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 /* Use sign extension! */
902 qrel
->r_addend
= (int)read32le(ptr
) + val
;
905 write32le(ptr
, read32le(ptr
) + val
);
909 if (s1
->output_type
== TCC_OUTPUT_DLL
) {
911 esym_index
= s1
->symtab_to_dynsym
[sym_index
];
913 qrel
->r_offset
= rel
->r_offset
;
914 qrel
->r_info
= ELFW(R_INFO
)(esym_index
, R_X86_64_PC32
);
915 /* Use sign extension! */
916 qrel
->r_addend
= (int)read32le(ptr
);
924 /* We've put the PLT slot offset into r_addend when generating
925 it, and that's what we must use as relocation value (adjusted
926 by section offset of course). */
927 val
= s1
->plt
->sh_addr
+ rel
->r_addend
;
933 diff
= (long long)val
- addr
;
934 if (diff
< -2147483648LL || diff
> 2147483647LL) {
935 tcc_error("internal error: relocation failed");
937 write32le(ptr
, read32le(ptr
) + diff
);
940 case R_X86_64_GLOB_DAT
:
941 case R_X86_64_JUMP_SLOT
:
942 /* They don't need addend */
943 write64le(ptr
, val
- rel
->r_addend
);
945 case R_X86_64_GOTPCREL
:
946 case R_X86_64_GOTPCRELX
:
947 case R_X86_64_REX_GOTPCRELX
:
948 write32le(ptr
, read32le(ptr
) +
949 (s1
->got
->sh_addr
- addr
+
950 s1
->sym_attrs
[sym_index
].got_offset
- 4));
952 case R_X86_64_GOTTPOFF
:
953 write32le(ptr
, read32le(ptr
) + val
- s1
->got
->sh_addr
);
956 /* we load the got offset */
957 write32le(ptr
, read32le(ptr
) + s1
->sym_attrs
[sym_index
].got_offset
);
960 #error unsupported processor
964 /* if the relocation is allocated, we change its symbol table */
965 if (sr
->sh_flags
& SHF_ALLOC
)
966 sr
->link
= s1
->dynsym
;
969 /* relocate relocation table in 'sr' */
970 static void relocate_rel(TCCState
*s1
, Section
*sr
)
975 s
= s1
->sections
[sr
->sh_info
];
976 for_each_elem(sr
, 0, rel
, ElfW_Rel
)
977 rel
->r_offset
+= s
->sh_addr
;
980 /* count the number of dynamic relocations so that we can reserve
982 static int prepare_dynamic_rel(TCCState
*s1
, Section
*sr
)
985 int sym_index
, esym_index
, type
, count
;
988 for_each_elem(sr
, 0, rel
, ElfW_Rel
) {
989 sym_index
= ELFW(R_SYM
)(rel
->r_info
);
990 type
= ELFW(R_TYPE
)(rel
->r_info
);
992 #if defined(TCC_TARGET_I386)
994 #elif defined(TCC_TARGET_X86_64)
1001 #if defined(TCC_TARGET_I386)
1003 #elif defined(TCC_TARGET_X86_64)
1006 esym_index
= s1
->symtab_to_dynsym
[sym_index
];
1015 /* allocate the section */
1016 sr
->sh_flags
|= SHF_ALLOC
;
1017 sr
->sh_size
= count
* sizeof(ElfW_Rel
);
1022 static struct sym_attr
*alloc_sym_attr(TCCState
*s1
, int index
)
1025 struct sym_attr
*tab
;
1027 if (index
>= s1
->nb_sym_attrs
) {
1028 /* find immediately bigger power of 2 and reallocate array */
1032 tab
= tcc_realloc(s1
->sym_attrs
, n
* sizeof(*s1
->sym_attrs
));
1033 s1
->sym_attrs
= tab
;
1034 memset(s1
->sym_attrs
+ s1
->nb_sym_attrs
, 0,
1035 (n
- s1
->nb_sym_attrs
) * sizeof(*s1
->sym_attrs
));
1036 s1
->nb_sym_attrs
= n
;
1038 return &s1
->sym_attrs
[index
];
1041 static void build_got(TCCState
*s1
)
1045 /* if no got, then create it */
1046 s1
->got
= new_section(s1
, ".got", SHT_PROGBITS
, SHF_ALLOC
| SHF_WRITE
);
1047 s1
->got
->sh_entsize
= 4;
1048 add_elf_sym(symtab_section
, 0, 4, ELFW(ST_INFO
)(STB_GLOBAL
, STT_OBJECT
),
1049 0, s1
->got
->sh_num
, "_GLOBAL_OFFSET_TABLE_");
1050 ptr
= section_ptr_add(s1
->got
, 3 * PTR_SIZE
);
1052 /* keep space for _DYNAMIC pointer, if present */
1054 /* two dummy got entries */
1055 write32le(ptr
+ 4, 0);
1056 write32le(ptr
+ 8, 0);
1058 /* keep space for _DYNAMIC pointer, if present */
1060 write32le(ptr
+ 4, 0);
1061 /* two dummy got entries */
1062 write32le(ptr
+ 8, 0);
1063 write32le(ptr
+ 12, 0);
1064 write32le(ptr
+ 16, 0);
1065 write32le(ptr
+ 20, 0);
1069 /* put a got or plt entry corresponding to a symbol in symtab_section. 'size'
1070 and 'info' can be modifed if more precise info comes from the DLL.
