x86_64/elf: only variadic calls need rax
[tinycc.git] / tccelf.c
blob2b50ff5262793d2a6fc0333951f18999564a15d1
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... may be -fcommon is needed?", name);
253 } else {
254 do_patch:
255 esym->st_info = ELFW(ST_INFO)(sym_bind, sym_type);
256 esym->st_shndx = sh_num;
257 new_undef_sym = 1;
258 esym->st_value = value;
259 esym->st_size = size;
260 esym->st_other = other;
262 } else {
263 do_def:
264 sym_index = put_elf_sym(s, value, size,
265 ELFW(ST_INFO)(sym_bind, sym_type), other,
266 sh_num, name);
268 return sym_index;
271 /* put relocation */
272 ST_FUNC void put_elf_reloca(Section *symtab, Section *s, unsigned long offset,
273 int type, int symbol, addr_t addend)
275 char buf[256];
276 Section *sr;
277 ElfW_Rel *rel;
279 sr = s->reloc;
280 if (!sr) {
281 /* if no relocation section, create it */
282 snprintf(buf, sizeof(buf), REL_SECTION_FMT, s->name);
283 /* if the symtab is allocated, then we consider the relocation
284 are also */
285 sr = new_section(tcc_state, buf, SHT_RELX, symtab->sh_flags);
286 sr->sh_entsize = sizeof(ElfW_Rel);
287 sr->link = symtab;
288 sr->sh_info = s->sh_num;
289 s->reloc = sr;
291 rel = section_ptr_add(sr, sizeof(ElfW_Rel));
292 rel->r_offset = offset;
293 rel->r_info = ELFW(R_INFO)(symbol, type);
294 #if defined(TCC_TARGET_ARM64) || defined(TCC_TARGET_X86_64)
295 rel->r_addend = addend;
296 #else
297 if (addend)
298 tcc_error("non-zero addend on REL architecture");
299 #endif
302 ST_FUNC void put_elf_reloc(Section *symtab, Section *s, unsigned long offset,
303 int type, int symbol)
305 put_elf_reloca(symtab, s, offset, type, symbol, 0);
308 /* put stab debug information */
310 ST_FUNC void put_stabs(const char *str, int type, int other, int desc,
311 unsigned long value)
313 Stab_Sym *sym;
315 sym = section_ptr_add(stab_section, sizeof(Stab_Sym));
316 if (str) {
317 sym->n_strx = put_elf_str(stabstr_section, str);
318 } else {
319 sym->n_strx = 0;
321 sym->n_type = type;
322 sym->n_other = other;
323 sym->n_desc = desc;
324 sym->n_value = value;
327 ST_FUNC void put_stabs_r(const char *str, int type, int other, int desc,
328 unsigned long value, Section *sec, int sym_index)
330 put_stabs(str, type, other, desc, value);
331 put_elf_reloc(symtab_section, stab_section,
332 stab_section->data_offset - sizeof(unsigned int),
333 R_DATA_32, sym_index);
336 ST_FUNC void put_stabn(int type, int other, int desc, int value)
338 put_stabs(NULL, type, other, desc, value);
341 ST_FUNC void put_stabd(int type, int other, int desc)
343 put_stabs(NULL, type, other, desc, 0);
346 /* Browse each elem of type <type> in section <sec> starting at elem <startoff>
347 using variable <elem> */
348 #define for_each_elem(sec, startoff, elem, type) \
349 for (elem = (type *) sec->data + startoff; \
350 elem < (type *) (sec->data + sec->data_offset); elem++)
352 /* In an ELF file symbol table, the local symbols must appear below
353 the global and weak ones. Since TCC cannot sort it while generating
354 the code, we must do it after. All the relocation tables are also
355 modified to take into account the symbol table sorting */
356 static void sort_syms(TCCState *s1, Section *s)
358 int *old_to_new_syms;
359 ElfW(Sym) *new_syms;
360 int nb_syms, i;
361 ElfW(Sym) *p, *q;
362 ElfW_Rel *rel;
363 Section *sr;
364 int type, sym_index;
366 nb_syms = s->data_offset / sizeof(ElfW(Sym));
367 new_syms = tcc_malloc(nb_syms * sizeof(ElfW(Sym)));
368 old_to_new_syms = tcc_malloc(nb_syms * sizeof(int));
370 /* first pass for local symbols */
371 p = (ElfW(Sym) *)s->data;
372 q = new_syms;
373 for(i = 0; i < nb_syms; i++) {
374 if (ELFW(ST_BIND)(p->st_info) == STB_LOCAL) {
375 old_to_new_syms[i] = q - new_syms;
376 *q++ = *p;
378 p++;
380 /* save the number of local symbols in section header */
381 s->sh_info = q - new_syms;
383 /* then second pass for non local symbols */
384 p = (ElfW(Sym) *)s->data;
385 for(i = 0; i < nb_syms; i++) {
386 if (ELFW(ST_BIND)(p->st_info) != STB_LOCAL) {
387 old_to_new_syms[i] = q - new_syms;
388 *q++ = *p;
390 p++;
393 /* we copy the new symbols to the old */
394 memcpy(s->data, new_syms, nb_syms * sizeof(ElfW(Sym)));
395 tcc_free(new_syms);
397 /* now we modify all the relocations */
398 for(i = 1; i < s1->nb_sections; i++) {
399 sr = s1->sections[i];
400 if (sr->sh_type == SHT_RELX && sr->link == s) {
401 for_each_elem(sr, 0, rel, ElfW_Rel) {
402 sym_index = ELFW(R_SYM)(rel->r_info);
403 type = ELFW(R_TYPE)(rel->r_info);
404 sym_index = old_to_new_syms[sym_index];
405 rel->r_info = ELFW(R_INFO)(sym_index, type);
410 tcc_free(old_to_new_syms);
413 /* relocate common symbols in the .bss section */
414 ST_FUNC void relocate_common_syms(void)
416 ElfW(Sym) *sym;
417 unsigned long offset, align;
419 for_each_elem(symtab_section, 1, sym, ElfW(Sym)) {
420 if (sym->st_shndx == SHN_COMMON) {
421 /* align symbol */
422 align = sym->st_value;
423 offset = bss_section->data_offset;
424 offset = (offset + align - 1) & -align;
425 sym->st_value = offset;
426 sym->st_shndx = bss_section->sh_num;
427 offset += sym->st_size;
428 bss_section->data_offset = offset;
433 /* relocate symbol table, resolve undefined symbols if do_resolve is
434 true and output error if undefined symbol. */
435 ST_FUNC void relocate_syms(TCCState *s1, int do_resolve)
437 ElfW(Sym) *sym, *esym;
438 int sym_bind, sh_num, sym_index;
439 const char *name;
441 for_each_elem(symtab_section, 1, sym, ElfW(Sym)) {
442 sh_num = sym->st_shndx;
443 if (sh_num == SHN_UNDEF) {
444 name = (char *) strtab_section->data + sym->st_name;
445 /* Use ld.so to resolve symbol for us (for tcc -run) */
446 if (do_resolve) {
447 #if defined TCC_IS_NATIVE && !defined _WIN32
448 void *addr;
449 name = (char *) symtab_section->link->data + sym->st_name;
450 addr = resolve_sym(s1, name);
451 if (addr) {
452 sym->st_value = (addr_t)addr;
453 #ifdef DEBUG_RELOC
454 printf ("relocate_sym: %s -> 0x%lx\n", name, sym->st_value);
455 #endif
456 goto found;
458 #endif
459 } else if (s1->dynsym) {
460 /* if dynamic symbol exist, then use it */
461 sym_index = find_elf_sym(s1->dynsym, name);
462 if (sym_index) {
463 esym = &((ElfW(Sym) *)s1->dynsym->data)[sym_index];
464 sym->st_value = esym->st_value;
465 goto found;
468 /* XXX: _fp_hw seems to be part of the ABI, so we ignore
469 it */
470 if (!strcmp(name, "_fp_hw"))
471 goto found;
472 /* only weak symbols are accepted to be undefined. Their
473 value is zero */
474 sym_bind = ELFW(ST_BIND)(sym->st_info);
475 if (sym_bind == STB_WEAK) {
476 sym->st_value = 0;
477 } else {
478 tcc_error_noabort("undefined symbol '%s'", name);
480 } else if (sh_num < SHN_LORESERVE) {
481 /* add section base */
482 sym->st_value += s1->sections[sym->st_shndx]->sh_addr;
484 found: ;
488 /* relocate a given section (CPU dependent) by applying the relocations
489 in the associated relocation section */
490 ST_FUNC void relocate_section(TCCState *s1, Section *s)
492 Section *sr = s->reloc;
493 ElfW_Rel *rel;
494 ElfW(Sym) *sym;
495 int type, sym_index;
496 unsigned char *ptr;
497 addr_t val, addr;
498 #if defined TCC_TARGET_I386 || defined TCC_TARGET_X86_64
499 ElfW_Rel *qrel = (ElfW_Rel *) sr->data; /* ptr to next reloc entry reused */
500 int esym_index;
501 #endif
503 for_each_elem(sr, 0, rel, ElfW_Rel) {
504 ptr = s->data + rel->r_offset;
506 sym_index = ELFW(R_SYM)(rel->r_info);
507 sym = &((ElfW(Sym) *)symtab_section->data)[sym_index];
508 val = sym->st_value;
509 #if defined(TCC_TARGET_ARM64) || defined(TCC_TARGET_X86_64)
510 val += rel->r_addend;
511 #endif
512 type = ELFW(R_TYPE)(rel->r_info);
513 addr = s->sh_addr + rel->r_offset;
515 /* CPU specific */
516 switch(type) {
517 #if defined(TCC_TARGET_I386)
518 case R_386_32:
519 if (s1->output_type == TCC_OUTPUT_DLL) {
520 esym_index = s1->symtab_to_dynsym[sym_index];
521 qrel->r_offset = rel->r_offset;
522 if (esym_index) {
523 qrel->r_info = ELFW(R_INFO)(esym_index, R_386_32);
524 qrel++;
525 break;
526 } else {
527 qrel->r_info = ELFW(R_INFO)(0, R_386_RELATIVE);
528 qrel++;
531 write32le(ptr, read32le(ptr) + val);
532 break;
533 case R_386_PC32:
534 if (s1->output_type == TCC_OUTPUT_DLL) {
535 /* DLL relocation */
536 esym_index = s1->symtab_to_dynsym[sym_index];
537 if (esym_index) {
538 qrel->r_offset = rel->r_offset;
539 qrel->r_info = ELFW(R_INFO)(esym_index, R_386_PC32);
540 qrel++;
541 break;
544 write32le(ptr, read32le(ptr) + val - addr);
545 break;
546 case R_386_PLT32:
547 write32le(ptr, read32le(ptr) + val - addr);
548 break;
549 case R_386_GLOB_DAT:
550 case R_386_JMP_SLOT:
551 write32le(ptr, val);
552 break;
553 case R_386_GOTPC:
554 write32le(ptr, read32le(ptr) + s1->got->sh_addr - addr);
555 break;
556 case R_386_GOTOFF:
557 write32le(ptr, read32le(ptr) + val - s1->got->sh_addr);
558 break;
559 case R_386_GOT32:
560 case R_386_GOT32X:
561 /* we load the got offset */
562 write32le(ptr, read32le(ptr) + s1->sym_attrs[sym_index].got_offset);
563 break;
564 case R_386_16:
565 if (s1->output_format != TCC_OUTPUT_FORMAT_BINARY) {
566 output_file:
567 tcc_error("can only produce 16-bit binary files");
569 write16le(ptr, read16le(ptr) + val);
570 break;
571 case R_386_PC16:
572 if (s1->output_format != TCC_OUTPUT_FORMAT_BINARY)
573 goto output_file;
574 write16le(ptr, read16le(ptr) + val - addr);
575 break;
576 case R_386_RELATIVE:
577 /* do nothing */
578 break;
579 case R_386_COPY:
580 /* This reloction must copy initialized data from the library
581 to the program .bss segment. Currently made like for ARM
582 (to remove noise of defaukt case). Is this true?
