Use RELA relocations properly for R_DATA_PTR on x86_64.
[tinycc.git] / tccelf.c
blobdc0a1443cacbcba549fff91e2704274713d86210
1 /*
2 * ELF file handling for TCC
4 * Copyright (c) 2001-2004 Fabrice Bellard
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.
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.
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
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)
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;
41 /* elf symbol hashing function */
42 static unsigned long elf_hash(const unsigned char *name)
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;
53 return h;
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)
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;
85 ptr++;
86 sym++;
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)
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, unsigned long 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 #ifdef 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%x\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 #ifdef 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_C67)
764 case R_C60_32:
765 *(int *)ptr += val;
766 break;
767 case R_C60LO16:
769 uint32_t orig;
771 /* put the low 16 bits of the absolute address
772 add to what is already there */
774 orig = ((*(int *)(ptr )) >> 7) & 0xffff;
775 orig |= (((*(int *)(ptr+4)) >> 7) & 0xffff) << 16;
777 /* patch both at once - assumes always in pairs Low - High */
779 *(int *) ptr = (*(int *) ptr & (~(0xffff << 7)) ) | (((val+orig) & 0xffff) << 7);
780 *(int *)(ptr+4) = (*(int *)(ptr+4) & (~(0xffff << 7)) ) | ((((val+orig)>>16) & 0xffff) << 7);
782 break;
783 case R_C60HI16:
784 break;
785 default:
786 fprintf(stderr,"FIXME: handle reloc type %x at %x [%p] to %x\n",
787 type, (unsigned)addr, ptr, (unsigned)val);
788 break;
789 #elif defined(TCC_TARGET_X86_64)
790 case R_X86_64_64:
791 if (s1->output_type == TCC_OUTPUT_DLL) {
792 esym_index = s1->symtab_to_dynsym[sym_index];
793 qrel->r_offset = rel->r_offset;
794 if (esym_index) {
795 qrel->r_info = ELFW(R_INFO)(esym_index, R_X86_64_64);
796 qrel->r_addend = rel->r_addend;
797 qrel++;
798 break;
799 } else {
800 qrel->r_info = ELFW(R_INFO)(0, R_X86_64_RELATIVE);
801 qrel->r_addend = *(long long *)ptr + val;
802 qrel++;
805 *(long long *)ptr += val;
806 break;
807 case R_X86_64_32:
808 case R_X86_64_32S:
809 if (s1->output_type == TCC_OUTPUT_DLL) {
810 /* XXX: this logic may depend on TCC's codegen
811 now TCC uses R_X86_64_32 even for a 64bit pointer */
812 qrel->r_info = ELFW(R_INFO)(0, R_X86_64_RELATIVE);
813 qrel->r_addend = *(int *)ptr + val;
814 qrel++;
816 *(int *)ptr += val;
817 break;
819 case R_X86_64_PC32:
820 if (s1->output_type == TCC_OUTPUT_DLL) {
821 /* DLL relocation */
822 esym_index = s1->symtab_to_dynsym[sym_index];
823 if (esym_index) {
824 qrel->r_offset = rel->r_offset;
825 qrel->r_info = ELFW(R_INFO)(esym_index, R_X86_64_PC32);
826 qrel->r_addend = *(int *)ptr;
827 qrel++;
828 break;
831 goto plt32pc32;
833 case R_X86_64_PLT32:
834 /* We've put the PLT slot offset into r_addend when generating
835 it, and that's what we must use as relocation value (adjusted
836 by section offset of course). */
837 val = s1->plt->sh_addr + rel->r_addend;
838 /* fallthrough. */
840 plt32pc32:
842 long long diff;
843 diff = (long long)val - addr;
844 if (diff <= -2147483647 || diff > 2147483647) {
845 tcc_error("internal error: relocation failed");
847 *(int *)ptr += diff;
849 break;
850 case R_X86_64_GLOB_DAT:
851 case R_X86_64_JUMP_SLOT:
852 /* They don't need addend */
853 *(addr_t *)ptr = val - rel->r_addend;
854 break;
855 case R_X86_64_GOTPCREL:
856 *(int *)ptr += (s1->got->sh_addr - addr +
857 s1->sym_attrs[sym_index].got_offset - 4);
858 break;
859 case R_X86_64_GOTTPOFF:
860 *(int *)ptr += val - s1->got->sh_addr;
861 break;
862 case R_X86_64_GOT32:
863 /* we load the got offset */
864 *(int *)ptr += s1->sym_attrs[sym_index].got_offset;
865 break;
866 #else
867 #error unsupported processor
868 #endif
871 /* if the relocation is allocated, we change its symbol table */
872 if (sr->sh_flags & SHF_ALLOC)
873 sr->link = s1->dynsym;
876 /* relocate relocation table in 'sr' */
877 static void relocate_rel(TCCState *s1, Section *sr)
879 Section *s;
880 ElfW_Rel *rel;
882 s = s1->sections[sr->sh_info];
883 for_each_elem(sr, 0, rel, ElfW_Rel)
884 rel->r_offset += s->sh_addr;
887 /* count the number of dynamic relocations so that we can reserve
888 their space */
889 static int prepare_dynamic_rel(TCCState *s1, Section *sr)
891 ElfW_Rel *rel;
892 int sym_index, esym_index, type, count;
894 count = 0;
895 for_each_elem(sr, 0, rel, ElfW_Rel) {
896 sym_index = ELFW(R_SYM)(rel->r_info);
897 type = ELFW(R_TYPE)(rel->r_info);
898 switch(type) {
899 #if defined(TCC_TARGET_I386)
900 case R_386_32:
901 #elif defined(TCC_TARGET_X86_64)
902 case R_X86_64_32:
903 case R_X86_64_32S:
904 case R_X86_64_64:
905 #endif
906 count++;
907 break;
908 #if defined(TCC_TARGET_I386)
909 case R_386_PC32:
910 #elif defined(TCC_TARGET_X86_64)
911 case R_X86_64_PC32:
912 #endif
913 esym_index = s1->symtab_to_dynsym[sym_index];
914 if (esym_index)
915 count++;
916 break;
917 default:
918 break;
921 if (count) {
922 /* allocate the section */
923 sr->sh_flags |= SHF_ALLOC;
924 sr->sh_size = count * sizeof(ElfW_Rel);
926 return count;
929 static struct sym_attr *alloc_sym_attr(TCCState *s1, int index)
931 int n;
932 struct sym_attr *tab;
934 if (index >= s1->nb_sym_attrs) {
935 /* find immediately bigger power of 2 and reallocate array */
936 n = 1;
937 while (index >= n)
938 n *= 2;
939 tab = tcc_realloc(s1->sym_attrs, n * sizeof(*s1->sym_attrs));
940 s1->sym_attrs = tab;
941 memset(s1->sym_attrs + s1->nb_sym_attrs, 0,
942 (n - s1->nb_sym_attrs) * sizeof(*s1->sym_attrs));
943 s1->nb_sym_attrs = n;
945 return &s1->sym_attrs[index];
948 /* XXX: suppress that */
949 static void put32(unsigned char *p, uint32_t val)
951 p[0] = val;
952 p[1] = val >> 8;
953 p[2] = val >> 16;
954 p[3] = val >> 24;
957 #if defined(TCC_TARGET_I386) || defined(TCC_TARGET_ARM) || \
958 defined(TCC_TARGET_X86_64)
959 static uint32_t get32(unsigned char *p)
961 return p[0] | (p[1] << 8) | (p[2] << 16) | (p[3] << 24);
963 #endif
965 static void build_got(TCCState *s1)
967 unsigned char *ptr;
969 /* if no got, then create it */
970 s1->got = new_section(s1, ".got", SHT_PROGBITS, SHF_ALLOC | SHF_WRITE);
971 s1->got->sh_entsize = 4;
972 add_elf_sym(symtab_section, 0, 4, ELFW(ST_INFO)(STB_GLOBAL, STT_OBJECT),
973 0, s1->got->sh_num, "_GLOBAL_OFFSET_TABLE_");
974 ptr = section_ptr_add(s1->got, 3 * PTR_SIZE);
975 #if PTR_SIZE == 4
976 /* keep space for _DYNAMIC pointer, if present */
977 put32(ptr, 0);
978 /* two dummy got entries */
979 put32(ptr + 4, 0);
980 put32(ptr + 8, 0);
981 #else
982 /* keep space for _DYNAMIC pointer, if present */
983 put32(ptr, 0);
984 put32(ptr + 4, 0);
985 /* two dummy got entries */
986 put32(ptr + 8, 0);
987 put32(ptr + 12, 0);
988 put32(ptr + 16, 0);
989 put32(ptr + 20, 0);
990 #endif
993 /* put a got or plt entry corresponding to a symbol in symtab_section. 'size'
994 and 'info' can be modifed if more precise info comes from the DLL.
