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)
604 int x
, is_thumb
, is_call
, h
, blx_avail
;
605 x
= (*(int *)ptr
)&0xffffff;
606 (*(int *)ptr
) &= 0xff000000;
610 blx_avail
= (TCC_ARM_VERSION
>= 5);
612 is_call
= (type
== R_ARM_CALL
);
615 #ifndef TCC_TARGET_PE
616 if ((x
& 3) || x
>= 0x4000000 || x
< -0x4000000)
617 if (!(x
& 3) || !blx_avail
|| !is_call
)
618 if (s1
->output_type
== TCC_OUTPUT_MEMORY
)
619 x
+= add_jmp_table(s1
, val
) - val
; /* add veneer */
621 if ((x
& 3) || x
>= 0x4000000 || x
< -0x4000000)
622 if (!(x
& 3) || !blx_avail
|| !is_call
)
623 tcc_error("can't relocate value at %x",addr
);
626 /* Only reached if blx is avail and it is a call */
629 (*(int *)ptr
) = 0xfa << 24; /* bl -> blx */
635 case R_ARM_MOVW_ABS_NC
:
638 if (type
== R_ARM_MOVT_ABS
)
641 imm4
= (val
>> 12) & 0xf;
642 x
= (imm4
<< 16) | imm12
;
643 if (type
== R_ARM_THM_MOVT_ABS
)
649 case R_ARM_THM_MOVT_ABS
:
650 case R_ARM_THM_MOVW_ABS_NC
:
652 int x
, i
, imm4
, imm3
, imm8
;
653 if (type
== R_ARM_THM_MOVT_ABS
)
656 imm3
= (val
>> 8) & 0x7;
658 imm4
= (val
>> 12) & 0xf;
659 x
= (imm3
<< 28) | (imm8
<< 16) | (i
<< 10) | imm4
;
660 if (type
== R_ARM_THM_MOVT_ABS
)
669 x
= (*(int *)ptr
) & 0x7fffffff;
670 (*(int *)ptr
) &= 0x80000000;
673 if((x
^(x
>>1))&0x40000000)
674 tcc_error("can't relocate value at %x",addr
);
675 (*(int *)ptr
) |= x
& 0x7fffffff;
681 *(int *)ptr
+= val
- addr
;
683 case R_ARM_BASE_PREL
:
684 *(int *)ptr
+= s1
->got
->sh_addr
- addr
;
687 *(int *)ptr
+= val
- s1
->got
->sh_addr
;
690 /* we load the got offset */
691 *(int *)ptr
+= s1
->got_offsets
[sym_index
];
696 /* trade Thumb support for ARMv4 support */
697 if ((0x0ffffff0 & *(int*)ptr
) == 0x012FFF10)
698 *(int*)ptr
^= 0xE12FFF10 ^ 0xE1A0F000; /* BX Rm -> MOV PC, Rm */
701 fprintf(stderr
,"FIXME: handle reloc type %x at %x [%.8x] to %x\n",
702 type
, (unsigned)addr
, (unsigned)(uplong
)ptr
, (unsigned)val
);
704 #elif defined(TCC_TARGET_C67)
712 /* put the low 16 bits of the absolute address */
713 // add to what is already there
715 orig
= ((*(int *)(ptr
)) >> 7) & 0xffff;
716 orig
|= (((*(int *)(ptr
+4)) >> 7) & 0xffff) << 16;
718 //patch both at once - assumes always in pairs Low - High
720 *(int *) ptr
= (*(int *) ptr
& (~(0xffff << 7)) ) | (((val
+orig
) & 0xffff) << 7);
721 *(int *)(ptr
+4) = (*(int *)(ptr
+4) & (~(0xffff << 7)) ) | ((((val
+orig
)>>16) & 0xffff) << 7);
727 fprintf(stderr
,"FIXME: handle reloc type %x at %x [%.8x] to %x\n",
728 type
, (unsigned)addr
, (unsigned)(uplong
)ptr
, (unsigned)val
);
730 #elif defined(TCC_TARGET_X86_64)
732 if (s1
->output_type
== TCC_OUTPUT_DLL
) {
733 qrel
->r_info
= ELFW(R_INFO
)(0, R_X86_64_RELATIVE
);
734 qrel
->r_addend
= *(long long *)ptr
+ val
;
737 *(long long *)ptr
+= val
;
741 if (s1
->output_type
== TCC_OUTPUT_DLL
) {
742 /* XXX: this logic may depend on TCC's codegen
743 now TCC uses R_X86_64_32 even for a 64bit pointer */
744 qrel
->r_info
= ELFW(R_INFO
)(0, R_X86_64_RELATIVE
);
745 qrel
->r_addend
= *(int *)ptr
+ val
;
752 if (s1
->output_type
== TCC_OUTPUT_DLL
) {
754 esym_index
= s1
->symtab_to_dynsym
[sym_index
];
756 qrel
->r_offset
= rel
->r_offset
;
757 qrel
->r_info
= ELFW(R_INFO
)(esym_index
, R_X86_64_PC32
);
758 qrel
->r_addend
= *(int *)ptr
;
764 case R_X86_64_PLT32
: {
766 diff
= (long long)val
- addr
;
767 if (diff
<= -2147483647 || diff
> 2147483647) {
768 #ifndef TCC_TARGET_PE
769 /* XXX: naive support for over 32bit jump */
770 if (s1
->output_type
== TCC_OUTPUT_MEMORY
) {
771 val
= (add_jmp_table(s1
, val
- rel
->r_addend
) +
776 if (diff
<= -2147483647 || diff
> 2147483647) {
777 tcc_error("internal error: relocation failed");
783 case R_X86_64_GLOB_DAT
:
784 case R_X86_64_JUMP_SLOT
:
785 /* They don't need addend */
786 *(int *)ptr
= val
- rel
->r_addend
;
788 case R_X86_64_GOTPCREL
:
789 #ifndef TCC_TARGET_PE
790 if (s1
->output_type
== TCC_OUTPUT_MEMORY
) {
791 val
= add_got_table(s1
, val
- rel
->r_addend
) + rel
->r_addend
;
792 *(int *)ptr
+= val
- addr
;
796 *(int *)ptr
+= (s1
->got
->sh_addr
- addr
+
797 s1
->got_offsets
[sym_index
] - 4);
799 case R_X86_64_GOTTPOFF
:
800 *(int *)ptr
+= val
- s1
->got
->sh_addr
;
803 /* we load the got offset */
804 *(int *)ptr
+= s1
->got_offsets
[sym_index
];
807 #error unsupported processor
811 /* if the relocation is allocated, we change its symbol table */
812 if (sr
->sh_flags
& SHF_ALLOC
)
813 sr
->link
= s1
->dynsym
;
816 /* relocate relocation table in 'sr' */
817 static void relocate_rel(TCCState
*s1
, Section
*sr
)
820 ElfW_Rel
*rel
, *rel_end
;
822 s
= s1
->sections
[sr
->sh_info
];
823 rel_end
= (ElfW_Rel
*)(sr
->data
+ sr
->data_offset
);
824 for(rel
= (ElfW_Rel
*)sr
->data
;
827 rel
->r_offset
+= s
->sh_addr
;
831 /* count the number of dynamic relocations so that we can reserve
833 static int prepare_dynamic_rel(TCCState
*s1
, Section
*sr
)
835 ElfW_Rel
*rel
, *rel_end
;
836 int sym_index
, esym_index
, type
, count
;
839 rel_end
= (ElfW_Rel
*)(sr
->data
+ sr
->data_offset
);
840 for(rel
= (ElfW_Rel
*)sr
->data
; rel
< rel_end
; rel
++) {
841 sym_index
= ELFW(R_SYM
)(rel
->r_info
);
842 type
= ELFW(R_TYPE
)(rel
->r_info
);
844 #if defined(TCC_TARGET_I386)
846 #elif defined(TCC_TARGET_X86_64)
853 #if defined(TCC_TARGET_I386)
855 #elif defined(TCC_TARGET_X86_64)
858 esym_index
= s1
->symtab_to_dynsym
[sym_index
];
867 /* allocate the section */
868 sr
->sh_flags
|= SHF_ALLOC
;
869 sr
->sh_size
= count
* sizeof(ElfW_Rel
);
874 static void put_got_offset(TCCState
*s1
, int index
, unsigned long val
)
879 if (index
>= s1
->nb_got_offsets
) {
880 /* find immediately bigger power of 2 and reallocate array */
884 tab
= tcc_realloc(s1
->got_offsets
, n
* sizeof(unsigned long));
885 s1
->got_offsets
= tab
;
886 memset(s1
->got_offsets
+ s1
->nb_got_offsets
, 0,
887 (n
- s1
->nb_got_offsets
) * sizeof(unsigned long));
888 s1
->nb_got_offsets
= n
;
890 s1
->got_offsets
[index
] = val
;
893 /* XXX: suppress that */
894 static void put32(unsigned char *p
, uint32_t val
)
902 #if defined(TCC_TARGET_I386) || defined(TCC_TARGET_ARM) || \
903 defined(TCC_TARGET_X86_64)
904 static uint32_t get32(unsigned char *p
)
906 return p
[0] | (p
[1] << 8) | (p
[2] << 16) | (p
[3] << 24);
910 static void build_got(TCCState
*s1
)
914 /* if no got, then create it */
915 s1
->got
= new_section(s1
, ".got", SHT_PROGBITS
, SHF_ALLOC
| SHF_WRITE
);
916 s1
->got
->sh_entsize
= 4;
917 add_elf_sym(symtab_section
, 0, 4, ELFW(ST_INFO
)(STB_GLOBAL
, STT_OBJECT
),
918 0, s1
->got
->sh_num
, "_GLOBAL_OFFSET_TABLE_");
919 ptr
= section_ptr_add(s1
->got
, 3 * PTR_SIZE
);
921 /* keep space for _DYNAMIC pointer, if present */
923 /* two dummy got entries */
927 /* keep space for _DYNAMIC pointer, if present */
930 /* two dummy got entries */
938 /* put a got entry corresponding to a symbol in symtab_section. 'size'
939 and 'info' can be modifed if more precise info comes from the DLL */
940 static void put_got_entry(TCCState
*s1
,
941 int reloc_type
