x86_64: Create proper PLT and GOT also for -run
[tinycc.git] / tccelf.c
blob4ed845bf8d4a8e474e90416668a6e7edb722efdd
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 case R_ARM_NONE:
773 /* Nothing to do. Normally used to indicate a dependency
774 on a certain symbol (like for exception handling under EABI). */
775 break;
776 default:
777 fprintf(stderr,"FIXME: handle reloc type %x at %x [%p] to %x\n",
778 type, (unsigned)addr, ptr, (unsigned)val);
779 break;
780 #elif defined(TCC_TARGET_C67)
781 case R_C60_32:
782 *(int *)ptr += val;
783 break;
784 case R_C60LO16:
786 uint32_t orig;
788 /* put the low 16 bits of the absolute address
789 add to what is already there */
791 orig = ((*(int *)(ptr )) >> 7) & 0xffff;
792 orig |= (((*(int *)(ptr+4)) >> 7) & 0xffff) << 16;
794 /* patch both at once - assumes always in pairs Low - High */
796 *(int *) ptr = (*(int *) ptr & (~(0xffff << 7)) ) | (((val+orig) & 0xffff) << 7);
797 *(int *)(ptr+4) = (*(int *)(ptr+4) & (~(0xffff << 7)) ) | ((((val+orig)>>16) & 0xffff) << 7);
799 break;
800 case R_C60HI16:
801 break;
802 default:
803 fprintf(stderr,"FIXME: handle reloc type %x at %x [%p] to %x\n",
804 type, (unsigned)addr, ptr, (unsigned)val);
805 break;
806 #elif defined(TCC_TARGET_X86_64)
807 case R_X86_64_64:
808 if (s1->output_type == TCC_OUTPUT_DLL) {
809 esym_index = s1->symtab_to_dynsym[sym_index];
810 qrel->r_offset = rel->r_offset;
811 if (esym_index) {
812 qrel->r_info = ELFW(R_INFO)(esym_index, R_X86_64_64);
813 qrel->r_addend = rel->r_addend;
814 qrel++;
815 break;
816 } else {
817 qrel->r_info = ELFW(R_INFO)(0, R_X86_64_RELATIVE);
818 qrel->r_addend = *(long long *)ptr + val;
819 qrel++;
822 *(long long *)ptr += val;
823 break;
824 case R_X86_64_32:
825 case R_X86_64_32S:
826 if (s1->output_type == TCC_OUTPUT_DLL) {
827 /* XXX: this logic may depend on TCC's codegen
828 now TCC uses R_X86_64_32 even for a 64bit pointer */
829 qrel->r_info = ELFW(R_INFO)(0, R_X86_64_RELATIVE);
830 qrel->r_addend = *(int *)ptr + val;
831 qrel++;
833 *(int *)ptr += val;
834 break;
836 case R_X86_64_PC32:
837 if (s1->output_type == TCC_OUTPUT_DLL) {
838 /* DLL relocation */
839 esym_index = s1->symtab_to_dynsym[sym_index];
840 if (esym_index) {
841 qrel->r_offset = rel->r_offset;
842 qrel->r_info = ELFW(R_INFO)(esym_index, R_X86_64_PC32);
843 qrel->r_addend = *(int *)ptr;
844 qrel++;
845 break;
848 goto plt32pc32;
850 case R_X86_64_PLT32:
851 /* We've put the PLT slot offset into r_addend when generating
852 it, and that's what we must use as relocation value (adjusted
853 by section offset of course). */
854 val = s1->plt->sh_addr + rel->r_addend;
855 /* fallthrough. */
857 plt32pc32:
859 long long diff;
860 diff = (long long)val - addr;
861 if (diff <= -2147483647 || diff > 2147483647) {
862 #ifdef TCC_HAS_RUNTIME_PLTGOT
863 /* XXX: naive support for over 32bit jump */
864 if (s1->output_type == TCC_OUTPUT_MEMORY) {
865 val = (add_jmp_table(s1, val - rel->r_addend) +
866 rel->r_addend);
867 diff = val - addr;
869 #endif
870 if (diff <= -2147483647 || diff > 2147483647) {
871 tcc_error("internal error: relocation failed");
874 *(int *)ptr += diff;
876 break;
877 case R_X86_64_GLOB_DAT:
878 case R_X86_64_JUMP_SLOT:
879 /* They don't need addend */
880 *(addr_t *)ptr = val - rel->r_addend;
881 break;
882 case R_X86_64_GOTPCREL:
883 #ifdef TCC_HAS_RUNTIME_PLTGOT
884 if (s1->output_type == TCC_OUTPUT_MEMORY) {
885 val = add_got_table(s1, val - rel->r_addend) + rel->r_addend;
886 *(int *)ptr += val - addr;
887 break;
889 #endif
890 *(int *)ptr += (s1->got->sh_addr - addr +
891 s1->sym_attrs[sym_index].got_offset - 4);
892 break;
893 case R_X86_64_GOTTPOFF:
894 *(int *)ptr += val - s1->got->sh_addr;
895 break;
896 case R_X86_64_GOT32:
897 /* we load the got offset */
898 *(int *)ptr += s1->sym_attrs[sym_index].got_offset;
899 break;
900 #else
901 #error unsupported processor
902 #endif
905 /* if the relocation is allocated, we change its symbol table */
906 if (sr->sh_flags & SHF_ALLOC)
907 sr->link = s1->dynsym;
910 /* relocate relocation table in 'sr' */
911 static void relocate_rel(TCCState *s1, Section *sr)
913 Section *s;
914 ElfW_Rel *rel;
916 s = s1->sections[sr->sh_info];
917 for_each_elem(sr, 0, rel, ElfW_Rel)
918 rel->r_offset += s->sh_addr;
921 /* count the number of dynamic relocations so that we can reserve
922 their space */
923 static int prepare_dynamic_rel(TCCState *s1, Section *sr)
925 ElfW_Rel *rel;
926 int sym_index, esym_index, type, count;
928 count = 0;
929 for_each_elem(sr, 0, rel, ElfW_Rel) {
930 sym_index = ELFW(R_SYM)(rel->r_info);
931 type = ELFW(R_TYPE)(rel->r_info);
932 switch(type) {
933 #if defined(TCC_TARGET_I386)
934 case R_386_32:
935 #elif defined(TCC_TARGET_X86_64)
936 case R_X86_64_32:
937 case R_X86_64_32S:
938 case R_X86_64_64:
939 #endif
940 count++;
941 break;
942 #if defined(TCC_TARGET_I386)
943 case R_386_PC32:
944 #elif defined(TCC_TARGET_X86_64)
945 case R_X86_64_PC32:
946 #endif
947 esym_index = s1->symtab_to_dynsym[sym_index];
948 if (esym_index)
949 count++;
950 break;
951 default:
952 break;
955 if (count) {
956 /* allocate the section */
957 sr->sh_flags |= SHF_ALLOC;
958 sr->sh_size = count * sizeof(ElfW_Rel);
960 return count;
963 static struct sym_attr *alloc_sym_attr(TCCState *s1, int index)
965 int n;
966 struct sym_attr *tab;
968 if (index >= s1->nb_sym_attrs) {
969 /* find immediately bigger power of 2 and reallocate array */
970 n = 1;
971 while (index >= n)
972 n *= 2;
973 tab = tcc_realloc(s1->sym_attrs, n * sizeof(*s1->sym_attrs));
974 s1->sym_attrs = tab;
975 memset(s1->sym_attrs + s1->nb_sym_attrs, 0,
976 (n - s1->nb_sym_attrs) * sizeof(*s1->sym_attrs));
977 s1->nb_sym_attrs = n;
979 return &s1->sym_attrs[index];
982 /* XXX: suppress that */
983 static void put32(unsigned char *p, uint32_t val)
985 p[0] = val;
986 p[1] = val >> 8;
987 p[2] = val >> 16;
988 p[3] = val >> 24;
991 #if defined(TCC_TARGET_I386) || defined(TCC_TARGET_ARM) || \
992 defined(TCC_TARGET_X86_64)
993 static uint32_t get32(unsigned char *p)
995 return p[0] | (p[1] << 8) | (p[2] << 16) | (p[3] << 24);
997 #endif
999 static void build_got(TCCState *s1)
1001 unsigned char *ptr;
1003 /* if no got, then create it */
1004 s1->got = new_section(s1, ".got", SHT_PROGBITS, SHF_ALLOC | SHF_WRITE);
1005 s1->got->sh_entsize = 4;
1006 add_elf_sym(symtab_section, 0, 4, ELFW(ST_INFO)(STB_GLOBAL, STT_OBJECT),
1007 0, s1->got->sh_num, "_GLOBAL_OFFSET_TABLE_");
1008 ptr = section_ptr_add(s1->got, 3 * PTR_SIZE);
1009 #if PTR_SIZE == 4
1010 /* keep space for _DYNAMIC pointer, if present */
1011 put32(ptr, 0);
1012 /* two dummy got entries */
1013 put32(ptr + 4, 0);
1014 put32(ptr + 8, 0);
1015 #else
1016 /* keep space for _DYNAMIC pointer, if present */
1017 put32(ptr, 0);
1018 put32(ptr + 4, 0);
1019 /* two dummy got entries */
1020 put32(ptr + 8, 0);
1021 put32(ptr + 12, 0);
1022 put32(ptr + 16, 0);
1023 put32(ptr + 20, 0);
1024 #endif
1027 /* put a got or plt entry corresponding to a symbol in symtab_section. 'size'
1028 and 'info' can be modifed if more precise info comes from the DLL.
