 4; /* interp->sh_addralign; */
2249 ph++;
2252 ph->p_type = PT_INTERP;
2253 ph->p_offset = interp->sh_offset;
2254 ph->p_vaddr = interp->sh_addr;
2255 ph->p_paddr = ph->p_vaddr;
2256 ph->p_filesz = interp->sh_size;
2257 ph->p_memsz = interp->sh_size;
2258 ph->p_flags = PF_R;
2259 ph->p_align = interp->sh_addralign;
2262 /* if dynamic section, then add corresponding program header */
2263 if (dynamic) {
2264 ph = &phdr[phnum - 1];
2266 ph->p_type = PT_DYNAMIC;
2267 ph->p_offset = dynamic->sh_offset;
2268 ph->p_vaddr = dynamic->sh_addr;
2269 ph->p_paddr = ph->p_vaddr;
2270 ph->p_filesz = dynamic->sh_size;
2271 ph->p_memsz = dynamic->sh_size;
2272 ph->p_flags = PF_R | PF_W;
2273 ph->p_align = dynamic->sh_addralign;
2277 /* Fill the dynamic section with tags describing the address and size of
2278 sections */
2279 static void fill_dynamic(TCCState *s1, struct dyn_inf *dyninf)
2281 Section *dynamic;
2283 dynamic = dyninf->dynamic;
2285 /* put dynamic section entries */
2286 dynamic->data_offset = dyninf->dyn_rel_off;
2287 put_dt(dynamic, DT_HASH, s1->dynsym->hash->sh_addr);
2288 put_dt(dynamic, DT_STRTAB, dyninf->dynstr->sh_addr);
2289 put_dt(dynamic, DT_SYMTAB, s1->dynsym->sh_addr);
2290 put_dt(dynamic, DT_STRSZ, dyninf->dynstr->data_offset);
2291 put_dt(dynamic, DT_SYMENT, sizeof(ElfW(Sym)));
2292 #if defined(TCC_TARGET_ARM64) || defined(TCC_TARGET_X86_64)
2293 put_dt(dynamic, DT_RELA, dyninf->rel_addr);
2294 put_dt(dynamic, DT_RELASZ, dyninf->rel_size);
2295 put_dt(dynamic, DT_RELAENT, sizeof(ElfW_Rel));
2296 #else
2297 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
2298 put_dt(dynamic, DT_PLTGOT, s1->got->sh_addr);
2299 put_dt(dynamic, DT_PLTRELSZ, dyninf->rel_size);
2300 put_dt(dynamic, DT_JMPREL, dyninf->rel_addr);
2301 put_dt(dynamic, DT_PLTREL, DT_REL);
2302 put_dt(dynamic, DT_REL, dyninf->bss_addr);
2303 put_dt(dynamic, DT_RELSZ, dyninf->bss_size);
2304 #else
2305 put_dt(dynamic, DT_REL, dyninf->rel_addr);
2306 put_dt(dynamic, DT_RELSZ, dyninf->rel_size);
2307 put_dt(dynamic, DT_RELENT, sizeof(ElfW_Rel));
2308 #endif
2309 #endif
2310 if (s1->do_debug)
2311 put_dt(dynamic, DT_DEBUG, 0);
2312 put_dt(dynamic, DT_NULL, 0);
2315 /* Relocate remaining sections and symbols (that is those not related to
2316 dynamic linking) */
2317 static int final_sections_reloc(TCCState *s1)
2319 int i;
2320 Section *s;
2322 relocate_syms(s1, 0);
2324 if (s1->nb_errors != 0)
2325 return -1;
2327 /* relocate sections */
2328 /* XXX: ignore sections with allocated relocations ? */
2329 for(i = 1; i < s1->nb_sections; i++) {
2330 s = s1->sections[i];
2331 #ifdef TCC_TARGET_I386
2332 if (s->reloc && s != s1->got && (s->sh_flags & SHF_ALLOC)) //gr
2333 /* On X86 gdb 7.3 works in any case but gdb 6.6 will crash if SHF_ALLOC
2334 checking is removed */
2335 #else
2336 if (s->reloc && s != s1->got)
2337 /* On X86_64 gdb 7.3 will crash if SHF_ALLOC checking is present */
2338 #endif
2339 relocate_section(s1, s);
2342 /* relocate relocation entries if the relocation tables are
2343 allocated in the executable */
2344 for(i = 1; i < s1->nb_sections; i++) {
2345 s = s1->sections[i];
2346 if ((s->sh_flags & SHF_ALLOC) &&
2347 s->sh_type == SHT_RELX) {
2348 relocate_rel(s1, s);
2351 return 0;
2354 /* Create an ELF file on disk.
