So that the generated code, and more short
[tinycc.git] / tccelf.c
blobf0ed22b72028f3895ac8fa657d3dd7792657e5f5
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_reloc(Section *symtab, Section *s, unsigned long offset,
273 int type, int symbol)
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 = 0;
296 #endif
299 /* put stab debug information */
301 ST_FUNC void put_stabs(const char *str, int type, int other, int desc,
302 unsigned long value)
304 Stab_Sym *sym;
306 sym = section_ptr_add(stab_section, sizeof(Stab_Sym));
307 if (str) {
308 sym->n_strx = put_elf_str(stabstr_section, str);
309 } else {
310 sym->n_strx = 0;
312 sym->n_type = type;
313 sym->n_other = other;
314 sym->n_desc = desc;
315 sym->n_value = value;
318 ST_FUNC void put_stabs_r(const char *str, int type, int other, int desc,
319 unsigned long value, Section *sec, int sym_index)
321 put_stabs(str, type, other, desc, value);
322 put_elf_reloc(symtab_section, stab_section,
323 stab_section->data_offset - sizeof(unsigned int),
324 R_DATA_32, sym_index);
327 ST_FUNC void put_stabn(int type, int other, int desc, int value)
329 put_stabs(NULL, type, other, desc, value);
332 ST_FUNC void put_stabd(int type, int other, int desc)
334 put_stabs(NULL, type, other, desc, 0);
337 /* Browse each elem of type <type> in section <sec> starting at elem <startoff>
338 using variable <elem> */
339 #define for_each_elem(sec, startoff, elem, type) \
340 for (elem = (type *) sec->data + startoff; \
341 elem < (type *) (sec->data + sec->data_offset); elem++)
343 /* In an ELF file symbol table, the local symbols must appear below
344 the global and weak ones. Since TCC cannot sort it while generating
345 the code, we must do it after. All the relocation tables are also
346 modified to take into account the symbol table sorting */
347 static void sort_syms(TCCState *s1, Section *s)
349 int *old_to_new_syms;
350 ElfW(Sym) *new_syms;
351 int nb_syms, i;
352 ElfW(Sym) *p, *q;
353 ElfW_Rel *rel;
354 Section *sr;
355 int type, sym_index;
357 nb_syms = s->data_offset / sizeof(ElfW(Sym));
358 new_syms = tcc_malloc(nb_syms * sizeof(ElfW(Sym)));
359 old_to_new_syms = tcc_malloc(nb_syms * sizeof(int));
361 /* first pass for local symbols */
362 p = (ElfW(Sym) *)s->data;
363 q = new_syms;
364 for(i = 0; i < nb_syms; i++) {
365 if (ELFW(ST_BIND)(p->st_info) == STB_LOCAL) {
366 old_to_new_syms[i] = q - new_syms;
367 *q++ = *p;
369 p++;
371 /* save the number of local symbols in section header */
372 s->sh_info = q - new_syms;
374 /* then second pass for non local symbols */
375 p = (ElfW(Sym) *)s->data;
376 for(i = 0; i < nb_syms; i++) {
377 if (ELFW(ST_BIND)(p->st_info) != STB_LOCAL) {
378 old_to_new_syms[i] = q - new_syms;
379 *q++ = *p;
381 p++;
384 /* we copy the new symbols to the old */
385 memcpy(s->data, new_syms, nb_syms * sizeof(ElfW(Sym)));
386 tcc_free(new_syms);
388 /* now we modify all the relocations */
389 for(i = 1; i < s1->nb_sections; i++) {
390 sr = s1->sections[i];
391 if (sr->sh_type == SHT_RELX && sr->link == s) {
392 for_each_elem(sr, 0, rel, ElfW_Rel) {
393 sym_index = ELFW(R_SYM)(rel->r_info);
394 type = ELFW(R_TYPE)(rel->r_info);
395 sym_index = old_to_new_syms[sym_index];
396 rel->r_info = ELFW(R_INFO)(sym_index, type);
401 tcc_free(old_to_new_syms);
404 /* relocate common symbols in the .bss section */
405 ST_FUNC void relocate_common_syms(void)
407 ElfW(Sym) *sym;
408 unsigned long offset, align;
410 for_each_elem(symtab_section, 1, sym, ElfW(Sym)) {
411 if (sym->st_shndx == SHN_COMMON) {
412 /* align symbol */
413 align = sym->st_value;
414 offset = bss_section->data_offset;
415 offset = (offset + align - 1) & -align;
416 sym->st_value = offset;
417 sym->st_shndx = bss_section->sh_num;
418 offset += sym->st_size;
419 bss_section->data_offset = offset;
424 /* relocate symbol table, resolve undefined symbols if do_resolve is
425 true and output error if undefined symbol. */
426 ST_FUNC void relocate_syms(TCCState *s1, int do_resolve)
428 ElfW(Sym) *sym, *esym;
429 int sym_bind, sh_num, sym_index;
430 const char *name;
432 for_each_elem(symtab_section, 1, sym, ElfW(Sym)) {
433 sh_num = sym->st_shndx;
434 if (sh_num == SHN_UNDEF) {
435 name = (char *) strtab_section->data + sym->st_name;
436 /* Use ld.so to resolve symbol for us (for tcc -run) */
437 if (do_resolve) {
438 #if defined TCC_IS_NATIVE && !defined _WIN32
439 void *addr;
440 name = (char *) symtab_section->link->data + sym->st_name;
441 addr = resolve_sym(s1, name);
442 if (addr) {
443 sym->st_value = (addr_t)addr;
444 #ifdef DEBUG_RELOC
445 printf ("relocate_sym: %s -> 0x%x\n", name, sym->st_value);
446 #endif
447 goto found;
449 #endif
450 } else if (s1->dynsym) {
451 /* if dynamic symbol exist, then use it */
452 sym_index = find_elf_sym(s1->dynsym, name);
453 if (sym_index) {
454 esym = &((ElfW(Sym) *)s1->dynsym->data)[sym_index];
455 sym->st_value = esym->st_value;
456 goto found;
459 /* XXX: _fp_hw seems to be part of the ABI, so we ignore
460 it */
461 if (!strcmp(name, "_fp_hw"))
462 goto found;
463 /* only weak symbols are accepted to be undefined. Their
464 value is zero */
465 sym_bind = ELFW(ST_BIND)(sym->st_info);
466 if (sym_bind == STB_WEAK) {
467 sym->st_value = 0;
468 } else {
469 tcc_error_noabort("undefined symbol '%s'", name);
471 } else if (sh_num < SHN_LORESERVE) {
472 /* add section base */
473 sym->st_value += s1->sections[sym->st_shndx]->sh_addr;
475 found: ;
479 /* relocate a given section (CPU dependent) by applying the relocations
480 in the associated relocation section */
481 ST_FUNC void relocate_section(TCCState *s1, Section *s)
483 Section *sr = s->reloc;
484 ElfW_Rel *rel;
485 ElfW(Sym) *sym;
486 int type, sym_index;
487 unsigned char *ptr;
488 addr_t val, addr;
489 #if defined TCC_TARGET_I386 || defined TCC_TARGET_X86_64
490 ElfW_Rel *qrel = (ElfW_Rel *) sr->data; /* ptr to next reloc entry reused */
491 int esym_index;
492 #endif
494 for_each_elem(sr, 0, rel, ElfW_Rel) {
495 ptr = s->data + rel->r_offset;
497 sym_index = ELFW(R_SYM)(rel->r_info);
498 sym = &((ElfW(Sym) *)symtab_section->data)[sym_index];
499 val = sym->st_value;
500 #ifdef TCC_TARGET_X86_64
501 val += rel->r_addend;
502 #endif
503 type = ELFW(R_TYPE)(rel->r_info);
504 addr = s->sh_addr + rel->r_offset;
506 /* CPU specific */
507 switch(type) {
508 #if defined(TCC_TARGET_I386)
509 case R_386_32:
510 if (s1->output_type == TCC_OUTPUT_DLL) {
511 esym_index = s1->symtab_to_dynsym[sym_index];
512 qrel->r_offset = rel->r_offset;
513 if (esym_index) {
514 qrel->r_info = ELFW(R_INFO)(esym_index, R_386_32);
515 qrel++;
516 break;
517 } else {
518 qrel->r_info = ELFW(R_INFO)(0, R_386_RELATIVE);
519 qrel++;
522 *(int *)ptr += val;
523 break;
524 case R_386_PC32:
525 if (s1->output_type == TCC_OUTPUT_DLL) {
526 /* DLL relocation */
527 esym_index = s1->symtab_to_dynsym[sym_index];
528 if (esym_index) {
529 qrel->r_offset = rel->r_offset;
530 qrel->r_info = ELFW(R_INFO)(esym_index, R_386_PC32);
531 qrel++;
532 break;
535 *(int *)ptr += val - addr;
536 break;
537 case R_386_PLT32:
538 *(int *)ptr += val - addr;
539 break;
540 case R_386_GLOB_DAT:
541 case R_386_JMP_SLOT:
542 *(int *)ptr = val;
543 break;
544 case R_386_GOTPC:
545 *(int *)ptr += s1->got->sh_addr - addr;
546 break;
547 case R_386_GOTOFF:
548 *(int *)ptr += val - s1->got->sh_addr;
549 break;
550 case R_386_GOT32:
551 /* we load the got offset */
552 *(int *)ptr += s1->sym_attrs[sym_index].got_offset;
553 break;
554 case R_386_16:
555 if (s1->output_format != TCC_OUTPUT_FORMAT_BINARY) {
556 output_file:
557 tcc_error("can only produce 16-bit binary files");
559 *(short *)ptr += val;
560 break;
561 case R_386_PC16:
562 if (s1->output_format != TCC_OUTPUT_FORMAT_BINARY)
563 goto output_file;
564 *(short *)ptr += val - addr;
565 break;
566 #elif defined(TCC_TARGET_ARM)
567 case R_ARM_PC24:
568 case R_ARM_CALL:
569 case R_ARM_JUMP24:
570 case R_ARM_PLT32:
572 int x, is_thumb, is_call, h, blx_avail, is_bl, th_ko;
573 x = (*(int *) ptr) & 0xffffff;
574 if (sym->st_shndx == SHN_UNDEF)
575 val = s1->plt->sh_addr;
576 #ifdef DEBUG_RELOC
577 printf ("reloc %d: x=0x%x val=0x%x ", type, x, val);
578 #endif
579 (*(int *)ptr) &= 0xff000000;
580 if (x & 0x800000)
581 x -= 0x1000000;
582 x <<= 2;
583 blx_avail = (TCC_ARM_VERSION >= 5);
584 is_thumb = val & 1;
585 is_bl = (*(unsigned *) ptr) >> 24 == 0xeb;
586 is_call = (type == R_ARM_CALL || (type == R_ARM_PC24 && is_bl));
587 x += val - addr;
588 #ifdef DEBUG_RELOC
589 printf (" newx=0x%x name=%s\n", x,
590 (char *) symtab_section->link->data + sym->st_name);
591 #endif
592 h = x & 2;
593 th_ko = (x & 3) && (!blx_avail || !is_call);
594 if (th_ko || x >= 0x2000000 || x < -0x2000000)
595 tcc_error("can't relocate value at %x,%d",addr, type);
596 x >>= 2;
597 x &= 0xffffff;
598 /* Only reached if blx is avail and it is a call */
599 if (is_thumb) {
600 x |= h << 24;
601 (*(int *)ptr) = 0xfa << 24; /* bl -> blx */
603 (*(int *) ptr) |= x;
605 break;
606 /* Since these relocations only concern Thumb-2 and blx instruction was
607 introduced before Thumb-2, we can assume blx is available and not
608 guard its use */
609 case R_ARM_THM_PC22:
610 case R_ARM_THM_JUMP24:
612 int x, hi, lo, s, j1, j2, i1, i2, imm10, imm11;
613 int to_thumb, is_call, to_plt, blx_bit = 1 << 12;
614 Section *plt;
616 /* weak reference */
617 if (sym->st_shndx == SHN_UNDEF &&
618 ELFW(ST_BIND)(sym->st_info) == STB_WEAK)
619 break;
621 /* Get initial offset */
622 hi = (*(uint16_t *)ptr);
623 lo = (*(uint16_t *)(ptr+2));
624 s = (hi >> 10) & 1;
625 j1 = (lo >> 13) & 1;
626 j2 = (lo >> 11) & 1;