, unsigned long size
, int info
,
947 unsigned long offset
;
953 /* if a got entry already exists for that symbol, no need to add one */
954 if (sym_index
< s1
->nb_got_offsets
&&
955 s1
->got_offsets
[sym_index
] != 0)
958 put_got_offset(s1
, sym_index
, s1
->got
->data_offset
);
961 sym
= &((ElfW(Sym
) *)symtab_section
->data
)[sym_index
];
962 name
= symtab_section
->link
->data
+ sym
->st_name
;
963 offset
= sym
->st_value
;
964 #if defined(TCC_TARGET_I386) || defined(TCC_TARGET_X86_64)
966 #ifdef TCC_TARGET_X86_64
976 #if defined(TCC_OUTPUT_DLL_WITH_PLT)
979 /* if we build a DLL, we add a %ebx offset */
980 if (s1
->output_type
== TCC_OUTPUT_DLL
)
986 /* add a PLT entry */
988 if (plt
->data_offset
== 0) {
989 /* first plt entry */
990 p
= section_ptr_add(plt
, 16);
991 p
[0] = 0xff; /* pushl got + PTR_SIZE */
993 put32(p
+ 2, PTR_SIZE
);
994 p
[6] = 0xff; /* jmp *(got + PTR_SIZE * 2) */
996 put32(p
+ 8, PTR_SIZE
* 2);
999 p
= section_ptr_add(plt
, 16);
1000 p
[0] = 0xff; /* jmp *(got + x) */
1002 put32(p
+ 2, s1
->got
->data_offset
);
1003 p
[6] = 0x68; /* push $xxx */
1004 put32(p
+ 7, (plt
->data_offset
- 32) >> 1);
1005 p
[11] = 0xe9; /* jmp plt_start */
1006 put32(p
+ 12, -(plt
->data_offset
));
1008 /* the symbol is modified so that it will be relocated to
1010 #if !defined(TCC_OUTPUT_DLL_WITH_PLT)
1011 if (s1
->output_type
== TCC_OUTPUT_EXE
)
1013 offset
= plt
->data_offset
- 16;
1015 #elif defined(TCC_TARGET_ARM)
1016 if (reloc_type
== R_ARM_JUMP_SLOT
) {
1020 /* if we build a DLL, we add a %ebx offset */
1021 if (s1
->output_type
== TCC_OUTPUT_DLL
)
1022 tcc_error("DLLs unimplemented!");
1024 /* add a PLT entry */
1026 if (plt
->data_offset
== 0) {
1027 /* first plt entry */
1028 p
= section_ptr_add(plt
, 16);
1029 put32(p
, 0xe52de004);
1030 put32(p
+ 4, 0xe59fe010);
1031 put32(p
+ 8, 0xe08fe00e);
1032 put32(p
+ 12, 0xe5bef008);
1035 p
= section_ptr_add(plt
, 16);
1036 put32(p
, 0xe59fc004);
1037 put32(p
+4, 0xe08fc00c);
1038 put32(p
+8, 0xe59cf000);
1039 put32(p
+12, s1
->got
->data_offset
);
1041 /* the symbol is modified so that it will be relocated to
1043 if (s1
->output_type
== TCC_OUTPUT_EXE
)
1044 offset
= plt
->data_offset
- 16;
1046 #elif defined(TCC_TARGET_C67)
1047 tcc_error("C67 got not implemented");
1049 #error unsupported CPU
1051 index
= put_elf_sym(s1
->dynsym
, offset
,
1052 size
, info
, 0, sym
->st_shndx
, name
);
1053 /* put a got entry */
1054 put_elf_reloc(s1
->dynsym
, s1
->got
,
1055 s1
->got
->data_offset
,
1058 ptr
= section_ptr_add(s1
->got
, PTR_SIZE
);
1062 /* build GOT and PLT entries */
1063 ST_FUNC
void build_got_entries(TCCState
*s1
)
1066 ElfW_Rel
*rel
, *rel_end
;
1068 int i
, type
, reloc_type
, sym_index
;
1070 for(i
= 1; i
< s1
->nb_sections
; i
++) {
1071 s
= s1
->sections
[i
];
1072 if (s
->sh_type
!= SHT_RELX
)
1074 /* no need to handle got relocations */
1075 if (s
->link
!= symtab_section
)
1077 rel_end
= (ElfW_Rel
*)(s
->data
+ s
->data_offset
);
1078 for(rel
= (ElfW_Rel
*)s
->data
;
1081 type
= ELFW(R_TYPE
)(rel
->r_info
);
1083 #if defined(TCC_TARGET_I386)
1090 if (type
== R_386_GOT32
|| type
== R_386_PLT32
) {
1091 sym_index
= ELFW(R_SYM
)(rel
->r_info
);
1092 sym
= &((ElfW(Sym
) *)symtab_section
->data
)[sym_index
];
1093 /* look at the symbol got offset. If none, then add one */
1094 if (type
== R_386_GOT32
)
1095 reloc_type
= R_386_GLOB_DAT
;
1097 reloc_type
= R_386_JMP_SLOT
;
1098 put_got_entry(s1
, reloc_type
, sym
->st_size
, sym
->st_info
,
1102 #elif defined(TCC_TARGET_ARM)
1103 case R_ARM_GOT_BREL
:
1104 case R_ARM_GOTOFF32
:
1105 case R_ARM_BASE_PREL
:
1109 if (type
== R_ARM_GOT_BREL
|| type
== R_ARM_PLT32
) {
1110 sym_index
= ELFW(R_SYM
)(rel
->r_info
);
1111 sym
= &((ElfW(Sym
) *)symtab_section
->data
)[sym_index
];
1112 /* look at the symbol got offset. If none, then add one */
1113 if (type
== R_ARM_GOT_BREL
)
1114 reloc_type
= R_ARM_GLOB_DAT
;
1116 reloc_type
= R_ARM_JUMP_SLOT
;
1117 put_got_entry(s1
, reloc_type
, sym
->st_size
, sym
->st_info
,
1121 #elif defined(TCC_TARGET_C67)
1128 if (type
== R_C60_GOT32
|| type
== R_C60_PLT32
) {
1129 sym_index
= ELFW(R_SYM
)(rel
->r_info
);
1130 sym
= &((ElfW(Sym
) *)symtab_section
->data
)[sym_index
];
1131 /* look at the symbol got offset. If none, then add one */
1132 if (type
== R_C60_GOT32
)
1133 reloc_type
= R_C60_GLOB_DAT
;
1135 reloc_type
= R_C60_JMP_SLOT
;
1136 put_got_entry(s1
, reloc_type
, sym
->st_size
, sym
->st_info
,
1140 #elif defined(TCC_TARGET_X86_64)
1141 case R_X86_64_GOT32
:
1142 case R_X86_64_GOTTPOFF
:
1143 case R_X86_64_GOTPCREL
:
1144 case R_X86_64_PLT32
:
1147 if (type
== R_X86_64_GOT32
|| type
== R_X86_64_GOTPCREL
||
1148 type
== R_X86_64_PLT32
) {
1149 sym_index
= ELFW(R_SYM
)(rel
->r_info
);
1150 sym
= &((ElfW(Sym
) *)symtab_section
->data
)[sym_index
];
1151 /* look at the symbol got offset. If none, then add one */
1152 if (type
== R_X86_64_GOT32
|| type
== R_X86_64_GOTPCREL
)
1153 reloc_type
= R_X86_64_GLOB_DAT
;
1155 reloc_type
= R_X86_64_JUMP_SLOT
;
1156 put_got_entry(s1
, reloc_type
, sym
->st_size
, sym
->st_info
,
1161 #error unsupported CPU
1170 ST_FUNC Section
*new_symtab(TCCState
*s1
,
1171 const char *symtab_name
, int sh_type
, int sh_flags
,
1172 const char *strtab_name
,
1173 const char *hash_name
, int hash_sh_flags
)
1175 Section
*symtab
, *strtab
, *hash
;
1176 int *ptr
, nb_buckets
;
1178 symtab
= new_section(s1
, symtab_name
, sh_type
, sh_flags
);
1179 symtab
->sh_entsize
= sizeof(ElfW(Sym
));
1180 strtab
= new_section(s1
, strtab_name
, SHT_STRTAB
, sh_flags
);
1181 put_elf_str(strtab
, "");
1182 symtab
->link
= strtab
;
1183 put_elf_sym(symtab
, 0, 0, 0, 0, 0, NULL
);
1187 hash
= new_section(s1
, hash_name
, SHT_HASH
, hash_sh_flags
);
1188 hash
->sh_entsize
= sizeof(int);
1189 symtab
->hash
= hash
;
1190 hash
->link
= symtab
;
1192 ptr
= section_ptr_add(hash
, (2 + nb_buckets
+ 1) * sizeof(int));
1193 ptr
[0] = nb_buckets
;
1195 memset(ptr
+ 2, 0, (nb_buckets
+ 1) * sizeof(int));
1199 /* put dynamic tag */
1200 static void put_dt(Section
*dynamic
, int dt
, uplong val
)
1203 dyn
= section_ptr_add(dynamic
, sizeof(ElfW(Dyn
)));
1205 dyn
->d_un
.d_val
= val
;
1208 static void add_init_array_defines(TCCState
*s1
, const char *section_name
)
1212 char sym_start
[1024];
1215 snprintf(sym_start
, sizeof(sym_start
), "__%s_start", section_name
+ 1);
1216 snprintf(sym_end
, sizeof(sym_end
), "__%s_end", section_name
+ 1);
1218 s
= find_section(s1
, section_name
);
1223 end_offset
= s
->data_offset
;
1226 add_elf_sym(symtab_section
,
1228 ELFW(ST_INFO
)(STB_GLOBAL
, STT_NOTYPE
), 0,
1229 s
->sh_num
, sym_start
);
1230 add_elf_sym(symtab_section
,
1232 ELFW(ST_INFO
)(STB_GLOBAL
, STT_NOTYPE
), 0,
1233 s
->sh_num
, sym_end
);
1236 static int tcc_add_support(TCCState
*s1
, const char *filename
)
1239 snprintf(buf
, sizeof(buf
), "%s/%s", s1
->tcc_lib_path
, filename
);
1240 return tcc_add_file(s1
, buf
);
1243 ST_FUNC
void tcc_add_bcheck(TCCState
*s1