1071 Returns offset of GOT or PLT slot. */
1072 static unsigned long put_got_entry(TCCState
*s1
,
1073 int reloc_type
, unsigned long size
, int info
,
1076 int index
, need_plt_entry
;
1079 unsigned long offset
;
1081 struct sym_attr
*symattr
;
1087 #ifdef TCC_TARGET_X86_64
1088 (reloc_type
== R_X86_64_JUMP_SLOT
);
1089 #elif defined(TCC_TARGET_I386)
1090 (reloc_type
== R_386_JMP_SLOT
);
1091 #elif defined(TCC_TARGET_ARM)
1092 (reloc_type
== R_ARM_JUMP_SLOT
);
1093 #elif defined(TCC_TARGET_ARM64)
1094 (reloc_type
== R_AARCH64_JUMP_SLOT
);
1099 if (need_plt_entry
&& !s1
->plt
) {
1101 s1
->plt
= new_section(s1
, ".plt", SHT_PROGBITS
,
1102 SHF_ALLOC
| SHF_EXECINSTR
);
1103 s1
->plt
->sh_entsize
= 4;
1106 /* If a got/plt entry already exists for that symbol, no need to add one */
1107 if (sym_index
< s1
->nb_sym_attrs
) {
1108 if (need_plt_entry
&& s1
->sym_attrs
[sym_index
].plt_offset
)
1109 return s1
->sym_attrs
[sym_index
].plt_offset
;
1110 else if (!need_plt_entry
&& s1
->sym_attrs
[sym_index
].got_offset
)
1111 return s1
->sym_attrs
[sym_index
].got_offset
;
1114 symattr
= alloc_sym_attr(s1
, sym_index
);
1116 /* Only store the GOT offset if it's not generated for the PLT entry. */
1117 if (!need_plt_entry
)
1118 symattr
->got_offset
= s1
->got
->data_offset
;
1120 sym
= &((ElfW(Sym
) *)symtab_section
->data
)[sym_index
];
1121 name
= (char *) symtab_section
->link
->data
+ sym
->st_name
;
1122 offset
= sym
->st_value
;
1123 #if defined(TCC_TARGET_I386) || defined(TCC_TARGET_X86_64)
1124 if (need_plt_entry
) {
1128 unsigned long relofs
;
1130 #if defined(TCC_OUTPUT_DLL_WITH_PLT)
1133 /* if we build a DLL, we add a %ebx offset */
1134 if (s1
->output_type
== TCC_OUTPUT_DLL
)
1140 /* add a PLT entry */
1142 if (plt
->data_offset
== 0) {
1143 /* first plt entry */
1144 p
= section_ptr_add(plt
, 16);
1145 p
[0] = 0xff; /* pushl got + PTR_SIZE */
1146 p
[1] = modrm
+ 0x10;
1147 write32le(p
+ 2, PTR_SIZE
);
1148 p
[6] = 0xff; /* jmp *(got + PTR_SIZE * 2) */
1150 write32le(p
+ 8, PTR_SIZE
* 2);
1153 /* The PLT slot refers to the relocation entry it needs
1154 via offset. The reloc entry is created below, so its
1155 offset is the current data_offset. */
1156 relofs
= s1
->got
->reloc
? s1
->got
->reloc
->data_offset
: 0;
1157 symattr
->plt_offset
= plt
->data_offset
;
1158 p
= section_ptr_add(plt
, 16);
1159 p
[0] = 0xff; /* jmp *(got + x) */
1161 write32le(p
+ 2, s1
->got
->data_offset
);
1162 p
[6] = 0x68; /* push $xxx */
1163 #ifdef TCC_TARGET_X86_64
1164 /* On x86-64, the relocation is referred to by _index_. */
1165 write32le(p
+ 7, relofs
/ sizeof (ElfW_Rel
));
1167 write32le(p
+ 7, relofs
);
1169 p
[11] = 0xe9; /* jmp plt_start */
1170 write32le(p
+ 12, -(plt
->data_offset
));
1172 /* If this was an UNDEF symbol set the offset in the
1173 dynsymtab to the PLT slot, so that PC32 relocs to it
1175 if (sym
->st_shndx
== SHN_UNDEF
)
1176 offset
= plt
->data_offset
- 16;
1178 #elif defined(TCC_TARGET_ARM)
1179 if (need_plt_entry
) {
1183 /* if we build a DLL, we add a %ebx offset */
1184 if (s1
->output_type
== TCC_OUTPUT_DLL
)
1185 tcc_error("DLLs unimplemented!");
1187 /* add a PLT entry */
1189 if (plt
->data_offset
== 0) {
1190 /* first plt entry */
1191 p
= section_ptr_add(plt
, 16);
1192 write32le(p
, 0xe52de004); /* push {lr} */
1193 write32le(p
+4, 0xe59fe010); /* ldr lr, [pc, #16] */
1194 write32le(p
+8, 0xe08fe00e); /* add lr, pc, lr */
1195 write32le(p
+12, 0xe5bef008); /* ldr pc, [lr, #8]! */
1198 symattr
->plt_offset
= plt
->data_offset
;
1199 if (symattr
->plt_thumb_stub
) {
1200 p
= section_ptr_add(plt
, 20);
1201 write32le(p
, 0x4778); /* bx pc */
1202 write32le(p
+2, 0x46c0); /* nop */
1205 p
= section_ptr_add(plt
, 16);
1206 write32le(p
, 0xe59fc004); /* ldr ip, [pc, #4] ; GOT entry offset */
1207 write32le(p
+4, 0xe08fc00c); /* add ip, pc, ip ; addr of GOT entry */
1208 write32le(p
+8, 0xe59cf000); /* ldr pc, [ip] ; jump to GOT entry */
1209 write32le(p
+12, s1
->got
->data_offset
); /* GOT entry off once patched */
1211 /* the symbol is modified so that it will be relocated to
1213 if (sym
->st_shndx
== SHN_UNDEF
)
1214 offset
= plt
->data_offset
- 16;
1216 #elif defined(TCC_TARGET_ARM64)
1217 if (need_plt_entry
) {
1221 if (s1
->output_type
== TCC_OUTPUT_DLL
)
1222 tcc_error("DLLs unimplemented!");
1225 if (plt
->data_offset
== 0)
1226 section_ptr_add(plt
, 32);
1227 symattr
->plt_offset
= plt
->data_offset
;
1228 p
= section_ptr_add(plt
, 16);
1229 write32le(p
, s1
->got
->data_offset
);