584 break;
585 default:
586 fprintf(stderr,"FIXME: handle reloc type %d at %x [%p] to %x\n",
587 type, (unsigned)addr, ptr, (unsigned)val);
588 break;
589 #elif defined(TCC_TARGET_ARM)
590 case R_ARM_PC24:
591 case R_ARM_CALL:
592 case R_ARM_JUMP24:
593 case R_ARM_PLT32:
595 int x, is_thumb, is_call, h, blx_avail, is_bl, th_ko;
596 x = (*(int *) ptr) & 0xffffff;
597 if (sym->st_shndx == SHN_UNDEF)
598 val = s1->plt->sh_addr;
599 #ifdef DEBUG_RELOC
600 printf ("reloc %d: x=0x%x val=0x%x ", type, x, val);
601 #endif
602 (*(int *)ptr) &= 0xff000000;
603 if (x & 0x800000)
604 x -= 0x1000000;
605 x <<= 2;
606 blx_avail = (TCC_ARM_VERSION >= 5);
607 is_thumb = val & 1;
608 is_bl = (*(unsigned *) ptr) >> 24 == 0xeb;
609 is_call = (type == R_ARM_CALL || (type == R_ARM_PC24 && is_bl));
610 x += val - addr;
611 #ifdef DEBUG_RELOC
612 printf (" newx=0x%x name=%s\n", x,
613 (char *) symtab_section->link->data + sym->st_name);
614 #endif
615 h = x & 2;
616 th_ko = (x & 3) && (!blx_avail || !is_call);
617 if (th_ko || x >= 0x2000000 || x < -0x2000000)
618 tcc_error("can't relocate value at %x,%d",addr, type);
619 x >>= 2;
620 x &= 0xffffff;
621 /* Only reached if blx is avail and it is a call */
622 if (is_thumb) {
623 x |= h << 24;
624 (*(int *)ptr) = 0xfa << 24; /* bl -> blx */
626 (*(int *) ptr) |= x;
628 break;
629 /* Since these relocations only concern Thumb-2 and blx instruction was
630 introduced before Thumb-2, we can assume blx is available and not
631 guard its use */
632 case R_ARM_THM_PC22:
633 case R_ARM_THM_JUMP24:
635 int x, hi, lo, s, j1, j2, i1, i2, imm10, imm11;
636 int to_thumb, is_call, to_plt, blx_bit = 1 << 12;
637 Section *plt;
639 /* weak reference */
640 if (sym->st_shndx == SHN_UNDEF &&
641 ELFW(ST_BIND)(sym->st_info) == STB_WEAK)
642 break;
644 /* Get initial offset */
645 hi = (*(uint16_t *)ptr);
646 lo = (*(uint16_t *)(ptr+2));
647 s = (hi >> 10) & 1;
648 j1 = (lo >> 13) & 1;
649 j2 = (lo >> 11) & 1;
650 i1 = (j1 ^ s) ^ 1;
651 i2 = (j2 ^ s) ^ 1;
652 imm10 = hi & 0x3ff;
653 imm11 = lo & 0x7ff;
654 x = (s << 24) | (i1 << 23) | (i2 << 22) |
655 (imm10 << 12) | (imm11 << 1);
656 if (x & 0x01000000)
657 x -= 0x02000000;
659 /* Relocation infos */
660 to_thumb = val & 1;
661 plt = s1->plt;
662 to_plt = (val >= plt->sh_addr) &&
663 (val < plt->sh_addr + plt->data_offset);
664 is_call = (type == R_ARM_THM_PC22);
666 /* Compute final offset */
667 if (to_plt && !is_call) /* Point to 1st instr of Thumb stub */
668 x -= 4;
669 x += val - addr;
670 if (!to_thumb && is_call) {
671 blx_bit = 0; /* bl -> blx */
672 x = (x + 3) & -4; /* Compute offset from aligned PC */
675 /* Check that relocation is possible
676 * offset must not be out of range
677 * if target is to be entered in arm mode:
678 - bit 1 must not set
679 - instruction must be a call (bl) or a jump to PLT */
680 if (!to_thumb || x >= 0x1000000 || x < -0x1000000)
681 if (to_thumb || (val & 2) || (!is_call && !to_plt))
682 tcc_error("can't relocate value at %x,%d",addr, type);
684 /* Compute and store final offset */
685 s = (x >> 24) & 1;
686 i1 = (x >> 23) & 1;
687 i2 = (x >> 22) & 1;
688 j1 = s ^ (i1 ^ 1);
689 j2 = s ^ (i2 ^ 1);
690 imm10 = (x >> 12) & 0x3ff;
691 imm11 = (x >> 1) & 0x7ff;
692 (*(uint16_t *)ptr) = (uint16_t) ((hi & 0xf800) |
693 (s << 10) | imm10);
694 (*(uint16_t *)(ptr+2)) = (uint16_t) ((lo & 0xc000) |
695 (j1 << 13) | blx_bit | (j2 << 11) |
696 imm11);
698 break;
699 case R_ARM_MOVT_ABS:
700 case R_ARM_MOVW_ABS_NC:
702 int x, imm4, imm12;
703 if (type == R_ARM_MOVT_ABS)
704 val >>= 16;
705 imm12 = val & 0xfff;
706 imm4 = (val >> 12) & 0xf;
707 x = (imm4 << 16) | imm12;
708 if (type == R_ARM_THM_MOVT_ABS)
709 *(int *)ptr |= x;
710 else
711 *(int *)ptr += x;
713 break;
714 case R_ARM_THM_MOVT_ABS:
715 case R_ARM_THM_MOVW_ABS_NC:
717 int x, i, imm4, imm3, imm8;
718 if (type == R_ARM_THM_MOVT_ABS)
719 val >>= 16;
720 imm8 = val & 0xff;
721 imm3 = (val >> 8) & 0x7;
722 i = (val >> 11) & 1;
723 imm4 = (val >> 12) & 0xf;
724 x = (imm3 << 28) | (imm8 << 16) | (i << 10) | imm4;
725 if (type == R_ARM_THM_MOVT_ABS)
726 *(int *)ptr |= x;
727 else
728 *(int *)ptr += x;
730 break;
731 case R_ARM_PREL31:
733 int x;
734 x = (*(int *)ptr) & 0x7fffffff;
735 (*(int *)ptr) &= 0x80000000;
736 x = (x * 2) / 2;
737 x += val - addr;
738 if((x^(x>>1))&0x40000000)
739 tcc_error("can't relocate value at %x,%d",addr, type);
740 (*(int *)ptr) |= x & 0x7fffffff;
742 case R_ARM_ABS32:
743 *(int *)ptr += val;
744 break;
745 case R_ARM_REL32:
746 *(int *)ptr += val - addr;
747 break;
748 case R_ARM_GOTPC:
749 *(int *)ptr += s1->got->sh_addr - addr;
750 break;
751 case R_ARM_GOTOFF:
752 *(int *)ptr += val - s1->got->sh_addr;
753 break;
754 case R_ARM_GOT32:
755 /* we load the got offset */
756 *(int *)ptr += s1->sym_attrs[sym_index].got_offset;
757 break;
758 case R_ARM_COPY:
759 break;
760 case R_ARM_V4BX:
761 /* trade Thumb support for ARMv4 support */
762 if ((0x0ffffff0 & *(int*)ptr) == 0x012FFF10)
763 *(int*)ptr ^= 0xE12FFF10 ^ 0xE1A0F000; /* BX Rm -> MOV PC, Rm */
764 break;
765 case R_ARM_GLOB_DAT:
766 case R_ARM_JUMP_SLOT:
767 *(addr_t *)ptr = val;
768 break;
769 case R_ARM_NONE:
770 /* Nothing to do. Normally used to indicate a dependency
771 on a certain symbol (like for exception handling under EABI). */
772 break;
773 default:
774 fprintf(stderr,"FIXME: handle reloc type %x at %x [%p] to %x\n",
775 type, (unsigned)addr, ptr, (unsigned)val);
776 break;
777 #elif defined(TCC_TARGET_ARM64)
778 case R_AARCH64_ABS64:
779 write64le(ptr, val);
780 break;
781 case R_AARCH64_ABS32:
782 write32le(ptr, val);
783 break;
784 case R_AARCH64_MOVW_UABS_G0_NC:
785 write32le(ptr, ((read32le(ptr) & 0xffe0001f) |
786 (val & 0xffff) << 5));
787 break;
788 case R_AARCH64_MOVW_UABS_G1_NC:
789 write32le(ptr, ((read32le(ptr) & 0xffe0001f) |
790 (val >> 16 & 0xffff) << 5));
791 break;
792 case R_AARCH64_MOVW_UABS_G2_NC:
793 write32le(ptr, ((read32le(ptr) & 0xffe0001f) |
794 (val >> 32 & 0xffff) << 5));
795 break;
796 case R_AARCH64_MOVW_UABS_G3:
797 write32le(ptr, ((read32le(ptr) & 0xffe0001f) |
798 (val >> 48 & 0xffff) << 5));
799 break;
800 case R_AARCH64_ADR_PREL_PG_HI21: {
801 uint64_t off = (val >> 12) - (addr >> 12);
802 if ((off + ((uint64_t)1 << 20)) >> 21)
803 tcc_error("R_AARCH64_ADR_PREL_PG_HI21 relocation failed");
804 write32le(ptr, ((read32le(ptr) & 0x9f00001f) |
805 (off & 0x1ffffc) << 3 | (off & 3) << 29));
806 break;
808 case R_AARCH64_ADD_ABS_LO12_NC:
809 write32le(ptr, ((read32le(ptr) & 0xffc003ff) |
810 (val & 0xfff) << 10));
811 break;
812 case R_AARCH64_JUMP26:
813 case R_AARCH64_CALL26:
814 /* This check must match the one in build_got_entries, testing
815 if we really need a PLT slot. */
816 if (sym->st_shndx == SHN_UNDEF)
817 /* We've put the PLT slot offset into r_addend when generating
818 it, and that's what we must use as relocation value (adjusted
819 by section offset of course). */
820 val = s1->plt->sh_addr + rel->r_addend;
821 #ifdef DEBUG_RELOC
822 printf ("reloc %d @ 0x%lx: val=0x%lx name=%s\n", type, addr, val,
823 (char *) symtab_section->link->data + sym->st_name);
824 #endif
825 if (((val - addr) + ((uint64_t)1 << 27)) & ~(uint64_t)0xffffffc)
827 tcc_error("R_AARCH64_(JUMP|CALL)26 relocation failed (val=%lx, addr=%lx)", addr, val);
829 write32le(ptr, (0x14000000 |
830 (uint32_t)(type == R_AARCH64_CALL26) << 31 |
831 ((val - addr) >> 2 & 0x3ffffff)));
832 break;
833 case R_AARCH64_ADR_GOT_PAGE: {
834 uint64_t off =
835 (((s1->got->sh_addr +
836 s1->sym_attrs[sym_index].got_offset) >> 12) - (addr >> 12));
837 if ((off + ((uint64_t)1 << 20)) >> 21)
838 tcc_error("R_AARCH64_ADR_GOT_PAGE relocation failed");
839 write32le(ptr, ((read32le(ptr) & 0x9f00001f) |
840 (off & 0x1ffffc) << 3 | (off & 3) << 29));
841 break;
843 case R_AARCH64_LD64_GOT_LO12_NC:
844 write32le(ptr,
845 ((read32le(ptr) & 0xfff803ff) |
846 ((s1->got->sh_addr +
847 s1->sym_attrs[sym_index].got_offset) & 0xff8) << 7));
848 break;
849 case R_AARCH64_COPY:
850 break;
851 case R_AARCH64_GLOB_DAT:
852 case R_AARCH64_JUMP_SLOT:
853 /* They don't need addend */
854 #ifdef DEBUG_RELOC
855 printf ("reloc %d @ 0x%lx: val=0x%lx name=%s\n", type, addr,
856 val - rel->r_addend,
857 (char *) symtab_section->link->data + sym->st_name);
858 #endif
859 write64le(ptr, val - rel->r_addend);
860 break;
861 default:
862 fprintf(stderr, "FIXME: handle reloc type %x at %x [%p] to %x\n",
863 type, (unsigned)addr, ptr, (unsigned)val);
864 break;
865 #elif defined(TCC_TARGET_C67)
866 case R_C60_32:
867 *(int *)ptr += val;
868 break;
869 case R_C60LO16:
871 uint32_t orig;
873 /* put the low 16 bits of the absolute address
874 add to what is already there */
876 orig = ((*(int *)(ptr )) >> 7) & 0xffff;
877 orig |= (((*(int *)(ptr+4)) >> 7) & 0xffff) << 16;
879 /* patch both at once - assumes always in pairs Low - High */
881 *(int *) ptr = (*(int *) ptr & (~(0xffff << 7)) ) | (((val+orig) & 0xffff) << 7);
882 *(int *)(ptr+4) = (*(int *)(ptr+4) & (~(0xffff << 7)) ) | ((((val+orig)>>16) & 0xffff) << 7);
884 break;
885 case R_C60HI16:
886 break;
887 default:
888 fprintf(stderr,"FIXME: handle reloc type %x at %x [%p] to %x\n",
889 type, (unsigned)addr, ptr, (unsigned)val);
890 break;
891 #elif defined(TCC_TARGET_X86_64)
892 case R_X86_64_64:
893 if (s1->output_type == TCC_OUTPUT_DLL) {
894 esym_index = s1->symtab_to_dynsym[sym_index];
895 qrel->r_offset = rel->r_offset;
896 if (esym_index) {
897 qrel->r_info = ELFW(R_INFO)(esym_index, R_X86_64_64);
898 qrel->r_addend = rel->r_addend;
899 qrel++;
900 break;
901 } else {
902 qrel->r_info = ELFW(R_INFO)(0, R_X86_64_RELATIVE);
903 qrel->r_addend = read64le(ptr) + val;
904 qrel++;
907 write64le(ptr, read64le(ptr) + val);
908 break;
909 case R_X86_64_32:
910 case R_X86_64_32S:
911 if (s1->output_type == TCC_OUTPUT_DLL) {
912 /* XXX: this logic may depend on TCC's codegen
913 now TCC uses R_X86_64_32 even for a 64bit pointer */
914 qrel->r_info = ELFW(R_INFO)(0, R_X86_64_RELATIVE);
915 /* Use sign extension! */
916 qrel->r_addend = (int)read32le(ptr) + val;
917 qrel++;
919 write32le(ptr, read32le(ptr) + val);
920 break;
922 case R_X86_64_PC32:
923 if (s1->output_type == TCC_OUTPUT_DLL) {
924 /* DLL relocation */
925 esym_index = s1->symtab_to_dynsym[sym_index];
926 if (esym_index) {
927 qrel->r_offset = rel->r_offset;
928 qrel->r_info = ELFW(R_INFO)(esym_index, R_X86_64_PC32);
929 /* Use sign extension! */
930 qrel->r_addend = (int)read32le(ptr) + rel->r_addend;
931 qrel++;
932 break;
935 goto plt32pc32;
937 case R_X86_64_PLT32:
938 /* We've put the PLT slot offset into r_addend when generating
939 it, and that's what we must use as relocation value (adjusted
940 by section offset of course). */
941 val = s1->plt->sh_addr + rel->r_addend;
942 /* fallthrough. */
944 plt32pc32:
946 long long diff;
947 diff = (long long)val - addr;
948 if (diff < -2147483648LL || diff > 2147483647LL) {
949 tcc_error("internal error: relocation failed");
951 write32le(ptr, read32le(ptr) + diff);
953 break;
954 case R_X86_64_GLOB_DAT:
955 case R_X86_64_JUMP_SLOT:
956 /* They don't need addend */
957 write64le(ptr, val - rel->r_addend);
958 break;
959 case R_X86_64_GOTPCREL:
960 case R_X86_64_GOTPCRELX:
961 case R_X86_64_REX_GOTPCRELX:
962 write32le(ptr, read32le(ptr) +
963 (s1->got->sh_addr - addr +
964 s1->sym_attrs[sym_index].got_offset - 4));
965 break;
966 case R_X86_64_GOTTPOFF:
967 write32le(ptr, read32le(ptr) + val - s1->got->sh_addr);
968 break;
969 case R_X86_64_GOT32:
970 /* we load the got offset */
971 write32le(ptr, read32le(ptr) + s1->sym_attrs[sym_index].got_offset);
972 break;
973 #else
974 #error unsupported processor
975 #endif
978 /* if the relocation is allocated, we change its symbol table */
979 if (sr->sh_flags & SHF_ALLOC)
980 sr->link = s1->dynsym;
983 /* relocate relocation table in 'sr' */
984 static void relocate_rel(TCCState *s1, Section *sr)
986 Section *s;
987 ElfW_Rel *rel;
989 s = s1->sections[sr->sh_info];
990 for_each_elem(sr, 0, rel, ElfW_Rel)
991 rel->r_offset += s->sh_addr;
994 /* count the number of dynamic relocations so that we can reserve
995 their space */
996 static int prepare_dynamic_rel(TCCState *s1, Section *sr)
998 ElfW_Rel *rel;
999 int sym_index, esym_index, type, count;
1001 count = 0;
1002 for_each_elem(sr, 0, rel, ElfW_Rel) {
1003 sym_index = ELFW(R_SYM)(rel->r_info);
1004 type = ELFW(R_TYPE)(rel->r_info);
1005 switch(type) {
1006 #if defined(TCC_TARGET_I386)
1007 case R_386_32:
1008 #elif defined(TCC_TARGET_X86_64)
1009 case R_X86_64_32:
1010 case R_X86_64_32S:
1011 case R_X86_64_64:
1012 #endif
1013 count++;
1014 break;
1015 #if defined(TCC_TARGET_I386)
1016 case R_386_PC32:
1017 #elif defined(TCC_TARGET_X86_64)
1018 case R_X86_64_PC32:
1019 #endif
1020 esym_index = s1->symtab_to_dynsym[sym_index];
1021 if (esym_index)
1022 count++;
1023 break;
1024 default:
1025 break;
1028 if (count) {
1029 /* allocate the section */
1030 sr->sh_flags |= SHF_ALLOC;
1031 sr->sh_size = count * sizeof(ElfW_Rel);
1033 return count;
1036 static struct sym_attr *alloc_sym_attr(TCCState *s1, int index)
1038 int n;
1039 struct sym_attr *tab;
1041 if (index >= s1->nb_sym_attrs) {
1042 /* find immediately bigger power of 2 and reallocate array */
1043 n = 1;
1044 while (index >= n)
1045 n *= 2;
1046 tab = tcc_realloc(s1->sym_attrs, n * sizeof(*s1->sym_attrs));
1047 s1->sym_attrs = tab;
1048 memset(s1->sym_attrs + s1->nb_sym_attrs, 0,
1049 (n - s1->nb_sym_attrs) * sizeof(*s1->sym_attrs));
1050 s1->nb_sym_attrs = n;
1052 return &s1->sym_attrs[index];
1055 static void build_got(TCCState *s1)
1057 unsigned char *ptr;
1059 /* if no got, then create it */
1060 s1->got = new_section(s1, ".got", SHT_PROGBITS, SHF_ALLOC | SHF_WRITE);
1061 s1->got->sh_entsize = 4;
1062 add_elf_sym(symtab_section, 0, 4, ELFW(ST_INFO)(STB_GLOBAL, STT_OBJECT),
1063 0, s1->got->sh_num, "_GLOBAL_OFFSET_TABLE_");
1064 ptr = section_ptr_add(s1->got, 3 * PTR_SIZE);
1065 #if PTR_SIZE == 4
1066 /* keep space for _DYNAMIC pointer, if present */
1067 write32le(ptr, 0);
1068 /* two dummy got entries */
1069 write32le(ptr + 4, 0);
1070 write32le(ptr + 8, 0);
1071 #else
1072 /* keep space for _DYNAMIC pointer, if present */
1073 write32le(ptr, 0);
1074 write32le(ptr + 4, 0);
1075 /* two dummy got entries */
1076 write32le(ptr + 8, 0);
1077 write32le(ptr + 12, 0);
1078 write32le(ptr + 16, 0);
1079 write32le(ptr + 20, 0);
1080 #endif
1083 /* put a got or plt entry corresponding to a symbol in symtab_section. 'size'
1084 and 'info' can be modifed if more precise info comes from the DLL.
