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
23 static int new_undef_sym
= 0; /* Is there a new undefined sym since last new_undef_sym() */
25 ST_FUNC
int put_elf_str(Section
*s
, const char *sym
)
30 len
= strlen(sym
) + 1;
31 offset
= s
->data_offset
;
32 ptr
= section_ptr_add(s
, len
);
33 memcpy(ptr
, sym
, len
);
37 /* elf symbol hashing function */
38 static unsigned long elf_hash(const unsigned char *name
)
40 unsigned long h
= 0, g
;
43 h
= (h
<< 4) + *name
++;
52 /* rebuild hash table of section s */
53 /* NOTE: we do factorize the hash table code to go faster */
54 static void rebuild_hash(Section
*s
, unsigned int nb_buckets
)
57 int *ptr
, *hash
, nb_syms
, sym_index
, h
;
60 strtab
= s
->link
->data
;
61 nb_syms
= s
->data_offset
/ sizeof(ElfW(Sym
));
63 s
->hash
->data_offset
= 0;
64 ptr
= section_ptr_add(s
->hash
, (2 + nb_buckets
+ nb_syms
) * sizeof(int));
69 memset(hash
, 0, (nb_buckets
+ 1) * sizeof(int));
70 ptr
+= nb_buckets
+ 1;
72 sym
= (ElfW(Sym
) *)s
->data
+ 1;
73 for(sym_index
= 1; sym_index
< nb_syms
; sym_index
++) {
74 if (ELFW(ST_BIND
)(sym
->st_info
) != STB_LOCAL
) {
75 h
= elf_hash(strtab
+ sym
->st_name
) % nb_buckets
;
86 /* return the symbol number */
87 ST_FUNC
int put_elf_sym(Section
*s
, uplong value
, unsigned long size
,
88 int info
, int other
, int shndx
, const char *name
)
90 int name_offset
, sym_index
;
95 sym
= section_ptr_add(s
, sizeof(ElfW(Sym
)));
97 name_offset
= put_elf_str(s
->link
, name
);
100 /* XXX: endianness */
101 sym
->st_name
= name_offset
;
102 sym
->st_value
= value
;
105 sym
->st_other
= other
;
106 sym
->st_shndx
= shndx
;
107 sym_index
= sym
- (ElfW(Sym
) *)s
->data
;
111 ptr
= section_ptr_add(hs
, sizeof(int));
112 base
= (int *)hs
->data
;
113 /* only add global or weak symbols */
114 if (ELFW(ST_BIND
)(info
) != STB_LOCAL
) {
115 /* add another hashing entry */
117 h
= elf_hash(name
) % nbuckets
;
119 base
[2 + h
] = sym_index
;
121 /* we resize the hash table */
122 hs
->nb_hashed_syms
++;
123 if (hs
->nb_hashed_syms
> 2 * nbuckets
) {
124 rebuild_hash(s
, 2 * nbuckets
);
134 /* find global ELF symbol 'name' and return its index. Return 0 if not
136 ST_FUNC
int find_elf_sym(Section
*s
, const char *name
)
140 int nbuckets
, sym_index
, h
;
146 nbuckets
= ((int *)hs
->data
)[0];
147 h
= elf_hash(name
) % nbuckets
;
148 sym_index
= ((int *)hs
->data
)[2 + h
];
149 while (sym_index
!= 0) {
150 sym
= &((ElfW(Sym
) *)s
->data
)[sym_index
];
151 name1
= s
->link
->data
+ sym
->st_name
;
152 if (!strcmp(name
, name1
))
154 sym_index
= ((int *)hs
->data
)[2 + nbuckets
+ sym_index
];
159 /* return elf symbol value, signal error if 'err' is nonzero */
160 static void *get_elf_sym_addr(TCCState
*s
, const char *name
, int err
)
165 sym_index
= find_elf_sym(s
->symtab
, name
);
166 sym
= &((ElfW(Sym
) *)s
->symtab
->data
)[sym_index
];
167 if (!sym_index
|| sym
->st_shndx
== SHN_UNDEF
) {
169 tcc_error("%s not defined", name
);
172 return (void*)(uplong
)sym
->st_value
;
175 /* return elf symbol value */
176 LIBTCCAPI
void *tcc_get_symbol(TCCState
*s
, const char *name
)
178 return get_elf_sym_addr(s
, name
, 0);
181 /* return elf symbol value or error */
182 ST_FUNC
void *tcc_get_symbol_err(TCCState
*s
, const char *name
)
184 return get_elf_sym_addr(s
, name
, 1);
187 /* add an elf symbol : check if it is already defined and patch
188 it. Return symbol index. NOTE that sh_num can be SHN_UNDEF. */
189 ST_FUNC
int add_elf_sym(Section
*s
, uplong value
, unsigned long size
,
190 int info
, int other
, int sh_num
, const char *name
)
193 int sym_bind
, sym_index
, sym_type
, esym_bind
;
194 unsigned char sym_vis
, esym_vis
, new_vis
;
196 sym_bind
= ELFW(ST_BIND
)(info
);
197 sym_type
= ELFW(ST_TYPE
)(info
);
198 sym_vis
= ELFW(ST_VISIBILITY
)(other
);
200 if (sym_bind
!= STB_LOCAL
) {
201 /* we search global or weak symbols */
202 sym_index
= find_elf_sym(s
, name
);
205 esym
= &((ElfW(Sym
) *)s
->data
)[sym_index
];
206 if (esym
->st_shndx
!= SHN_UNDEF
) {
207 esym_bind
= ELFW(ST_BIND
)(esym
->st_info
);
208 /* propagate the most constraining visibility */
209 /* STV_DEFAULT(0)<STV_PROTECTED(3)<STV_HIDDEN(2)<STV_INTERNAL(1) */
210 esym_vis
= ELFW(ST_VISIBILITY
)(esym
->st_other
);
211 if (esym_vis
== STV_DEFAULT
) {
213 } else if (sym_vis
== STV_DEFAULT
) {
216 new_vis
= (esym_vis
< sym_vis
) ? esym_vis
: sym_vis
;
218 esym
->st_other
= (esym
->st_other
& ~ELFW(ST_VISIBILITY
)(-1))
220 other
= esym
->st_other
; /* in case we have to patch esym */
221 if (sh_num
== SHN_UNDEF
) {
222 /* ignore adding of undefined symbol if the
223 corresponding symbol is already defined */
224 } else if (sym_bind
== STB_GLOBAL
&& esym_bind
== STB_WEAK
) {
225 /* global overrides weak, so patch */
227 } else if (sym_bind
== STB_WEAK
&& esym_bind
== STB_GLOBAL
) {
228 /* weak is ignored if already global */
229 } else if (sym_bind
== STB_WEAK
&& esym_bind
== STB_WEAK
) {
230 /* keep first-found weak definition, ignore subsequents */
231 } else if (sym_vis
== STV_HIDDEN
|| sym_vis
== STV_INTERNAL
) {
232 /* ignore hidden symbols after */
233 } else if (esym
->st_shndx
== SHN_COMMON
234 && (sh_num
< SHN_LORESERVE
|| sh_num
== SHN_COMMON
)) {
235 /* gr: Happens with 'tcc ... -static tcctest.c' on e.g. Ubuntu 6.01
236 No idea if this is the correct solution ... */
238 } else if (s
== tcc_state
->dynsymtab_section
) {
239 /* we accept that two DLL define the same symbol */
242 printf("new_bind=%x new_shndx=%x new_vis=%x old_bind=%x old_shndx=%x old_vis=%x\n",
243 sym_bind
, sh_num
, new_vis
, esym_bind
, esym
->st_shndx
, esym_vis
);
245 tcc_error_noabort("'%s' defined twice", name
);
249 esym
->st_info
= ELFW(ST_INFO
)(sym_bind
, sym_type
);
250 esym
->st_shndx
= sh_num
;
252 esym
->st_value
= value
;
253 esym
->st_size
= size
;
254 esym
->st_other
= other
;
258 sym_index
= put_elf_sym(s
, value
, size
,
259 ELFW(ST_INFO
)(sym_bind
, sym_type
), other
,
266 ST_FUNC
void put_elf_reloc(Section
*symtab
, Section
*s
, unsigned long offset
,
267 int type
, int symbol
)
275 /* if no relocation section, create it */
276 snprintf(buf
, sizeof(buf
), REL_SECTION_FMT
, s
->name
);
277 /* if the symtab is allocated, then we consider the relocation
279 sr
= new_section(tcc_state
, buf
, SHT_RELX
, symtab
->sh_flags
);
280 sr
->sh_entsize
= sizeof(ElfW_Rel
);
282 sr
->sh_info
= s
->sh_num
;
285 rel
= section_ptr_add(sr
, sizeof(ElfW_Rel
));
286 rel
->r_offset
= offset
;
287 rel
->r_info
= ELFW(R_INFO
)(symbol
, type
);
288 #ifdef TCC_TARGET_X86_64
293 /* put stab debug information */
295 ST_FUNC
void put_stabs(const char *str
, int type
, int other
, int desc
,
300 sym
= section_ptr_add(stab_section
, sizeof(Stab_Sym
));
302 sym
->n_strx
= put_elf_str(stabstr_section
, str
);
307 sym
->n_other
= other
;
309 sym
->n_value
= value
;
312 ST_FUNC
void put_stabs_r(const char *str
, int type
, int other
, int desc
,
313 unsigned long value
, Section
*sec
, int sym_index
)
315 put_stabs(str
, type
, other
, desc
, value
);
316 put_elf_reloc(symtab_section
, stab_section
,
317 stab_section
->data_offset
- sizeof(unsigned int),
318 R_DATA_32
, sym_index
);
321 ST_FUNC
void put_stabn(int type
, int other
, int desc
, int value
)
323 put_stabs(NULL
, type
, other
, desc
, value
);
326 ST_FUNC
void put_stabd(int type
, int other
, int desc
)
328 put_stabs(NULL
, type
, other
, desc
, 0);
331 /* In an ELF file symbol table, the local symbols must appear below
332 the global and weak ones. Since TCC cannot sort it while generating
333 the code, we must do it after. All the relocation tables are also
334 modified to take into account the symbol table sorting */
335 static void sort_syms(TCCState
*s1
, Section
*s
)
337 int *old_to_new_syms
;
341 ElfW_Rel
*rel
, *rel_end
;
345 nb_syms
= s
->data_offset
/ sizeof(ElfW(Sym
));
346 new_syms
= tcc_malloc(nb_syms
* sizeof(ElfW(Sym
)));
347 old_to_new_syms
= tcc_malloc(nb_syms
* sizeof(int));
349 /* first pass for local symbols */
350 p
= (ElfW(Sym
) *)s
->data
;
352 for(i
= 0; i
< nb_syms
; i
++) {
353 if (ELFW(ST_BIND
)(p
->st_info
) == STB_LOCAL
) {
354 old_to_new_syms
[i
] = q
- new_syms
;
359 /* save the number of local symbols in section header */
360 s
->sh_info
= q
- new_syms
;
362 /* then second pass for non local symbols */
363 p
= (ElfW(Sym
) *)s
->data
;
364 for(i
= 0; i
< nb_syms
; i
++) {
365 if (ELFW(ST_BIND
)(p
->st_info
) != STB_LOCAL
) {
366 old_to_new_syms
[i
] = q
- new_syms
;
372 /* we copy the new symbols to the old */
373 memcpy(s
->data
, new_syms
, nb_syms
* sizeof(ElfW(Sym
)));
376 /* now we modify all the relocations */
377 for(i
= 1; i
< s1
->nb_sections
