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 /* Define this to get some debug output during relocation processing. */
26 /* XXX: avoid static variable */
27 static int new_undef_sym
= 0; /* Is there a new undefined sym since last new_undef_sym() */
29 ST_FUNC
int put_elf_str(Section
*s
, const char *sym
)
34 len
= strlen(sym
) + 1;
35 offset
= s
->data_offset
;
36 ptr
= section_ptr_add(s
, len
);
37 memcpy(ptr
, sym
, len
);
41 /* elf symbol hashing function */
42 static unsigned long elf_hash(const unsigned char *name
)
44 unsigned long h
= 0, g
;
47 h
= (h
<< 4) + *name
++;
56 /* rebuild hash table of section s */
57 /* NOTE: we do factorize the hash table code to go faster */
58 static void rebuild_hash(Section
*s
, unsigned int nb_buckets
)
61 int *ptr
, *hash
, nb_syms
, sym_index
, h
;
62 unsigned char *strtab
;
64 strtab
= s
->link
->data
;
65 nb_syms
= s
->data_offset
/ sizeof(ElfW(Sym
));
67 s
->hash
->data_offset
= 0;
68 ptr
= section_ptr_add(s
->hash
, (2 + nb_buckets
+ nb_syms
) * sizeof(int));
73 memset(hash
, 0, (nb_buckets
+ 1) * sizeof(int));
74 ptr
+= nb_buckets
+ 1;
76 sym
= (ElfW(Sym
) *)s
->data
+ 1;
77 for(sym_index
= 1; sym_index
< nb_syms
; sym_index
++) {
78 if (ELFW(ST_BIND
)(sym
->st_info
) != STB_LOCAL
) {
79 h
= elf_hash(strtab
+ sym
->st_name
) % nb_buckets
;
90 /* return the symbol number */
91 ST_FUNC
int put_elf_sym(Section
*s
, addr_t value
, unsigned long size
,
92 int info
, int other
, int shndx
, const char *name
)
94 int name_offset
, sym_index
;
99 sym
= section_ptr_add(s
, sizeof(ElfW(Sym
)));
101 name_offset
= put_elf_str(s
->link
, name
);
104 /* XXX: endianness */
105 sym
->st_name
= name_offset
;
106 sym
->st_value
= value
;
109 sym
->st_other
= other
;
110 sym
->st_shndx
= shndx
;
111 sym_index
= sym
- (ElfW(Sym
) *)s
->data
;
115 ptr
= section_ptr_add(hs
, sizeof(int));
116 base
= (int *)hs
->data
;
117 /* only add global or weak symbols */
118 if (ELFW(ST_BIND
)(info
) != STB_LOCAL
) {
119 /* add another hashing entry */
121 h
= elf_hash((unsigned char *) name
) % nbuckets
;
123 base
[2 + h
] = sym_index
;
125 /* we resize the hash table */
126 hs
->nb_hashed_syms
++;
127 if (hs
->nb_hashed_syms
> 2 * nbuckets
) {
128 rebuild_hash(s
, 2 * nbuckets
);
138 /* find global ELF symbol 'name' and return its index. Return 0 if not
140 ST_FUNC
int find_elf_sym(Section
*s
, const char *name
)
144 int nbuckets
, sym_index
, h
;
150 nbuckets
= ((int *)hs
->data
)[0];
151 h
= elf_hash((unsigned char *) name
) % nbuckets
;
152 sym_index
= ((int *)hs
->data
)[2 + h
];
153 while (sym_index
!= 0) {
154 sym
= &((ElfW(Sym
) *)s
->data
)[sym_index
];
155 name1
= (char *) s
->link
->data
+ sym
->st_name
;
156 if (!strcmp(name
, name1
))
158 sym_index
= ((int *)hs
->data
)[2 + nbuckets
+ sym_index
];
163 /* return elf symbol value, signal error if 'err' is nonzero */
164 ST_FUNC addr_t
get_elf_sym_addr(TCCState
*s
, const char *name
, int err
)
169 sym_index
= find_elf_sym(s
->symtab
, name
);
170 sym
= &((ElfW(Sym
) *)s
->symtab
->data
)[sym_index
];
171 if (!sym_index
|| sym
->st_shndx
== SHN_UNDEF
) {
173 tcc_error("%s not defined", name
);
176 return sym
->st_value
;
179 /* return elf symbol value */
180 LIBTCCAPI
void *tcc_get_symbol(TCCState
*s
, const char *name
)
182 return (void*)(uintptr_t)get_elf_sym_addr(s
, name
, 0);
185 #if defined TCC_IS_NATIVE || defined TCC_TARGET_PE
186 /* return elf symbol value or error */
187 ST_FUNC
void* tcc_get_symbol_err(TCCState
*s
, const char *name
)
189 return (void*)(uintptr_t)get_elf_sym_addr(s
, name
, 1);
193 /* add an elf symbol : check if it is already defined and patch
194 it. Return symbol index. NOTE that sh_num can be SHN_UNDEF. */
195 ST_FUNC
int add_elf_sym(Section
*s
, addr_t value
, unsigned long size
,
196 int info
, int other
, int sh_num
, const char *name
)
199 int sym_bind
, sym_index
, sym_type
, esym_bind
;
200 unsigned char sym_vis
, esym_vis
, new_vis
;
202 sym_bind
= ELFW(ST_BIND
)(info
);
203 sym_type
= ELFW(ST_TYPE
)(info
);
204 sym_vis
= ELFW(ST_VISIBILITY
)(other
);
206 if (sym_bind
!= STB_LOCAL
) {
207 /* we search global or weak symbols */
208 sym_index
= find_elf_sym(s
, name
);
211 esym
= &((ElfW(Sym
) *)s
->data
)[sym_index
];
212 if (esym
->st_shndx
!= SHN_UNDEF
) {
213 esym_bind
= ELFW(ST_BIND
)(esym
->st_info
);
214 /* propagate the most constraining visibility */
215 /* STV_DEFAULT(0)<STV_PROTECTED(3)<STV_HIDDEN(2)<STV_INTERNAL(1) */
216 esym_vis
= ELFW(ST_VISIBILITY
)(esym
->st_other
);
217 if (esym_vis
== STV_DEFAULT
) {
219 } else if (sym_vis
== STV_DEFAULT
) {
222 new_vis
= (esym_vis
< sym_vis
) ? esym_vis
: sym_vis
;
224 esym
->st_other
= (esym
->st_other
& ~ELFW(ST_VISIBILITY
)(-1))
226 other
= esym
->st_other
; /* in case we have to patch esym */
227 if (sh_num
== SHN_UNDEF
) {
228 /* ignore adding of undefined symbol if the
229 corresponding symbol is already defined */
230 } else if (sym_bind
== STB_GLOBAL
&& esym_bind
== STB_WEAK
) {
231 /* global overrides weak, so patch */
233 } else if (sym_bind
== STB_WEAK
&& esym_bind
== STB_GLOBAL
) {
234 /* weak is ignored if already global */
235 } else if (sym_bind
== STB_WEAK
&& esym_bind
== STB_WEAK
) {
236 /* keep first-found weak definition, ignore subsequents */
237 } else if (sym_vis
== STV_HIDDEN
|| sym_vis
== STV_INTERNAL
) {
238 /* ignore hidden symbols after */
239 } else if (esym
->st_shndx
== SHN_COMMON
240 && (sh_num
< SHN_LORESERVE
|| sh_num
== SHN_COMMON
)) {
241 /* gr: Happens with 'tcc ... -static tcctest.c' on e.g. Ubuntu 6.01
242 No idea if this is the correct solution ... */
244 } else if (s
== tcc_state
->dynsymtab_section
) {
245 /* we accept that two DLL define the same symbol */
248 printf("new_bind=%x new_shndx=%x new_vis=%x old_bind=%x old_shndx=%x old_vis=%x\n",
249 sym_bind
, sh_num
, new_vis
, esym_bind
, esym
->st_shndx
, esym_vis
);
251 tcc_error_noabort("'%s' defined twice", name
);
255 esym
->st_info
= ELFW(ST_INFO
)(sym_bind
, sym_type
);
256 esym
->st_shndx
= sh_num
;
258 esym
->st_value
= value
;
259 esym
->st_size
= size
;
260 esym
->st_other
= other
;
264 sym_index
= put_elf_sym(s
, value
, size
,
265 ELFW(ST_INFO
)(sym_bind
, sym_type
), other
,
272 ST_FUNC
void put_elf_reloca(Section
*symtab
, Section
*s
, unsigned long offset
,
273 int type
, int symbol
, addr_t addend
)
281 /* if no relocation section, create it */
282 snprintf(buf
, sizeof(buf
), REL_SECTION_FMT
, s
->name
);
283 /* if the symtab is allocated, then we consider the relocation
285 sr
= new_section(tcc_state
, buf
, SHT_RELX
, symtab
->sh_flags
);
286 sr
->sh_entsize
= sizeof(ElfW_Rel
);
288 sr
->sh_info
= s
->sh_num
;
291 rel
= section_ptr_add(sr
, sizeof(ElfW_Rel
));
292 rel
->r_offset
= offset
;
293 rel
->r_info
= ELFW(R_INFO
)(symbol
, type
);
294 #if defined(TCC_TARGET_ARM64) || defined(TCC_TARGET_X86_64)
295 rel
->r_addend
= addend
;
298 tcc_error("non-zero addend on REL architecture");
302 ST_FUNC
void put_elf_reloc(Section
*symtab
, Section
*s
, unsigned long offset
,
303 int type
, int symbol
)
305 put_elf_reloca(symtab
, s
, offset
, type
, symbol
, 0);
308 /* put stab debug information */
310 ST_FUNC
void put_stabs(const char *str
, int type
, int other
, int desc
,
315 sym
= section_ptr_add(stab_section
, sizeof(Stab_Sym
));
317 sym
->n_strx
= put_elf_str(stabstr_section
, str
);
322 sym
->n_other
= other
;
324 sym
->n_value
= value
;
327 ST_FUNC
void put_stabs_r(const char *str
, int type
, int other
, int desc
,
328 unsigned long value
, Section
*sec
, int sym_index
)
330 put_stabs(str
, type
, other
, desc
, value
);
331 put_elf_reloc(symtab_section
, stab_section
,
332 stab_section
->data_offset
- sizeof(unsigned int),
333 R_DATA_32
, sym_index
);
336 ST_FUNC
void put_stabn(int type
, int other
, int desc
, int value
)
338 put_stabs(NULL
, type
, other
, desc
, value
);
341 ST_FUNC
void put_stabd(int type
, int other
, int desc
)
343 put_stabs(NULL
, type
, other
, desc
, 0);
346 /* Browse each elem of type <type> in section <sec> starting at elem <startoff>
347 using variable <elem> */
348 #define for_each_elem(sec, startoff, elem, type) \
349 for (elem = (type *) sec->data + startoff; \
350 elem < (type *) (sec->data + sec->data_offset); elem++)
352 /* In an ELF file symbol table, the local symbols must appear below
353 the global and weak ones. Since TCC cannot sort it while generating
354 the code, we must do it after. All the relocation tables are also
355 modified to take into account the symbol table sorting */
356 static void sort_syms(TCCState
*s1
, Section
*s
)
358 int *old_to_new_syms
;
366 nb_syms
= s
->data_offset
/ sizeof(ElfW(Sym
));
367 new_syms
= tcc_malloc(nb_syms
* sizeof(ElfW(Sym
)));
368 old_to_new_syms
= tcc_malloc(nb_syms
* sizeof(int));
370 /* first pass for local symbols */
371 p
= (ElfW(Sym
) *)s
->data
;
373 for(i
= 0; i
< nb_syms
; i
++) {
374 if (ELFW(ST_BIND
)(p
->st_info
) == STB_LOCAL
) {
375 old_to_new_syms
[i
] = q
- new_syms
;
380 /* save the number of local symbols in section header */
381 s
->sh_info
= q
- new_syms
;
383 /* then second pass for non local symbols */
384 p
= (ElfW(Sym
) *)s
->data
;
385 for(i
= 0; i
< nb_syms
; i
++) {
386 if (ELFW(ST_BIND
)(p
->st_info
) != STB_LOCAL
) {
387 old_to_new_syms
[i
] = q
- new_syms
;
393 /* we copy the new symbols to the old */
394 memcpy(s
->data
, new_syms
, nb_syms
* sizeof(ElfW(Sym
)));
397 /* now we modify all the relocations */
398 for(i
= 1; i
< s1
->nb_sections
; i
++) {
399 sr
= s1
->sections
[i
];
400 if (sr
->sh_type
== SHT_RELX
&& sr
->link
== s
) {
401 for_each_elem(sr
, 0, rel
, ElfW_Rel
) {
402 sym_index
= ELFW(R_SYM
)(rel
->r_info
);
403 type
= ELFW(R_TYPE
)(rel
->r_info
);
404 sym_index
= old_to_new_syms
[sym_index
];
405 rel
->r_info
= ELFW(R_INFO
)(sym_index
, type
);
410 tcc_free(old_to_new_syms
);
413 /* relocate common symbols in the .bss section */
414 ST_FUNC
void relocate_common_syms(void)
417 unsigned long offset
, align
;
419 for_each_elem(symtab_section
, 1, sym
, ElfW(Sym
)) {
420 if (sym
->st_shndx
== SHN_COMMON
) {
422 align
= sym
->st_value
;
423 offset
= bss_section
->data_offset
;
424 offset
= (offset
+ align
- 1) & -align
;
425 sym
->st_value
= offset
;
426 sym
->st_shndx
= bss_section
->sh_num
;
427 offset
+= sym
->st_size
;
428 bss_section
->data_offset
= offset
;
433 /* relocate symbol table, resolve undefined symbols if do_resolve is
434 true and output error if undefined symbol. */
435 ST_FUNC
void relocate_syms(TCCState
*s1
, int do_resolve
)
437 ElfW(Sym
) *sym
, *esym
;
438 int sym_bind
, sh_num
, sym_index
;
441 for_each_elem(symtab_section
, 1, sym
, ElfW(Sym
)) {
442 sh_num
= sym
->st_shndx
;
443 if (sh_num
== SHN_UNDEF
) {
444 name
= (char *) strtab_section
->data
+ sym
->st_name
;
445 /* Use ld.so to resolve symbol for us (for tcc -run) */
447 #if defined TCC_IS_NATIVE && !defined _WIN32