1230 write32le(p
+ 4, (uint64_t)s1
->got
->data_offset
>> 32);
1232 if (sym
->st_shndx
== SHN_UNDEF
)
1233 offset
= plt
->data_offset
- 16;
1235 #elif defined(TCC_TARGET_C67)
1237 tcc_error("C67 got not implemented");
1240 #error unsupported CPU
1243 /* XXX This might generate multiple syms for name. */
1244 index
= put_elf_sym(s1
->dynsym
, offset
,
1245 size
, info
, 0, sym
->st_shndx
, name
);
1246 /* Create the relocation (it's against the GOT for PLT
1248 put_elf_reloc(s1
->dynsym
, s1
->got
,
1249 s1
->got
->data_offset
,
1252 /* Without .dynsym (i.e. static link or memory output) we
1253 still need relocs against the generated got, so as to fill
1254 the entries with the symbol values (determined later). */
1255 put_elf_reloc(symtab_section
, s1
->got
,
1256 s1
->got
->data_offset
,
1257 reloc_type
, sym_index
);
1259 /* And now create the GOT slot itself. */
1260 ptr
= section_ptr_add(s1
->got
, PTR_SIZE
);
1263 return symattr
->plt_offset
;
1265 return symattr
->got_offset
;
1268 /* build GOT and PLT entries */
1269 ST_FUNC
void build_got_entries(TCCState
*s1
)
1274 int i
, type
, reloc_type
, sym_index
;
1276 for(i
= 1; i
< s1
->nb_sections
; i
++) {
1277 s
= s1
->sections
[i
];
1278 if (s
->sh_type
!= SHT_RELX
)
1280 /* no need to handle got relocations */
1281 if (s
->link
!= symtab_section
)
1283 for_each_elem(s
, 0, rel
, ElfW_Rel
) {
1284 type
= ELFW(R_TYPE
)(rel
->r_info
);
1286 #if defined(TCC_TARGET_I386)
1293 if (type
== R_386_GOT32
|| type
== R_386_PLT32
) {
1294 sym_index
= ELFW(R_SYM
)(rel
->r_info
);
1295 sym
= &((ElfW(Sym
) *)symtab_section
->data
)[sym_index
];
1296 /* look at the symbol got offset. If none, then add one */
1297 if (type
== R_386_GOT32
)
1298 reloc_type
= R_386_GLOB_DAT
;
1300 reloc_type
= R_386_JMP_SLOT
;
1301 put_got_entry(s1
, reloc_type
, sym
->st_size
, sym
->st_info
,
1305 #elif defined(TCC_TARGET_ARM)
1315 sym_index
= ELFW(R_SYM
)(rel
->r_info
);
1316 sym
= &((ElfW(Sym
) *)symtab_section
->data
)[sym_index
];
1317 if (type
!= R_ARM_GOTOFF
&& type
!= R_ARM_GOTPC
1318 && sym
->st_shndx
== SHN_UNDEF
) {
1320 /* look at the symbol got offset. If none, then add one */
1321 if (type
== R_ARM_GOT32
)
1322 reloc_type
= R_ARM_GLOB_DAT
;
1324 reloc_type
= R_ARM_JUMP_SLOT
;
1325 ofs
= put_got_entry(s1
, reloc_type
, sym
->st_size
,
1326 sym
->st_info
, sym_index
);
1328 printf ("maybegot: %s, %d, %d --> ofs=0x%x\n",
1329 (char *) symtab_section
->link
->data
+ sym
->st_name
,
1330 type
, sym
->st_shndx
, ofs
);
1332 if (type
!= R_ARM_GOT32
) {
1333 addr_t
*ptr
= (addr_t
*)(s1
->sections
[s
->sh_info
]->data
1335 /* x must be signed! */
1336 int x
= *ptr
& 0xffffff;
1342 printf ("insn=0x%x --> 0x%x (x==0x%x)\n", *ptr
,
1343 (*ptr
& 0xff000000) | x
, x
);
1345 *ptr
= (*ptr
& 0xff000000) | x
;
1349 case R_ARM_THM_JUMP24
:
1350 sym_index
= ELFW(R_SYM
)(rel
->r_info
);
1351 sym
= &((ElfW(Sym
) *)symtab_section
->data
)[sym_index
];
1352 /* We are relocating a jump from thumb code to arm code */
1353 if (sym
->st_shndx
!= SHN_UNDEF
&& !(sym
->st_value
& 1)) {
1356 char *name
, buf
[1024];
1357 Section
*text_section
;
1359 name
= (char *) symtab_section
->link
->data
+ sym
->st_name
;
1360 text_section
= s1
->sections
[sym
->st_shndx
];
1361 /* Modify reloc to target a thumb stub to switch to ARM */
1362 snprintf(buf
, sizeof(buf
), "%s_from_thumb", name
);
1363 index
= put_elf_sym(symtab_section
,
1364 text_section
->data_offset
+ 1,
1365 sym
->st_size
, sym
->st_info
, 0,
1366 sym
->st_shndx
, buf
);
1367 rel
->r_info
= ELFW(R_INFO
)(index
, type
);
1368 /* Create a thumb stub fonction to switch to ARM mode */
1369 put_elf_reloc(symtab_section
, text_section
,
1370 text_section
->data_offset
+ 4, R_ARM_JUMP24
,
1372 p
= section_ptr_add(text_section
, 8);
1373 write32le(p
, 0x4778); /* bx pc */
1374 write32le(p
+2, 0x46c0); /* nop */
1375 write32le(p
+4, 0xeafffffe); /* b $sym */
1377 #elif defined(TCC_TARGET_ARM64)
1378 //xx Other cases may be required here:
1379 case R_AARCH64_ADR_GOT_PAGE
:
1380 case R_AARCH64_LD64_GOT_LO12_NC
:
1383 sym_index
= ELFW(R_SYM
)(rel
->r_info
);
1384 sym
= &((ElfW(Sym
) *)symtab_section
->data
)[sym_index
];
1385 reloc_type
= R_AARCH64_GLOB_DAT
;
1386 put_got_entry(s1
, reloc_type
, sym
->st_size
, sym
->st_info
,
1390 case R_AARCH64_JUMP26
:
1391 case R_AARCH64_CALL26
:
1394 sym_index
= ELFW(R_SYM
)(rel
->r_info
);
1395 sym
= &((ElfW(Sym
) *)symtab_section
->data
)[sym_index
];
1396 if (sym
->st_shndx
== SHN_UNDEF
) {
1398 reloc_type
= R_AARCH64_JUMP_SLOT
