arm: Handle R_ARM_NONE relocs
[tinycc.git] / tccelf.c
blobb55dbc9034fe09445abcc8ae9da57684f6cf2728
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 if (s1->output_type != TCC_OUTPUT_MEMORY)
855 val = s1->plt->sh_addr + rel->r_addend;
856 /* fallthrough. */
858 plt32pc32:
860 long long diff;
861 diff = (long long)val - addr;
862 if (diff <= -2147483647 || diff > 2147483647) {
863 #ifdef TCC_HAS_RUNTIME_PLTGOT
864 /* XXX: naive support for over 32bit jump */
865 if (s1->output_type == TCC_OUTPUT_MEMORY) {
866 val = (add_jmp_table(s1, val - rel->r_addend) +
867 rel->r_addend);
868 diff = val - addr;
870 #endif
871 if (diff <= -2147483647 || diff > 2147483647) {
872 tcc_error("internal error: relocation failed");
875 *(int *)ptr += diff;
877 break;
878 case R_X86_64_GLOB_DAT:
879 case R_X86_64_JUMP_SLOT:
880 /* They don't need addend */
881 *(int *)ptr = val - rel->r_addend;
882 break;
883 case R_X86_64_GOTPCREL:
884 #ifdef TCC_HAS_RUNTIME_PLTGOT
885 if (s1->output_type == TCC_OUTPUT_MEMORY) {
886 val = add_got_table(s1, val - rel->r_addend) + rel->r_addend;
887 *(int *)ptr += val - addr;
888 break;
890 #endif
891 *(int *)ptr += (s1->got->sh_addr - addr +
892 s1->sym_attrs[sym_index].got_offset - 4);
893 break;
894 case R_X86_64_GOTTPOFF:
895 *(int *)ptr += val - s1->got->sh_addr;
896 break;
897 case R_X86_64_GOT32:
898 /* we load the got offset */
899 *(int *)ptr += s1->sym_attrs[sym_index].got_offset;
900 break;
901 #else
902 #error unsupported processor
903 #endif
906 /* if the relocation is allocated, we change its symbol table */
907 if (sr->sh_flags & SHF_ALLOC)
908 sr->link = s1->dynsym;
911 /* relocate relocation table in 'sr' */
912 static void relocate_rel(TCCState *s1, Section *sr)
914 Section *s;
915 ElfW_Rel *rel;
917 s = s1->sections[sr->sh_info];
918 for_each_elem(sr, 0, rel, ElfW_Rel)
919 rel->r_offset += s->sh_addr;
922 /* count the number of dynamic relocations so that we can reserve
923 their space */
924 static int prepare_dynamic_rel(TCCState *s1, Section *sr)
926 ElfW_Rel *rel;
927 int sym_index, esym_index, type, count;
929 count = 0;
930 for_each_elem(sr, 0, rel, ElfW_Rel) {
931 sym_index = ELFW(R_SYM)(rel->r_info);
932 type = ELFW(R_TYPE)(rel->r_info);
933 switch(type) {
934 #if defined(TCC_TARGET_I386)
935 case R_386_32:
936 #elif defined(TCC_TARGET_X86_64)
937 case R_X86_64_32:
938 case R_X86_64_32S:
939 case R_X86_64_64:
940 #endif
941 count++;
942 break;
943 #if defined(TCC_TARGET_I386)
944 case R_386_PC32:
945 #elif defined(TCC_TARGET_X86_64)
946 case R_X86_64_PC32:
947 #endif
948 esym_index = s1->symtab_to_dynsym[sym_index];
949 if (esym_index)
950 count++;
951 break;
952 default:
953 break;
956 if (count) {
957 /* allocate the section */
958 sr->sh_flags |= SHF_ALLOC;
959 sr->sh_size = count * sizeof(ElfW_Rel);
961 return count;
964 static struct sym_attr *alloc_sym_attr(TCCState *s1, int index)
966 int n;
967 struct sym_attr *tab;
969 if (index >= s1->nb_sym_attrs) {
970 /* find immediately bigger power of 2 and reallocate array */
971 n = 1;
972 while (index >= n)
973 n *= 2;
974 tab = tcc_realloc(s1->sym_attrs, n * sizeof(*s1->sym_attrs));
975 s1->sym_attrs = tab;
976 memset(s1->sym_attrs + s1->nb_sym_attrs, 0,
977 (n - s1->nb_sym_attrs) * sizeof(*s1->sym_attrs));
978 s1->nb_sym_attrs = n;
980 return &s1->sym_attrs[index];
983 /* XXX: suppress that */
984 static void put32(unsigned char *p, uint32_t val)
986 p[0] = val;
987 p[1] = val >> 8;
988 p[2] = val >> 16;
989 p[3] = val >> 24;
992 #if defined(TCC_TARGET_I386) || defined(TCC_TARGET_ARM) || \
993 defined(TCC_TARGET_X86_64)
994 static uint32_t get32(unsigned char *p)
996 return p[0] | (p[1] << 8) | (p[2] << 16) | (p[3] << 24);
998 #endif
1000 static void build_got(TCCState *s1)
1002 unsigned char *ptr;
1004 /* if no got, then create it */
1005 s1->got = new_section(s1, ".got", SHT_PROGBITS, SHF_ALLOC | SHF_WRITE);
1006 s1->got->sh_entsize = 4;
1007 add_elf_sym(symtab_section, 0, 4, ELFW(ST_INFO)(STB_GLOBAL, STT_OBJECT),
1008 0, s1->got->sh_num, "_GLOBAL_OFFSET_TABLE_");
1009 ptr = section_ptr_add(s1->got, 3 * PTR_SIZE);
1010 #if PTR_SIZE == 4
1011 /* keep space for _DYNAMIC pointer, if present */
1012 put32(ptr, 0);
1013 /* two dummy got entries */
1014 put32(ptr + 4, 0);
1015 put32(ptr + 8, 0);
1016 #else
1017 /* keep space for _DYNAMIC pointer, if present */
1018 put32(ptr, 0);
1019 put32(ptr + 4, 0);
1020 /* two dummy got entries */
1021 put32(ptr + 8, 0);
1022 put32(ptr + 12, 0);
1023 put32(ptr + 16, 0);
1024 put32(ptr + 20, 0);
1025 #endif
1028 /* put a got or plt entry corresponding to a symbol in symtab_section. 'size'
1029 and 'info' can be modifed if more precise info comes from the DLL.
1030 Returns offset of GOT or PLT slot. */
1031 static unsigned long put_got_entry(TCCState *s1,
1032 int reloc_type, unsigned long size, int info,
1033 int sym_index)
1035 int index, need_plt_entry;
1036 const char *name;
1037 ElfW(Sym) *sym;
1038 unsigned long offset;
1039 int *ptr;
1040 struct sym_attr *symattr;
1042 if (!s1->got)
1043 build_got(s1);
1045 need_plt_entry = s1->dynsym &&
1046 #ifdef TCC_TARGET_X86_64
1047 (reloc_type == R_X86_64_JUMP_SLOT);
1048 #elif defined(TCC_TARGET_I386)
1049 (reloc_type == R_386_JMP_SLOT);
1050 #elif defined(TCC_TARGET_ARM)
1051 (reloc_type == R_ARM_JUMP_SLOT);
1052 #else
1054 #endif
1056 /* If a got/plt entry already exists for that symbol, no need to add one */
1057 if (sym_index < s1->nb_sym_attrs) {
1058 if (need_plt_entry && s1->sym_attrs[sym_index].plt_offset)
1059 return s1->sym_attrs[sym_index].plt_offset;
1060 else if (!need_plt_entry && s1->sym_attrs[sym_index].got_offset)
1061 return s1->sym_attrs[sym_index].got_offset;
1064 symattr = alloc_sym_attr(s1, sym_index);
1066 /* Only store the GOT offset if it's not generated for the PLT entry. */
1067 if (!need_plt_entry)
1068 symattr->got_offset = s1->got->data_offset;
1070 if (s1->dynsym) {
1071 sym = &((ElfW(Sym) *)symtab_section->data)[sym_index];
1072 name = (char *) symtab_section->link->data + sym->st_name;
1073 offset = sym->st_value;
1074 #if defined(TCC_TARGET_I386) || defined(TCC_TARGET_X86_64)
1075 if (need_plt_entry) {
1076 Section *plt;
1077 uint8_t *p;
1078 int modrm;
1079 unsigned long relofs;
1081 #if defined(TCC_OUTPUT_DLL_WITH_PLT)
1082 modrm = 0x25;
1083 #else
1084 /* if we build a DLL, we add a %ebx offset */
1085 if (s1->output_type == TCC_OUTPUT_DLL)
1086 modrm = 0xa3;
1087 else
1088 modrm = 0x25;
1089 #endif
1091 /* add a PLT entry */
1092 plt = s1->plt;
1093 if (plt->data_offset == 0) {
1094 /* first plt entry */
1095 p = section_ptr_add(plt, 16);
1096 p[0] = 0xff; /* pushl got + PTR_SIZE */
1097 p[1] = modrm + 0x10;
1098 put32(p + 2, PTR_SIZE);
1099 p[6] = 0xff; /* jmp *(got + PTR_SIZE * 2) */
1100 p[7] = modrm;
1101 put32(p + 8, PTR_SIZE * 2);
1104 /* The PLT slot refers to the relocation entry it needs
1105 via offset. The reloc entry is created below, so its
1106 offset is the current data_offset. */
1107 relofs = s1->got->reloc ? s1->got->reloc->data_offset : 0;
1108 symattr->plt_offset = plt->data_offset;
1109 p = section_ptr_add(plt, 16);
1110 p[0] = 0xff; /* jmp *(got + x) */
1111 p[1] = modrm;
1112 put32(p + 2, s1->got->data_offset);
1113 p[6] = 0x68; /* push $xxx */
1114 #ifdef TCC_TARGET_X86_64
1115 /* On x86-64, the relocation is referred to by _index_. */
1116 put32(p + 7, relofs / sizeof (ElfW_Rel));
1117 #else
1118 put32(p + 7, relofs);
1119 #endif
1120 p[11] = 0xe9; /* jmp plt_start */
1121 put32(p + 12, -(plt->data_offset));
1123 /* If this was an UNDEF symbol set the offset in the
1124 dynsymtab to the PLT slot, so that PC32 relocs to it
1125 can be resolved. */
1126 if (sym->st_shndx == SHN_UNDEF)
1127 offset = plt->data_offset - 16;
1129 #elif defined(TCC_TARGET_ARM)
1130 if (need_plt_entry) {
1131 Section *plt;
1132 uint8_t *p;
1134 /* if we build a DLL, we add a %ebx offset */
1135 if (s1->output_type == TCC_OUTPUT_DLL)
1136 tcc_error("DLLs unimplemented!");
1138 /* add a PLT entry */
1139 plt = s1->plt;
1140 if (plt->data_offset == 0) {
1141 /* first plt entry */
1142 p = section_ptr_add(plt, 16);
1143 put32(p, 0xe52de004); /* push {lr} */
1144 put32(p+4, 0xe59fe010); /* ldr lr, [pc, #16] */
1145 put32(p+8, 0xe08fe00e); /* add lr, pc, lr */
1146 put32(p+12, 0xe5bef008); /* ldr pc, [lr, #8]! */
1149 symattr->plt_offset = plt->data_offset;
1150 if (symattr->plt_thumb_stub) {
1151 p = section_ptr_add(plt, 20);
1152 put32(p, 0x4778); /* bx pc */
1153 put32(p+2, 0x46c0); /* nop */
1154 p += 4;
1155 } else
1156 p = section_ptr_add(plt, 16);
1157 put32(p, 0xe59fc004); /* ldr ip, [pc, #4] ; GOT entry offset */
1158 put32(p+4, 0xe08fc00c); /* add ip, pc, ip ; addr of GOT entry */
1159 put32(p+8, 0xe59cf000); /* ldr pc, [ip] ; jump to GOT entry */
1160 put32(p+12, s1->got->data_offset); /* GOT entry off once patched */
1162 /* the symbol is modified so that it will be relocated to
1163 the PLT */
1164 if (s1->output_type == TCC_OUTPUT_EXE)
1165 offset = plt->data_offset - 16;
1167 #elif defined(TCC_TARGET_C67)
1168 tcc_error("C67 got not implemented");
1169 #else
1170 #error unsupported CPU
1171 #endif
1172 /* XXX This might generate multiple syms for name. */
1173 index = put_elf_sym(s1->dynsym, offset,
1174 size, info, 0, sym->st_shndx, name);
1175 /* Create the relocation (it's against the GOT for PLT
1176 and GOT relocs). */
1177 put_elf_reloc(s1->dynsym, s1->got,
1178 s1->got->data_offset,
1179 reloc_type, index);
1180 } else {