2355 This function handle ELF specific layout requirements */
2356 static void tcc_output_elf(TCCState *s1, FILE *f, int phnum, ElfW(Phdr) *phdr,
2357 int file_offset, int *sec_order)
2359 int i, shnum, offset, size, file_type;
2360 Section *s;
2361 ElfW(Ehdr) ehdr;
2362 ElfW(Shdr) shdr, *sh;
2364 file_type = s1->output_type;
2365 shnum = s1->nb_sections;
2367 memset(&ehdr, 0, sizeof(ehdr));
2369 if (phnum > 0) {
2370 ehdr.e_phentsize = sizeof(ElfW(Phdr));
2371 ehdr.e_phnum = phnum;
2372 ehdr.e_phoff = sizeof(ElfW(Ehdr));
2375 /* align to 4 */
2376 file_offset = (file_offset + 3) & -4;
2378 /* fill header */
2379 ehdr.e_ident[0] = ELFMAG0;
2380 ehdr.e_ident[1] = ELFMAG1;
2381 ehdr.e_ident[2] = ELFMAG2;
2382 ehdr.e_ident[3] = ELFMAG3;
2383 ehdr.e_ident[4] = ELFCLASSW;
2384 ehdr.e_ident[5] = ELFDATA2LSB;
2385 ehdr.e_ident[6] = EV_CURRENT;
2386 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
2387 ehdr.e_ident[EI_OSABI] = ELFOSABI_FREEBSD;
2388 #endif
2389 #ifdef TCC_TARGET_ARM
2390 #ifdef TCC_ARM_EABI
2391 ehdr.e_ident[EI_OSABI] = 0;
2392 ehdr.e_flags = EF_ARM_EABI_VER4;
2393 if (file_type == TCC_OUTPUT_EXE || file_type == TCC_OUTPUT_DLL)
2394 ehdr.e_flags |= EF_ARM_HASENTRY;
2395 if (s1->float_abi == ARM_HARD_FLOAT)
2396 ehdr.e_flags |= EF_ARM_VFP_FLOAT;
2397 else
2398 ehdr.e_flags |= EF_ARM_SOFT_FLOAT;
2399 #else
2400 ehdr.e_ident[EI_OSABI] = ELFOSABI_ARM;
2401 #endif
2402 #endif
2403 switch(file_type) {
2404 default:
2405 case TCC_OUTPUT_EXE:
2406 ehdr.e_type = ET_EXEC;
2407 ehdr.e_entry = get_elf_sym_addr(s1, "_start", 1);
2408 break;
2409 case TCC_OUTPUT_DLL:
2410 ehdr.e_type = ET_DYN;
2411 ehdr.e_entry = text_section->sh_addr; /* XXX: is it correct ? */
2412 break;
2413 case TCC_OUTPUT_OBJ:
2414 ehdr.e_type = ET_REL;
2415 break;
2417 ehdr.e_machine = EM_TCC_TARGET;
2418 ehdr.e_version = EV_CURRENT;
2419 ehdr.e_shoff = file_offset;
2420 ehdr.e_ehsize = sizeof(ElfW(Ehdr));
2421 ehdr.e_shentsize = sizeof(ElfW(Shdr));
2422 ehdr.e_shnum = shnum;
2423 ehdr.e_shstrndx = shnum - 1;
2425 fwrite(&ehdr, 1, sizeof(ElfW(Ehdr)), f);
2426 fwrite(phdr, 1, phnum * sizeof(ElfW(Phdr)), f);
2427 offset = sizeof(ElfW(Ehdr)) + phnum * sizeof(ElfW(Phdr));
2429 sort_syms(s1, symtab_section);
2430 for(i = 1; i < s1->nb_sections; i++) {
2431 s = s1->sections[sec_order[i]];
2432 if (s->sh_type != SHT_NOBITS) {
2433 if (s->sh_type == SHT_DYNSYM)
2434 patch_dynsym_undef(s1, s);
2435 while (offset < s->sh_offset) {
2436 fputc(0, f);
2437 offset++;
2439 size = s->sh_size;
2440 if (size)
2441 fwrite(s->data, 1, size, f);
2442 offset += size;
2446 /* output section headers */
2447 while (offset < ehdr.e_shoff) {
2448 fputc(0, f);
2449 offset++;
2452 for(i = 0; i < s1->nb_sections; i++) {
2453 sh = &shdr;
2454 memset(sh, 0, sizeof(ElfW(Shdr)));
2455 s = s1->sections[i];
2456 if (s) {
2457 sh->sh_name = s->sh_name;
2458 sh->sh_type = s->sh_type;
2459 sh->sh_flags = s->sh_flags;
2460 sh->sh_entsize = s->sh_entsize;
2461 sh->sh_info = s->sh_info;
2462 if (s->link)
2463 sh->sh_link = s->link->sh_num;
2464 sh->sh_addralign = s->sh_addralign;
2465 sh->sh_addr = s->sh_addr;
2466 sh->sh_offset = s->sh_offset;
2467 sh->sh_size = s->sh_size;
2469 fwrite(sh, 1, sizeof(ElfW(Shdr)), f);
2473 /* Write an elf, coff or "binary" file */
2474 static int tcc_write_elf_file(TCCState *s1, const char *filename, int phnum,
2475 ElfW(Phdr) *phdr, int file_offset, int *sec_order)
2477 int fd, mode, file_type;
2478 FILE *f;
2480 file_type = s1->output_type;
2481 if (file_type == TCC_OUTPUT_OBJ)
2482 mode = 0666;
2483 else
2484 mode = 0777;
2485 unlink(filename);
2486 fd = open(filename, O_WRONLY | O_CREAT | O_TRUNC | O_BINARY, mode);
2487 if (fd < 0) {
2488 tcc_error_noabort("could not write '%s'", filename);
2489 return -1;
2491 f = fdopen(fd, "wb");
2492 if (s1->verbose)
2493 printf("<- %s\n", filename);
2495 #ifdef TCC_TARGET_COFF
2496 if (s1->output_format == TCC_OUTPUT_FORMAT_COFF)
2497 tcc_output_coff(s1, f);
2498 else
2499 #endif
2500 if (s1->output_format == TCC_OUTPUT_FORMAT_ELF)
2501 tcc_output_elf(s1, f, phnum, phdr, file_offset, sec_order);
2502 else
2503 tcc_output_binary(s1, f, sec_order);
2504 fclose(f);
2506 return 0;
2509 /* Output an elf, coff or binary file */
2510 /* XXX: suppress unneeded sections */
2511 static int elf_output_file(TCCState *s1, const char *filename)
2513 int i, ret, phnum, shnum, file_type, file_offset, *sec_order;
2514 struct dyn_inf dyninf;
2515 ElfW(Phdr) *phdr;
2516 ElfW(Sym) *sym;
2517 Section *strsec, *interp, *dynamic, *dynstr;
2519 file_type = s1->output_type;
2520 s1->nb_errors = 0;
2522 /* if linking, also link in runtime libraries (libc, libgcc, etc.) */