995 Returns offset of GOT or PLT slot. */
996 static unsigned long put_got_entry(TCCState *s1,
997 int reloc_type, unsigned long size, int info,
998 int sym_index)
1000 int index, need_plt_entry;
1001 const char *name;
1002 ElfW(Sym) *sym;
1003 unsigned long offset;
1004 int *ptr;
1005 struct sym_attr *symattr;
1007 if (!s1->got)
1008 build_got(s1);
1010 need_plt_entry =
1011 #ifdef TCC_TARGET_X86_64
1012 (reloc_type == R_X86_64_JUMP_SLOT);
1013 #elif defined(TCC_TARGET_I386)
1014 (reloc_type == R_386_JMP_SLOT);
1015 #elif defined(TCC_TARGET_ARM)
1016 (reloc_type == R_ARM_JUMP_SLOT);
1017 #else
1019 #endif
1021 if (need_plt_entry && !s1->plt) {
1022 /* add PLT */
1023 s1->plt = new_section(s1, ".plt", SHT_PROGBITS,
1024 SHF_ALLOC | SHF_EXECINSTR);
1025 s1->plt->sh_entsize = 4;
1028 /* If a got/plt entry already exists for that symbol, no need to add one */
1029 if (sym_index < s1->nb_sym_attrs) {
1030 if (need_plt_entry && s1->sym_attrs[sym_index].plt_offset)
1031 return s1->sym_attrs[sym_index].plt_offset;
1032 else if (!need_plt_entry && s1->sym_attrs[sym_index].got_offset)
1033 return s1->sym_attrs[sym_index].got_offset;
1036 symattr = alloc_sym_attr(s1, sym_index);
1038 /* Only store the GOT offset if it's not generated for the PLT entry. */
1039 if (!need_plt_entry)
1040 symattr->got_offset = s1->got->data_offset;
1042 sym = &((ElfW(Sym) *)symtab_section->data)[sym_index];
1043 name = (char *) symtab_section->link->data + sym->st_name;
1044 offset = sym->st_value;
1045 #if defined(TCC_TARGET_I386) || defined(TCC_TARGET_X86_64)
1046 if (need_plt_entry) {
1047 Section *plt;
1048 uint8_t *p;
1049 int modrm;
1050 unsigned long relofs;
1052 #if defined(TCC_OUTPUT_DLL_WITH_PLT)
1053 modrm = 0x25;
1054 #else
1055 /* if we build a DLL, we add a %ebx offset */
1056 if (s1->output_type == TCC_OUTPUT_DLL)
1057 modrm = 0xa3;
1058 else
1059 modrm = 0x25;
1060 #endif
1062 /* add a PLT entry */
1063 plt = s1->plt;
1064 if (plt->data_offset == 0) {
1065 /* first plt entry */
1066 p = section_ptr_add(plt, 16);
1067 p[0] = 0xff; /* pushl got + PTR_SIZE */
1068 p[1] = modrm + 0x10;
1069 put32(p + 2, PTR_SIZE);
1070 p[6] = 0xff; /* jmp *(got + PTR_SIZE * 2) */
1071 p[7] = modrm;
1072 put32(p + 8, PTR_SIZE * 2);
1075 /* The PLT slot refers to the relocation entry it needs
1076 via offset. The reloc entry is created below, so its
1077 offset is the current data_offset. */
1078 relofs = s1->got->reloc ? s1->got->reloc->data_offset : 0;
1079 symattr->plt_offset = plt->data_offset;
1080 p = section_ptr_add(plt, 16);
1081 p[0] = 0xff; /* jmp *(got + x) */
1082 p[1] = modrm;
1083 put32(p + 2, s1->got->data_offset);
1084 p[6] = 0x68; /* push $xxx */
1085 #ifdef TCC_TARGET_X86_64
1086 /* On x86-64, the relocation is referred to by _index_. */
1087 put32(p + 7, relofs / sizeof (ElfW_Rel));
1088 #else
1089 put32(p + 7, relofs);
1090 #endif
1091 p[11] = 0xe9; /* jmp plt_start */
1092 put32(p + 12, -(plt->data_offset));
1094 /* If this was an UNDEF symbol set the offset in the
1095 dynsymtab to the PLT slot, so that PC32 relocs to it
1096 can be resolved. */
1097 if (sym->st_shndx == SHN_UNDEF)
1098 offset = plt->data_offset - 16;
1100 #elif defined(TCC_TARGET_ARM)
1101 if (need_plt_entry) {
1102 Section *plt;
1103 uint8_t *p;
1105 /* if we build a DLL, we add a %ebx offset */
1106 if (s1->output_type == TCC_OUTPUT_DLL)
1107 tcc_error("DLLs unimplemented!");
1109 /* add a PLT entry */
1110 plt = s1->plt;
1111 if (plt->data_offset == 0) {
1112 /* first plt entry */
1113 p = section_ptr_add(plt, 16);
1114 put32(p, 0xe52de004); /* push {lr} */
1115 put32(p+4, 0xe59fe010); /* ldr lr, [pc, #16] */
1116 put32(p+8, 0xe08fe00e); /* add lr, pc, lr */
1117 put32(p+12, 0xe5bef008); /* ldr pc, [lr, #8]! */
1120 symattr->plt_offset = plt->data_offset;
1121 if (symattr->plt_thumb_stub) {
1122 p = section_ptr_add(plt, 20);
1123 put32(p, 0x4778); /* bx pc */
1124 put32(p+2, 0x46c0); /* nop */
1125 p += 4;
1126 } else
1127 p = section_ptr_add(plt, 16);
1128 put32(p, 0xe59fc004); /* ldr ip, [pc, #4] ; GOT entry offset */
1129 put32(p+4, 0xe08fc00c); /* add ip, pc, ip ; addr of GOT entry */
1130 put32(p+8, 0xe59cf000); /* ldr pc, [ip] ; jump to GOT entry */
1131 put32(p+12, s1->got->data_offset); /* GOT entry off once patched */
1133 /* the symbol is modified so that it will be relocated to
1134 the PLT */
1135 if (sym->st_shndx == SHN_UNDEF)
1136 offset = plt->data_offset - 16;
1138 #elif defined(TCC_TARGET_C67)
1139 if (s1->dynsym) {
1140 tcc_error("C67 got not implemented");
1142 #else
1143 #error unsupported CPU
1144 #endif
1145 if (s1->dynsym) {
1146 /* XXX This might generate multiple syms for name. */
1147 index = put_elf_sym(s1->dynsym, offset,
1148 size, info, 0, sym->st_shndx, name);
1149 /* Create the relocation (it's against the GOT for PLT
1150 and GOT relocs). */
1151 put_elf_reloc(s1->dynsym, s1->got,
1152 s1->got->data_offset,
1153 reloc_type, index);
1154 } else {
1155 /* Without .dynsym (i.e. static link or memory output) we
1156 still need relocs against the generated got, so as to fill
1157 the entries with the symbol values (determined later). */
1158 put_elf_reloc(symtab_section, s1->got,
1159 s1->got->data_offset,
1160 reloc_type, sym_index);
1162 /* And now create the GOT slot itself. */
1163 ptr = section_ptr_add(s1->got, PTR_SIZE);
1164 *ptr = 0;
1165 if (need_plt_entry)
1166 return symattr->plt_offset;
1167 else
1168 return symattr->got_offset;
1171 /* build GOT and PLT entries */
1172 ST_FUNC void build_got_entries(TCCState *s1)
1174 Section *s;
1175 ElfW_Rel *rel;
1176 ElfW(Sym) *sym;
1177 int i, type, reloc_type, sym_index;
1179 for(i = 1; i < s1->nb_sections; i++) {
1180 s = s1->sections[i];
1181 if (s->sh_type != SHT_RELX)
1182 continue;
1183 /* no need to handle got relocations */
1184 if (s->link != symtab_section)
1185 continue;
1186 for_each_elem(s, 0, rel, ElfW_Rel) {
1187 type = ELFW(R_TYPE)(rel->r_info);
1188 switch(type) {
1189 #if defined(TCC_TARGET_I386)
1190 case R_386_GOT32:
1191 case R_386_GOTOFF:
1192 case R_386_GOTPC:
1193 case R_386_PLT32:
1194 if (!s1->got)
1195 build_got(s1);
1196 if (type == R_386_GOT32 || type == R_386_PLT32) {
1197 sym_index = ELFW(R_SYM)(rel->r_info);
1198 sym = &((ElfW(Sym) *)symtab_section->data)[sym_index];
1199 /* look at the symbol got offset. If none, then add one */
1200 if (type == R_386_GOT32)
1201 reloc_type = R_386_GLOB_DAT;
1202 else
1203 reloc_type = R_386_JMP_SLOT;
1204 put_got_entry(s1, reloc_type, sym->st_size, sym->st_info,
1205 sym_index);
1207 break;
1208 #elif defined(TCC_TARGET_ARM)
1209 case R_ARM_PC24:
1210 case R_ARM_CALL:
1211 case R_ARM_JUMP24:
1212 case R_ARM_GOT32:
1213 case R_ARM_GOTOFF:
1214 case R_ARM_GOTPC:
1215 case R_ARM_PLT32:
1216 if (!s1->got)
1217 build_got(s1);
1218 sym_index = ELFW(R_SYM)(rel->r_info);
1219 sym = &((ElfW(Sym) *)symtab_section->data)[sym_index];
1220 if (type != R_ARM_GOTOFF && type != R_ARM_GOTPC
1221 && sym->st_shndx == SHN_UNDEF) {
1222 unsigned long ofs;
1223 /* look at the symbol got offset. If none, then add one */
1224 if (type == R_ARM_GOT32)
1225 reloc_type = R_ARM_GLOB_DAT;
1226 else
1227 reloc_type = R_ARM_JUMP_SLOT;
1228 ofs = put_got_entry(s1, reloc_type, sym->st_size,
1229 sym->st_info, sym_index);
1230 #ifdef DEBUG_RELOC
1231 printf ("maybegot: %s, %d, %d --> ofs=0x%x\n",
1232 (char *) symtab_section->link->data + sym->st_name,
1233 type, sym->st_shndx, ofs);
1234 #endif
1235 if (type != R_ARM_GOT32) {
1236 addr_t *ptr = (addr_t*)(s1->sections[s->sh_info]->data
1237 + rel->r_offset);
1238 /* x must be signed! */
1239 int x = *ptr & 0xffffff;
1240 x = (x << 8) >> 8;
1241 x <<= 2;
1242 x += ofs;
1243 x >>= 2;
1244 #ifdef DEBUG_RELOC
1245 printf ("insn=0x%x --> 0x%x (x==0x%x)\n", *ptr,
1246 (*ptr & 0xff000000) | x, x);
1247 #endif
1248 *ptr = (*ptr & 0xff000000) | x;
1251 break;
1252 case R_ARM_THM_JUMP24:
1253 sym_index = ELFW(R_SYM)(rel->r_info);
1254 sym = &((ElfW(Sym) *)symtab_section->data)[sym_index];
1255 /* We are relocating a jump from thumb code to arm code */
1256 if (sym->st_shndx != SHN_UNDEF && !(sym->st_value & 1)) {
1257 int index;
1258 uint8_t *p;
1259 char *name, buf[1024];
1260 Section *text_section;
1262 name = (char *) symtab_section->link->data + sym->st_name;
1263 text_section = s1->sections[sym->st_shndx];
1264 /* Modify reloc to target a thumb stub to switch to ARM */
1265 snprintf(buf, sizeof(buf), "%s_from_thumb", name);
1266 index = put_elf_sym(symtab_section,
1267 text_section->data_offset + 1,
1268 sym->st_size, sym->st_info, 0,
1269 sym->st_shndx, buf);
1270 rel->r_info = ELFW(R_INFO)(index, type);
1271 /* Create a thumb stub fonction to switch to ARM mode */
1272 put_elf_reloc(symtab_section, text_section,
1273 text_section->data_offset + 4, R_ARM_JUMP24,
1274 sym_index);
1275 p = section_ptr_add(text_section, 8);
1276 put32(p, 0x4778); /* bx pc */
1277 put32(p+2, 0x46c0); /* nop */
1278 put32(p+4, 0xeafffffe); /* b $sym */
1280 #elif defined(TCC_TARGET_C67)
1281 case R_C60_GOT32:
1282 case R_C60_GOTOFF:
1283 case R_C60_GOTPC:
1284 case R_C60_PLT32:
1285 if (!s1->got)
1286 build_got(s1);
1287 if (type == R_C60_GOT32 || type == R_C60_PLT32) {
1288 sym_index = ELFW(R_SYM)(rel->r_info);
1289 sym = &((ElfW(Sym) *)symtab_section->data)[sym_index];
1290 /* look at the symbol got offset. If none, then add one */
1291 if (type == R_C60_GOT32)