1085 Returns offset of GOT or PLT slot. */
1086 static unsigned long put_got_entry(TCCState *s1,
1087 int reloc_type, unsigned long size, int info,
1088 int sym_index)
1090 int index, need_plt_entry;
1091 const char *name;
1092 ElfW(Sym) *sym;
1093 unsigned long offset;
1094 int *ptr;
1095 struct sym_attr *symattr;
1097 if (!s1->got)
1098 build_got(s1);
1100 need_plt_entry =
1101 #ifdef TCC_TARGET_X86_64
1102 (reloc_type == R_X86_64_JUMP_SLOT);
1103 #elif defined(TCC_TARGET_I386)
1104 (reloc_type == R_386_JMP_SLOT);
1105 #elif defined(TCC_TARGET_ARM)
1106 (reloc_type == R_ARM_JUMP_SLOT);
1107 #elif defined(TCC_TARGET_ARM64)
1108 (reloc_type == R_AARCH64_JUMP_SLOT);
1109 #else
1111 #endif
1113 if (need_plt_entry && !s1->plt) {
1114 /* add PLT */
1115 s1->plt = new_section(s1, ".plt", SHT_PROGBITS,
1116 SHF_ALLOC | SHF_EXECINSTR);
1117 s1->plt->sh_entsize = 4;
1120 /* If a got/plt entry already exists for that symbol, no need to add one */
1121 if (sym_index < s1->nb_sym_attrs) {
1122 if (need_plt_entry && s1->sym_attrs[sym_index].plt_offset)
1123 return s1->sym_attrs[sym_index].plt_offset;
1124 else if (!need_plt_entry && s1->sym_attrs[sym_index].got_offset)
1125 return s1->sym_attrs[sym_index].got_offset;
1128 symattr = alloc_sym_attr(s1, sym_index);
1130 /* Only store the GOT offset if it's not generated for the PLT entry. */
1131 if (!need_plt_entry)
1132 symattr->got_offset = s1->got->data_offset;
1134 sym = &((ElfW(Sym) *)symtab_section->data)[sym_index];
1135 name = (char *) symtab_section->link->data + sym->st_name;
1136 offset = sym->st_value;
1137 #if defined(TCC_TARGET_I386) || defined(TCC_TARGET_X86_64)
1138 if (need_plt_entry) {
1139 Section *plt;
1140 uint8_t *p;
1141 int modrm;
1142 unsigned long relofs;
1144 #if defined(TCC_OUTPUT_DLL_WITH_PLT)
1145 modrm = 0x25;
1146 #else
1147 /* if we build a DLL, we add a %ebx offset */
1148 if (s1->output_type == TCC_OUTPUT_DLL)
1149 modrm = 0xa3;
1150 else
1151 modrm = 0x25;
1152 #endif
1154 /* add a PLT entry */
1155 plt = s1->plt;
1156 if (plt->data_offset == 0) {
1157 /* first plt entry */
1158 p = section_ptr_add(plt, 16);
1159 p[0] = 0xff; /* pushl got + PTR_SIZE */
1160 p[1] = modrm + 0x10;
1161 write32le(p + 2, PTR_SIZE);
1162 p[6] = 0xff; /* jmp *(got + PTR_SIZE * 2) */
1163 p[7] = modrm;
1164 write32le(p + 8, PTR_SIZE * 2);
1167 /* The PLT slot refers to the relocation entry it needs
1168 via offset. The reloc entry is created below, so its
1169 offset is the current data_offset. */
1170 relofs = s1->got->reloc ? s1->got->reloc->data_offset : 0;
1171 symattr->plt_offset = plt->data_offset;
1172 p = section_ptr_add(plt, 16);
1173 p[0] = 0xff; /* jmp *(got + x) */
1174 p[1] = modrm;
1175 write32le(p + 2, s1->got->data_offset);
1176 p[6] = 0x68; /* push $xxx */
1177 #ifdef TCC_TARGET_X86_64
1178 /* On x86-64, the relocation is referred to by _index_. */
1179 write32le(p + 7, relofs / sizeof (ElfW_Rel));
1180 #else
1181 write32le(p + 7, relofs);
1182 #endif
1183 p[11] = 0xe9; /* jmp plt_start */
1184 write32le(p + 12, -(plt->data_offset));
1186 /* If this was an UNDEF symbol set the offset in the
1187 dynsymtab to the PLT slot, so that PC32 relocs to it
1188 can be resolved. */
1189 if (sym->st_shndx == SHN_UNDEF)
1190 offset = plt->data_offset - 16;
1192 #elif defined(TCC_TARGET_ARM)
1193 if (need_plt_entry) {
1194 Section *plt;
1195 uint8_t *p;
1197 /* if we build a DLL, we add a %ebx offset */
1198 if (s1->output_type == TCC_OUTPUT_DLL)
1199 tcc_error("DLLs unimplemented!");
1201 /* add a PLT entry */
1202 plt = s1->plt;
1203 if (plt->data_offset == 0) {
1204 /* first plt entry */
1205 p = section_ptr_add(plt, 16);
1206 write32le(p, 0xe52de004); /* push {lr} */
1207 write32le(p+4, 0xe59fe010); /* ldr lr, [pc, #16] */
1208 write32le(p+8, 0xe08fe00e); /* add lr, pc, lr */
1209 write32le(p+12, 0xe5bef008); /* ldr pc, [lr, #8]! */
1212 symattr->plt_offset = plt->data_offset;
1213 if (symattr->plt_thumb_stub) {
1214 p = section_ptr_add(plt, 20);
1215 write32le(p, 0x4778); /* bx pc */
1216 write32le(p+2, 0x46c0); /* nop */
1217 p += 4;
1218 } else
1219 p = section_ptr_add(plt, 16);
1220 write32le(p, 0xe59fc004); /* ldr ip, [pc, #4] ; GOT entry offset */
1221 write32le(p+4, 0xe08fc00c); /* add ip, pc, ip ; addr of GOT entry */
1222 write32le(p+8, 0xe59cf000); /* ldr pc, [ip] ; jump to GOT entry */
1223 write32le(p+12, s1->got->data_offset); /* GOT entry off once patched */
1225 /* the symbol is modified so that it will be relocated to
1226 the PLT */
1227 if (sym->st_shndx == SHN_UNDEF)
1228 offset = plt->data_offset - 16;
1230 #elif defined(TCC_TARGET_ARM64)
1231 if (need_plt_entry) {
1232 Section *plt;
1233 uint8_t *p;
1235 if (s1->output_type == TCC_OUTPUT_DLL)
1236 tcc_error("DLLs unimplemented!");
1238 plt = s1->plt;
1239 if (plt->data_offset == 0)
1240 section_ptr_add(plt, 32);
1241 symattr->plt_offset = plt->data_offset;
1242 p = section_ptr_add(plt, 16);
1243 write32le(p, s1->got->data_offset);
1244 write32le(p + 4, (uint64_t)s1->got->data_offset >> 32);
1246 if (sym->st_shndx == SHN_UNDEF)
1247 offset = plt->data_offset - 16;
1249 #elif defined(TCC_TARGET_C67)
1250 if (s1->dynsym) {
1251 tcc_error("C67 got not implemented");
1253 #else
1254 #error unsupported CPU
1255 #endif
1256 if (s1->dynsym) {
1257 /* XXX This might generate multiple syms for name. */
1258 index = put_elf_sym(s1->dynsym, offset,
1259 size, info, 0, sym->st_shndx, name);
1260 /* Create the relocation (it's against the GOT for PLT
1261 and GOT relocs). */
1262 put_elf_reloc(s1->dynsym, s1->got,
1263 s1->got->data_offset,
1264 reloc_type, index);
1265 } else {
1266 /* Without .dynsym (i.e. static link or memory output) we
1267 still need relocs against the generated got, so as to fill
1268 the entries with the symbol values (determined later). */
1269 put_elf_reloc(symtab_section, s1->got,
1270 s1->got->data_offset,
1271 reloc_type, sym_index);
1273 /* And now create the GOT slot itself. */
1274 ptr = section_ptr_add(s1->got, PTR_SIZE);
1275 *ptr = 0;
1276 if (need_plt_entry)
1277 return symattr->plt_offset;
1278 else
1279 return symattr->got_offset;
1282 /* build GOT and PLT entries */
1283 ST_FUNC void build_got_entries(TCCState *s1)
1285 Section *s;
1286 ElfW_Rel *rel;
1287 ElfW(Sym) *sym;
1288 int i, type, reloc_type, sym_index;
1290 for(i = 1; i < s1->nb_sections; i++) {
1291 s = s1->sections[i];
1292 if (s->sh_type != SHT_RELX)
1293 continue;
1294 /* no need to handle got relocations */
1295 if (s->link != symtab_section)
1296 continue;
1297 for_each_elem(s, 0, rel, ElfW_Rel) {
1298 type = ELFW(R_TYPE)(rel->r_info);
1299 switch(type) {
1300 #if defined(TCC_TARGET_I386)
1301 case R_386_GOT32:
1302 case R_386_GOT32X:
1303 case R_386_GOTOFF:
1304 case R_386_GOTPC:
1305 case R_386_PLT32:
1306 if (!s1->got)
1307 build_got(s1);
1308 if (type == R_386_GOT32 || type == R_386_GOT32X ||
1309 type == R_386_PLT32) {
1310 sym_index = ELFW(R_SYM)(rel->r_info);
1311 sym = &((ElfW(Sym) *)symtab_section->data)[sym_index];
1312 /* look at the symbol got offset. If none, then add one */
1313 if (type == R_386_GOT32 || type == R_386_GOT32X)
1314 reloc_type = R_386_GLOB_DAT;
1315 else
1316 reloc_type = R_386_JMP_SLOT;
1317 put_got_entry(s1, reloc_type, sym->st_size, sym->st_info,
1318 sym_index);
1320 break;
1321 #elif defined(TCC_TARGET_ARM)
1322 case R_ARM_PC24:
1323 case R_ARM_CALL:
1324 case R_ARM_JUMP24:
1325 case R_ARM_GOT32:
1326 case R_ARM_GOTOFF:
1327 case R_ARM_GOTPC:
1328 case R_ARM_PLT32:
1329 if (!s1->got)
1330 build_got(s1);
1331 sym_index = ELFW(R_SYM)(rel->r_info);
1332 sym = &((ElfW(Sym) *)symtab_section->data)[sym_index];
1333 if (type != R_ARM_GOTOFF && type != R_ARM_GOTPC
1334 && sym->st_shndx == SHN_UNDEF) {
1335 unsigned long ofs;
1336 /* look at the symbol got offset. If none, then add one */
1337 if (type == R_ARM_GOT32)
1338 reloc_type = R_ARM_GLOB_DAT;
1339 else
1340 reloc_type = R_ARM_JUMP_SLOT;
1341 ofs = put_got_entry(s1, reloc_type, sym->st_size,
1342 sym->st_info, sym_index);
1343 #ifdef DEBUG_RELOC
1344 printf ("maybegot: %s, %d, %d --> ofs=0x%x\n",
1345 (char *) symtab_section->link->data + sym->st_name,
1346 type, sym->st_shndx, ofs);
1347 #endif
1348 if (type != R_ARM_GOT32) {
1349 addr_t *ptr = (addr_t*)(s1->sections[s->sh_info]->data
1350 + rel->r_offset);
1351 /* x must be signed! */
1352 int x = *ptr & 0xffffff;
1353 x = (x << 8) >> 8;
1354 x <<= 2;
1355 x += ofs;
1356 x >>= 2;
1357 #ifdef DEBUG_RELOC
1358 printf ("insn=0x%x --> 0x%x (x==0x%x)\n", *ptr,
1359 (*ptr & 0xff000000) | x, x);
1360 #endif
1361 *ptr = (*ptr & 0xff000000) | x;
1364 break;
1365 case R_ARM_THM_JUMP24:
1366 sym_index = ELFW(R_SYM)(rel->r_info);
1367 sym = &((ElfW(Sym) *)symtab_section->data)[sym_index];
1368 /* We are relocating a jump from thumb code to arm code */
1369 if (sym->st_shndx != SHN_UNDEF && !(sym->st_value & 1)) {
1370 int index;
1371 uint8_t *p;
1372 char *name, buf[1024];
1373 Section *text_section;
1375 name = (char *) symtab_section->link->data + sym->st_name;
1376 text_section = s1->sections[sym->st_shndx];
1377 /* Modify reloc to target a thumb stub to switch to ARM */
1378 snprintf(buf, sizeof(buf), "%s_from_thumb", name);
1379 index = put_elf_sym(symtab_section,
1380 text_section->data_offset + 1,
1381 sym->st_size, sym->st_info, 0,
1382 sym->st_shndx, buf);
1383 rel->r_info = ELFW(R_INFO)(index, type);
1384 /* Create a thumb stub fonction to switch to ARM mode */
1385 put_elf_reloc(symtab_section, text_section,
1386 text_section->data_offset + 4, R_ARM_JUMP24,
1387 sym_index);
1388 p = section_ptr_add(text_section, 8);
1389 write32le(p, 0x4778); /* bx pc */
1390 write32le(p+2, 0x46c0); /* nop */
1391 write32le(p+4, 0xeafffffe); /* b $sym */
1393 #elif defined(TCC_TARGET_ARM64)
1394 //xx Other cases may be required here:
1395 case R_AARCH64_ADR_GOT_PAGE:
1396 case R_AARCH64_LD64_GOT_LO12_NC:
1397 if (!s1->got)
1398 build_got(s1);
1399 sym_index = ELFW(R_SYM)(rel->r_info);
1400 sym = &((ElfW(Sym) *)symtab_section->data)[sym_index];
1401 reloc_type = R_AARCH64_GLOB_DAT;
1402 put_got_entry(s1, reloc_type, sym->st_size, sym->st_info,
1403 sym_index);
1404 break;
1406 case R_AARCH64_JUMP26:
1407 case R_AARCH64_CALL26:
1408 if (!s1->got)
1409 build_got(s1);
1410 sym_index = ELFW(R_SYM)(rel->r_info);
1411 sym = &((ElfW(Sym) *)symtab_section->data)[sym_index];
1412 if (sym->st_shndx == SHN_UNDEF) {
1413 unsigned long ofs;
1414 reloc_type = R_AARCH64_JUMP_SLOT;
1415 ofs = put_got_entry(s1, reloc_type, sym->st_size,
1416 sym->st_info, sym_index);
1417 /* We store the place of the generated PLT slot
1418 in our addend. */
1419 rel->r_addend += ofs;
1421 break;
1422 #elif defined(TCC_TARGET_C67)
1423 case R_C60_GOT32:
1424 case R_C60_GOTOFF:
1425 case R_C60_GOTPC:
1426 case R_C60_PLT32:
1427 if (!s1->got)
1428 build_got(s1);
1429 if (type == R_C60_GOT32 || type == R_C60_PLT32) {
1430 sym_index = ELFW(R_SYM)(rel->r_info);
1431 sym = &((ElfW(Sym) *)symtab_section->data)[sym_index];
1432 /* look at the symbol got offset. If none, then add one */
1433 if (type == R_C60_GOT32)
1434 reloc_type = R_C60_GLOB_DAT;
1435 else
1436 reloc_type = R_C60_JMP_SLOT;
1437 put_got_entry(s1, reloc_type, sym->st_size, sym->st_info,
1438 sym_index);
1440 break;
1441 #elif defined(TCC_TARGET_X86_64)
1442 case R_X86_64_GOT32:
1443 case R_X86_64_GOTTPOFF:
1444 case R_X86_64_GOTPCREL:
1445 case R_X86_64_GOTPCRELX:
1446 case R_X86_64_REX_GOTPCRELX:
1447 case R_X86_64_PLT32:
1448 sym_index = ELFW(R_SYM)(rel->r_info);
1449 sym = &((ElfW(Sym) *)symtab_section->data)[sym_index];
1450 if (type == R_X86_64_PLT32 &&