)
1245 #ifdef CONFIG_TCC_BCHECK
1247 Section
*init_section
;
1248 unsigned char *pinit
;
1251 if (0 == s1
->do_bounds_check
)
1254 /* XXX: add an object file to do that */
1255 ptr
= section_ptr_add(bounds_section
, sizeof(unsigned long));
1257 add_elf_sym(symtab_section
, 0, 0,
1258 ELFW(ST_INFO
)(STB_GLOBAL
, STT_NOTYPE
), 0,
1259 bounds_section
->sh_num
, "__bounds_start");
1260 /* add bound check code */
1261 #ifndef TCC_TARGET_PE
1262 tcc_add_support(s1
, "bcheck.o");
1264 #ifdef TCC_TARGET_I386
1265 if (s1
->output_type
!= TCC_OUTPUT_MEMORY
) {
1266 /* add 'call __bound_init()' in .init section */
1267 init_section
= find_section(s1
, ".init");
1268 pinit
= section_ptr_add(init_section
, 5);
1270 put32(pinit
+ 1, -4);
1271 sym_index
= find_elf_sym(symtab_section
, "__bound_init");
1272 put_elf_reloc(symtab_section
, init_section
,
1273 init_section
->data_offset
- 4, R_386_PC32
, sym_index
);
1279 /* add tcc runtime libraries */
1280 ST_FUNC
void tcc_add_runtime(TCCState
*s1
)
1285 if (!s1
->nostdlib
) {
1286 tcc_add_library(s1
, "c");
1287 #ifdef CONFIG_USE_LIBGCC
1288 tcc_add_file(s1
, TCC_LIBGCC
);
1289 #elif !defined WITHOUT_LIBTCC
1290 tcc_add_support(s1
, "libtcc1.a");
1292 /* add crt end if not memory output */
1293 if (s1
->output_type
!= TCC_OUTPUT_MEMORY
)
1294 tcc_add_crt(s1
, "crtn.o");
1298 /* add various standard linker symbols (must be done after the
1299 sections are filled (for example after allocating common
1301 ST_FUNC
void tcc_add_linker_symbols(TCCState
*s1
)
1307 add_elf_sym(symtab_section
,
1308 text_section
->data_offset
, 0,
1309 ELFW(ST_INFO
)(STB_GLOBAL
, STT_NOTYPE
), 0,
1310 text_section
->sh_num
, "_etext");
1311 add_elf_sym(symtab_section
,
1312 data_section
->data_offset
, 0,
1313 ELFW(ST_INFO
)(STB_GLOBAL
, STT_NOTYPE
), 0,
1314 data_section
->sh_num
, "_edata");
1315 add_elf_sym(symtab_section
,
1316 bss_section
->data_offset
, 0,
1317 ELFW(ST_INFO
)(STB_GLOBAL
, STT_NOTYPE
), 0,
1318 bss_section
->sh_num
, "_end");
1319 /* horrible new standard ldscript defines */
1320 add_init_array_defines(s1
, ".preinit_array");
1321 add_init_array_defines(s1
, ".init_array");
1322 add_init_array_defines(s1
, ".fini_array");
1324 /* add start and stop symbols for sections whose name can be
1326 for(i
= 1; i
< s1
->nb_sections
; i
++) {
1327 s
= s1
->sections
[i
];
1328 if (s
->sh_type
== SHT_PROGBITS
&&
1329 (s
->sh_flags
& SHF_ALLOC
)) {
1333 /* check if section name can be expressed in C */
1339 if (!isid(ch
) && !isnum(ch
))
1343 snprintf(buf
, sizeof(buf
), "__start_%s", s
->name
);
1344 add_elf_sym(symtab_section
,
1346 ELFW(ST_INFO
)(STB_GLOBAL
, STT_NOTYPE
), 0,
1348 snprintf(buf
, sizeof(buf
), "__stop_%s", s
->name
);
1349 add_elf_sym(symtab_section
,
1351 ELFW(ST_INFO
)(STB_GLOBAL
, STT_NOTYPE
), 0,
1358 static void tcc_output_binary(TCCState
*s1
, FILE *f
,
1359 const int *section_order
)
1362 int i
, offset
, size
;
1365 for(i
=1;i
<s1
->nb_sections
;i
++) {
1366 s
= s1
->sections
[section_order
[i
]];
1367 if (s
->sh_type
!= SHT_NOBITS
&&
1368 (s
->sh_flags
& SHF_ALLOC
)) {
1369 while (offset
< s
->sh_offset
) {
1374 fwrite(s
->data
, 1, size
, f
);
1380 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
1382 #define EXTRA_RELITEMS 14
1384 /* move the relocation value from .dynsym to .got */
1385 void patch_dynsym_undef(TCCState
*s1
, Section
*s
)
1387 uint32_t *gotd
= (void *)s1
->got
->data
;
1388 ElfW(Sym
) *sym
, *sym_end
;
1390 gotd
+= 3; // dummy entries in .got
1391 /* relocate symbols in .dynsym */
1392 sym_end
= (ElfW(Sym
) *)(s
->data
+ s
->data_offset
);
1393 for (sym
= (ElfW(Sym
) *)s
->data
+ 1; sym
< sym_end
; sym
++) {
1394 if (sym
->st_shndx
== SHN_UNDEF
) {
1395 *gotd
++ = sym
->st_value
+ 6; // XXX 6 is magic ?
1402 #define EXTRA_RELITEMS 9
1404 /* zero plt offsets of weak symbols in .dynsym */
1405 void patch_dynsym_undef(TCCState
*s1
, Section
*s
)
1407 ElfW(Sym
) *sym
, *sym_end
;
1409 sym_end
= (ElfW(Sym
) *)(s
->data
+ s
->data_offset
);
1410 for (sym
= (ElfW(Sym
) *)s
->data
+ 1; sym
< sym_end
; sym
++)
1411 if (sym
->st_shndx
== SHN_UNDEF
&& ELFW(ST_BIND
)(sym
->st_info
) == STB_WEAK
)
1416 ST_FUNC
void fill_got_entry(TCCState
*s1
, ElfW_Rel
*rel
)
1418 int sym_index
= ELFW(R_SYM
) (rel
->r_info
);
1419 ElfW(Sym
) *sym
= &((ElfW(Sym
) *) symtab_section
->data
)[sym_index
];
1420 unsigned long offset
;
1422 if (sym_index
>= s1
->nb_got_offsets
)
1424 offset
= s1
->got_offsets
[sym_index
];
1425 section_reserve(s1
->got
, offset
+ PTR_SIZE
);
1426 #ifdef TCC_TARGET_X86_64
1427 /* only works for x86-64 */
1428 put32(s1
->got
->data
+ offset
+ 4, sym
->st_value
>> 32);
1430 put32(s1
->got
->data
+ offset
, sym
->st_value
& 0xffffffff);
1433 ST_FUNC
void fill_got(TCCState
*s1
)
1436 ElfW_Rel
*rel
, *rel_end
;
1439 for(i
= 1; i
< s1
->nb_sections
; i
++) {
1440 s
= s1
->sections
[i
];
1441 if (s
->sh_type
!= SHT_RELX
)
1443 /* no need to handle got relocations */
1444 if (s
->link
!= symtab_section
)
1446 rel_end
= (ElfW_Rel
*) (s
->data
+ s
->data_offset
);
1447 for(rel
= (ElfW_Rel
*) s
->data
; rel
< rel_end
; rel
++) {
1448 switch (ELFW(R_TYPE
) (rel
->r_info
)) {
1449 case R_X86_64_GOT32
:
1450 case R_X86_64_GOTPCREL
:
1451 case R_X86_64_PLT32
:
1452 fill_got_entry(s1
, rel
);
1460 /* output an ELF file */
1461 /* XXX: suppress unneeded sections */
1462 static int elf_output_file(TCCState
*s1
, const char *filename
)
1468 int shnum
, i
, phnum
, file_offset
, offset
, size
, j
, sh_order_index
, k
;
1471 Section
*strsec
, *s
;
1472 ElfW(Shdr
) shdr
, *sh
;
1473 ElfW(Phdr
) *phdr
, *ph
;
1474 Section
*interp
, *dynamic
, *dynstr
;
1475 unsigned long saved_dynamic_data_offset
;
1477 int type
, file_type
;
1478 uplong rel_addr
, rel_size
;
1479 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
1480 uplong bss_addr
, bss_size
;
1483 file_type
= s1
->output_type
;
1486 if (file_type
!= TCC_OUTPUT_OBJ
) {
1487 tcc_add_runtime(s1
);
1491 section_order
= NULL
;
1494 dynstr
= NULL
; /* avoid warning */
1495 saved_dynamic_data_offset
= 0; /* avoid warning */
1497 if (file_type
!= TCC_OUTPUT_OBJ
) {
1498 relocate_common_syms();
1500 tcc_add_linker_symbols(s1
);
1502 if (!s1
->static_link
) {
1504 int sym_index
, index
;
1505 ElfW(Sym
) *esym
, *sym_end
;
1507 if (file_type
== TCC_OUTPUT_EXE
) {
1509 /* allow override the dynamic loader */
1510 const char *elfint
= getenv("LD_SO");
1512 elfint
= CONFIG_TCC_ELFINTERP
;
1513 /* add interpreter section only if executable */
1514 interp
= new_section(s1
, ".interp", SHT_PROGBITS
, SHF_ALLOC
);
1515 interp
->sh_addralign
= 1;
1516 ptr
= section_ptr_add(interp
, 1+strlen(elfint
));
1517 strcpy(ptr
, elfint
);
1520 /* add dynamic symbol table */
1521 s1
->dynsym
= new_symtab(s1
, ".dynsym", SHT_DYNSYM
, SHF_ALLOC
,
1523 ".hash", SHF_ALLOC
);
1524 dynstr
= s1
->dynsym
->link
;
1526 /* add dynamic section */
1527 dynamic
= new_section(s1
, ".dynamic", SHT_DYNAMIC
,
1528 SHF_ALLOC
| SHF_WRITE
);
1529 dynamic
->link
