A possible fix for the memory leak reported by valgrind when running tcctest.c with...
[tinycc.git] / tccelf.c
blob6bde7d5f56901d81c848ac05c16bd76eeed581fa
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 (!find_elf_sym(s1->dynsym, name))
1046 need_plt_entry = 1;
1047 else
1048 return;
1049 offset = sym->st_value;
1050 #if defined(TCC_TARGET_I386) || defined(TCC_TARGET_X86_64)
1051 if (need_plt_entry) {
1052 Section *plt;
1053 uint8_t *p;
1054 int modrm;
1056 #if defined(TCC_OUTPUT_DLL_WITH_PLT)
1057 modrm = 0x25;
1058 #else
1059 /* if we build a DLL, we add a %ebx offset */
1060 if (s1->output_type == TCC_OUTPUT_DLL)
1061 modrm = 0xa3;
1062 else
1063 modrm = 0x25;
1064 #endif
1066 /* add a PLT entry */
1067 plt = s1->plt;
1068 if (plt->data_offset == 0) {
1069 /* first plt entry */
1070 p = section_ptr_add(plt, 16);
1071 p[0] = 0xff; /* pushl got + PTR_SIZE */
1072 p[1] = modrm + 0x10;
1073 put32(p + 2, PTR_SIZE);
1074 p[6] = 0xff; /* jmp *(got + PTR_SIZE * 2) */
1075 p[7] = modrm;
1076 put32(p + 8, PTR_SIZE * 2);
1079 p = section_ptr_add(plt, 16);
1080 p[0] = 0xff; /* jmp *(got + x) */
1081 p[1] = modrm;
1082 put32(p + 2, s1->got->data_offset);
1083 p[6] = 0x68; /* push $xxx */
1084 put32(p + 7, (plt->data_offset - 32) >> 1);
1085 p[11] = 0xe9; /* jmp plt_start */
1086 put32(p + 12, -(plt->data_offset));
1088 /* the symbol is modified so that it will be relocated to
1089 the PLT */
1090 #if !defined(TCC_OUTPUT_DLL_WITH_PLT)
1091 if (s1->output_type == TCC_OUTPUT_EXE)
1092 #endif
1093 offset = plt->data_offset - 16;
1095 #elif defined(TCC_TARGET_ARM)
1096 if (need_plt_entry) {
1097 Section *plt;
1098 uint8_t *p;
1100 /* if we build a DLL, we add a %ebx offset */
1101 if (s1->output_type == TCC_OUTPUT_DLL)
1102 tcc_error("DLLs unimplemented!");
1104 /* add a PLT entry */
1105 plt = s1->plt;
1106 if (plt->data_offset == 0) {
1107 /* first plt entry */
1108 p = section_ptr_add(plt, 16);
1109 put32(p, 0xe52de004); /* push {lr} */
1110 put32(p+4, 0xe59fe010); /* ldr lr, [pc, #16] */
1111 put32(p+8, 0xe08fe00e); /* add lr, pc, lr */
1112 put32(p+12, 0xe5bef008); /* ldr pc, [lr, #8]! */
1115 if (s1->sym_attrs[sym_index].plt_thumb_stub) {
1116 p = section_ptr_add(plt, 20);
1117 put32(p, 0x4778); /* bx pc */
1118 put32(p+2, 0x46c0); /* nop */
1119 p += 4;
1120 } else
1121 p = section_ptr_add(plt, 16);
1122 put32(p, 0xe59fc004); /* ldr ip, [pc, #4] ; GOT entry offset */
1123 put32(p+4, 0xe08fc00c); /* add ip, pc, ip ; addr of GOT entry */
1124 put32(p+8, 0xe59cf000); /* ldr pc, [ip] ; jump to GOT entry */
1125 put32(p+12, s1->got->data_offset); /* GOT entry off once patched */
1127 /* the symbol is modified so that it will be relocated to
1128 the PLT */
1129 if (s1->output_type == TCC_OUTPUT_EXE)
1130 offset = plt->data_offset - 16;
1132 #elif defined(TCC_TARGET_C67)
1133 tcc_error("C67 got not implemented");
1134 #else
1135 #error unsupported CPU
1136 #endif
1137 index = put_elf_sym(s1->dynsym, offset,
1138 size, info, 0, sym->st_shndx, name);
1139 if (got_entry_present) {
1140 put_elf_reloc(s1->dynsym, s1->got,
1141 s1->sym_attrs[sym_index].got_offset,
1142 reloc_type, index);
1143 return;
1146 /* put a got entry */
1147 put_elf_reloc(s1->dynsym, s1->got,
1148 s1->got->data_offset,
1149 reloc_type, index);
1151 ptr = section_ptr_add(s1->got, PTR_SIZE);
1152 *ptr = 0;
1155 /* build GOT and PLT entries */
1156 ST_FUNC void build_got_entries(TCCState *s1)
1158 Section *s;
1159 ElfW_Rel *rel;
1160 ElfW(Sym) *sym;
1161 int i, type, reloc_type, sym_index;
1163 for(i = 1; i < s1->nb_sections; i++) {
1164 s = s1->sections[i];
1165 if (s->sh_type != SHT_RELX)
1166 continue;
1167 /* no need to handle got relocations */
1168 if (s->link != symtab_section)
1169 continue;
1170 for_each_elem(s, 0, rel, ElfW_Rel) {
1171 type = ELFW(R_TYPE)(rel->r_info);
1172 switch(type) {
1173 #if defined(TCC_TARGET_I386)
1174 case R_386_GOT32:
1175 case R_386_GOTOFF:
1176 case R_386_GOTPC:
1177 case R_386_PLT32:
1178 if (!s1->got)
1179 build_got(s1);
1180 if (type == R_386_GOT32 || type == R_386_PLT32) {
1181 sym_index = ELFW(R_SYM)(rel->r_info);
1182 sym = &((ElfW(Sym) *)symtab_section->data)[sym_index];
1183 /* look at the symbol got offset. If none, then add one */
1184 if (type == R_386_GOT32)
1185 reloc_type = R_386_GLOB_DAT;
1186 else
1187 reloc_type = R_386_JMP_SLOT;
1188 put_got_entry(s1, reloc_type, sym->st_size, sym->st_info,
1189 sym_index);
1191 break;
1192 #elif defined(TCC_TARGET_ARM)
1193 case R_ARM_GOT32:
1194 case R_ARM_GOTOFF:
1195 case R_ARM_GOTPC:
1196 case R_ARM_PLT32:
1197 if (!s1->got)
1198 build_got(s1);
1199 if (type == R_ARM_GOT32 || type == R_ARM_PLT32) {
1200 sym_index = ELFW(R_SYM)(rel->r_info);
1201 sym = &((ElfW(Sym) *)symtab_section->data)[sym_index];
1202 /* look at the symbol got offset. If none, then add one */
1203 if (type == R_ARM_GOT32)
1204 reloc_type = R_ARM_GLOB_DAT;
1205 else
1206 reloc_type = R_ARM_JUMP_SLOT;
1207 put_got_entry(s1, reloc_type, sym->st_size, sym->st_info,
1208 sym_index);
1210 break;
1211 case R_ARM_THM_JUMP24:
1212 sym_index = ELFW(R_SYM)(rel->r_info);
1213 sym = &((ElfW(Sym) *)symtab_section->data)[sym_index];
1214 /* We are relocating a jump from thumb code to arm code */
1215 if (sym->st_shndx != SHN_UNDEF && !(sym->st_value & 1)) {
1216 int index;
1217 uint8_t *p;
1218 char *name, buf[1024];
1219 Section *text_section;
1221 name = (char *) symtab_section->link->data + sym->st_name;
1222 text_section = s1->sections[sym->st_shndx];
1223 /* Modify reloc to target a thumb stub to switch to ARM */
1224 snprintf(buf, sizeof(buf), "%s_from_thumb", name);
1225 index = put_elf_sym(symtab_section,
1226 text_section->data_offset + 1,
1227 sym->st_size, sym->st_info, 0,
1228 sym->st_shndx, buf);
1229 rel->r_info = ELFW(R_INFO)(index, type);
1230 /* Create a thumb stub fonction to switch to ARM mode */
1231 put_elf_reloc(symtab_section, text_section,
1232 text_section->data_offset + 4, R_ARM_JUMP24,
1233 sym_index);
1234 p = section_ptr_add(text_section, 8);
1235 put32(p, 0x4778); /* bx pc */
1236 put32(p+2, 0x46c0); /* nop */
1237 put32(p+4, 0xeafffffe); /* b $sym */
1239 #elif defined(TCC_TARGET_C67)
1240 case R_C60_GOT32:
1241 case R_C60_GOTOFF:
1242 case R_C60_GOTPC:
1243 case R_C60_PLT32:
1244 if (!s1->got)
1245 build_got(s1);
1246 if (type == R_C60_GOT32 || type == R_C60_PLT32) {
1247 sym_index = ELFW(R_SYM)(rel->r_info);
1248 sym = &((ElfW(Sym) *)symtab_section->data)[sym_index];
1249 /* look at the symbol got offset. If none, then add one */
1250 if (type == R_C60_GOT32)
1251 reloc_type = R_C60_GLOB_DAT;
1252 else
1253 reloc_type = R_C60_JMP_SLOT;
1254 put_got_entry(s1, reloc_type, sym->st_size, sym->st_info,
1255 sym_index);
1257 break;
1258 #elif defined(TCC_TARGET_X86_64)
1259 case R_X86_64_GOT32:
1260 case R_X86_64_GOTTPOFF:
1261 case R_X86_64_GOTPCREL:
1262 case R_X86_64_PLT32:
1263 if (!s1->got)
1264 build_got(s1);
1265 if (type == R_X86_64_GOT32 || type == R_X86_64_GOTPCREL ||
1266 type == R_X86_64_PLT32) {
1267 sym_index = ELFW(R_SYM)(rel->r_info);
1268 sym = &((ElfW(Sym) *)symtab_section->data)[sym_index];
1269 /* look at the symbol got offset. If none, then add one */
1270 if (type == R_X86_64_GOT32 || type == R_X86_64_GOTPCREL)
1271 reloc_type = R_X86_64_GLOB_DAT;
1272 else
1273 reloc_type = R_X86_64_JUMP_SLOT;
1274 put_got_entry(s1, reloc_type, sym->st_size, sym->st_info,