1451 ELFW(ST_VISIBILITY)(sym->st_other) != STV_DEFAULT)
1453 rel->r_info = ELFW(R_INFO)(sym_index, R_X86_64_PC32);
1454 break;
1457 if (!s1->got) {
1458 build_got(s1);
1459 sym = &((ElfW(Sym) *)symtab_section->data)[sym_index];
1461 if (type == R_X86_64_GOT32 || type == R_X86_64_GOTPCREL ||
1462 type == R_X86_64_GOTPCRELX ||
1463 type == R_X86_64_REX_GOTPCRELX ||
1464 type == R_X86_64_PLT32) {
1465 unsigned long ofs;
1466 /* look at the symbol got offset. If none, then add one */
1467 if (type == R_X86_64_PLT32)
1468 reloc_type = R_X86_64_JUMP_SLOT;
1469 else
1470 reloc_type = R_X86_64_GLOB_DAT;
1471 ofs = put_got_entry(s1, reloc_type, sym->st_size,
1472 sym->st_info, sym_index);
1473 if (type == R_X86_64_PLT32)
1474 /* We store the place of the generated PLT slot
1475 in our addend. */
1476 rel->r_addend += ofs;
1478 break;
1479 #else
1480 #error unsupported CPU
1481 #endif
1482 default:
1483 break;
1489 ST_FUNC Section *new_symtab(TCCState *s1,
1490 const char *symtab_name, int sh_type, int sh_flags,
1491 const char *strtab_name,
1492 const char *hash_name, int hash_sh_flags)
1494 Section *symtab, *strtab, *hash;
1495 int *ptr, nb_buckets;
1497 symtab = new_section(s1, symtab_name, sh_type, sh_flags);
1498 symtab->sh_entsize = sizeof(ElfW(Sym));
1499 strtab = new_section(s1, strtab_name, SHT_STRTAB, sh_flags);
1500 put_elf_str(strtab, "");
1501 symtab->link = strtab;
1502 put_elf_sym(symtab, 0, 0, 0, 0, 0, NULL);
1504 nb_buckets = 1;
1506 hash = new_section(s1, hash_name, SHT_HASH, hash_sh_flags);
1507 hash->sh_entsize = sizeof(int);
1508 symtab->hash = hash;
1509 hash->link = symtab;
1511 ptr = section_ptr_add(hash, (2 + nb_buckets + 1) * sizeof(int));
1512 ptr[0] = nb_buckets;
1513 ptr[1] = 1;
1514 memset(ptr + 2, 0, (nb_buckets + 1) * sizeof(int));
1515 return symtab;
1518 /* put dynamic tag */
1519 static void put_dt(Section *dynamic, int dt, addr_t val)
1521 ElfW(Dyn) *dyn;
1522 dyn = section_ptr_add(dynamic, sizeof(ElfW(Dyn)));
1523 dyn->d_tag = dt;
1524 dyn->d_un.d_val = val;
1527 static void add_init_array_defines(TCCState *s1, const char *section_name)
1529 Section *s;
1530 long end_offset;
1531 char sym_start[1024];
1532 char sym_end[1024];
1534 snprintf(sym_start, sizeof(sym_start), "__%s_start", section_name + 1);
1535 snprintf(sym_end, sizeof(sym_end), "__%s_end", section_name + 1);
1537 s = find_section(s1, section_name);
1538 if (!s) {
1539 end_offset = 0;
1540 s = data_section;
1541 } else {
1542 end_offset = s->data_offset;
1545 add_elf_sym(symtab_section,
1546 0, 0,
1547 ELFW(ST_INFO)(STB_GLOBAL, STT_NOTYPE), 0,
1548 s->sh_num, sym_start);
1549 add_elf_sym(symtab_section,
1550 end_offset, 0,
1551 ELFW(ST_INFO)(STB_GLOBAL, STT_NOTYPE), 0,
1552 s->sh_num, sym_end);
1555 static int tcc_add_support(TCCState *s1, const char *filename)
1557 char buf[1024];
1558 snprintf(buf, sizeof(buf), "%s/%s/%s", s1->tcc_lib_path,
1559 /* an cpu specific path inside tcc_lib_path, mainly for keeping libtcc1.a */
1560 #ifdef TCC_TARGET_I386
1561 "i386"
1562 #endif
1563 #ifdef TCC_TARGET_X86_64
1564 "x86-64"
1565 #endif
1566 #ifdef TCC_TARGET_ARM
1567 "arm"
1568 #endif
1569 #ifdef TCC_TARGET_ARM64
1570 "arm64"
1571 #endif
1572 #ifdef TCC_TARGET_C67
1573 "C67"
1574 #endif
1575 ,filename);
1577 return tcc_add_file(s1, buf, TCC_FILETYPE_BINARY);
1580 ST_FUNC void tcc_add_bcheck(TCCState *s1)
1582 #ifdef CONFIG_TCC_BCHECK
1583 addr_t *ptr;
1585 if (0 == s1->do_bounds_check)
1586 return;
1588 /* XXX: add an object file to do that */
1589 ptr = section_ptr_add(bounds_section, sizeof(*ptr));
1590 *ptr = 0;
1591 add_elf_sym(symtab_section, 0, 0,
1592 ELFW(ST_INFO)(STB_GLOBAL, STT_NOTYPE), 0,
1593 bounds_section->sh_num, "__bounds_start");
1594 if (s1->output_type != TCC_OUTPUT_MEMORY) {
1595 /* add 'call __bound_init()' in .init section */
1597 /* XXX not called on MSYS, reason is unknown. For this
1598 case a call to __bound_init is performed in bcheck.c
1599 when __bound_ptr_add, __bound_new_region,
1600 __bound_delete_region called */
1602 int sym_index = find_elf_sym(symtab_section, "__bound_init");
1603 if (sym_index) {
1604 Section *init_section = find_section(s1, ".init");
1605 unsigned char *pinit = section_ptr_add(init_section, 5);
1606 pinit[0] = 0xe8;
1607 write32le(pinit + 1, -4);
1608 put_elf_reloc(symtab_section, init_section,
1609 init_section->data_offset - 4, R_386_PC32, sym_index);
1611 else
1612 tcc_warning("__bound_init not defined");
1614 #endif
1617 /* add tcc runtime libraries */
1618 ST_FUNC void tcc_add_runtime(TCCState *s1)
1620 tcc_add_pragma_libs(s1);
1622 /* add libc */
1623 if (!s1->nostdlib) {
1624 tcc_add_library(s1, "c");
1625 #ifdef CONFIG_USE_LIBGCC
1626 if (!s1->static_link) {
1627 tcc_add_file(s1, TCC_LIBGCC, TCC_FILETYPE_BINARY);
1629 #endif
1630 tcc_add_support(s1, "libtcc1.a");
1633 /* tcc_add_bcheck tries to relocate a call to __bound_init in _init so
1634 libtcc1.a must be loaded before for __bound_init to be defined and
1635 crtn.o must be loaded after to not finalize _init too early. */
1636 tcc_add_bcheck(s1);
1638 if (!s1->nostdlib) {
1639 /* add crt end if not memory output */
1640 if (s1->output_type != TCC_OUTPUT_MEMORY)
1641 tcc_add_crt(s1, "crtn.o");
1645 /* add various standard linker symbols (must be done after the
1646 sections are filled (for example after allocating common
1647 symbols)) */
1648 ST_FUNC void tcc_add_linker_symbols(TCCState *s1)
1650 char buf[1024];
1651 int i;
1652 Section *s;
1654 add_elf_sym(symtab_section,
1655 text_section->data_offset, 0,
1656 ELFW(ST_INFO)(STB_GLOBAL, STT_NOTYPE), 0,
1657 text_section->sh_num, "_etext");
1658 add_elf_sym(symtab_section,
1659 data_section->data_offset, 0,
1660 ELFW(ST_INFO)(STB_GLOBAL, STT_NOTYPE), 0,
1661 data_section->sh_num, "_edata");
1662 add_elf_sym(symtab_section,
1663 bss_section->data_offset, 0,
1664 ELFW(ST_INFO)(STB_GLOBAL, STT_NOTYPE), 0,
1665 bss_section->sh_num, "_end");
1666 /* horrible new standard ldscript defines */
1667 add_init_array_defines(s1, ".preinit_array");
1668 add_init_array_defines(s1, ".init_array");
1669 add_init_array_defines(s1, ".fini_array");
1671 /* add start and stop symbols for sections whose name can be
1672 expressed in C */
1673 for(i = 1; i < s1->nb_sections; i++) {
1674 s = s1->sections[i];
1675 if (s->sh_type == SHT_PROGBITS &&
1676 (s->sh_flags & SHF_ALLOC)) {
1677 const char *p;
1678 int ch;
1680 /* check if section name can be expressed in C */
1681 p = s->name;
1682 for(;;) {
1683 ch = *p;
1684 if (!ch)
1685 break;
1686 if (!isid(ch) && !isnum(ch))
1687 goto next_sec;
1688 p++;
1690 snprintf(buf, sizeof(buf), "__start_%s", s->name);
1691 add_elf_sym(symtab_section,
1692 0, 0,
1693 ELFW(ST_INFO)(STB_GLOBAL, STT_NOTYPE), 0,
1694 s->sh_num, buf);
1695 snprintf(buf, sizeof(buf), "__stop_%s", s->name);
1696 add_elf_sym(symtab_section,
1697 s->data_offset, 0,
1698 ELFW(ST_INFO)(STB_GLOBAL, STT_NOTYPE), 0,
1699 s->sh_num, buf);
1701 next_sec: ;
1705 static void tcc_output_binary(TCCState *s1, FILE *f,
1706 const int *sec_order)
1708 Section *s;
1709 int i, offset, size;
1711 offset = 0;
1712 for(i=1;i<s1->nb_sections;i++) {
1713 s = s1->sections[sec_order[i]];
1714 if (s->sh_type != SHT_NOBITS &&
1715 (s->sh_flags & SHF_ALLOC)) {
1716 while (offset < s->sh_offset) {
1717 fputc(0, f);
1718 offset++;
1720 size = s->sh_size;
1721 fwrite(s->data, 1, size, f);
1722 offset += size;
1727 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
1728 #define HAVE_PHDR 1
1729 #define EXTRA_RELITEMS 14
1731 /* move the relocation value from .dynsym to .got */
1732 void patch_dynsym_undef(TCCState *s1, Section *s)
1734 uint32_t *gotd = (void *)s1->got->data;
1735 ElfW(Sym) *sym;
1737 gotd += 3; /* dummy entries in .got */
1738 /* relocate symbols in .dynsym */
1739 for_each_elem(s, 1, sym, ElfW(Sym)) {
1740 if (sym->st_shndx == SHN_UNDEF) {
1741 *gotd++ = sym->st_value + 6; /* XXX 6 is magic ? */
1742 sym->st_value = 0;
1746 #else
1747 #define HAVE_PHDR 1
1748 #define EXTRA_RELITEMS 9
1750 /* zero plt offsets of weak symbols in .dynsym */
1751 void patch_dynsym_undef(TCCState *s1, Section *s)
1753 ElfW(Sym) *sym;
1755 for_each_elem(s, 1, sym, ElfW(Sym))
1756 if (sym->st_shndx == SHN_UNDEF && ELFW(ST_BIND)(sym->st_info) == STB_WEAK)
1757 sym->st_value = 0;
1759 #endif
1761 ST_FUNC void fill_got_entry(TCCState *s1, ElfW_Rel *rel)
1763 int sym_index = ELFW(R_SYM) (rel->r_info);
1764 ElfW(Sym) *sym = &((ElfW(Sym) *) symtab_section->data)[sym_index];
1765 unsigned long offset;
1767 if (sym_index >= s1->nb_sym_attrs)
1768 return;
1769 offset = s1->sym_attrs[sym_index].got_offset;
1770 section_reserve(s1->got, offset + PTR_SIZE);
1771 #ifdef TCC_TARGET_X86_64
1772 /* only works for x86-64 */
1773 write32le(s1->got->data + offset + 4, sym->st_value >> 32);
1774 #endif
1775 write32le(s1->got->data + offset, sym->st_value & 0xffffffff);
1778 /* Perform relocation to GOT or PLT entries */
1779 ST_FUNC void fill_got(TCCState *s1)
1781 Section *s;
1782 ElfW_Rel *rel;
1783 int i;
1785 for(i = 1; i < s1->nb_sections; i++) {
1786 s = s1->sections[i];
1787 if (s->sh_type != SHT_RELX)
1788 continue;
1789 /* no need to handle got relocations */
1790 if (s->link != symtab_section)
1791 continue;
1792 for_each_elem(s, 0, rel, ElfW_Rel) {
1793 switch (ELFW(R_TYPE) (rel->r_info)) {
1794 case R_X86_64_GOT32:
1795 case R_X86_64_GOTPCREL:
1796 case R_X86_64_GOTPCRELX:
1797 case R_X86_64_REX_GOTPCRELX:
1798 case R_X86_64_PLT32:
1799 fill_got_entry(s1, rel);
1800 break;
1806 /* Bind symbols of executable: resolve undefined symbols from exported symbols
1807 in shared libraries and export non local defined symbols to shared libraries
1808 if -rdynamic switch was given on command line */
1809 static void bind_exe_dynsyms(TCCState *s1)
1811 const char *name;
1812 int sym_index, index;
1813 ElfW(Sym) *sym, *esym;
1814 int type;
1816 /* Resolve undefined symbols from dynamic symbols. When there is a match:
1817 - if STT_FUNC or STT_GNU_IFUNC symbol -> add it in PLT
1818 - if STT_OBJECT symbol -> add it in .bss section with suitable reloc */
1819 for_each_elem(symtab_section, 1, sym, ElfW(Sym)) {
1820 if (sym->st_shndx == SHN_UNDEF) {
1821 name = (char *) symtab_section->link->data + sym->st_name;
1822 sym_index = find_elf_sym(s1->dynsymtab_section, name);
1823 if (sym_index) {
1824 esym = &((ElfW(Sym) *)s1->dynsymtab_section->data)[sym_index];
1825 type = ELFW(ST_TYPE)(esym->st_info);
1826 if ((type == STT_FUNC) || (type == STT_GNU_IFUNC)) {
1827 /* Indirect functions shall have STT_FUNC type in executable
1828 * dynsym section. Indeed, a dlsym call following a lazy
1829 * resolution would pick the symbol value from the
1830 * executable dynsym entry which would contain the address
1831 * of the function wanted by the caller of dlsym instead of
1832 * the address of the function that would return that
1833 * address */
1834 put_got_entry(s1, R_JMP_SLOT, esym->st_size,
1835 ELFW(ST_INFO)(STB_GLOBAL,STT_FUNC),
1836 sym - (ElfW(Sym) *)symtab_section->data);
1837 } else if (type == STT_OBJECT) {
1838 unsigned long offset;
1839 ElfW(Sym) *dynsym;
1840 offset = bss_section->data_offset;
1841 /* XXX: which alignment ? */
1842 offset = (offset + 16 - 1) & -16;
1843 index = put_elf_sym(s1->dynsym, offset, esym->st_size,
1844 esym->st_info, 0, bss_section->sh_num,
1845 name);
1846 /* Ensure R_COPY works for weak symbol aliases */
1847 if (ELFW(ST_BIND)(esym->st_info) == STB_WEAK) {
1848 for_each_elem(s1->dynsymtab_section, 1, dynsym, ElfW(Sym)) {
1849 if ((dynsym->st_value == esym->st_value)
1850 && (ELFW(ST_BIND)(dynsym->st_info) == STB_GLOBAL)) {