1292 reloc_type = R_C60_GLOB_DAT;
1293 else
1294 reloc_type = R_C60_JMP_SLOT;
1295 put_got_entry(s1, reloc_type, sym->st_size, sym->st_info,
1296 sym_index);
1298 break;
1299 #elif defined(TCC_TARGET_X86_64)
1300 case R_X86_64_GOT32:
1301 case R_X86_64_GOTTPOFF:
1302 case R_X86_64_GOTPCREL:
1303 case R_X86_64_PLT32:
1304 sym_index = ELFW(R_SYM)(rel->r_info);
1305 sym = &((ElfW(Sym) *)symtab_section->data)[sym_index];
1306 if (type == R_X86_64_PLT32 &&
1307 ELFW(ST_VISIBILITY)(sym->st_other) != STV_DEFAULT)
1309 rel->r_info = ELFW(R_INFO)(sym_index, R_X86_64_PC32);
1310 break;
1313 if (!s1->got)
1314 build_got(s1);
1315 if (type == R_X86_64_GOT32 || type == R_X86_64_GOTPCREL ||
1316 type == R_X86_64_PLT32) {
1317 unsigned long ofs;
1318 /* look at the symbol got offset. If none, then add one */
1319 if (type == R_X86_64_GOT32 || type == R_X86_64_GOTPCREL)
1320 reloc_type = R_X86_64_GLOB_DAT;
1321 else
1322 reloc_type = R_X86_64_JUMP_SLOT;
1323 ofs = put_got_entry(s1, reloc_type, sym->st_size,
1324 sym->st_info, sym_index);
1325 if (type == R_X86_64_PLT32)
1326 /* We store the place of the generated PLT slot
1327 in our addend. */
1328 rel->r_addend += ofs;
1330 break;
1331 #else
1332 #error unsupported CPU
1333 #endif
1334 default:
1335 break;
1341 ST_FUNC Section *new_symtab(TCCState *s1,
1342 const char *symtab_name, int sh_type, int sh_flags,
1343 const char *strtab_name,
1344 const char *hash_name, int hash_sh_flags)
1346 Section *symtab, *strtab, *hash;
1347 int *ptr, nb_buckets;
1349 symtab = new_section(s1, symtab_name, sh_type, sh_flags);
1350 symtab->sh_entsize = sizeof(ElfW(Sym));
1351 strtab = new_section(s1, strtab_name, SHT_STRTAB, sh_flags);
1352 put_elf_str(strtab, "");
1353 symtab->link = strtab;
1354 put_elf_sym(symtab, 0, 0, 0, 0, 0, NULL);
1356 nb_buckets = 1;
1358 hash = new_section(s1, hash_name, SHT_HASH, hash_sh_flags);
1359 hash->sh_entsize = sizeof(int);
1360 symtab->hash = hash;
1361 hash->link = symtab;
1363 ptr = section_ptr_add(hash, (2 + nb_buckets + 1) * sizeof(int));
1364 ptr[0] = nb_buckets;
1365 ptr[1] = 1;
1366 memset(ptr + 2, 0, (nb_buckets + 1) * sizeof(int));
1367 return symtab;
1370 /* put dynamic tag */
1371 static void put_dt(Section *dynamic, int dt, addr_t val)
1373 ElfW(Dyn) *dyn;
1374 dyn = section_ptr_add(dynamic, sizeof(ElfW(Dyn)));
1375 dyn->d_tag = dt;
1376 dyn->d_un.d_val = val;
1379 static void add_init_array_defines(TCCState *s1, const char *section_name)
1381 Section *s;
1382 long end_offset;
1383 char sym_start[1024];
1384 char sym_end[1024];
1386 snprintf(sym_start, sizeof(sym_start), "__%s_start", section_name + 1);
1387 snprintf(sym_end, sizeof(sym_end), "__%s_end", section_name + 1);
1389 s = find_section(s1, section_name);
1390 if (!s) {
1391 end_offset = 0;
1392 s = data_section;
1393 } else {
1394 end_offset = s->data_offset;
1397 add_elf_sym(symtab_section,
1398 0, 0,
1399 ELFW(ST_INFO)(STB_GLOBAL, STT_NOTYPE), 0,
1400 s->sh_num, sym_start);
1401 add_elf_sym(symtab_section,
1402 end_offset, 0,
1403 ELFW(ST_INFO)(STB_GLOBAL, STT_NOTYPE), 0,
1404 s->sh_num, sym_end);
1407 ST_FUNC void tcc_add_bcheck(TCCState *s1)
1409 #ifdef CONFIG_TCC_BCHECK
1410 unsigned long *ptr;
1411 Section *init_section;
1412 unsigned char *pinit;
1413 int sym_index;
1415 if (0 == s1->do_bounds_check)
1416 return;
1418 /* XXX: add an object file to do that */
1419 ptr = section_ptr_add(bounds_section, sizeof(unsigned long));
1420 *ptr = 0;
1421 add_elf_sym(symtab_section, 0, 0,
1422 ELFW(ST_INFO)(STB_GLOBAL, STT_NOTYPE), 0,
1423 bounds_section->sh_num, "__bounds_start");
1424 #ifdef TCC_TARGET_I386
1425 if (s1->output_type != TCC_OUTPUT_MEMORY) {
1426 /* add 'call __bound_init()' in .init section */
1427 init_section = find_section(s1, ".init");
1428 pinit = section_ptr_add(init_section, 5);
1429 pinit[0] = 0xe8;
1430 put32(pinit + 1, -4);
1431 sym_index = find_elf_sym(symtab_section, "__bound_init");
1432 put_elf_reloc(symtab_section, init_section,
1433 init_section->data_offset - 4, R_386_PC32, sym_index);
1435 #endif
1436 #endif
1439 static inline int tcc_add_support(TCCState *s1, const char *filename)
1441 char buf[1024];
1442 snprintf(buf, sizeof(buf), "%s/%s", s1->tcc_lib_path, filename);
1443 return tcc_add_file(s1, buf);
1446 /* add tcc runtime libraries */
1447 ST_FUNC void tcc_add_runtime(TCCState *s1)
1449 /* add libc */
1450 if (!s1->nostdlib) {
1451 tcc_add_library(s1, "c");
1452 #ifdef CONFIG_USE_LIBGCC
1453 if (!s1->static_link) {
1454 tcc_add_file(s1, TCC_LIBGCC);
1455 tcc_add_support(s1, "libtcc1.a");
1456 } else
1457 tcc_add_support(s1, "libtcc1.a");
1458 #else
1459 tcc_add_support(s1, "libtcc1.a");
1460 #endif
1463 /* tcc_add_bcheck tries to relocate a call to __bound_init in _init so
1464 libtcc1.a must be loaded before for __bound_init to be defined and
1465 crtn.o must be loaded after to not finalize _init too early. */
1466 tcc_add_bcheck(s1);
1468 if (!s1->nostdlib) {
1469 /* add crt end if not memory output */
1470 if (s1->output_type != TCC_OUTPUT_MEMORY)
1471 tcc_add_crt(s1, "crtn.o");
1475 /* add various standard linker symbols (must be done after the
1476 sections are filled (for example after allocating common
1477 symbols)) */
1478 ST_FUNC void tcc_add_linker_symbols(TCCState *s1)
1480 char buf[1024];
1481 int i;
1482 Section *s;
1484 add_elf_sym(symtab_section,
1485 text_section->data_offset, 0,
1486 ELFW(ST_INFO)(STB_GLOBAL, STT_NOTYPE), 0,
1487 text_section->sh_num, "_etext");
1488 add_elf_sym(symtab_section,
1489 data_section->data_offset, 0,
1490 ELFW(ST_INFO)(STB_GLOBAL, STT_NOTYPE), 0,
1491 data_section->sh_num, "_edata");
1492 add_elf_sym(symtab_section,
1493 bss_section->data_offset, 0,
1494 ELFW(ST_INFO)(STB_GLOBAL, STT_NOTYPE), 0,
1495 bss_section->sh_num, "_end");
1496 /* horrible new standard ldscript defines */
1497 add_init_array_defines(s1, ".preinit_array");
1498 add_init_array_defines(s1, ".init_array");
1499 add_init_array_defines(s1, ".fini_array");
1501 /* add start and stop symbols for sections whose name can be
1502 expressed in C */
1503 for(i = 1; i < s1->nb_sections; i++) {
1504 s = s1->sections[i];
1505 if (s->sh_type == SHT_PROGBITS &&
1506 (s->sh_flags & SHF_ALLOC)) {
1507 const char *p;
1508 int ch;
1510 /* check if section name can be expressed in C */
1511 p = s->name;
1512 for(;;) {
1513 ch = *p;
1514 if (!ch)
1515 break;
1516 if (!isid(ch) && !isnum(ch))
1517 goto next_sec;
1518 p++;
1520 snprintf(buf, sizeof(buf), "__start_%s", s->name);
1521 add_elf_sym(symtab_section,
1522 0, 0,
1523 ELFW(ST_INFO)(STB_GLOBAL, STT_NOTYPE), 0,
1524 s->sh_num, buf);
1525 snprintf(buf, sizeof(buf), "__stop_%s", s->name);
1526 add_elf_sym(symtab_section,
1527 s->data_offset, 0,
1528 ELFW(ST_INFO)(STB_GLOBAL, STT_NOTYPE), 0,
1529 s->sh_num, buf);
1531 next_sec: ;
1535 static void tcc_output_binary(TCCState *s1, FILE *f,
1536 const int *sec_order)
1538 Section *s;
1539 int i, offset, size;
1541 offset = 0;
1542 for(i=1;i<s1->nb_sections;i++) {
1543 s = s1->sections[sec_order[i]];
1544 if (s->sh_type != SHT_NOBITS &&
1545 (s->sh_flags & SHF_ALLOC)) {
1546 while (offset < s->sh_offset) {
1547 fputc(0, f);
1548 offset++;
1550 size = s->sh_size;
1551 fwrite(s->data, 1, size, f);
1552 offset += size;
1557 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
1558 #define HAVE_PHDR 1
1559 #define EXTRA_RELITEMS 14
1561 /* move the relocation value from .dynsym to .got */
1562 void patch_dynsym_undef(TCCState *s1, Section *s)
1564 uint32_t *gotd = (void *)s1->got->data;
1565 ElfW(Sym) *sym;
1567 gotd += 3; /* dummy entries in .got */
1568 /* relocate symbols in .dynsym */
1569 for_each_elem(s, 1, sym, ElfW(Sym)) {
1570 if (sym->st_shndx == SHN_UNDEF) {
1571 *gotd++ = sym->st_value + 6; /* XXX 6 is magic ? */
1572 sym->st_value = 0;
1576 #else
1577 #define HAVE_PHDR 1
1578 #define EXTRA_RELITEMS 9
1580 /* zero plt offsets of weak symbols in .dynsym */
1581 void patch_dynsym_undef(TCCState *s1, Section *s)
1583 ElfW(Sym) *sym;
1585 for_each_elem(s, 1, sym, ElfW(Sym))
1586 if (sym->st_shndx == SHN_UNDEF && ELFW(ST_BIND)(sym->st_info) == STB_WEAK)
1587 sym->st_value = 0;
1589 #endif
1591 ST_FUNC void fill_got_entry(TCCState *s1, ElfW_Rel *rel)
1593 int sym_index = ELFW(R_SYM) (rel->r_info);
1594 ElfW(Sym) *sym = &((ElfW(Sym) *) symtab_section->data)[sym_index];
1595 unsigned long offset;
1597 if (sym_index >= s1->nb_sym_attrs)
1598 return;
1599 offset = s1->sym_attrs[sym_index].got_offset;
1600 section_reserve(s1->got, offset + PTR_SIZE);
1601 #ifdef TCC_TARGET_X86_64
1602 /* only works for x86-64 */
1603 put32(s1->got->data + offset + 4, sym->st_value >> 32);
1604 #endif
1605 put32(s1->got->data + offset, sym->st_value & 0xffffffff);
1608 /* Perform relocation to GOT or PLT entries */
1609 ST_FUNC void fill_got(TCCState *s1)
1611 Section *s;
1612 ElfW_Rel *rel;
1613 int i;
1615 for(i = 1; i < s1->nb_sections; i++) {
1616 s = s1->sections[i];
1617 if (s->sh_type != SHT_RELX)
1618 continue;
1619 /* no need to handle got relocations */
1620 if (s->link != symtab_section)
1621 continue;
1622 for_each_elem(s, 0, rel, ElfW_Rel) {
1623 switch (ELFW(R_TYPE) (rel->r_info)) {
1624 case R_X86_64_GOT32:
1625 case R_X86_64_GOTPCREL:
1626 case R_X86_64_PLT32:
1627 fill_got_entry(s1, rel);
1628 break;
1634 /* Bind symbols of executable: resolve undefined symbols from exported symbols