627 i1 = (j1 ^ s) ^ 1;
628 i2 = (j2 ^ s) ^ 1;
629 imm10 = hi & 0x3ff;
630 imm11 = lo & 0x7ff;
631 x = (s << 24) | (i1 << 23) | (i2 << 22) |
632 (imm10 << 12) | (imm11 << 1);
633 if (x & 0x01000000)
634 x -= 0x02000000;
636 /* Relocation infos */
637 to_thumb = val & 1;
638 plt = s1->plt;
639 to_plt = (val >= plt->sh_addr) &&
640 (val < plt->sh_addr + plt->data_offset);
641 is_call = (type == R_ARM_THM_PC22);
643 /* Compute final offset */
644 if (to_plt && !is_call) /* Point to 1st instr of Thumb stub */
645 x -= 4;
646 x += val - addr;
647 if (!to_thumb && is_call) {
648 blx_bit = 0; /* bl -> blx */
649 x = (x + 3) & -4; /* Compute offset from aligned PC */
652 /* Check that relocation is possible
653 * offset must not be out of range
654 * if target is to be entered in arm mode:
655 - bit 1 must not set
656 - instruction must be a call (bl) or a jump to PLT */
657 if (!to_thumb || x >= 0x1000000 || x < -0x1000000)
658 if (to_thumb || (val & 2) || (!is_call && !to_plt))
659 tcc_error("can't relocate value at %x,%d",addr, type);
661 /* Compute and store final offset */
662 s = (x >> 24) & 1;
663 i1 = (x >> 23) & 1;
664 i2 = (x >> 22) & 1;
665 j1 = s ^ (i1 ^ 1);
666 j2 = s ^ (i2 ^ 1);
667 imm10 = (x >> 12) & 0x3ff;
668 imm11 = (x >> 1) & 0x7ff;
669 (*(uint16_t *)ptr) = (uint16_t) ((hi & 0xf800) |
670 (s << 10) | imm10);
671 (*(uint16_t *)(ptr+2)) = (uint16_t) ((lo & 0xc000) |
672 (j1 << 13) | blx_bit | (j2 << 11) |
673 imm11);
675 break;
676 case R_ARM_MOVT_ABS:
677 case R_ARM_MOVW_ABS_NC:
679 int x, imm4, imm12;
680 if (type == R_ARM_MOVT_ABS)
681 val >>= 16;
682 imm12 = val & 0xfff;
683 imm4 = (val >> 12) & 0xf;
684 x = (imm4 << 16) | imm12;
685 if (type == R_ARM_THM_MOVT_ABS)
686 *(int *)ptr |= x;
687 else
688 *(int *)ptr += x;
690 break;
691 case R_ARM_THM_MOVT_ABS:
692 case R_ARM_THM_MOVW_ABS_NC:
694 int x, i, imm4, imm3, imm8;
695 if (type == R_ARM_THM_MOVT_ABS)
696 val >>= 16;
697 imm8 = val & 0xff;
698 imm3 = (val >> 8) & 0x7;
699 i = (val >> 11) & 1;
700 imm4 = (val >> 12) & 0xf;
701 x = (imm3 << 28) | (imm8 << 16) | (i << 10) | imm4;
702 if (type == R_ARM_THM_MOVT_ABS)
703 *(int *)ptr |= x;
704 else
705 *(int *)ptr += x;
707 break;
708 case R_ARM_PREL31:
710 int x;
711 x = (*(int *)ptr) & 0x7fffffff;
712 (*(int *)ptr) &= 0x80000000;
713 x = (x * 2) / 2;
714 x += val - addr;
715 if((x^(x>>1))&0x40000000)
716 tcc_error("can't relocate value at %x,%d",addr, type);
717 (*(int *)ptr) |= x & 0x7fffffff;
719 case R_ARM_ABS32:
720 *(int *)ptr += val;
721 break;
722 case R_ARM_REL32:
723 *(int *)ptr += val - addr;
724 break;
725 case R_ARM_GOTPC:
726 *(int *)ptr += s1->got->sh_addr - addr;
727 break;
728 case R_ARM_GOTOFF:
729 *(int *)ptr += val - s1->got->sh_addr;
730 break;
731 case R_ARM_GOT32:
732 /* we load the got offset */
733 *(int *)ptr += s1->sym_attrs[sym_index].got_offset;
734 break;
735 case R_ARM_COPY:
736 break;
737 case R_ARM_V4BX:
738 /* trade Thumb support for ARMv4 support */
739 if ((0x0ffffff0 & *(int*)ptr) == 0x012FFF10)
740 *(int*)ptr ^= 0xE12FFF10 ^ 0xE1A0F000; /* BX Rm -> MOV PC, Rm */
741 break;
742 case R_ARM_GLOB_DAT:
743 case R_ARM_JUMP_SLOT:
744 *(addr_t *)ptr = val;
745 break;
746 case R_ARM_NONE:
747 /* Nothing to do. Normally used to indicate a dependency
748 on a certain symbol (like for exception handling under EABI). */
749 break;
750 default:
751 fprintf(stderr,"FIXME: handle reloc type %x at %x [%p] to %x\n",
752 type, (unsigned)addr, ptr, (unsigned)val);
753 break;
754 #elif defined(TCC_TARGET_C67)
755 case R_C60_32:
756 *(int *)ptr += val;
757 break;
758 case R_C60LO16:
760 uint32_t orig;
762 /* put the low 16 bits of the absolute address
763 add to what is already there */
765 orig = ((*(int *)(ptr )) >> 7) & 0xffff;
766 orig |= (((*(int *)(ptr+4)) >> 7) & 0xffff) << 16;
768 /* patch both at once - assumes always in pairs Low - High */
770 *(int *) ptr = (*(int *) ptr & (~(0xffff << 7)) ) | (((val+orig) & 0xffff) << 7);
771 *(int *)(ptr+4) = (*(int *)(ptr+4) & (~(0xffff << 7)) ) | ((((val+orig)>>16) & 0xffff) << 7);
773 break;
774 case R_C60HI16:
775 break;
776 default:
777 fprintf(stderr,"FIXME: handle reloc type %x at %x [%p] to %x\n",
778 type, (unsigned)addr, ptr, (unsigned)val);
779 break;
780 #elif defined(TCC_TARGET_X86_64)
781 case R_X86_64_64:
782 if (s1->output_type == TCC_OUTPUT_DLL) {
783 esym_index = s1->symtab_to_dynsym[sym_index];
784 qrel->r_offset = rel->r_offset;
785 if (esym_index) {
786 qrel->r_info = ELFW(R_INFO)(esym_index, R_X86_64_64);
787 qrel->r_addend = rel->r_addend;
788 qrel++;
789 break;
790 } else {
791 qrel->r_info = ELFW(R_INFO)(0, R_X86_64_RELATIVE);
792 qrel->r_addend = *(long long *)ptr + val;
793 qrel++;
796 *(long long *)ptr += val;
797 break;
798 case R_X86_64_32:
799 case R_X86_64_32S:
800 if (s1->output_type == TCC_OUTPUT_DLL) {
801 /* XXX: this logic may depend on TCC's codegen
802 now TCC uses R_X86_64_32 even for a 64bit pointer */
803 qrel->r_info = ELFW(R_INFO)(0, R_X86_64_RELATIVE);
804 qrel->r_addend = *(int *)ptr + val;
805 qrel++;
807 *(int *)ptr += val;
808 break;
810 case R_X86_64_PC32:
811 if (s1->output_type == TCC_OUTPUT_DLL) {
812 /* DLL relocation */
813 esym_index = s1->symtab_to_dynsym[sym_index];
814 if (esym_index) {
815 qrel->r_offset = rel->r_offset;
816 qrel->r_info = ELFW(R_INFO)(esym_index, R_X86_64_PC32);
817 qrel->r_addend = *(int *)ptr;
818 qrel++;
819 break;
822 goto plt32pc32;
824 case R_X86_64_PLT32:
825 /* We've put the PLT slot offset into r_addend when generating
826 it, and that's what we must use as relocation value (adjusted
827 by section offset of course). */
828 val = s1->plt->sh_addr + rel->r_addend;
829 /* fallthrough. */
831 plt32pc32:
833 long long diff;
834 diff = (long long)val - addr;
835 if (diff <= -2147483647 || diff > 2147483647) {
836 tcc_error("internal error: relocation failed");
838 *(int *)ptr += diff;
840 break;
841 case R_X86_64_GLOB_DAT:
842 case R_X86_64_JUMP_SLOT:
843 /* They don't need addend */
844 *(addr_t *)ptr = val - rel->r_addend;
845 break;
846 case R_X86_64_GOTPCREL:
847 *(int *)ptr += (s1->got->sh_addr - addr +
848 s1->sym_attrs[sym_index].got_offset - 4);
849 break;
850 case R_X86_64_GOTTPOFF:
851 *(int *)ptr += val - s1->got->sh_addr;
852 break;
853 case R_X86_64_GOT32:
854 /* we load the got offset */
855 *(int *)ptr += s1->sym_attrs[sym_index].got_offset;
856 break;
857 #else
858 #error unsupported processor
859 #endif
862 /* if the relocation is allocated, we change its symbol table */
863 if (sr->sh_flags & SHF_ALLOC)
864 sr->link = s1->dynsym;
867 /* relocate relocation table in 'sr' */
868 static void relocate_rel(TCCState *s1, Section *sr)
870 Section *s;
871 ElfW_Rel *rel;
873 s = s1->sections[sr->sh_info];
874 for_each_elem(sr, 0, rel, ElfW_Rel)
875 rel->r_offset += s->sh_addr;
878 /* count the number of dynamic relocations so that we can reserve
879 their space */
880 static int prepare_dynamic_rel(TCCState *s1, Section *sr)
882 ElfW_Rel *rel;
883 int sym_index, esym_index, type, count;
885 count = 0;
886 for_each_elem(sr, 0, rel, ElfW_Rel) {
887 sym_index = ELFW(R_SYM)(rel->r_info);
888 type = ELFW(R_TYPE)(rel->r_info);
889 switch(type) {
890 #if defined(TCC_TARGET_I386)
891 case R_386_32:
892 #elif defined(TCC_TARGET_X86_64)
893 case R_X86_64_32:
894 case R_X86_64_32S:
895 case R_X86_64_64:
896 #endif
897 count++;
898 break;
899 #if defined(TCC_TARGET_I386)
900 case R_386_PC32:
901 #elif defined(TCC_TARGET_X86_64)
902 case R_X86_64_PC32:
903 #endif
904 esym_index = s1->symtab_to_dynsym[sym_index];
905 if (esym_index)
906 count++;
907 break;
908 default:
909 break;
912 if (count) {
913 /* allocate the section */
914 sr->sh_flags |= SHF_ALLOC;
915 sr->sh_size = count * sizeof(ElfW_Rel);
917 return count;
920 static struct sym_attr *alloc_sym_attr(TCCState *s1, int index)
922 int n;
923 struct sym_attr *tab;
925 if (index >= s1->nb_sym_attrs) {
926 /* find immediately bigger power of 2 and reallocate array */
927 n = 1;
928 while (index >= n)
929 n *= 2;
930 tab = tcc_realloc(s1->sym_attrs, n * sizeof(*s1->sym_attrs));
931 s1->sym_attrs = tab;
932 memset(s1->sym_attrs + s1->nb_sym_attrs, 0,
933 (n - s1->nb_sym_attrs) * sizeof(*s1->sym_attrs));
934 s1->nb_sym_attrs = n;
936 return &s1->sym_attrs[index];
939 /* XXX: suppress that */
940 static void put32(unsigned char *p, uint32_t val)
942 p[0] = val;
943 p[1] = val >> 8;
944 p[2] = val >> 16;
945 p[3] = val >> 24;
948 #if defined(TCC_TARGET_I386) || defined(TCC_TARGET_ARM) || \
949 defined(TCC_TARGET_X86_64)
950 static uint32_t get32(unsigned char *p)
952 return p[0] | (p[1] << 8) | (p[2] << 16) | (p[3] << 24);
954 #endif
956 static void build_got(TCCState *s1)
958 unsigned char *ptr;
960 /* if no got, then create it */
961 s1->got = new_section(s1, ".got", SHT_PROGBITS, SHF_ALLOC | SHF_WRITE);
962 s1->got->sh_entsize = 4;
963 add_elf_sym(symtab_section, 0, 4, ELFW(ST_INFO)(STB_GLOBAL, STT_OBJECT),
964 0, s1->got->sh_num, "_GLOBAL_OFFSET_TABLE_");
965 ptr = section_ptr_add(s1->got, 3 * PTR_SIZE);
966 #if PTR_SIZE == 4
967 /* keep space for _DYNAMIC pointer, if present */
968 put32(ptr, 0);
969 /* two dummy got entries */
970 put32(ptr + 4, 0);
971 put32(ptr + 8, 0);
972 #else
973 /* keep space for _DYNAMIC pointer, if present */
974 put32(ptr, 0);
975 put32(ptr + 4, 0);
976 /* two dummy got entries */
977 put32(ptr + 8, 0);
978 put32(ptr + 12, 0);
979 put32(ptr + 16, 0);
980 put32(ptr + 20, 0);
981 #endif
984 /* put a got or plt entry corresponding to a symbol in symtab_section. 'size'
985 and 'info' can be modifed if more precise info comes from the DLL.