1851 char *dynname = (char *) s1->dynsymtab_section->link->data
1852 + dynsym->st_name;
1853 put_elf_sym(s1->dynsym, offset, dynsym->st_size,
1854 dynsym->st_info, 0,
1855 bss_section->sh_num, dynname);
1856 break;
1860 put_elf_reloc(s1->dynsym, bss_section,
1861 offset, R_COPY, index);
1862 offset += esym->st_size;
1863 bss_section->data_offset = offset;
1865 } else {
1866 /* STB_WEAK undefined symbols are accepted */
1867 /* XXX: _fp_hw seems to be part of the ABI, so we ignore it */
1868 if (ELFW(ST_BIND)(sym->st_info) == STB_WEAK ||
1869 !strcmp(name, "_fp_hw")) {
1870 } else {
1871 tcc_error_noabort("undefined symbol '%s'", name);
1874 } else if (s1->rdynamic && ELFW(ST_BIND)(sym->st_info) != STB_LOCAL) {
1875 /* if -rdynamic option, then export all non local symbols */
1876 name = (char *) symtab_section->link->data + sym->st_name;
1877 put_elf_sym(s1->dynsym, sym->st_value, sym->st_size, sym->st_info,
1878 0, sym->st_shndx, name);
1883 /* Bind symbols of libraries: export non local symbols of executable that
1884 resolve undefined symbols of shared libraries */
1885 static void bind_libs_dynsyms(TCCState *s1)
1887 const char *name;
1888 int sym_index;
1889 ElfW(Sym) *sym, *esym;
1891 /* now look at unresolved dynamic symbols and export
1892 corresponding symbol */
1893 for_each_elem(s1->dynsymtab_section, 1, esym, ElfW(Sym)) {
1894 name = (char *) s1->dynsymtab_section->link->data + esym->st_name;
1895 sym_index = find_elf_sym(symtab_section, name);
1896 if (sym_index) {
1897 /* XXX: avoid adding a symbol if already present because of
1898 -rdynamic ? */
1899 sym = &((ElfW(Sym) *)symtab_section->data)[sym_index];
1900 if (sym->st_shndx != SHN_UNDEF)
1901 put_elf_sym(s1->dynsym, sym->st_value, sym->st_size,
1902 sym->st_info, 0, sym->st_shndx, name);
1903 } else if (esym->st_shndx == SHN_UNDEF) {
1904 /* weak symbols can stay undefined */
1905 if (ELFW(ST_BIND)(esym->st_info) != STB_WEAK)
1906 tcc_warning("undefined dynamic symbol '%s'", name);
1911 /* Export all non local symbols (for shared libraries) */
1912 static void export_global_syms(TCCState *s1)
1914 int nb_syms, dynindex, index;
1915 const char *name;
1916 ElfW(Sym) *sym;
1918 nb_syms = symtab_section->data_offset / sizeof(ElfW(Sym));
1919 s1->symtab_to_dynsym = tcc_mallocz(sizeof(int) * nb_syms);
1920 for_each_elem(symtab_section, 1, sym, ElfW(Sym)) {
1921 if (ELFW(ST_BIND)(sym->st_info) != STB_LOCAL) {
1922 name = (char *) symtab_section->link->data + sym->st_name;
1923 dynindex = put_elf_sym(s1->dynsym, sym->st_value, sym->st_size,
1924 sym->st_info, 0, sym->st_shndx, name);
1925 index = sym - (ElfW(Sym) *) symtab_section->data;
1926 s1->symtab_to_dynsym[index] = dynindex;
1931 /* relocate the PLT: compute addresses and offsets in the PLT now that final
1932 address for PLT and GOT are known (see fill_program_header) */
1933 ST_FUNC void relocate_plt(TCCState *s1)
1935 uint8_t *p, *p_end;
1937 if (!s1->plt)
1938 return;
1940 p = s1->plt->data;
1941 p_end = p + s1->plt->data_offset;
1942 if (p < p_end) {
1943 #if defined(TCC_TARGET_I386)
1944 write32le(p + 2, read32le(p + 2) + s1->got->sh_addr);
1945 write32le(p + 8, read32le(p + 8) + s1->got->sh_addr);
1946 p += 16;
1947 while (p < p_end) {
1948 write32le(p + 2, read32le(p + 2) + s1->got->sh_addr);
1949 p += 16;
1951 #elif defined(TCC_TARGET_X86_64)
1952 int x = s1->got->sh_addr - s1->plt->sh_addr - 6;
1953 write32le(p + 2, read32le(p + 2) + x);
1954 write32le(p + 8, read32le(p + 8) + x - 6);
1955 p += 16;
1956 while (p < p_end) {
1957 write32le(p + 2, read32le(p + 2) + x + s1->plt->data - p);
1958 p += 16;
1960 #elif defined(TCC_TARGET_ARM)
1961 int x;
1962 x=s1->got->sh_addr - s1->plt->sh_addr - 12;
1963 p += 16;
1964 while (p < p_end) {
1965 if (read32le(p) == 0x46c04778) /* PLT Thumb stub present */
1966 p += 4;
1967 write32le(p + 12, x + read32le(p + 12) + s1->plt->data - p);
1968 p += 16;
1970 #elif defined(TCC_TARGET_ARM64)
1971 uint64_t plt = s1->plt->sh_addr;
1972 uint64_t got = s1->got->sh_addr;
1973 uint64_t off = (got >> 12) - (plt >> 12);
1974 if ((off + ((uint32_t)1 << 20)) >> 21)
1975 tcc_error("Failed relocating PLT (off=0x%lx, got=0x%lx, plt=0x%lx)", off, got, plt);
1976 write32le(p, 0xa9bf7bf0); // stp x16,x30,[sp,#-16]!
1977 write32le(p + 4, (0x90000010 | // adrp x16,...
1978 (off & 0x1ffffc) << 3 | (off & 3) << 29));
1979 write32le(p + 8, (0xf9400211 | // ldr x17,[x16,#...]
1980 (got & 0xff8) << 7));
1981 write32le(p + 12, (0x91000210 | // add x16,x16,#...
1982 (got & 0xfff) << 10));
1983 write32le(p + 16, 0xd61f0220); // br x17
1984 write32le(p + 20, 0xd503201f); // nop
1985 write32le(p + 24, 0xd503201f); // nop
1986 write32le(p + 28, 0xd503201f); // nop
1987 p += 32;
1988 while (p < p_end) {
1989 uint64_t pc = plt + (p - s1->plt->data);
1990 uint64_t addr = got + read64le(p);
1991 uint64_t off = (addr >> 12) - (pc >> 12);
1992 if ((off + ((uint32_t)1 << 20)) >> 21)
1993 tcc_error("Failed relocating PLT (off=0x%lx, addr=0x%lx, pc=0x%lx)", off, addr, pc);
1994 write32le(p, (0x90000010 | // adrp x16,...
1995 (off & 0x1ffffc) << 3 | (off & 3) << 29));
1996 write32le(p + 4, (0xf9400211 | // ldr x17,[x16,#...]
1997 (addr & 0xff8) << 7));
1998 write32le(p + 8, (0x91000210 | // add x16,x16,#...
1999 (addr & 0xfff) << 10));
2000 write32le(p + 12, 0xd61f0220); // br x17
2001 p += 16;
2003 #elif defined(TCC_TARGET_C67)
2004 /* XXX: TODO */
2005 #else
2006 #error unsupported CPU
2007 #endif
2011 /* Allocate strings for section names and decide if an unallocated section
2012 should be output.
2014 NOTE: the strsec section comes last, so its size is also correct ! */
2015 static void alloc_sec_names(TCCState *s1, int file_type, Section *strsec)
2017 int i;
2018 Section *s;
2020 /* Allocate strings for section names */
2021 for(i = 1; i < s1->nb_sections; i++) {
2022 s = s1->sections[i];
2023 s->sh_name = put_elf_str(strsec, s->name);
2024 /* when generating a DLL, we include relocations but we may
2025 patch them */
2026 if (file_type == TCC_OUTPUT_DLL &&
2027 s->sh_type == SHT_RELX &&
2028 !(s->sh_flags & SHF_ALLOC)) {
2029 /* gr: avoid bogus relocs for empty (debug) sections */
2030 if (s1->sections[s->sh_info]->sh_flags & SHF_ALLOC)
2031 prepare_dynamic_rel(s1, s);
2032 else if (s1->do_debug)
2033 s->sh_size = s->data_offset;
2034 } else if (s1->do_debug ||
2035 file_type == TCC_OUTPUT_OBJ ||
2036 file_type == TCC_OUTPUT_EXE ||
2037 (s->sh_flags & SHF_ALLOC) ||
2038 i == (s1->nb_sections - 1)) {
2039 /* we output all sections if debug or object file */
2040 s->sh_size = s->data_offset;
2045 /* Info to be copied in dynamic section */
2046 struct dyn_inf {
2047 Section *dynamic;
2048 Section *dynstr;
2049 unsigned long dyn_rel_off;
2050 addr_t rel_addr;
2051 addr_t rel_size;
2052 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
2053 addr_t bss_addr;
2054 addr_t bss_size;
2055 #endif
2058 /* Assign sections to segments and decide how are sections laid out when loaded
2059 in memory. This function also fills corresponding program headers. */
2060 static int layout_sections(TCCState *s1, ElfW(Phdr) *phdr, int phnum,
2061 Section *interp, Section* strsec,
2062 struct dyn_inf *dyninf, int *sec_order)
2064 int i, j, k, file_type, sh_order_index, file_offset;
2065 unsigned long s_align;
2066 long long tmp;
2067 addr_t addr;
2068 ElfW(Phdr) *ph;
2069 Section *s;
2071 file_type = s1->output_type;
2072 sh_order_index = 1;
2073 file_offset = 0;
2074 if (s1->output_format == TCC_OUTPUT_FORMAT_ELF)
2075 file_offset = sizeof(ElfW(Ehdr)) + phnum * sizeof(ElfW(Phdr));
2076 s_align = ELF_PAGE_SIZE;
2077 if (s1->section_align)
2078 s_align = s1->section_align;
2080 if (phnum > 0) {
2081 if (s1->has_text_addr) {
2082 int a_offset, p_offset;
2083 addr = s1->text_addr;
2084 /* we ensure that (addr % ELF_PAGE_SIZE) == file_offset %
2085 ELF_PAGE_SIZE */
2086 a_offset = (int) (addr & (s_align - 1));
2087 p_offset = file_offset & (s_align - 1);
2088 if (a_offset < p_offset)
2089 a_offset += s_align;
2090 file_offset += (a_offset - p_offset);
2091 } else {
2092 if (file_type == TCC_OUTPUT_DLL)
2093 addr = 0;
2094 else
2095 addr = ELF_START_ADDR;
2096 /* compute address after headers */
2097 addr += (file_offset & (s_align - 1));
2100 ph = &phdr[0];
2101 /* Leave one program headers for the program interpreter and one for
2102 the program header table itself if needed. These are done later as
2103 they require section layout to be done first. */
2104 if (interp)
2105 ph += 1 + HAVE_PHDR;
2107 /* dynamic relocation table information, for .dynamic section */
2108 dyninf->rel_addr = dyninf->rel_size = 0;
2109 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
2110 dyninf->bss_addr = dyninf->bss_size = 0;
2111 #endif
2113 for(j = 0; j < 2; j++) {
2114 ph->p_type = PT_LOAD;
2115 if (j == 0)
2116 ph->p_flags = PF_R | PF_X;
2117 else
2118 ph->p_flags = PF_R | PF_W;
2119 ph->p_align = s_align;
2121 /* Decide the layout of sections loaded in memory. This must
2122 be done before program headers are filled since they contain
2123 info about the layout. We do the following ordering: interp,
2124 symbol tables, relocations, progbits, nobits */
2125 /* XXX: do faster and simpler sorting */
2126 for(k = 0; k < 5; k++) {
2127 for(i = 1; i < s1->nb_sections; i++) {
2128 s = s1->sections[i];
2129 /* compute if section should be included */
2130 if (j == 0) {
2131 if ((s->sh_flags & (SHF_ALLOC | SHF_WRITE)) !=
2132 SHF_ALLOC)
2133 continue;
2134 } else {
2135 if ((s->sh_flags & (SHF_ALLOC | SHF_WRITE)) !=
2136 (SHF_ALLOC | SHF_WRITE))
2137 continue;
2139 if (s == interp) {
2140 if (k != 0)
2141 continue;
2142 } else if (s->sh_type == SHT_DYNSYM ||
2143 s->sh_type == SHT_STRTAB ||
2144 s->sh_type == SHT_HASH) {
2145 if (k != 1)
2146 continue;
2147 } else if (s->sh_type == SHT_RELX) {
2148 if (k != 2)
2149 continue;
2150 } else if (s->sh_type == SHT_NOBITS) {
2151 if (k != 4)
2152 continue;
2153 } else {
2154 if (k != 3)
2155 continue;
2157 sec_order[sh_order_index++] = i;
2159 /* section matches: we align it and add its size */
2160 tmp = addr;
2161 addr = (addr + s->sh_addralign - 1) &
2162 ~(s->sh_addralign - 1);
2163 file_offset += (int) ( addr - tmp );
2164 s->sh_offset = file_offset;
2165 s->sh_addr = addr;
2167 /* update program header infos */
2168 if (ph->p_offset == 0) {
2169 ph->p_offset = file_offset;
2170 ph->p_vaddr = addr;
2171 ph->p_paddr = ph->p_vaddr;
2173 /* update dynamic relocation infos */
2174 if (s->sh_type == SHT_RELX) {
2175 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
2176 if (!strcmp(strsec->data + s->sh_name, ".rel.got")) {
2177 dyninf->rel_addr = addr;
2178 dyninf->rel_size += s->sh_size; /* XXX only first rel. */
2180 if (!strcmp(strsec->data + s->sh_name, ".rel.bss")) {
2181 dyninf->bss_addr = addr;
2182 dyninf->bss_size = s->sh_size; /* XXX only first rel. */
2184 #else
2185 if (dyninf->rel_size == 0)
2186 dyninf->rel_addr = addr;
2187 dyninf->rel_size += s->sh_size;
2188 #endif
2190 addr += s->sh_size;
2191 if (s->sh_type != SHT_NOBITS)
2192 file_offset += s->sh_size;
2195 if (j == 0) {
2196 /* Make the first PT_LOAD segment include the program
2197 headers itself (and the ELF header as well), it'll
2198 come out with same memory use but will make various
2199 tools like binutils strip work better. */
2200 ph->p_offset &= ~(ph->p_align - 1);
2201 ph->p_vaddr &= ~(ph->p_align - 1);
2202 ph->p_paddr &= ~(ph->p_align - 1);
2204 ph->p_filesz = file_offset - ph->p_offset;
2205 ph->p_memsz = addr - ph->p_vaddr;
2206 ph++;
2207 if (j == 0) {
2208 if (s1->output_format == TCC_OUTPUT_FORMAT_ELF) {
2209 /* if in the middle of a page, we duplicate the page in