1635 in shared libraries and export non local defined symbols to shared libraries
1636 if -rdynamic switch was given on command line */
1637 static void bind_exe_dynsyms(TCCState *s1)
1639 const char *name;
1640 int sym_index, index;
1641 ElfW(Sym) *sym, *esym;
1642 int type;
1644 /* Resolve undefined symbols from dynamic symbols. When there is a match:
1645 - if STT_FUNC or STT_GNU_IFUNC symbol -> add it in PLT
1646 - if STT_OBJECT symbol -> add it in .bss section with suitable reloc */
1647 for_each_elem(symtab_section, 1, sym, ElfW(Sym)) {
1648 if (sym->st_shndx == SHN_UNDEF) {
1649 name = (char *) symtab_section->link->data + sym->st_name;
1650 sym_index = find_elf_sym(s1->dynsymtab_section, name);
1651 if (sym_index) {
1652 esym = &((ElfW(Sym) *)s1->dynsymtab_section->data)[sym_index];
1653 type = ELFW(ST_TYPE)(esym->st_info);
1654 if ((type == STT_FUNC) || (type == STT_GNU_IFUNC)) {
1655 /* Indirect functions shall have STT_FUNC type in executable
1656 * dynsym section. Indeed, a dlsym call following a lazy
1657 * resolution would pick the symbol value from the
1658 * executable dynsym entry which would contain the address
1659 * of the function wanted by the caller of dlsym instead of
1660 * the address of the function that would return that
1661 * address */
1662 put_got_entry(s1, R_JMP_SLOT, esym->st_size,
1663 ELFW(ST_INFO)(STB_GLOBAL,STT_FUNC),
1664 sym - (ElfW(Sym) *)symtab_section->data);
1665 } else if (type == STT_OBJECT) {
1666 unsigned long offset;
1667 ElfW(Sym) *dynsym;
1668 offset = bss_section->data_offset;
1669 /* XXX: which alignment ? */
1670 offset = (offset + 16 - 1) & -16;
1671 index = put_elf_sym(s1->dynsym, offset, esym->st_size,
1672 esym->st_info, 0, bss_section->sh_num,
1673 name);
1674 /* Ensure R_COPY works for weak symbol aliases */
1675 if (ELFW(ST_BIND)(esym->st_info) == STB_WEAK) {
1676 for_each_elem(s1->dynsymtab_section, 1, dynsym, ElfW(Sym)) {
1677 if ((dynsym->st_value == esym->st_value)
1678 && (ELFW(ST_BIND)(dynsym->st_info) == STB_GLOBAL)) {
1679 char *dynname = (char *) s1->dynsymtab_section->link->data
1680 + dynsym->st_name;
1681 put_elf_sym(s1->dynsym, offset, dynsym->st_size,
1682 dynsym->st_info, 0,
1683 bss_section->sh_num, dynname);
1684 break;
1688 put_elf_reloc(s1->dynsym, bss_section,
1689 offset, R_COPY, index);
1690 offset += esym->st_size;
1691 bss_section->data_offset = offset;
1693 } else {
1694 /* STB_WEAK undefined symbols are accepted */
1695 /* XXX: _fp_hw seems to be part of the ABI, so we ignore it */
1696 if (ELFW(ST_BIND)(sym->st_info) == STB_WEAK ||
1697 !strcmp(name, "_fp_hw")) {
1698 } else {
1699 tcc_error_noabort("undefined symbol '%s'", name);
1702 } else if (s1->rdynamic && ELFW(ST_BIND)(sym->st_info) != STB_LOCAL) {
1703 /* if -rdynamic option, then export all non local symbols */
1704 name = (char *) symtab_section->link->data + sym->st_name;
1705 put_elf_sym(s1->dynsym, sym->st_value, sym->st_size, sym->st_info,
1706 0, sym->st_shndx, name);
1711 /* Bind symbols of libraries: export non local symbols of executable that
1712 resolve undefined symbols of shared libraries */
1713 static void bind_libs_dynsyms(TCCState *s1)
1715 const char *name;
1716 int sym_index;
1717 ElfW(Sym) *sym, *esym;
1719 /* now look at unresolved dynamic symbols and export
1720 corresponding symbol */
1721 for_each_elem(s1->dynsymtab_section, 1, esym, ElfW(Sym)) {
1722 if (esym->st_shndx == SHN_UNDEF) {
1723 name = (char *) s1->dynsymtab_section->link->data + esym->st_name;
1724 sym_index = find_elf_sym(symtab_section, name);
1725 if (sym_index) {
1726 /* XXX: avoid adding a symbol if already present because of
1727 -rdynamic ? */
1728 sym = &((ElfW(Sym) *)symtab_section->data)[sym_index];
1729 put_elf_sym(s1->dynsym, sym->st_value, sym->st_size,
1730 sym->st_info, 0, sym->st_shndx, name);
1731 } else {
1732 /* weak symbols can stay undefined */
1733 if (ELFW(ST_BIND)(esym->st_info) != STB_WEAK)
1734 tcc_warning("undefined dynamic symbol '%s'", name);
1740 /* Export all non local symbols (for shared libraries) */
1741 static void export_global_syms(TCCState *s1)
1743 int nb_syms, dynindex, index;
1744 const char *name;
1745 ElfW(Sym) *sym;
1747 nb_syms = symtab_section->data_offset / sizeof(ElfW(Sym));
1748 s1->symtab_to_dynsym = tcc_mallocz(sizeof(int) * nb_syms);
1749 for_each_elem(symtab_section, 1, sym, ElfW(Sym)) {
1750 if (ELFW(ST_BIND)(sym->st_info) != STB_LOCAL) {
1751 name = (char *) symtab_section->link->data + sym->st_name;
1752 dynindex = put_elf_sym(s1->dynsym, sym->st_value, sym->st_size,
1753 sym->st_info, 0, sym->st_shndx, name);
1754 index = sym - (ElfW(Sym) *) symtab_section->data;
1755 s1->symtab_to_dynsym[index] = dynindex;
1760 /* relocate the PLT: compute addresses and offsets in the PLT now that final
1761 address for PLT and GOT are known (see fill_program_header) */
1762 ST_FUNC void relocate_plt(TCCState *s1)
1764 uint8_t *p, *p_end;
1766 if (!s1->plt)
1767 return;
1769 p = s1->plt->data;
1770 p_end = p + s1->plt->data_offset;
1771 if (p < p_end) {
1772 #if defined(TCC_TARGET_I386)
1773 put32(p + 2, get32(p + 2) + s1->got->sh_addr);
1774 put32(p + 8, get32(p + 8) + s1->got->sh_addr);
1775 p += 16;
1776 while (p < p_end) {
1777 put32(p + 2, get32(p + 2) + s1->got->sh_addr);
1778 p += 16;
1780 #elif defined(TCC_TARGET_X86_64)
1781 int x = s1->got->sh_addr - s1->plt->sh_addr - 6;
1782 put32(p + 2, get32(p + 2) + x);
1783 put32(p + 8, get32(p + 8) + x - 6);
1784 p += 16;
1785 while (p < p_end) {
1786 put32(p + 2, get32(p + 2) + x + s1->plt->data - p);
1787 p += 16;
1789 #elif defined(TCC_TARGET_ARM)
1790 int x;
1791 x=s1->got->sh_addr - s1->plt->sh_addr - 12;
1792 p += 16;
1793 while (p < p_end) {
1794 if (get32(p) == 0x46c04778) /* PLT Thumb stub present */
1795 p += 4;
1796 put32(p + 12, x + get32(p + 12) + s1->plt->data - p);
1797 p += 16;
1799 #elif defined(TCC_TARGET_C67)
1800 /* XXX: TODO */
1801 #else
1802 #error unsupported CPU
1803 #endif
1807 /* Allocate strings for section names and decide if an unallocated section
1808 should be output.
1810 NOTE: the strsec section comes last, so its size is also correct ! */
1811 static void alloc_sec_names(TCCState *s1, int file_type, Section *strsec)
1813 int i;
1814 Section *s;
1816 /* Allocate strings for section names */
1817 for(i = 1; i < s1->nb_sections; i++) {
1818 s = s1->sections[i];
1819 s->sh_name = put_elf_str(strsec, s->name);
1820 /* when generating a DLL, we include relocations but we may
1821 patch them */
1822 if (file_type == TCC_OUTPUT_DLL &&
1823 s->sh_type == SHT_RELX &&
1824 !(s->sh_flags & SHF_ALLOC)) {
1825 /* gr: avoid bogus relocs for empty (debug) sections */
1826 if (s1->sections[s->sh_info]->sh_flags & SHF_ALLOC)
1827 prepare_dynamic_rel(s1, s);
1828 else if (s1->do_debug)
1829 s->sh_size = s->data_offset;
1830 } else if (s1->do_debug ||
1831 file_type == TCC_OUTPUT_OBJ ||
1832 (s->sh_flags & SHF_ALLOC) ||
1833 i == (s1->nb_sections - 1)) {
1834 /* we output all sections if debug or object file */
1835 s->sh_size = s->data_offset;
1840 /* Info to be copied in dynamic section */
1841 struct dyn_inf {
1842 Section *dynamic;
1843 Section *dynstr;
1844 unsigned long dyn_rel_off;
1845 addr_t rel_addr;
1846 addr_t rel_size;
1847 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
1848 addr_t bss_addr;
1849 addr_t bss_size;
1850 #endif
1853 /* Assign sections to segments and decide how are sections laid out when loaded
1854 in memory. This function also fills corresponding program headers. */
1855 static int layout_sections(TCCState *s1, ElfW(Phdr) *phdr, int phnum,
1856 Section *interp, Section* strsec,
1857 struct dyn_inf *dyninf, int *sec_order)
1859 int i, j, k, file_type, sh_order_index, file_offset;
1860 unsigned long s_align;
1861 long long tmp;
1862 addr_t addr;
1863 ElfW(Phdr) *ph;
1864 Section *s;
1866 file_type = s1->output_type;
1867 sh_order_index = 1;
1868 file_offset = 0;
1869 if (s1->output_format == TCC_OUTPUT_FORMAT_ELF)
1870 file_offset = sizeof(ElfW(Ehdr)) + phnum * sizeof(ElfW(Phdr));
1871 s_align = ELF_PAGE_SIZE;
1872 if (s1->section_align)
1873 s_align = s1->section_align;
1875 if (phnum > 0) {
1876 if (s1->has_text_addr) {
1877 int a_offset, p_offset;
1878 addr = s1->text_addr;
1879 /* we ensure that (addr % ELF_PAGE_SIZE) == file_offset %
1880 ELF_PAGE_SIZE */
1881 a_offset = (int) (addr & (s_align - 1));
1882 p_offset = file_offset & (s_align - 1);
1883 if (a_offset < p_offset)
1884 a_offset += s_align;
1885 file_offset += (a_offset - p_offset);
1886 } else {
1887 if (file_type == TCC_OUTPUT_DLL)
1888 addr = 0;
1889 else
1890 addr = ELF_START_ADDR;
1891 /* compute address after headers */
1892 addr += (file_offset & (s_align - 1));
1895 ph = &phdr[0];
1896 /* Leave one program headers for the program interpreter and one for
1897 the program header table itself if needed. These are done later as
1898 they require section layout to be done first. */
1899 if (interp)
1900 ph += 1 + HAVE_PHDR;
1902 /* dynamic relocation table information, for .dynamic section */
1903 dyninf->rel_addr = dyninf->rel_size = 0;
1904 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
1905 dyninf->bss_addr = dyninf->bss_size = 0;
1906 #endif
1908 for(j = 0; j < 2; j++) {
1909 ph->p_type = PT_LOAD;
1910 if (j == 0)
1911 ph->p_flags = PF_R | PF_X;
1912 else
1913 ph->p_flags = PF_R | PF_W;