;
1399 ofs
= put_got_entry(s1
, reloc_type
, sym
->st_size
,
1400 sym
->st_info
, sym_index
);
1401 /* We store the place of the generated PLT slot
1403 rel
->r_addend
+= ofs
;
1406 #elif defined(TCC_TARGET_C67)
1413 if (type
== R_C60_GOT32
|| type
== R_C60_PLT32
) {
1414 sym_index
= ELFW(R_SYM
)(rel
->r_info
);
1415 sym
= &((ElfW(Sym
) *)symtab_section
->data
)[sym_index
];
1416 /* look at the symbol got offset. If none, then add one */
1417 if (type
== R_C60_GOT32
)
1418 reloc_type
= R_C60_GLOB_DAT
;
1420 reloc_type
= R_C60_JMP_SLOT
;
1421 put_got_entry(s1
, reloc_type
, sym
->st_size
, sym
->st_info
,
1425 #elif defined(TCC_TARGET_X86_64)
1426 case R_X86_64_GOT32
:
1427 case R_X86_64_GOTTPOFF
:
1428 case R_X86_64_GOTPCREL
:
1429 case R_X86_64_GOTPCRELX
:
1430 case R_X86_64_REX_GOTPCRELX
:
1431 case R_X86_64_PLT32
:
1432 sym_index
= ELFW(R_SYM
)(rel
->r_info
);
1433 sym
= &((ElfW(Sym
) *)symtab_section
->data
)[sym_index
];
1434 if (type
== R_X86_64_PLT32
&&
1435 ELFW(ST_VISIBILITY
)(sym
->st_other
) != STV_DEFAULT
)
1437 rel
->r_info
= ELFW(R_INFO
)(sym_index
, R_X86_64_PC32
);
1443 sym
= &((ElfW(Sym
) *)symtab_section
->data
)[sym_index
];
1445 if (type
== R_X86_64_GOT32
|| type
== R_X86_64_GOTPCREL
||
1446 type
== R_X86_64_GOTPCRELX
||
1447 type
== R_X86_64_REX_GOTPCRELX
||
1448 type
== R_X86_64_PLT32
) {
1450 /* look at the symbol got offset. If none, then add one */
1451 if (type
== R_X86_64_PLT32
)
1452 reloc_type
= R_X86_64_JUMP_SLOT
;
1454 reloc_type
= R_X86_64_GLOB_DAT
;
1455 ofs
= put_got_entry(s1
, reloc_type
, sym
->st_size
,
1456 sym
->st_info
, sym_index
);
1457 if (type
== R_X86_64_PLT32
)
1458 /* We store the place of the generated PLT slot
1460 rel
->r_addend
+= ofs
;
1464 #error unsupported CPU
1473 ST_FUNC Section
*new_symtab(TCCState
*s1
,
1474 const char *symtab_name
, int sh_type
, int sh_flags
,
1475 const char *strtab_name
,
1476 const char *hash_name
, int hash_sh_flags
)
1478 Section
*symtab
, *strtab
, *hash
;
1479 int *ptr
, nb_buckets
;
1481 symtab
= new_section(s1
, symtab_name
, sh_type
, sh_flags
);
1482 symtab
->sh_entsize
= sizeof(ElfW(Sym
));
1483 strtab
= new_section(s1
, strtab_name
, SHT_STRTAB
, sh_flags
);
1484 put_elf_str(strtab
, "");
1485 symtab
->link
= strtab
;
1486 put_elf_sym(symtab
, 0, 0, 0, 0, 0, NULL
);
1490 hash
= new_section(s1
, hash_name
, SHT_HASH
, hash_sh_flags
);
1491 hash
->sh_entsize
= sizeof(int);
1492 symtab
->hash
= hash
;
1493 hash
->link
= symtab
;
1495 ptr
= section_ptr_add(hash
, (2 + nb_buckets
+ 1) * sizeof(int));
1496 ptr
[0] = nb_buckets
;
1498 memset(ptr
+ 2, 0, (nb_buckets
+ 1) * sizeof(int));
1502 /* put dynamic tag */
1503 static void put_dt(Section
*dynamic
, int dt
, addr_t val
)
1506 dyn
= section_ptr_add(dynamic
, sizeof(ElfW(Dyn
)));
1508 dyn
->d_un
.d_val
= val
;
1511 static void add_init_array_defines(TCCState
*s1
, const char *section_name
)
1515 char sym_start
[1024];
1518 snprintf(sym_start
, sizeof(sym_start
), "__%s_start", section_name
+ 1);
1519 snprintf(sym_end
, sizeof(sym_end
), "__%s_end", section_name
+ 1);
1521 s
= find_section(s1
, section_name
);
1526 end_offset
= s
->data_offset
;
1529 add_elf_sym(symtab_section
,
1531 ELFW(ST_INFO
)(STB_GLOBAL
, STT_NOTYPE
), 0,
1532 s
->sh_num
, sym_start
);
1533 add_elf_sym(symtab_section
,
1535 ELFW(ST_INFO
)(STB_GLOBAL
, STT_NOTYPE
), 0,
1536 s
->sh_num
, sym_end
);
1539 static int tcc_add_support(TCCState
*s1
, const char *filename
)
1542 snprintf(buf
, sizeof(buf
), "%s/%s/%s", s1
->tcc_lib_path
,
1543 /* an cpu specific path inside tcc_lib_path, mainly for keeping libtcc1.a */
1544 #ifdef TCC_TARGET_I386
1547 #ifdef TCC_TARGET_X86_64
1550 #ifdef TCC_TARGET_ARM
1553 #ifdef TCC_TARGET_ARM64
1556 #ifdef TCC_TARGET_C67
1561 return tcc_add_file(s1
, buf
, TCC_FILETYPE_BINARY
);
1564 ST_FUNC
void tcc_add_bcheck(TCCState
*s1
)
1566 #ifdef CONFIG_TCC_BCHECK
1569 if (0 == s1
->do_bounds_check
)
1572 /* XXX: add an object file to do that */
1573 ptr
= section_ptr_add(bounds_section
, sizeof(*ptr
));
1575 add_elf_sym(symtab_section
, 0, 0,
1576 ELFW(ST_INFO
)(STB_GLOBAL
, STT_NOTYPE
), 0,
1577 bounds_section
->sh_num
, "__bounds_start");
1578 if (s1
->output_type
!= TCC_OUTPUT_MEMORY
) {
1579 /* add 'call __bound_init()' in .init section */
1581 /* XXX not called on MSYS, reason is unknown. For this
1582 case a call to __bound_init is performed in bcheck.c
1583 when __bound_ptr_add, __bound_new_region,
1584 __bound_delete_region called */
1586 int sym_index
= find_elf_sym(symtab_section