; i
++) {
378 sr
= s1
->sections
[i
];
379 if (sr
->sh_type
== SHT_RELX
&& sr
->link
== s
) {
380 rel_end
= (ElfW_Rel
*)(sr
->data
+ sr
->data_offset
);
381 for(rel
= (ElfW_Rel
*)sr
->data
;
384 sym_index
= ELFW(R_SYM
)(rel
->r_info
);
385 type
= ELFW(R_TYPE
)(rel
->r_info
);
386 sym_index
= old_to_new_syms
[sym_index
];
387 rel
->r_info
= ELFW(R_INFO
)(sym_index
, type
);
392 tcc_free(old_to_new_syms
);
395 /* relocate common symbols in the .bss section */
396 ST_FUNC
void relocate_common_syms(void)
398 ElfW(Sym
) *sym
, *sym_end
;
399 unsigned long offset
, align
;
401 sym_end
= (ElfW(Sym
) *)(symtab_section
->data
+ symtab_section
->data_offset
);
402 for(sym
= (ElfW(Sym
) *)symtab_section
->data
+ 1;
405 if (sym
->st_shndx
== SHN_COMMON
) {
407 align
= sym
->st_value
;
408 offset
= bss_section
->data_offset
;
409 offset
= (offset
+ align
- 1) & -align
;
410 sym
->st_value
= offset
;
411 sym
->st_shndx
= bss_section
->sh_num
;
412 offset
+= sym
->st_size
;
413 bss_section
->data_offset
= offset
;
418 /* relocate symbol table, resolve undefined symbols if do_resolve is
419 true and output error if undefined symbol. */
420 ST_FUNC
void relocate_syms(TCCState
*s1
, int do_resolve
)
422 ElfW(Sym
) *sym
, *esym
, *sym_end
;
423 int sym_bind
, sh_num
, sym_index
;
426 sym_end
= (ElfW(Sym
) *)(symtab_section
->data
+ symtab_section
->data_offset
);
427 for(sym
= (ElfW(Sym
) *)symtab_section
->data
+ 1;
430 sh_num
= sym
->st_shndx
;
431 if (sh_num
== SHN_UNDEF
) {
432 name
= strtab_section
->data
+ sym
->st_name
;
434 #if !defined TCC_TARGET_PE || !defined _WIN32
436 name
= symtab_section
->link
->data
+ sym
->st_name
;
437 addr
= resolve_sym(s1
, name
);
439 sym
->st_value
= (uplong
)addr
;
443 } else if (s1
->dynsym
) {
444 /* if dynamic symbol exist, then use it */
445 sym_index
= find_elf_sym(s1
->dynsym
, name
);
447 esym
= &((ElfW(Sym
) *)s1
->dynsym
->data
)[sym_index
];
448 sym
->st_value
= esym
->st_value
;
452 /* XXX: _fp_hw seems to be part of the ABI, so we ignore
454 if (!strcmp(name
, "_fp_hw"))
456 /* only weak symbols are accepted to be undefined. Their
458 sym_bind
= ELFW(ST_BIND
)(sym
->st_info
);
459 if (sym_bind
== STB_WEAK
) {
462 tcc_error_noabort("undefined symbol '%s'", name
);
464 } else if (sh_num
< SHN_LORESERVE
) {
465 /* add section base */
466 sym
->st_value
+= s1
->sections
[sym
->st_shndx
]->sh_addr
;
472 #ifndef TCC_TARGET_PE
473 #ifdef TCC_TARGET_X86_64
474 #define JMP_TABLE_ENTRY_SIZE 14
475 static uplong
add_jmp_table(TCCState
*s1
, uplong val
)
477 char *p
= s1
->runtime_plt_and_got
+ s1
->runtime_plt_and_got_offset
;
478 s1
->runtime_plt_and_got_offset
+= JMP_TABLE_ENTRY_SIZE
;
483 *(uplong
*)(p
+ 6) = val
;
487 static uplong
add_got_table(TCCState
*s1
, uplong val
)
489 uplong
*p
= (uplong
*)(s1
->runtime_plt_and_got
+ s1
->runtime_plt_and_got_offset
);
490 s1
->runtime_plt_and_got_offset
+= sizeof(uplong
);
494 #elif defined TCC_TARGET_ARM
495 #define JMP_TABLE_ENTRY_SIZE 8
496 static uplong
add_jmp_table(TCCState
*s1
, int val
)
498 uint32_t *p
= (uint32_t *)(s1
->runtime_plt_and_got
+ s1
->runtime_plt_and_got_offset
);
499 s1
->runtime_plt_and_got_offset
+= JMP_TABLE_ENTRY_SIZE
;
500 /* ldr pc, [pc, #-4] */
508 /* relocate a given section (CPU dependent) */
509 ST_FUNC
void relocate_section(TCCState
*s1
, Section
*s
)
512 ElfW_Rel
*rel
, *rel_end
, *qrel
;
517 #if defined TCC_TARGET_I386 || defined TCC_TARGET_X86_64
522 rel_end
= (ElfW_Rel
*)(sr
->data
+ sr
->data_offset
);
523 qrel
= (ElfW_Rel
*)sr
->data
;
527 ptr
= s
->data
+ rel
->r_offset
;
529 sym_index
= ELFW(R_SYM
)(rel
->r_info
);
530 sym
= &((ElfW(Sym
) *)symtab_section
->data
)[sym_index
];
532 #ifdef TCC_TARGET_X86_64
533 val
+= rel
->r_addend
;
535 type
= ELFW(R_TYPE
)(rel
->r_info
);
536 addr
= s
->sh_addr
+ rel
->r_offset
;
540 #if defined(TCC_TARGET_I386)
542 if (s1
->output_type
== TCC_OUTPUT_DLL
) {
543 esym_index
= s1
->symtab_to_dynsym
[sym_index
];
544 qrel
->r_offset
= rel
->r_offset
;
546 qrel
->r_info
= ELFW(R_INFO
)(esym_index
, R_386_32
);
550 qrel
->r_info
= ELFW(R_INFO
)(0, R_386_RELATIVE
);
557 if (s1
->output_type
== TCC_OUTPUT_DLL
) {
559 esym_index
= s1
->symtab_to_dynsym
[sym_index
];
561 qrel
->r_offset
= rel
->r_offset
;
562 qrel
->r_info
= ELFW(R_INFO
)(esym_index
, R_386_PC32
);
567 *(int *)ptr
+= val
- addr
;
570 *(int *)ptr
+= val
- addr
;
577 *(int *)ptr
+= s1
->got
->sh_addr
- addr
;
580 *(int *)ptr
+= val
- s1
->got
->sh_addr
;
583 /* we load the got offset */
584 *(int *)ptr
+= s1
->got_offsets
[sym_index
];
587 if (s1
->output_format
!= TCC_OUTPUT_FORMAT_BINARY
) {
589 tcc_error("can only produce 16-bit binary files");
591 *(short *)ptr
+= val
;
594 if (s1
->output_format
!= TCC_OUTPUT_FORMAT_BINARY
)
596 *(short *)ptr
+= val
- addr
;
598 #elif defined(TCC_TARGET_ARM)
605 x
= (*(int *)ptr
)&0xffffff;
606 (*(int *)ptr
) &= 0xff000000;
611 #ifndef TCC_TARGET_PE
612 if((x
& 3) != 0 || x
>= 0x4000000 || x
< -0x4000000)
613 if (s1
->output_type
== TCC_OUTPUT_MEMORY
)
614 x
+= add_jmp_table(s1
, val
) - val
; /* add veneer */
616 if((x
& 3) != 0 || x
>= 0x4000000 || x
< -0x4000000)
617 tcc_error("can't relocate value at %x",addr
);
626 x
= (*(int *)ptr
) & 0x7fffffff;
627 (*(int *)ptr
) &= 0x80000000;
630 if((x
^(x
>>1))&0x40000000)
631 tcc_error("can't relocate value at %x",addr
);
632 (*(int *)ptr
) |= x
& 0x7fffffff;
638 *(int *)ptr
+= val
- addr
;
640 case R_ARM_BASE_PREL
:
641 *(int *)ptr
+= s1
->got
->sh_addr
- addr
;
644 *(int *)ptr
+= val
- s1
->got
->sh_addr
;
647 /* we load the got offset */
648 *(int *)ptr
+= s1
->got_offsets
[sym_index
];
653 /* trade Thumb support for ARMv4 support */
654 if ((0x0ffffff0 & *(int*)ptr
) == 0x012FFF10)
655 *(int*)ptr
^= 0xE12FFF10 ^ 0xE1A0F000; /* BX Rm -> MOV PC, Rm */
658 fprintf(stderr
,"FIXME: handle reloc type %x at %x [%.8x] to %x\n",
659 type
, (unsigned)addr
, (unsigned)(uplong
)ptr
, (unsigned)val
);
661 #elif defined(TCC_TARGET_C67)
669 /* put the low 16 bits of the absolute address */
670 // add to what is already there
672 orig
= ((*(int *)(ptr
)) >> 7) & 0xffff;
673 orig
|= (((*(int *)(ptr
+4)) >> 7) & 0xffff) << 16;
675 //patch both at once - assumes always in pairs Low - High
677 *(int *) ptr
= (*(int *) ptr
& (~(0xffff << 7)) ) | (((val
+orig
) & 0xffff) << 7);
678 *(int *)(ptr
+4) = (*(int *)(ptr
+4) & (~(0xffff << 7)) ) | ((((val
+orig
)>>16) & 0xffff) << 7);
684 fprintf(stderr
,"FIXME: handle reloc type %x at %x [%.8x] to %x\n",
685 type
, (unsigned)addr
, (unsigned)(uplong
)ptr
, (unsigned)val
);
687 #elif defined(TCC_TARGET_X86_64)
689 if (s1
->output_type
== TCC_OUTPUT_DLL
) {
690 qrel
->r_info
= ELFW(R_INFO
)(0, R_X86_64_RELATIVE
);
691 qrel
->r_addend
= *(long long *)ptr
+ val
;
694 *(long long *)ptr
+= val
;
698 if (s1
->output_type
== TCC_OUTPUT_DLL
) {
699 /* XXX: this logic may depend on TCC's codegen
700 now TCC uses R_X86_64_32 even for a 64bit pointer */
701 qrel
->r_info
= ELFW(R_INFO
)(0, R_X86_64_RELATIVE
);
702 qrel
->r_addend
= *(int *)ptr
+ val
;
709 if (s1
->output_type
== TCC_OUTPUT_DLL
) {
711 esym_index
= s1
->symtab_to_dynsym
[sym_index
];
713 qrel
->r_offset
= rel
->r_offset
;
714 qrel
->r_info
= ELFW(R_INFO
)(esym_index
, R_X86_64_PC32
);
715 qrel
->r_addend
= *(int *)ptr
;
721 case R_X86_64_PLT32
: {
723 diff
= (long long)val
- addr
;
724 if (diff
<= -2147483647 || diff
> 2147483647) {
725 #ifndef TCC_TARGET_PE
726 /* XXX: naive support for over 32bit jump */
727 if (s1
->output_type
== TCC_OUTPUT_MEMORY
) {
728 val
= (add_jmp_table(s1
, val
- rel
->r_addend
) +
733 if (diff
<= -2147483647 || diff
> 2147483647) {
734 tcc_error("internal error: relocation failed");
740 case R_X86_64_GLOB_DAT
:
741 case R_X86_64_JUMP_SLOT
:
742 /* They don't need addend */
743 *(int *)ptr
= val
- rel
->r_addend
;
745 case R_X86_64_GOTPCREL
:
746 #ifndef TCC_TARGET_PE
747 if (s1
->output_type
== TCC_OUTPUT_MEMORY
) {
748 val
= add_got_table(s1
, val
- rel
->r_addend
) + rel
->r_addend
;
749 *(int *)ptr
+= val
- addr
;
753 *(int *)ptr
+= (s1
->got
->sh_addr
- addr
+
754 s1
->got_offsets
[sym_index
] - 4);
756 case R_X86_64_GOTTPOFF
:
757 *(int *)ptr
+= val
- s1
->got
->sh_addr
;
760 /* we load the got offset */
761 *(int *)ptr
+= s1
->got_offsets
[sym_index
];
764 #error unsupported processor
768 /* if the relocation is allocated, we change its symbol table */
769 if (sr
->sh_flags
& SHF_ALLOC
)
770 sr
->link
= s1
->dynsym
;
773 /* relocate relocation table in 'sr' */
774 static void relocate_rel(TCCState
*s1
, Section
*sr
)
777 ElfW_Rel
*rel
, *rel_end
;
779 s
= s1
->sections
[sr
->sh_info
];
780 rel_end
= (ElfW_Rel
*)(sr
->data
+ sr