2523 if (file_type != TCC_OUTPUT_OBJ) {
2524 tcc_add_runtime(s1);
2527 phdr = NULL;
2528 sec_order = NULL;
2529 interp = dynamic = dynstr = NULL; /* avoid warning */
2530 dyninf.dyn_rel_off = 0; /* avoid warning */
2532 if (file_type != TCC_OUTPUT_OBJ) {
2533 relocate_common_syms();
2535 tcc_add_linker_symbols(s1);
2537 if (!s1->static_link) {
2538 if (file_type == TCC_OUTPUT_EXE) {
2539 char *ptr;
2540 /* allow override the dynamic loader */
2541 const char *elfint = getenv("LD_SO");
2542 if (elfint == NULL)
2543 elfint = DEFAULT_ELFINTERP(s1);
2544 /* add interpreter section only if executable */
2545 interp = new_section(s1, ".interp", SHT_PROGBITS, SHF_ALLOC);
2546 interp->sh_addralign = 1;
2547 ptr = section_ptr_add(interp, 1 + strlen(elfint));
2548 strcpy(ptr, elfint);
2551 /* add dynamic symbol table */
2552 s1->dynsym = new_symtab(s1, ".dynsym", SHT_DYNSYM, SHF_ALLOC,
2553 ".dynstr",
2554 ".hash", SHF_ALLOC);
2555 dynstr = s1->dynsym->link;
2557 /* add dynamic section */
2558 dynamic = new_section(s1, ".dynamic", SHT_DYNAMIC,
2559 SHF_ALLOC | SHF_WRITE);
2560 dynamic->link = dynstr;
2561 dynamic->sh_entsize = sizeof(ElfW(Dyn));
2563 build_got(s1);
2565 if (file_type == TCC_OUTPUT_EXE) {
2566 bind_exe_dynsyms(s1);
2568 if (s1->nb_errors) {
2569 ret = -1;
2570 goto the_end;
2573 bind_libs_dynsyms(s1);
2574 } else /* shared library case: simply export all global symbols */
2575 export_global_syms(s1);
2577 build_got_entries(s1);
2579 /* add a list of needed dlls */
2580 for(i = 0; i < s1->nb_loaded_dlls; i++) {
2581 DLLReference *dllref = s1->loaded_dlls[i];
2582 if (dllref->level == 0)
2583 put_dt(dynamic, DT_NEEDED, put_elf_str(dynstr, dllref->name));
2586 if (s1->rpath)
2587 put_dt(dynamic, DT_RPATH, put_elf_str(dynstr, s1->rpath));
2589 /* XXX: currently, since we do not handle PIC code, we
2590 must relocate the readonly segments */
2591 if (file_type == TCC_OUTPUT_DLL) {
2592 if (s1->soname)
2593 put_dt(dynamic, DT_SONAME, put_elf_str(dynstr, s1->soname));
2594 put_dt(dynamic, DT_TEXTREL, 0);
2597 if (s1->symbolic)
2598 put_dt(dynamic, DT_SYMBOLIC, 0);
2600 /* add necessary space for other entries */
2601 dyninf.dyn_rel_off = dynamic->data_offset;
2602 dynamic->data_offset += sizeof(ElfW(Dyn)) * EXTRA_RELITEMS;
2603 } else {
2604 /* still need to build got entries in case of static link */
2605 build_got_entries(s1);
2609 /* we add a section for symbols */
2610 strsec = new_section(s1, ".shstrtab", SHT_STRTAB, 0);
2611 put_elf_str(strsec, "");
2613 /* compute number of sections */
2614 shnum = s1->nb_sections;
2616 /* this array is used to reorder sections in the output file */
2617 sec_order = tcc_malloc(sizeof(int) * shnum);
2618 sec_order[0] = 0;
2620 /* compute number of program headers */
2621 switch(file_type) {
2622 default:
2623 case TCC_OUTPUT_OBJ:
2624 phnum = 0;
2625 break;
2626 case TCC_OUTPUT_EXE:
2627 if (!s1->static_link)
2628 phnum = 4 + HAVE_PHDR;
2629 else
2630 phnum = 2;
2631 break;
2632 case TCC_OUTPUT_DLL:
2633 phnum = 3;
2634 break;
2637 /* Allocate strings for section names */
2638 alloc_sec_names(s1, file_type, strsec);
2640 /* allocate program segment headers */
2641 phdr = tcc_mallocz(phnum * sizeof(ElfW(Phdr)));
2643 /* compute section to program header mapping */
2644 file_offset = layout_sections(s1, phdr, phnum, interp, strsec, &dyninf,
2645 sec_order);
2647 /* Fill remaining program header and finalize relocation related to dynamic
2648 linking. */
2649 if (phnum > 0) {
2650 fill_unloadable_phdr(phdr, phnum, interp, dynamic);
2651 if (dynamic) {
2652 dyninf.dynamic = dynamic;
2653 dyninf.dynstr = dynstr;
2655 fill_dynamic(s1, &dyninf);
2657 /* put in GOT the dynamic section address and relocate PLT */
2658 write32le(s1->got->data, dynamic->sh_addr);
2659 if (file_type == TCC_OUTPUT_EXE
2660 #if defined(TCC_OUTPUT_DLL_WITH_PLT)
2661 || file_type == TCC_OUTPUT_DLL
2662 #endif
2664 relocate_plt(s1);
2666 /* relocate symbols in .dynsym now that final addresses are known */
2667 for_each_elem(s1->dynsym, 1, sym, ElfW(Sym)) {
2668 if (sym->st_shndx == SHN_UNDEF) {
2669 /* relocate to PLT if symbol corresponds to a PLT entry,
2670 but not if it's a weak symbol */
2671 if (ELFW(ST_BIND)(sym->st_info) == STB_WEAK)
2672 sym->st_value = 0;
2673 else if (sym->st_value)
2674 sym->st_value += s1->plt->sh_addr;
2675 } else if (sym->st_shndx < SHN_LORESERVE) {
2676 /* do symbol relocation */
2677 sym->st_value += s1->sections[sym->st_shndx]->sh_addr;
2683 /* if building executable or DLL, then relocate each section
2684 except the GOT which is already relocated */
2685 if (file_type != TCC_OUTPUT_OBJ) {
2686 ret = final_sections_reloc(s1);
2687 if (ret)
2688 goto the_end;