, "__bound_init");
1588 Section
*init_section
= find_section(s1
, ".init");
1589 unsigned char *pinit
= section_ptr_add(init_section
, 5);
1591 write32le(pinit
+ 1, -4);
1592 put_elf_reloc(symtab_section
, init_section
,
1593 init_section
->data_offset
- 4, R_386_PC32
, sym_index
);
1596 tcc_warning("__bound_init not defined");
1601 /* add tcc runtime libraries */
1602 ST_FUNC
void tcc_add_runtime(TCCState
*s1
)
1604 tcc_add_pragma_libs(s1
);
1607 if (!s1
->nostdlib
) {
1608 tcc_add_library(s1
, "c");
1609 #ifdef CONFIG_USE_LIBGCC
1610 if (!s1
->static_link
) {
1611 tcc_add_file(s1
, TCC_LIBGCC
, TCC_FILETYPE_BINARY
);
1614 tcc_add_support(s1
, "libtcc1.a");
1617 /* tcc_add_bcheck tries to relocate a call to __bound_init in _init so
1618 libtcc1.a must be loaded before for __bound_init to be defined and
1619 crtn.o must be loaded after to not finalize _init too early. */
1622 if (!s1
->nostdlib
) {
1623 /* add crt end if not memory output */
1624 if (s1
->output_type
!= TCC_OUTPUT_MEMORY
)
1625 tcc_add_crt(s1
, "crtn.o");
1629 /* add various standard linker symbols (must be done after the
1630 sections are filled (for example after allocating common
1632 ST_FUNC
void tcc_add_linker_symbols(TCCState
*s1
)
1638 add_elf_sym(symtab_section
,
1639 text_section
->data_offset
, 0,
1640 ELFW(ST_INFO
)(STB_GLOBAL
, STT_NOTYPE
), 0,
1641 text_section
->sh_num
, "_etext");
1642 add_elf_sym(symtab_section
,
1643 data_section
->data_offset
, 0,
1644 ELFW(ST_INFO
)(STB_GLOBAL
, STT_NOTYPE
), 0,
1645 data_section
->sh_num
, "_edata");
1646 add_elf_sym(symtab_section
,
1647 bss_section
->data_offset
, 0,
1648 ELFW(ST_INFO
)(STB_GLOBAL
, STT_NOTYPE
), 0,
1649 bss_section
->sh_num
, "_end");
1650 /* horrible new standard ldscript defines */
1651 add_init_array_defines(s1
, ".preinit_array");
1652 add_init_array_defines(s1
, ".init_array");
1653 add_init_array_defines(s1
, ".fini_array");
1655 /* add start and stop symbols for sections whose name can be
1657 for(i
= 1; i
< s1
->nb_sections
; i
++) {
1658 s
= s1
->sections
[i
];
1659 if (s
->sh_type
== SHT_PROGBITS
&&
1660 (s
->sh_flags
& SHF_ALLOC
)) {
1664 /* check if section name can be expressed in C */
1670 if (!isid(ch
) && !isnum(ch
))
1674 snprintf(buf
, sizeof(buf
), "__start_%s", s
->name
);
1675 add_elf_sym(symtab_section
,
1677 ELFW(ST_INFO
)(STB_GLOBAL
, STT_NOTYPE
), 0,
1679 snprintf(buf
, sizeof(buf
), "__stop_%s", s
->name
);
1680 add_elf_sym(symtab_section
,
1682 ELFW(ST_INFO
)(STB_GLOBAL
, STT_NOTYPE
), 0,
1689 static void tcc_output_binary(TCCState
*s1
, FILE *f
,
1690 const int *sec_order
)
1693 int i
, offset
, size
;
1696 for(i
=1;i
<s1
->nb_sections
;i
++) {
1697 s
= s1
->sections
[sec_order
[i
]];
1698 if (s
->sh_type
!= SHT_NOBITS
&&
1699 (s
->sh_flags
& SHF_ALLOC
)) {
1700 while (offset
< s
->sh_offset
) {
1705 fwrite(s
->data
, 1, size
, f
);
1711 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
1713 #define EXTRA_RELITEMS 14
1715 /* move the relocation value from .dynsym to .got */
1716 void patch_dynsym_undef(TCCState
*s1
, Section
*s
)
1718 uint32_t *gotd
= (void *)s1
->got
->data
;
1721 gotd
+= 3; /* dummy entries in .got */
1722 /* relocate symbols in .dynsym */
1723 for_each_elem(s
, 1, sym
, ElfW(Sym
)) {
1724 if (sym
->st_shndx
== SHN_UNDEF
) {
1725 *gotd
++ = sym
->st_value
+ 6; /* XXX 6 is magic ? */
1732 #define EXTRA_RELITEMS 9
1734 /* zero plt offsets of weak symbols in .dynsym */
1735 void patch_dynsym_undef(TCCState
*s1
, Section
*s
)
1739 for_each_elem(s
, 1, sym
, ElfW(Sym
))
1740 if (sym
->st_shndx
== SHN_UNDEF
&& ELFW(ST_BIND
)(sym
->st_info
) == STB_WEAK
)
1745 ST_FUNC
void fill_got_entry(TCCState
*s1
, ElfW_Rel
*rel
)
1747 int sym_index
= ELFW(R_SYM
) (rel
->r_info
);
1748 ElfW(Sym
) *sym
= &((ElfW(Sym
) *) symtab_section
->data
)[sym_index
];
1749 unsigned long offset
;
1751 if (sym_index
>= s1
->nb_sym_attrs
)
1753 offset
= s1
->sym_attrs
[sym_index
].got_offset
;
1754 section_reserve(s1
->got
, offset
+ PTR_SIZE
);
1755 #ifdef TCC_TARGET_X86_64
1756 /* only works for x86-64 */
1757 write32le(s1
->got
->data
+ offset
+ 4, sym
->st_value
>> 32);
1759 write32le(s1
->got
->data
+ offset
, sym
->st_value
& 0xffffffff);
1762 /* Perform relocation to GOT or PLT entries */
1763 ST_FUNC
void fill_got(TCCState
*s1
)
1769 for(i
= 1; i
< s1
->nb_sections
; i
++) {
1770 s
= s1
->sections
[i
];
1771 if (s
->sh_type
!= SHT_RELX
)
1773 /* no need to handle got relocations */
1774 if (s
->link
!= symtab_section
)
1776 for_each_elem(s
