The hack to allow valgrind works with tcc compiled programs
[tinycc.git] / tccelf.c
blobf5f9eed2e4b38110cdecfd1980244fae88d5a9a3
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 /* XXX: avoid static variable */
24 static int new_undef_sym = 0; /* Is there a new undefined sym since last new_undef_sym() */
26 ST_FUNC int put_elf_str(Section *s, const char *sym)
28 int offset, len;
29 char *ptr;
31 len = strlen(sym) + 1;
32 offset = s->data_offset;
33 ptr = section_ptr_add(s, len);
34 memcpy(ptr, sym, len);
35 return offset;
38 /* elf symbol hashing function */
39 static unsigned long elf_hash(const unsigned char *name)
41 unsigned long h = 0, g;
43 while (*name) {
44 h = (h << 4) + *name++;
45 g = h & 0xf0000000;
46 if (g)
47 h ^= g >> 24;
48 h &= ~g;
50 return h;
53 /* rebuild hash table of section s */
54 /* NOTE: we do factorize the hash table code to go faster */
55 static void rebuild_hash(Section *s, unsigned int nb_buckets)
57 ElfW(Sym) *sym;
58 int *ptr, *hash, nb_syms, sym_index, h;
59 unsigned char *strtab;
61 strtab = s->link->data;
62 nb_syms = s->data_offset / sizeof(ElfW(Sym));
64 s->hash->data_offset = 0;
65 ptr = section_ptr_add(s->hash, (2 + nb_buckets + nb_syms) * sizeof(int));
66 ptr[0] = nb_buckets;
67 ptr[1] = nb_syms;
68 ptr += 2;
69 hash = ptr;
70 memset(hash, 0, (nb_buckets + 1) * sizeof(int));
71 ptr += nb_buckets + 1;
73 sym = (ElfW(Sym) *)s->data + 1;
74 for(sym_index = 1; sym_index < nb_syms; sym_index++) {
75 if (ELFW(ST_BIND)(sym->st_info) != STB_LOCAL) {
76 h = elf_hash(strtab + sym->st_name) % nb_buckets;
77 *ptr = hash[h];
78 hash[h] = sym_index;
79 } else {
80 *ptr = 0;
82 ptr++;
83 sym++;
87 /* return the symbol number */
88 ST_FUNC int put_elf_sym(Section *s, addr_t value, unsigned long size,
89 int info, int other, int shndx, const char *name)
91 int name_offset, sym_index;
92 int nbuckets, h;
93 ElfW(Sym) *sym;
94 Section *hs;
96 sym = section_ptr_add(s, sizeof(ElfW(Sym)));
97 if (name)
98 name_offset = put_elf_str(s->link, name);
99 else
100 name_offset = 0;
101 /* XXX: endianness */
102 sym->st_name = name_offset;
103 sym->st_value = value;
104 sym->st_size = size;
105 sym->st_info = info;
106 sym->st_other = other;
107 sym->st_shndx = shndx;
108 sym_index = sym - (ElfW(Sym) *)s->data;
109 hs = s->hash;
110 if (hs) {
111 int *ptr, *base;
112 ptr = section_ptr_add(hs, sizeof(int));
113 base = (int *)hs->data;
114 /* only add global or weak symbols */
115 if (ELFW(ST_BIND)(info) != STB_LOCAL) {
116 /* add another hashing entry */
117 nbuckets = base[0];
118 h = elf_hash((unsigned char *) name) % nbuckets;
119 *ptr = base[2 + h];
120 base[2 + h] = sym_index;
121 base[1]++;
122 /* we resize the hash table */
123 hs->nb_hashed_syms++;
124 if (hs->nb_hashed_syms > 2 * nbuckets) {
125 rebuild_hash(s, 2 * nbuckets);
127 } else {
128 *ptr = 0;
129 base[1]++;
132 return sym_index;
135 /* find global ELF symbol 'name' and return its index. Return 0 if not
136 found. */
137 ST_FUNC int find_elf_sym(Section *s, const char *name)
139 ElfW(Sym) *sym;
140 Section *hs;
141 int nbuckets, sym_index, h;
142 const char *name1;
144 hs = s->hash;
145 if (!hs)
146 return 0;
147 nbuckets = ((int *)hs->data)[0];
148 h = elf_hash((unsigned char *) name) % nbuckets;
149 sym_index = ((int *)hs->data)[2 + h];
150 while (sym_index != 0) {
151 sym = &((ElfW(Sym) *)s->data)[sym_index];
152 name1 = (char *) s->link->data + sym->st_name;
153 if (!strcmp(name, name1))
154 return sym_index;
155 sym_index = ((int *)hs->data)[2 + nbuckets + sym_index];
157 return 0;
160 /* return elf symbol value, signal error if 'err' is nonzero */
161 ST_FUNC addr_t get_elf_sym_addr(TCCState *s, const char *name, int err)
163 int sym_index;
164 ElfW(Sym) *sym;
166 sym_index = find_elf_sym(s->symtab, name);
167 sym = &((ElfW(Sym) *)s->symtab->data)[sym_index];
168 if (!sym_index || sym->st_shndx == SHN_UNDEF) {
169 if (err)
170 tcc_error("%s not defined", name);
171 return 0;
173 return sym->st_value;
176 /* return elf symbol value */
177 LIBTCCAPI void *tcc_get_symbol(TCCState *s, const char *name)
179 return (void*)(uintptr_t)get_elf_sym_addr(s, name, 0);
182 #if defined TCC_IS_NATIVE || defined TCC_TARGET_PE
183 /* return elf symbol value or error */
184 ST_FUNC void* tcc_get_symbol_err(TCCState *s, const char *name)
186 return (void*)(uintptr_t)get_elf_sym_addr(s, name, 1);
188 #endif
190 /* add an elf symbol : check if it is already defined and patch
191 it. Return symbol index. NOTE that sh_num can be SHN_UNDEF. */
192 ST_FUNC int add_elf_sym(Section *s, addr_t value, unsigned long size,
193 int info, int other, int sh_num, const char *name)
195 ElfW(Sym) *esym;
196 int sym_bind, sym_index, sym_type, esym_bind;
197 unsigned char sym_vis, esym_vis, new_vis;
199 sym_bind = ELFW(ST_BIND)(info);
200 sym_type = ELFW(ST_TYPE)(info);
201 sym_vis = ELFW(ST_VISIBILITY)(other);
203 if (sym_bind != STB_LOCAL) {
204 /* we search global or weak symbols */
205 sym_index = find_elf_sym(s, name);
206 if (!sym_index)
207 goto do_def;
208 esym = &((ElfW(Sym) *)s->data)[sym_index];
209 if (esym->st_shndx != SHN_UNDEF) {
210 esym_bind = ELFW(ST_BIND)(esym->st_info);
211 /* propagate the most constraining visibility */
212 /* STV_DEFAULT(0)<STV_PROTECTED(3)<STV_HIDDEN(2)<STV_INTERNAL(1) */
213 esym_vis = ELFW(ST_VISIBILITY)(esym->st_other);
214 if (esym_vis == STV_DEFAULT) {
215 new_vis = sym_vis;
216 } else if (sym_vis == STV_DEFAULT) {
217 new_vis = esym_vis;
218 } else {
219 new_vis = (esym_vis < sym_vis) ? esym_vis : sym_vis;
221 esym->st_other = (esym->st_other & ~ELFW(ST_VISIBILITY)(-1))
222 | new_vis;
223 other = esym->st_other; /* in case we have to patch esym */
224 if (sh_num == SHN_UNDEF) {
225 /* ignore adding of undefined symbol if the
226 corresponding symbol is already defined */
227 } else if (sym_bind == STB_GLOBAL && esym_bind == STB_WEAK) {
228 /* global overrides weak, so patch */
229 goto do_patch;
230 } else if (sym_bind == STB_WEAK && esym_bind == STB_GLOBAL) {
231 /* weak is ignored if already global */
232 } else if (sym_bind == STB_WEAK && esym_bind == STB_WEAK) {
233 /* keep first-found weak definition, ignore subsequents */
234 } else if (sym_vis == STV_HIDDEN || sym_vis == STV_INTERNAL) {
235 /* ignore hidden symbols after */
236 } else if (esym->st_shndx == SHN_COMMON
237 && (sh_num < SHN_LORESERVE || sh_num == SHN_COMMON)) {
238 /* gr: Happens with 'tcc ... -static tcctest.c' on e.g. Ubuntu 6.01
239 No idea if this is the correct solution ... */
240 goto do_patch;
241 } else if (s == tcc_state->dynsymtab_section) {
242 /* we accept that two DLL define the same symbol */
243 } else {
244 #if 0
245 printf("new_bind=%x new_shndx=%x new_vis=%x old_bind=%x old_shndx=%x old_vis=%x\n",
246 sym_bind, sh_num, new_vis, esym_bind, esym->st_shndx, esym_vis);
247 #endif
248 tcc_error_noabort("'%s' defined twice", name);
250 } else {
251 do_patch:
252 esym->st_info = ELFW(ST_INFO)(sym_bind, sym_type);
253 esym->st_shndx = sh_num;
254 new_undef_sym = 1;
255 esym->st_value = value;
256 esym->st_size = size;
257 esym->st_other = other;
259 } else {
260 do_def:
261 sym_index = put_elf_sym(s, value, size,
262 ELFW(ST_INFO)(sym_bind, sym_type), other,
263 sh_num, name);
265 return sym_index;
268 /* put relocation */
269 ST_FUNC void put_elf_reloc(Section *symtab, Section *s, unsigned long offset,
270 int type, int symbol)
272 char buf[256];
273 Section *sr;
274 ElfW_Rel *rel;
276 sr = s->reloc;
277 if (!sr) {
278 /* if no relocation section, create it */
279 snprintf(buf, sizeof(buf), REL_SECTION_FMT, s->name);
280 /* if the symtab is allocated, then we consider the relocation
281 are also */
282 sr = new_section(tcc_state, buf, SHT_RELX, symtab->sh_flags);
283 sr->sh_entsize = sizeof(ElfW_Rel);
284 sr->link = symtab;
285 sr->sh_info = s->sh_num;
286 s->reloc = sr;
288 rel = section_ptr_add(sr, sizeof(ElfW_Rel));
289 rel->r_offset = offset;
290 rel->r_info = ELFW(R_INFO)(symbol, type);
291 #ifdef TCC_TARGET_X86_64
292 rel->r_addend = 0;
293 #endif
296 /* put stab debug information */
298 ST_FUNC void put_stabs(const char *str, int type, int other, int desc,
299 unsigned long value)
301 Stab_Sym *sym;
303 sym = section_ptr_add(stab_section, sizeof(Stab_Sym));
304 if (str) {
305 sym->n_strx = put_elf_str(stabstr_section, str);
306 } else {
307 sym->n_strx = 0;
309 sym->n_type = type;
310 sym->n_other = other;
311 sym->n_desc = desc;
312 sym->n_value = value;
315 ST_FUNC void put_stabs_r(const char *str, int type, int other, int desc,
316 unsigned long value, Section *sec, int sym_index)
318 put_stabs(str, type, other, desc, value);
319 put_elf_reloc(symtab_section, stab_section,
320 stab_section->data_offset - sizeof(unsigned int),
321 R_DATA_32, sym_index);
324 ST_FUNC void put_stabn(int type, int other, int desc, int value)
326 put_stabs(NULL, type, other, desc, value);
329 ST_FUNC void put_stabd(int type, int other, int desc)
331 put_stabs(NULL, type, other, desc, 0);
334 /* Browse each elem of type <type> in section <sec> starting at elem <startoff>
335 using variable <elem> */
336 #define for_each_elem(sec, startoff, elem, type) \
337 for (elem = (type *) sec->data + startoff; \
338 elem < (type *) (sec->data + sec->data_offset); elem++)
340 /* In an ELF file symbol table, the local symbols must appear below
341 the global and weak ones. Since TCC cannot sort it while generating
342 the code, we must do it after. All the relocation tables are also
343 modified to take into account the symbol table sorting */
344 static void sort_syms(TCCState *s1, Section *s)
346 int *old_to_new_syms;
347 ElfW(Sym) *new_syms;
348 int nb_syms, i;
349 ElfW(Sym) *p, *q;
350 ElfW_Rel *rel;
351 Section *sr;
352 int type, sym_index;
354 nb_syms = s->data_offset / sizeof(ElfW(Sym));
355 new_syms = tcc_malloc(nb_syms * sizeof(ElfW(Sym)));
356 old_to_new_syms = tcc_malloc(nb_syms * sizeof(int));
358 /* first pass for local symbols */
359 p = (ElfW(Sym) *)s->data;
360 q = new_syms;
361 for(i = 0; i < nb_syms; i++) {
362 if (ELFW(ST_BIND)(p->st_info) == STB_LOCAL) {
363 old_to_new_syms[i] = q - new_syms;
364 *q++ = *p;
366 p++;
368 /* save the number of local symbols in section header */
369 s->sh_info = q - new_syms;
371 /* then second pass for non local symbols */
372 p = (ElfW(Sym) *)s->data;
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++;
381 /* we copy the new symbols to the old */
382 memcpy(s->data, new_syms, nb_syms * sizeof(ElfW(Sym)));
383 tcc_free(new_syms);
385 /* now we modify all the relocations */
386 for(i = 1; i < s1->nb_sections; i++) {
387 sr = s1->sections[i];
388 if (sr->sh_type == SHT_RELX && sr->link == s) {
389 for_each_elem(sr, 0, rel, ElfW_Rel) {
390 sym_index = ELFW(R_SYM)(rel->r_info);
391 type = ELFW(R_TYPE)(rel->r_info);
392 sym_index = old_to_new_syms[sym_index];
393 rel->r_info = ELFW(R_INFO)(sym_index, type);
398 tcc_free(old_to_new_syms);
401 /* relocate common symbols in the .bss section */
402 ST_FUNC void relocate_common_syms(void)
404 ElfW(Sym) *sym;
405 unsigned long offset, align;
407 for_each_elem(symtab_section, 1, sym, ElfW(Sym)) {
408 if (sym->st_shndx == SHN_COMMON) {
409 /* align symbol */
410 align = sym->st_value;
411 offset = bss_section->data_offset;
412 offset = (offset + align - 1) & -align;
413 sym->st_value = offset;
414 sym->st_shndx = bss_section->sh_num;
415 offset += sym->st_size;
416 bss_section->data_offset = offset;
421 /* relocate symbol table, resolve undefined symbols if do_resolve is
422 true and output error if undefined symbol. */
423 ST_FUNC void relocate_syms(TCCState *s1, int do_resolve)
425 ElfW(Sym) *sym, *esym;
426 int sym_bind, sh_num, sym_index;
427 const char *name;
429 for_each_elem(symtab_section, 1, sym, ElfW(Sym)) {
430 sh_num = sym->st_shndx;
431 if (sh_num == SHN_UNDEF) {
432 name = (char *) strtab_section->data + sym->st_name;
433 /* Use ld.so to resolve symbol for us (for tcc -run) */
434 if (do_resolve) {
435 #if defined TCC_IS_NATIVE && !defined _WIN32
436 void *addr;
437 name = (char *) symtab_section->link->data + sym->st_name;
438 addr = resolve_sym(s1, name);
439 if (addr) {