1914 ph->p_align = s_align;
1916 /* Decide the layout of sections loaded in memory. This must
1917 be done before program headers are filled since they contain
1918 info about the layout. We do the following ordering: interp,
1919 symbol tables, relocations, progbits, nobits */
1920 /* XXX: do faster and simpler sorting */
1921 for(k = 0; k < 5; k++) {
1922 for(i = 1; i < s1->nb_sections; i++) {
1923 s = s1->sections[i];
1924 /* compute if section should be included */
1925 if (j == 0) {
1926 if ((s->sh_flags & (SHF_ALLOC | SHF_WRITE)) !=
1927 SHF_ALLOC)
1928 continue;
1929 } else {
1930 if ((s->sh_flags & (SHF_ALLOC | SHF_WRITE)) !=
1931 (SHF_ALLOC | SHF_WRITE))
1932 continue;
1934 if (s == interp) {
1935 if (k != 0)
1936 continue;
1937 } else if (s->sh_type == SHT_DYNSYM ||
1938 s->sh_type == SHT_STRTAB ||
1939 s->sh_type == SHT_HASH) {
1940 if (k != 1)
1941 continue;
1942 } else if (s->sh_type == SHT_RELX) {
1943 if (k != 2)
1944 continue;
1945 } else if (s->sh_type == SHT_NOBITS) {
1946 if (k != 4)
1947 continue;
1948 } else {
1949 if (k != 3)
1950 continue;
1952 sec_order[sh_order_index++] = i;
1954 /* section matches: we align it and add its size */
1955 tmp = addr;
1956 addr = (addr + s->sh_addralign - 1) &
1957 ~(s->sh_addralign - 1);
1958 file_offset += (int) ( addr - tmp );
1959 s->sh_offset = file_offset;
1960 s->sh_addr = addr;
1962 /* update program header infos */
1963 if (ph->p_offset == 0) {
1964 ph->p_offset = file_offset;
1965 ph->p_vaddr = addr;
1966 ph->p_paddr = ph->p_vaddr;
1968 /* update dynamic relocation infos */
1969 if (s->sh_type == SHT_RELX) {
1970 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
1971 if (!strcmp(strsec->data + s->sh_name, ".rel.got")) {
1972 dyninf->rel_addr = addr;
1973 dyninf->rel_size += s->sh_size; /* XXX only first rel. */
1975 if (!strcmp(strsec->data + s->sh_name, ".rel.bss")) {
1976 dyninf->bss_addr = addr;
1977 dyninf->bss_size = s->sh_size; /* XXX only first rel. */
1979 #else
1980 if (dyninf->rel_size == 0)
1981 dyninf->rel_addr = addr;
1982 dyninf->rel_size += s->sh_size;
1983 #endif
1985 addr += s->sh_size;
1986 if (s->sh_type != SHT_NOBITS)
1987 file_offset += s->sh_size;
1990 if (j == 0) {
1991 /* Make the first PT_LOAD segment include the program
1992 headers itself (and the ELF header as well), it'll
1993 come out with same memory use but will make various
1994 tools like binutils strip work better. */
1995 ph->p_offset &= ~(ph->p_align - 1);
1996 ph->p_vaddr &= ~(ph->p_align - 1);
1997 ph->p_paddr &= ~(ph->p_align - 1);
1999 ph->p_filesz = file_offset - ph->p_offset;
2000 ph->p_memsz = addr - ph->p_vaddr;
2001 ph++;
2002 if (j == 0) {
2003 if (s1->output_format == TCC_OUTPUT_FORMAT_ELF) {
2004 /* if in the middle of a page, we duplicate the page in
2005 memory so that one copy is RX and the other is RW */
2006 if ((addr & (s_align - 1)) != 0)
2007 addr += s_align;
2008 } else {
2009 addr = (addr + s_align - 1) & ~(s_align - 1);
2010 file_offset = (file_offset + s_align - 1) & ~(s_align - 1);
2016 /* all other sections come after */
2017 for(i = 1; i < s1->nb_sections; i++) {
2018 s = s1->sections[i];
2019 if (phnum > 0 && (s->sh_flags & SHF_ALLOC))
2020 continue;
2021 sec_order[sh_order_index++] = i;
2023 file_offset = (file_offset + s->sh_addralign - 1) &
2024 ~(s->sh_addralign - 1);
2025 s->sh_offset = file_offset;
2026 if (s->sh_type != SHT_NOBITS)
2027 file_offset += s->sh_size;
2030 return file_offset;
2033 static void fill_unloadable_phdr(ElfW(Phdr) *phdr, int phnum, Section *interp,
2034 Section *dynamic)
2036 ElfW(Phdr) *ph;
2038 /* if interpreter, then add corresponding program header */
2039 if (interp) {
2040 ph = &phdr[0];
2042 if (HAVE_PHDR)
2044 int len = phnum * sizeof(ElfW(Phdr));
2046 ph->p_type = PT_PHDR;
2047 ph->p_offset = sizeof(ElfW(Ehdr));
2048 ph->p_vaddr = interp->sh_addr - len;
2049 ph->p_paddr = ph->p_vaddr;
2050 ph->p_filesz = ph->p_memsz = len;
2051 ph->p_flags = PF_R | PF_X;
2052 ph->p_align = 4; /* interp->sh_addralign; */
2053 ph++;
2056 ph->p_type = PT_INTERP;
2057 ph->p_offset = interp->sh_offset;
2058 ph->p_vaddr = interp->sh_addr;
2059 ph->p_paddr = ph->p_vaddr;
2060 ph->p_filesz = interp->sh_size;
2061 ph->p_memsz = interp->sh_size;
2062 ph->p_flags = PF_R;
2063 ph->p_align = interp->sh_addralign;
2066 /* if dynamic section, then add corresponding program header */
2067 if (dynamic) {
2068 ph = &phdr[phnum - 1];
2070 ph->p_type = PT_DYNAMIC;
2071 ph->p_offset = dynamic->sh_offset;
2072 ph->p_vaddr = dynamic->sh_addr;
2073 ph->p_paddr = ph->p_vaddr;
2074 ph->p_filesz = dynamic->sh_size;
2075 ph->p_memsz = dynamic->sh_size;
2076 ph->p_flags = PF_R | PF_W;
2077 ph->p_align = dynamic->sh_addralign;
2081 /* Fill the dynamic section with tags describing the address and size of
2082 sections */
2083 static void fill_dynamic(TCCState *s1, struct dyn_inf *dyninf)
2085 Section *dynamic;
2087 dynamic = dyninf->dynamic;
2089 /* put dynamic section entries */
2090 dynamic->data_offset = dyninf->dyn_rel_off;
2091 put_dt(dynamic, DT_HASH, s1->dynsym->hash->sh_addr);
2092 put_dt(dynamic, DT_STRTAB, dyninf->dynstr->sh_addr);
2093 put_dt(dynamic, DT_SYMTAB, s1->dynsym->sh_addr);
2094 put_dt(dynamic, DT_STRSZ, dyninf->dynstr->data_offset);
2095 put_dt(dynamic, DT_SYMENT, sizeof(ElfW(Sym)));
2096 #ifdef TCC_TARGET_X86_64
2097 put_dt(dynamic, DT_RELA, dyninf->rel_addr);
2098 put_dt(dynamic, DT_RELASZ, dyninf->rel_size);
2099 put_dt(dynamic, DT_RELAENT, sizeof(ElfW_Rel));
2100 #else
2101 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
2102 put_dt(dynamic, DT_PLTGOT, s1->got->sh_addr);
2103 put_dt(dynamic, DT_PLTRELSZ, dyninf->rel_size);
2104 put_dt(dynamic, DT_JMPREL, dyninf->rel_addr);
2105 put_dt(dynamic, DT_PLTREL, DT_REL);
2106 put_dt(dynamic, DT_REL, dyninf->bss_addr);
2107 put_dt(dynamic, DT_RELSZ, dyninf->bss_size);
2108 #else
2109 put_dt(dynamic, DT_REL, dyninf->rel_addr);
2110 put_dt(dynamic, DT_RELSZ, dyninf->rel_size);
2111 put_dt(dynamic, DT_RELENT, sizeof(ElfW_Rel));
2112 #endif
2113 #endif
2114 if (s1->do_debug)
2115 put_dt(dynamic, DT_DEBUG, 0);
2116 put_dt(dynamic, DT_NULL, 0);
2119 /* Relocate remaining sections and symbols (that is those not related to
2120 dynamic linking) */
2121 static int final_sections_reloc(TCCState *s1)
2123 int i;
2124 Section *s;
2126 relocate_syms(s1, 0);
2128 if (s1->nb_errors != 0)
2129 return -1;
2131 /* relocate sections */
2132 /* XXX: ignore sections with allocated relocations ? */
2133 for(i = 1; i < s1->nb_sections; i++) {
2134 s = s1->sections[i];
2135 if (s->reloc && s != s1->got)
2136 relocate_section(s1, s);
2139 /* relocate relocation entries if the relocation tables are
2140 allocated in the executable */
2141 for(i = 1; i < s1->nb_sections; i++) {
2142 s = s1->sections[i];
2143 if ((s->sh_flags & SHF_ALLOC) &&
2144 s->sh_type == SHT_RELX) {
2145 relocate_rel(s1, s);
2148 return 0;
2151 /* Create an ELF file on disk.
2152 This function handle ELF specific layout requirements */
2153 static void tcc_output_elf(TCCState *s1, FILE *f, int phnum, ElfW(Phdr) *phdr,
2154 int file_offset, int *sec_order)
2156 int i, shnum, offset, size, file_type;
2157 Section *s;
2158 ElfW(Ehdr) ehdr;
2159 ElfW(Shdr) shdr, *sh;
2161 file_type = s1->output_type;
2162 shnum = s1->nb_sections;
2164 memset(&ehdr, 0, sizeof(ehdr));
2166 if (phnum > 0) {
2167 ehdr.e_phentsize = sizeof(ElfW(Phdr));
2168 ehdr.e_phnum = phnum;
2169 ehdr.e_phoff = sizeof(ElfW(Ehdr));
2172 /* align to 4 */
2173 file_offset = (file_offset + 3) & -4;
2175 /* fill header */
2176 ehdr.e_ident[0] = ELFMAG0;
2177 ehdr.e_ident[1] = ELFMAG1;
2178 ehdr.e_ident[2] = ELFMAG2;
2179 ehdr.e_ident[3] = ELFMAG3;
2180 ehdr.e_ident[4] = ELFCLASSW;
2181 ehdr.e_ident[5] = ELFDATA2LSB;
2182 ehdr.e_ident[6] = EV_CURRENT;
2183 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
2184 ehdr.e_ident[EI_OSABI] = ELFOSABI_FREEBSD;
2185 #endif
2186 #ifdef TCC_TARGET_ARM
2187 #ifdef TCC_ARM_EABI
2188 ehdr.e_ident[EI_OSABI] = 0;
2189 ehdr.e_flags = EF_ARM_EABI_VER4;
2190 if (file_type == TCC_OUTPUT_EXE || file_type == TCC_OUTPUT_DLL)
2191 ehdr.e_flags |= EF_ARM_HASENTRY;
2192 if (s1->float_abi == ARM_HARD_FLOAT)
2193 ehdr.e_flags |= EF_ARM_VFP_FLOAT;
2194 else
2195 ehdr.e_flags |= EF_ARM_SOFT_FLOAT;
2196 #else
2197 ehdr.e_ident[EI_OSABI] = ELFOSABI_ARM;
2198 #endif
2199 #endif
2200 switch(file_type) {
2201 default:
2202 case TCC_OUTPUT_EXE:
2203 ehdr.e_type = ET_EXEC;
2204 ehdr.e_entry = get_elf_sym_addr(s1, "_start", 1);
2205 break;
2206 case TCC_OUTPUT_DLL:
2207 ehdr.e_type = ET_DYN;
2208 ehdr.e_entry = text_section->sh_addr; /* XXX: is it correct ? */
2209 break;
2210 case TCC_OUTPUT_OBJ:
2211 ehdr.e_type = ET_REL;
2212 break;
2214 ehdr.e_machine = EM_TCC_TARGET;
2215 ehdr.e_version = EV_CURRENT;
2216 ehdr.e_shoff = file_offset;
2217 ehdr.e_ehsize = sizeof(ElfW(Ehdr));
2218 ehdr.e_shentsize = sizeof(ElfW(Shdr));
2219 ehdr.e_shnum = shnum;
2220 ehdr.e_shstrndx = shnum - 1;
2222 fwrite(&ehdr, 1, sizeof(ElfW(Ehdr)), f);
2223 fwrite(phdr, 1, phnum * sizeof(ElfW(Phdr)), f);
2224 offset = sizeof(ElfW(Ehdr)) + phnum * sizeof(ElfW(Phdr));
2226 sort_syms(s1, symtab_section);
2227 for(i = 1; i < s1->nb_sections; i++) {
2228 s = s1->sections[sec_order[i]];
2229 if (s->sh_type != SHT_NOBITS) {
2230 if (s->sh_type == SHT_DYNSYM)
2231 patch_dynsym_undef(s1, s);
2232 while (offset < s->sh_offset) {