1181 /* Without .dynsym (i.e. static link or memory output) we
1182 still need relocs against the generated got, so as to fill
1183 the entries with the symbol values (determined later). */
1184 put_elf_reloc(symtab_section, s1->got,
1185 s1->got->data_offset,
1186 reloc_type, sym_index);
1188 /* And now create the GOT slot itself. */
1189 ptr = section_ptr_add(s1->got, PTR_SIZE);
1190 *ptr = 0;
1191 if (need_plt_entry)
1192 return symattr->plt_offset;
1193 else
1194 return symattr->got_offset;
1197 /* build GOT and PLT entries */
1198 ST_FUNC void build_got_entries(TCCState *s1)
1200 Section *s;
1201 ElfW_Rel *rel;
1202 ElfW(Sym) *sym;
1203 int i, type, reloc_type, sym_index;
1205 for(i = 1; i < s1->nb_sections; i++) {
1206 s = s1->sections[i];
1207 if (s->sh_type != SHT_RELX)
1208 continue;
1209 /* no need to handle got relocations */
1210 if (s->link != symtab_section)
1211 continue;
1212 for_each_elem(s, 0, rel, ElfW_Rel) {
1213 type = ELFW(R_TYPE)(rel->r_info);
1214 switch(type) {
1215 #if defined(TCC_TARGET_I386)
1216 case R_386_GOT32:
1217 case R_386_GOTOFF:
1218 case R_386_GOTPC:
1219 case R_386_PLT32:
1220 if (!s1->got)
1221 build_got(s1);
1222 if (type == R_386_GOT32 || type == R_386_PLT32) {
1223 sym_index = ELFW(R_SYM)(rel->r_info);
1224 sym = &((ElfW(Sym) *)symtab_section->data)[sym_index];
1225 /* look at the symbol got offset. If none, then add one */
1226 if (type == R_386_GOT32)
1227 reloc_type = R_386_GLOB_DAT;
1228 else
1229 reloc_type = R_386_JMP_SLOT;
1230 put_got_entry(s1, reloc_type, sym->st_size, sym->st_info,
1231 sym_index);
1233 break;
1234 #elif defined(TCC_TARGET_ARM)
1235 case R_ARM_GOT32:
1236 case R_ARM_GOTOFF:
1237 case R_ARM_GOTPC:
1238 case R_ARM_PLT32:
1239 if (!s1->got)
1240 build_got(s1);
1241 if (type == R_ARM_GOT32 || type == R_ARM_PLT32) {
1242 sym_index = ELFW(R_SYM)(rel->r_info);
1243 sym = &((ElfW(Sym) *)symtab_section->data)[sym_index];
1244 /* look at the symbol got offset. If none, then add one */
1245 if (type == R_ARM_GOT32)
1246 reloc_type = R_ARM_GLOB_DAT;
1247 else
1248 reloc_type = R_ARM_JUMP_SLOT;
1249 put_got_entry(s1, reloc_type, sym->st_size, sym->st_info,
1250 sym_index);
1252 break;
1253 case R_ARM_THM_JUMP24:
1254 sym_index = ELFW(R_SYM)(rel->r_info);
1255 sym = &((ElfW(Sym) *)symtab_section->data)[sym_index];
1256 /* We are relocating a jump from thumb code to arm code */
1257 if (sym->st_shndx != SHN_UNDEF && !(sym->st_value & 1)) {
1258 int index;
1259 uint8_t *p;
1260 char *name, buf[1024];
1261 Section *text_section;
1263 name = (char *) symtab_section->link->data + sym->st_name;
1264 text_section = s1->sections[sym->st_shndx];
1265 /* Modify reloc to target a thumb stub to switch to ARM */
1266 snprintf(buf, sizeof(buf), "%s_from_thumb", name);
1267 index = put_elf_sym(symtab_section,
1268 text_section->data_offset + 1,
1269 sym->st_size, sym->st_info, 0,
1270 sym->st_shndx, buf);
1271 rel->r_info = ELFW(R_INFO)(index, type);
1272 /* Create a thumb stub fonction to switch to ARM mode */
1273 put_elf_reloc(symtab_section, text_section,
1274 text_section->data_offset + 4, R_ARM_JUMP24,
1275 sym_index);
1276 p = section_ptr_add(text_section, 8);
1277 put32(p, 0x4778); /* bx pc */
1278 put32(p+2, 0x46c0); /* nop */
1279 put32(p+4, 0xeafffffe); /* b $sym */
1281 #elif defined(TCC_TARGET_C67)
1282 case R_C60_GOT32:
1283 case R_C60_GOTOFF:
1284 case R_C60_GOTPC:
1285 case R_C60_PLT32:
1286 if (!s1->got)
1287 build_got(s1);
1288 if (type == R_C60_GOT32 || type == R_C60_PLT32) {
1289 sym_index = ELFW(R_SYM)(rel->r_info);
1290 sym = &((ElfW(Sym) *)symtab_section->data)[sym_index];
1291 /* look at the symbol got offset. If none, then add one */
1292 if (type == R_C60_GOT32)
1293 reloc_type = R_C60_GLOB_DAT;
1294 else
1295 reloc_type = R_C60_JMP_SLOT;
1296 put_got_entry(s1, reloc_type, sym->st_size, sym->st_info,
1297 sym_index);
1299 break;
1300 #elif defined(TCC_TARGET_X86_64)
1301 case R_X86_64_GOT32:
1302 case R_X86_64_GOTTPOFF:
1303 case R_X86_64_GOTPCREL:
1304 case R_X86_64_PLT32:
1305 if (!s1->got)
1306 build_got(s1);
1307 if (type == R_X86_64_GOT32 || type == R_X86_64_GOTPCREL ||
1308 type == R_X86_64_PLT32) {
1309 unsigned long ofs;
1310 sym_index = ELFW(R_SYM)(rel->r_info);
1311 sym = &((ElfW(Sym) *)symtab_section->data)[sym_index];
1312 /* look at the symbol got offset. If none, then add one */
1313 if (type == R_X86_64_GOT32 || type == R_X86_64_GOTPCREL)
1314 reloc_type = R_X86_64_GLOB_DAT;
1315 else
1316 reloc_type = R_X86_64_JUMP_SLOT;
1317 ofs = put_got_entry(s1, reloc_type, sym->st_size,
1318 sym->st_info, sym_index);
1319 if (type == R_X86_64_PLT32
1320 && s1->output_type != TCC_OUTPUT_MEMORY)
1321 /* We store the place of the generated PLT slot
1322 in our addend. */
1323 rel->r_addend += ofs;
1325 break;
1326 #else
1327 #error unsupported CPU
1328 #endif
1329 default:
1330 break;
1336 ST_FUNC Section *new_symtab(TCCState *s1,
1337 const char *symtab_name, int sh_type, int sh_flags,
1338 const char *strtab_name,
1339 const char *hash_name, int hash_sh_flags)
1341 Section *symtab, *strtab, *hash;
1342 int *ptr, nb_buckets;
1344 symtab = new_section(s1, symtab_name, sh_type, sh_flags);
1345 symtab->sh_entsize = sizeof(ElfW(Sym));
1346 strtab = new_section(s1, strtab_name, SHT_STRTAB, sh_flags);
1347 put_elf_str(strtab, "");
1348 symtab->link = strtab;
1349 put_elf_sym(symtab, 0, 0, 0, 0, 0, NULL);
1351 nb_buckets = 1;
1353 hash = new_section(s1, hash_name, SHT_HASH, hash_sh_flags);
1354 hash->sh_entsize = sizeof(int);
1355 symtab->hash = hash;
1356 hash->link = symtab;
1358 ptr = section_ptr_add(hash, (2 + nb_buckets + 1) * sizeof(int));
1359 ptr[0] = nb_buckets;
1360 ptr[1] = 1;
1361 memset(ptr + 2, 0, (nb_buckets + 1) * sizeof(int));
1362 return symtab;
1365 /* put dynamic tag */
1366 static void put_dt(Section *dynamic, int dt, addr_t val)
1368 ElfW(Dyn) *dyn;
1369 dyn = section_ptr_add(dynamic, sizeof(ElfW(Dyn)));
1370 dyn->d_tag = dt;
1371 dyn->d_un.d_val = val;
1374 static void add_init_array_defines(TCCState *s1, const char *section_name)
1376 Section *s;
1377 long end_offset;
1378 char sym_start[1024];
1379 char sym_end[1024];
1381 snprintf(sym_start, sizeof(sym_start), "__%s_start", section_name + 1);
1382 snprintf(sym_end, sizeof(sym_end), "__%s_end", section_name + 1);
1384 s = find_section(s1, section_name);
1385 if (!s) {
1386 end_offset = 0;
1387 s = data_section;
1388 } else {
1389 end_offset = s->data_offset;
1392 add_elf_sym(symtab_section,
1393 0, 0,
1394 ELFW(ST_INFO)(STB_GLOBAL, STT_NOTYPE), 0,
1395 s->sh_num, sym_start);
1396 add_elf_sym(symtab_section,
1397 end_offset, 0,
1398 ELFW(ST_INFO)(STB_GLOBAL, STT_NOTYPE), 0,
1399 s->sh_num, sym_end);
1402 ST_FUNC void tcc_add_bcheck(TCCState *s1)
1404 #ifdef CONFIG_TCC_BCHECK
1405 unsigned long *ptr;
1406 Section *init_section;
1407 unsigned char *pinit;
1408 int sym_index;
1410 if (0 == s1->do_bounds_check)
1411 return;
1413 /* XXX: add an object file to do that */
1414 ptr = section_ptr_add(bounds_section, sizeof(unsigned long));
1415 *ptr = 0;
1416 add_elf_sym(symtab_section, 0, 0,
1417 ELFW(ST_INFO)(STB_GLOBAL, STT_NOTYPE), 0,
1418 bounds_section->sh_num, "__bounds_start");
1419 #ifdef TCC_TARGET_I386
1420 if (s1->output_type != TCC_OUTPUT_MEMORY) {
1421 /* add 'call __bound_init()' in .init section */
1422 init_section = find_section(s1, ".init");
1423 pinit = section_ptr_add(init_section, 5);
1424 pinit[0] = 0xe8;
1425 put32(pinit + 1, -4);
1426 sym_index = find_elf_sym(symtab_section, "__bound_init");
1427 put_elf_reloc(symtab_section, init_section,
1428 init_section->data_offset - 4, R_386_PC32, sym_index);
1430 #endif
1431 #endif
1434 static inline int tcc_add_support(TCCState *s1, const char *filename)
1436 char buf[1024];
1437 snprintf(buf, sizeof(buf), "%s/%s", s1->tcc_lib_path, filename);
1438 return tcc_add_file(s1, buf);
1441 /* add tcc runtime libraries */
1442 ST_FUNC void tcc_add_runtime(TCCState *s1)
1444 /* add libc */
1445 if (!s1->nostdlib) {
1446 tcc_add_library(s1, "c");
1447 #ifdef CONFIG_USE_LIBGCC
1448 if (!s1->static_link) {
1449 tcc_add_file(s1, TCC_LIBGCC);
1450 tcc_add_support(s1, "libtcc1.a");
1451 } else
1452 tcc_add_support(s1, "libtcc1.a");
1453 #else
1454 tcc_add_support(s1, "libtcc1.a");
1455 #endif
1458 /* tcc_add_bcheck tries to relocate a call to __bound_init in _init so
1459 libtcc1.a must be loaded before for __bound_init to be defined and
1460 crtn.o must be loaded after to not finalize _init too early. */
1461 tcc_add_bcheck(s1);
1463 if (!s1->nostdlib) {
1464 /* add crt end if not memory output */
1465 if (s1->output_type != TCC_OUTPUT_MEMORY)
1466 tcc_add_crt(s1, "crtn.o");
1470 /* add various standard linker symbols (must be done after the
1471 sections are filled (for example after allocating common
1472 symbols)) */
1473 ST_FUNC void tcc_add_linker_symbols(TCCState *s1)
1475 char buf[1024];
1476 int i;
1477 Section *s;
1479 add_elf_sym(symtab_section,
1480 text_section->data_offset, 0,
1481 ELFW(ST_INFO)(STB_GLOBAL, STT_NOTYPE), 0,
1482 text_section->sh_num, "_etext");
1483 add_elf_sym(symtab_section,
1484 data_section->data_offset, 0,
1485 ELFW(ST_INFO)(STB_GLOBAL, STT_NOTYPE), 0,
1486 data_section->sh_num, "_edata");
1487 add_elf_sym(symtab_section,
1488 bss_section->data_offset, 0,
1489 ELFW(ST_INFO)(STB_GLOBAL, STT_NOTYPE), 0,
1490 bss_section->sh_num, "_end");
1491 /* horrible new standard ldscript defines */
1492 add_init_array_defines(s1, ".preinit_array");
1493 add_init_array_defines(s1, ".init_array");