440 sym->st_value = (addr_t)addr;
441 goto found;
443 #endif
444 } else if (s1->dynsym) {
445 /* if dynamic symbol exist, then use it */
446 sym_index = find_elf_sym(s1->dynsym, name);
447 if (sym_index) {
448 esym = &((ElfW(Sym) *)s1->dynsym->data)[sym_index];
449 sym->st_value = esym->st_value;
450 goto found;
453 /* XXX: _fp_hw seems to be part of the ABI, so we ignore
454 it */
455 if (!strcmp(name, "_fp_hw"))
456 goto found;
457 /* only weak symbols are accepted to be undefined. Their
458 value is zero */
459 sym_bind = ELFW(ST_BIND)(sym->st_info);
460 if (sym_bind == STB_WEAK) {
461 sym->st_value = 0;
462 } else {
463 tcc_error_noabort("undefined symbol '%s'", name);
465 } else if (sh_num < SHN_LORESERVE) {
466 /* add section base */
467 sym->st_value += s1->sections[sym->st_shndx]->sh_addr;
469 found: ;
473 #ifdef TCC_HAS_RUNTIME_PLTGOT
474 #ifdef TCC_TARGET_X86_64
475 #define JMP_TABLE_ENTRY_SIZE 14
476 static addr_t add_jmp_table(TCCState *s1, addr_t val)
478 char *p = s1->runtime_plt_and_got + s1->runtime_plt_and_got_offset;
479 s1->runtime_plt_and_got_offset += JMP_TABLE_ENTRY_SIZE;
480 /* jmp *0x0(%rip) */
481 p[0] = 0xff;
482 p[1] = 0x25;
483 *(int *)(p + 2) = 0;
484 *(addr_t *)(p + 6) = val;
485 return (addr_t)p;
488 static addr_t add_got_table(TCCState *s1, addr_t val)
490 addr_t *p = (addr_t *)(s1->runtime_plt_and_got + s1->runtime_plt_and_got_offset);
491 s1->runtime_plt_and_got_offset += sizeof(addr_t);
492 *p = val;
493 return (addr_t)p;
495 #elif defined TCC_TARGET_ARM
496 #define JMP_TABLE_ENTRY_SIZE 8
497 static addr_t add_jmp_table(TCCState *s1, int val)
499 uint32_t *p = (uint32_t *)(s1->runtime_plt_and_got + s1->runtime_plt_and_got_offset);
500 s1->runtime_plt_and_got_offset += JMP_TABLE_ENTRY_SIZE;
501 /* ldr pc, [pc, #-4] */
502 p[0] = 0xE51FF004;
503 p[1] = val;
504 return (addr_t)p;
506 #endif
507 #endif /* def TCC_HAS_RUNTIME_PLTGOT */
509 /* relocate a given section (CPU dependent) by applying the relocations
510 in the associated relocation section */
511 ST_FUNC void relocate_section(TCCState *s1, Section *s)
513 Section *sr = s->reloc;
514 ElfW_Rel *rel;
515 ElfW(Sym) *sym;
516 int type, sym_index;
517 unsigned char *ptr;
518 addr_t val, addr;
519 #if defined TCC_TARGET_I386 || defined TCC_TARGET_X86_64
520 ElfW_Rel *qrel = (ElfW_Rel *) sr->data; /* ptr to next reloc entry reused */
521 int esym_index;
522 #endif
524 for_each_elem(sr, 0, rel, ElfW_Rel) {
525 ptr = s->data + rel->r_offset;
527 sym_index = ELFW(R_SYM)(rel->r_info);
528 sym = &((ElfW(Sym) *)symtab_section->data)[sym_index];
529 val = sym->st_value;
530 #ifdef TCC_TARGET_X86_64
531 val += rel->r_addend;
532 #endif
533 type = ELFW(R_TYPE)(rel->r_info);
534 addr = s->sh_addr + rel->r_offset;
536 /* CPU specific */
537 switch(type) {
538 #if defined(TCC_TARGET_I386)
539 case R_386_32:
540 if (s1->output_type == TCC_OUTPUT_DLL) {
541 esym_index = s1->symtab_to_dynsym[sym_index];
542 qrel->r_offset = rel->r_offset;
543 if (esym_index) {
544 qrel->r_info = ELFW(R_INFO)(esym_index, R_386_32);
545 qrel++;
546 break;
547 } else {
548 qrel->r_info = ELFW(R_INFO)(0, R_386_RELATIVE);
549 qrel++;
552 *(int *)ptr += val;
553 break;
554 case R_386_PC32:
555 if (s1->output_type == TCC_OUTPUT_DLL) {
556 /* DLL relocation */
557 esym_index = s1->symtab_to_dynsym[sym_index];
558 if (esym_index) {
559 qrel->r_offset = rel->r_offset;
560 qrel->r_info = ELFW(R_INFO)(esym_index, R_386_PC32);
561 qrel++;
562 break;
565 *(int *)ptr += val - addr;
566 break;
567 case R_386_PLT32:
568 *(int *)ptr += val - addr;
569 break;
570 case R_386_GLOB_DAT:
571 case R_386_JMP_SLOT:
572 *(int *)ptr = val;
573 break;
574 case R_386_GOTPC:
575 *(int *)ptr += s1->got->sh_addr - addr;
576 break;
577 case R_386_GOTOFF:
578 *(int *)ptr += val - s1->got->sh_addr;
579 break;
580 case R_386_GOT32:
581 /* we load the got offset */
582 *(int *)ptr += s1->sym_attrs[sym_index].got_offset;
583 break;
584 case R_386_16:
585 if (s1->output_format != TCC_OUTPUT_FORMAT_BINARY) {
586 output_file:
587 tcc_error("can only produce 16-bit binary files");
589 *(short *)ptr += val;
590 break;
591 case R_386_PC16:
592 if (s1->output_format != TCC_OUTPUT_FORMAT_BINARY)
593 goto output_file;
594 *(short *)ptr += val - addr;
595 break;
596 #elif defined(TCC_TARGET_ARM)
597 case R_ARM_PC24:
598 case R_ARM_CALL:
599 case R_ARM_JUMP24:
600 case R_ARM_PLT32:
602 int x, is_thumb, is_call, h, blx_avail, is_bl, th_ko;
603 x = (*(int *) ptr) & 0xffffff;
604 (*(int *)ptr) &= 0xff000000;
605 if (x & 0x800000)
606 x -= 0x1000000;
607 x <<= 2;
608 blx_avail = (TCC_ARM_VERSION >= 5);
609 is_thumb = val & 1;
610 is_bl = (*(unsigned *) ptr) >> 24 == 0xeb;
611 is_call = (type == R_ARM_CALL || (type == R_ARM_PC24 && is_bl));
612 x += val - addr;
613 h = x & 2;
614 th_ko = (x & 3) && (!blx_avail || !is_call);
615 #ifdef TCC_HAS_RUNTIME_PLTGOT
616 if (s1->output_type == TCC_OUTPUT_MEMORY) {
617 if (th_ko || x >= 0x2000000 || x < -0x2000000) {
618 x += add_jmp_table(s1, val) - val; /* add veneer */
619 th_ko = (x & 3) && (!blx_avail || !is_call);
620 is_thumb = 0; /* Veneer uses ARM instructions */
623 #endif
624 if (th_ko || x >= 0x2000000 || x < -0x2000000)
625 tcc_error("can't relocate value at %x",addr);
626 x >>= 2;
627 x &= 0xffffff;
628 /* Only reached if blx is avail and it is a call */
629 if (is_thumb) {
630 x |= h << 24;
631 (*(int *)ptr) = 0xfa << 24; /* bl -> blx */
633 (*(int *) ptr) |= x;
635 break;
636 /* Since these relocations only concern Thumb-2 and blx instruction was
637 introduced before Thumb-2, we can assume blx is available and not
638 guard its use */
639 case R_ARM_THM_PC22:
640 case R_ARM_THM_JUMP24:
642 int x, hi, lo, s, j1, j2, i1, i2, imm10, imm11;
643 int to_thumb, is_call, to_plt, blx_bit = 1 << 12;
644 Section *plt;
646 /* weak reference */
647 if (sym->st_shndx == SHN_UNDEF &&
648 ELFW(ST_BIND)(sym->st_info) == STB_WEAK)
649 break;
651 /* Get initial offset */
652 hi = (*(uint16_t *)ptr);
653 lo = (*(uint16_t *)(ptr+2));
654 s = (hi >> 10) & 1;
655 j1 = (lo >> 13) & 1;
656 j2 = (lo >> 11) & 1;
657 i1 = (j1 ^ s) ^ 1;
658 i2 = (j2 ^ s) ^ 1;
659 imm10 = hi & 0x3ff;
660 imm11 = lo & 0x7ff;
661 x = (s << 24) | (i1 << 23) | (i2 << 22) |
662 (imm10 << 12) | (imm11 << 1);
663 if (x & 0x01000000)
664 x -= 0x02000000;
666 /* Relocation infos */
667 to_thumb = val & 1;
668 plt = s1->plt;
669 to_plt = (val >= plt->sh_addr) &&
670 (val < plt->sh_addr + plt->data_offset);
671 is_call = (type == R_ARM_THM_PC22);
673 /* Compute final offset */
674 if (to_plt && !is_call) /* Point to 1st instr of Thumb stub */
675 x -= 4;
676 x += val - addr;
677 if (!to_thumb && is_call) {
678 blx_bit = 0; /* bl -> blx */
679 x = (x + 3) & -4; /* Compute offset from aligned PC */
682 /* Check that relocation is possible
683 * offset must not be out of range
684 * if target is to be entered in arm mode:
685 - bit 1 must not set
686 - instruction must be a call (bl) or a jump to PLT */
687 if (!to_thumb || x >= 0x1000000 || x < -0x1000000)
688 if (to_thumb || (val & 2) || (!is_call && !to_plt))
689 tcc_error("can't relocate value at %x",addr);
691 /* Compute and store final offset */
692 s = (x >> 24) & 1;
693 i1 = (x >> 23) & 1;
694 i2 = (x >> 22) & 1;
695 j1 = s ^ (i1 ^ 1);
696 j2 = s ^ (i2 ^ 1);
697 imm10 = (x >> 12) & 0x3ff;
698 imm11 = (x >> 1) & 0x7ff;
699 (*(uint16_t *)ptr) = (uint16_t) ((hi & 0xf800) |
700 (s << 10) | imm10);
701 (*(uint16_t *)(ptr+2)) = (uint16_t) ((lo & 0xc000) |
702 (j1 << 13) | blx_bit | (j2 << 11) |
703 imm11);
705 break;
706 case R_ARM_MOVT_ABS:
707 case R_ARM_MOVW_ABS_NC:
709 int x, imm4, imm12;
710 if (type == R_ARM_MOVT_ABS)
711 val >>= 16;
712 imm12 = val & 0xfff;
713 imm4 = (val >> 12) & 0xf;
714 x = (imm4 << 16) | imm12;
715 if (type == R_ARM_THM_MOVT_ABS)
716 *(int *)ptr |= x;
717 else
718 *(int *)ptr += x;
720 break;
721 case R_ARM_THM_MOVT_ABS:
722 case R_ARM_THM_MOVW_ABS_NC:
724 int x, i, imm4, imm3, imm8;
725 if (type == R_ARM_THM_MOVT_ABS)
726 val >>= 16;
727 imm8 = val & 0xff;
728 imm3 = (val >> 8) & 0x7;
729 i = (val >> 11) & 1;
730 imm4 = (val >> 12) & 0xf;
731 x = (imm3 << 28) | (imm8 << 16) | (i << 10) | imm4;
732 if (type == R_ARM_THM_MOVT_ABS)
733 *(int *)ptr |= x;
734 else
735 *(int *)ptr += x;
737 break;
738 case R_ARM_PREL31:
740 int x;
741 x = (*(int *)ptr) & 0x7fffffff;
742 (*(int *)ptr) &= 0x80000000;
743 x = (x * 2) / 2;
744 x += val - addr;
745 if((x^(x>>1))&0x40000000)
746 tcc_error("can't relocate value at %x",addr);
747 (*(int *)ptr) |= x & 0x7fffffff;
749 case R_ARM_ABS32:
750 *(int *)ptr += val;
751 break;
752 case R_ARM_REL32:
753 *(int *)ptr += val - addr;
754 break;
755 case R_ARM_GOTPC:
756 *(int *)ptr += s1->got->sh_addr - addr;
757 break;
758 case R_ARM_GOTOFF:
759 *(int *)ptr += val - s1->got->sh_addr;
760 break;
761 case R_ARM_GOT32:
762 /* we load the got offset */
763 *(int *)ptr += s1->sym_attrs[sym_index].got_offset;
764 break;
765 case R_ARM_COPY:
766 break;
767 case R_ARM_V4BX:
768 /* trade Thumb support for ARMv4 support */
769 if ((0x0ffffff0 & *(int*)ptr) == 0x012FFF10)
770 *(int*)ptr ^= 0xE12FFF10 ^ 0xE1A0F000; /* BX Rm -> MOV PC, Rm */
771 break;
772 default:
773 fprintf(stderr,"FIXME: handle reloc type %x at %x [%p] to %x\n",
774 type, (unsigned)addr, ptr, (unsigned)val);
775 break;
776 #elif defined(TCC_TARGET_C67)
777 case R_C60_32:
778 *(int *)ptr += val;
779 break;
780 case R_C60LO16:
782 uint32_t orig;
784 /* put the low 16 bits of the absolute address
785 add to what is already there */
787 orig = ((*(int *)(ptr )) >> 7) & 0xffff;
788 orig |= (((*(int *)(ptr+4)) >> 7) & 0xffff) << 16;
790 /* patch both at once - assumes always in pairs Low - High */
792 *(int *) ptr = (*(int *) ptr & (~(0xffff << 7)) ) | (((val+orig) & 0xffff) << 7);
793 *(int *)(ptr+4) = (*(int *)(ptr+4) & (~(0xffff << 7)) ) | ((((val+orig)>>16) & 0xffff) << 7);
795 break;
796 case R_C60HI16:
797 break;
798 default:
799 fprintf(stderr,"FIXME: handle reloc type %x at %x [%p] to %x\n",
800 type, (unsigned)addr, ptr, (unsigned)val);
801 break;
802 #elif defined(TCC_TARGET_X86_64)
803 case R_X86_64_64:
804 if (s1->output_type == TCC_OUTPUT_DLL) {
805 qrel->r_info = ELFW(R_INFO)(0, R_X86_64_RELATIVE);
806 qrel->r_addend = *(long long *)ptr + val;
807 qrel++;
809 *(long long *)ptr += val;
810 break;
811 case R_X86_64_32:
812 case R_X86_64_32S:
813 if (s1->output_type == TCC_OUTPUT_DLL) {
814 /* XXX: this logic may depend on TCC's codegen
815 now TCC uses R_X86_64_32 even for a 64bit pointer */
816 qrel->r_info = ELFW(R_INFO)(0, R_X86_64_RELATIVE);
817 qrel->r_addend = *(int *)ptr + val;
818 qrel++;
820 *(int *)ptr += val;
821 break;
823 case R_X86_64_PC32:
824 if (s1->output_type == TCC_OUTPUT_DLL) {
825 /* DLL relocation */
826 esym_index = s1->symtab_to_dynsym[sym_index];
827 if (esym_index) {
828 qrel->r_offset = rel->r_offset;
829 qrel->r_info = ELFW(R_INFO)(esym_index, R_X86_64_PC32);
830 qrel->r_addend = *(int *)ptr;
831 qrel++;
832 break;
835 /* fall through */
836 case R_X86_64_PLT32: {
837 long long diff;
838 diff = (long long)val - addr;
839 if (diff <= -2147483647 || diff > 2147483647) {
840 #ifdef TCC_HAS_RUNTIME_PLTGOT
841 /* XXX: naive support for over 32bit jump */
842 if (s1->output_type == TCC_OUTPUT_MEMORY) {
843 val = (add_jmp_table(s1, val - rel->r_addend) +
844 rel->r_addend);
845 diff = val - addr;
847 #endif
848 if (diff <= -2147483647 || diff > 2147483647) {
849 tcc_error("internal error: relocation failed");
852 *(int *)ptr += diff;
854 break;
855 case R_X86_64_GLOB_DAT:
856 case R_X86_64_JUMP_SLOT:
857 /* They don't need addend */
858 *(int *)ptr = val - rel->r_addend;
859 break;
860 case R_X86_64_GOTPCREL:
861 #ifdef TCC_HAS_RUNTIME_PLTGOT
862 if (s1->output_type == TCC_OUTPUT_MEMORY) {
863 val = add_got_table(s1, val - rel->r_addend) + rel->r_addend;
864 *(int *)ptr += val - addr;
865 break;
867 #endif
868 *(int *)ptr += (s1->got->sh_addr - addr +
869 s1->sym_attrs[sym_index].got_offset - 4);
870 break;
871 case R_X86_64_GOTTPOFF:
872 *(int *)ptr += val - s1->got->sh_addr;
873 break;
874 case R_X86_64_GOT32:
875 /* we load the got offset */