2233 fputc(0, f);
2234 offset++;
2236 size = s->sh_size;
2237 fwrite(s->data, 1, size, f);
2238 offset += size;
2242 /* output section headers */
2243 while (offset < ehdr.e_shoff) {
2244 fputc(0, f);
2245 offset++;
2248 for(i = 0; i < s1->nb_sections; i++) {
2249 sh = &shdr;
2250 memset(sh, 0, sizeof(ElfW(Shdr)));
2251 s = s1->sections[i];
2252 if (s) {
2253 sh->sh_name = s->sh_name;
2254 sh->sh_type = s->sh_type;
2255 sh->sh_flags = s->sh_flags;
2256 sh->sh_entsize = s->sh_entsize;
2257 sh->sh_info = s->sh_info;
2258 if (s->link)
2259 sh->sh_link = s->link->sh_num;
2260 sh->sh_addralign = s->sh_addralign;
2261 sh->sh_addr = s->sh_addr;
2262 sh->sh_offset = s->sh_offset;
2263 sh->sh_size = s->sh_size;
2265 fwrite(sh, 1, sizeof(ElfW(Shdr)), f);
2269 /* Write an elf, coff or "binary" file */
2270 static int tcc_write_elf_file(TCCState *s1, const char *filename, int phnum,
2271 ElfW(Phdr) *phdr, int file_offset, int *sec_order)
2273 int fd, mode, file_type;
2274 FILE *f;
2276 file_type = s1->output_type;
2277 if (file_type == TCC_OUTPUT_OBJ)
2278 mode = 0666;
2279 else
2280 mode = 0777;
2281 unlink(filename);
2282 fd = open(filename, O_WRONLY | O_CREAT | O_TRUNC | O_BINARY, mode);
2283 if (fd < 0) {
2284 tcc_error_noabort("could not write '%s'", filename);
2285 return -1;
2287 f = fdopen(fd, "wb");
2288 if (s1->verbose)
2289 printf("<- %s\n", filename);
2291 #ifdef TCC_TARGET_COFF
2292 if (s1->output_format == TCC_OUTPUT_FORMAT_COFF)
2293 tcc_output_coff(s1, f);
2294 else
2295 #endif
2296 if (s1->output_format == TCC_OUTPUT_FORMAT_ELF)
2297 tcc_output_elf(s1, f, phnum, phdr, file_offset, sec_order);
2298 else
2299 tcc_output_binary(s1, f, sec_order);
2300 fclose(f);
2302 return 0;
2305 /* Output an elf, coff or binary file */
2306 /* XXX: suppress unneeded sections */
2307 static int elf_output_file(TCCState *s1, const char *filename)
2309 int i, ret, phnum, shnum, file_type, file_offset, *sec_order;
2310 struct dyn_inf dyninf;
2311 ElfW(Phdr) *phdr;
2312 ElfW(Sym) *sym;
2313 Section *strsec, *interp, *dynamic, *dynstr;
2315 file_type = s1->output_type;
2316 s1->nb_errors = 0;
2318 /* if linking, also link in runtime libraries (libc, libgcc, etc.) */
2319 if (file_type != TCC_OUTPUT_OBJ) {
2320 tcc_add_runtime(s1);
2323 phdr = NULL;
2324 sec_order = NULL;
2325 interp = dynamic = dynstr = NULL; /* avoid warning */
2326 dyninf.dyn_rel_off = 0; /* avoid warning */
2328 if (file_type != TCC_OUTPUT_OBJ) {
2329 relocate_common_syms();
2331 tcc_add_linker_symbols(s1);
2333 if (!s1->static_link) {
2334 if (file_type == TCC_OUTPUT_EXE) {
2335 char *ptr;
2336 /* allow override the dynamic loader */
2337 const char *elfint = getenv("LD_SO");
2338 if (elfint == NULL)
2339 elfint = DEFAULT_ELFINTERP(s1);
2340 /* add interpreter section only if executable */
2341 interp = new_section(s1, ".interp", SHT_PROGBITS, SHF_ALLOC);
2342 interp->sh_addralign = 1;
2343 ptr = section_ptr_add(interp, 1 + strlen(elfint));
2344 strcpy(ptr, elfint);
2347 /* add dynamic symbol table */
2348 s1->dynsym = new_symtab(s1, ".dynsym", SHT_DYNSYM, SHF_ALLOC,
2349 ".dynstr",
2350 ".hash", SHF_ALLOC);
2351 dynstr = s1->dynsym->link;
2353 /* add dynamic section */
2354 dynamic = new_section(s1, ".dynamic", SHT_DYNAMIC,
2355 SHF_ALLOC | SHF_WRITE);
2356 dynamic->link = dynstr;
2357 dynamic->sh_entsize = sizeof(ElfW(Dyn));
2359 build_got(s1);
2361 if (file_type == TCC_OUTPUT_EXE) {
2362 bind_exe_dynsyms(s1);
2364 if (s1->nb_errors) {
2365 ret = -1;
2366 goto the_end;
2369 bind_libs_dynsyms(s1);
2370 } else /* shared library case: simply export all global symbols */
2371 export_global_syms(s1);
2373 build_got_entries(s1);
2375 /* add a list of needed dlls */
2376 for(i = 0; i < s1->nb_loaded_dlls; i++) {
2377 DLLReference *dllref = s1->loaded_dlls[i];
2378 if (dllref->level == 0)
2379 put_dt(dynamic, DT_NEEDED, put_elf_str(dynstr, dllref->name));
2382 if (s1->rpath)
2383 put_dt(dynamic, DT_RPATH, put_elf_str(dynstr, s1->rpath));
2385 /* XXX: currently, since we do not handle PIC code, we
2386 must relocate the readonly segments */
2387 if (file_type == TCC_OUTPUT_DLL) {
2388 if (s1->soname)
2389 put_dt(dynamic, DT_SONAME, put_elf_str(dynstr, s1->soname));
2390 put_dt(dynamic, DT_TEXTREL, 0);
2393 if (s1->symbolic)
2394 put_dt(dynamic, DT_SYMBOLIC, 0);
2396 /* add necessary space for other entries */
2397 dyninf.dyn_rel_off = dynamic->data_offset;
2398 dynamic->data_offset += sizeof(ElfW(Dyn)) * EXTRA_RELITEMS;
2399 } else {
2400 /* still need to build got entries in case of static link */
2401 build_got_entries(s1);
2405 /* we add a section for symbols */
2406 strsec = new_section(s1, ".shstrtab", SHT_STRTAB, 0);
2407 put_elf_str(strsec, "");
2409 /* compute number of sections */
2410 shnum = s1->nb_sections;
2412 /* this array is used to reorder sections in the output file */
2413 sec_order = tcc_malloc(sizeof(int) * shnum);
2414 sec_order[0] = 0;
2416 /* compute number of program headers */
2417 switch(file_type) {
2418 default:
2419 case TCC_OUTPUT_OBJ:
2420 phnum = 0;
2421 break;
2422 case TCC_OUTPUT_EXE:
2423 if (!s1->static_link)
2424 phnum = 4 + HAVE_PHDR;
2425 else
2426 phnum = 2;
2427 break;
2428 case TCC_OUTPUT_DLL:
2429 phnum = 3;
2430 break;
2433 /* Allocate strings for section names */
2434 alloc_sec_names(s1, file_type, strsec);
2436 /* allocate program segment headers */
2437 phdr = tcc_mallocz(phnum * sizeof(ElfW(Phdr)));
2439 /* compute section to program header mapping */
2440 file_offset = layout_sections(s1, phdr, phnum, interp, strsec, &dyninf,
2441 sec_order);
2443 /* Fill remaining program header and finalize relocation related to dynamic
2444 linking. */
2445 if (phnum > 0) {
2446 fill_unloadable_phdr(phdr, phnum, interp, dynamic);
2447 if (dynamic) {
2448 dyninf.dynamic = dynamic;
2449 dyninf.dynstr = dynstr;
2451 fill_dynamic(s1, &dyninf);
2453 /* put in GOT the dynamic section address and relocate PLT */
2454 put32(s1->got->data, dynamic->sh_addr);
2455 if (file_type == TCC_OUTPUT_EXE
2456 #if defined(TCC_OUTPUT_DLL_WITH_PLT)
2457 || file_type == TCC_OUTPUT_DLL
2458 #endif
2460 relocate_plt(s1);
2462 /* relocate symbols in .dynsym now that final addresses are known */
2463 for_each_elem(s1->dynsym, 1, sym, ElfW(Sym)) {
2464 /* relocate to PLT if symbol corresponds to a PLT entry */
2465 if (sym->st_shndx == SHN_UNDEF) {
2466 if (sym->st_value)
2467 sym->st_value += s1->plt->sh_addr;
2468 } else if (sym->st_shndx < SHN_LORESERVE) {
2469 /* do symbol relocation */
2470 sym->st_value += s1->sections[sym->st_shndx]->sh_addr;
2476 /* if building executable or DLL, then relocate each section
2477 except the GOT which is already relocated */
2478 if (file_type != TCC_OUTPUT_OBJ) {
2479 ret = final_sections_reloc(s1);
2480 if (ret)
2481 goto the_end;
2484 /* Perform relocation to GOT or PLT entries */
2485 if (file_type == TCC_OUTPUT_EXE && s1->static_link)
2486 fill_got(s1);
2488 /* Create the ELF file with name 'filename' */
2489 ret = tcc_write_elf_file(s1, filename, phnum, phdr, file_offset, sec_order);
2490 the_end:
2491 tcc_free(s1->symtab_to_dynsym);
2492 tcc_free(sec_order);
2493 tcc_free(phdr);
2494 tcc_free(s1->sym_attrs);
2495 s1->sym_attrs = NULL;
2496 return ret;
2499 LIBTCCAPI int tcc_output_file(TCCState *s, const char *filename)
2501 int ret;
2502 #ifdef TCC_TARGET_PE
2503 if (s->output_type != TCC_OUTPUT_OBJ) {
2504 ret = pe_output_file(s, filename);
2505 } else
2506 #endif
2507 ret = elf_output_file(s, filename);
2508 return ret;
2511 static void *load_data(int fd, unsigned long file_offset, unsigned long size)
2513 void *data;
2515 data = tcc_malloc(size);
2516 lseek(fd, file_offset, SEEK_SET);
2517 read(fd, data, size);
2518 return data;
2521 typedef struct SectionMergeInfo {
2522 Section *s; /* corresponding existing section */
2523 unsigned long offset; /* offset of the new section in the existing section */
2524 uint8_t new_section; /* true if section 's' was added */
2525 uint8_t link_once; /* true if link once section */
2526 } SectionMergeInfo;
2528 /* load an object file and merge it with current files */
2529 /* XXX: handle correctly stab (debug) info */
2530 ST_FUNC int tcc_load_object_file(TCCState *s1,
2531 int fd, unsigned long file_offset)
2533 ElfW(Ehdr) ehdr;
2534 ElfW(Shdr) *shdr, *sh;
2535 int size, i, j, offset, offseti, nb_syms, sym_index, ret;
2536 unsigned char *strsec, *strtab;
2537 int *old_to_new_syms;
2538 char *sh_name, *name;
2539 SectionMergeInfo *sm_table, *sm;
2540 ElfW(Sym) *sym, *symtab;
2541 ElfW_Rel *rel;
2542 Section *s;
2544 int stab_index;
2545 int stabstr_index;
2547 stab_index = stabstr_index = 0;
2549 if (read(fd, &ehdr, sizeof(ehdr)) != sizeof(ehdr))
2550 goto fail1;
2551 if (ehdr.e_ident[0] != ELFMAG0 ||
2552 ehdr.e_ident[1] != ELFMAG1 ||
2553 ehdr.e_ident[2] != ELFMAG2 ||
2554 ehdr.e_ident[3] != ELFMAG3)
2555 goto fail1;
2556 /* test if object file */
2557 if (ehdr.e_type != ET_REL)
2558 goto fail1;
2559 /* test CPU specific stuff */
2560 if (ehdr.e_ident[5] != ELFDATA2LSB ||
2561 ehdr.e_machine != EM_TCC_TARGET) {
2562 fail1:
2563 tcc_error_noabort("invalid object file");
2564 return -1;
2566 /* read sections */
2567 shdr = load_data(fd, file_offset + ehdr.e_shoff,
2568 sizeof(ElfW(Shdr)) * ehdr.e_shnum);
2569 sm_table = tcc_mallocz(sizeof(SectionMergeInfo) * ehdr.e_shnum);
2571 /* load section names */
