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... may be -fcommon is needed?", 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
);
531 write32le(ptr
, read32le(ptr
) + val
);
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 write32le(ptr
, read32le(ptr
) + val
- addr
);
547 write32le(ptr
, read32le(ptr
) + val
- addr
);
554 write32le(ptr
, read32le(ptr
) + s1
->got
->sh_addr
- addr
);
557 write32le(ptr
, read32le(ptr
) + val
- s1
->got
->sh_addr
);
560 /* we load the got offset */
561 write32le(ptr
, read32le(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 write16le(ptr
, read16le(ptr
) + val
);
571 if (s1
->output_format
!= TCC_OUTPUT_FORMAT_BINARY
)
573 write16le(ptr
, read16le(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
:
767 case R_AARCH64_ABS32
:
770 case R_AARCH64_MOVW_UABS_G0_NC
:
771 write32le(ptr
, ((read32le(ptr
) & 0xffe0001f) |
772 (val
& 0xffff) << 5));
774 case R_AARCH64_MOVW_UABS_G1_NC
:
775 write32le(ptr
, ((read32le(ptr
) & 0xffe0001f) |
776 (val
>> 16 & 0xffff) << 5));
778 case R_AARCH64_MOVW_UABS_G2_NC
:
779 write32le(ptr
, ((read32le(ptr
) & 0xffe0001f) |
780 (val
>> 32 & 0xffff) << 5));
782 case R_AARCH64_MOVW_UABS_G3
:
783 write32le(ptr
, ((read32le(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 write32le(ptr
, ((read32le(ptr
) & 0x9f00001f) |
791 (off
& 0x1ffffc) << 3 | (off
& 3) << 29));
794 case R_AARCH64_ADD_ABS_LO12_NC
:
795 write32le(ptr
, ((read32le(ptr
) & 0xffc003ff) |
796 (val
& 0xfff) << 10));
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 write32le(ptr
, (0x14000000 |
816 (uint32_t)(type
== R_AARCH64_CALL26
) << 31 |
817 ((val
- addr
) >> 2 & 0x3ffffff)));
819 case R_AARCH64_ADR_GOT_PAGE
: {
821 (((s1
->got
->sh_addr
+
822 s1
->sym_attrs
[sym_index
].got_offset
) >> 12) - (addr
>> 12));
823 if ((off
+ ((uint64_t)1 << 20)) >> 21)
824 tcc_error("R_AARCH64_ADR_GOT_PAGE relocation failed");
825 write32le(ptr
, ((read32le(ptr
) & 0x9f00001f) |
826 (off
& 0x1ffffc) << 3 | (off
& 3) << 29));
829 case R_AARCH64_LD64_GOT_LO12_NC
:
831 ((read32le(ptr
) & 0xfff803ff) |
833 s1
->sym_attrs
[sym_index
].got_offset
) & 0xff8) << 7));
837 case R_AARCH64_GLOB_DAT
:
838 case R_AARCH64_JUMP_SLOT
:
839 /* They don't need addend */
841 printf ("reloc %d @ 0x%lx: val=0x%lx name=%s\n", type
, addr
,
843 (char *) symtab_section
->link
->data
+ sym
->st_name
);
845 write64le(ptr
, val
- rel
->r_addend
);
848 fprintf(stderr
, "FIXME: handle reloc type %x at %x [%p] to %x\n",
849 type
, (unsigned)addr
, ptr
, (unsigned)val
);
851 #elif defined(TCC_TARGET_C67)
859 /* put the low 16 bits of the absolute address
860 add to what is already there */
862 orig
= ((*(int *)(ptr
)) >> 7) & 0xffff;
863 orig
|= (((*(int *)(ptr
+4)) >> 7) & 0xffff) << 16;
865 /* patch both at once - assumes always in pairs Low - High */
867 *(int *) ptr
= (*(int *) ptr
& (~(0xffff << 7)) ) | (((val
+orig
) & 0xffff) << 7);
868 *(int *)(ptr
+4) = (*(int *)(ptr
+4) & (~(0xffff << 7)) ) | ((((val
+orig
)>>16) & 0xffff) << 7);
874 fprintf(stderr
,"FIXME: handle reloc type %x at %x [%p] to %x\n",
875 type
, (unsigned)addr
, ptr
, (unsigned)val
);
877 #elif defined(TCC_TARGET_X86_64)
879 if (s1
->output_type
== TCC_OUTPUT_DLL
) {
880 esym_index
= s1
->symtab_to_dynsym
[sym_index
];
881 qrel
->r_offset
= rel
->r_offset
;
883 qrel
->r_info
= ELFW(R_INFO
)(esym_index
, R_X86_64_64
);
884 qrel
->r_addend
= rel
->r_addend
;
888 qrel
->r_info
= ELFW(R_INFO
)(0, R_X86_64_RELATIVE
);
889 qrel
->r_addend
= read64le(ptr
) + val
;
893 write64le(ptr
, read64le(ptr
) + val
);
897 if (s1
->output_type
== TCC_OUTPUT_DLL
) {
898 /* XXX: this logic may depend on TCC's codegen
899 now TCC uses R_X86_64_32 even for a 64bit pointer */
900 qrel
->r_info
= ELFW(R_INFO
)(0, R_X86_64_RELATIVE
);
901 qrel
->r_addend
= read32le(ptr
) + val
;
904 write32le(ptr
, read32le(ptr
) + val
);
908 if (s1
->output_type
== TCC_OUTPUT_DLL
) {
910 esym_index
= s1
->symtab_to_dynsym
[sym_index
];
912 qrel
->r_offset
= rel
->r_offset
;
913 qrel
->r_info
= ELFW(R_INFO
)(esym_index
, R_X86_64_PC32
);
914 qrel
->r_addend
= read32le(ptr
);
922 /* We've put the PLT slot offset into r_addend when generating
923 it, and that's what we must use as relocation value (adjusted
924 by section offset of course). */
925 val
= s1
->plt
->sh_addr
+ rel
->r_addend
;
931 diff
= (long long)val
- addr
;
932 if (diff
< -2147483648LL || diff
> 2147483647LL) {
933 tcc_error("internal error: relocation failed");
935 write32le(ptr
, read32le(ptr
) + diff
);
938 case R_X86_64_GLOB_DAT
:
939 case R_X86_64_JUMP_SLOT
:
940 /* They don't need addend */
941 write64le(ptr
, val
- rel
->r_addend
);
943 case R_X86_64_GOTPCREL
:
944 case 41 /* R_X86_64_GOTPCRELX */:
945 case 42 /* R_X86_64_REX_GOTPCRELX */:
946 write32le(ptr
, read32le(ptr
) +
947 (s1
->got
->sh_addr
- addr
+
948 s1
->sym_attrs
[sym_index
].got_offset
- 4));
950 case R_X86_64_GOTTPOFF
:
951 write32le(ptr
, read32le(ptr
) + val
- s1
->got
->sh_addr
);
954 /* we load the got offset */
955 write32le(ptr
, read32le(ptr
) + s1
->sym_attrs
[sym_index
].got_offset
);
958 #error unsupported processor
962 /* if the relocation is allocated, we change its symbol table */
963 if (sr
->sh_flags
& SHF_ALLOC
)
964 sr
->link
= s1
->dynsym
;
967 /* relocate relocation table in 'sr' */
968 static void relocate_rel(TCCState
*s1
, Section
*sr
)
973 s
= s1
->sections
[sr
->sh_info
];
974 for_each_elem(sr
, 0, rel
, ElfW_Rel
)
975 rel
->r_offset
+= s
->sh_addr
;
978 /* count the number of dynamic relocations so that we can reserve
980 static int prepare_dynamic_rel(TCCState
*s1
, Section
*sr
)
983 int sym_index
, esym_index
, type
, count
;
986 for_each_elem(sr
, 0, rel
, ElfW_Rel
) {
987 sym_index
= ELFW(R_SYM
)(rel
->r_info
);
988 type
= ELFW(R_TYPE
)(rel
->r_info
);
990 #if defined(TCC_TARGET_I386)
992 #elif defined(TCC_TARGET_X86_64)
999 #if defined(TCC_TARGET_I386)
1001 #elif defined(TCC_TARGET_X86_64)
1004 esym_index
= s1
->symtab_to_dynsym
[sym_index
];
1013 /* allocate the section */
1014 sr
->sh_flags
|= SHF_ALLOC
;
1015 sr
->sh_size
= count
* sizeof(ElfW_Rel
);
1020 static struct sym_attr
*alloc_sym_attr(TCCState
*s1
, int index
)
1023 struct sym_attr
*tab
;
1025 if (index
>= s1
->nb_sym_attrs
) {
1026 /* find immediately bigger power of 2 and reallocate array */
1030 tab
= tcc_realloc(s1
->sym_attrs
, n
* sizeof(*s1
->sym_attrs
));
1031 s1
->sym_attrs
= tab
;
1032 memset(s1
->sym_attrs
+ s1
->nb_sym_attrs
, 0,
1033 (n
- s1
->nb_sym_attrs
) * sizeof(*s1
->sym_attrs
));
1034 s1
->nb_sym_attrs
= n
;
1036 return &s1
->sym_attrs
[index
];
1039 static void build_got(TCCState
*s1
)
1043 /* if no got, then create it */
1044 s1
->got
= new_section(s1
, ".got", SHT_PROGBITS
, SHF_ALLOC
| SHF_WRITE
);
1045 s1
->got
->sh_entsize
= 4;
1046 add_elf_sym(symtab_section
, 0, 4, ELFW(ST_INFO
)(STB_GLOBAL
, STT_OBJECT
),
1047 0, s1
->got
->sh_num
, "_GLOBAL_OFFSET_TABLE_");
1048 ptr
= section_ptr_add(s1
->got
, 3 * PTR_SIZE
);
1050 /* keep space for _DYNAMIC pointer, if present */
1052 /* two dummy got entries */
1053 write32le(ptr
+ 4, 0);
1054 write32le(ptr
+ 8, 0);
1056 /* keep space for _DYNAMIC pointer, if present */
1058 write32le(ptr
+ 4, 0);
1059 /* two dummy got entries */
1060 write32le(ptr
+ 8, 0);
1061 write32le(ptr
+ 12, 0);
1062 write32le(ptr
+ 16, 0);
1063 write32le(ptr
+ 20, 0);
1067 /* put a got or plt entry corresponding to a symbol in symtab_section. 'size'
1068 and 'info' can be modifed if more precise info comes from the DLL.