1275 sym_index);
1277 break;
1278 #else
1279 #error unsupported CPU
1280 #endif
1281 default:
1282 break;
1288 ST_FUNC Section *new_symtab(TCCState *s1,
1289 const char *symtab_name, int sh_type, int sh_flags,
1290 const char *strtab_name,
1291 const char *hash_name, int hash_sh_flags)
1293 Section *symtab, *strtab, *hash;
1294 int *ptr, nb_buckets;
1296 symtab = new_section(s1, symtab_name, sh_type, sh_flags);
1297 symtab->sh_entsize = sizeof(ElfW(Sym));
1298 strtab = new_section(s1, strtab_name, SHT_STRTAB, sh_flags);
1299 put_elf_str(strtab, "");
1300 symtab->link = strtab;
1301 put_elf_sym(symtab, 0, 0, 0, 0, 0, NULL);
1303 nb_buckets = 1;
1305 hash = new_section(s1, hash_name, SHT_HASH, hash_sh_flags);
1306 hash->sh_entsize = sizeof(int);
1307 symtab->hash = hash;
1308 hash->link = symtab;
1310 ptr = section_ptr_add(hash, (2 + nb_buckets + 1) * sizeof(int));
1311 ptr[0] = nb_buckets;
1312 ptr[1] = 1;
1313 memset(ptr + 2, 0, (nb_buckets + 1) * sizeof(int));
1314 return symtab;
1317 /* put dynamic tag */
1318 static void put_dt(Section *dynamic, int dt, addr_t val)
1320 ElfW(Dyn) *dyn;
1321 dyn = section_ptr_add(dynamic, sizeof(ElfW(Dyn)));
1322 dyn->d_tag = dt;
1323 dyn->d_un.d_val = val;
1326 static void add_init_array_defines(TCCState *s1, const char *section_name)
1328 Section *s;
1329 long end_offset;
1330 char sym_start[1024];
1331 char sym_end[1024];
1333 snprintf(sym_start, sizeof(sym_start), "__%s_start", section_name + 1);
1334 snprintf(sym_end, sizeof(sym_end), "__%s_end", section_name + 1);
1336 s = find_section(s1, section_name);
1337 if (!s) {
1338 end_offset = 0;
1339 s = data_section;
1340 } else {
1341 end_offset = s->data_offset;
1344 add_elf_sym(symtab_section,
1345 0, 0,
1346 ELFW(ST_INFO)(STB_GLOBAL, STT_NOTYPE), 0,
1347 s->sh_num, sym_start);
1348 add_elf_sym(symtab_section,
1349 end_offset, 0,
1350 ELFW(ST_INFO)(STB_GLOBAL, STT_NOTYPE), 0,
1351 s->sh_num, sym_end);
1354 ST_FUNC void tcc_add_bcheck(TCCState *s1)
1356 #ifdef CONFIG_TCC_BCHECK
1357 unsigned long *ptr;
1358 Section *init_section;
1359 unsigned char *pinit;
1360 int sym_index;
1362 if (0 == s1->do_bounds_check)
1363 return;
1365 /* XXX: add an object file to do that */
1366 ptr = section_ptr_add(bounds_section, sizeof(unsigned long));
1367 *ptr = 0;
1368 add_elf_sym(symtab_section, 0, 0,
1369 ELFW(ST_INFO)(STB_GLOBAL, STT_NOTYPE), 0,
1370 bounds_section->sh_num, "__bounds_start");
1371 #ifdef TCC_TARGET_I386
1372 if (s1->output_type != TCC_OUTPUT_MEMORY) {
1373 /* add 'call __bound_init()' in .init section */
1374 init_section = find_section(s1, ".init");
1375 pinit = section_ptr_add(init_section, 5);
1376 pinit[0] = 0xe8;
1377 put32(pinit + 1, -4);
1378 sym_index = find_elf_sym(symtab_section, "__bound_init");
1379 put_elf_reloc(symtab_section, init_section,
1380 init_section->data_offset - 4, R_386_PC32, sym_index);
1382 #endif
1383 #endif
1386 static inline int tcc_add_support(TCCState *s1, const char *filename)
1388 char buf[1024];
1389 snprintf(buf, sizeof(buf), "%s/%s", s1->tcc_lib_path, filename);
1390 return tcc_add_file(s1, buf);
1393 /* add tcc runtime libraries */
1394 ST_FUNC void tcc_add_runtime(TCCState *s1)
1396 /* add libc */
1397 if (!s1->nostdlib) {
1398 tcc_add_library(s1, "c");
1399 #ifdef CONFIG_USE_LIBGCC
1400 if (!s1->static_link) {
1401 tcc_add_file(s1, TCC_LIBGCC);
1402 tcc_add_support(s1, "libtcc1.a");
1403 } else
1404 tcc_add_support(s1, "libtcc1.a");
1405 #else
1406 tcc_add_support(s1, "libtcc1.a");
1407 #endif
1410 /* tcc_add_bcheck tries to relocate a call to __bound_init in _init so
1411 libtcc1.a must be loaded before for __bound_init to be defined and
1412 crtn.o must be loaded after to not finalize _init too early. */
1413 tcc_add_bcheck(s1);
1415 if (!s1->nostdlib) {
1416 /* add crt end if not memory output */
1417 if (s1->output_type != TCC_OUTPUT_MEMORY)
1418 tcc_add_crt(s1, "crtn.o");
1422 /* add various standard linker symbols (must be done after the
1423 sections are filled (for example after allocating common
1424 symbols)) */
1425 ST_FUNC void tcc_add_linker_symbols(TCCState *s1)
1427 char buf[1024];
1428 int i;
1429 Section *s;
1431 add_elf_sym(symtab_section,
1432 text_section->data_offset, 0,
1433 ELFW(ST_INFO)(STB_GLOBAL, STT_NOTYPE), 0,
1434 text_section->sh_num, "_etext");
1435 add_elf_sym(symtab_section,
1436 data_section->data_offset, 0,
1437 ELFW(ST_INFO)(STB_GLOBAL, STT_NOTYPE), 0,
1438 data_section->sh_num, "_edata");
1439 add_elf_sym(symtab_section,
1440 bss_section->data_offset, 0,
1441 ELFW(ST_INFO)(STB_GLOBAL, STT_NOTYPE), 0,
1442 bss_section->sh_num, "_end");
1443 /* horrible new standard ldscript defines */
1444 add_init_array_defines(s1, ".preinit_array");
1445 add_init_array_defines(s1, ".init_array");
1446 add_init_array_defines(s1, ".fini_array");
1448 /* add start and stop symbols for sections whose name can be
1449 expressed in C */
1450 for(i = 1; i < s1->nb_sections; i++) {
1451 s = s1->sections[i];
1452 if (s->sh_type == SHT_PROGBITS &&
1453 (s->sh_flags & SHF_ALLOC)) {
1454 const char *p;
1455 int ch;
1457 /* check if section name can be expressed in C */
1458 p = s->name;
1459 for(;;) {
1460 ch = *p;
1461 if (!ch)
1462 break;
1463 if (!isid(ch) && !isnum(ch))
1464 goto next_sec;
1465 p++;
1467 snprintf(buf, sizeof(buf), "__start_%s", s->name);
1468 add_elf_sym(symtab_section,
1469 0, 0,
1470 ELFW(ST_INFO)(STB_GLOBAL, STT_NOTYPE), 0,
1471 s->sh_num, buf);
1472 snprintf(buf, sizeof(buf), "__stop_%s", s->name);
1473 add_elf_sym(symtab_section,
1474 s->data_offset, 0,
1475 ELFW(ST_INFO)(STB_GLOBAL, STT_NOTYPE), 0,
1476 s->sh_num, buf);
1478 next_sec: ;
1482 static void tcc_output_binary(TCCState *s1, FILE *f,
1483 const int *sec_order)
1485 Section *s;
1486 int i, offset, size;
1488 offset = 0;
1489 for(i=1;i<s1->nb_sections;i++) {
1490 s = s1->sections[sec_order[i]];
1491 if (s->sh_type != SHT_NOBITS &&
1492 (s->sh_flags & SHF_ALLOC)) {
1493 while (offset < s->sh_offset) {
1494 fputc(0, f);
1495 offset++;
1497 size = s->sh_size;
1498 fwrite(s->data, 1, size, f);
1499 offset += size;
1504 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
1505 #define HAVE_PHDR 1
1506 #define EXTRA_RELITEMS 14
1508 /* move the relocation value from .dynsym to .got */
1509 void patch_dynsym_undef(TCCState *s1, Section *s)
1511 uint32_t *gotd = (void *)s1->got->data;
1512 ElfW(Sym) *sym;
1514 gotd += 3; /* dummy entries in .got */
1515 /* relocate symbols in .dynsym */
1516 for_each_elem(s, 1, sym, ElfW(Sym)) {
1517 if (sym->st_shndx == SHN_UNDEF) {
1518 *gotd++ = sym->st_value + 6; /* XXX 6 is magic ? */
1519 sym->st_value = 0;
1523 #else
1524 #define HAVE_PHDR 0
1525 #define EXTRA_RELITEMS 9
1527 /* zero plt offsets of weak symbols in .dynsym */
1528 void patch_dynsym_undef(TCCState *s1, Section *s)
1530 ElfW(Sym) *sym;
1532 for_each_elem(s, 1, sym, ElfW(Sym))
1533 if (sym->st_shndx == SHN_UNDEF && ELFW(ST_BIND)(sym->st_info) == STB_WEAK)
1534 sym->st_value = 0;
1536 #endif
1538 ST_FUNC void fill_got_entry(TCCState *s1, ElfW_Rel *rel)
1540 int sym_index = ELFW(R_SYM) (rel->r_info);
1541 ElfW(Sym) *sym = &((ElfW(Sym) *) symtab_section->data)[sym_index];
1542 unsigned long offset;
1544 if (sym_index >= s1->nb_sym_attrs)
1545 return;
1546 offset = s1->sym_attrs[sym_index].got_offset;
1547 section_reserve(s1->got, offset + PTR_SIZE);
1548 #ifdef TCC_TARGET_X86_64