2572 sh = &shdr[ehdr.e_shstrndx];
2573 strsec = load_data(fd, file_offset + sh->sh_offset, sh->sh_size);
2575 /* load symtab and strtab */
2576 old_to_new_syms = NULL;
2577 symtab = NULL;
2578 strtab = NULL;
2579 nb_syms = 0;
2580 for(i = 1; i < ehdr.e_shnum; i++) {
2581 sh = &shdr[i];
2582 if (sh->sh_type == SHT_SYMTAB) {
2583 if (symtab) {
2584 tcc_error_noabort("object must contain only one symtab");
2585 fail:
2586 ret = -1;
2587 goto the_end;
2589 nb_syms = sh->sh_size / sizeof(ElfW(Sym));
2590 symtab = load_data(fd, file_offset + sh->sh_offset, sh->sh_size);
2591 sm_table[i].s = symtab_section;
2593 /* now load strtab */
2594 sh = &shdr[sh->sh_link];
2595 strtab = load_data(fd, file_offset + sh->sh_offset, sh->sh_size);
2599 /* now examine each section and try to merge its content with the
2600 ones in memory */
2601 for(i = 1; i < ehdr.e_shnum; i++) {
2602 /* no need to examine section name strtab */
2603 if (i == ehdr.e_shstrndx)
2604 continue;
2605 sh = &shdr[i];
2606 sh_name = (char *) strsec + sh->sh_name;
2607 /* ignore sections types we do not handle */
2608 if (sh->sh_type != SHT_PROGBITS &&
2609 sh->sh_type != SHT_RELX &&
2610 #ifdef TCC_ARM_EABI
2611 sh->sh_type != SHT_ARM_EXIDX &&
2612 #endif
2613 sh->sh_type != SHT_NOBITS &&
2614 sh->sh_type != SHT_PREINIT_ARRAY &&
2615 sh->sh_type != SHT_INIT_ARRAY &&
2616 sh->sh_type != SHT_FINI_ARRAY &&
2617 strcmp(sh_name, ".stabstr")
2619 continue;
2620 if (sh->sh_addralign < 1)
2621 sh->sh_addralign = 1;
2622 /* find corresponding section, if any */
2623 for(j = 1; j < s1->nb_sections;j++) {
2624 s = s1->sections[j];
2625 if (!strcmp(s->name, sh_name)) {
2626 if (!strncmp(sh_name, ".gnu.linkonce",
2627 sizeof(".gnu.linkonce") - 1)) {
2628 /* if a 'linkonce' section is already present, we
2629 do not add it again. It is a little tricky as
2630 symbols can still be defined in
2631 it. */
2632 sm_table[i].link_once = 1;
2633 goto next;
2634 } else {
2635 goto found;
2639 /* not found: create new section */
2640 s = new_section(s1, sh_name, sh->sh_type, sh->sh_flags);
2641 /* take as much info as possible from the section. sh_link and
2642 sh_info will be updated later */
2643 s->sh_addralign = sh->sh_addralign;
2644 s->sh_entsize = sh->sh_entsize;
2645 sm_table[i].new_section = 1;
2646 found:
2647 if (sh->sh_type != s->sh_type) {
2648 tcc_error_noabort("invalid section type");
2649 goto fail;
2652 /* align start of section */
2653 offset = s->data_offset;
2655 if (0 == strcmp(sh_name, ".stab")) {
2656 stab_index = i;
2657 goto no_align;
2659 if (0 == strcmp(sh_name, ".stabstr")) {
2660 stabstr_index = i;
2661 goto no_align;
2664 size = sh->sh_addralign - 1;
2665 offset = (offset + size) & ~size;
2666 if (sh->sh_addralign > s->sh_addralign)
2667 s->sh_addralign = sh->sh_addralign;
2668 s->data_offset = offset;
2669 no_align:
2670 sm_table[i].offset = offset;
2671 sm_table[i].s = s;
2672 /* concatenate sections */
2673 size = sh->sh_size;
2674 if (sh->sh_type != SHT_NOBITS) {
2675 unsigned char *ptr;
2676 lseek(fd, file_offset + sh->sh_offset, SEEK_SET);
2677 ptr = section_ptr_add(s, size);
2678 read(fd, ptr, size);
2679 } else {
2680 s->data_offset += size;
2682 next: ;
2685 /* gr relocate stab strings */
2686 if (stab_index && stabstr_index) {
2687 Stab_Sym *a, *b;
2688 unsigned o;
2689 s = sm_table[stab_index].s;
2690 a = (Stab_Sym *)(s->data + sm_table[stab_index].offset);
2691 b = (Stab_Sym *)(s->data + s->data_offset);
2692 o = sm_table[stabstr_index].offset;
2693 while (a < b)
2694 a->n_strx += o, a++;
2697 /* second short pass to update sh_link and sh_info fields of new
2698 sections */
2699 for(i = 1; i < ehdr.e_shnum; i++) {
2700 s = sm_table[i].s;
2701 if (!s || !sm_table[i].new_section)
2702 continue;
2703 sh = &shdr[i];
2704 if (sh->sh_link > 0)
2705 s->link = sm_table[sh->sh_link].s;
2706 if (sh->sh_type == SHT_RELX) {
2707 s->sh_info = sm_table[sh->sh_info].s->sh_num;
2708 /* update backward link */
2709 s1->sections[s->sh_info]->reloc = s;
2712 sm = sm_table;
2714 /* resolve symbols */
2715 old_to_new_syms = tcc_mallocz(nb_syms * sizeof(int));
2717 sym = symtab + 1;
2718 for(i = 1; i < nb_syms; i++, sym++) {
2719 if (sym->st_shndx != SHN_UNDEF &&
2720 sym->st_shndx < SHN_LORESERVE) {
2721 sm = &sm_table[sym->st_shndx];
2722 if (sm->link_once) {
2723 /* if a symbol is in a link once section, we use the
2724 already defined symbol. It is very important to get
2725 correct relocations */
2726 if (ELFW(ST_BIND)(sym->st_info) != STB_LOCAL) {
2727 name = (char *) strtab + sym->st_name;
2728 sym_index = find_elf_sym(symtab_section, name);
2729 if (sym_index)
2730 old_to_new_syms[i] = sym_index;
2732 continue;
2734 /* if no corresponding section added, no need to add symbol */
2735 if (!sm->s)
2736 continue;
2737 /* convert section number */
2738 sym->st_shndx = sm->s->sh_num;
2739 /* offset value */
2740 sym->st_value += sm->offset;
2742 /* add symbol */
2743 name = (char *) strtab + sym->st_name;
2744 sym_index = add_elf_sym(symtab_section, sym->st_value, sym->st_size,
2745 sym->st_info, sym->st_other,
2746 sym->st_shndx, name);
2747 old_to_new_syms[i] = sym_index;
2750 /* third pass to patch relocation entries */
2751 for(i = 1; i < ehdr.e_shnum; i++) {
2752 s = sm_table[i].s;
2753 if (!s)
2754 continue;
2755 sh = &shdr[i];
2756 offset = sm_table[i].offset;
2757 switch(s->sh_type) {
2758 case SHT_RELX:
2759 /* take relocation offset information */
2760 offseti = sm_table[sh->sh_info].offset;
2761 for_each_elem(s, (offset / sizeof(*rel)), rel, ElfW_Rel) {
2762 int type;
2763 unsigned sym_index;
2764 /* convert symbol index */
2765 type = ELFW(R_TYPE)(rel->r_info);
2766 sym_index = ELFW(R_SYM)(rel->r_info);
2767 /* NOTE: only one symtab assumed */
2768 if (sym_index >= nb_syms)
2769 goto invalid_reloc;
2770 sym_index = old_to_new_syms[sym_index];
2771 /* ignore link_once in rel section. */
2772 if (!sym_index && !sm->link_once
2773 #ifdef TCC_TARGET_ARM
2774 && type != R_ARM_V4BX
2775 #endif
2777 invalid_reloc:
2778 tcc_error_noabort("Invalid relocation entry [%2d] '%s' @ %.8x",
2779 i, strsec + sh->sh_name, rel->r_offset);
2780 goto fail;
2782 rel->r_info = ELFW(R_INFO)(sym_index, type);
2783 /* offset the relocation offset */
2784 rel->r_offset += offseti;
2785 #ifdef TCC_TARGET_ARM
2786 /* Jumps and branches from a Thumb code to a PLT entry need
2787 special handling since PLT entries are ARM code.
2788 Unconditional bl instructions referencing PLT entries are
2789 handled by converting these instructions into blx
2790 instructions. Other case of instructions referencing a PLT
2791 entry require to add a Thumb stub before the PLT entry to
2792 switch to ARM mode. We set bit plt_thumb_stub of the
2793 attribute of a symbol to indicate such a case. */
2794 if (type == R_ARM_THM_JUMP24)
2795 alloc_sym_attr(s1, sym_index)->plt_thumb_stub = 1;
2796 #endif
2798 break;
2799 default:
2800 break;
2804 ret = 0;
2805 the_end:
2806 tcc_free(symtab);
2807 tcc_free(strtab);
2808 tcc_free(old_to_new_syms);
2809 tcc_free(sm_table);
2810 tcc_free(strsec);
2811 tcc_free(shdr);
2812 return ret;
2815 typedef struct ArchiveHeader {
2816 char ar_name[16]; /* name of this member */
2817 char ar_date[12]; /* file mtime */
2818 char ar_uid[6]; /* owner uid; printed as decimal */
2819 char ar_gid[6]; /* owner gid; printed as decimal */
2820 char ar_mode[8]; /* file mode, printed as octal */
2821 char ar_size[10]; /* file size, printed as decimal */
2822 char ar_fmag[2]; /* should contain ARFMAG */
2823 } ArchiveHeader;
2825 static int get_be32(const uint8_t *b)
2827 return b[3] | (b[2] << 8) | (b[1] << 16) | (b[0] << 24);
2830 /* load only the objects which resolve undefined symbols */
2831 static int tcc_load_alacarte(TCCState *s1, int fd, int size)
2833 int i, bound, nsyms, sym_index, off, ret;
2834 uint8_t *data;
2835 const char *ar_names, *p;
2836 const uint8_t *ar_index;
2837 ElfW(Sym) *sym;
2839 data = tcc_malloc(size);
2840 if (read(fd, data, size) != size)
2841 goto fail;
2842 nsyms = get_be32(data);
2843 ar_index = data + 4;
2844 ar_names = (char *) ar_index + nsyms * 4;
2846 do {
2847 bound = 0;
2848 for(p = ar_names, i = 0; i < nsyms; i++, p += strlen(p)+1) {
2849 sym_index = find_elf_sym(symtab_section, p);
2850 if(sym_index) {
2851 sym = &((ElfW(Sym) *)symtab_section->data)[sym_index];
2852 if(sym->st_shndx == SHN_UNDEF) {
2853 off = get_be32(ar_index + i * 4) + sizeof(ArchiveHeader);
2854 ++bound;
2855 lseek(fd, off, SEEK_SET);
2856 if(tcc_load_object_file(s1, fd, off) < 0) {
2857 fail:
2858 ret = -1;
2859 goto the_end;
2864 } while(bound);
2865 ret = 0;
2866 the_end:
2867 tcc_free(data);
2868 return ret;
2871 /* load a '.a' file */
2872 ST_FUNC int tcc_load_archive(TCCState *s1, int fd)
2874 ArchiveHeader hdr;
2875 char ar_size[11];
2876 char ar_name[17];
2877 char magic[8];
2878 int size, len, i;
2879 unsigned long file_offset;
2881 /* skip magic which was already checked */
2882 read(fd, magic, sizeof(magic));
2884 for(;;) {
2885 len = read(fd, &hdr, sizeof(hdr));
2886 if (len == 0)
2887 break;
2888 if (len != sizeof(hdr)) {
2889 tcc_error_noabort("invalid archive");
2890 return -1;
2892 memcpy(ar_size, hdr.ar_size, sizeof(hdr.ar_size));
2893 ar_size[sizeof(hdr.ar_size)] = '\0';
2894 size = strtol(ar_size, NULL, 0);
2895 memcpy(ar_name, hdr.ar_name, sizeof(hdr.ar_name));