1069 Returns offset of GOT or PLT slot. */
1070 static unsigned long put_got_entry(TCCState
*s1
,
1071 int reloc_type
, unsigned long size
, int info
,
1074 int index
, need_plt_entry
;
1077 unsigned long offset
;
1079 struct sym_attr
*symattr
;
1085 #ifdef TCC_TARGET_X86_64
1086 (reloc_type
== R_X86_64_JUMP_SLOT
);
1087 #elif defined(TCC_TARGET_I386)
1088 (reloc_type
== R_386_JMP_SLOT
);
1089 #elif defined(TCC_TARGET_ARM)
1090 (reloc_type
== R_ARM_JUMP_SLOT
);
1091 #elif defined(TCC_TARGET_ARM64)
1092 (reloc_type
== R_AARCH64_JUMP_SLOT
);
1097 if (need_plt_entry
&& !s1
->plt
) {
1099 s1
->plt
= new_section(s1
, ".plt", SHT_PROGBITS
,
1100 SHF_ALLOC
| SHF_EXECINSTR
);
1101 s1
->plt
->sh_entsize
= 4;
1104 /* If a got/plt entry already exists for that symbol, no need to add one */
1105 if (sym_index
< s1
->nb_sym_attrs
) {
1106 if (need_plt_entry
&& s1
->sym_attrs
[sym_index
].plt_offset
)
1107 return s1
->sym_attrs
[sym_index
].plt_offset
;
1108 else if (!need_plt_entry
&& s1
->sym_attrs
[sym_index
].got_offset
)
1109 return s1
->sym_attrs
[sym_index
].got_offset
;
1112 symattr
= alloc_sym_attr(s1
, sym_index
);
1114 /* Only store the GOT offset if it's not generated for the PLT entry. */
1115 if (!need_plt_entry
)
1116 symattr
->got_offset
= s1
->got
->data_offset
;
1118 sym
= &((ElfW(Sym
) *)symtab_section
->data
)[sym_index
];
1119 name
= (char *) symtab_section
->link
->data
+ sym
->st_name
;
1120 offset
= sym
->st_value
;
1121 #if defined(TCC_TARGET_I386) || defined(TCC_TARGET_X86_64)
1122 if (need_plt_entry
) {
1126 unsigned long relofs
;
1128 #if defined(TCC_OUTPUT_DLL_WITH_PLT)
1131 /* if we build a DLL, we add a %ebx offset */
1132 if (s1
->output_type
== TCC_OUTPUT_DLL
)
1138 /* add a PLT entry */
1140 if (plt
->data_offset
== 0) {
1141 /* first plt entry */
1142 p
= section_ptr_add(plt
, 16);
1143 p
[0] = 0xff; /* pushl got + PTR_SIZE */
1144 p
[1] = modrm
+ 0x10;
1145 write32le(p
+ 2, PTR_SIZE
);
1146 p
[6] = 0xff; /* jmp *(got + PTR_SIZE * 2) */
1148 write32le(p
+ 8, PTR_SIZE
* 2);
1151 /* The PLT slot refers to the relocation entry it needs
1152 via offset. The reloc entry is created below, so its
1153 offset is the current data_offset. */
1154 relofs
= s1
->got
->reloc
? s1
->got
->reloc
->data_offset
: 0;
1155 symattr
->plt_offset
= plt
->data_offset
;
1156 p
= section_ptr_add(plt
, 16);
1157 p
[0] = 0xff; /* jmp *(got + x) */
1159 write32le(p
+ 2, s1
->got
->data_offset
);
1160 p
[6] = 0x68; /* push $xxx */
1161 #ifdef TCC_TARGET_X86_64
1162 /* On x86-64, the relocation is referred to by _index_. */
1163 write32le(p
+ 7, relofs
/ sizeof (ElfW_Rel
));
1165 write32le(p
+ 7, relofs
);
1167 p
[11] = 0xe9; /* jmp plt_start */
1168 write32le(p
+ 12, -(plt
->data_offset
));
1170 /* If this was an UNDEF symbol set the offset in the
1171 dynsymtab to the PLT slot, so that PC32 relocs to it
1173 if (sym
->st_shndx
== SHN_UNDEF
)
1174 offset
= plt
->data_offset
- 16;
1176 #elif defined(TCC_TARGET_ARM)
1177 if (need_plt_entry
) {
1181 /* if we build a DLL, we add a %ebx offset */
1182 if (s1
->output_type
== TCC_OUTPUT_DLL
)
1183 tcc_error("DLLs unimplemented!");
1185 /* add a PLT entry */
1187 if (plt
->data_offset
== 0) {
1188 /* first plt entry */
1189 p
= section_ptr_add(plt
, 16);
1190 write32le(p
, 0xe52de004); /* push {lr} */
1191 write32le(p
+4, 0xe59fe010); /* ldr lr, [pc, #16] */
1192 write32le(p
+8, 0xe08fe00e); /* add lr, pc, lr */
1193 write32le(p
+12, 0xe5bef008); /* ldr pc, [lr, #8]! */
1196 symattr
->plt_offset
= plt
->data_offset
;
1197 if (symattr
->plt_thumb_stub
) {
1198 p
= section_ptr_add(plt
, 20);
1199 write32le(p
, 0x4778); /* bx pc */
1200 write32le(p
+2, 0x46c0); /* nop */
1203 p
= section_ptr_add(plt
, 16);
1204 write32le(p
, 0xe59fc004); /* ldr ip, [pc, #4] ; GOT entry offset */
1205 write32le(p
+4, 0xe08fc00c); /* add ip, pc, ip ; addr of GOT entry */
1206 write32le(p
+8, 0xe59cf000); /* ldr pc, [ip] ; jump to GOT entry */
1207 write32le(p
+12, s1
->got
->data_offset
); /* GOT entry off once patched */
1209 /* the symbol is modified so that it will be relocated to
1211 if (sym
->st_shndx
== SHN_UNDEF
)
1212 offset
= plt
->data_offset
- 16;
1214 #elif defined(TCC_TARGET_ARM64)
1215 if (need_plt_entry
) {
1219 if (s1
->output_type
== TCC_OUTPUT_DLL
)
1220 tcc_error("DLLs unimplemented!");
1223 if (plt
->data_offset
== 0)
1224 section_ptr_add(plt
, 32);
1225 symattr
->plt_offset
= plt
->data_offset
;
1226 p
= section_ptr_add(plt
, 16);
1227 write32le(p
, s1
->got
->data_offset
);
1228 write32le(p
+ 4, (uint64_t)s1
->got
->data_offset
>> 32);
1230 if (sym
->st_shndx
== SHN_UNDEF
)
1231 offset
= plt
->data_offset
- 16;
1233 #elif defined(TCC_TARGET_C67)
1235 tcc_error("C67 got not implemented");
1238 #error unsupported CPU
1241 /* XXX This might generate multiple syms for name. */
1242 index
= put_elf_sym(s1
->dynsym
, offset
,
1243 size
, info
, 0, sym
->st_shndx
, name
);
1244 /* Create the relocation (it's against the GOT for PLT
1246 put_elf_reloc(s1
->dynsym
, s1
->got
,
1247 s1
->got
->data_offset
,
1250 /* Without .dynsym (i.e. static link or memory output) we
1251 still need relocs against the generated got, so as to fill
1252 the entries with the symbol values (determined later). */
1253 put_elf_reloc(symtab_section
, s1
->got
,
1254 s1
->got
->data_offset
,
1255 reloc_type
, sym_index
);
1257 /* And now create the GOT slot itself. */
1258 ptr
= section_ptr_add(s1
->got
, PTR_SIZE
);
1261 return symattr
->plt_offset
;
1263 return symattr
->got_offset
;
1266 /* build GOT and PLT entries */
1267 ST_FUNC
void build_got_entries(TCCState
*s1
)
1272 int i
, type
, reloc_type
, sym_index
;
1274 for(i
= 1; i
< s1
->nb_sections
; i
++) {
1275 s
= s1
->sections
[i
];
1276 if (s
->sh_type
!= SHT_RELX
)
1278 /* no need to handle got relocations */
1279 if (s
->link
!= symtab_section
)
1281 for_each_elem(s
, 0, rel
, ElfW_Rel
) {
1282 type
= ELFW(R_TYPE
)(rel
->r_info
);
1284 #if defined(TCC_TARGET_I386)
1291 if (type
== R_386_GOT32
|| type
== R_386_PLT32
) {
1292 sym_index
= ELFW(R_SYM
)(rel
->r_info
);
1293 sym
= &((ElfW(Sym
) *)symtab_section
->data
)[sym_index
];
1294 /* look at the symbol got offset. If none, then add one */
1295 if (type
== R_386_GOT32
)
1296 reloc_type
= R_386_GLOB_DAT
;
1298 reloc_type
= R_386_JMP_SLOT
;
1299 put_got_entry(s1
, reloc_type
, sym
->st_size
, sym
->st_info
,
1303 #elif defined(TCC_TARGET_ARM)
1313 sym_index
= ELFW(R_SYM
)(rel
->r_info
);
1314 sym
= &((ElfW(Sym
) *)symtab_section
->data
)[sym_index
];
1315 if (type
!= R_ARM_GOTOFF
&& type
!= R_ARM_GOTPC
1316 && sym
->st_shndx
== SHN_UNDEF
) {
1318 /* look at the symbol got offset. If none, then add one */
1319 if (type
== R_ARM_GOT32
)
1320 reloc_type
= R_ARM_GLOB_DAT
;
1322 reloc_type
= R_ARM_JUMP_SLOT
;
1323 ofs
= put_got_entry(s1
, reloc_type
, sym
->st_size
,
1324 sym
->st_info
, sym_index
);
1326 printf ("maybegot: %s, %d, %d --> ofs=0x%x\n",
1327 (char *) symtab_section
->link
->data
+ sym
->st_name
,
1328 type
, sym
->st_shndx
, ofs
);
1330 if (type
!= R_ARM_GOT32
) {
1331 addr_t
*ptr
= (addr_t
*)(s1
->sections
[s
->sh_info
]->data
1333 /* x must be signed! */
1334 int x
= *ptr
& 0xffffff;
1340 printf ("insn=0x%x --> 0x%x (x==0x%x)\n", *ptr
,
1341 (*ptr
& 0xff000000) | x
, x
);
1343 *ptr
= (*ptr
& 0xff000000) | x
;
1347 case R_ARM_THM_JUMP24
:
1348 sym_index
= ELFW(R_SYM
)(rel
->r_info
);
1349 sym
= &((ElfW(Sym
) *)symtab_section
->data
)[sym_index
];
1350 /* We are relocating a jump from thumb code to arm code */
1351 if (sym
->st_shndx
!= SHN_UNDEF
&& !(sym
->st_value
& 1)) {
1354 char *name
, buf
[1024];
1355 Section
*text_section
;
1357 name
= (char *) symtab_section
->link
->data
+ sym
->st_name
;
1358 text_section
= s1
->sections
[sym
->st_shndx
];
1359 /* Modify reloc to target a thumb stub to switch to ARM */
1360 snprintf(buf
, sizeof(buf
), "%s_from_thumb", name
);
1361 index
= put_elf_sym(symtab_section
,
1362 text_section
->data_offset
+ 1,
1363 sym
->st_size
, sym
->st_info
, 0,
1364 sym
->st_shndx
, buf
);
1365 rel
->r_info
= ELFW(R_INFO
)(index
, type
);
1366 /* Create a thumb stub fonction to switch to ARM mode */
1367 put_elf_reloc(symtab_section
, text_section
,
1368 text_section
->data_offset
+ 4, R_ARM_JUMP24
,
1370 p
= section_ptr_add(text_section
, 8);
1371 write32le(p
, 0x4778); /* bx pc */
1372 write32le(p
+2, 0x46c0); /* nop */
1373 write32le(p
+4, 0xeafffffe); /* b $sym */
