fix-mixed-struct (patch by Pip Cet)
[tinycc.git] / tccelf.c
blob1f879465d3ee2207cf292de3096922aa441676ea
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", 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 *(int *)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 *(int *)ptr += val - addr;
545 break;
546 case R_386_PLT32:
547 *(int *)ptr += val - addr;
548 break;
549 case R_386_GLOB_DAT:
550 case R_386_JMP_SLOT:
551 *(int *)ptr = val;
552 break;
553 case R_386_GOTPC:
554 *(int *)ptr += s1->got->sh_addr - addr;
555 break;
556 case R_386_GOTOFF:
557 *(int *)ptr += val - s1->got->sh_addr;
558 break;
559 case R_386_GOT32:
560 /* we load the got offset */
561 *(int *)ptr += s1->sym_attrs[sym_index].got_offset;
562 break;
563 case R_386_16:
564 if (s1->output_format != TCC_OUTPUT_FORMAT_BINARY) {
565 output_file:
566 tcc_error("can only produce 16-bit binary files");
568 *(short *)ptr += val;
569 break;
570 case R_386_PC16:
571 if (s1->output_format != TCC_OUTPUT_FORMAT_BINARY)
572 goto output_file;
573 *(short *)ptr += val - addr;
574 break;
575 #elif defined(TCC_TARGET_ARM)
576 case R_ARM_PC24:
577 case R_ARM_CALL:
578 case R_ARM_JUMP24:
579 case R_ARM_PLT32:
581 int x, is_thumb, is_call, h, blx_avail, is_bl, th_ko;
582 x = (*(int *) ptr) & 0xffffff;
583 if (sym->st_shndx == SHN_UNDEF)
584 val = s1->plt->sh_addr;
585 #ifdef DEBUG_RELOC
586 printf ("reloc %d: x=0x%x val=0x%x ", type, x, val);
587 #endif
588 (*(int *)ptr) &= 0xff000000;
589 if (x & 0x800000)
590 x -= 0x1000000;
591 x <<= 2;
592 blx_avail = (TCC_ARM_VERSION >= 5);
593 is_thumb = val & 1;
594 is_bl = (*(unsigned *) ptr) >> 24 == 0xeb;
595 is_call = (type == R_ARM_CALL || (type == R_ARM_PC24 && is_bl));
596 x += val - addr;
597 #ifdef DEBUG_RELOC
598 printf (" newx=0x%x name=%s\n", x,
599 (char *) symtab_section->link->data + sym->st_name);
600 #endif
601 h = x & 2;
602 th_ko = (x & 3) && (!blx_avail || !is_call);
603 if (th_ko || x >= 0x2000000 || x < -0x2000000)
604 tcc_error("can't relocate value at %x,%d",addr, type);
605 x >>= 2;
606 x &= 0xffffff;
607 /* Only reached if blx is avail and it is a call */
608 if (is_thumb) {
609 x |= h << 24;
610 (*(int *)ptr) = 0xfa << 24; /* bl -> blx */
612 (*(int *) ptr) |= x;
614 break;
615 /* Since these relocations only concern Thumb-2 and blx instruction was
616 introduced before Thumb-2, we can assume blx is available and not
617 guard its use */
618 case R_ARM_THM_PC22:
619 case R_ARM_THM_JUMP24:
621 int x, hi, lo, s, j1, j2, i1, i2, imm10, imm11;
622 int to_thumb, is_call, to_plt, blx_bit = 1 << 12;
623 Section *plt;
625 /* weak reference */
626 if (sym->st_shndx == SHN_UNDEF &&
627 ELFW(ST_BIND)(sym->st_info) == STB_WEAK)
628 break;
630 /* Get initial offset */
631 hi = (*(uint16_t *)ptr);
632 lo = (*(uint16_t *)(ptr+2));
633 s = (hi >> 10) & 1;
634 j1 = (lo >> 13) & 1;
635 j2 = (lo >> 11) & 1;
636 i1 = (j1 ^ s) ^ 1;
637 i2 = (j2 ^ s) ^ 1;
638 imm10 = hi & 0x3ff;
639 imm11 = lo & 0x7ff;
640 x = (s << 24) | (i1 << 23) | (i2 << 22) |
641 (imm10 << 12) | (imm11 << 1);
642 if (x & 0x01000000)
643 x -= 0x02000000;
645 /* Relocation infos */
646 to_thumb = val & 1;
647 plt = s1->plt;
648 to_plt = (val >= plt->sh_addr) &&
649 (val < plt->sh_addr + plt->data_offset);
650 is_call = (type == R_ARM_THM_PC22);
652 /* Compute final offset */
653 if (to_plt && !is_call) /* Point to 1st instr of Thumb stub */
654 x -= 4;
655 x += val - addr;
656 if (!to_thumb && is_call) {
657 blx_bit = 0; /* bl -> blx */
658 x = (x + 3) & -4; /* Compute offset from aligned PC */
661 /* Check that relocation is possible
662 * offset must not be out of range
663 * if target is to be entered in arm mode:
664 - bit 1 must not set
665 - instruction must be a call (bl) or a jump to PLT */
666 if (!to_thumb || x >= 0x1000000 || x < -0x1000000)
667 if (to_thumb || (val & 2) || (!is_call && !to_plt))
668 tcc_error("can't relocate value at %x,%d",addr, type);
670 /* Compute and store final offset */
671 s = (x >> 24) & 1;
672 i1 = (x >> 23) & 1;
673 i2 = (x >> 22) & 1;
674 j1 = s ^ (i1 ^ 1);
675 j2 = s ^ (i2 ^ 1);
676 imm10 = (x >> 12) & 0x3ff;
677 imm11 = (x >> 1) & 0x7ff;
678 (*(uint16_t *)ptr) = (uint16_t) ((hi & 0xf800) |
679 (s << 10) | imm10);
680 (*(uint16_t *)(ptr+2)) = (uint16_t) ((lo & 0xc000) |
681 (j1 << 13) | blx_bit | (j2 << 11) |
682 imm11);
684 break;
685 case R_ARM_MOVT_ABS:
686 case R_ARM_MOVW_ABS_NC:
688 int x, imm4, imm12;
689 if (type == R_ARM_MOVT_ABS)
690 val >>= 16;
691 imm12 = val & 0xfff;
692 imm4 = (val >> 12) & 0xf;
693 x = (imm4 << 16) | imm12;
694 if (type == R_ARM_THM_MOVT_ABS)
695 *(int *)ptr |= x;
696 else
697 *(int *)ptr += x;
699 break;
700 case R_ARM_THM_MOVT_ABS:
701 case R_ARM_THM_MOVW_ABS_NC:
703 int x, i, imm4, imm3, imm8;
704 if (type == R_ARM_THM_MOVT_ABS)
705 val >>= 16;
706 imm8 = val & 0xff;
707 imm3 = (val >> 8) & 0x7;
708 i = (val >> 11) & 1;
709 imm4 = (val >> 12) & 0xf;
710 x = (imm3 << 28) | (imm8 << 16) | (i << 10) | imm4;
711 if (type == R_ARM_THM_MOVT_ABS)
712 *(int *)ptr |= x;
713 else
714 *(int *)ptr += x;
716 break;
717 case R_ARM_PREL31:
719 int x;
720 x = (*(int *)ptr) & 0x7fffffff;
721 (*(int *)ptr) &= 0x80000000;
722 x = (x * 2) / 2;
723 x += val - addr;
724 if((x^(x>>1))&0x40000000)
725 tcc_error("can't relocate value at %x,%d",addr, type);
726 (*(int *)ptr) |= x & 0x7fffffff;
728 case R_ARM_ABS32:
729 *(int *)ptr += val;
730 break;
731 case R_ARM_REL32:
732 *(int *)ptr += val - addr;
733 break;
734 case R_ARM_GOTPC:
735 *(int *)ptr += s1->got->sh_addr - addr;
736 break;
737 case R_ARM_GOTOFF:
738 *(int *)ptr += val - s1->got->sh_addr;
739 break;
740 case R_ARM_GOT32:
741 /* we load the got offset */
742 *(int *)ptr += s1->sym_attrs[sym_index].got_offset;
743 break;
744 case R_ARM_COPY:
745 break;
746 case R_ARM_V4BX:
747 /* trade Thumb support for ARMv4 support */
748 if ((0x0ffffff0 & *(int*)ptr) == 0x012FFF10)
749 *(int*)ptr ^= 0xE12FFF10 ^ 0xE1A0F000; /* BX Rm -> MOV PC, Rm */
750 break;
751 case R_ARM_GLOB_DAT:
752 case R_ARM_JUMP_SLOT:
753 *(addr_t *)ptr = val;
754 break;
755 case R_ARM_NONE:
756 /* Nothing to do. Normally used to indicate a dependency
757 on a certain symbol (like for exception handling under EABI). */
758 break;
759 default:
760 fprintf(stderr,"FIXME: handle reloc type %x at %x [%p] to %x\n",
761 type, (unsigned)addr, ptr, (unsigned)val);
762 break;
763 #elif defined(TCC_TARGET_ARM64)
764 case R_AARCH64_ABS64:
765 *(uint64_t *)ptr = val;
766 break;
767 case R_AARCH64_ABS32:
768 *(uint32_t *)ptr = val;
769 break;
770 case R_AARCH64_MOVW_UABS_G0_NC:
771 *(uint32_t *)ptr = (*(uint32_t *)ptr & 0xffe0001f) |
772 (val & 0xffff) << 5;
773 break;
774 case R_AARCH64_MOVW_UABS_G1_NC:
775 *(uint32_t *)ptr = (*(uint32_t *)ptr & 0xffe0001f) |
776 (val >> 16 & 0xffff) << 5;
777 break;
778 case R_AARCH64_MOVW_UABS_G2_NC:
779 *(uint32_t *)ptr = (*(uint32_t *)ptr & 0xffe0001f) |
780 (val >> 32 & 0xffff) << 5;
781 break;
782 case R_AARCH64_MOVW_UABS_G3:
783 *(uint32_t *)ptr = (*(uint32_t *)ptr & 0xffe0001f) |
784 (val >> 48 & 0xffff) << 5;
785 break;
786 case R_AARCH64_ADR_PREL_PG_HI21: {
787 uint64_t off = (val >> 12) - (addr >> 12);
788 if ((off + ((uint64_t)1 << 20)) >> 21)
789 tcc_error("R_AARCH64_ADR_PREL_PG_HI21 relocation failed");
790 *(uint32_t *)ptr = (*(uint32_t *)ptr & 0x9f00001f) |
791 (off & 0x1ffffc) << 3 | (off & 3) << 29;
792 break;
794 case R_AARCH64_ADD_ABS_LO12_NC:
795 *(uint32_t *)ptr = (*(uint32_t *)ptr & 0xffc003ff) |
796 (val & 0xfff) << 10;
797 break;
798 case R_AARCH64_JUMP26:
799 case R_AARCH64_CALL26:
800 /* This check must match the one in build_got_entries, testing
801 if we really need a PLT slot. */
802 if (sym->st_shndx == SHN_UNDEF)
803 /* We've put the PLT slot offset into r_addend when generating
804 it, and that's what we must use as relocation value (adjusted
805 by section offset of course). */
806 val = s1->plt->sh_addr + rel->r_addend;
807 #ifdef DEBUG_RELOC
808 printf ("reloc %d @ 0x%lx: val=0x%lx name=%s\n", type, addr, val,
809 (char *) symtab_section->link->data + sym->st_name);
810 #endif
811 if (((val - addr) + ((uint64_t)1 << 27)) & ~(uint64_t)0xffffffc)
813 tcc_error("R_AARCH64_(JUMP|CALL)26 relocation failed (val=%lx, addr=%lx)", addr, val);
815 *(uint32_t *)ptr = 0x14000000 | (type == R_AARCH64_CALL26) << 31 |
816 ((val - addr) >> 2 & 0x3ffffff);
817 break;
818 case R_AARCH64_ADR_GOT_PAGE: {
819 uint64_t off =
820 (((s1->got->sh_addr +
821 s1->sym_attrs[sym_index].got_offset) >> 12) - (addr >> 12));
822 if ((off + ((uint64_t)1 << 20)) >> 21)
823 tcc_error("R_AARCH64_ADR_GOT_PAGE relocation failed");
824 *(uint32_t *)ptr = (*(uint32_t *)ptr & 0x9f00001f) |
825 (off & 0x1ffffc) << 3 | (off & 3) << 29;
826 break;
828 case R_AARCH64_LD64_GOT_LO12_NC:
829 *(uint32_t *)ptr = (*(uint32_t *)ptr & 0xfff803ff) |
830 ((s1->got->sh_addr + s1->sym_attrs[sym_index].got_offset)
831 & 0xff8) << 7;
832 break;
833 case R_AARCH64_COPY:
834 break;
835 case R_AARCH64_GLOB_DAT:
836 case R_AARCH64_JUMP_SLOT:
837 /* They don't need addend */
838 #ifdef DEBUG_RELOC
839 printf ("reloc %d @ 0x%lx: val=0x%lx name=%s\n", type, addr,
840 val - rel->r_addend,
841 (char *) symtab_section->link->data + sym->st_name);
842 #endif
843 *(addr_t *)ptr = val - rel->r_addend;
844 break;
845 default:
846 fprintf(stderr, "FIXME: handle reloc type %x at %x [%p] to %x\n",
847 type, (unsigned)addr, ptr, (unsigned)val);
848 break;
849 #elif defined(TCC_TARGET_C67)
850 case R_C60_32:
851 *(int *)ptr += val;
852 break;
853 case R_C60LO16:
855 uint32_t orig;
857 /* put the low 16 bits of the absolute address
858 add to what is already there */
860 orig = ((*(int *)(ptr )) >> 7) & 0xffff;
861 orig |= (((*(int *)(ptr+4)) >> 7) & 0xffff) << 16;
863 /* patch both at once - assumes always in pairs Low - High */
865 *(int *) ptr = (*(int *) ptr & (~(0xffff << 7)) ) | (((val+orig) & 0xffff) << 7);
866 *(int *)(ptr+4) = (*(int *)(ptr+4) & (~(0xffff << 7)) ) | ((((val+orig)>>16) & 0xffff) << 7);
868 break;
869 case R_C60HI16:
870 break;
871 default:
872 fprintf(stderr,"FIXME: handle reloc type %x at %x [%p] to %x\n",
873 type, (unsigned)addr, ptr, (unsigned)val);
874 break;
875 #elif defined(TCC_TARGET_X86_64)
876 case R_X86_64_64:
877 if (s1->output_type == TCC_OUTPUT_DLL) {
878 esym_index = s1->symtab_to_dynsym[sym_index];
879 qrel->r_offset = rel->r_offset;
880 if (esym_index) {
881 qrel->r_info = ELFW(R_INFO)(esym_index, R_X86_64_64);
882 qrel->r_addend = rel->r_addend;
883 qrel++;
884 break;
885 } else {
886 qrel->r_info = ELFW(R_INFO)(0, R_X86_64_RELATIVE);
887 qrel->r_addend = *(long long *)ptr + val;
888 qrel++;
891 *(long long *)ptr += val;
892 break;