1494 add_init_array_defines(s1, ".fini_array");
1496 /* add start and stop symbols for sections whose name can be
1497 expressed in C */
1498 for(i = 1; i < s1->nb_sections; i++) {
1499 s = s1->sections[i];
1500 if (s->sh_type == SHT_PROGBITS &&
1501 (s->sh_flags & SHF_ALLOC)) {
1502 const char *p;
1503 int ch;
1505 /* check if section name can be expressed in C */
1506 p = s->name;
1507 for(;;) {
1508 ch = *p;
1509 if (!ch)
1510 break;
1511 if (!isid(ch) && !isnum(ch))
1512 goto next_sec;
1513 p++;
1515 snprintf(buf, sizeof(buf), "__start_%s", s->name);
1516 add_elf_sym(symtab_section,
1517 0, 0,
1518 ELFW(ST_INFO)(STB_GLOBAL, STT_NOTYPE), 0,
1519 s->sh_num, buf);
1520 snprintf(buf, sizeof(buf), "__stop_%s", s->name);
1521 add_elf_sym(symtab_section,
1522 s->data_offset, 0,
1523 ELFW(ST_INFO)(STB_GLOBAL, STT_NOTYPE), 0,
1524 s->sh_num, buf);
1526 next_sec: ;
1530 static void tcc_output_binary(TCCState *s1, FILE *f,
1531 const int *sec_order)
1533 Section *s;
1534 int i, offset, size;
1536 offset = 0;
1537 for(i=1;i<s1->nb_sections;i++) {
1538 s = s1->sections[sec_order[i]];
1539 if (s->sh_type != SHT_NOBITS &&
1540 (s->sh_flags & SHF_ALLOC)) {
1541 while (offset < s->sh_offset) {
1542 fputc(0, f);
1543 offset++;
1545 size = s->sh_size;
1546 fwrite(s->data, 1, size, f);
1547 offset += size;
1552 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
1553 #define HAVE_PHDR 1
1554 #define EXTRA_RELITEMS 14
1556 /* move the relocation value from .dynsym to .got */
1557 void patch_dynsym_undef(TCCState *s1, Section *s)
1559 uint32_t *gotd = (void *)s1->got->data;
1560 ElfW(Sym) *sym;
1562 gotd += 3; /* dummy entries in .got */
1563 /* relocate symbols in .dynsym */
1564 for_each_elem(s, 1, sym, ElfW(Sym)) {
1565 if (sym->st_shndx == SHN_UNDEF) {
1566 *gotd++ = sym->st_value + 6; /* XXX 6 is magic ? */
1567 sym->st_value = 0;
1571 #else
1572 #define HAVE_PHDR 1
1573 #define EXTRA_RELITEMS 9
1575 /* zero plt offsets of weak symbols in .dynsym */
1576 void patch_dynsym_undef(TCCState *s1, Section *s)
1578 ElfW(Sym) *sym;
1580 for_each_elem(s, 1, sym, ElfW(Sym))
1581 if (sym->st_shndx == SHN_UNDEF && ELFW(ST_BIND)(sym->st_info) == STB_WEAK)
1582 sym->st_value = 0;
1584 #endif
1586 ST_FUNC void fill_got_entry(TCCState *s1, ElfW_Rel *rel)
1588 int sym_index = ELFW(R_SYM) (rel->r_info);
1589 ElfW(Sym) *sym = &((ElfW(Sym) *) symtab_section->data)[sym_index];
1590 unsigned long offset;
1592 if (sym_index >= s1->nb_sym_attrs)
1593 return;
1594 offset = s1->sym_attrs[sym_index].got_offset;
1595 section_reserve(s1->got, offset + PTR_SIZE);
1596 #ifdef TCC_TARGET_X86_64
1597 /* only works for x86-64 */
1598 put32(s1->got->data + offset + 4, sym->st_value >> 32);
1599 #endif
1600 put32(s1->got->data + offset, sym->st_value & 0xffffffff);
1603 /* Perform relocation to GOT or PLT entries */
1604 ST_FUNC void fill_got(TCCState *s1)
1606 Section *s;
1607 ElfW_Rel *rel;
1608 int i;
1610 for(i = 1; i < s1->nb_sections; i++) {
1611 s = s1->sections[i];
1612 if (s->sh_type != SHT_RELX)
1613 continue;
1614 /* no need to handle got relocations */
1615 if (s->link != symtab_section)
1616 continue;
1617 for_each_elem(s, 0, rel, ElfW_Rel) {
1618 switch (ELFW(R_TYPE) (rel->r_info)) {
1619 case R_X86_64_GOT32:
1620 case R_X86_64_GOTPCREL:
1621 case R_X86_64_PLT32:
1622 fill_got_entry(s1, rel);
1623 break;
1629 /* Bind symbols of executable: resolve undefined symbols from exported symbols
1630 in shared libraries and export non local defined symbols to shared libraries
1631 if -rdynamic switch was given on command line */
1632 static void bind_exe_dynsyms(TCCState *s1)
1634 const char *name;
1635 int sym_index, index;
1636 ElfW(Sym) *sym, *esym;
1637 int type;
1639 /* Resolve undefined symbols from dynamic symbols. When there is a match:
1640 - if STT_FUNC or STT_GNU_IFUNC symbol -> add it in PLT
1641 - if STT_OBJECT symbol -> add it in .bss section with suitable reloc */
1642 for_each_elem(symtab_section, 1, sym, ElfW(Sym)) {
1643 if (sym->st_shndx == SHN_UNDEF) {
1644 name = (char *) symtab_section->link->data + sym->st_name;
1645 sym_index = find_elf_sym(s1->dynsymtab_section, name);
1646 if (sym_index) {
1647 esym = &((ElfW(Sym) *)s1->dynsymtab_section->data)[sym_index];
1648 type = ELFW(ST_TYPE)(esym->st_info);
1649 if ((type == STT_FUNC) || (type == STT_GNU_IFUNC)) {
1650 /* Indirect functions shall have STT_FUNC type in executable
1651 * dynsym section. Indeed, a dlsym call following a lazy
1652 * resolution would pick the symbol value from the
1653 * executable dynsym entry which would contain the address
1654 * of the function wanted by the caller of dlsym instead of
1655 * the address of the function that would return that
1656 * address */
1657 put_got_entry(s1, R_JMP_SLOT, esym->st_size,
1658 ELFW(ST_INFO)(STB_GLOBAL,STT_FUNC),
1659 sym - (ElfW(Sym) *)symtab_section->data);
1660 } else if (type == STT_OBJECT) {
1661 unsigned long offset;
1662 ElfW(Sym) *dynsym;
1663 offset = bss_section->data_offset;
1664 /* XXX: which alignment ? */
1665 offset = (offset + 16 - 1) & -16;
1666 index = put_elf_sym(s1->dynsym, offset, esym->st_size,
1667 esym->st_info, 0, bss_section->sh_num,
1668 name);
1669 /* Ensure R_COPY works for weak symbol aliases */
1670 if (ELFW(ST_BIND)(esym->st_info) == STB_WEAK) {
1671 for_each_elem(s1->dynsymtab_section, 1, dynsym, ElfW(Sym)) {
1672 if ((dynsym->st_value == esym->st_value)
1673 && (ELFW(ST_BIND)(dynsym->st_info) == STB_GLOBAL)) {
1674 char *dynname = (char *) s1->dynsymtab_section->link->data
1675 + dynsym->st_name;
1676 put_elf_sym(s1->dynsym, offset, dynsym->st_size,
1677 dynsym->st_info, 0,
1678 bss_section->sh_num, dynname);
1679 break;
1683 put_elf_reloc(s1->dynsym, bss_section,
1684 offset, R_COPY, index);
1685 offset += esym->st_size;
1686 bss_section->data_offset = offset;
1688 } else {
1689 /* STB_WEAK undefined symbols are accepted */
1690 /* XXX: _fp_hw seems to be part of the ABI, so we ignore it */
1691 if (ELFW(ST_BIND)(sym->st_info) == STB_WEAK ||
1692 !strcmp(name, "_fp_hw")) {
1693 } else {
1694 tcc_error_noabort("undefined symbol '%s'", name);
1697 } else if (s1->rdynamic && ELFW(ST_BIND)(sym->st_info) != STB_LOCAL) {
1698 /* if -rdynamic option, then export all non local symbols */
1699 name = (char *) symtab_section->link->data + sym->st_name;
1700 put_elf_sym(s1->dynsym, sym->st_value, sym->st_size, sym->st_info,
1701 0, sym->st_shndx, name);
1706 /* Bind symbols of libraries: export non local symbols of executable that
1707 resolve undefined symbols of shared libraries */
1708 static void bind_libs_dynsyms(TCCState *s1)
1710 const char *name;
1711 int sym_index;
1712 ElfW(Sym) *sym, *esym;
1714 /* now look at unresolved dynamic symbols and export
1715 corresponding symbol */
1716 for_each_elem(s1->dynsymtab_section, 1, esym, ElfW(Sym)) {
1717 if (esym->st_shndx == SHN_UNDEF) {
1718 name = (char *) s1->dynsymtab_section->link->data + esym->st_name;
1719 sym_index = find_elf_sym(symtab_section, name);
1720 if (sym_index) {
1721 /* XXX: avoid adding a symbol if already present because of
1722 -rdynamic ? */
1723 sym = &((ElfW(Sym) *)symtab_section->data)[sym_index];
1724 put_elf_sym(s1->dynsym, sym->st_value, sym->st_size,
1725 sym->st_info, 0, sym->st_shndx, name);
1726 } else {
1727 /* weak symbols can stay undefined */
1728 if (ELFW(ST_BIND)(esym->st_info) != STB_WEAK)
1729 tcc_warning("undefined dynamic symbol '%s'", name);
1735 /* Export all non local symbols (for shared libraries) */
1736 static void export_global_syms(TCCState *s1)
1738 int nb_syms, dynindex, index;
1739 const char *name;
1740 ElfW(Sym) *sym;
1742 nb_syms = symtab_section->data_offset / sizeof(ElfW(Sym));
1743 s1->symtab_to_dynsym = tcc_mallocz(sizeof(int) * nb_syms);
1744 for_each_elem(symtab_section, 1, sym, ElfW(Sym)) {
1745 if (ELFW(ST_BIND)(sym->st_info) != STB_LOCAL) {
1746 name = (char *) symtab_section->link->data + sym->st_name;
1747 dynindex = put_elf_sym(s1->dynsym, sym->st_value, sym->st_size,
1748 sym->st_info, 0, sym->st_shndx, name);
1749 index = sym - (ElfW(Sym) *) symtab_section->data;
1750 s1->symtab_to_dynsym[index] = dynindex;
1755 /* relocate the PLT: compute addresses and offsets in the PLT now that final
1756 address for PLT and GOT are known (see fill_program_header) */
1757 static void relocate_plt(TCCState *s1)
1759 uint8_t *p, *p_end;
1761 p = s1->plt->data;
1762 p_end = p + s1->plt->data_offset;
1763 if (p < p_end) {
1764 #if defined(TCC_TARGET_I386)
1765 put32(p + 2, get32(p + 2) + s1->got->sh_addr);
1766 put32(p + 8, get32(p + 8) + s1->got->sh_addr);
1767 p += 16;
1768 while (p < p_end) {
1769 put32(p + 2, get32(p + 2) + s1->got->sh_addr);
1770 p += 16;
1772 #elif defined(TCC_TARGET_X86_64)
1773 int x = s1->got->sh_addr - s1->plt->sh_addr - 6;
1774 put32(p + 2, get32(p + 2) + x);
1775 put32(p + 8, get32(p + 8) + x - 6);
1776 p += 16;
1777 while (p < p_end) {
1778 put32(p + 2, get32(p + 2) + x + s1->plt->data - p);
1779 p += 16;
1781 #elif defined(TCC_TARGET_ARM)
1782 int x;
1783 x=s1->got->sh_addr - s1->plt->sh_addr - 12;
1784 p += 16;
1785 while (p < p_end) {
1786 if (get32(p) == 0x46c04778) /* PLT Thumb stub present */
1787 p += 4;
1788 put32(p + 12, x + get32(p + 12) + s1->plt->data - p);
1789 p += 16;
1791 #elif defined(TCC_TARGET_C67)
1792 /* XXX: TODO */
1793 #else
1794 #error unsupported CPU
1795 #endif
1799 /* Allocate strings for section names and decide if an unallocated section
1800 should be output.