2896 for(i = sizeof(hdr.ar_name) - 1; i >= 0; i--) {
2897 if (ar_name[i] != ' ')
2898 break;
2900 ar_name[i + 1] = '\0';
2901 file_offset = lseek(fd, 0, SEEK_CUR);
2902 /* align to even */
2903 size = (size + 1) & ~1;
2904 if (!strcmp(ar_name, "/")) {
2905 /* coff symbol table : we handle it */
2906 if(s1->alacarte_link)
2907 return tcc_load_alacarte(s1, fd, size);
2908 } else if (!strcmp(ar_name, "//") ||
2909 !strcmp(ar_name, "__.SYMDEF") ||
2910 !strcmp(ar_name, "__.SYMDEF/") ||
2911 !strcmp(ar_name, "ARFILENAMES/")) {
2912 /* skip symbol table or archive names */
2913 } else {
2914 if (tcc_load_object_file(s1, fd, file_offset) < 0)
2915 return -1;
2917 lseek(fd, file_offset + size, SEEK_SET);
2919 return 0;
2922 #ifndef TCC_TARGET_PE
2923 /* load a DLL and all referenced DLLs. 'level = 0' means that the DLL
2924 is referenced by the user (so it should be added as DT_NEEDED in
2925 the generated ELF file) */
2926 ST_FUNC int tcc_load_dll(TCCState *s1, int fd, const char *filename, int level)
2928 ElfW(Ehdr) ehdr;
2929 ElfW(Shdr) *shdr, *sh, *sh1;
2930 int i, j, nb_syms, nb_dts, sym_bind, ret;
2931 ElfW(Sym) *sym, *dynsym;
2932 ElfW(Dyn) *dt, *dynamic;
2933 unsigned char *dynstr;
2934 const char *name, *soname;
2935 DLLReference *dllref;
2937 read(fd, &ehdr, sizeof(ehdr));
2939 /* test CPU specific stuff */
2940 if (ehdr.e_ident[5] != ELFDATA2LSB ||
2941 ehdr.e_machine != EM_TCC_TARGET) {
2942 tcc_error_noabort("bad architecture");
2943 return -1;
2946 /* read sections */
2947 shdr = load_data(fd, ehdr.e_shoff, sizeof(ElfW(Shdr)) * ehdr.e_shnum);
2949 /* load dynamic section and dynamic symbols */
2950 nb_syms = 0;
2951 nb_dts = 0;
2952 dynamic = NULL;
2953 dynsym = NULL; /* avoid warning */
2954 dynstr = NULL; /* avoid warning */
2955 for(i = 0, sh = shdr; i < ehdr.e_shnum; i++, sh++) {
2956 switch(sh->sh_type) {
2957 case SHT_DYNAMIC:
2958 nb_dts = sh->sh_size / sizeof(ElfW(Dyn));
2959 dynamic = load_data(fd, sh->sh_offset, sh->sh_size);
2960 break;
2961 case SHT_DYNSYM:
2962 nb_syms = sh->sh_size / sizeof(ElfW(Sym));
2963 dynsym = load_data(fd, sh->sh_offset, sh->sh_size);
2964 sh1 = &shdr[sh->sh_link];
2965 dynstr = load_data(fd, sh1->sh_offset, sh1->sh_size);
2966 break;
2967 default:
2968 break;
2972 /* compute the real library name */
2973 soname = tcc_basename(filename);
2975 for(i = 0, dt = dynamic; i < nb_dts; i++, dt++) {
2976 if (dt->d_tag == DT_SONAME) {
2977 soname = (char *) dynstr + dt->d_un.d_val;
2981 /* if the dll is already loaded, do not load it */
2982 for(i = 0; i < s1->nb_loaded_dlls; i++) {
2983 dllref = s1->loaded_dlls[i];
2984 if (!strcmp(soname, dllref->name)) {
2985 /* but update level if needed */
2986 if (level < dllref->level)
2987 dllref->level = level;
2988 ret = 0;
2989 goto the_end;
2993 /* add the dll and its level */
2994 dllref = tcc_mallocz(sizeof(DLLReference) + strlen(soname));
2995 dllref->level = level;
2996 strcpy(dllref->name, soname);
2997 dynarray_add((void ***)&s1->loaded_dlls, &s1->nb_loaded_dlls, dllref);
2999 /* add dynamic symbols in dynsym_section */
3000 for(i = 1, sym = dynsym + 1; i < nb_syms; i++, sym++) {
3001 sym_bind = ELFW(ST_BIND)(sym->st_info);
3002 if (sym_bind == STB_LOCAL)
3003 continue;
3004 name = (char *) dynstr + sym->st_name;
3005 add_elf_sym(s1->dynsymtab_section, sym->st_value, sym->st_size,
3006 sym->st_info, sym->st_other, sym->st_shndx, name);
3009 /* load all referenced DLLs */
3010 for(i = 0, dt = dynamic; i < nb_dts; i++, dt++) {
3011 switch(dt->d_tag) {
3012 case DT_NEEDED:
3013 name = (char *) dynstr + dt->d_un.d_val;
3014 for(j = 0; j < s1->nb_loaded_dlls; j++) {
3015 dllref = s1->loaded_dlls[j];
3016 if (!strcmp(name, dllref->name))
3017 goto already_loaded;
3019 if (tcc_add_dll(s1, name, AFF_REFERENCED_DLL) < 0) {
3020 tcc_error_noabort("referenced dll '%s' not found", name);
3021 ret = -1;
3022 goto the_end;
3024 already_loaded:
3025 break;
3028 ret = 0;
3029 the_end:
3030 tcc_free(dynstr);
3031 tcc_free(dynsym);
3032 tcc_free(dynamic);
3033 tcc_free(shdr);
3034 return ret;
3037 #define LD_TOK_NAME 256
3038 #define LD_TOK_EOF (-1)
3040 /* return next ld script token */
3041 static int ld_next(TCCState *s1, char *name, int name_size)
3043 int c;
3044 char *q;
3046 redo:
3047 switch(ch) {
3048 case ' ':
3049 case '\t':
3050 case '\f':
3051 case '\v':
3052 case '\r':
3053 case '\n':
3054 inp();
3055 goto redo;
3056 case '/':
3057 minp();
3058 if (ch == '*') {
3059 file->buf_ptr = parse_comment(file->buf_ptr);
3060 ch = file->buf_ptr[0];
3061 goto redo;
3062 } else {
3063 q = name;
3064 *q++ = '/';
3065 goto parse_name;
3067 break;
3068 /* case 'a' ... 'z': */
3069 case 'a':
3070 case 'b':
3071 case 'c':
3072 case 'd':
3073 case 'e':
3074 case 'f':
3075 case 'g':
3076 case 'h':
3077 case 'i':
3078 case 'j':
3079 case 'k':
3080 case 'l':
3081 case 'm':
3082 case 'n':
3083 case 'o':
3084 case 'p':
3085 case 'q':
3086 case 'r':
3087 case 's':
3088 case 't':
3089 case 'u':
3090 case 'v':
3091 case 'w':
3092 case 'x':
3093 case 'y':
3094 case 'z':
3095 /* case 'A' ... 'z': */
3096 case 'A':
3097 case 'B':
3098 case 'C':
3099 case 'D':
3100 case 'E':
3101 case 'F':
3102 case 'G':
3103 case 'H':
3104 case 'I':
3105 case 'J':
3106 case 'K':
3107 case 'L':
3108 case 'M':
3109 case 'N':
3110 case 'O':
3111 case 'P':
3112 case 'Q':
3113 case 'R':
3114 case 'S':
3115 case 'T':
3116 case 'U':
3117 case 'V':
3118 case 'W':
3119 case 'X':
3120 case 'Y':
3121 case 'Z':
3122 case '_':
3123 case '\\':
3124 case '.':
3125 case '$':
3126 case '~':
3127 q = name;
3128 parse_name:
3129 for(;;) {
3130 if (!((ch >= 'a' && ch <= 'z') ||
3131 (ch >= 'A' && ch <= 'Z') ||
3132 (ch >= '0' && ch <= '9') ||
3133 strchr("/.-_+=$:\\,~", ch)))
3134 break;
3135 if ((q - name) < name_size - 1) {
3136 *q++ = ch;
3138 minp();
3140 *q = '\0';
3141 c = LD_TOK_NAME;
3142 break;
3143 case CH_EOF:
3144 c = LD_TOK_EOF;
3145 break;
3146 default:
3147 c = ch;
3148 inp();
3149 break;
3151 return c;
3154 static int ld_add_file(TCCState *s1, const char filename[])
3156 int ret;
3158 ret = tcc_add_file_internal(s1, filename, 0);
3159 if (ret)
3160 ret = tcc_add_dll(s1, filename, 0);
3161 return ret;
3164 static inline int new_undef_syms(void)
3166 int ret = 0;
3167 ret = new_undef_sym;
3168 new_undef_sym = 0;
3169 return ret;
3172 static int ld_add_file_list(TCCState *s1, const char *cmd, int as_needed)
3174 char filename[1024], libname[1024];
3175 int t, group, nblibs = 0, ret = 0;
3176 char **libs = NULL;
3178 group = !strcmp(cmd, "GROUP");
3179 if (!as_needed)
3180 new_undef_syms();
3181 t = ld_next(s1, filename, sizeof(filename));
3182 if (t != '(')
3183 expect("(");
3184 t = ld_next(s1, filename, sizeof(filename));
3185 for(;;) {
3186 libname[0] = '\0';
3187 if (t == LD_TOK_EOF) {
3188 tcc_error_noabort("unexpected end of file");
3189 ret = -1;
3190 goto lib_parse_error;
3191 } else if (t == ')') {
3192 break;
3193 } else if (t == '-') {
3194 t = ld_next(s1, filename, sizeof(filename));
3195 if ((t != LD_TOK_NAME) || (filename[0] != 'l')) {
3196 tcc_error_noabort("library name expected");
3197 ret = -1;
3198 goto lib_parse_error;
3200 pstrcpy(libname, sizeof libname, &filename[1]);
3201 if (s1->static_link) {
3202 snprintf(filename, sizeof filename, "lib%s.a", libname);
3203 } else {
3204 snprintf(filename, sizeof filename, "lib%s.so", libname);
3206 } else if (t != LD_TOK_NAME) {
3207 tcc_error_noabort("filename expected");
3208 ret = -1;
3209 goto lib_parse_error;
3211 if (!strcmp(filename, "AS_NEEDED")) {
3212 ret = ld_add_file_list(s1, cmd, 1);
3213 if (ret)
3214 goto lib_parse_error;
3215 } else {
3216 /* TODO: Implement AS_NEEDED support. Ignore it for now */
3217 if (!as_needed) {
3218 ret = ld_add_file(s1, filename);
3219 if (ret)
3220 goto lib_parse_error;
3221 if (group) {
3222 /* Add the filename *and* the libname to avoid future conversions */
3223 dynarray_add((void ***) &libs, &nblibs, tcc_strdup(filename));
3224 if (libname[0] != '\0')
3225 dynarray_add((void ***) &libs, &nblibs, tcc_strdup(libname));
3229 t = ld_next(s1, filename, sizeof(filename));
3230 if (t == ',') {
3231 t = ld_next(s1, filename, sizeof(filename));
3234 if (group && !as_needed) {
3235 while (new_undef_syms()) {
3236 int i;
3238 for (i = 0; i < nblibs; i ++)
3239 ld_add_file(s1, libs[i]);
3242 lib_parse_error:
3243 dynarray_reset(&libs, &nblibs);
3244 return ret;
3247 /* interpret a subset of GNU ldscripts to handle the dummy libc.so
3248 files */
3249 ST_FUNC int tcc_load_ldscript(TCCState *s1)
3251 char cmd[64];
3252 char filename[1024];
3253 int t, ret;
3255 ch = file->buf_ptr[0];
3256 ch = handle_eob();
3257 for(;;) {
3258 t = ld_next(s1, cmd, sizeof(cmd));
3259 if (t == LD_TOK_EOF)
3260 return 0;
3261 else if (t != LD_TOK_NAME)
3262 return -1;
3263 if (!strcmp(cmd, "INPUT") ||
3264 !strcmp(cmd, "GROUP")) {
3265 ret = ld_add_file_list(s1, cmd, 0);
3266 if (ret)
3267 return ret;
3268 } else if (!strcmp(cmd, "OUTPUT_FORMAT") ||
3269 !strcmp(cmd, "TARGET")) {
3270 /* ignore some commands */
3271 t = ld_next(s1, cmd, sizeof(cmd));
3272 if (t != '(')
3273 expect("(");
3274 for(;;) {
3275 t = ld_next(s1, filename, sizeof(filename));
3276 if (t == LD_TOK_EOF) {
3277 tcc_error_noabort("unexpected end of file");
3278 return -1;
3279 } else if (t == ')') {
3280 break;
3283 } else {
3284 return -1;
3287 return 0;
3289 #endif /* !TCC_TARGET_PE */