986 Returns offset of GOT or PLT slot. */
987 static unsigned long put_got_entry(TCCState *s1,
988 int reloc_type, unsigned long size, int info,
989 int sym_index)
991 int index, need_plt_entry;
992 const char *name;
993 ElfW(Sym) *sym;
994 unsigned long offset;
995 int *ptr;
996 struct sym_attr *symattr;
998 if (!s1->got)
999 build_got(s1);
1001 need_plt_entry =
1002 #ifdef TCC_TARGET_X86_64
1003 (reloc_type == R_X86_64_JUMP_SLOT);
1004 #elif defined(TCC_TARGET_I386)
1005 (reloc_type == R_386_JMP_SLOT);
1006 #elif defined(TCC_TARGET_ARM)
1007 (reloc_type == R_ARM_JUMP_SLOT);
1008 #else
1010 #endif
1012 if (need_plt_entry && !s1->plt) {
1013 /* add PLT */
1014 s1->plt = new_section(s1, ".plt", SHT_PROGBITS,
1015 SHF_ALLOC | SHF_EXECINSTR);
1016 s1->plt->sh_entsize = 4;
1019 /* If a got/plt entry already exists for that symbol, no need to add one */
1020 if (sym_index < s1->nb_sym_attrs) {
1021 if (need_plt_entry && s1->sym_attrs[sym_index].plt_offset)
1022 return s1->sym_attrs[sym_index].plt_offset;
1023 else if (!need_plt_entry && s1->sym_attrs[sym_index].got_offset)
1024 return s1->sym_attrs[sym_index].got_offset;
1027 symattr = alloc_sym_attr(s1, sym_index);
1029 /* Only store the GOT offset if it's not generated for the PLT entry. */
1030 if (!need_plt_entry)
1031 symattr->got_offset = s1->got->data_offset;
1033 sym = &((ElfW(Sym) *)symtab_section->data)[sym_index];
1034 name = (char *) symtab_section->link->data + sym->st_name;
1035 offset = sym->st_value;
1036 #if defined(TCC_TARGET_I386) || defined(TCC_TARGET_X86_64)
1037 if (need_plt_entry) {
1038 Section *plt;
1039 uint8_t *p;
1040 int modrm;
1041 unsigned long relofs;
1043 #if defined(TCC_OUTPUT_DLL_WITH_PLT)
1044 modrm = 0x25;
1045 #else
1046 /* if we build a DLL, we add a %ebx offset */
1047 if (s1->output_type == TCC_OUTPUT_DLL)
1048 modrm = 0xa3;
1049 else
1050 modrm = 0x25;
1051 #endif
1053 /* add a PLT entry */
1054 plt = s1->plt;
1055 if (plt->data_offset == 0) {
1056 /* first plt entry */
1057 p = section_ptr_add(plt, 16);
1058 p[0] = 0xff; /* pushl got + PTR_SIZE */
1059 p[1] = modrm + 0x10;
1060 put32(p + 2, PTR_SIZE);
1061 p[6] = 0xff; /* jmp *(got + PTR_SIZE * 2) */
1062 p[7] = modrm;
1063 put32(p + 8, PTR_SIZE * 2);
1066 /* The PLT slot refers to the relocation entry it needs
1067 via offset. The reloc entry is created below, so its
1068 offset is the current data_offset. */
1069 relofs = s1->got->reloc ? s1->got->reloc->data_offset : 0;
1070 symattr->plt_offset = plt->data_offset;
1071 p = section_ptr_add(plt, 16);
1072 p[0] = 0xff; /* jmp *(got + x) */
1073 p[1] = modrm;
1074 put32(p + 2, s1->got->data_offset);
1075 p[6] = 0x68; /* push $xxx */
1076 #ifdef TCC_TARGET_X86_64
1077 /* On x86-64, the relocation is referred to by _index_. */
1078 put32(p + 7, relofs / sizeof (ElfW_Rel));
1079 #else
1080 put32(p + 7, relofs);
1081 #endif
1082 p[11] = 0xe9; /* jmp plt_start */
1083 put32(p + 12, -(plt->data_offset));
1085 /* If this was an UNDEF symbol set the offset in the
1086 dynsymtab to the PLT slot, so that PC32 relocs to it
1087 can be resolved. */
1088 if (sym->st_shndx == SHN_UNDEF)
1089 offset = plt->data_offset - 16;
1091 #elif defined(TCC_TARGET_ARM)
1092 if (need_plt_entry) {
1093 Section *plt;
1094 uint8_t *p;
1096 /* if we build a DLL, we add a %ebx offset */
1097 if (s1->output_type == TCC_OUTPUT_DLL)
1098 tcc_error("DLLs unimplemented!");
1100 /* add a PLT entry */
1101 plt = s1->plt;
1102 if (plt->data_offset == 0) {
1103 /* first plt entry */
1104 p = section_ptr_add(plt, 16);
1105 put32(p, 0xe52de004); /* push {lr} */
1106 put32(p+4, 0xe59fe010); /* ldr lr, [pc, #16] */
1107 put32(p+8, 0xe08fe00e); /* add lr, pc, lr */
1108 put32(p+12, 0xe5bef008); /* ldr pc, [lr, #8]! */
1111 symattr->plt_offset = plt->data_offset;
1112 if (symattr->plt_thumb_stub) {
1113 p = section_ptr_add(plt, 20);
1114 put32(p, 0x4778); /* bx pc */
1115 put32(p+2, 0x46c0); /* nop */
1116 p += 4;
1117 } else
1118 p = section_ptr_add(plt, 16);
1119 put32(p, 0xe59fc004); /* ldr ip, [pc, #4] ; GOT entry offset */
1120 put32(p+4, 0xe08fc00c); /* add ip, pc, ip ; addr of GOT entry */
1121 put32(p+8, 0xe59cf000); /* ldr pc, [ip] ; jump to GOT entry */
1122 put32(p+12, s1->got->data_offset); /* GOT entry off once patched */
1124 /* the symbol is modified so that it will be relocated to
1125 the PLT */
1126 if (sym->st_shndx == SHN_UNDEF)
1127 offset = plt->data_offset - 16;
1129 #elif defined(TCC_TARGET_C67)
1130 if (s1->dynsym) {
1131 tcc_error("C67 got not implemented");
1133 #else
1134 #error unsupported CPU
1135 #endif
1136 if (s1->dynsym) {
1137 /* XXX This might generate multiple syms for name. */
1138 index = put_elf_sym(s1->dynsym, offset,
1139 size, info, 0, sym->st_shndx, name);
1140 /* Create the relocation (it's against the GOT for PLT
1141 and GOT relocs). */
1142 put_elf_reloc(s1->dynsym, s1->got,
1143 s1->got->data_offset,
1144 reloc_type, index);
1145 } else {
1146 /* Without .dynsym (i.e. static link or memory output) we
1147 still need relocs against the generated got, so as to fill
1148 the entries with the symbol values (determined later). */
1149 put_elf_reloc(symtab_section, s1->got,
1150 s1->got->data_offset,
1151 reloc_type, sym_index);
1153 /* And now create the GOT slot itself. */
1154 ptr = section_ptr_add(s1->got, PTR_SIZE);
1155 *ptr = 0;
1156 if (need_plt_entry)
1157 return symattr->plt_offset;
1158 else
1159 return symattr->got_offset;
1162 /* build GOT and PLT entries */
1163 ST_FUNC void build_got_entries(TCCState *s1)
1165 Section *s;
1166 ElfW_Rel *rel;
1167 ElfW(Sym) *sym;
1168 int i, type, reloc_type, sym_index;
1170 for(i = 1; i < s1->nb_sections; i++) {
1171 s = s1->sections[i];
1172 if (s->sh_type != SHT_RELX)
1173 continue;
1174 /* no need to handle got relocations */
1175 if (s->link != symtab_section)
1176 continue;
1177 for_each_elem(s, 0, rel, ElfW_Rel) {
1178 type = ELFW(R_TYPE)(rel->r_info);
1179 switch(type) {
1180 #if defined(TCC_TARGET_I386)
1181 case R_386_GOT32:
1182 case R_386_GOTOFF:
1183 case R_386_GOTPC:
1184 case R_386_PLT32:
1185 if (!s1->got)
1186 build_got(s1);
1187 if (type == R_386_GOT32 || type == R_386_PLT32) {
1188 sym_index = ELFW(R_SYM)(rel->r_info);
1189 sym = &((ElfW(Sym) *)symtab_section->data)[sym_index];
1190 /* look at the symbol got offset. If none, then add one */
1191 if (type == R_386_GOT32)
1192 reloc_type = R_386_GLOB_DAT;
1193 else
1194 reloc_type = R_386_JMP_SLOT;
1195 put_got_entry(s1, reloc_type, sym->st_size, sym->st_info,
1196 sym_index);
1198 break;
1199 #elif defined(TCC_TARGET_ARM)
1200 case R_ARM_PC24:
1201 case R_ARM_CALL:
1202 case R_ARM_JUMP24:
1203 case R_ARM_GOT32:
1204 case R_ARM_GOTOFF:
1205 case R_ARM_GOTPC:
1206 case R_ARM_PLT32:
1207 if (!s1->got)
1208 build_got(s1);
1209 sym_index = ELFW(R_SYM)(rel->r_info);
1210 sym = &((ElfW(Sym) *)symtab_section->data)[sym_index];
1211 if (type != R_ARM_GOTOFF && type != R_ARM_GOTPC
1212 && sym->st_shndx == SHN_UNDEF) {
1213 unsigned long ofs;
1214 /* look at the symbol got offset. If none, then add one */
1215 if (type == R_ARM_GOT32)
1216 reloc_type = R_ARM_GLOB_DAT;
1217 else
1218 reloc_type = R_ARM_JUMP_SLOT;
1219 ofs = put_got_entry(s1, reloc_type, sym->st_size,
1220 sym->st_info, sym_index);
1221 #ifdef DEBUG_RELOC
1222 printf ("maybegot: %s, %d, %d --> ofs=0x%x\n",
1223 (char *) symtab_section->link->data + sym->st_name,
1224 type, sym->st_shndx, ofs);
1225 #endif
1226 if (type != R_ARM_GOT32) {
1227 addr_t *ptr = (addr_t*)(s1->sections[s->sh_info]->data
1228 + rel->r_offset);
1229 /* x must be signed! */
1230 int x = *ptr & 0xffffff;
1231 x = (x << 8) >> 8;
1232 x <<= 2;
1233 x += ofs;
1234 x >>= 2;
1235 #ifdef DEBUG_RELOC
1236 printf ("insn=0x%x --> 0x%x (x==0x%x)\n", *ptr,
1237 (*ptr & 0xff000000) | x, x);
1238 #endif
1239 *ptr = (*ptr & 0xff000000) | x;
1242 break;
1243 case R_ARM_THM_JUMP24:
1244 sym_index = ELFW(R_SYM)(rel->r_info);
1245 sym = &((ElfW(Sym) *)symtab_section->data)[sym_index];
1246 /* We are relocating a jump from thumb code to arm code */
1247 if (sym->st_shndx != SHN_UNDEF && !(sym->st_value & 1)) {
1248 int index;
1249 uint8_t *p;
1250 char *name, buf[1024];
1251 Section *text_section;
1253 name = (char *) symtab_section->link->data + sym->st_name;
1254 text_section = s1->sections[sym->st_shndx];
1255 /* Modify reloc to target a thumb stub to switch to ARM */
1256 snprintf(buf, sizeof(buf), "%s_from_thumb", name);
1257 index = put_elf_sym(symtab_section,
1258 text_section->data_offset + 1,
1259 sym->st_size, sym->st_info, 0,
1260 sym->st_shndx, buf);
1261 rel->r_info = ELFW(R_INFO)(index, type);
1262 /* Create a thumb stub fonction to switch to ARM mode */
1263 put_elf_reloc(symtab_section, text_section,
1264 text_section->data_offset + 4, R_ARM_JUMP24,
1265 sym_index);
1266 p = section_ptr_add(text_section, 8);
1267 put32(p, 0x4778); /* bx pc */
1268 put32(p+2, 0x46c0); /* nop */
1269 put32(p+4, 0xeafffffe); /* b $sym */
1271 #elif defined(TCC_TARGET_C67)
1272 case R_C60_GOT32:
1273 case R_C60_GOTOFF:
1274 case R_C60_GOTPC:
1275 case R_C60_PLT32:
1276 if (!s1->got)
1277 build_got(s1);
1278 if (type == R_C60_GOT32 || type == R_C60_PLT32) {
1279 sym_index = ELFW(R_SYM)(rel->r_info);
1280 sym = &((ElfW(Sym) *)symtab_section->data)[sym_index];
1281 /* look at the symbol got offset. If none, then add one */
1282 if (type == R_C60_GOT32)
1283 reloc_type = R_C60_GLOB_DAT;
1284 else
1285 reloc_type = R_C60_JMP_SLOT;
1286 put_got_entry(s1, reloc_type, sym->st_size, sym->st_info,
1287 sym_index);
1289 break;
1290 #elif defined(TCC_TARGET_X86_64)
1291 case R_X86_64_GOT32:
1292 case R_X86_64_GOTTPOFF:
1293 case R_X86_64_GOTPCREL:
1294 case R_X86_64_PLT32:
1295 sym_index = ELFW(R_SYM)(rel->r_info);
1296 sym = &((ElfW(Sym) *)symtab_section->data)[sym_index];
1297 if (type == R_X86_64_PLT32 &&
1298 ELFW(ST_VISIBILITY)(sym->st_other) != STV_DEFAULT)
1300 rel->r_info = ELFW(R_INFO)(sym_index, R_X86_64_PC32);
1301 break;
1304 if (!s1->got)
1305 build_got(s1);
1306 if (type == R_X86_64_GOT32 || type == R_X86_64_GOTPCREL ||
1307 type == R_X86_64_PLT32) {
1308 unsigned long ofs;
1309 /* look at the symbol got offset. If none, then add one */
1310 if (type == R_X86_64_GOT32 || type == R_X86_64_GOTPCREL)
1311 reloc_type = R_X86_64_GLOB_DAT;
1312 else
1313 reloc_type = R_X86_64_JUMP_SLOT;
1314 ofs = put_got_entry(s1, reloc_type, sym->st_size,
1315 sym->st_info, sym_index);
1316 if (type == R_X86_64_PLT32)
1317 /* We store the place of the generated PLT slot
1318 in our addend. */
1319 rel->r_addend += ofs;
1321 break;
1322 #else
1323 #error unsupported CPU
1324 #endif
1325 default:
1326 break;
1332 ST_FUNC Section *new_symtab(TCCState *s1,
1333 const char *symtab_name, int sh_type, int sh_flags,
1334 const char *strtab_name,
1335 const char *hash_name, int hash_sh_flags)
1337 Section *symtab, *strtab, *hash;
1338 int *ptr, nb_buckets;
1340 symtab = new_section(s1, symtab_name, sh_type, sh_flags);
1341 symtab->sh_entsize = sizeof(ElfW(Sym));
1342 strtab = new_section(s1, strtab_name, SHT_STRTAB, sh_flags);
1343 put_elf_str(strtab, "");
1344 symtab->link = strtab;
1345 put_elf_sym(symtab, 0, 0, 0, 0, 0, NULL);
1347 nb_buckets = 1;
1349 hash = new_section(s1, hash_name, SHT_HASH, hash_sh_flags);
1350 hash->sh_entsize = sizeof(int);
1351 symtab->hash = hash;
1352 hash->link = symtab;
1354 ptr = section_ptr_add(hash, (2 + nb_buckets + 1) * sizeof(int));
1355 ptr[0] = nb_buckets;
1356 ptr[1] = 1;
1357 memset(ptr + 2, 0, (nb_buckets + 1) * sizeof(int));
1358 return symtab;
1361 /* put dynamic tag */
1362 static void put_dt(Section *dynamic, int dt, addr_t val)
1364 ElfW(Dyn) *dyn;
1365 dyn = section_ptr_add(dynamic, sizeof(ElfW(Dyn)));
1366 dyn->d_tag = dt;
1367 dyn->d_un.d_val = val;
1370 static void add_init_array_defines(TCCState *s1, const char *section_name)
1372 Section *s;
1373 long end_offset;
1374 char sym_start[1024];
1375 char sym_end[1024];
1377 snprintf(sym_start, sizeof(sym_start), "__%s_start", section_name + 1);
1378 snprintf(sym_end, sizeof(sym_end), "__%s_end", section_name + 1);
1380 s = find_section(s1, section_name);
1381 if (!s) {
1382 end_offset = 0;
1383 s = data_section;
1384 } else {
1385 end_offset = s->data_offset;
1388 add_elf_sym(symtab_section,
1389 0, 0,
1390 ELFW(ST_INFO)(STB_GLOBAL, STT_NOTYPE), 0,
1391 s->sh_num, sym_start);
1392 add_elf_sym(symtab_section,
1393 end_offset, 0,