893 case R_X86_64_32:
894 case R_X86_64_32S:
895 if (s1->output_type == TCC_OUTPUT_DLL) {
896 /* XXX: this logic may depend on TCC's codegen
897 now TCC uses R_X86_64_32 even for a 64bit pointer */
898 qrel->r_info = ELFW(R_INFO)(0, R_X86_64_RELATIVE);
899 qrel->r_addend = *(int *)ptr + val;
900 qrel++;
902 *(int *)ptr += val;
903 break;
905 case R_X86_64_PC32:
906 if (s1->output_type == TCC_OUTPUT_DLL) {
907 /* DLL relocation */
908 esym_index = s1->symtab_to_dynsym[sym_index];
909 if (esym_index) {
910 qrel->r_offset = rel->r_offset;
911 qrel->r_info = ELFW(R_INFO)(esym_index, R_X86_64_PC32);
912 qrel->r_addend = *(int *)ptr;
913 qrel++;
914 break;
917 goto plt32pc32;
919 case R_X86_64_PLT32:
920 /* We've put the PLT slot offset into r_addend when generating
921 it, and that's what we must use as relocation value (adjusted
922 by section offset of course). */
923 val = s1->plt->sh_addr + rel->r_addend;
924 /* fallthrough. */
926 plt32pc32:
928 long long diff;
929 diff = (long long)val - addr;
930 if (diff < -2147483648LL || diff > 2147483647LL) {
931 tcc_error("internal error: relocation failed");
933 *(int *)ptr += diff;
935 break;
936 case R_X86_64_GLOB_DAT:
937 case R_X86_64_JUMP_SLOT:
938 /* They don't need addend */
939 *(addr_t *)ptr = val - rel->r_addend;
940 break;
941 case R_X86_64_GOTPCREL:
942 *(int *)ptr += (s1->got->sh_addr - addr +
943 s1->sym_attrs[sym_index].got_offset - 4);
944 break;
945 case R_X86_64_GOTTPOFF:
946 *(int *)ptr += val - s1->got->sh_addr;
947 break;
948 case R_X86_64_GOT32:
949 /* we load the got offset */
950 *(int *)ptr += s1->sym_attrs[sym_index].got_offset;
951 break;
952 #else
953 #error unsupported processor
954 #endif
957 /* if the relocation is allocated, we change its symbol table */
958 if (sr->sh_flags & SHF_ALLOC)
959 sr->link = s1->dynsym;
962 /* relocate relocation table in 'sr' */
963 static void relocate_rel(TCCState *s1, Section *sr)
965 Section *s;
966 ElfW_Rel *rel;
968 s = s1->sections[sr->sh_info];
969 for_each_elem(sr, 0, rel, ElfW_Rel)
970 rel->r_offset += s->sh_addr;
973 /* count the number of dynamic relocations so that we can reserve
974 their space */
975 static int prepare_dynamic_rel(TCCState *s1, Section *sr)
977 ElfW_Rel *rel;
978 int sym_index, esym_index, type, count;
980 count = 0;
981 for_each_elem(sr, 0, rel, ElfW_Rel) {
982 sym_index = ELFW(R_SYM)(rel->r_info);
983 type = ELFW(R_TYPE)(rel->r_info);
984 switch(type) {
985 #if defined(TCC_TARGET_I386)
986 case R_386_32:
987 #elif defined(TCC_TARGET_X86_64)
988 case R_X86_64_32:
989 case R_X86_64_32S:
990 case R_X86_64_64:
991 #endif
992 count++;
993 break;
994 #if defined(TCC_TARGET_I386)
995 case R_386_PC32:
996 #elif defined(TCC_TARGET_X86_64)
997 case R_X86_64_PC32:
998 #endif
999 esym_index = s1->symtab_to_dynsym[sym_index];
1000 if (esym_index)
1001 count++;
1002 break;
1003 default:
1004 break;
1007 if (count) {
1008 /* allocate the section */
1009 sr->sh_flags |= SHF_ALLOC;
1010 sr->sh_size = count * sizeof(ElfW_Rel);
1012 return count;
1015 static struct sym_attr *alloc_sym_attr(TCCState *s1, int index)
1017 int n;
1018 struct sym_attr *tab;
1020 if (index >= s1->nb_sym_attrs) {
1021 /* find immediately bigger power of 2 and reallocate array */
1022 n = 1;
1023 while (index >= n)
1024 n *= 2;
1025 tab = tcc_realloc(s1->sym_attrs, n * sizeof(*s1->sym_attrs));
1026 s1->sym_attrs = tab;
1027 memset(s1->sym_attrs + s1->nb_sym_attrs, 0,
1028 (n - s1->nb_sym_attrs) * sizeof(*s1->sym_attrs));
1029 s1->nb_sym_attrs = n;
1031 return &s1->sym_attrs[index];
1034 /* XXX: suppress that */
1035 static void put32(unsigned char *p, uint32_t val)
1037 p[0] = val;
1038 p[1] = val >> 8;
1039 p[2] = val >> 16;
1040 p[3] = val >> 24;
1043 #if defined(TCC_TARGET_I386) || defined(TCC_TARGET_ARM) || \
1044 defined(TCC_TARGET_ARM64) || defined(TCC_TARGET_X86_64)
1045 static uint32_t get32(unsigned char *p)
1047 return p[0] | (p[1] << 8) | (p[2] << 16) | (p[3] << 24);
1049 #endif
1051 static void build_got(TCCState *s1)
1053 unsigned char *ptr;
1055 /* if no got, then create it */
1056 s1->got = new_section(s1, ".got", SHT_PROGBITS, SHF_ALLOC | SHF_WRITE);
1057 s1->got->sh_entsize = 4;
1058 add_elf_sym(symtab_section, 0, 4, ELFW(ST_INFO)(STB_GLOBAL, STT_OBJECT),
1059 0, s1->got->sh_num, "_GLOBAL_OFFSET_TABLE_");
1060 ptr = section_ptr_add(s1->got, 3 * PTR_SIZE);
1061 #if PTR_SIZE == 4
1062 /* keep space for _DYNAMIC pointer, if present */
1063 put32(ptr, 0);
1064 /* two dummy got entries */
1065 put32(ptr + 4, 0);
1066 put32(ptr + 8, 0);
1067 #else
1068 /* keep space for _DYNAMIC pointer, if present */
1069 put32(ptr, 0);
1070 put32(ptr + 4, 0);
1071 /* two dummy got entries */
1072 put32(ptr + 8, 0);
1073 put32(ptr + 12, 0);
1074 put32(ptr + 16, 0);
1075 put32(ptr + 20, 0);
1076 #endif
1079 /* put a got or plt entry corresponding to a symbol in symtab_section. 'size'
1080 and 'info' can be modifed if more precise info comes from the DLL.
1081 Returns offset of GOT or PLT slot. */
1082 static unsigned long put_got_entry(TCCState *s1,
1083 int reloc_type, unsigned long size, int info,
1084 int sym_index)
1086 int index, need_plt_entry;
1087 const char *name;
1088 ElfW(Sym) *sym;
1089 unsigned long offset;
1090 int *ptr;
1091 struct sym_attr *symattr;
1093 if (!s1->got)
1094 build_got(s1);
1096 need_plt_entry =
1097 #ifdef TCC_TARGET_X86_64
1098 (reloc_type == R_X86_64_JUMP_SLOT);
1099 #elif defined(TCC_TARGET_I386)
1100 (reloc_type == R_386_JMP_SLOT);
1101 #elif defined(TCC_TARGET_ARM)
1102 (reloc_type == R_ARM_JUMP_SLOT);
1103 #elif defined(TCC_TARGET_ARM64)
1104 (reloc_type == R_AARCH64_JUMP_SLOT);
1105 #else
1106 0;
1107 #endif
1109 if (need_plt_entry && !s1->plt) {
1110 /* add PLT */
1111 s1->plt = new_section(s1, ".plt", SHT_PROGBITS,
1112 SHF_ALLOC | SHF_EXECINSTR);
1113 s1->plt->sh_entsize = 4;
1116 /* If a got/plt entry already exists for that symbol, no need to add one */
1117 if (sym_index < s1->nb_sym_attrs) {
1118 if (need_plt_entry && s1->sym_attrs[sym_index].plt_offset)
1119 return s1->sym_attrs[sym_index].plt_offset;
1120 else if (!need_plt_entry && s1->sym_attrs[sym_index].got_offset)
1121 return s1->sym_attrs[sym_index].got_offset;
1124 symattr = alloc_sym_attr(s1, sym_index);
1126 /* Only store the GOT offset if it's not generated for the PLT entry. */
1127 if (!need_plt_entry)
1128 symattr->got_offset = s1->got->data_offset;
1130 sym = &((ElfW(Sym) *)symtab_section->data)[sym_index];
1131 name = (char *) symtab_section->link->data + sym->st_name;
1132 offset = sym->st_value;
1133 #if defined(TCC_TARGET_I386) || defined(TCC_TARGET_X86_64)
1134 if (need_plt_entry) {
1135 Section *plt;
1136 uint8_t *p;
1137 int modrm;
1138 unsigned long relofs;
1140 #if defined(TCC_OUTPUT_DLL_WITH_PLT)
1141 modrm = 0x25;
1142 #else
1143 /* if we build a DLL, we add a %ebx offset */
1144 if (s1->output_type == TCC_OUTPUT_DLL)
1145 modrm = 0xa3;
1146 else
1147 modrm = 0x25;
1148 #endif
1150 /* add a PLT entry */
1151 plt = s1->plt;
1152 if (plt->data_offset == 0) {
1153 /* first plt entry */
1154 p = section_ptr_add(plt, 16);
1155 p[0] = 0xff; /* pushl got + PTR_SIZE */
1156 p[1] = modrm + 0x10;
1157 put32(p + 2, PTR_SIZE);
1158 p[6] = 0xff; /* jmp *(got + PTR_SIZE * 2) */
1159 p[7] = modrm;
1160 put32(p + 8, PTR_SIZE * 2);
1163 /* The PLT slot refers to the relocation entry it needs
1164 via offset. The reloc entry is created below, so its
1165 offset is the current data_offset. */
1166 relofs = s1->got->reloc ? s1->got->reloc->data_offset : 0;
1167 symattr->plt_offset = plt->data_offset;
1168 p = section_ptr_add(plt, 16);
1169 p[0] = 0xff; /* jmp *(got + x) */
1170 p[1] = modrm;
1171 put32(p + 2, s1->got->data_offset);
1172 p[6] = 0x68; /* push $xxx */
1173 #ifdef TCC_TARGET_X86_64
1174 /* On x86-64, the relocation is referred to by _index_. */
1175 put32(p + 7, relofs / sizeof (ElfW_Rel));
1176 #else
1177 put32(p + 7, relofs);
1178 #endif
1179 p[11] = 0xe9; /* jmp plt_start */
1180 put32(p + 12, -(plt->data_offset));
1182 /* If this was an UNDEF symbol set the offset in the
1183 dynsymtab to the PLT slot, so that PC32 relocs to it
1184 can be resolved. */
1185 if (sym->st_shndx == SHN_UNDEF)
1186 offset = plt->data_offset - 16;
1188 #elif defined(TCC_TARGET_ARM)
1189 if (need_plt_entry) {
1190 Section *plt;
1191 uint8_t *p;
1193 /* if we build a DLL, we add a %ebx offset */
1194 if (s1->output_type == TCC_OUTPUT_DLL)
1195 tcc_error("DLLs unimplemented!");
1197 /* add a PLT entry */
1198 plt = s1->plt;
1199 if (plt->data_offset == 0) {
1200 /* first plt entry */
1201 p = section_ptr_add(plt, 16);
1202 put32(p, 0xe52de004); /* push {lr} */
1203 put32(p+4, 0xe59fe010); /* ldr lr, [pc, #16] */
1204 put32(p+8, 0xe08fe00e); /* add lr, pc, lr */
1205 put32(p+12, 0xe5bef008); /* ldr pc, [lr, #8]! */
1208 symattr->plt_offset = plt->data_offset;
1209 if (symattr->plt_thumb_stub) {
1210 p = section_ptr_add(plt, 20);
1211 put32(p, 0x4778); /* bx pc */
1212 put32(p+2, 0x46c0); /* nop */
1213 p += 4;
1214 } else
1215 p = section_ptr_add(plt, 16);
1216 put32(p, 0xe59fc004); /* ldr ip, [pc, #4] ; GOT entry offset */
1217 put32(p+4, 0xe08fc00c); /* add ip, pc, ip ; addr of GOT entry */
1218 put32(p+8, 0xe59cf000); /* ldr pc, [ip] ; jump to GOT entry */
1219 put32(p+12, s1->got->data_offset); /* GOT entry off once patched */
1221 /* the symbol is modified so that it will be relocated to
1222 the PLT */
1223 if (sym->st_shndx == SHN_UNDEF)
1224 offset = plt->data_offset - 16;
1226 #elif defined(TCC_TARGET_ARM64)
1227 if (need_plt_entry) {
1228 Section *plt;
1229 uint8_t *p;
1231 if (s1->output_type == TCC_OUTPUT_DLL)
1232 tcc_error("DLLs unimplemented!");
1234 plt = s1->plt;
1235 if (plt->data_offset == 0)
1236 section_ptr_add(plt, 32);
1237 symattr->plt_offset = plt->data_offset;
1238 p = section_ptr_add(plt, 16);
1239 put32(p, s1->got->data_offset);
1240 put32(p + 4, (uint64_t)s1->got->data_offset >> 32);
1242 if (sym->st_shndx == SHN_UNDEF)
1243 offset = plt->data_offset - 16;
1245 #elif defined(TCC_TARGET_C67)
1246 if (s1->dynsym) {
1247 tcc_error("C67 got not implemented");
1249 #else
1250 #error unsupported CPU
1251 #endif
1252 if (s1->dynsym) {
1253 /* XXX This might generate multiple syms for name. */
1254 index = put_elf_sym(s1->dynsym, offset,
1255 size, info, 0, sym->st_shndx, name);
1256 /* Create the relocation (it's against the GOT for PLT
1257 and GOT relocs). */
1258 put_elf_reloc(s1->dynsym, s1->got,
1259 s1->got->data_offset,
1260 reloc_type, index);
1261 } else {
1262 /* Without .dynsym (i.e. static link or memory output) we
1263 still need relocs against the generated got, so as to fill