2691 /* Perform relocation to GOT or PLT entries */
2692 if (file_type == TCC_OUTPUT_EXE && s1->static_link)
2693 fill_got(s1);
2695 /* Create the ELF file with name 'filename' */
2696 ret = tcc_write_elf_file(s1, filename, phnum, phdr, file_offset, sec_order);
2697 if (s1->do_strip) {
2698 int rc;
2699 const char *strip_cmd = "sstrip "; // super strip utility from ELFkickers
2700 const char *null_dev = " 2> /dev/null";
2701 char buf[1050];
2702 snprintf(buf, sizeof(buf), "%s%s%s", strip_cmd, filename, null_dev);
2703 rc = system(buf);
2704 if (rc)
2705 system(buf+1); // call a strip utility from binutils
2707 the_end:
2708 tcc_free(s1->symtab_to_dynsym);
2709 tcc_free(sec_order);
2710 tcc_free(phdr);
2711 tcc_free(s1->sym_attrs);
2712 s1->sym_attrs = NULL;
2713 return ret;
2716 LIBTCCAPI int tcc_output_file(TCCState *s, const char *filename)
2718 int ret;
2719 #ifdef TCC_TARGET_PE
2720 if (s->output_type != TCC_OUTPUT_OBJ) {
2721 ret = pe_output_file(s, filename);
2722 } else
2723 #endif
2724 ret = elf_output_file(s, filename);
2725 return ret;
2728 static void *load_data(int fd, unsigned long file_offset, unsigned long size)
2730 void *data;
2732 data = tcc_malloc(size);
2733 lseek(fd, file_offset, SEEK_SET);
2734 read(fd, data, size);
2735 return data;
2738 typedef struct SectionMergeInfo {
2739 Section *s; /* corresponding existing section */
2740 unsigned long offset; /* offset of the new section in the existing section */
2741 uint8_t new_section; /* true if section 's' was added */
2742 uint8_t link_once; /* true if link once section */
2743 } SectionMergeInfo;
2745 /* load an object file and merge it with current files */
2746 /* XXX: handle correctly stab (debug) info */
2747 ST_FUNC int tcc_load_object_file(TCCState *s1,
2748 int fd, unsigned long file_offset)
2750 ElfW(Ehdr) ehdr;
2751 ElfW(Shdr) *shdr, *sh;
2752 int size, i, j, offset, offseti, nb_syms, sym_index, ret;
2753 unsigned char *strsec, *strtab;
2754 int *old_to_new_syms;
2755 char *sh_name, *name;
2756 SectionMergeInfo *sm_table, *sm;
2757 ElfW(Sym) *sym, *symtab;
2758 ElfW_Rel *rel;
2759 Section *s;
2761 int stab_index;
2762 int stabstr_index;
2764 stab_index = stabstr_index = 0;
2766 if (read(fd, &ehdr, sizeof(ehdr)) != sizeof(ehdr))
2767 goto fail1;
2768 if (ehdr.e_ident[0] != ELFMAG0 ||
2769 ehdr.e_ident[1] != ELFMAG1 ||
2770 ehdr.e_ident[2] != ELFMAG2 ||
2771 ehdr.e_ident[3] != ELFMAG3)
2772 goto fail1;
2773 /* test if object file */
2774 if (ehdr.e_type != ET_REL)
2775 goto fail1;
2776 /* test CPU specific stuff */
2777 if (ehdr.e_ident[5] != ELFDATA2LSB ||
2778 ehdr.e_machine != EM_TCC_TARGET) {
2779 fail1:
2780 tcc_error_noabort("invalid object file");
2781 return -1;
2783 /* read sections */
2784 shdr = load_data(fd, file_offset + ehdr.e_shoff,
2785 sizeof(ElfW(Shdr)) * ehdr.e_shnum);
2786 sm_table = tcc_mallocz(sizeof(SectionMergeInfo) * ehdr.e_shnum);
2788 /* load section names */
2789 sh = &shdr[ehdr.e_shstrndx];
2790 strsec = load_data(fd, file_offset + sh->sh_offset, sh->sh_size);
2792 /* load symtab and strtab */
2793 old_to_new_syms = NULL;
2794 symtab = NULL;
2795 strtab = NULL;
2796 nb_syms = 0;
2797 for(i = 1; i < ehdr.e_shnum; i++) {
2798 sh = &shdr[i];
2799 if (sh->sh_type == SHT_SYMTAB) {
2800 if (symtab) {
2801 tcc_error_noabort("object must contain only one symtab");
2802 fail:
2803 ret = -1;
2804 goto the_end;
2806 nb_syms = sh->sh_size / sizeof(ElfW(Sym));
2807 symtab = load_data(fd, file_offset + sh->sh_offset, sh->sh_size);
2808 sm_table[i].s = symtab_section;
2810 /* now load strtab */
2811 sh = &shdr[sh->sh_link];
2812 strtab = load_data(fd, file_offset + sh->sh_offset, sh->sh_size);
2816 /* now examine each section and try to merge its content with the
2817 ones in memory */
2818 for(i = 1; i < ehdr.e_shnum; i++) {
2819 /* no need to examine section name strtab */
2820 if (i == ehdr.e_shstrndx)
2821 continue;
2822 sh = &shdr[i];
2823 sh_name = (char *) strsec + sh->sh_name;
2824 /* ignore sections types we do not handle */
2825 if (sh->sh_type != SHT_PROGBITS &&
2826 sh->sh_type != SHT_RELX &&
2827 #ifdef TCC_ARM_EABI
2828 sh->sh_type != SHT_ARM_EXIDX &&
2829 #endif
2830 sh->sh_type != SHT_NOBITS &&
2831 sh->sh_type != SHT_PREINIT_ARRAY &&
2832 sh->sh_type != SHT_INIT_ARRAY &&
2833 sh->sh_type != SHT_FINI_ARRAY &&
2834 strcmp(sh_name, ".stabstr")
2836 continue;
2837 if (sh->sh_addralign < 1)
2838 sh->sh_addralign = 1;
2839 /* find corresponding section, if any */
2840 for(j = 1; j < s1->nb_sections;j++) {
2841 s = s1->sections[j];
2842 if (!strcmp(s->name, sh_name)) {
2843 if (!strncmp(sh_name, ".gnu.linkonce",
2844 sizeof(".gnu.linkonce") - 1)) {