449 name
= (char *) symtab_section
->link
->data
+ sym
->st_name
;
450 addr
= resolve_sym(s1
, name
);
452 sym
->st_value
= (addr_t
)addr
;
454 printf ("relocate_sym: %s -> 0x%lx\n", name
, sym
->st_value
);
459 } else if (s1
->dynsym
) {
460 /* if dynamic symbol exist, then use it */
461 sym_index
= find_elf_sym(s1
->dynsym
, name
);
463 esym
= &((ElfW(Sym
) *)s1
->dynsym
->data
)[sym_index
];
464 sym
->st_value
= esym
->st_value
;
468 /* XXX: _fp_hw seems to be part of the ABI, so we ignore
470 if (!strcmp(name
, "_fp_hw"))
472 /* only weak symbols are accepted to be undefined. Their
474 sym_bind
= ELFW(ST_BIND
)(sym
->st_info
);
475 if (sym_bind
== STB_WEAK
) {
478 tcc_error_noabort("undefined symbol '%s'", name
);
480 } else if (sh_num
< SHN_LORESERVE
) {
481 /* add section base */
482 sym
->st_value
+= s1
->sections
[sym
->st_shndx
]->sh_addr
;
488 /* relocate a given section (CPU dependent) by applying the relocations
489 in the associated relocation section */
490 ST_FUNC
void relocate_section(TCCState
*s1
, Section
*s
)
492 Section
*sr
= s
->reloc
;
498 #if defined TCC_TARGET_I386 || defined TCC_TARGET_X86_64
499 ElfW_Rel
*qrel
= (ElfW_Rel
*) sr
->data
; /* ptr to next reloc entry reused */
503 for_each_elem(sr
, 0, rel
, ElfW_Rel
) {
504 ptr
= s
->data
+ rel
->r_offset
;
506 sym_index
= ELFW(R_SYM
)(rel
->r_info
);
507 sym
= &((ElfW(Sym
) *)symtab_section
->data
)[sym_index
];
509 #if defined(TCC_TARGET_ARM64) || defined(TCC_TARGET_X86_64)
510 val
+= rel
->r_addend
;
512 type
= ELFW(R_TYPE
)(rel
->r_info
);
513 addr
= s
->sh_addr
+ rel
->r_offset
;
517 #if defined(TCC_TARGET_I386)
519 if (s1
->output_type
== TCC_OUTPUT_DLL
) {
520 esym_index
= s1
->symtab_to_dynsym
[sym_index
];
521 qrel
->r_offset
= rel
->r_offset
;
523 qrel
->r_info
= ELFW(R_INFO
)(esym_index
, R_386_32
);
527 qrel
->r_info
= ELFW(R_INFO
)(0, R_386_RELATIVE
);
534 if (s1
->output_type
== TCC_OUTPUT_DLL
) {
536 esym_index
= s1
->symtab_to_dynsym
[sym_index
];
538 qrel
->r_offset
= rel
->r_offset
;
539 qrel
->r_info
= ELFW(R_INFO
)(esym_index
, R_386_PC32
);
544 *(int *)ptr
+= val
- addr
;
547 *(int *)ptr
+= val
- addr
;
554 *(int *)ptr
+= s1
->got
->sh_addr
- addr
;
557 *(int *)ptr
+= val
- s1
->got
->sh_addr
;
560 /* we load the got offset */
561 *(int *)ptr
+= s1
->sym_attrs
[sym_index
].got_offset
;
564 if (s1
->output_format
!= TCC_OUTPUT_FORMAT_BINARY
) {
566 tcc_error("can only produce 16-bit binary files");
568 *(short *)ptr
+= val
;
571 if (s1
->output_format
!= TCC_OUTPUT_FORMAT_BINARY
)
573 *(short *)ptr
+= val
- addr
;
575 #elif defined(TCC_TARGET_ARM)
581 int x
, is_thumb
, is_call
, h
, blx_avail
, is_bl
, th_ko
;
582 x
= (*(int *) ptr
) & 0xffffff;
583 if (sym
->st_shndx
== SHN_UNDEF
)
584 val
= s1
->plt
->sh_addr
;
586 printf ("reloc %d: x=0x%x val=0x%x ", type
, x
, val
);
588 (*(int *)ptr
) &= 0xff000000;
592 blx_avail
= (TCC_ARM_VERSION
>= 5);
594 is_bl
= (*(unsigned *) ptr
) >> 24 == 0xeb;
595 is_call
= (type
== R_ARM_CALL
|| (type
== R_ARM_PC24
&& is_bl
));
598 printf (" newx=0x%x name=%s\n", x
,
599 (char *) symtab_section
->link
->data
+ sym
->st_name
);
602 th_ko
= (x
& 3) && (!blx_avail
|| !is_call
);
603 if (th_ko
|| x
>= 0x2000000 || x
< -0x2000000)
604 tcc_error("can't relocate value at %x,%d",addr
, type
);
607 /* Only reached if blx is avail and it is a call */
610 (*(int *)ptr
) = 0xfa << 24; /* bl -> blx */
615 /* Since these relocations only concern Thumb-2 and blx instruction was
616 introduced before Thumb-2, we can assume blx is available and not
619 case R_ARM_THM_JUMP24
:
621 int x
, hi
, lo
, s
, j1
, j2
, i1
, i2
, imm10
, imm11
;
622 int to_thumb
, is_call
, to_plt
, blx_bit
= 1 << 12;
626 if (sym
->st_shndx
== SHN_UNDEF
&&
627 ELFW(ST_BIND
)(sym
->st_info
) == STB_WEAK
)
630 /* Get initial offset */
631 hi
= (*(uint16_t *)ptr
);
632 lo
= (*(uint16_t *)(ptr
+2));
640 x
= (s
<< 24) | (i1
<< 23) | (i2
<< 22) |
641 (imm10
<< 12) | (imm11
<< 1);
645 /* Relocation infos */
648 to_plt
= (val
>= plt
->sh_addr
) &&
649 (val
< plt
->sh_addr
+ plt
->data_offset
);
650 is_call
= (type
== R_ARM_THM_PC22
);
652 /* Compute final offset */
653 if (to_plt
&& !is_call
) /* Point to 1st instr of Thumb stub */
656 if (!to_thumb
&& is_call
) {
657 blx_bit
= 0; /* bl -> blx */
658 x
= (x
+ 3) & -4; /* Compute offset from aligned PC */
661 /* Check that relocation is possible
662 * offset must not be out of range
663 * if target is to be entered in arm mode:
665 - instruction must be a call (bl) or a jump to PLT */
666 if (!to_thumb
|| x
>= 0x1000000 || x
< -0x1000000)
667 if (to_thumb
|| (val
& 2) || (!is_call
&& !to_plt
))
668 tcc_error("can't relocate value at %x,%d",addr
, type
);
670 /* Compute and store final offset */
676 imm10
= (x
>> 12) & 0x3ff;
677 imm11
= (x
>> 1) & 0x7ff;
678 (*(uint16_t *)ptr
) = (uint16_t) ((hi
& 0xf800) |
680 (*(uint16_t *)(ptr
+2)) = (uint16_t) ((lo
& 0xc000) |
681 (j1
<< 13) | blx_bit
| (j2
<< 11) |
686 case R_ARM_MOVW_ABS_NC
:
689 if (type
== R_ARM_MOVT_ABS
)
692 imm4
= (val
>> 12) & 0xf;
693 x
= (imm4
<< 16) | imm12
;
694 if (type
== R_ARM_THM_MOVT_ABS
)
700 case R_ARM_THM_MOVT_ABS
:
701 case R_ARM_THM_MOVW_ABS_NC
:
703 int x
, i
, imm4
, imm3
, imm8
;
704 if (type
== R_ARM_THM_MOVT_ABS
)
707 imm3
= (val
>> 8) & 0x7;
709 imm4
= (val
>> 12) & 0xf;
710 x
= (imm3
<< 28) | (imm8
<< 16) | (i
<< 10) | imm4
;
711 if (type
== R_ARM_THM_MOVT_ABS
)
720 x
= (*(int *)ptr
) & 0x7fffffff;
721 (*(int *)ptr
) &= 0x80000000;
724 if((x
^(x
>>1))&0x40000000)
725 tcc_error("can't relocate value at %x,%d",addr
, type
);
726 (*(int *)ptr
) |= x
& 0x7fffffff;
732 *(int *)ptr
+= val
- addr
;
735 *(int *)ptr
+= s1
->got
->sh_addr
- addr
;
738 *(int *)ptr
+= val
- s1
->got
->sh_addr
;
741 /* we load the got offset */
742 *(int *)ptr
+= s1
->sym_attrs
[sym_index
].got_offset
;
747 /* trade Thumb support for ARMv4 support */
748 if ((0x0ffffff0 & *(int*)ptr
) == 0x012FFF10)
749 *(int*)ptr
^= 0xE12FFF10 ^ 0xE1A0F000; /* BX Rm -> MOV PC, Rm */
752 case R_ARM_JUMP_SLOT
:
753 *(addr_t
*)ptr
= val
;
756 /* Nothing to do. Normally used to indicate a dependency
757 on a certain symbol (like for exception handling under EABI). */
760 fprintf(stderr
,"FIXME: handle reloc type %x at %x [%p] to %x\n",
761 type
, (unsigned)addr
, ptr
, (unsigned)val
);
763 #elif defined(TCC_TARGET_ARM64)
764 case R_AARCH64_ABS64
:
765 *(uint64_t *)ptr
= val
;
767 case R_AARCH64_ABS32
:
768 *(uint32_t *)ptr
= val
;
770 case R_AARCH64_MOVW_UABS_G0_NC
:
771 *(uint32_t *)ptr
= (*(uint32_t *)ptr
& 0xffe0001f) |
774 case R_AARCH64_MOVW_UABS_G1_NC
:
775 *(uint32_t *)ptr
= (*(uint32_t *)ptr
& 0xffe0001f) |
776 (val
>> 16 & 0xffff) << 5;
778 case R_AARCH64_MOVW_UABS_G2_NC
:
779 *(uint32_t *)ptr
= (*(uint32_t *)ptr
& 0xffe0001f) |
780 (val
>> 32 & 0xffff) << 5;
782 case R_AARCH64_MOVW_UABS_G3
:
783 *(uint32_t *)ptr
= (*(uint32_t *)ptr
& 0xffe0001f) |
784 (val
>> 48 & 0xffff) << 5;
786 case R_AARCH64_ADR_PREL_PG_HI21
: {
787 uint64_t off
= (val
>> 12) - (addr
>> 12);
788 if ((off
+ ((uint64_t)1 << 20)) >> 21)
789 tcc_error("R_AARCH64_ADR_PREL_PG_HI21 relocation failed");
790 *(uint32_t *)ptr
= (*(uint32_t *)ptr
& 0x9f00001f) |
791 (off
& 0x1ffffc) << 3 | (off
& 3) << 29;
794 case R_AARCH64_ADD_ABS_LO12_NC
:
795 *(uint32_t *)ptr
= (*(uint32_t *)ptr
& 0xffc003ff) |
798 case R_AARCH64_JUMP26
:
799 case R_AARCH64_CALL26
:
800 /* This check must match the one in build_got_entries, testing
801 if we really need a PLT slot. */
802 if (sym
->st_shndx
== SHN_UNDEF
)
803 /* We've put the PLT slot offset into r_addend when generating
804 it, and that's what we must use as relocation value (adjusted
805 by section offset of course). */
806 val
= s1
->plt
->sh_addr
+ rel
->r_addend
;
808 printf ("reloc %d @ 0x%lx: val=0x%lx name=%s\n", type
, addr
, val
,
809 (char *) symtab_section
->link
->data
+ sym
->st_name
);
811 if (((val
- addr
) + ((uint64_t)1 << 27)) & ~(uint64_t)0xffffffc)
813 tcc_error("R_AARCH64_(JUMP|CALL)26 relocation failed (val=%lx, addr=%lx)", addr
, val
);
815 *(uint32_t *)ptr
= 0x14000000 | (type
== R_AARCH64_CALL26
) << 31 |
816 ((val
- addr
) >> 2 & 0x3ffffff);
818 case R_AARCH64_ADR_GOT_PAGE
: {
820 (((s1
->got
->sh_addr
+
821 s1
->sym_attrs
[sym_index
].got_offset
) >> 12) - (addr
>> 12));
822 if ((off
+ ((uint64_t)1 << 20)) >> 21)
823 tcc_error("R_AARCH64_ADR_GOT_PAGE relocation failed");
824 *(uint32_t *)ptr
= (*(uint32_t *)ptr
& 0x9f00001f) |
825 (off
& 0x1ffffc) << 3 | (off
& 3) << 29;
828 case R_AARCH64_LD64_GOT_LO12_NC
:
829 *(uint32_t *)ptr
= (*(uint32_t *)ptr
& 0xfff803ff) |
830 ((s1
->got
->sh_addr
+ s1
->sym_attrs
[sym_index
].got_offset
)
835 case R_AARCH64_GLOB_DAT
:
836 case R_AARCH64_JUMP_SLOT
:
837 /* They don't need addend */
839 printf ("reloc %d @ 0x%lx: val=0x%lx name=%s\n", type
, addr
,
841 (char *) symtab_section
->link
->data
+ sym
->st_name
);
843 *(addr_t
*)ptr
= val
- rel
->r_addend
;
846 fprintf(stderr
, "FIXME: handle reloc type %x at %x [%p] to %x\n",
847 type
, (unsigned)addr
, ptr
, (unsigned)val
);
849 #elif defined(TCC_TARGET_C67)
857 /* put the low 16 bits of the absolute address
858 add to what is already there */
860 orig
= ((*(int *)(ptr
)) >> 7) & 0xffff;
861 orig
|= (((*(int *)(ptr
+4)) >> 7) & 0xffff) << 16;
863 /* patch both at once - assumes always in pairs Low - High */
865 *(int *) ptr
= (*(int *) ptr
& (~(0xffff << 7)) ) | (((val
+orig
) & 0xffff) << 7);
866 *(int *)(ptr
+4) = (*(int *)(ptr
+4) & (~(0xffff << 7)) ) | ((((val
+orig
)>>16) & 0xffff) << 7);
872 fprintf(stderr
,"FIXME: handle reloc type %x at %x [%p] to %x\n",
873 type
, (unsigned)addr
, ptr
, (unsigned)val
);
875 #elif defined(TCC_TARGET_X86_64)
877 if (s1
->output_type
== TCC_OUTPUT_DLL
) {
878 esym_index
= s1
->symtab_to_dynsym
[sym_index
];
879 qrel
->r_offset
= rel
->r_offset
;
881 qrel
->r_info
= ELFW(R_INFO
)(esym_index
, R_X86_64_64
);
882 qrel
->r_addend
= rel
->r_addend
;
886 qrel
->r_info
= ELFW(R_INFO
)(0, R_X86_64_RELATIVE
);
887 qrel
->r_addend
= *(long long *)ptr
+ val
;
891 *(long long *)ptr
+= val
;
895 if (s1
->output_type
== TCC_OUTPUT_DLL
) {
896 /* XXX: this logic may depend on TCC's codegen
897 now TCC uses R_X86_64_32 even for a 64bit pointer */
898 qrel
->r_info
= ELFW(R_INFO
)(0, R_X86_64_RELATIVE
);
899 qrel
->r_addend
= *(int *)ptr
+ val
;
906 if (s1
->output_type
== TCC_OUTPUT_DLL
) {
908 esym_index
= s1
->symtab_to_dynsym
[sym_index
];
910 qrel
->r_offset
= rel
->r_offset
;
911 qrel
->r_info
= ELFW(R_INFO
)(esym_index
, R_X86_64_PC32
);
912 qrel
->r_addend
= *(int *)ptr
;
920 /* We've put the PLT slot offset into r_addend when generating
921 it, and that's what we must use as relocation value (adjusted
922 by section offset of course). */
923 val
= s1
->plt
->sh_addr
+ rel
->r_addend
;
929 diff
= (long long)val
- addr
;
930 if (diff
< -2147483648LL || diff