1264 the entries with the symbol values (determined later). */
1265 put_elf_reloc(symtab_section, s1->got,
1266 s1->got->data_offset,
1267 reloc_type, sym_index);
1269 /* And now create the GOT slot itself. */
1270 ptr = section_ptr_add(s1->got, PTR_SIZE);
1271 *ptr = 0;
1272 if (need_plt_entry)
1273 return symattr->plt_offset;
1274 else
1275 return symattr->got_offset;
1278 /* build GOT and PLT entries */
1279 ST_FUNC void build_got_entries(TCCState *s1)
1281 Section *s;
1282 ElfW_Rel *rel;
1283 ElfW(Sym) *sym;
1284 int i, type, reloc_type, sym_index;
1286 for(i = 1; i < s1->nb_sections; i++) {
1287 s = s1->sections[i];
1288 if (s->sh_type != SHT_RELX)
1289 continue;
1290 /* no need to handle got relocations */
1291 if (s->link != symtab_section)
1292 continue;
1293 for_each_elem(s, 0, rel, ElfW_Rel) {
1294 type = ELFW(R_TYPE)(rel->r_info);
1295 switch(type) {
1296 #if defined(TCC_TARGET_I386)
1297 case R_386_GOT32:
1298 case R_386_GOTOFF:
1299 case R_386_GOTPC:
1300 case R_386_PLT32:
1301 if (!s1->got)
1302 build_got(s1);
1303 if (type == R_386_GOT32 || type == R_386_PLT32) {
1304 sym_index = ELFW(R_SYM)(rel->r_info);
1305 sym = &((ElfW(Sym) *)symtab_section->data)[sym_index];
1306 /* look at the symbol got offset. If none, then add one */
1307 if (type == R_386_GOT32)
1308 reloc_type = R_386_GLOB_DAT;
1309 else
1310 reloc_type = R_386_JMP_SLOT;
1311 put_got_entry(s1, reloc_type, sym->st_size, sym->st_info,
1312 sym_index);
1314 break;
1315 #elif defined(TCC_TARGET_ARM)
1316 case R_ARM_PC24:
1317 case R_ARM_CALL:
1318 case R_ARM_JUMP24:
1319 case R_ARM_GOT32:
1320 case R_ARM_GOTOFF:
1321 case R_ARM_GOTPC:
1322 case R_ARM_PLT32:
1323 if (!s1->got)
1324 build_got(s1);
1325 sym_index = ELFW(R_SYM)(rel->r_info);
1326 sym = &((ElfW(Sym) *)symtab_section->data)[sym_index];
1327 if (type != R_ARM_GOTOFF && type != R_ARM_GOTPC
1328 && sym->st_shndx == SHN_UNDEF) {
1329 unsigned long ofs;
1330 /* look at the symbol got offset. If none, then add one */
1331 if (type == R_ARM_GOT32)
1332 reloc_type = R_ARM_GLOB_DAT;
1333 else
1334 reloc_type = R_ARM_JUMP_SLOT;
1335 ofs = put_got_entry(s1, reloc_type, sym->st_size,
1336 sym->st_info, sym_index);
1337 #ifdef DEBUG_RELOC
1338 printf ("maybegot: %s, %d, %d --> ofs=0x%x\n",
1339 (char *) symtab_section->link->data + sym->st_name,
1340 type, sym->st_shndx, ofs);
1341 #endif
1342 if (type != R_ARM_GOT32) {
1343 addr_t *ptr = (addr_t*)(s1->sections[s->sh_info]->data
1344 + rel->r_offset);
1345 /* x must be signed! */
1346 int x = *ptr & 0xffffff;
1347 x = (x << 8) >> 8;
1348 x <<= 2;
1349 x += ofs;
1350 x >>= 2;
1351 #ifdef DEBUG_RELOC
1352 printf ("insn=0x%x --> 0x%x (x==0x%x)\n", *ptr,
1353 (*ptr & 0xff000000) | x, x);
1354 #endif
1355 *ptr = (*ptr & 0xff000000) | x;
1358 break;
1359 case R_ARM_THM_JUMP24:
1360 sym_index = ELFW(R_SYM)(rel->r_info);
1361 sym = &((ElfW(Sym) *)symtab_section->data)[sym_index];
1362 /* We are relocating a jump from thumb code to arm code */
1363 if (sym->st_shndx != SHN_UNDEF && !(sym->st_value & 1)) {
1364 int index;
1365 uint8_t *p;
1366 char *name, buf[1024];
1367 Section *text_section;
1369 name = (char *) symtab_section->link->data + sym->st_name;
1370 text_section = s1->sections[sym->st_shndx];
1371 /* Modify reloc to target a thumb stub to switch to ARM */
1372 snprintf(buf, sizeof(buf), "%s_from_thumb", name);
1373 index = put_elf_sym(symtab_section,
1374 text_section->data_offset + 1,
1375 sym->st_size, sym->st_info, 0,
1376 sym->st_shndx, buf);
1377 rel->r_info = ELFW(R_INFO)(index, type);
1378 /* Create a thumb stub fonction to switch to ARM mode */
1379 put_elf_reloc(symtab_section, text_section,
1380 text_section->data_offset + 4, R_ARM_JUMP24,
1381 sym_index);
1382 p = section_ptr_add(text_section, 8);
1383 put32(p, 0x4778); /* bx pc */
1384 put32(p+2, 0x46c0); /* nop */
1385 put32(p+4, 0xeafffffe); /* b $sym */
1387 #elif defined(TCC_TARGET_ARM64)
1388 //xx Other cases may be required here:
1389 case R_AARCH64_ADR_GOT_PAGE:
1390 case R_AARCH64_LD64_GOT_LO12_NC:
1391 if (!s1->got)
1392 build_got(s1);
1393 sym_index = ELFW(R_SYM)(rel->r_info);
1394 sym = &((ElfW(Sym) *)symtab_section->data)[sym_index];
1395 reloc_type = R_AARCH64_GLOB_DAT;
1396 put_got_entry(s1, reloc_type, sym->st_size, sym->st_info,
1397 sym_index);
1398 break;
1400 case R_AARCH64_JUMP26:
1401 case R_AARCH64_CALL26:
1402 if (!s1->got)
1403 build_got(s1);
1404 sym_index = ELFW(R_SYM)(rel->r_info);
1405 sym = &((ElfW(Sym) *)symtab_section->data)[sym_index];
1406 if (sym->st_shndx == SHN_UNDEF) {
1407 unsigned long ofs;
1408 reloc_type = R_AARCH64_JUMP_SLOT;
1409 ofs = put_got_entry(s1, reloc_type, sym->st_size,
1410 sym->st_info, sym_index);
1411 /* We store the place of the generated PLT slot
1412 in our addend. */
1413 rel->r_addend += ofs;
1415 break;
1416 #elif defined(TCC_TARGET_C67)
1417 case R_C60_GOT32:
1418 case R_C60_GOTOFF:
1419 case R_C60_GOTPC:
1420 case R_C60_PLT32:
1421 if (!s1->got)
1422 build_got(s1);
1423 if (type == R_C60_GOT32 || type == R_C60_PLT32) {
1424 sym_index = ELFW(R_SYM)(rel->r_info);
1425 sym = &((ElfW(Sym) *)symtab_section->data)[sym_index];
1426 /* look at the symbol got offset. If none, then add one */
1427 if (type == R_C60_GOT32)
1428 reloc_type = R_C60_GLOB_DAT;
1429 else
1430 reloc_type = R_C60_JMP_SLOT;
1431 put_got_entry(s1, reloc_type, sym->st_size, sym->st_info,
1432 sym_index);
1434 break;
1435 #elif defined(TCC_TARGET_X86_64)
1436 case R_X86_64_GOT32:
1437 case R_X86_64_GOTTPOFF:
1438 case R_X86_64_GOTPCREL:
1439 case R_X86_64_PLT32:
1440 sym_index = ELFW(R_SYM)(rel->r_info);
1441 sym = &((ElfW(Sym) *)symtab_section->data)[sym_index];
1442 if (type == R_X86_64_PLT32 &&
1443 ELFW(ST_VISIBILITY)(sym->st_other) != STV_DEFAULT)
1445 rel->r_info = ELFW(R_INFO)(sym_index, R_X86_64_PC32);
1446 break;
1449 if (!s1->got)
1450 build_got(s1);
1451 if (type == R_X86_64_GOT32 || type == R_X86_64_GOTPCREL ||
1452 type == R_X86_64_PLT32) {
1453 unsigned long ofs;
1454 /* look at the symbol got offset. If none, then add one */
1455 if (type == R_X86_64_GOT32 || type == R_X86_64_GOTPCREL)
1456 reloc_type = R_X86_64_GLOB_DAT;
1457 else
1458 reloc_type = R_X86_64_JUMP_SLOT;
1459 ofs = put_got_entry(s1, reloc_type, sym->st_size,
1460 sym->st_info, sym_index);
1461 if (type == R_X86_64_PLT32)
1462 /* We store the place of the generated PLT slot
1463 in our addend. */
1464 rel->r_addend += ofs;
1466 break;
1467 #else
1468 #error unsupported CPU
1469 #endif
1470 default:
1471 break;
1477 ST_FUNC Section *new_symtab(TCCState *s1,
1478 const char *symtab_name, int sh_type, int sh_flags,
1479 const char *strtab_name,
1480 const char *hash_name, int hash_sh_flags)
1482 Section *symtab, *strtab, *hash;
1483 int *ptr, nb_buckets;
1485 symtab = new_section(s1, symtab_name, sh_type, sh_flags);
1486 symtab->sh_entsize = sizeof(ElfW(Sym));
1487 strtab = new_section(s1, strtab_name, SHT_STRTAB, sh_flags);
1488 put_elf_str(strtab, "");
1489 symtab->link = strtab;
1490 put_elf_sym(symtab, 0, 0, 0, 0, 0, NULL);
1492 nb_buckets = 1;
1494 hash = new_section(s1, hash_name, SHT_HASH, hash_sh_flags);
1495 hash->sh_entsize = sizeof(int);
1496 symtab->hash = hash;
1497 hash->link = symtab;
1499 ptr = section_ptr_add(hash, (2 + nb_buckets + 1) * sizeof(int));
1500 ptr[0] = nb_buckets;
1501 ptr[1] = 1;
1502 memset(ptr + 2, 0, (nb_buckets + 1) * sizeof(int));
1503 return symtab;
1506 /* put dynamic tag */
1507 static void put_dt(Section *dynamic, int dt, addr_t val)
1509 ElfW(Dyn) *dyn;
1510 dyn = section_ptr_add(dynamic, sizeof(ElfW(Dyn)));
1511 dyn->d_tag = dt;
1512 dyn->d_un.d_val = val;
1515 static void add_init_array_defines(TCCState *s1, const char *section_name)
1517 Section *s;
1518 long end_offset;
1519 char sym_start[1024];
1520 char sym_end[1024];
1522 snprintf(sym_start, sizeof(sym_start), "__%s_start", section_name + 1);
1523 snprintf(sym_end, sizeof(sym_end), "__%s_end", section_name + 1);
1525 s = find_section(s1, section_name);
1526 if (!s) {
1527 end_offset = 0;
1528 s = data_section;
1529 } else {
1530 end_offset = s->data_offset;
1533 add_elf_sym(symtab_section,
1534 0, 0,
1535 ELFW(ST_INFO)(STB_GLOBAL, STT_NOTYPE), 0,
1536 s->sh_num, sym_start);
1537 add_elf_sym(symtab_section,
1538 end_offset, 0,
1539 ELFW(ST_INFO)(STB_GLOBAL, STT_NOTYPE), 0,
1540 s->sh_num, sym_end);
1543 static int tcc_add_support(TCCState *s1, const char *filename)
1545 char buf[1024];
1546 snprintf(buf, sizeof(buf), "%s/%s/%s", s1->tcc_lib_path,
1547 /* an cpu specific path inside tcc_lib_path, mainly for keeping libtcc1.a */
1548 #ifdef TCC_TARGET_I386
1549 "i386"
1550 #endif
1551 #ifdef TCC_TARGET_X86_64
1552 "x86-64"
1553 #endif
1554 #ifdef TCC_TARGET_ARM
1555 "arm"
1556 #endif
1557 #ifdef TCC_TARGET_ARM64
1558 "arm64"
1559 #endif
1560 #ifdef TCC_TARGET_C67
1561 "C67"
1562 #endif
1563 ,filename);
1565 return tcc_add_file(s1, buf, TCC_FILETYPE_BINARY);
1568 ST_FUNC void tcc_add_bcheck(TCCState *s1)
1570 #ifdef CONFIG_TCC_BCHECK
1571 addr_t *ptr;
1573 if (0 == s1->do_bounds_check)
1574 return;
1576 /* XXX: add an object file to do that */
1577 ptr = section_ptr_add(bounds_section, sizeof(*ptr));
1578 *ptr = 0;
1579 add_elf_sym(symtab_section, 0, 0,
1580 ELFW(ST_INFO)(STB_GLOBAL, STT_NOTYPE), 0,
1581 bounds_section->sh_num, "__bounds_start");
1582 if (s1->output_type != TCC_OUTPUT_MEMORY) {
1583 /* add 'call __bound_init()' in .init section */
1585 /* XXX not called on MSYS, reason is unknown. For this
1586 case a call to __bound_init is performed in bcheck.c
1587 when __bound_ptr_add, __bound_new_region,
1588 __bound_delete_region called */
1590 int sym_index = find_elf_sym(symtab_section, "__bound_init");
1591 if (sym_index) {
1592 Section *init_section = find_section(s1, ".init");
1593 unsigned char *pinit = section_ptr_add(init_section, 5);
1594 pinit[0] = 0xe8;
1595 put32(pinit + 1, -4);
1596 put_elf_reloc(symtab_section, init_section,
1597 init_section->data_offset - 4, R_386_PC32, sym_index);
1599 else
1600 tcc_warning("__bound_init not defined");
1602 #endif
1605 /* add tcc runtime libraries */
1606 ST_FUNC void tcc_add_runtime(TCCState *s1)
1608 tcc_add_pragma_libs(s1);
1610 /* add libc */
1611 if (!s1->nostdlib) {
1612 tcc_add_library(s1, "c");
1613 #ifdef CONFIG_USE_LIBGCC
1614 if (!s1->static_link) {
1615 tcc_add_file(s1, TCC_LIBGCC, TCC_FILETYPE_BINARY);
1617 #endif
1618 tcc_add_support(s1, "libtcc1.a");
1621 /* tcc_add_bcheck tries to relocate a call to __bound_init in _init so
1622 libtcc1.a must be loaded before for __bound_init to be defined and
1623 crtn.o must be loaded after to not finalize _init too early. */
1624 tcc_add_bcheck(s1);
1626 if (!s1->nostdlib) {