1802 NOTE: the strsec section comes last, so its size is also correct ! */
1803 static void alloc_sec_names(TCCState *s1, int file_type, Section *strsec)
1805 int i;
1806 Section *s;
1808 /* Allocate strings for section names */
1809 for(i = 1; i < s1->nb_sections; i++) {
1810 s = s1->sections[i];
1811 s->sh_name = put_elf_str(strsec, s->name);
1812 /* when generating a DLL, we include relocations but we may
1813 patch them */
1814 if (file_type == TCC_OUTPUT_DLL &&
1815 s->sh_type == SHT_RELX &&
1816 !(s->sh_flags & SHF_ALLOC)) {
1817 /* gr: avoid bogus relocs for empty (debug) sections */
1818 if (s1->sections[s->sh_info]->sh_flags & SHF_ALLOC)
1819 prepare_dynamic_rel(s1, s);
1820 else if (s1->do_debug)
1821 s->sh_size = s->data_offset;
1822 } else if (s1->do_debug ||
1823 file_type == TCC_OUTPUT_OBJ ||
1824 (s->sh_flags & SHF_ALLOC) ||
1825 i == (s1->nb_sections - 1)) {
1826 /* we output all sections if debug or object file */
1827 s->sh_size = s->data_offset;
1832 /* Info to be copied in dynamic section */
1833 struct dyn_inf {
1834 Section *dynamic;
1835 Section *dynstr;
1836 unsigned long dyn_rel_off;
1837 addr_t rel_addr;
1838 addr_t rel_size;
1839 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
1840 addr_t bss_addr;
1841 addr_t bss_size;
1842 #endif
1845 /* Assign sections to segments and decide how are sections laid out when loaded
1846 in memory. This function also fills corresponding program headers. */
1847 static int layout_sections(TCCState *s1, ElfW(Phdr) *phdr, int phnum,
1848 Section *interp, struct dyn_inf *dyninf,
1849 int *sec_order)
1851 int i, j, k, file_type, sh_order_index, file_offset;
1852 long long tmp;
1853 addr_t addr;
1854 ElfW(Phdr) *ph;
1855 Section *s;
1857 file_type = s1->output_type;
1858 sh_order_index = 1;
1859 if (s1->output_format == TCC_OUTPUT_FORMAT_ELF)
1860 file_offset = sizeof(ElfW(Ehdr)) + phnum * sizeof(ElfW(Phdr));
1861 else
1862 file_offset = 0;
1864 if (phnum > 0) {
1865 if (s1->has_text_addr) {
1866 int a_offset, p_offset;
1867 addr = s1->text_addr;
1868 /* we ensure that (addr % ELF_PAGE_SIZE) == file_offset %
1869 ELF_PAGE_SIZE */
1870 a_offset = (int) (addr & (s1->section_align - 1));
1871 p_offset = file_offset & (s1->section_align - 1);
1872 if (a_offset < p_offset)
1873 a_offset += s1->section_align;
1874 file_offset += (a_offset - p_offset);
1875 } else {
1876 if (file_type == TCC_OUTPUT_DLL)
1877 addr = 0;
1878 else
1879 addr = ELF_START_ADDR;
1880 /* compute address after headers */
1881 addr += (file_offset & (s1->section_align - 1));
1884 ph = &phdr[0];
1885 /* Leave one program headers for the program interpreter and one for
1886 the program header table itself if needed. These are done later as
1887 they require section layout to be done first. */
1888 if (interp)
1889 ph += 1 + HAVE_PHDR;
1891 /* dynamic relocation table information, for .dynamic section */
1892 dyninf->rel_addr = dyninf->rel_size = 0;
1893 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
1894 dyninf->bss_addr = dyninf->bss_size = 0;
1895 #endif
1897 for(j = 0; j < 2; j++) {
1898 ph->p_type = PT_LOAD;
1899 if (j == 0)
1900 ph->p_flags = PF_R | PF_X;
1901 else
1902 ph->p_flags = PF_R | PF_W;
1903 ph->p_align = s1->section_align;
1905 /* Decide the layout of sections loaded in memory. This must
1906 be done before program headers are filled since they contain
1907 info about the layout. We do the following ordering: interp,
1908 symbol tables, relocations, progbits, nobits */
1909 /* XXX: do faster and simpler sorting */
1910 for(k = 0; k < 5; k++) {
1911 for(i = 1; i < s1->nb_sections; i++) {
1912 s = s1->sections[i];
1913 /* compute if section should be included */
1914 if (j == 0) {
1915 if ((s->sh_flags & (SHF_ALLOC | SHF_WRITE)) !=
1916 SHF_ALLOC)
1917 continue;
1918 } else {
1919 if ((s->sh_flags & (SHF_ALLOC | SHF_WRITE)) !=
1920 (SHF_ALLOC | SHF_WRITE))
1921 continue;
1923 if (s == interp) {
1924 if (k != 0)
1925 continue;
1926 } else if (s->sh_type == SHT_DYNSYM ||
1927 s->sh_type == SHT_STRTAB ||
1928 s->sh_type == SHT_HASH) {
1929 if (k != 1)
1930 continue;
1931 } else if (s->sh_type == SHT_RELX) {
1932 if (k != 2)
1933 continue;
1934 } else if (s->sh_type == SHT_NOBITS) {
1935 if (k != 4)
1936 continue;
1937 } else {
1938 if (k != 3)
1939 continue;
1941 sec_order[sh_order_index++] = i;
1943 /* section matches: we align it and add its size */
1944 tmp = addr;
1945 addr = (addr + s->sh_addralign - 1) &
1946 ~(s->sh_addralign - 1);
1947 file_offset += (int) ( addr - tmp );
1948 s->sh_offset = file_offset;
1949 s->sh_addr = addr;
1951 /* update program header infos */
1952 if (ph->p_offset == 0) {
1953 ph->p_offset = file_offset;
1954 ph->p_vaddr = addr;
1955 ph->p_paddr = ph->p_vaddr;
1957 /* update dynamic relocation infos */
1958 if (s->sh_type == SHT_RELX) {
1959 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
1960 if (!strcmp(strsec->data + s->sh_name, ".rel.got")) {
1961 dyninf->rel_addr = addr;
1962 dyninf->rel_size += s->sh_size; /* XXX only first rel. */
1964 if (!strcmp(strsec->data + s->sh_name, ".rel.bss")) {
1965 dyninf->bss_addr = addr;
1966 dyninf->bss_size = s->sh_size; /* XXX only first rel. */
1968 #else
1969 if (dyninf->rel_size == 0)
1970 dyninf->rel_addr = addr;
1971 dyninf->rel_size += s->sh_size;
1972 #endif
1974 addr += s->sh_size;
1975 if (s->sh_type != SHT_NOBITS)
1976 file_offset += s->sh_size;
1979 if (j == 0) {
1980 /* Make the first PT_LOAD segment include the program
1981 headers itself (and the ELF header as well), it'll
1982 come out with same memory use but will make various
1983 tools like binutils strip work better. */
1984 ph->p_offset &= ~(ph->p_align - 1);
1985 ph->p_vaddr &= ~(ph->p_align - 1);
1986 ph->p_paddr &= ~(ph->p_align - 1);
1988 ph->p_filesz = file_offset - ph->p_offset;
1989 ph->p_memsz = addr - ph->p_vaddr;
1990 ph++;
1991 if (j == 0) {
1992 if (s1->output_format == TCC_OUTPUT_FORMAT_ELF) {
1993 /* if in the middle of a page, we duplicate the page in
1994 memory so that one copy is RX and the other is RW */
1995 if ((addr & (s1->section_align - 1)) != 0)
1996 addr += s1->section_align;
1997 } else {
1998 addr = (addr + s1->section_align - 1) & ~(s1->section_align - 1);
1999 file_offset = (file_offset + s1->section_align - 1) &
2000 ~(s1->section_align - 1);
2006 /* all other sections come after */
2007 for(i = 1; i < s1->nb_sections; i++) {
2008 s = s1->sections[i];
2009 if (phnum > 0 && (s->sh_flags & SHF_ALLOC))
2010 continue;
2011 sec_order[sh_order_index++] = i;
2013 file_offset = (file_offset + s->sh_addralign - 1) &
2014 ~(s->sh_addralign - 1);
2015 s->sh_offset = file_offset;
2016 if (s->sh_type != SHT_NOBITS)
2017 file_offset += s->sh_size;
2020 return file_offset;
2023 static void fill_unloadable_phdr(ElfW(Phdr) *phdr, int phnum, Section *interp,
2024 Section *dynamic)
2026 ElfW(Phdr) *ph;
2028 /* if interpreter, then add corresponding program header */
2029 if (interp) {
2030 ph = &phdr[0];
2032 if (HAVE_PHDR)
2034 int len = phnum * sizeof(ElfW(Phdr));
2036 ph->p_type = PT_PHDR;
2037 ph->p_offset = sizeof(ElfW(Ehdr));
2038 ph->p_vaddr = interp->sh_addr - len;
2039 ph->p_paddr = ph->p_vaddr;
2040 ph->p_filesz = ph->p_memsz = len;
2041 ph->p_flags = PF_R | PF_X;
2042 ph->p_align = 4; /* interp->sh_addralign; */
2043 ph++;
2046 ph->p_type = PT_INTERP;
2047 ph->p_offset = interp->sh_offset;
2048 ph->p_vaddr = interp->sh_addr;
2049 ph->p_paddr = ph->p_vaddr;
2050 ph->p_filesz = interp->sh_size;
2051 ph->p_memsz = interp->sh_size;
2052 ph->p_flags = PF_R;
2053 ph->p_align = interp->sh_addralign;
2056 /* if dynamic section, then add corresponding program header */
2057 if (dynamic) {
2058 ph = &phdr[phnum - 1];
2060 ph->p_type = PT_DYNAMIC;
2061 ph->p_offset = dynamic->sh_offset;
2062 ph->p_vaddr = dynamic->sh_addr;
2063 ph->p_paddr = ph->p_vaddr;
2064 ph->p_filesz = dynamic->sh_size;
2065 ph->p_memsz = dynamic->sh_size;
2066 ph->p_flags = PF_R | PF_W;
2067 ph->p_align = dynamic->sh_addralign;
2071 /* Fill the dynamic section with tags describing the address and size of
2072 sections */
2073 static void fill_dynamic(TCCState *s1, struct dyn_inf *dyninf)
2075 Section *dynamic;