1549 /* only works for x86-64 */
1550 put32(s1->got->data + offset + 4, sym->st_value >> 32);
1551 #endif
1552 put32(s1->got->data + offset, sym->st_value & 0xffffffff);
1555 /* Perform relocation to GOT or PLT entries */
1556 ST_FUNC void fill_got(TCCState *s1)
1558 Section *s;
1559 ElfW_Rel *rel;
1560 int i;
1562 for(i = 1; i < s1->nb_sections; i++) {
1563 s = s1->sections[i];
1564 if (s->sh_type != SHT_RELX)
1565 continue;
1566 /* no need to handle got relocations */
1567 if (s->link != symtab_section)
1568 continue;
1569 for_each_elem(s, 0, rel, ElfW_Rel) {
1570 switch (ELFW(R_TYPE) (rel->r_info)) {
1571 case R_X86_64_GOT32:
1572 case R_X86_64_GOTPCREL:
1573 case R_X86_64_PLT32:
1574 fill_got_entry(s1, rel);
1575 break;
1581 /* Bind symbols of executable: resolve undefined symbols from exported symbols
1582 in shared libraries and export non local defined symbols to shared libraries
1583 if -rdynamic switch was given on command line */
1584 static void bind_exe_dynsyms(TCCState *s1)
1586 const char *name;
1587 int sym_index, index;
1588 ElfW(Sym) *sym, *esym;
1589 int type;
1591 /* Resolve undefined symbols from dynamic symbols. When there is a match:
1592 - if STT_FUNC or STT_GNU_IFUNC symbol -> add it in PLT
1593 - if STT_OBJECT symbol -> add it in .bss section with suitable reloc */
1594 for_each_elem(symtab_section, 1, sym, ElfW(Sym)) {
1595 if (sym->st_shndx == SHN_UNDEF) {
1596 name = (char *) symtab_section->link->data + sym->st_name;
1597 sym_index = find_elf_sym(s1->dynsymtab_section, name);
1598 if (sym_index) {
1599 esym = &((ElfW(Sym) *)s1->dynsymtab_section->data)[sym_index];
1600 type = ELFW(ST_TYPE)(esym->st_info);
1601 if ((type == STT_FUNC) || (type == STT_GNU_IFUNC)) {
1602 /* Indirect functions shall have STT_FUNC type in executable
1603 * dynsym section. Indeed, a dlsym call following a lazy
1604 * resolution would pick the symbol value from the
1605 * executable dynsym entry which would contain the address
1606 * of the function wanted by the caller of dlsym instead of
1607 * the address of the function that would return that
1608 * address */
1609 put_got_entry(s1, R_JMP_SLOT, esym->st_size,
1610 ELFW(ST_INFO)(STB_GLOBAL,STT_FUNC),
1611 sym - (ElfW(Sym) *)symtab_section->data);
1612 } else if (type == STT_OBJECT) {
1613 unsigned long offset;
1614 ElfW(Sym) *dynsym;
1615 offset = bss_section->data_offset;
1616 /* XXX: which alignment ? */
1617 offset = (offset + 16 - 1) & -16;
1618 index = put_elf_sym(s1->dynsym, offset, esym->st_size,
1619 esym->st_info, 0, bss_section->sh_num,
1620 name);
1621 /* Ensure R_COPY works for weak symbol aliases */
1622 if (ELFW(ST_BIND)(esym->st_info) == STB_WEAK) {
1623 for_each_elem(s1->dynsymtab_section, 1, dynsym, ElfW(Sym)) {
1624 if ((dynsym->st_value == esym->st_value)
1625 && (ELFW(ST_BIND)(dynsym->st_info) == STB_GLOBAL)) {
1626 char *dynname = (char *) s1->dynsymtab_section->link->data
1627 + dynsym->st_name;
1628 put_elf_sym(s1->dynsym, offset, dynsym->st_size,
1629 dynsym->st_info, 0,
1630 bss_section->sh_num, dynname);
1631 break;
1635 put_elf_reloc(s1->dynsym, bss_section,
1636 offset, R_COPY, index);
1637 offset += esym->st_size;
1638 bss_section->data_offset = offset;
1640 } else {
1641 /* STB_WEAK undefined symbols are accepted */
1642 /* XXX: _fp_hw seems to be part of the ABI, so we ignore it */
1643 if (ELFW(ST_BIND)(sym->st_info) == STB_WEAK ||
1644 !strcmp(name, "_fp_hw")) {
1645 } else {
1646 tcc_error_noabort("undefined symbol '%s'", name);
1649 } else if (s1->rdynamic && ELFW(ST_BIND)(sym->st_info) != STB_LOCAL) {
1650 /* if -rdynamic option, then export all non local symbols */
1651 name = (char *) symtab_section->link->data + sym->st_name;
1652 put_elf_sym(s1->dynsym, sym->st_value, sym->st_size, sym->st_info,
1653 0, sym->st_shndx, name);
1658 /* Bind symbols of libraries: export non local symbols of executable that
1659 resolve undefined symbols of shared libraries */
1660 static void bind_libs_dynsyms(TCCState *s1)
1662 const char *name;
1663 int sym_index;
1664 ElfW(Sym) *sym, *esym;
1666 /* now look at unresolved dynamic symbols and export
1667 corresponding symbol */
1668 for_each_elem(s1->dynsymtab_section, 1, esym, ElfW(Sym)) {
1669 if (esym->st_shndx == SHN_UNDEF) {
1670 name = (char *) s1->dynsymtab_section->link->data + esym->st_name;
1671 sym_index = find_elf_sym(symtab_section, name);
1672 if (sym_index) {
1673 /* XXX: avoid adding a symbol if already present because of
1674 -rdynamic ? */
1675 sym = &((ElfW(Sym) *)symtab_section->data)[sym_index];
1676 put_elf_sym(s1->dynsym, sym->st_value, sym->st_size,
1677 sym->st_info, 0, sym->st_shndx, name);
1678 } else {
1679 /* weak symbols can stay undefined */
1680 if (ELFW(ST_BIND)(esym->st_info) != STB_WEAK)
1681 tcc_warning("undefined dynamic symbol '%s'", name);
1687 /* Export all non local symbols (for shared libraries) */
1688 static void export_global_syms(TCCState *s1)
1690 int nb_syms, dynindex, index;
1691 const char *name;
1692 ElfW(Sym) *sym;
1694 nb_syms = symtab_section->data_offset / sizeof(ElfW(Sym));
1695 s1->symtab_to_dynsym = tcc_mallocz(sizeof(int) * nb_syms);
1696 for_each_elem(symtab_section, 1, sym, ElfW(Sym)) {
1697 if (ELFW(ST_BIND)(sym->st_info) != STB_LOCAL) {
1698 #if defined(TCC_OUTPUT_DLL_WITH_PLT)
1699 int type = ELFW(ST_TYPE)(sym->st_info);
1700 if ((type == STT_FUNC || type == STT_GNU_IFUNC)
1701 && sym->st_shndx == SHN_UNDEF) {
1702 int visibility = ELFW(ST_BIND)(sym->st_info);
1703 put_got_entry(s1, R_JMP_SLOT, sym->st_size,
1704 ELFW(ST_INFO)(visibility, STT_FUNC),
1705 sym - (ElfW(Sym) *) symtab_section->data);
1706 } else if (type == STT_OBJECT) {
1707 put_got_entry(s1, R_X86_64_GLOB_DAT, sym->st_size, sym->st_info,
1708 sym - (ElfW(Sym) *) symtab_section->data);
1709 } else
1710 #endif
1712 name = (char *) symtab_section->link->data + sym->st_name;
1713 dynindex = put_elf_sym(s1->dynsym, sym->st_value, sym->st_size,
1714 sym->st_info, 0, sym->st_shndx, name);
1715 index = sym - (ElfW(Sym) *) symtab_section->data;
1716 s1->symtab_to_dynsym[index] = dynindex;
1722 /* relocate the PLT: compute addresses and offsets in the PLT now that final
1723 address for PLT and GOT are known (see fill_program_header) */
1724 static void relocate_plt(TCCState *s1)
1726 uint8_t *p, *p_end;
1728 p = s1->plt->data;
1729 p_end = p + s1->plt->data_offset;
1730 if (p < p_end) {
1731 #if defined(TCC_TARGET_I386)
1732 put32(p + 2, get32(p + 2) + s1->got->sh_addr);
1733 put32(p + 8, get32(p + 8) + s1->got->sh_addr);
1734 p += 16;
1735 while (p < p_end) {
1736 put32(p + 2, get32(p + 2) + s1->got->sh_addr);
1737 p += 16;
1739 #elif defined(TCC_TARGET_X86_64)
1740 int x = s1->got->sh_addr - s1->plt->sh_addr - 6;
1741 put32(p + 2, get32(p + 2) + x);
1742 put32(p + 8, get32(p + 8) + x - 6);
1743 p += 16;
1744 while (p < p_end) {
1745 put32(p + 2, get32(p + 2) + x + s1->plt->data - p);
1746 p += 16;
1748 #elif defined(TCC_TARGET_ARM)
1749 int x;
1750 x=s1->got->sh_addr - s1->plt->sh_addr - 12;
1751 p += 16;
1752 while (p < p_end) {
1753 if (get32(p) == 0x46c04778) /* PLT Thumb stub present */
1754 p += 4;
1755 put32(p + 12, x + get32(p + 12) + s1->plt->data - p);
1756 p += 16;
1758 #elif defined(TCC_TARGET_C67)
1759 /* XXX: TODO */
1760 #else
1761 #error unsupported CPU
1762 #endif
1766 /* Allocate strings for section names and decide if an unallocated section
1767 should be output.