1394 ELFW(ST_INFO)(STB_GLOBAL, STT_NOTYPE), 0,
1395 s->sh_num, sym_end);
1398 ST_FUNC void tcc_add_bcheck(TCCState *s1)
1400 #ifdef CONFIG_TCC_BCHECK
1401 unsigned long *ptr;
1402 Section *init_section;
1403 unsigned char *pinit;
1404 int sym_index;
1406 if (0 == s1->do_bounds_check)
1407 return;
1409 /* XXX: add an object file to do that */
1410 ptr = section_ptr_add(bounds_section, sizeof(unsigned long));
1411 *ptr = 0;
1412 add_elf_sym(symtab_section, 0, 0,
1413 ELFW(ST_INFO)(STB_GLOBAL, STT_NOTYPE), 0,
1414 bounds_section->sh_num, "__bounds_start");
1415 #ifdef TCC_TARGET_I386
1416 if (s1->output_type != TCC_OUTPUT_MEMORY) {
1417 /* add 'call __bound_init()' in .init section */
1418 init_section = find_section(s1, ".init");
1419 pinit = section_ptr_add(init_section, 5);
1420 pinit[0] = 0xe8;
1421 put32(pinit + 1, -4);
1422 sym_index = find_elf_sym(symtab_section, "__bound_init");
1423 put_elf_reloc(symtab_section, init_section,
1424 init_section->data_offset - 4, R_386_PC32, sym_index);
1426 #endif
1427 #endif
1430 static inline int tcc_add_support(TCCState *s1, const char *filename)
1432 char buf[1024];
1433 snprintf(buf, sizeof(buf), "%s/%s", s1->tcc_lib_path, filename);
1434 return tcc_add_file(s1, buf);
1437 /* add tcc runtime libraries */
1438 ST_FUNC void tcc_add_runtime(TCCState *s1)
1440 /* add libc */
1441 if (!s1->nostdlib) {
1442 tcc_add_library(s1, "c");
1443 #ifdef CONFIG_USE_LIBGCC
1444 if (!s1->static_link) {
1445 tcc_add_file(s1, TCC_LIBGCC);
1446 tcc_add_support(s1, "libtcc1.a");
1447 } else
1448 tcc_add_support(s1, "libtcc1.a");
1449 #else
1450 tcc_add_support(s1, "libtcc1.a");
1451 #endif
1454 /* tcc_add_bcheck tries to relocate a call to __bound_init in _init so
1455 libtcc1.a must be loaded before for __bound_init to be defined and
1456 crtn.o must be loaded after to not finalize _init too early. */
1457 tcc_add_bcheck(s1);
1459 if (!s1->nostdlib) {
1460 /* add crt end if not memory output */
1461 if (s1->output_type != TCC_OUTPUT_MEMORY)
1462 tcc_add_crt(s1, "crtn.o");
1466 /* add various standard linker symbols (must be done after the
1467 sections are filled (for example after allocating common
1468 symbols)) */
1469 ST_FUNC void tcc_add_linker_symbols(TCCState *s1)
1471 char buf[1024];
1472 int i;
1473 Section *s;
1475 add_elf_sym(symtab_section,
1476 text_section->data_offset, 0,
1477 ELFW(ST_INFO)(STB_GLOBAL, STT_NOTYPE), 0,
1478 text_section->sh_num, "_etext");
1479 add_elf_sym(symtab_section,
1480 data_section->data_offset, 0,
1481 ELFW(ST_INFO)(STB_GLOBAL, STT_NOTYPE), 0,
1482 data_section->sh_num, "_edata");
1483 add_elf_sym(symtab_section,
1484 bss_section->data_offset, 0,
1485 ELFW(ST_INFO)(STB_GLOBAL, STT_NOTYPE), 0,
1486 bss_section->sh_num, "_end");
1487 /* horrible new standard ldscript defines */
1488 add_init_array_defines(s1, ".preinit_array");
1489 add_init_array_defines(s1, ".init_array");
1490 add_init_array_defines(s1, ".fini_array");
1492 /* add start and stop symbols for sections whose name can be
1493 expressed in C */
1494 for(i = 1; i < s1->nb_sections; i++) {
1495 s = s1->sections[i];
1496 if (s->sh_type == SHT_PROGBITS &&
1497 (s->sh_flags & SHF_ALLOC)) {
1498 const char *p;
1499 int ch;
1501 /* check if section name can be expressed in C */
1502 p = s->name;
1503 for(;;) {
1504 ch = *p;
1505 if (!ch)
1506 break;
1507 if (!isid(ch) && !isnum(ch))
1508 goto next_sec;
1509 p++;
1511 snprintf(buf, sizeof(buf), "__start_%s", s->name);
1512 add_elf_sym(symtab_section,
1513 0, 0,
1514 ELFW(ST_INFO)(STB_GLOBAL, STT_NOTYPE), 0,
1515 s->sh_num, buf);
1516 snprintf(buf, sizeof(buf), "__stop_%s", s->name);
1517 add_elf_sym(symtab_section,
1518 s->data_offset, 0,
1519 ELFW(ST_INFO)(STB_GLOBAL, STT_NOTYPE), 0,
1520 s->sh_num, buf);
1522 next_sec: ;
1526 static void tcc_output_binary(TCCState *s1, FILE *f,
1527 const int *sec_order)
1529 Section *s;
1530 int i, offset, size;
1532 offset = 0;
1533 for(i=1;i<s1->nb_sections;i++) {
1534 s = s1->sections[sec_order[i]];
1535 if (s->sh_type != SHT_NOBITS &&
1536 (s->sh_flags & SHF_ALLOC)) {
1537 while (offset < s->sh_offset) {
1538 fputc(0, f);
1539 offset++;
1541 size = s->sh_size;
1542 fwrite(s->data, 1, size, f);
1543 offset += size;
1548 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
1549 #define HAVE_PHDR 1
1550 #define EXTRA_RELITEMS 14
1552 /* move the relocation value from .dynsym to .got */
1553 void patch_dynsym_undef(TCCState *s1, Section *s)
1555 uint32_t *gotd = (void *)s1->got->data;
1556 ElfW(Sym) *sym;
1558 gotd += 3; /* dummy entries in .got */
1559 /* relocate symbols in .dynsym */
1560 for_each_elem(s, 1, sym, ElfW(Sym)) {
1561 if (sym->st_shndx == SHN_UNDEF) {
1562 *gotd++ = sym->st_value + 6; /* XXX 6 is magic ? */
1563 sym->st_value = 0;
1567 #else
1568 #define HAVE_PHDR 1
1569 #define EXTRA_RELITEMS 9
1571 /* zero plt offsets of weak symbols in .dynsym */
1572 void patch_dynsym_undef(TCCState *s1, Section *s)
1574 ElfW(Sym) *sym;
1576 for_each_elem(s, 1, sym, ElfW(Sym))
1577 if (sym->st_shndx == SHN_UNDEF && ELFW(ST_BIND)(sym->st_info) == STB_WEAK)
1578 sym->st_value = 0;
1580 #endif
1582 ST_FUNC void fill_got_entry(TCCState *s1, ElfW_Rel *rel)
1584 int sym_index = ELFW(R_SYM) (rel->r_info);
1585 ElfW(Sym) *sym = &((ElfW(Sym) *) symtab_section->data)[sym_index];
1586 unsigned long offset;
1588 if (sym_index >= s1->nb_sym_attrs)
1589 return;
1590 offset = s1->sym_attrs[sym_index].got_offset;
1591 section_reserve(s1->got, offset + PTR_SIZE);
1592 #ifdef TCC_TARGET_X86_64
1593 /* only works for x86-64 */
1594 put32(s1->got->data + offset + 4, sym->st_value >> 32);
1595 #endif
1596 put32(s1->got->data + offset, sym->st_value & 0xffffffff);
1599 /* Perform relocation to GOT or PLT entries */
1600 ST_FUNC void fill_got(TCCState *s1)
1602 Section *s;
1603 ElfW_Rel *rel;
1604 int i;
1606 for(i = 1; i < s1->nb_sections; i++) {
1607 s = s1->sections[i];
1608 if (s->sh_type != SHT_RELX)
1609 continue;
1610 /* no need to handle got relocations */
1611 if (s->link != symtab_section)
1612 continue;
1613 for_each_elem(s, 0, rel, ElfW_Rel) {
1614 switch (ELFW(R_TYPE) (rel->r_info)) {
1615 case R_X86_64_GOT32:
1616 case R_X86_64_GOTPCREL:
1617 case R_X86_64_PLT32:
1618 fill_got_entry(s1, rel);
1619 break;
1625 /* Bind symbols of executable: resolve undefined symbols from exported symbols
1626 in shared libraries and export non local defined symbols to shared libraries
1627 if -rdynamic switch was given on command line */
1628 static void bind_exe_dynsyms(TCCState *s1)
1630 const char *name;
1631 int sym_index, index;
1632 ElfW(Sym) *sym, *esym;
1633 int type;
1635 /* Resolve undefined symbols from dynamic symbols. When there is a match:
1636 - if STT_FUNC or STT_GNU_IFUNC symbol -> add it in PLT
1637 - if STT_OBJECT symbol -> add it in .bss section with suitable reloc */
1638 for_each_elem(symtab_section, 1, sym, ElfW(Sym)) {
1639 if (sym->st_shndx == SHN_UNDEF) {
1640 name = (char *) symtab_section->link->data + sym->st_name;
1641 sym_index = find_elf_sym(s1->dynsymtab_section, name);
1642 if (sym_index) {
1643 esym = &((ElfW(Sym) *)s1->dynsymtab_section->data)[sym_index];
1644 type = ELFW(ST_TYPE)(esym->st_info);
1645 if ((type == STT_FUNC) || (type == STT_GNU_IFUNC)) {
1646 /* Indirect functions shall have STT_FUNC type in executable
1647 * dynsym section. Indeed, a dlsym call following a lazy
1648 * resolution would pick the symbol value from the
1649 * executable dynsym entry which would contain the address
1650 * of the function wanted by the caller of dlsym instead of
1651 * the address of the function that would return that
1652 * address */
1653 put_got_entry(s1, R_JMP_SLOT, esym->st_size,
1654 ELFW(ST_INFO)(STB_GLOBAL,STT_FUNC),
1655 sym - (ElfW(Sym) *)symtab_section->data);
1656 } else if (type == STT_OBJECT) {
1657 unsigned long offset;
1658 ElfW(Sym) *dynsym;
1659 offset = bss_section->data_offset;
1660 /* XXX: which alignment ? */
1661 offset = (offset + 16 - 1) & -16;
1662 index = put_elf_sym(s1->dynsym, offset, esym->st_size,
1663 esym->st_info, 0, bss_section->sh_num,
1664 name);
1665 /* Ensure R_COPY works for weak symbol aliases */
1666 if (ELFW(ST_BIND)(esym->st_info) == STB_WEAK) {
1667 for_each_elem(s1->dynsymtab_section, 1, dynsym, ElfW(Sym)) {
1668 if ((dynsym->st_value == esym->st_value)
1669 && (ELFW(ST_BIND)(dynsym->st_info) == STB_GLOBAL)) {
1670 char *dynname = (char *) s1->dynsymtab_section->link->data
1671 + dynsym->st_name;
1672 put_elf_sym(s1->dynsym, offset, dynsym->st_size,
1673 dynsym->st_info, 0,
1674 bss_section->sh_num, dynname);
1675 break;
1679 put_elf_reloc(s1->dynsym, bss_section,
1680 offset, R_COPY, index);
1681 offset += esym->st_size;
1682 bss_section->data_offset = offset;
1684 } else {
1685 /* STB_WEAK undefined symbols are accepted */
1686 /* XXX: _fp_hw seems to be part of the ABI, so we ignore it */
1687 if (ELFW(ST_BIND)(sym->st_info) == STB_WEAK ||
1688 !strcmp(name, "_fp_hw")) {
1689 } else {
1690 tcc_error_noabort("undefined symbol '%s'", name);
1693 } else if (s1->rdynamic && ELFW(ST_BIND)(sym->st_info) != STB_LOCAL) {
1694 /* if -rdynamic option, then export all non local symbols */
1695 name = (char *) symtab_section->link->data + sym->st_name;
1696 put_elf_sym(s1->dynsym, sym->st_value, sym->st_size, sym->st_info,
1697 0, sym->st_shndx, name);
1702 /* Bind symbols of libraries: export non local symbols of executable that
1703 resolve undefined symbols of shared libraries */
1704 static void bind_libs_dynsyms(TCCState *s1)
1706 const char *name;
1707 int sym_index;
1708 ElfW(Sym) *sym, *esym;
1710 /* now look at unresolved dynamic symbols and export
1711 corresponding symbol */
1712 for_each_elem(s1->dynsymtab_section, 1, esym, ElfW(Sym)) {
1713 if (esym->st_shndx == SHN_UNDEF) {
1714 name = (char *) s1->dynsymtab_section->link->data + esym->st_name;
1715 sym_index = find_elf_sym(symtab_section, name);
1716 if (sym_index) {
1717 /* XXX: avoid adding a symbol if already present because of
1718 -rdynamic ? */
1719 sym = &((ElfW(Sym) *)symtab_section->data)[sym_index];
1720 put_elf_sym(s1->dynsym, sym->st_value, sym->st_size,
1721 sym->st_info, 0, sym->st_shndx, name);
1722 } else {
1723 /* weak symbols can stay undefined */
1724 if (ELFW(ST_BIND)(esym->st_info) != STB_WEAK)
1725 tcc_warning("undefined dynamic symbol '%s'", name);
1731 /* Export all non local symbols (for shared libraries) */
1732 static void export_global_syms(TCCState *s1)
1734 int nb_syms, dynindex, index;
1735 const char *name;
1736 ElfW(Sym) *sym;
1738 nb_syms = symtab_section->data_offset / sizeof(ElfW(Sym));
1739 s1->symtab_to_dynsym = tcc_mallocz(sizeof(int) * nb_syms);
1740 for_each_elem(symtab_section, 1, sym, ElfW(Sym)) {
1741 if (ELFW(ST_BIND)(sym->st_info) != STB_LOCAL) {
1742 name = (char *) symtab_section->link->data + sym->st_name;
1743 dynindex = put_elf_sym(s1->dynsym, sym->st_value, sym->st_size,
1744 sym->st_info, 0, sym->st_shndx, name);
1745 index = sym - (ElfW(Sym) *) symtab_section->data;
1746 s1->symtab_to_dynsym[index] = dynindex;
1751 /* relocate the PLT: compute addresses and offsets in the PLT now that final
1752 address for PLT and GOT are known (see fill_program_header) */
1753 ST_FUNC void relocate_plt(TCCState *s1)
1755 uint8_t *p, *p_end;
1757 if (!s1->plt)
1758 return;
1760 p = s1->plt->data;
1761 p_end = p + s1->plt->data_offset;
1762 if (p < p_end) {
1763 #if defined(TCC_TARGET_I386)
1764 put32(p + 2, get32(p + 2) + s1->got->sh_addr);
1765 put32(p + 8, get32(p + 8) + s1->got->sh_addr);
1766 p += 16;
1767 while (p < p_end) {
1768 put32(p + 2, get32(p + 2) + s1->got->sh_addr);
1769 p += 16;
1771 #elif defined(TCC_TARGET_X86_64)
1772 int x = s1->got->sh_addr - s1->plt->sh_addr - 6;
1773 put32(p + 2, get32(p + 2) + x);
1774 put32(p + 8, get32(p + 8) + x - 6);
1775 p += 16;
1776 while (p < p_end) {
1777 put32(p + 2, get32(p + 2) + x + s1->plt->data - p);
1778 p += 16;
1780 #elif defined(TCC_TARGET_ARM)
1781 int x;
1782 x=s1->got->sh_addr - s1->plt->sh_addr - 12;
1783 p += 16;
1784 while (p < p_end) {
1785 if (get32(p) == 0x46c04778) /* PLT Thumb stub present */
1786 p += 4;
1787 put32(p + 12, x + get32(p + 12) + s1->plt->data - p);
1788 p += 16;
1790 #elif defined(TCC_TARGET_C67)
1791 /* XXX: TODO */
1792 #else
1793 #error unsupported CPU
1794 #endif
1798 /* Allocate strings for section names and decide if an unallocated section
1799 should be output.