1375 #elif defined(TCC_TARGET_ARM64)
1376 //xx Other cases may be required here:
1377 case R_AARCH64_ADR_GOT_PAGE
:
1378 case R_AARCH64_LD64_GOT_LO12_NC
:
1381 sym_index
= ELFW(R_SYM
)(rel
->r_info
);
1382 sym
= &((ElfW(Sym
) *)symtab_section
->data
)[sym_index
];
1383 reloc_type
= R_AARCH64_GLOB_DAT
;
1384 put_got_entry(s1
, reloc_type
, sym
->st_size
, sym
->st_info
,
1388 case R_AARCH64_JUMP26
:
1389 case R_AARCH64_CALL26
:
1392 sym_index
= ELFW(R_SYM
)(rel
->r_info
);
1393 sym
= &((ElfW(Sym
) *)symtab_section
->data
)[sym_index
];
1394 if (sym
->st_shndx
== SHN_UNDEF
) {
1396 reloc_type
= R_AARCH64_JUMP_SLOT
;
1397 ofs
= put_got_entry(s1
, reloc_type
, sym
->st_size
,
1398 sym
->st_info
, sym_index
);
1399 /* We store the place of the generated PLT slot
1401 rel
->r_addend
+= ofs
;
1404 #elif defined(TCC_TARGET_C67)
1411 if (type
== R_C60_GOT32
|| type
== R_C60_PLT32
) {
1412 sym_index
= ELFW(R_SYM
)(rel
->r_info
);
1413 sym
= &((ElfW(Sym
) *)symtab_section
->data
)[sym_index
];
1414 /* look at the symbol got offset. If none, then add one */
1415 if (type
== R_C60_GOT32
)
1416 reloc_type
= R_C60_GLOB_DAT
;
1418 reloc_type
= R_C60_JMP_SLOT
;
1419 put_got_entry(s1
, reloc_type
, sym
->st_size
, sym
->st_info
,
1423 #elif defined(TCC_TARGET_X86_64)
1424 case R_X86_64_GOT32
:
1425 case R_X86_64_GOTTPOFF
:
1426 case R_X86_64_GOTPCREL
:
1427 case 41 /* R_X86_64_GOTPCRELX */:
1428 case 42 /* R_X86_64_REX_GOTPCRELX */:
1429 case R_X86_64_PLT32
:
1430 sym_index
= ELFW(R_SYM
)(rel
->r_info
);
1431 sym
= &((ElfW(Sym
) *)symtab_section
->data
)[sym_index
];
1432 if (type
== R_X86_64_PLT32
&&
1433 ELFW(ST_VISIBILITY
)(sym
->st_other
) != STV_DEFAULT
)
1435 rel
->r_info
= ELFW(R_INFO
)(sym_index
, R_X86_64_PC32
);
1441 sym
= &((ElfW(Sym
) *)symtab_section
->data
)[sym_index
];
1443 if (type
== R_X86_64_GOT32
|| type
== R_X86_64_GOTPCREL
||
1444 type
== R_X86_64_PLT32
|| type
== 41 || type
== 42) {
1446 /* look at the symbol got offset. If none, then add one */
1447 if (type
== R_X86_64_PLT32
)
1448 reloc_type
= R_X86_64_JUMP_SLOT
;
1450 reloc_type
= R_X86_64_GLOB_DAT
;
1451 ofs
= put_got_entry(s1
, reloc_type
, sym
->st_size
,
1452 sym
->st_info
, sym_index
);
1453 if (type
== R_X86_64_PLT32
)
1454 /* We store the place of the generated PLT slot
1456 rel
->r_addend
+= ofs
;
1460 #error unsupported CPU
1469 ST_FUNC Section
*new_symtab(TCCState
*s1
,
1470 const char *symtab_name
, int sh_type
, int sh_flags
,
1471 const char *strtab_name
,
1472 const char *hash_name
, int hash_sh_flags
)
1474 Section
*symtab
, *strtab
, *hash
;
1475 int *ptr
, nb_buckets
;
1477 symtab
= new_section(s1
, symtab_name
, sh_type
, sh_flags
);
1478 symtab
->sh_entsize
= sizeof(ElfW(Sym
));
1479 strtab
= new_section(s1
, strtab_name
, SHT_STRTAB
, sh_flags
);
1480 put_elf_str(strtab
, "");
1481 symtab
->link
= strtab
;
1482 put_elf_sym(symtab
, 0, 0, 0, 0, 0, NULL
);
1486 hash
= new_section(s1
, hash_name
, SHT_HASH
, hash_sh_flags
);
1487 hash
->sh_entsize
= sizeof(int);
1488 symtab
->hash
= hash
;
1489 hash
->link
= symtab
;
1491 ptr
= section_ptr_add(hash
, (2 + nb_buckets
+ 1) * sizeof(int));
1492 ptr
[0] = nb_buckets
;
1494 memset(ptr
+ 2, 0, (nb_buckets
+ 1) * sizeof(int));
1498 /* put dynamic tag */
1499 static void put_dt(Section
*dynamic
, int dt
, addr_t val
)
1502 dyn
= section_ptr_add(dynamic
, sizeof(ElfW(Dyn
)));
1504 dyn
->d_un
.d_val
= val
;
1507 static void add_init_array_defines(TCCState
*s1
, const char *section_name
)
1511 char sym_start
[1024];
1514 snprintf(sym_start
, sizeof(sym_start
), "__%s_start", section_name
+ 1);
1515 snprintf(sym_end
, sizeof(sym_end
), "__%s_end", section_name
+ 1);
1517 s
= find_section(s1
, section_name
);
1522 end_offset
= s
->data_offset
;
1525 add_elf_sym(symtab_section
,
1527 ELFW(ST_INFO
)(STB_GLOBAL
, STT_NOTYPE
), 0,
1528 s
->sh_num
, sym_start
);
1529 add_elf_sym(symtab_section
,
1531 ELFW(ST_INFO
)(STB_GLOBAL
, STT_NOTYPE
), 0,
1532 s
->sh_num
, sym_end
);
1535 static int tcc_add_support(TCCState
*s1
, const char *filename
)
1538 snprintf(buf
, sizeof(buf
), "%s/%s/%s", s1
->tcc_lib_path
,
1539 /* an cpu specific path inside tcc_lib_path, mainly for keeping libtcc1.a */
1540 #ifdef TCC_TARGET_I386
1543 #ifdef TCC_TARGET_X86_64
1546 #ifdef TCC_TARGET_ARM
1549 #ifdef TCC_TARGET_ARM64
1552 #ifdef TCC_TARGET_C67
1557 return tcc_add_file(s1
, buf
, TCC_FILETYPE_BINARY
);
1560 ST_FUNC
void tcc_add_bcheck(TCCState
*s1
)
1562 #ifdef CONFIG_TCC_BCHECK
1565 if (0 == s1
->do_bounds_check
)
1568 /* XXX: add an object file to do that */
1569 ptr
= section_ptr_add(bounds_section
, sizeof(*ptr
));
1571 add_elf_sym(symtab_section
, 0, 0,
1572 ELFW(ST_INFO
)(STB_GLOBAL
, STT_NOTYPE
), 0,
1573 bounds_section
->sh_num
, "__bounds_start");
1574 if (s1
->output_type
!= TCC_OUTPUT_MEMORY
) {
1575 /* add 'call __bound_init()' in .init section */
1577 /* XXX not called on MSYS, reason is unknown. For this
1578 case a call to __bound_init is performed in bcheck.c
1579 when __bound_ptr_add, __bound_new_region,
1580 __bound_delete_region called */
1582 int sym_index
= find_elf_sym(symtab_section
, "__bound_init");
1584 Section
*init_section
= find_section(s1
, ".init");
1585 unsigned char *pinit
= section_ptr_add(init_section
, 5);
1587 write32le(pinit
+ 1, -4);
1588 put_elf_reloc(symtab_section
, init_section
,
1589 init_section
->data_offset
- 4, R_386_PC32
, sym_index
);
1592 tcc_warning("__bound_init not defined");
1597 /* add tcc runtime libraries */
1598 ST_FUNC
void tcc_add_runtime(TCCState
*s1
)
1600 tcc_add_pragma_libs(s1
);
1603 if (!s1
->nostdlib
) {
1604 tcc_add_library(s1
, "c");
1605 #ifdef CONFIG_USE_LIBGCC
1606 if (!s1
->static_link
) {
1607 tcc_add_file(s1
, TCC_LIBGCC
, TCC_FILETYPE_BINARY
);
1610 tcc_add_support(s1
, "libtcc1.a");
1613 /* tcc_add_bcheck tries to relocate a call to __bound_init in _init so
1614 libtcc1.a must be loaded before for __bound_init to be defined and
1615 crtn.o must be loaded after to not finalize _init too early. */
1618 if (!s1
->nostdlib
) {
1619 /* add crt end if not memory output */
1620 if (s1
->output_type
!= TCC_OUTPUT_MEMORY
)
1621 tcc_add_crt(s1
, "crtn.o");
1625 /* add various standard linker symbols (must be done after the
1626 sections are filled (for example after allocating common
1628 ST_FUNC
void tcc_add_linker_symbols(TCCState
*s1
)
1634 add_elf_sym(symtab_section
,
1635 text_section
->data_offset
, 0,
1636 ELFW(ST_INFO
)(STB_GLOBAL
, STT_NOTYPE
), 0,
1637 text_section
->sh_num
, "_etext");
1638 add_elf_sym(symtab_section
,
1639 data_section
->data_offset
, 0,
1640 ELFW(ST_INFO
)(STB_GLOBAL
, STT_NOTYPE
), 0,
1641 data_section
->sh_num
, "_edata");
1642 add_elf_sym(symtab_section
,
1643 bss_section
->data_offset
, 0,
1644 ELFW(ST_INFO
)(STB_GLOBAL
, STT_NOTYPE
), 0,
1645 bss_section
->sh_num
, "_end");
1646 /* horrible new standard ldscript defines */
1647 add_init_array_defines(s1
, ".preinit_array");
1648 add_init_array_defines(s1
, ".init_array");
1649 add_init_array_defines(s1
, ".fini_array");
1651 /* add start and stop symbols for sections whose name can be
1653 for(i
= 1; i
< s1
->nb_sections
; i
++) {
1654 s
= s1
->sections
[i
];
1655 if (s
->sh_type
== SHT_PROGBITS
&&
1656 (s
->sh_flags
& SHF_ALLOC
)) {
1660 /* check if section name can be expressed in C */
1666 if (!isid(ch
) && !isnum(ch
))
1670 snprintf(buf
, sizeof(buf
), "__start_%s", s
->name
);
1671 add_elf_sym(symtab_section
,
1673 ELFW(ST_INFO
)(STB_GLOBAL
, STT_NOTYPE
), 0,
1675 snprintf(buf
, sizeof(buf
), "__stop_%s", s
->name
);
1676 add_elf_sym(symtab_section
,
1678 ELFW(ST_INFO
)(STB_GLOBAL
, STT_NOTYPE
), 0,
1685 static void tcc_output_binary(TCCState
*s1
, FILE *f
,
1686 const int *sec_order
)
1689 int i
, offset
, size
;
1692 for(i
=1;i
<s1
->nb_sections
;i
++) {
1693 s
= s1
->sections
[sec_order
[i
]];
1694 if (s
->sh_type
!= SHT_NOBITS
&&
1695 (s
->sh_flags
& SHF_ALLOC
)) {
1696 while (offset
< s
->sh_offset
) {
1701 fwrite(s
->data
, 1, size
, f
);
1707 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
1709 #define EXTRA_RELITEMS 14
1711 /* move the relocation value from .dynsym to .got */
1712 void patch_dynsym_undef(TCCState
*s1
, Section
*s
)
1714 uint32_t *gotd
= (void *)s1
->got
->data
;
1717 gotd
+= 3; /* dummy entries in .got */
1718 /* relocate symbols in .dynsym */
1719 for_each_elem(s
, 1, sym
, ElfW(Sym
)) {
1720 if (sym
->st_shndx
== SHN_UNDEF
) {
1721 *gotd