1769 NOTE: the strsec section comes last, so its size is also correct ! */
1770 static void alloc_sec_names(TCCState *s1, int file_type, Section *strsec)
1772 int i;
1773 Section *s;
1775 /* Allocate strings for section names */
1776 for(i = 1; i < s1->nb_sections; i++) {
1777 s = s1->sections[i];
1778 s->sh_name = put_elf_str(strsec, s->name);
1779 /* when generating a DLL, we include relocations but we may
1780 patch them */
1781 if (file_type == TCC_OUTPUT_DLL &&
1782 s->sh_type == SHT_RELX &&
1783 !(s->sh_flags & SHF_ALLOC)) {
1784 /* gr: avoid bogus relocs for empty (debug) sections */
1785 if (s1->sections[s->sh_info]->sh_flags & SHF_ALLOC)
1786 prepare_dynamic_rel(s1, s);
1787 else if (s1->do_debug)
1788 s->sh_size = s->data_offset;
1789 } else if (s1->do_debug ||
1790 file_type == TCC_OUTPUT_OBJ ||
1791 (s->sh_flags & SHF_ALLOC) ||
1792 i == (s1->nb_sections - 1)) {
1793 /* we output all sections if debug or object file */
1794 s->sh_size = s->data_offset;
1799 /* Info to be copied in dynamic section */
1800 struct dyn_inf {
1801 Section *dynamic;
1802 Section *dynstr;
1803 unsigned long dyn_rel_off;
1804 addr_t rel_addr;
1805 addr_t rel_size;
1806 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
1807 addr_t bss_addr;
1808 addr_t bss_size;
1809 #endif
1812 /* Assign sections to segments and decide how are sections laid out when loaded
1813 in memory. This function also fills corresponding program headers. */
1814 static int layout_sections(TCCState *s1, ElfW(Phdr) *phdr, int phnum,
1815 Section *interp, struct dyn_inf *dyninf,
1816 int *sec_order)
1818 int i, j, k, file_type, sh_order_index, file_offset;
1819 long long tmp;
1820 addr_t addr;
1821 ElfW(Phdr) *ph;
1822 Section *s;
1824 file_type = s1->output_type;
1825 sh_order_index = 1;
1826 if (s1->output_format == TCC_OUTPUT_FORMAT_ELF)
1827 file_offset = sizeof(ElfW(Ehdr)) + phnum * sizeof(ElfW(Phdr));
1828 else
1829 file_offset = 0;
1831 if (phnum > 0) {
1832 if (s1->has_text_addr) {
1833 int a_offset, p_offset;
1834 addr = s1->text_addr;
1835 /* we ensure that (addr % ELF_PAGE_SIZE) == file_offset %
1836 ELF_PAGE_SIZE */
1837 a_offset = (int) (addr & (s1->section_align - 1));
1838 p_offset = file_offset & (s1->section_align - 1);
1839 if (a_offset < p_offset)
1840 a_offset += s1->section_align;
1841 file_offset += (a_offset - p_offset);
1842 } else {
1843 if (file_type == TCC_OUTPUT_DLL)
1844 addr = 0;
1845 else
1846 addr = ELF_START_ADDR;
1847 /* compute address after headers */
1848 addr += (file_offset & (s1->section_align - 1));
1851 ph = &phdr[0];
1852 /* Leave one program headers for the program interpreter and one for
1853 the program header table itself if needed. These are done later as
1854 they require section layout to be done first. */
1855 if (interp)
1856 ph += 1 + HAVE_PHDR;
1858 /* dynamic relocation table information, for .dynamic section */
1859 dyninf->rel_addr = dyninf->rel_size = 0;
1860 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
1861 dyninf->bss_addr = dyninf->bss_size = 0;
1862 #endif
1864 for(j = 0; j < 2; j++) {
1865 ph->p_type = PT_LOAD;
1866 if (j == 0)
1867 ph->p_flags = PF_R | PF_X;
1868 else
1869 ph->p_flags = PF_R | PF_W;
1870 ph->p_align = s1->section_align;
1872 /* Decide the layout of sections loaded in memory. This must
1873 be done before program headers are filled since they contain
1874 info about the layout. We do the following ordering: interp,
1875 symbol tables, relocations, progbits, nobits */
1876 /* XXX: do faster and simpler sorting */
1877 for(k = 0; k < 5; k++) {
1878 for(i = 1; i < s1->nb_sections; i++) {
1879 s = s1->sections[i];
1880 /* compute if section should be included */
1881 if (j == 0) {
1882 if ((s->sh_flags & (SHF_ALLOC | SHF_WRITE)) !=
1883 SHF_ALLOC)
1884 continue;
1885 } else {
1886 if ((s->sh_flags & (SHF_ALLOC | SHF_WRITE)) !=
1887 (SHF_ALLOC | SHF_WRITE))
1888 continue;
1890 if (s == interp) {
1891 if (k != 0)
1892 continue;
1893 } else if (s->sh_type == SHT_DYNSYM ||
1894 s->sh_type == SHT_STRTAB ||
1895 s->sh_type == SHT_HASH) {
1896 if (k != 1)
1897 continue;
1898 } else if (s->sh_type == SHT_RELX) {
1899 if (k != 2)
1900 continue;
1901 } else if (s->sh_type == SHT_NOBITS) {
1902 if (k != 4)
1903 continue;
1904 } else {
1905 if (k != 3)
1906 continue;
1908 sec_order[sh_order_index++] = i;
1910 /* section matches: we align it and add its size */
1911 tmp = addr;
1912 addr = (addr + s->sh_addralign - 1) &
1913 ~(s->sh_addralign - 1);
1914 file_offset += (int) ( addr - tmp );
1915 s->sh_offset = file_offset;
1916 s->sh_addr = addr;
1918 /* update program header infos */
1919 if (ph->p_offset == 0) {
1920 ph->p_offset = file_offset;
1921 ph->p_vaddr = addr;
1922 ph->p_paddr = ph->p_vaddr;
1924 /* update dynamic relocation infos */
1925 if (s->sh_type == SHT_RELX) {
1926 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
1927 if (!strcmp(strsec->data + s->sh_name, ".rel.got")) {
1928 dyninf->rel_addr = addr;
1929 dyninf->rel_size += s->sh_size; /* XXX only first rel. */
1931 if (!strcmp(strsec->data + s->sh_name, ".rel.bss")) {
1932 dyninf->bss_addr = addr;
1933 dyninf->bss_size = s->sh_size; /* XXX only first rel. */
1935 #else
1936 if (dyninf->rel_size == 0)
1937 dyninf->rel_addr = addr;
1938 dyninf->rel_size += s->sh_size;
1939 #endif
1941 addr += s->sh_size;
1942 if (s->sh_type != SHT_NOBITS)
1943 file_offset += s->sh_size;
1946 ph->p_filesz = file_offset - ph->p_offset;
1947 ph->p_memsz = addr - ph->p_vaddr;
1948 ph++;
1949 if (j == 0) {
1950 if (s1->output_format == TCC_OUTPUT_FORMAT_ELF) {
1951 /* if in the middle of a page, we duplicate the page in
1952 memory so that one copy is RX and the other is RW */
1953 if ((addr & (s1->section_align - 1)) != 0)
1954 addr += s1->section_align;
1955 } else {
1956 addr = (addr + s1->section_align - 1) & ~(s1->section_align - 1);
1957 file_offset = (file_offset + s1->section_align - 1) &
1958 ~(s1->section_align - 1);
1964 /* all other sections come after */
1965 for(i = 1; i < s1->nb_sections; i++) {
1966 s = s1->sections[i];
1967 if (phnum > 0 && (s->sh_flags & SHF_ALLOC))
1968 continue;
1969 sec_order[sh_order_index++] = i;
1971 file_offset = (file_offset + s->sh_addralign - 1) &
1972 ~(s->sh_addralign - 1);
1973 s->sh_offset = file_offset;
1974 if (s->sh_type != SHT_NOBITS)
1975 file_offset += s->sh_size;
1978 return file_offset;
1981 static void fill_unloadable_phdr(ElfW(Phdr) *phdr, int phnum, Section *interp,
1982 Section *dynamic)
1984 ElfW(Phdr) *ph;
1986 /* if interpreter, then add corresponding program header */
1987 if (interp) {
1988 ph = &phdr[0];
1990 if (HAVE_PHDR)
1992 int len = phnum * sizeof(ElfW(Phdr));
1994 ph->p_type = PT_PHDR;
1995 ph->p_offset = sizeof(ElfW(Ehdr));
1996 ph->p_vaddr = interp->sh_addr - len;
1997 ph->p_paddr = ph->p_vaddr;
1998 ph->p_filesz = ph->p_memsz = len;
1999 ph->p_flags = PF_R | PF_X;
2000 ph->p_align = 4; /* interp->sh_addralign; */
2001 ph++;
2004 ph->p_type = PT_INTERP;
2005 ph->p_offset = interp->sh_offset;
2006 ph->p_vaddr = interp->sh_addr;
2007 ph->p_paddr = ph->p_vaddr;
2008 ph->p_filesz = interp->sh_size;
2009 ph->p_memsz = interp->sh_size;
2010 ph->p_flags = PF_R;
2011 ph->p_align = interp->sh_addralign;
2014 /* if dynamic section, then add corresponding program header */
2015 if (dynamic) {
2016 ph = &phdr[phnum - 1];
2018 ph->p_type = PT_DYNAMIC;
2019 ph->p_offset = dynamic->sh_offset;
2020 ph->p_vaddr = dynamic->sh_addr;
2021 ph->p_paddr = ph->p_vaddr;
2022 ph->p_filesz = dynamic->sh_size;
2023 ph->p_memsz = dynamic->sh_size;
2024 ph->p_flags = PF_R | PF_W;
2025 ph->p_align = dynamic->sh_addralign;
2029 /* Fill the dynamic section with tags describing the address and size of
2030 sections */
2031 static void fill_dynamic(TCCState *s1, struct dyn_inf *dyninf)
2033 Section *dynamic;
2035 dynamic = dyninf->dynamic;
2037 /* put dynamic section entries */
2038 dynamic->data_offset = dyninf->dyn_rel_off;
2039 put_dt(dynamic, DT_HASH, s1->dynsym->hash->sh_addr);
2040 put_dt(dynamic, DT_STRTAB, dyninf->dynstr->sh_addr);
2041 put_dt(dynamic, DT_SYMTAB, s1->dynsym->sh_addr);
2042 put_dt(dynamic, DT_STRSZ, dyninf->dynstr->data_offset);
2043 put_dt(dynamic, DT_SYMENT, sizeof(ElfW(Sym)));
2044 #ifdef TCC_TARGET_X86_64
2045 put_dt(dynamic, DT_RELA, dyninf->rel_addr);
2046 put_dt(dynamic, DT_RELASZ, dyninf->rel_size);
2047 put_dt(dynamic, DT_RELAENT, sizeof(ElfW_Rel));
2048 #else
2049 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
2050 put_dt(dynamic, DT_PLTGOT, s1->got->sh_addr);
2051 put_dt(dynamic, DT_PLTRELSZ, dyninf->rel_size);
2052 put_dt(dynamic, DT_JMPREL, dyninf->rel_addr);
2053 put_dt(dynamic, DT_PLTREL, DT_REL);
2054 put_dt(dynamic, DT_REL, dyninf->bss_addr);
2055 put_dt(dynamic, DT_RELSZ, dyninf->bss_size);
2056 #else
2057 put_dt(dynamic, DT_REL, dyninf->rel_addr);
2058 put_dt(dynamic, DT_RELSZ, dyninf->rel_size);
2059 put_dt(dynamic, DT_RELENT, sizeof(ElfW_Rel));
2060 #endif
2061 #endif
2062 if (s1->do_debug)
2063 put_dt(dynamic, DT_DEBUG, 0);
2064 put_dt(dynamic, DT_NULL, 0);
2067 /* Relocate remaining sections and symbols (that is those not related to
2068 dynamic linking) */
2069 static int final_sections_reloc(TCCState *s1)
2071 int i;
2072 Section *s;
2074 relocate_syms(s1, 0);
2076 if (s1->nb_errors != 0)
2077 return -1;
2079 /* relocate sections */
2080 /* XXX: ignore sections with allocated relocations ? */
2081 for(i = 1; i < s1->nb_sections; i++) {
2082 s = s1->sections[i];
2083 if (s->reloc && s != s1->got)
2084 relocate_section(s1, s);
2087 /* relocate relocation entries if the relocation tables are
2088 allocated in the executable */
2089 for(i = 1; i < s1->nb_sections; i++) {
2090 s = s1->sections[i];
2091 if ((s->sh_flags & SHF_ALLOC) &&
2092 s->sh_type == SHT_RELX) {
2093 relocate_rel(s1, s);
2096 return 0;
2099 /* Create an ELF file on disk.