2845 /* if a 'linkonce' section is already present, we
2846 do not add it again. It is a little tricky as
2847 symbols can still be defined in
2848 it. */
2849 sm_table[i].link_once = 1;
2850 goto next;
2851 } else {
2852 goto found;
2856 /* not found: create new section */
2857 s = new_section(s1, sh_name, sh->sh_type, sh->sh_flags);
2858 /* take as much info as possible from the section. sh_link and
2859 sh_info will be updated later */
2860 s->sh_addralign = sh->sh_addralign;
2861 s->sh_entsize = sh->sh_entsize;
2862 sm_table[i].new_section = 1;
2863 found:
2864 if (sh->sh_type != s->sh_type) {
2865 tcc_error_noabort("invalid section type");
2866 goto fail;
2869 /* align start of section */
2870 offset = s->data_offset;
2872 if (0 == strcmp(sh_name, ".stab")) {
2873 stab_index = i;
2874 goto no_align;
2876 if (0 == strcmp(sh_name, ".stabstr")) {
2877 stabstr_index = i;
2878 goto no_align;
2881 size = sh->sh_addralign - 1;
2882 offset = (offset + size) & ~size;
2883 if (sh->sh_addralign > s->sh_addralign)
2884 s->sh_addralign = sh->sh_addralign;
2885 s->data_offset = offset;
2886 no_align:
2887 sm_table[i].offset = offset;
2888 sm_table[i].s = s;
2889 /* concatenate sections */
2890 size = sh->sh_size;
2891 if (sh->sh_type != SHT_NOBITS) {
2892 unsigned char *ptr;
2893 lseek(fd, file_offset + sh->sh_offset, SEEK_SET);
2894 ptr = section_ptr_add(s, size);
2895 read(fd, ptr, size);
2896 } else {
2897 s->data_offset += size;
2899 next: ;
2902 /* gr relocate stab strings */
2903 if (stab_index && stabstr_index) {
2904 Stab_Sym *a, *b;
2905 unsigned o;
2906 s = sm_table[stab_index].s;
2907 a = (Stab_Sym *)(s->data + sm_table[stab_index].offset);
2908 b = (Stab_Sym *)(s->data + s->data_offset);
2909 o = sm_table[stabstr_index].offset;
2910 while (a < b)
2911 a->n_strx += o, a++;
2914 /* second short pass to update sh_link and sh_info fields of new
2915 sections */
2916 for(i = 1; i < ehdr.e_shnum; i++) {
2917 s = sm_table[i].s;
2918 if (!s || !sm_table[i].new_section)
2919 continue;
2920 sh = &shdr[i];
2921 if (sh->sh_link > 0)
2922 s->link = sm_table[sh->sh_link].s;
2923 if (sh->sh_type == SHT_RELX) {
2924 s->sh_info = sm_table[sh->sh_info].s->sh_num;
2925 /* update backward link */
2926 s1->sections[s->sh_info]->reloc = s;
2929 sm = sm_table;
2931 /* resolve symbols */
2932 old_to_new_syms = tcc_mallocz(nb_syms * sizeof(int));
2934 sym = symtab + 1;
2935 for(i = 1; i < nb_syms; i++, sym++) {
2936 if (sym->st_shndx != SHN_UNDEF &&
2937 sym->st_shndx < SHN_LORESERVE) {
2938 sm = &sm_table[sym->st_shndx];
2939 if (sm->link_once) {
2940 /* if a symbol is in a link once section, we use the
2941 already defined symbol. It is very important to get
2942 correct relocations */
2943 if (ELFW(ST_BIND)(sym->st_info) != STB_LOCAL) {
2944 name = (char *) strtab + sym->st_name;
2945 sym_index = find_elf_sym(symtab_section, name);
2946 if (sym_index)
2947 old_to_new_syms[i] = sym_index;
2949 continue;
2951 /* if no corresponding section added, no need to add symbol */
2952 if (!sm->s)
2953 continue;
2954 /* convert section number */
2955 sym->st_shndx = sm->s->sh_num;
2956 /* offset value */
2957 sym->st_value += sm->offset;
2959 /* add symbol */
2960 name = (char *) strtab + sym->st_name;
2961 sym_index = add_elf_sym(symtab_section, sym->st_value, sym->st_size,
2962 sym->st_info, sym->st_other,
2963 sym->st_shndx, name);
2964 old_to_new_syms[i] = sym_index;
2967 /* third pass to patch relocation entries */
2968 for(i = 1; i < ehdr.e_shnum; i++) {
2969 s = sm_table[i].s;
2970 if (!s)
2971 continue;
2972 sh = &shdr[i];
2973 offset = sm_table[i].offset;
2974 switch(s->sh_type) {
2975 case SHT_RELX:
2976 /* take relocation offset information */
2977 offseti = sm_table[sh->sh_info].offset;
2978 for_each_elem(s, (offset / sizeof(*rel)), rel, ElfW_Rel) {
2979 int type;
2980 unsigned sym_index;
2981 /* convert symbol index */
2982 type = ELFW(R_TYPE)(rel->r_info);
2983 sym_index = ELFW(R_SYM)(rel->r_info);
2984 /* NOTE: only one symtab assumed */
2985 if (sym_index >= nb_syms)
2986 goto invalid_reloc;
2987 sym_index = old_to_new_syms[sym_index];
2988 /* ignore link_once in rel section. */
2989 if (!sym_index && !sm->link_once
2990 #ifdef TCC_TARGET_ARM
2991 && type != R_ARM_V4BX
2992 #endif
2993 ) {
2994 invalid_reloc:
2995 tcc_error_noabort("Invalid relocation entry [%2d] '%s' @ %.8x",
2996 i, strsec + sh->sh_name, rel->r_offset);
2997 goto fail;
2999 rel->r_info = ELFW(R_INFO)(sym_index, type);
3000 /* offset the relocation offset */
3001 rel->r_offset += offseti;
3002 #ifdef TCC_TARGET_ARM
3003 /* Jumps and branches from a Thumb code to a PLT entry need
3004 special handling since PLT entries are ARM code.