++ = sym
->st_value
+ 6; /* XXX 6 is magic ? */
1728 #define EXTRA_RELITEMS 9
1730 /* zero plt offsets of weak symbols in .dynsym */
1731 void patch_dynsym_undef(TCCState
*s1
, Section
*s
)
1735 for_each_elem(s
, 1, sym
, ElfW(Sym
))
1736 if (sym
->st_shndx
== SHN_UNDEF
&& ELFW(ST_BIND
)(sym
->st_info
) == STB_WEAK
)
1741 ST_FUNC
void fill_got_entry(TCCState
*s1
, ElfW_Rel
*rel
)
1743 int sym_index
= ELFW(R_SYM
) (rel
->r_info
);
1744 ElfW(Sym
) *sym
= &((ElfW(Sym
) *) symtab_section
->data
)[sym_index
];
1745 unsigned long offset
;
1747 if (sym_index
>= s1
->nb_sym_attrs
)
1749 offset
= s1
->sym_attrs
[sym_index
].got_offset
;
1750 section_reserve(s1
->got
, offset
+ PTR_SIZE
);
1751 #ifdef TCC_TARGET_X86_64
1752 /* only works for x86-64 */
1753 write32le(s1
->got
->data
+ offset
+ 4, sym
->st_value
>> 32);
1755 write32le(s1
->got
->data
+ offset
, sym
->st_value
& 0xffffffff);
1758 /* Perform relocation to GOT or PLT entries */
1759 ST_FUNC
void fill_got(TCCState
*s1
)
1765 for(i
= 1; i
< s1
->nb_sections
; i
++) {
1766 s
= s1
->sections
[i
];
1767 if (s
->sh_type
!= SHT_RELX
)
1769 /* no need to handle got relocations */
1770 if (s
->link
!= symtab_section
)
1772 for_each_elem(s
, 0, rel
, ElfW_Rel
) {
1773 switch (ELFW(R_TYPE
) (rel
->r_info
)) {
1774 case R_X86_64_GOT32
:
1775 case R_X86_64_GOTPCREL
:
1776 case 41 /* R_X86_64_GOTPCRELX */:
1777 case 42 /* R_X86_64_REX_GOTPCRELX */:
1778 case R_X86_64_PLT32
:
1779 fill_got_entry(s1
, rel
);
1786 /* Bind symbols of executable: resolve undefined symbols from exported symbols
1787 in shared libraries and export non local defined symbols to shared libraries
1788 if -rdynamic switch was given on command line */
1789 static void bind_exe_dynsyms(TCCState
*s1
)
1792 int sym_index
, index
;
1793 ElfW(Sym
) *sym
, *esym
;
1796 /* Resolve undefined symbols from dynamic symbols. When there is a match:
1797 - if STT_FUNC or STT_GNU_IFUNC symbol -> add it in PLT
1798 - if STT_OBJECT symbol -> add it in .bss section with suitable reloc */
1799 for_each_elem(symtab_section
, 1, sym
, ElfW(Sym
)) {
1800 if (sym
->st_shndx
== SHN_UNDEF
) {
1801 name
= (char *) symtab_section
->link
->data
+ sym
->st_name
;
1802 sym_index
= find_elf_sym(s1
->dynsymtab_section
, name
);
1804 esym
= &((ElfW(Sym
) *)s1
->dynsymtab_section
->data
)[sym_index
];
1805 type
= ELFW(ST_TYPE
)(esym
->st_info
);
1806 if ((type
== STT_FUNC
) || (type
== STT_GNU_IFUNC
)) {
1807 /* Indirect functions shall have STT_FUNC type in executable
1808 * dynsym section. Indeed, a dlsym call following a lazy
1809 * resolution would pick the symbol value from the
1810 * executable dynsym entry which would contain the address
1811 * of the function wanted by the caller of dlsym instead of
1812 * the address of the function that would return that
1814 put_got_entry(s1
, R_JMP_SLOT
, esym
->st_size
,
1815 ELFW(ST_INFO
)(STB_GLOBAL
,STT_FUNC
),
1816 sym
- (ElfW(Sym
) *)symtab_section
->data
);
1817 } else if (type
== STT_OBJECT
) {
1818 unsigned long offset
;
1820 offset
= bss_section
->data_offset
;
1821 /* XXX: which alignment ? */
1822 offset
= (offset
+ 16 - 1) & -16;
1823 index
= put_elf_sym(s1
->dynsym
, offset
, esym
->st_size
,
1824 esym
->st_info
, 0, bss_section
->sh_num
,
1826 /* Ensure R_COPY works for weak symbol aliases */
1827 if (ELFW(ST_BIND
)(esym
->st_info
) == STB_WEAK
) {
1828 for_each_elem(s1
->dynsymtab_section
, 1, dynsym
, ElfW(Sym
)) {
1829 if ((dynsym
->st_value
== esym
->st_value
)
1830 && (ELFW(ST_BIND
)(dynsym
->st_info
) == STB_GLOBAL
)) {
1831 char *dynname
= (char *) s1
->dynsymtab_section
->link
->data
1833 put_elf_sym(s1
->dynsym
, offset
, dynsym
->st_size
,
1835 bss_section
->sh_num
, dynname
);
1840 put_elf_reloc(s1
->dynsym
, bss_section
,
1841 offset
, R_COPY
, index
);
1842 offset
+= esym
->st_size
;
1843 bss_section
->data_offset
= offset
;
1846 /* STB_WEAK undefined symbols are accepted */
1847 /* XXX: _fp_hw seems to be part of the ABI, so we ignore it */
1848 if (ELFW(ST_BIND
)(sym
->st_info
) == STB_WEAK
||
1849 !strcmp(name
, "_fp_hw")) {
1851 tcc_error_noabort("undefined symbol '%s'", name
);
1854 } else if (s1
->rdynamic
&& ELFW(ST_BIND
)(sym
->st_info
) != STB_LOCAL
) {
1855 /* if -rdynamic option, then export all non local symbols */
1856 name
= (char *) symtab_section
->link
->data
+ sym
->st_name
;
1857 put_elf_sym(s1
->dynsym
, sym
->st_value
, sym
->st_size
, sym
->st_info
,
1858 0, sym
->st_shndx
, name
);
1863 /* Bind symbols of libraries: export non local symbols of executable that
1864 resolve undefined symbols of shared libraries */
1865 static void bind_libs_dynsyms(TCCState
*s1
)
1869 ElfW(Sym
) *sym
, *esym
;
1871 /* now look at unresolved dynamic symbols and export
1872 corresponding symbol */
1873 for_each_elem(s1
->dynsymtab_section
, 1, esym
, ElfW(Sym
)) {
1874 name
= (char *) s1
->dynsymtab_section
->link
->data
+ esym
->st_name
;
1875 sym_index
= find_elf_sym(symtab_section
, name
);
1877 /* XXX: avoid adding a symbol if already present because of
1879 sym
= &((ElfW(Sym
) *)symtab_section
->data
)[sym_index
];
1880 if (sym
->st_shndx
!= SHN_UNDEF
)
1881 put_elf_sym(s1
->dynsym
, sym
->st_value
, sym
->st_size
,
1882 sym
->st_info
, 0, sym
->st_shndx
, name
);
1883 } else if (esym
->st_shndx
== SHN_UNDEF
) {
1884 /* weak symbols can stay undefined */
1885 if (ELFW(ST_BIND
)(esym
->st_info
) != STB_WEAK
)
1886 tcc_warning("undefined dynamic symbol '%s'", name
);
1891 /* Export all non local symbols (for shared libraries) */
1892 static void export_global_syms(TCCState
*s1
)
1894 int nb_syms
, dynindex
, index
;
1898 nb_syms
= symtab_section
->data_offset
/ sizeof(ElfW(Sym
));
1899 s1
->symtab_to_dynsym
= tcc_mallocz(sizeof(int) * nb_syms
);
1900 for_each_elem(symtab_section
, 1, sym
, ElfW(Sym
)) {
1901 if (ELFW(ST_BIND
)(sym
->st_info
) != STB_LOCAL
) {
1902 name
= (char *) symtab_section
->link
->data
+ sym
->st_name
;
1903 dynindex
= put_elf_sym(s1
->dynsym
, sym
->st_value
, sym
->st_size
,
1904 sym
->st_info
, 0, sym
->st_shndx
, name
);
1905 index
= sym
- (ElfW(Sym
) *) symtab_section
->data
;
1906 s1
->symtab_to_dynsym
[index
] = dynindex
;
1911 /* relocate the PLT: compute addresses and offsets in the PLT now that final
1912 address for PLT and GOT are known (see fill_program_header) */
1913 ST_FUNC
void relocate_plt(TCCState
*s1
)
1921 p_end
= p
+ s1
->plt
->data_offset
;
1923 #if defined(TCC_TARGET_I386)
1924 write32le(p
+ 2, read32le(p
+ 2) + s1
->got
->sh_addr
);
1925 write32le(p
+ 8, read32le(p
+ 8) + s1
->got
->sh_addr
);
1928 write32le(p
+ 2, read32le(p
+ 2) + s1
->got
->sh_addr
);
1931 #elif defined(TCC_TARGET_X86_64)
1932 int x
= s1
->got
->sh_addr
- s1
->plt
->sh_addr
- 6;
1933 write32le(p
+ 2, read32le(p
+ 2) + x
);
1934 write32le(p
+ 8, read32le(p
+ 8) + x
- 6);
1937 write32le(p
+ 2, read32le(p
+ 2) + x
+ s1
->plt
->data
- p
);
1940 #elif defined(TCC_TARGET_ARM)
1942 x
=s1
->got
->sh_addr
- s1
->plt
->sh_addr
- 12;
1945 if (read32le(p
) == 0x46c04778) /* PLT Thumb stub present */
1947 write32le(p
+ 12, x
+ read32le(p
+ 12) + s1
->plt
->data
- p
);
1950 #elif defined(TCC_TARGET_ARM64)
1951 uint64_t plt
= s1
->plt
->sh_addr
;
1952 uint64_t got
= s1
->got
->sh_addr
;
1953 uint64_t off
= (got
>> 12) - (plt
>> 12);
1954 if ((off
+ ((uint32_t)1 << 20)) >> 21)
1955 tcc_error("Failed relocating PLT (off=0x%lx, got=0x%lx, plt=0x%lx)", off
, got
, plt
);
1956 write32le(p
, 0xa9bf7bf0); // stp x16,x30,[sp,#-16]!
1957 write32le(p
+ 4, (0x90000010 | // adrp x16,...
1958 (off
& 0x1ffffc) << 3 | (off
& 3) << 29));
1959 write32le(p
+ 8, (0xf9400211 | // ldr x17,[x16,#...]
1960 (got
& 0xff8) << 7));
1961 write32le(p
+ 12, (0x91000210 | // add x16,x16,#...
1962 (got
& 0xfff) << 10));
1963 write32le(p
+ 16, 0xd61f0220); // br x17
1964 write32le(p
+ 20, 0xd503201f); // nop
1965 write32le(p
+ 24, 0xd503201f); // nop
1966 write32le(p
+ 28, 0xd503201f); // nop
1969 uint64_t pc
= plt
+ (p
- s1
->plt
->data
);
1970 uint64_t addr
= got
+ read64le(p
);
1971 uint64_t off
= (addr
>> 12) - (pc
>> 12);
1972 if ((off
+ ((uint32_t)1 << 20)) >> 21)
1973 tcc_error("Failed relocating PLT (off=0x%lx, addr=0x%lx, pc=0x%lx)", off
, addr
, pc
);
1974 write32le(p
, (0x90000010 | // adrp x16,...
1975 (off
& 0x1ffffc) << 3 | (off
& 3) << 29));
1976 write32le(p
+ 4, (0xf9400211 | // ldr x17,[x16,#...]
1977 (addr
& 0xff8) << 7));
1978 write32le(p
+ 8, (0x91000210 | // add x16,x16,#...