2100 This function handle ELF specific layout requirements */
2101 static void tcc_output_elf(TCCState *s1, FILE *f, int phnum, ElfW(Phdr) *phdr,
2102 int file_offset, int *sec_order)
2104 int i, shnum, offset, size, file_type;
2105 Section *s;
2106 ElfW(Ehdr) ehdr;
2107 ElfW(Shdr) shdr, *sh;
2109 file_type = s1->output_type;
2110 shnum = s1->nb_sections;
2112 memset(&ehdr, 0, sizeof(ehdr));
2114 if (phnum > 0) {
2115 ehdr.e_phentsize = sizeof(ElfW(Phdr));
2116 ehdr.e_phnum = phnum;
2117 ehdr.e_phoff = sizeof(ElfW(Ehdr));
2120 /* align to 4 */
2121 file_offset = (file_offset + 3) & -4;
2123 /* fill header */
2124 ehdr.e_ident[0] = ELFMAG0;
2125 ehdr.e_ident[1] = ELFMAG1;
2126 ehdr.e_ident[2] = ELFMAG2;
2127 ehdr.e_ident[3] = ELFMAG3;
2128 ehdr.e_ident[4] = ELFCLASSW;
2129 ehdr.e_ident[5] = ELFDATA2LSB;
2130 ehdr.e_ident[6] = EV_CURRENT;
2131 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
2132 ehdr.e_ident[EI_OSABI] = ELFOSABI_FREEBSD;
2133 #endif
2134 #ifdef TCC_TARGET_ARM
2135 #ifdef TCC_ARM_EABI
2136 ehdr.e_ident[EI_OSABI] = 0;
2137 ehdr.e_flags = EF_ARM_EABI_VER4;
2138 if (file_type == TCC_OUTPUT_EXE || file_type == TCC_OUTPUT_DLL)
2139 ehdr.e_flags |= EF_ARM_HASENTRY;
2140 if (s1->float_abi == ARM_HARD_FLOAT)
2141 ehdr.e_flags |= EF_ARM_VFP_FLOAT;
2142 else
2143 ehdr.e_flags |= EF_ARM_SOFT_FLOAT;
2144 #else
2145 ehdr.e_ident[EI_OSABI] = ELFOSABI_ARM;
2146 #endif
2147 #endif
2148 switch(file_type) {
2149 default:
2150 case TCC_OUTPUT_EXE:
2151 ehdr.e_type = ET_EXEC;
2152 ehdr.e_entry = get_elf_sym_addr(s1, "_start", 1);
2153 break;
2154 case TCC_OUTPUT_DLL:
2155 ehdr.e_type = ET_DYN;
2156 ehdr.e_entry = text_section->sh_addr; /* XXX: is it correct ? */
2157 break;
2158 case TCC_OUTPUT_OBJ:
2159 ehdr.e_type = ET_REL;
2160 break;
2162 ehdr.e_machine = EM_TCC_TARGET;
2163 ehdr.e_version = EV_CURRENT;
2164 ehdr.e_shoff = file_offset;
2165 ehdr.e_ehsize = sizeof(ElfW(Ehdr));
2166 ehdr.e_shentsize = sizeof(ElfW(Shdr));
2167 ehdr.e_shnum = shnum;
2168 ehdr.e_shstrndx = shnum - 1;
2170 fwrite(&ehdr, 1, sizeof(ElfW(Ehdr)), f);
2171 fwrite(phdr, 1, phnum * sizeof(ElfW(Phdr)), f);
2172 offset = sizeof(ElfW(Ehdr)) + phnum * sizeof(ElfW(Phdr));
2174 sort_syms(s1, symtab_section);
2175 for(i = 1; i < s1->nb_sections; i++) {
2176 s = s1->sections[sec_order[i]];
2177 if (s->sh_type != SHT_NOBITS) {
2178 if (s->sh_type == SHT_DYNSYM)
2179 patch_dynsym_undef(s1, s);
2180 while (offset < s->sh_offset) {
2181 fputc(0, f);
2182 offset++;
2184 size = s->sh_size;
2185 fwrite(s->data, 1, size, f);
2186 offset += size;
2190 /* output section headers */
2191 while (offset < ehdr.e_shoff) {
2192 fputc(0, f);
2193 offset++;
2196 for(i = 0; i < s1->nb_sections; i++) {
2197 sh = &shdr;
2198 memset(sh, 0, sizeof(ElfW(Shdr)));
2199 s = s1->sections[i];
2200 if (s) {
2201 sh->sh_name = s->sh_name;
2202 sh->sh_type = s->sh_type;
2203 sh->sh_flags = s->sh_flags;
2204 sh->sh_entsize = s->sh_entsize;
2205 sh->sh_info = s->sh_info;
2206 if (s->link)
2207 sh->sh_link = s->link->sh_num;
2208 sh->sh_addralign = s->sh_addralign;
2209 sh->sh_addr = s->sh_addr;
2210 sh->sh_offset = s->sh_offset;
2211 sh->sh_size = s->sh_size;
2213 fwrite(sh, 1, sizeof(ElfW(Shdr)), f);
2217 /* Write an elf, coff or "binary" file */
2218 static int tcc_write_elf_file(TCCState *s1, const char *filename, int phnum,
2219 ElfW(Phdr) *phdr, int file_offset, int *sec_order)
2221 int fd, mode, file_type;
2222 FILE *f;
2224 file_type = s1->output_type;
2225 if (file_type == TCC_OUTPUT_OBJ)
2226 mode = 0666;
2227 else
2228 mode = 0777;
2229 unlink(filename);
2230 fd = open(filename, O_WRONLY | O_CREAT | O_TRUNC | O_BINARY, mode);
2231 if (fd < 0) {
2232 tcc_error_noabort("could not write '%s'", filename);
2233 return -1;
2235 f = fdopen(fd, "wb");
2236 if (s1->verbose)
2237 printf("<- %s\n", filename);
2239 #ifdef TCC_TARGET_COFF
2240 if (s1->output_format == TCC_OUTPUT_FORMAT_COFF)
2241 tcc_output_coff(s1, f);
2242 else
2243 #endif
2244 if (s1->output_format == TCC_OUTPUT_FORMAT_ELF)
2245 tcc_output_elf(s1, f, phnum, phdr, file_offset, sec_order);
2246 else
2247 tcc_output_binary(s1, f, sec_order);
2248 fclose(f);
2250 return 0;
2253 /* Output an elf, coff or binary file */
2254 /* XXX: suppress unneeded sections */
2255 static int elf_output_file(TCCState *s1, const char *filename)
2257 int i, ret, phnum, shnum, file_type, file_offset, *sec_order;
2258 struct dyn_inf dyninf;
2259 ElfW(Phdr) *phdr;
2260 ElfW(Sym) *sym;
2261 Section *strsec, *interp, *dynamic, *dynstr;
2263 file_type = s1->output_type;
2264 s1->nb_errors = 0;
2266 /* if linking, also link in runtime libraries (libc, libgcc, etc.) */
2267 if (file_type != TCC_OUTPUT_OBJ) {
2268 tcc_add_runtime(s1);
2271 phdr = NULL;
2272 sec_order = NULL;
2273 interp = dynamic = dynstr = NULL; /* avoid warning */
2274 dyninf.dyn_rel_off = 0; /* avoid warning */
2276 if (file_type != TCC_OUTPUT_OBJ) {
2277 relocate_common_syms();
2279 tcc_add_linker_symbols(s1);
2281 if (!s1->static_link) {
2282 if (file_type == TCC_OUTPUT_EXE) {
2283 char *ptr;
2284 /* allow override the dynamic loader */
2285 const char *elfint = getenv("LD_SO");
2286 if (elfint == NULL)
2287 elfint = DEFAULT_ELFINTERP(s1);
2288 /* add interpreter section only if executable */
2289 interp = new_section(s1, ".interp", SHT_PROGBITS, SHF_ALLOC);
2290 interp->sh_addralign = 1;
2291 ptr = section_ptr_add(interp, 1 + strlen(elfint));
2292 strcpy(ptr, elfint);
2295 /* add dynamic symbol table */
2296 s1->dynsym = new_symtab(s1, ".dynsym", SHT_DYNSYM, SHF_ALLOC,
2297 ".dynstr",
2298 ".hash", SHF_ALLOC);
2299 dynstr = s1->dynsym->link;
2301 /* add dynamic section */
2302 dynamic = new_section(s1, ".dynamic", SHT_DYNAMIC,
2303 SHF_ALLOC | SHF_WRITE);
2304 dynamic->link = dynstr;
2305 dynamic->sh_entsize = sizeof(ElfW(Dyn));
2307 /* add PLT */
2308 s1->plt = new_section(s1, ".plt", SHT_PROGBITS,
2309 SHF_ALLOC | SHF_EXECINSTR);
2310 s1->plt->sh_entsize = 4;
2312 build_got(s1);
2314 if (file_type == TCC_OUTPUT_EXE) {
2315 bind_exe_dynsyms(s1);
2317 if (s1->nb_errors) {
2318 ret = -1;
2319 goto the_end;
2322 bind_libs_dynsyms(s1);
2323 } else /* shared library case: simply export all global symbols */
2324 export_global_syms(s1);
2326 build_got_entries(s1);
2328 /* add a list of needed dlls */
2329 for(i = 0; i < s1->nb_loaded_dlls; i++) {
2330 DLLReference *dllref = s1->loaded_dlls[i];
2331 if (dllref->level == 0)
2332 put_dt(dynamic, DT_NEEDED, put_elf_str(dynstr, dllref->name));
2335 if (s1->rpath)
2336 put_dt(dynamic, DT_RPATH, put_elf_str(dynstr, s1->rpath));
2338 /* XXX: currently, since we do not handle PIC code, we
2339 must relocate the readonly segments */
2340 if (file_type == TCC_OUTPUT_DLL) {
2341 if (s1->soname)
2342 put_dt(dynamic, DT_SONAME, put_elf_str(dynstr, s1->soname));
2343 put_dt(dynamic, DT_TEXTREL, 0);
2346 if (s1->symbolic)
2347 put_dt(dynamic, DT_SYMBOLIC, 0);
2349 /* add necessary space for other entries */
2350 dyninf.dyn_rel_off = dynamic->data_offset;
2351 dynamic->data_offset += sizeof(ElfW(Dyn)) * EXTRA_RELITEMS;
2352 } else {
2353 /* still need to build got entries in case of static link */
2354 build_got_entries(s1);
2358 /* we add a section for symbols */
2359 strsec = new_section(s1, ".shstrtab", SHT_STRTAB, 0);
2360 put_elf_str(strsec, "");
2362 /* compute number of sections */
2363 shnum = s1->nb_sections;
2365 /* this array is used to reorder sections in the output file */
2366 sec_order = tcc_malloc(sizeof(int) * shnum);
2367 sec_order[0] = 0;
2369 /* compute number of program headers */
2370 switch(file_type) {
2371 default:
2372 case TCC_OUTPUT_OBJ:
2373 phnum = 0;
2374 break;
2375 case TCC_OUTPUT_EXE:
2376 if (!s1->static_link)
2377 phnum = 4 + HAVE_PHDR;
2378 else
2379 phnum = 2;
2380 break;
2381 case TCC_OUTPUT_DLL:
2382 phnum = 3;
2383 break;
2386 /* Allocate strings for section names */
2387 alloc_sec_names(s1, file_type, strsec);
2389 /* allocate program segment headers */
2390 phdr = tcc_mallocz(phnum * sizeof(ElfW(Phdr)));
2392 /* compute section to program header mapping */
2393 file_offset = layout_sections(s1, phdr, phnum, interp, &dyninf, sec_order);
2395 /* Fill remaining program header and finalize relocation related to dynamic
2396 linking. */
2397 if (phnum > 0) {
2398 fill_unloadable_phdr(phdr, phnum, interp, dynamic);
2399 if (dynamic) {
2400 dyninf.dynamic = dynamic;
2401 dyninf.dynstr = dynstr;
2403 fill_dynamic(s1, &dyninf);
2405 /* put in GOT the dynamic section address and relocate PLT */
2406 put32(s1->got->data, dynamic->sh_addr);
2407 if (file_type == TCC_OUTPUT_EXE
2408 #if defined(TCC_OUTPUT_DLL_WITH_PLT)
2409 || file_type == TCC_OUTPUT_DLL
2410 #endif
2412 relocate_plt(s1);
2414 /* relocate symbols in .dynsym now that final addresses are known */
2415 for_each_elem(s1->dynsym, 1, sym, ElfW(Sym)) {
2416 /* relocate to PLT if symbol corresponds to a PLT entry */
2417 if (sym->st_shndx == SHN_UNDEF) {
2418 if (sym->st_value)
2419 sym->st_value += s1->plt->sh_addr;
2420 } else if (sym->st_shndx < SHN_LORESERVE) {
2421 /* do symbol relocation */
2422 sym->st_value += s1->sections[sym->st_shndx]->sh_addr;
2428 /* if building executable or DLL, then relocate each section
2429 except the GOT which is already relocated */
2430 if (file_type != TCC_OUTPUT_OBJ) {
2431 ret = final_sections_reloc(s1);
2432 if (ret)
2433 goto the_end;
2436 /* Perform relocation to GOT or PLT entries */
2437 if (file_type == TCC_OUTPUT_EXE && s1->static_link)
2438 fill_got(s1);
2440 /* Create the ELF file with name 'filename' */
2441 ret = tcc_write_elf_file(s1, filename, phnum, phdr, file_offset, sec_order);
2442 the_end:
2443 tcc_free(s1->symtab_to_dynsym);
2444 tcc_free(sec_order);
2445 tcc_free(phdr);
2446 tcc_free(s1->sym_attrs);
2447 s1->sym_attrs = NULL;
2448 return ret;
2451 LIBTCCAPI int tcc_output_file(TCCState *s, const char *filename)
2453 int ret;
2454 #ifdef TCC_TARGET_PE
2455 if (s->output_type != TCC_OUTPUT_OBJ) {
2456 ret = pe_output_file(s, filename);
2457 } else
2458 #endif
2459 ret = elf_output_file(s, filename);
2460 return ret;
2463 static void *load_data(int fd, unsigned long file_offset, unsigned long size)
2465 void *data;
2467 data = tcc_malloc(size);
2468 lseek(fd, file_offset, SEEK_SET);
2469 read(fd, data, size);
2470 return data;
2473 typedef struct SectionMergeInfo {
2474 Section *s; /* corresponding existing section */
2475 unsigned long offset; /* offset of the new section in the existing section */
2476 uint8_t new_section; /* true if section 's' was added */
2477 uint8_t link_once; /* true if link once section */
2478 } SectionMergeInfo;
2480 /* load an object file and merge it with current files */
2481 /* XXX: handle correctly stab (debug) info */
2482 ST_FUNC int tcc_load_object_file(TCCState *s1,
2483 int fd, unsigned long file_offset)
2485 ElfW(Ehdr) ehdr;
2486 ElfW(Shdr) *shdr, *sh;
2487 int size, i, j, offset, offseti, nb_syms, sym_index, ret;
2488 unsigned char *strsec, *strtab;
2489 int *old_to_new_syms;
2490 char *sh_name, *name;
2491 SectionMergeInfo *sm_table, *sm;
2492 ElfW(Sym) *sym, *symtab;
2493 ElfW_Rel *rel;
2494 Section *s;
2496 int stab_index;
2497 int stabstr_index;
2499 stab_index = stabstr_index = 0;
2501 if (read(fd, &ehdr, sizeof(ehdr)) != sizeof(ehdr))
2502 goto fail1;
2503 if (ehdr.e_ident[0] != ELFMAG0 ||
2504 ehdr.e_ident[1] != ELFMAG1 ||
2505 ehdr.e_ident[2] != ELFMAG2 ||
2506 ehdr.e_ident[3] != ELFMAG3)
2507 goto fail1;
2508 /* test if object file */
2509 if (ehdr.e_type != ET_REL)
2510 goto fail1;
2511 /* test CPU specific stuff */
2512 if (ehdr.e_ident[5] != ELFDATA2LSB ||
2513 ehdr.e_machine != EM_TCC_TARGET) {
2514 fail1:
2515 tcc_error_noabort("invalid object file");
2516 return -1;
2518 /* read sections */
2519 shdr = load_data(fd, file_offset + ehdr.e_shoff,
2520 sizeof(ElfW(Shdr)) * ehdr.e_shnum);
2521 sm_table = tcc_mallocz(sizeof(SectionMergeInfo) * ehdr.e_shnum);
2523 /* load section names */
2524 sh = &shdr[ehdr.e_shstrndx];
2525 strsec = load_data(fd, file_offset + sh->sh_offset, sh->sh_size);
2527 /* load symtab and strtab */
2528 old_to_new_syms = NULL;
2529 symtab = NULL;
2530 strtab = NULL;
2531 nb_syms = 0;
2532 for(i = 1; i < ehdr.e_shnum; i++) {
2533 sh = &shdr[i];
2534 if (sh->sh_type == SHT_SYMTAB) {
2535 if (symtab) {
2536 tcc_error_noabort("object must contain only one symtab");
2537 fail:
2538 ret = -1;
2539 goto the_end;
2541 nb_syms = sh->sh_size / sizeof(ElfW(Sym));
2542 symtab = load_data(fd, file_offset + sh->sh_offset, sh->sh_size);
2543 sm_table[i].s = symtab_section;
2545 /* now load strtab */
2546 sh = &shdr[sh->sh_link];
2547 strtab = load_data(fd, file_offset + sh->sh_offset, sh->sh_size);
2551 /* now examine each section and try to merge its content with the
2552 ones in memory */
2553 for(i = 1; i < ehdr.e_shnum; i++) {
2554 /* no need to examine section name strtab */
2555 if (i == ehdr.e_shstrndx)
2556 continue;
2557 sh = &shdr[i];
2558 sh_name = (char *) strsec + sh->sh_name;
2559 /* ignore sections types we do not handle */
2560 if (sh->sh_type != SHT_PROGBITS &&
2561 sh->sh_type != SHT_RELX &&
2562 #ifdef TCC_ARM_EABI
2563 sh->sh_type != SHT_ARM_EXIDX &&
2564 #endif
2565 sh->sh_type != SHT_NOBITS &&
2566 sh->sh_type != SHT_PREINIT_ARRAY &&
2567 sh->sh_type != SHT_INIT_ARRAY &&
2568 sh->sh_type != SHT_FINI_ARRAY &&
2569 strcmp(sh_name, ".stabstr")
2571 continue;
2572 if (sh->sh_addralign < 1)
2573 sh->sh_addralign = 1;
2574 /* find corresponding section, if any */
2575 for(j = 1; j < s1->nb_sections;j++) {
2576 s = s1->sections[j];
2577 if (!strcmp(s->name, sh_name)) {
2578 if (!strncmp(sh_name, ".gnu.linkonce",
2579 sizeof(".gnu.linkonce") - 1)) {
2580 /* if a 'linkonce' section is already present, we
2581 do not add it again. It is a little tricky as
2582 symbols can still be defined in
2583 it. */
2584 sm_table[i].link_once = 1;
2585 goto next;
2586 } else {
2587 goto found;
2591 /* not found: create new section */
2592 s = new_section(s1, sh_name, sh->sh_type, sh->sh_flags);
2593 /* take as much info as possible from the section. sh_link and
2594 sh_info will be updated later */
2595 s->sh_addralign = sh->sh_addralign;
2596 s->sh_entsize = sh->sh_entsize;
2597 sm_table[i].new_section = 1;
2598 found:
2599 if (sh->sh_type != s->sh_type) {
2600 tcc_error_noabort("invalid section type");
2601 goto fail;
2604 /* align start of section */
2605 offset = s->data_offset;
2607 if (0 == strcmp(sh_name, ".stab")) {
2608 stab_index = i;
2609 goto no_align;
2611 if (0 == strcmp(sh_name, ".stabstr")) {
2612 stabstr_index = i;
2613 goto no_align;
2616 size = sh->sh_addralign - 1;
2617 offset = (offset + size) & ~size;
2618 if (sh->sh_addralign > s->sh_addralign)
2619 s->sh_addralign = sh->sh_addralign;
2620 s->data_offset = offset;
2621 no_align:
2622 sm_table[i].offset = offset;
2623 sm_table[i].s = s;
2624 /* concatenate sections */
2625 size = sh->sh_size;
2626 if (sh->sh_type != SHT_NOBITS) {
2627 unsigned char *ptr;
2628 lseek(fd, file_offset + sh->sh_offset, SEEK_SET);
2629 ptr = section_ptr_add(s, size);
2630 read(fd, ptr, size);
2631 } else {
2632 s->data_offset += size;
2634 next: ;
2637 /* gr relocate stab strings */
2638 if (stab_index && stabstr_index) {
2639 Stab_Sym *a, *b;
2640 unsigned o;
2641 s = sm_table[stab_index].s;
2642 a = (Stab_Sym *)(s->data + sm_table[stab_index].offset);