1029 Returns offset of GOT or PLT slot. */
1030 static unsigned long put_got_entry(TCCState *s1,
1031 int reloc_type, unsigned long size, int info,
1032 int sym_index)
1034 int index, need_plt_entry;
1035 const char *name;
1036 ElfW(Sym) *sym;
1037 unsigned long offset;
1038 int *ptr;
1039 struct sym_attr *symattr;
1041 if (!s1->got)
1042 build_got(s1);
1044 need_plt_entry =
1045 #ifdef TCC_TARGET_X86_64
1046 (reloc_type == R_X86_64_JUMP_SLOT);
1047 #elif defined(TCC_TARGET_I386)
1048 (reloc_type == R_386_JMP_SLOT);
1049 #elif defined(TCC_TARGET_ARM)
1050 (reloc_type == R_ARM_JUMP_SLOT);
1051 #else
1053 #endif
1055 if (need_plt_entry && !s1->plt) {
1056 /* add PLT */
1057 s1->plt = new_section(s1, ".plt", SHT_PROGBITS,
1058 SHF_ALLOC | SHF_EXECINSTR);
1059 s1->plt->sh_entsize = 4;
1062 /* If a got/plt entry already exists for that symbol, no need to add one */
1063 if (sym_index < s1->nb_sym_attrs) {
1064 if (need_plt_entry && s1->sym_attrs[sym_index].plt_offset)
1065 return s1->sym_attrs[sym_index].plt_offset;
1066 else if (!need_plt_entry && s1->sym_attrs[sym_index].got_offset)
1067 return s1->sym_attrs[sym_index].got_offset;
1070 symattr = alloc_sym_attr(s1, sym_index);
1072 /* Only store the GOT offset if it's not generated for the PLT entry. */
1073 if (!need_plt_entry)
1074 symattr->got_offset = s1->got->data_offset;
1076 sym = &((ElfW(Sym) *)symtab_section->data)[sym_index];
1077 name = (char *) symtab_section->link->data + sym->st_name;
1078 offset = sym->st_value;
1079 #if defined(TCC_TARGET_I386) || defined(TCC_TARGET_X86_64)
1080 if (need_plt_entry) {
1081 Section *plt;
1082 uint8_t *p;
1083 int modrm;
1084 unsigned long relofs;
1086 #if defined(TCC_OUTPUT_DLL_WITH_PLT)
1087 modrm = 0x25;
1088 #else
1089 /* if we build a DLL, we add a %ebx offset */
1090 if (s1->output_type == TCC_OUTPUT_DLL)
1091 modrm = 0xa3;
1092 else
1093 modrm = 0x25;
1094 #endif
1096 /* add a PLT entry */
1097 plt = s1->plt;
1098 if (plt->data_offset == 0) {
1099 /* first plt entry */
1100 p = section_ptr_add(plt, 16);
1101 p[0] = 0xff; /* pushl got + PTR_SIZE */
1102 p[1] = modrm + 0x10;
1103 put32(p + 2, PTR_SIZE);
1104 p[6] = 0xff; /* jmp *(got + PTR_SIZE * 2) */
1105 p[7] = modrm;
1106 put32(p + 8, PTR_SIZE * 2);
1109 /* The PLT slot refers to the relocation entry it needs
1110 via offset. The reloc entry is created below, so its
1111 offset is the current data_offset. */
1112 relofs = s1->got->reloc ? s1->got->reloc->data_offset : 0;
1113 symattr->plt_offset = plt->data_offset;
1114 p = section_ptr_add(plt, 16);
1115 p[0] = 0xff; /* jmp *(got + x) */
1116 p[1] = modrm;
1117 put32(p + 2, s1->got->data_offset);
1118 p[6] = 0x68; /* push $xxx */
1119 #ifdef TCC_TARGET_X86_64
1120 /* On x86-64, the relocation is referred to by _index_. */
1121 put32(p + 7, relofs / sizeof (ElfW_Rel));
1122 #else
1123 put32(p + 7, relofs);
1124 #endif
1125 p[11] = 0xe9; /* jmp plt_start */
1126 put32(p + 12, -(plt->data_offset));
1128 /* If this was an UNDEF symbol set the offset in the
1129 dynsymtab to the PLT slot, so that PC32 relocs to it
1130 can be resolved. */
1131 if (sym->st_shndx == SHN_UNDEF)
1132 offset = plt->data_offset - 16;
1134 #elif defined(TCC_TARGET_ARM)
1135 if (need_plt_entry) {
1136 Section *plt;
1137 uint8_t *p;
1139 /* if we build a DLL, we add a %ebx offset */
1140 if (s1->output_type == TCC_OUTPUT_DLL)
1141 tcc_error("DLLs unimplemented!");
1143 /* add a PLT entry */
1144 plt = s1->plt;
1145 if (plt->data_offset == 0) {
1146 /* first plt entry */
1147 p = section_ptr_add(plt, 16);
1148 put32(p, 0xe52de004); /* push {lr} */
1149 put32(p+4, 0xe59fe010); /* ldr lr, [pc, #16] */
1150 put32(p+8, 0xe08fe00e); /* add lr, pc, lr */
1151 put32(p+12, 0xe5bef008); /* ldr pc, [lr, #8]! */
1154 symattr->plt_offset = plt->data_offset;
1155 if (symattr->plt_thumb_stub) {
1156 p = section_ptr_add(plt, 20);
1157 put32(p, 0x4778); /* bx pc */
1158 put32(p+2, 0x46c0); /* nop */
1159 p += 4;
1160 } else
1161 p = section_ptr_add(plt, 16);
1162 put32(p, 0xe59fc004); /* ldr ip, [pc, #4] ; GOT entry offset */
1163 put32(p+4, 0xe08fc00c); /* add ip, pc, ip ; addr of GOT entry */
1164 put32(p+8, 0xe59cf000); /* ldr pc, [ip] ; jump to GOT entry */
1165 put32(p+12, s1->got->data_offset); /* GOT entry off once patched */
1167 /* the symbol is modified so that it will be relocated to
1168 the PLT */
1169 if (s1->output_type == TCC_OUTPUT_EXE)
1170 offset = plt->data_offset - 16;
1172 #elif defined(TCC_TARGET_C67)
1173 if (s1->dynsym) {
1174 tcc_error("C67 got not implemented");
1176 #else
1177 #error unsupported CPU
1178 #endif
1179 if (s1->dynsym) {
1180 /* XXX This might generate multiple syms for name. */
1181 index = put_elf_sym(s1->dynsym, offset,
1182 size, info, 0, sym->st_shndx, name);
1183 /* Create the relocation (it's against the GOT for PLT
1184 and GOT relocs). */
1185 put_elf_reloc(s1->dynsym, s1->got,
1186 s1->got->data_offset,
1187 reloc_type, index);
1188 } else {
1189 /* Without .dynsym (i.e. static link or memory output) we
1190 still need relocs against the generated got, so as to fill
1191 the entries with the symbol values (determined later). */
1192 put_elf_reloc(symtab_section, s1->got,
1193 s1->got->data_offset,
1194 reloc_type, sym_index);
1196 /* And now create the GOT slot itself. */
1197 ptr = section_ptr_add(s1->got, PTR_SIZE);
1198 *ptr = 0;
1199 if (need_plt_entry)
1200 return symattr->plt_offset;
1201 else
1202 return symattr->got_offset;
1205 /* build GOT and PLT entries */
1206 ST_FUNC void build_got_entries(TCCState *s1)
1208 Section *s;
1209 ElfW_Rel *rel;
1210 ElfW(Sym) *sym;
1211 int i, type, reloc_type, sym_index;
1213 for(i = 1; i < s1->nb_sections; i++) {
1214 s = s1->sections[i];
1215 if (s->sh_type != SHT_RELX)
1216 continue;
1217 /* no need to handle got relocations */
1218 if (s->link != symtab_section)
1219 continue;
1220 for_each_elem(s, 0, rel, ElfW_Rel) {
1221 type = ELFW(R_TYPE)(rel->r_info);
1222 switch(type) {
1223 #if defined(TCC_TARGET_I386)
1224 case R_386_GOT32:
1225 case R_386_GOTOFF:
1226 case R_386_GOTPC:
1227 case R_386_PLT32:
1228 if (!s1->got)
1229 build_got(s1);
1230 if (type == R_386_GOT32 || type == R_386_PLT32) {
1231 sym_index = ELFW(R_SYM)(rel->r_info);
1232 sym = &((ElfW(Sym) *)symtab_section->data)[sym_index];
1233 /* look at the symbol got offset. If none, then add one */
1234 if (type == R_386_GOT32)
1235 reloc_type = R_386_GLOB_DAT;
1236 else
1237 reloc_type = R_386_JMP_SLOT;
1238 put_got_entry(s1, reloc_type, sym->st_size, sym->st_info,
1239 sym_index);
1241 break;
1242 #elif defined(TCC_TARGET_ARM)
1243 case R_ARM_GOT32:
1244 case R_ARM_GOTOFF:
1245 case R_ARM_GOTPC:
1246 case R_ARM_PLT32:
1247 if (!s1->got)
1248 build_got(s1);
1249 if (type == R_ARM_GOT32 || type == R_ARM_PLT32) {
1250 sym_index = ELFW(R_SYM)(rel->r_info);
1251 sym = &((ElfW(Sym) *)symtab_section->data)[sym_index];
1252 /* look at the symbol got offset. If none, then add one */
1253 if (type == R_ARM_GOT32)
1254 reloc_type = R_ARM_GLOB_DAT;
1255 else
1256 reloc_type = R_ARM_JUMP_SLOT;
1257 put_got_entry(s1, reloc_type, sym->st_size, sym->st_info,
1258 sym_index);
1260 break;
1261 case R_ARM_THM_JUMP24:
1262 sym_index = ELFW(R_SYM)(rel->r_info);
1263 sym = &((ElfW(Sym) *)symtab_section->data)[sym_index];
1264 /* We are relocating a jump from thumb code to arm code */
1265 if (sym->st_shndx != SHN_UNDEF && !(sym->st_value & 1)) {
1266 int index;
1267 uint8_t *p;
1268 char *name, buf[1024];
1269 Section *text_section;
1271 name = (char *) symtab_section->link->data + sym->st_name;
1272 text_section = s1->sections[sym->st_shndx];
1273 /* Modify reloc to target a thumb stub to switch to ARM */
1274 snprintf(buf, sizeof(buf), "%s_from_thumb", name);
1275 index = put_elf_sym(symtab_section,
1276 text_section->data_offset + 1,
1277 sym->st_size, sym->st_info, 0,
1278 sym->st_shndx, buf);
1279 rel->r_info = ELFW(R_INFO)(index, type);
1280 /* Create a thumb stub fonction to switch to ARM mode */
1281 put_elf_reloc(symtab_section, text_section,
1282 text_section->data_offset + 4, R_ARM_JUMP24,
1283 sym_index);
1284 p = section_ptr_add(text_section, 8);
1285 put32(p, 0x4778); /* bx pc */
1286 put32(p+2, 0x46c0); /* nop */
1287 put32(p+4, 0xeafffffe); /* b $sym */
1289 #elif defined(TCC_TARGET_C67)
1290 case R_C60_GOT32:
1291 case R_C60_GOTOFF:
1292 case R_C60_GOTPC:
1293 case R_C60_PLT32:
1294 if (!s1->got)
1295 build_got(s1);
1296 if (type == R_C60_GOT32 || type == R_C60_PLT32) {
1297 sym_index = ELFW(R_SYM)(rel->r_info);
1298 sym = &((ElfW(Sym) *)symtab_section->data)[sym_index];
1299 /* look at the symbol got offset. If none, then add one */
1300 if (type == R_C60_GOT32)
1301 reloc_type = R_C60_GLOB_DAT;
1302 else
1303 reloc_type = R_C60_JMP_SLOT;
1304 put_got_entry(s1, reloc_type, sym->st_size, sym->st_info,
1305 sym_index);
1307 break;
1308 #elif defined(TCC_TARGET_X86_64)
1309 case R_X86_64_GOT32:
1310 case R_X86_64_GOTTPOFF:
1311 case R_X86_64_GOTPCREL:
1312 case R_X86_64_PLT32:
1313 if (!s1->got)
1314 build_got(s1);
1315 if (type == R_X86_64_GOT32 || type == R_X86_64_GOTPCREL ||
1316 type == R_X86_64_PLT32) {
1317 unsigned long ofs;
1318 sym_index = ELFW(R_SYM)(rel->r_info);
1319 sym = &((ElfW(Sym) *)symtab_section->data)[sym_index];