> 2147483647LL) {
931 tcc_error("internal error: relocation failed");
936 case R_X86_64_GLOB_DAT
:
937 case R_X86_64_JUMP_SLOT
:
938 /* They don't need addend */
939 *(addr_t
*)ptr
= val
- rel
->r_addend
;
941 case R_X86_64_GOTPCREL
:
942 *(int *)ptr
+= (s1
->got
->sh_addr
- addr
+
943 s1
->sym_attrs
[sym_index
].got_offset
- 4);
945 case R_X86_64_GOTTPOFF
:
946 *(int *)ptr
+= val
- s1
->got
->sh_addr
;
949 /* we load the got offset */
950 *(int *)ptr
+= s1
->sym_attrs
[sym_index
].got_offset
;
953 #error unsupported processor
957 /* if the relocation is allocated, we change its symbol table */
958 if (sr
->sh_flags
& SHF_ALLOC
)
959 sr
->link
= s1
->dynsym
;
962 /* relocate relocation table in 'sr' */
963 static void relocate_rel(TCCState
*s1
, Section
*sr
)
968 s
= s1
->sections
[sr
->sh_info
];
969 for_each_elem(sr
, 0, rel
, ElfW_Rel
)
970 rel
->r_offset
+= s
->sh_addr
;
973 /* count the number of dynamic relocations so that we can reserve
975 static int prepare_dynamic_rel(TCCState
*s1
, Section
*sr
)
978 int sym_index
, esym_index
, type
, count
;
981 for_each_elem(sr
, 0, rel
, ElfW_Rel
) {
982 sym_index
= ELFW(R_SYM
)(rel
->r_info
);
983 type
= ELFW(R_TYPE
)(rel
->r_info
);
985 #if defined(TCC_TARGET_I386)
987 #elif defined(TCC_TARGET_X86_64)
994 #if defined(TCC_TARGET_I386)
996 #elif defined(TCC_TARGET_X86_64)
999 esym_index
= s1
->symtab_to_dynsym
[sym_index
];
1008 /* allocate the section */
1009 sr
->sh_flags
|= SHF_ALLOC
;
1010 sr
->sh_size
= count
* sizeof(ElfW_Rel
);
1015 static struct sym_attr
*alloc_sym_attr(TCCState
*s1
, int index
)
1018 struct sym_attr
*tab
;
1020 if (index
>= s1
->nb_sym_attrs
) {
1021 /* find immediately bigger power of 2 and reallocate array */
1025 tab
= tcc_realloc(s1
->sym_attrs
, n
* sizeof(*s1
->sym_attrs
));
1026 s1
->sym_attrs
= tab
;
1027 memset(s1
->sym_attrs
+ s1
->nb_sym_attrs
, 0,
1028 (n
- s1
->nb_sym_attrs
) * sizeof(*s1
->sym_attrs
));
1029 s1
->nb_sym_attrs
= n
;
1031 return &s1
->sym_attrs
[index
];
1034 /* XXX: suppress that */
1035 static void put32(unsigned char *p
, uint32_t val
)
1043 #if defined(TCC_TARGET_I386) || defined(TCC_TARGET_ARM) || \
1044 defined(TCC_TARGET_ARM64) || defined(TCC_TARGET_X86_64)
1045 static uint32_t get32(unsigned char *p
)
1047 return p
[0] | (p
[1] << 8) | (p
[2] << 16) | (p
[3] << 24);
1051 static void build_got(TCCState
*s1
)
1055 /* if no got, then create it */
1056 s1
->got
= new_section(s1
, ".got", SHT_PROGBITS
, SHF_ALLOC
| SHF_WRITE
);
1057 s1
->got
->sh_entsize
= 4;
1058 add_elf_sym(symtab_section
, 0, 4, ELFW(ST_INFO
)(STB_GLOBAL
, STT_OBJECT
),
1059 0, s1
->got
->sh_num
, "_GLOBAL_OFFSET_TABLE_");
1060 ptr
= section_ptr_add(s1
->got
, 3 * PTR_SIZE
);
1062 /* keep space for _DYNAMIC pointer, if present */
1064 /* two dummy got entries */
1068 /* keep space for _DYNAMIC pointer, if present */
1071 /* two dummy got entries */
1079 /* put a got or plt entry corresponding to a symbol in symtab_section. 'size'
1080 and 'info' can be modifed if more precise info comes from the DLL.
1081 Returns offset of GOT or PLT slot. */
1082 static unsigned long put_got_entry(TCCState
*s1
,
1083 int reloc_type
, unsigned long size
, int info
,
1086 int index
, need_plt_entry
;
1089 unsigned long offset
;
1091 struct sym_attr
*symattr
;
1097 #ifdef TCC_TARGET_X86_64
1098 (reloc_type
== R_X86_64_JUMP_SLOT
);
1099 #elif defined(TCC_TARGET_I386)
1100 (reloc_type
== R_386_JMP_SLOT
);
1101 #elif defined(TCC_TARGET_ARM)
1102 (reloc_type
== R_ARM_JUMP_SLOT
);
1103 #elif defined(TCC_TARGET_ARM64)
1104 (reloc_type
== R_AARCH64_JUMP_SLOT
);
1109 if (need_plt_entry
&& !s1
->plt
) {
1111 s1
->plt
= new_section(s1
, ".plt", SHT_PROGBITS
,
1112 SHF_ALLOC
| SHF_EXECINSTR
);
1113 s1
->plt
->sh_entsize
= 4;
1116 /* If a got/plt entry already exists for that symbol, no need to add one */
1117 if (sym_index
< s1
->nb_sym_attrs
) {
1118 if (need_plt_entry
&& s1
->sym_attrs
[sym_index
].plt_offset
)
1119 return s1
->sym_attrs
[sym_index
].plt_offset
;
1120 else if (!need_plt_entry
&& s1
->sym_attrs
[sym_index
].got_offset
)
1121 return s1
->sym_attrs
[sym_index
].got_offset
;
1124 symattr
= alloc_sym_attr(s1
, sym_index
);
1126 /* Only store the GOT offset if it's not generated for the PLT entry. */
1127 if (!need_plt_entry
)
1128 symattr
->got_offset
= s1
->got
->data_offset
;
1130 sym
= &((ElfW(Sym
) *)symtab_section
->data
)[sym_index
];
1131 name
= (char *) symtab_section
->link
->data
+ sym
->st_name
;
1132 offset
= sym
->st_value
;
1133 #if defined(TCC_TARGET_I386) || defined(TCC_TARGET_X86_64)
1134 if (need_plt_entry
) {
1138 unsigned long relofs
;
1140 #if defined(TCC_OUTPUT_DLL_WITH_PLT)
1143 /* if we build a DLL, we add a %ebx offset */
1144 if (s1
->output_type
== TCC_OUTPUT_DLL
)
1150 /* add a PLT entry */
1152 if (plt
->data_offset
== 0) {
1153 /* first plt entry */
1154 p
= section_ptr_add(plt
, 16);
1155 p
[0] = 0xff; /* pushl got + PTR_SIZE */
1156 p
[1] = modrm
+ 0x10;
1157 put32(p
+ 2, PTR_SIZE
);
1158 p
[6] = 0xff; /* jmp *(got + PTR_SIZE * 2) */
1160 put32(p
+ 8, PTR_SIZE
* 2);
1163 /* The PLT slot refers to the relocation entry it needs
1164 via offset. The reloc entry is created below, so its
1165 offset is the current data_offset. */
1166 relofs
= s1
->got
->reloc
? s1
->got
->reloc
->data_offset
: 0;
1167 symattr
->plt_offset
= plt
->data_offset
;
1168 p
= section_ptr_add(plt
, 16);
1169 p
[0] = 0xff; /* jmp *(got + x) */
1171 put32(p
+ 2, s1
->got
->data_offset
);
1172 p
[6] = 0x68; /* push $xxx */
1173 #ifdef TCC_TARGET_X86_64
1174 /* On x86-64, the relocation is referred to by _index_. */
1175 put32(p
+ 7, relofs
/ sizeof (ElfW_Rel
));
1177 put32(p
+ 7, relofs
);
1179 p
[11] = 0xe9; /* jmp plt_start */
1180 put32(p
+ 12, -(plt
->data_offset
));
1182 /* If this was an UNDEF symbol set the offset in the
1183 dynsymtab to the PLT slot, so that PC32 relocs to it
1185 if (sym
->st_shndx
== SHN_UNDEF
)
1186 offset
= plt
->data_offset
- 16;
1188 #elif defined(TCC_TARGET_ARM)
1189 if (need_plt_entry
) {
1193 /* if we build a DLL, we add a %ebx offset */
1194 if (s1
->output_type
== TCC_OUTPUT_DLL
)
1195 tcc_error("DLLs unimplemented!");
1197 /* add a PLT entry */
1199 if (plt
->data_offset
== 0) {
1200 /* first plt entry */
1201 p
= section_ptr_add(plt
, 16);
1202 put32(p
, 0xe52de004); /* push {lr} */
1203 put32(p
+4, 0xe59fe010); /* ldr lr, [pc, #16] */
1204 put32(p
+8, 0xe08fe00e); /* add lr, pc, lr */
1205 put32(p
+12, 0xe5bef008); /* ldr pc, [lr, #8]! */
1208 symattr
->plt_offset
= plt
->data_offset
;
1209 if (symattr
->plt_thumb_stub
) {
1210 p
= section_ptr_add(plt
, 20);
1211 put32(p
, 0x4778); /* bx pc */
1212 put32(p
+2, 0x46c0); /* nop */
1215 p
= section_ptr_add(plt
, 16);
1216 put32(p
, 0xe59fc004); /* ldr ip, [pc, #4] ; GOT entry offset */
1217 put32(p
+4, 0xe08fc00c); /* add ip, pc, ip ; addr of GOT entry */
1218 put32(p
+8, 0xe59cf000); /* ldr pc, [ip] ; jump to GOT entry */
1219 put32(p
+12, s1
->got
->data_offset
); /* GOT entry off once patched */
1221 /* the symbol is modified so that it will be relocated to
1223 if (sym
->st_shndx
== SHN_UNDEF
)
1224 offset
= plt
->data_offset
- 16;
1226 #elif defined(TCC_TARGET_ARM64)
1227 if (need_plt_entry
) {
1231 if (s1
->output_type
== TCC_OUTPUT_DLL
)
1232 tcc_error("DLLs unimplemented!");
1235 if (plt
->data_offset
== 0)
1236 section_ptr_add(plt
, 32);
1237 symattr
->plt_offset
= plt
->data_offset
;
1238 p
= section_ptr_add(plt
, 16);
1239 put32(p
, s1
->got
->data_offset
);
1240 put32(p
+ 4, (uint64_t)s1
->got
->data_offset
>> 32);
1242 if (sym
->st_shndx
== SHN_UNDEF
)
1243 offset
= plt
->data_offset
- 16;
1245 #elif defined(TCC_TARGET_C67)
1247 tcc_error("C67 got not implemented");
1250 #error unsupported CPU
1253 /* XXX This might generate multiple syms for name. */
1254 index
= put_elf_sym(s1
->dynsym
, offset
,
1255 size
, info
, 0, sym
->st_shndx
, name
);
1256 /* Create the relocation (it's against the GOT for PLT
1258 put_elf_reloc(s1
->dynsym
, s1
->got
,
1259 s1
->got
->data_offset
,
1262 /* Without .dynsym (i.e. static link or memory output) we
1263 still need relocs against the generated got, so as to fill
1264 the entries with the symbol values (determined later). */
1265 put_elf_reloc(symtab_section
, s1
->got
,
1266 s1
->got
->data_offset
,
1267 reloc_type
, sym_index
);
1269 /* And now create the GOT slot itself. */
1270 ptr
= section_ptr_add(s1
->got
, PTR_SIZE
);
1273 return symattr
->plt_offset
;
1275 return symattr
->got_offset
;
1278 /* build GOT and PLT entries */
1279 ST_FUNC
void build_got_entries(TCCState
*s1
)
1284 int i
, type
, reloc_type
, sym_index
;
1286 for(i
= 1; i
< s1
->nb_sections
; i
++) {
1287 s
= s1
->sections
[i
];
1288 if (s
->sh_type
!= SHT_RELX
)
1290 /* no need to handle got relocations */
1291 if (s
->link
!= symtab_section
)
1293 for_each_elem(s
, 0, rel
, ElfW_Rel
) {
1294 type
= ELFW(R_TYPE
)(rel
->r_info
);
1296 #if defined(TCC_TARGET_I386)
1303 if (type
== R_386_GOT32
|| type
== R_386_PLT32
) {
1304 sym_index
= ELFW(R_SYM
)(rel
->r_info
);
1305 sym
= &((ElfW(Sym
) *)symtab_section
->data
)[sym_index
];
1306 /* look at the symbol got offset. If none, then add one */
1307 if (type
== R_386_GOT32
)
1308 reloc_type
= R_386_GLOB_DAT
;
1310 reloc_type
= R_386_JMP_SLOT
;
1311 put_got_entry(s1
, reloc_type
, sym
->st_size
, sym
->st_info
,
1315 #elif defined(TCC_TARGET_ARM)
1325 sym_index
= ELFW(R_SYM
)(rel
->r_info
);
1326 sym
= &((ElfW(Sym
) *)symtab_section
->data
)[sym_index
];
1327 if (type
!= R_ARM_GOTOFF
&& type
!= R_ARM_GOTPC
1328 && sym
->st_shndx
== SHN_UNDEF
) {
1330 /* look at the symbol got offset. If none, then add one */
1331 if (type
== R_ARM_GOT32
)
1332 reloc_type
= R_ARM_GLOB_DAT
;
1334 reloc_type
= R_ARM_JUMP_SLOT
;
1335 ofs
= put_got_entry(s1
, reloc_type
, sym
->st_size
,
1336 sym
->st_info
, sym_index
);
1338 printf ("maybegot: %s, %d, %d --> ofs=0x%x\n",
1339 (char *) symtab_section
->link
->data
+ sym
->st_name
,
1340 type
, sym
->st_shndx
, ofs
);
1342 if (type
!= R_ARM_GOT32
) {
1343 addr_t
*ptr
= (addr_t
*)(s1
->sections
[s
->sh_info
]->data
1345 /* x must be signed! */
1346 int x
= *ptr
& 0xffffff;
1352 printf ("insn=0x%x --> 0x%x (x==0x%x)\n", *ptr
,
1353 (*ptr
& 0xff000000) | x
, x
);
1355 *ptr
= (*ptr
& 0xff000000) | x
;
1359 case R_ARM_THM_JUMP24
:
1360 sym_index
= ELFW(R_SYM
)(rel
->r_info
);
1361 sym
= &((ElfW(Sym
) *)symtab_section
->data
)[sym_index
];
1362 /* We are relocating a jump from thumb code to arm code */
1363 if (sym
->st_shndx
!= SHN_UNDEF
&& !(sym
->st_value
& 1)) {
1366 char *name
, buf
[1024];
1367 Section
*text_section
;
1369 name
= (char *) symtab_section
->link
->data
+ sym
->st_name
;
1370 text_section
= s1
->sections
[sym
->st_shndx
];
1371 /* Modify reloc to target a thumb stub to switch to ARM */
1372 snprintf(buf
, sizeof(buf
), "%s_from_thumb", name