1979 (addr
& 0xfff) << 10));
1980 write32le(p
+ 12, 0xd61f0220); // br x17
1983 #elif defined(TCC_TARGET_C67)
1986 #error unsupported CPU
1991 /* Allocate strings for section names and decide if an unallocated section
1994 NOTE: the strsec section comes last, so its size is also correct ! */
1995 static void alloc_sec_names(TCCState
*s1
, int file_type
, Section
*strsec
)
2000 /* Allocate strings for section names */
2001 for(i
= 1; i
< s1
->nb_sections
; i
++) {
2002 s
= s1
->sections
[i
];
2003 s
->sh_name
= put_elf_str(strsec
, s
->name
);
2004 /* when generating a DLL, we include relocations but we may
2006 if (file_type
== TCC_OUTPUT_DLL
&&
2007 s
->sh_type
== SHT_RELX
&&
2008 !(s
->sh_flags
& SHF_ALLOC
)) {
2009 /* gr: avoid bogus relocs for empty (debug) sections */
2010 if (s1
->sections
[s
->sh_info
]->sh_flags
& SHF_ALLOC
)
2011 prepare_dynamic_rel(s1
, s
);
2012 else if (s1
->do_debug
)
2013 s
->sh_size
= s
->data_offset
;
2014 } else if (s1
->do_debug
||
2015 file_type
== TCC_OUTPUT_OBJ
||
2016 file_type
== TCC_OUTPUT_EXE
||
2017 (s
->sh_flags
& SHF_ALLOC
) ||
2018 i
== (s1
->nb_sections
- 1)) {
2019 /* we output all sections if debug or object file */
2020 s
->sh_size
= s
->data_offset
;
2025 /* Info to be copied in dynamic section */
2029 unsigned long dyn_rel_off
;
2032 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
2038 /* Assign sections to segments and decide how are sections laid out when loaded
2039 in memory. This function also fills corresponding program headers. */
2040 static int layout_sections(TCCState
*s1
, ElfW(Phdr
) *phdr
, int phnum
,
2041 Section
*interp
, Section
* strsec
,
2042 struct dyn_inf
*dyninf
, int *sec_order
)
2044 int i
, j
, k
, file_type
, sh_order_index
, file_offset
;
2045 unsigned long s_align
;
2051 file_type
= s1
->output_type
;
2054 if (s1
->output_format
== TCC_OUTPUT_FORMAT_ELF
)
2055 file_offset
= sizeof(ElfW(Ehdr
)) + phnum
* sizeof(ElfW(Phdr
));
2056 s_align
= ELF_PAGE_SIZE
;
2057 if (s1
->section_align
)
2058 s_align
= s1
->section_align
;
2061 if (s1
->has_text_addr
) {
2062 int a_offset
, p_offset
;
2063 addr
= s1
->text_addr
;
2064 /* we ensure that (addr % ELF_PAGE_SIZE) == file_offset %
2066 a_offset
= (int) (addr
& (s_align
- 1));
2067 p_offset
= file_offset
& (s_align
- 1);
2068 if (a_offset
< p_offset
)
2069 a_offset
+= s_align
;
2070 file_offset
+= (a_offset
- p_offset
);
2072 if (file_type
== TCC_OUTPUT_DLL
)
2075 addr
= ELF_START_ADDR
;
2076 /* compute address after headers */
2077 addr
+= (file_offset
& (s_align
- 1));
2081 /* Leave one program headers for the program interpreter and one for
2082 the program header table itself if needed. These are done later as
2083 they require section layout to be done first. */
2085 ph
+= 1 + HAVE_PHDR
;
2087 /* dynamic relocation table information, for .dynamic section */
2088 dyninf
->rel_addr
= dyninf
->rel_size
= 0;
2089 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
2090 dyninf
->bss_addr
= dyninf
->bss_size
= 0;
2093 for(j
= 0; j
< 2; j
++) {
2094 ph
->p_type
= PT_LOAD
;
2096 ph
->p_flags
= PF_R
| PF_X
;
2098 ph
->p_flags
= PF_R
| PF_W
;
2099 ph
->p_align
= s_align
;
2101 /* Decide the layout of sections loaded in memory. This must
2102 be done before program headers are filled since they contain
2103 info about the layout. We do the following ordering: interp,
2104 symbol tables, relocations, progbits, nobits */
2105 /* XXX: do faster and simpler sorting */
2106 for(k
= 0; k
< 5; k
++) {
2107 for(i
= 1; i
< s1
->nb_sections
; i
++) {
2108 s
= s1
->sections
[i
];
2109 /* compute if section should be included */
2111 if ((s
->sh_flags
& (SHF_ALLOC
| SHF_WRITE
)) !=
2115 if ((s
->sh_flags
& (SHF_ALLOC
| SHF_WRITE
)) !=
2116 (SHF_ALLOC
| SHF_WRITE
))
2122 } else if (s
->sh_type
== SHT_DYNSYM
||
2123 s
->sh_type
== SHT_STRTAB
||
2124 s
->sh_type
== SHT_HASH
) {
2127 } else if (s
->sh_type
== SHT_RELX
) {
2130 } else if (s
->sh_type
== SHT_NOBITS
) {
2137 sec_order
[sh_order_index
++] = i
;
2139 /* section matches: we align it and add its size */
2141 addr
= (addr
+ s
->sh_addralign
- 1) &
2142 ~(s
->sh_addralign
- 1);
2143 file_offset
+= (int) ( addr
- tmp
);
2144 s
->sh_offset
= file_offset
;
2147 /* update program header infos */
2148 if (ph
->p_offset
== 0) {
2149 ph
->p_offset
= file_offset
;
2151 ph
->p_paddr
= ph
->p_vaddr
;
2153 /* update dynamic relocation infos */
2154 if (s
->sh_type
== SHT_RELX
) {
2155 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
2156 if (!strcmp(strsec
->data
+ s
->sh_name
, ".rel.got")) {
2157 dyninf
->rel_addr
= addr
;
2158 dyninf
->rel_size
+= s
->sh_size
; /* XXX only first rel. */
2160 if (!strcmp(strsec
->data
+ s
->sh_name
, ".rel.bss")) {
2161 dyninf
->bss_addr
= addr
;
2162 dyninf
->bss_size
= s
->sh_size
; /* XXX only first rel. */
2165 if (dyninf
->rel_size
== 0)
2166 dyninf
->rel_addr
= addr
;
2167 dyninf
->rel_size
+= s
->sh_size
;
2171 if (s
->sh_type
!= SHT_NOBITS
)
2172 file_offset
+= s
->sh_size
;
2176 /* Make the first PT_LOAD segment include the program
2177 headers itself (and the ELF header as well), it'll
2178 come out with same memory use but will make various
2179 tools like binutils strip work better. */
2180 ph
->p_offset
&= ~(ph
->p_align
- 1);
2181 ph
->p_vaddr
&= ~(ph
->p_align
- 1);
2182 ph
->p_paddr
&= ~(ph
->p_align
- 1);
2184 ph
->p_filesz
= file_offset
- ph
->p_offset
;
2185 ph
->p_memsz
= addr
- ph
->p_vaddr
;
2188 if (s1
->output_format
== TCC_OUTPUT_FORMAT_ELF
) {
2189 /* if in the middle of a page, we duplicate the page in
2190 memory so that one copy is RX and the other is RW */
2191 if ((addr
& (s_align
- 1)) != 0)
2194 addr
= (addr
+ s_align
- 1) & ~(s_align
- 1);
2195 file_offset
= (file_offset
+ s_align
- 1) & ~(s_align
- 1);
2201 /* all other sections come after */
2202 for(i
= 1; i
< s1
->nb_sections
; i
++) {
2203 s
= s1
->sections
[i
];
2204 if (phnum
> 0 && (s
->sh_flags
& SHF_ALLOC
))
2206 sec_order
[sh_order_index
++] = i
;
2208 file_offset
= (file_offset
+ s
->sh_addralign
- 1) &
2209 ~(s
->sh_addralign
- 1);
2210 s
->sh_offset
= file_offset
;
2211 if (s
->sh_type
!= SHT_NOBITS
)
2212 file_offset
+= s
->sh_size
;
2218 static void fill_unloadable_phdr(ElfW(Phdr
) *phdr
, int phnum
, Section
*interp
,
2223 /* if interpreter, then add corresponding program header */
2229 int len
= phnum
* sizeof(ElfW(Phdr
));
2231 ph
->p_type
= PT_PHDR
;
2232 ph
->p_offset
= sizeof(ElfW(Ehdr
));
2233 ph
->p_vaddr
= interp
->sh_addr
- len
;
2234 ph
->p_paddr
= ph
->p_vaddr
;
2235 ph
->p_filesz
= ph
->p_memsz
= len
;
2236 ph
->p_flags
= PF_R
| PF_X
;
2237 ph
->p_align
= 4; /* interp->sh_addralign; */
2241 ph
->p_type
= PT_INTERP
;
2242 ph
->p_offset
= interp
->sh_offset
;
2243 ph
->p_vaddr
= interp
->sh_addr
;
2244 ph
->p_paddr
= ph
->p_vaddr
;
2245 ph
->p_filesz
= interp
->sh_size
;
2246 ph
->p_memsz
= interp
->sh_size
;
2248 ph
->p_align
= interp
->sh_addralign
;
2251 /* if dynamic section, then add corresponding program header */
2253 ph
= &phdr
[phnum
- 1];
2255 ph
->p_type
= PT_DYNAMIC
;
2256 ph
->p_offset
= dynamic
->sh_offset
;
2257 ph
->p_vaddr
= dynamic
->sh_addr
;
2258 ph
->p_paddr
= ph
->p_vaddr
;
2259 ph
->p_filesz
= dynamic
->sh_size
;
2260 ph
->p_memsz
= dynamic
->sh_size
;
2261 ph
->p_flags
= PF_R
| PF_W
;
2262 ph
->p_align
= dynamic
->sh_addralign
;
2266 /* Fill the dynamic section with tags describing the address and size of
2268 static void fill_dynamic(TCCState
*s1
, struct dyn_inf
*dyninf
)
2272 dynamic
= dyninf
->dynamic
;
2274 /* put dynamic section entries */
2275 dynamic
->data_offset
= dyninf
->dyn_rel_off
;
2276 put_dt(dynamic
, DT_HASH
, s1
->dynsym
->hash
->sh_addr
);
2277 put_dt(dynamic
, DT_STRTAB
, dyninf
->dynstr
->sh_addr
);
2278 put_dt(dynamic
, DT_SYMTAB
, s1
->dynsym
->sh_addr
);
2279 put_dt(dynamic
, DT_STRSZ
, dyninf
->dynstr
->data_offset
);
2280 put_dt(dynamic
, DT_SYMENT
, sizeof(ElfW(Sym
)));
2281 #if defined(TCC_TARGET_ARM64) || defined(TCC_TARGET_X86_64)
2282 put_dt(dynamic
, DT_RELA
, dyninf
->rel_addr
);
2283 put_dt(dynamic
, DT_RELASZ
, dyninf
->rel_size
);
2284 put_dt(dynamic
, DT_RELAENT
, sizeof(ElfW_Rel
));
2286 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
2287 put_dt(dynamic
, DT_PLTGOT
, s1
->got
->sh_addr
);
2288 put_dt(dynamic
, DT_PLTRELSZ
, dyninf
->rel_size
);
2289 put_dt(dynamic
, DT_JMPREL
, dyninf
->rel_addr
);
2290 put_dt(dynamic
, DT_PLTREL
, DT_REL
);
2291 put_dt(dynamic
, DT_REL
, dyninf
->bss_addr
);
2292 put_dt(dynamic
, DT_RELSZ
, dyninf
->bss_size
);
2294 put_dt(dynamic
, DT_REL
, dyninf
->rel_addr
);
2295 put_dt(dynamic
, DT_RELSZ
, dyninf
->rel_size
);
2296 put_dt(dynamic
, DT_RELENT
, sizeof(ElfW_Rel
));
2300 put_dt(dynamic
, DT_DEBUG
, 0);
2301 put_dt(dynamic
, DT_NULL
, 0);
2304 /* Relocate remaining sections and symbols (that is those not related to
2306 static int final_sections_reloc(TCCState
*s1
)
2311 relocate_syms(s1
, 0);
2313 if (s1
->nb_errors
!= 0)
2316 /* relocate sections */
2317 /* XXX: ignore sections with allocated relocations ? */
2318 for(i
= 1; i
< s1
->nb_sections
; i
++) {
2319 s
= s1
->sections
[i
];
2320 #ifdef TCC_TARGET_I386
2321 if (s
->reloc
&& s
!= s1
->got
&& (s
->sh_flags
& SHF_ALLOC
)) //gr
2322 /* On X86 gdb 7.3 works in any case but gdb 6.6 will crash if SHF_ALLOC
2323 checking is removed */
2325 if (s
->reloc
&& s
!= s1
->got
)
2326 /* On X86_64 gdb 7.3 will crash if SHF_ALLOC checking is present */
2328 relocate_section(s1
, s
);
2331 /* relocate relocation entries if the relocation tables are
2332 allocated in the executable */
2333 for(i
= 1; i
< s1
->nb_sections
; i
++) {
2334 s
= s1
->sections
[i
];
2335 if ((s
->sh_flags
& SHF_ALLOC
) &&
2336 s
->sh_type
== SHT_RELX
) {
2337 relocate_rel(s1
, s
);
2343 /* Create an ELF file on disk.