1320 /* look at the symbol got offset. If none, then add one */
1321 if (type == R_X86_64_GOT32 || type == R_X86_64_GOTPCREL)
1322 reloc_type = R_X86_64_GLOB_DAT;
1323 else
1324 reloc_type = R_X86_64_JUMP_SLOT;
1325 ofs = put_got_entry(s1, reloc_type, sym->st_size,
1326 sym->st_info, sym_index);
1327 if (type == R_X86_64_PLT32)
1328 /* We store the place of the generated PLT slot
1329 in our addend. */
1330 rel->r_addend += ofs;
1332 break;
1333 #else
1334 #error unsupported CPU
1335 #endif
1336 default:
1337 break;
1343 ST_FUNC Section *new_symtab(TCCState *s1,
1344 const char *symtab_name, int sh_type, int sh_flags,
1345 const char *strtab_name,
1346 const char *hash_name, int hash_sh_flags)
1348 Section *symtab, *strtab, *hash;
1349 int *ptr, nb_buckets;
1351 symtab = new_section(s1, symtab_name, sh_type, sh_flags);
1352 symtab->sh_entsize = sizeof(ElfW(Sym));
1353 strtab = new_section(s1, strtab_name, SHT_STRTAB, sh_flags);
1354 put_elf_str(strtab, "");
1355 symtab->link = strtab;
1356 put_elf_sym(symtab, 0, 0, 0, 0, 0, NULL);
1358 nb_buckets = 1;
1360 hash = new_section(s1, hash_name, SHT_HASH, hash_sh_flags);
1361 hash->sh_entsize = sizeof(int);
1362 symtab->hash = hash;
1363 hash->link = symtab;
1365 ptr = section_ptr_add(hash, (2 + nb_buckets + 1) * sizeof(int));
1366 ptr[0] = nb_buckets;
1367 ptr[1] = 1;
1368 memset(ptr + 2, 0, (nb_buckets + 1) * sizeof(int));
1369 return symtab;
1372 /* put dynamic tag */
1373 static void put_dt(Section *dynamic, int dt, addr_t val)
1375 ElfW(Dyn) *dyn;
1376 dyn = section_ptr_add(dynamic, sizeof(ElfW(Dyn)));
1377 dyn->d_tag = dt;
1378 dyn->d_un.d_val = val;
1381 static void add_init_array_defines(TCCState *s1, const char *section_name)
1383 Section *s;
1384 long end_offset;
1385 char sym_start[1024];
1386 char sym_end[1024];
1388 snprintf(sym_start, sizeof(sym_start), "__%s_start", section_name + 1);
1389 snprintf(sym_end, sizeof(sym_end), "__%s_end", section_name + 1);
1391 s = find_section(s1, section_name);
1392 if (!s) {
1393 end_offset = 0;
1394 s = data_section;
1395 } else {
1396 end_offset = s->data_offset;
1399 add_elf_sym(symtab_section,
1400 0, 0,
1401 ELFW(ST_INFO)(STB_GLOBAL, STT_NOTYPE), 0,
1402 s->sh_num, sym_start);
1403 add_elf_sym(symtab_section,
1404 end_offset, 0,
1405 ELFW(ST_INFO)(STB_GLOBAL, STT_NOTYPE), 0,
1406 s->sh_num, sym_end);
1409 ST_FUNC void tcc_add_bcheck(TCCState *s1)
1411 #ifdef CONFIG_TCC_BCHECK
1412 unsigned long *ptr;
1413 Section *init_section;
1414 unsigned char *pinit;
1415 int sym_index;
1417 if (0 == s1->do_bounds_check)
1418 return;
1420 /* XXX: add an object file to do that */
1421 ptr = section_ptr_add(bounds_section, sizeof(unsigned long));
1422 *ptr = 0;
1423 add_elf_sym(symtab_section, 0, 0,
1424 ELFW(ST_INFO)(STB_GLOBAL, STT_NOTYPE), 0,
1425 bounds_section->sh_num, "__bounds_start");
1426 #ifdef TCC_TARGET_I386
1427 if (s1->output_type != TCC_OUTPUT_MEMORY) {
1428 /* add 'call __bound_init()' in .init section */
1429 init_section = find_section(s1, ".init");
1430 pinit = section_ptr_add(init_section, 5);
1431 pinit[0] = 0xe8;
1432 put32(pinit + 1, -4);
1433 sym_index = find_elf_sym(symtab_section, "__bound_init");
1434 put_elf_reloc(symtab_section, init_section,
1435 init_section->data_offset - 4, R_386_PC32, sym_index);
1437 #endif
1438 #endif
1441 static inline int tcc_add_support(TCCState *s1, const char *filename)
1443 char buf[1024];
1444 snprintf(buf, sizeof(buf), "%s/%s", s1->tcc_lib_path, filename);
1445 return tcc_add_file(s1, buf);
1448 /* add tcc runtime libraries */
1449 ST_FUNC void tcc_add_runtime(TCCState *s1)
1451 /* add libc */
1452 if (!s1->nostdlib) {
1453 tcc_add_library(s1, "c");
1454 #ifdef CONFIG_USE_LIBGCC
1455 if (!s1->static_link) {
1456 tcc_add_file(s1, TCC_LIBGCC);
1457 tcc_add_support(s1, "libtcc1.a");
1458 } else
1459 tcc_add_support(s1, "libtcc1.a");
1460 #else
1461 tcc_add_support(s1, "libtcc1.a");
1462 #endif
1465 /* tcc_add_bcheck tries to relocate a call to __bound_init in _init so
1466 libtcc1.a must be loaded before for __bound_init to be defined and
1467 crtn.o must be loaded after to not finalize _init too early. */
1468 tcc_add_bcheck(s1);
1470 if (!s1->nostdlib) {
1471 /* add crt end if not memory output */
1472 if (s1->output_type != TCC_OUTPUT_MEMORY)
1473 tcc_add_crt(s1, "crtn.o");
1477 /* add various standard linker symbols (must be done after the
1478 sections are filled (for example after allocating common
1479 symbols)) */
1480 ST_FUNC void tcc_add_linker_symbols(TCCState *s1)
1482 char buf[1024];
1483 int i;
1484 Section *s;
1486 add_elf_sym(symtab_section,
1487 text_section->data_offset, 0,
1488 ELFW(ST_INFO)(STB_GLOBAL, STT_NOTYPE), 0,
1489 text_section->sh_num, "_etext");
1490 add_elf_sym(symtab_section,
1491 data_section->data_offset, 0,
1492 ELFW(ST_INFO)(STB_GLOBAL, STT_NOTYPE), 0,
1493 data_section->sh_num, "_edata");
1494 add_elf_sym(symtab_section,
1495 bss_section->data_offset, 0,
1496 ELFW(ST_INFO)(STB_GLOBAL, STT_NOTYPE), 0,
1497 bss_section->sh_num, "_end");
1498 /* horrible new standard ldscript defines */
1499 add_init_array_defines(s1, ".preinit_array");
1500 add_init_array_defines(s1, ".init_array");
1501 add_init_array_defines(s1, ".fini_array");
1503 /* add start and stop symbols for sections whose name can be
1504 expressed in C */
1505 for(i = 1; i < s1->nb_sections; i++) {
1506 s = s1->sections[i];
1507 if (s->sh_type == SHT_PROGBITS &&
1508 (s->sh_flags & SHF_ALLOC)) {
1509 const char *p;
1510 int ch;
1512 /* check if section name can be expressed in C */
1513 p = s->name;
1514 for(;;) {
1515 ch = *p;
1516 if (!ch)
1517 break;
1518 if (!isid(ch) && !isnum(ch))
1519 goto next_sec;
1520 p++;
1522 snprintf(buf, sizeof(buf), "__start_%s", s->name);
1523 add_elf_sym(symtab_section,
1524 0, 0,
1525 ELFW(ST_INFO)(STB_GLOBAL, STT_NOTYPE), 0,
1526 s->sh_num, buf);
1527 snprintf(buf, sizeof(buf), "__stop_%s", s->name);
1528 add_elf_sym(symtab_section,
1529 s->data_offset, 0,
1530 ELFW(ST_INFO)(STB_GLOBAL, STT_NOTYPE), 0,
1531 s->sh_num, buf);
1533 next_sec: ;
1537 static void tcc_output_binary(TCCState *s1, FILE *f,
1538 const int *sec_order)
1540 Section *s;
1541 int i, offset, size;
1543 offset = 0;
1544 for(i=1;i<s1->nb_sections;i++) {
1545 s = s1->sections[sec_order[i]];
1546 if (s->sh_type != SHT_NOBITS &&
1547 (s->sh_flags & SHF_ALLOC)) {
1548 while (offset < s->sh_offset) {
1549 fputc(0, f);
1550 offset++;
1552 size = s->sh_size;
1553 fwrite(s->data, 1, size, f);
1554 offset += size;
1559 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
1560 #define HAVE_PHDR 1
1561 #define EXTRA_RELITEMS 14
1563 /* move the relocation value from .dynsym to .got */
1564 void patch_dynsym_undef(TCCState *s1, Section *s)
1566 uint32_t *gotd = (void *)s1->got->data;
1567 ElfW(Sym) *sym;
1569 gotd += 3; /* dummy entries in .got */
1570 /* relocate symbols in .dynsym */
1571 for_each_elem(s, 1, sym, ElfW(Sym)) {
1572 if (sym->st_shndx == SHN_UNDEF) {
1573 *gotd++ = sym->st_value + 6; /* XXX 6 is magic ? */
1574 sym->st_value = 0;
1578 #else
1579 #define HAVE_PHDR 1
1580 #define EXTRA_RELITEMS 9
1582 /* zero plt offsets of weak symbols in .dynsym */
1583 void patch_dynsym_undef(TCCState *s1, Section *s)
1585 ElfW(Sym) *sym;
1587 for_each_elem(s, 1, sym, ElfW(Sym))
1588 if (sym->st_shndx == SHN_UNDEF && ELFW(ST_BIND)(sym->st_info) == STB_WEAK)
1589 sym->st_value = 0;
1591 #endif
1593 ST_FUNC void fill_got_entry(TCCState *s1, ElfW_Rel *rel)
1595 int sym_index = ELFW(R_SYM) (rel->r_info);
1596 ElfW(Sym) *sym = &((ElfW(Sym) *) symtab_section->data)[sym_index];
1597 unsigned long offset;
1599 if (sym_index >= s1->nb_sym_attrs)
1600 return;
1601 offset = s1->sym_attrs[sym_index].got_offset;
1602 section_reserve(s1->got, offset + PTR_SIZE);
1603 #ifdef TCC_TARGET_X86_64
1604 /* only works for x86-64 */
1605 put32(s1->got->data + offset + 4, sym->st_value >> 32);
1606 #endif
1607 put32(s1->got->data + offset, sym->st_value & 0xffffffff);
1610 /* Perform relocation to GOT or PLT entries */
1611 ST_FUNC void fill_got(TCCState *s1)
1613 Section *s;
1614 ElfW_Rel *rel;
1615 int i;
1617 for(i = 1; i < s1->nb_sections; i++) {
1618 s = s1->sections[i];
1619 if (s->sh_type != SHT_RELX)
1620 continue;
1621 /* no need to handle got relocations */
1622 if (s->link != symtab_section)
1623 continue;
1624 for_each_elem(s, 0, rel, ElfW_Rel) {
1625 switch (ELFW(R_TYPE) (rel->r_info)) {
1626 case R_X86_64_GOT32:
1627 case R_X86_64_GOTPCREL:
1628 case R_X86_64_PLT32:
1629 fill_got_entry(s1, rel);
1630 break;
1636 /* Bind symbols of executable: resolve undefined symbols from exported symbols
1637 in shared libraries and export non local defined symbols to shared libraries
1638 if -rdynamic switch was given on command line */
1639 static void bind_exe_dynsyms(TCCState *s1)
1641 const char *name;
1642 int sym_index, index;
1643 ElfW(Sym) *sym, *esym;
1644 int type;
1646 /* Resolve undefined symbols from dynamic symbols. When there is a match:
1647 - if STT_FUNC or STT_GNU_IFUNC symbol -> add it in PLT
1648 - if STT_OBJECT symbol -> add it in .bss section with suitable reloc */
1649 for_each_elem(symtab_section, 1, sym, ElfW(Sym)) {
1650 if (sym->st_shndx == SHN_UNDEF) {