876 *(int *)ptr += s1->sym_attrs[sym_index].got_offset;
877 break;
878 #else
879 #error unsupported processor
880 #endif
883 /* if the relocation is allocated, we change its symbol table */
884 if (sr->sh_flags & SHF_ALLOC)
885 sr->link = s1->dynsym;
888 /* relocate relocation table in 'sr' */
889 static void relocate_rel(TCCState *s1, Section *sr)
891 Section *s;
892 ElfW_Rel *rel;
894 s = s1->sections[sr->sh_info];
895 for_each_elem(sr, 0, rel, ElfW_Rel)
896 rel->r_offset += s->sh_addr;
899 /* count the number of dynamic relocations so that we can reserve
900 their space */
901 static int prepare_dynamic_rel(TCCState *s1, Section *sr)
903 ElfW_Rel *rel;
904 int sym_index, esym_index, type, count;
906 count = 0;
907 for_each_elem(sr, 0, rel, ElfW_Rel) {
908 sym_index = ELFW(R_SYM)(rel->r_info);
909 type = ELFW(R_TYPE)(rel->r_info);
910 switch(type) {
911 #if defined(TCC_TARGET_I386)
912 case R_386_32:
913 #elif defined(TCC_TARGET_X86_64)
914 case R_X86_64_32:
915 case R_X86_64_32S:
916 case R_X86_64_64:
917 #endif
918 count++;
919 break;
920 #if defined(TCC_TARGET_I386)
921 case R_386_PC32:
922 #elif defined(TCC_TARGET_X86_64)
923 case R_X86_64_PC32:
924 #endif
925 esym_index = s1->symtab_to_dynsym[sym_index];
926 if (esym_index)
927 count++;
928 break;
929 default:
930 break;
933 if (count) {
934 /* allocate the section */
935 sr->sh_flags |= SHF_ALLOC;
936 sr->sh_size = count * sizeof(ElfW_Rel);
938 return count;
941 static struct sym_attr *alloc_sym_attr(TCCState *s1, int index)
943 int n;
944 struct sym_attr *tab;
946 if (index >= s1->nb_sym_attrs) {
947 /* find immediately bigger power of 2 and reallocate array */
948 n = 1;
949 while (index >= n)
950 n *= 2;
951 tab = tcc_realloc(s1->sym_attrs, n * sizeof(*s1->sym_attrs));
952 s1->sym_attrs = tab;
953 memset(s1->sym_attrs + s1->nb_sym_attrs, 0,
954 (n - s1->nb_sym_attrs) * sizeof(*s1->sym_attrs));
955 s1->nb_sym_attrs = n;
957 return &s1->sym_attrs[index];
960 /* XXX: suppress that */
961 static void put32(unsigned char *p, uint32_t val)
963 p[0] = val;
964 p[1] = val >> 8;
965 p[2] = val >> 16;
966 p[3] = val >> 24;
969 #if defined(TCC_TARGET_I386) || defined(TCC_TARGET_ARM) || \
970 defined(TCC_TARGET_X86_64)
971 static uint32_t get32(unsigned char *p)
973 return p[0] | (p[1] << 8) | (p[2] << 16) | (p[3] << 24);
975 #endif
977 static void build_got(TCCState *s1)
979 unsigned char *ptr;
981 /* if no got, then create it */
982 s1->got = new_section(s1, ".got", SHT_PROGBITS, SHF_ALLOC | SHF_WRITE);
983 s1->got->sh_entsize = 4;
984 add_elf_sym(symtab_section, 0, 4, ELFW(ST_INFO)(STB_GLOBAL, STT_OBJECT),
985 0, s1->got->sh_num, "_GLOBAL_OFFSET_TABLE_");
986 ptr = section_ptr_add(s1->got, 3 * PTR_SIZE);
987 #if PTR_SIZE == 4
988 /* keep space for _DYNAMIC pointer, if present */
989 put32(ptr, 0);
990 /* two dummy got entries */
991 put32(ptr + 4, 0);
992 put32(ptr + 8, 0);
993 #else
994 /* keep space for _DYNAMIC pointer, if present */
995 put32(ptr, 0);
996 put32(ptr + 4, 0);
997 /* two dummy got entries */
998 put32(ptr + 8, 0);
999 put32(ptr + 12, 0);
1000 put32(ptr + 16, 0);
1001 put32(ptr + 20, 0);
1002 #endif
1005 /* put a got entry corresponding to a symbol in symtab_section. 'size'
1006 and 'info' can be modifed if more precise info comes from the DLL */
1007 static void put_got_entry(TCCState *s1,
1008 int reloc_type, unsigned long size, int info,
1009 int sym_index)
1011 int index, need_plt_entry, got_entry_present = 0;
1012 const char *name;
1013 ElfW(Sym) *sym;
1014 unsigned long offset;
1015 int *ptr;
1017 if (!s1->got)
1018 build_got(s1);
1020 need_plt_entry = s1->dynsym &&
1021 #ifdef TCC_TARGET_X86_64
1022 (reloc_type == R_X86_64_JUMP_SLOT);
1023 #elif defined(TCC_TARGET_I386)
1024 (reloc_type == R_386_JMP_SLOT);
1025 #elif defined(TCC_TARGET_ARM)
1026 (reloc_type == R_ARM_JUMP_SLOT);
1027 #else
1029 #endif
1031 /* if a got entry already exists for that symbol, no need to add one */
1032 if (sym_index < s1->nb_sym_attrs &&
1033 s1->sym_attrs[sym_index].got_offset) {
1034 if (!need_plt_entry)
1035 return;
1036 else
1037 got_entry_present = 1;
1040 alloc_sym_attr(s1, sym_index)->got_offset = s1->got->data_offset;
1042 if (s1->dynsym) {
1043 sym = &((ElfW(Sym) *)symtab_section->data)[sym_index];
1044 name = (char *) symtab_section->link->data + sym->st_name;
1045 if (s1->sym_attrs[sym_index].has_plt_entry)
1046 return;
1047 s1->sym_attrs[sym_index].has_plt_entry = 1;
1048 offset = sym->st_value;
1049 #if defined(TCC_TARGET_I386) || defined(TCC_TARGET_X86_64)
1050 if (need_plt_entry) {
1051 Section *plt;
1052 uint8_t *p;
1053 int modrm;
1055 #if defined(TCC_OUTPUT_DLL_WITH_PLT)
1056 modrm = 0x25;
1057 #else
1058 /* if we build a DLL, we add a %ebx offset */
1059 if (s1->output_type == TCC_OUTPUT_DLL)
1060 modrm = 0xa3;
1061 else
1062 modrm = 0x25;
1063 #endif
1065 /* add a PLT entry */
1066 plt = s1->plt;
1067 if (plt->data_offset == 0) {
1068 /* first plt entry */
1069 p = section_ptr_add(plt, 16);
1070 p[0] = 0xff; /* pushl got + PTR_SIZE */
1071 p[1] = modrm + 0x10;
1072 put32(p + 2, PTR_SIZE);
1073 p[6] = 0xff; /* jmp *(got + PTR_SIZE * 2) */
1074 p[7] = modrm;
1075 put32(p + 8, PTR_SIZE * 2);
1078 p = section_ptr_add(plt, 16);
1079 p[0] = 0xff; /* jmp *(got + x) */
1080 p[1] = modrm;
1081 put32(p + 2, s1->got->data_offset);
1082 p[6] = 0x68; /* push $xxx */
1083 put32(p + 7, (plt->data_offset - 32) >> 1);
1084 p[11] = 0xe9; /* jmp plt_start */
1085 put32(p + 12, -(plt->data_offset));
1087 /* the symbol is modified so that it will be relocated to
1088 the PLT */
1089 #if !defined(TCC_OUTPUT_DLL_WITH_PLT)
1090 if (s1->output_type == TCC_OUTPUT_EXE)
1091 #endif
1092 offset = plt->data_offset - 16;
1094 #elif defined(TCC_TARGET_ARM)
1095 if (need_plt_entry) {
1096 Section *plt;
1097 uint8_t *p;
1099 /* if we build a DLL, we add a %ebx offset */
1100 if (s1->output_type == TCC_OUTPUT_DLL)
1101 tcc_error("DLLs unimplemented!");
1103 /* add a PLT entry */
1104 plt = s1->plt;
1105 if (plt->data_offset == 0) {
1106 /* first plt entry */
1107 p = section_ptr_add(plt, 16);
1108 put32(p, 0xe52de004); /* push {lr} */
1109 put32(p+4, 0xe59fe010); /* ldr lr, [pc, #16] */
1110 put32(p+8, 0xe08fe00e); /* add lr, pc, lr */
1111 put32(p+12, 0xe5bef008); /* ldr pc, [lr, #8]! */
1114 if (s1->sym_attrs[sym_index].plt_thumb_stub) {
1115 p = section_ptr_add(plt, 20);
1116 put32(p, 0x4778); /* bx pc */
1117 put32(p+2, 0x46c0); /* nop */
1118 p += 4;
1119 } else
1120 p = section_ptr_add(plt, 16);
1121 put32(p, 0xe59fc004); /* ldr ip, [pc, #4] ; GOT entry offset */
1122 put32(p+4, 0xe08fc00c); /* add ip, pc, ip ; addr of GOT entry */
1123 put32(p+8, 0xe59cf000); /* ldr pc, [ip] ; jump to GOT entry */
1124 put32(p+12, s1->got->data_offset); /* GOT entry off once patched */
1126 /* the symbol is modified so that it will be relocated to
1127 the PLT */
1128 if (s1->output_type == TCC_OUTPUT_EXE)
1129 offset = plt->data_offset - 16;
1131 #elif defined(TCC_TARGET_C67)
1132 tcc_error("C67 got not implemented");
1133 #else
1134 #error unsupported CPU
1135 #endif
1136 index = put_elf_sym(s1->dynsym, offset,
1137 size, info, 0, sym->st_shndx, name);
1138 if (got_entry_present) {
1139 put_elf_reloc(s1->dynsym, s1->got,
1140 s1->sym_attrs[sym_index].got_offset,
1141 reloc_type, index);
1142 return;
1145 /* put a got entry */
1146 put_elf_reloc(s1->dynsym, s1->got,
1147 s1->got->data_offset,
1148 reloc_type, index);
1150 ptr = section_ptr_add(s1->got, PTR_SIZE);
1151 *ptr = 0;
1154 /* build GOT and PLT entries */
1155 ST_FUNC void build_got_entries(TCCState *s1)
1157 Section *s;
1158 ElfW_Rel *rel;
1159 ElfW(Sym) *sym;
1160 int i, type, reloc_type, sym_index;
1162 for(i = 1; i < s1->nb_sections; i++) {
1163 s = s1->sections[i];
1164 if (s->sh_type != SHT_RELX)
1165 continue;
1166 /* no need to handle got relocations */
1167 if (s->link != symtab_section)
1168 continue;
1169 for_each_elem(s, 0, rel, ElfW_Rel) {
1170 type = ELFW(R_TYPE)(rel->r_info);
1171 switch(type) {
1172 #if defined(TCC_TARGET_I386)
1173 case R_386_GOT32:
1174 case R_386_GOTOFF:
1175 case R_386_GOTPC:
1176 case R_386_PLT32:
1177 if (!s1->got)
1178 build_got(s1);
1179 if (type == R_386_GOT32 || type == R_386_PLT32) {
1180 sym_index = ELFW(R_SYM)(rel->r_info);
1181 sym = &((ElfW(Sym) *)symtab_section->data)[sym_index];
1182 /* look at the symbol got offset. If none, then add one */
1183 if (type == R_386_GOT32)
1184 reloc_type = R_386_GLOB_DAT;
1185 else
1186 reloc_type = R_386_JMP_SLOT;
1187 put_got_entry(s1, reloc_type, sym->st_size, sym->st_info,
1188 sym_index);
1190 break;
1191 #elif defined(TCC_TARGET_ARM)
1192 case R_ARM_GOT32:
1193 case R_ARM_GOTOFF:
1194 case R_ARM_GOTPC:
1195 case R_ARM_PLT32:
1196 if (!s1->got)
1197 build_got(s1);
1198 if (type == R_ARM_GOT32 || type == R_ARM_PLT32) {
1199 sym_index = ELFW(R_SYM)(rel->r_info);
1200 sym = &((ElfW(Sym) *)symtab_section->data)[sym_index];
1201 /* look at the symbol got offset. If none, then add one */
1202 if (type == R_ARM_GOT32)
1203 reloc_type = R_ARM_GLOB_DAT;
1204 else
1205 reloc_type = R_ARM_JUMP_SLOT;
1206 put_got_entry(s1, reloc_type, sym->st_size, sym->st_info,
1207 sym_index);
1209 break;
1210 case R_ARM_THM_JUMP24:
1211 sym_index = ELFW(R_SYM)(rel->r_info);
1212 sym = &((ElfW(Sym) *)symtab_section->data)[sym_index];
1213 /* We are relocating a jump from thumb code to arm code */
1214 if (sym->st_shndx != SHN_UNDEF && !(sym->st_value & 1)) {
1215 int index;
1216 uint8_t *p;
1217 char *name, buf[1024];
1218 Section *text_section;
1220 name = (char *) symtab_section->link->data + sym->st_name;
1221 text_section = s1->sections[sym->st_shndx];
1222 /* Modify reloc to target a thumb stub to switch to ARM */
1223 snprintf(buf, sizeof(buf), "%s_from_thumb", name);
1224 index = put_elf_sym(symtab_section,
1225 text_section->data_offset + 1,
1226 sym->st_size, sym->st_info, 0,
1227 sym->st_shndx, buf);
1228 rel->r_info = ELFW(R_INFO)(index, type);
1229 /* Create a thumb stub fonction to switch to ARM mode */
1230 put_elf_reloc(symtab_section, text_section,
1231 text_section->data_offset + 4, R_ARM_JUMP24,
1232 sym_index);
1233 p = section_ptr_add(text_section, 8);
1234 put32(p, 0x4778); /* bx pc */
1235 put32(p+2, 0x46c0); /* nop */
1236 put32(p+4, 0xeafffffe); /* b $sym */
1238 #elif defined(TCC_TARGET_C67)
1239 case R_C60_GOT32:
1240 case R_C60_GOTOFF:
1241 case R_C60_GOTPC:
1242 case R_C60_PLT32:
1243 if (!s1->got)
1244 build_got(s1);
1245 if (type == R_C60_GOT32 || type == R_C60_PLT32) {
1246 sym_index = ELFW(R_SYM)(rel->r_info);
1247 sym = &((ElfW(Sym) *)symtab_section->data)[sym_index];
1248 /* look at the symbol got offset. If none, then add one */
1249 if (type == R_C60_GOT32)
1250 reloc_type = R_C60_GLOB_DAT;
1251 else
1252 reloc_type = R_C60_JMP_SLOT;
1253 put_got_entry(s1, reloc_type, sym->st_size, sym->st_info,
1254 sym_index);
1256 break;
1257 #elif defined(TCC_TARGET_X86_64)
1258 case R_X86_64_GOT32:
1259 case R_X86_64_GOTTPOFF:
1260 case R_X86_64_GOTPCREL:
1261 case R_X86_64_PLT32:
1262 if (!s1->got)
1263 build_got(s1);
1264 if (type == R_X86_64_GOT32 || type == R_X86_64_GOTPCREL ||
1265 type == R_X86_64_PLT32) {
1266 sym_index = ELFW(R_SYM)(rel->r_info);
1267 sym = &((ElfW(Sym) *)symtab_section->data)[sym_index];
1268 /* look at the symbol got offset. If none, then add one */
1269 if (type == R_X86_64_GOT32 || type == R_X86_64_GOTPCREL)
1270 reloc_type = R_X86_64_GLOB_DAT;
1271 else
1272 reloc_type = R_X86_64_JUMP_SLOT;
1273 put_got_entry(s1, reloc_type, sym->st_size, sym->st_info,
1274 sym_index);
1276 break;
1277 #else
1278 #error unsupported CPU
1279 #endif
1280 default:
1281 break;
1287 ST_FUNC Section *new_symtab(TCCState *s1,
1288 const char *symtab_name, int sh_type, int sh_flags,
1289 const char *strtab_name,
1290 const char *hash_name, int hash_sh_flags)
1292 Section *symtab, *strtab, *hash;
1293 int *ptr, nb_buckets;
1295 symtab = new_section(s1, symtab_name, sh_type, sh_flags);
1296 symtab->sh_entsize = sizeof(ElfW(Sym));
1297 strtab = new_section(s1, strtab_name, SHT_STRTAB, sh_flags);
1298 put_elf_str(strtab, "");
1299 symtab->link = strtab;