2344 This function handle ELF specific layout requirements */
2345 static void tcc_output_elf(TCCState
*s1
, FILE *f
, int phnum
, ElfW(Phdr
) *phdr
,
2346 int file_offset
, int *sec_order
)
2348 int i
, shnum
, offset
, size
, file_type
;
2351 ElfW(Shdr
) shdr
, *sh
;
2353 file_type
= s1
->output_type
;
2354 shnum
= s1
->nb_sections
;
2356 memset(&ehdr
, 0, sizeof(ehdr
));
2359 ehdr
.e_phentsize
= sizeof(ElfW(Phdr
));
2360 ehdr
.e_phnum
= phnum
;
2361 ehdr
.e_phoff
= sizeof(ElfW(Ehdr
));
2365 file_offset
= (file_offset
+ 3) & -4;
2368 ehdr
.e_ident
[0] = ELFMAG0
;
2369 ehdr
.e_ident
[1] = ELFMAG1
;
2370 ehdr
.e_ident
[2] = ELFMAG2
;
2371 ehdr
.e_ident
[3] = ELFMAG3
;
2372 ehdr
.e_ident
[4] = ELFCLASSW
;
2373 ehdr
.e_ident
[5] = ELFDATA2LSB
;
2374 ehdr
.e_ident
[6] = EV_CURRENT
;
2375 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
2376 ehdr
.e_ident
[EI_OSABI
] = ELFOSABI_FREEBSD
;
2378 #ifdef TCC_TARGET_ARM
2380 ehdr
.e_ident
[EI_OSABI
] = 0;
2381 ehdr
.e_flags
= EF_ARM_EABI_VER4
;
2382 if (file_type
== TCC_OUTPUT_EXE
|| file_type
== TCC_OUTPUT_DLL
)
2383 ehdr
.e_flags
|= EF_ARM_HASENTRY
;
2384 if (s1
->float_abi
== ARM_HARD_FLOAT
)
2385 ehdr
.e_flags
|= EF_ARM_VFP_FLOAT
;
2387 ehdr
.e_flags
|= EF_ARM_SOFT_FLOAT
;
2389 ehdr
.e_ident
[EI_OSABI
] = ELFOSABI_ARM
;
2394 case TCC_OUTPUT_EXE
:
2395 ehdr
.e_type
= ET_EXEC
;
2396 ehdr
.e_entry
= get_elf_sym_addr(s1
, "_start", 1);
2398 case TCC_OUTPUT_DLL
:
2399 ehdr
.e_type
= ET_DYN
;
2400 ehdr
.e_entry
= text_section
->sh_addr
; /* XXX: is it correct ? */
2402 case TCC_OUTPUT_OBJ
:
2403 ehdr
.e_type
= ET_REL
;
2406 ehdr
.e_machine
= EM_TCC_TARGET
;
2407 ehdr
.e_version
= EV_CURRENT
;
2408 ehdr
.e_shoff
= file_offset
;
2409 ehdr
.e_ehsize
= sizeof(ElfW(Ehdr
));
2410 ehdr
.e_shentsize
= sizeof(ElfW(Shdr
));
2411 ehdr
.e_shnum
= shnum
;
2412 ehdr
.e_shstrndx
= shnum
- 1;
2414 fwrite(&ehdr
, 1, sizeof(ElfW(Ehdr
)), f
);
2415 fwrite(phdr
, 1, phnum
* sizeof(ElfW(Phdr
)), f
);
2416 offset
= sizeof(ElfW(Ehdr
)) + phnum
* sizeof(ElfW(Phdr
));
2418 sort_syms(s1
, symtab_section
);
2419 for(i
= 1; i
< s1
->nb_sections
; i
++) {
2420 s
= s1
->sections
[sec_order
[i
]];
2421 if (s
->sh_type
!= SHT_NOBITS
) {
2422 if (s
->sh_type
== SHT_DYNSYM
)
2423 patch_dynsym_undef(s1
, s
);
2424 while (offset
< s
->sh_offset
) {
2430 fwrite(s
->data
, 1, size
, f
);
2435 /* output section headers */
2436 while (offset
< ehdr
.e_shoff
) {
2441 for(i
= 0; i
< s1
->nb_sections
; i
++) {
2443 memset(sh
, 0, sizeof(ElfW(Shdr
)));
2444 s
= s1
->sections
[i
];
2446 sh
->sh_name
= s
->sh_name
;
2447 sh
->sh_type
= s
->sh_type
;
2448 sh
->sh_flags
= s
->sh_flags
;
2449 sh
->sh_entsize
= s
->sh_entsize
;
2450 sh
->sh_info
= s
->sh_info
;
2452 sh
->sh_link
= s
->link
->sh_num
;
2453 sh
->sh_addralign
= s
->sh_addralign
;
2454 sh
->sh_addr
= s
->sh_addr
;
2455 sh
->sh_offset
= s
->sh_offset
;
2456 sh
->sh_size
= s
->sh_size
;
2458 fwrite(sh
, 1, sizeof(ElfW(Shdr
)), f
);
2462 /* Write an elf, coff or "binary" file */
2463 static int tcc_write_elf_file(TCCState
*s1
, const char *filename
, int phnum
,
2464 ElfW(Phdr
) *phdr
, int file_offset
, int *sec_order
)
2466 int fd
, mode
, file_type
;
2469 file_type
= s1
->output_type
;
2470 if (file_type
== TCC_OUTPUT_OBJ
)
2475 fd
= open(filename
, O_WRONLY
| O_CREAT
| O_TRUNC
| O_BINARY
, mode
);
2477 tcc_error_noabort("could not write '%s'", filename
);
2480 f
= fdopen(fd
, "wb");
2482 printf("<- %s\n", filename
);
2484 #ifdef TCC_TARGET_COFF
2485 if (s1
->output_format
== TCC_OUTPUT_FORMAT_COFF
)
2486 tcc_output_coff(s1
, f
);
2489 if (s1
->output_format
== TCC_OUTPUT_FORMAT_ELF
)
2490 tcc_output_elf(s1
, f
, phnum
, phdr
, file_offset
, sec_order
);
2492 tcc_output_binary(s1
, f
, sec_order
);
2498 /* Output an elf, coff or binary file */
2499 /* XXX: suppress unneeded sections */
2500 static int elf_output_file(TCCState
*s1
, const char *filename
)
2502 int i
, ret
, phnum
, shnum
, file_type
, file_offset
, *sec_order
;
2503 struct dyn_inf dyninf
;
2506 Section
*strsec
, *interp
, *dynamic
, *dynstr
;
2508 file_type
= s1
->output_type
;
2511 /* if linking, also link in runtime libraries (libc, libgcc, etc.) */
2512 if (file_type
!= TCC_OUTPUT_OBJ
) {
2513 tcc_add_runtime(s1
);
2518 interp
= dynamic
= dynstr
= NULL
; /* avoid warning */
2519 dyninf
.dyn_rel_off
= 0; /* avoid warning */
2521 if (file_type
!= TCC_OUTPUT_OBJ
) {
2522 relocate_common_syms();
2524 tcc_add_linker_symbols(s1
);
2526 if (!s1
->static_link
) {
2527 if (file_type
== TCC_OUTPUT_EXE
) {
2529 /* allow override the dynamic loader */
2530 const char *elfint
= getenv("LD_SO");
2532 elfint
= DEFAULT_ELFINTERP(s1
);
2533 /* add interpreter section only if executable */
2534 interp
= new_section(s1
, ".interp", SHT_PROGBITS
, SHF_ALLOC
);
2535 interp
->sh_addralign
= 1;
2536 ptr
= section_ptr_add(interp
, 1 + strlen(elfint
));
2537 strcpy(ptr
, elfint
);
2540 /* add dynamic symbol table */
2541 s1
->dynsym
= new_symtab(s1
, ".dynsym", SHT_DYNSYM
, SHF_ALLOC
,
2543 ".hash", SHF_ALLOC
);
2544 dynstr
= s1
->dynsym
->link
;
2546 /* add dynamic section */
2547 dynamic
= new_section(s1
, ".dynamic", SHT_DYNAMIC
,
2548 SHF_ALLOC
| SHF_WRITE
);
2549 dynamic
->link
= dynstr
;
2550 dynamic
->sh_entsize
= sizeof(ElfW(Dyn
));
2554 if (file_type
== TCC_OUTPUT_EXE
) {
2555 bind_exe_dynsyms(s1
);
2557 if (s1
->nb_errors
) {
2562 bind_libs_dynsyms(s1
);
2563 } else /* shared library case: simply export all global symbols */
2564 export_global_syms(s1
);
2566 build_got_entries(s1
);
2568 /* add a list of needed dlls */
2569 for(i
= 0; i
< s1
->nb_loaded_dlls
; i
++) {
2570 DLLReference
*dllref
= s1
->loaded_dlls
[i
];
2571 if (dllref
->level
== 0)
2572 put_dt(dynamic
, DT_NEEDED
, put_elf_str(dynstr
, dllref
->name
));
2576 put_dt(dynamic
, DT_RPATH
, put_elf_str(dynstr
, s1
->rpath
));
2578 /* XXX: currently, since we do not handle PIC code, we
2579 must relocate the readonly segments */
2580 if (file_type
== TCC_OUTPUT_DLL
) {
2582 put_dt(dynamic
, DT_SONAME
, put_elf_str(dynstr
, s1
->soname
));
2583 put_dt(dynamic
, DT_TEXTREL
, 0);
2587 put_dt(dynamic
, DT_SYMBOLIC
, 0);
2589 /* add necessary space for other entries */
2590 dyninf
.dyn_rel_off
= dynamic
->data_offset
;
2591 dynamic
->data_offset
+= sizeof(ElfW(Dyn
)) * EXTRA_RELITEMS
;
2593 /* still need to build got entries in case of static link */
2594 build_got_entries(s1
);
2598 /* we add a section for symbols */
2599 strsec
= new_section(s1
, ".shstrtab", SHT_STRTAB
, 0);
2600 put_elf_str(strsec
, "");
2602 /* compute number of sections */
2603 shnum
= s1
->nb_sections
;
2605 /* this array is used to reorder sections in the output file */
2606 sec_order
= tcc_malloc(sizeof(int) * shnum
);
2609 /* compute number of program headers */
2612 case TCC_OUTPUT_OBJ
:
2615 case TCC_OUTPUT_EXE
:
2616 if (!s1
->static_link
)
2617 phnum
= 4 + HAVE_PHDR
;
2621 case TCC_OUTPUT_DLL
:
2626 /* Allocate strings for section names */
2627 alloc_sec_names(s1
, file_type
, strsec
);
2629 /* allocate program segment headers */
2630 phdr
= tcc_mallocz(phnum
* sizeof(ElfW(Phdr
)));
2632 /* compute section to program header mapping */
2633 file_offset
= layout_sections(s1
, phdr
, phnum
, interp
, strsec
, &dyninf
,
2636 /* Fill remaining program header and finalize relocation related to dynamic
2639 fill_unloadable_phdr(phdr
, phnum
, interp
, dynamic
);
2641 dyninf
.dynamic
= dynamic
;
2642 dyninf
.dynstr
= dynstr
;
2644 fill_dynamic(s1
, &dyninf
);
2646 /* put in GOT the dynamic section address and relocate PLT */
2647 write32le(s1
->got
->data
, dynamic
->sh_addr
);
2648 if (file_type
== TCC_OUTPUT_EXE
2649 #if defined(TCC_OUTPUT_DLL_WITH_PLT)
2650 || file_type
== TCC_OUTPUT_DLL
2655 /* relocate symbols in .dynsym now that final addresses are known */
2656 for_each_elem(s1
->dynsym
, 1, sym
, ElfW(Sym
)) {
2657 if (sym
->st_shndx
== SHN_UNDEF
) {
2658 /* relocate to PLT if symbol corresponds to a PLT entry,
2659 but not if it's a weak symbol */
2660 if (ELFW(ST_BIND
)(sym
->st_info
) == STB_WEAK
)
2662 else if (sym
->st_value
)
2663 sym
->st_value
+= s1
->plt
->sh_addr
;
2664 } else if (sym
->st_shndx
< SHN_LORESERVE
) {
2665 /* do symbol relocation */
2666 sym
->st_value
+= s1
->sections
[sym
->st_shndx
]->sh_addr
;
2672 /* if building executable or DLL, then relocate each section