2077 dynamic = dyninf->dynamic;
2079 /* put dynamic section entries */
2080 dynamic->data_offset = dyninf->dyn_rel_off;
2081 put_dt(dynamic, DT_HASH, s1->dynsym->hash->sh_addr);
2082 put_dt(dynamic, DT_STRTAB, dyninf->dynstr->sh_addr);
2083 put_dt(dynamic, DT_SYMTAB, s1->dynsym->sh_addr);
2084 put_dt(dynamic, DT_STRSZ, dyninf->dynstr->data_offset);
2085 put_dt(dynamic, DT_SYMENT, sizeof(ElfW(Sym)));
2086 #ifdef TCC_TARGET_X86_64
2087 put_dt(dynamic, DT_RELA, dyninf->rel_addr);
2088 put_dt(dynamic, DT_RELASZ, dyninf->rel_size);
2089 put_dt(dynamic, DT_RELAENT, sizeof(ElfW_Rel));
2090 #else
2091 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
2092 put_dt(dynamic, DT_PLTGOT, s1->got->sh_addr);
2093 put_dt(dynamic, DT_PLTRELSZ, dyninf->rel_size);
2094 put_dt(dynamic, DT_JMPREL, dyninf->rel_addr);
2095 put_dt(dynamic, DT_PLTREL, DT_REL);
2096 put_dt(dynamic, DT_REL, dyninf->bss_addr);
2097 put_dt(dynamic, DT_RELSZ, dyninf->bss_size);
2098 #else
2099 put_dt(dynamic, DT_REL, dyninf->rel_addr);
2100 put_dt(dynamic, DT_RELSZ, dyninf->rel_size);
2101 put_dt(dynamic, DT_RELENT, sizeof(ElfW_Rel));
2102 #endif
2103 #endif
2104 if (s1->do_debug)
2105 put_dt(dynamic, DT_DEBUG, 0);
2106 put_dt(dynamic, DT_NULL, 0);
2109 /* Relocate remaining sections and symbols (that is those not related to
2110 dynamic linking) */
2111 static int final_sections_reloc(TCCState *s1)
2113 int i;
2114 Section *s;
2116 relocate_syms(s1, 0);
2118 if (s1->nb_errors != 0)
2119 return -1;
2121 /* relocate sections */
2122 /* XXX: ignore sections with allocated relocations ? */
2123 for(i = 1; i < s1->nb_sections; i++) {
2124 s = s1->sections[i];
2125 if (s->reloc && s != s1->got)
2126 relocate_section(s1, s);
2129 /* relocate relocation entries if the relocation tables are
2130 allocated in the executable */
2131 for(i = 1; i < s1->nb_sections; i++) {
2132 s = s1->sections[i];
2133 if ((s->sh_flags & SHF_ALLOC) &&
2134 s->sh_type == SHT_RELX) {
2135 relocate_rel(s1, s);
2138 return 0;
2141 /* Create an ELF file on disk.
2142 This function handle ELF specific layout requirements */
2143 static void tcc_output_elf(TCCState *s1, FILE *f, int phnum, ElfW(Phdr) *phdr,
2144 int file_offset, int *sec_order)
2146 int i, shnum, offset, size, file_type;
2147 Section *s;
2148 ElfW(Ehdr) ehdr;
2149 ElfW(Shdr) shdr, *sh;
2151 file_type = s1->output_type;
2152 shnum = s1->nb_sections;
2154 memset(&ehdr, 0, sizeof(ehdr));
2156 if (phnum > 0) {
2157 ehdr.e_phentsize = sizeof(ElfW(Phdr));
2158 ehdr.e_phnum = phnum;
2159 ehdr.e_phoff = sizeof(ElfW(Ehdr));
2162 /* align to 4 */
2163 file_offset = (file_offset + 3) & -4;
2165 /* fill header */
2166 ehdr.e_ident[0] = ELFMAG0;
2167 ehdr.e_ident[1] = ELFMAG1;
2168 ehdr.e_ident[2] = ELFMAG2;
2169 ehdr.e_ident[3] = ELFMAG3;
2170 ehdr.e_ident[4] = ELFCLASSW;
2171 ehdr.e_ident[5] = ELFDATA2LSB;
2172 ehdr.e_ident[6] = EV_CURRENT;
2173 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
2174 ehdr.e_ident[EI_OSABI] = ELFOSABI_FREEBSD;
2175 #endif
2176 #ifdef TCC_TARGET_ARM
2177 #ifdef TCC_ARM_EABI
2178 ehdr.e_ident[EI_OSABI] = 0;
2179 ehdr.e_flags = EF_ARM_EABI_VER4;
2180 if (file_type == TCC_OUTPUT_EXE || file_type == TCC_OUTPUT_DLL)
2181 ehdr.e_flags |= EF_ARM_HASENTRY;
2182 if (s1->float_abi == ARM_HARD_FLOAT)
2183 ehdr.e_flags |= EF_ARM_VFP_FLOAT;
2184 else
2185 ehdr.e_flags |= EF_ARM_SOFT_FLOAT;
2186 #else
2187 ehdr.e_ident[EI_OSABI] = ELFOSABI_ARM;
2188 #endif
2189 #endif
2190 switch(file_type) {
2191 default:
2192 case TCC_OUTPUT_EXE:
2193 ehdr.e_type = ET_EXEC;
2194 ehdr.e_entry = get_elf_sym_addr(s1, "_start", 1);
2195 break;
2196 case TCC_OUTPUT_DLL:
2197 ehdr.e_type = ET_DYN;
2198 ehdr.e_entry = text_section->sh_addr; /* XXX: is it correct ? */
2199 break;
2200 case TCC_OUTPUT_OBJ:
2201 ehdr.e_type = ET_REL;
2202 break;
2204 ehdr.e_machine = EM_TCC_TARGET;
2205 ehdr.e_version = EV_CURRENT;
2206 ehdr.e_shoff = file_offset;
2207 ehdr.e_ehsize = sizeof(ElfW(Ehdr));
2208 ehdr.e_shentsize = sizeof(ElfW(Shdr));
2209 ehdr.e_shnum = shnum;
2210 ehdr.e_shstrndx = shnum - 1;
2212 fwrite(&ehdr, 1, sizeof(ElfW(Ehdr)), f);
2213 fwrite(phdr, 1, phnum * sizeof(ElfW(Phdr)), f);
2214 offset = sizeof(ElfW(Ehdr)) + phnum * sizeof(ElfW(Phdr));
2216 sort_syms(s1, symtab_section);
2217 for(i = 1; i < s1->nb_sections; i++) {
2218 s = s1->sections[sec_order[i]];
2219 if (s->sh_type != SHT_NOBITS) {
2220 if (s->sh_type == SHT_DYNSYM)
2221 patch_dynsym_undef(s1, s);
2222 while (offset < s->sh_offset) {
2223 fputc(0, f);
2224 offset++;
2226 size = s->sh_size;
2227 fwrite(s->data, 1, size, f);
2228 offset += size;
2232 /* output section headers */
2233 while (offset < ehdr.e_shoff) {
2234 fputc(0, f);
2235 offset++;
2238 for(i = 0; i < s1->nb_sections; i++) {
2239 sh = &shdr;
2240 memset(sh, 0, sizeof(ElfW(Shdr)));
2241 s = s1->sections[i];
2242 if (s) {
2243 sh->sh_name = s->sh_name;
2244 sh->sh_type = s->sh_type;
2245 sh->sh_flags = s->sh_flags;
2246 sh->sh_entsize = s->sh_entsize;
2247 sh->sh_info = s->sh_info;
2248 if (s->link)
2249 sh->sh_link = s->link->sh_num;
2250 sh->sh_addralign = s->sh_addralign;
2251 sh->sh_addr = s->sh_addr;
2252 sh->sh_offset = s->sh_offset;
2253 sh->sh_size = s->sh_size;
2255 fwrite(sh, 1, sizeof(ElfW(Shdr)), f);
2259 /* Write an elf, coff or "binary" file */
2260 static int tcc_write_elf_file(TCCState *s1, const char *filename, int phnum,
2261 ElfW(Phdr) *phdr, int file_offset, int *sec_order)
2263 int fd, mode, file_type;
2264 FILE *f;
2266 file_type = s1->output_type;
2267 if (file_type == TCC_OUTPUT_OBJ)
2268 mode = 0666;
2269 else
2270 mode = 0777;
2271 unlink(filename);
2272 fd = open(filename, O_WRONLY | O_CREAT | O_TRUNC | O_BINARY, mode);
2273 if (fd < 0) {
2274 tcc_error_noabort("could not write '%s'", filename);
2275 return -1;
2277 f = fdopen(fd, "wb");
2278 if (s1->verbose)
2279 printf("<- %s\n", filename);
2281 #ifdef TCC_TARGET_COFF
2282 if (s1->output_format == TCC_OUTPUT_FORMAT_COFF)
2283 tcc_output_coff(s1, f);
2284 else
2285 #endif
2286 if (s1->output_format == TCC_OUTPUT_FORMAT_ELF)
2287 tcc_output_elf(s1, f, phnum, phdr, file_offset, sec_order);
2288 else
2289 tcc_output_binary(s1, f, sec_order);
2290 fclose(f);
2292 return 0;
2295 /* Output an elf, coff or binary file */
2296 /* XXX: suppress unneeded sections */
2297 static int elf_output_file(TCCState *s1, const char *filename)
2299 int i, ret, phnum, shnum, file_type, file_offset, *sec_order;
2300 struct dyn_inf dyninf;
2301 ElfW(Phdr) *phdr;
2302 ElfW(Sym) *sym;
2303 Section *strsec, *interp, *dynamic, *dynstr;
2305 file_type = s1->output_type;
2306 s1->nb_errors = 0;
2308 /* if linking, also link in runtime libraries (libc, libgcc, etc.) */
2309 if (file_type != TCC_OUTPUT_OBJ) {
2310 tcc_add_runtime(s1);
2313 phdr = NULL;
2314 sec_order = NULL;
2315 interp = dynamic = dynstr = NULL; /* avoid warning */
2316 dyninf.dyn_rel_off = 0; /* avoid warning */
2318 if (file_type != TCC_OUTPUT_OBJ) {
2319 relocate_common_syms();
2321 tcc_add_linker_symbols(s1);
2323 if (!s1->static_link) {
2324 if (file_type == TCC_OUTPUT_EXE) {
2325 char *ptr;
2326 /* allow override the dynamic loader */
2327 const char *elfint = getenv("LD_SO");
2328 if (elfint == NULL)
2329 elfint = DEFAULT_ELFINTERP(s1);
2330 /* add interpreter section only if executable */
2331 interp = new_section(s1, ".interp", SHT_PROGBITS, SHF_ALLOC);
2332 interp->sh_addralign = 1;
2333 ptr = section_ptr_add(interp, 1 + strlen(elfint));
2334 strcpy(ptr, elfint);
2337 /* add dynamic symbol table */
2338 s1->dynsym = new_symtab(s1, ".dynsym", SHT_DYNSYM, SHF_ALLOC,
2339 ".dynstr",
2340 ".hash", SHF_ALLOC);
2341 dynstr = s1->dynsym->link;
2343 /* add dynamic section */
2344 dynamic = new_section(s1, ".dynamic", SHT_DYNAMIC,
2345 SHF_ALLOC | SHF_WRITE);
2346 dynamic->link = dynstr;
2347 dynamic->sh_entsize = sizeof(ElfW(Dyn));
2349 /* add PLT */
2350 s1->plt = new_section(s1, ".plt", SHT_PROGBITS,
2351 SHF_ALLOC | SHF_EXECINSTR);
2352 s1->plt->sh_entsize = 4;
2354 build_got(s1);
2356 if (file_type == TCC_OUTPUT_EXE) {
2357 bind_exe_dynsyms(s1);
2359 if (s1->nb_errors) {
2360 ret = -1;
2361 goto the_end;
2364 bind_libs_dynsyms(s1);