1300 put_elf_sym(symtab, 0, 0, 0, 0, 0, NULL);
1302 nb_buckets = 1;
1304 hash = new_section(s1, hash_name, SHT_HASH, hash_sh_flags);
1305 hash->sh_entsize = sizeof(int);
1306 symtab->hash = hash;
1307 hash->link = symtab;
1309 ptr = section_ptr_add(hash, (2 + nb_buckets + 1) * sizeof(int));
1310 ptr[0] = nb_buckets;
1311 ptr[1] = 1;
1312 memset(ptr + 2, 0, (nb_buckets + 1) * sizeof(int));
1313 return symtab;
1316 /* put dynamic tag */
1317 static void put_dt(Section *dynamic, int dt, addr_t val)
1319 ElfW(Dyn) *dyn;
1320 dyn = section_ptr_add(dynamic, sizeof(ElfW(Dyn)));
1321 dyn->d_tag = dt;
1322 dyn->d_un.d_val = val;
1325 static void add_init_array_defines(TCCState *s1, const char *section_name)
1327 Section *s;
1328 long end_offset;
1329 char sym_start[1024];
1330 char sym_end[1024];
1332 snprintf(sym_start, sizeof(sym_start), "__%s_start", section_name + 1);
1333 snprintf(sym_end, sizeof(sym_end), "__%s_end", section_name + 1);
1335 s = find_section(s1, section_name);
1336 if (!s) {
1337 end_offset = 0;
1338 s = data_section;
1339 } else {
1340 end_offset = s->data_offset;
1343 add_elf_sym(symtab_section,
1344 0, 0,
1345 ELFW(ST_INFO)(STB_GLOBAL, STT_NOTYPE), 0,
1346 s->sh_num, sym_start);
1347 add_elf_sym(symtab_section,
1348 end_offset, 0,
1349 ELFW(ST_INFO)(STB_GLOBAL, STT_NOTYPE), 0,
1350 s->sh_num, sym_end);
1353 ST_FUNC void tcc_add_bcheck(TCCState *s1)
1355 #ifdef CONFIG_TCC_BCHECK
1356 unsigned long *ptr;
1357 Section *init_section;
1358 unsigned char *pinit;
1359 int sym_index;
1361 if (0 == s1->do_bounds_check)
1362 return;
1364 /* XXX: add an object file to do that */
1365 ptr = section_ptr_add(bounds_section, sizeof(unsigned long));
1366 *ptr = 0;
1367 add_elf_sym(symtab_section, 0, 0,
1368 ELFW(ST_INFO)(STB_GLOBAL, STT_NOTYPE), 0,
1369 bounds_section->sh_num, "__bounds_start");
1370 #ifdef TCC_TARGET_I386
1371 if (s1->output_type != TCC_OUTPUT_MEMORY) {
1372 /* add 'call __bound_init()' in .init section */
1373 init_section = find_section(s1, ".init");
1374 pinit = section_ptr_add(init_section, 5);
1375 pinit[0] = 0xe8;
1376 put32(pinit + 1, -4);
1377 sym_index = find_elf_sym(symtab_section, "__bound_init");
1378 put_elf_reloc(symtab_section, init_section,
1379 init_section->data_offset - 4, R_386_PC32, sym_index);
1381 #endif
1382 #endif
1385 static inline int tcc_add_support(TCCState *s1, const char *filename)
1387 char buf[1024];
1388 snprintf(buf, sizeof(buf), "%s/%s", s1->tcc_lib_path, filename);
1389 return tcc_add_file(s1, buf);
1392 /* add tcc runtime libraries */
1393 ST_FUNC void tcc_add_runtime(TCCState *s1)
1395 /* add libc */
1396 if (!s1->nostdlib) {
1397 tcc_add_library(s1, "c");
1398 #ifdef CONFIG_USE_LIBGCC
1399 if (!s1->static_link) {
1400 tcc_add_file(s1, TCC_LIBGCC);
1401 tcc_add_support(s1, "libtcc1.a");
1402 } else
1403 tcc_add_support(s1, "libtcc1.a");
1404 #else
1405 tcc_add_support(s1, "libtcc1.a");
1406 #endif
1409 /* tcc_add_bcheck tries to relocate a call to __bound_init in _init so
1410 libtcc1.a must be loaded before for __bound_init to be defined and
1411 crtn.o must be loaded after to not finalize _init too early. */
1412 tcc_add_bcheck(s1);
1414 if (!s1->nostdlib) {
1415 /* add crt end if not memory output */
1416 if (s1->output_type != TCC_OUTPUT_MEMORY)
1417 tcc_add_crt(s1, "crtn.o");
1421 /* add various standard linker symbols (must be done after the
1422 sections are filled (for example after allocating common
1423 symbols)) */
1424 ST_FUNC void tcc_add_linker_symbols(TCCState *s1)
1426 char buf[1024];
1427 int i;
1428 Section *s;
1430 add_elf_sym(symtab_section,
1431 text_section->data_offset, 0,
1432 ELFW(ST_INFO)(STB_GLOBAL, STT_NOTYPE), 0,
1433 text_section->sh_num, "_etext");
1434 add_elf_sym(symtab_section,
1435 data_section->data_offset, 0,
1436 ELFW(ST_INFO)(STB_GLOBAL, STT_NOTYPE), 0,
1437 data_section->sh_num, "_edata");
1438 add_elf_sym(symtab_section,
1439 bss_section->data_offset, 0,
1440 ELFW(ST_INFO)(STB_GLOBAL, STT_NOTYPE), 0,
1441 bss_section->sh_num, "_end");
1442 /* horrible new standard ldscript defines */
1443 add_init_array_defines(s1, ".preinit_array");
1444 add_init_array_defines(s1, ".init_array");
1445 add_init_array_defines(s1, ".fini_array");
1447 /* add start and stop symbols for sections whose name can be
1448 expressed in C */
1449 for(i = 1; i < s1->nb_sections; i++) {
1450 s = s1->sections[i];
1451 if (s->sh_type == SHT_PROGBITS &&
1452 (s->sh_flags & SHF_ALLOC)) {
1453 const char *p;
1454 int ch;
1456 /* check if section name can be expressed in C */
1457 p = s->name;
1458 for(;;) {
1459 ch = *p;
1460 if (!ch)
1461 break;
1462 if (!isid(ch) && !isnum(ch))
1463 goto next_sec;
1464 p++;
1466 snprintf(buf, sizeof(buf), "__start_%s", s->name);
1467 add_elf_sym(symtab_section,
1468 0, 0,
1469 ELFW(ST_INFO)(STB_GLOBAL, STT_NOTYPE), 0,
1470 s->sh_num, buf);
1471 snprintf(buf, sizeof(buf), "__stop_%s", s->name);
1472 add_elf_sym(symtab_section,
1473 s->data_offset, 0,
1474 ELFW(ST_INFO)(STB_GLOBAL, STT_NOTYPE), 0,
1475 s->sh_num, buf);
1477 next_sec: ;
1481 static void tcc_output_binary(TCCState *s1, FILE *f,
1482 const int *sec_order)
1484 Section *s;
1485 int i, offset, size;
1487 offset = 0;
1488 for(i=1;i<s1->nb_sections;i++) {
1489 s = s1->sections[sec_order[i]];
1490 if (s->sh_type != SHT_NOBITS &&
1491 (s->sh_flags & SHF_ALLOC)) {
1492 while (offset < s->sh_offset) {
1493 fputc(0, f);
1494 offset++;
1496 size = s->sh_size;
1497 fwrite(s->data, 1, size, f);
1498 offset += size;
1503 // making this evaluate to true allow valgrind to work on linux
1504 // but when compiled with debug info and then striped
1505 // the compiled programs segfault
1506 // more tought must be applyed here
1507 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
1508 #define HAVE_PHDR 1
1509 #define EXTRA_RELITEMS 14
1511 /* move the relocation value from .dynsym to .got */
1512 void patch_dynsym_undef(TCCState *s1, Section *s)
1514 uint32_t *gotd = (void *)s1->got->data;
1515 ElfW(Sym) *sym;
1517 gotd += 3; /* dummy entries in .got */
1518 /* relocate symbols in .dynsym */
1519 for_each_elem(s, 1, sym, ElfW(Sym)) {
1520 if (sym->st_shndx == SHN_UNDEF) {
1521 *gotd++ = sym->st_value + 6; /* XXX 6 is magic ? */
1522 sym->st_value = 0;
1526 #else
1527 #define HAVE_PHDR 0
1528 #define EXTRA_RELITEMS 9
1530 /* zero plt offsets of weak symbols in .dynsym */
1531 void patch_dynsym_undef(TCCState *s1, Section *s)
1533 ElfW(Sym) *sym;
1535 for_each_elem(s, 1, sym, ElfW(Sym))
1536 if (sym->st_shndx == SHN_UNDEF && ELFW(ST_BIND)(sym->st_info) == STB_WEAK)
1537 sym->st_value = 0;
1539 #endif
1541 ST_FUNC void fill_got_entry(TCCState *s1, ElfW_Rel *rel)
1543 int sym_index = ELFW(R_SYM) (rel->r_info);
1544 ElfW(Sym) *sym = &((ElfW(Sym) *) symtab_section->data)[sym_index];
1545 unsigned long offset;
1547 if (sym_index >= s1->nb_sym_attrs)
1548 return;
1549 offset = s1->sym_attrs[sym_index].got_offset;
1550 section_reserve(s1->got, offset + PTR_SIZE);
1551 #ifdef TCC_TARGET_X86_64
1552 /* only works for x86-64 */
1553 put32(s1->got->data + offset + 4, sym->st_value >> 32);
1554 #endif
1555 put32(s1->got->data + offset, sym->st_value & 0xffffffff);
1558 /* Perform relocation to GOT or PLT entries */
1559 ST_FUNC void fill_got(TCCState *s1)
1561 Section *s;
1562 ElfW_Rel *rel;
1563 int i;
1565 for(i = 1; i < s1->nb_sections; i++) {
1566 s = s1->sections[i];
1567 if (s->sh_type != SHT_RELX)
1568 continue;
1569 /* no need to handle got relocations */
1570 if (s->link != symtab_section)
1571 continue;
1572 for_each_elem(s, 0, rel, ElfW_Rel) {
1573 switch (ELFW(R_TYPE) (rel->r_info)) {
1574 case R_X86_64_GOT32:
1575 case R_X86_64_GOTPCREL:
1576 case R_X86_64_PLT32:
1577 fill_got_entry(s1, rel);
1578 break;
1584 /* Bind symbols of executable: resolve undefined symbols from exported symbols
1585 in shared libraries and export non local defined symbols to shared libraries
1586 if -rdynamic switch was given on command line */
1587 static void bind_exe_dynsyms(TCCState *s1)
1589 const char *name;
1590 int sym_index, index;
1591 ElfW(Sym) *sym, *esym;
1592 int type;
1594 /* Resolve undefined symbols from dynamic symbols. When there is a match:
1595 - if STT_FUNC or STT_GNU_IFUNC symbol -> add it in PLT
1596 - if STT_OBJECT symbol -> add it in .bss section with suitable reloc */
1597 for_each_elem(symtab_section, 1, sym, ElfW(Sym)) {
1598 if (sym->st_shndx == SHN_UNDEF) {
1599 name = (char *) symtab_section->link->data + sym->st_name;
1600 sym_index = find_elf_sym(s1->dynsymtab_section, name);
1601 if (sym_index) {
1602 esym = &((ElfW(Sym) *)s1->dynsymtab_section->data)[sym_index];
1603 type = ELFW(ST_TYPE)(esym->st_info);
1604 if ((type == STT_FUNC) || (type == STT_GNU_IFUNC)) {
1605 /* Indirect functions shall have STT_FUNC type in executable
1606 * dynsym section. Indeed, a dlsym call following a lazy
1607 * resolution would pick the symbol value from the
1608 * executable dynsym entry which would contain the address
1609 * of the function wanted by the caller of dlsym instead of
1610 * the address of the function that would return that
1611 * address */
1612 put_got_entry(s1, R_JMP_SLOT, esym->st_size,
1613 ELFW(ST_INFO)(STB_GLOBAL,STT_FUNC),
1614 sym - (ElfW(Sym) *)symtab_section->data);
1615 } else if (type == STT_OBJECT) {
1616 unsigned long offset;
1617 ElfW(Sym) *dynsym;
1618 offset = bss_section->data_offset;
1619 /* XXX: which alignment ? */
1620 offset = (offset + 16 - 1) & -16;
1621 index = put_elf_sym(s1->dynsym, offset, esym->st_size,
1622 esym->st_info, 0, bss_section->sh_num,
1623 name);
1624 /* Ensure R_COPY works for weak symbol aliases */
1625 if (ELFW(ST_BIND)(esym->st_info) == STB_WEAK) {
1626 for_each_elem(s1->dynsymtab_section, 1, dynsym, ElfW(Sym)) {
1627 if ((dynsym->st_value == esym->st_value)
1628 && (ELFW(ST_BIND)(dynsym->st_info) == STB_GLOBAL)) {
1629 char *dynname = (char *) s1->dynsymtab_section->link->data
1630 + dynsym->st_name;
1631 put_elf_sym(s1->dynsym, offset, dynsym->st_size,
1632 dynsym->st_info, 0,
1633 bss_section->sh_num, dynname);
1634 break;
1638 put_elf_reloc(s1->dynsym, bss_section,
1639 offset, R_COPY, index);
1640 offset += esym->st_size;
1641 bss_section->data_offset = offset;
1643 } else {
1644 /* STB_WEAK undefined symbols are accepted */
1645 /* XXX: _fp_hw seems to be part of the ABI, so we ignore it */
1646 if (ELFW(ST_BIND)(sym->st_info) == STB_WEAK ||
1647 !strcmp(name, "_fp_hw")) {
1648 } else {
1649 tcc_error_noabort("undefined symbol '%s'", name);
1652 } else if (s1->rdynamic && ELFW(ST_BIND)(sym->st_info) != STB_LOCAL) {
1653 /* if -rdynamic option, then export all non local symbols */
1654 name = (char *) symtab_section->link->data + sym->st_name;
1655 put_elf_sym(s1->dynsym, sym->st_value, sym->st_size, sym->st_info,
1656 0, sym->st_shndx, name);
1661 /* Bind symbols of libraries: export non local symbols of executable that
1662 resolve undefined symbols of shared libraries */
1663 static void bind_libs_dynsyms(TCCState *s1)
1665 const char *name;
1666 int sym_index;
1667 ElfW(Sym) *sym, *esym;
1669 /* now look at unresolved dynamic symbols and export
1670 corresponding symbol */
1671 for_each_elem(s1->dynsymtab_section, 1, esym, ElfW(Sym)) {
1672 if (esym->st_shndx == SHN_UNDEF) {
1673 name = (char *) s1->dynsymtab_section->link->data + esym->st_name;
1674 sym_index = find_elf_sym(symtab_section, name);
1675 if (sym_index) {
1676 /* XXX: avoid adding a symbol if already present because of
1677 -rdynamic ? */
1678 sym = &((ElfW(Sym) *)symtab_section->data)[sym_index];
1679 put_elf_sym(s1->dynsym, sym->st_value, sym->st_size,
1680 sym->st_info, 0, sym->st_shndx, name);
1681 } else {
1682 /* weak symbols can stay undefined */
1683 if (ELFW(ST_BIND)(esym->st_info) != STB_WEAK)
1684 tcc_warning("undefined dynamic symbol '%s'", name);
1690 /* Export all non local symbols (for shared libraries) */
1691 static void export_global_syms(TCCState *s1)
1693 int nb_syms, dynindex, index;
1694 const char *name;
1695 ElfW(Sym) *sym;
1697 nb_syms = symtab_section->data_offset / sizeof(ElfW(Sym));
1698 s1->symtab_to_dynsym = tcc_mallocz(sizeof(int) * nb_syms);