3005 Unconditional bl instructions referencing PLT entries are
3006 handled by converting these instructions into blx
3007 instructions. Other case of instructions referencing a PLT
3008 entry require to add a Thumb stub before the PLT entry to
3009 switch to ARM mode. We set bit plt_thumb_stub of the
3010 attribute of a symbol to indicate such a case. */
3011 if (type == R_ARM_THM_JUMP24)
3012 alloc_sym_attr(s1, sym_index)->plt_thumb_stub = 1;
3013 #endif
3015 break;
3016 default:
3017 break;
3021 ret = 0;
3022 the_end:
3023 tcc_free(symtab);
3024 tcc_free(strtab);
3025 tcc_free(old_to_new_syms);
3026 tcc_free(sm_table);
3027 tcc_free(strsec);
3028 tcc_free(shdr);
3029 return ret;
3032 typedef struct ArchiveHeader {
3033 char ar_name[16]; /* name of this member */
3034 char ar_date[12]; /* file mtime */
3035 char ar_uid[6]; /* owner uid; printed as decimal */
3036 char ar_gid[6]; /* owner gid; printed as decimal */
3037 char ar_mode[8]; /* file mode, printed as octal */
3038 char ar_size[10]; /* file size, printed as decimal */
3039 char ar_fmag[2]; /* should contain ARFMAG */
3040 } ArchiveHeader;
3042 static int get_be32(const uint8_t *b)
3044 return b[3] | (b[2] << 8) | (b[1] << 16) | (b[0] << 24);
3047 /* load only the objects which resolve undefined symbols */
3048 static int tcc_load_alacarte(TCCState *s1, int fd, int size)
3050 int i, bound, nsyms, sym_index, off, ret;
3051 uint8_t *data;
3052 const char *ar_names, *p;
3053 const uint8_t *ar_index;
3054 ElfW(Sym) *sym;
3056 data = tcc_malloc(size);
3057 if (read(fd, data, size) != size)
3058 goto fail;
3059 nsyms = get_be32(data);
3060 ar_index = data + 4;
3061 ar_names = (char *) ar_index + nsyms * 4;
3063 do {
3064 bound = 0;
3065 for(p = ar_names, i = 0; i < nsyms; i++, p += strlen(p)+1) {
3066 sym_index = find_elf_sym(symtab_section, p);
3067 if(sym_index) {
3068 sym = &((ElfW(Sym) *)symtab_section->data)[sym_index];
3069 if(sym->st_shndx == SHN_UNDEF) {
3070 off = get_be32(ar_index + i * 4) + sizeof(ArchiveHeader);
3071 ++bound;
3072 lseek(fd, off, SEEK_SET);
3073 if(tcc_load_object_file(s1, fd, off) < 0) {
3074 fail:
3075 ret = -1;
3076 goto the_end;
3081 } while(bound);
3082 ret = 0;
3083 the_end:
3084 tcc_free(data);
3085 return ret;
3088 /* load a '.a' file */
3089 ST_FUNC int tcc_load_archive(TCCState *s1, int fd)
3091 ArchiveHeader hdr;
3092 char ar_size[11];
3093 char ar_name[17];
3094 char magic[8];
3095 int size, len, i;
3096 unsigned long file_offset;
3098 /* skip magic which was already checked */
3099 read(fd, magic, sizeof(magic));
3101 for(;;) {
3102 len = read(fd, &hdr, sizeof(hdr));
3103 if (len == 0)
3104 break;
3105 if (len != sizeof(hdr)) {
3106 tcc_error_noabort("invalid archive");
3107 return -1;
3109 memcpy(ar_size, hdr.ar_size, sizeof(hdr.ar_size));
3110 ar_size[sizeof(hdr.ar_size)] = '\0';
3111 size = strtol(ar_size, NULL, 0);
3112 memcpy(ar_name, hdr.ar_name, sizeof(hdr.ar_name));
3113 for(i = sizeof(hdr.ar_name) - 1; i >= 0; i--) {
3114 if (ar_name[i] != ' ')
3115 break;
3117 ar_name[i + 1] = '\0';
3118 file_offset = lseek(fd, 0, SEEK_CUR);
3119 /* align to even */
3120 size = (size + 1) & ~1;
3121 if (!strcmp(ar_name, "/")) {
3122 /* coff symbol table : we handle it */
3123 if(s1->alacarte_link)
3124 return tcc_load_alacarte(s1, fd, size);
3125 } else if (!strcmp(ar_name, "//") ||
3126 !strcmp(ar_name, "__.SYMDEF") ||
3127 !strcmp(ar_name, "__.SYMDEF/") ||
3128 !strcmp(ar_name, "ARFILENAMES/")) {
3129 /* skip symbol table or archive names */
3130 } else {
3131 if (tcc_load_object_file(s1, fd, file_offset) < 0)
3132 return -1;
3134 lseek(fd, file_offset + size, SEEK_SET);
3136 return 0;
3139 #ifndef TCC_TARGET_PE
3140 /* load a DLL and all referenced DLLs. 'level = 0' means that the DLL
3141 is referenced by the user (so it should be added as DT_NEEDED in
3142 the generated ELF file) */
3143 ST_FUNC int tcc_load_dll(TCCState *s1, int fd, const char *filename, int level)
3145 ElfW(Ehdr) ehdr;
3146 ElfW(Shdr) *shdr, *sh, *sh1;
3147 int i, j, nb_syms, nb_dts, sym_bind, ret;
3148 ElfW(Sym) *sym, *dynsym;
3149 ElfW(Dyn) *dt, *dynamic;
3150 unsigned char *dynstr;