= dynstr
;
1530 dynamic
->sh_entsize
= sizeof(ElfW(Dyn
));
1533 s1
->plt
= new_section(s1
, ".plt", SHT_PROGBITS
,
1534 SHF_ALLOC
| SHF_EXECINSTR
);
1535 s1
->plt
->sh_entsize
= 4;
1539 /* scan for undefined symbols and see if they are in the
1540 dynamic symbols. If a symbol STT_FUNC or STT_GNU_IFUNC
1541 is found, then we add it in the PLT. If a symbol
1542 STT_OBJECT is found, we add it in the .bss section with
1543 a suitable relocation */
1544 sym_end
= (ElfW(Sym
) *)(symtab_section
->data
+
1545 symtab_section
->data_offset
);
1546 if (file_type
== TCC_OUTPUT_EXE
) {
1547 for(sym
= (ElfW(Sym
) *)symtab_section
->data
+ 1;
1550 if (sym
->st_shndx
== SHN_UNDEF
) {
1551 name
= symtab_section
->link
->data
+ sym
->st_name
;
1552 sym_index
= find_elf_sym(s1
->dynsymtab_section
, name
);
1554 esym
= &((ElfW(Sym
) *)s1
->dynsymtab_section
->data
)[sym_index
];
1555 type
= ELFW(ST_TYPE
)(esym
->st_info
);
1556 if ((type
== STT_FUNC
) || (type
== STT_GNU_IFUNC
)) {
1557 put_got_entry(s1
, R_JMP_SLOT
, esym
->st_size
,
1558 ELFW(ST_INFO
)(STB_GLOBAL
,type
),
1559 sym
- (ElfW(Sym
) *)symtab_section
->data
);
1560 } else if (type
== STT_OBJECT
) {
1561 unsigned long offset
;
1562 ElfW(Sym
) *dynsym
, *dynsym_end
;
1563 offset
= bss_section
->data_offset
;
1564 /* XXX: which alignment ? */
1565 offset
= (offset
+ 16 - 1) & -16;
1566 index
= put_elf_sym(s1
->dynsym
, offset
, esym
->st_size
,
1568 bss_section
->sh_num
, name
);
1569 // Ensure R_COPY works for weak symbol aliases
1570 if (ELFW(ST_BIND
)(esym
->st_info
) == STB_WEAK
) {
1571 dynsym_end
= (ElfW(Sym
) *)
1572 (s1
->dynsymtab_section
->data
+
1573 s1
->dynsymtab_section
->data_offset
);
1574 for(dynsym
= (ElfW(Sym
) *)s1
->dynsymtab_section
->data
+ 1;
1575 dynsym
< dynsym_end
; dynsym
++) {
1576 if ((dynsym
->st_value
== esym
->st_value
)
1577 && (ELFW(ST_BIND
)(dynsym
->st_info
) == STB_GLOBAL
)) {
1579 dynname
= s1
->dynsymtab_section
->link
->data
1581 put_elf_sym(s1
->dynsym
, offset
,
1584 bss_section
->sh_num
,
1590 put_elf_reloc(s1
->dynsym
, bss_section
,
1591 offset
, R_COPY
, index
);
1592 offset
+= esym
->st_size
;
1593 bss_section
->data_offset
= offset
;
1596 /* STB_WEAK undefined symbols are accepted */
1597 /* XXX: _fp_hw seems to be part of the ABI, so we ignore
1599 if (ELFW(ST_BIND
)(sym
->st_info
) == STB_WEAK
||
1600 !strcmp(name
, "_fp_hw")) {
1602 tcc_error_noabort("undefined symbol '%s'", name
);
1605 } else if (s1
->rdynamic
&&
1606 ELFW(ST_BIND
)(sym
->st_info
) != STB_LOCAL
) {
1607 /* if -rdynamic option, then export all non
1609 name
= symtab_section
->link
->data
+ sym
->st_name
;
1610 put_elf_sym(s1
->dynsym
, sym
->st_value
, sym
->st_size
,
1612 sym
->st_shndx
, name
);
1619 /* now look at unresolved dynamic symbols and export
1620 corresponding symbol */
1621 sym_end
= (ElfW(Sym
) *)(s1
->dynsymtab_section
->data
+
1622 s1
->dynsymtab_section
->data_offset
);
1623 for(esym
= (ElfW(Sym
) *)s1
->dynsymtab_section
->data
+ 1;
1626 if (esym
->st_shndx
== SHN_UNDEF
) {
1627 name
= s1
->dynsymtab_section
->link
->data
+ esym
->st_name
;
1628 sym_index
= find_elf_sym(symtab_section
, name
);
1630 /* XXX: avoid adding a symbol if already
1631 present because of -rdynamic ? */
1632 sym
= &((ElfW(Sym
) *)symtab_section
->data
)[sym_index
];
1633 put_elf_sym(s1
->dynsym
, sym
->st_value
, sym
->st_size
,
1635 sym
->st_shndx
, name
);
1637 if (ELFW(ST_BIND
)(esym
->st_info
) == STB_WEAK
) {
1638 /* weak symbols can stay undefined */
1640 tcc_warning("undefined dynamic symbol '%s'", name
);
1647 /* shared library case : we simply export all the global symbols */
1648 nb_syms
= symtab_section
->data_offset
/ sizeof(ElfW(Sym
));
1649 s1
->symtab_to_dynsym
= tcc_mallocz(sizeof(int) * nb_syms
);
1650 for(sym
= (ElfW(Sym
) *)symtab_section
->data
+ 1;
1653 if (ELFW(ST_BIND
)(sym
->st_info
) != STB_LOCAL
) {
1654 #if defined(TCC_OUTPUT_DLL_WITH_PLT)
1655 if ((ELFW(ST_TYPE
)(sym
->st_info
) == STT_FUNC
||
1656 ELFW(ST_TYPE
)(sym
->st_info
) == STT_GNU_IFUNC
)
1657 && sym
->st_shndx
== SHN_UNDEF
) {
1658 put_got_entry(s1
, R_JMP_SLOT
, sym
->st_size
,
1660 sym
- (ElfW(Sym
) *)symtab_section
->data
);
1662 else if (ELFW(ST_TYPE
)(sym
->st_info
) == STT_OBJECT
) {
1663 put_got_entry(s1
, R_X86_64_GLOB_DAT
, sym
->st_size
,
1665 sym
- (ElfW(Sym
) *)symtab_section
->data
);
1670 name
= symtab_section
->link
->data
+ sym
->st_name
;
1671 index
= put_elf_sym(s1
->dynsym
, sym
->st_value
, sym
->st_size
,
1673 sym
->st_shndx
, name
);
1674 s1
->symtab_to_dynsym
[sym
-
1675 (ElfW(Sym
) *)symtab_section
->data
] =
1682 build_got_entries(s1
);
1684 /* add a list of needed dlls */
1685 for(i
= 0; i
< s1
->nb_loaded_dlls
; i
++) {
1686 DLLReference
*dllref
= s1
->loaded_dlls
[i
];
1687 if (dllref
->level
== 0)
1688 put_dt(dynamic
, DT_NEEDED
, put_elf_str(dynstr
, dllref
->name
));
1692 put_dt(dynamic
, DT_RPATH
, put_elf_str(dynstr
, s1
->rpath
));
1694 /* XXX: currently, since we do not handle PIC code, we
1695 must relocate the readonly segments */
1696 if (file_type
== TCC_OUTPUT_DLL
) {
1698 put_dt(dynamic
, DT_SONAME
, put_elf_str(dynstr
, s1
->soname
));
1699 put_dt(dynamic
, DT_TEXTREL
, 0);
1703 put_dt(dynamic
, DT_SYMBOLIC
, 0);
1705 /* add necessary space for other entries */
1706 saved_dynamic_data_offset
= dynamic
->data_offset
;
1707 dynamic
->data_offset
+= sizeof(ElfW(Dyn
)) * EXTRA_RELITEMS
;
1709 /* still need to build got entries in case of static link */
1710 build_got_entries(s1
);
1714 memset(&ehdr
, 0, sizeof(ehdr
));
1716 /* we add a section for symbols */
1717 strsec
= new_section(s1
, ".shstrtab", SHT_STRTAB
, 0);
1718 put_elf_str(strsec
, "");
1720 /* compute number of sections */
1721 shnum
= s1
->nb_sections
;
1723 /* this array is used to reorder sections in the output file */
1724 section_order
= tcc_malloc(sizeof(int) * shnum
);
1725 section_order
[0] = 0;
1728 /* compute number of program headers */
1731 case TCC_OUTPUT_OBJ
:
1734 case TCC_OUTPUT_EXE
:
1735 if (!s1
->static_link
)
1736 phnum
= 4 + HAVE_PHDR
;
1740 case TCC_OUTPUT_DLL
:
1745 /* allocate strings for section names and decide if an unallocated
1746 section should be output */
1747 /* NOTE: the strsec section comes last, so its size is also
1749 for(i
= 1; i
< s1
->nb_sections
; i
++) {
1750 s
= s1
->sections
[i
];
1751 s
->sh_name
= put_elf_str(strsec
, s
->name
);
1753 printf("section: f=%08x t=%08x i=%08x %s %s\n",
1758 s
->reloc
? s
->reloc
->name
: "n"
1761 /* when generating a DLL, we include relocations but we may
1763 if (file_type
== TCC_OUTPUT_DLL
&&
1764 s
->sh_type
== SHT_RELX
&&
1765 !(s
->sh_flags
& SHF_ALLOC
)) {
1766 /* //gr: avoid bogus relocs for empty (debug) sections */
1767 if (s1
->sections
[s
->sh_info
]->sh_flags
& SHF_ALLOC
)
1768 prepare_dynamic_rel(s1
, s
);
1769 else if (s1
->do_debug
)
1770 s
->sh_size
= s
->data_offset
;
1771 } else if (s1
->do_debug
||
1772 file_type
== TCC_OUTPUT_OBJ
||
1773 (s
->sh_flags
& SHF_ALLOC
) ||
1774 i