2643 b = (Stab_Sym *)(s->data + s->data_offset);
2644 o = sm_table[stabstr_index].offset;
2645 while (a < b)
2646 a->n_strx += o, a++;
2649 /* second short pass to update sh_link and sh_info fields of new
2650 sections */
2651 for(i = 1; i < ehdr.e_shnum; i++) {
2652 s = sm_table[i].s;
2653 if (!s || !sm_table[i].new_section)
2654 continue;
2655 sh = &shdr[i];
2656 if (sh->sh_link > 0)
2657 s->link = sm_table[sh->sh_link].s;
2658 if (sh->sh_type == SHT_RELX) {
2659 s->sh_info = sm_table[sh->sh_info].s->sh_num;
2660 /* update backward link */
2661 s1->sections[s->sh_info]->reloc = s;
2664 sm = sm_table;
2666 /* resolve symbols */
2667 old_to_new_syms = tcc_mallocz(nb_syms * sizeof(int));
2669 sym = symtab + 1;
2670 for(i = 1; i < nb_syms; i++, sym++) {
2671 if (sym->st_shndx != SHN_UNDEF &&
2672 sym->st_shndx < SHN_LORESERVE) {
2673 sm = &sm_table[sym->st_shndx];
2674 if (sm->link_once) {
2675 /* if a symbol is in a link once section, we use the
2676 already defined symbol. It is very important to get
2677 correct relocations */
2678 if (ELFW(ST_BIND)(sym->st_info) != STB_LOCAL) {
2679 name = (char *) strtab + sym->st_name;
2680 sym_index = find_elf_sym(symtab_section, name);
2681 if (sym_index)
2682 old_to_new_syms[i] = sym_index;
2684 continue;
2686 /* if no corresponding section added, no need to add symbol */
2687 if (!sm->s)
2688 continue;
2689 /* convert section number */
2690 sym->st_shndx = sm->s->sh_num;
2691 /* offset value */
2692 sym->st_value += sm->offset;
2694 /* add symbol */
2695 name = (char *) strtab + sym->st_name;
2696 sym_index = add_elf_sym(symtab_section, sym->st_value, sym->st_size,
2697 sym->st_info, sym->st_other,
2698 sym->st_shndx, name);
2699 old_to_new_syms[i] = sym_index;
2702 /* third pass to patch relocation entries */
2703 for(i = 1; i < ehdr.e_shnum; i++) {
2704 s = sm_table[i].s;
2705 if (!s)
2706 continue;
2707 sh = &shdr[i];
2708 offset = sm_table[i].offset;
2709 switch(s->sh_type) {
2710 case SHT_RELX:
2711 /* take relocation offset information */
2712 offseti = sm_table[sh->sh_info].offset;
2713 for_each_elem(s, (offset / sizeof(*rel)), rel, ElfW_Rel) {
2714 int type;
2715 unsigned sym_index;
2716 /* convert symbol index */
2717 type = ELFW(R_TYPE)(rel->r_info);
2718 sym_index = ELFW(R_SYM)(rel->r_info);
2719 /* NOTE: only one symtab assumed */
2720 if (sym_index >= nb_syms)
2721 goto invalid_reloc;
2722 sym_index = old_to_new_syms[sym_index];
2723 /* ignore link_once in rel section. */
2724 if (!sym_index && !sm->link_once
2725 #ifdef TCC_TARGET_ARM
2726 && type != R_ARM_V4BX
2727 #endif
2729 invalid_reloc:
2730 tcc_error_noabort("Invalid relocation entry [%2d] '%s' @ %.8x",
2731 i, strsec + sh->sh_name, rel->r_offset);
2732 goto fail;
2734 rel->r_info = ELFW(R_INFO)(sym_index, type);
2735 /* offset the relocation offset */
2736 rel->r_offset += offseti;
2737 #ifdef TCC_TARGET_ARM
2738 /* Jumps and branches from a Thumb code to a PLT entry need
2739 special handling since PLT entries are ARM code.
2740 Unconditional bl instructions referencing PLT entries are
2741 handled by converting these instructions into blx
2742 instructions. Other case of instructions referencing a PLT
2743 entry require to add a Thumb stub before the PLT entry to
2744 switch to ARM mode. We set bit plt_thumb_stub of the
2745 attribute of a symbol to indicate such a case. */
2746 if (type == R_ARM_THM_JUMP24)
2747 alloc_sym_attr(s1, sym_index)->plt_thumb_stub = 1;
2748 #endif
2750 break;
2751 default:
2752 break;
2756 ret = 0;
2757 the_end:
2758 tcc_free(symtab);
2759 tcc_free(strtab);
2760 tcc_free(old_to_new_syms);
2761 tcc_free(sm_table);
2762 tcc_free(strsec);
2763 tcc_free(shdr);
2764 return ret;
2767 typedef struct ArchiveHeader {
2768 char ar_name[16]; /* name of this member */
2769 char ar_date[12]; /* file mtime */
2770 char ar_uid[6]; /* owner uid; printed as decimal */
2771 char ar_gid[6]; /* owner gid; printed as decimal */
2772 char ar_mode[8]; /* file mode, printed as octal */
2773 char ar_size[10]; /* file size, printed as decimal */
2774 char ar_fmag[2]; /* should contain ARFMAG */
2775 } ArchiveHeader;
2777 static int get_be32(const uint8_t *b)
2779 return b[3] | (b[2] << 8) | (b[1] << 16) | (b[0] << 24);
2782 /* load only the objects which resolve undefined symbols */
2783 static int tcc_load_alacarte(TCCState *s1, int fd, int size)
2785 int i, bound, nsyms, sym_index, off, ret;
2786 uint8_t *data;
2787 const char *ar_names, *p;
2788 const uint8_t *ar_index;
2789 ElfW(Sym) *sym;
2791 data = tcc_malloc(size);
2792 if (read(fd, data, size) != size)
2793 goto fail;
2794 nsyms = get_be32(data);
2795 ar_index = data + 4;
2796 ar_names = (char *) ar_index + nsyms * 4;
2798 do {
2799 bound = 0;
2800 for(p = ar_names, i = 0; i < nsyms; i++, p += strlen(p)+1) {
2801 sym_index = find_elf_sym(symtab_section, p);
2802 if(sym_index) {
2803 sym = &((ElfW(Sym) *)symtab_section->data)[sym_index];
2804 if(sym->st_shndx == SHN_UNDEF) {
2805 off = get_be32(ar_index + i * 4) + sizeof(ArchiveHeader);
2806 ++bound;
2807 lseek(fd, off, SEEK_SET);
2808 if(tcc_load_object_file(s1, fd, off) < 0) {
2809 fail:
2810 ret = -1;
2811 goto the_end;
2816 } while(bound);
2817 ret = 0;
2818 the_end:
2819 tcc_free(data);
2820 return ret;
2823 /* load a '.a' file */
2824 ST_FUNC int tcc_load_archive(TCCState *s1, int fd)
2826 ArchiveHeader hdr;
2827 char ar_size[11];
2828 char ar_name[17];
2829 char magic[8];
2830 int size, len, i;
2831 unsigned long file_offset;
2833 /* skip magic which was already checked */
2834 read(fd, magic, sizeof(magic));
2836 for(;;) {
2837 len = read(fd, &hdr, sizeof(hdr));
2838 if (len == 0)
2839 break;
2840 if (len != sizeof(hdr)) {
2841 tcc_error_noabort("invalid archive");
2842 return -1;
2844 memcpy(ar_size, hdr.ar_size, sizeof(hdr.ar_size));
2845 ar_size[sizeof(hdr.ar_size)] = '\0';
2846 size = strtol(ar_size, NULL, 0);
2847 memcpy(ar_name, hdr.ar_name, sizeof(hdr.ar_name));
2848 for(i = sizeof(hdr.ar_name) - 1; i >= 0; i--) {
2849 if (ar_name[i] != ' ')
2850 break;
2852 ar_name[i + 1] = '\0';
2853 file_offset = lseek(fd, 0, SEEK_CUR);
2854 /* align to even */
2855 size = (size + 1) & ~1;
2856 if (!strcmp(ar_name, "/")) {
2857 /* coff symbol table : we handle it */
2858 if(s1->alacarte_link)
2859 return tcc_load_alacarte(s1, fd, size);
2860 } else if (!strcmp(ar_name, "//") ||
2861 !strcmp(ar_name, "__.SYMDEF") ||
2862 !strcmp(ar_name, "__.SYMDEF/") ||
2863 !strcmp(ar_name, "ARFILENAMES/")) {
2864 /* skip symbol table or archive names */
2865 } else {
2866 if (tcc_load_object_file(s1, fd, file_offset) < 0)
2867 return -1;
2869 lseek(fd, file_offset + size, SEEK_SET);
2871 return 0;
2874 #ifndef TCC_TARGET_PE
2875 /* load a DLL and all referenced DLLs. 'level = 0' means that the DLL
2876 is referenced by the user (so it should be added as DT_NEEDED in
2877 the generated ELF file) */
2878 ST_FUNC int tcc_load_dll(TCCState *s1, int fd, const char *filename, int level)
2880 ElfW(Ehdr) ehdr;
2881 ElfW(Shdr) *shdr, *sh, *sh1;
2882 int i, j, nb_syms, nb_dts, sym_bind, ret;
2883 ElfW(Sym) *sym, *dynsym;
2884 ElfW(Dyn) *dt, *dynamic;
2885 unsigned char *dynstr;
2886 const char *name, *soname;
2887 DLLReference *dllref;
2889 read(fd, &ehdr, sizeof(ehdr));
2891 /* test CPU specific stuff */
2892 if (ehdr.e_ident[5] != ELFDATA2LSB ||
2893 ehdr.e_machine != EM_TCC_TARGET) {
2894 tcc_error_noabort("bad architecture");
2895 return -1;
2898 /* read sections */
2899 shdr = load_data(fd, ehdr.e_shoff, sizeof(ElfW(Shdr)) * ehdr.e_shnum);
2901 /* load dynamic section and dynamic symbols */
2902 nb_syms = 0;
2903 nb_dts = 0;
2904 dynamic = NULL;
2905 dynsym = NULL; /* avoid warning */