2365 } else /* shared library case: simply export all global symbols */
2366 export_global_syms(s1);
2368 build_got_entries(s1);
2370 /* add a list of needed dlls */
2371 for(i = 0; i < s1->nb_loaded_dlls; i++) {
2372 DLLReference *dllref = s1->loaded_dlls[i];
2373 if (dllref->level == 0)
2374 put_dt(dynamic, DT_NEEDED, put_elf_str(dynstr, dllref->name));
2377 if (s1->rpath)
2378 put_dt(dynamic, DT_RPATH, put_elf_str(dynstr, s1->rpath));
2380 /* XXX: currently, since we do not handle PIC code, we
2381 must relocate the readonly segments */
2382 if (file_type == TCC_OUTPUT_DLL) {
2383 if (s1->soname)
2384 put_dt(dynamic, DT_SONAME, put_elf_str(dynstr, s1->soname));
2385 put_dt(dynamic, DT_TEXTREL, 0);
2388 if (s1->symbolic)
2389 put_dt(dynamic, DT_SYMBOLIC, 0);
2391 /* add necessary space for other entries */
2392 dyninf.dyn_rel_off = dynamic->data_offset;
2393 dynamic->data_offset += sizeof(ElfW(Dyn)) * EXTRA_RELITEMS;
2394 } else {
2395 /* still need to build got entries in case of static link */
2396 build_got_entries(s1);
2400 /* we add a section for symbols */
2401 strsec = new_section(s1, ".shstrtab", SHT_STRTAB, 0);
2402 put_elf_str(strsec, "");
2404 /* compute number of sections */
2405 shnum = s1->nb_sections;
2407 /* this array is used to reorder sections in the output file */
2408 sec_order = tcc_malloc(sizeof(int) * shnum);
2409 sec_order[0] = 0;
2411 /* compute number of program headers */
2412 switch(file_type) {
2413 default:
2414 case TCC_OUTPUT_OBJ:
2415 phnum = 0;
2416 break;
2417 case TCC_OUTPUT_EXE:
2418 if (!s1->static_link)
2419 phnum = 4 + HAVE_PHDR;
2420 else
2421 phnum = 2;
2422 break;
2423 case TCC_OUTPUT_DLL:
2424 phnum = 3;
2425 break;
2428 /* Allocate strings for section names */
2429 alloc_sec_names(s1, file_type, strsec);
2431 /* allocate program segment headers */
2432 phdr = tcc_mallocz(phnum * sizeof(ElfW(Phdr)));
2434 /* compute section to program header mapping */
2435 file_offset = layout_sections(s1, phdr, phnum, interp, &dyninf, sec_order);
2437 /* Fill remaining program header and finalize relocation related to dynamic
2438 linking. */
2439 if (phnum > 0) {
2440 fill_unloadable_phdr(phdr, phnum, interp, dynamic);
2441 if (dynamic) {
2442 dyninf.dynamic = dynamic;
2443 dyninf.dynstr = dynstr;
2445 fill_dynamic(s1, &dyninf);
2447 /* put in GOT the dynamic section address and relocate PLT */
2448 put32(s1->got->data, dynamic->sh_addr);
2449 if (file_type == TCC_OUTPUT_EXE
2450 #if defined(TCC_OUTPUT_DLL_WITH_PLT)
2451 || file_type == TCC_OUTPUT_DLL
2452 #endif
2454 relocate_plt(s1);
2456 /* relocate symbols in .dynsym now that final addresses are known */
2457 for_each_elem(s1->dynsym, 1, sym, ElfW(Sym)) {
2458 /* relocate to PLT if symbol corresponds to a PLT entry */
2459 if (sym->st_shndx == SHN_UNDEF) {
2460 if (sym->st_value)
2461 sym->st_value += s1->plt->sh_addr;
2462 } else if (sym->st_shndx < SHN_LORESERVE) {
2463 /* do symbol relocation */
2464 sym->st_value += s1->sections[sym->st_shndx]->sh_addr;
2470 /* if building executable or DLL, then relocate each section
2471 except the GOT which is already relocated */
2472 if (file_type != TCC_OUTPUT_OBJ) {
2473 ret = final_sections_reloc(s1);
2474 if (ret)
2475 goto the_end;
2478 /* Perform relocation to GOT or PLT entries */
2479 if (file_type == TCC_OUTPUT_EXE && s1->static_link)
2480 fill_got(s1);
2482 /* Create the ELF file with name 'filename' */
2483 ret = tcc_write_elf_file(s1, filename, phnum, phdr, file_offset, sec_order);
2484 the_end:
2485 tcc_free(s1->symtab_to_dynsym);
2486 tcc_free(sec_order);
2487 tcc_free(phdr);
2488 tcc_free(s1->sym_attrs);
2489 s1->sym_attrs = NULL;
2490 return ret;
2493 LIBTCCAPI int tcc_output_file(TCCState *s, const char *filename)
2495 int ret;
2496 #ifdef TCC_TARGET_PE
2497 if (s->output_type != TCC_OUTPUT_OBJ) {
2498 ret = pe_output_file(s, filename);
2499 } else
2500 #endif
2501 ret = elf_output_file(s, filename);
2502 return ret;
2505 static void *load_data(int fd, unsigned long file_offset, unsigned long size)
2507 void *data;
2509 data = tcc_malloc(size);
2510 lseek(fd, file_offset, SEEK_SET);
2511 read(fd, data, size);
2512 return data;
2515 typedef struct SectionMergeInfo {
2516 Section *s; /* corresponding existing section */
2517 unsigned long offset; /* offset of the new section in the existing section */
2518 uint8_t new_section; /* true if section 's' was added */
2519 uint8_t link_once; /* true if link once section */
2520 } SectionMergeInfo;
2522 /* load an object file and merge it with current files */
2523 /* XXX: handle correctly stab (debug) info */
2524 ST_FUNC int tcc_load_object_file(TCCState *s1,
2525 int fd, unsigned long file_offset)
2527 ElfW(Ehdr) ehdr;
2528 ElfW(Shdr) *shdr, *sh;
2529 int size, i, j, offset, offseti, nb_syms, sym_index, ret;
2530 unsigned char *strsec, *strtab;
2531 int *old_to_new_syms;
2532 char *sh_name, *name;
2533 SectionMergeInfo *sm_table, *sm;
2534 ElfW(Sym) *sym, *symtab;
2535 ElfW_Rel *rel;
2536 Section *s;
2538 int stab_index;
2539 int stabstr_index;
2541 stab_index = stabstr_index = 0;
2543 if (read(fd, &ehdr, sizeof(ehdr)) != sizeof(ehdr))
2544 goto fail1;
2545 if (ehdr.e_ident[0] != ELFMAG0 ||
2546 ehdr.e_ident[1] != ELFMAG1 ||
2547 ehdr.e_ident[2] != ELFMAG2 ||
2548 ehdr.e_ident[3] != ELFMAG3)
2549 goto fail1;
2550 /* test if object file */
2551 if (ehdr.e_type != ET_REL)
2552 goto fail1;
2553 /* test CPU specific stuff */
2554 if (ehdr.e_ident[5] != ELFDATA2LSB ||
2555 ehdr.e_machine != EM_TCC_TARGET) {
2556 fail1:
2557 tcc_error_noabort("invalid object file");
2558 return -1;
2560 /* read sections */
2561 shdr = load_data(fd, file_offset + ehdr.e_shoff,
2562 sizeof(ElfW(Shdr)) * ehdr.e_shnum);
2563 sm_table = tcc_mallocz(sizeof(SectionMergeInfo) * ehdr.e_shnum);
2565 /* load section names */
2566 sh = &shdr[ehdr.e_shstrndx];
2567 strsec = load_data(fd, file_offset + sh->sh_offset, sh->sh_size);
2569 /* load symtab and strtab */
2570 old_to_new_syms = NULL;
2571 symtab = NULL;
2572 strtab = NULL;
2573 nb_syms = 0;
2574 for(i = 1; i < ehdr.e_shnum; i++) {
2575 sh = &shdr[i];
2576 if (sh->sh_type == SHT_SYMTAB) {
2577 if (symtab) {
2578 tcc_error_noabort("object must contain only one symtab");
2579 fail:
2580 ret = -1;
2581 goto the_end;
2583 nb_syms = sh->sh_size / sizeof(ElfW(Sym));
2584 symtab = load_data(fd, file_offset + sh->sh_offset, sh->sh_size);
2585 sm_table[i].s = symtab_section;
2587 /* now load strtab */
2588 sh = &shdr[sh->sh_link];
2589 strtab = load_data(fd, file_offset + sh->sh_offset, sh->sh_size);
2593 /* now examine each section and try to merge its content with the
2594 ones in memory */
2595 for(i = 1; i < ehdr.e_shnum; i++) {
2596 /* no need to examine section name strtab */
2597 if (i == ehdr.e_shstrndx)
2598 continue;
2599 sh = &shdr[i];
2600 sh_name = (char *) strsec + sh->sh_name;
2601 /* ignore sections types we do not handle */
2602 if (sh->sh_type != SHT_PROGBITS &&
2603 sh->sh_type != SHT_RELX &&
2604 #ifdef TCC_ARM_EABI
2605 sh->sh_type != SHT_ARM_EXIDX &&
2606 #endif
2607 sh->sh_type != SHT_NOBITS &&
2608 sh->sh_type != SHT_PREINIT_ARRAY &&
2609 sh->sh_type != SHT_INIT_ARRAY &&
2610 sh->sh_type != SHT_FINI_ARRAY &&
2611 strcmp(sh_name, ".stabstr")
2613 continue;
2614 if (sh->sh_addralign < 1)
2615 sh->sh_addralign = 1;
2616 /* find corresponding section, if any */
2617 for(j = 1; j < s1->nb_sections;j++) {
2618 s = s1->sections[j];
2619 if (!strcmp(s->name, sh_name)) {
2620 if (!strncmp(sh_name, ".gnu.linkonce",
2621 sizeof(".gnu.linkonce") - 1)) {
2622 /* if a 'linkonce' section is already present, we
2623 do not add it again. It is a little tricky as
2624 symbols can still be defined in
2625 it. */
2626 sm_table[i].link_once = 1;
2627 goto next;
2628 } else {
2629 goto found;
2633 /* not found: create new section */
2634 s = new_section(s1, sh_name, sh->sh_type, sh->sh_flags);
2635 /* take as much info as possible from the section. sh_link and
2636 sh_info will be updated later */
2637 s->sh_addralign = sh->sh_addralign;
2638 s->sh_entsize = sh->sh_entsize;
2639 sm_table[i].new_section = 1;
2640 found:
2641 if (sh->sh_type != s->sh_type) {
2642 tcc_error_noabort("invalid section type");