== (s1
->nb_sections
- 1)) {
1775 /* we output all sections if debug or object file */
1776 s
->sh_size
= s
->data_offset
;
1780 /* allocate program segment headers */
1781 phdr
= tcc_mallocz(phnum
* sizeof(ElfW(Phdr
)));
1783 if (s1
->output_format
== TCC_OUTPUT_FORMAT_ELF
) {
1784 file_offset
= sizeof(ElfW(Ehdr
)) + phnum
* sizeof(ElfW(Phdr
));
1789 /* compute section to program header mapping */
1790 if (s1
->has_text_addr
) {
1791 int a_offset
, p_offset
;
1792 addr
= s1
->text_addr
;
1793 /* we ensure that (addr % ELF_PAGE_SIZE) == file_offset %
1795 a_offset
= (int) (addr
& (s1
->section_align
- 1));
1796 p_offset
= file_offset
& (s1
->section_align
- 1);
1797 if (a_offset
< p_offset
)
1798 a_offset
+= s1
->section_align
;
1799 file_offset
+= (a_offset
- p_offset
);
1801 if (file_type
== TCC_OUTPUT_DLL
)
1804 addr
= ELF_START_ADDR
;
1805 /* compute address after headers */
1806 addr
+= (file_offset
& (s1
->section_align
- 1));
1809 /* dynamic relocation table information, for .dynamic section */
1813 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
1814 bss_addr
= bss_size
= 0;
1816 /* leave one program header for the program interpreter */
1819 ph
+= 1 + HAVE_PHDR
;
1821 for(j
= 0; j
< 2; j
++) {
1822 ph
->p_type
= PT_LOAD
;
1824 ph
->p_flags
= PF_R
| PF_X
;
1826 ph
->p_flags
= PF_R
| PF_W
;
1827 ph
->p_align
= s1
->section_align
;
1829 /* we do the following ordering: interp, symbol tables,
1830 relocations, progbits, nobits */
1831 /* XXX: do faster and simpler sorting */
1832 for(k
= 0; k
< 5; k
++) {
1833 for(i
= 1; i
< s1
->nb_sections
; i
++) {
1834 s
= s1
->sections
[i
];
1835 /* compute if section should be included */
1837 if ((s
->sh_flags
& (SHF_ALLOC
| SHF_WRITE
)) !=
1841 if ((s
->sh_flags
& (SHF_ALLOC
| SHF_WRITE
)) !=
1842 (SHF_ALLOC
| SHF_WRITE
))
1848 } else if (s
->sh_type
== SHT_DYNSYM
||
1849 s
->sh_type
== SHT_STRTAB
||
1850 s
->sh_type
== SHT_HASH
) {
1853 } else if (s
->sh_type
== SHT_RELX
) {
1856 } else if (s
->sh_type
== SHT_NOBITS
) {
1863 section_order
[sh_order_index
++] = i
;
1865 /* section matches: we align it and add its size */
1867 addr
= (addr
+ s
->sh_addralign
- 1) &
1868 ~(s
->sh_addralign
- 1);
1869 file_offset
+= (int) ( addr
- tmp
);
1870 s
->sh_offset
= file_offset
;
1873 /* update program header infos */
1874 if (ph
->p_offset
== 0) {
1875 ph
->p_offset
= file_offset
;
1877 ph
->p_paddr
= ph
->p_vaddr
;
1879 /* update dynamic relocation infos */
1880 if (s
->sh_type
== SHT_RELX
) {
1881 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
1882 if (!strcmp(strsec
->data
+ s
->sh_name
, ".rel.got")) { // rel_size == 0) {
1884 rel_size
+= s
->sh_size
; // XXX only first rel.
1886 if (!strcmp(strsec
->data
+ s
->sh_name
, ".rel.bss")) { // rel_size == 0) {
1888 bss_size
= s
->sh_size
; // XXX only first rel.
1893 rel_size
+= s
->sh_size
;
1897 if (s
->sh_type
!= SHT_NOBITS
)
1898 file_offset
+= s
->sh_size
;
1901 ph
->p_filesz
= file_offset
- ph
->p_offset
;
1902 ph
->p_memsz
= addr
- ph
->p_vaddr
;
1905 if (s1
->output_format
== TCC_OUTPUT_FORMAT_ELF
) {
1906 /* if in the middle of a page, we duplicate the page in
1907 memory so that one copy is RX and the other is RW */
1908 if ((addr
& (s1
->section_align
- 1)) != 0)
1909 addr
+= s1
->section_align
;
1911 addr
= (addr
+ s1
->section_align
- 1) & ~(s1
->section_align
- 1);
1912 file_offset
= (file_offset
+ s1
->section_align
- 1) &
1913 ~(s1
->section_align
- 1);
1918 /* if interpreter, then add corresponing program header */
1922 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
1924 int len
= phnum
* sizeof(ElfW(Phdr
));
1926 ph
->p_type
= PT_PHDR
;
1927 ph
->p_offset
= sizeof(ElfW(Ehdr
));
1928 ph
->p_vaddr
= interp
->sh_addr
- len
;
1929 ph
->p_paddr
= ph
->p_vaddr
;
1930 ph
->p_filesz
= ph
->p_memsz
= len
;
1931 ph
->p_flags
= PF_R
| PF_X
;
1932 ph
->p_align
= 4; // interp->sh_addralign;
1937 ph
->p_type
= PT_INTERP
;
1938 ph
->p_offset
= interp
->sh_offset
;
1939 ph
->p_vaddr
= interp
->sh_addr
;
1940 ph
->p_paddr
= ph
->p_vaddr
;
1941 ph
->p_filesz
= interp
->sh_size
;
1942 ph
->p_memsz
= interp
->sh_size
;
1944 ph
->p_align
= interp
->sh_addralign
;
1947 /* if dynamic section, then add corresponing program header */
1951 ph
= &phdr
[phnum
- 1];
1953 ph
->p_type
= PT_DYNAMIC
;
1954 ph
->p_offset
= dynamic
->sh_offset
;
1955 ph
->p_vaddr
= dynamic
->sh_addr
;
1956 ph
->p_paddr
= ph
->p_vaddr
;
1957 ph
->p_filesz
= dynamic
->sh_size
;
1958 ph
->p_memsz
= dynamic
->sh_size
;
1959 ph
->p_flags
= PF_R
| PF_W
;
1960 ph
->p_align
= dynamic
->sh_addralign
;
1962 /* put GOT dynamic section address */
1963 put32(s1
->got
->data
, dynamic
->sh_addr
);
1965 /* relocate the PLT */
1966 if (file_type
== TCC_OUTPUT_EXE
1967 #if defined(TCC_OUTPUT_DLL_WITH_PLT)
1968 || file_type
== TCC_OUTPUT_DLL
1974 p_end
= p
+ s1
->plt
->data_offset
;
1976 #if defined(TCC_TARGET_I386)
1977 put32(p
+ 2, get32(p
+ 2) + s1
->got
->sh_addr
);
1978 put32(p
+ 8, get32(p
+ 8) + s1
->got
->sh_addr
);
1981 put32(p
+ 2, get32(p
+ 2) + s1
->got
->sh_addr
);
1984 #elif defined(TCC_TARGET_X86_64)
1985 int x
= s1
->got
->sh_addr
- s1
->plt
->sh_addr
- 6;
1986 put32(p
+ 2, get32(p
+ 2) + x
);
1987 put32(p
+ 8, get32(p
+ 8) + x
- 6);
1990 put32(p
+ 2, get32(p
+ 2) + x
+ s1
->plt
->data
- p
);
1993 #elif defined(TCC_TARGET_ARM)
1995 x
=s1
->got
->sh_addr
- s1
->plt
->sh_addr
- 12;
1998 put32(p
+ 12, x
+ get32(p
+ 12) + s1
->plt
->data
- p
);
2001 #elif defined(TCC_TARGET_C67)
2004 #error unsupported CPU
2009 /* relocate symbols in .dynsym */
2010 sym_end
= (ElfW(Sym
) *)(s1
->dynsym
->data
+ s1
->dynsym
->data_offset
);
2011 for(sym
= (ElfW(Sym
) *)s1
->dynsym
->data
+ 1;
2014 if (sym
->st_shndx
== SHN_UNDEF
) {
2015 /* relocate to the PLT if the symbol corresponds
2018 sym
->st_value
+= s1
->plt
->sh_addr
;
2019 } else if (sym
->st_shndx
< SHN_LORESERVE
) {
2020 /* do symbol relocation */
2021 sym
->st_value
+= s1
->sections
[sym
->st_shndx
]->sh_addr
;
2025 /* put dynamic section entries */
2026 dynamic
->data_offset
= saved_dynamic_data_offset
;
2027 put_dt(dynamic
, DT_HASH
, s1
->dynsym
->hash
->sh_addr
);
2028 put_dt(dynamic
, DT_STRTAB
, dynstr
->sh_addr
);
2029 put_dt(dynamic
, DT_SYMTAB
, s1
->dynsym
->sh_addr
);
2030 put_dt(dynamic
, DT_STRSZ
, dynstr
->data_offset
);
2031 put_dt(dynamic
, DT_SYMENT
, sizeof(ElfW(Sym
)));
2032 #ifdef TCC_TARGET_X86_64
2033 put_dt(dynamic
, DT_RELA
, rel_addr
);
2034 put_dt(dynamic
, DT_RELASZ
, rel_size
);
2035 put_dt(dynamic
, DT_RELAENT
, sizeof(ElfW_Rel
));
2037 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
2038 put_dt(dynamic
, DT_PLTGOT
, s1
->got
->sh_addr
);
2039 put_dt(dynamic
, DT_PLTRELSZ
, rel_size
);
2040 put_dt(dynamic
, DT_JMPREL
, rel_addr
);
2041 put_dt(dynamic
, DT_PLTREL
, DT_REL
);
2042 put_dt(dynamic
, DT_REL
, bss_addr
);
2043 put_dt(dynamic
, DT_RELSZ
, bss_size
);
2045 put_dt(dynamic
, DT_REL
, rel_addr
);
2046 put_dt(dynamic
, DT_RELSZ
, rel_size
);
2047 put_dt(dynamic