1699 for_each_elem(symtab_section, 1, sym, ElfW(Sym)) {
1700 if (ELFW(ST_BIND)(sym->st_info) != STB_LOCAL) {
1701 #if defined(TCC_OUTPUT_DLL_WITH_PLT)
1702 int type = ELFW(ST_TYPE)(sym->st_info);
1703 if ((type == STT_FUNC || type == STT_GNU_IFUNC)
1704 && sym->st_shndx == SHN_UNDEF) {
1705 int visibility = ELFW(ST_BIND)(sym->st_info);
1706 put_got_entry(s1, R_JMP_SLOT, sym->st_size,
1707 ELFW(ST_INFO)(visibility, STT_FUNC),
1708 sym - (ElfW(Sym) *) symtab_section->data);
1709 } else if (type == STT_OBJECT) {
1710 put_got_entry(s1, R_X86_64_GLOB_DAT, sym->st_size, sym->st_info,
1711 sym - (ElfW(Sym) *) symtab_section->data);
1712 } else
1713 #endif
1715 name = (char *) symtab_section->link->data + sym->st_name;
1716 dynindex = put_elf_sym(s1->dynsym, sym->st_value, sym->st_size,
1717 sym->st_info, 0, sym->st_shndx, name);
1718 index = sym - (ElfW(Sym) *) symtab_section->data;
1719 s1->symtab_to_dynsym[index] = dynindex;
1725 /* relocate the PLT: compute addresses and offsets in the PLT now that final
1726 address for PLT and GOT are known (see fill_program_header) */
1727 static void relocate_plt(TCCState *s1)
1729 uint8_t *p, *p_end;
1731 p = s1->plt->data;
1732 p_end = p + s1->plt->data_offset;
1733 if (p < p_end) {
1734 #if defined(TCC_TARGET_I386)
1735 put32(p + 2, get32(p + 2) + s1->got->sh_addr);
1736 put32(p + 8, get32(p + 8) + s1->got->sh_addr);
1737 p += 16;
1738 while (p < p_end) {
1739 put32(p + 2, get32(p + 2) + s1->got->sh_addr);
1740 p += 16;
1742 #elif defined(TCC_TARGET_X86_64)
1743 int x = s1->got->sh_addr - s1->plt->sh_addr - 6;
1744 put32(p + 2, get32(p + 2) + x);
1745 put32(p + 8, get32(p + 8) + x - 6);
1746 p += 16;
1747 while (p < p_end) {
1748 put32(p + 2, get32(p + 2) + x + s1->plt->data - p);
1749 p += 16;
1751 #elif defined(TCC_TARGET_ARM)
1752 int x;
1753 x=s1->got->sh_addr - s1->plt->sh_addr - 12;
1754 p += 16;
1755 while (p < p_end) {
1756 if (get32(p) == 0x46c04778) /* PLT Thumb stub present */
1757 p += 4;
1758 put32(p + 12, x + get32(p + 12) + s1->plt->data - p);
1759 p += 16;
1761 #elif defined(TCC_TARGET_C67)
1762 /* XXX: TODO */
1763 #else
1764 #error unsupported CPU
1765 #endif
1769 /* Allocate strings for section names and decide if an unallocated section
1770 should be output.
1772 NOTE: the strsec section comes last, so its size is also correct ! */
1773 static void alloc_sec_names(TCCState *s1, int file_type, Section *strsec)
1775 int i;
1776 Section *s;
1778 /* Allocate strings for section names */
1779 for(i = 1; i < s1->nb_sections; i++) {
1780 s = s1->sections[i];
1781 s->sh_name = put_elf_str(strsec, s->name);
1782 /* when generating a DLL, we include relocations but we may
1783 patch them */
1784 if (file_type == TCC_OUTPUT_DLL &&
1785 s->sh_type == SHT_RELX &&
1786 !(s->sh_flags & SHF_ALLOC)) {
1787 /* gr: avoid bogus relocs for empty (debug) sections */
1788 if (s1->sections[s->sh_info]->sh_flags & SHF_ALLOC)
1789 prepare_dynamic_rel(s1, s);
1790 else if (s1->do_debug)
1791 s->sh_size = s->data_offset;
1792 } else if (s1->do_debug ||
1793 file_type == TCC_OUTPUT_OBJ ||
1794 (s->sh_flags & SHF_ALLOC) ||
1795 i == (s1->nb_sections - 1)) {
1796 /* we output all sections if debug or object file */
1797 s->sh_size = s->data_offset;
1802 /* Info to be copied in dynamic section */
1803 struct dyn_inf {
1804 Section *dynamic;
1805 Section *dynstr;
1806 unsigned long dyn_rel_off;
1807 addr_t rel_addr;
1808 addr_t rel_size;
1809 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
1810 addr_t bss_addr;
1811 addr_t bss_size;
1812 #endif
1815 /* Assign sections to segments and decide how are sections laid out when loaded
1816 in memory. This function also fills corresponding program headers. */
1817 static int layout_sections(TCCState *s1, ElfW(Phdr) *phdr, int phnum,
1818 Section *interp, struct dyn_inf *dyninf,
1819 int *sec_order)
1821 int i, j, k, file_type, sh_order_index, file_offset;
1822 long long tmp;
1823 addr_t addr;
1824 ElfW(Phdr) *ph;
1825 Section *s;
1827 file_type = s1->output_type;
1828 sh_order_index = 1;
1829 if (s1->output_format == TCC_OUTPUT_FORMAT_ELF)
1830 file_offset = sizeof(ElfW(Ehdr)) + phnum * sizeof(ElfW(Phdr));
1831 else
1832 file_offset = 0;
1834 if (phnum > 0) {
1835 if (s1->has_text_addr) {
1836 int a_offset, p_offset;
1837 addr = s1->text_addr;
1838 /* we ensure that (addr % ELF_PAGE_SIZE) == file_offset %
1839 ELF_PAGE_SIZE */
1840 a_offset = (int) (addr & (s1->section_align - 1));
1841 p_offset = file_offset & (s1->section_align - 1);
1842 if (a_offset < p_offset)
1843 a_offset += s1->section_align;
1844 file_offset += (a_offset - p_offset);
1845 } else {
1846 if (file_type == TCC_OUTPUT_DLL)
1847 addr = 0;
1848 else
1849 addr = ELF_START_ADDR;
1850 /* compute address after headers */
1851 addr += (file_offset & (s1->section_align - 1));
1854 ph = &phdr[0];
1855 /* Leave one program headers for the program interpreter and one for
1856 the program header table itself if needed. These are done later as
1857 they require section layout to be done first. */
1858 if (interp)
1859 ph += 1 + HAVE_PHDR;
1861 /* dynamic relocation table information, for .dynamic section */
1862 dyninf->rel_addr = dyninf->rel_size = 0;
1863 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
1864 dyninf->bss_addr = dyninf->bss_size = 0;
1865 #endif
1867 for(j = 0; j < 2; j++) {
1868 ph->p_type = PT_LOAD;
1869 if (j == 0)
1870 ph->p_flags = PF_R | PF_X;
1871 else
1872 ph->p_flags = PF_R | PF_W;
1873 ph->p_align = s1->section_align;
1875 /* Decide the layout of sections loaded in memory. This must
1876 be done before program headers are filled since they contain
1877 info about the layout. We do the following ordering: interp,
1878 symbol tables, relocations, progbits, nobits */
1879 /* XXX: do faster and simpler sorting */
1880 for(k = 0; k < 5; k++) {
1881 for(i = 1; i < s1->nb_sections; i++) {
1882 s = s1->sections[i];
1883 /* compute if section should be included */
1884 if (j == 0) {
1885 if ((s->sh_flags & (SHF_ALLOC | SHF_WRITE)) !=
1886 SHF_ALLOC)
1887 continue;
1888 } else {
1889 if ((s->sh_flags & (SHF_ALLOC | SHF_WRITE)) !=
1890 (SHF_ALLOC | SHF_WRITE))
1891 continue;
1893 if (s == interp) {
1894 if (k != 0)
1895 continue;
1896 } else if (s->sh_type == SHT_DYNSYM ||
1897 s->sh_type == SHT_STRTAB ||
1898 s->sh_type == SHT_HASH) {
1899 if (k != 1)
1900 continue;
1901 } else if (s->sh_type == SHT_RELX) {
1902 if (k != 2)
1903 continue;
1904 } else if (s->sh_type == SHT_NOBITS) {
1905 if (k != 4)
1906 continue;
1907 } else {
1908 if (k != 3)
1909 continue;
1911 sec_order[sh_order_index++] = i;
1913 /* section matches: we align it and add its size */
1914 tmp = addr;
1915 addr = (addr + s->sh_addralign - 1) &
1916 ~(s->sh_addralign - 1);
1917 file_offset += (int) ( addr - tmp );
1918 s->sh_offset = file_offset;
1919 s->sh_addr = addr;
1921 /* update program header infos */
1922 if (ph->p_offset == 0) {
1923 ph->p_offset = file_offset;
1924 ph->p_vaddr = addr;
1925 ph->p_paddr = ph->p_vaddr;
1927 /* update dynamic relocation infos */
1928 if (s->sh_type == SHT_RELX) {
1929 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
1930 if (!strcmp(strsec->data + s->sh_name, ".rel.got")) {
1931 dyninf->rel_addr = addr;
1932 dyninf->rel_size += s->sh_size; /* XXX only first rel. */
1934 if (!strcmp(strsec->data + s->sh_name, ".rel.bss")) {
1935 dyninf->bss_addr = addr;
1936 dyninf->bss_size = s->sh_size; /* XXX only first rel. */
1938 #else
1939 if (dyninf->rel_size == 0)
1940 dyninf->rel_addr = addr;
1941 dyninf->rel_size += s->sh_size;
1942 #endif
1944 addr += s->sh_size;
1945 if (s->sh_type != SHT_NOBITS)
1946 file_offset += s->sh_size;
1949 ph->p_filesz = file_offset - ph->p_offset;
1950 ph->p_memsz = addr - ph->p_vaddr;
1951 ph++;
1952 if (j == 0) {
1953 if (s1->output_format == TCC_OUTPUT_FORMAT_ELF) {
1954 /* if in the middle of a page, we duplicate the page in
1955 memory so that one copy is RX and the other is RW */
1956 if ((addr & (s1->section_align - 1)) != 0)
1957 addr += s1->section_align;
1958 } else {
1959 addr = (addr + s1->section_align - 1) & ~(s1->section_align - 1);
1960 file_offset = (file_offset + s1->section_align - 1) &
1961 ~(s1->section_align - 1);
1967 /* all other sections come after */
1968 for(i = 1; i < s1->nb_sections; i++) {
1969 s = s1->sections[i];
1970 if (phnum > 0 && (s->sh_flags & SHF_ALLOC))
1971 continue;
1972 sec_order[sh_order_index++] = i;
1974 file_offset = (file_offset + s->sh_addralign - 1) &
1975 ~(s->sh_addralign - 1);
1976 s->sh_offset = file_offset;
1977 if (s->sh_type != SHT_NOBITS)
1978 file_offset += s->sh_size;
1981 return file_offset;
1984 static void fill_unloadable_phdr(ElfW(Phdr) *phdr, int phnum, Section *interp,
1985 Section *dynamic)
1987 ElfW(Phdr) *ph;
1989 /* if interpreter, then add corresponding program header */
1990 if (interp) {
1991 ph = &phdr[0];
1993 if (HAVE_PHDR)
1995 int len = phnum * sizeof(ElfW(Phdr));
1997 ph->p_type = PT_PHDR;
1998 ph->p_offset = sizeof(ElfW(Ehdr));
1999 ph->p_vaddr = interp->sh_addr - len;
2000 ph->p_paddr = ph->p_vaddr;
2001 ph->p_filesz = ph->p_memsz = len;
2002 ph->p_flags = PF_R | PF_X;
2003 ph->p_align = 4; /* interp->sh_addralign; */
2004 ph++;
2007 ph->p_type = PT_INTERP;
2008 ph->p_offset = interp->sh_offset;
2009 ph->p_vaddr = interp->sh_addr;
2010 ph->p_paddr = ph->p_vaddr;
2011 ph->p_filesz = interp->sh_size;
2012 ph->p_memsz = interp->sh_size;
2013 ph->p_flags = PF_R;
2014 ph->p_align = interp->sh_addralign;
2017 /* if dynamic section, then add corresponding program header */
2018 if (dynamic) {
2019 ph = &phdr[phnum - 1];
2021 ph->p_type = PT_DYNAMIC;
2022 ph->p_offset = dynamic->sh_offset;
2023 ph->p_vaddr = dynamic->sh_addr;
2024 ph->p_paddr = ph->p_vaddr;
2025 ph->p_filesz = dynamic->sh_size;
2026 ph->p_memsz = dynamic->sh_size;
2027 ph->p_flags = PF_R | PF_W;
2028 ph->p_align = dynamic->sh_addralign;
2032 /* Fill the dynamic section with tags describing the address and size of
2033 sections */
2034 static void fill_dynamic(TCCState *s1, struct dyn_inf *dyninf)
2036 Section *dynamic;
2038 dynamic = dyninf->dynamic;
2040 /* put dynamic section entries */
2041 dynamic->data_offset = dyninf->dyn_rel_off;
2042 put_dt(dynamic, DT_HASH, s1->dynsym->hash->sh_addr);
2043 put_dt(dynamic, DT_STRTAB, dyninf->dynstr->sh_addr);
2044 put_dt(dynamic, DT_SYMTAB, s1->dynsym->sh_addr);
2045 put_dt(dynamic, DT_STRSZ, dyninf->dynstr->data_offset);
2046 put_dt(dynamic, DT_SYMENT, sizeof(ElfW(Sym)));
2047 #ifdef TCC_TARGET_X86_64
2048 put_dt(dynamic, DT_RELA, dyninf->rel_addr);
2049 put_dt(dynamic, DT_RELASZ, dyninf->rel_size);
2050 put_dt(dynamic, DT_RELAENT, sizeof(ElfW_Rel));
2051 #else
2052 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
2053 put_dt(dynamic, DT_PLTGOT, s1->got->sh_addr);
2054 put_dt(dynamic, DT_PLTRELSZ, dyninf->rel_size);
2055 put_dt(dynamic, DT_JMPREL, dyninf->rel_addr);
2056 put_dt(dynamic, DT_PLTREL, DT_REL);
2057 put_dt(dynamic, DT_REL, dyninf->bss_addr);
2058 put_dt(dynamic, DT_RELSZ, dyninf->bss_size);
2059 #else
2060 put_dt(dynamic, DT_REL, dyninf->rel_addr);
2061 put_dt(dynamic, DT_RELSZ, dyninf->rel_size);
2062 put_dt(dynamic, DT_RELENT, sizeof(ElfW_Rel));
2063 #endif
2064 #endif
2065 if (s1->do_debug)
2066 put_dt(dynamic, DT_DEBUG, 0);
2067 put_dt(dynamic, DT_NULL, 0);
2070 /* Relocate remaining sections and symbols (that is those not related to
2071 dynamic linking) */
2072 static int final_sections_reloc(TCCState *s1)
2074 int i;
2075 Section *s;
2077 relocate_syms(s1, 0);
2079 if (s1->nb_errors != 0)
2080 return -1;
2082 /* relocate sections */
2083 /* XXX: ignore sections with allocated relocations ? */
2084 for(i = 1; i < s1->nb_sections; i++) {
2085 s = s1->sections[i];
2086 if (s->reloc && s != s1->got)
2087 relocate_section(s1, s);
2090 /* relocate relocation entries if the relocation tables are
2091 allocated in the executable */
2092 for(i = 1; i < s1->nb_sections; i++) {
2093 s = s1->sections[i];
2094 if ((s->sh_flags & SHF_ALLOC) &&
2095 s->sh_type == SHT_RELX) {
2096 relocate_rel(s1, s);
2099 return 0;
2102 /* Create an ELF file on disk.