1627 /* add crt end if not memory output */
1628 if (s1->output_type != TCC_OUTPUT_MEMORY)
1629 tcc_add_crt(s1, "crtn.o");
1633 /* add various standard linker symbols (must be done after the
1634 sections are filled (for example after allocating common
1635 symbols)) */
1636 ST_FUNC void tcc_add_linker_symbols(TCCState *s1)
1638 char buf[1024];
1639 int i;
1640 Section *s;
1642 add_elf_sym(symtab_section,
1643 text_section->data_offset, 0,
1644 ELFW(ST_INFO)(STB_GLOBAL, STT_NOTYPE), 0,
1645 text_section->sh_num, "_etext");
1646 add_elf_sym(symtab_section,
1647 data_section->data_offset, 0,
1648 ELFW(ST_INFO)(STB_GLOBAL, STT_NOTYPE), 0,
1649 data_section->sh_num, "_edata");
1650 add_elf_sym(symtab_section,
1651 bss_section->data_offset, 0,
1652 ELFW(ST_INFO)(STB_GLOBAL, STT_NOTYPE), 0,
1653 bss_section->sh_num, "_end");
1654 /* horrible new standard ldscript defines */
1655 add_init_array_defines(s1, ".preinit_array");
1656 add_init_array_defines(s1, ".init_array");
1657 add_init_array_defines(s1, ".fini_array");
1659 /* add start and stop symbols for sections whose name can be
1660 expressed in C */
1661 for(i = 1; i < s1->nb_sections; i++) {
1662 s = s1->sections[i];
1663 if (s->sh_type == SHT_PROGBITS &&
1664 (s->sh_flags & SHF_ALLOC)) {
1665 const char *p;
1666 int ch;
1668 /* check if section name can be expressed in C */
1669 p = s->name;
1670 for(;;) {
1671 ch = *p;
1672 if (!ch)
1673 break;
1674 if (!isid(ch) && !isnum(ch))
1675 goto next_sec;
1676 p++;
1678 snprintf(buf, sizeof(buf), "__start_%s", s->name);
1679 add_elf_sym(symtab_section,
1680 0, 0,
1681 ELFW(ST_INFO)(STB_GLOBAL, STT_NOTYPE), 0,
1682 s->sh_num, buf);
1683 snprintf(buf, sizeof(buf), "__stop_%s", s->name);
1684 add_elf_sym(symtab_section,
1685 s->data_offset, 0,
1686 ELFW(ST_INFO)(STB_GLOBAL, STT_NOTYPE), 0,
1687 s->sh_num, buf);
1689 next_sec: ;
1693 static void tcc_output_binary(TCCState *s1, FILE *f,
1694 const int *sec_order)
1696 Section *s;
1697 int i, offset, size;
1699 offset = 0;
1700 for(i=1;i<s1->nb_sections;i++) {
1701 s = s1->sections[sec_order[i]];
1702 if (s->sh_type != SHT_NOBITS &&
1703 (s->sh_flags & SHF_ALLOC)) {
1704 while (offset < s->sh_offset) {
1705 fputc(0, f);
1706 offset++;
1708 size = s->sh_size;
1709 fwrite(s->data, 1, size, f);
1710 offset += size;
1715 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
1716 #define HAVE_PHDR 1
1717 #define EXTRA_RELITEMS 14
1719 /* move the relocation value from .dynsym to .got */
1720 void patch_dynsym_undef(TCCState *s1, Section *s)
1722 uint32_t *gotd = (void *)s1->got->data;
1723 ElfW(Sym) *sym;
1725 gotd += 3; /* dummy entries in .got */
1726 /* relocate symbols in .dynsym */
1727 for_each_elem(s, 1, sym, ElfW(Sym)) {
1728 if (sym->st_shndx == SHN_UNDEF) {
1729 *gotd++ = sym->st_value + 6; /* XXX 6 is magic ? */
1730 sym->st_value = 0;
1734 #else
1735 #define HAVE_PHDR 1
1736 #define EXTRA_RELITEMS 9
1738 /* zero plt offsets of weak symbols in .dynsym */
1739 void patch_dynsym_undef(TCCState *s1, Section *s)
1741 ElfW(Sym) *sym;
1743 for_each_elem(s, 1, sym, ElfW(Sym))
1744 if (sym->st_shndx == SHN_UNDEF && ELFW(ST_BIND)(sym->st_info) == STB_WEAK)
1745 sym->st_value = 0;
1747 #endif
1749 ST_FUNC void fill_got_entry(TCCState *s1, ElfW_Rel *rel)
1751 int sym_index = ELFW(R_SYM) (rel->r_info);
1752 ElfW(Sym) *sym = &((ElfW(Sym) *) symtab_section->data)[sym_index];
1753 unsigned long offset;
1755 if (sym_index >= s1->nb_sym_attrs)
1756 return;
1757 offset = s1->sym_attrs[sym_index].got_offset;
1758 section_reserve(s1->got, offset + PTR_SIZE);
1759 #ifdef TCC_TARGET_X86_64
1760 /* only works for x86-64 */
1761 put32(s1->got->data + offset + 4, sym->st_value >> 32);
1762 #endif
1763 put32(s1->got->data + offset, sym->st_value & 0xffffffff);
1766 /* Perform relocation to GOT or PLT entries */
1767 ST_FUNC void fill_got(TCCState *s1)
1769 Section *s;
1770 ElfW_Rel *rel;
1771 int i;
1773 for(i = 1; i < s1->nb_sections; i++) {
1774 s = s1->sections[i];
1775 if (s->sh_type != SHT_RELX)
1776 continue;
1777 /* no need to handle got relocations */
1778 if (s->link != symtab_section)
1779 continue;
1780 for_each_elem(s, 0, rel, ElfW_Rel) {
1781 switch (ELFW(R_TYPE) (rel->r_info)) {
1782 case R_X86_64_GOT32:
1783 case R_X86_64_GOTPCREL:
1784 case R_X86_64_PLT32:
1785 fill_got_entry(s1, rel);
1786 break;
1792 /* Bind symbols of executable: resolve undefined symbols from exported symbols
1793 in shared libraries and export non local defined symbols to shared libraries
1794 if -rdynamic switch was given on command line */
1795 static void bind_exe_dynsyms(TCCState *s1)
1797 const char *name;
1798 int sym_index, index;
1799 ElfW(Sym) *sym, *esym;
1800 int type;
1802 /* Resolve undefined symbols from dynamic symbols. When there is a match:
1803 - if STT_FUNC or STT_GNU_IFUNC symbol -> add it in PLT
1804 - if STT_OBJECT symbol -> add it in .bss section with suitable reloc */
1805 for_each_elem(symtab_section, 1, sym, ElfW(Sym)) {
1806 if (sym->st_shndx == SHN_UNDEF) {
1807 name = (char *) symtab_section->link->data + sym->st_name;
1808 sym_index = find_elf_sym(s1->dynsymtab_section, name);
1809 if (sym_index) {
1810 esym = &((ElfW(Sym) *)s1->dynsymtab_section->data)[sym_index];
1811 type = ELFW(ST_TYPE)(esym->st_info);
1812 if ((type == STT_FUNC) || (type == STT_GNU_IFUNC)) {
1813 /* Indirect functions shall have STT_FUNC type in executable
1814 * dynsym section. Indeed, a dlsym call following a lazy
1815 * resolution would pick the symbol value from the
1816 * executable dynsym entry which would contain the address
1817 * of the function wanted by the caller of dlsym instead of
1818 * the address of the function that would return that
1819 * address */
1820 put_got_entry(s1, R_JMP_SLOT, esym->st_size,
1821 ELFW(ST_INFO)(STB_GLOBAL,STT_FUNC),
1822 sym - (ElfW(Sym) *)symtab_section->data);
1823 } else if (type == STT_OBJECT) {
1824 unsigned long offset;
1825 ElfW(Sym) *dynsym;
1826 offset = bss_section->data_offset;
1827 /* XXX: which alignment ? */
1828 offset = (offset + 16 - 1) & -16;
1829 index = put_elf_sym(s1->dynsym, offset, esym->st_size,
1830 esym->st_info, 0, bss_section->sh_num,
1831 name);
1832 /* Ensure R_COPY works for weak symbol aliases */
1833 if (ELFW(ST_BIND)(esym->st_info) == STB_WEAK) {
1834 for_each_elem(s1->dynsymtab_section, 1, dynsym, ElfW(Sym)) {
1835 if ((dynsym->st_value == esym->st_value)
1836 && (ELFW(ST_BIND)(dynsym->st_info) == STB_GLOBAL)) {
1837 char *dynname = (char *) s1->dynsymtab_section->link->data
1838 + dynsym->st_name;
1839 put_elf_sym(s1->dynsym, offset, dynsym->st_size,
1840 dynsym->st_info, 0,
1841 bss_section->sh_num, dynname);
1842 break;
1846 put_elf_reloc(s1->dynsym, bss_section,
1847 offset, R_COPY, index);
1848 offset += esym->st_size;
1849 bss_section->data_offset = offset;
1851 } else {
1852 /* STB_WEAK undefined symbols are accepted */
1853 /* XXX: _fp_hw seems to be part of the ABI, so we ignore it */
1854 if (ELFW(ST_BIND)(sym->st_info) == STB_WEAK ||
1855 !strcmp(name, "_fp_hw")) {
1856 } else {
1857 tcc_error_noabort("undefined symbol '%s'", name);
1860 } else if (s1->rdynamic && ELFW(ST_BIND)(sym->st_info) != STB_LOCAL) {
1861 /* if -rdynamic option, then export all non local symbols */
1862 name = (char *) symtab_section->link->data + sym->st_name;
1863 put_elf_sym(s1->dynsym, sym->st_value, sym->st_size, sym->st_info,
1864 0, sym->st_shndx, name);
1869 /* Bind symbols of libraries: export non local symbols of executable that
1870 resolve undefined symbols of shared libraries */
1871 static void bind_libs_dynsyms(TCCState *s1)
1873 const char *name;
1874 int sym_index;
1875 ElfW(Sym) *sym, *esym;
1877 /* now look at unresolved dynamic symbols and export
1878 corresponding symbol */
1879 for_each_elem(s1->dynsymtab_section, 1, esym, ElfW(Sym)) {
1880 name = (char *) s1->dynsymtab_section->link->data + esym->st_name;
1881 sym_index = find_elf_sym(symtab_section, name);
1882 if (sym_index) {
1883 /* XXX: avoid adding a symbol if already present because of
1884 -rdynamic ? */
1885 sym = &((ElfW(Sym) *)symtab_section->data)[sym_index];
1886 if (sym->st_shndx != SHN_UNDEF)
1887 put_elf_sym(s1->dynsym, sym->st_value, sym->st_size,
1888 sym->st_info, 0, sym->st_shndx, name);
1889 } else if (esym->st_shndx == SHN_UNDEF) {
1890 /* weak symbols can stay undefined */
1891 if (ELFW(ST_BIND)(esym->st_info) != STB_WEAK)
1892 tcc_warning("undefined dynamic symbol '%s'", name);
1897 /* Export all non local symbols (for shared libraries) */
1898 static void export_global_syms(TCCState *s1)
1900 int nb_syms, dynindex, index;
1901 const char *name;
1902 ElfW(Sym) *sym;
1904 nb_syms = symtab_section->data_offset / sizeof(ElfW(Sym));
1905 s1->symtab_to_dynsym = tcc_mallocz(sizeof(int) * nb_syms);
1906 for_each_elem(symtab_section, 1, sym, ElfW(Sym)) {
1907 if (ELFW(ST_BIND)(sym->st_info) != STB_LOCAL) {
1908 name = (char *) symtab_section->link->data + sym->st_name;
1909 dynindex = put_elf_sym(s1->dynsym, sym->st_value, sym->st_size,
1910 sym->st_info, 0, sym->st_shndx, name);
1911 index = sym - (ElfW(Sym) *) symtab_section->data;
1912 s1->symtab_to_dynsym[index] = dynindex;
1917 /* relocate the PLT: compute addresses and offsets in the PLT now that final
1918 address for PLT and GOT are known (see fill_program_header) */
1919 ST_FUNC void relocate_plt(TCCState *s1)
1921 uint8_t *p, *p_end;
1923 if (!s1->plt)
1924 return;
1926 p = s1->plt->data;
1927 p_end = p + s1->plt->data_offset;
1928 if (p < p_end) {
1929 #if defined(TCC_TARGET_I386)
1930 put32(p + 2, get32(p + 2) + s1->got->sh_addr);
1931 put32(p + 8, get32(p + 8) + s1->got->sh_addr);
1932 p += 16;
1933 while (p < p_end) {
1934 put32(p + 2, get32(p + 2) + s1->got->sh_addr);
1935 p += 16;
1937 #elif defined(TCC_TARGET_X86_64)
1938 int x = s1->got->sh_addr - s1->plt->sh_addr - 6;
1939 put32(p + 2, get32(p + 2) + x);
1940 put32(p + 8, get32(p + 8) + x - 6);
1941 p += 16;
1942 while (p < p_end) {
1943 put32(p + 2, get32(p + 2) + x + s1->plt->data - p);
1944 p += 16;
1946 #elif defined(TCC_TARGET_ARM)
1947 int x;
1948 x=s1->got->sh_addr - s1->plt->sh_addr - 12;
1949 p += 16;
1950 while (p < p_end) {
1951 if (get32(p) == 0x46c04778) /* PLT Thumb stub present */
1952 p += 4;
1953 put32(p + 12, x + get32(p + 12) + s1->plt->data - p);
1954 p += 16;
1956 #elif defined(TCC_TARGET_ARM64)
1957 uint64_t plt = s1->plt->sh_addr;
1958 uint64_t got = s1->got->sh_addr;
1959 uint64_t off = (got >> 12) - (plt >> 12);
1960 if ((off + ((uint32_t)1 << 20)) >> 21)
1961 tcc_error("Failed relocating PLT (off=0x%lx, got=0x%lx, plt=0x%lx)", off, got, plt);
1962 put32(p, 0xa9bf7bf0); // stp x16,x30,[sp,#-16]!