2673 except the GOT which is already relocated */
2674 if (file_type
!= TCC_OUTPUT_OBJ
) {
2675 ret
= final_sections_reloc(s1
);
2680 /* Perform relocation to GOT or PLT entries */
2681 if (file_type
== TCC_OUTPUT_EXE
&& s1
->static_link
)
2684 /* Create the ELF file with name 'filename' */
2685 ret
= tcc_write_elf_file(s1
, filename
, phnum
, phdr
, file_offset
, sec_order
);
2688 const char *strip_cmd
= "sstrip "; // super strip utility from ELFkickers
2689 const char *null_dev
= " 2> /dev/null";
2691 snprintf(buf
, sizeof(buf
), "%s%s%s", strip_cmd
, filename
, null_dev
);
2694 system(buf
+1); // call a strip utility from binutils
2697 tcc_free(s1
->symtab_to_dynsym
);
2698 tcc_free(sec_order
);
2700 tcc_free(s1
->sym_attrs
);
2701 s1
->sym_attrs
= NULL
;
2705 LIBTCCAPI
int tcc_output_file(TCCState
*s
, const char *filename
)
2708 #ifdef TCC_TARGET_PE
2709 if (s
->output_type
!= TCC_OUTPUT_OBJ
) {
2710 ret
= pe_output_file(s
, filename
);
2713 ret
= elf_output_file(s
, filename
);
2717 static void *load_data(int fd
, unsigned long file_offset
, unsigned long size
)
2721 data
= tcc_malloc(size
);
2722 lseek(fd
, file_offset
, SEEK_SET
);
2723 read(fd
, data
, size
);
2727 typedef struct SectionMergeInfo
{
2728 Section
*s
; /* corresponding existing section */
2729 unsigned long offset
; /* offset of the new section in the existing section */
2730 uint8_t new_section
; /* true if section 's' was added */
2731 uint8_t link_once
; /* true if link once section */
2734 /* load an object file and merge it with current files */
2735 /* XXX: handle correctly stab (debug) info */
2736 ST_FUNC
int tcc_load_object_file(TCCState
*s1
,
2737 int fd
, unsigned long file_offset
)
2740 ElfW(Shdr
) *shdr
, *sh
;
2741 int size
, i
, j
, offset
, offseti
, nb_syms
, sym_index
, ret
;
2742 unsigned char *strsec
, *strtab
;
2743 int *old_to_new_syms
;
2744 char *sh_name
, *name
;
2745 SectionMergeInfo
*sm_table
, *sm
;
2746 ElfW(Sym
) *sym
, *symtab
;
2753 stab_index
= stabstr_index
= 0;
2755 if (read(fd
, &ehdr
, sizeof(ehdr
)) != sizeof(ehdr
))
2757 if (ehdr
.e_ident
[0] != ELFMAG0
||
2758 ehdr
.e_ident
[1] != ELFMAG1
||
2759 ehdr
.e_ident
[2] != ELFMAG2
||
2760 ehdr
.e_ident
[3] != ELFMAG3
)
2762 /* test if object file */
2763 if (ehdr
.e_type
!= ET_REL
)
2765 /* test CPU specific stuff */
2766 if (ehdr
.e_ident
[5] != ELFDATA2LSB
||
2767 ehdr
.e_machine
!= EM_TCC_TARGET
) {
2769 tcc_error_noabort("invalid object file");
2773 shdr
= load_data(fd
, file_offset
+ ehdr
.e_shoff
,
2774 sizeof(ElfW(Shdr
)) * ehdr
.e_shnum
);
2775 sm_table
= tcc_mallocz(sizeof(SectionMergeInfo
) * ehdr
.e_shnum
);
2777 /* load section names */
2778 sh
= &shdr
[ehdr
.e_shstrndx
];
2779 strsec
= load_data(fd
, file_offset
+ sh
->sh_offset
, sh
->sh_size
);
2781 /* load symtab and strtab */
2782 old_to_new_syms
= NULL
;
2786 for(i
= 1; i
< ehdr
.e_shnum
; i
++) {
2788 if (sh
->sh_type
== SHT_SYMTAB
) {
2790 tcc_error_noabort("object must contain only one symtab");
2795 nb_syms
= sh
->sh_size
/ sizeof(ElfW(Sym
));
2796 symtab
= load_data(fd
, file_offset
+ sh
->sh_offset
, sh
->sh_size
);
2797 sm_table
[i
].s
= symtab_section
;
2799 /* now load strtab */
2800 sh
= &shdr
[sh
->sh_link
];
2801 strtab
= load_data(fd
, file_offset
+ sh
->sh_offset
, sh
->sh_size
);
2805 /* now examine each section and try to merge its content with the
2807 for(i
= 1; i
< ehdr
.e_shnum
; i
++) {
2808 /* no need to examine section name strtab */
2809 if (i
== ehdr
.e_shstrndx
)
2812 sh_name
= (char *) strsec
+ sh
->sh_name
;
2813 /* ignore sections types we do not handle */
2814 if (sh
->sh_type
!= SHT_PROGBITS
&&
2815 sh
->sh_type
!= SHT_RELX
&&
2817 sh
->sh_type
!= SHT_ARM_EXIDX
&&
2819 sh
->sh_type
!= SHT_NOBITS
&&
2820 sh
->sh_type
!= SHT_PREINIT_ARRAY
&&
2821 sh
->sh_type
!= SHT_INIT_ARRAY
&&
2822 sh
->sh_type
!= SHT_FINI_ARRAY
&&
2823 strcmp(sh_name
, ".stabstr")
2826 if (sh
->sh_addralign
< 1)
2827 sh
->sh_addralign
= 1;
2828 /* find corresponding section, if any */
2829 for(j
= 1; j
< s1
->nb_sections
;j
++) {
2830 s
= s1
->sections
[j
];
2831 if (!strcmp(s
->name
, sh_name
)) {
2832 if (!strncmp(sh_name
, ".gnu.linkonce",
2833 sizeof(".gnu.linkonce") - 1)) {
2834 /* if a 'linkonce' section is already present, we
2835 do not add it again. It is a little tricky as
2836 symbols can still be defined in
2838 sm_table
[i
].link_once
= 1;
2845 /* not found: create new section */
2846 s
= new_section(s1
, sh_name
, sh
->sh_type
, sh
->sh_flags
);
2847 /* take as much info as possible from the section. sh_link and
2848 sh_info will be updated later */
2849 s
->sh_addralign
= sh
->sh_addralign
;
2850 s
->sh_entsize
= sh
->sh_entsize
;
2851 sm_table
[i
].new_section
= 1;
2853 if (sh
->sh_type
!= s
->sh_type
) {
2854 tcc_error_noabort("invalid section type");
2858 /* align start of section */
2859 offset
= s
->data_offset
;
2861 if (0 == strcmp(sh_name
, ".stab")) {
2865 if (0 == strcmp(sh_name
, ".stabstr")) {
2870 size
= sh
->sh_addralign
- 1;
2871 offset
= (offset
+ size
) & ~size
;
2872 if (sh
->sh_addralign
> s
->sh_addralign
)
2873 s
->sh_addralign
= sh
->sh_addralign
;
2874 s
->data_offset
= offset
;
2876 sm_table
[i
].offset
= offset
;
2878 /* concatenate sections */
2880 if (sh
->sh_type
!= SHT_NOBITS
) {
2882 lseek(fd
, file_offset
+ sh
->sh_offset
, SEEK_SET
);
2883 ptr
= section_ptr_add(s
, size
);
2884 read(fd
, ptr
, size
);
2886 s
->data_offset
+= size
;
2891 /* gr relocate stab strings */
2892 if (stab_index
&& stabstr_index
) {
2895 s
= sm_table
[stab_index
].s
;
2896 a
= (Stab_Sym
*)(s
->data
+ sm_table
[stab_index
].offset
);
2897 b
= (Stab_Sym
*)(s
->data
+ s
->data_offset
);
2898 o
= sm_table
[stabstr_index
].offset
;
2900 a
->n_strx
+= o
, a
++;
2903 /* second short pass to update sh_link and sh_info fields of new
2905 for(i
= 1; i
< ehdr
.e_shnum
; i
++) {
2907 if (!s
|| !sm_table
[i
].new_section
)
2910 if (sh
->sh_link
> 0)
2911 s
->link
= sm_table
[sh
->sh_link
].s
;
2912 if (sh
->sh_type
== SHT_RELX
) {
2913 s
->sh_info
= sm_table
[sh
->sh_info
].s
->sh_num
;
2914 /* update backward link */
2915 s1
->sections
[s
->sh_info
]->reloc
= s
;
2920 /* resolve symbols */
2921 old_to_new_syms
= tcc_mallocz(nb_syms
* sizeof(int));
2924 for(i
= 1; i
< nb_syms
; i
++, sym
++) {
2925 if (sym
->st_shndx
!= SHN_UNDEF
&&
2926 sym
->st_shndx
< SHN_LORESERVE
) {
2927 sm
= &sm_table
[sym
->st_shndx
];
2928 if (sm
->link_once
) {
2929 /* if a symbol is in a link once section, we use the
2930 already defined symbol. It is very important to get
2931 correct relocations */
2932 if (ELFW(ST_BIND
)(sym
->st_info
) != STB_LOCAL
) {
2933 name
= (char *) strtab
+ sym
->st_name
;
2934 sym_index
= find_elf_sym(symtab_section
, name
);
2936 old_to_new_syms
[i
] = sym_index
;
2940 /* if no corresponding section added, no need to add symbol */
2943 /* convert section number */
2944 sym
->st_shndx
= sm
->s
->sh_num
;
2946 sym
->st_value
+= sm
->offset
;
2949 name
= (char *) strtab
+ sym
->st_name
;
2950 sym_index
= add_elf_sym(symtab_section
, sym
->st_value
, sym
->st_size
,
2951 sym
->st_info
, sym
->st_other
,
2952 sym
->st_shndx
, name
);
2953 old_to_new_syms
[i
] = sym_index
;
2956 /* third pass to patch relocation entries */
2957 for(i
= 1; i
< ehdr
.e_shnum
; i
++) {
2962 offset
= sm_table
[i
].offset
;
2963 switch(s
->sh_type
) {
2965 /* take relocation offset information */
2966 offseti
= sm_table
[sh
->sh_info
].offset
;
2967 for_each_elem(s
, (offset
/ sizeof(*rel
)), rel
, ElfW_Rel
) {
2970 /* convert symbol index */
2971 type
= ELFW(R_TYPE
)(rel
->r_info
);
2972 sym_index
= ELFW(R_SYM
)(rel
->r_info
);
2973 /* NOTE: only one symtab assumed */
2974 if (sym_index
>= nb_syms
)
2976 sym_index
= old_to_new_syms
[sym_index
];
2977 /* ignore link_once in rel section. */
2978 if (!sym_index
&& !sm
->link_once
2979 #ifdef TCC_TARGET_ARM
2980 && type
!= R_ARM_V4BX
2984 tcc_error_noabort("Invalid relocation entry [%2d] '%s' @ %.8x",
2985 i
, strsec
+ sh
->sh_name
, rel
->r_offset
);
2988 rel
->r_info
= ELFW(R_INFO
)(sym_index
, type
);
2989 /* offset the relocation offset */
2990 rel
->r_offset
+= offseti
;
2991 #ifdef TCC_TARGET_ARM
2992 /* Jumps and branches from a Thumb code to a PLT entry need
2993 special handling since PLT entries are ARM code.