2103 This function handle ELF specific layout requirements */
2104 static void tcc_output_elf(TCCState *s1, FILE *f, int phnum, ElfW(Phdr) *phdr,
2105 int file_offset, int *sec_order)
2107 int i, shnum, offset, size, file_type;
2108 Section *s;
2109 ElfW(Ehdr) ehdr;
2110 ElfW(Shdr) shdr, *sh;
2112 file_type = s1->output_type;
2113 shnum = s1->nb_sections;
2115 memset(&ehdr, 0, sizeof(ehdr));
2117 if (phnum > 0) {
2118 ehdr.e_phentsize = sizeof(ElfW(Phdr));
2119 ehdr.e_phnum = phnum;
2120 ehdr.e_phoff = sizeof(ElfW(Ehdr));
2123 /* align to 4 */
2124 file_offset = (file_offset + 3) & -4;
2126 /* fill header */
2127 ehdr.e_ident[0] = ELFMAG0;
2128 ehdr.e_ident[1] = ELFMAG1;
2129 ehdr.e_ident[2] = ELFMAG2;
2130 ehdr.e_ident[3] = ELFMAG3;
2131 ehdr.e_ident[4] = ELFCLASSW;
2132 ehdr.e_ident[5] = ELFDATA2LSB;
2133 ehdr.e_ident[6] = EV_CURRENT;
2134 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
2135 ehdr.e_ident[EI_OSABI] = ELFOSABI_FREEBSD;
2136 #endif
2137 #ifdef TCC_TARGET_ARM
2138 #ifdef TCC_ARM_EABI
2139 ehdr.e_ident[EI_OSABI] = 0;
2140 ehdr.e_flags = EF_ARM_EABI_VER4;
2141 if (file_type == TCC_OUTPUT_EXE || file_type == TCC_OUTPUT_DLL)
2142 ehdr.e_flags |= EF_ARM_HASENTRY;
2143 if (s1->float_abi == ARM_HARD_FLOAT)
2144 ehdr.e_flags |= EF_ARM_VFP_FLOAT;
2145 else
2146 ehdr.e_flags |= EF_ARM_SOFT_FLOAT;
2147 #else
2148 ehdr.e_ident[EI_OSABI] = ELFOSABI_ARM;
2149 #endif
2150 #endif
2151 switch(file_type) {
2152 default:
2153 case TCC_OUTPUT_EXE:
2154 ehdr.e_type = ET_EXEC;
2155 ehdr.e_entry = get_elf_sym_addr(s1, "_start", 1);
2156 break;
2157 case TCC_OUTPUT_DLL:
2158 ehdr.e_type = ET_DYN;
2159 ehdr.e_entry = text_section->sh_addr; /* XXX: is it correct ? */
2160 break;
2161 case TCC_OUTPUT_OBJ:
2162 ehdr.e_type = ET_REL;
2163 break;
2165 ehdr.e_machine = EM_TCC_TARGET;
2166 ehdr.e_version = EV_CURRENT;
2167 ehdr.e_shoff = file_offset;
2168 ehdr.e_ehsize = sizeof(ElfW(Ehdr));
2169 ehdr.e_shentsize = sizeof(ElfW(Shdr));
2170 ehdr.e_shnum = shnum;
2171 ehdr.e_shstrndx = shnum - 1;
2173 fwrite(&ehdr, 1, sizeof(ElfW(Ehdr)), f);
2174 fwrite(phdr, 1, phnum * sizeof(ElfW(Phdr)), f);
2175 offset = sizeof(ElfW(Ehdr)) + phnum * sizeof(ElfW(Phdr));
2177 sort_syms(s1, symtab_section);
2178 for(i = 1; i < s1->nb_sections; i++) {
2179 s = s1->sections[sec_order[i]];
2180 if (s->sh_type != SHT_NOBITS) {
2181 if (s->sh_type == SHT_DYNSYM)
2182 patch_dynsym_undef(s1, s);
2183 while (offset < s->sh_offset) {
2184 fputc(0, f);
2185 offset++;
2187 size = s->sh_size;
2188 fwrite(s->data, 1, size, f);
2189 offset += size;
2193 /* output section headers */
2194 while (offset < ehdr.e_shoff) {
2195 fputc(0, f);
2196 offset++;
2199 for(i = 0; i < s1->nb_sections; i++) {
2200 sh = &shdr;
2201 memset(sh, 0, sizeof(ElfW(Shdr)));
2202 s = s1->sections[i];
2203 if (s) {
2204 sh->sh_name = s->sh_name;
2205 sh->sh_type = s->sh_type;
2206 sh->sh_flags = s->sh_flags;
2207 sh->sh_entsize = s->sh_entsize;
2208 sh->sh_info = s->sh_info;
2209 if (s->link)
2210 sh->sh_link = s->link->sh_num;
2211 sh->sh_addralign = s->sh_addralign;
2212 sh->sh_addr = s->sh_addr;
2213 sh->sh_offset = s->sh_offset;
2214 sh->sh_size = s->sh_size;
2216 fwrite(sh, 1, sizeof(ElfW(Shdr)), f);
2220 /* Write an elf, coff or "binary" file */
2221 static int tcc_write_elf_file(TCCState *s1, const char *filename, int phnum,
2222 ElfW(Phdr) *phdr, int file_offset, int *sec_order)
2224 int fd, mode, file_type;
2225 FILE *f;
2227 file_type = s1->output_type;
2228 if (file_type == TCC_OUTPUT_OBJ)
2229 mode = 0666;
2230 else
2231 mode = 0777;
2232 unlink(filename);
2233 fd = open(filename, O_WRONLY | O_CREAT | O_TRUNC | O_BINARY, mode);
2234 if (fd < 0) {
2235 tcc_error_noabort("could not write '%s'", filename);
2236 return -1;
2238 f = fdopen(fd, "wb");
2239 if (s1->verbose)
2240 printf("<- %s\n", filename);
2242 #ifdef TCC_TARGET_COFF
2243 if (s1->output_format == TCC_OUTPUT_FORMAT_COFF)
2244 tcc_output_coff(s1, f);
2245 else
2246 #endif
2247 if (s1->output_format == TCC_OUTPUT_FORMAT_ELF)
2248 tcc_output_elf(s1, f, phnum, phdr, file_offset, sec_order);
2249 else
2250 tcc_output_binary(s1, f, sec_order);
2251 fclose(f);
2253 return 0;
2256 /* Output an elf, coff or binary file */
2257 /* XXX: suppress unneeded sections */
2258 static int elf_output_file(TCCState *s1, const char *filename)
2260 int i, ret, phnum, shnum, file_type, file_offset, *sec_order;
2261 struct dyn_inf dyninf;
2262 ElfW(Phdr) *phdr;
2263 ElfW(Sym) *sym;
2264 Section *strsec, *interp, *dynamic, *dynstr;
2266 file_type = s1->output_type;
2267 s1->nb_errors = 0;
2269 /* if linking, also link in runtime libraries (libc, libgcc, etc.) */
2270 if (file_type != TCC_OUTPUT_OBJ) {
2271 tcc_add_runtime(s1);
2274 phdr = NULL;
2275 sec_order = NULL;
2276 interp = dynamic = dynstr = NULL; /* avoid warning */
2277 dyninf.dyn_rel_off = 0; /* avoid warning */
2279 if (file_type != TCC_OUTPUT_OBJ) {
2280 relocate_common_syms();
2282 tcc_add_linker_symbols(s1);
2284 if (!s1->static_link) {
2285 if (file_type == TCC_OUTPUT_EXE) {
2286 char *ptr;
2287 /* allow override the dynamic loader */
2288 const char *elfint = getenv("LD_SO");
2289 if (elfint == NULL)
2290 elfint = DEFAULT_ELFINTERP(s1);
2291 /* add interpreter section only if executable */
2292 interp = new_section(s1, ".interp", SHT_PROGBITS, SHF_ALLOC);
2293 interp->sh_addralign = 1;
2294 ptr = section_ptr_add(interp, 1 + strlen(elfint));
2295 strcpy(ptr, elfint);
2298 /* add dynamic symbol table */
2299 s1->dynsym = new_symtab(s1, ".dynsym", SHT_DYNSYM, SHF_ALLOC,
2300 ".dynstr",
2301 ".hash", SHF_ALLOC);
2302 dynstr = s1->dynsym->link;
2304 /* add dynamic section */
2305 dynamic = new_section(s1, ".dynamic", SHT_DYNAMIC,
2306 SHF_ALLOC | SHF_WRITE);
2307 dynamic->link = dynstr;
2308 dynamic->sh_entsize = sizeof(ElfW(Dyn));
2310 /* add PLT */
2311 s1->plt = new_section(s1, ".plt", SHT_PROGBITS,
2312 SHF_ALLOC | SHF_EXECINSTR);
2313 s1->plt->sh_entsize = 4;
2315 build_got(s1);
2317 if (file_type == TCC_OUTPUT_EXE) {
2318 bind_exe_dynsyms(s1);
2320 if (s1->nb_errors) {
2321 ret = -1;
2322 goto the_end;
2325 bind_libs_dynsyms(s1);
2326 } else /* shared library case: simply export all global symbols */
2327 export_global_syms(s1);
2329 build_got_entries(s1);
2331 /* add a list of needed dlls */
2332 for(i = 0; i < s1->nb_loaded_dlls; i++) {
2333 DLLReference *dllref = s1->loaded_dlls[i];
2334 if (dllref->level == 0)
2335 put_dt(dynamic, DT_NEEDED, put_elf_str(dynstr, dllref->name));
2338 if (s1->rpath)
2339 put_dt(dynamic, DT_RPATH, put_elf_str(dynstr, s1->rpath));
2341 /* XXX: currently, since we do not handle PIC code, we
2342 must relocate the readonly segments */
2343 if (file_type == TCC_OUTPUT_DLL) {
2344 if (s1->soname)
2345 put_dt(dynamic, DT_SONAME, put_elf_str(dynstr, s1->soname));
2346 put_dt(dynamic, DT_TEXTREL, 0);
2349 if (s1->symbolic)
2350 put_dt(dynamic, DT_SYMBOLIC, 0);
2352 /* add necessary space for other entries */
2353 dyninf.dyn_rel_off = dynamic->data_offset;
2354 dynamic->data_offset += sizeof(ElfW(Dyn)) * EXTRA_RELITEMS;
2355 } else {
2356 /* still need to build got entries in case of static link */
2357 build_got_entries(s1);
2361 /* we add a section for symbols */
2362 strsec = new_section(s1, ".shstrtab", SHT_STRTAB, 0);
2363 put_elf_str(strsec, "");
2365 /* compute number of sections */
2366 shnum = s1->nb_sections;
2368 /* this array is used to reorder sections in the output file */
2369 sec_order = tcc_malloc(sizeof(int) * shnum);
2370 sec_order[0] = 0;
2372 /* compute number of program headers */
2373 switch(file_type) {
2374 default:
2375 case TCC_OUTPUT_OBJ:
2376 phnum = 0;
2377 break;
2378 case TCC_OUTPUT_EXE:
2379 if (!s1->static_link)
2380 phnum = 4 + HAVE_PHDR;
2381 else
2382 phnum = 2;
2383 break;
2384 case TCC_OUTPUT_DLL:
2385 phnum = 3;
2386 break;
2389 /* Allocate strings for section names */
2390 alloc_sec_names(s1, file_type, strsec);
2392 /* allocate program segment headers */
2393 phdr = tcc_mallocz(phnum * sizeof(ElfW(Phdr)));
2395 /* compute section to program header mapping */
2396 file_offset = layout_sections(s1, phdr, phnum, interp, &dyninf, sec_order);
2398 /* Fill remaining program header and finalize relocation related to dynamic
2399 linking. */
2400 if (phnum > 0) {
2401 fill_unloadable_phdr(phdr, phnum, interp, dynamic);
2402 if (dynamic) {
2403 dyninf.dynamic = dynamic;
2404 dyninf.dynstr = dynstr;
2406 fill_dynamic(s1, &dyninf);
2408 /* put in GOT the dynamic section address and relocate PLT */
2409 put32(s1->got->data, dynamic->sh_addr);
2410 if (file_type == TCC_OUTPUT_EXE
2411 #if defined(TCC_OUTPUT_DLL_WITH_PLT)
2412 || file_type == TCC_OUTPUT_DLL
2413 #endif
2415 relocate_plt(s1);
2417 /* relocate symbols in .dynsym now that final addresses are known */
2418 for_each_elem(s1->dynsym, 1, sym, ElfW(Sym)) {
2419 /* relocate to PLT if symbol corresponds to a PLT entry */
2420 if (sym->st_shndx == SHN_UNDEF) {
2421 if (sym->st_value)
2422 sym->st_value += s1->plt->sh_addr;
2423 } else if (sym->st_shndx < SHN_LORESERVE) {
2424 /* do symbol relocation */
2425 sym->st_value += s1->sections[sym->st_shndx]->sh_addr;
2431 /* if building executable or DLL, then relocate each section