, 0, rel
, ElfW_Rel
) {
1777 switch (ELFW(R_TYPE
) (rel
->r_info
)) {
1778 case R_X86_64_GOT32
:
1779 case R_X86_64_GOTPCREL
:
1780 case R_X86_64_GOTPCRELX
:
1781 case R_X86_64_REX_GOTPCRELX
:
1782 case R_X86_64_PLT32
:
1783 fill_got_entry(s1
, rel
);
1790 /* Bind symbols of executable: resolve undefined symbols from exported symbols
1791 in shared libraries and export non local defined symbols to shared libraries
1792 if -rdynamic switch was given on command line */
1793 static void bind_exe_dynsyms(TCCState
*s1
)
1796 int sym_index
, index
;
1797 ElfW(Sym
) *sym
, *esym
;
1800 /* Resolve undefined symbols from dynamic symbols. When there is a match:
1801 - if STT_FUNC or STT_GNU_IFUNC symbol -> add it in PLT
1802 - if STT_OBJECT symbol -> add it in .bss section with suitable reloc */
1803 for_each_elem(symtab_section
, 1, sym
, ElfW(Sym
)) {
1804 if (sym
->st_shndx
== SHN_UNDEF
) {
1805 name
= (char *) symtab_section
->link
->data
+ sym
->st_name
;
1806 sym_index
= find_elf_sym(s1
->dynsymtab_section
, name
);
1808 esym
= &((ElfW(Sym
) *)s1
->dynsymtab_section
->data
)[sym_index
];
1809 type
= ELFW(ST_TYPE
)(esym
->st_info
);
1810 if ((type
== STT_FUNC
) || (type
== STT_GNU_IFUNC
)) {
1811 /* Indirect functions shall have STT_FUNC type in executable
1812 * dynsym section. Indeed, a dlsym call following a lazy
1813 * resolution would pick the symbol value from the
1814 * executable dynsym entry which would contain the address
1815 * of the function wanted by the caller of dlsym instead of
1816 * the address of the function that would return that
1818 put_got_entry(s1
, R_JMP_SLOT
, esym
->st_size
,
1819 ELFW(ST_INFO
)(STB_GLOBAL
,STT_FUNC
),
1820 sym
- (ElfW(Sym
) *)symtab_section
->data
);
1821 } else if (type
== STT_OBJECT
) {
1822 unsigned long offset
;
1824 offset
= bss_section
->data_offset
;
1825 /* XXX: which alignment ? */
1826 offset
= (offset
+ 16 - 1) & -16;
1827 index
= put_elf_sym(s1
->dynsym
, offset
, esym
->st_size
,
1828 esym
->st_info
, 0, bss_section
->sh_num
,
1830 /* Ensure R_COPY works for weak symbol aliases */
1831 if (ELFW(ST_BIND
)(esym
->st_info
) == STB_WEAK
) {
1832 for_each_elem(s1
->dynsymtab_section
, 1, dynsym
, ElfW(Sym
)) {
1833 if ((dynsym
->st_value
== esym
->st_value
)
1834 && (ELFW(ST_BIND
)(dynsym
->st_info
) == STB_GLOBAL
)) {
1835 char *dynname
= (char *) s1
->dynsymtab_section
->link
->data
1837 put_elf_sym(s1
->dynsym
, offset
, dynsym
->st_size
,
1839 bss_section
->sh_num
, dynname
);
1844 put_elf_reloc(s1
->dynsym
, bss_section
,
1845 offset
, R_COPY
, index
);
1846 offset
+= esym
->st_size
;
1847 bss_section
->data_offset
= offset
;
1850 /* STB_WEAK undefined symbols are accepted */
1851 /* XXX: _fp_hw seems to be part of the ABI, so we ignore it */
1852 if (ELFW(ST_BIND
)(sym
->st_info
) == STB_WEAK
||
1853 !strcmp(name
, "_fp_hw")) {
1855 tcc_error_noabort("undefined symbol '%s'", name
);
1858 } else if (s1
->rdynamic
&& ELFW(ST_BIND
)(sym
->st_info
) != STB_LOCAL
) {
1859 /* if -rdynamic option, then export all non local symbols */
1860 name
= (char *) symtab_section
->link
->data
+ sym
->st_name
;
1861 put_elf_sym(s1
->dynsym
, sym
->st_value
, sym
->st_size
, sym
->st_info
,
1862 0, sym
->st_shndx
, name
);
1867 /* Bind symbols of libraries: export non local symbols of executable that
1868 resolve undefined symbols of shared libraries */
1869 static void bind_libs_dynsyms(TCCState
*s1
)
1873 ElfW(Sym
) *sym
, *esym
;
1875 /* now look at unresolved dynamic symbols and export
1876 corresponding symbol */
1877 for_each_elem(s1
->dynsymtab_section
, 1, esym
, ElfW(Sym
)) {
1878 name
= (char *) s1
->dynsymtab_section
->link
->data
+ esym
->st_name
;
1879 sym_index
= find_elf_sym(symtab_section
, name
);
1881 /* XXX: avoid adding a symbol if already present because of
1883 sym
= &((ElfW(Sym
) *)symtab_section
->data
)[sym_index
];
1884 if (sym
->st_shndx
!= SHN_UNDEF
)
1885 put_elf_sym(s1
->dynsym
, sym
->st_value
, sym
->st_size
,
1886 sym
->st_info
, 0, sym
->st_shndx
, name
);
1887 } else if (esym
->st_shndx
== SHN_UNDEF
) {
1888 /* weak symbols can stay undefined */
1889 if (ELFW(ST_BIND
)(esym
->st_info
) != STB_WEAK
)
1890 tcc_warning("undefined dynamic symbol '%s'", name
);
1895 /* Export all non local symbols (for shared libraries) */
1896 static void export_global_syms(TCCState
*s1
)
1898 int nb_syms
, dynindex
, index
;
1902 nb_syms
= symtab_section
->data_offset
/ sizeof(ElfW(Sym
));
1903 s1
->symtab_to_dynsym
= tcc_mallocz(sizeof(int) * nb_syms
);