1651 name = (char *) symtab_section->link->data + sym->st_name;
1652 sym_index = find_elf_sym(s1->dynsymtab_section, name);
1653 if (sym_index) {
1654 esym = &((ElfW(Sym) *)s1->dynsymtab_section->data)[sym_index];
1655 type = ELFW(ST_TYPE)(esym->st_info);
1656 if ((type == STT_FUNC) || (type == STT_GNU_IFUNC)) {
1657 /* Indirect functions shall have STT_FUNC type in executable
1658 * dynsym section. Indeed, a dlsym call following a lazy
1659 * resolution would pick the symbol value from the
1660 * executable dynsym entry which would contain the address
1661 * of the function wanted by the caller of dlsym instead of
1662 * the address of the function that would return that
1663 * address */
1664 put_got_entry(s1, R_JMP_SLOT, esym->st_size,
1665 ELFW(ST_INFO)(STB_GLOBAL,STT_FUNC),
1666 sym - (ElfW(Sym) *)symtab_section->data);
1667 } else if (type == STT_OBJECT) {
1668 unsigned long offset;
1669 ElfW(Sym) *dynsym;
1670 offset = bss_section->data_offset;
1671 /* XXX: which alignment ? */
1672 offset = (offset + 16 - 1) & -16;
1673 index = put_elf_sym(s1->dynsym, offset, esym->st_size,
1674 esym->st_info, 0, bss_section->sh_num,
1675 name);
1676 /* Ensure R_COPY works for weak symbol aliases */
1677 if (ELFW(ST_BIND)(esym->st_info) == STB_WEAK) {
1678 for_each_elem(s1->dynsymtab_section, 1, dynsym, ElfW(Sym)) {
1679 if ((dynsym->st_value == esym->st_value)
1680 && (ELFW(ST_BIND)(dynsym->st_info) == STB_GLOBAL)) {
1681 char *dynname = (char *) s1->dynsymtab_section->link->data
1682 + dynsym->st_name;
1683 put_elf_sym(s1->dynsym, offset, dynsym->st_size,
1684 dynsym->st_info, 0,
1685 bss_section->sh_num, dynname);
1686 break;
1690 put_elf_reloc(s1->dynsym, bss_section,
1691 offset, R_COPY, index);
1692 offset += esym->st_size;
1693 bss_section->data_offset = offset;
1695 } else {
1696 /* STB_WEAK undefined symbols are accepted */
1697 /* XXX: _fp_hw seems to be part of the ABI, so we ignore it */
1698 if (ELFW(ST_BIND)(sym->st_info) == STB_WEAK ||
1699 !strcmp(name, "_fp_hw")) {
1700 } else {
1701 tcc_error_noabort("undefined symbol '%s'", name);
1704 } else if (s1->rdynamic && ELFW(ST_BIND)(sym->st_info) != STB_LOCAL) {
1705 /* if -rdynamic option, then export all non local symbols */
1706 name = (char *) symtab_section->link->data + sym->st_name;
1707 put_elf_sym(s1->dynsym, sym->st_value, sym->st_size, sym->st_info,
1708 0, sym->st_shndx, name);
1713 /* Bind symbols of libraries: export non local symbols of executable that
1714 resolve undefined symbols of shared libraries */
1715 static void bind_libs_dynsyms(TCCState *s1)
1717 const char *name;
1718 int sym_index;
1719 ElfW(Sym) *sym, *esym;
1721 /* now look at unresolved dynamic symbols and export
1722 corresponding symbol */
1723 for_each_elem(s1->dynsymtab_section, 1, esym, ElfW(Sym)) {
1724 if (esym->st_shndx == SHN_UNDEF) {
1725 name = (char *) s1->dynsymtab_section->link->data + esym->st_name;
1726 sym_index = find_elf_sym(symtab_section, name);
1727 if (sym_index) {
1728 /* XXX: avoid adding a symbol if already present because of
1729 -rdynamic ? */
1730 sym = &((ElfW(Sym) *)symtab_section->data)[sym_index];
1731 put_elf_sym(s1->dynsym, sym->st_value, sym->st_size,
1732 sym->st_info, 0, sym->st_shndx, name);
1733 } else {
1734 /* weak symbols can stay undefined */
1735 if (ELFW(ST_BIND)(esym->st_info) != STB_WEAK)
1736 tcc_warning("undefined dynamic symbol '%s'", name);
1742 /* Export all non local symbols (for shared libraries) */
1743 static void export_global_syms(TCCState *s1)
1745 int nb_syms, dynindex, index;
1746 const char *name;
1747 ElfW(Sym) *sym;
1749 nb_syms = symtab_section->data_offset / sizeof(ElfW(Sym));
1750 s1->symtab_to_dynsym = tcc_mallocz(sizeof(int) * nb_syms);
1751 for_each_elem(symtab_section, 1, sym, ElfW(Sym)) {
1752 if (ELFW(ST_BIND)(sym->st_info) != STB_LOCAL) {
1753 name = (char *) symtab_section->link->data + sym->st_name;
1754 dynindex = put_elf_sym(s1->dynsym, sym->st_value, sym->st_size,
1755 sym->st_info, 0, sym->st_shndx, name);
1756 index = sym - (ElfW(Sym) *) symtab_section->data;
1757 s1->symtab_to_dynsym[index] = dynindex;
1762 /* relocate the PLT: compute addresses and offsets in the PLT now that final
1763 address for PLT and GOT are known (see fill_program_header) */
1764 ST_FUNC void relocate_plt(TCCState *s1)
1766 uint8_t *p, *p_end;
1768 if (!s1->plt)
1769 return;
1771 p = s1->plt->data;
1772 p_end = p + s1->plt->data_offset;
1773 if (p < p_end) {
1774 #if defined(TCC_TARGET_I386)
1775 put32(p + 2, get32(p + 2) + s1->got->sh_addr);
1776 put32(p + 8, get32(p + 8) + s1->got->sh_addr);
1777 p += 16;
1778 while (p < p_end) {
1779 put32(p + 2, get32(p + 2) + s1->got->sh_addr);
1780 p += 16;
1782 #elif defined(TCC_TARGET_X86_64)
1783 int x = s1->got->sh_addr - s1->plt->sh_addr - 6;
1784 put32(p + 2, get32(p + 2) + x);
1785 put32(p + 8, get32(p + 8) + x - 6);
1786 p += 16;
1787 while (p < p_end) {
1788 put32(p + 2, get32(p + 2) + x + s1->plt->data - p);
1789 p += 16;
1791 #elif defined(TCC_TARGET_ARM)
1792 int x;
1793 x=s1->got->sh_addr - s1->plt->sh_addr - 12;
1794 p += 16;
1795 while (p < p_end) {
1796 if (get32(p) == 0x46c04778) /* PLT Thumb stub present */
1797 p += 4;
1798 put32(p + 12, x + get32(p + 12) + s1->plt->data - p);
1799 p += 16;
1801 #elif defined(TCC_TARGET_C67)
1802 /* XXX: TODO */
1803 #else
1804 #error unsupported CPU
1805 #endif
1809 /* Allocate strings for section names and decide if an unallocated section
1810 should be output.
1812 NOTE: the strsec section comes last, so its size is also correct ! */
1813 static void alloc_sec_names(TCCState *s1, int file_type, Section *strsec)
1815 int i;
1816 Section *s;
1818 /* Allocate strings for section names */
1819 for(i = 1; i < s1->nb_sections; i++) {
1820 s = s1->sections[i];
1821 s->sh_name = put_elf_str(strsec, s->name);
1822 /* when generating a DLL, we include relocations but we may
1823 patch them */
1824 if (file_type == TCC_OUTPUT_DLL &&
1825 s->sh_type == SHT_RELX &&
1826 !(s->sh_flags & SHF_ALLOC)) {
1827 /* gr: avoid bogus relocs for empty (debug) sections */
1828 if (s1->sections[s->sh_info]->sh_flags & SHF_ALLOC)
1829 prepare_dynamic_rel(s1, s);
1830 else if (s1->do_debug)
1831 s->sh_size = s->data_offset;
1832 } else if (s1->do_debug ||
1833 file_type == TCC_OUTPUT_OBJ ||
1834 (s->sh_flags & SHF_ALLOC) ||
1835 i == (s1->nb_sections - 1)) {
1836 /* we output all sections if debug or object file */
1837 s->sh_size = s->data_offset;
1842 /* Info to be copied in dynamic section */
1843 struct dyn_inf {
1844 Section *dynamic;
1845 Section *dynstr;
1846 unsigned long dyn_rel_off;
1847 addr_t rel_addr;
1848 addr_t rel_size;
1849 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
1850 addr_t bss_addr;
1851 addr_t bss_size;
1852 #endif
1855 /* Assign sections to segments and decide how are sections laid out when loaded
1856 in memory. This function also fills corresponding program headers. */
1857 static int layout_sections(TCCState *s1, ElfW(Phdr) *phdr, int phnum,
1858 Section *interp, struct dyn_inf *dyninf,
1859 int *sec_order)
1861 int i, j, k, file_type, sh_order_index, file_offset;
1862 long long tmp;
1863 addr_t addr;
1864 ElfW(Phdr) *ph;
1865 Section *s;
1867 file_type = s1->output_type;
1868 sh_order_index = 1;
1869 if (s1->output_format == TCC_OUTPUT_FORMAT_ELF)
1870 file_offset = sizeof(ElfW(Ehdr)) + phnum * sizeof(ElfW(Phdr));
1871 else
1872 file_offset = 0;
1874 if (phnum > 0) {
1875 if (s1->has_text_addr) {
1876 int a_offset, p_offset;
1877 addr = s1->text_addr;
1878 /* we ensure that (addr % ELF_PAGE_SIZE) == file_offset %
1879 ELF_PAGE_SIZE */
1880 a_offset = (int) (addr & (s1->section_align - 1));
1881 p_offset = file_offset & (s1->section_align - 1);
1882 if (a_offset < p_offset)
1883 a_offset += s1->section_align;
1884 file_offset += (a_offset - p_offset);
1885 } else {
1886 if (file_type == TCC_OUTPUT_DLL)
1887 addr = 0;
1888 else
1889 addr = ELF_START_ADDR;
1890 /* compute address after headers */
1891 addr += (file_offset & (s1->section_align - 1));
1894 ph = &phdr[0];
1895 /* Leave one program headers for the program interpreter and one for
1896 the program header table itself if needed. These are done later as
1897 they require section layout to be done first. */
1898 if (interp)
1899 ph += 1 + HAVE_PHDR;
1901 /* dynamic relocation table information, for .dynamic section */
1902 dyninf->rel_addr = dyninf->rel_size = 0;
1903 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
1904 dyninf->bss_addr = dyninf->bss_size = 0;
1905 #endif
1907 for(j = 0; j < 2; j++) {
1908 ph->p_type = PT_LOAD;
1909 if (j == 0)
1910 ph->p_flags = PF_R | PF_X;
1911 else
1912 ph->p_flags = PF_R | PF_W;
1913 ph->p_align = s1->section_align;
1915 /* Decide the layout of sections loaded in memory. This must
1916 be done before program headers are filled since they contain
1917 info about the layout. We do the following ordering: interp,
1918 symbol tables, relocations, progbits, nobits */
1919 /* XXX: do faster and simpler sorting */
1920 for(k = 0; k < 5; k++) {
1921 for(i = 1; i < s1->nb_sections; i++) {
1922 s = s1->sections[i];
1923 /* compute if section should be included */
1924 if (j == 0) {
1925 if ((s->sh_flags & (SHF_ALLOC | SHF_WRITE)) !=
1926 SHF_ALLOC)