);
1373 index
= put_elf_sym(symtab_section
,
1374 text_section
->data_offset
+ 1,
1375 sym
->st_size
, sym
->st_info
, 0,
1376 sym
->st_shndx
, buf
);
1377 rel
->r_info
= ELFW(R_INFO
)(index
, type
);
1378 /* Create a thumb stub fonction to switch to ARM mode */
1379 put_elf_reloc(symtab_section
, text_section
,
1380 text_section
->data_offset
+ 4, R_ARM_JUMP24
,
1382 p
= section_ptr_add(text_section
, 8);
1383 put32(p
, 0x4778); /* bx pc */
1384 put32(p
+2, 0x46c0); /* nop */
1385 put32(p
+4, 0xeafffffe); /* b $sym */
1387 #elif defined(TCC_TARGET_ARM64)
1388 //xx Other cases may be required here:
1389 case R_AARCH64_ADR_GOT_PAGE
:
1390 case R_AARCH64_LD64_GOT_LO12_NC
:
1393 sym_index
= ELFW(R_SYM
)(rel
->r_info
);
1394 sym
= &((ElfW(Sym
) *)symtab_section
->data
)[sym_index
];
1395 reloc_type
= R_AARCH64_GLOB_DAT
;
1396 put_got_entry(s1
, reloc_type
, sym
->st_size
, sym
->st_info
,
1400 case R_AARCH64_JUMP26
:
1401 case R_AARCH64_CALL26
:
1404 sym_index
= ELFW(R_SYM
)(rel
->r_info
);
1405 sym
= &((ElfW(Sym
) *)symtab_section
->data
)[sym_index
];
1406 if (sym
->st_shndx
== SHN_UNDEF
) {
1408 reloc_type
= R_AARCH64_JUMP_SLOT
;
1409 ofs
= put_got_entry(s1
, reloc_type
, sym
->st_size
,
1410 sym
->st_info
, sym_index
);
1411 /* We store the place of the generated PLT slot
1413 rel
->r_addend
+= ofs
;
1416 #elif defined(TCC_TARGET_C67)
1423 if (type
== R_C60_GOT32
|| type
== R_C60_PLT32
) {
1424 sym_index
= ELFW(R_SYM
)(rel
->r_info
);
1425 sym
= &((ElfW(Sym
) *)symtab_section
->data
)[sym_index
];
1426 /* look at the symbol got offset. If none, then add one */
1427 if (type
== R_C60_GOT32
)
1428 reloc_type
= R_C60_GLOB_DAT
;
1430 reloc_type
= R_C60_JMP_SLOT
;
1431 put_got_entry(s1
, reloc_type
, sym
->st_size
, sym
->st_info
,
1435 #elif defined(TCC_TARGET_X86_64)
1436 case R_X86_64_GOT32
:
1437 case R_X86_64_GOTTPOFF
:
1438 case R_X86_64_GOTPCREL
:
1439 case R_X86_64_PLT32
:
1440 sym_index
= ELFW(R_SYM
)(rel
->r_info
);
1441 sym
= &((ElfW(Sym
) *)symtab_section
->data
)[sym_index
];
1442 if (type
== R_X86_64_PLT32
&&
1443 ELFW(ST_VISIBILITY
)(sym
->st_other
) != STV_DEFAULT
)
1445 rel
->r_info
= ELFW(R_INFO
)(sym_index
, R_X86_64_PC32
);
1451 if (type
== R_X86_64_GOT32
|| type
== R_X86_64_GOTPCREL
||
1452 type
== R_X86_64_PLT32
) {
1454 /* look at the symbol got offset. If none, then add one */
1455 if (type
== R_X86_64_GOT32
|| type
== R_X86_64_GOTPCREL
)
1456 reloc_type
= R_X86_64_GLOB_DAT
;
1458 reloc_type
= R_X86_64_JUMP_SLOT
;
1459 ofs
= put_got_entry(s1
, reloc_type
, sym
->st_size
,
1460 sym
->st_info
, sym_index
);
1461 if (type
== R_X86_64_PLT32
)
1462 /* We store the place of the generated PLT slot
1464 rel
->r_addend
+= ofs
;
1468 #error unsupported CPU
1477 ST_FUNC Section
*new_symtab(TCCState
*s1
,
1478 const char *symtab_name
, int sh_type
, int sh_flags
,
1479 const char *strtab_name
,
1480 const char *hash_name
, int hash_sh_flags
)
1482 Section
*symtab
, *strtab
, *hash
;
1483 int *ptr
, nb_buckets
;
1485 symtab
= new_section(s1
, symtab_name
, sh_type
, sh_flags
);
1486 symtab
->sh_entsize
= sizeof(ElfW(Sym
));
1487 strtab
= new_section(s1
, strtab_name
, SHT_STRTAB
, sh_flags
);
1488 put_elf_str(strtab
, "");
1489 symtab
->link
= strtab
;
1490 put_elf_sym(symtab
, 0, 0, 0, 0, 0, NULL
);
1494 hash
= new_section(s1
, hash_name
, SHT_HASH
, hash_sh_flags
);
1495 hash
->sh_entsize
= sizeof(int);
1496 symtab
->hash
= hash
;
1497 hash
->link
= symtab
;
1499 ptr
= section_ptr_add(hash
, (2 + nb_buckets
+ 1) * sizeof(int));
1500 ptr
[0] = nb_buckets
;
1502 memset(ptr
+ 2, 0, (nb_buckets
+ 1) * sizeof(int));
1506 /* put dynamic tag */
1507 static void put_dt(Section
*dynamic
, int dt
, addr_t val
)
1510 dyn
= section_ptr_add(dynamic
, sizeof(ElfW(Dyn
)));
1512 dyn
->d_un
.d_val
= val
;
1515 static void add_init_array_defines(TCCState
*s1
, const char *section_name
)
1519 char sym_start
[1024];
1522 snprintf(sym_start
, sizeof(sym_start
), "__%s_start", section_name
+ 1);
1523 snprintf(sym_end
, sizeof(sym_end
), "__%s_end", section_name
+ 1);
1525 s
= find_section(s1
, section_name
);
1530 end_offset
= s
->data_offset
;
1533 add_elf_sym(symtab_section
,
1535 ELFW(ST_INFO
)(STB_GLOBAL
, STT_NOTYPE
), 0,
1536 s
->sh_num
, sym_start
);
1537 add_elf_sym(symtab_section
,
1539 ELFW(ST_INFO
)(STB_GLOBAL
, STT_NOTYPE
), 0,
1540 s
->sh_num
, sym_end
);
1543 static int tcc_add_support(TCCState
*s1
, const char *filename
)
1546 snprintf(buf
, sizeof(buf
), "%s/%s/%s", s1
->tcc_lib_path
,
1547 /* an cpu specific path inside tcc_lib_path, mainly for keeping libtcc1.a */
1548 #ifdef TCC_TARGET_I386
1551 #ifdef TCC_TARGET_X86_64
1554 #ifdef TCC_TARGET_ARM
1557 #ifdef TCC_TARGET_ARM64
1560 #ifdef TCC_TARGET_C67
1565 return tcc_add_file(s1
, buf
);
1568 ST_FUNC
void tcc_add_bcheck(TCCState
*s1
)
1570 #ifdef CONFIG_TCC_BCHECK
1573 if (0 == s1
->do_bounds_check
)
1576 /* XXX: add an object file to do that */
1577 ptr
= section_ptr_add(bounds_section
, sizeof(*ptr
));
1579 add_elf_sym(symtab_section
, 0, 0,
1580 ELFW(ST_INFO
)(STB_GLOBAL
, STT_NOTYPE
), 0,
1581 bounds_section
->sh_num
, "__bounds_start");
1582 #ifdef TCC_TARGET_I386
1583 if (s1
->output_type
!= TCC_OUTPUT_MEMORY
) {
1584 /* add 'call __bound_init()' in .init section */
1586 /* XXX not called on MSYS, reason is unknown. For this
1587 case a call to __bound_init is performed in bcheck.c
1588 when __bound_ptr_add, __bound_new_region,
1589 __bound_delete_region called */
1591 int sym_index
= find_elf_sym(symtab_section
, "__bound_init");
1593 Section
*init_section
= find_section(s1
, ".init");
1594 unsigned char *pinit
= section_ptr_add(init_section
, 5);
1596 put32(pinit
+ 1, -4);
1597 put_elf_reloc(symtab_section
, init_section
,
1598 init_section
->data_offset
- 4, R_386_PC32
, sym_index
);
1601 tcc_warning("__bound_init not defined");
1607 /* add tcc runtime libraries */
1608 ST_FUNC
void tcc_add_runtime(TCCState
*s1
)
1611 if (!s1
->nostdlib
) {
1612 tcc_add_library(s1
, "c");
1613 #ifdef CONFIG_USE_LIBGCC
1614 if (!s1
->static_link
) {
1615 tcc_add_file(s1
, TCC_LIBGCC
);
1618 tcc_add_support(s1
, "libtcc1.a");
1621 /* tcc_add_bcheck tries to relocate a call to __bound_init in _init so
1622 libtcc1.a must be loaded before for __bound_init to be defined and
1623 crtn.o must be loaded after to not finalize _init too early. */
1626 if (!s1
->nostdlib
) {
1627 /* add crt end if not memory output */
1628 if (s1
->output_type
!= TCC_OUTPUT_MEMORY
)
1629 tcc_add_crt(s1
, "crtn.o");
1633 /* add various standard linker symbols (must be done after the
1634 sections are filled (for example after allocating common
1636 ST_FUNC
void tcc_add_linker_symbols(TCCState
*s1
)
1642 add_elf_sym(symtab_section
,
1643 text_section
->data_offset
, 0,
1644 ELFW(ST_INFO
)(STB_GLOBAL
, STT_NOTYPE
), 0,
1645 text_section
->sh_num
, "_etext");
1646 add_elf_sym(symtab_section
,
1647 data_section
->data_offset
, 0,
1648 ELFW(ST_INFO
)(STB_GLOBAL
, STT_NOTYPE
), 0,
1649 data_section
->sh_num
, "_edata");
1650 add_elf_sym(symtab_section
,
1651 bss_section
->data_offset
, 0,
1652 ELFW(ST_INFO
)(STB_GLOBAL
, STT_NOTYPE
), 0,
1653 bss_section
->sh_num
, "_end");
1654 /* horrible new standard ldscript defines */
1655 add_init_array_defines(s1
, ".preinit_array");
1656 add_init_array_defines(s1
, ".init_array");
1657 add_init_array_defines(s1
, ".fini_array");
1659 /* add start and stop symbols for sections whose name can be
1661 for(i
= 1; i
< s1
->nb_sections
; i
++) {
1662 s
= s1
->sections
[i
];
1663 if (s
->sh_type
== SHT_PROGBITS
&&
1664 (s
->sh_flags
& SHF_ALLOC
)) {
1668 /* check if section name can be expressed in C */
1674 if (!isid(ch
) && !isnum(ch
))
1678 snprintf(buf
, sizeof(buf
), "__start_%s", s
->name
);
1679 add_elf_sym(symtab_section
,
1681 ELFW(ST_INFO
)(STB_GLOBAL
, STT_NOTYPE
), 0,
1683 snprintf(buf
, sizeof(buf
), "__stop_%s", s
->name
);
1684 add_elf_sym(symtab_section
,
1686 ELFW(ST_INFO
)(STB_GLOBAL
, STT_NOTYPE
), 0,
1693 static void tcc_output_binary(TCCState
*s1
, FILE *f
,
1694 const int *sec_order
)
1697 int i
, offset
, size
;
1700 for(i
=1;i
<s1
->nb_sections
;i
++) {
1701 s
= s1
->sections
[sec_order
[i
]];
1702 if (s
->sh_type
!= SHT_NOBITS
&&
1703 (s
->sh_flags
& SHF_ALLOC
)) {
1704 while (offset
< s
->sh_offset
) {
1709 fwrite(s
->data
, 1, size
, f
);
1715 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
1717 #define EXTRA_RELITEMS 14
1719 /* move the relocation value from .dynsym to .got */
1720 void patch_dynsym_undef(TCCState
*s1
, Section
*s
)
1722 uint32_t *gotd
= (void *)s1
->got
->data
;
1725 gotd
+= 3; /* dummy entries in .got */
1726 /* relocate symbols in .dynsym */
1727 for_each_elem(s
, 1, sym
, ElfW(Sym
)) {
1728 if (sym
->st_shndx
== SHN_UNDEF
) {
1729 *gotd
++ = sym
->st_value
+ 6; /* XXX 6 is magic ? */
1736 #define EXTRA_RELITEMS 9
1738 /* zero plt offsets of weak symbols in .dynsym */
1739 void patch_dynsym_undef(TCCState
*s1
, Section
*s
)
1743 for_each_elem(s
, 1, sym
, ElfW(Sym
))
1744 if (sym
->st_shndx
== SHN_UNDEF
&& ELFW(ST_BIND
)(sym
->st_info
) == STB_WEAK
)
1749 ST_FUNC
void fill_got_entry(TCCState
*s1
, ElfW_Rel
*rel
)
1751 int sym_index
= ELFW(R_SYM
) (rel
->r_info
);
1752 ElfW(Sym
) *sym
= &((ElfW(Sym
) *) symtab_section
->data
)[sym_index
];
1753 unsigned long offset
;
1755 if (sym_index
>= s1
->nb_sym_attrs
)
1757 offset
= s1
->sym_attrs
[sym_index
].got_offset
;
1758 section_reserve(s1
->got
, offset
+ PTR_SIZE
);
1759 #ifdef TCC_TARGET_X86_64
1760 /* only works for x86-64 */
1761 put32(s1
->got
->data
+ offset
+ 4, sym
->st_value
>> 32);
1763 put32(s1
->got
->data
+ offset
, sym
->st_value
& 0xffffffff);
1766 /* Perform relocation to GOT or PLT entries */
1767 ST_FUNC
void fill_got(TCCState
*s1
)
1773 for(i
= 1; i
< s1
->nb_sections
; i
++) {
1774 s
= s1
->sections
[i
];
1775 if (s
->sh_type
!= SHT_RELX
)
1777 /* no need to handle got relocations */
1778 if (s
->link
!= symtab_section
)
1780 for_each_elem(s
, 0, rel
, ElfW_Rel
) {
1781 switch (ELFW(R_TYPE
) (rel
->r_info
)) {
1782 case R_X86_64_GOT32
:
1783 case R_X86_64_GOTPCREL
:
1784 case R_X86_64_PLT32
:
1785 fill_got_entry(s1
, rel
);
1792 /* Bind symbols of executable: resolve undefined symbols from exported symbols
1793 in shared libraries and export non local defined symbols to shared libraries
1794 if -rdynamic switch was given on command line */
1795 static void bind_exe_dynsyms(TCCState
*s1
)
1798 int sym_index
, index
;
1799 ElfW(Sym
) *sym
, *esym
;
1802 /* Resolve undefined symbols from dynamic symbols. When there is a match:
1803 - if STT_FUNC or STT_GNU_IFUNC symbol -> add it in PLT
1804 - if STT_OBJECT symbol -> add it in .bss section with suitable reloc */
1805 for_each_elem(symtab_section
, 1, sym
, ElfW(Sym
)) {
1806 if (sym
->st_shndx
== SHN_UNDEF
) {
1807 name
= (char *) symtab_section
->link
->data
+ sym