, DT_RELENT
, sizeof(ElfW_Rel
));
2051 put_dt(dynamic
, DT_DEBUG
, 0);
2052 put_dt(dynamic
, DT_NULL
, 0);
2055 ehdr
.e_phentsize
= sizeof(ElfW(Phdr
));
2056 ehdr
.e_phnum
= phnum
;
2057 ehdr
.e_phoff
= sizeof(ElfW(Ehdr
));
2060 /* all other sections come after */
2061 for(i
= 1; i
< s1
->nb_sections
; i
++) {
2062 s
= s1
->sections
[i
];
2063 if (phnum
> 0 && (s
->sh_flags
& SHF_ALLOC
))
2065 section_order
[sh_order_index
++] = i
;
2067 file_offset
= (file_offset
+ s
->sh_addralign
- 1) &
2068 ~(s
->sh_addralign
- 1);
2069 s
->sh_offset
= file_offset
;
2070 if (s
->sh_type
!= SHT_NOBITS
)
2071 file_offset
+= s
->sh_size
;
2074 /* if building executable or DLL, then relocate each section
2075 except the GOT which is already relocated */
2076 if (file_type
!= TCC_OUTPUT_OBJ
) {
2077 relocate_syms(s1
, 0);
2079 if (s1
->nb_errors
!= 0) {
2085 /* relocate sections */
2086 /* XXX: ignore sections with allocated relocations ? */
2087 for(i
= 1; i
< s1
->nb_sections
; i
++) {
2088 s
= s1
->sections
[i
];
2089 if (s
->reloc
&& s
!= s1
->got
&& (s
->sh_flags
& SHF_ALLOC
)) //gr
2090 relocate_section(s1
, s
);
2093 /* relocate relocation entries if the relocation tables are
2094 allocated in the executable */
2095 for(i
= 1; i
< s1
->nb_sections
; i
++) {
2096 s
= s1
->sections
[i
];
2097 if ((s
->sh_flags
& SHF_ALLOC
) &&
2098 s
->sh_type
== SHT_RELX
) {
2099 relocate_rel(s1
, s
);
2103 /* get entry point address */
2104 if (file_type
== TCC_OUTPUT_EXE
)
2105 ehdr
.e_entry
= (uplong
)tcc_get_symbol_err(s1
, "_start");
2107 ehdr
.e_entry
= text_section
->sh_addr
; /* XXX: is it correct ? */
2109 if (file_type
== TCC_OUTPUT_EXE
&& s1
->static_link
)
2112 /* write elf file */
2113 if (file_type
== TCC_OUTPUT_OBJ
)
2118 fd
= open(filename
, O_WRONLY
| O_CREAT
| O_TRUNC
| O_BINARY
, mode
);
2120 tcc_error_noabort("could not write '%s'", filename
);
2123 f
= fdopen(fd
, "wb");
2125 printf("<- %s\n", filename
);
2127 #ifdef TCC_TARGET_COFF
2128 if (s1
->output_format
== TCC_OUTPUT_FORMAT_COFF
) {
2129 tcc_output_coff(s1
, f
);
2132 if (s1
->output_format
== TCC_OUTPUT_FORMAT_ELF
) {
2133 sort_syms(s1
, symtab_section
);
2136 file_offset
= (file_offset
+ 3) & -4;
2139 ehdr
.e_ident
[0] = ELFMAG0
;
2140 ehdr
.e_ident
[1] = ELFMAG1
;
2141 ehdr
.e_ident
[2] = ELFMAG2
;
2142 ehdr
.e_ident
[3] = ELFMAG3
;
2143 ehdr
.e_ident
[4] = ELFCLASSW
;
2144 ehdr
.e_ident
[5] = ELFDATA2LSB
;
2145 ehdr
.e_ident
[6] = EV_CURRENT
;
2146 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
2147 ehdr
.e_ident
[EI_OSABI
] = ELFOSABI_FREEBSD
;
2149 #ifdef TCC_TARGET_ARM
2151 ehdr
.e_ident
[EI_OSABI
] = 0;
2152 ehdr
.e_flags
= 4 << 24;
2154 ehdr
.e_ident
[EI_OSABI
] = ELFOSABI_ARM
;
2159 case TCC_OUTPUT_EXE
:
2160 ehdr
.e_type
= ET_EXEC
;
2162 case TCC_OUTPUT_DLL
:
2163 ehdr
.e_type
= ET_DYN
;
2165 case TCC_OUTPUT_OBJ
:
2166 ehdr
.e_type
= ET_REL
;
2169 ehdr
.e_machine
= EM_TCC_TARGET
;
2170 ehdr
.e_version
= EV_CURRENT
;
2171 ehdr
.e_shoff
= file_offset
;
2172 ehdr
.e_ehsize
= sizeof(ElfW(Ehdr
));
2173 ehdr
.e_shentsize
= sizeof(ElfW(Shdr
));
2174 ehdr
.e_shnum
= shnum
;
2175 ehdr
.e_shstrndx
= shnum
- 1;
2177 fwrite(&ehdr
, 1, sizeof(ElfW(Ehdr
)), f
);
2178 fwrite(phdr
, 1, phnum
* sizeof(ElfW(Phdr
)), f
);
2179 offset
= sizeof(ElfW(Ehdr
)) + phnum
* sizeof(ElfW(Phdr
));
2181 for(i
=1;i
<s1
->nb_sections
;i
++) {
2182 s
= s1
->sections
[section_order
[i
]];
2183 if (s
->sh_type
!= SHT_NOBITS
) {
2184 if (s
->sh_type
== SHT_DYNSYM
)
2185 patch_dynsym_undef(s1
, s
);
2186 while (offset
< s
->sh_offset
) {
2191 fwrite(s
->data
, 1, size
, f
);
2196 /* output section headers */
2197 while (offset
< ehdr
.e_shoff
) {
2202 for(i
=0;i
<s1
->nb_sections
;i
++) {
2204 memset(sh
, 0, sizeof(ElfW(Shdr
)));
2205 s
= s1
->sections
[i
];
2207 sh
->sh_name
= s
->sh_name
;
2208 sh
->sh_type
= s
->sh_type
;
2209 sh
->sh_flags
= s
->sh_flags
;
2210 sh
->sh_entsize
= s
->sh_entsize
;
2211 sh
->sh_info
= s
->sh_info
;
2213 sh
->sh_link
= s
->link
->sh_num
;
2214 sh
->sh_addralign
= s
->sh_addralign
;
2215 sh
->sh_addr
= s
->sh_addr
;
2216 sh
->sh_offset
= s
->sh_offset
;
2217 sh
->sh_size
= s
->sh_size
;
2219 fwrite(sh
, 1, sizeof(ElfW(Shdr
)), f
);
2222 tcc_output_binary(s1
, f
, section_order
);
2228 tcc_free(s1
->symtab_to_dynsym
);
2229 tcc_free(section_order
);
2231 tcc_free(s1
->got_offsets
);
2235 LIBTCCAPI
int tcc_output_file(TCCState
*s
, const char *filename
)
2238 #ifdef TCC_TARGET_PE
2239 if (s
->output_type
!= TCC_OUTPUT_OBJ
) {
2240 ret
= pe_output_file(s
, filename
);
2244 ret
= elf_output_file(s
, filename
);
2249 static void *load_data(int fd
, unsigned long file_offset
, unsigned long size
)
2253 data
= tcc_malloc(size
);
2254 lseek(fd
, file_offset
, SEEK_SET
);
2255 read(fd
, data
, size
);
2259 typedef struct SectionMergeInfo
{
2260 Section
*s
; /* corresponding existing section */
2261 unsigned long offset
; /* offset of the new section in the existing section */
2262 uint8_t new_section
; /* true if section 's' was added */
2263 uint8_t link_once
; /* true if link once section */
2266 /* load an object file and merge it with current files */
2267 /* XXX: handle correctly stab (debug) info */
2268 ST_FUNC
int tcc_load_object_file(TCCState
*s1
,
2269 int fd
, unsigned long file_offset
)
2272 ElfW(Shdr
) *shdr
, *sh
;
2273 int size
, i
, j
, offset
, offseti
, nb_syms
, sym_index
, ret
;
2274 unsigned char *strsec
, *strtab
;
2275 int *old_to_new_syms
;
2276 char *sh_name
, *name
;
2277 SectionMergeInfo
*sm_table
, *sm
;
2278 ElfW(Sym
) *sym
, *symtab
;
2279 ElfW_Rel
*rel
, *rel_end
;
2285 stab_index
= stabstr_index
= 0;
2287 if (read(fd
, &ehdr
, sizeof(ehdr
)) != sizeof(ehdr
))
2289 if (ehdr
.e_ident
[0] != ELFMAG0
||
2290 ehdr
.e_ident
[1] != ELFMAG1
||
2291 ehdr
.e_ident
[2] != ELFMAG2
||
2292 ehdr
.e_ident
[3] != ELFMAG3
)
2294 /* test if object file */
2295 if (ehdr
.e_type
!= ET_REL
)
2297 /* test CPU specific stuff */
2298 if (ehdr
.e_ident
[5] != ELFDATA2LSB
||
2299 ehdr
.e_machine
!= EM_TCC_TARGET
) {
2301 tcc_error_noabort("invalid object file");
2305 shdr
= load_data(fd
, file_offset
+ ehdr
.e_shoff
,
2306 sizeof(ElfW(Shdr
)) * ehdr
.e_shnum
);
2307 sm_table
= tcc_mallocz(sizeof(SectionMergeInfo
) * ehdr
.e_shnum
);
2309 /* load section names */
2310 sh
= &shdr
[ehdr
.e_shstrndx
];
2311 strsec
= load_data(fd
, file_offset
+ sh
->sh_offset
, sh
->sh_size
);
2313 /* load symtab and strtab */
2314 old_to_new_syms
= NULL
;
2318 for(i
= 1; i
< ehdr
.e_shnum
; i
++) {
2320 if (sh
->sh_type
== SHT_SYMTAB
) {
2322 tcc_error_noabort("object must contain only one symtab");
2327 nb_syms
= sh
->sh_size
/ sizeof(ElfW(Sym
));
2328 symtab
= load_data(fd
, file_offset
+ sh
->sh_offset
, sh
->sh_size
);
2329 sm_table
[i
].s
= symtab_section
;
2331 /* now load strtab */
2332 sh
= &shdr
[sh
->sh_link
];
2333 strtab
= load_data(fd
, file_offset
+ sh
->sh_offset
, sh
->sh_size
);
2337 /* now examine each section and try to merge its content with the
2339 for(i