1927 continue;
1928 } else {
1929 if ((s->sh_flags & (SHF_ALLOC | SHF_WRITE)) !=
1930 (SHF_ALLOC | SHF_WRITE))
1931 continue;
1933 if (s == interp) {
1934 if (k != 0)
1935 continue;
1936 } else if (s->sh_type == SHT_DYNSYM ||
1937 s->sh_type == SHT_STRTAB ||
1938 s->sh_type == SHT_HASH) {
1939 if (k != 1)
1940 continue;
1941 } else if (s->sh_type == SHT_RELX) {
1942 if (k != 2)
1943 continue;
1944 } else if (s->sh_type == SHT_NOBITS) {
1945 if (k != 4)
1946 continue;
1947 } else {
1948 if (k != 3)
1949 continue;
1951 sec_order[sh_order_index++] = i;
1953 /* section matches: we align it and add its size */
1954 tmp = addr;
1955 addr = (addr + s->sh_addralign - 1) &
1956 ~(s->sh_addralign - 1);
1957 file_offset += (int) ( addr - tmp );
1958 s->sh_offset = file_offset;
1959 s->sh_addr = addr;
1961 /* update program header infos */
1962 if (ph->p_offset == 0) {
1963 ph->p_offset = file_offset;
1964 ph->p_vaddr = addr;
1965 ph->p_paddr = ph->p_vaddr;
1967 /* update dynamic relocation infos */
1968 if (s->sh_type == SHT_RELX) {
1969 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
1970 if (!strcmp(strsec->data + s->sh_name, ".rel.got")) {
1971 dyninf->rel_addr = addr;
1972 dyninf->rel_size += s->sh_size; /* XXX only first rel. */
1974 if (!strcmp(strsec->data + s->sh_name, ".rel.bss")) {
1975 dyninf->bss_addr = addr;
1976 dyninf->bss_size = s->sh_size; /* XXX only first rel. */
1978 #else
1979 if (dyninf->rel_size == 0)
1980 dyninf->rel_addr = addr;
1981 dyninf->rel_size += s->sh_size;
1982 #endif
1984 addr += s->sh_size;
1985 if (s->sh_type != SHT_NOBITS)
1986 file_offset += s->sh_size;
1989 if (j == 0) {
1990 /* Make the first PT_LOAD segment include the program
1991 headers itself (and the ELF header as well), it'll
1992 come out with same memory use but will make various
1993 tools like binutils strip work better. */
1994 ph->p_offset &= ~(ph->p_align - 1);
1995 ph->p_vaddr &= ~(ph->p_align - 1);
1996 ph->p_paddr &= ~(ph->p_align - 1);
1998 ph->p_filesz = file_offset - ph->p_offset;
1999 ph->p_memsz = addr - ph->p_vaddr;
2000 ph++;
2001 if (j == 0) {
2002 if (s1->output_format == TCC_OUTPUT_FORMAT_ELF) {
2003 /* if in the middle of a page, we duplicate the page in
2004 memory so that one copy is RX and the other is RW */
2005 if ((addr & (s1->section_align - 1)) != 0)
2006 addr += s1->section_align;
2007 } else {
2008 addr = (addr + s1->section_align - 1) & ~(s1->section_align - 1);
2009 file_offset = (file_offset + s1->section_align - 1) &
2010 ~(s1->section_align - 1);
2016 /* all other sections come after */
2017 for(i = 1; i < s1->nb_sections; i++) {
2018 s = s1->sections[i];
2019 if (phnum > 0 && (s->sh_flags & SHF_ALLOC))
2020 continue;
2021 sec_order[sh_order_index++] = i;
2023 file_offset = (file_offset + s->sh_addralign - 1) &
2024 ~(s->sh_addralign - 1);
2025 s->sh_offset = file_offset;
2026 if (s->sh_type != SHT_NOBITS)
2027 file_offset += s->sh_size;
2030 return file_offset;
2033 static void fill_unloadable_phdr(ElfW(Phdr) *phdr, int phnum, Section *interp,
2034 Section *dynamic)
2036 ElfW(Phdr) *ph;
2038 /* if interpreter, then add corresponding program header */
2039 if (interp) {
2040 ph = &phdr[0];
2042 if (HAVE_PHDR)
2044 int len = phnum * sizeof(ElfW(Phdr));
2046 ph->p_type = PT_PHDR;
2047 ph->p_offset = sizeof(ElfW(Ehdr));
2048 ph->p_vaddr = interp->sh_addr - len;
2049 ph->p_paddr = ph->p_vaddr;
2050 ph->p_filesz = ph->p_memsz = len;
2051 ph->p_flags = PF_R | PF_X;
2052 ph->p_align = 4; /* interp->sh_addralign; */
2053 ph++;
2056 ph->p_type = PT_INTERP;
2057 ph->p_offset = interp->sh_offset;
2058 ph->p_vaddr = interp->sh_addr;
2059 ph->p_paddr = ph->p_vaddr;
2060 ph->p_filesz = interp->sh_size;
2061 ph->p_memsz = interp->sh_size;
2062 ph->p_flags = PF_R;
2063 ph->p_align = interp->sh_addralign;
2066 /* if dynamic section, then add corresponding program header */
2067 if (dynamic) {
2068 ph = &phdr[phnum - 1];
2070 ph->p_type = PT_DYNAMIC;
2071 ph->p_offset = dynamic->sh_offset;
2072 ph->p_vaddr = dynamic->sh_addr;
2073 ph->p_paddr = ph->p_vaddr;
2074 ph->p_filesz = dynamic->sh_size;
2075 ph->p_memsz = dynamic->sh_size;
2076 ph->p_flags = PF_R | PF_W;
2077 ph->p_align = dynamic->sh_addralign;
2081 /* Fill the dynamic section with tags describing the address and size of
2082 sections */
2083 static void fill_dynamic(TCCState *s1, struct dyn_inf *dyninf)
2085 Section *dynamic;
2087 dynamic = dyninf->dynamic;
2089 /* put dynamic section entries */
2090 dynamic->data_offset = dyninf->dyn_rel_off;
2091 put_dt(dynamic, DT_HASH, s1->dynsym->hash->sh_addr);
2092 put_dt(dynamic, DT_STRTAB, dyninf->dynstr->sh_addr);
2093 put_dt(dynamic, DT_SYMTAB, s1->dynsym->sh_addr);
2094 put_dt(dynamic, DT_STRSZ, dyninf->dynstr->data_offset);
2095 put_dt(dynamic, DT_SYMENT, sizeof(ElfW(Sym)));
2096 #ifdef TCC_TARGET_X86_64
2097 put_dt(dynamic, DT_RELA, dyninf->rel_addr);
2098 put_dt(dynamic, DT_RELASZ, dyninf->rel_size);
2099 put_dt(dynamic, DT_RELAENT, sizeof(ElfW_Rel));
2100 #else
2101 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
2102 put_dt(dynamic, DT_PLTGOT, s1->got->sh_addr);
2103 put_dt(dynamic, DT_PLTRELSZ, dyninf->rel_size);
2104 put_dt(dynamic, DT_JMPREL, dyninf->rel_addr);
2105 put_dt(dynamic, DT_PLTREL, DT_REL);
2106 put_dt(dynamic, DT_REL, dyninf->bss_addr);
2107 put_dt(dynamic, DT_RELSZ, dyninf->bss_size);
2108 #else
2109 put_dt(dynamic, DT_REL, dyninf->rel_addr);
2110 put_dt(dynamic, DT_RELSZ, dyninf->rel_size);
2111 put_dt(dynamic, DT_RELENT, sizeof(ElfW_Rel));
2112 #endif
2113 #endif
2114 if (s1->do_debug)
2115 put_dt(dynamic, DT_DEBUG, 0);
2116 put_dt(dynamic, DT_NULL, 0);
2119 /* Relocate remaining sections and symbols (that is those not related to
2120 dynamic linking) */
2121 static int final_sections_reloc(TCCState *s1)
2123 int i;
2124 Section *s;
2126 relocate_syms(s1, 0);
2128 if (s1->nb_errors != 0)
2129 return -1;
2131 /* relocate sections */
2132 /* XXX: ignore sections with allocated relocations ? */
2133 for(i = 1; i < s1->nb_sections; i++) {
2134 s = s1->sections[i];
2135 if (s->reloc && s != s1->got)
2136 relocate_section(s1, s);
2139 /* relocate relocation entries if the relocation tables are
2140 allocated in the executable */
2141 for(i = 1; i < s1->nb_sections; i++) {
2142 s = s1->sections[i];
2143 if ((s->sh_flags & SHF_ALLOC) &&
2144 s->sh_type == SHT_RELX) {
2145 relocate_rel(s1, s);
2148 return 0;
2151 /* Create an ELF file on disk.
2152 This function handle ELF specific layout requirements */
2153 static void tcc_output_elf(TCCState *s1, FILE *f, int phnum, ElfW(Phdr) *phdr,
2154 int file_offset, int *sec_order)
2156 int i, shnum, offset, size, file_type;
2157 Section *s;
2158 ElfW(Ehdr) ehdr;
2159 ElfW(Shdr) shdr, *sh;
2161 file_type = s1->output_type;
2162 shnum = s1->nb_sections;
2164 memset(&ehdr, 0, sizeof(ehdr));
2166 if (phnum > 0) {
2167 ehdr.e_phentsize = sizeof(ElfW(Phdr));
2168 ehdr.e_phnum = phnum;
2169 ehdr.e_phoff = sizeof(ElfW(Ehdr));
2172 /* align to 4 */
2173 file_offset = (file_offset + 3) & -4;
2175 /* fill header */
2176 ehdr.e_ident[0] = ELFMAG0;
2177 ehdr.e_ident[1] = ELFMAG1;
2178 ehdr.e_ident[2] = ELFMAG2;
2179 ehdr.e_ident[3] = ELFMAG3;
2180 ehdr.e_ident[4] = ELFCLASSW;
2181 ehdr.e_ident[5] = ELFDATA2LSB;
2182 ehdr.e_ident[6] = EV_CURRENT;
2183 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
2184 ehdr.e_ident[EI_OSABI] = ELFOSABI_FREEBSD;
2185 #endif
2186 #ifdef TCC_TARGET_ARM
2187 #ifdef TCC_ARM_EABI
2188 ehdr.e_ident[EI_OSABI] = 0;
2189 ehdr.e_flags = EF_ARM_EABI_VER4;
2190 if (file_type == TCC_OUTPUT_EXE || file_type == TCC_OUTPUT_DLL)
2191 ehdr.e_flags |= EF_ARM_HASENTRY;
2192 if (s1->float_abi == ARM_HARD_FLOAT)
2193 ehdr.e_flags |= EF_ARM_VFP_FLOAT;
2194 else
2195 ehdr.e_flags |= EF_ARM_SOFT_FLOAT;
2196 #else
2197 ehdr.e_ident[EI_OSABI] = ELFOSABI_ARM;
2198 #endif
2199 #endif
2200 switch(file_type) {
2201 default:
2202 case TCC_OUTPUT_EXE:
2203 ehdr.e_type = ET_EXEC;
2204 ehdr.e_entry = get_elf_sym_addr(s1, "_start", 1);
2205 break;
2206 case TCC_OUTPUT_DLL:
2207 ehdr.e_type = ET_DYN;
2208 ehdr.e_entry = text_section->sh_addr; /* XXX: is it correct ? */
2209 break;
2210 case TCC_OUTPUT_OBJ:
2211 ehdr.e_type = ET_REL;
2212 break;
2214 ehdr.e_machine = EM_TCC_TARGET;
2215 ehdr.e_version = EV_CURRENT;
2216 ehdr.e_shoff = file_offset;
2217 ehdr.e_ehsize = sizeof(ElfW(Ehdr));
2218 ehdr.e_shentsize = sizeof(ElfW(Shdr));
2219 ehdr.e_shnum = shnum;
2220 ehdr.e_shstrndx = shnum - 1;
2222 fwrite(&ehdr, 1, sizeof(ElfW(Ehdr)), f);
2223 fwrite(phdr, 1, phnum * sizeof(ElfW(Phdr)), f);
2224 offset = sizeof(ElfW(Ehdr)) + phnum * sizeof(ElfW(Phdr));
2226 sort_syms(s1, symtab_section);
2227 for(i = 1; i < s1->nb_sections; i++) {
2228 s = s1->sections[sec_order[i]];
2229 if (s->sh_type != SHT_NOBITS) {
2230 if (s->sh_type == SHT_DYNSYM)
2231 patch_dynsym_undef(s1, s);
2232 while (offset < s->sh_offset) {
2233 fputc(0, f);
2234 offset++;
2236 size = s->sh_size;
2237 fwrite(s->data, 1, size, f);
2238 offset += size;
2242 /* output section headers */
2243 while (offset < ehdr.e_shoff) {
2244 fputc(0, f);
2245 offset++;
2248 for(i = 0; i < s1->nb_sections; i++) {
2249 sh = &shdr;
2250 memset(sh, 0, sizeof(ElfW(Shdr)));
2251 s = s1->sections[i];
2252 if (s) {