1801 NOTE: the strsec section comes last, so its size is also correct ! */
1802 static void alloc_sec_names(TCCState *s1, int file_type, Section *strsec)
1804 int i;
1805 Section *s;
1807 /* Allocate strings for section names */
1808 for(i = 1; i < s1->nb_sections; i++) {
1809 s = s1->sections[i];
1810 s->sh_name = put_elf_str(strsec, s->name);
1811 /* when generating a DLL, we include relocations but we may
1812 patch them */
1813 if (file_type == TCC_OUTPUT_DLL &&
1814 s->sh_type == SHT_RELX &&
1815 !(s->sh_flags & SHF_ALLOC)) {
1816 /* gr: avoid bogus relocs for empty (debug) sections */
1817 if (s1->sections[s->sh_info]->sh_flags & SHF_ALLOC)
1818 prepare_dynamic_rel(s1, s);
1819 else if (s1->do_debug)
1820 s->sh_size = s->data_offset;
1821 } else if (s1->do_debug ||
1822 file_type == TCC_OUTPUT_OBJ ||
1823 (s->sh_flags & SHF_ALLOC) ||
1824 i == (s1->nb_sections - 1)) {
1825 /* we output all sections if debug or object file */
1826 s->sh_size = s->data_offset;
1831 /* Info to be copied in dynamic section */
1832 struct dyn_inf {
1833 Section *dynamic;
1834 Section *dynstr;
1835 unsigned long dyn_rel_off;
1836 addr_t rel_addr;
1837 addr_t rel_size;
1838 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
1839 addr_t bss_addr;
1840 addr_t bss_size;
1841 #endif
1844 /* Assign sections to segments and decide how are sections laid out when loaded
1845 in memory. This function also fills corresponding program headers. */
1846 static int layout_sections(TCCState *s1, ElfW(Phdr) *phdr, int phnum,
1847 Section *interp, struct dyn_inf *dyninf,
1848 int *sec_order)
1850 int i, j, k, file_type, sh_order_index, file_offset;
1851 long long tmp;
1852 addr_t addr;
1853 ElfW(Phdr) *ph;
1854 Section *s;
1856 file_type = s1->output_type;
1857 sh_order_index = 1;
1858 if (s1->output_format == TCC_OUTPUT_FORMAT_ELF)
1859 file_offset = sizeof(ElfW(Ehdr)) + phnum * sizeof(ElfW(Phdr));
1860 else
1861 file_offset = 0;
1863 if (phnum > 0) {
1864 if (s1->has_text_addr) {
1865 int a_offset, p_offset;
1866 addr = s1->text_addr;
1867 /* we ensure that (addr % ELF_PAGE_SIZE) == file_offset %
1868 ELF_PAGE_SIZE */
1869 a_offset = (int) (addr & (s1->section_align - 1));
1870 p_offset = file_offset & (s1->section_align - 1);
1871 if (a_offset < p_offset)
1872 a_offset += s1->section_align;
1873 file_offset += (a_offset - p_offset);
1874 } else {
1875 if (file_type == TCC_OUTPUT_DLL)
1876 addr = 0;
1877 else
1878 addr = ELF_START_ADDR;
1879 /* compute address after headers */
1880 addr += (file_offset & (s1->section_align - 1));
1883 ph = &phdr[0];
1884 /* Leave one program headers for the program interpreter and one for
1885 the program header table itself if needed. These are done later as
1886 they require section layout to be done first. */
1887 if (interp)
1888 ph += 1 + HAVE_PHDR;
1890 /* dynamic relocation table information, for .dynamic section */
1891 dyninf->rel_addr = dyninf->rel_size = 0;
1892 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
1893 dyninf->bss_addr = dyninf->bss_size = 0;
1894 #endif
1896 for(j = 0; j < 2; j++) {
1897 ph->p_type = PT_LOAD;
1898 if (j == 0)
1899 ph->p_flags = PF_R | PF_X;
1900 else
1901 ph->p_flags = PF_R | PF_W;
1902 ph->p_align = s1->section_align;
1904 /* Decide the layout of sections loaded in memory. This must
1905 be done before program headers are filled since they contain
1906 info about the layout. We do the following ordering: interp,
1907 symbol tables, relocations, progbits, nobits */
1908 /* XXX: do faster and simpler sorting */
1909 for(k = 0; k < 5; k++) {
1910 for(i = 1; i < s1->nb_sections; i++) {
1911 s = s1->sections[i];
1912 /* compute if section should be included */
1913 if (j == 0) {
1914 if ((s->sh_flags & (SHF_ALLOC | SHF_WRITE)) !=
1915 SHF_ALLOC)
1916 continue;
1917 } else {
1918 if ((s->sh_flags & (SHF_ALLOC | SHF_WRITE)) !=
1919 (SHF_ALLOC | SHF_WRITE))
1920 continue;
1922 if (s == interp) {
1923 if (k != 0)
1924 continue;
1925 } else if (s->sh_type == SHT_DYNSYM ||
1926 s->sh_type == SHT_STRTAB ||
1927 s->sh_type == SHT_HASH) {
1928 if (k != 1)
1929 continue;
1930 } else if (s->sh_type == SHT_RELX) {
1931 if (k != 2)
1932 continue;
1933 } else if (s->sh_type == SHT_NOBITS) {
1934 if (k != 4)
1935 continue;
1936 } else {
1937 if (k != 3)
1938 continue;
1940 sec_order[sh_order_index++] = i;
1942 /* section matches: we align it and add its size */
1943 tmp = addr;
1944 addr = (addr + s->sh_addralign - 1) &
1945 ~(s->sh_addralign - 1);
1946 file_offset += (int) ( addr - tmp );
1947 s->sh_offset = file_offset;
1948 s->sh_addr = addr;
1950 /* update program header infos */
1951 if (ph->p_offset == 0) {
1952 ph->p_offset = file_offset;
1953 ph->p_vaddr = addr;
1954 ph->p_paddr = ph->p_vaddr;
1956 /* update dynamic relocation infos */
1957 if (s->sh_type == SHT_RELX) {
1958 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
1959 if (!strcmp(strsec->data + s->sh_name, ".rel.got")) {
1960 dyninf->rel_addr = addr;
1961 dyninf->rel_size += s->sh_size; /* XXX only first rel. */
1963 if (!strcmp(strsec->data + s->sh_name, ".rel.bss")) {
1964 dyninf->bss_addr = addr;
1965 dyninf->bss_size = s->sh_size; /* XXX only first rel. */
1967 #else
1968 if (dyninf->rel_size == 0)
1969 dyninf->rel_addr = addr;
1970 dyninf->rel_size += s->sh_size;
1971 #endif
1973 addr += s->sh_size;
1974 if (s->sh_type != SHT_NOBITS)
1975 file_offset += s->sh_size;
1978 if (j == 0) {
1979 /* Make the first PT_LOAD segment include the program
1980 headers itself (and the ELF header as well), it'll
1981 come out with same memory use but will make various
1982 tools like binutils strip work better. */
1983 ph->p_offset &= ~(ph->p_align - 1);
1984 ph->p_vaddr &= ~(ph->p_align - 1);
1985 ph->p_paddr &= ~(ph->p_align - 1);
1987 ph->p_filesz = file_offset - ph->p_offset;
1988 ph->p_memsz = addr - ph->p_vaddr;
1989 ph++;
1990 if (j == 0) {
1991 if (s1->output_format == TCC_OUTPUT_FORMAT_ELF) {
1992 /* if in the middle of a page, we duplicate the page in
1993 memory so that one copy is RX and the other is RW */
1994 if ((addr & (s1->section_align - 1)) != 0)
1995 addr += s1->section_align;
1996 } else {
1997 addr = (addr + s1->section_align - 1) & ~(s1->section_align - 1);
1998 file_offset = (file_offset + s1->section_align - 1) &
1999 ~(s1->section_align - 1);
2005 /* all other sections come after */
2006 for(i = 1; i < s1->nb_sections; i++) {
2007 s = s1->sections[i];
2008 if (phnum > 0 && (s->sh_flags & SHF_ALLOC))
2009 continue;
2010 sec_order[sh_order_index++] = i;
2012 file_offset = (file_offset + s->sh_addralign - 1) &
2013 ~(s->sh_addralign - 1);
2014 s->sh_offset = file_offset;
2015 if (s->sh_type != SHT_NOBITS)
2016 file_offset += s->sh_size;
2019 return file_offset;
2022 static void fill_unloadable_phdr(ElfW(Phdr) *phdr, int phnum, Section *interp,
2023 Section *dynamic)
2025 ElfW(Phdr) *ph;
2027 /* if interpreter, then add corresponding program header */
2028 if (interp) {
2029 ph = &phdr[0];
2031 if (HAVE_PHDR)
2033 int len = phnum * sizeof(ElfW(Phdr));
2035 ph->p_type = PT_PHDR;
2036 ph->p_offset = sizeof(ElfW(Ehdr));
2037 ph->p_vaddr = interp->sh_addr - len;
2038 ph->p_paddr = ph->p_vaddr;
2039 ph->p_filesz = ph->p_memsz = len;
2040 ph->p_flags = PF_R | PF_X;
2041 ph->p_align = 4; /* interp->sh_addralign; */
2042 ph++;
2045 ph->p_type = PT_INTERP;
2046 ph->p_offset = interp->sh_offset;
2047 ph->p_vaddr = interp->sh_addr;
2048 ph->p_paddr = ph->p_vaddr;
2049 ph->p_filesz = interp->sh_size;
2050 ph->p_memsz = interp->sh_size;
2051 ph->p_flags = PF_R;
2052 ph->p_align = interp->sh_addralign;
2055 /* if dynamic section, then add corresponding program header */
2056 if (dynamic) {
2057 ph = &phdr[phnum - 1];
2059 ph->p_type = PT_DYNAMIC;
2060 ph->p_offset = dynamic->sh_offset;
2061 ph->p_vaddr = dynamic->sh_addr;
2062 ph->p_paddr = ph->p_vaddr;
2063 ph->p_filesz = dynamic->sh_size;
2064 ph->p_memsz = dynamic->sh_size;
2065 ph->p_flags = PF_R | PF_W;
2066 ph->p_align = dynamic->sh_addralign;
2070 /* Fill the dynamic section with tags describing the address and size of
2071 sections */
2072 static void fill_dynamic(TCCState *s1, struct dyn_inf *dyninf)
2074 Section *dynamic;
2076 dynamic = dyninf->dynamic;
2078 /* put dynamic section entries */
2079 dynamic->data_offset = dyninf->dyn_rel_off;
2080 put_dt(dynamic, DT_HASH, s1->dynsym->hash->sh_addr);
2081 put_dt(dynamic, DT_STRTAB, dyninf->dynstr->sh_addr);
2082 put_dt(dynamic, DT_SYMTAB, s1->dynsym->sh_addr);
2083 put_dt(dynamic, DT_STRSZ, dyninf->dynstr->data_offset);
2084 put_dt(dynamic, DT_SYMENT, sizeof(ElfW(Sym)));
2085 #ifdef TCC_TARGET_X86_64
2086 put_dt(dynamic, DT_RELA, dyninf->rel_addr);
2087 put_dt(dynamic, DT_RELASZ, dyninf->rel_size);
2088 put_dt(dynamic, DT_RELAENT, sizeof(ElfW_Rel));
2089 #else
2090 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
2091 put_dt(dynamic, DT_PLTGOT, s1->got->sh_addr);
2092 put_dt(dynamic, DT_PLTRELSZ, dyninf->rel_size);
2093 put_dt(dynamic, DT_JMPREL, dyninf->rel_addr);
2094 put_dt(dynamic, DT_PLTREL, DT_REL);
2095 put_dt(dynamic, DT_REL, dyninf->bss_addr);
2096 put_dt(dynamic, DT_RELSZ, dyninf->bss_size);
2097 #else
2098 put_dt(dynamic, DT_REL, dyninf->rel_addr);
2099 put_dt(dynamic, DT_RELSZ, dyninf->rel_size);
2100 put_dt(dynamic, DT_RELENT, sizeof(ElfW_Rel));
2101 #endif
2102 #endif
2103 if (s1->do_debug)
2104 put_dt(dynamic, DT_DEBUG, 0);
2105 put_dt(dynamic, DT_NULL, 0);
2108 /* Relocate remaining sections and symbols (that is those not related to
2109 dynamic linking) */
2110 static int final_sections_reloc(TCCState *s1)
2112 int i;
2113 Section *s;
2115 relocate_syms(s1, 0);
2117 if (s1->nb_errors != 0)
2118 return -1;
2120 /* relocate sections */
2121 /* XXX: ignore sections with allocated relocations ? */
2122 for(i = 1; i < s1->nb_sections; i++) {
2123 s = s1->sections[i];
2124 if (s->reloc && s != s1->got)
2125 relocate_section(s1, s);
2128 /* relocate relocation entries if the relocation tables are
2129 allocated in the executable */
2130 for(i = 1; i < s1->nb_sections; i++) {
2131 s = s1->sections[i];
2132 if ((s->sh_flags & SHF_ALLOC) &&
2133 s->sh_type == SHT_RELX) {
2134 relocate_rel(s1, s);
2137 return 0;
2140 /* Create an ELF file on disk.