->data_offset
);
781 for(rel
= (ElfW_Rel
*)sr
->data
;
784 rel
->r_offset
+= s
->sh_addr
;
788 /* count the number of dynamic relocations so that we can reserve
790 static int prepare_dynamic_rel(TCCState
*s1
, Section
*sr
)
792 ElfW_Rel
*rel
, *rel_end
;
793 int sym_index
, esym_index
, type
, count
;
796 rel_end
= (ElfW_Rel
*)(sr
->data
+ sr
->data_offset
);
797 for(rel
= (ElfW_Rel
*)sr
->data
; rel
< rel_end
; rel
++) {
798 sym_index
= ELFW(R_SYM
)(rel
->r_info
);
799 type
= ELFW(R_TYPE
)(rel
->r_info
);
801 #if defined(TCC_TARGET_I386)
803 #elif defined(TCC_TARGET_X86_64)
810 #if defined(TCC_TARGET_I386)
812 #elif defined(TCC_TARGET_X86_64)
815 esym_index
= s1
->symtab_to_dynsym
[sym_index
];
824 /* allocate the section */
825 sr
->sh_flags
|= SHF_ALLOC
;
826 sr
->sh_size
= count
* sizeof(ElfW_Rel
);
831 static void put_got_offset(TCCState
*s1
, int index
, unsigned long val
)
836 if (index
>= s1
->nb_got_offsets
) {
837 /* find immediately bigger power of 2 and reallocate array */
841 tab
= tcc_realloc(s1
->got_offsets
, n
* sizeof(unsigned long));
842 s1
->got_offsets
= tab
;
843 memset(s1
->got_offsets
+ s1
->nb_got_offsets
, 0,
844 (n
- s1
->nb_got_offsets
) * sizeof(unsigned long));
845 s1
->nb_got_offsets
= n
;
847 s1
->got_offsets
[index
] = val
;
850 /* XXX: suppress that */
851 static void put32(unsigned char *p
, uint32_t val
)
859 #if defined(TCC_TARGET_I386) || defined(TCC_TARGET_ARM) || \
860 defined(TCC_TARGET_X86_64)
861 static uint32_t get32(unsigned char *p
)
863 return p
[0] | (p
[1] << 8) | (p
[2] << 16) | (p
[3] << 24);
867 static void build_got(TCCState
*s1
)
871 /* if no got, then create it */
872 s1
->got
= new_section(s1
, ".got", SHT_PROGBITS
, SHF_ALLOC
| SHF_WRITE
);
873 s1
->got
->sh_entsize
= 4;
874 add_elf_sym(symtab_section
, 0, 4, ELFW(ST_INFO
)(STB_GLOBAL
, STT_OBJECT
),
875 0, s1
->got
->sh_num
, "_GLOBAL_OFFSET_TABLE_");
876 ptr
= section_ptr_add(s1
->got
, 3 * PTR_SIZE
);
878 /* keep space for _DYNAMIC pointer, if present */
880 /* two dummy got entries */
884 /* keep space for _DYNAMIC pointer, if present */
887 /* two dummy got entries */
895 /* put a got entry corresponding to a symbol in symtab_section. 'size'
896 and 'info' can be modifed if more precise info comes from the DLL */
897 static void put_got_entry(TCCState
*s1
,
898 int reloc_type
, unsigned long size
, int info
,
904 unsigned long offset
;
910 /* if a got entry already exists for that symbol, no need to add one */
911 if (sym_index
< s1
->nb_got_offsets
&&
912 s1
->got_offsets
[sym_index
] != 0)
915 put_got_offset(s1
, sym_index
, s1
->got
->data_offset
);
918 sym
= &((ElfW(Sym
) *)symtab_section
->data
)[sym_index
];
919 name
= symtab_section
->link
->data
+ sym
->st_name
;
920 offset
= sym
->st_value
;
921 #if defined(TCC_TARGET_I386) || defined(TCC_TARGET_X86_64)
923 #ifdef TCC_TARGET_X86_64
933 #if defined(TCC_OUTPUT_DLL_WITH_PLT)
936 /* if we build a DLL, we add a %ebx offset */
937 if (s1
->output_type
== TCC_OUTPUT_DLL
)
943 /* add a PLT entry */
945 if (plt
->data_offset
== 0) {
946 /* first plt entry */
947 p
= section_ptr_add(plt
, 16);
948 p
[0] = 0xff; /* pushl got + PTR_SIZE */
950 put32(p
+ 2, PTR_SIZE
);
951 p
[6] = 0xff; /* jmp *(got + PTR_SIZE * 2) */
953 put32(p
+ 8, PTR_SIZE
* 2);
956 p
= section_ptr_add(plt
, 16);
957 p
[0] = 0xff; /* jmp *(got + x) */
959 put32(p
+ 2, s1
->got
->data_offset
);
960 p
[6] = 0x68; /* push $xxx */
961 put32(p
+ 7, (plt
->data_offset
- 32) >> 1);
962 p
[11] = 0xe9; /* jmp plt_start */
963 put32(p
+ 12, -(plt
->data_offset
));
965 /* the symbol is modified so that it will be relocated to
967 #if !defined(TCC_OUTPUT_DLL_WITH_PLT)
968 if (s1
->output_type
== TCC_OUTPUT_EXE
)
970 offset
= plt
->data_offset
- 16;
972 #elif defined(TCC_TARGET_ARM)
973 if (reloc_type
== R_ARM_JUMP_SLOT
) {
977 /* if we build a DLL, we add a %ebx offset */
978 if (s1
->output_type
== TCC_OUTPUT_DLL
)
979 tcc_error("DLLs unimplemented!");
981 /* add a PLT entry */
983 if (plt
->data_offset
== 0) {
984 /* first plt entry */
985 p
= section_ptr_add(plt
, 16);
986 put32(p
, 0xe52de004);
987 put32(p
+ 4, 0xe59fe010);
988 put32(p
+ 8, 0xe08fe00e);
989 put32(p
+ 12, 0xe5bef008);
992 p
= section_ptr_add(plt
, 16);
993 put32(p
, 0xe59fc004);
994 put32(p
+4, 0xe08fc00c);
995 put32(p
+8, 0xe59cf000);
996 put32(p
+12, s1
->got
->data_offset
);
998 /* the symbol is modified so that it will be relocated to
1000 if (s1
->output_type
== TCC_OUTPUT_EXE
)
1001 offset
= plt
->data_offset
- 16;
1003 #elif defined(TCC_TARGET_C67)
1004 tcc_error("C67 got not implemented");
1006 #error unsupported CPU
1008 index
= put_elf_sym(s1
->dynsym
, offset
,
1009 size
, info
, 0, sym
->st_shndx
, name
);
1010 /* put a got entry */
1011 put_elf_reloc(s1
->dynsym
, s1
->got
,
1012 s1
->got
->data_offset
,
1015 ptr
= section_ptr_add(s1
->got
, PTR_SIZE
);
1019 /* build GOT and PLT entries */
1020 ST_FUNC
void build_got_entries(TCCState
*s1
)
1023 ElfW_Rel
*rel
, *rel_end
;
1025 int i
, type
, reloc_type
, sym_index
;
1027 for(i
= 1; i
< s1
->nb_sections
; i
++) {
1028 s
= s1
->sections
[i
];
1029 if (s
->sh_type
!= SHT_RELX
)
1031 /* no need to handle got relocations */
1032 if (s
->link
!= symtab_section
)
1034 rel_end
= (ElfW_Rel
*)(s
->data
+ s
->data_offset
);
1035 for(rel
= (ElfW_Rel
*)s
->data
;
1038 type
= ELFW(R_TYPE
)(rel
->r_info
);
1040 #if defined(TCC_TARGET_I386)
1047 if (type
== R_386_GOT32
|| type
== R_386_PLT32
) {
1048 sym_index
= ELFW(R_SYM
)(rel
->r_info
);
1049 sym
= &((ElfW(Sym
) *)symtab_section
->data
)[sym_index
];
1050 /* look at the symbol got offset. If none, then add one */
1051 if (type
== R_386_GOT32
)
1052 reloc_type
= R_386_GLOB_DAT
;
1054 reloc_type
= R_386_JMP_SLOT
;
1055 put_got_entry(s1
, reloc_type
, sym
->st_size
, sym
->st_info
,
1059 #elif defined(TCC_TARGET_ARM)
1060 case R_ARM_GOT_BREL
:
1061 case R_ARM_GOTOFF32
:
1062 case R_ARM_BASE_PREL
:
1066 if (type
== R_ARM_GOT_BREL
|| type
== R_ARM_PLT32
) {
1067 sym_index
= ELFW(R_SYM
)(rel
->r_info
);
1068 sym
= &((ElfW(Sym
) *)symtab_section
->data
)[sym_index
];
1069 /* look at the symbol got offset. If none, then add one */
1070 if (type
== R_ARM_GOT_BREL
)
1071 reloc_type
= R_ARM_GLOB_DAT
;
1073 reloc_type
= R_ARM_JUMP_SLOT
;
1074 put_got_entry(s1
, reloc_type
, sym
->st_size
, sym
->st_info
,
1078 #elif defined(TCC_TARGET_C67)
1085 if (type
== R_C60_GOT32
|| type
== R_C60_PLT32
) {
1086 sym_index
= ELFW(R_SYM
)(rel
->r_info
);
1087 sym
= &((ElfW(Sym
) *)symtab_section
->data
)[sym_index
];
1088 /* look at the symbol got offset. If none, then add one */
1089 if (type
== R_C60_GOT32
)
1090 reloc_type
= R_C60_GLOB_DAT
;
1092 reloc_type
= R_C60_JMP_SLOT
;
1093 put_got_entry(s1
, reloc_type
, sym
->st_size
, sym
->st_info
,
1097 #elif defined(TCC_TARGET_X86_64)
1098 case R_X86_64_GOT32
:
1099 case R_X86_64_GOTTPOFF
:
1100 case R_X86_64_GOTPCREL
:
1101 case R_X86_64_PLT32
:
1104 if (type
== R_X86_64_GOT32
|| type
== R_X86_64_GOTPCREL
||
1105 type
== R_X86_64_PLT32
) {
1106 sym_index
= ELFW(R_SYM
)(rel
->r_info
);
1107 sym
= &((ElfW(Sym
) *)symtab_section
->data
)[sym_index
];
1108 /* look at the symbol got offset. If none, then add one */
1109 if (type
== R_X86_64_GOT32
|| type
== R_X86_64_GOTPCREL
)
1110 reloc_type
= R_X86_64_GLOB_DAT
;
1112 reloc_type
= R_X86_64_JUMP_SLOT
;
1113 put_got_entry(s1
, reloc_type
, sym
->st_size
, sym
->st_info
,
1118 #error unsupported CPU
1127 ST_FUNC Section
*new_symtab(TCCState
*s1
,
1128 const char *symtab_name
, int sh_type
, int sh_flags
,
1129 const char *strtab_name
,
1130 const char *hash_name
, int hash_sh_flags
)
1132 Section
*symtab
, *strtab
, *hash
;
1133 int *ptr
, nb_buckets
;
1135 symtab
= new_section(s1
, symtab_name
, sh_type
, sh_flags
);
1136 symtab
->sh_entsize
= sizeof(ElfW(Sym
));
1137 strtab
= new_section(s1
, strtab_name
, SHT_STRTAB
, sh_flags
);
1138 put_elf_str(strtab
, "");
1139 symtab
->link
= strtab
;
1140 put_elf_sym(symtab
, 0, 0, 0, 0, 0, NULL
);
1144 hash
= new_section(s1
, hash_name
, SHT_HASH
, hash_sh_flags
);
1145 hash
->sh_entsize
= sizeof(int);
1146 symtab
->hash
= hash
;
1147 hash
->link
= symtab
;
1149 ptr
= section_ptr_add(hash
, (2 + nb_buckets
+ 1) * sizeof(int));
1150 ptr
[0] = nb_buckets
;
1152 memset(ptr
+ 2, 0, (nb_buckets
+ 1) * sizeof(int));
1156 /* put dynamic tag */
1157 static void put_dt(Section
*dynamic
, int dt
, uplong val
)
1160 dyn
= section_ptr_add(dynamic
, sizeof(ElfW(Dyn
)));
1162 dyn
->d_un
.d_val
= val
;
1165 static void add_init_array_defines(TCCState
*s1
, const char *section_name
)
1169 char sym_start
[1024];
1172 snprintf(sym_start
, sizeof(sym_start