2253 sh->sh_name = s->sh_name;
2254 sh->sh_type = s->sh_type;
2255 sh->sh_flags = s->sh_flags;
2256 sh->sh_entsize = s->sh_entsize;
2257 sh->sh_info = s->sh_info;
2258 if (s->link)
2259 sh->sh_link = s->link->sh_num;
2260 sh->sh_addralign = s->sh_addralign;
2261 sh->sh_addr = s->sh_addr;
2262 sh->sh_offset = s->sh_offset;
2263 sh->sh_size = s->sh_size;
2265 fwrite(sh, 1, sizeof(ElfW(Shdr)), f);
2269 /* Write an elf, coff or "binary" file */
2270 static int tcc_write_elf_file(TCCState *s1, const char *filename, int phnum,
2271 ElfW(Phdr) *phdr, int file_offset, int *sec_order)
2273 int fd, mode, file_type;
2274 FILE *f;
2276 file_type = s1->output_type;
2277 if (file_type == TCC_OUTPUT_OBJ)
2278 mode = 0666;
2279 else
2280 mode = 0777;
2281 unlink(filename);
2282 fd = open(filename, O_WRONLY | O_CREAT | O_TRUNC | O_BINARY, mode);
2283 if (fd < 0) {
2284 tcc_error_noabort("could not write '%s'", filename);
2285 return -1;
2287 f = fdopen(fd, "wb");
2288 if (s1->verbose)
2289 printf("<- %s\n", filename);
2291 #ifdef TCC_TARGET_COFF
2292 if (s1->output_format == TCC_OUTPUT_FORMAT_COFF)
2293 tcc_output_coff(s1, f);
2294 else
2295 #endif
2296 if (s1->output_format == TCC_OUTPUT_FORMAT_ELF)
2297 tcc_output_elf(s1, f, phnum, phdr, file_offset, sec_order);
2298 else
2299 tcc_output_binary(s1, f, sec_order);
2300 fclose(f);
2302 return 0;
2305 /* Output an elf, coff or binary file */
2306 /* XXX: suppress unneeded sections */
2307 static int elf_output_file(TCCState *s1, const char *filename)
2309 int i, ret, phnum, shnum, file_type, file_offset, *sec_order;
2310 struct dyn_inf dyninf;
2311 ElfW(Phdr) *phdr;
2312 ElfW(Sym) *sym;
2313 Section *strsec, *interp, *dynamic, *dynstr;
2315 file_type = s1->output_type;
2316 s1->nb_errors = 0;
2318 /* if linking, also link in runtime libraries (libc, libgcc, etc.) */
2319 if (file_type != TCC_OUTPUT_OBJ) {
2320 tcc_add_runtime(s1);
2323 phdr = NULL;
2324 sec_order = NULL;
2325 interp = dynamic = dynstr = NULL; /* avoid warning */
2326 dyninf.dyn_rel_off = 0; /* avoid warning */
2328 if (file_type != TCC_OUTPUT_OBJ) {
2329 relocate_common_syms();
2331 tcc_add_linker_symbols(s1);
2333 if (!s1->static_link) {
2334 if (file_type == TCC_OUTPUT_EXE) {
2335 char *ptr;
2336 /* allow override the dynamic loader */
2337 const char *elfint = getenv("LD_SO");
2338 if (elfint == NULL)
2339 elfint = DEFAULT_ELFINTERP(s1);
2340 /* add interpreter section only if executable */
2341 interp = new_section(s1, ".interp", SHT_PROGBITS, SHF_ALLOC);
2342 interp->sh_addralign = 1;
2343 ptr = section_ptr_add(interp, 1 + strlen(elfint));
2344 strcpy(ptr, elfint);
2347 /* add dynamic symbol table */
2348 s1->dynsym = new_symtab(s1, ".dynsym", SHT_DYNSYM, SHF_ALLOC,
2349 ".dynstr",
2350 ".hash", SHF_ALLOC);
2351 dynstr = s1->dynsym->link;
2353 /* add dynamic section */
2354 dynamic = new_section(s1, ".dynamic", SHT_DYNAMIC,
2355 SHF_ALLOC | SHF_WRITE);
2356 dynamic->link = dynstr;
2357 dynamic->sh_entsize = sizeof(ElfW(Dyn));
2359 build_got(s1);
2361 if (file_type == TCC_OUTPUT_EXE) {
2362 bind_exe_dynsyms(s1);
2364 if (s1->nb_errors) {
2365 ret = -1;
2366 goto the_end;
2369 bind_libs_dynsyms(s1);
2370 } else /* shared library case: simply export all global symbols */
2371 export_global_syms(s1);
2373 build_got_entries(s1);
2375 /* add a list of needed dlls */
2376 for(i = 0; i < s1->nb_loaded_dlls; i++) {
2377 DLLReference *dllref = s1->loaded_dlls[i];
2378 if (dllref->level == 0)
2379 put_dt(dynamic, DT_NEEDED, put_elf_str(dynstr, dllref->name));
2382 if (s1->rpath)
2383 put_dt(dynamic, DT_RPATH, put_elf_str(dynstr, s1->rpath));
2385 /* XXX: currently, since we do not handle PIC code, we
2386 must relocate the readonly segments */
2387 if (file_type == TCC_OUTPUT_DLL) {
2388 if (s1->soname)
2389 put_dt(dynamic, DT_SONAME, put_elf_str(dynstr, s1->soname));
2390 put_dt(dynamic, DT_TEXTREL, 0);
2393 if (s1->symbolic)
2394 put_dt(dynamic, DT_SYMBOLIC, 0);
2396 /* add necessary space for other entries */
2397 dyninf.dyn_rel_off = dynamic->data_offset;
2398 dynamic->data_offset += sizeof(ElfW(Dyn)) * EXTRA_RELITEMS;
2399 } else {
2400 /* still need to build got entries in case of static link */
2401 build_got_entries(s1);
2405 /* we add a section for symbols */
2406 strsec = new_section(s1, ".shstrtab", SHT_STRTAB, 0);
2407 put_elf_str(strsec, "");
2409 /* compute number of sections */
2410 shnum = s1->nb_sections;
2412 /* this array is used to reorder sections in the output file */
2413 sec_order = tcc_malloc(sizeof(int) * shnum);
2414 sec_order[0] = 0;
2416 /* compute number of program headers */
2417 switch(file_type) {
2418 default:
2419 case TCC_OUTPUT_OBJ:
2420 phnum = 0;
2421 break;
2422 case TCC_OUTPUT_EXE:
2423 if (!s1->static_link)
2424 phnum = 4 + HAVE_PHDR;
2425 else
2426 phnum = 2;
2427 break;
2428 case TCC_OUTPUT_DLL:
2429 phnum = 3;
2430 break;
2433 /* Allocate strings for section names */
2434 alloc_sec_names(s1, file_type, strsec);
2436 /* allocate program segment headers */
2437 phdr = tcc_mallocz(phnum * sizeof(ElfW(Phdr)));
2439 /* compute section to program header mapping */
2440 file_offset = layout_sections(s1, phdr, phnum, interp, &dyninf, sec_order);
2442 /* Fill remaining program header and finalize relocation related to dynamic
2443 linking. */
2444 if (phnum > 0) {
2445 fill_unloadable_phdr(phdr, phnum, interp, dynamic);
2446 if (dynamic) {
2447 dyninf.dynamic = dynamic;
2448 dyninf.dynstr = dynstr;
2450 fill_dynamic(s1, &dyninf);
2452 /* put in GOT the dynamic section address and relocate PLT */
2453 put32(s1->got->data, dynamic->sh_addr);
2454 if (file_type == TCC_OUTPUT_EXE
2455 #if defined(TCC_OUTPUT_DLL_WITH_PLT)
2456 || file_type == TCC_OUTPUT_DLL
2457 #endif
2459 relocate_plt(s1);
2461 /* relocate symbols in .dynsym now that final addresses are known */
2462 for_each_elem(s1->dynsym, 1, sym, ElfW(Sym)) {
2463 /* relocate to PLT if symbol corresponds to a PLT entry */
2464 if (sym->st_shndx == SHN_UNDEF) {
2465 if (sym->st_value)
2466 sym->st_value += s1->plt->sh_addr;
2467 } else if (sym->st_shndx < SHN_LORESERVE) {
2468 /* do symbol relocation */
2469 sym->st_value += s1->sections[sym->st_shndx]->sh_addr;
2475 /* if building executable or DLL, then relocate each section
2476 except the GOT which is already relocated */
2477 if (file_type != TCC_OUTPUT_OBJ) {
2478 ret = final_sections_reloc(s1);
2479 if (ret)
2480 goto the_end;
2483 /* Perform relocation to GOT or PLT entries */
2484 if (file_type == TCC_OUTPUT_EXE && s1->static_link)
2485 fill_got(s1);
2487 /* Create the ELF file with name 'filename' */
2488 ret = tcc_write_elf_file(s1, filename, phnum, phdr, file_offset, sec_order);
2489 the_end:
2490 tcc_free(s1->symtab_to_dynsym);
2491 tcc_free(sec_order);
2492 tcc_free(phdr);
2493 tcc_free(s1->sym_attrs);
2494 s1->sym_attrs = NULL;
2495 return ret;
2498 LIBTCCAPI int tcc_output_file(TCCState *s, const char *filename)
2500 int ret;
2501 #ifdef TCC_TARGET_PE
2502 if (s->output_type != TCC_OUTPUT_OBJ) {
2503 ret = pe_output_file(s, filename);
2504 } else
2505 #endif
2506 ret = elf_output_file(s, filename);
2507 return ret;
2510 static void *load_data(int fd, unsigned long file_offset, unsigned long size)
2512 void *data;
2514 data = tcc_malloc(size);
2515 lseek(fd, file_offset, SEEK_SET);
2516 read(fd, data, size);
2517 return data;
2520 typedef struct SectionMergeInfo {
2521 Section *s; /* corresponding existing section */
2522 unsigned long offset; /* offset of the new section in the existing section */
2523 uint8_t new_section; /* true if section 's' was added */
2524 uint8_t link_once; /* true if link once section */
2525 } SectionMergeInfo;
2527 /* load an object file and merge it with current files */
2528 /* XXX: handle correctly stab (debug) info */
2529 ST_FUNC int tcc_load_object_file(TCCState *s1,
2530 int fd, unsigned long file_offset)
2532 ElfW(Ehdr) ehdr;
2533 ElfW(Shdr) *shdr, *sh;
2534 int size, i, j, offset, offseti, nb_syms, sym_index, ret;
2535 unsigned char *strsec, *strtab;
2536 int *old_to_new_syms;
2537 char *sh_name, *name;
2538 SectionMergeInfo *sm_table, *sm;
2539 ElfW(Sym) *sym, *symtab;
2540 ElfW_Rel *rel;
2541 Section *s;
2543 int stab_index;
2544 int stabstr_index;
2546 stab_index = stabstr_index = 0;
2548 if (read(fd, &ehdr, sizeof(ehdr)) != sizeof(ehdr))
2549 goto fail1;
2550 if (ehdr.e_ident[0] != ELFMAG0 ||
2551 ehdr.e_ident[1] != ELFMAG1 ||
2552 ehdr.e_ident[2] != ELFMAG2 ||
2553 ehdr.e_ident[3] != ELFMAG3)
2554 goto fail1;
2555 /* test if object file */
2556 if (ehdr.e_type != ET_REL)
2557 goto fail1;
2558 /* test CPU specific stuff */
2559 if (ehdr.e_ident[5] != ELFDATA2LSB ||
2560 ehdr.e_machine != EM_TCC_TARGET) {
2561 fail1:
2562 tcc_error_noabort("invalid object file");
2563 return -1;
2565 /* read sections */
2566 shdr = load_data(fd, file_offset + ehdr.e_shoff,
2567 sizeof(ElfW(Shdr)) * ehdr.e_shnum);
2568 sm_table = tcc_mallocz(sizeof(SectionMergeInfo) * ehdr.e_shnum);
2570 /* load section names */
2571 sh = &shdr[ehdr.e_shstrndx];
2572 strsec = load_data(fd, file_offset + sh->sh_offset, sh->sh_size);
2574 /* load symtab and strtab */
2575 old_to_new_syms = NULL;
2576 symtab = NULL;
2577 strtab = NULL;
2578 nb_syms = 0;
2579 for(i = 1; i < ehdr.e_shnum; i++) {