2141 This function handle ELF specific layout requirements */
2142 static void tcc_output_elf(TCCState *s1, FILE *f, int phnum, ElfW(Phdr) *phdr,
2143 int file_offset, int *sec_order)
2145 int i, shnum, offset, size, file_type;
2146 Section *s;
2147 ElfW(Ehdr) ehdr;
2148 ElfW(Shdr) shdr, *sh;
2150 file_type = s1->output_type;
2151 shnum = s1->nb_sections;
2153 memset(&ehdr, 0, sizeof(ehdr));
2155 if (phnum > 0) {
2156 ehdr.e_phentsize = sizeof(ElfW(Phdr));
2157 ehdr.e_phnum = phnum;
2158 ehdr.e_phoff = sizeof(ElfW(Ehdr));
2161 /* align to 4 */
2162 file_offset = (file_offset + 3) & -4;
2164 /* fill header */
2165 ehdr.e_ident[0] = ELFMAG0;
2166 ehdr.e_ident[1] = ELFMAG1;
2167 ehdr.e_ident[2] = ELFMAG2;
2168 ehdr.e_ident[3] = ELFMAG3;
2169 ehdr.e_ident[4] = ELFCLASSW;
2170 ehdr.e_ident[5] = ELFDATA2LSB;
2171 ehdr.e_ident[6] = EV_CURRENT;
2172 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
2173 ehdr.e_ident[EI_OSABI] = ELFOSABI_FREEBSD;
2174 #endif
2175 #ifdef TCC_TARGET_ARM
2176 #ifdef TCC_ARM_EABI
2177 ehdr.e_ident[EI_OSABI] = 0;
2178 ehdr.e_flags = EF_ARM_EABI_VER4;
2179 if (file_type == TCC_OUTPUT_EXE || file_type == TCC_OUTPUT_DLL)
2180 ehdr.e_flags |= EF_ARM_HASENTRY;
2181 if (s1->float_abi == ARM_HARD_FLOAT)
2182 ehdr.e_flags |= EF_ARM_VFP_FLOAT;
2183 else
2184 ehdr.e_flags |= EF_ARM_SOFT_FLOAT;
2185 #else
2186 ehdr.e_ident[EI_OSABI] = ELFOSABI_ARM;
2187 #endif
2188 #endif
2189 switch(file_type) {
2190 default:
2191 case TCC_OUTPUT_EXE:
2192 ehdr.e_type = ET_EXEC;
2193 ehdr.e_entry = get_elf_sym_addr(s1, "_start", 1);
2194 break;
2195 case TCC_OUTPUT_DLL:
2196 ehdr.e_type = ET_DYN;
2197 ehdr.e_entry = text_section->sh_addr; /* XXX: is it correct ? */
2198 break;
2199 case TCC_OUTPUT_OBJ:
2200 ehdr.e_type = ET_REL;
2201 break;
2203 ehdr.e_machine = EM_TCC_TARGET;
2204 ehdr.e_version = EV_CURRENT;
2205 ehdr.e_shoff = file_offset;
2206 ehdr.e_ehsize = sizeof(ElfW(Ehdr));
2207 ehdr.e_shentsize = sizeof(ElfW(Shdr));
2208 ehdr.e_shnum = shnum;
2209 ehdr.e_shstrndx = shnum - 1;
2211 fwrite(&ehdr, 1, sizeof(ElfW(Ehdr)), f);
2212 fwrite(phdr, 1, phnum * sizeof(ElfW(Phdr)), f);
2213 offset = sizeof(ElfW(Ehdr)) + phnum * sizeof(ElfW(Phdr));
2215 sort_syms(s1, symtab_section);
2216 for(i = 1; i < s1->nb_sections; i++) {
2217 s = s1->sections[sec_order[i]];
2218 if (s->sh_type != SHT_NOBITS) {
2219 if (s->sh_type == SHT_DYNSYM)
2220 patch_dynsym_undef(s1, s);
2221 while (offset < s->sh_offset) {
2222 fputc(0, f);
2223 offset++;
2225 size = s->sh_size;
2226 fwrite(s->data, 1, size, f);
2227 offset += size;
2231 /* output section headers */
2232 while (offset < ehdr.e_shoff) {
2233 fputc(0, f);
2234 offset++;
2237 for(i = 0; i < s1->nb_sections; i++) {
2238 sh = &shdr;
2239 memset(sh, 0, sizeof(ElfW(Shdr)));
2240 s = s1->sections[i];
2241 if (s) {
2242 sh->sh_name = s->sh_name;
2243 sh->sh_type = s->sh_type;
2244 sh->sh_flags = s->sh_flags;
2245 sh->sh_entsize = s->sh_entsize;
2246 sh->sh_info = s->sh_info;
2247 if (s->link)
2248 sh->sh_link = s->link->sh_num;
2249 sh->sh_addralign = s->sh_addralign;
2250 sh->sh_addr = s->sh_addr;
2251 sh->sh_offset = s->sh_offset;
2252 sh->sh_size = s->sh_size;
2254 fwrite(sh, 1, sizeof(ElfW(Shdr)), f);
2258 /* Write an elf, coff or "binary" file */
2259 static int tcc_write_elf_file(TCCState *s1, const char *filename, int phnum,
2260 ElfW(Phdr) *phdr, int file_offset, int *sec_order)
2262 int fd, mode, file_type;
2263 FILE *f;
2265 file_type = s1->output_type;
2266 if (file_type == TCC_OUTPUT_OBJ)
2267 mode = 0666;
2268 else
2269 mode = 0777;
2270 unlink(filename);
2271 fd = open(filename, O_WRONLY | O_CREAT | O_TRUNC | O_BINARY, mode);
2272 if (fd < 0) {
2273 tcc_error_noabort("could not write '%s'", filename);
2274 return -1;
2276 f = fdopen(fd, "wb");
2277 if (s1->verbose)
2278 printf("<- %s\n", filename);
2280 #ifdef TCC_TARGET_COFF
2281 if (s1->output_format == TCC_OUTPUT_FORMAT_COFF)
2282 tcc_output_coff(s1, f);
2283 else
2284 #endif
2285 if (s1->output_format == TCC_OUTPUT_FORMAT_ELF)
2286 tcc_output_elf(s1, f, phnum, phdr, file_offset, sec_order);
2287 else
2288 tcc_output_binary(s1, f, sec_order);
2289 fclose(f);
2291 return 0;
2294 /* Output an elf, coff or binary file */
2295 /* XXX: suppress unneeded sections */
2296 static int elf_output_file(TCCState *s1, const char *filename)
2298 int i, ret, phnum, shnum, file_type, file_offset, *sec_order;
2299 struct dyn_inf dyninf;
2300 ElfW(Phdr) *phdr;
2301 ElfW(Sym) *sym;
2302 Section *strsec, *interp, *dynamic, *dynstr;
2304 file_type = s1->output_type;
2305 s1->nb_errors = 0;
2307 /* if linking, also link in runtime libraries (libc, libgcc, etc.) */
2308 if (file_type != TCC_OUTPUT_OBJ) {
2309 tcc_add_runtime(s1);
2312 phdr = NULL;
2313 sec_order = NULL;
2314 interp = dynamic = dynstr = NULL; /* avoid warning */
2315 dyninf.dyn_rel_off = 0; /* avoid warning */
2317 if (file_type != TCC_OUTPUT_OBJ) {
2318 relocate_common_syms();
2320 tcc_add_linker_symbols(s1);
2322 if (!s1->static_link) {
2323 if (file_type == TCC_OUTPUT_EXE) {
2324 char *ptr;
2325 /* allow override the dynamic loader */
2326 const char *elfint = getenv("LD_SO");
2327 if (elfint == NULL)
2328 elfint = DEFAULT_ELFINTERP(s1);
2329 /* add interpreter section only if executable */
2330 interp = new_section(s1, ".interp", SHT_PROGBITS, SHF_ALLOC);
2331 interp->sh_addralign = 1;
2332 ptr = section_ptr_add(interp, 1 + strlen(elfint));
2333 strcpy(ptr, elfint);
2336 /* add dynamic symbol table */
2337 s1->dynsym = new_symtab(s1, ".dynsym", SHT_DYNSYM, SHF_ALLOC,
2338 ".dynstr",
2339 ".hash", SHF_ALLOC);
2340 dynstr = s1->dynsym->link;
2342 /* add dynamic section */
2343 dynamic = new_section(s1, ".dynamic", SHT_DYNAMIC,
2344 SHF_ALLOC | SHF_WRITE);
2345 dynamic->link = dynstr;
2346 dynamic->sh_entsize = sizeof(ElfW(Dyn));
2348 build_got(s1);
2350 if (file_type == TCC_OUTPUT_EXE) {
2351 bind_exe_dynsyms(s1);
2353 if (s1->nb_errors) {
2354 ret = -1;
2355 goto the_end;
2358 bind_libs_dynsyms(s1);
2359 } else /* shared library case: simply export all global symbols */
2360 export_global_syms(s1);
2362 build_got_entries(s1);
2364 /* add a list of needed dlls */
2365 for(i = 0; i < s1->nb_loaded_dlls; i++) {
2366 DLLReference *dllref = s1->loaded_dlls[i];
2367 if (dllref->level == 0)
2368 put_dt(dynamic, DT_NEEDED, put_elf_str(dynstr, dllref->name));
2371 if (s1->rpath)
2372 put_dt(dynamic, DT_RPATH, put_elf_str(dynstr, s1->rpath));
2374 /* XXX: currently, since we do not handle PIC code, we
2375 must relocate the readonly segments */
2376 if (file_type == TCC_OUTPUT_DLL) {
2377 if (s1->soname)
2378 put_dt(dynamic, DT_SONAME, put_elf_str(dynstr, s1->soname));
2379 put_dt(dynamic, DT_TEXTREL, 0);
2382 if (s1->symbolic)
2383 put_dt(dynamic, DT_SYMBOLIC, 0);
2385 /* add necessary space for other entries */
2386 dyninf.dyn_rel_off = dynamic->data_offset;
2387 dynamic->data_offset += sizeof(ElfW(Dyn)) * EXTRA_RELITEMS;
2388 } else {
2389 /* still need to build got entries in case of static link */
2390 build_got_entries(s1);
2394 /* we add a section for symbols */
2395 strsec = new_section(s1, ".shstrtab", SHT_STRTAB, 0);
2396 put_elf_str(strsec, "");
2398 /* compute number of sections */
2399 shnum = s1->nb_sections;
2401 /* this array is used to reorder sections in the output file */
2402 sec_order = tcc_malloc(sizeof(int) * shnum);
2403 sec_order[0] = 0;
2405 /* compute number of program headers */
2406 switch(file_type) {
2407 default:
2408 case TCC_OUTPUT_OBJ:
2409 phnum = 0;
2410 break;
2411 case TCC_OUTPUT_EXE:
2412 if (!s1->static_link)
2413 phnum = 4 + HAVE_PHDR;
2414 else
2415 phnum = 2;
2416 break;
2417 case TCC_OUTPUT_DLL:
2418 phnum = 3;
2419 break;
2422 /* Allocate strings for section names */
2423 alloc_sec_names(s1, file_type, strsec);
2425 /* allocate program segment headers */
2426 phdr = tcc_mallocz(phnum * sizeof(ElfW(Phdr)));
2428 /* compute section to program header mapping */
2429 file_offset = layout_sections(s1, phdr, phnum, interp, &dyninf, sec_order);
2431 /* Fill remaining program header and finalize relocation related to dynamic
2432 linking. */
2433 if (phnum > 0) {
2434 fill_unloadable_phdr(phdr, phnum, interp, dynamic);
2435 if (dynamic) {
2436 dyninf.dynamic = dynamic;
2437 dyninf.dynstr = dynstr;
2439 fill_dynamic(s1, &dyninf);
2441 /* put in GOT the dynamic section address and relocate PLT */
2442 put32(s1->got->data, dynamic->sh_addr);
2443 if (file_type == TCC_OUTPUT_EXE
2444 #if defined(TCC_OUTPUT_DLL_WITH_PLT)
2445 || file_type == TCC_OUTPUT_DLL
2446 #endif
2448 relocate_plt(s1);
2450 /* relocate symbols in .dynsym now that final addresses are known */
2451 for_each_elem(s1->dynsym, 1, sym, ElfW(Sym)) {
2452 /* relocate to PLT if symbol corresponds to a PLT entry */
2453 if (sym->st_shndx == SHN_UNDEF) {
2454 if (sym->st_value)
2455 sym->st_value += s1->plt->sh_addr;
2456 } else if (sym->st_shndx < SHN_LORESERVE) {
2457 /* do symbol relocation */
2458 sym->st_value += s1->sections[sym->st_shndx]->sh_addr;
2464 /* if building executable or DLL, then relocate each section
2465 except the GOT which is already relocated */