1963 put32(p + 4, (0x90000010 | // adrp x16,...
1964 (off & 0x1ffffc) << 3 | (off & 3) << 29));
1965 put32(p + 8, (0xf9400211 | // ldr x17,[x16,#...]
1966 (got & 0xff8) << 7));
1967 put32(p + 12, (0x91000210 | // add x16,x16,#...
1968 (got & 0xfff) << 10));
1969 put32(p + 16, 0xd61f0220); // br x17
1970 put32(p + 20, 0xd503201f); // nop
1971 put32(p + 24, 0xd503201f); // nop
1972 put32(p + 28, 0xd503201f); // nop
1973 p += 32;
1974 while (p < p_end) {
1975 uint64_t pc = plt + (p - s1->plt->data);
1976 uint64_t addr = got +
1977 (get32(p) | (uint64_t)get32(p + 4) << 32);
1978 uint32_t off = (addr >> 12) - (pc >> 12);
1979 if ((off + ((uint32_t)1 << 20)) >> 21)
1980 tcc_error("Failed relocating PLT (off=0x%lx, addr=0x%lx, pc=0x%lx)", off, addr, pc);
1981 put32(p, (0x90000010 | // adrp x16,...
1982 (off & 0x1ffffc) << 3 | (off & 3) << 29));
1983 put32(p + 4, (0xf9400211 | // ldr x17,[x16,#...]
1984 (addr & 0xff8) << 7));
1985 put32(p + 8, (0x91000210 | // add x16,x16,#...
1986 (addr & 0xfff) << 10));
1987 put32(p + 12, 0xd61f0220); // br x17
1988 p += 16;
1990 #elif defined(TCC_TARGET_C67)
1991 /* XXX: TODO */
1992 #else
1993 #error unsupported CPU
1994 #endif
1998 /* Allocate strings for section names and decide if an unallocated section
1999 should be output.
2001 NOTE: the strsec section comes last, so its size is also correct ! */
2002 static void alloc_sec_names(TCCState *s1, int file_type, Section *strsec)
2004 int i;
2005 Section *s;
2007 /* Allocate strings for section names */
2008 for(i = 1; i < s1->nb_sections; i++) {
2009 s = s1->sections[i];
2010 s->sh_name = put_elf_str(strsec, s->name);
2011 /* when generating a DLL, we include relocations but we may
2012 patch them */
2013 if (file_type == TCC_OUTPUT_DLL &&
2014 s->sh_type == SHT_RELX &&
2015 !(s->sh_flags & SHF_ALLOC)) {
2016 /* gr: avoid bogus relocs for empty (debug) sections */
2017 if (s1->sections[s->sh_info]->sh_flags & SHF_ALLOC)
2018 prepare_dynamic_rel(s1, s);
2019 else if (s1->do_debug)
2020 s->sh_size = s->data_offset;
2021 } else if (s1->do_debug ||
2022 file_type == TCC_OUTPUT_OBJ ||
2023 file_type == TCC_OUTPUT_EXE ||
2024 (s->sh_flags & SHF_ALLOC) ||
2025 i == (s1->nb_sections - 1)) {
2026 /* we output all sections if debug or object file */
2027 s->sh_size = s->data_offset;
2032 /* Info to be copied in dynamic section */
2033 struct dyn_inf {
2034 Section *dynamic;
2035 Section *dynstr;
2036 unsigned long dyn_rel_off;
2037 addr_t rel_addr;
2038 addr_t rel_size;
2039 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
2040 addr_t bss_addr;
2041 addr_t bss_size;
2042 #endif
2043 };
2045 /* Assign sections to segments and decide how are sections laid out when loaded
2046 in memory. This function also fills corresponding program headers. */
2047 static int layout_sections(TCCState *s1, ElfW(Phdr) *phdr, int phnum,
2048 Section *interp, Section* strsec,
2049 struct dyn_inf *dyninf, int *sec_order)
2051 int i, j, k, file_type, sh_order_index, file_offset;
2052 unsigned long s_align;
2053 long long tmp;
2054 addr_t addr;
2055 ElfW(Phdr) *ph;
2056 Section *s;
2058 file_type = s1->output_type;
2059 sh_order_index = 1;
2060 file_offset = 0;
2061 if (s1->output_format == TCC_OUTPUT_FORMAT_ELF)
2062 file_offset = sizeof(ElfW(Ehdr)) + phnum * sizeof(ElfW(Phdr));
2063 s_align = ELF_PAGE_SIZE;
2064 if (s1->section_align)
2065 s_align = s1->section_align;
2067 if (phnum > 0) {
2068 if (s1->has_text_addr) {
2069 int a_offset, p_offset;
2070 addr = s1->text_addr;
2071 /* we ensure that (addr % ELF_PAGE_SIZE) == file_offset %
2072 ELF_PAGE_SIZE */
2073 a_offset = (int) (addr & (s_align - 1));
2074 p_offset = file_offset & (s_align - 1);
2075 if (a_offset < p_offset)
2076 a_offset += s_align;
2077 file_offset += (a_offset - p_offset);
2078 } else {
2079 if (file_type == TCC_OUTPUT_DLL)
2080 addr = 0;
2081 else
2082 addr = ELF_START_ADDR;
2083 /* compute address after headers */
2084 addr += (file_offset & (s_align - 1));
2087 ph = &phdr[0];
2088 /* Leave one program headers for the program interpreter and one for
2089 the program header table itself if needed. These are done later as
2090 they require section layout to be done first. */
2091 if (interp)
2092 ph += 1 + HAVE_PHDR;
2094 /* dynamic relocation table information, for .dynamic section */
2095 dyninf->rel_addr = dyninf->rel_size = 0;
2096 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
2097 dyninf->bss_addr = dyninf->bss_size = 0;
2098 #endif
2100 for(j = 0; j < 2; j++) {
2101 ph->p_type = PT_LOAD;
2102 if (j == 0)
2103 ph->p_flags = PF_R | PF_X;
2104 else
2105 ph->p_flags = PF_R | PF_W;
2106 ph->p_align = s_align;
2108 /* Decide the layout of sections loaded in memory. This must
2109 be done before program headers are filled since they contain
2110 info about the layout. We do the following ordering: interp,
2111 symbol tables, relocations, progbits, nobits */
2112 /* XXX: do faster and simpler sorting */
2113 for(k = 0; k < 5; k++) {
2114 for(i = 1; i < s1->nb_sections; i++) {
2115 s = s1->sections[i];
2116 /* compute if section should be included */
2117 if (j == 0) {
2118 if ((s->sh_flags & (SHF_ALLOC | SHF_WRITE)) !=
2119 SHF_ALLOC)
2120 continue;
2121 } else {
2122 if ((s->sh_flags & (SHF_ALLOC | SHF_WRITE)) !=
2123 (SHF_ALLOC | SHF_WRITE))
2124 continue;
2126 if (s == interp) {
2127 if (k != 0)
2128 continue;
2129 } else if (s->sh_type == SHT_DYNSYM ||
2130 s->sh_type == SHT_STRTAB ||
2131 s->sh_type == SHT_HASH) {
2132 if (k != 1)
2133 continue;
2134 } else if (s->sh_type == SHT_RELX) {
2135 if (k != 2)
2136 continue;
2137 } else if (s->sh_type == SHT_NOBITS) {
2138 if (k != 4)
2139 continue;
2140 } else {
2141 if (k != 3)
2142 continue;
2144 sec_order[sh_order_index++] = i;
2146 /* section matches: we align it and add its size */
2147 tmp = addr;
2148 addr = (addr + s->sh_addralign - 1) &
2149 ~(s->sh_addralign - 1);
2150 file_offset += (int) ( addr - tmp );
2151 s->sh_offset = file_offset;
2152 s->sh_addr = addr;
2154 /* update program header infos */
2155 if (ph->p_offset == 0) {
2156 ph->p_offset = file_offset;
2157 ph->p_vaddr = addr;
2158 ph->p_paddr = ph->p_vaddr;
2160 /* update dynamic relocation infos */
2161 if (s->sh_type == SHT_RELX) {
2162 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
2163 if (!strcmp(strsec->data + s->sh_name, ".rel.got")) {
2164 dyninf->rel_addr = addr;
2165 dyninf->rel_size += s->sh_size; /* XXX only first rel. */
2167 if (!strcmp(strsec->data + s->sh_name, ".rel.bss")) {
2168 dyninf->bss_addr = addr;
2169 dyninf->bss_size = s->sh_size; /* XXX only first rel. */
2171 #else
2172 if (dyninf->rel_size == 0)
2173 dyninf->rel_addr = addr;
2174 dyninf->rel_size += s->sh_size;
2175 #endif
2177 addr += s->sh_size;
2178 if (s->sh_type != SHT_NOBITS)
2179 file_offset += s->sh_size;
2182 if (j == 0) {
2183 /* Make the first PT_LOAD segment include the program
2184 headers itself (and the ELF header as well), it'll
2185 come out with same memory use but will make various
2186 tools like binutils strip work better. */
2187 ph->p_offset &= ~(ph->p_align - 1);
2188 ph->p_vaddr &= ~(ph->p_align - 1);
2189 ph->p_paddr &= ~(ph->p_align - 1);
2191 ph->p_filesz = file_offset - ph->p_offset;
2192 ph->p_memsz = addr - ph->p_vaddr;
2193 ph++;
2194 if (j == 0) {
2195 if (s1->output_format == TCC_OUTPUT_FORMAT_ELF) {
2196 /* if in the middle of a page, we duplicate the page in
2197 memory so that one copy is RX and the other is RW */
2198 if ((addr & (s_align - 1)) != 0)
2199 addr += s_align;
2200 } else {
2201 addr = (addr + s_align - 1) & ~(s_align - 1);
2202 file_offset = (file_offset + s_align - 1) & ~(s_align - 1);
2208 /* all other sections come after */
2209 for(i = 1; i < s1->nb_sections; i++) {
2210 s = s1->sections[i];
2211 if (phnum > 0 && (s->sh_flags & SHF_ALLOC))
2212 continue;
2213 sec_order[sh_order_index++] = i;
2215 file_offset = (file_offset + s->sh_addralign - 1) &
2216 ~(s->sh_addralign - 1);
2217 s->sh_offset = file_offset;
2218 if (s->sh_type != SHT_NOBITS)
2219 file_offset += s->sh_size;
2222 return file_offset;
2225 static void fill_unloadable_phdr(ElfW(Phdr) *phdr, int phnum, Section *interp,
2226 Section *dynamic)
2228 ElfW(Phdr) *ph;
2230 /* if interpreter, then add corresponding program header */
2231 if (interp) {
2232 ph = &phdr[0];
2234 if (HAVE_PHDR)
2236 int len = phnum * sizeof(ElfW(Phdr));
2238 ph->p_type = PT_PHDR;
2239 ph->p_offset = sizeof(ElfW(Ehdr));
2240 ph->p_vaddr = interp->sh_addr - len;
2241 ph->p_paddr = ph->p_vaddr;
2242 ph->p_filesz = ph->p_memsz = len;
2243 ph->p_flags = PF_R | PF_X;
2244 ph->p_align = 4; /* interp->sh_addralign; */
2245 ph++;
2248 ph->p_type = PT_INTERP;
2249 ph->p_offset = interp->sh_offset;
2250 ph->p_vaddr = interp->sh_addr;
2251 ph->p_paddr = ph->p_vaddr;
2252 ph->p_filesz = interp->sh_size;
2253 ph->p_memsz = interp->sh_size;
2254 ph->p_flags = PF_R;
2255 ph->p_align = interp->sh_addralign;
2258 /* if dynamic section, then add corresponding program header */
2259 if (dynamic) {
2260 ph = &phdr[phnum - 1];
2262 ph->p_type = PT_DYNAMIC;
2263 ph->p_offset = dynamic->sh_offset;
2264 ph->p_vaddr = dynamic->sh_addr;
2265 ph->p_paddr = ph->p_vaddr;
2266 ph->p_filesz = dynamic->sh_size;
2267 ph->p_memsz = dynamic->sh_size;
2268 ph->p_flags = PF_R | PF_W;
2269 ph->p_align = dynamic->sh_addralign;
2273 /* Fill the dynamic section with tags describing the address and size of
2274 sections */
2275 static void fill_dynamic(TCCState *s1, struct dyn_inf *dyninf)
2277 Section *dynamic;
2279 dynamic = dyninf->dynamic;
2281 /* put dynamic section entries */
2282 dynamic->data_offset = dyninf->dyn_rel_off;
2283 put_dt(dynamic, DT_HASH, s1->dynsym->hash->sh_addr);
2284 put_dt(dynamic, DT_STRTAB, dyninf->dynstr->sh_addr);
2285 put_dt(dynamic, DT_SYMTAB, s1->dynsym->sh_addr);
2286 put_dt(dynamic, DT_STRSZ, dyninf->dynstr->data_offset);
2287 put_dt(dynamic, DT_SYMENT, sizeof(ElfW(Sym)));
2288 #if defined(TCC_TARGET_ARM64) || defined(TCC_TARGET_X86_64)
2289 put_dt(dynamic, DT_RELA, dyninf->rel_addr);
2290 put_dt(dynamic, DT_RELASZ, dyninf->rel_size);
2291 put_dt(dynamic, DT_RELAENT, sizeof(ElfW_Rel));
2292 #else
2293 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
2294 put_dt(dynamic, DT_PLTGOT, s1->got->sh_addr);
2295 put_dt(dynamic, DT_PLTRELSZ, dyninf->rel_size);
2296 put_dt(dynamic, DT_JMPREL, dyninf->rel_addr);
2297 put_dt(dynamic, DT_PLTREL, DT_REL);
2298 put_dt(dynamic, DT_REL, dyninf->bss_addr);
2299 put_dt(dynamic, DT_RELSZ, dyninf->bss_size);
2300 #else
2301 put_dt(dynamic, DT_REL, dyninf->rel_addr);
2302 put_dt(dynamic, DT_RELSZ, dyninf->rel_size);
2303 put_dt(dynamic, DT_RELENT, sizeof(ElfW_Rel));
2304 #endif
2305 #endif
2306 if (s1->do_debug)
2307 put_dt(dynamic, DT_DEBUG, 0);
2308 put_dt(dynamic, DT_NULL, 0);
2311 /* Relocate remaining sections and symbols (that is those not related to
2312 dynamic linking) */
2313 static int final_sections_reloc(TCCState *s1)
2315 int i;
2316 Section *s;
2318 relocate_syms(s1, 0);
2320 if (s1->nb_errors != 0)
2321 return -1;
2323 /* relocate sections */
2324 /* XXX: ignore sections with allocated relocations ? */
2325 for(i = 1; i < s1->nb_sections; i++) {
2326 s = s1->sections[i];
2327 #ifdef TCC_TARGET_I386
2328 if (s->reloc && s != s1->got && (s->sh_flags & SHF_ALLOC)) //gr
2329 /* On X86 gdb 7.3 works in any case but gdb 6.6 will crash if SHF_ALLOC
2330 checking is removed */
2331 #else
2332 if (s->reloc && s != s1->got)
2333 /* On X86_64 gdb 7.3 will crash if SHF_ALLOC checking is present */
2334 #endif
2335 relocate_section(s1, s);
2338 /* relocate relocation entries if the relocation tables are
2339 allocated in the executable */
2340 for(i = 1; i < s1->nb_sections; i++) {
2341 s = s1->sections[i];
2342 if ((s->sh_flags & SHF_ALLOC) &&
2343 s->sh_type == SHT_RELX) {
2344 relocate_rel(s1, s);
2347 return 0;
2350 /* Create an ELF file on disk.