2994 Unconditional bl instructions referencing PLT entries are
2995 handled by converting these instructions into blx
2996 instructions. Other case of instructions referencing a PLT
2997 entry require to add a Thumb stub before the PLT entry to
2998 switch to ARM mode. We set bit plt_thumb_stub of the
2999 attribute of a symbol to indicate such a case. */
3000 if (type
== R_ARM_THM_JUMP24
)
3001 alloc_sym_attr(s1
, sym_index
)->plt_thumb_stub
= 1;
3014 tcc_free(old_to_new_syms
);
3021 typedef struct ArchiveHeader
{
3022 char ar_name
[16]; /* name of this member */
3023 char ar_date
[12]; /* file mtime */
3024 char ar_uid
[6]; /* owner uid; printed as decimal */
3025 char ar_gid
[6]; /* owner gid; printed as decimal */
3026 char ar_mode
[8]; /* file mode, printed as octal */
3027 char ar_size
[10]; /* file size, printed as decimal */
3028 char ar_fmag
[2]; /* should contain ARFMAG */
3031 static int get_be32(const uint8_t *b
)
3033 return b
[3] | (b
[2] << 8) | (b
[1] << 16) | (b
[0] << 24);
3036 /* load only the objects which resolve undefined symbols */
3037 static int tcc_load_alacarte(TCCState
*s1
, int fd
, int size
)
3039 int i
, bound
, nsyms
, sym_index
, off
, ret
;
3041 const char *ar_names
, *p
;
3042 const uint8_t *ar_index
;
3045 data
= tcc_malloc(size
);
3046 if (read(fd
, data
, size
) != size
)
3048 nsyms
= get_be32(data
);
3049 ar_index
= data
+ 4;
3050 ar_names
= (char *) ar_index
+ nsyms
* 4;
3054 for(p
= ar_names
, i
= 0; i
< nsyms
; i
++, p
+= strlen(p
)+1) {
3055 sym_index
= find_elf_sym(symtab_section
, p
);
3057 sym
= &((ElfW(Sym
) *)symtab_section
->data
)[sym_index
];
3058 if(sym
->st_shndx
== SHN_UNDEF
) {
3059 off
= get_be32(ar_index
+ i
* 4) + sizeof(ArchiveHeader
);
3061 lseek(fd
, off
, SEEK_SET
);
3062 if(tcc_load_object_file(s1
, fd
, off
) < 0) {
3077 /* load a '.a' file */
3078 ST_FUNC
int tcc_load_archive(TCCState
*s1
, int fd
)
3085 unsigned long file_offset
;
3087 /* skip magic which was already checked */
3088 read(fd
, magic
, sizeof(magic
));
3091 len
= read(fd
, &hdr
, sizeof(hdr
));
3094 if (len
!= sizeof(hdr
)) {
3095 tcc_error_noabort("invalid archive");
3098 memcpy(ar_size
, hdr
.ar_size
, sizeof(hdr
.ar_size
));
3099 ar_size
[sizeof(hdr
.ar_size
)] = '\0';
3100 size
= strtol(ar_size
, NULL
, 0);
3101 memcpy(ar_name
, hdr
.ar_name
, sizeof(hdr
.ar_name
));
3102 for(i
= sizeof(hdr
.ar_name
) - 1; i
>= 0; i
--) {
3103 if (ar_name
[i
] != ' ')
3106 ar_name
[i
+ 1] = '\0';
3107 file_offset
= lseek(fd
, 0, SEEK_CUR
);
3109 size
= (size
+ 1) & ~1;
3110 if (!strcmp(ar_name
, "/")) {
3111 /* coff symbol table : we handle it */
3112 if(s1
->alacarte_link
)
3113 return tcc_load_alacarte(s1
, fd
, size
);
3114 } else if (!strcmp(ar_name
, "//") ||
3115 !strcmp(ar_name
, "__.SYMDEF") ||
3116 !strcmp(ar_name
, "__.SYMDEF/") ||
3117 !strcmp(ar_name
, "ARFILENAMES/")) {
3118 /* skip symbol table or archive names */
3120 if (tcc_load_object_file(s1
, fd
, file_offset
) < 0)
3123 lseek(fd
, file_offset
+ size
, SEEK_SET
);
3128 #ifndef TCC_TARGET_PE
3129 /* load a DLL and all referenced DLLs. 'level = 0' means that the DLL
3130 is referenced by the user (so it should be added as DT_NEEDED in
3131 the generated ELF file) */
3132 ST_FUNC
int tcc_load_dll(TCCState
*s1
, int fd
, const char *filename
, int level
)
3135 ElfW(Shdr
) *shdr
, *sh
, *sh1
;
3136 int i
, j
, nb_syms
, nb_dts
, sym_bind
, ret
;
3137 ElfW(Sym
) *sym
, *dynsym
;
3138 ElfW(Dyn
) *dt
, *dynamic
;
3139 unsigned char *dynstr
;
3140 const char *name
, *soname
;
3141 DLLReference
*dllref
;
3143 read(fd
, &ehdr
, sizeof(ehdr
));
3145 /* test CPU specific stuff */
3146 if (ehdr
.e_ident
[5] != ELFDATA2LSB
||
3147 ehdr
.e_machine
!= EM_TCC_TARGET
) {
3148 tcc_error_noabort("bad architecture");
3153 shdr
= load_data(fd
, ehdr
.e_shoff
, sizeof(ElfW(Shdr
)) * ehdr
.e_shnum
);
3155 /* load dynamic section and dynamic symbols */
3159 dynsym
= NULL
; /* avoid warning */
3160 dynstr
= NULL
; /* avoid warning */
3161 for(i
= 0, sh
= shdr
; i
< ehdr
.e_shnum
; i
++, sh
++) {
3162 switch(sh
->sh_type
) {
3164 nb_dts
= sh
->sh_size
/ sizeof(ElfW(Dyn
));
3165 dynamic
= load_data(fd
, sh
->sh_offset
, sh
->sh_size
);
3168 nb_syms
= sh
->sh_size
/ sizeof(ElfW(Sym
));
3169 dynsym
= load_data(fd
, sh
->sh_offset
, sh
->sh_size
);
3170 sh1
= &shdr
[sh
->sh_link
];
3171 dynstr
= load_data(fd
, sh1
->sh_offset
, sh1
->sh_size
);
3178 /* compute the real library name */
3179 soname
= tcc_basename(filename
);
3181 for(i
= 0, dt
= dynamic
; i
< nb_dts
; i
++, dt
++) {
3182 if (dt
->d_tag
== DT_SONAME
) {
3183 soname
= (char *) dynstr
+ dt
->d_un
.d_val
;
3187 /* if the dll is already loaded, do not load it */
3188 for(i
= 0; i
< s1
->nb_loaded_dlls
; i
++) {
3189 dllref
= s1
->loaded_dlls
[i
];
3190 if (!strcmp(soname
, dllref
->name
)) {
3191 /* but update level if needed */
3192 if (level
< dllref
->level
)
3193 dllref
->level
= level
;
3199 /* add the dll and its level */
3200 dllref
= tcc_mallocz(sizeof(DLLReference
) + strlen(soname
));
3201 dllref
->level
= level
;
3202 strcpy(dllref
->name
, soname
);
3203 dynarray_add((void ***)&s1
->loaded_dlls
, &s1
->nb_loaded_dlls
, dllref
);
3205 /* add dynamic symbols in dynsym_section */
3206 for(i
= 1, sym
= dynsym
+ 1; i
< nb_syms
; i
++, sym
++) {
3207 sym_bind
= ELFW(ST_BIND
)(sym
->st_info
);
3208 if (sym_bind
== STB_LOCAL
)
3210 name
= (char *) dynstr
+ sym
->st_name
;
3211 add_elf_sym(s1
->dynsymtab_section
, sym
->st_value
, sym
->st_size
,
3212 sym
->st_info
, sym
->st_other
, sym
->st_shndx
, name
);
3215 /* load all referenced DLLs */
3216 for(i
= 0, dt
= dynamic
; i
< nb_dts
; i
++, dt
++) {
3219 name
= (char *) dynstr
+ dt
->d_un
.d_val
;
3220 for(j
= 0; j
< s1
->nb_loaded_dlls
; j
++) {
3221 dllref
= s1
->loaded_dlls
[j
];
3222 if (!strcmp(name
, dllref
->name
))
3223 goto already_loaded
;
3225 if (tcc_add_dll(s1
, name
, AFF_REFERENCED_DLL
) < 0) {
3226 tcc_error_noabort("referenced dll '%s' not found", name
);
3243 #define LD_TOK_NAME 256
3244 #define LD_TOK_EOF (-1)
3246 /* return next ld script token */
3247 static int ld_next(TCCState
*s1
, char *name
, int name_size
)
3265 file
->buf_ptr
= parse_comment(file
->buf_ptr
);
3266 ch
= file
->buf_ptr
[0];
3279 /* case 'a' ... 'z': */
3306 /* case 'A' ... 'z': */
3340 if (!((ch
>= 'a' && ch
<= 'z') ||
3341 (ch
>= 'A' && ch
<= 'Z') ||
3342 (ch
>= '0' && ch
<= '9') ||
3343 strchr("/.-_+=$:\\,~", ch
)))
3345 if ((q
- name
) < name_size
- 1) {
3364 static int ld_add_file(TCCState
*s1
, const char filename
[])
3368 ret
= tcc_add_file_internal(s1
, filename
, 0, TCC_FILETYPE_BINARY
);
3370 ret
= tcc_add_dll(s1
, filename
, 0);
3374 static inline int new_undef_syms(void)
3377 ret
= new_undef_sym
;
3382 static int ld_add_file_list(TCCState
*s1
, const char *cmd
, int as_needed
)
3384 char filename
[1024], libname
[1024];
3385 int t
, group
, nblibs
= 0, ret
= 0;
3388 group
= !strcmp(cmd
, "GROUP");
3391 t
= ld_next(s1
, filename
, sizeof(filename
));
3394 t
= ld_next(s1
, filename
, sizeof(filename
));
3397 if (t
== LD_TOK_EOF
) {
3398 tcc_error_noabort("unexpected end of file");
3400 goto lib_parse_error
;
3401 } else if (t
== ')') {
3403 } else if (t
== '-') {
3404 t
= ld_next(s1
, filename
, sizeof(filename
));
3405 if ((t
!= LD_TOK_NAME
) || (filename
[0] != 'l')) {
3406 tcc_error_noabort("library name expected");
3408 goto lib_parse_error
;
3410 pstrcpy(libname
, sizeof libname
, &filename
[1]);
3411 if (s1
->static_link
) {
3412 snprintf(filename
, sizeof filename
, "lib%s.a", libname
);
3414 snprintf(filename
, sizeof filename
, "lib%s.so", libname
);
3416 } else if (t
!= LD_TOK_NAME
) {
3417 tcc_error_noabort("filename expected");
3419 goto lib_parse_error
;
3421 if (!strcmp(filename
, "AS_NEEDED")) {
3422 ret
= ld_add_file_list(s1
, cmd
, 1);
3424 goto lib_parse_error
;
3426 /* TODO: Implement AS_NEEDED support. Ignore it for now */
3428 ret
= ld_add_file(s1
, filename
);
3430 goto lib_parse_error
;
3432 /* Add the filename *and* the libname to avoid future conversions */
3433 dynarray_add((void ***) &libs
, &nblibs
, tcc_strdup(filename
));
3434 if (libname
[0] != '\0')
3435 dynarray_add((void ***) &libs
, &nblibs
, tcc_strdup(libname
));
3439 t
= ld_next(s1
, filename
, sizeof(filename
));
3441 t
= ld_next(s1
, filename
, sizeof(filename
));
3444 if (group
&& !as_needed
) {
3445 while (new_undef_syms()) {
3448 for (i
= 0; i
< nblibs
; i
++)
3449 ld_add_file(s1
, libs
[i
]);
3453 dynarray_reset(&libs
, &nblibs
);
3457 /* interpret a subset of GNU ldscripts to handle the dummy libc.so
3459 ST_FUNC
int tcc_load_ldscript(TCCState
*s1
)
3462 char filename
[1024];
3467 t
= ld_next(s1
, cmd
, sizeof(cmd
));
3468 if (t
== LD_TOK_EOF
)
3470 else if (t
!= LD_TOK_NAME
)
3472 if (!strcmp(cmd
, "INPUT") ||
3473 !strcmp(cmd
, "GROUP")) {
3474 ret
= ld_add_file_list(s1
, cmd
, 0);
3477 } else if (!strcmp(cmd
, "OUTPUT_FORMAT") ||
3478 !strcmp(cmd
, "TARGET")) {
3479 /* ignore some commands */
3480 t
= ld_next(s1
, cmd
, sizeof(cmd
));
3484 t
= ld_next(s1
, filename
, sizeof(filename
));
3485 if (t
== LD_TOK_EOF
) {
3486 tcc_error_noabort("unexpected end of file");
3488 } else if (t
== ')') {
3498 #endif /* !TCC_TARGET_PE */