= 1; i
< ehdr
.e_shnum
; i
++) {
2340 /* no need to examine section name strtab */
2341 if (i
== ehdr
.e_shstrndx
)
2344 sh_name
= strsec
+ sh
->sh_name
;
2345 /* ignore sections types we do not handle */
2346 if (sh
->sh_type
!= SHT_PROGBITS
&&
2347 sh
->sh_type
!= SHT_RELX
&&
2349 sh
->sh_type
!= SHT_ARM_EXIDX
&&
2351 sh
->sh_type
!= SHT_NOBITS
&&
2352 sh
->sh_type
!= SHT_PREINIT_ARRAY
&&
2353 sh
->sh_type
!= SHT_INIT_ARRAY
&&
2354 sh
->sh_type
!= SHT_FINI_ARRAY
&&
2355 strcmp(sh_name
, ".stabstr")
2358 if (sh
->sh_addralign
< 1)
2359 sh
->sh_addralign
= 1;
2360 /* find corresponding section, if any */
2361 for(j
= 1; j
< s1
->nb_sections
;j
++) {
2362 s
= s1
->sections
[j
];
2363 if (!strcmp(s
->name
, sh_name
)) {
2364 if (!strncmp(sh_name
, ".gnu.linkonce",
2365 sizeof(".gnu.linkonce") - 1)) {
2366 /* if a 'linkonce' section is already present, we
2367 do not add it again. It is a little tricky as
2368 symbols can still be defined in
2370 sm_table
[i
].link_once
= 1;
2377 /* not found: create new section */
2378 s
= new_section(s1
, sh_name
, sh
->sh_type
, sh
->sh_flags
);
2379 /* take as much info as possible from the section. sh_link and
2380 sh_info will be updated later */
2381 s
->sh_addralign
= sh
->sh_addralign
;
2382 s
->sh_entsize
= sh
->sh_entsize
;
2383 sm_table
[i
].new_section
= 1;
2385 if (sh
->sh_type
!= s
->sh_type
) {
2386 tcc_error_noabort("invalid section type");
2390 /* align start of section */
2391 offset
= s
->data_offset
;
2393 if (0 == strcmp(sh_name
, ".stab")) {
2397 if (0 == strcmp(sh_name
, ".stabstr")) {
2402 size
= sh
->sh_addralign
- 1;
2403 offset
= (offset
+ size
) & ~size
;
2404 if (sh
->sh_addralign
> s
->sh_addralign
)
2405 s
->sh_addralign
= sh
->sh_addralign
;
2406 s
->data_offset
= offset
;
2408 sm_table
[i
].offset
= offset
;
2410 /* concatenate sections */
2412 if (sh
->sh_type
!= SHT_NOBITS
) {
2414 lseek(fd
, file_offset
+ sh
->sh_offset
, SEEK_SET
);
2415 ptr
= section_ptr_add(s
, size
);
2416 read(fd
, ptr
, size
);
2418 s
->data_offset
+= size
;
2423 /* //gr relocate stab strings */
2424 if (stab_index
&& stabstr_index
) {
2427 s
= sm_table
[stab_index
].s
;
2428 a
= (Stab_Sym
*)(s
->data
+ sm_table
[stab_index
].offset
);
2429 b
= (Stab_Sym
*)(s
->data
+ s
->data_offset
);
2430 o
= sm_table
[stabstr_index
].offset
;
2432 a
->n_strx
+= o
, a
++;
2435 /* second short pass to update sh_link and sh_info fields of new
2437 for(i
= 1; i
< ehdr
.e_shnum
; i
++) {
2439 if (!s
|| !sm_table
[i
].new_section
)
2442 if (sh
->sh_link
> 0)
2443 s
->link
= sm_table
[sh
->sh_link
].s
;
2444 if (sh
->sh_type
== SHT_RELX
) {
2445 s
->sh_info
= sm_table
[sh
->sh_info
].s
->sh_num
;
2446 /* update backward link */
2447 s1
->sections
[s
->sh_info
]->reloc
= s
;
2452 /* resolve symbols */
2453 old_to_new_syms
= tcc_mallocz(nb_syms
* sizeof(int));
2456 for(i
= 1; i
< nb_syms
; i
++, sym
++) {
2457 if (sym
->st_shndx
!= SHN_UNDEF
&&
2458 sym
->st_shndx
< SHN_LORESERVE
) {
2459 sm
= &sm_table
[sym
->st_shndx
];
2460 if (sm
->link_once
) {
2461 /* if a symbol is in a link once section, we use the
2462 already defined symbol. It is very important to get
2463 correct relocations */
2464 if (ELFW(ST_BIND
)(sym
->st_info
) != STB_LOCAL
) {
2465 name
= strtab
+ sym
->st_name
;
2466 sym_index
= find_elf_sym(symtab_section
, name
);
2468 old_to_new_syms
[i
] = sym_index
;
2472 /* if no corresponding section added, no need to add symbol */
2475 /* convert section number */
2476 sym
->st_shndx
= sm
->s
->sh_num
;
2478 sym
->st_value
+= sm
->offset
;
2481 name
= strtab
+ sym
->st_name
;
2482 sym_index
= add_elf_sym(symtab_section
, sym
->st_value
, sym
->st_size
,
2483 sym
->st_info
, sym
->st_other
,
2484 sym
->st_shndx
, name
);
2485 old_to_new_syms
[i
] = sym_index
;
2488 /* third pass to patch relocation entries */
2489 for(i
= 1; i
< ehdr
.e_shnum
; i
++) {
2494 offset
= sm_table
[i
].offset
;
2495 switch(s
->sh_type
) {
2497 /* take relocation offset information */
2498 offseti
= sm_table
[sh
->sh_info
].offset
;
2499 rel_end
= (ElfW_Rel
*)(s
->data
+ s
->data_offset
);
2500 for(rel
= (ElfW_Rel
*)(s
->data
+ offset
);
2505 /* convert symbol index */
2506 type
= ELFW(R_TYPE
)(rel
->r_info
);
2507 sym_index
= ELFW(R_SYM
)(rel
->r_info
);
2508 /* NOTE: only one symtab assumed */
2509 if (sym_index
>= nb_syms
)
2511 sym_index
= old_to_new_syms
[sym_index
];
2512 /* ignore link_once in rel section. */
2513 if (!sym_index
&& !sm
->link_once
2514 #ifdef TCC_TARGET_ARM
2515 && type
!= R_ARM_V4BX
2519 tcc_error_noabort("Invalid relocation entry [%2d] '%s' @ %.8x",
2520 i
, strsec
+ sh
->sh_name
, rel
->r_offset
);
2523 rel
->r_info
= ELFW(R_INFO
)(sym_index
, type
);
2524 /* offset the relocation offset */
2525 rel
->r_offset
+= offseti
;
2537 tcc_free(old_to_new_syms
);
2544 typedef struct ArchiveHeader
{
2545 char ar_name
[16]; /* name of this member */
2546 char ar_date
[12]; /* file mtime */
2547 char ar_uid
[6]; /* owner uid; printed as decimal */
2548 char ar_gid
[6]; /* owner gid; printed as decimal */
2549 char ar_mode
[8]; /* file mode, printed as octal */
2550 char ar_size
[10]; /* file size, printed as decimal */
2551 char ar_fmag
[2]; /* should contain ARFMAG */
2554 static int get_be32(const uint8_t *b
)
2556 return b
[3] | (b
[2] << 8) | (b
[1] << 16) | (b
[0] << 24);
2559 /* load only the objects which resolve undefined symbols */
2560 static int tcc_load_alacarte(TCCState
*s1
, int fd
, int size
)
2562 int i
, bound
, nsyms
, sym_index
, off
, ret
;
2564 const char *ar_names
, *p
;
2565 const uint8_t *ar_index
;
2568 data
= tcc_malloc(size
);
2569 if (read(fd
, data
, size
) != size
)
2571 nsyms
= get_be32(data
);
2572 ar_index
= data
+ 4;
2573 ar_names
= ar_index
+ nsyms
* 4;
2577 for(p
= ar_names
, i
= 0; i
< nsyms
; i
++, p
+= strlen(p
)+1) {
2578 sym_index
= find_elf_sym(symtab_section
, p
);
2580 sym
= &((ElfW(Sym
) *)symtab_section
->data
)[sym_index
];
2581 if(sym
->st_shndx
== SHN_UNDEF
) {
2582 off
= get_be32(ar_index
+ i
* 4) + sizeof(ArchiveHeader
);
2584 printf("%5d\t%s\t%08x\n", i
, p
, sym
->st_shndx
);
2587 lseek(fd
, off
, SEEK_SET
);
2588 if(tcc_load_object_file(s1
, fd
, off
) < 0) {
2603 /* load a '.a' file */
2604 ST_FUNC
int tcc_load_archive(TCCState
*s1
, int fd
)
2611 unsigned long file_offset
;
2613 /* skip magic which was already checked */
2614 read(fd
, magic
, sizeof(magic
));
2617 len
= read(fd
, &hdr
, sizeof(hdr
));
2620 if (len
!= sizeof(hdr
)) {
2621 tcc_error_noabort("invalid archive");
2624 memcpy(ar_size
, hdr
.ar_size
, sizeof(hdr
.ar_size
));
2625 ar_size
[sizeof(hdr
.ar_size
)] = '\0';
2626 size
= strtol(ar_size
, NULL
, 0);
2627 memcpy(ar_name
, hdr
.ar_name
, sizeof(hdr
.ar_name
));
2628 for(i
= sizeof(hdr
.ar_name
) - 1; i
>= 0; i
--) {
2629 if (ar_name
[i
] != ' ')
2632 ar_name
[i
+ 1] = '\0';
2633 // printf("name='%s' size=%d %s\n", ar_name, size, ar_size);
2634 file_offset
= lseek(fd
, 0, SEEK_CUR
);
2636 size
= (size
+ 1) & ~1;