2351 This function handle ELF specific layout requirements */
2352 static void tcc_output_elf(TCCState *s1, FILE *f, int phnum, ElfW(Phdr) *phdr,
2353 int file_offset, int *sec_order)
2355 int i, shnum, offset, size, file_type;
2356 Section *s;
2357 ElfW(Ehdr) ehdr;
2358 ElfW(Shdr) shdr, *sh;
2360 file_type = s1->output_type;
2361 shnum = s1->nb_sections;
2363 memset(&ehdr, 0, sizeof(ehdr));
2365 if (phnum > 0) {
2366 ehdr.e_phentsize = sizeof(ElfW(Phdr));
2367 ehdr.e_phnum = phnum;
2368 ehdr.e_phoff = sizeof(ElfW(Ehdr));
2371 /* align to 4 */
2372 file_offset = (file_offset + 3) & -4;
2374 /* fill header */
2375 ehdr.e_ident[0] = ELFMAG0;
2376 ehdr.e_ident[1] = ELFMAG1;
2377 ehdr.e_ident[2] = ELFMAG2;
2378 ehdr.e_ident[3] = ELFMAG3;
2379 ehdr.e_ident[4] = ELFCLASSW;
2380 ehdr.e_ident[5] = ELFDATA2LSB;
2381 ehdr.e_ident[6] = EV_CURRENT;
2382 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
2383 ehdr.e_ident[EI_OSABI] = ELFOSABI_FREEBSD;
2384 #endif
2385 #ifdef TCC_TARGET_ARM
2386 #ifdef TCC_ARM_EABI
2387 ehdr.e_ident[EI_OSABI] = 0;
2388 ehdr.e_flags = EF_ARM_EABI_VER4;
2389 if (file_type == TCC_OUTPUT_EXE || file_type == TCC_OUTPUT_DLL)
2390 ehdr.e_flags |= EF_ARM_HASENTRY;
2391 if (s1->float_abi == ARM_HARD_FLOAT)
2392 ehdr.e_flags |= EF_ARM_VFP_FLOAT;
2393 else
2394 ehdr.e_flags |= EF_ARM_SOFT_FLOAT;
2395 #else
2396 ehdr.e_ident[EI_OSABI] = ELFOSABI_ARM;
2397 #endif
2398 #endif
2399 switch(file_type) {
2400 default:
2401 case TCC_OUTPUT_EXE:
2402 ehdr.e_type = ET_EXEC;
2403 ehdr.e_entry = get_elf_sym_addr(s1, "_start", 1);
2404 break;
2405 case TCC_OUTPUT_DLL:
2406 ehdr.e_type = ET_DYN;
2407 ehdr.e_entry = text_section->sh_addr; /* XXX: is it correct ? */
2408 break;
2409 case TCC_OUTPUT_OBJ:
2410 ehdr.e_type = ET_REL;
2411 break;
2413 ehdr.e_machine = EM_TCC_TARGET;
2414 ehdr.e_version = EV_CURRENT;
2415 ehdr.e_shoff = file_offset;
2416 ehdr.e_ehsize = sizeof(ElfW(Ehdr));
2417 ehdr.e_shentsize = sizeof(ElfW(Shdr));
2418 ehdr.e_shnum = shnum;
2419 ehdr.e_shstrndx = shnum - 1;
2421 fwrite(&ehdr, 1, sizeof(ElfW(Ehdr)), f);
2422 fwrite(phdr, 1, phnum * sizeof(ElfW(Phdr)), f);
2423 offset = sizeof(ElfW(Ehdr)) + phnum * sizeof(ElfW(Phdr));
2425 sort_syms(s1, symtab_section);
2426 for(i = 1; i < s1->nb_sections; i++) {
2427 s = s1->sections[sec_order[i]];
2428 if (s->sh_type != SHT_NOBITS) {
2429 if (s->sh_type == SHT_DYNSYM)
2430 patch_dynsym_undef(s1, s);
2431 while (offset < s->sh_offset) {
2432 fputc(0, f);
2433 offset++;
2435 size = s->sh_size;
2436 fwrite(s->data, 1, size, f);
2437 offset += size;
2441 /* output section headers */
2442 while (offset < ehdr.e_shoff) {
2443 fputc(0, f);
2444 offset++;
2447 for(i = 0; i < s1->nb_sections; i++) {
2448 sh = &shdr;
2449 memset(sh, 0, sizeof(ElfW(Shdr)));
2450 s = s1->sections[i];
2451 if (s) {
2452 sh->sh_name = s->sh_name;
2453 sh->sh_type = s->sh_type;
2454 sh->sh_flags = s->sh_flags;
2455 sh->sh_entsize = s->sh_entsize;
2456 sh->sh_info = s->sh_info;
2457 if (s->link)
2458 sh->sh_link = s->link->sh_num;
2459 sh->sh_addralign = s->sh_addralign;
2460 sh->sh_addr = s->sh_addr;
2461 sh->sh_offset = s->sh_offset;
2462 sh->sh_size = s->sh_size;
2464 fwrite(sh, 1, sizeof(ElfW(Shdr)), f);
2468 /* Write an elf, coff or "binary" file */
2469 static int tcc_write_elf_file(TCCState *s1, const char *filename, int phnum,
2470 ElfW(Phdr) *phdr, int file_offset, int *sec_order)
2472 int fd, mode, file_type;
2473 FILE *f;
2475 file_type = s1->output_type;
2476 if (file_type == TCC_OUTPUT_OBJ)
2477 mode = 0666;
2478 else
2479 mode = 0777;
2480 unlink(filename);
2481 fd = open(filename, O_WRONLY | O_CREAT | O_TRUNC | O_BINARY, mode);
2482 if (fd < 0) {
2483 tcc_error_noabort("could not write '%s'", filename);
2484 return -1;
2486 f = fdopen(fd, "wb");
2487 if (s1->verbose)
2488 printf("<- %s\n", filename);
2490 #ifdef TCC_TARGET_COFF
2491 if (s1->output_format == TCC_OUTPUT_FORMAT_COFF)
2492 tcc_output_coff(s1, f);
2493 else
2494 #endif
2495 if (s1->output_format == TCC_OUTPUT_FORMAT_ELF)
2496 tcc_output_elf(s1, f, phnum, phdr, file_offset, sec_order);
2497 else
2498 tcc_output_binary(s1, f, sec_order);
2499 fclose(f);
2501 return 0;
2504 /* Output an elf, coff or binary file */
2505 /* XXX: suppress unneeded sections */
2506 static int elf_output_file(TCCState *s1, const char *filename)
2508 int i, ret, phnum, shnum, file_type, file_offset, *sec_order;
2509 struct dyn_inf dyninf;
2510 ElfW(Phdr) *phdr;
2511 ElfW(Sym) *sym;
2512 Section *strsec, *interp, *dynamic, *dynstr;
2514 file_type = s1->output_type;
2515 s1->nb_errors = 0;
2517 /* if linking, also link in runtime libraries (libc, libgcc, etc.) */
2518 if (file_type != TCC_OUTPUT_OBJ) {
2519 tcc_add_runtime(s1);
2522 phdr = NULL;
2523 sec_order = NULL;
2524 interp = dynamic = dynstr = NULL; /* avoid warning */
2525 dyninf.dyn_rel_off = 0; /* avoid warning */
2527 if (file_type != TCC_OUTPUT_OBJ) {
2528 relocate_common_syms();
2530 tcc_add_linker_symbols(s1);
2532 if (!s1->static_link) {
2533 if (file_type == TCC_OUTPUT_EXE) {
2534 char *ptr;
2535 /* allow override the dynamic loader */
2536 const char *elfint = getenv("LD_SO");
2537 if (elfint == NULL)
2538 elfint = DEFAULT_ELFINTERP(s1);
2539 /* add interpreter section only if executable */
2540 interp = new_section(s1, ".interp", SHT_PROGBITS, SHF_ALLOC);
2541 interp->sh_addralign = 1;
2542 ptr = section_ptr_add(interp, 1 + strlen(elfint));
2543 strcpy(ptr, elfint);
2546 /* add dynamic symbol table */
2547 s1->dynsym = new_symtab(s1, ".dynsym", SHT_DYNSYM, SHF_ALLOC,
2548 ".dynstr",
2549 ".hash", SHF_ALLOC);
2550 dynstr = s1->dynsym->link;
2552 /* add dynamic section */
2553 dynamic = new_section(s1, ".dynamic", SHT_DYNAMIC,
2554 SHF_ALLOC | SHF_WRITE);
2555 dynamic->link = dynstr;
2556 dynamic->sh_entsize = sizeof(ElfW(Dyn));
2558 build_got(s1);
2560 if (file_type == TCC_OUTPUT_EXE) {
2561 bind_exe_dynsyms(s1);
2563 if (s1->nb_errors) {
2564 ret = -1;
2565 goto the_end;
2568 bind_libs_dynsyms(s1);
2569 } else /* shared library case: simply export all global symbols */
2570 export_global_syms(s1);
2572 build_got_entries(s1);
2574 /* add a list of needed dlls */
2575 for(i = 0; i < s1->nb_loaded_dlls; i++) {
2576 DLLReference *dllref = s1->loaded_dlls[i];
2577 if (dllref->level == 0)
2578 put_dt(dynamic, DT_NEEDED, put_elf_str(dynstr, dllref->name));
2581 if (s1->rpath)
2582 put_dt(dynamic, DT_RPATH, put_elf_str(dynstr, s1->rpath));
2584 /* XXX: currently, since we do not handle PIC code, we
2585 must relocate the readonly segments */
2586 if (file_type == TCC_OUTPUT_DLL) {
2587 if (s1->soname)
2588 put_dt(dynamic, DT_SONAME, put_elf_str(dynstr, s1->soname));
2589 put_dt(dynamic, DT_TEXTREL, 0);
2592 if (s1->symbolic)
2593 put_dt(dynamic, DT_SYMBOLIC, 0);
2595 /* add necessary space for other entries */
2596 dyninf.dyn_rel_off = dynamic->data_offset;
2597 dynamic->data_offset += sizeof(ElfW(Dyn)) * EXTRA_RELITEMS;
2598 } else {
2599 /* still need to build got entries in case of static link */
2600 build_got_entries(s1);
2604 /* we add a section for symbols */
2605 strsec = new_section(s1, ".shstrtab", SHT_STRTAB, 0);
2606 put_elf_str(strsec, "");
2608 /* compute number of sections */
2609 shnum = s1->nb_sections;
2611 /* this array is used to reorder sections in the output file */
2612 sec_order = tcc_malloc(sizeof(int) * shnum);
2613 sec_order[0] = 0;
2615 /* compute number of program headers */
2616 switch(file_type) {
2617 default:
2618 case TCC_OUTPUT_OBJ:
2619 phnum = 0;
2620 break;
2621 case TCC_OUTPUT_EXE:
2622 if (!s1->static_link)
2623 phnum = 4 + HAVE_PHDR;
2624 else
2625 phnum = 2;
2626 break;
2627 case TCC_OUTPUT_DLL:
2628 phnum = 3;
2629 break;
2632 /* Allocate strings for section names */
2633 alloc_sec_names(s1, file_type, strsec);
2635 /* allocate program segment headers */
2636 phdr = tcc_mallocz(phnum * sizeof(ElfW(Phdr)));
2638 /* compute section to program header mapping */
2639 file_offset = layout_sections(s1, phdr, phnum, interp, strsec, &dyninf,
2640 sec_order);
2642 /* Fill remaining program header and finalize relocation related to dynamic
2643 linking. */
2644 if (phnum > 0) {
2645 fill_unloadable_phdr(phdr, phnum, interp, dynamic);
2646 if (dynamic) {
2647 dyninf.dynamic = dynamic;
2648 dyninf.dynstr = dynstr;
2650 fill_dynamic(s1, &dyninf);
2652 /* put in GOT the dynamic section address and relocate PLT */
2653 put32(s1->got->data, dynamic->sh_addr);
2654 if (file_type == TCC_OUTPUT_EXE
2655 #if defined(TCC_OUTPUT_DLL_WITH_PLT)
2656 || file_type == TCC_OUTPUT_DLL
2657 #endif
2659 relocate_plt(s1);
2661 /* relocate symbols in .dynsym now that final addresses are known */
2662 for_each_elem(s1->dynsym, 1, sym, ElfW(Sym)) {
2663 if (sym->st_shndx == SHN_UNDEF) {
2664 /* relocate to PLT if symbol corresponds to a PLT entry,
2665 but not if it's a weak symbol */
2666 if (ELFW(ST_BIND)(sym->st_info) == STB_WEAK)
2667 sym->st_value = 0;
2668 else if (sym->st_value)
2669 sym->st_value += s1->plt->sh_addr;
2670 } else if (sym->st_shndx < SHN_LORESERVE) {
2671 /* do symbol relocation */
2672 sym->st_value += s1->sections[sym->st_shndx]->sh_addr;
2678 /* if building executable or DLL, then relocate each section
2679 except the GOT which is already relocated */
2680 if (file_type != TCC_OUTPUT_OBJ) {
2681 ret = final_sections_reloc(s1);
2682 if (ret)
2683 goto the_end;