2906 dynstr = NULL; /* avoid warning */
2907 for(i = 0, sh = shdr; i < ehdr.e_shnum; i++, sh++) {
2908 switch(sh->sh_type) {
2909 case SHT_DYNAMIC:
2910 nb_dts = sh->sh_size / sizeof(ElfW(Dyn));
2911 dynamic = load_data(fd, sh->sh_offset, sh->sh_size);
2912 break;
2913 case SHT_DYNSYM:
2914 nb_syms = sh->sh_size / sizeof(ElfW(Sym));
2915 dynsym = load_data(fd, sh->sh_offset, sh->sh_size);
2916 sh1 = &shdr[sh->sh_link];
2917 dynstr = load_data(fd, sh1->sh_offset, sh1->sh_size);
2918 break;
2919 default:
2920 break;
2924 /* compute the real library name */
2925 soname = tcc_basename(filename);
2927 for(i = 0, dt = dynamic; i < nb_dts; i++, dt++) {
2928 if (dt->d_tag == DT_SONAME) {
2929 soname = (char *) dynstr + dt->d_un.d_val;
2933 /* if the dll is already loaded, do not load it */
2934 for(i = 0; i < s1->nb_loaded_dlls; i++) {
2935 dllref = s1->loaded_dlls[i];
2936 if (!strcmp(soname, dllref->name)) {
2937 /* but update level if needed */
2938 if (level < dllref->level)
2939 dllref->level = level;
2940 ret = 0;
2941 goto the_end;
2945 /* add the dll and its level */
2946 dllref = tcc_mallocz(sizeof(DLLReference) + strlen(soname));
2947 dllref->level = level;
2948 strcpy(dllref->name, soname);
2949 dynarray_add((void ***)&s1->loaded_dlls, &s1->nb_loaded_dlls, dllref);
2951 /* add dynamic symbols in dynsym_section */
2952 for(i = 1, sym = dynsym + 1; i < nb_syms; i++, sym++) {
2953 sym_bind = ELFW(ST_BIND)(sym->st_info);
2954 if (sym_bind == STB_LOCAL)
2955 continue;
2956 name = (char *) dynstr + sym->st_name;
2957 add_elf_sym(s1->dynsymtab_section, sym->st_value, sym->st_size,
2958 sym->st_info, sym->st_other, sym->st_shndx, name);
2961 /* load all referenced DLLs */
2962 for(i = 0, dt = dynamic; i < nb_dts; i++, dt++) {
2963 switch(dt->d_tag) {
2964 case DT_NEEDED:
2965 name = (char *) dynstr + dt->d_un.d_val;
2966 for(j = 0; j < s1->nb_loaded_dlls; j++) {
2967 dllref = s1->loaded_dlls[j];
2968 if (!strcmp(name, dllref->name))
2969 goto already_loaded;
2971 if (tcc_add_dll(s1, name, AFF_REFERENCED_DLL) < 0) {
2972 tcc_error_noabort("referenced dll '%s' not found", name);
2973 ret = -1;
2974 goto the_end;
2976 already_loaded:
2977 break;
2980 ret = 0;
2981 the_end:
2982 tcc_free(dynstr);
2983 tcc_free(dynsym);
2984 tcc_free(dynamic);
2985 tcc_free(shdr);
2986 return ret;
2989 #define LD_TOK_NAME 256
2990 #define LD_TOK_EOF (-1)
2992 /* return next ld script token */
2993 static int ld_next(TCCState *s1, char *name, int name_size)
2995 int c;
2996 char *q;
2998 redo:
2999 switch(ch) {
3000 case ' ':
3001 case '\t':
3002 case '\f':
3003 case '\v':
3004 case '\r':
3005 case '\n':
3006 inp();
3007 goto redo;
3008 case '/':
3009 minp();
3010 if (ch == '*') {
3011 file->buf_ptr = parse_comment(file->buf_ptr);
3012 ch = file->buf_ptr[0];
3013 goto redo;
3014 } else {
3015 q = name;
3016 *q++ = '/';
3017 goto parse_name;
3019 break;
3020 /* case 'a' ... 'z': */
3021 case 'a':
3022 case 'b':
3023 case 'c':
3024 case 'd':
3025 case 'e':
3026 case 'f':
3027 case 'g':
3028 case 'h':
3029 case 'i':
3030 case 'j':
3031 case 'k':
3032 case 'l':
3033 case 'm':
3034 case 'n':
3035 case 'o':
3036 case 'p':
3037 case 'q':
3038 case 'r':
3039 case 's':
3040 case 't':
3041 case 'u':
3042 case 'v':
3043 case 'w':
3044 case 'x':
3045 case 'y':
3046 case 'z':
3047 /* case 'A' ... 'z': */
3048 case 'A':
3049 case 'B':
3050 case 'C':
3051 case 'D':
3052 case 'E':
3053 case 'F':
3054 case 'G':
3055 case 'H':
3056 case 'I':
3057 case 'J':
3058 case 'K':
3059 case 'L':
3060 case 'M':
3061 case 'N':
3062 case 'O':
3063 case 'P':
3064 case 'Q':
3065 case 'R':
3066 case 'S':
3067 case 'T':
3068 case 'U':
3069 case 'V':
3070 case 'W':
3071 case 'X':
3072 case 'Y':
3073 case 'Z':
3074 case '_':
3075 case '\\':
3076 case '.':
3077 case '$':
3078 case '~':
3079 q = name;
3080 parse_name:
3081 for(;;) {
3082 if (!((ch >= 'a' && ch <= 'z') ||
3083 (ch >= 'A' && ch <= 'Z') ||
3084 (ch >= '0' && ch <= '9') ||
3085 strchr("/.-_+=$:\\,~", ch)))
3086 break;
3087 if ((q - name) < name_size - 1) {
3088 *q++ = ch;
3090 minp();
3092 *q = '\0';
3093 c = LD_TOK_NAME;
3094 break;
3095 case CH_EOF:
3096 c = LD_TOK_EOF;
3097 break;
3098 default:
3099 c = ch;
3100 inp();
3101 break;
3103 return c;
3106 static int ld_add_file(TCCState *s1, const char filename[])
3108 int ret;
3110 ret = tcc_add_file_internal(s1, filename, 0);
3111 if (ret)
3112 ret = tcc_add_dll(s1, filename, 0);
3113 return ret;
3116 static inline int new_undef_syms(void)
3118 int ret = 0;
3119 ret = new_undef_sym;
3120 new_undef_sym = 0;
3121 return ret;
3124 static int ld_add_file_list(TCCState *s1, const char *cmd, int as_needed)
3126 char filename[1024], libname[1024];
3127 int t, group, nblibs = 0, ret = 0;
3128 char **libs = NULL;
3130 group = !strcmp(cmd, "GROUP");
3131 if (!as_needed)
3132 new_undef_syms();
3133 t = ld_next(s1, filename, sizeof(filename));
3134 if (t != '(')
3135 expect("(");
3136 t = ld_next(s1, filename, sizeof(filename));
3137 for(;;) {
3138 libname[0] = '\0';
3139 if (t == LD_TOK_EOF) {
3140 tcc_error_noabort("unexpected end of file");
3141 ret = -1;
3142 goto lib_parse_error;
3143 } else if (t == ')') {
3144 break;
3145 } else if (t == '-') {
3146 t = ld_next(s1, filename, sizeof(filename));
3147 if ((t != LD_TOK_NAME) || (filename[0] != 'l')) {
3148 tcc_error_noabort("library name expected");
3149 ret = -1;
3150 goto lib_parse_error;
3152 pstrcpy(libname, sizeof libname, &filename[1]);
3153 if (s1->static_link) {
3154 snprintf(filename, sizeof filename, "lib%s.a", libname);
3155 } else {
3156 snprintf(filename, sizeof filename, "lib%s.so", libname);
3158 } else if (t != LD_TOK_NAME) {
3159 tcc_error_noabort("filename expected");
3160 ret = -1;
3161 goto lib_parse_error;
3163 if (!strcmp(filename, "AS_NEEDED")) {
3164 ret = ld_add_file_list(s1, cmd, 1);
3165 if (ret)
3166 goto lib_parse_error;
3167 } else {
3168 /* TODO: Implement AS_NEEDED support. Ignore it for now */
3169 if (!as_needed) {
3170 ret = ld_add_file(s1, filename);
3171 if (ret)
3172 goto lib_parse_error;
3173 if (group) {
3174 /* Add the filename *and* the libname to avoid future conversions */
3175 dynarray_add((void ***) &libs, &nblibs, tcc_strdup(filename));
3176 if (libname[0] != '\0')
3177 dynarray_add((void ***) &libs, &nblibs, tcc_strdup(libname));
3181 t = ld_next(s1, filename, sizeof(filename));
3182 if (t == ',') {
3183 t = ld_next(s1, filename, sizeof(filename));
3186 if (group && !as_needed) {
3187 while (new_undef_syms()) {
3188 int i;
3190 for (i = 0; i < nblibs; i ++)
3191 ld_add_file(s1, libs[i]);
3194 lib_parse_error:
3195 dynarray_reset(&libs, &nblibs);
3196 return ret;
3199 /* interpret a subset of GNU ldscripts to handle the dummy libc.so
3200 files */
3201 ST_FUNC int tcc_load_ldscript(TCCState *s1)
3203 char cmd[64];
3204 char filename[1024];
3205 int t, ret;
3207 ch = file->buf_ptr[0];
3208 ch = handle_eob();
3209 for(;;) {
3210 t = ld_next(s1, cmd, sizeof(cmd));
3211 if (t == LD_TOK_EOF)
3212 return 0;
3213 else if (t != LD_TOK_NAME)
3214 return -1;
3215 if (!strcmp(cmd, "INPUT") ||
3216 !strcmp(cmd, "GROUP")) {
3217 ret = ld_add_file_list(s1, cmd, 0);
3218 if (ret)
3219 return ret;
3220 } else if (!strcmp(cmd, "OUTPUT_FORMAT") ||
3221 !strcmp(cmd, "TARGET")) {
3222 /* ignore some commands */
3223 t = ld_next(s1, cmd, sizeof(cmd));
3224 if (t != '(')
3225 expect("(");
3226 for(;;) {
3227 t = ld_next(s1, filename, sizeof(filename));
3228 if (t == LD_TOK_EOF) {
3229 tcc_error_noabort("unexpected end of file");
3230 return -1;
3231 } else if (t == ')') {
3232 break;
3235 } else {
3236 return -1;
3239 return 0;
3241 #endif /* !TCC_TARGET_PE */