3151 const char *name, *soname;
3152 DLLReference *dllref;
3154 read(fd, &ehdr, sizeof(ehdr));
3156 /* test CPU specific stuff */
3157 if (ehdr.e_ident[5] != ELFDATA2LSB ||
3158 ehdr.e_machine != EM_TCC_TARGET) {
3159 tcc_error_noabort("bad architecture");
3160 return -1;
3163 /* read sections */
3164 shdr = load_data(fd, ehdr.e_shoff, sizeof(ElfW(Shdr)) * ehdr.e_shnum);
3166 /* load dynamic section and dynamic symbols */
3167 nb_syms = 0;
3168 nb_dts = 0;
3169 dynamic = NULL;
3170 dynsym = NULL; /* avoid warning */
3171 dynstr = NULL; /* avoid warning */
3172 for(i = 0, sh = shdr; i < ehdr.e_shnum; i++, sh++) {
3173 switch(sh->sh_type) {
3174 case SHT_DYNAMIC:
3175 nb_dts = sh->sh_size / sizeof(ElfW(Dyn));
3176 dynamic = load_data(fd, sh->sh_offset, sh->sh_size);
3177 break;
3178 case SHT_DYNSYM:
3179 nb_syms = sh->sh_size / sizeof(ElfW(Sym));
3180 dynsym = load_data(fd, sh->sh_offset, sh->sh_size);
3181 sh1 = &shdr[sh->sh_link];
3182 dynstr = load_data(fd, sh1->sh_offset, sh1->sh_size);
3183 break;
3184 default:
3185 break;
3189 /* compute the real library name */
3190 soname = tcc_basename(filename);
3192 for(i = 0, dt = dynamic; i < nb_dts; i++, dt++) {
3193 if (dt->d_tag == DT_SONAME) {
3194 soname = (char *) dynstr + dt->d_un.d_val;
3198 /* if the dll is already loaded, do not load it */
3199 for(i = 0; i < s1->nb_loaded_dlls; i++) {
3200 dllref = s1->loaded_dlls[i];
3201 if (!strcmp(soname, dllref->name)) {
3202 /* but update level if needed */
3203 if (level < dllref->level)
3204 dllref->level = level;
3205 ret = 0;
3206 goto the_end;
3210 /* add the dll and its level */
3211 dllref = tcc_mallocz(sizeof(DLLReference) + strlen(soname));
3212 dllref->level = level;
3213 strcpy(dllref->name, soname);
3214 dynarray_add((void ***)&s1->loaded_dlls, &s1->nb_loaded_dlls, dllref);
3216 /* add dynamic symbols in dynsym_section */
3217 for(i = 1, sym = dynsym + 1; i < nb_syms; i++, sym++) {
3218 sym_bind = ELFW(ST_BIND)(sym->st_info);
3219 if (sym_bind == STB_LOCAL)
3220 continue;
3221 name = (char *) dynstr + sym->st_name;
3222 add_elf_sym(s1->dynsymtab_section, sym->st_value, sym->st_size,
3223 sym->st_info, sym->st_other, sym->st_shndx, name);
3226 /* load all referenced DLLs */
3227 for(i = 0, dt = dynamic; i < nb_dts; i++, dt++) {
3228 switch(dt->d_tag) {
3229 case DT_NEEDED:
3230 name = (char *) dynstr + dt->d_un.d_val;
3231 for(j = 0; j < s1->nb_loaded_dlls; j++) {
3232 dllref = s1->loaded_dlls[j];
3233 if (!strcmp(name, dllref->name))
3234 goto already_loaded;
3236 if (tcc_add_dll(s1, name, AFF_REFERENCED_DLL) < 0) {
3237 tcc_error_noabort("referenced dll '%s' not found", name);
3238 ret = -1;
3239 goto the_end;
3241 already_loaded:
3242 break;
3245 ret = 0;
3246 the_end:
3247 tcc_free(dynstr);
3248 tcc_free(dynsym);
3249 tcc_free(dynamic);
3250 tcc_free(shdr);
3251 return ret;
3254 #define LD_TOK_NAME 256
3255 #define LD_TOK_EOF (-1)
3257 /* return next ld script token */
3258 static int ld_next(TCCState *s1, char *name, int name_size)
3260 int c;
3261 char *q;
3263 redo:
3264 switch(ch) {
3265 case ' ':
3266 case '\t':
3267 case '\f':
3268 case '\v':
3269 case '\r':
3270 case '\n':
3271 inp();
3272 goto redo;
3273 case '/':
3274 minp();
3275 if (ch == '*') {
3276 file->buf_ptr = parse_comment(file->buf_ptr);
3277 ch = file->buf_ptr[0];
3278 goto redo;
3279 } else {
3280 q = name;
3281 *q++ = '/';
3282 goto parse_name;
3284 break;
3285 case '\\':
3286 ch = handle_eob();
3287 if (ch != '\\')
3288 goto redo;
3289 /* fall through */
3290 /* case 'a' ... 'z': */
3291 case 'a':
3292 case 'b':
3293 case 'c':
3294 case 'd':
3295 case 'e':
3296 case 'f':
3297 case 'g':
3298 case 'h':
3299 case 'i':
3300 case 'j':
3301 case 'k':
3302 case 'l':
3303 case 'm':
3304 case 'n':
3305 case 'o':
3306 case 'p':
3307 case 'q':
3308 case 'r':
3309 case 's':
3310 case 't':
3311 case 'u':
3312 case 'v':
3313 case 'w':
3314 case 'x':
3315 case 'y':
3316 case 'z':
3317 /* case 'A' ... 'z': */
3318 case 'A':
3319 case 'B':
3320 case 'C':
3321 case 'D':
3322 case 'E':
3323 case 'F':
3324 case 'G':
3325 case 'H':
3326 case 'I':
3327 case 'J':