), "__%s_start", section_name
+ 1);
1173 snprintf(sym_end
, sizeof(sym_end
), "__%s_end", section_name
+ 1);
1175 s
= find_section(s1
, section_name
);
1180 end_offset
= s
->data_offset
;
1183 add_elf_sym(symtab_section
,
1185 ELFW(ST_INFO
)(STB_GLOBAL
, STT_NOTYPE
), 0,
1186 s
->sh_num
, sym_start
);
1187 add_elf_sym(symtab_section
,
1189 ELFW(ST_INFO
)(STB_GLOBAL
, STT_NOTYPE
), 0,
1190 s
->sh_num
, sym_end
);
1193 static int tcc_add_support(TCCState
*s1
, const char *filename
)
1196 snprintf(buf
, sizeof(buf
), "%s/%s", s1
->tcc_lib_path
, filename
);
1197 return tcc_add_file(s1
, buf
);
1200 ST_FUNC
void tcc_add_bcheck(TCCState
*s1
)
1202 #ifdef CONFIG_TCC_BCHECK
1204 Section
*init_section
;
1205 unsigned char *pinit
;
1208 if (0 == s1
->do_bounds_check
)
1211 /* XXX: add an object file to do that */
1212 ptr
= section_ptr_add(bounds_section
, sizeof(unsigned long));
1214 add_elf_sym(symtab_section
, 0, 0,
1215 ELFW(ST_INFO
)(STB_GLOBAL
, STT_NOTYPE
), 0,
1216 bounds_section
->sh_num
, "__bounds_start");
1217 /* add bound check code */
1218 #ifndef TCC_TARGET_PE
1219 tcc_add_support(s1
, "bcheck.o");
1221 #ifdef TCC_TARGET_I386
1222 if (s1
->output_type
!= TCC_OUTPUT_MEMORY
) {
1223 /* add 'call __bound_init()' in .init section */
1224 init_section
= find_section(s1
, ".init");
1225 pinit
= section_ptr_add(init_section
, 5);
1227 put32(pinit
+ 1, -4);
1228 sym_index
= find_elf_sym(symtab_section
, "__bound_init");
1229 put_elf_reloc(symtab_section
, init_section
,
1230 init_section
->data_offset
- 4, R_386_PC32
, sym_index
);
1236 /* add tcc runtime libraries */
1237 ST_FUNC
void tcc_add_runtime(TCCState
*s1
)
1242 if (!s1
->nostdlib
) {
1243 tcc_add_library(s1
, "c");
1244 #ifdef CONFIG_USE_LIBGCC
1245 tcc_add_file(s1
, TCC_LIBGCC
);
1246 #elif !defined WITHOUT_LIBTCC
1247 tcc_add_support(s1
, "libtcc1.a");
1249 /* add crt end if not memory output */
1250 if (s1
->output_type
!= TCC_OUTPUT_MEMORY
)
1251 tcc_add_crt(s1
, "crtn.o");
1255 /* add various standard linker symbols (must be done after the
1256 sections are filled (for example after allocating common
1258 ST_FUNC
void tcc_add_linker_symbols(TCCState
*s1
)
1264 add_elf_sym(symtab_section
,
1265 text_section
->data_offset
, 0,
1266 ELFW(ST_INFO
)(STB_GLOBAL
, STT_NOTYPE
), 0,
1267 text_section
->sh_num
, "_etext");
1268 add_elf_sym(symtab_section
,
1269 data_section
->data_offset
, 0,
1270 ELFW(ST_INFO
)(STB_GLOBAL
, STT_NOTYPE
), 0,
1271 data_section
->sh_num
, "_edata");
1272 add_elf_sym(symtab_section
,
1273 bss_section
->data_offset
, 0,
1274 ELFW(ST_INFO
)(STB_GLOBAL
, STT_NOTYPE
), 0,
1275 bss_section
->sh_num
, "_end");
1276 /* horrible new standard ldscript defines */
1277 add_init_array_defines(s1
, ".preinit_array");
1278 add_init_array_defines(s1
, ".init_array");
1279 add_init_array_defines(s1
, ".fini_array");
1281 /* add start and stop symbols for sections whose name can be
1283 for(i
= 1; i
< s1
->nb_sections
; i
++) {
1284 s
= s1
->sections
[i
];
1285 if (s
->sh_type
== SHT_PROGBITS
&&
1286 (s
->sh_flags
& SHF_ALLOC
)) {
1290 /* check if section name can be expressed in C */
1296 if (!isid(ch
) && !isnum(ch
))
1300 snprintf(buf
, sizeof(buf
), "__start_%s", s
->name
);
1301 add_elf_sym(symtab_section
,
1303 ELFW(ST_INFO
)(STB_GLOBAL
, STT_NOTYPE
), 0,
1305 snprintf(buf
, sizeof(buf
), "__stop_%s", s
->name
);
1306 add_elf_sym(symtab_section
,
1308 ELFW(ST_INFO
)(STB_GLOBAL
, STT_NOTYPE
), 0,
1315 static void tcc_output_binary(TCCState
*s1
, FILE *f
,
1316 const int *section_order
)
1319 int i
, offset
, size
;
1322 for(i
=1;i
<s1
->nb_sections
;i
++) {
1323 s
= s1
->sections
[section_order
[i
]];
1324 if (s
->sh_type
!= SHT_NOBITS
&&
1325 (s
->sh_flags
& SHF_ALLOC
)) {
1326 while (offset
< s
->sh_offset
) {
1331 fwrite(s
->data
, 1, size
, f
);
1337 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
1339 #define EXTRA_RELITEMS 14
1341 /* move the relocation value from .dynsym to .got */
1342 void patch_dynsym_undef(TCCState
*s1
, Section
*s
)
1344 uint32_t *gotd
= (void *)s1
->got
->data
;
1345 ElfW(Sym
) *sym
, *sym_end
;
1347 gotd
+= 3; // dummy entries in .got
1348 /* relocate symbols in .dynsym */
1349 sym_end
= (ElfW(Sym
) *)(s
->data
+ s
->data_offset
);
1350 for (sym
= (ElfW(Sym
) *)s
->data
+ 1; sym
< sym_end
; sym
++) {
1351 if (sym
->st_shndx
== SHN_UNDEF
) {
1352 *gotd
++ = sym
->st_value
+ 6; // XXX 6 is magic ?
1359 #define EXTRA_RELITEMS 9
1361 /* zero plt offsets of weak symbols in .dynsym */
1362 void patch_dynsym_undef(TCCState
*s1
, Section
*s
)
1364 ElfW(Sym
) *sym
, *sym_end
;
1366 sym_end
= (ElfW(Sym
) *)(s
->data
+ s
->data_offset
);
1367 for (sym
= (ElfW(Sym
) *)s
->data
+ 1; sym
< sym_end
; sym
++)
1368 if (sym
->st_shndx
== SHN_UNDEF
&& ELFW(ST_BIND
)(sym
->st_info
) == STB_WEAK
)
1373 ST_FUNC
void fill_got_entry(TCCState
*s1
, ElfW_Rel
*rel
)
1375 int sym_index
= ELFW(R_SYM
) (rel
->r_info
);
1376 ElfW(Sym
) *sym
= &((ElfW(Sym
) *) symtab_section
->data
)[sym_index
];
1377 unsigned long offset
;
1379 if (sym_index
>= s1
->nb_got_offsets
)
1381 offset
= s1
->got_offsets
[sym_index
];
1382 section_reserve(s1
->got
, offset
+ PTR_SIZE
);
1383 #ifdef TCC_TARGET_X86_64
1384 /* only works for x86-64 */
1385 put32(s1
->got
->data
+ offset
+ 4, sym
->st_value
>> 32);
1387 put32(s1
->got
->data
+ offset
, sym
->st_value
& 0xffffffff);
1390 ST_FUNC
void fill_got(TCCState
*s1
)
1393 ElfW_Rel
*rel
, *rel_end
;
1396 for(i
= 1; i
< s1
->nb_sections
; i
++) {
1397 s
= s1
->sections
[i
];
1398 if (s
->sh_type
!= SHT_RELX
)
1400 /* no need to handle got relocations */
1401 if (s
->link
!= symtab_section
)
1403 rel_end
= (ElfW_Rel
*) (s
->data
+ s
->data_offset
);
1404 for(rel
= (ElfW_Rel
*) s
->data
; rel
< rel_end
; rel
++) {
1405 switch (ELFW(R_TYPE
) (rel
->r_info
)) {
1406 case R_X86_64_GOT32
:
1407 case R_X86_64_GOTPCREL
:
1408 case R_X86_64_PLT32
:
1409 fill_got_entry(s1
, rel
);
1417 /* output an ELF file */
1418 /* XXX: suppress unneeded sections */
1419 static int elf_output_file(TCCState
*s1
, const char *filename
)
1425 int shnum
, i
, phnum
, file_offset
, offset
, size
, j
, sh_order_index
, k
;
1428 Section
*strsec
, *s
;
1429 ElfW(Shdr
) shdr
, *sh
;
1430 ElfW(Phdr
) *phdr
, *ph
;
1431 Section
*interp
, *dynamic
, *dynstr
;
1432 unsigned long saved_dynamic_data_offset
;
1434 int type
, file_type
;
1435 uplong rel_addr
, rel_size
;
1436 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
1437 uplong bss_addr
, bss_size
;
1440 file_type
= s1
->output_type
;
1443 if (file_type
!= TCC_OUTPUT_OBJ
) {
1444 tcc_add_runtime(s1
);
1448 section_order
= NULL
;
1451 dynstr
= NULL
; /* avoid warning */
1452 saved_dynamic_data_offset
= 0; /* avoid warning */
1454 if (file_type
!= TCC_OUTPUT_OBJ
) {
1455 relocate_common_syms();
1457 tcc_add_linker_symbols(s1
);
1459 if (!s1
->static_link
) {
1461 int sym_index
, index
;
1462 ElfW(Sym
) *esym
, *sym_end
;
1464 if (file_type
== TCC_OUTPUT_EXE
) {
1466 /* allow override the dynamic loader */
1467 const char *elfint
= getenv("LD_SO");
1469 elfint
= CONFIG_TCC_ELFINTERP
;
1470 /* add interpreter section only if executable */
1471 interp
= new_section(s1
, ".interp", SHT_PROGBITS
, SHF_ALLOC
);
1472 interp
->sh_addralign
= 1;
1473 ptr
= section_ptr_add(interp
, 1+strlen(elfint
));
1474 strcpy(ptr
, elfint
);
1477 /* add dynamic symbol table */
1478 s1
->dynsym
= new_symtab(s1
, ".dynsym", SHT_DYNSYM
, SHF_ALLOC
,
1480 ".hash", SHF_ALLOC
);
1481 dynstr
= s1
->dynsym
->link
;
1483 /* add dynamic section */
1484 dynamic
= new_section(s1
, ".dynamic", SHT_DYNAMIC
,
1485 SHF_ALLOC
| SHF_WRITE
);
1486 dynamic
->link
= dynstr
;
1487 dynamic
->sh_entsize
= sizeof(ElfW(Dyn
));
1490 s1
->plt
= new_section(s1
, ".plt", SHT_PROGBITS
,
1491 SHF_ALLOC
| SHF_EXECINSTR
);
1492 s1
->plt
->sh_entsize
= 4;
1496 /* scan for undefined symbols and see if they are in the
1497 dynamic symbols. If a symbol STT_FUNC or STT_GNU_IFUNC
1498 is found, then we add it in the PLT. If a symbol
1499 STT_OBJECT is found, we add it in the .bss section with
1500 a suitable relocation */
1501 sym_end
= (ElfW(Sym
) *)(symtab_section
->data
+
1502 symtab_section
->data_offset
);
1503 if (file_type
== TCC_OUTPUT_EXE
) {
1504 for(sym
= (ElfW(Sym
) *)symtab_section
->data
+ 1;
1507 if (sym
->st_shndx
== SHN_UNDEF
) {
1508 name
= symtab_section
->link
->data
+ sym
->st_name
;
1509 sym_index
= find_elf_sym(s1
->dynsymtab_section
, name
);
1511 esym
= &((ElfW(Sym
) *)s1
->dynsymtab_section
->data
)[sym_index
];
1512 type
= ELFW(ST_TYPE
)(esym
->st_info
);
1513 if ((type
== STT_FUNC
) || (type
== STT_GNU_IFUNC
)) {
1514 put_got_entry(s1
, R_JMP_SLOT
, esym