2210 memory so that one copy is RX and the other is RW */
2211 if ((addr & (s_align - 1)) != 0)
2212 addr += s_align;
2213 } else {
2214 addr = (addr + s_align - 1) & ~(s_align - 1);
2215 file_offset = (file_offset + s_align - 1) & ~(s_align - 1);
2221 /* all other sections come after */
2222 for(i = 1; i < s1->nb_sections; i++) {
2223 s = s1->sections[i];
2224 if (phnum > 0 && (s->sh_flags & SHF_ALLOC))
2225 continue;
2226 sec_order[sh_order_index++] = i;
2228 file_offset = (file_offset + s->sh_addralign - 1) &
2229 ~(s->sh_addralign - 1);
2230 s->sh_offset = file_offset;
2231 if (s->sh_type != SHT_NOBITS)
2232 file_offset += s->sh_size;
2235 return file_offset;
2238 static void fill_unloadable_phdr(ElfW(Phdr) *phdr, int phnum, Section *interp,
2239 Section *dynamic)
2241 ElfW(Phdr) *ph;
2243 /* if interpreter, then add corresponding program header */
2244 if (interp) {
2245 ph = &phdr[0];
2247 if (HAVE_PHDR)
2249 int len = phnum * sizeof(ElfW(Phdr));
2251 ph->p_type = PT_PHDR;
2252 ph->p_offset = sizeof(ElfW(Ehdr));
2253 ph->p_vaddr = interp->sh_addr - len;
2254 ph->p_paddr = ph->p_vaddr;
2255 ph->p_filesz = ph->p_memsz = len;
2256 ph->p_flags = PF_R | PF_X;
2257 ph->p_align = 4; /* interp->sh_addralign; */
2258 ph++;
2261 ph->p_type = PT_INTERP;
2262 ph->p_offset = interp->sh_offset;
2263 ph->p_vaddr = interp->sh_addr;
2264 ph->p_paddr = ph->p_vaddr;
2265 ph->p_filesz = interp->sh_size;
2266 ph->p_memsz = interp->sh_size;
2267 ph->p_flags = PF_R;
2268 ph->p_align = interp->sh_addralign;
2271 /* if dynamic section, then add corresponding program header */
2272 if (dynamic) {
2273 ph = &phdr[phnum - 1];
2275 ph->p_type = PT_DYNAMIC;
2276 ph->p_offset = dynamic->sh_offset;
2277 ph->p_vaddr = dynamic->sh_addr;
2278 ph->p_paddr = ph->p_vaddr;
2279 ph->p_filesz = dynamic->sh_size;
2280 ph->p_memsz = dynamic->sh_size;
2281 ph->p_flags = PF_R | PF_W;
2282 ph->p_align = dynamic->sh_addralign;
2286 /* Fill the dynamic section with tags describing the address and size of
2287 sections */
2288 static void fill_dynamic(TCCState *s1, struct dyn_inf *dyninf)
2290 Section *dynamic;
2292 dynamic = dyninf->dynamic;
2294 /* put dynamic section entries */
2295 dynamic->data_offset = dyninf->dyn_rel_off;
2296 put_dt(dynamic, DT_HASH, s1->dynsym->hash->sh_addr);
2297 put_dt(dynamic, DT_STRTAB, dyninf->dynstr->sh_addr);
2298 put_dt(dynamic, DT_SYMTAB, s1->dynsym->sh_addr);
2299 put_dt(dynamic, DT_STRSZ, dyninf->dynstr->data_offset);
2300 put_dt(dynamic, DT_SYMENT, sizeof(ElfW(Sym)));
2301 #if defined(TCC_TARGET_ARM64) || defined(TCC_TARGET_X86_64)
2302 put_dt(dynamic, DT_RELA, dyninf->rel_addr);
2303 put_dt(dynamic, DT_RELASZ, dyninf->rel_size);
2304 put_dt(dynamic, DT_RELAENT, sizeof(ElfW_Rel));
2305 #else
2306 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
2307 put_dt(dynamic, DT_PLTGOT, s1->got->sh_addr);
2308 put_dt(dynamic, DT_PLTRELSZ, dyninf->rel_size);
2309 put_dt(dynamic, DT_JMPREL, dyninf->rel_addr);
2310 put_dt(dynamic, DT_PLTREL, DT_REL);
2311 put_dt(dynamic, DT_REL, dyninf->bss_addr);
2312 put_dt(dynamic, DT_RELSZ, dyninf->bss_size);
2313 #else
2314 put_dt(dynamic, DT_REL, dyninf->rel_addr);
2315 put_dt(dynamic, DT_RELSZ, dyninf->rel_size);
2316 put_dt(dynamic, DT_RELENT, sizeof(ElfW_Rel));
2317 #endif
2318 #endif
2319 if (s1->do_debug)
2320 put_dt(dynamic, DT_DEBUG, 0);
2321 put_dt(dynamic, DT_NULL, 0);
2324 /* Relocate remaining sections and symbols (that is those not related to
2325 dynamic linking) */
2326 static int final_sections_reloc(TCCState *s1)
2328 int i;
2329 Section *s;
2331 relocate_syms(s1, 0);
2333 if (s1->nb_errors != 0)
2334 return -1;
2336 /* relocate sections */
2337 /* XXX: ignore sections with allocated relocations ? */
2338 for(i = 1; i < s1->nb_sections; i++) {
2339 s = s1->sections[i];
2340 #ifdef TCC_TARGET_I386
2341 if (s->reloc && s != s1->got && (s->sh_flags & SHF_ALLOC)) //gr
2342 /* On X86 gdb 7.3 works in any case but gdb 6.6 will crash if SHF_ALLOC
2343 checking is removed */
2344 #else
2345 if (s->reloc && s != s1->got)
2346 /* On X86_64 gdb 7.3 will crash if SHF_ALLOC checking is present */
2347 #endif
2348 relocate_section(s1, s);
2351 /* relocate relocation entries if the relocation tables are
2352 allocated in the executable */
2353 for(i = 1; i < s1->nb_sections; i++) {
2354 s = s1->sections[i];
2355 if ((s->sh_flags & SHF_ALLOC) &&
2356 s->sh_type == SHT_RELX) {
2357 relocate_rel(s1, s);
2360 return 0;
2363 /* Create an ELF file on disk.
2364 This function handle ELF specific layout requirements */
2365 static void tcc_output_elf(TCCState *s1, FILE *f, int phnum, ElfW(Phdr) *phdr,
2366 int file_offset, int *sec_order)
2368 int i, shnum, offset, size, file_type;
2369 Section *s;
2370 ElfW(Ehdr) ehdr;
2371 ElfW(Shdr) shdr, *sh;
2373 file_type = s1->output_type;
2374 shnum = s1->nb_sections;
2376 memset(&ehdr, 0, sizeof(ehdr));
2378 if (phnum > 0) {
2379 ehdr.e_phentsize = sizeof(ElfW(Phdr));
2380 ehdr.e_phnum = phnum;
2381 ehdr.e_phoff = sizeof(ElfW(Ehdr));
2384 /* align to 4 */
2385 file_offset = (file_offset + 3) & -4;
2387 /* fill header */
2388 ehdr.e_ident[0] = ELFMAG0;
2389 ehdr.e_ident[1] = ELFMAG1;
2390 ehdr.e_ident[2] = ELFMAG2;
2391 ehdr.e_ident[3] = ELFMAG3;
2392 ehdr.e_ident[4] = ELFCLASSW;
2393 ehdr.e_ident[5] = ELFDATA2LSB;
2394 ehdr.e_ident[6] = EV_CURRENT;
2395 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
2396 ehdr.e_ident[EI_OSABI] = ELFOSABI_FREEBSD;
2397 #endif
2398 #ifdef TCC_TARGET_ARM
2399 #ifdef TCC_ARM_EABI
2400 ehdr.e_ident[EI_OSABI] = 0;
2401 ehdr.e_flags = EF_ARM_EABI_VER4;
2402 if (file_type == TCC_OUTPUT_EXE || file_type == TCC_OUTPUT_DLL)
2403 ehdr.e_flags |= EF_ARM_HASENTRY;
2404 if (s1->float_abi == ARM_HARD_FLOAT)
2405 ehdr.e_flags |= EF_ARM_VFP_FLOAT;
2406 else
2407 ehdr.e_flags |= EF_ARM_SOFT_FLOAT;
2408 #else
2409 ehdr.e_ident[EI_OSABI] = ELFOSABI_ARM;
2410 #endif
2411 #endif
2412 switch(file_type) {
2413 default:
2414 case TCC_OUTPUT_EXE:
2415 ehdr.e_type = ET_EXEC;
2416 ehdr.e_entry = get_elf_sym_addr(s1, "_start", 1);
2417 break;
2418 case TCC_OUTPUT_DLL:
2419 ehdr.e_type = ET_DYN;
2420 ehdr.e_entry = text_section->sh_addr; /* XXX: is it correct ? */
2421 break;
2422 case TCC_OUTPUT_OBJ:
2423 ehdr.e_type = ET_REL;
2424 break;
2426 ehdr.e_machine = EM_TCC_TARGET;
2427 ehdr.e_version = EV_CURRENT;
2428 ehdr.e_shoff = file_offset;
2429 ehdr.e_ehsize = sizeof(ElfW(Ehdr));
2430 ehdr.e_shentsize = sizeof(ElfW(Shdr));
2431 ehdr.e_shnum = shnum;
2432 ehdr.e_shstrndx = shnum - 1;
2434 fwrite(&ehdr, 1, sizeof(ElfW(Ehdr)), f);
2435 fwrite(phdr, 1, phnum * sizeof(ElfW(Phdr)), f);
2436 offset = sizeof(ElfW(Ehdr)) + phnum * sizeof(ElfW(Phdr));
2438 sort_syms(s1, symtab_section);
2439 for(i = 1; i < s1->nb_sections; i++) {
2440 s = s1->sections[sec_order[i]];
2441 if (s->sh_type != SHT_NOBITS) {
2442 if (s->sh_type == SHT_DYNSYM)
2443 patch_dynsym_undef(s1, s);
2444 while (offset < s->sh_offset) {
2445 fputc(0, f);
2446 offset++;
2448 size = s->sh_size;
2449 if (size)
2450 fwrite(s->data, 1, size, f);
2451 offset += size;
2455 /* output section headers */
2456 while (offset < ehdr.e_shoff) {
2457 fputc(0, f);
2458 offset++;
2461 for(i = 0; i < s1->nb_sections; i++) {
2462 sh = &shdr;
2463 memset(sh, 0, sizeof(ElfW(Shdr)));
2464 s = s1->sections[i];
2465 if (s) {
2466 sh->sh_name = s->sh_name;
2467 sh->sh_type = s->sh_type;
2468 sh->sh_flags = s->sh_flags;
2469 sh->sh_entsize = s->sh_entsize;
2470 sh->sh_info = s->sh_info;
2471 if (s->link)
2472 sh->sh_link = s->link->sh_num;
2473 sh->sh_addralign = s->sh_addralign;
2474 sh->sh_addr = s->sh_addr;
2475 sh->sh_offset = s->sh_offset;
2476 sh->sh_size = s->sh_size;
2478 fwrite(sh, 1, sizeof(ElfW(Shdr)), f);
2482 /* Write an elf, coff or "binary" file */
2483 static int tcc_write_elf_file(TCCState *s1, const char *filename, int phnum,
2484 ElfW(Phdr) *phdr, int file_offset, int *sec_order)
2486 int fd, mode, file_type;
2487 FILE *f;
2489 file_type = s1->output_type;
2490 if (file_type == TCC_OUTPUT_OBJ)
2491 mode = 0666;
2492 else
2493 mode = 0777;
2494 unlink(filename);
2495 fd = open(filename, O_WRONLY | O_CREAT | O_TRUNC | O_BINARY, mode);
2496 if (fd < 0) {
2497 tcc_error_noabort("could not write '%s'", filename);
2498 return -1;
2500 f = fdopen(fd, "wb");
2501 if (s1->verbose)
2502 printf("<- %s\n", filename);
2504 #ifdef TCC_TARGET_COFF
2505 if (s1->output_format == TCC_OUTPUT_FORMAT_COFF)
2506 tcc_output_coff(s1, f);
2507 else
2508 #endif
2509 if (s1->output_format == TCC_OUTPUT_FORMAT_ELF)
2510 tcc_output_elf(s1, f, phnum, phdr, file_offset, sec_order);
2511 else
2512 tcc_output_binary(s1, f, sec_order);
2513 fclose(f);
2515 return 0;
2518 /* Output an elf, coff or binary file */
2519 /* XXX: suppress unneeded sections */
2520 static int elf_output_file(TCCState *s1, const char *filename)
2522 int i, ret, phnum, shnum, file_type, file_offset, *sec_order;
2523 struct dyn_inf dyninf;
2524 ElfW(Phdr) *phdr;
2525 ElfW(Sym) *sym;
2526 Section *strsec, *interp, *dynamic, *dynstr;
2528 file_type = s1->output_type;
2529 s1->nb_errors = 0;
2531 /* if linking, also link in runtime libraries (libc, libgcc, etc.) */
2532 if (file_type != TCC_OUTPUT_OBJ) {
2533 tcc_add_runtime(s1);
2536 phdr = NULL;
2537 sec_order = NULL;
2538 interp = dynamic = dynstr = NULL; /* avoid warning */
2539 dyninf.dyn_rel_off = 0; /* avoid warning */
2541 if (file_type != TCC_OUTPUT_OBJ) {
2542 relocate_common_syms();
2544 tcc_add_linker_symbols(s1);
2546 if (!s1->static_link) {
2547 if (file_type == TCC_OUTPUT_EXE) {
2548 char *ptr;
2549 /* allow override the dynamic loader */
2550 const char *elfint = getenv("LD_SO");
2551 if (elfint == NULL)
2552 elfint = DEFAULT_ELFINTERP(s1);
2553 /* add interpreter section only if executable */
2554 interp = new_section(s1, ".interp", SHT_PROGBITS, SHF_ALLOC);
2555 interp->sh_addralign = 1;
2556 ptr = section_ptr_add(interp, 1 + strlen(elfint));
2557 strcpy(ptr, elfint);
2560 /* add dynamic symbol table */
2561 s1->dynsym = new_symtab(s1, ".dynsym", SHT_DYNSYM, SHF_ALLOC,
2562 ".dynstr",
2563 ".hash", SHF_ALLOC);
2564 dynstr = s1->dynsym->link;
2566 /* add dynamic section */
2567 dynamic = new_section(s1, ".dynamic", SHT_DYNAMIC,
2568 SHF_ALLOC | SHF_WRITE);
2569 dynamic->link = dynstr;
2570 dynamic->sh_entsize = sizeof(ElfW(Dyn));
2572 build_got(s1);
2574 if (file_type == TCC_OUTPUT_EXE) {
2575 bind_exe_dynsyms(s1);
2577 if (s1->nb_errors) {
2578 ret = -1;
2579 goto the_end;
2582 bind_libs_dynsyms(s1);
2583 } else /* shared library case: simply export all global symbols */
2584 export_global_syms(s1);
2586 build_got_entries(s1);
2588 /* add a list of needed dlls */
2589 for(i = 0; i < s1->nb_loaded_dlls; i++) {
2590 DLLReference *dllref = s1->loaded_dlls[i];
2591 if (dllref->level == 0)
2592 put_dt(dynamic, DT_NEEDED, put_elf_str(dynstr, dllref->name));
2595 if (s1->rpath)
2596 put_dt(dynamic, DT_RPATH, put_elf_str(dynstr, s1->rpath));
2598 /* XXX: currently, since we do not handle PIC code, we
2599 must relocate the readonly segments */
2600 if (file_type == TCC_OUTPUT_DLL) {
2601 if (s1->soname)
2602 put_dt(dynamic, DT_SONAME, put_elf_str(dynstr, s1->soname));
2603 put_dt(dynamic, DT_TEXTREL, 0);
2606 if (s1->symbolic)
2607 put_dt(dynamic, DT_SYMBOLIC, 0);
2609 /* add necessary space for other entries */
2610 dyninf.dyn_rel_off = dynamic->data_offset;
2611 dynamic->data_offset += sizeof(ElfW(Dyn)) * EXTRA_RELITEMS;
2612 } else {
2613 /* still need to build got entries in case of static link */
2614 build_got_entries(s1);
2618 /* we add a section for symbols */