2432 except the GOT which is already relocated */
2433 if (file_type != TCC_OUTPUT_OBJ) {
2434 ret = final_sections_reloc(s1);
2435 if (ret)
2436 goto the_end;
2439 /* Perform relocation to GOT or PLT entries */
2440 if (file_type == TCC_OUTPUT_EXE && s1->static_link)
2441 fill_got(s1);
2443 /* Create the ELF file with name 'filename' */
2444 ret = tcc_write_elf_file(s1, filename, phnum, phdr, file_offset, sec_order);
2445 the_end:
2446 tcc_free(s1->symtab_to_dynsym);
2447 tcc_free(sec_order);
2448 tcc_free(phdr);
2449 tcc_free(s1->sym_attrs);
2450 s1->sym_attrs = NULL;
2451 return ret;
2454 LIBTCCAPI int tcc_output_file(TCCState *s, const char *filename)
2456 int ret;
2457 #ifdef TCC_TARGET_PE
2458 if (s->output_type != TCC_OUTPUT_OBJ) {
2459 ret = pe_output_file(s, filename);
2460 } else
2461 #endif
2462 ret = elf_output_file(s, filename);
2463 return ret;
2466 static void *load_data(int fd, unsigned long file_offset, unsigned long size)
2468 void *data;
2470 data = tcc_malloc(size);
2471 lseek(fd, file_offset, SEEK_SET);
2472 read(fd, data, size);
2473 return data;
2476 typedef struct SectionMergeInfo {
2477 Section *s; /* corresponding existing section */
2478 unsigned long offset; /* offset of the new section in the existing section */
2479 uint8_t new_section; /* true if section 's' was added */
2480 uint8_t link_once; /* true if link once section */
2481 } SectionMergeInfo;
2483 /* load an object file and merge it with current files */
2484 /* XXX: handle correctly stab (debug) info */
2485 ST_FUNC int tcc_load_object_file(TCCState *s1,
2486 int fd, unsigned long file_offset)
2488 ElfW(Ehdr) ehdr;
2489 ElfW(Shdr) *shdr, *sh;
2490 int size, i, j, offset, offseti, nb_syms, sym_index, ret;
2491 unsigned char *strsec, *strtab;
2492 int *old_to_new_syms;
2493 char *sh_name, *name;
2494 SectionMergeInfo *sm_table, *sm;
2495 ElfW(Sym) *sym, *symtab;
2496 ElfW_Rel *rel;
2497 Section *s;
2499 int stab_index;
2500 int stabstr_index;
2502 stab_index = stabstr_index = 0;
2504 if (read(fd, &ehdr, sizeof(ehdr)) != sizeof(ehdr))
2505 goto fail1;
2506 if (ehdr.e_ident[0] != ELFMAG0 ||
2507 ehdr.e_ident[1] != ELFMAG1 ||
2508 ehdr.e_ident[2] != ELFMAG2 ||
2509 ehdr.e_ident[3] != ELFMAG3)
2510 goto fail1;
2511 /* test if object file */
2512 if (ehdr.e_type != ET_REL)
2513 goto fail1;
2514 /* test CPU specific stuff */
2515 if (ehdr.e_ident[5] != ELFDATA2LSB ||
2516 ehdr.e_machine != EM_TCC_TARGET) {
2517 fail1:
2518 tcc_error_noabort("invalid object file");
2519 return -1;
2521 /* read sections */
2522 shdr = load_data(fd, file_offset + ehdr.e_shoff,
2523 sizeof(ElfW(Shdr)) * ehdr.e_shnum);
2524 sm_table = tcc_mallocz(sizeof(SectionMergeInfo) * ehdr.e_shnum);
2526 /* load section names */
2527 sh = &shdr[ehdr.e_shstrndx];
2528 strsec = load_data(fd, file_offset + sh->sh_offset, sh->sh_size);
2530 /* load symtab and strtab */
2531 old_to_new_syms = NULL;
2532 symtab = NULL;
2533 strtab = NULL;
2534 nb_syms = 0;
2535 for(i = 1; i < ehdr.e_shnum; i++) {
2536 sh = &shdr[i];
2537 if (sh->sh_type == SHT_SYMTAB) {
2538 if (symtab) {
2539 tcc_error_noabort("object must contain only one symtab");
2540 fail:
2541 ret = -1;
2542 goto the_end;
2544 nb_syms = sh->sh_size / sizeof(ElfW(Sym));
2545 symtab = load_data(fd, file_offset + sh->sh_offset, sh->sh_size);
2546 sm_table[i].s = symtab_section;
2548 /* now load strtab */
2549 sh = &shdr[sh->sh_link];
2550 strtab = load_data(fd, file_offset + sh->sh_offset, sh->sh_size);
2554 /* now examine each section and try to merge its content with the
2555 ones in memory */
2556 for(i = 1; i < ehdr.e_shnum; i++) {
2557 /* no need to examine section name strtab */
2558 if (i == ehdr.e_shstrndx)
2559 continue;
2560 sh = &shdr[i];
2561 sh_name = (char *) strsec + sh->sh_name;
2562 /* ignore sections types we do not handle */
2563 if (sh->sh_type != SHT_PROGBITS &&
2564 sh->sh_type != SHT_RELX &&
2565 #ifdef TCC_ARM_EABI
2566 sh->sh_type != SHT_ARM_EXIDX &&
2567 #endif
2568 sh->sh_type != SHT_NOBITS &&
2569 sh->sh_type != SHT_PREINIT_ARRAY &&
2570 sh->sh_type != SHT_INIT_ARRAY &&
2571 sh->sh_type != SHT_FINI_ARRAY &&
2572 strcmp(sh_name, ".stabstr")
2574 continue;
2575 if (sh->sh_addralign < 1)
2576 sh->sh_addralign = 1;
2577 /* find corresponding section, if any */
2578 for(j = 1; j < s1->nb_sections;j++) {
2579 s = s1->sections[j];
2580 if (!strcmp(s->name, sh_name)) {
2581 if (!strncmp(sh_name, ".gnu.linkonce",
2582 sizeof(".gnu.linkonce") - 1)) {
2583 /* if a 'linkonce' section is already present, we
2584 do not add it again. It is a little tricky as
2585 symbols can still be defined in
2586 it. */
2587 sm_table[i].link_once = 1;
2588 goto next;
2589 } else {
2590 goto found;
2594 /* not found: create new section */
2595 s = new_section(s1, sh_name, sh->sh_type, sh->sh_flags);
2596 /* take as much info as possible from the section. sh_link and
2597 sh_info will be updated later */
2598 s->sh_addralign = sh->sh_addralign;
2599 s->sh_entsize = sh->sh_entsize;
2600 sm_table[i].new_section = 1;
2601 found:
2602 if (sh->sh_type != s->sh_type) {
2603 tcc_error_noabort("invalid section type");
2604 goto fail;
2607 /* align start of section */
2608 offset = s->data_offset;
2610 if (0 == strcmp(sh_name, ".stab")) {
2611 stab_index = i;
2612 goto no_align;
2614 if (0 == strcmp(sh_name, ".stabstr")) {
2615 stabstr_index = i;
2616 goto no_align;
2619 size = sh->sh_addralign - 1;
2620 offset = (offset + size) & ~size;
2621 if (sh->sh_addralign > s->sh_addralign)
2622 s->sh_addralign = sh->sh_addralign;
2623 s->data_offset = offset;
2624 no_align:
2625 sm_table[i].offset = offset;
2626 sm_table[i].s = s;
2627 /* concatenate sections */
2628 size = sh->sh_size;
2629 if (sh->sh_type != SHT_NOBITS) {
2630 unsigned char *ptr;
2631 lseek(fd, file_offset + sh->sh_offset, SEEK_SET);
2632 ptr = section_ptr_add(s, size);
2633 read(fd, ptr, size);
2634 } else {
2635 s->data_offset += size;
2637 next: ;
2640 /* gr relocate stab strings */
2641 if (stab_index && stabstr_index) {
2642 Stab_Sym *a, *b;
2643 unsigned o;
2644 s = sm_table[stab_index].s;
2645 a = (Stab_Sym *)(s->data + sm_table[stab_index].offset);
2646 b = (Stab_Sym *)(s->data + s->data_offset);
2647 o = sm_table[stabstr_index].offset;
2648 while (a < b)
2649 a->n_strx += o, a++;
2652 /* second short pass to update sh_link and sh_info fields of new
2653 sections */
2654 for(i = 1; i < ehdr.e_shnum; i++) {
2655 s = sm_table[i].s;
2656 if (!s || !sm_table[i].new_section)
2657 continue;
2658 sh = &shdr[i];
2659 if (sh->sh_link > 0)
2660 s->link = sm_table[sh->sh_link].s;
2661 if (sh->sh_type == SHT_RELX) {
2662 s->sh_info = sm_table[sh->sh_info].s->sh_num;
2663 /* update backward link */
2664 s1->sections[s->sh_info]->reloc = s;
2667 sm = sm_table;
2669 /* resolve symbols */
2670 old_to_new_syms = tcc_mallocz(nb_syms * sizeof(int));
2672 sym = symtab + 1;
2673 for(i = 1; i < nb_syms; i++, sym++) {
2674 if (sym->st_shndx != SHN_UNDEF &&
2675 sym->st_shndx < SHN_LORESERVE) {
2676 sm = &sm_table[sym->st_shndx];
2677 if (sm->link_once) {
2678 /* if a symbol is in a link once section, we use the
2679 already defined symbol. It is very important to get
2680 correct relocations */
2681 if (ELFW(ST_BIND)(sym->st_info) != STB_LOCAL) {
2682 name = (char *) strtab + sym->st_name;
2683 sym_index = find_elf_sym(symtab_section, name);
2684 if (sym_index)
2685 old_to_new_syms[i] = sym_index;
2687 continue;
2689 /* if no corresponding section added, no need to add symbol */
2690 if (!sm->s)
2691 continue;
2692 /* convert section number */
2693 sym->st_shndx = sm->s->sh_num;
2694 /* offset value */
2695 sym->st_value += sm->offset;
2697 /* add symbol */
2698 name = (char *) strtab + sym->st_name;
2699 sym_index = add_elf_sym(symtab_section, sym->st_value, sym->st_size,
2700 sym->st_info, sym->st_other,
2701 sym->st_shndx, name);
2702 old_to_new_syms[i] = sym_index;
2705 /* third pass to patch relocation entries */
2706 for(i = 1; i < ehdr.e_shnum; i++) {
2707 s = sm_table[i].s;
2708 if (!s)
2709 continue;
2710 sh = &shdr[i];
2711 offset = sm_table[i].offset;
2712 switch(s->sh_type) {
2713 case SHT_RELX:
2714 /* take relocation offset information */
2715 offseti = sm_table[sh->sh_info].offset;
2716 for_each_elem(s, (offset / sizeof(*rel)), rel, ElfW_Rel) {
2717 int type;
2718 unsigned sym_index;
2719 /* convert symbol index */
2720 type = ELFW(R_TYPE)(rel->r_info);
2721 sym_index = ELFW(R_SYM)(rel->r_info);
2722 /* NOTE: only one symtab assumed */
2723 if (sym_index >= nb_syms)
2724 goto invalid_reloc;
2725 sym_index = old_to_new_syms[sym_index];
2726 /* ignore link_once in rel section. */
2727 if (!sym_index && !sm->link_once
2728 #ifdef TCC_TARGET_ARM
2729 && type != R_ARM_V4BX
2730 #endif
2732 invalid_reloc:
2733 tcc_error_noabort("Invalid relocation entry [%2d] '%s' @ %.8x",
2734 i, strsec + sh->sh_name, rel->r_offset);
2735 goto fail;
2737 rel->r_info = ELFW(R_INFO)(sym_index, type);
2738 /* offset the relocation offset */
2739 rel->r_offset += offseti;
2740 #ifdef TCC_TARGET_ARM
2741 /* Jumps and branches from a Thumb code to a PLT entry need
2742 special handling since PLT entries are ARM code.