2580 sh = &shdr[i];
2581 if (sh->sh_type == SHT_SYMTAB) {
2582 if (symtab) {
2583 tcc_error_noabort("object must contain only one symtab");
2584 fail:
2585 ret = -1;
2586 goto the_end;
2588 nb_syms = sh->sh_size / sizeof(ElfW(Sym));
2589 symtab = load_data(fd, file_offset + sh->sh_offset, sh->sh_size);
2590 sm_table[i].s = symtab_section;
2592 /* now load strtab */
2593 sh = &shdr[sh->sh_link];
2594 strtab = load_data(fd, file_offset + sh->sh_offset, sh->sh_size);
2598 /* now examine each section and try to merge its content with the
2599 ones in memory */
2600 for(i = 1; i < ehdr.e_shnum; i++) {
2601 /* no need to examine section name strtab */
2602 if (i == ehdr.e_shstrndx)
2603 continue;
2604 sh = &shdr[i];
2605 sh_name = (char *) strsec + sh->sh_name;
2606 /* ignore sections types we do not handle */
2607 if (sh->sh_type != SHT_PROGBITS &&
2608 sh->sh_type != SHT_RELX &&
2609 #ifdef TCC_ARM_EABI
2610 sh->sh_type != SHT_ARM_EXIDX &&
2611 #endif
2612 sh->sh_type != SHT_NOBITS &&
2613 sh->sh_type != SHT_PREINIT_ARRAY &&
2614 sh->sh_type != SHT_INIT_ARRAY &&
2615 sh->sh_type != SHT_FINI_ARRAY &&
2616 strcmp(sh_name, ".stabstr")
2618 continue;
2619 if (sh->sh_addralign < 1)
2620 sh->sh_addralign = 1;
2621 /* find corresponding section, if any */
2622 for(j = 1; j < s1->nb_sections;j++) {
2623 s = s1->sections[j];
2624 if (!strcmp(s->name, sh_name)) {
2625 if (!strncmp(sh_name, ".gnu.linkonce",
2626 sizeof(".gnu.linkonce") - 1)) {
2627 /* if a 'linkonce' section is already present, we
2628 do not add it again. It is a little tricky as
2629 symbols can still be defined in
2630 it. */
2631 sm_table[i].link_once = 1;
2632 goto next;
2633 } else {
2634 goto found;
2638 /* not found: create new section */
2639 s = new_section(s1, sh_name, sh->sh_type, sh->sh_flags);
2640 /* take as much info as possible from the section. sh_link and
2641 sh_info will be updated later */
2642 s->sh_addralign = sh->sh_addralign;
2643 s->sh_entsize = sh->sh_entsize;
2644 sm_table[i].new_section = 1;
2645 found:
2646 if (sh->sh_type != s->sh_type) {
2647 tcc_error_noabort("invalid section type");
2648 goto fail;
2651 /* align start of section */
2652 offset = s->data_offset;
2654 if (0 == strcmp(sh_name, ".stab")) {
2655 stab_index = i;
2656 goto no_align;
2658 if (0 == strcmp(sh_name, ".stabstr")) {
2659 stabstr_index = i;
2660 goto no_align;
2663 size = sh->sh_addralign - 1;
2664 offset = (offset + size) & ~size;
2665 if (sh->sh_addralign > s->sh_addralign)
2666 s->sh_addralign = sh->sh_addralign;
2667 s->data_offset = offset;
2668 no_align:
2669 sm_table[i].offset = offset;
2670 sm_table[i].s = s;
2671 /* concatenate sections */
2672 size = sh->sh_size;
2673 if (sh->sh_type != SHT_NOBITS) {
2674 unsigned char *ptr;
2675 lseek(fd, file_offset + sh->sh_offset, SEEK_SET);
2676 ptr = section_ptr_add(s, size);
2677 read(fd, ptr, size);
2678 } else {
2679 s->data_offset += size;
2681 next: ;
2684 /* gr relocate stab strings */
2685 if (stab_index && stabstr_index) {
2686 Stab_Sym *a, *b;
2687 unsigned o;
2688 s = sm_table[stab_index].s;
2689 a = (Stab_Sym *)(s->data + sm_table[stab_index].offset);
2690 b = (Stab_Sym *)(s->data + s->data_offset);
2691 o = sm_table[stabstr_index].offset;
2692 while (a < b)
2693 a->n_strx += o, a++;
2696 /* second short pass to update sh_link and sh_info fields of new
2697 sections */
2698 for(i = 1; i < ehdr.e_shnum; i++) {
2699 s = sm_table[i].s;
2700 if (!s || !sm_table[i].new_section)
2701 continue;
2702 sh = &shdr[i];
2703 if (sh->sh_link > 0)
2704 s->link = sm_table[sh->sh_link].s;
2705 if (sh->sh_type == SHT_RELX) {
2706 s->sh_info = sm_table[sh->sh_info].s->sh_num;
2707 /* update backward link */
2708 s1->sections[s->sh_info]->reloc = s;
2711 sm = sm_table;
2713 /* resolve symbols */
2714 old_to_new_syms = tcc_mallocz(nb_syms * sizeof(int));
2716 sym = symtab + 1;
2717 for(i = 1; i < nb_syms; i++, sym++) {
2718 if (sym->st_shndx != SHN_UNDEF &&
2719 sym->st_shndx < SHN_LORESERVE) {
2720 sm = &sm_table[sym->st_shndx];
2721 if (sm->link_once) {
2722 /* if a symbol is in a link once section, we use the
2723 already defined symbol. It is very important to get
2724 correct relocations */
2725 if (ELFW(ST_BIND)(sym->st_info) != STB_LOCAL) {
2726 name = (char *) strtab + sym->st_name;
2727 sym_index = find_elf_sym(symtab_section, name);
2728 if (sym_index)
2729 old_to_new_syms[i] = sym_index;
2731 continue;
2733 /* if no corresponding section added, no need to add symbol */
2734 if (!sm->s)
2735 continue;
2736 /* convert section number */
2737 sym->st_shndx = sm->s->sh_num;
2738 /* offset value */
2739 sym->st_value += sm->offset;
2741 /* add symbol */
2742 name = (char *) strtab + sym->st_name;
2743 sym_index = add_elf_sym(symtab_section, sym->st_value, sym->st_size,
2744 sym->st_info, sym->st_other,
2745 sym->st_shndx, name);
2746 old_to_new_syms[i] = sym_index;
2749 /* third pass to patch relocation entries */
2750 for(i = 1; i < ehdr.e_shnum; i++) {
2751 s = sm_table[i].s;
2752 if (!s)
2753 continue;
2754 sh = &shdr[i];
2755 offset = sm_table[i].offset;
2756 switch(s->sh_type) {
2757 case SHT_RELX:
2758 /* take relocation offset information */
2759 offseti = sm_table[sh->sh_info].offset;
2760 for_each_elem(s, (offset / sizeof(*rel)), rel, ElfW_Rel) {
2761 int type;
2762 unsigned sym_index;
2763 /* convert symbol index */
2764 type = ELFW(R_TYPE)(rel->r_info);
2765 sym_index = ELFW(R_SYM)(rel->r_info);
2766 /* NOTE: only one symtab assumed */
2767 if (sym_index >= nb_syms)
2768 goto invalid_reloc;
2769 sym_index = old_to_new_syms[sym_index];
2770 /* ignore link_once in rel section. */
2771 if (!sym_index && !sm->link_once
2772 #ifdef TCC_TARGET_ARM
2773 && type != R_ARM_V4BX
2774 #endif
2776 invalid_reloc:
2777 tcc_error_noabort("Invalid relocation entry [%2d] '%s' @ %.8x",
2778 i, strsec + sh->sh_name, rel->r_offset);
2779 goto fail;
2781 rel->r_info = ELFW(R_INFO)(sym_index, type);
2782 /* offset the relocation offset */
2783 rel->r_offset += offseti;
2784 #ifdef TCC_TARGET_ARM
2785 /* Jumps and branches from a Thumb code to a PLT entry need
2786 special handling since PLT entries are ARM code.
2787 Unconditional bl instructions referencing PLT entries are
2788 handled by converting these instructions into blx
2789 instructions. Other case of instructions referencing a PLT
2790 entry require to add a Thumb stub before the PLT entry to
2791 switch to ARM mode. We set bit plt_thumb_stub of the
2792 attribute of a symbol to indicate such a case. */
2793 if (type == R_ARM_THM_JUMP24)
2794 alloc_sym_attr(s1, sym_index)->plt_thumb_stub = 1;
2795 #endif
2797 break;
2798 default:
2799 break;
2803 ret = 0;
2804 the_end:
2805 tcc_free(symtab);
2806 tcc_free(strtab);
2807 tcc_free(old_to_new_syms);
2808 tcc_free(sm_table);
2809 tcc_free(strsec);
2810 tcc_free(shdr);
2811 return ret;
2814 typedef struct ArchiveHeader {
2815 char ar_name[16]; /* name of this member */
2816 char ar_date[12]; /* file mtime */
2817 char ar_uid[6]; /* owner uid; printed as decimal */
2818 char ar_gid[6]; /* owner gid; printed as decimal */
2819 char ar_mode[8]; /* file mode, printed as octal */
2820 char ar_size[10]; /* file size, printed as decimal */
2821 char ar_fmag[2]; /* should contain ARFMAG */
2822 } ArchiveHeader;
2824 static int get_be32(const uint8_t *b)
2826 return b[3] | (b[2] << 8) | (b[1] << 16) | (b[0] << 24);
2829 /* load only the objects which resolve undefined symbols */
2830 static int tcc_load_alacarte(TCCState *s1, int fd, int size)
2832 int i, bound, nsyms, sym_index, off, ret;
2833 uint8_t *data;
2834 const char *ar_names, *p;
2835 const uint8_t *ar_index;
2836 ElfW(Sym) *sym;
2838 data = tcc_malloc(size);
2839 if (read(fd, data, size) != size)
2840 goto fail;
2841 nsyms = get_be32(data);
2842 ar_index = data + 4;
2843 ar_names = (char *) ar_index + nsyms * 4;
2845 do {
2846 bound = 0;
2847 for(p = ar_names, i = 0; i < nsyms; i++, p += strlen(p)+1) {
2848 sym_index = find_elf_sym(symtab_section, p);
2849 if(sym_index) {
2850 sym = &((ElfW(Sym) *)symtab_section->data)[sym_index];
2851 if(sym->st_shndx == SHN_UNDEF) {
2852 off = get_be32(ar_index + i * 4) + sizeof(ArchiveHeader);
2853 ++bound;
2854 lseek(fd, off, SEEK_SET);
2855 if(tcc_load_object_file(s1, fd, off) < 0) {
2856 fail:
2857 ret = -1;
2858 goto the_end;
2863 } while(bound);
2864 ret = 0;
2865 the_end:
2866 tcc_free(data);
2867 return ret;
2870 /* load a '.a' file */
2871 ST_FUNC int tcc_load_archive(TCCState *s1, int fd)
2873 ArchiveHeader hdr;
2874 char ar_size[11];
2875 char ar_name[17];
2876 char magic[8];
2877 int size, len, i;
2878 unsigned long file_offset;
2880 /* skip magic which was already checked */
2881 read(fd, magic, sizeof(magic));
2883 for(;;) {
2884 len = read(fd, &hdr, sizeof(hdr));
2885 if (len == 0)
2886 break;
2887 if (len != sizeof(hdr)) {
2888 tcc_error_noabort("invalid archive");
2889 return -1;
2891 memcpy(ar_size, hdr.ar_size, sizeof(hdr.ar_size));
2892 ar_size[sizeof(hdr.ar_size)] = '\0';
2893 size = strtol(ar_size, NULL, 0);
2894 memcpy(ar_name, hdr.ar_name, sizeof(hdr.ar_name));
2895 for(i = sizeof(hdr.ar_name) - 1; i >= 0; i--) {
2896 if (ar_name[i] != ' ')
2897 break;
2899 ar_name[i + 1] = '\0';
2900 file_offset = lseek(fd, 0, SEEK_CUR);