->st_name
;
1808 sym_index
= find_elf_sym(s1
->dynsymtab_section
, name
);
1810 esym
= &((ElfW(Sym
) *)s1
->dynsymtab_section
->data
)[sym_index
];
1811 type
= ELFW(ST_TYPE
)(esym
->st_info
);
1812 if ((type
== STT_FUNC
) || (type
== STT_GNU_IFUNC
)) {
1813 /* Indirect functions shall have STT_FUNC type in executable
1814 * dynsym section. Indeed, a dlsym call following a lazy
1815 * resolution would pick the symbol value from the
1816 * executable dynsym entry which would contain the address
1817 * of the function wanted by the caller of dlsym instead of
1818 * the address of the function that would return that
1820 put_got_entry(s1
, R_JMP_SLOT
, esym
->st_size
,
1821 ELFW(ST_INFO
)(STB_GLOBAL
,STT_FUNC
),
1822 sym
- (ElfW(Sym
) *)symtab_section
->data
);
1823 } else if (type
== STT_OBJECT
) {
1824 unsigned long offset
;
1826 offset
= bss_section
->data_offset
;
1827 /* XXX: which alignment ? */
1828 offset
= (offset
+ 16 - 1) & -16;
1829 index
= put_elf_sym(s1
->dynsym
, offset
, esym
->st_size
,
1830 esym
->st_info
, 0, bss_section
->sh_num
,
1832 /* Ensure R_COPY works for weak symbol aliases */
1833 if (ELFW(ST_BIND
)(esym
->st_info
) == STB_WEAK
) {
1834 for_each_elem(s1
->dynsymtab_section
, 1, dynsym
, ElfW(Sym
)) {
1835 if ((dynsym
->st_value
== esym
->st_value
)
1836 && (ELFW(ST_BIND
)(dynsym
->st_info
) == STB_GLOBAL
)) {
1837 char *dynname
= (char *) s1
->dynsymtab_section
->link
->data
1839 put_elf_sym(s1
->dynsym
, offset
, dynsym
->st_size
,
1841 bss_section
->sh_num
, dynname
);
1846 put_elf_reloc(s1
->dynsym
, bss_section
,
1847 offset
, R_COPY
, index
);
1848 offset
+= esym
->st_size
;
1849 bss_section
->data_offset
= offset
;
1852 /* STB_WEAK undefined symbols are accepted */
1853 /* XXX: _fp_hw seems to be part of the ABI, so we ignore it */
1854 if (ELFW(ST_BIND
)(sym
->st_info
) == STB_WEAK
||
1855 !strcmp(name
, "_fp_hw")) {
1857 tcc_error_noabort("undefined symbol '%s'", name
);
1860 } else if (s1
->rdynamic
&& ELFW(ST_BIND
)(sym
->st_info
) != STB_LOCAL
) {
1861 /* if -rdynamic option, then export all non local symbols */
1862 name
= (char *) symtab_section
->link
->data
+ sym
->st_name
;
1863 put_elf_sym(s1
->dynsym
, sym
->st_value
, sym
->st_size
, sym
->st_info
,
1864 0, sym
->st_shndx
, name
);
1869 /* Bind symbols of libraries: export non local symbols of executable that
1870 resolve undefined symbols of shared libraries */
1871 static void bind_libs_dynsyms(TCCState
*s1
)
1875 ElfW(Sym
) *sym
, *esym
;
1877 /* now look at unresolved dynamic symbols and export
1878 corresponding symbol */
1879 for_each_elem(s1
->dynsymtab_section
, 1, esym
, ElfW(Sym
)) {
1880 if (esym
->st_shndx
== SHN_UNDEF
) {
1881 name
= (char *) s1
->dynsymtab_section
->link
->data
+ esym
->st_name
;
1882 sym_index
= find_elf_sym(symtab_section
, name
);
1884 /* XXX: avoid adding a symbol if already present because of
1886 sym
= &((ElfW(Sym
) *)symtab_section
->data
)[sym_index
];
1887 put_elf_sym(s1
->dynsym
, sym
->st_value
, sym
->st_size
,
1888 sym
->st_info
, 0, sym
->st_shndx
, name
);
1890 /* weak symbols can stay undefined */
1891 if (ELFW(ST_BIND
)(esym
->st_info
) != STB_WEAK
)
1892 tcc_warning("undefined dynamic symbol '%s'", name
);
1898 /* Export all non local symbols (for shared libraries) */
1899 static void export_global_syms(TCCState
*s1
)
1901 int nb_syms
, dynindex
, index
;
1905 nb_syms
= symtab_section
->data_offset
/ sizeof(ElfW(Sym
));
1906 s1
->symtab_to_dynsym
= tcc_mallocz(sizeof(int) * nb_syms
);
1907 for_each_elem(symtab_section
, 1, sym
, ElfW(Sym
)) {
1908 if (ELFW(ST_BIND
)(sym
->st_info
) != STB_LOCAL
) {
1909 name
= (char *) symtab_section
->link
->data
+ sym
->st_name
;
1910 dynindex
= put_elf_sym(s1
->dynsym
, sym
->st_value
, sym
->st_size
,
1911 sym
->st_info
, 0, sym
->st_shndx
, name
);
1912 index
= sym
- (ElfW(Sym
) *) symtab_section
->data
;
1913 s1
->symtab_to_dynsym
[index
] = dynindex
;
1918 /* relocate the PLT: compute addresses and offsets in the PLT now that final
1919 address for PLT and GOT are known (see fill_program_header) */
1920 ST_FUNC
void relocate_plt(TCCState
*s1
)
1928 p_end
= p
+ s1
->plt
->data_offset
;
1930 #if defined(TCC_TARGET_I386)
1931 put32(p
+ 2, get32(p
+ 2) + s1
->got
->sh_addr
);
1932 put32(p
+ 8, get32(p
+ 8) + s1
->got
->sh_addr
);
1935 put32(p
+ 2, get32(p
+ 2) + s1
->got
->sh_addr
);
1938 #elif defined(TCC_TARGET_X86_64)
1939 int x
= s1
->got
->sh_addr
- s1
->plt
->sh_addr
- 6;
1940 put32(p
+ 2, get32(p
+ 2) + x
);
1941 put32(p
+ 8, get32(p
+ 8) + x
- 6);
1944 put32(p
+ 2, get32(p
+ 2) + x
+ s1
->plt
->data
- p
);
1947 #elif defined(TCC_TARGET_ARM)
1949 x
=s1
->got
->sh_addr
- s1
->plt
->sh_addr
- 12;
1952 if (get32(p
) == 0x46c04778) /* PLT Thumb stub present */
1954 put32(p
+ 12, x
+ get32(p
+ 12) + s1
->plt
->data
- p
);
1957 #elif defined(TCC_TARGET_ARM64)
1958 uint64_t plt
= s1
->plt
->sh_addr
;
1959 uint64_t got
= s1
->got
->sh_addr
;
1960 uint64_t off
= (got
>> 12) - (plt
>> 12);
1961 if ((off
+ ((uint32_t)1 << 20)) >> 21)
1962 tcc_error("Failed relocating PLT (off=0x%lx, got=0x%lx, plt=0x%lx)", off
, got
, plt
);
1963 put32(p
, 0xa9bf7bf0); // stp x16,x30,[sp,#-16]!
1964 put32(p
+ 4, (0x90000010 | // adrp x16,...
1965 (off
& 0x1ffffc) << 3 | (off
& 3) << 29));
1966 put32(p
+ 8, (0xf9400211 | // ldr x17,[x16,#...]
1967 (got
& 0xff8) << 7));
1968 put32(p
+ 12, (0x91000210 | // add x16,x16,#...
1969 (got
& 0xfff) << 10));
1970 put32(p
+ 16, 0xd61f0220); // br x17
1971 put32(p
+ 20, 0xd503201f); // nop
1972 put32(p
+ 24, 0xd503201f); // nop
1973 put32(p
+ 28, 0xd503201f); // nop
1976 uint64_t pc
= plt
+ (p
- s1
->plt
->data
);
1977 uint64_t addr
= got
+
1978 (get32(p
) | (uint64_t)get32(p
+ 4) << 32);
1979 uint32_t off
= (addr
>> 12) - (pc
>> 12);
1980 if ((off
+ ((uint32_t)1 << 20)) >> 21)
1981 tcc_error("Failed relocating PLT (off=0x%lx, addr=0x%lx, pc=0x%lx)", off
, addr
, pc
);
1982 put32(p
, (0x90000010 | // adrp x16,...
1983 (off
& 0x1ffffc) << 3 | (off
& 3) << 29));
1984 put32(p
+ 4, (0xf9400211 | // ldr x17,[x16,#...]
1985 (addr
& 0xff8) << 7));
1986 put32(p
+ 8, (0x91000210 | // add x16,x16,#...
1987 (addr
& 0xfff) << 10));
1988 put32(p
+ 12, 0xd61f0220); // br x17
1991 #elif defined(TCC_TARGET_C67)
1994 #error unsupported CPU
1999 /* Allocate strings for section names and decide if an unallocated section
2002 NOTE: the strsec section comes last, so its size is also correct ! */
2003 static void alloc_sec_names(TCCState
*s1
, int file_type
, Section
*strsec
)
2008 /* Allocate strings for section names */
2009 for(i
= 1; i
< s1
->nb_sections
; i
++) {
2010 s
= s1
->sections
[i
];
2011 s
->sh_name
= put_elf_str(strsec
, s
->name
);
2012 /* when generating a DLL, we include relocations but we may
2014 if (file_type
== TCC_OUTPUT_DLL
&&
2015 s
->sh_type
== SHT_RELX
&&
2016 !(s
->sh_flags
& SHF_ALLOC
)) {
2017 /* gr: avoid bogus relocs for empty (debug) sections */
2018 if (s1
->sections
[s
->sh_info
]->sh_flags
& SHF_ALLOC
)
2019 prepare_dynamic_rel(s1
, s
);
2020 else if (s1
->do_debug
)
2021 s
->sh_size
= s
->data_offset
;
2022 } else if (s1
->do_debug
||
2023 file_type
== TCC_OUTPUT_OBJ
||
2024 file_type
== TCC_OUTPUT_EXE
||
2025 (s
->sh_flags
& SHF_ALLOC
) ||
2026 i
== (s1
->nb_sections
- 1)) {
2027 /* we output all sections if debug or object file */
2028 s
->sh_size
= s
->data_offset
;
2033 /* Info to be copied in dynamic section */
2037 unsigned long dyn_rel_off
;
2040 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
2046 /* Assign sections to segments and decide how are sections laid out when loaded
2047 in memory. This function also fills corresponding program headers. */
2048 static int layout_sections(TCCState
*s1
, ElfW(Phdr
) *phdr
, int phnum
,
2049 Section
*interp
, Section
* strsec
,
2050 struct dyn_inf
*dyninf
, int *sec_order
)
2052 int i
, j
, k
, file_type
, sh_order_index
, file_offset
;
2053 unsigned long s_align
;
2059 file_type
= s1
->output_type
;
2062 if (s1
->output_format
== TCC_OUTPUT_FORMAT_ELF
)
2063 file_offset
= sizeof(ElfW(Ehdr
)) + phnum
* sizeof(ElfW(Phdr
));
2064 s_align
= ELF_PAGE_SIZE
;
2065 if (s1
->section_align
)
2066 s_align
= s1
->section_align
;
2069 if (s1
->has_text_addr
) {
2070 int a_offset
, p_offset
;
2071 addr
= s1
->text_addr
;
2072 /* we ensure that (addr % ELF_PAGE_SIZE) == file_offset %
2074 a_offset
= (int) (addr
& (s_align
- 1));
2075 p_offset
= file_offset
& (s_align
- 1);
2076 if (a_offset
< p_offset
)
2077 a_offset
+= s_align
;
2078 file_offset
+= (a_offset
- p_offset
);
2080 if (file_type
== TCC_OUTPUT_DLL
)
2083 addr
= ELF_START_ADDR
;
2084 /* compute address after headers */
2085 addr
+= (file_offset
& (s_align
- 1));
2089 /* Leave one program headers for the program interpreter and one for
2090 the program header table itself if needed. These are done later as
2091 they require section layout to be done first. */
2093 ph
+= 1 + HAVE_PHDR
;
2095 /* dynamic relocation table information, for .dynamic section */
2096 dyninf
->rel_addr
= dyninf
->rel_size
= 0;
2097 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
2098 dyninf
->bss_addr
= dyninf
->bss_size
= 0;
2101 for(j
= 0; j
< 2; j
++) {
2102 ph
->p_type
= PT_LOAD
;
2104 ph
->p_flags
= PF_R
| PF_X
;
2106 ph
->p_flags
= PF_R
| PF_W
;
2107 ph
->p_align
= s_align
;
2109 /* Decide the layout of sections loaded in memory. This must
2110 be done before program headers are filled since they contain
2111 info about the layout. We do the following ordering: interp,
2112 symbol tables, relocations, progbits, nobits */
2113 /* XXX: do faster and simpler sorting */
2114 for(k
= 0; k
< 5; k
++) {
2115 for(i
= 1; i
< s1
->nb_sections
; i
++) {
2116 s
= s1
->sections
[i
];
2117 /* compute if section should be included */
2119 if ((s
->sh_flags
& (SHF_ALLOC
| SHF_WRITE
)) !=
2123 if ((s
->sh_flags
& (SHF_ALLOC
| SHF_WRITE
)) !=
2124 (SHF_ALLOC
| SHF_WRITE
))
2130 } else if (s
->sh_type
== SHT_DYNSYM
||
2131 s
->sh_type
== SHT_STRTAB
||
2132 s
->sh_type
== SHT_HASH
) {
2135 } else if (s
->sh_type
== SHT_RELX
) {
2138 } else if (s
->sh_type
== SHT_NOBITS
) {
2145 sec_order
[sh_order_index
++] = i
;
2147 /* section matches: we align it and add its size */
2149 addr
= (addr
+ s
->sh_addralign
- 1) &
2150 ~(s
->sh_addralign
- 1);
2151 file_offset
+= (int) ( addr
- tmp
);
2152 s
->sh_offset
= file_offset
;
2155 /* update program header infos */
2156 if (ph
->p_offset
== 0) {
2157 ph
->p_offset
= file_offset
;
2159 ph
->p_paddr
= ph
->p_vaddr
;
2161 /* update dynamic relocation infos */
2162 if (s
->sh_type
== SHT_RELX
) {
2163 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
2164 if (!strcmp(strsec
->data
+ s
->sh_name
, ".rel.got")) {
2165 dyninf
->rel_addr
= addr
;
2166 dyninf
->rel_size
+= s
->sh_size
; /* XXX only first rel. */
2168 if (!strcmp(strsec
->data
+ s
->sh_name
, ".rel.bss")) {
2169 dyninf