3328 case 'K':
3329 case 'L':
3330 case 'M':
3331 case 'N':
3332 case 'O':
3333 case 'P':
3334 case 'Q':
3335 case 'R':
3336 case 'S':
3337 case 'T':
3338 case 'U':
3339 case 'V':
3340 case 'W':
3341 case 'X':
3342 case 'Y':
3343 case 'Z':
3344 case '_':
3345 case '.':
3346 case '$':
3347 case '~':
3348 q = name;
3349 parse_name:
3350 for(;;) {
3351 if (!((ch >= 'a' && ch <= 'z') ||
3352 (ch >= 'A' && ch <= 'Z') ||
3353 (ch >= '0' && ch <= '9') ||
3354 strchr("/.-_+=$:\\,~", ch)))
3355 break;
3356 if ((q - name) < name_size - 1) {
3357 *q++ = ch;
3359 minp();
3361 *q = '\0';
3362 c = LD_TOK_NAME;
3363 break;
3364 case CH_EOF:
3365 c = LD_TOK_EOF;
3366 break;
3367 default:
3368 c = ch;
3369 inp();
3370 break;
3372 return c;
3375 static int ld_add_file(TCCState *s1, const char filename[])
3377 int ret;
3379 ret = tcc_add_file_internal(s1, filename, 0, TCC_FILETYPE_BINARY);
3380 if (ret)
3381 ret = tcc_add_dll(s1, filename, 0);
3382 return ret;
3385 static inline int new_undef_syms(void)
3387 int ret = 0;
3388 ret = new_undef_sym;
3389 new_undef_sym = 0;
3390 return ret;
3393 static int ld_add_file_list(TCCState *s1, const char *cmd, int as_needed)
3395 char filename[1024], libname[1024];
3396 int t, group, nblibs = 0, ret = 0;
3397 char **libs = NULL;
3399 group = !strcmp(cmd, "GROUP");
3400 if (!as_needed)
3401 new_undef_syms();
3402 t = ld_next(s1, filename, sizeof(filename));
3403 if (t != '(')
3404 expect("(");
3405 t = ld_next(s1, filename, sizeof(filename));
3406 for(;;) {
3407 libname[0] = '\0';
3408 if (t == LD_TOK_EOF) {
3409 tcc_error_noabort("unexpected end of file");
3410 ret = -1;
3411 goto lib_parse_error;
3412 } else if (t == ')') {
3413 break;
3414 } else if (t == '-') {
3415 t = ld_next(s1, filename, sizeof(filename));
3416 if ((t != LD_TOK_NAME) || (filename[0] != 'l')) {
3417 tcc_error_noabort("library name expected");
3418 ret = -1;
3419 goto lib_parse_error;
3421 pstrcpy(libname, sizeof libname, &filename[1]);
3422 if (s1->static_link) {
3423 snprintf(filename, sizeof filename, "lib%s.a", libname);
3424 } else {
3425 snprintf(filename, sizeof filename, "lib%s.so", libname);
3427 } else if (t != LD_TOK_NAME) {
3428 tcc_error_noabort("filename expected");
3429 ret = -1;
3430 goto lib_parse_error;
3432 if (!strcmp(filename, "AS_NEEDED")) {
3433 ret = ld_add_file_list(s1, cmd, 1);
3434 if (ret)
3435 goto lib_parse_error;
3436 } else {
3437 /* TODO: Implement AS_NEEDED support. Ignore it for now */
3438 if (!as_needed) {
3439 ret = ld_add_file(s1, filename);
3440 if (ret)
3441 goto lib_parse_error;
3442 if (group) {
3443 /* Add the filename *and* the libname to avoid future conversions */
3444 dynarray_add((void ***) &libs, &nblibs, tcc_strdup(filename));
3445 if (libname[0] != '\0')
3446 dynarray_add((void ***) &libs, &nblibs, tcc_strdup(libname));
3450 t = ld_next(s1, filename, sizeof(filename));
3451 if (t == ',') {
3452 t = ld_next(s1, filename, sizeof(filename));
3455 if (group && !as_needed) {
3456 while (new_undef_syms()) {
3457 int i;
3459 for (i = 0; i < nblibs; i ++)
3460 ld_add_file(s1, libs[i]);
3463 lib_parse_error:
3464 dynarray_reset(&libs, &nblibs);
3465 return ret;
3468 /* interpret a subset of GNU ldscripts to handle the dummy libc.so
3469 files */
3470 ST_FUNC int tcc_load_ldscript(TCCState *s1)
3472 char cmd[64];
3473 char filename[1024];
3474 int t, ret;
3476 ch = handle_eob();
3477 for(;;) {
3478 t = ld_next(s1, cmd, sizeof(cmd));
3479 if (t == LD_TOK_EOF)
3480 return 0;
3481 else if (t != LD_TOK_NAME)
3482 return -1;
3483 if (!strcmp(cmd, "INPUT") ||
3484 !strcmp(cmd, "GROUP")) {
3485 ret = ld_add_file_list(s1, cmd, 0);
3486 if (ret)
3487 return ret;
3488 } else if (!strcmp(cmd, "OUTPUT_FORMAT") ||
3489 !strcmp(cmd, "TARGET")) {
3490 /* ignore some commands */
3491 t = ld_next(s1, cmd, sizeof(cmd));
3492 if (t != '(')
3493 expect("(");
3494 for(;;) {
3495 t = ld_next(s1, filename, sizeof(filename));
3496 if (t == LD_TOK_EOF) {
3497 tcc_error_noabort("unexpected end of file");
3498 return -1;
3499 } else if (t == ')') {
3500 break;
3503 } else {
3504 return -1;
3507 return 0;
3509 #endif /* !TCC_TARGET_PE */