2901 /* align to even */
2902 size = (size + 1) & ~1;
2903 if (!strcmp(ar_name, "/")) {
2904 /* coff symbol table : we handle it */
2905 if(s1->alacarte_link)
2906 return tcc_load_alacarte(s1, fd, size);
2907 } else if (!strcmp(ar_name, "//") ||
2908 !strcmp(ar_name, "__.SYMDEF") ||
2909 !strcmp(ar_name, "__.SYMDEF/") ||
2910 !strcmp(ar_name, "ARFILENAMES/")) {
2911 /* skip symbol table or archive names */
2912 } else {
2913 if (tcc_load_object_file(s1, fd, file_offset) < 0)
2914 return -1;
2916 lseek(fd, file_offset + size, SEEK_SET);
2918 return 0;
2921 #ifndef TCC_TARGET_PE
2922 /* load a DLL and all referenced DLLs. 'level = 0' means that the DLL
2923 is referenced by the user (so it should be added as DT_NEEDED in
2924 the generated ELF file) */
2925 ST_FUNC int tcc_load_dll(TCCState *s1, int fd, const char *filename, int level)
2927 ElfW(Ehdr) ehdr;
2928 ElfW(Shdr) *shdr, *sh, *sh1;
2929 int i, j, nb_syms, nb_dts, sym_bind, ret;
2930 ElfW(Sym) *sym, *dynsym;
2931 ElfW(Dyn) *dt, *dynamic;
2932 unsigned char *dynstr;
2933 const char *name, *soname;
2934 DLLReference *dllref;
2936 read(fd, &ehdr, sizeof(ehdr));
2938 /* test CPU specific stuff */
2939 if (ehdr.e_ident[5] != ELFDATA2LSB ||
2940 ehdr.e_machine != EM_TCC_TARGET) {
2941 tcc_error_noabort("bad architecture");
2942 return -1;
2945 /* read sections */
2946 shdr = load_data(fd, ehdr.e_shoff, sizeof(ElfW(Shdr)) * ehdr.e_shnum);
2948 /* load dynamic section and dynamic symbols */
2949 nb_syms = 0;
2950 nb_dts = 0;
2951 dynamic = NULL;
2952 dynsym = NULL; /* avoid warning */
2953 dynstr = NULL; /* avoid warning */
2954 for(i = 0, sh = shdr; i < ehdr.e_shnum; i++, sh++) {
2955 switch(sh->sh_type) {
2956 case SHT_DYNAMIC:
2957 nb_dts = sh->sh_size / sizeof(ElfW(Dyn));
2958 dynamic = load_data(fd, sh->sh_offset, sh->sh_size);
2959 break;
2960 case SHT_DYNSYM:
2961 nb_syms = sh->sh_size / sizeof(ElfW(Sym));
2962 dynsym = load_data(fd, sh->sh_offset, sh->sh_size);
2963 sh1 = &shdr[sh->sh_link];
2964 dynstr = load_data(fd, sh1->sh_offset, sh1->sh_size);
2965 break;
2966 default:
2967 break;
2971 /* compute the real library name */
2972 soname = tcc_basename(filename);
2974 for(i = 0, dt = dynamic; i < nb_dts; i++, dt++) {
2975 if (dt->d_tag == DT_SONAME) {
2976 soname = (char *) dynstr + dt->d_un.d_val;
2980 /* if the dll is already loaded, do not load it */
2981 for(i = 0; i < s1->nb_loaded_dlls; i++) {
2982 dllref = s1->loaded_dlls[i];
2983 if (!strcmp(soname, dllref->name)) {
2984 /* but update level if needed */
2985 if (level < dllref->level)
2986 dllref->level = level;
2987 ret = 0;
2988 goto the_end;
2992 /* add the dll and its level */
2993 dllref = tcc_mallocz(sizeof(DLLReference) + strlen(soname));
2994 dllref->level = level;
2995 strcpy(dllref->name, soname);
2996 dynarray_add((void ***)&s1->loaded_dlls, &s1->nb_loaded_dlls, dllref);
2998 /* add dynamic symbols in dynsym_section */
2999 for(i = 1, sym = dynsym + 1; i < nb_syms; i++, sym++) {
3000 sym_bind = ELFW(ST_BIND)(sym->st_info);
3001 if (sym_bind == STB_LOCAL)
3002 continue;
3003 name = (char *) dynstr + sym->st_name;
3004 add_elf_sym(s1->dynsymtab_section, sym->st_value, sym->st_size,
3005 sym->st_info, sym->st_other, sym->st_shndx, name);
3008 /* load all referenced DLLs */
3009 for(i = 0, dt = dynamic; i < nb_dts; i++, dt++) {
3010 switch(dt->d_tag) {
3011 case DT_NEEDED:
3012 name = (char *) dynstr + dt->d_un.d_val;
3013 for(j = 0; j < s1->nb_loaded_dlls; j++) {
3014 dllref = s1->loaded_dlls[j];
3015 if (!strcmp(name, dllref->name))
3016 goto already_loaded;
3018 if (tcc_add_dll(s1, name, AFF_REFERENCED_DLL) < 0) {
3019 tcc_error_noabort("referenced dll '%s' not found", name);
3020 ret = -1;
3021 goto the_end;
3023 already_loaded:
3024 break;
3027 ret = 0;
3028 the_end:
3029 tcc_free(dynstr);
3030 tcc_free(dynsym);
3031 tcc_free(dynamic);
3032 tcc_free(shdr);
3033 return ret;
3036 #define LD_TOK_NAME 256
3037 #define LD_TOK_EOF (-1)
3039 /* return next ld script token */
3040 static int ld_next(TCCState *s1, char *name, int name_size)
3042 int c;
3043 char *q;
3045 redo:
3046 switch(ch) {
3047 case ' ':
3048 case '\t':
3049 case '\f':
3050 case '\v':
3051 case '\r':
3052 case '\n':
3053 inp();
3054 goto redo;
3055 case '/':
3056 minp();
3057 if (ch == '*') {
3058 file->buf_ptr = parse_comment(file->buf_ptr);
3059 ch = file->buf_ptr[0];
3060 goto redo;
3061 } else {
3062 q = name;
3063 *q++ = '/';
3064 goto parse_name;
3066 break;
3067 /* case 'a' ... 'z': */
3068 case 'a':
3069 case 'b':
3070 case 'c':
3071 case 'd':
3072 case 'e':
3073 case 'f':
3074 case 'g':
3075 case 'h':
3076 case 'i':
3077 case 'j':
3078 case 'k':
3079 case 'l':
3080 case 'm':
3081 case 'n':
3082 case 'o':
3083 case 'p':
3084 case 'q':
3085 case 'r':
3086 case 's':
3087 case 't':
3088 case 'u':
3089 case 'v':
3090 case 'w':
3091 case 'x':
3092 case 'y':
3093 case 'z':
3094 /* case 'A' ... 'z': */
3095 case 'A':
3096 case 'B':
3097 case 'C':
3098 case 'D':
3099 case 'E':
3100 case 'F':
3101 case 'G':
3102 case 'H':
3103 case 'I':
3104 case 'J':
3105 case 'K':
3106 case 'L':
3107 case 'M':
3108 case 'N':
3109 case 'O':
3110 case 'P':
3111 case 'Q':
3112 case 'R':
3113 case 'S':
3114 case 'T':
3115 case 'U':
3116 case 'V':
3117 case 'W':
3118 case 'X':
3119 case 'Y':
3120 case 'Z':
3121 case '_':
3122 case '\\':
3123 case '.':
3124 case '$':
3125 case '~':
3126 q = name;
3127 parse_name:
3128 for(;;) {
3129 if (!((ch >= 'a' && ch <= 'z') ||
3130 (ch >= 'A' && ch <= 'Z') ||
3131 (ch >= '0' && ch <= '9') ||
3132 strchr("/.-_+=$:\\,~", ch)))
3133 break;
3134 if ((q - name) < name_size - 1) {
3135 *q++ = ch;
3137 minp();
3139 *q = '\0';
3140 c = LD_TOK_NAME;
3141 break;
3142 case CH_EOF:
3143 c = LD_TOK_EOF;
3144 break;
3145 default:
3146 c = ch;
3147 inp();
3148 break;
3150 return c;
3153 static int ld_add_file(TCCState *s1, const char filename[])
3155 int ret;
3157 ret = tcc_add_file_internal(s1, filename, 0);
3158 if (ret)
3159 ret = tcc_add_dll(s1, filename, 0);
3160 return ret;
3163 static inline int new_undef_syms(void)
3165 int ret = 0;
3166 ret = new_undef_sym;
3167 new_undef_sym = 0;
3168 return ret;
3171 static int ld_add_file_list(TCCState *s1, const char *cmd, int as_needed)
3173 char filename[1024], libname[1024];
3174 int t, group, nblibs = 0, ret = 0;
3175 char **libs = NULL;
3177 group = !strcmp(cmd, "GROUP");
3178 if (!as_needed)
3179 new_undef_syms();
3180 t = ld_next(s1, filename, sizeof(filename));
3181 if (t != '(')
3182 expect("(");
3183 t = ld_next(s1, filename, sizeof(filename));
3184 for(;;) {
3185 libname[0] = '\0';
3186 if (t == LD_TOK_EOF) {
3187 tcc_error_noabort("unexpected end of file");
3188 ret = -1;
3189 goto lib_parse_error;
3190 } else if (t == ')') {
3191 break;
3192 } else if (t == '-') {
3193 t = ld_next(s1, filename, sizeof(filename));
3194 if ((t != LD_TOK_NAME) || (filename[0] != 'l')) {
3195 tcc_error_noabort("library name expected");
3196 ret = -1;
3197 goto lib_parse_error;
3199 pstrcpy(libname, sizeof libname, &filename[1]);
3200 if (s1->static_link) {
3201 snprintf(filename, sizeof filename, "lib%s.a", libname);
3202 } else {
3203 snprintf(filename, sizeof filename, "lib%s.so", libname);
3205 } else if (t != LD_TOK_NAME) {
3206 tcc_error_noabort("filename expected");
3207 ret = -1;
3208 goto lib_parse_error;
3210 if (!strcmp(filename, "AS_NEEDED")) {
3211 ret = ld_add_file_list(s1, cmd, 1);
3212 if (ret)
3213 goto lib_parse_error;
3214 } else {
3215 /* TODO: Implement AS_NEEDED support. Ignore it for now */
3216 if (!as_needed) {
3217 ret = ld_add_file(s1, filename);
3218 if (ret)
3219 goto lib_parse_error;
3220 if (group) {
3221 /* Add the filename *and* the libname to avoid future conversions */
3222 dynarray_add((void ***) &libs, &nblibs, tcc_strdup(filename));
3223 if (libname[0] != '\0')
3224 dynarray_add((void ***) &libs, &nblibs, tcc_strdup(libname));
3228 t = ld_next(s1, filename, sizeof(filename));
3229 if (t == ',') {
3230 t = ld_next(s1, filename, sizeof(filename));
3233 if (group && !as_needed) {
3234 while (new_undef_syms()) {
3235 int i;
3237 for (i = 0; i < nblibs; i ++)
3238 ld_add_file(s1, libs[i]);
3241 lib_parse_error:
3242 dynarray_reset(&libs, &nblibs);
3243 return ret;
3246 /* interpret a subset of GNU ldscripts to handle the dummy libc.so
3247 files */
3248 ST_FUNC int tcc_load_ldscript(TCCState *s1)
3250 char cmd[64];
3251 char filename[1024];
3252 int t, ret;
3254 ch = file->buf_ptr[0];
3255 ch = handle_eob();
3256 for(;;) {
3257 t = ld_next(s1, cmd, sizeof(cmd));
3258 if (t == LD_TOK_EOF)
3259 return 0;
3260 else if (t != LD_TOK_NAME)
3261 return -1;
3262 if (!strcmp(cmd, "INPUT") ||
3263 !strcmp(cmd, "GROUP")) {
3264 ret = ld_add_file_list(s1, cmd, 0);
3265 if (ret)
3266 return ret;
3267 } else if (!strcmp(cmd, "OUTPUT_FORMAT") ||
3268 !strcmp(cmd, "TARGET")) {
3269 /* ignore some commands */
3270 t = ld_next(s1, cmd, sizeof(cmd));
3271 if (t != '(')
3272 expect("(");
3273 for(;;) {
3274 t = ld_next(s1, filename, sizeof(filename));
3275 if (t == LD_TOK_EOF) {
3276 tcc_error_noabort("unexpected end of file");
3277 return -1;
3278 } else if (t == ')') {
3279 break;
3282 } else {
3283 return -1;
3286 return 0;
3288 #endif /* !TCC_TARGET_PE */