2686 /* Perform relocation to GOT or PLT entries */
2687 if (file_type == TCC_OUTPUT_EXE && s1->static_link)
2688 fill_got(s1);
2690 /* Create the ELF file with name 'filename' */
2691 ret = tcc_write_elf_file(s1, filename, phnum, phdr, file_offset, sec_order);
2692 if (s1->do_strip) {
2693 int rc;
2694 const char *strip_cmd = "sstrip "; // super strip utility from ELFkickers
2695 const char *null_dev = " 2> /dev/null";
2696 char buf[1050];
2697 snprintf(buf, sizeof(buf), "%s%s%s", strip_cmd, filename, null_dev);
2698 rc = system(buf);
2699 if (rc)
2700 system(buf+1); // call a strip utility from binutils
2702 the_end:
2703 tcc_free(s1->symtab_to_dynsym);
2704 tcc_free(sec_order);
2705 tcc_free(phdr);
2706 tcc_free(s1->sym_attrs);
2707 s1->sym_attrs = NULL;
2708 return ret;
2711 LIBTCCAPI int tcc_output_file(TCCState *s, const char *filename)
2713 int ret;
2714 #ifdef TCC_TARGET_PE
2715 if (s->output_type != TCC_OUTPUT_OBJ) {
2716 ret = pe_output_file(s, filename);
2717 } else
2718 #endif
2719 ret = elf_output_file(s, filename);
2720 return ret;
2723 static void *load_data(int fd, unsigned long file_offset, unsigned long size)
2725 void *data;
2727 data = tcc_malloc(size);
2728 lseek(fd, file_offset, SEEK_SET);
2729 read(fd, data, size);
2730 return data;
2733 typedef struct SectionMergeInfo {
2734 Section *s; /* corresponding existing section */
2735 unsigned long offset; /* offset of the new section in the existing section */
2736 uint8_t new_section; /* true if section 's' was added */
2737 uint8_t link_once; /* true if link once section */
2738 } SectionMergeInfo;
2740 /* load an object file and merge it with current files */
2741 /* XXX: handle correctly stab (debug) info */
2742 ST_FUNC int tcc_load_object_file(TCCState *s1,
2743 int fd, unsigned long file_offset)
2745 ElfW(Ehdr) ehdr;
2746 ElfW(Shdr) *shdr, *sh;
2747 int size, i, j, offset, offseti, nb_syms, sym_index, ret;
2748 unsigned char *strsec, *strtab;
2749 int *old_to_new_syms;
2750 char *sh_name, *name;
2751 SectionMergeInfo *sm_table, *sm;
2752 ElfW(Sym) *sym, *symtab;
2753 ElfW_Rel *rel;
2754 Section *s;
2756 int stab_index;
2757 int stabstr_index;
2759 stab_index = stabstr_index = 0;
2761 if (read(fd, &ehdr, sizeof(ehdr)) != sizeof(ehdr))
2762 goto fail1;
2763 if (ehdr.e_ident[0] != ELFMAG0 ||
2764 ehdr.e_ident[1] != ELFMAG1 ||
2765 ehdr.e_ident[2] != ELFMAG2 ||
2766 ehdr.e_ident[3] != ELFMAG3)
2767 goto fail1;
2768 /* test if object file */
2769 if (ehdr.e_type != ET_REL)
2770 goto fail1;
2771 /* test CPU specific stuff */
2772 if (ehdr.e_ident[5] != ELFDATA2LSB ||
2773 ehdr.e_machine != EM_TCC_TARGET) {
2774 fail1:
2775 tcc_error_noabort("invalid object file");
2776 return -1;
2778 /* read sections */
2779 shdr = load_data(fd, file_offset + ehdr.e_shoff,
2780 sizeof(ElfW(Shdr)) * ehdr.e_shnum);
2781 sm_table = tcc_mallocz(sizeof(SectionMergeInfo) * ehdr.e_shnum);
2783 /* load section names */
2784 sh = &shdr[ehdr.e_shstrndx];
2785 strsec = load_data(fd, file_offset + sh->sh_offset, sh->sh_size);
2787 /* load symtab and strtab */
2788 old_to_new_syms = NULL;
2789 symtab = NULL;
2790 strtab = NULL;
2791 nb_syms = 0;
2792 for(i = 1; i < ehdr.e_shnum; i++) {
2793 sh = &shdr[i];
2794 if (sh->sh_type == SHT_SYMTAB) {
2795 if (symtab) {
2796 tcc_error_noabort("object must contain only one symtab");
2797 fail:
2798 ret = -1;
2799 goto the_end;
2801 nb_syms = sh->sh_size / sizeof(ElfW(Sym));
2802 symtab = load_data(fd, file_offset + sh->sh_offset, sh->sh_size);
2803 sm_table[i].s = symtab_section;
2805 /* now load strtab */
2806 sh = &shdr[sh->sh_link];
2807 strtab = load_data(fd, file_offset + sh->sh_offset, sh->sh_size);
2811 /* now examine each section and try to merge its content with the
2812 ones in memory */
2813 for(i = 1; i < ehdr.e_shnum; i++) {
2814 /* no need to examine section name strtab */
2815 if (i == ehdr.e_shstrndx)
2816 continue;
2817 sh = &shdr[i];
2818 sh_name = (char *) strsec + sh->sh_name;
2819 /* ignore sections types we do not handle */
2820 if (sh->sh_type != SHT_PROGBITS &&
2821 sh->sh_type != SHT_RELX &&
2822 #ifdef TCC_ARM_EABI
2823 sh->sh_type != SHT_ARM_EXIDX &&
2824 #endif
2825 sh->sh_type != SHT_NOBITS &&
2826 sh->sh_type != SHT_PREINIT_ARRAY &&
2827 sh->sh_type != SHT_INIT_ARRAY &&
2828 sh->sh_type != SHT_FINI_ARRAY &&
2829 strcmp(sh_name, ".stabstr")
2831 continue;
2832 if (sh->sh_addralign < 1)
2833 sh->sh_addralign = 1;
2834 /* find corresponding section, if any */
2835 for(j = 1; j < s1->nb_sections;j++) {
2836 s = s1->sections[j];
2837 if (!strcmp(s->name, sh_name)) {
2838 if (!strncmp(sh_name, ".gnu.linkonce",
2839 sizeof(".gnu.linkonce") - 1)) {
2840 /* if a 'linkonce' section is already present, we
2841 do not add it again. It is a little tricky as
2842 symbols can still be defined in
2843 it. */
2844 sm_table[i].link_once = 1;
2845 goto next;
2846 } else {
2847 goto found;
2851 /* not found: create new section */
2852 s = new_section(s1, sh_name, sh->sh_type, sh->sh_flags);
2853 /* take as much info as possible from the section. sh_link and
2854 sh_info will be updated later */
2855 s->sh_addralign = sh->sh_addralign;
2856 s->sh_entsize = sh->sh_entsize;
2857 sm_table[i].new_section = 1;
2858 found:
2859 if (sh->sh_type != s->sh_type) {
2860 tcc_error_noabort("invalid section type");
2861 goto fail;
2864 /* align start of section */
2865 offset = s->data_offset;
2867 if (0 == strcmp(sh_name, ".stab")) {
2868 stab_index = i;
2869 goto no_align;
2871 if (0 == strcmp(sh_name, ".stabstr")) {
2872 stabstr_index = i;
2873 goto no_align;
2876 size = sh->sh_addralign - 1;
2877 offset = (offset + size) & ~size;
2878 if (sh->sh_addralign > s->sh_addralign)
2879 s->sh_addralign = sh->sh_addralign;
2880 s->data_offset = offset;
2881 no_align:
2882 sm_table[i].offset = offset;
2883 sm_table[i].s = s;
2884 /* concatenate sections */
2885 size = sh->sh_size;
2886 if (sh->sh_type != SHT_NOBITS) {
2887 unsigned char *ptr;
2888 lseek(fd, file_offset + sh->sh_offset, SEEK_SET);
2889 ptr = section_ptr_add(s, size);
2890 read(fd, ptr, size);
2891 } else {
2892 s->data_offset += size;
2894 next: ;
2897 /* gr relocate stab strings */
2898 if (stab_index && stabstr_index) {
2899 Stab_Sym *a, *b;
2900 unsigned o;
2901 s = sm_table[stab_index].s;
2902 a = (Stab_Sym *)(s->data + sm_table[stab_index].offset);
2903 b = (Stab_Sym *)(s->data + s->data_offset);
2904 o = sm_table[stabstr_index].offset;
2905 while (a < b)
2906 a->n_strx += o, a++;
2909 /* second short pass to update sh_link and sh_info fields of new
2910 sections */
2911 for(i = 1; i < ehdr.e_shnum; i++) {
2912 s = sm_table[i].s;
2913 if (!s || !sm_table[i].new_section)
2914 continue;
2915 sh = &shdr[i];
2916 if (sh->sh_link > 0)
2917 s->link = sm_table[sh->sh_link].s;
2918 if (sh->sh_type == SHT_RELX) {
2919 s->sh_info = sm_table[sh->sh_info].s->sh_num;
2920 /* update backward link */
2921 s1->sections[s->sh_info]->reloc = s;
2924 sm = sm_table;
2926 /* resolve symbols */
2927 old_to_new_syms = tcc_mallocz(nb_syms * sizeof(int));
2929 sym = symtab + 1;
2930 for(i = 1; i < nb_syms; i++, sym++) {
2931 if (sym->st_shndx != SHN_UNDEF &&
2932 sym->st_shndx < SHN_LORESERVE) {
2933 sm = &sm_table[sym->st_shndx];
2934 if (sm->link_once) {
2935 /* if a symbol is in a link once section, we use the
2936 already defined symbol. It is very important to get
2937 correct relocations */
2938 if (ELFW(ST_BIND)(sym->st_info) != STB_LOCAL) {
2939 name = (char *) strtab + sym->st_name;
2940 sym_index = find_elf_sym(symtab_section, name);
2941 if (sym_index)
2942 old_to_new_syms[i] = sym_index;
2944 continue;
2946 /* if no corresponding section added, no need to add symbol */
2947 if (!sm->s)
2948 continue;
2949 /* convert section number */
2950 sym->st_shndx = sm->s->sh_num;
2951 /* offset value */
2952 sym->st_value += sm->offset;
2954 /* add symbol */
2955 name = (char *) strtab + sym->st_name;
2956 sym_index = add_elf_sym(symtab_section, sym->st_value, sym->st_size,
2957 sym->st_info, sym->st_other,
2958 sym->st_shndx, name);
2959 old_to_new_syms[i] = sym_index;
2962 /* third pass to patch relocation entries */
2963 for(i = 1; i < ehdr.e_shnum; i++) {
2964 s = sm_table[i].s;
2965 if (!s)
2966 continue;
2967 sh = &shdr[i];
2968 offset = sm_table[i].offset;
2969 switch(s->sh_type) {
2970 case SHT_RELX:
2971 /* take relocation offset information */
2972 offseti = sm_table[sh->sh_info].offset;
2973 for_each_elem(s, (offset / sizeof(*rel)), rel, ElfW_Rel) {
2974 int type;
2975 unsigned sym_index;
2976 /* convert symbol index */
2977 type = ELFW(R_TYPE)(rel->r_info);
2978 sym_index = ELFW(R_SYM)(rel->r_info);
2979 /* NOTE: only one symtab assumed */
2980 if (sym_index >= nb_syms)
2981 goto invalid_reloc;
2982 sym_index = old_to_new_syms[sym_index];
2983 /* ignore link_once in rel section. */
2984 if (!sym_index && !sm->link_once
2985 #ifdef TCC_TARGET_ARM
2986 && type != R_ARM_V4BX
2987 #endif
2988 ) {
2989 invalid_reloc:
2990 tcc_error_noabort("Invalid relocation entry [%2d] '%s' @ %.8x",
2991 i, strsec + sh->sh_name, rel->r_offset);
2992 goto fail;
2994 rel->r_info = ELFW(R_INFO)(sym_index, type);
2995 /* offset the relocation offset */
2996 rel->r_offset += offseti;
2997 #ifdef TCC_TARGET_ARM
2998 /* Jumps and branches from a Thumb code to a PLT entry need
2999 special handling since PLT entries are ARM code.