->bss_addr
= addr
;
2170 dyninf
->bss_size
= s
->sh_size
; /* XXX only first rel. */
2173 if (dyninf
->rel_size
== 0)
2174 dyninf
->rel_addr
= addr
;
2175 dyninf
->rel_size
+= s
->sh_size
;
2179 if (s
->sh_type
!= SHT_NOBITS
)
2180 file_offset
+= s
->sh_size
;
2184 /* Make the first PT_LOAD segment include the program
2185 headers itself (and the ELF header as well), it'll
2186 come out with same memory use but will make various
2187 tools like binutils strip work better. */
2188 ph
->p_offset
&= ~(ph
->p_align
- 1);
2189 ph
->p_vaddr
&= ~(ph
->p_align
- 1);
2190 ph
->p_paddr
&= ~(ph
->p_align
- 1);
2192 ph
->p_filesz
= file_offset
- ph
->p_offset
;
2193 ph
->p_memsz
= addr
- ph
->p_vaddr
;
2196 if (s1
->output_format
== TCC_OUTPUT_FORMAT_ELF
) {
2197 /* if in the middle of a page, we duplicate the page in
2198 memory so that one copy is RX and the other is RW */
2199 if ((addr
& (s_align
- 1)) != 0)
2202 addr
= (addr
+ s_align
- 1) & ~(s_align
- 1);
2203 file_offset
= (file_offset
+ s_align
- 1) & ~(s_align
- 1);
2209 /* all other sections come after */
2210 for(i
= 1; i
< s1
->nb_sections
; i
++) {
2211 s
= s1
->sections
[i
];
2212 if (phnum
> 0 && (s
->sh_flags
& SHF_ALLOC
))
2214 sec_order
[sh_order_index
++] = i
;
2216 file_offset
= (file_offset
+ s
->sh_addralign
- 1) &
2217 ~(s
->sh_addralign
- 1);
2218 s
->sh_offset
= file_offset
;
2219 if (s
->sh_type
!= SHT_NOBITS
)
2220 file_offset
+= s
->sh_size
;
2226 static void fill_unloadable_phdr(ElfW(Phdr
) *phdr
, int phnum
, Section
*interp
,
2231 /* if interpreter, then add corresponding program header */
2237 int len
= phnum
* sizeof(ElfW(Phdr
));
2239 ph
->p_type
= PT_PHDR
;
2240 ph
->p_offset
= sizeof(ElfW(Ehdr
));
2241 ph
->p_vaddr
= interp
->sh_addr
- len
;
2242 ph
->p_paddr
= ph
->p_vaddr
;
2243 ph
->p_filesz
= ph
->p_memsz
= len
;
2244 ph
->p_flags
= PF_R
| PF_X
;
2245 ph
->p_align
= 4; /* interp->sh_addralign; */
2249 ph
->p_type
= PT_INTERP
;
2250 ph
->p_offset
= interp
->sh_offset
;
2251 ph
->p_vaddr
= interp
->sh_addr
;
2252 ph
->p_paddr
= ph
->p_vaddr
;
2253 ph
->p_filesz
= interp
->sh_size
;
2254 ph
->p_memsz
= interp
->sh_size
;
2256 ph
->p_align
= interp
->sh_addralign
;
2259 /* if dynamic section, then add corresponding program header */
2261 ph
= &phdr
[phnum
- 1];
2263 ph
->p_type
= PT_DYNAMIC
;
2264 ph
->p_offset
= dynamic
->sh_offset
;
2265 ph
->p_vaddr
= dynamic
->sh_addr
;
2266 ph
->p_paddr
= ph
->p_vaddr
;
2267 ph
->p_filesz
= dynamic
->sh_size
;
2268 ph
->p_memsz
= dynamic
->sh_size
;
2269 ph
->p_flags
= PF_R
| PF_W
;
2270 ph
->p_align
= dynamic
->sh_addralign
;
2274 /* Fill the dynamic section with tags describing the address and size of
2276 static void fill_dynamic(TCCState
*s1
, struct dyn_inf
*dyninf
)
2280 dynamic
= dyninf
->dynamic
;
2282 /* put dynamic section entries */
2283 dynamic
->data_offset
= dyninf
->dyn_rel_off
;
2284 put_dt(dynamic
, DT_HASH
, s1
->dynsym
->hash
->sh_addr
);
2285 put_dt(dynamic
, DT_STRTAB
, dyninf
->dynstr
->sh_addr
);
2286 put_dt(dynamic
, DT_SYMTAB
, s1
->dynsym
->sh_addr
);
2287 put_dt(dynamic
, DT_STRSZ
, dyninf
->dynstr
->data_offset
);
2288 put_dt(dynamic
, DT_SYMENT
, sizeof(ElfW(Sym
)));
2289 #if defined(TCC_TARGET_ARM64) || defined(TCC_TARGET_X86_64)
2290 put_dt(dynamic
, DT_RELA
, dyninf
->rel_addr
);
2291 put_dt(dynamic
, DT_RELASZ
, dyninf
->rel_size
);
2292 put_dt(dynamic
, DT_RELAENT
, sizeof(ElfW_Rel
));
2294 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
2295 put_dt(dynamic
, DT_PLTGOT
, s1
->got
->sh_addr
);
2296 put_dt(dynamic
, DT_PLTRELSZ
, dyninf
->rel_size
);
2297 put_dt(dynamic
, DT_JMPREL
, dyninf
->rel_addr
);
2298 put_dt(dynamic
, DT_PLTREL
, DT_REL
);
2299 put_dt(dynamic
, DT_REL
, dyninf
->bss_addr
);
2300 put_dt(dynamic
, DT_RELSZ
, dyninf
->bss_size
);
2302 put_dt(dynamic
, DT_REL
, dyninf
->rel_addr
);
2303 put_dt(dynamic
, DT_RELSZ
, dyninf
->rel_size
);
2304 put_dt(dynamic
, DT_RELENT
, sizeof(ElfW_Rel
));
2308 put_dt(dynamic
, DT_DEBUG
, 0);
2309 put_dt(dynamic
, DT_NULL
, 0);
2312 /* Relocate remaining sections and symbols (that is those not related to
2314 static int final_sections_reloc(TCCState
*s1
)
2319 relocate_syms(s1
, 0);
2321 if (s1
->nb_errors
!= 0)
2324 /* relocate sections */
2325 /* XXX: ignore sections with allocated relocations ? */
2326 for(i
= 1; i
< s1
->nb_sections
; i
++) {
2327 s
= s1
->sections
[i
];
2328 #ifdef TCC_TARGET_I386
2329 if (s
->reloc
&& s
!= s1
->got
&& (s
->sh_flags
& SHF_ALLOC
)) //gr
2330 /* On X86 gdb 7.3 works in any case but gdb 6.6 will crash if SHF_ALLOC
2331 checking is removed */
2333 if (s
->reloc
&& s
!= s1
->got
)
2334 /* On X86_64 gdb 7.3 will crash if SHF_ALLOC checking is present */
2336 relocate_section(s1
, s
);
2339 /* relocate relocation entries if the relocation tables are
2340 allocated in the executable */
2341 for(i
= 1; i
< s1
->nb_sections
; i
++) {
2342 s
= s1
->sections
[i
];
2343 if ((s
->sh_flags
& SHF_ALLOC
) &&
2344 s
->sh_type
== SHT_RELX
) {
2345 relocate_rel(s1
, s
);
2351 /* Create an ELF file on disk.
2352 This function handle ELF specific layout requirements */
2353 static void tcc_output_elf(TCCState
*s1
, FILE *f
, int phnum
, ElfW(Phdr
) *phdr
,
2354 int file_offset
, int *sec_order
)
2356 int i
, shnum
, offset
, size
, file_type
;
2359 ElfW(Shdr
) shdr
, *sh
;
2361 file_type
= s1
->output_type
;
2362 shnum
= s1
->nb_sections
;
2364 memset(&ehdr
, 0, sizeof(ehdr
));
2367 ehdr
.e_phentsize
= sizeof(ElfW(Phdr
));
2368 ehdr
.e_phnum
= phnum
;
2369 ehdr
.e_phoff
= sizeof(ElfW(Ehdr
));
2373 file_offset
= (file_offset
+ 3) & -4;
2376 ehdr
.e_ident
[0] = ELFMAG0
;
2377 ehdr
.e_ident
[1] = ELFMAG1
;
2378 ehdr
.e_ident
[2] = ELFMAG2
;
2379 ehdr
.e_ident
[3] = ELFMAG3
;
2380 ehdr
.e_ident
[4] = ELFCLASSW
;
2381 ehdr
.e_ident
[5] = ELFDATA2LSB
;
2382 ehdr
.e_ident
[6] = EV_CURRENT
;
2383 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
2384 ehdr
.e_ident
[EI_OSABI
] = ELFOSABI_FREEBSD
;
2386 #ifdef TCC_TARGET_ARM
2388 ehdr
.e_ident
[EI_OSABI
] = 0;
2389 ehdr
.e_flags
= EF_ARM_EABI_VER4
;
2390 if (file_type
== TCC_OUTPUT_EXE
|| file_type
== TCC_OUTPUT_DLL
)
2391 ehdr
.e_flags
|= EF_ARM_HASENTRY
;
2392 if (s1
->float_abi
== ARM_HARD_FLOAT
)
2393 ehdr
.e_flags
|= EF_ARM_VFP_FLOAT
;
2395 ehdr
.e_flags
|= EF_ARM_SOFT_FLOAT
;
2397 ehdr
.e_ident
[EI_OSABI
] = ELFOSABI_ARM
;
2402 case TCC_OUTPUT_EXE
:
2403 ehdr
.e_type
= ET_EXEC
;
2404 ehdr
.e_entry
= get_elf_sym_addr(s1
, "_start", 1);
2406 case TCC_OUTPUT_DLL
:
2407 ehdr
.e_type
= ET_DYN
;
2408 ehdr
.e_entry
= text_section
->sh_addr
; /* XXX: is it correct ? */
2410 case TCC_OUTPUT_OBJ
:
2411 ehdr
.e_type
= ET_REL
;
2414 ehdr
.e_machine
= EM_TCC_TARGET
;
2415 ehdr
.e_version
= EV_CURRENT
;
2416 ehdr
.e_shoff
= file_offset
;
2417 ehdr
.e_ehsize
= sizeof(ElfW(Ehdr
));
2418 ehdr
.e_shentsize
= sizeof(ElfW(Shdr
));
2419 ehdr
.e_shnum
= shnum
;
2420 ehdr
.e_shstrndx
= shnum
- 1;
2422 fwrite(&ehdr
, 1, sizeof(ElfW(Ehdr
)), f
);
2423 fwrite(phdr
, 1, phnum
* sizeof(ElfW(Phdr
)), f
);
2424 offset
= sizeof(ElfW(Ehdr
)) + phnum
* sizeof(ElfW(Phdr
));
2426 sort_syms(s1
, symtab_section
);
2427 for(i
= 1; i
< s1
->nb_sections
; i
++) {
2428 s
= s1
->sections
[sec_order
[i
]];
2429 if (s
->sh_type
!= SHT_NOBITS
) {
2430 if (s
->sh_type
== SHT_DYNSYM
)
2431 patch_dynsym_undef(s1
, s
);
2432 while (offset
< s
->sh_offset
) {
2437 fwrite(s
->data
, 1, size
, f
);
2442 /* output section headers */
2443 while (offset
< ehdr
.e_shoff
) {
2448 for(i
= 0; i
< s1
->nb_sections
; i
++) {
2450 memset(sh
, 0, sizeof(ElfW(Shdr
)));
2451 s
= s1
->sections
[i
];
2453 sh
->sh_name
= s
->sh_name
;
2454 sh
->sh_type
= s
->sh_type
;
2455 sh
->sh_flags
= s
->sh_flags
;
2456 sh
->sh_entsize
= s
->sh_entsize
;
2457 sh
->sh_info
= s
->sh_info
;
2459 sh
->sh_link
= s
->link
->sh_num
;
2460 sh
->sh_addralign
= s
->sh_addralign
;
2461 sh
->sh_addr
= s
->sh_addr
;
2462 sh
->sh_offset
= s
->sh_offset
;
2463 sh
->sh_size
= s
->sh_size
;
2465 fwrite(sh
, 1, sizeof(ElfW(Shdr
)), f
);
2469 /* Write an elf, coff or "binary" file */
2470 static int tcc_write_elf_file(TCCState
*s1
, const char *filename
, int phnum
,
2471 ElfW(Phdr
) *phdr
, int file_offset
, int *sec_order
)
2473 int fd
, mode
, file_type
;
2476 file_type
= s1
->output_type
;
2477 if (file_type
== TCC_OUTPUT_OBJ
)
2482 fd
= open(filename
, O_WRONLY
| O_CREAT
| O_TRUNC
| O_BINARY
, mode
);
2484 tcc_error_noabort("could not write '%s'", filename
);
2487 f
= fdopen(fd
, "wb");
2489 printf("<- %s\n", filename
);
2491 #ifdef TCC_TARGET_COFF
2492 if (s1
->output_format
== TCC_OUTPUT_FORMAT_COFF
)
2493 tcc_output_coff(s1
, f
);
2496 if (s1
->output_format
== TCC_OUTPUT_FORMAT_ELF
)
2497 tcc_output_elf(s1
, f
, phnum
, phdr
, file_offset
, sec_order
);
2499 tcc_output_binary(s1
, f
, sec_order
);
2505 /* Output an elf, coff or binary file */
2506 /* XXX: suppress unneeded sections */
2507 static int elf_output_file(TCCState
*s1
, const char *filename
)
2509 int i
, ret
, phnum
, shnum
, file_type
, file_offset
, *sec_order
;
2510 struct dyn_inf dyninf
;
2513 Section
*strsec
, *interp
, *dynamic
, *dynstr
;
2515 file_type
= s1
->output_type
;
2518 /* if linking, also link in runtime libraries (libc, libgcc, etc.) */
2519 if (file_type
!= TCC_OUTPUT_OBJ
) {
2520 tcc_add_runtime(s1
);
2525 interp
= dynamic
= dynstr
= NULL
; /* avoid warning */
2526 dyninf
.dyn_rel_off
= 0; /* avoid warning */
2528 if (file_type
!= TCC_OUTPUT_OBJ
) {
2529 relocate_common_syms();
2531 tcc_add_linker_symbols(s1
);
2533 if (!s1
->static_link
) {
2534 if (file_type
== TCC_OUTPUT_EXE
) {
2536 /* allow override the dynamic loader */
2537 const char *elfint
= getenv("LD_SO");
2539 elfint
= DEFAULT_ELFINTERP(s1
);
2540 /* add interpreter section only if executable */
2541 interp
= new_section(s1
, ".interp", SHT_PROGBITS
, SHF_ALLOC
);
2542 interp
->sh_addralign
= 1;
2543 ptr
= section_ptr_add(interp
, 1 + strlen(elfint
));
2544 strcpy(ptr
, elfint
);
2547 /* add dynamic symbol table */
2548 s1
->dynsym
= new_symtab(s1
, ".dynsym", SHT_DYNSYM
, SHF_ALLOC
,
2550 ".hash", SHF_ALLOC
);
2551 dynstr
= s1
->dynsym
->link
;
2553 /* add dynamic section */
2554 dynamic
= new_section(s1
, ".dynamic", SHT_DYNAMIC
,
2555 SHF_ALLOC
| SHF_WRITE
);
2556 dynamic
->link
= dynstr
;
2557 dynamic
->sh_entsize
= sizeof(ElfW(Dyn
));
2561 if (file_type
== TCC_OUTPUT_EXE
) {
2562 bind_exe_dynsyms(s1
);
2564 if (s1
->nb_errors
) {
2569 bind_libs_dynsyms(s1
);
2570 } else /* shared library case: simply export all global symbols */
2571 export_global_syms(s1
);
2573 build_got_entries(s1
);
2575 /* add a list of needed dlls */
2576 for(i
= 0; i
< s1
->nb_loaded_dlls
; i
++) {
2577 DLLReference
*dllref
= s1
->loaded_dlls
[i
];
2578 if (dllref
->level
== 0)
2579 put_dt(dynamic
, DT_NEEDED
, put_elf_str(dynstr
, dllref
->name
));
2583 put_dt(dynamic
, DT_RPATH