2637 if (!strcmp(ar_name
, "/")) {
2638 /* coff symbol table : we handle it */
2639 if(s1
->alacarte_link
)
2640 return tcc_load_alacarte(s1
, fd
, size
);
2641 } else if (!strcmp(ar_name
, "//") ||
2642 !strcmp(ar_name
, "__.SYMDEF") ||
2643 !strcmp(ar_name
, "__.SYMDEF/") ||
2644 !strcmp(ar_name
, "ARFILENAMES/")) {
2645 /* skip symbol table or archive names */
2647 if (tcc_load_object_file(s1
, fd
, file_offset
) < 0)
2650 lseek(fd
, file_offset
+ size
, SEEK_SET
);
2655 #ifndef TCC_TARGET_PE
2656 /* load a DLL and all referenced DLLs. 'level = 0' means that the DLL
2657 is referenced by the user (so it should be added as DT_NEEDED in
2658 the generated ELF file) */
2659 ST_FUNC
int tcc_load_dll(TCCState
*s1
, int fd
, const char *filename
, int level
)
2662 ElfW(Shdr
) *shdr
, *sh
, *sh1
;
2663 int i
, j
, nb_syms
, nb_dts
, sym_bind
, ret
;
2664 ElfW(Sym
) *sym
, *dynsym
;
2665 ElfW(Dyn
) *dt
, *dynamic
;
2666 unsigned char *dynstr
;
2667 const char *name
, *soname
;
2668 DLLReference
*dllref
;
2670 read(fd
, &ehdr
, sizeof(ehdr
));
2672 /* test CPU specific stuff */
2673 if (ehdr
.e_ident
[5] != ELFDATA2LSB
||
2674 ehdr
.e_machine
!= EM_TCC_TARGET
) {
2675 tcc_error_noabort("bad architecture");
2680 shdr
= load_data(fd
, ehdr
.e_shoff
, sizeof(ElfW(Shdr
)) * ehdr
.e_shnum
);
2682 /* load dynamic section and dynamic symbols */
2686 dynsym
= NULL
; /* avoid warning */
2687 dynstr
= NULL
; /* avoid warning */
2688 for(i
= 0, sh
= shdr
; i
< ehdr
.e_shnum
; i
++, sh
++) {
2689 switch(sh
->sh_type
) {
2691 nb_dts
= sh
->sh_size
/ sizeof(ElfW(Dyn
));
2692 dynamic
= load_data(fd
, sh
->sh_offset
, sh
->sh_size
);
2695 nb_syms
= sh
->sh_size
/ sizeof(ElfW(Sym
));
2696 dynsym
= load_data(fd
, sh
->sh_offset
, sh
->sh_size
);
2697 sh1
= &shdr
[sh
->sh_link
];
2698 dynstr
= load_data(fd
, sh1
->sh_offset
, sh1
->sh_size
);
2705 /* compute the real library name */
2706 soname
= tcc_basename(filename
);
2708 for(i
= 0, dt
= dynamic
; i
< nb_dts
; i
++, dt
++) {
2709 if (dt
->d_tag
== DT_SONAME
) {
2710 soname
= dynstr
+ dt
->d_un
.d_val
;
2714 /* if the dll is already loaded, do not load it */
2715 for(i
= 0; i
< s1
->nb_loaded_dlls
; i
++) {
2716 dllref
= s1
->loaded_dlls
[i
];
2717 if (!strcmp(soname
, dllref
->name
)) {
2718 /* but update level if needed */
2719 if (level
< dllref
->level
)
2720 dllref
->level
= level
;
2726 // printf("loading dll '%s'\n", soname);
2728 /* add the dll and its level */
2729 dllref
= tcc_mallocz(sizeof(DLLReference
) + strlen(soname
));
2730 dllref
->level
= level
;
2731 strcpy(dllref
->name
, soname
);
2732 dynarray_add((void ***)&s1
->loaded_dlls
, &s1
->nb_loaded_dlls
, dllref
);
2734 /* add dynamic symbols in dynsym_section */
2735 for(i
= 1, sym
= dynsym
+ 1; i
< nb_syms
; i
++, sym
++) {
2736 sym_bind
= ELFW(ST_BIND
)(sym
->st_info
);
2737 if (sym_bind
== STB_LOCAL
)
2739 name
= dynstr
+ sym
->st_name
;
2740 add_elf_sym(s1
->dynsymtab_section
, sym
->st_value
, sym
->st_size
,
2741 sym
->st_info
, sym
->st_other
, sym
->st_shndx
, name
);
2744 /* load all referenced DLLs */
2745 for(i
= 0, dt
= dynamic
; i
< nb_dts
; i
++, dt
++) {
2748 name
= dynstr
+ dt
->d_un
.d_val
;
2749 for(j
= 0; j
< s1
->nb_loaded_dlls
; j
++) {
2750 dllref
= s1
->loaded_dlls
[j
];
2751 if (!strcmp(name
, dllref
->name
))
2752 goto already_loaded
;
2754 if (tcc_add_dll(s1
, name
, AFF_REFERENCED_DLL
) < 0) {
2755 tcc_error_noabort("referenced dll '%s' not found", name
);
2772 #define LD_TOK_NAME 256
2773 #define LD_TOK_EOF (-1)
2775 /* return next ld script token */
2776 static int ld_next(TCCState
*s1
, char *name
, int name_size
)
2794 file
->buf_ptr
= parse_comment(file
->buf_ptr
);
2795 ch
= file
->buf_ptr
[0];
2803 /* case 'a' ... 'z': */
2830 /* case 'A' ... 'z': */
2865 if (!((ch
>= 'a' && ch
<= 'z') ||
2866 (ch
>= 'A' && ch
<= 'Z') ||
2867 (ch
>= '0' && ch
<= '9') ||
2868 strchr("/.-_+=$:\\,~", ch
)))
2870 if ((q
- name
) < name_size
- 1) {
2887 printf("tok=%c %d\n", c
, c
);
2888 if (c
== LD_TOK_NAME
)
2889 printf(" name=%s\n", name
);
2895 * Extract the file name from the library name
2897 * /!\ No test on filename capacity, be careful
2899 static void libname_to_filename(TCCState
*s1
, const char libname
[], char filename
[])
2901 if (!s1
->static_link
) {
2902 sprintf(filename
, "lib%s.so", libname
);
2904 sprintf(filename
, "lib%s.a", libname
);
2908 static int ld_add_file(TCCState
*s1
, const char filename
[])
2912 ret
= tcc_add_file_internal(s1
, filename
, 0);
2914 ret
= tcc_add_dll(s1
, filename
, 0);
2918 static inline int new_undef_syms(void)
2921 ret
= new_undef_sym
;
2926 static int ld_add_file_list(TCCState
*s1
, const char *cmd
, int as_needed
)
2928 char filename
[1024], libname
[1024];
2929 int t
, group
, nblibs
= 0, ret
= 0;
2932 group
= !strcmp(cmd
, "GROUP");
2935 t
= ld_next(s1
, filename
, sizeof(filename
));
2938 t
= ld_next(s1
, filename
, sizeof(filename
));
2941 if (t
== LD_TOK_EOF
) {
2942 tcc_error_noabort("unexpected end of file");
2944 goto lib_parse_error
;
2945 } else if (t
== ')') {
2947 } else if (t
== '-') {
2948 t
= ld_next(s1
, filename
, sizeof(filename
));
2949 if ((t
!= LD_TOK_NAME
) || (filename
[0] != 'l')) {
2950 tcc_error_noabort("library name expected");
2952 goto lib_parse_error
;
2954 strcpy(libname
, &filename
[1]);
2955 libname_to_filename(s1
, libname
, filename
);
2956 } else if (t
!= LD_TOK_NAME
) {
2957 tcc_error_noabort("filename expected");
2959 goto lib_parse_error
;
2961 if (!strcmp(filename
, "AS_NEEDED")) {
2962 ret
= ld_add_file_list(s1
, cmd
, 1);
2964 goto lib_parse_error
;
2966 /* TODO: Implement AS_NEEDED support. Ignore it for now */
2968 ret
= ld_add_file(s1
, filename
);
2970 goto lib_parse_error
;
2972 /* Add the filename *and* the libname to avoid future conversions */
2973 dynarray_add((void ***) &libs
, &nblibs
, tcc_strdup(filename
));
2974 if (libname
[0] != '\0')
2975 dynarray_add((void ***) &libs
, &nblibs
, tcc_strdup(libname
));
2979 t
= ld_next(s1
, filename
, sizeof(filename
));
2981 t
= ld_next(s1
, filename
, sizeof(filename
));
2984 if (group
&& !as_needed
) {
2985 while (new_undef_syms()) {
2988 for (i
= 0; i
< nblibs
; i
++)
2989 ld_add_file(s1
, libs
[i
]);
2993 dynarray_reset(&libs
, &nblibs
);
2997 /* interpret a subset of GNU ldscripts to handle the dummy libc.so
2999 ST_FUNC
int tcc_load_ldscript(TCCState
*s1
)
3002 char filename
[1024];
3005 ch
= file
->buf_ptr
[0];
3008 t
= ld_next(s1
, cmd
, sizeof(cmd
));
3009 if (t
== LD_TOK_EOF
)
3011 else if (t
!= LD_TOK_NAME
)
3013 if (!strcmp(cmd
, "INPUT") ||
3014 !strcmp(cmd
, "GROUP")) {
3015 ret
= ld_add_file_list(s1
, cmd
, 0);
3018 } else if (!strcmp(cmd
, "OUTPUT_FORMAT") ||
3019 !strcmp(cmd
, "TARGET")) {
3020 /* ignore some commands */
3021 t
= ld_next(s1
, cmd
, sizeof(cmd
));
3025 t
= ld_next(s1
, filename
, sizeof(filename
));
3026 if (t
== LD_TOK_EOF
) {
3027 tcc_error_noabort("unexpected end of file");
3029 } else if (t
== ')') {
3039 #endif /* ndef TCC_TARGET_PE */