2643 goto fail;
2646 /* align start of section */
2647 offset = s->data_offset;
2649 if (0 == strcmp(sh_name, ".stab")) {
2650 stab_index = i;
2651 goto no_align;
2653 if (0 == strcmp(sh_name, ".stabstr")) {
2654 stabstr_index = i;
2655 goto no_align;
2658 size = sh->sh_addralign - 1;
2659 offset = (offset + size) & ~size;
2660 if (sh->sh_addralign > s->sh_addralign)
2661 s->sh_addralign = sh->sh_addralign;
2662 s->data_offset = offset;
2663 no_align:
2664 sm_table[i].offset = offset;
2665 sm_table[i].s = s;
2666 /* concatenate sections */
2667 size = sh->sh_size;
2668 if (sh->sh_type != SHT_NOBITS) {
2669 unsigned char *ptr;
2670 lseek(fd, file_offset + sh->sh_offset, SEEK_SET);
2671 ptr = section_ptr_add(s, size);
2672 read(fd, ptr, size);
2673 } else {
2674 s->data_offset += size;
2676 next: ;
2679 /* gr relocate stab strings */
2680 if (stab_index && stabstr_index) {
2681 Stab_Sym *a, *b;
2682 unsigned o;
2683 s = sm_table[stab_index].s;
2684 a = (Stab_Sym *)(s->data + sm_table[stab_index].offset);
2685 b = (Stab_Sym *)(s->data + s->data_offset);
2686 o = sm_table[stabstr_index].offset;
2687 while (a < b)
2688 a->n_strx += o, a++;
2691 /* second short pass to update sh_link and sh_info fields of new
2692 sections */
2693 for(i = 1; i < ehdr.e_shnum; i++) {
2694 s = sm_table[i].s;
2695 if (!s || !sm_table[i].new_section)
2696 continue;
2697 sh = &shdr[i];
2698 if (sh->sh_link > 0)
2699 s->link = sm_table[sh->sh_link].s;
2700 if (sh->sh_type == SHT_RELX) {
2701 s->sh_info = sm_table[sh->sh_info].s->sh_num;
2702 /* update backward link */
2703 s1->sections[s->sh_info]->reloc = s;
2706 sm = sm_table;
2708 /* resolve symbols */
2709 old_to_new_syms = tcc_mallocz(nb_syms * sizeof(int));
2711 sym = symtab + 1;
2712 for(i = 1; i < nb_syms; i++, sym++) {
2713 if (sym->st_shndx != SHN_UNDEF &&
2714 sym->st_shndx < SHN_LORESERVE) {
2715 sm = &sm_table[sym->st_shndx];
2716 if (sm->link_once) {
2717 /* if a symbol is in a link once section, we use the
2718 already defined symbol. It is very important to get
2719 correct relocations */
2720 if (ELFW(ST_BIND)(sym->st_info) != STB_LOCAL) {
2721 name = (char *) strtab + sym->st_name;
2722 sym_index = find_elf_sym(symtab_section, name);
2723 if (sym_index)
2724 old_to_new_syms[i] = sym_index;
2726 continue;
2728 /* if no corresponding section added, no need to add symbol */
2729 if (!sm->s)
2730 continue;
2731 /* convert section number */
2732 sym->st_shndx = sm->s->sh_num;
2733 /* offset value */
2734 sym->st_value += sm->offset;
2736 /* add symbol */
2737 name = (char *) strtab + sym->st_name;
2738 sym_index = add_elf_sym(symtab_section, sym->st_value, sym->st_size,
2739 sym->st_info, sym->st_other,
2740 sym->st_shndx, name);
2741 old_to_new_syms[i] = sym_index;
2744 /* third pass to patch relocation entries */
2745 for(i = 1; i < ehdr.e_shnum; i++) {
2746 s = sm_table[i].s;
2747 if (!s)
2748 continue;
2749 sh = &shdr[i];
2750 offset = sm_table[i].offset;
2751 switch(s->sh_type) {
2752 case SHT_RELX:
2753 /* take relocation offset information */
2754 offseti = sm_table[sh->sh_info].offset;
2755 for_each_elem(s, (offset / sizeof(*rel)), rel, ElfW_Rel) {
2756 int type;
2757 unsigned sym_index;
2758 /* convert symbol index */
2759 type = ELFW(R_TYPE)(rel->r_info);
2760 sym_index = ELFW(R_SYM)(rel->r_info);
2761 /* NOTE: only one symtab assumed */
2762 if (sym_index >= nb_syms)
2763 goto invalid_reloc;
2764 sym_index = old_to_new_syms[sym_index];
2765 /* ignore link_once in rel section. */
2766 if (!sym_index && !sm->link_once
2767 #ifdef TCC_TARGET_ARM
2768 && type != R_ARM_V4BX
2769 #endif
2771 invalid_reloc:
2772 tcc_error_noabort("Invalid relocation entry [%2d] '%s' @ %.8x",
2773 i, strsec + sh->sh_name, rel->r_offset);
2774 goto fail;
2776 rel->r_info = ELFW(R_INFO)(sym_index, type);
2777 /* offset the relocation offset */
2778 rel->r_offset += offseti;
2779 #ifdef TCC_TARGET_ARM
2780 /* Jumps and branches from a Thumb code to a PLT entry need
2781 special handling since PLT entries are ARM code.
2782 Unconditional bl instructions referencing PLT entries are
2783 handled by converting these instructions into blx
2784 instructions. Other case of instructions referencing a PLT
2785 entry require to add a Thumb stub before the PLT entry to
2786 switch to ARM mode. We set bit plt_thumb_stub of the
2787 attribute of a symbol to indicate such a case. */
2788 if (type == R_ARM_THM_JUMP24)
2789 alloc_sym_attr(s1, sym_index)->plt_thumb_stub = 1;
2790 #endif
2792 break;
2793 default:
2794 break;
2798 ret = 0;
2799 the_end:
2800 tcc_free(symtab);
2801 tcc_free(strtab);
2802 tcc_free(old_to_new_syms);
2803 tcc_free(sm_table);
2804 tcc_free(strsec);
2805 tcc_free(shdr);
2806 return ret;
2809 typedef struct ArchiveHeader {
2810 char ar_name[16]; /* name of this member */
2811 char ar_date[12]; /* file mtime */
2812 char ar_uid[6]; /* owner uid; printed as decimal */
2813 char ar_gid[6]; /* owner gid; printed as decimal */
2814 char ar_mode[8]; /* file mode, printed as octal */
2815 char ar_size[10]; /* file size, printed as decimal */
2816 char ar_fmag[2]; /* should contain ARFMAG */
2817 } ArchiveHeader;
2819 static int get_be32(const uint8_t *b)
2821 return b[3] | (b[2] << 8) | (b[1] << 16) | (b[0] << 24);
2824 /* load only the objects which resolve undefined symbols */
2825 static int tcc_load_alacarte(TCCState *s1, int fd, int size)
2827 int i, bound, nsyms, sym_index, off, ret;
2828 uint8_t *data;
2829 const char *ar_names, *p;
2830 const uint8_t *ar_index;
2831 ElfW(Sym) *sym;
2833 data = tcc_malloc(size);
2834 if (read(fd, data, size) != size)
2835 goto fail;
2836 nsyms = get_be32(data);
2837 ar_index = data + 4;
2838 ar_names = (char *) ar_index + nsyms * 4;
2840 do {
2841 bound = 0;
2842 for(p = ar_names, i = 0; i < nsyms; i++, p += strlen(p)+1) {
2843 sym_index = find_elf_sym(symtab_section, p);
2844 if(sym_index) {
2845 sym = &((ElfW(Sym) *)symtab_section->data)[sym_index];
2846 if(sym->st_shndx == SHN_UNDEF) {
2847 off = get_be32(ar_index + i * 4) + sizeof(ArchiveHeader);
2848 ++bound;
2849 lseek(fd, off, SEEK_SET);
2850 if(tcc_load_object_file(s1, fd, off) < 0) {
2851 fail:
2852 ret = -1;
2853 goto the_end;
2858 } while(bound);
2859 ret = 0;
2860 the_end:
2861 tcc_free(data);
2862 return ret;
2865 /* load a '.a' file */
2866 ST_FUNC int tcc_load_archive(TCCState *s1, int fd)
2868 ArchiveHeader hdr;
2869 char ar_size[11];
2870 char ar_name[17];
2871 char magic[8];
2872 int size, len, i;
2873 unsigned long file_offset;
2875 /* skip magic which was already checked */
2876 read(fd, magic, sizeof(magic));
2878 for(;;) {
2879 len = read(fd, &hdr, sizeof(hdr));
2880 if (len == 0)
2881 break;
2882 if (len != sizeof(hdr)) {
2883 tcc_error_noabort("invalid archive");
2884 return -1;
2886 memcpy(ar_size, hdr.ar_size, sizeof(hdr.ar_size));
2887 ar_size[sizeof(hdr.ar_size)] = '\0';
2888 size = strtol(ar_size, NULL, 0);
2889 memcpy(ar_name, hdr.ar_name, sizeof(hdr.ar_name));
2890 for(i = sizeof(hdr.ar_name) - 1; i >= 0; i--) {
2891 if (ar_name[i] != ' ')
2892 break;
2894 ar_name[i + 1] = '\0';
2895 file_offset = lseek(fd, 0, SEEK_CUR);
2896 /* align to even */
2897 size = (size + 1) & ~1;
2898 if (!strcmp(ar_name, "/")) {
2899 /* coff symbol table : we handle it */
2900 if(s1->alacarte_link)
2901 return tcc_load_alacarte(s1, fd, size);
2902 } else if (!strcmp(ar_name, "//") ||
2903 !strcmp(ar_name, "__.SYMDEF") ||
2904 !strcmp(ar_name, "__.SYMDEF/") ||
2905 !strcmp(ar_name, "ARFILENAMES/")) {
2906 /* skip symbol table or archive names */
2907 } else {
2908 if (tcc_load_object_file(s1, fd, file_offset) < 0)
2909 return -1;
2911 lseek(fd, file_offset + size, SEEK_SET);
2913 return 0;
2916 #ifndef TCC_TARGET_PE
2917 /* load a DLL and all referenced DLLs. 'level = 0' means that the DLL
2918 is referenced by the user (so it should be added as DT_NEEDED in
2919 the generated ELF file) */
2920 ST_FUNC int tcc_load_dll(TCCState *s1, int fd, const char *filename, int level)
2922 ElfW(Ehdr) ehdr;
2923 ElfW(Shdr) *shdr, *sh, *sh1;
2924 int i, j, nb_syms, nb_dts, sym_bind, ret;
2925 ElfW(Sym) *sym, *dynsym;
2926 ElfW(Dyn) *dt, *dynamic;
2927 unsigned char *dynstr;