3000 Unconditional bl instructions referencing PLT entries are
3001 handled by converting these instructions into blx
3002 instructions. Other case of instructions referencing a PLT
3003 entry require to add a Thumb stub before the PLT entry to
3004 switch to ARM mode. We set bit plt_thumb_stub of the
3005 attribute of a symbol to indicate such a case. */
3006 if (type == R_ARM_THM_JUMP24)
3007 alloc_sym_attr(s1, sym_index)->plt_thumb_stub = 1;
3008 #endif
3010 break;
3011 default:
3012 break;
3016 ret = 0;
3017 the_end:
3018 tcc_free(symtab);
3019 tcc_free(strtab);
3020 tcc_free(old_to_new_syms);
3021 tcc_free(sm_table);
3022 tcc_free(strsec);
3023 tcc_free(shdr);
3024 return ret;
3027 typedef struct ArchiveHeader {
3028 char ar_name[16]; /* name of this member */
3029 char ar_date[12]; /* file mtime */
3030 char ar_uid[6]; /* owner uid; printed as decimal */
3031 char ar_gid[6]; /* owner gid; printed as decimal */
3032 char ar_mode[8]; /* file mode, printed as octal */
3033 char ar_size[10]; /* file size, printed as decimal */
3034 char ar_fmag[2]; /* should contain ARFMAG */
3035 } ArchiveHeader;
3037 static int get_be32(const uint8_t *b)
3039 return b[3] | (b[2] << 8) | (b[1] << 16) | (b[0] << 24);
3042 /* load only the objects which resolve undefined symbols */
3043 static int tcc_load_alacarte(TCCState *s1, int fd, int size)
3045 int i, bound, nsyms, sym_index, off, ret;
3046 uint8_t *data;
3047 const char *ar_names, *p;
3048 const uint8_t *ar_index;
3049 ElfW(Sym) *sym;
3051 data = tcc_malloc(size);
3052 if (read(fd, data, size) != size)
3053 goto fail;
3054 nsyms = get_be32(data);
3055 ar_index = data + 4;
3056 ar_names = (char *) ar_index + nsyms * 4;
3058 do {
3059 bound = 0;
3060 for(p = ar_names, i = 0; i < nsyms; i++, p += strlen(p)+1) {
3061 sym_index = find_elf_sym(symtab_section, p);
3062 if(sym_index) {
3063 sym = &((ElfW(Sym) *)symtab_section->data)[sym_index];
3064 if(sym->st_shndx == SHN_UNDEF) {
3065 off = get_be32(ar_index + i * 4) + sizeof(ArchiveHeader);
3066 ++bound;
3067 lseek(fd, off, SEEK_SET);
3068 if(tcc_load_object_file(s1, fd, off) < 0) {
3069 fail:
3070 ret = -1;
3071 goto the_end;
3076 } while(bound);
3077 ret = 0;
3078 the_end:
3079 tcc_free(data);
3080 return ret;
3083 /* load a '.a' file */
3084 ST_FUNC int tcc_load_archive(TCCState *s1, int fd)
3086 ArchiveHeader hdr;
3087 char ar_size[11];
3088 char ar_name[17];
3089 char magic[8];
3090 int size, len, i;
3091 unsigned long file_offset;
3093 /* skip magic which was already checked */
3094 read(fd, magic, sizeof(magic));
3096 for(;;) {
3097 len = read(fd, &hdr, sizeof(hdr));
3098 if (len == 0)
3099 break;
3100 if (len != sizeof(hdr)) {
3101 tcc_error_noabort("invalid archive");
3102 return -1;
3104 memcpy(ar_size, hdr.ar_size, sizeof(hdr.ar_size));
3105 ar_size[sizeof(hdr.ar_size)] = '\0';
3106 size = strtol(ar_size, NULL, 0);
3107 memcpy(ar_name, hdr.ar_name, sizeof(hdr.ar_name));
3108 for(i = sizeof(hdr.ar_name) - 1; i >= 0; i--) {
3109 if (ar_name[i] != ' ')
3110 break;
3112 ar_name[i + 1] = '\0';
3113 file_offset = lseek(fd, 0, SEEK_CUR);
3114 /* align to even */
3115 size = (size + 1) & ~1;
3116 if (!strcmp(ar_name, "/")) {
3117 /* coff symbol table : we handle it */
3118 if(s1->alacarte_link)
3119 return tcc_load_alacarte(s1, fd, size);
3120 } else if (!strcmp(ar_name, "//") ||
3121 !strcmp(ar_name, "__.SYMDEF") ||
3122 !strcmp(ar_name, "__.SYMDEF/") ||
3123 !strcmp(ar_name, "ARFILENAMES/")) {
3124 /* skip symbol table or archive names */
3125 } else {
3126 if (tcc_load_object_file(s1, fd, file_offset) < 0)
3127 return -1;
3129 lseek(fd, file_offset + size, SEEK_SET);
3131 return 0;
3134 #ifndef TCC_TARGET_PE
3135 /* load a DLL and all referenced DLLs. 'level = 0' means that the DLL
3136 is referenced by the user (so it should be added as DT_NEEDED in
3137 the generated ELF file) */
3138 ST_FUNC int tcc_load_dll(TCCState *s1, int fd, const char *filename, int level)
3140 ElfW(Ehdr) ehdr;
3141 ElfW(Shdr) *shdr, *sh, *sh1;
3142 int i, j, nb_syms, nb_dts, sym_bind, ret;
3143 ElfW(Sym) *sym, *dynsym;
3144 ElfW(Dyn) *dt, *dynamic;
3145 unsigned char *dynstr;
3146 const char *name, *soname;
3147 DLLReference *dllref;
3149 read(fd, &ehdr, sizeof(ehdr));
3151 /* test CPU specific stuff */
3152 if (ehdr.e_ident[5] != ELFDATA2LSB ||
3153 ehdr.e_machine != EM_TCC_TARGET) {
3154 tcc_error_noabort("bad architecture");
3155 return -1;
3158 /* read sections */
3159 shdr = load_data(fd, ehdr.e_shoff, sizeof(ElfW(Shdr)) * ehdr.e_shnum);
3161 /* load dynamic section and dynamic symbols */
3162 nb_syms = 0;
3163 nb_dts = 0;
3164 dynamic = NULL;
3165 dynsym = NULL; /* avoid warning */
3166 dynstr = NULL; /* avoid warning */
3167 for(i = 0, sh = shdr; i < ehdr.e_shnum; i++, sh++) {
3168 switch(sh->sh_type) {
3169 case SHT_DYNAMIC:
3170 nb_dts = sh->sh_size / sizeof(ElfW(Dyn));
3171 dynamic = load_data(fd, sh->sh_offset, sh->sh_size);
3172 break;
3173 case SHT_DYNSYM:
3174 nb_syms = sh->sh_size / sizeof(ElfW(Sym));
3175 dynsym = load_data(fd, sh->sh_offset, sh->sh_size);
3176 sh1 = &shdr[sh->sh_link];
3177 dynstr = load_data(fd, sh1->sh_offset, sh1->sh_size);
3178 break;
3179 default:
3180 break;
3184 /* compute the real library name */
3185 soname = tcc_basename(filename);
3187 for(i = 0, dt = dynamic; i < nb_dts; i++, dt++) {
3188 if (dt->d_tag == DT_SONAME) {
3189 soname = (char *) dynstr + dt->d_un.d_val;
3193 /* if the dll is already loaded, do not load it */
3194 for(i = 0; i < s1->nb_loaded_dlls; i++) {
3195 dllref = s1->loaded_dlls[i];
3196 if (!strcmp(soname, dllref->name)) {
3197 /* but update level if needed */
3198 if (level < dllref->level)
3199 dllref->level = level;
3200 ret = 0;
3201 goto the_end;
3205 /* add the dll and its level */
3206 dllref = tcc_mallocz(sizeof(DLLReference) + strlen(soname));
3207 dllref->level = level;
3208 strcpy(dllref->name, soname);
3209 dynarray_add((void ***)&s1->loaded_dlls, &s1->nb_loaded_dlls, dllref);
3211 /* add dynamic symbols in dynsym_section */
3212 for(i = 1, sym = dynsym + 1; i < nb_syms; i++, sym++) {
3213 sym_bind = ELFW(ST_BIND)(sym->st_info);
3214 if (sym_bind == STB_LOCAL)
3215 continue;
3216 name = (char *) dynstr + sym->st_name;
3217 add_elf_sym(s1->dynsymtab_section, sym->st_value, sym->st_size,
3218 sym->st_info, sym->st_other, sym->st_shndx, name);
3221 /* load all referenced DLLs */
3222 for(i = 0, dt = dynamic; i < nb_dts; i++, dt++) {
3223 switch(dt->d_tag) {
3224 case DT_NEEDED:
3225 name = (char *) dynstr + dt->d_un.d_val;
3226 for(j = 0; j < s1->nb_loaded_dlls; j++) {
3227 dllref = s1->loaded_dlls[j];
3228 if (!strcmp(name, dllref->name))
3229 goto already_loaded;
3231 if (tcc_add_dll(s1, name, AFF_REFERENCED_DLL) < 0) {
3232 tcc_error_noabort("referenced dll '%s' not found", name);
3233 ret = -1;
3234 goto the_end;
3236 already_loaded:
3237 break;
3240 ret = 0;
3241 the_end:
3242 tcc_free(dynstr);
3243 tcc_free(dynsym);
3244 tcc_free(dynamic);
3245 tcc_free(shdr);
3246 return ret;
3249 #define LD_TOK_NAME 256
3250 #define LD_TOK_EOF (-1)
3252 /* return next ld script token */
3253 static int ld_next(TCCState *s1, char *name, int name_size)
3255 int c;
3256 char *q;
3258 redo:
3259 switch(ch) {
3260 case ' ':
3261 case '\t':
3262 case '\f':
3263 case '\v':
3264 case '\r':
3265 case '\n':
3266 inp();
3267 goto redo;
3268 case '/':
3269 minp();
3270 if (ch == '*') {
3271 file->buf_ptr = parse_comment(file->buf_ptr);
3272 ch = file->buf_ptr[0];
3273 goto redo;
3274 } else {
3275 q = name;
3276 *q++ = '/';
3277 goto parse_name;
3279 break;
3280 case '\\':
3281 ch = handle_eob();
3282 if (ch != '\\')
3283 goto redo;
3284 /* fall through */
3285 /* case 'a' ... 'z': */
3286 case 'a':
3287 case 'b':
3288 case 'c':
3289 case 'd':
3290 case 'e':
3291 case 'f':
3292 case 'g':
3293 case 'h':
3294 case 'i':
3295 case 'j':
3296 case 'k':
3297 case 'l':
3298 case 'm':
3299 case 'n':
3300 case 'o':
3301 case 'p':
3302 case 'q':
3303 case 'r':
3304 case 's':
3305 case 't':
3306 case 'u':
3307 case 'v':
3308 case 'w':
3309 case 'x':
3310 case 'y':
3311 case 'z':
3312 /* case 'A' ... 'z': */
3313 case 'A':
3314 case 'B':
3315 case 'C':
3316 case 'D':
3317 case 'E':
3318 case 'F':
3319 case 'G':
3320 case 'H':
3321 case 'I':
3322 case 'J':
3323 case 'K':
3324 case 'L':
3325 case 'M':
3326 case 'N':
3327 case 'O':
3328 case 'P':
3329 case 'Q':
3330 case 'R':
3331 case 'S':
3332 case 'T':
3333 case 'U':
3334 case 'V':
3335 case 'W':
3336 case 'X':
3337 case 'Y':
3338 case 'Z':
3339 case '_':
3340 case '.':
3341 case '$':
3342 case '~':
3343 q = name;
3344 parse_name:
3345 for(;;) {
3346 if (!((ch >= 'a' && ch <= 'z') ||
3347 (ch >= 'A' && ch <= 'Z') ||
3348 (ch >= '0' && ch <= '9') ||
3349 strchr("/.-_+=$:\\,~", ch)))
3350 break;
3351 if ((q - name) < name_size - 1) {
3352 *q++ = ch;
3354 minp();
3356 *q = '\0';
3357 c = LD_TOK_NAME;
3358 break;
3359 case CH_EOF:
3360 c = LD_TOK_EOF;
3361 break;
3362 default:
3363 c = ch;
3364 inp();
3365 break;
3367 return c;
3370 static int ld_add_file(TCCState *s1, const char filename[])
3372 int ret;
3374 ret = tcc_add_file_internal(s1, filename, 0, TCC_FILETYPE_BINARY);
3375 if (ret)
3376 ret = tcc_add_dll(s1, filename, 0);
3377 return ret;
3380 static inline int new_undef_syms(void)
3382 int ret = 0;
3383 ret = new_undef_sym;
3384 new_undef_sym = 0;
3385 return ret;
3388 static int ld_add_file_list(TCCState *s1, const char *cmd, int as_needed)
3390 char filename[1024], libname[1024];
3391 int t, group, nblibs = 0, ret = 0;
3392 char **libs = NULL;
3394 group = !strcmp(cmd, "GROUP");
3395 if (!as_needed)
3396 new_undef_syms();
3397 t = ld_next(s1, filename, sizeof(filename));
3398 if (t != '(')
3399 expect("(");
3400 t = ld_next(s1, filename, sizeof(filename));
3401 for(;;) {
3402 libname[0] = '\0';
3403 if (t == LD_TOK_EOF) {
3404 tcc_error_noabort("unexpected end of file");
3405 ret = -1;
3406 goto lib_parse_error;
3407 } else if (t == ')') {
3408 break;
3409 } else if (t == '-') {
3410 t = ld_next(s1, filename, sizeof(filename));
3411 if ((t != LD_TOK_NAME) || (filename[0] != 'l')) {
3412 tcc_error_noabort("library name expected");
3413 ret = -1;
3414 goto lib_parse_error;
3416 pstrcpy(libname, sizeof libname, &filename[1]);
3417 if (s1->static_link) {
3418 snprintf(filename, sizeof filename, "lib%s.a", libname);
3419 } else {
3420 snprintf(filename, sizeof filename, "lib%s.so", libname);
3422 } else if (t != LD_TOK_NAME) {
3423 tcc_error_noabort("filename expected");
3424 ret = -1;
3425 goto lib_parse_error;
3427 if (!strcmp(filename, "AS_NEEDED")) {
3428 ret = ld_add_file_list(s1, cmd, 1);
3429 if (ret)
3430 goto lib_parse_error;
3431 } else {
3432 /* TODO: Implement AS_NEEDED support. Ignore it for now */
3433 if (!as_needed) {
3434 ret = ld_add_file(s1, filename);
3435 if (ret)
3436 goto lib_parse_error;
3437 if (group) {
3438 /* Add the filename *and* the libname to avoid future conversions */
3439 dynarray_add((void ***) &libs, &nblibs, tcc_strdup(filename));
3440 if (libname[0] != '\0')
3441 dynarray_add((void ***) &libs, &nblibs, tcc_strdup(libname));
3445 t = ld_next(s1, filename, sizeof(filename));
3446 if (t == ',') {
3447 t = ld_next(s1, filename, sizeof(filename));
3450 if (group && !as_needed) {
3451 while (new_undef_syms()) {
3452 int i;
3454 for (i = 0; i < nblibs; i ++)
3455 ld_add_file(s1, libs[i]);
3458 lib_parse_error:
3459 dynarray_reset(&libs, &nblibs);
3460 return ret;
3463 /* interpret a subset of GNU ldscripts to handle the dummy libc.so
3464 files */
3465 ST_FUNC int tcc_load_ldscript(TCCState *s1)
3467 char cmd[64];
3468 char filename[1024];
3469 int t, ret;
3471 ch = handle_eob();
3472 for(;;) {
3473 t = ld_next(s1, cmd, sizeof(cmd));
3474 if (t == LD_TOK_EOF)
3475 return 0;
3476 else if (t != LD_TOK_NAME)
3477 return -1;
3478 if (!strcmp(cmd, "INPUT") ||
3479 !strcmp(cmd, "GROUP")) {
3480 ret = ld_add_file_list(s1, cmd, 0);
3481 if (ret)
3482 return ret;
3483 } else if (!strcmp(cmd, "OUTPUT_FORMAT") ||
3484 !strcmp(cmd, "TARGET")) {
3485 /* ignore some commands */
3486 t = ld_next(s1, cmd, sizeof(cmd));
3487 if (t != '(')
3488 expect("(");
3489 for(;;) {
3490 t = ld_next(s1, filename, sizeof(filename));
3491 if (t == LD_TOK_EOF) {
3492 tcc_error_noabort("unexpected end of file");
3493 return -1;
3494 } else if (t == ')') {
3495 break;
3498 } else {
3499 return -1;
3502 return 0;
3504 #endif /* !TCC_TARGET_PE */