, put_elf_str(dynstr
, s1
->rpath
));
2585 /* XXX: currently, since we do not handle PIC code, we
2586 must relocate the readonly segments */
2587 if (file_type
== TCC_OUTPUT_DLL
) {
2589 put_dt(dynamic
, DT_SONAME
, put_elf_str(dynstr
, s1
->soname
));
2590 put_dt(dynamic
, DT_TEXTREL
, 0);
2594 put_dt(dynamic
, DT_SYMBOLIC
, 0);
2596 /* add necessary space for other entries */
2597 dyninf
.dyn_rel_off
= dynamic
->data_offset
;
2598 dynamic
->data_offset
+= sizeof(ElfW(Dyn
)) * EXTRA_RELITEMS
;
2600 /* still need to build got entries in case of static link */
2601 build_got_entries(s1
);
2605 /* we add a section for symbols */
2606 strsec
= new_section(s1
, ".shstrtab", SHT_STRTAB
, 0);
2607 put_elf_str(strsec
, "");
2609 /* compute number of sections */
2610 shnum
= s1
->nb_sections
;
2612 /* this array is used to reorder sections in the output file */
2613 sec_order
= tcc_malloc(sizeof(int) * shnum
);
2616 /* compute number of program headers */
2619 case TCC_OUTPUT_OBJ
:
2622 case TCC_OUTPUT_EXE
:
2623 if (!s1
->static_link
)
2624 phnum
= 4 + HAVE_PHDR
;
2628 case TCC_OUTPUT_DLL
:
2633 /* Allocate strings for section names */
2634 alloc_sec_names(s1
, file_type
, strsec
);
2636 /* allocate program segment headers */
2637 phdr
= tcc_mallocz(phnum
* sizeof(ElfW(Phdr
)));
2639 /* compute section to program header mapping */
2640 file_offset
= layout_sections(s1
, phdr
, phnum
, interp
, strsec
, &dyninf
,
2643 /* Fill remaining program header and finalize relocation related to dynamic
2646 fill_unloadable_phdr(phdr
, phnum
, interp
, dynamic
);
2648 dyninf
.dynamic
= dynamic
;
2649 dyninf
.dynstr
= dynstr
;
2651 fill_dynamic(s1
, &dyninf
);
2653 /* put in GOT the dynamic section address and relocate PLT */
2654 put32(s1
->got
->data
, dynamic
->sh_addr
);
2655 if (file_type
== TCC_OUTPUT_EXE
2656 #if defined(TCC_OUTPUT_DLL_WITH_PLT)
2657 || file_type
== TCC_OUTPUT_DLL
2662 /* relocate symbols in .dynsym now that final addresses are known */
2663 for_each_elem(s1
->dynsym
, 1, sym
, ElfW(Sym
)) {
2664 if (sym
->st_shndx
== SHN_UNDEF
) {
2665 /* relocate to PLT if symbol corresponds to a PLT entry,
2666 but not if it's a weak symbol */
2667 if (ELFW(ST_BIND
)(sym
->st_info
) == STB_WEAK
)
2669 else if (sym
->st_value
)
2670 sym
->st_value
+= s1
->plt
->sh_addr
;
2671 } else if (sym
->st_shndx
< SHN_LORESERVE
) {
2672 /* do symbol relocation */
2673 sym
->st_value
+= s1
->sections
[sym
->st_shndx
]->sh_addr
;
2679 /* if building executable or DLL, then relocate each section
2680 except the GOT which is already relocated */
2681 if (file_type
!= TCC_OUTPUT_OBJ
) {
2682 ret
= final_sections_reloc(s1
);
2687 /* Perform relocation to GOT or PLT entries */
2688 if (file_type
== TCC_OUTPUT_EXE
&& s1
->static_link
)
2691 /* Create the ELF file with name 'filename' */
2692 ret
= tcc_write_elf_file(s1
, filename
, phnum
, phdr
, file_offset
, sec_order
);
2694 const char *strip
= "sstrip "; // super strip utility from ELFkickers
2695 const char *null
= " 2> /dev/null";
2696 int len
= strlen(strip
) + strlen(filename
) + strlen(null
) + 1;
2700 sprintf(buf
, "%s%s%s", strip
, filename
, null
);
2703 system(buf
+1); // call a strip utility from binutils
2708 tcc_free(s1
->symtab_to_dynsym
);
2709 tcc_free(sec_order
);
2711 tcc_free(s1
->sym_attrs
);
2712 s1
->sym_attrs
= NULL
;
2716 LIBTCCAPI
int tcc_output_file(TCCState
*s
, const char *filename
)
2719 #ifdef TCC_TARGET_PE
2720 if (s
->output_type
!= TCC_OUTPUT_OBJ
) {
2721 ret
= pe_output_file(s
, filename
);
2724 ret
= elf_output_file(s
, filename
);
2728 static void *load_data(int fd
, unsigned long file_offset
, unsigned long size
)
2732 data
= tcc_malloc(size
);
2733 lseek(fd
, file_offset
, SEEK_SET
);
2734 read(fd
, data
, size
);
2738 typedef struct SectionMergeInfo
{
2739 Section
*s
; /* corresponding existing section */
2740 unsigned long offset
; /* offset of the new section in the existing section */
2741 uint8_t new_section
; /* true if section 's' was added */
2742 uint8_t link_once
; /* true if link once section */
2745 /* load an object file and merge it with current files */
2746 /* XXX: handle correctly stab (debug) info */
2747 ST_FUNC
int tcc_load_object_file(TCCState
*s1
,
2748 int fd
, unsigned long file_offset
)
2751 ElfW(Shdr
) *shdr
, *sh
;
2752 int size
, i
, j
, offset
, offseti
, nb_syms
, sym_index
, ret
;
2753 unsigned char *strsec
, *strtab
;
2754 int *old_to_new_syms
;
2755 char *sh_name
, *name
;
2756 SectionMergeInfo
*sm_table
, *sm
;
2757 ElfW(Sym
) *sym
, *symtab
;
2764 stab_index
= stabstr_index
= 0;
2766 if (read(fd
, &ehdr
, sizeof(ehdr
)) != sizeof(ehdr
))
2768 if (ehdr
.e_ident
[0] != ELFMAG0
||
2769 ehdr
.e_ident
[1] != ELFMAG1
||
2770 ehdr
.e_ident
[2] != ELFMAG2
||
2771 ehdr
.e_ident
[3] != ELFMAG3
)
2773 /* test if object file */
2774 if (ehdr
.e_type
!= ET_REL
)
2776 /* test CPU specific stuff */
2777 if (ehdr
.e_ident
[5] != ELFDATA2LSB
||
2778 ehdr
.e_machine
!= EM_TCC_TARGET
) {
2780 tcc_error_noabort("invalid object file");
2784 shdr
= load_data(fd
, file_offset
+ ehdr
.e_shoff
,
2785 sizeof(ElfW(Shdr
)) * ehdr
.e_shnum
);
2786 sm_table
= tcc_mallocz(sizeof(SectionMergeInfo
) * ehdr
.e_shnum
);
2788 /* load section names */
2789 sh
= &shdr
[ehdr
.e_shstrndx
];
2790 strsec
= load_data(fd
, file_offset
+ sh
->sh_offset
, sh
->sh_size
);
2792 /* load symtab and strtab */
2793 old_to_new_syms
= NULL
;
2797 for(i
= 1; i
< ehdr
.e_shnum
; i
++) {
2799 if (sh
->sh_type
== SHT_SYMTAB
) {
2801 tcc_error_noabort("object must contain only one symtab");
2806 nb_syms
= sh
->sh_size
/ sizeof(ElfW(Sym
));
2807 symtab
= load_data(fd
, file_offset
+ sh
->sh_offset
, sh
->sh_size
);
2808 sm_table
[i
].s
= symtab_section
;
2810 /* now load strtab */
2811 sh
= &shdr
[sh
->sh_link
];
2812 strtab
= load_data(fd
, file_offset
+ sh
->sh_offset
, sh
->sh_size
);
2816 /* now examine each section and try to merge its content with the
2818 for(i
= 1; i
< ehdr
.e_shnum
; i
++) {
2819 /* no need to examine section name strtab */
2820 if (i
== ehdr
.e_shstrndx
)
2823 sh_name
= (char *) strsec
+ sh
->sh_name
;
2824 /* ignore sections types we do not handle */
2825 if (sh
->sh_type
!= SHT_PROGBITS
&&
2826 sh
->sh_type
!= SHT_RELX
&&
2828 sh
->sh_type
!= SHT_ARM_EXIDX
&&
2830 sh
->sh_type
!= SHT_NOBITS
&&
2831 sh
->sh_type
!= SHT_PREINIT_ARRAY
&&
2832 sh
->sh_type
!= SHT_INIT_ARRAY
&&
2833 sh
->sh_type
!= SHT_FINI_ARRAY
&&
2834 strcmp(sh_name
, ".stabstr")
2837 if (sh
->sh_addralign
< 1)
2838 sh
->sh_addralign
= 1;
2839 /* find corresponding section, if any */
2840 for(j
= 1; j
< s1
->nb_sections
;j
++) {
2841 s
= s1
->sections
[j
];
2842 if (!strcmp(s
->name
, sh_name
)) {
2843 if (!strncmp(sh_name
, ".gnu.linkonce",
2844 sizeof(".gnu.linkonce") - 1)) {
2845 /* if a 'linkonce' section is already present, we
2846 do not add it again. It is a little tricky as
2847 symbols can still be defined in
2849 sm_table
[i
].link_once
= 1;
2856 /* not found: create new section */
2857 s
= new_section(s1
, sh_name
, sh
->sh_type
, sh
->sh_flags
);
2858 /* take as much info as possible from the section. sh_link and
2859 sh_info will be updated later */
2860 s
->sh_addralign
= sh
->sh_addralign
;
2861 s
->sh_entsize
= sh
->sh_entsize
;
2862 sm_table
[i
].new_section
= 1;
2864 if (sh
->sh_type
!= s
->sh_type
) {
2865 tcc_error_noabort("invalid section type");
2869 /* align start of section */
2870 offset
= s
->data_offset
;
2872 if (0 == strcmp(sh_name
, ".stab")) {
2876 if (0 == strcmp(sh_name
, ".stabstr")) {
2881 size
= sh
->sh_addralign
- 1;
2882 offset
= (offset
+ size
) & ~size
;
2883 if (sh
->sh_addralign
> s
->sh_addralign
)
2884 s
->sh_addralign
= sh
->sh_addralign
;
2885 s
->data_offset
= offset
;
2887 sm_table
[i
].offset
= offset
;
2889 /* concatenate sections */
2891 if (sh
->sh_type
!= SHT_NOBITS
) {
2893 lseek(fd
, file_offset
+ sh
->sh_offset
, SEEK_SET
);
2894 ptr
= section_ptr_add(s
, size
);
2895 read(fd
, ptr
, size
);
2897 s
->data_offset
+= size
;
2902 /* gr relocate stab strings */
2903 if (stab_index
&& stabstr_index
) {
2906 s
= sm_table
[stab_index
].s
;
2907 a
= (Stab_Sym
*)(s
->data
+ sm_table
[stab_index
].offset
);
2908 b
= (Stab_Sym
*)(s
->data
+ s
->data_offset
);
2909 o
= sm_table
[stabstr_index
].offset
;
2911 a
->n_strx
+= o
, a
++;
2914 /* second short pass to update sh_link and sh_info fields of new
2916 for(i
= 1; i
< ehdr
.e_shnum
; i
++) {
2918 if (!s
|| !sm_table
[i
].new_section
)
2921 if (sh
->sh_link
> 0)
2922 s
->link
= sm_table
[sh
->sh_link
].s
;
2923 if (sh
->sh_type
== SHT_RELX
) {
2924 s
->sh_info
= sm_table
[sh
->sh_info
].s
->sh_num
;
2925 /* update backward link */
2926 s1
->sections
[s
->sh_info
]->reloc
= s
;
2931 /* resolve symbols */
2932 old_to_new_syms
= tcc_mallocz(nb_syms
* sizeof(int));
2935 for(i
= 1; i
< nb_syms
; i
++, sym
++) {
2936 if (sym
->st_shndx
!= SHN_UNDEF
&&
2937 sym
->st_shndx
< SHN_LORESERVE
) {
2938 sm
= &sm_table
[sym
->st_shndx
];
2939 if (sm
->link_once
) {
2940 /* if a symbol is in a link once section, we use the
2941 already defined symbol. It is very important to get
2942 correct relocations */
2943 if (ELFW(ST_BIND
)(sym
->st_info
) != STB_LOCAL
) {
2944 name
= (char *) strtab
+ sym
->st_name
;
2945 sym_index
= find_elf_sym(symtab_section
, name
);
2947 old_to_new_syms
[i
] = sym_index
;
2951 /* if no corresponding section added, no need to add symbol */
2954 /* convert section number */
2955 sym
->st_shndx
= sm
->s
->sh_num
;
2957 sym
->st_value
+= sm
->offset
;
2960 name
= (char *) strtab
+ sym
->st_name
;
2961 sym_index
= add_elf_sym(symtab_section
, sym
->st_value
, sym
->st_size
,
2962 sym
->st_info
, sym
->st_other
,
2963 sym
->st_shndx
, name
);
2964 old_to_new_syms
[i
] = sym_index
;
2967 /* third pass to patch relocation entries */
2968 for(i
= 1; i
< ehdr
.e_shnum
; i
++) {
2973 offset
= sm_table
[i
].offset
;
2974 switch(s
->sh_type
) {
2976 /* take relocation offset information */
2977 offseti
= sm_table
[sh
->sh_info
].offset
;
2978 for_each_elem(s
, (offset
/ sizeof(*rel
)), rel
, ElfW_Rel
) {
2981 /* convert symbol index */
2982 type
= ELFW(R_TYPE
)(rel
->r_info
);
2983 sym_index
= ELFW(R_SYM
)(rel
->r_info
);
2984 /* NOTE: only one symtab assumed */
2985 if (sym_index
>= nb_syms
)
2987 sym_index
= old_to_new_syms
[sym_index
];
2988 /* ignore link_once in rel section. */
2989 if (!sym_index
&& !sm
->link_once
2990 #ifdef TCC_TARGET_ARM
2991 && type
!= R_ARM_V4BX
2995 tcc_error_noabort("Invalid relocation entry [%2d] '%s' @ %.8x",
2996 i
, strsec
+ sh
->sh_name
, rel
->r_offset
);
2999 rel
->r_info
= ELFW(R_INFO
)(sym_index
, type
);
3000 /* offset the relocation offset */
3001 rel
->r_offset
+= offseti
;
3002 #ifdef TCC_TARGET_ARM
3003 /* Jumps and branches from a Thumb code to a PLT entry need
3004 special handling since PLT entries are ARM code.