2466 if (file_type != TCC_OUTPUT_OBJ) {
2467 ret = final_sections_reloc(s1);
2468 if (ret)
2469 goto the_end;
2472 /* Perform relocation to GOT or PLT entries */
2473 if (file_type == TCC_OUTPUT_EXE && s1->static_link)
2474 fill_got(s1);
2476 /* Create the ELF file with name 'filename' */
2477 ret = tcc_write_elf_file(s1, filename, phnum, phdr, file_offset, sec_order);
2478 the_end:
2479 tcc_free(s1->symtab_to_dynsym);
2480 tcc_free(sec_order);
2481 tcc_free(phdr);
2482 tcc_free(s1->sym_attrs);
2483 s1->sym_attrs = NULL;
2484 return ret;
2487 LIBTCCAPI int tcc_output_file(TCCState *s, const char *filename)
2489 int ret;
2490 #ifdef TCC_TARGET_PE
2491 if (s->output_type != TCC_OUTPUT_OBJ) {
2492 ret = pe_output_file(s, filename);
2493 } else
2494 #endif
2495 ret = elf_output_file(s, filename);
2496 return ret;
2499 static void *load_data(int fd, unsigned long file_offset, unsigned long size)
2501 void *data;
2503 data = tcc_malloc(size);
2504 lseek(fd, file_offset, SEEK_SET);
2505 read(fd, data, size);
2506 return data;
2509 typedef struct SectionMergeInfo {
2510 Section *s; /* corresponding existing section */
2511 unsigned long offset; /* offset of the new section in the existing section */
2512 uint8_t new_section; /* true if section 's' was added */
2513 uint8_t link_once; /* true if link once section */
2514 } SectionMergeInfo;
2516 /* load an object file and merge it with current files */
2517 /* XXX: handle correctly stab (debug) info */
2518 ST_FUNC int tcc_load_object_file(TCCState *s1,
2519 int fd, unsigned long file_offset)
2521 ElfW(Ehdr) ehdr;
2522 ElfW(Shdr) *shdr, *sh;
2523 int size, i, j, offset, offseti, nb_syms, sym_index, ret;
2524 unsigned char *strsec, *strtab;
2525 int *old_to_new_syms;
2526 char *sh_name, *name;
2527 SectionMergeInfo *sm_table, *sm;
2528 ElfW(Sym) *sym, *symtab;
2529 ElfW_Rel *rel;
2530 Section *s;
2532 int stab_index;
2533 int stabstr_index;
2535 stab_index = stabstr_index = 0;
2537 if (read(fd, &ehdr, sizeof(ehdr)) != sizeof(ehdr))
2538 goto fail1;
2539 if (ehdr.e_ident[0] != ELFMAG0 ||
2540 ehdr.e_ident[1] != ELFMAG1 ||
2541 ehdr.e_ident[2] != ELFMAG2 ||
2542 ehdr.e_ident[3] != ELFMAG3)
2543 goto fail1;
2544 /* test if object file */
2545 if (ehdr.e_type != ET_REL)
2546 goto fail1;
2547 /* test CPU specific stuff */
2548 if (ehdr.e_ident[5] != ELFDATA2LSB ||
2549 ehdr.e_machine != EM_TCC_TARGET) {
2550 fail1:
2551 tcc_error_noabort("invalid object file");
2552 return -1;
2554 /* read sections */
2555 shdr = load_data(fd, file_offset + ehdr.e_shoff,
2556 sizeof(ElfW(Shdr)) * ehdr.e_shnum);
2557 sm_table = tcc_mallocz(sizeof(SectionMergeInfo) * ehdr.e_shnum);
2559 /* load section names */
2560 sh = &shdr[ehdr.e_shstrndx];
2561 strsec = load_data(fd, file_offset + sh->sh_offset, sh->sh_size);
2563 /* load symtab and strtab */
2564 old_to_new_syms = NULL;
2565 symtab = NULL;
2566 strtab = NULL;
2567 nb_syms = 0;
2568 for(i = 1; i < ehdr.e_shnum; i++) {
2569 sh = &shdr[i];
2570 if (sh->sh_type == SHT_SYMTAB) {
2571 if (symtab) {
2572 tcc_error_noabort("object must contain only one symtab");
2573 fail:
2574 ret = -1;
2575 goto the_end;
2577 nb_syms = sh->sh_size / sizeof(ElfW(Sym));
2578 symtab = load_data(fd, file_offset + sh->sh_offset, sh->sh_size);
2579 sm_table[i].s = symtab_section;
2581 /* now load strtab */
2582 sh = &shdr[sh->sh_link];
2583 strtab = load_data(fd, file_offset + sh->sh_offset, sh->sh_size);
2587 /* now examine each section and try to merge its content with the
2588 ones in memory */
2589 for(i = 1; i < ehdr.e_shnum; i++) {
2590 /* no need to examine section name strtab */
2591 if (i == ehdr.e_shstrndx)
2592 continue;
2593 sh = &shdr[i];
2594 sh_name = (char *) strsec + sh->sh_name;
2595 /* ignore sections types we do not handle */
2596 if (sh->sh_type != SHT_PROGBITS &&
2597 sh->sh_type != SHT_RELX &&
2598 #ifdef TCC_ARM_EABI
2599 sh->sh_type != SHT_ARM_EXIDX &&
2600 #endif
2601 sh->sh_type != SHT_NOBITS &&
2602 sh->sh_type != SHT_PREINIT_ARRAY &&
2603 sh->sh_type != SHT_INIT_ARRAY &&
2604 sh->sh_type != SHT_FINI_ARRAY &&
2605 strcmp(sh_name, ".stabstr")
2607 continue;
2608 if (sh->sh_addralign < 1)
2609 sh->sh_addralign = 1;
2610 /* find corresponding section, if any */
2611 for(j = 1; j < s1->nb_sections;j++) {
2612 s = s1->sections[j];
2613 if (!strcmp(s->name, sh_name)) {
2614 if (!strncmp(sh_name, ".gnu.linkonce",
2615 sizeof(".gnu.linkonce") - 1)) {
2616 /* if a 'linkonce' section is already present, we
2617 do not add it again. It is a little tricky as
2618 symbols can still be defined in
2619 it. */
2620 sm_table[i].link_once = 1;
2621 goto next;
2622 } else {
2623 goto found;
2627 /* not found: create new section */
2628 s = new_section(s1, sh_name, sh->sh_type, sh->sh_flags);
2629 /* take as much info as possible from the section. sh_link and
2630 sh_info will be updated later */
2631 s->sh_addralign = sh->sh_addralign;
2632 s->sh_entsize = sh->sh_entsize;
2633 sm_table[i].new_section = 1;
2634 found:
2635 if (sh->sh_type != s->sh_type) {
2636 tcc_error_noabort("invalid section type");
2637 goto fail;
2640 /* align start of section */
2641 offset = s->data_offset;
2643 if (0 == strcmp(sh_name, ".stab")) {
2644 stab_index = i;
2645 goto no_align;
2647 if (0 == strcmp(sh_name, ".stabstr")) {
2648 stabstr_index = i;
2649 goto no_align;
2652 size = sh->sh_addralign - 1;
2653 offset = (offset + size) & ~size;
2654 if (sh->sh_addralign > s->sh_addralign)
2655 s->sh_addralign = sh->sh_addralign;
2656 s->data_offset = offset;
2657 no_align:
2658 sm_table[i].offset = offset;
2659 sm_table[i].s = s;
2660 /* concatenate sections */
2661 size = sh->sh_size;
2662 if (sh->sh_type != SHT_NOBITS) {
2663 unsigned char *ptr;
2664 lseek(fd, file_offset + sh->sh_offset, SEEK_SET);
2665 ptr = section_ptr_add(s, size);
2666 read(fd, ptr, size);
2667 } else {
2668 s->data_offset += size;
2670 next: ;
2673 /* gr relocate stab strings */
2674 if (stab_index && stabstr_index) {
2675 Stab_Sym *a, *b;
2676 unsigned o;
2677 s = sm_table[stab_index].s;
2678 a = (Stab_Sym *)(s->data + sm_table[stab_index].offset);
2679 b = (Stab_Sym *)(s->data + s->data_offset);
2680 o = sm_table[stabstr_index].offset;
2681 while (a < b)
2682 a->n_strx += o, a++;
2685 /* second short pass to update sh_link and sh_info fields of new
2686 sections */
2687 for(i = 1; i < ehdr.e_shnum; i++) {
2688 s = sm_table[i].s;
2689 if (!s || !sm_table[i].new_section)
2690 continue;
2691 sh = &shdr[i];
2692 if (sh->sh_link > 0)
2693 s->link = sm_table[sh->sh_link].s;
2694 if (sh->sh_type == SHT_RELX) {
2695 s->sh_info = sm_table[sh->sh_info].s->sh_num;
2696 /* update backward link */
2697 s1->sections[s->sh_info]->reloc = s;
2700 sm = sm_table;
2702 /* resolve symbols */
2703 old_to_new_syms = tcc_mallocz(nb_syms * sizeof(int));
2705 sym = symtab + 1;
2706 for(i = 1; i < nb_syms; i++, sym++) {
2707 if (sym->st_shndx != SHN_UNDEF &&
2708 sym->st_shndx < SHN_LORESERVE) {
2709 sm = &sm_table[sym->st_shndx];
2710 if (sm->link_once) {
2711 /* if a symbol is in a link once section, we use the
2712 already defined symbol. It is very important to get
2713 correct relocations */
2714 if (ELFW(ST_BIND)(sym->st_info) != STB_LOCAL) {
2715 name = (char *) strtab + sym->st_name;
2716 sym_index = find_elf_sym(symtab_section, name);
2717 if (sym_index)
2718 old_to_new_syms[i] = sym_index;
2720 continue;
2722 /* if no corresponding section added, no need to add symbol */
2723 if (!sm->s)
2724 continue;
2725 /* convert section number */
2726 sym->st_shndx = sm->s->sh_num;
2727 /* offset value */
2728 sym->st_value += sm->offset;
2730 /* add symbol */
2731 name = (char *) strtab + sym->st_name;
2732 sym_index = add_elf_sym(symtab_section, sym->st_value, sym->st_size,
2733 sym->st_info, sym->st_other,
2734 sym->st_shndx, name);
2735 old_to_new_syms[i] = sym_index;
2738 /* third pass to patch relocation entries */
2739 for(i = 1; i < ehdr.e_shnum; i++) {
2740 s = sm_table[i].s;
2741 if (!s)
2742 continue;
2743 sh = &shdr[i];
2744 offset = sm_table[i].offset;
2745 switch(s->sh_type) {
2746 case SHT_RELX:
2747 /* take relocation offset information */
2748 offseti = sm_table[sh->sh_info].offset;
2749 for_each_elem(s, (offset / sizeof(*rel)), rel, ElfW_Rel) {
2750 int type;
2751 unsigned sym_index;
2752 /* convert symbol index */
2753 type = ELFW(R_TYPE)(rel->r_info);
2754 sym_index = ELFW(R_SYM)(rel->r_info);
2755 /* NOTE: only one symtab assumed */
2756 if (sym_index >= nb_syms)
2757 goto invalid_reloc;
2758 sym_index = old_to_new_syms[sym_index];
2759 /* ignore link_once in rel section. */
2760 if (!sym_index && !sm->link_once
2761 #ifdef TCC_TARGET_ARM
2762 && type != R_ARM_V4BX
2763 #endif
2765 invalid_reloc:
2766 tcc_error_noabort("Invalid relocation entry [%2d] '%s' @ %.8x",
2767 i, strsec + sh->sh_name, rel->r_offset);
2768 goto fail;
2770 rel->r_info = ELFW(R_INFO)(sym_index, type);
2771 /* offset the relocation offset */
2772 rel->r_offset += offseti;
2773 #ifdef TCC_TARGET_ARM
2774 /* Jumps and branches from a Thumb code to a PLT entry need
2775 special handling since PLT entries are ARM code.