->st_size
,
1515 ELFW(ST_INFO
)(STB_GLOBAL
,type
),
1516 sym
- (ElfW(Sym
) *)symtab_section
->data
);
1517 } else if (type
== STT_OBJECT
) {
1518 unsigned long offset
;
1519 ElfW(Sym
) *dynsym
, *dynsym_end
;
1520 offset
= bss_section
->data_offset
;
1521 /* XXX: which alignment ? */
1522 offset
= (offset
+ 16 - 1) & -16;
1523 index
= put_elf_sym(s1
->dynsym
, offset
, esym
->st_size
,
1525 bss_section
->sh_num
, name
);
1526 // Ensure R_COPY works for weak symbol aliases
1527 if (ELFW(ST_BIND
)(esym
->st_info
) == STB_WEAK
) {
1528 dynsym_end
= (ElfW(Sym
) *)
1529 (s1
->dynsymtab_section
->data
+
1530 s1
->dynsymtab_section
->data_offset
);
1531 for(dynsym
= (ElfW(Sym
) *)s1
->dynsymtab_section
->data
+ 1;
1532 dynsym
< dynsym_end
; dynsym
++) {
1533 if ((dynsym
->st_value
== esym
->st_value
)
1534 && (ELFW(ST_BIND
)(dynsym
->st_info
) == STB_GLOBAL
)) {
1536 dynname
= s1
->dynsymtab_section
->link
->data
1538 put_elf_sym(s1
->dynsym
, offset
,
1541 bss_section
->sh_num
,
1547 put_elf_reloc(s1
->dynsym
, bss_section
,
1548 offset
, R_COPY
, index
);
1549 offset
+= esym
->st_size
;
1550 bss_section
->data_offset
= offset
;
1553 /* STB_WEAK undefined symbols are accepted */
1554 /* XXX: _fp_hw seems to be part of the ABI, so we ignore
1556 if (ELFW(ST_BIND
)(sym
->st_info
) == STB_WEAK
||
1557 !strcmp(name
, "_fp_hw")) {
1559 tcc_error_noabort("undefined symbol '%s'", name
);
1562 } else if (s1
->rdynamic
&&
1563 ELFW(ST_BIND
)(sym
->st_info
) != STB_LOCAL
) {
1564 /* if -rdynamic option, then export all non
1566 name
= symtab_section
->link
->data
+ sym
->st_name
;
1567 put_elf_sym(s1
->dynsym
, sym
->st_value
, sym
->st_size
,
1569 sym
->st_shndx
, name
);
1576 /* now look at unresolved dynamic symbols and export
1577 corresponding symbol */
1578 sym_end
= (ElfW(Sym
) *)(s1
->dynsymtab_section
->data
+
1579 s1
->dynsymtab_section
->data_offset
);
1580 for(esym
= (ElfW(Sym
) *)s1
->dynsymtab_section
->data
+ 1;
1583 if (esym
->st_shndx
== SHN_UNDEF
) {
1584 name
= s1
->dynsymtab_section
->link
->data
+ esym
->st_name
;
1585 sym_index
= find_elf_sym(symtab_section
, name
);
1587 /* XXX: avoid adding a symbol if already
1588 present because of -rdynamic ? */
1589 sym
= &((ElfW(Sym
) *)symtab_section
->data
)[sym_index
];
1590 put_elf_sym(s1
->dynsym
, sym
->st_value
, sym
->st_size
,
1592 sym
->st_shndx
, name
);
1594 if (ELFW(ST_BIND
)(esym
->st_info
) == STB_WEAK
) {
1595 /* weak symbols can stay undefined */
1597 tcc_warning("undefined dynamic symbol '%s'", name
);
1604 /* shared library case : we simply export all the global symbols */
1605 nb_syms
= symtab_section
->data_offset
/ sizeof(ElfW(Sym
));
1606 s1
->symtab_to_dynsym
= tcc_mallocz(sizeof(int) * nb_syms
);
1607 for(sym
= (ElfW(Sym
) *)symtab_section
->data
+ 1;
1610 if (ELFW(ST_BIND
)(sym
->st_info
) != STB_LOCAL
) {
1611 #if defined(TCC_OUTPUT_DLL_WITH_PLT)
1612 if ((ELFW(ST_TYPE
)(sym
->st_info
) == STT_FUNC
||
1613 ELFW(ST_TYPE
)(sym
->st_info
) == STT_GNU_IFUNC
)
1614 && sym
->st_shndx
== SHN_UNDEF
) {
1615 put_got_entry(s1
, R_JMP_SLOT
, sym
->st_size
,
1617 sym
- (ElfW(Sym
) *)symtab_section
->data
);
1619 else if (ELFW(ST_TYPE
)(sym
->st_info
) == STT_OBJECT
) {
1620 put_got_entry(s1
, R_X86_64_GLOB_DAT
, sym
->st_size
,
1622 sym
- (ElfW(Sym
) *)symtab_section
->data
);
1627 name
= symtab_section
->link
->data
+ sym
->st_name
;
1628 index
= put_elf_sym(s1
->dynsym
, sym
->st_value
, sym
->st_size
,
1630 sym
->st_shndx
, name
);
1631 s1
->symtab_to_dynsym
[sym
-
1632 (ElfW(Sym
) *)symtab_section
->data
] =
1639 build_got_entries(s1
);
1641 /* add a list of needed dlls */
1642 for(i
= 0; i
< s1
->nb_loaded_dlls
; i
++) {
1643 DLLReference
*dllref
= s1
->loaded_dlls
[i
];
1644 if (dllref
->level
== 0)
1645 put_dt(dynamic
, DT_NEEDED
, put_elf_str(dynstr
, dllref
->name
));
1649 put_dt(dynamic
, DT_RPATH
, put_elf_str(dynstr
, s1
->rpath
));
1651 /* XXX: currently, since we do not handle PIC code, we
1652 must relocate the readonly segments */
1653 if (file_type
== TCC_OUTPUT_DLL
) {
1655 put_dt(dynamic
, DT_SONAME
, put_elf_str(dynstr
, s1
->soname
));
1656 put_dt(dynamic
, DT_TEXTREL
, 0);
1660 put_dt(dynamic
, DT_SYMBOLIC
, 0);
1662 /* add necessary space for other entries */
1663 saved_dynamic_data_offset
= dynamic
->data_offset
;
1664 dynamic
->data_offset
+= sizeof(ElfW(Dyn
)) * EXTRA_RELITEMS
;
1666 /* still need to build got entries in case of static link */
1667 build_got_entries(s1
);
1671 memset(&ehdr
, 0, sizeof(ehdr
));
1673 /* we add a section for symbols */
1674 strsec
= new_section(s1
, ".shstrtab", SHT_STRTAB
, 0);
1675 put_elf_str(strsec
, "");
1677 /* compute number of sections */
1678 shnum
= s1
->nb_sections
;
1680 /* this array is used to reorder sections in the output file */
1681 section_order
= tcc_malloc(sizeof(int) * shnum
);
1682 section_order
[0] = 0;
1685 /* compute number of program headers */
1688 case TCC_OUTPUT_OBJ
:
1691 case TCC_OUTPUT_EXE
:
1692 if (!s1
->static_link
)
1693 phnum
= 4 + HAVE_PHDR
;
1697 case TCC_OUTPUT_DLL
:
1702 /* allocate strings for section names and decide if an unallocated
1703 section should be output */
1704 /* NOTE: the strsec section comes last, so its size is also
1706 for(i
= 1; i
< s1
->nb_sections
; i
++) {
1707 s
= s1
->sections
[i
];
1708 s
->sh_name
= put_elf_str(strsec
, s
->name
);
1710 printf("section: f=%08x t=%08x i=%08x %s %s\n",
1715 s
->reloc
? s
->reloc
->name
: "n"
1718 /* when generating a DLL, we include relocations but we may
1720 if (file_type
== TCC_OUTPUT_DLL
&&
1721 s
->sh_type
== SHT_RELX
&&
1722 !(s
->sh_flags
& SHF_ALLOC
)) {
1723 /* //gr: avoid bogus relocs for empty (debug) sections */
1724 if (s1
->sections
[s
->sh_info
]->sh_flags
& SHF_ALLOC
)
1725 prepare_dynamic_rel(s1
, s
);
1726 else if (s1
->do_debug
)
1727 s
->sh_size
= s
->data_offset
;
1728 } else if (s1
->do_debug
||
1729 file_type
== TCC_OUTPUT_OBJ
||
1730 (s
->sh_flags
& SHF_ALLOC
) ||
1731 i
== (s1
->nb_sections
- 1)) {
1732 /* we output all sections if debug or object file */
1733 s
->sh_size
= s
->data_offset
;
1737 /* allocate program segment headers */
1738 phdr
= tcc_mallocz(phnum
* sizeof(ElfW(Phdr
)));
1740 if (s1
->output_format
== TCC_OUTPUT_FORMAT_ELF
) {
1741 file_offset
= sizeof(ElfW(Ehdr
)) + phnum
* sizeof(ElfW(Phdr
));
1746 /* compute section to program header mapping */
1747 if (s1
->has_text_addr
) {
1748 int a_offset
, p_offset
;
1749 addr
= s1
->text_addr
;
1750 /* we ensure that (addr % ELF_PAGE_SIZE) == file_offset %
1752 a_offset
= (int) (addr
& (s1
->section_align
- 1));
1753 p_offset
= file_offset
& (s1
->section_align
- 1);
1754 if (a_offset
< p_offset
)
1755 a_offset
+= s1
->section_align
;
1756 file_offset
+= (a_offset
- p_offset
);
1758 if (file_type
== TCC_OUTPUT_DLL
)
1761 addr
= ELF_START_ADDR
;
1762 /* compute address after headers */
1763 addr
+= (file_offset
& (s1
->section_align
- 1));
1766 /* dynamic relocation table information, for .dynamic section */
1770 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
1771 bss_addr
= bss_size
= 0;
1773 /* leave one program header for the program interpreter */
1776 ph
+= 1 + HAVE_PHDR
;
1778 for(j
= 0; j
< 2; j
++) {
1779 ph
->p_type
= PT_LOAD
;
1781 ph
->p_flags
= PF_R
| PF_X
;
1783 ph
->p_flags
= PF_R
| PF_W
;
1784 ph
->p_align
= s1
->section_align
;
1786 /* we do the following ordering: interp, symbol tables,
1787 relocations, progbits, nobits */
1788 /* XXX: do faster and simpler sorting */
1789 for(k
= 0; k
< 5; k
++) {
1790 for(i
= 1; i
< s1
->nb_sections
; i
++) {
1791 s
= s1
->sections
[i
];
1792 /* compute if section should be included */
1794 if ((s
->sh_flags
& (SHF_ALLOC
| SHF_WRITE
)) !=
1798 if ((s
->sh_flags
& (SHF_ALLOC
| SHF_WRITE
)) !=
1799 (SHF_ALLOC
| SHF_WRITE
))
1805 } else if (s
->sh_type
== SHT_DYNSYM
||
1806 s
->sh_type
== SHT_STRTAB
||
1807 s
->sh_type
== SHT_HASH
) {
1810 } else if (s
->sh_type
== SHT_RELX
) {
1813 } else if (s
->sh_type
== SHT_NOBITS
) {
1820 section_order
[sh_order_index
++] = i
;
1822 /* section matches: we align it and add its size */
1824 addr
= (addr
+ s
->sh_addralign
- 1) &
1825 ~(s
->sh_addralign
- 1);
1826 file_offset
+= (int) ( addr
- tmp
);
1827 s
->sh_offset
= file_offset
;
1830 /* update program header infos */
1831 if (ph
->p_offset
== 0) {
1832 ph
->p_offset
= file_offset
;
1834 ph
->p_paddr
= ph
->p_vaddr
;
1836 /* update dynamic relocation infos */
1837 if (s
->sh_type
== SHT_RELX
) {
1838 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
1839 if (!strcmp(strsec
->data
+ s
->sh_name
, ".rel.got")) { // rel_size == 0) {
1841 rel_size
+= s
->sh_size
; // XXX only first rel.