2619 strsec = new_section(s1, ".shstrtab", SHT_STRTAB, 0);
2620 put_elf_str(strsec, "");
2622 /* compute number of sections */
2623 shnum = s1->nb_sections;
2625 /* this array is used to reorder sections in the output file */
2626 sec_order = tcc_malloc(sizeof(int) * shnum);
2627 sec_order[0] = 0;
2629 /* compute number of program headers */
2630 switch(file_type) {
2631 default:
2632 case TCC_OUTPUT_OBJ:
2633 phnum = 0;
2634 break;
2635 case TCC_OUTPUT_EXE:
2636 if (!s1->static_link)
2637 phnum = 4 + HAVE_PHDR;
2638 else
2639 phnum = 2;
2640 break;
2641 case TCC_OUTPUT_DLL:
2642 phnum = 3;
2643 break;
2646 /* Allocate strings for section names */
2647 alloc_sec_names(s1, file_type, strsec);
2649 /* allocate program segment headers */
2650 phdr = tcc_mallocz(phnum * sizeof(ElfW(Phdr)));
2652 /* compute section to program header mapping */
2653 file_offset = layout_sections(s1, phdr, phnum, interp, strsec, &dyninf,
2654 sec_order);
2656 /* Fill remaining program header and finalize relocation related to dynamic
2657 linking. */
2658 if (phnum > 0) {
2659 fill_unloadable_phdr(phdr, phnum, interp, dynamic);
2660 if (dynamic) {
2661 dyninf.dynamic = dynamic;
2662 dyninf.dynstr = dynstr;
2664 fill_dynamic(s1, &dyninf);
2666 /* put in GOT the dynamic section address and relocate PLT */
2667 write32le(s1->got->data, dynamic->sh_addr);
2668 if (file_type == TCC_OUTPUT_EXE
2669 #if defined(TCC_OUTPUT_DLL_WITH_PLT)
2670 || file_type == TCC_OUTPUT_DLL
2671 #endif
2673 relocate_plt(s1);
2675 /* relocate symbols in .dynsym now that final addresses are known */
2676 for_each_elem(s1->dynsym, 1, sym, ElfW(Sym)) {
2677 if (sym->st_shndx == SHN_UNDEF) {
2678 /* relocate to PLT if symbol corresponds to a PLT entry,
2679 but not if it's a weak symbol */
2680 if (ELFW(ST_BIND)(sym->st_info) == STB_WEAK)
2681 sym->st_value = 0;
2682 else if (sym->st_value)
2683 sym->st_value += s1->plt->sh_addr;
2684 } else if (sym->st_shndx < SHN_LORESERVE) {
2685 /* do symbol relocation */
2686 sym->st_value += s1->sections[sym->st_shndx]->sh_addr;
2692 /* if building executable or DLL, then relocate each section
2693 except the GOT which is already relocated */
2694 if (file_type != TCC_OUTPUT_OBJ) {
2695 ret = final_sections_reloc(s1);
2696 if (ret)
2697 goto the_end;
2700 /* Perform relocation to GOT or PLT entries */
2701 if (file_type == TCC_OUTPUT_EXE && s1->static_link)
2702 fill_got(s1);
2704 /* Create the ELF file with name 'filename' */
2705 ret = tcc_write_elf_file(s1, filename, phnum, phdr, file_offset, sec_order);
2706 if (s1->do_strip) {
2707 int rc;
2708 const char *strip_cmd = "sstrip "; // super strip utility from ELFkickers
2709 const char *null_dev = " 2> /dev/null";
2710 char buf[1050];
2711 snprintf(buf, sizeof(buf), "%s%s%s", strip_cmd, filename, null_dev);
2712 rc = system(buf);
2713 if (rc)
2714 system(buf+1); // call a strip utility from binutils
2716 the_end:
2717 tcc_free(s1->symtab_to_dynsym);
2718 tcc_free(sec_order);
2719 tcc_free(phdr);
2720 tcc_free(s1->sym_attrs);
2721 s1->sym_attrs = NULL;
2722 return ret;
2725 LIBTCCAPI int tcc_output_file(TCCState *s, const char *filename)
2727 int ret;
2728 #ifdef TCC_TARGET_PE
2729 if (s->output_type != TCC_OUTPUT_OBJ) {
2730 ret = pe_output_file(s, filename);
2731 } else
2732 #endif
2733 ret = elf_output_file(s, filename);
2734 return ret;
2737 static void *load_data(int fd, unsigned long file_offset, unsigned long size)
2739 void *data;
2741 data = tcc_malloc(size);
2742 lseek(fd, file_offset, SEEK_SET);
2743 read(fd, data, size);
2744 return data;
2747 typedef struct SectionMergeInfo {
2748 Section *s; /* corresponding existing section */
2749 unsigned long offset; /* offset of the new section in the existing section */
2750 uint8_t new_section; /* true if section 's' was added */
2751 uint8_t link_once; /* true if link once section */
2752 } SectionMergeInfo;
2754 /* load an object file and merge it with current files */
2755 /* XXX: handle correctly stab (debug) info */
2756 ST_FUNC int tcc_load_object_file(TCCState *s1,
2757 int fd, unsigned long file_offset)
2759 ElfW(Ehdr) ehdr;
2760 ElfW(Shdr) *shdr, *sh;
2761 int size, i, j, offset, offseti, nb_syms, sym_index, ret;
2762 unsigned char *strsec, *strtab;
2763 int *old_to_new_syms;
2764 char *sh_name, *name;
2765 SectionMergeInfo *sm_table, *sm;
2766 ElfW(Sym) *sym, *symtab;
2767 ElfW_Rel *rel;
2768 Section *s;
2770 int stab_index;
2771 int stabstr_index;
2773 stab_index = stabstr_index = 0;
2775 if (read(fd, &ehdr, sizeof(ehdr)) != sizeof(ehdr))
2776 goto fail1;
2777 if (ehdr.e_ident[0] != ELFMAG0 ||
2778 ehdr.e_ident[1] != ELFMAG1 ||
2779 ehdr.e_ident[2] != ELFMAG2 ||
2780 ehdr.e_ident[3] != ELFMAG3)
2781 goto fail1;
2782 /* test if object file */
2783 if (ehdr.e_type != ET_REL)
2784 goto fail1;
2785 /* test CPU specific stuff */
2786 if (ehdr.e_ident[5] != ELFDATA2LSB ||
2787 ehdr.e_machine != EM_TCC_TARGET) {
2788 fail1:
2789 tcc_error_noabort("invalid object file");
2790 return -1;
2792 /* read sections */
2793 shdr = load_data(fd, file_offset + ehdr.e_shoff,
2794 sizeof(ElfW(Shdr)) * ehdr.e_shnum);
2795 sm_table = tcc_mallocz(sizeof(SectionMergeInfo) * ehdr.e_shnum);
2797 /* load section names */
2798 sh = &shdr[ehdr.e_shstrndx];
2799 strsec = load_data(fd, file_offset + sh->sh_offset, sh->sh_size);
2801 /* load symtab and strtab */
2802 old_to_new_syms = NULL;
2803 symtab = NULL;
2804 strtab = NULL;
2805 nb_syms = 0;
2806 for(i = 1; i < ehdr.e_shnum; i++) {
2807 sh = &shdr[i];
2808 if (sh->sh_type == SHT_SYMTAB) {
2809 if (symtab) {
2810 tcc_error_noabort("object must contain only one symtab");
2811 fail:
2812 ret = -1;
2813 goto the_end;
2815 nb_syms = sh->sh_size / sizeof(ElfW(Sym));
2816 symtab = load_data(fd, file_offset + sh->sh_offset, sh->sh_size);
2817 sm_table[i].s = symtab_section;
2819 /* now load strtab */
2820 sh = &shdr[sh->sh_link];
2821 strtab = load_data(fd, file_offset + sh->sh_offset, sh->sh_size);
2825 /* now examine each section and try to merge its content with the
2826 ones in memory */
2827 for(i = 1; i < ehdr.e_shnum; i++) {
2828 /* no need to examine section name strtab */
2829 if (i == ehdr.e_shstrndx)
2830 continue;
2831 sh = &shdr[i];
2832 sh_name = (char *) strsec + sh->sh_name;
2833 /* ignore sections types we do not handle */
2834 if (sh->sh_type != SHT_PROGBITS &&
2835 sh->sh_type != SHT_RELX &&
2836 #ifdef TCC_ARM_EABI
2837 sh->sh_type != SHT_ARM_EXIDX &&
2838 #endif
2839 sh->sh_type != SHT_NOBITS &&
2840 sh->sh_type != SHT_PREINIT_ARRAY &&
2841 sh->sh_type != SHT_INIT_ARRAY &&
2842 sh->sh_type != SHT_FINI_ARRAY &&
2843 strcmp(sh_name, ".stabstr")
2845 continue;
2846 if (sh->sh_addralign < 1)
2847 sh->sh_addralign = 1;
2848 /* find corresponding section, if any */
2849 for(j = 1; j < s1->nb_sections;j++) {
2850 s = s1->sections[j];
2851 if (!strcmp(s->name, sh_name)) {
2852 if (!strncmp(sh_name, ".gnu.linkonce",
2853 sizeof(".gnu.linkonce") - 1)) {
2854 /* if a 'linkonce' section is already present, we
2855 do not add it again. It is a little tricky as
2856 symbols can still be defined in
2857 it. */
2858 sm_table[i].link_once = 1;
2859 goto next;
2860 } else {
2861 goto found;
2865 /* not found: create new section */
2866 s = new_section(s1, sh_name, sh->sh_type, sh->sh_flags);
2867 /* take as much info as possible from the section. sh_link and
2868 sh_info will be updated later */
2869 s->sh_addralign = sh->sh_addralign;
2870 s->sh_entsize = sh->sh_entsize;
2871 sm_table[i].new_section = 1;
2872 found:
2873 if (sh->sh_type != s->sh_type) {
2874 tcc_error_noabort("invalid section type");
2875 goto fail;
2878 /* align start of section */
2879 offset = s->data_offset;
2881 if (0 == strcmp(sh_name, ".stab")) {
2882 stab_index = i;
2883 goto no_align;
2885 if (0 == strcmp(sh_name, ".stabstr")) {
2886 stabstr_index = i;
2887 goto no_align;
2890 size = sh->sh_addralign - 1;
2891 offset = (offset + size) & ~size;
2892 if (sh->sh_addralign > s->sh_addralign)
2893 s->sh_addralign = sh->sh_addralign;
2894 s->data_offset = offset;
2895 no_align:
2896 sm_table[i].offset = offset;
2897 sm_table[i].s = s;
2898 /* concatenate sections */
2899 size = sh->sh_size;
2900 if (sh->sh_type != SHT_NOBITS) {
2901 unsigned char *ptr;
2902 lseek(fd, file_offset + sh->sh_offset, SEEK_SET);
2903 ptr = section_ptr_add(s, size);
2904 read(fd, ptr, size);
2905 } else {
2906 s->data_offset += size;
2908 next: ;
2911 /* gr relocate stab strings */
2912 if (stab_index && stabstr_index) {
2913 Stab_Sym *a, *b;
2914 unsigned o;
2915 s = sm_table[stab_index].s;
2916 a = (Stab_Sym *)(s->data + sm_table[stab_index].offset);
2917 b = (Stab_Sym *)(s->data + s->data_offset);
2918 o = sm_table[stabstr_index].offset;
2919 while (a < b)
2920 a->n_strx += o, a++;
2923 /* second short pass to update sh_link and sh_info fields of new
2924 sections */
2925 for(i = 1; i < ehdr.e_shnum; i++) {
2926 s = sm_table[i].s;
2927 if (!s || !sm_table[i].new_section)
2928 continue;
2929 sh = &shdr[i];
2930 if (sh->sh_link > 0)
2931 s->link = sm_table[sh->sh_link].s;
2932 if (sh->sh_type == SHT_RELX) {
2933 s->sh_info = sm_table[sh->sh_info].s->sh_num;
2934 /* update backward link */
2935 s1->sections[s->sh_info]->reloc = s;
2938 sm = sm_table;
2940 /* resolve symbols */
2941 old_to_new_syms = tcc_mallocz(nb_syms * sizeof(int));
2943 sym = symtab + 1;
2944 for(i = 1; i < nb_syms; i++, sym++) {
2945 if (sym->st_shndx != SHN_UNDEF &&
2946 sym->st_shndx < SHN_LORESERVE) {
2947 sm = &sm_table[sym->st_shndx];
2948 if (sm->link_once) {
2949 /* if a symbol is in a link once section, we use the
2950 already defined symbol. It is very important to get
2951 correct relocations */
2952 if (ELFW(ST_BIND)(sym->st_info) != STB_LOCAL) {
2953 name = (char *) strtab + sym->st_name;
2954 sym_index = find_elf_sym(symtab_section, name);
2955 if (sym_index)
2956 old_to_new_syms[i] = sym_index;
2958 continue;
2960 /* if no corresponding section added, no need to add symbol */
2961 if (!sm->s)
2962 continue;
2963 /* convert section number */
2964 sym->st_shndx = sm->s->sh_num;
2965 /* offset value */
2966 sym->st_value += sm->offset;
2968 /* add symbol */
2969 name = (char *) strtab + sym->st_name;
2970 sym_index = add_elf_sym(symtab_section, sym->st_value, sym->st_size,
2971 sym->st_info, sym->st_other,
2972 sym->st_shndx, name);
2973 old_to_new_syms[i] = sym_index;
2976 /* third pass to patch relocation entries */
2977 for(i = 1; i < ehdr.e_shnum; i++) {
2978 s = sm_table[i].s;
2979 if (!s)
2980 continue;
2981 sh = &shdr[i];
2982 offset = sm_table[i].offset;
2983 switch(s->sh_type) {
2984 case SHT_RELX:
2985 /* take relocation offset information */
2986 offseti = sm_table[sh->sh_info].offset;
2987 for_each_elem(s, (offset / sizeof(*rel)), rel, ElfW_Rel) {
2988 int type;
2989 unsigned sym_index;
2990 /* convert symbol index */
2991 type = ELFW(R_TYPE)(rel->r_info);
2992 sym_index = ELFW(R_SYM)(rel->r_info);
2993 /* NOTE: only one symtab assumed */
2994 if (sym_index >= nb_syms)
2995 goto invalid_reloc;
2996 sym_index = old_to_new_syms[sym_index];
2997 /* ignore link_once in rel section. */
2998 if (!sym_index && !sm->link_once
2999 #ifdef TCC_TARGET_ARM
3000 && type != R_ARM_V4BX
3001 #endif
3003 invalid_reloc:
3004 tcc_error_noabort("Invalid relocation entry [%2d] '%s' @ %.8x",
3005 i, strsec + sh->sh_name, rel->r_offset);
3006 goto fail;
3008 rel->r_info = ELFW(R_INFO)(sym_index, type);
3009 /* offset the relocation offset */
3010 rel->r_offset += offseti;
3011 #ifdef TCC_TARGET_ARM
3012 /* Jumps and branches from a Thumb code to a PLT entry need
3013 special handling since PLT entries are ARM code.