2928 const char *name, *soname;
2929 DLLReference *dllref;
2931 read(fd, &ehdr, sizeof(ehdr));
2933 /* test CPU specific stuff */
2934 if (ehdr.e_ident[5] != ELFDATA2LSB ||
2935 ehdr.e_machine != EM_TCC_TARGET) {
2936 tcc_error_noabort("bad architecture");
2937 return -1;
2940 /* read sections */
2941 shdr = load_data(fd, ehdr.e_shoff, sizeof(ElfW(Shdr)) * ehdr.e_shnum);
2943 /* load dynamic section and dynamic symbols */
2944 nb_syms = 0;
2945 nb_dts = 0;
2946 dynamic = NULL;
2947 dynsym = NULL; /* avoid warning */
2948 dynstr = NULL; /* avoid warning */
2949 for(i = 0, sh = shdr; i < ehdr.e_shnum; i++, sh++) {
2950 switch(sh->sh_type) {
2951 case SHT_DYNAMIC:
2952 nb_dts = sh->sh_size / sizeof(ElfW(Dyn));
2953 dynamic = load_data(fd, sh->sh_offset, sh->sh_size);
2954 break;
2955 case SHT_DYNSYM:
2956 nb_syms = sh->sh_size / sizeof(ElfW(Sym));
2957 dynsym = load_data(fd, sh->sh_offset, sh->sh_size);
2958 sh1 = &shdr[sh->sh_link];
2959 dynstr = load_data(fd, sh1->sh_offset, sh1->sh_size);
2960 break;
2961 default:
2962 break;
2966 /* compute the real library name */
2967 soname = tcc_basename(filename);
2969 for(i = 0, dt = dynamic; i < nb_dts; i++, dt++) {
2970 if (dt->d_tag == DT_SONAME) {
2971 soname = (char *) dynstr + dt->d_un.d_val;
2975 /* if the dll is already loaded, do not load it */
2976 for(i = 0; i < s1->nb_loaded_dlls; i++) {
2977 dllref = s1->loaded_dlls[i];
2978 if (!strcmp(soname, dllref->name)) {
2979 /* but update level if needed */
2980 if (level < dllref->level)
2981 dllref->level = level;
2982 ret = 0;
2983 goto the_end;
2987 /* add the dll and its level */
2988 dllref = tcc_mallocz(sizeof(DLLReference) + strlen(soname));
2989 dllref->level = level;
2990 strcpy(dllref->name, soname);
2991 dynarray_add((void ***)&s1->loaded_dlls, &s1->nb_loaded_dlls, dllref);
2993 /* add dynamic symbols in dynsym_section */
2994 for(i = 1, sym = dynsym + 1; i < nb_syms; i++, sym++) {
2995 sym_bind = ELFW(ST_BIND)(sym->st_info);
2996 if (sym_bind == STB_LOCAL)
2997 continue;
2998 name = (char *) dynstr + sym->st_name;
2999 add_elf_sym(s1->dynsymtab_section, sym->st_value, sym->st_size,
3000 sym->st_info, sym->st_other, sym->st_shndx, name);
3003 /* load all referenced DLLs */
3004 for(i = 0, dt = dynamic; i < nb_dts; i++, dt++) {
3005 switch(dt->d_tag) {
3006 case DT_NEEDED:
3007 name = (char *) dynstr + dt->d_un.d_val;
3008 for(j = 0; j < s1->nb_loaded_dlls; j++) {
3009 dllref = s1->loaded_dlls[j];
3010 if (!strcmp(name, dllref->name))
3011 goto already_loaded;
3013 if (tcc_add_dll(s1, name, AFF_REFERENCED_DLL) < 0) {
3014 tcc_error_noabort("referenced dll '%s' not found", name);
3015 ret = -1;
3016 goto the_end;
3018 already_loaded:
3019 break;
3022 ret = 0;
3023 the_end:
3024 tcc_free(dynstr);
3025 tcc_free(dynsym);
3026 tcc_free(dynamic);
3027 tcc_free(shdr);
3028 return ret;
3031 #define LD_TOK_NAME 256
3032 #define LD_TOK_EOF (-1)
3034 /* return next ld script token */
3035 static int ld_next(TCCState *s1, char *name, int name_size)
3037 int c;
3038 char *q;
3040 redo:
3041 switch(ch) {
3042 case ' ':
3043 case '\t':
3044 case '\f':
3045 case '\v':
3046 case '\r':
3047 case '\n':
3048 inp();
3049 goto redo;
3050 case '/':
3051 minp();
3052 if (ch == '*') {
3053 file->buf_ptr = parse_comment(file->buf_ptr);
3054 ch = file->buf_ptr[0];
3055 goto redo;
3056 } else {
3057 q = name;
3058 *q++ = '/';
3059 goto parse_name;
3061 break;
3062 /* case 'a' ... 'z': */
3063 case 'a':
3064 case 'b':
3065 case 'c':
3066 case 'd':
3067 case 'e':
3068 case 'f':
3069 case 'g':
3070 case 'h':
3071 case 'i':
3072 case 'j':
3073 case 'k':
3074 case 'l':
3075 case 'm':
3076 case 'n':
3077 case 'o':
3078 case 'p':
3079 case 'q':
3080 case 'r':
3081 case 's':
3082 case 't':
3083 case 'u':
3084 case 'v':
3085 case 'w':
3086 case 'x':
3087 case 'y':
3088 case 'z':
3089 /* case 'A' ... 'z': */
3090 case 'A':
3091 case 'B':
3092 case 'C':
3093 case 'D':
3094 case 'E':
3095 case 'F':
3096 case 'G':
3097 case 'H':
3098 case 'I':
3099 case 'J':
3100 case 'K':
3101 case 'L':
3102 case 'M':
3103 case 'N':
3104 case 'O':
3105 case 'P':
3106 case 'Q':
3107 case 'R':
3108 case 'S':
3109 case 'T':
3110 case 'U':
3111 case 'V':
3112 case 'W':
3113 case 'X':
3114 case 'Y':
3115 case 'Z':
3116 case '_':
3117 case '\\':
3118 case '.':
3119 case '$':
3120 case '~':
3121 q = name;
3122 parse_name:
3123 for(;;) {
3124 if (!((ch >= 'a' && ch <= 'z') ||
3125 (ch >= 'A' && ch <= 'Z') ||
3126 (ch >= '0' && ch <= '9') ||
3127 strchr("/.-_+=$:\\,~", ch)))
3128 break;
3129 if ((q - name) < name_size - 1) {
3130 *q++ = ch;
3132 minp();
3134 *q = '\0';
3135 c = LD_TOK_NAME;
3136 break;
3137 case CH_EOF:
3138 c = LD_TOK_EOF;
3139 break;
3140 default:
3141 c = ch;
3142 inp();
3143 break;
3145 return c;
3148 static int ld_add_file(TCCState *s1, const char filename[])
3150 int ret;
3152 ret = tcc_add_file_internal(s1, filename, 0);
3153 if (ret)
3154 ret = tcc_add_dll(s1, filename, 0);
3155 return ret;
3158 static inline int new_undef_syms(void)
3160 int ret = 0;
3161 ret = new_undef_sym;
3162 new_undef_sym = 0;
3163 return ret;
3166 static int ld_add_file_list(TCCState *s1, const char *cmd, int as_needed)
3168 char filename[1024], libname[1024];
3169 int t, group, nblibs = 0, ret = 0;
3170 char **libs = NULL;
3172 group = !strcmp(cmd, "GROUP");
3173 if (!as_needed)
3174 new_undef_syms();
3175 t = ld_next(s1, filename, sizeof(filename));
3176 if (t != '(')
3177 expect("(");
3178 t = ld_next(s1, filename, sizeof(filename));
3179 for(;;) {
3180 libname[0] = '\0';
3181 if (t == LD_TOK_EOF) {
3182 tcc_error_noabort("unexpected end of file");
3183 ret = -1;
3184 goto lib_parse_error;
3185 } else if (t == ')') {
3186 break;
3187 } else if (t == '-') {
3188 t = ld_next(s1, filename, sizeof(filename));
3189 if ((t != LD_TOK_NAME) || (filename[0] != 'l')) {
3190 tcc_error_noabort("library name expected");
3191 ret = -1;
3192 goto lib_parse_error;
3194 pstrcpy(libname, sizeof libname, &filename[1]);
3195 if (s1->static_link) {
3196 snprintf(filename, sizeof filename, "lib%s.a", libname);
3197 } else {
3198 snprintf(filename, sizeof filename, "lib%s.so", libname);
3200 } else if (t != LD_TOK_NAME) {
3201 tcc_error_noabort("filename expected");
3202 ret = -1;
3203 goto lib_parse_error;
3205 if (!strcmp(filename, "AS_NEEDED")) {
3206 ret = ld_add_file_list(s1, cmd, 1);
3207 if (ret)
3208 goto lib_parse_error;
3209 } else {
3210 /* TODO: Implement AS_NEEDED support. Ignore it for now */
3211 if (!as_needed) {
3212 ret = ld_add_file(s1, filename);
3213 if (ret)
3214 goto lib_parse_error;
3215 if (group) {
3216 /* Add the filename *and* the libname to avoid future conversions */
3217 dynarray_add((void ***) &libs, &nblibs, tcc_strdup(filename));
3218 if (libname[0] != '\0')
3219 dynarray_add((void ***) &libs, &nblibs, tcc_strdup(libname));
3223 t = ld_next(s1, filename, sizeof(filename));
3224 if (t == ',') {
3225 t = ld_next(s1, filename, sizeof(filename));
3228 if (group && !as_needed) {
3229 while (new_undef_syms()) {
3230 int i;
3232 for (i = 0; i < nblibs; i ++)
3233 ld_add_file(s1, libs[i]);
3236 lib_parse_error:
3237 dynarray_reset(&libs, &nblibs);
3238 return ret;
3241 /* interpret a subset of GNU ldscripts to handle the dummy libc.so
3242 files */
3243 ST_FUNC int tcc_load_ldscript(TCCState *s1)
3245 char cmd[64];
3246 char filename[1024];
3247 int t, ret;
3249 ch = file->buf_ptr[0];
3250 ch = handle_eob();
3251 for(;;) {
3252 t = ld_next(s1, cmd, sizeof(cmd));
3253 if (t == LD_TOK_EOF)
3254 return 0;
3255 else if (t != LD_TOK_NAME)
3256 return -1;
3257 if (!strcmp(cmd, "INPUT") ||
3258 !strcmp(cmd, "GROUP")) {
3259 ret = ld_add_file_list(s1, cmd, 0);
3260 if (ret)
3261 return ret;
3262 } else if (!strcmp(cmd, "OUTPUT_FORMAT") ||
3263 !strcmp(cmd, "TARGET")) {
3264 /* ignore some commands */
3265 t = ld_next(s1, cmd, sizeof(cmd));
3266 if (t != '(')
3267 expect("(");
3268 for(;;) {
3269 t = ld_next(s1, filename, sizeof(filename));
3270 if (t == LD_TOK_EOF) {
3271 tcc_error_noabort("unexpected end of file");
3272 return -1;
3273 } else if (t == ')') {
3274 break;
3277 } else {
3278 return -1;
3281 return 0;
3283 #endif /* !TCC_TARGET_PE */