1904 for_each_elem(symtab_section
, 1, sym
, ElfW(Sym
)) {
1905 if (ELFW(ST_BIND
)(sym
->st_info
) != STB_LOCAL
) {
1906 name
= (char *) symtab_section
->link
->data
+ sym
->st_name
;
1907 dynindex
= put_elf_sym(s1
->dynsym
, sym
->st_value
, sym
->st_size
,
1908 sym
->st_info
, 0, sym
->st_shndx
, name
);
1909 index
= sym
- (ElfW(Sym
) *) symtab_section
->data
;
1910 s1
->symtab_to_dynsym
[index
] = dynindex
;
1915 /* relocate the PLT: compute addresses and offsets in the PLT now that final
1916 address for PLT and GOT are known (see fill_program_header) */
1917 ST_FUNC
void relocate_plt(TCCState
*s1
)
1925 p_end
= p
+ s1
->plt
->data_offset
;
1927 #if defined(TCC_TARGET_I386)
1928 write32le(p
+ 2, read32le(p
+ 2) + s1
->got
->sh_addr
);
1929 write32le(p
+ 8, read32le(p
+ 8) + s1
->got
->sh_addr
);
1932 write32le(p
+ 2, read32le(p
+ 2) + s1
->got
->sh_addr
);
1935 #elif defined(TCC_TARGET_X86_64)
1936 int x
= s1
->got
->sh_addr
- s1
->plt
->sh_addr
- 6;
1937 write32le(p
+ 2, read32le(p
+ 2) + x
);
1938 write32le(p
+ 8, read32le(p
+ 8) + x
- 6);
1941 write32le(p
+ 2, read32le(p
+ 2) + x
+ s1
->plt
->data
- p
);
1944 #elif defined(TCC_TARGET_ARM)
1946 x
=s1
->got
->sh_addr
- s1
->plt
->sh_addr
- 12;
1949 if (read32le(p
) == 0x46c04778) /* PLT Thumb stub present */
1951 write32le(p
+ 12, x
+ read32le(p
+ 12) + s1
->plt
->data
- p
);
1954 #elif defined(TCC_TARGET_ARM64)
1955 uint64_t plt
= s1
->plt
->sh_addr
;
1956 uint64_t got
= s1
->got
->sh_addr
;
1957 uint64_t off
= (got
>> 12) - (plt
>> 12);
1958 if ((off
+ ((uint32_t)1 << 20)) >> 21)
1959 tcc_error("Failed relocating PLT (off=0x%lx, got=0x%lx, plt=0x%lx)", off
, got
, plt
);
1960 write32le(p
, 0xa9bf7bf0); // stp x16,x30,[sp,#-16]!
1961 write32le(p
+ 4, (0x90000010 | // adrp x16,...
1962 (off
& 0x1ffffc) << 3 | (off
& 3) << 29));
1963 write32le(p
+ 8, (0xf9400211 | // ldr x17,[x16,#...]
1964 (got
& 0xff8) << 7));
1965 write32le(p
+ 12, (0x91000210 | // add x16,x16,#...
1966 (got
& 0xfff) << 10));
1967 write32le(p
+ 16, 0xd61f0220); // br x17
1968 write32le(p
+ 20, 0xd503201f); // nop
1969 write32le(p
+ 24, 0xd503201f); // nop
1970 write32le(p
+ 28, 0xd503201f); // nop
1973 uint64_t pc
= plt
+ (p
- s1
->plt
->data
);
1974 uint64_t addr
= got
+ read64le(p
);
1975 uint64_t off
= (addr
>> 12) - (pc
>> 12);
1976 if ((off
+ ((uint32_t)1 << 20)) >> 21)
1977 tcc_error("Failed relocating PLT (off=0x%lx, addr=0x%lx, pc=0x%lx)", off
, addr
, pc
);
1978 write32le(p
, (0x90000010 | // adrp x16,...
1979 (off
& 0x1ffffc) << 3 | (off
& 3) << 29));
1980 write32le(p
+ 4, (0xf9400211 | // ldr x17,[x16,#...]
1981 (addr
& 0xff8) << 7));
1982 write32le(p
+ 8, (0x91000210 | // add x16,x16,#...
1983 (addr
& 0xfff) << 10));
1984 write32le(p
+ 12, 0xd61f0220); // br x17
1987 #elif defined(TCC_TARGET_C67)
1990 #error unsupported CPU
1995 /* Allocate strings for section names and decide if an unallocated section
1998 NOTE: the strsec section comes last, so its size is also correct ! */
1999 static void alloc_sec_names(TCCState
*s1
, int file_type
, Section
*strsec
)
2004 /* Allocate strings for section names */
2005 for(i
= 1; i
< s1
->nb_sections
; i
++) {
2006 s
= s1
->sections
[i
];
2007 s
->sh_name
= put_elf_str(strsec
, s
->name
);
2008 /* when generating a DLL, we include relocations but we may
2010 if (file_type
== TCC_OUTPUT_DLL
&&
2011 s
->sh_type
== SHT_RELX
&&
2012 !(s
->sh_flags
& SHF_ALLOC
)) {
2013 /* gr: avoid bogus relocs for empty (debug) sections */
2014 if (s1
->sections
[s
->sh_info
]->sh_flags
& SHF_ALLOC
)
2015 prepare_dynamic_rel(s1
, s
);
2016 else if (s1
->do_debug
)
2017 s
->sh_size
= s
->data_offset
;
2018 } else if (s1
->do_debug
||
2019 file_type
== TCC_OUTPUT_OBJ
||
2020 file_type
== TCC_OUTPUT_EXE
||
2021 (s
->sh_flags
& SHF_ALLOC
) ||
2022 i
== (s1
->nb_sections
- 1)) {
2023 /* we output all sections if debug or object file */
2024 s
->sh_size
= s
->data_offset
;
2029 /* Info to be copied in dynamic section */
2033 unsigned long dyn_rel_off
;
2036 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
2042 /* Assign sections to segments and decide how are sections laid out when loaded
2043 in memory. This function also fills corresponding program headers. */
2044 static int layout_sections(TCCState
*s1
, ElfW(Phdr
) *phdr
, int phnum
,
2045 Section
*interp
, Section
* strsec
,
2046 struct dyn_inf
*dyninf
, int *sec_order
)
2048 int i
, j
, k
, file_type
, sh_order_index
, file_offset
;
2049 unsigned long s_align
;
2055 file_type
= s1
->output_type
;
2058 if (s1