3005 Unconditional bl instructions referencing PLT entries are
3006 handled by converting these instructions into blx
3007 instructions. Other case of instructions referencing a PLT
3008 entry require to add a Thumb stub before the PLT entry to
3009 switch to ARM mode. We set bit plt_thumb_stub of the
3010 attribute of a symbol to indicate such a case. */
3011 if (type
== R_ARM_THM_JUMP24
)
3012 alloc_sym_attr(s1
, sym_index
)->plt_thumb_stub
= 1;
3025 tcc_free(old_to_new_syms
);
3032 typedef struct ArchiveHeader
{
3033 char ar_name
[16]; /* name of this member */
3034 char ar_date
[12]; /* file mtime */
3035 char ar_uid
[6]; /* owner uid; printed as decimal */
3036 char ar_gid
[6]; /* owner gid; printed as decimal */
3037 char ar_mode
[8]; /* file mode, printed as octal */
3038 char ar_size
[10]; /* file size, printed as decimal */
3039 char ar_fmag
[2]; /* should contain ARFMAG */
3042 static int get_be32(const uint8_t *b
)
3044 return b
[3] | (b
[2] << 8) | (b
[1] << 16) | (b
[0] << 24);
3047 /* load only the objects which resolve undefined symbols */
3048 static int tcc_load_alacarte(TCCState
*s1
, int fd
, int size
)
3050 int i
, bound
, nsyms
, sym_index
, off
, ret
;
3052 const char *ar_names
, *p
;
3053 const uint8_t *ar_index
;
3056 data
= tcc_malloc(size
);
3057 if (read(fd
, data
, size
) != size
)
3059 nsyms
= get_be32(data
);
3060 ar_index
= data
+ 4;
3061 ar_names
= (char *) ar_index
+ nsyms
* 4;
3065 for(p
= ar_names
, i
= 0; i
< nsyms
; i
++, p
+= strlen(p
)+1) {
3066 sym_index
= find_elf_sym(symtab_section
, p
);
3068 sym
= &((ElfW(Sym
) *)symtab_section
->data
)[sym_index
];
3069 if(sym
->st_shndx
== SHN_UNDEF
) {
3070 off
= get_be32(ar_index
+ i
* 4) + sizeof(ArchiveHeader
);
3072 lseek(fd
, off
, SEEK_SET
);
3073 if(tcc_load_object_file(s1
, fd
, off
) < 0) {
3088 /* load a '.a' file */
3089 ST_FUNC
int tcc_load_archive(TCCState
*s1
, int fd
)
3096 unsigned long file_offset
;
3098 /* skip magic which was already checked */
3099 read(fd
, magic
, sizeof(magic
));
3102 len
= read(fd
, &hdr
, sizeof(hdr
));
3105 if (len
!= sizeof(hdr
)) {
3106 tcc_error_noabort("invalid archive");
3109 memcpy(ar_size
, hdr
.ar_size
, sizeof(hdr
.ar_size
));
3110 ar_size
[sizeof(hdr
.ar_size
)] = '\0';
3111 size
= strtol(ar_size
, NULL
, 0);
3112 memcpy(ar_name
, hdr
.ar_name
, sizeof(hdr
.ar_name
));
3113 for(i
= sizeof(hdr
.ar_name
) - 1; i
>= 0; i
--) {
3114 if (ar_name
[i
] != ' ')
3117 ar_name
[i
+ 1] = '\0';
3118 file_offset
= lseek(fd
, 0, SEEK_CUR
);
3120 size
= (size
+ 1) & ~1;
3121 if (!strcmp(ar_name
, "/")) {
3122 /* coff symbol table : we handle it */
3123 if(s1
->alacarte_link
)
3124 return tcc_load_alacarte(s1
, fd
, size
);
3125 } else if (!strcmp(ar_name
, "//") ||
3126 !strcmp(ar_name
, "__.SYMDEF") ||
3127 !strcmp(ar_name
, "__.SYMDEF/") ||
3128 !strcmp(ar_name
, "ARFILENAMES/")) {
3129 /* skip symbol table or archive names */
3131 if (tcc_load_object_file(s1
, fd
, file_offset
) < 0)
3134 lseek(fd
, file_offset
+ size
, SEEK_SET
);
3139 #ifndef TCC_TARGET_PE
3140 /* load a DLL and all referenced DLLs. 'level = 0' means that the DLL
3141 is referenced by the user (so it should be added as DT_NEEDED in
3142 the generated ELF file) */
3143 ST_FUNC
int tcc_load_dll(TCCState
*s1
, int fd
, const char *filename
, int level
)
3146 ElfW(Shdr
) *shdr
, *sh
, *sh1
;
3147 int i
, j
, nb_syms
, nb_dts
, sym_bind
, ret
;
3148 ElfW(Sym
) *sym
, *dynsym
;
3149 ElfW(Dyn
) *dt
, *dynamic
;
3150 unsigned char *dynstr
;
3151 const char *name
, *soname
;
3152 DLLReference
*dllref
;
3154 read(fd
, &ehdr
, sizeof(ehdr
));
3156 /* test CPU specific stuff */
3157 if (ehdr
.e_ident
[5] != ELFDATA2LSB
||
3158 ehdr
.e_machine
!= EM_TCC_TARGET
) {
3159 tcc_error_noabort("bad architecture");
3164 shdr
= load_data(fd
, ehdr
.e_shoff
, sizeof(ElfW(Shdr
)) * ehdr
.e_shnum
);
3166 /* load dynamic section and dynamic symbols */
3170 dynsym
= NULL
; /* avoid warning */
3171 dynstr
= NULL
; /* avoid warning */
3172 for(i
= 0, sh
= shdr
; i
< ehdr
.e_shnum
; i
++, sh
++) {
3173 switch(sh
->sh_type
) {
3175 nb_dts
= sh
->sh_size
/ sizeof(ElfW(Dyn
));
3176 dynamic
= load_data(fd
, sh
->sh_offset
, sh
->sh_size
);
3179 nb_syms
= sh
->sh_size
/ sizeof(ElfW(Sym
));
3180 dynsym
= load_data(fd
, sh
->sh_offset
, sh
->sh_size
);
3181 sh1
= &shdr
[sh
->sh_link
];
3182 dynstr
= load_data(fd
, sh1
->sh_offset
, sh1
->sh_size
);
3189 /* compute the real library name */
3190 soname
= tcc_basename(filename
);
3192 for(i
= 0, dt
= dynamic
; i
< nb_dts
; i
++, dt
++) {
3193 if (dt
->d_tag
== DT_SONAME
) {
3194 soname
= (char *) dynstr
+ dt
->d_un
.d_val
;
3198 /* if the dll is already loaded, do not load it */
3199 for(i
= 0; i
< s1
->nb_loaded_dlls
; i
++) {
3200 dllref
= s1
->loaded_dlls
[i
];
3201 if (!strcmp(soname
, dllref
->name
)) {
3202 /* but update level if needed */
3203 if (level
< dllref
->level
)
3204 dllref
->level
= level
;
3210 /* add the dll and its level */
3211 dllref
= tcc_mallocz(sizeof(DLLReference
) + strlen(soname
));
3212 dllref
->level
= level
;
3213 strcpy(dllref
->name
, soname
);
3214 dynarray_add((void ***)&s1
->loaded_dlls
, &s1
->nb_loaded_dlls
, dllref
);
3216 /* add dynamic symbols in dynsym_section */
3217 for(i
= 1, sym
= dynsym
+ 1; i
< nb_syms
; i
++, sym
++) {
3218 sym_bind
= ELFW(ST_BIND
)(sym
->st_info
);
3219 if (sym_bind
== STB_LOCAL
)
3221 name
= (char *) dynstr
+ sym
->st_name
;
3222 add_elf_sym(s1
->dynsymtab_section
, sym
->st_value
, sym
->st_size
,
3223 sym
->st_info
, sym
->st_other
, sym
->st_shndx
, name
);
3226 /* load all referenced DLLs */
3227 for(i
= 0, dt
= dynamic
; i
< nb_dts
; i
++, dt
++) {
3230 name
= (char *) dynstr
+ dt
->d_un
.d_val
;
3231 for(j
= 0; j
< s1
->nb_loaded_dlls
; j
++) {
3232 dllref
= s1
->loaded_dlls
[j
];
3233 if (!strcmp(name
, dllref
->name
))
3234 goto already_loaded
;
3236 if (tcc_add_dll(s1
, name
, AFF_REFERENCED_DLL
) < 0) {
3237 tcc_error_noabort("referenced dll '%s' not found", name
);
3254 #define LD_TOK_NAME 256
3255 #define LD_TOK_EOF (-1)
3257 /* return next ld script token */
3258 static int ld_next(TCCState
*s1
, char *name
, int name_size
)
3276 file
->buf_ptr
= parse_comment(file
->buf_ptr
);
3277 ch
= file
->buf_ptr
[0];
3285 /* case 'a' ... 'z': */
3312 /* case 'A' ... 'z': */
3347 if (!((ch
>= 'a' && ch
<= 'z') ||
3348 (ch
>= 'A' && ch
<= 'Z') ||
3349 (ch
>= '0' && ch
<= '9') ||
3350 strchr("/.-_+=$:\\,~", ch
)))
3352 if ((q
- name
) < name_size
- 1) {
3371 static int ld_add_file(TCCState
*s1
, const char filename
[])
3375 ret
= tcc_add_file_internal(s1
, filename
, 0);
3377 ret
= tcc_add_dll(s1
, filename
, 0);
3381 static inline int new_undef_syms(void)
3384 ret
= new_undef_sym
;
3389 static int ld_add_file_list(TCCState
*s1
, const char *cmd
, int as_needed
)
3391 char filename
[1024], libname
[1024];
3392 int t
, group
, nblibs
= 0, ret
= 0;
3395 group
= !strcmp(cmd
, "GROUP");
3398 t
= ld_next(s1
, filename
, sizeof(filename
));
3401 t
= ld_next(s1
, filename
, sizeof(filename
));
3404 if (t
== LD_TOK_EOF
) {
3405 tcc_error_noabort("unexpected end of file");
3407 goto lib_parse_error
;
3408 } else if (t
== ')') {
3410 } else if (t
== '-') {
3411 t
= ld_next(s1
, filename
, sizeof(filename
));
3412 if ((t
!= LD_TOK_NAME
) || (filename
[0] != 'l')) {
3413 tcc_error_noabort("library name expected");
3415 goto lib_parse_error
;
3417 pstrcpy(libname
, sizeof libname
, &filename
[1]);
3418 if (s1
->static_link
) {
3419 snprintf(filename
, sizeof filename
, "lib%s.a", libname
);
3421 snprintf(filename
, sizeof filename
, "lib%s.so", libname
);
3423 } else if (t
!= LD_TOK_NAME
) {
3424 tcc_error_noabort("filename expected");
3426 goto lib_parse_error
;
3428 if (!strcmp(filename
, "AS_NEEDED")) {
3429 ret
= ld_add_file_list(s1
, cmd
, 1);
3431 goto lib_parse_error
;
3433 /* TODO: Implement AS_NEEDED support. Ignore it for now */
3435 ret
= ld_add_file(s1
, filename
);
3437 goto lib_parse_error
;
3439 /* Add the filename *and* the libname to avoid future conversions */
3440 dynarray_add((void ***) &libs
, &nblibs
, tcc_strdup(filename
));
3441 if (libname
[0] != '\0')
3442 dynarray_add((void ***) &libs
, &nblibs
, tcc_strdup(libname
));
3446 t
= ld_next(s1
, filename
, sizeof(filename
));
3448 t
= ld_next(s1
, filename
, sizeof(filename
));
3451 if (group
&& !as_needed
) {
3452 while (new_undef_syms()) {
3455 for (i
= 0; i
< nblibs
; i
++)
3456 ld_add_file(s1
, libs
[i
]);
3460 dynarray_reset(&libs
, &nblibs
);
3464 /* interpret a subset of GNU ldscripts to handle the dummy libc.so
3466 ST_FUNC
int tcc_load_ldscript(TCCState
*s1
)
3469 char filename
[1024];
3472 ch
= file
->buf_ptr
[0];
3475 t
= ld_next(s1
, cmd
, sizeof(cmd
));
3476 if (t
== LD_TOK_EOF
)
3478 else if (t
!= LD_TOK_NAME
)
3480 if (!strcmp(cmd
, "INPUT") ||
3481 !strcmp(cmd
, "GROUP")) {
3482 ret
= ld_add_file_list(s1
, cmd
, 0);
3485 } else if (!strcmp(cmd
, "OUTPUT_FORMAT") ||
3486 !strcmp(cmd
, "TARGET")) {
3487 /* ignore some commands */
3488 t
= ld_next(s1
, cmd
, sizeof(cmd
));
3492 t
= ld_next(s1
, filename
, sizeof(filename
));
3493 if (t
== LD_TOK_EOF
) {
3494 tcc_error_noabort("unexpected end of file");
3496 } else if (t
== ')') {
3506 #endif /* !TCC_TARGET_PE */