1843 if (!strcmp(strsec
->data
+ s
->sh_name
, ".rel.bss")) { // rel_size == 0) {
1845 bss_size
= s
->sh_size
; // XXX only first rel.
1850 rel_size
+= s
->sh_size
;
1854 if (s
->sh_type
!= SHT_NOBITS
)
1855 file_offset
+= s
->sh_size
;
1858 ph
->p_filesz
= file_offset
- ph
->p_offset
;
1859 ph
->p_memsz
= addr
- ph
->p_vaddr
;
1862 if (s1
->output_format
== TCC_OUTPUT_FORMAT_ELF
) {
1863 /* if in the middle of a page, we duplicate the page in
1864 memory so that one copy is RX and the other is RW */
1865 if ((addr
& (s1
->section_align
- 1)) != 0)
1866 addr
+= s1
->section_align
;
1868 addr
= (addr
+ s1
->section_align
- 1) & ~(s1
->section_align
- 1);
1869 file_offset
= (file_offset
+ s1
->section_align
- 1) &
1870 ~(s1
->section_align
- 1);
1875 /* if interpreter, then add corresponing program header */
1879 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
1881 int len
= phnum
* sizeof(ElfW(Phdr
));
1883 ph
->p_type
= PT_PHDR
;
1884 ph
->p_offset
= sizeof(ElfW(Ehdr
));
1885 ph
->p_vaddr
= interp
->sh_addr
- len
;
1886 ph
->p_paddr
= ph
->p_vaddr
;
1887 ph
->p_filesz
= ph
->p_memsz
= len
;
1888 ph
->p_flags
= PF_R
| PF_X
;
1889 ph
->p_align
= 4; // interp->sh_addralign;
1894 ph
->p_type
= PT_INTERP
;
1895 ph
->p_offset
= interp
->sh_offset
;
1896 ph
->p_vaddr
= interp
->sh_addr
;
1897 ph
->p_paddr
= ph
->p_vaddr
;
1898 ph
->p_filesz
= interp
->sh_size
;
1899 ph
->p_memsz
= interp
->sh_size
;
1901 ph
->p_align
= interp
->sh_addralign
;
1904 /* if dynamic section, then add corresponing program header */
1908 ph
= &phdr
[phnum
- 1];
1910 ph
->p_type
= PT_DYNAMIC
;
1911 ph
->p_offset
= dynamic
->sh_offset
;
1912 ph
->p_vaddr
= dynamic
->sh_addr
;
1913 ph
->p_paddr
= ph
->p_vaddr
;
1914 ph
->p_filesz
= dynamic
->sh_size
;
1915 ph
->p_memsz
= dynamic
->sh_size
;
1916 ph
->p_flags
= PF_R
| PF_W
;
1917 ph
->p_align
= dynamic
->sh_addralign
;
1919 /* put GOT dynamic section address */
1920 put32(s1
->got
->data
, dynamic
->sh_addr
);
1922 /* relocate the PLT */
1923 if (file_type
== TCC_OUTPUT_EXE
1924 #if defined(TCC_OUTPUT_DLL_WITH_PLT)
1925 || file_type
== TCC_OUTPUT_DLL
1931 p_end
= p
+ s1
->plt
->data_offset
;
1933 #if defined(TCC_TARGET_I386)
1934 put32(p
+ 2, get32(p
+ 2) + s1
->got
->sh_addr
);
1935 put32(p
+ 8, get32(p
+ 8) + s1
->got
->sh_addr
);
1938 put32(p
+ 2, get32(p
+ 2) + s1
->got
->sh_addr
);
1941 #elif defined(TCC_TARGET_X86_64)
1942 int x
= s1
->got
->sh_addr
- s1
->plt
->sh_addr
- 6;
1943 put32(p
+ 2, get32(p
+ 2) + x
);
1944 put32(p
+ 8, get32(p
+ 8) + x
- 6);
1947 put32(p
+ 2, get32(p
+ 2) + x
+ s1
->plt
->data
- p
);
1950 #elif defined(TCC_TARGET_ARM)
1952 x
=s1
->got
->sh_addr
- s1
->plt
->sh_addr
- 12;
1955 put32(p
+ 12, x
+ get32(p
+ 12) + s1
->plt
->data
- p
);
1958 #elif defined(TCC_TARGET_C67)
1961 #error unsupported CPU
1966 /* relocate symbols in .dynsym */
1967 sym_end
= (ElfW(Sym
) *)(s1
->dynsym
->data
+ s1
->dynsym
->data_offset
);
1968 for(sym
= (ElfW(Sym
) *)s1
->dynsym
->data
+ 1;
1971 if (sym
->st_shndx
== SHN_UNDEF
) {
1972 /* relocate to the PLT if the symbol corresponds
1975 sym
->st_value
+= s1
->plt
->sh_addr
;
1976 } else if (sym
->st_shndx
< SHN_LORESERVE
) {
1977 /* do symbol relocation */
1978 sym
->st_value
+= s1
->sections
[sym
->st_shndx
]->sh_addr
;
1982 /* put dynamic section entries */
1983 dynamic
->data_offset
= saved_dynamic_data_offset
;
1984 put_dt(dynamic
, DT_HASH
, s1
->dynsym
->hash
->sh_addr
);
1985 put_dt(dynamic
, DT_STRTAB
, dynstr
->sh_addr
);
1986 put_dt(dynamic
, DT_SYMTAB
, s1
->dynsym
->sh_addr
);
1987 put_dt(dynamic
, DT_STRSZ
, dynstr
->data_offset
);
1988 put_dt(dynamic
, DT_SYMENT
, sizeof(ElfW(Sym
)));
1989 #ifdef TCC_TARGET_X86_64
1990 put_dt(dynamic
, DT_RELA
, rel_addr
);
1991 put_dt(dynamic
, DT_RELASZ
, rel_size
);
1992 put_dt(dynamic
, DT_RELAENT
, sizeof(ElfW_Rel
));
1994 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
1995 put_dt(dynamic
, DT_PLTGOT
, s1
->got
->sh_addr
);
1996 put_dt(dynamic
, DT_PLTRELSZ
, rel_size
);
1997 put_dt(dynamic
, DT_JMPREL
, rel_addr
);
1998 put_dt(dynamic
, DT_PLTREL
, DT_REL
);
1999 put_dt(dynamic
, DT_REL
, bss_addr
);
2000 put_dt(dynamic
, DT_RELSZ
, bss_size
);
2002 put_dt(dynamic
, DT_REL
, rel_addr
);
2003 put_dt(dynamic
, DT_RELSZ
, rel_size
);
2004 put_dt(dynamic
, DT_RELENT
, sizeof(ElfW_Rel
));
2008 put_dt(dynamic
, DT_DEBUG
, 0);
2009 put_dt(dynamic
, DT_NULL
, 0);
2012 ehdr
.e_phentsize
= sizeof(ElfW(Phdr
));
2013 ehdr
.e_phnum
= phnum
;
2014 ehdr
.e_phoff
= sizeof(ElfW(Ehdr
));
2017 /* all other sections come after */
2018 for(i
= 1; i
< s1
->nb_sections
; i
++) {
2019 s
= s1
->sections
[i
];
2020 if (phnum
> 0 && (s
->sh_flags
& SHF_ALLOC
))
2022 section_order
[sh_order_index
++] = i
;
2024 file_offset
= (file_offset
+ s
->sh_addralign
- 1) &
2025 ~(s
->sh_addralign
- 1);
2026 s
->sh_offset
= file_offset
;
2027 if (s
->sh_type
!= SHT_NOBITS
)
2028 file_offset
+= s
->sh_size
;
2031 /* if building executable or DLL, then relocate each section
2032 except the GOT which is already relocated */
2033 if (file_type
!= TCC_OUTPUT_OBJ
) {
2034 relocate_syms(s1
, 0);
2036 if (s1
->nb_errors
!= 0) {
2042 /* relocate sections */
2043 /* XXX: ignore sections with allocated relocations ? */
2044 for(i
= 1; i
< s1
->nb_sections
; i
++) {
2045 s
= s1
->sections
[i
];
2046 if (s
->reloc
&& s
!= s1
->got
&& (s
->sh_flags
& SHF_ALLOC
)) //gr
2047 relocate_section(s1
, s
);
2050 /* relocate relocation entries if the relocation tables are
2051 allocated in the executable */
2052 for(i
= 1; i
< s1
->nb_sections
; i
++) {
2053 s
= s1
->sections
[i
];
2054 if ((s
->sh_flags
& SHF_ALLOC
) &&
2055 s
->sh_type
== SHT_RELX
) {
2056 relocate_rel(s1
, s
);
2060 /* get entry point address */
2061 if (file_type
== TCC_OUTPUT_EXE
)
2062 ehdr
.e_entry
= (uplong
)tcc_get_symbol_err(s1
, "_start");
2064 ehdr
.e_entry
= text_section
->sh_addr
; /* XXX: is it correct ? */
2066 if (file_type
== TCC_OUTPUT_EXE
&& s1
->static_link
)
2069 /* write elf file */
2070 if (file_type
== TCC_OUTPUT_OBJ
)
2075 fd
= open(filename
, O_WRONLY
| O_CREAT
| O_TRUNC
| O_BINARY
, mode
);
2077 tcc_error_noabort("could not write '%s'", filename
);
2080 f
= fdopen(fd
, "wb");
2082 printf("<- %s\n", filename
);
2084 #ifdef TCC_TARGET_COFF
2085 if (s1
->output_format
== TCC_OUTPUT_FORMAT_COFF
) {
2086 tcc_output_coff(s1
, f
);
2089 if (s1
->output_format
== TCC_OUTPUT_FORMAT_ELF
) {
2090 sort_syms(s1
, symtab_section
);
2093 file_offset
= (file_offset
+ 3) & -4;
2096 ehdr
.e_ident
[0] = ELFMAG0
;
2097 ehdr
.e_ident
[1] = ELFMAG1
;
2098 ehdr
.e_ident
[2] = ELFMAG2
;
2099 ehdr
.e_ident
[3] = ELFMAG3
;
2100 ehdr
.e_ident
[4] = ELFCLASSW
;
2101 ehdr
.e_ident
[5] = ELFDATA2LSB
;
2102 ehdr
.e_ident
[6] = EV_CURRENT
;
2103 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
2104 ehdr
.e_ident
[EI_OSABI
] = ELFOSABI_FREEBSD
;
2106 #ifdef TCC_TARGET_ARM
2108 ehdr
.e_ident
[EI_OSABI
] = 0;
2109 ehdr
.e_flags
= 4 << 24;
2111 ehdr
.e_ident
[EI_OSABI
] = ELFOSABI_ARM
;
2116 case TCC_OUTPUT_EXE
:
2117 ehdr
.e_type
= ET_EXEC
;
2119 case TCC_OUTPUT_DLL
:
2120 ehdr
.e_type
= ET_DYN
;
2122 case TCC_OUTPUT_OBJ
:
2123 ehdr
.e_type
= ET_REL
;
2126 ehdr
.e_machine
= EM_TCC_TARGET
;
2127 ehdr
.e_version
= EV_CURRENT
;
2128 ehdr
.e_shoff
= file_offset
;
2129 ehdr
.e_ehsize
= sizeof(ElfW(Ehdr
));
2130 ehdr
.e_shentsize
= sizeof(ElfW(Shdr
));
2131 ehdr
.e_shnum
= shnum
;
2132 ehdr
.e_shstrndx
= shnum
- 1;
2134 fwrite(&ehdr
, 1, sizeof(ElfW(Ehdr
)), f
);
2135 fwrite(phdr
, 1, phnum
* sizeof(ElfW(Phdr
)), f
);
2136 offset
= sizeof(ElfW(Ehdr
)) + phnum
* sizeof(ElfW(Phdr
));
2138 for(i
=1;i
<s1
->nb_sections
;i
++) {
2139 s
= s1
->sections
[section_order
[i
]];
2140 if (s
->sh_type
!= SHT_NOBITS
) {
2141 if (s
->sh_type
== SHT_DYNSYM
)
2142 patch_dynsym_undef(s1
, s
);
2143 while (offset
< s
->sh_offset
) {
2148 fwrite(s
->data
, 1, size
, f
);
2153 /* output section headers */
2154 while (offset
< ehdr
.e_shoff
) {
2159 for(i
=0;i
<s1
->nb_sections
;i
++) {
2161 memset(sh
, 0, sizeof(ElfW(Shdr
)));
2162 s
= s1
->sections
[i
];
2164 sh
->sh_name
= s
->sh_name
;
2165 sh
->sh_type
= s
->sh_type
;
2166 sh
->sh_flags
= s
->sh_flags
;
2167 sh
->sh_entsize
= s
->sh_entsize
;
2168 sh
->sh_info
= s
->sh_info
;
2170 sh
->sh_link
= s
->link
->sh_num
;
2171 sh
->sh_addralign
= s
->sh_addralign
;
2172 sh
->sh_addr
= s
->sh_addr
;
2173 sh
->sh_offset
= s
->sh_offset
;
2174 sh