2743 Unconditional bl instructions referencing PLT entries are
2744 handled by converting these instructions into blx
2745 instructions. Other case of instructions referencing a PLT
2746 entry require to add a Thumb stub before the PLT entry to
2747 switch to ARM mode. We set bit plt_thumb_stub of the
2748 attribute of a symbol to indicate such a case. */
2749 if (type == R_ARM_THM_JUMP24)
2750 alloc_sym_attr(s1, sym_index)->plt_thumb_stub = 1;
2751 #endif
2753 break;
2754 default:
2755 break;
2759 ret = 0;
2760 the_end:
2761 tcc_free(symtab);
2762 tcc_free(strtab);
2763 tcc_free(old_to_new_syms);
2764 tcc_free(sm_table);
2765 tcc_free(strsec);
2766 tcc_free(shdr);
2767 return ret;
2770 typedef struct ArchiveHeader {
2771 char ar_name[16]; /* name of this member */
2772 char ar_date[12]; /* file mtime */
2773 char ar_uid[6]; /* owner uid; printed as decimal */
2774 char ar_gid[6]; /* owner gid; printed as decimal */
2775 char ar_mode[8]; /* file mode, printed as octal */
2776 char ar_size[10]; /* file size, printed as decimal */
2777 char ar_fmag[2]; /* should contain ARFMAG */
2778 } ArchiveHeader;
2780 static int get_be32(const uint8_t *b)
2782 return b[3] | (b[2] << 8) | (b[1] << 16) | (b[0] << 24);
2785 /* load only the objects which resolve undefined symbols */
2786 static int tcc_load_alacarte(TCCState *s1, int fd, int size)
2788 int i, bound, nsyms, sym_index, off, ret;
2789 uint8_t *data;
2790 const char *ar_names, *p;
2791 const uint8_t *ar_index;
2792 ElfW(Sym) *sym;
2794 data = tcc_malloc(size);
2795 if (read(fd, data, size) != size)
2796 goto fail;
2797 nsyms = get_be32(data);
2798 ar_index = data + 4;
2799 ar_names = (char *) ar_index + nsyms * 4;
2801 do {
2802 bound = 0;
2803 for(p = ar_names, i = 0; i < nsyms; i++, p += strlen(p)+1) {
2804 sym_index = find_elf_sym(symtab_section, p);
2805 if(sym_index) {
2806 sym = &((ElfW(Sym) *)symtab_section->data)[sym_index];
2807 if(sym->st_shndx == SHN_UNDEF) {
2808 off = get_be32(ar_index + i * 4) + sizeof(ArchiveHeader);
2809 ++bound;
2810 lseek(fd, off, SEEK_SET);
2811 if(tcc_load_object_file(s1, fd, off) < 0) {
2812 fail:
2813 ret = -1;
2814 goto the_end;
2819 } while(bound);
2820 ret = 0;
2821 the_end:
2822 tcc_free(data);
2823 return ret;
2826 /* load a '.a' file */
2827 ST_FUNC int tcc_load_archive(TCCState *s1, int fd)
2829 ArchiveHeader hdr;
2830 char ar_size[11];
2831 char ar_name[17];
2832 char magic[8];
2833 int size, len, i;
2834 unsigned long file_offset;
2836 /* skip magic which was already checked */
2837 read(fd, magic, sizeof(magic));
2839 for(;;) {
2840 len = read(fd, &hdr, sizeof(hdr));
2841 if (len == 0)
2842 break;
2843 if (len != sizeof(hdr)) {
2844 tcc_error_noabort("invalid archive");
2845 return -1;
2847 memcpy(ar_size, hdr.ar_size, sizeof(hdr.ar_size));
2848 ar_size[sizeof(hdr.ar_size)] = '\0';
2849 size = strtol(ar_size, NULL, 0);
2850 memcpy(ar_name, hdr.ar_name, sizeof(hdr.ar_name));
2851 for(i = sizeof(hdr.ar_name) - 1; i >= 0; i--) {
2852 if (ar_name[i] != ' ')
2853 break;
2855 ar_name[i + 1] = '\0';
2856 file_offset = lseek(fd, 0, SEEK_CUR);
2857 /* align to even */
2858 size = (size + 1) & ~1;
2859 if (!strcmp(ar_name, "/")) {
2860 /* coff symbol table : we handle it */
2861 if(s1->alacarte_link)
2862 return tcc_load_alacarte(s1, fd, size);
2863 } else if (!strcmp(ar_name, "//") ||
2864 !strcmp(ar_name, "__.SYMDEF") ||
2865 !strcmp(ar_name, "__.SYMDEF/") ||
2866 !strcmp(ar_name, "ARFILENAMES/")) {
2867 /* skip symbol table or archive names */
2868 } else {
2869 if (tcc_load_object_file(s1, fd, file_offset) < 0)
2870 return -1;
2872 lseek(fd, file_offset + size, SEEK_SET);
2874 return 0;
2877 #ifndef TCC_TARGET_PE
2878 /* load a DLL and all referenced DLLs. 'level = 0' means that the DLL
2879 is referenced by the user (so it should be added as DT_NEEDED in
2880 the generated ELF file) */
2881 ST_FUNC int tcc_load_dll(TCCState *s1, int fd, const char *filename, int level)
2883 ElfW(Ehdr) ehdr;
2884 ElfW(Shdr) *shdr, *sh, *sh1;
2885 int i, j, nb_syms, nb_dts, sym_bind, ret;
2886 ElfW(Sym) *sym, *dynsym;
2887 ElfW(Dyn) *dt, *dynamic;
2888 unsigned char *dynstr;
2889 const char *name, *soname;
2890 DLLReference *dllref;
2892 read(fd, &ehdr, sizeof(ehdr));
2894 /* test CPU specific stuff */
2895 if (ehdr.e_ident[5] != ELFDATA2LSB ||
2896 ehdr.e_machine != EM_TCC_TARGET) {
2897 tcc_error_noabort("bad architecture");
2898 return -1;
2901 /* read sections */
2902 shdr = load_data(fd, ehdr.e_shoff, sizeof(ElfW(Shdr)) * ehdr.e_shnum);
2904 /* load dynamic section and dynamic symbols */
2905 nb_syms = 0;
2906 nb_dts = 0;
2907 dynamic = NULL;
2908 dynsym = NULL; /* avoid warning */
2909 dynstr = NULL; /* avoid warning */
2910 for(i = 0, sh = shdr; i < ehdr.e_shnum; i++, sh++) {
2911 switch(sh->sh_type) {
2912 case SHT_DYNAMIC:
2913 nb_dts = sh->sh_size / sizeof(ElfW(Dyn));
2914 dynamic = load_data(fd, sh->sh_offset, sh->sh_size);
2915 break;
2916 case SHT_DYNSYM:
2917 nb_syms = sh->sh_size / sizeof(ElfW(Sym));
2918 dynsym = load_data(fd, sh->sh_offset, sh->sh_size);
2919 sh1 = &shdr[sh->sh_link];
2920 dynstr = load_data(fd, sh1->sh_offset, sh1->sh_size);
2921 break;
2922 default:
2923 break;
2927 /* compute the real library name */
2928 soname = tcc_basename(filename);
2930 for(i = 0, dt = dynamic; i < nb_dts; i++, dt++) {
2931 if (dt->d_tag == DT_SONAME) {
2932 soname = (char *) dynstr + dt->d_un.d_val;
2936 /* if the dll is already loaded, do not load it */
2937 for(i = 0; i < s1->nb_loaded_dlls; i++) {
2938 dllref = s1->loaded_dlls[i];
2939 if (!strcmp(soname, dllref->name)) {
2940 /* but update level if needed */
2941 if (level < dllref->level)
2942 dllref->level = level;
2943 ret = 0;
2944 goto the_end;
2948 /* add the dll and its level */
2949 dllref = tcc_mallocz(sizeof(DLLReference) + strlen(soname));
2950 dllref->level = level;
2951 strcpy(dllref->name, soname);
2952 dynarray_add((void ***)&s1->loaded_dlls, &s1->nb_loaded_dlls, dllref);
2954 /* add dynamic symbols in dynsym_section */
2955 for(i = 1, sym = dynsym + 1; i < nb_syms; i++, sym++) {
2956 sym_bind = ELFW(ST_BIND)(sym->st_info);
2957 if (sym_bind == STB_LOCAL)
2958 continue;
2959 name = (char *) dynstr + sym->st_name;
2960 add_elf_sym(s1->dynsymtab_section, sym->st_value, sym->st_size,
2961 sym->st_info, sym->st_other, sym->st_shndx, name);
2964 /* load all referenced DLLs */
2965 for(i = 0, dt = dynamic; i < nb_dts; i++, dt++) {
2966 switch(dt->d_tag) {
2967 case DT_NEEDED:
2968 name = (char *) dynstr + dt->d_un.d_val;
2969 for(j = 0; j < s1->nb_loaded_dlls; j++) {
2970 dllref = s1->loaded_dlls[j];
2971 if (!strcmp(name, dllref->name))
2972 goto already_loaded;
2974 if (tcc_add_dll(s1, name, AFF_REFERENCED_DLL) < 0) {
2975 tcc_error_noabort("referenced dll '%s' not found", name);
2976 ret = -1;
2977 goto the_end;
2979 already_loaded:
2980 break;
2983 ret = 0;
2984 the_end:
2985 tcc_free(dynstr);
2986 tcc_free(dynsym);
2987 tcc_free(dynamic);
2988 tcc_free(shdr);
2989 return ret;
2992 #define LD_TOK_NAME 256
2993 #define LD_TOK_EOF (-1)
2995 /* return next ld script token */
2996 static int ld_next(TCCState *s1, char *name, int name_size)
2998 int c;
2999 char *q;
3001 redo:
3002 switch(ch) {
3003 case ' ':
3004 case '\t':
3005 case '\f':
3006 case '\v':
3007 case '\r':
3008 case '\n':
3009 inp();
3010 goto redo;
3011 case '/':
3012 minp();
3013 if (ch == '*') {
3014 file->buf_ptr = parse_comment(file->buf_ptr);
3015 ch = file->buf_ptr[0];
3016 goto redo;
3017 } else {
3018 q = name;
3019 *q++ = '/';
3020 goto parse_name;
3022 break;
3023 /* case 'a' ... 'z': */
3024 case 'a':
3025 case 'b':
3026 case 'c':
3027 case 'd':
3028 case 'e':
3029 case 'f':
3030 case 'g':
3031 case 'h':
3032 case 'i':
3033 case 'j':
3034 case 'k':
3035 case 'l':
3036 case 'm':
3037 case 'n':
3038 case 'o':
3039 case 'p':
3040 case 'q':
3041 case 'r':
3042 case 's':
3043 case 't':
3044 case 'u':
3045 case 'v':
3046 case 'w':
3047 case 'x':
3048 case 'y':
3049 case 'z':
3050 /* case 'A' ... 'z': */
3051 case 'A':
3052 case 'B':
3053 case 'C':
3054 case 'D':
3055 case 'E':
3056 case 'F':
3057 case 'G':
3058 case 'H':
3059 case 'I':
3060 case 'J':
3061 case 'K':
3062 case 'L':
3063 case 'M':
3064 case 'N':
3065 case 'O':
3066 case 'P':
3067 case 'Q':
3068 case 'R':
3069 case 'S':
3070 case 'T':
3071 case 'U':
3072 case 'V':
3073 case 'W':
3074 case 'X':
3075 case 'Y':
3076 case 'Z':
3077 case '_':
3078 case '\\':
3079 case '.':
3080 case '$':
3081 case '~':
3082 q = name;
3083 parse_name:
3084 for(;;) {
3085 if (!((ch >= 'a' && ch <= 'z') ||
3086 (ch >= 'A' && ch <= 'Z') ||
3087 (ch >= '0' && ch <= '9') ||
3088 strchr("/.-_+=$:\\,~", ch)))
3089 break;
3090 if ((q - name) < name_size - 1) {
3091 *q++ = ch;
3093 minp();
3095 *q = '\0';
3096 c = LD_TOK_NAME;
3097 break;
3098 case CH_EOF:
3099 c = LD_TOK_EOF;
3100 break;
3101 default:
3102 c = ch;
3103 inp();
3104 break;
3106 return c;
3109 static int ld_add_file(TCCState *s1, const char filename[])
3111 int ret;
3113 ret = tcc_add_file_internal(s1, filename, 0);
3114 if (ret)
3115 ret = tcc_add_dll(s1, filename, 0);
3116 return ret;
3119 static inline int new_undef_syms(void)
3121 int ret = 0;
3122 ret = new_undef_sym;
3123 new_undef_sym = 0;
3124 return ret;
3127 static int ld_add_file_list(TCCState *s1, const char *cmd, int as_needed)
3129 char filename[1024], libname[1024];
3130 int t, group, nblibs = 0, ret = 0;
3131 char **libs = NULL;
3133 group = !strcmp(cmd, "GROUP");
3134 if (!as_needed)
3135 new_undef_syms();
3136 t = ld_next(s1, filename, sizeof(filename));
3137 if (t != '(')
3138 expect("(");
3139 t = ld_next(s1, filename, sizeof(filename));
3140 for(;;) {
3141 libname[0] = '\0';
3142 if (t == LD_TOK_EOF) {
3143 tcc_error_noabort("unexpected end of file");
3144 ret = -1;
3145 goto lib_parse_error;
3146 } else if (t == ')') {
3147 break;
3148 } else if (t == '-') {
3149 t = ld_next(s1, filename, sizeof(filename));
3150 if ((t != LD_TOK_NAME) || (filename[0] != 'l')) {
3151 tcc_error_noabort("library name expected");
3152 ret = -1;
3153 goto lib_parse_error;
3155 pstrcpy(libname, sizeof libname, &filename[1]);
3156 if (s1->static_link) {
3157 snprintf(filename, sizeof filename, "lib%s.a", libname);
3158 } else {
3159 snprintf(filename, sizeof filename, "lib%s.so", libname);
3161 } else if (t != LD_TOK_NAME) {
3162 tcc_error_noabort("filename expected");
3163 ret = -1;
3164 goto lib_parse_error;
3166 if (!strcmp(filename, "AS_NEEDED")) {
3167 ret = ld_add_file_list(s1, cmd, 1);
3168 if (ret)
3169 goto lib_parse_error;
3170 } else {
3171 /* TODO: Implement AS_NEEDED support. Ignore it for now */
3172 if (!as_needed) {
3173 ret = ld_add_file(s1, filename);
3174 if (ret)
3175 goto lib_parse_error;
3176 if (group) {
3177 /* Add the filename *and* the libname to avoid future conversions */
3178 dynarray_add((void ***) &libs, &nblibs, tcc_strdup(filename));
3179 if (libname[0] != '\0')
3180 dynarray_add((void ***) &libs, &nblibs, tcc_strdup(libname));
3184 t = ld_next(s1, filename, sizeof(filename));
3185 if (t == ',') {
3186 t = ld_next(s1, filename, sizeof(filename));
3189 if (group && !as_needed) {
3190 while (new_undef_syms()) {
3191 int i;
3193 for (i = 0; i < nblibs; i ++)
3194 ld_add_file(s1, libs[i]);
3197 lib_parse_error:
3198 dynarray_reset(&libs, &nblibs);
3199 return ret;
3202 /* interpret a subset of GNU ldscripts to handle the dummy libc.so
3203 files */
3204 ST_FUNC int tcc_load_ldscript(TCCState *s1)
3206 char cmd[64];
3207 char filename[1024];
3208 int t, ret;
3210 ch = file->buf_ptr[0];
3211 ch = handle_eob();
3212 for(;;) {
3213 t = ld_next(s1, cmd, sizeof(cmd));
3214 if (t == LD_TOK_EOF)
3215 return 0;
3216 else if (t != LD_TOK_NAME)
3217 return -1;
3218 if (!strcmp(cmd, "INPUT") ||
3219 !strcmp(cmd, "GROUP")) {
3220 ret = ld_add_file_list(s1, cmd, 0);
3221 if (ret)
3222 return ret;
3223 } else if (!strcmp(cmd, "OUTPUT_FORMAT") ||
3224 !strcmp(cmd, "TARGET")) {
3225 /* ignore some commands */
3226 t = ld_next(s1, cmd, sizeof(cmd));
3227 if (t != '(')
3228 expect("(");
3229 for(;;) {
3230 t = ld_next(s1, filename, sizeof(filename));
3231 if (t == LD_TOK_EOF) {
3232 tcc_error_noabort("unexpected end of file");
3233 return -1;
3234 } else if (t == ')') {
3235 break;
3238 } else {
3239 return -1;
3242 return 0;
3244 #endif /* !TCC_TARGET_PE */