3014 Unconditional bl instructions referencing PLT entries are
3015 handled by converting these instructions into blx
3016 instructions. Other case of instructions referencing a PLT
3017 entry require to add a Thumb stub before the PLT entry to
3018 switch to ARM mode. We set bit plt_thumb_stub of the
3019 attribute of a symbol to indicate such a case. */
3020 if (type == R_ARM_THM_JUMP24)
3021 alloc_sym_attr(s1, sym_index)->plt_thumb_stub = 1;
3022 #endif
3024 break;
3025 default:
3026 break;
3030 ret = 0;
3031 the_end:
3032 tcc_free(symtab);
3033 tcc_free(strtab);
3034 tcc_free(old_to_new_syms);
3035 tcc_free(sm_table);
3036 tcc_free(strsec);
3037 tcc_free(shdr);
3038 return ret;
3041 typedef struct ArchiveHeader {
3042 char ar_name[16]; /* name of this member */
3043 char ar_date[12]; /* file mtime */
3044 char ar_uid[6]; /* owner uid; printed as decimal */
3045 char ar_gid[6]; /* owner gid; printed as decimal */
3046 char ar_mode[8]; /* file mode, printed as octal */
3047 char ar_size[10]; /* file size, printed as decimal */
3048 char ar_fmag[2]; /* should contain ARFMAG */
3049 } ArchiveHeader;
3051 static int get_be32(const uint8_t *b)
3053 return b[3] | (b[2] << 8) | (b[1] << 16) | (b[0] << 24);
3056 /* load only the objects which resolve undefined symbols */
3057 static int tcc_load_alacarte(TCCState *s1, int fd, int size)
3059 int i, bound, nsyms, sym_index, off, ret;
3060 uint8_t *data;
3061 const char *ar_names, *p;
3062 const uint8_t *ar_index;
3063 ElfW(Sym) *sym;
3065 data = tcc_malloc(size);
3066 if (read(fd, data, size) != size)
3067 goto fail;
3068 nsyms = get_be32(data);
3069 ar_index = data + 4;
3070 ar_names = (char *) ar_index + nsyms * 4;
3072 do {
3073 bound = 0;
3074 for(p = ar_names, i = 0; i < nsyms; i++, p += strlen(p)+1) {
3075 sym_index = find_elf_sym(symtab_section, p);
3076 if(sym_index) {
3077 sym = &((ElfW(Sym) *)symtab_section->data)[sym_index];
3078 if(sym->st_shndx == SHN_UNDEF) {
3079 load_obj:
3080 off = get_be32(ar_index + i * 4) + sizeof(ArchiveHeader);
3081 ++bound;
3082 lseek(fd, off, SEEK_SET);
3083 if(tcc_load_object_file(s1, fd, off) < 0) {
3084 fail:
3085 ret = -1;
3086 goto the_end;
3089 } else if (s1->whole_archive) {
3090 goto load_obj;
3093 } while(bound);
3094 ret = 0;
3095 the_end:
3096 tcc_free(data);
3097 return ret;
3100 /* load a '.a' file */
3101 ST_FUNC int tcc_load_archive(TCCState *s1, int fd)
3103 ArchiveHeader hdr;
3104 char ar_size[11];
3105 char ar_name[17];
3106 char magic[8];
3107 int size, len, i;
3108 unsigned long file_offset;
3110 /* skip magic which was already checked */
3111 read(fd, magic, sizeof(magic));
3113 for(;;) {
3114 len = read(fd, &hdr, sizeof(hdr));
3115 if (len == 0)
3116 break;
3117 if (len != sizeof(hdr)) {
3118 tcc_error_noabort("invalid archive");
3119 return -1;
3121 memcpy(ar_size, hdr.ar_size, sizeof(hdr.ar_size));
3122 ar_size[sizeof(hdr.ar_size)] = '\0';
3123 size = strtol(ar_size, NULL, 0);
3124 memcpy(ar_name, hdr.ar_name, sizeof(hdr.ar_name));
3125 for(i = sizeof(hdr.ar_name) - 1; i >= 0; i--) {
3126 if (ar_name[i] != ' ')
3127 break;
3129 ar_name[i + 1] = '\0';
3130 file_offset = lseek(fd, 0, SEEK_CUR);
3131 /* align to even */
3132 size = (size + 1) & ~1;
3133 if (!strcmp(ar_name, "/")) {
3134 /* coff symbol table : we handle it */
3135 if(s1->alacarte_link)
3136 return tcc_load_alacarte(s1, fd, size);
3137 } else if (!strcmp(ar_name, "//") ||
3138 !strcmp(ar_name, "__.SYMDEF") ||
3139 !strcmp(ar_name, "__.SYMDEF/") ||
3140 !strcmp(ar_name, "ARFILENAMES/")) {
3141 /* skip symbol table or archive names */
3142 } else {
3143 if (tcc_load_object_file(s1, fd, file_offset) < 0)
3144 return -1;
3146 lseek(fd, file_offset + size, SEEK_SET);
3148 return 0;
3151 #ifndef TCC_TARGET_PE
3152 /* load a DLL and all referenced DLLs. 'level = 0' means that the DLL
3153 is referenced by the user (so it should be added as DT_NEEDED in
3154 the generated ELF file) */
3155 ST_FUNC int tcc_load_dll(TCCState *s1, int fd, const char *filename, int level)
3157 ElfW(Ehdr) ehdr;
3158 ElfW(Shdr) *shdr, *sh, *sh1;
3159 int i, j, nb_syms, nb_dts, sym_bind, ret;
3160 ElfW(Sym) *sym, *dynsym;
3161 ElfW(Dyn) *dt, *dynamic;
3162 unsigned char *dynstr;
3163 const char *name, *soname;
3164 DLLReference *dllref;
3166 read(fd, &ehdr, sizeof(ehdr));
3168 /* test CPU specific stuff */
3169 if (ehdr.e_ident[5] != ELFDATA2LSB ||
3170 ehdr.e_machine != EM_TCC_TARGET) {
3171 tcc_error_noabort("bad architecture");
3172 return -1;
3175 /* read sections */
3176 shdr = load_data(fd, ehdr.e_shoff, sizeof(ElfW(Shdr)) * ehdr.e_shnum);
3178 /* load dynamic section and dynamic symbols */
3179 nb_syms = 0;
3180 nb_dts = 0;
3181 dynamic = NULL;
3182 dynsym = NULL; /* avoid warning */
3183 dynstr = NULL; /* avoid warning */
3184 for(i = 0, sh = shdr; i < ehdr.e_shnum; i++, sh++) {
3185 switch(sh->sh_type) {
3186 case SHT_DYNAMIC:
3187 nb_dts = sh->sh_size / sizeof(ElfW(Dyn));
3188 dynamic = load_data(fd, sh->sh_offset, sh->sh_size);
3189 break;
3190 case SHT_DYNSYM:
3191 nb_syms = sh->sh_size / sizeof(ElfW(Sym));
3192 dynsym = load_data(fd, sh->sh_offset, sh->sh_size);
3193 sh1 = &shdr[sh->sh_link];
3194 dynstr = load_data(fd, sh1->sh_offset, sh1->sh_size);
3195 break;
3196 default:
3197 break;
3201 /* compute the real library name */
3202 soname = tcc_basename(filename);
3204 for(i = 0, dt = dynamic; i < nb_dts; i++, dt++) {
3205 if (dt->d_tag == DT_SONAME) {
3206 soname = (char *) dynstr + dt->d_un.d_val;
3210 /* if the dll is already loaded, do not load it */
3211 for(i = 0; i < s1->nb_loaded_dlls; i++) {
3212 dllref = s1->loaded_dlls[i];
3213 if (!strcmp(soname, dllref->name)) {
3214 /* but update level if needed */
3215 if (level < dllref->level)
3216 dllref->level = level;
3217 ret = 0;
3218 goto the_end;
3222 /* add the dll and its level */
3223 dllref = tcc_mallocz(sizeof(DLLReference) + strlen(soname));
3224 dllref->level = level;
3225 strcpy(dllref->name, soname);
3226 dynarray_add((void ***)&s1->loaded_dlls, &s1->nb_loaded_dlls, dllref);
3228 /* add dynamic symbols in dynsym_section */
3229 for(i = 1, sym = dynsym + 1; i < nb_syms; i++, sym++) {
3230 sym_bind = ELFW(ST_BIND)(sym->st_info);
3231 if (sym_bind == STB_LOCAL)
3232 continue;
3233 name = (char *) dynstr + sym->st_name;
3234 add_elf_sym(s1->dynsymtab_section, sym->st_value, sym->st_size,
3235 sym->st_info, sym->st_other, sym->st_shndx, name);
3238 /* load all referenced DLLs */
3239 for(i = 0, dt = dynamic; i < nb_dts; i++, dt++) {
3240 switch(dt->d_tag) {
3241 case DT_NEEDED:
3242 name = (char *) dynstr + dt->d_un.d_val;
3243 for(j = 0; j < s1->nb_loaded_dlls; j++) {
3244 dllref = s1->loaded_dlls[j];
3245 if (!strcmp(name, dllref->name))
3246 goto already_loaded;
3248 if (tcc_add_dll(s1, name, AFF_REFERENCED_DLL) < 0) {
3249 tcc_error_noabort("referenced dll '%s' not found", name);
3250 ret = -1;
3251 goto the_end;
3253 already_loaded:
3254 break;
3257 ret = 0;
3258 the_end:
3259 tcc_free(dynstr);
3260 tcc_free(dynsym);
3261 tcc_free(dynamic);
3262 tcc_free(shdr);
3263 return ret;
3266 #define LD_TOK_NAME 256
3267 #define LD_TOK_EOF (-1)
3269 /* return next ld script token */
3270 static int ld_next(TCCState *s1, char *name, int name_size)
3272 int c;
3273 char *q;
3275 redo:
3276 switch(ch) {
3277 case ' ':
3278 case '\t':
3279 case '\f':
3280 case '\v':
3281 case '\r':
3282 case '\n':
3283 inp();
3284 goto redo;
3285 case '/':
3286 minp();
3287 if (ch == '*') {
3288 file->buf_ptr = parse_comment(file->buf_ptr);
3289 ch = file->buf_ptr[0];
3290 goto redo;
3291 } else {
3292 q = name;
3293 *q++ = '/';
3294 goto parse_name;
3296 break;
3297 case '\\':
3298 ch = handle_eob();
3299 if (ch != '\\')
3300 goto redo;
3301 /* fall through */
3302 /* case 'a' ... 'z': */
3303 case 'a':
3304 case 'b':
3305 case 'c':
3306 case 'd':
3307 case 'e':
3308 case 'f':
3309 case 'g':
3310 case 'h':
3311 case 'i':
3312 case 'j':
3313 case 'k':
3314 case 'l':
3315 case 'm':
3316 case 'n':
3317 case 'o':
3318 case 'p':
3319 case 'q':
3320 case 'r':
3321 case 's':
3322 case 't':
3323 case 'u':
3324 case 'v':
3325 case 'w':
3326 case 'x':
3327 case 'y':
3328 case 'z':
3329 /* case 'A' ... 'z': */
3330 case 'A':
3331 case 'B':
3332 case 'C':
3333 case 'D':
3334 case 'E':
3335 case 'F':
3336 case 'G':
3337 case 'H':
3338 case 'I':
3339 case 'J':
3340 case 'K':
3341 case 'L':
3342 case 'M':
3343 case 'N':
3344 case 'O':
3345 case 'P':
3346 case 'Q':
3347 case 'R':
3348 case 'S':
3349 case 'T':
3350 case 'U':
3351 case 'V':
3352 case 'W':
3353 case 'X':
3354 case 'Y':
3355 case 'Z':
3356 case '_':
3357 case '.':
3358 case '$':
3359 case '~':
3360 q = name;
3361 parse_name:
3362 for(;;) {
3363 if (!((ch >= 'a' && ch <= 'z') ||
3364 (ch >= 'A' && ch <= 'Z') ||
3365 (ch >= '0' && ch <= '9') ||
3366 strchr("/.-_+=$:\\,~", ch)))
3367 break;
3368 if ((q - name) < name_size - 1) {
3369 *q++ = ch;
3371 minp();
3373 *q = '\0';
3374 c = LD_TOK_NAME;
3375 break;
3376 case CH_EOF:
3377 c = LD_TOK_EOF;
3378 break;
3379 default:
3380 c = ch;
3381 inp();
3382 break;
3384 return c;
3387 static int ld_add_file(TCCState *s1, const char filename[])
3389 int ret;
3391 ret = tcc_add_file_internal(s1, filename, 0, TCC_FILETYPE_BINARY);
3392 if (ret)
3393 ret = tcc_add_dll(s1, filename, 0);
3394 return ret;
3397 static inline int new_undef_syms(void)
3399 int ret = 0;
3400 ret = new_undef_sym;
3401 new_undef_sym = 0;
3402 return ret;
3405 static int ld_add_file_list(TCCState *s1, const char *cmd, int as_needed)
3407 char filename[1024], libname[1024];
3408 int t, group, nblibs = 0, ret = 0;
3409 char **libs = NULL;
3411 group = !strcmp(cmd, "GROUP");
3412 if (!as_needed)
3413 new_undef_syms();
3414 t = ld_next(s1, filename, sizeof(filename));
3415 if (t != '(')
3416 expect("(");
3417 t = ld_next(s1, filename, sizeof(filename));
3418 for(;;) {
3419 libname[0] = '\0';
3420 if (t == LD_TOK_EOF) {
3421 tcc_error_noabort("unexpected end of file");
3422 ret = -1;
3423 goto lib_parse_error;
3424 } else if (t == ')') {
3425 break;
3426 } else if (t == '-') {
3427 t = ld_next(s1, filename, sizeof(filename));
3428 if ((t != LD_TOK_NAME) || (filename[0] != 'l')) {
3429 tcc_error_noabort("library name expected");
3430 ret = -1;
3431 goto lib_parse_error;
3433 pstrcpy(libname, sizeof libname, &filename[1]);
3434 if (s1->static_link) {
3435 snprintf(filename, sizeof filename, "lib%s.a", libname);
3436 } else {
3437 snprintf(filename, sizeof filename, "lib%s.so", libname);
3439 } else if (t != LD_TOK_NAME) {
3440 tcc_error_noabort("filename expected");
3441 ret = -1;
3442 goto lib_parse_error;
3444 if (!strcmp(filename, "AS_NEEDED")) {
3445 ret = ld_add_file_list(s1, cmd, 1);
3446 if (ret)
3447 goto lib_parse_error;
3448 } else {
3449 /* TODO: Implement AS_NEEDED support. Ignore it for now */
3450 if (!as_needed) {
3451 ret = ld_add_file(s1, filename);
3452 if (ret)
3453 goto lib_parse_error;
3454 if (group) {
3455 /* Add the filename *and* the libname to avoid future conversions */
3456 dynarray_add((void ***) &libs, &nblibs, tcc_strdup(filename));
3457 if (libname[0] != '\0')
3458 dynarray_add((void ***) &libs, &nblibs, tcc_strdup(libname));
3462 t = ld_next(s1, filename, sizeof(filename));
3463 if (t == ',') {
3464 t = ld_next(s1, filename, sizeof(filename));
3467 if (group && !as_needed) {
3468 while (new_undef_syms()) {
3469 int i;
3471 for (i = 0; i < nblibs; i ++)
3472 ld_add_file(s1, libs[i]);
3475 lib_parse_error:
3476 dynarray_reset(&libs, &nblibs);
3477 return ret;
3480 /* interpret a subset of GNU ldscripts to handle the dummy libc.so
3481 files */
3482 ST_FUNC int tcc_load_ldscript(TCCState *s1)
3484 char cmd[64];
3485 char filename[1024];
3486 int t, ret;
3488 ch = handle_eob();
3489 for(;;) {
3490 t = ld_next(s1, cmd, sizeof(cmd));
3491 if (t == LD_TOK_EOF)
3492 return 0;
3493 else if (t != LD_TOK_NAME)
3494 return -1;
3495 if (!strcmp(cmd, "INPUT") ||
3496 !strcmp(cmd, "GROUP")) {
3497 ret = ld_add_file_list(s1, cmd, 0);
3498 if (ret)
3499 return ret;
3500 } else if (!strcmp(cmd, "OUTPUT_FORMAT") ||
3501 !strcmp(cmd, "TARGET")) {
3502 /* ignore some commands */
3503 t = ld_next(s1, cmd, sizeof(cmd));
3504 if (t != '(')
3505 expect("(");
3506 for(;;) {
3507 t = ld_next(s1, filename, sizeof(filename));
3508 if (t == LD_TOK_EOF) {
3509 tcc_error_noabort("unexpected end of file");
3510 return -1;
3511 } else if (t == ')') {
3512 break;
3515 } else {
3516 return -1;
3519 return 0;
3521 #endif /* !TCC_TARGET_PE */