->sh_size
= s
->sh_size
;
2176 fwrite(sh
, 1, sizeof(ElfW(Shdr
)), f
);
2179 tcc_output_binary(s1
, f
, section_order
);
2185 tcc_free(s1
->symtab_to_dynsym
);
2186 tcc_free(section_order
);
2188 tcc_free(s1
->got_offsets
);
2192 LIBTCCAPI
int tcc_output_file(TCCState
*s
, const char *filename
)
2195 #ifdef TCC_TARGET_PE
2196 if (s
->output_type
!= TCC_OUTPUT_OBJ
) {
2197 ret
= pe_output_file(s
, filename
);
2201 ret
= elf_output_file(s
, filename
);
2206 static void *load_data(int fd
, unsigned long file_offset
, unsigned long size
)
2210 data
= tcc_malloc(size
);
2211 lseek(fd
, file_offset
, SEEK_SET
);
2212 read(fd
, data
, size
);
2216 typedef struct SectionMergeInfo
{
2217 Section
*s
; /* corresponding existing section */
2218 unsigned long offset
; /* offset of the new section in the existing section */
2219 uint8_t new_section
; /* true if section 's' was added */
2220 uint8_t link_once
; /* true if link once section */
2223 /* load an object file and merge it with current files */
2224 /* XXX: handle correctly stab (debug) info */
2225 ST_FUNC
int tcc_load_object_file(TCCState
*s1
,
2226 int fd
, unsigned long file_offset
)
2229 ElfW(Shdr
) *shdr
, *sh
;
2230 int size
, i
, j
, offset
, offseti
, nb_syms
, sym_index
, ret
;
2231 unsigned char *strsec
, *strtab
;
2232 int *old_to_new_syms
;
2233 char *sh_name
, *name
;
2234 SectionMergeInfo
*sm_table
, *sm
;
2235 ElfW(Sym
) *sym
, *symtab
;
2236 ElfW_Rel
*rel
, *rel_end
;
2242 stab_index
= stabstr_index
= 0;
2244 if (read(fd
, &ehdr
, sizeof(ehdr
)) != sizeof(ehdr
))
2246 if (ehdr
.e_ident
[0] != ELFMAG0
||
2247 ehdr
.e_ident
[1] != ELFMAG1
||
2248 ehdr
.e_ident
[2] != ELFMAG2
||
2249 ehdr
.e_ident
[3] != ELFMAG3
)
2251 /* test if object file */
2252 if (ehdr
.e_type
!= ET_REL
)
2254 /* test CPU specific stuff */
2255 if (ehdr
.e_ident
[5] != ELFDATA2LSB
||
2256 ehdr
.e_machine
!= EM_TCC_TARGET
) {
2258 tcc_error_noabort("invalid object file");
2262 shdr
= load_data(fd
, file_offset
+ ehdr
.e_shoff
,
2263 sizeof(ElfW(Shdr
)) * ehdr
.e_shnum
);
2264 sm_table
= tcc_mallocz(sizeof(SectionMergeInfo
) * ehdr
.e_shnum
);
2266 /* load section names */
2267 sh
= &shdr
[ehdr
.e_shstrndx
];
2268 strsec
= load_data(fd
, file_offset
+ sh
->sh_offset
, sh
->sh_size
);
2270 /* load symtab and strtab */
2271 old_to_new_syms
= NULL
;
2275 for(i
= 1; i
< ehdr
.e_shnum
; i
++) {
2277 if (sh
->sh_type
== SHT_SYMTAB
) {
2279 tcc_error_noabort("object must contain only one symtab");
2284 nb_syms
= sh
->sh_size
/ sizeof(ElfW(Sym
));
2285 symtab
= load_data(fd
, file_offset
+ sh
->sh_offset
, sh
->sh_size
);
2286 sm_table
[i
].s
= symtab_section
;
2288 /* now load strtab */
2289 sh
= &shdr
[sh
->sh_link
];
2290 strtab
= load_data(fd
, file_offset
+ sh
->sh_offset
, sh
->sh_size
);
2294 /* now examine each section and try to merge its content with the
2296 for(i
= 1; i
< ehdr
.e_shnum
; i
++) {
2297 /* no need to examine section name strtab */
2298 if (i
== ehdr
.e_shstrndx
)
2301 sh_name
= strsec
+ sh
->sh_name
;
2302 /* ignore sections types we do not handle */
2303 if (sh
->sh_type
!= SHT_PROGBITS
&&
2304 sh
->sh_type
!= SHT_RELX
&&
2306 sh
->sh_type
!= SHT_ARM_EXIDX
&&
2308 sh
->sh_type
!= SHT_NOBITS
&&
2309 sh
->sh_type
!= SHT_PREINIT_ARRAY
&&
2310 sh
->sh_type
!= SHT_INIT_ARRAY
&&
2311 sh
->sh_type
!= SHT_FINI_ARRAY
&&
2312 strcmp(sh_name
, ".stabstr")
2315 if (sh
->sh_addralign
< 1)
2316 sh
->sh_addralign
= 1;
2317 /* find corresponding section, if any */
2318 for(j
= 1; j
< s1
->nb_sections
;j
++) {
2319 s
= s1
->sections
[j
];
2320 if (!strcmp(s
->name
, sh_name
)) {
2321 if (!strncmp(sh_name
, ".gnu.linkonce",
2322 sizeof(".gnu.linkonce") - 1)) {
2323 /* if a 'linkonce' section is already present, we
2324 do not add it again. It is a little tricky as
2325 symbols can still be defined in
2327 sm_table
[i
].link_once
= 1;
2334 /* not found: create new section */
2335 s
= new_section(s1
, sh_name
, sh
->sh_type
, sh
->sh_flags
);
2336 /* take as much info as possible from the section. sh_link and
2337 sh_info will be updated later */
2338 s
->sh_addralign
= sh
->sh_addralign
;
2339 s
->sh_entsize
= sh
->sh_entsize
;
2340 sm_table
[i
].new_section
= 1;
2342 if (sh
->sh_type
!= s
->sh_type
) {
2343 tcc_error_noabort("invalid section type");
2347 /* align start of section */
2348 offset
= s
->data_offset
;
2350 if (0 == strcmp(sh_name
, ".stab")) {
2354 if (0 == strcmp(sh_name
, ".stabstr")) {
2359 size
= sh
->sh_addralign
- 1;
2360 offset
= (offset
+ size
) & ~size
;
2361 if (sh
->sh_addralign
> s
->sh_addralign
)
2362 s
->sh_addralign
= sh
->sh_addralign
;
2363 s
->data_offset
= offset
;
2365 sm_table
[i
].offset
= offset
;
2367 /* concatenate sections */
2369 if (sh
->sh_type
!= SHT_NOBITS
) {
2371 lseek(fd
, file_offset
+ sh
->sh_offset
, SEEK_SET
);
2372 ptr
= section_ptr_add(s
, size
);
2373 read(fd
, ptr
, size
);
2375 s
->data_offset
+= size
;
2380 /* //gr relocate stab strings */
2381 if (stab_index
&& stabstr_index
) {
2384 s
= sm_table
[stab_index
].s
;
2385 a
= (Stab_Sym
*)(s
->data
+ sm_table
[stab_index
].offset
);
2386 b
= (Stab_Sym
*)(s
->data
+ s
->data_offset
);
2387 o
= sm_table
[stabstr_index
].offset
;
2389 a
->n_strx
+= o
, a
++;
2392 /* second short pass to update sh_link and sh_info fields of new
2394 for(i
= 1; i
< ehdr
.e_shnum
; i
++) {
2396 if (!s
|| !sm_table
[i
].new_section
)
2399 if (sh
->sh_link
> 0)
2400 s
->link
= sm_table
[sh
->sh_link
].s
;
2401 if (sh
->sh_type
== SHT_RELX
) {
2402 s
->sh_info
= sm_table
[sh
->sh_info
].s
->sh_num
;
2403 /* update backward link */
2404 s1
->sections
[s
->sh_info
]->reloc
= s
;
2409 /* resolve symbols */
2410 old_to_new_syms
= tcc_mallocz(nb_syms
* sizeof(int));
2413 for(i
= 1; i
< nb_syms
; i
++, sym
++) {
2414 if (sym
->st_shndx
!= SHN_UNDEF
&&
2415 sym
->st_shndx
< SHN_LORESERVE
) {
2416 sm
= &sm_table
[sym
->st_shndx
];
2417 if (sm
->link_once
) {
2418 /* if a symbol is in a link once section, we use the
2419 already defined symbol. It is very important to get
2420 correct relocations */
2421 if (ELFW(ST_BIND
)(sym
->st_info
) != STB_LOCAL
) {
2422 name
= strtab
+ sym
->st_name
;
2423 sym_index
= find_elf_sym(symtab_section
, name
);
2425 old_to_new_syms
[i
] = sym_index
;
2429 /* if no corresponding section added, no need to add symbol */
2432 /* convert section number */
2433 sym
->st_shndx
= sm
->s
->sh_num
;
2435 sym
->st_value
+= sm
->offset
;
2438 name
= strtab
+ sym
->st_name
;
2439 sym_index
= add_elf_sym(symtab_section
, sym
->st_value
, sym
->st_size
,
2440 sym
->st_info
, sym
->st_other
,
2441 sym
->st_shndx
, name
);
2442 old_to_new_syms
[i
] = sym_index
;
2445 /* third pass to patch relocation entries */
2446 for(i
= 1; i
< ehdr
.e_shnum
; i
++) {
2451 offset
= sm_table
[i
].offset
;
2452 switch(s
->sh_type
) {
2454 /* take relocation offset information */
2455 offseti
= sm_table
[sh
->sh_info
].offset
;
2456 rel_end
= (ElfW_Rel
*)(s
->data
+ s
->data_offset
);
2457 for(rel
= (ElfW_Rel
*)(s
->data
+ offset
);
2462 /* convert symbol index */
2463 type
= ELFW(R_TYPE
)(rel
->r_info
);
2464 sym_index
= ELFW(R_SYM
)(rel
->r_info
);
2465 /* NOTE: only one symtab assumed */
2466 if (sym_index
>= nb_syms
)
2468 sym_index
= old_to_new_syms
[sym_index
];
2469 /* ignore link_once in rel section. */
2470 if (!sym_index
&& !sm
->link_once
2471 #ifdef TCC_TARGET_ARM
2472 && type
!= R_ARM_V4BX
2476 tcc_error_noabort("Invalid relocation entry [%2d] '%s' @ %.8x",
2477 i
, strsec
+ sh
->sh_name
, rel
->r_offset
);
2480 rel
->r_info
= ELFW(R_INFO
)(sym_index
, type
);
2481 /* offset the relocation offset */
2482 rel
->r_offset
+= offseti
;
2494 tcc_free(old_to_new_syms
);
2501 typedef struct ArchiveHeader
{
2502 char ar_name
[16]; /* name of this member */
2503 char ar_date
[12]; /* file mtime */
2504 char ar_uid
[6]; /* owner uid; printed as decimal */
2505 char ar_gid
[6]; /* owner gid; printed as decimal */
2506 char ar_mode
[8]; /* file mode, printed as octal */
2507 char ar_size
[10]; /* file size, printed as decimal */
2508 char ar_fmag
[2]; /* should contain ARFMAG */
2511 static int get_be32(const uint8_t *b
)
2513 return b
[3] | (b
[2] << 8) | (b
[1] << 16) | (b
[0] << 24);
2516 /* load only the objects which resolve undefined symbols */
2517 static int tcc_load_alacarte(TCCState
*s1
, int fd
, int size
)
2519 int i
, bound
, nsyms
, sym_index
, off
, ret
;
2521 const char *ar_names
, *p
;