2776 Unconditional bl instructions referencing PLT entries are
2777 handled by converting these instructions into blx
2778 instructions. Other case of instructions referencing a PLT
2779 entry require to add a Thumb stub before the PLT entry to
2780 switch to ARM mode. We set bit plt_thumb_stub of the
2781 attribute of a symbol to indicate such a case. */
2782 if (type == R_ARM_THM_JUMP24)
2783 alloc_sym_attr(s1, sym_index)->plt_thumb_stub = 1;
2784 #endif
2786 break;
2787 default:
2788 break;
2792 ret = 0;
2793 the_end:
2794 tcc_free(symtab);
2795 tcc_free(strtab);
2796 tcc_free(old_to_new_syms);
2797 tcc_free(sm_table);
2798 tcc_free(strsec);
2799 tcc_free(shdr);
2800 return ret;
2803 typedef struct ArchiveHeader {
2804 char ar_name[16]; /* name of this member */
2805 char ar_date[12]; /* file mtime */
2806 char ar_uid[6]; /* owner uid; printed as decimal */
2807 char ar_gid[6]; /* owner gid; printed as decimal */
2808 char ar_mode[8]; /* file mode, printed as octal */
2809 char ar_size[10]; /* file size, printed as decimal */
2810 char ar_fmag[2]; /* should contain ARFMAG */
2811 } ArchiveHeader;
2813 static int get_be32(const uint8_t *b)
2815 return b[3] | (b[2] << 8) | (b[1] << 16) | (b[0] << 24);
2818 /* load only the objects which resolve undefined symbols */
2819 static int tcc_load_alacarte(TCCState *s1, int fd, int size)
2821 int i, bound, nsyms, sym_index, off, ret;
2822 uint8_t *data;
2823 const char *ar_names, *p;
2824 const uint8_t *ar_index;
2825 ElfW(Sym) *sym;
2827 data = tcc_malloc(size);
2828 if (read(fd, data, size) != size)
2829 goto fail;
2830 nsyms = get_be32(data);
2831 ar_index = data + 4;
2832 ar_names = (char *) ar_index + nsyms * 4;
2834 do {
2835 bound = 0;
2836 for(p = ar_names, i = 0; i < nsyms; i++, p += strlen(p)+1) {
2837 sym_index = find_elf_sym(symtab_section, p);
2838 if(sym_index) {
2839 sym = &((ElfW(Sym) *)symtab_section->data)[sym_index];
2840 if(sym->st_shndx == SHN_UNDEF) {
2841 off = get_be32(ar_index + i * 4) + sizeof(ArchiveHeader);
2842 ++bound;
2843 lseek(fd, off, SEEK_SET);
2844 if(tcc_load_object_file(s1, fd, off) < 0) {
2845 fail:
2846 ret = -1;
2847 goto the_end;
2852 } while(bound);
2853 ret = 0;
2854 the_end:
2855 tcc_free(data);
2856 return ret;
2859 /* load a '.a' file */
2860 ST_FUNC int tcc_load_archive(TCCState *s1, int fd)
2862 ArchiveHeader hdr;
2863 char ar_size[11];
2864 char ar_name[17];
2865 char magic[8];
2866 int size, len, i;
2867 unsigned long file_offset;
2869 /* skip magic which was already checked */
2870 read(fd, magic, sizeof(magic));
2872 for(;;) {
2873 len = read(fd, &hdr, sizeof(hdr));
2874 if (len == 0)
2875 break;
2876 if (len != sizeof(hdr)) {
2877 tcc_error_noabort("invalid archive");
2878 return -1;
2880 memcpy(ar_size, hdr.ar_size, sizeof(hdr.ar_size));
2881 ar_size[sizeof(hdr.ar_size)] = '\0';
2882 size = strtol(ar_size, NULL, 0);
2883 memcpy(ar_name, hdr.ar_name, sizeof(hdr.ar_name));
2884 for(i = sizeof(hdr.ar_name) - 1; i >= 0; i--) {
2885 if (ar_name[i] != ' ')
2886 break;
2888 ar_name[i + 1] = '\0';
2889 file_offset = lseek(fd, 0, SEEK_CUR);
2890 /* align to even */
2891 size = (size + 1) & ~1;
2892 if (!strcmp(ar_name, "/")) {
2893 /* coff symbol table : we handle it */
2894 if(s1->alacarte_link)
2895 return tcc_load_alacarte(s1, fd, size);
2896 } else if (!strcmp(ar_name, "//") ||
2897 !strcmp(ar_name, "__.SYMDEF") ||
2898 !strcmp(ar_name, "__.SYMDEF/") ||
2899 !strcmp(ar_name, "ARFILENAMES/")) {
2900 /* skip symbol table or archive names */
2901 } else {
2902 if (tcc_load_object_file(s1, fd, file_offset) < 0)
2903 return -1;
2905 lseek(fd, file_offset + size, SEEK_SET);
2907 return 0;
2910 #ifndef TCC_TARGET_PE
2911 /* load a DLL and all referenced DLLs. 'level = 0' means that the DLL
2912 is referenced by the user (so it should be added as DT_NEEDED in
2913 the generated ELF file) */
2914 ST_FUNC int tcc_load_dll(TCCState *s1, int fd, const char *filename, int level)
2916 ElfW(Ehdr) ehdr;
2917 ElfW(Shdr) *shdr, *sh, *sh1;
2918 int i, j, nb_syms, nb_dts, sym_bind, ret;
2919 ElfW(Sym) *sym, *dynsym;
2920 ElfW(Dyn) *dt, *dynamic;
2921 unsigned char *dynstr;
2922 const char *name, *soname;
2923 DLLReference *dllref;
2925 read(fd, &ehdr, sizeof(ehdr));
2927 /* test CPU specific stuff */
2928 if (ehdr.e_ident[5] != ELFDATA2LSB ||
2929 ehdr.e_machine != EM_TCC_TARGET) {
2930 tcc_error_noabort("bad architecture");
2931 return -1;
2934 /* read sections */
2935 shdr = load_data(fd, ehdr.e_shoff, sizeof(ElfW(Shdr)) * ehdr.e_shnum);
2937 /* load dynamic section and dynamic symbols */
2938 nb_syms = 0;
2939 nb_dts = 0;
2940 dynamic = NULL;
2941 dynsym = NULL; /* avoid warning */
2942 dynstr = NULL; /* avoid warning */
2943 for(i = 0, sh = shdr; i < ehdr.e_shnum; i++, sh++) {
2944 switch(sh->sh_type) {
2945 case SHT_DYNAMIC:
2946 nb_dts = sh->sh_size / sizeof(ElfW(Dyn));
2947 dynamic = load_data(fd, sh->sh_offset, sh->sh_size);
2948 break;
2949 case SHT_DYNSYM:
2950 nb_syms = sh->sh_size / sizeof(ElfW(Sym));
2951 dynsym = load_data(fd, sh->sh_offset, sh->sh_size);
2952 sh1 = &shdr[sh->sh_link];
2953 dynstr = load_data(fd, sh1->sh_offset, sh1->sh_size);
2954 break;
2955 default:
2956 break;
2960 /* compute the real library name */
2961 soname = tcc_basename(filename);
2963 for(i = 0, dt = dynamic; i < nb_dts; i++, dt++) {
2964 if (dt->d_tag == DT_SONAME) {
2965 soname = (char *) dynstr + dt->d_un.d_val;
2969 /* if the dll is already loaded, do not load it */
2970 for(i = 0; i < s1->nb_loaded_dlls; i++) {
2971 dllref = s1->loaded_dlls[i];
2972 if (!strcmp(soname, dllref->name)) {
2973 /* but update level if needed */
2974 if (level < dllref->level)
2975 dllref->level = level;
2976 ret = 0;
2977 goto the_end;
2981 /* add the dll and its level */
2982 dllref = tcc_mallocz(sizeof(DLLReference) + strlen(soname));
2983 dllref->level = level;
2984 strcpy(dllref->name, soname);
2985 dynarray_add((void ***)&s1->loaded_dlls, &s1->nb_loaded_dlls, dllref);
2987 /* add dynamic symbols in dynsym_section */
2988 for(i = 1, sym = dynsym + 1; i < nb_syms; i++, sym++) {
2989 sym_bind = ELFW(ST_BIND)(sym->st_info);
2990 if (sym_bind == STB_LOCAL)
2991 continue;
2992 name = (char *) dynstr + sym->st_name;
2993 add_elf_sym(s1->dynsymtab_section, sym->st_value, sym->st_size,
2994 sym->st_info, sym->st_other, sym->st_shndx, name);
2997 /* load all referenced DLLs */
2998 for(i = 0, dt = dynamic; i < nb_dts; i++, dt++) {
2999 switch(dt->d_tag) {
3000 case DT_NEEDED:
3001 name = (char *) dynstr + dt->d_un.d_val;
3002 for(j = 0; j < s1->nb_loaded_dlls; j++) {
3003 dllref = s1->loaded_dlls[j];
3004 if (!strcmp(name, dllref->name))
3005 goto already_loaded;
3007 if (tcc_add_dll(s1, name, AFF_REFERENCED_DLL) < 0) {
3008 tcc_error_noabort("referenced dll '%s' not found", name);
3009 ret = -1;
3010 goto the_end;
3012 already_loaded:
3013 break;
3016 ret = 0;
3017 the_end:
3018 tcc_free(dynstr);
3019 tcc_free(dynsym);
3020 tcc_free(dynamic);
3021 tcc_free(shdr);
3022 return ret;
3025 #define LD_TOK_NAME 256
3026 #define LD_TOK_EOF (-1)
3028 /* return next ld script token */
3029 static int ld_next(TCCState *s1, char *name, int name_size)
3031 int c;
3032 char *q;
3034 redo:
3035 switch(ch) {
3036 case ' ':
3037 case '\t':
3038 case '\f':
3039 case '\v':
3040 case '\r':
3041 case '\n':
3042 inp();
3043 goto redo;
3044 case '/':
3045 minp();
3046 if (ch == '*') {
3047 file->buf_ptr = parse_comment(file->buf_ptr);
3048 ch = file->buf_ptr[0];
3049 goto redo;
3050 } else {
3051 q = name;
3052 *q++ = '/';
3053 goto parse_name;
3055 break;
3056 /* case 'a' ... 'z': */
3057 case 'a':
3058 case 'b':
3059 case 'c':
3060 case 'd':
3061 case 'e':
3062 case 'f':
3063 case 'g':
3064 case 'h':
3065 case 'i':
3066 case 'j':
3067 case 'k':
3068 case 'l':
3069 case 'm':
3070 case 'n':
3071 case 'o':
3072 case 'p':
3073 case 'q':
3074 case 'r':
3075 case 's':
3076 case 't':
3077 case 'u':
3078 case 'v':
3079 case 'w':
3080 case 'x':
3081 case 'y':
3082 case 'z':
3083 /* case 'A' ... 'z': */
3084 case 'A':
3085 case 'B':
3086 case 'C':
3087 case 'D':
3088 case 'E':
3089 case 'F':
3090 case 'G':
3091 case 'H':
3092 case 'I':
3093 case 'J':
3094 case 'K':
3095 case 'L':
3096 case 'M':
3097 case 'N':
3098 case 'O':
3099 case 'P':
3100 case 'Q':
3101 case 'R':
3102 case 'S':
3103 case 'T':
3104 case 'U':
3105 case 'V':
3106 case 'W':
3107 case 'X':
3108 case 'Y':
3109 case 'Z':
3110 case '_':
3111 case '\\':
3112 case '.':
3113 case '$':
3114 case '~':
3115 q = name;
3116 parse_name:
3117 for(;;) {
3118 if (!((ch >= 'a' && ch <= 'z') ||
3119 (ch >= 'A' && ch <= 'Z') ||
3120 (ch >= '0' && ch <= '9') ||
3121 strchr("/.-_+=$:\\,~", ch)))
3122 break;
3123 if ((q - name) < name_size - 1) {
3124 *q++ = ch;
3126 minp();
3128 *q = '\0';
3129 c = LD_TOK_NAME;
3130 break;
3131 case CH_EOF:
3132 c = LD_TOK_EOF;
3133 break;
3134 default:
3135 c = ch;
3136 inp();
3137 break;
3139 return c;
3142 static int ld_add_file(TCCState *s1, const char filename[])
3144 int ret;
3146 ret = tcc_add_file_internal(s1, filename, 0);
3147 if (ret)
3148 ret = tcc_add_dll(s1, filename, 0);
3149 return ret;
3152 static inline int new_undef_syms(void)
3154 int ret = 0;
3155 ret = new_undef_sym;
3156 new_undef_sym = 0;
3157 return ret;
3160 static int ld_add_file_list(TCCState *s1, const char *cmd, int as_needed)
3162 char filename[1024], libname[1024];
3163 int t, group, nblibs = 0, ret = 0;
3164 char **libs = NULL;
3166 group = !strcmp(cmd, "GROUP");
3167 if (!as_needed)
3168 new_undef_syms();
3169 t = ld_next(s1, filename, sizeof(filename));
3170 if (t != '(')
3171 expect("(");
3172 t = ld_next(s1, filename, sizeof(filename));
3173 for(;;) {
3174 libname[0] = '\0';
3175 if (t == LD_TOK_EOF) {
3176 tcc_error_noabort("unexpected end of file");
3177 ret = -1;
3178 goto lib_parse_error;
3179 } else if (t == ')') {
3180 break;
3181 } else if (t == '-') {
3182 t = ld_next(s1, filename, sizeof(filename));
3183 if ((t != LD_TOK_NAME) || (filename[0] != 'l')) {
3184 tcc_error_noabort("library name expected");
3185 ret = -1;
3186 goto lib_parse_error;
3188 pstrcpy(libname, sizeof libname, &filename[1]);
3189 if (s1->static_link) {
3190 snprintf(filename, sizeof filename, "lib%s.a", libname);
3191 } else {
3192 snprintf(filename, sizeof filename, "lib%s.so", libname);
3194 } else if (t != LD_TOK_NAME) {
3195 tcc_error_noabort("filename expected");
3196 ret = -1;
3197 goto lib_parse_error;
3199 if (!strcmp(filename, "AS_NEEDED")) {
3200 ret = ld_add_file_list(s1, cmd, 1);
3201 if (ret)
3202 goto lib_parse_error;
3203 } else {
3204 /* TODO: Implement AS_NEEDED support. Ignore it for now */
3205 if (!as_needed) {
3206 ret = ld_add_file(s1, filename);
3207 if (ret)
3208 goto lib_parse_error;
3209 if (group) {
3210 /* Add the filename *and* the libname to avoid future conversions */
3211 dynarray_add((void ***) &libs, &nblibs, tcc_strdup(filename));
3212 if (libname[0] != '\0')
3213 dynarray_add((void ***) &libs, &nblibs, tcc_strdup(libname));
3217 t = ld_next(s1, filename, sizeof(filename));
3218 if (t == ',') {
3219 t = ld_next(s1, filename, sizeof(filename));
3222 if (group && !as_needed) {
3223 while (new_undef_syms()) {
3224 int i;
3226 for (i = 0; i < nblibs; i ++)
3227 ld_add_file(s1, libs[i]);
3230 lib_parse_error:
3231 dynarray_reset(&libs, &nblibs);
3232 return ret;
3235 /* interpret a subset of GNU ldscripts to handle the dummy libc.so
3236 files */
3237 ST_FUNC int tcc_load_ldscript(TCCState *s1)
3239 char cmd[64];
3240 char filename[1024];
3241 int t, ret;
3243 ch = file->buf_ptr[0];
3244 ch = handle_eob();
3245 for(;;) {
3246 t = ld_next(s1, cmd, sizeof(cmd));
3247 if (t == LD_TOK_EOF)
3248 return 0;
3249 else if (t != LD_TOK_NAME)
3250 return -1;
3251 if (!strcmp(cmd, "INPUT") ||
3252 !strcmp(cmd, "GROUP")) {
3253 ret = ld_add_file_list(s1, cmd, 0);
3254 if (ret)
3255 return ret;
3256 } else if (!strcmp(cmd, "OUTPUT_FORMAT") ||
3257 !strcmp(cmd, "TARGET")) {
3258 /* ignore some commands */
3259 t = ld_next(s1, cmd, sizeof(cmd));
3260 if (t != '(')
3261 expect("(");
3262 for(;;) {
3263 t = ld_next(s1, filename, sizeof(filename));
3264 if (t == LD_TOK_EOF) {
3265 tcc_error_noabort("unexpected end of file");
3266 return -1;
3267 } else if (t == ')') {
3268 break;
3271 } else {
3272 return -1;
3275 return 0;
3277 #endif /* !TCC_TARGET_PE */