2522 const uint8_t *ar_index
;
2525 data
= tcc_malloc(size
);
2526 if (read(fd
, data
, size
) != size
)
2528 nsyms
= get_be32(data
);
2529 ar_index
= data
+ 4;
2530 ar_names
= ar_index
+ nsyms
* 4;
2534 for(p
= ar_names
, i
= 0; i
< nsyms
; i
++, p
+= strlen(p
)+1) {
2535 sym_index
= find_elf_sym(symtab_section
, p
);
2537 sym
= &((ElfW(Sym
) *)symtab_section
->data
)[sym_index
];
2538 if(sym
->st_shndx
== SHN_UNDEF
) {
2539 off
= get_be32(ar_index
+ i
* 4) + sizeof(ArchiveHeader
);
2541 printf("%5d\t%s\t%08x\n", i
, p
, sym
->st_shndx
);
2544 lseek(fd
, off
, SEEK_SET
);
2545 if(tcc_load_object_file(s1
, fd
, off
) < 0) {
2560 /* load a '.a' file */
2561 ST_FUNC
int tcc_load_archive(TCCState
*s1
, int fd
)
2568 unsigned long file_offset
;
2570 /* skip magic which was already checked */
2571 read(fd
, magic
, sizeof(magic
));
2574 len
= read(fd
, &hdr
, sizeof(hdr
));
2577 if (len
!= sizeof(hdr
)) {
2578 tcc_error_noabort("invalid archive");
2581 memcpy(ar_size
, hdr
.ar_size
, sizeof(hdr
.ar_size
));
2582 ar_size
[sizeof(hdr
.ar_size
)] = '\0';
2583 size
= strtol(ar_size
, NULL
, 0);
2584 memcpy(ar_name
, hdr
.ar_name
, sizeof(hdr
.ar_name
));
2585 for(i
= sizeof(hdr
.ar_name
) - 1; i
>= 0; i
--) {
2586 if (ar_name
[i
] != ' ')
2589 ar_name
[i
+ 1] = '\0';
2590 // printf("name='%s' size=%d %s\n", ar_name, size, ar_size);
2591 file_offset
= lseek(fd
, 0, SEEK_CUR
);
2593 size
= (size
+ 1) & ~1;
2594 if (!strcmp(ar_name
, "/")) {
2595 /* coff symbol table : we handle it */
2596 if(s1
->alacarte_link
)
2597 return tcc_load_alacarte(s1
, fd
, size
);
2598 } else if (!strcmp(ar_name
, "//") ||
2599 !strcmp(ar_name
, "__.SYMDEF") ||
2600 !strcmp(ar_name
, "__.SYMDEF/") ||
2601 !strcmp(ar_name
, "ARFILENAMES/")) {
2602 /* skip symbol table or archive names */
2604 if (tcc_load_object_file(s1
, fd
, file_offset
) < 0)
2607 lseek(fd
, file_offset
+ size
, SEEK_SET
);
2612 #ifndef TCC_TARGET_PE
2613 /* load a DLL and all referenced DLLs. 'level = 0' means that the DLL
2614 is referenced by the user (so it should be added as DT_NEEDED in
2615 the generated ELF file) */
2616 ST_FUNC
int tcc_load_dll(TCCState
*s1
, int fd
, const char *filename
, int level
)
2619 ElfW(Shdr
) *shdr
, *sh
, *sh1
;
2620 int i
, j
, nb_syms
, nb_dts
, sym_bind
, ret
;
2621 ElfW(Sym
) *sym
, *dynsym
;
2622 ElfW(Dyn
) *dt
, *dynamic
;
2623 unsigned char *dynstr
;
2624 const char *name
, *soname
;
2625 DLLReference
*dllref
;
2627 read(fd
, &ehdr
, sizeof(ehdr
));
2629 /* test CPU specific stuff */
2630 if (ehdr
.e_ident
[5] != ELFDATA2LSB
||
2631 ehdr
.e_machine
!= EM_TCC_TARGET
) {
2632 tcc_error_noabort("bad architecture");
2637 shdr
= load_data(fd
, ehdr
.e_shoff
, sizeof(ElfW(Shdr
)) * ehdr
.e_shnum
);
2639 /* load dynamic section and dynamic symbols */
2643 dynsym
= NULL
; /* avoid warning */
2644 dynstr
= NULL
; /* avoid warning */
2645 for(i
= 0, sh
= shdr
; i
< ehdr
.e_shnum
; i
++, sh
++) {
2646 switch(sh
->sh_type
) {
2648 nb_dts
= sh
->sh_size
/ sizeof(ElfW(Dyn
));
2649 dynamic
= load_data(fd
, sh
->sh_offset
, sh
->sh_size
);
2652 nb_syms
= sh
->sh_size
/ sizeof(ElfW(Sym
));
2653 dynsym
= load_data(fd
, sh
->sh_offset
, sh
->sh_size
);
2654 sh1
= &shdr
[sh
->sh_link
];
2655 dynstr
= load_data(fd
, sh1
->sh_offset
, sh1
->sh_size
);
2662 /* compute the real library name */
2663 soname
= tcc_basename(filename
);
2665 for(i
= 0, dt
= dynamic
; i
< nb_dts
; i
++, dt
++) {
2666 if (dt
->d_tag
== DT_SONAME
) {
2667 soname
= dynstr
+ dt
->d_un
.d_val
;
2671 /* if the dll is already loaded, do not load it */
2672 for(i
= 0; i
< s1
->nb_loaded_dlls
; i
++) {
2673 dllref
= s1
->loaded_dlls
[i
];
2674 if (!strcmp(soname
, dllref
->name
)) {
2675 /* but update level if needed */
2676 if (level
< dllref
->level
)
2677 dllref
->level
= level
;
2683 // printf("loading dll '%s'\n", soname);
2685 /* add the dll and its level */
2686 dllref
= tcc_mallocz(sizeof(DLLReference
) + strlen(soname
));
2687 dllref
->level
= level
;
2688 strcpy(dllref
->name
, soname
);
2689 dynarray_add((void ***)&s1
->loaded_dlls
, &s1
->nb_loaded_dlls
, dllref
);
2691 /* add dynamic symbols in dynsym_section */
2692 for(i
= 1, sym
= dynsym
+ 1; i
< nb_syms
; i
++, sym
++) {
2693 sym_bind
= ELFW(ST_BIND
)(sym
->st_info
);
2694 if (sym_bind
== STB_LOCAL
)
2696 name
= dynstr
+ sym
->st_name
;
2697 add_elf_sym(s1
->dynsymtab_section
, sym
->st_value
, sym
->st_size
,
2698 sym
->st_info
, sym
->st_other
, sym
->st_shndx
, name
);
2701 /* load all referenced DLLs */
2702 for(i
= 0, dt
= dynamic
; i
< nb_dts
; i
++, dt
++) {
2705 name
= dynstr
+ dt
->d_un
.d_val
;
2706 for(j
= 0; j
< s1
->nb_loaded_dlls
; j
++) {
2707 dllref
= s1
->loaded_dlls
[j
];
2708 if (!strcmp(name
, dllref
->name
))
2709 goto already_loaded
;
2711 if (tcc_add_dll(s1
, name
, AFF_REFERENCED_DLL
) < 0) {
2712 tcc_error_noabort("referenced dll '%s' not found", name
);
2729 #define LD_TOK_NAME 256
2730 #define LD_TOK_EOF (-1)
2732 /* return next ld script token */
2733 static int ld_next(TCCState
*s1
, char *name
, int name_size
)
2751 file
->buf_ptr
= parse_comment(file
->buf_ptr
);
2752 ch
= file
->buf_ptr
[0];
2760 /* case 'a' ... 'z': */
2787 /* case 'A' ... 'z': */
2822 if (!((ch
>= 'a' && ch
<= 'z') ||
2823 (ch
>= 'A' && ch
<= 'Z') ||
2824 (ch
>= '0' && ch
<= '9') ||
2825 strchr("/.-_+=$:\\,~", ch
)))
2827 if ((q
- name
) < name_size
- 1) {
2844 printf("tok=%c %d\n", c
, c
);
2845 if (c
== LD_TOK_NAME
)
2846 printf(" name=%s\n", name
);
2852 * Extract the file name from the library name
2854 * /!\ No test on filename capacity, be careful
2856 static void libname_to_filename(TCCState
*s1
, const char libname
[], char filename
[])
2858 if (!s1
->static_link
) {
2859 sprintf(filename
, "lib%s.so", libname
);
2861 sprintf(filename
, "lib%s.a", libname
);
2865 static int ld_add_file(TCCState
*s1
, const char filename
[])
2869 ret
= tcc_add_file_internal(s1
, filename
, 0);
2871 ret
= tcc_add_dll(s1
, filename
, 0);
2875 static inline int new_undef_syms(void)
2878 ret
= new_undef_sym
;
2883 static int ld_add_file_list(TCCState
*s1
, const char *cmd
, int as_needed
)
2885 char filename
[1024], libname
[1024];
2886 int t
, group
, nblibs
= 0, ret
= 0;
2889 group
= !strcmp(cmd
, "GROUP");
2892 t
= ld_next(s1
, filename
, sizeof(filename
));
2895 t
= ld_next(s1
, filename
, sizeof(filename
));
2898 if (t
== LD_TOK_EOF
) {
2899 tcc_error_noabort("unexpected end of file");
2901 goto lib_parse_error
;
2902 } else if (t
== ')') {
2904 } else if (t
== '-') {
2905 t
= ld_next(s1
, filename
, sizeof(filename
));
2906 if ((t
!= LD_TOK_NAME
) || (filename
[0] != 'l')) {
2907 tcc_error_noabort("library name expected");
2909 goto lib_parse_error
;
2911 strcpy(libname
, &filename
[1]);
2912 libname_to_filename(s1
, libname
, filename
);
2913 } else if (t
!= LD_TOK_NAME
) {
2914 tcc_error_noabort("filename expected");
2916 goto lib_parse_error
;
2918 if (!strcmp(filename
, "AS_NEEDED")) {
2919 ret
= ld_add_file_list(s1
, cmd
, 1);
2921 goto lib_parse_error
;
2923 /* TODO: Implement AS_NEEDED support. Ignore it for now */
2925 ret
= ld_add_file(s1
, filename
);
2927 goto lib_parse_error
;
2929 /* Add the filename *and* the libname to avoid future conversions */
2930 dynarray_add((void ***) &libs
, &nblibs
, tcc_strdup(filename
));
2931 if (libname
[0] != '\0')
2932 dynarray_add((void ***) &libs
, &nblibs
, tcc_strdup(libname
));
2936 t
= ld_next(s1
, filename
, sizeof(filename
));
2938 t
= ld_next(s1
, filename
, sizeof(filename
));
2941 if (group
&& !as_needed
) {
2942 while (new_undef_syms()) {
2945 for (i
= 0; i
< nblibs
; i
++)
2946 ld_add_file(s1
, libs
[i
]);
2950 dynarray_reset(&libs
, &nblibs
);
2954 /* interpret a subset of GNU ldscripts to handle the dummy libc.so
2956 ST_FUNC
int tcc_load_ldscript(TCCState
*s1
)
2959 char filename
[1024];
2962 ch
= file
->buf_ptr
[0];
2965 t
= ld_next(s1
, cmd
, sizeof(cmd
));
2966 if (t
== LD_TOK_EOF
)
2968 else if (t
!= LD_TOK_NAME
)
2970 if (!strcmp(cmd
, "INPUT") ||
2971 !strcmp(cmd
, "GROUP")) {
2972 ret
= ld_add_file_list(s1
, cmd
, 0);
2975 } else if (!strcmp(cmd
, "OUTPUT_FORMAT") ||
2976 !strcmp(cmd
, "TARGET")) {
2977 /* ignore some commands */
2978 t
= ld_next(s1
, cmd
, sizeof(cmd
));
2982 t
= ld_next(s1
, filename
, sizeof(filename
));
2983 if (t
== LD_TOK_EOF
) {
2984 tcc_error_noabort("unexpected end of file");
2986 } else if (t
== ')') {
2996 #endif /* ndef TCC_TARGET_PE */