->output_format
== TCC_OUTPUT_FORMAT_ELF
)
2059 file_offset
= sizeof(ElfW(Ehdr
)) + phnum
* sizeof(ElfW(Phdr
));
2060 s_align
= ELF_PAGE_SIZE
;
2061 if (s1
->section_align
)
2062 s_align
= s1
->section_align
;
2065 if (s1
->has_text_addr
) {
2066 int a_offset
, p_offset
;
2067 addr
= s1
->text_addr
;
2068 /* we ensure that (addr % ELF_PAGE_SIZE) == file_offset %
2070 a_offset
= (int) (addr
& (s_align
- 1));
2071 p_offset
= file_offset
& (s_align
- 1);
2072 if (a_offset
< p_offset
)
2073 a_offset
+= s_align
;
2074 file_offset
+= (a_offset
- p_offset
);
2076 if (file_type
== TCC_OUTPUT_DLL
)
2079 addr
= ELF_START_ADDR
;
2080 /* compute address after headers */
2081 addr
+= (file_offset
& (s_align
- 1));
2085 /* Leave one program headers for the program interpreter and one for
2086 the program header table itself if needed. These are done later as
2087 they require section layout to be done first. */
2089 ph
+= 1 + HAVE_PHDR
;
2091 /* dynamic relocation table information, for .dynamic section */
2092 dyninf
->rel_addr
= dyninf
->rel_size
= 0;
2093 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
2094 dyninf
->bss_addr
= dyninf
->bss_size
= 0;
2097 for(j
= 0; j
< 2; j
++) {
2098 ph
->p_type
= PT_LOAD
;
2100 ph
->p_flags
= PF_R
| PF_X
;
2102 ph
->p_flags
= PF_R
| PF_W
;
2103 ph
->p_align
= s_align
;
2105 /* Decide the layout of sections loaded in memory. This must
2106 be done before program headers are filled since they contain
2107 info about the layout. We do the following ordering: interp,
2108 symbol tables, relocations, progbits, nobits */
2109 /* XXX: do faster and simpler sorting */
2110 for(k
= 0; k
< 5; k
++) {
2111 for(i
= 1; i
< s1
->nb_sections
; i
++) {
2112 s
= s1
->sections
[i
];
2113 /* compute if section should be included */
2115 if ((s
->sh_flags
& (SHF_ALLOC
| SHF_WRITE
)) !=
2119 if ((s
->sh_flags
& (SHF_ALLOC
| SHF_WRITE
)) !=
2120 (SHF_ALLOC
| SHF_WRITE
))
2126 } else if (s
->sh_type
== SHT_DYNSYM
||
2127 s
->sh_type
== SHT_STRTAB
||
2128 s
->sh_type
== SHT_HASH
) {
2131 } else if (s
->sh_type
== SHT_RELX
) {
2134 } else if (s
->sh_type
== SHT_NOBITS
) {
2141 sec_order
[sh_order_index
++] = i
;
2143 /* section matches: we align it and add its size */
2145 addr
= (addr
+ s
->sh_addralign
- 1) &
2146 ~(s
->sh_addralign
- 1);
2147 file_offset
+= (int) ( addr
- tmp
);
2148 s
->sh_offset
= file_offset
;
2151 /* update program header infos */
2152 if (ph
->p_offset
== 0) {
2153 ph
->p_offset
= file_offset
;
2155 ph
->p_paddr
= ph
->p_vaddr
;
2157 /* update dynamic relocation infos */
2158 if (s
->sh_type
== SHT_RELX
) {
2159 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
2160 if (!strcmp(strsec
->data
+ s
->sh_name
, ".rel.got")) {
2161 dyninf
->rel_addr
= addr
;
2162 dyninf
->rel_size
+= s
->sh_size
; /* XXX only first rel. */
2164 if (!strcmp(strsec
->data
+ s
->sh_name
, ".rel.bss")) {
2165 dyninf
->bss_addr
= addr
;
2166 dyninf
->bss_size
= s
->sh_size
; /* XXX only first rel. */
2169 if (dyninf
->rel_size
== 0)
2170 dyninf
->rel_addr
= addr
;
2171 dyninf
->rel_size
+= s
->sh_size
;
2175 if (s
->sh_type
!= SHT_NOBITS
)
2176 file_offset
+= s
->sh_size
;
2180 /* Make the first PT_LOAD segment include the program
2181 headers itself (and the ELF header as well), it'll
2182 come out with same memory use but will make various
2183 tools like binutils strip work better. */
2184 ph
->p_offset
&= ~(ph
->p_align
- 1);
2185 ph
->p_vaddr
&= ~(ph
->p_align
- 1);
2186 ph
->p_paddr
&= ~(ph
->p_align
- 1);
2188 ph
->p_filesz
= file_offset
- ph
->p_offset
;
2189 ph
->p_memsz
= addr
- ph
->p_vaddr
;
2192 if (s1
->output_format
== TCC_OUTPUT_FORMAT_ELF
) {
2193 /* if in the middle of a page, we duplicate the page in
2194 memory so that one copy is RX and the other is RW */
2195 if ((addr
& (s_align
- 1)) != 0)
2198 addr
= (addr
+ s_align
- 1) & ~(s_align
- 1);
2199 file_offset
= (file_offset
+ s_align
- 1) & ~(s_align
- 1);
2205 /* all other sections come after */
2206 for(i
= 1; i
< s1
->nb_sections
; i
++) {
2207 s
= s1
->sections
[i
];
2208 if (phnum
> 0 && (s
->sh_flags
& SHF_ALLOC
))
2210 sec_order
[sh_order_index
++] = i
;
2212 file_offset
= (file_offset
+ s
->sh_addralign
- 1) &
2213 ~(s
->sh_addralign
- 1);
2214 s
->sh_offset
= file_offset
;
2215 if (s
->sh_type
!= SHT_NOBITS
)
2216 file_offset
+= s
->sh_size
;
2222 static void fill_unloadable_phdr(ElfW(Phdr
) *phdr
, int phnum
, Section
*interp
,
2227 /* if interpreter, then add corresponding program header */
2233 int len
= phnum
* sizeof(ElfW(Phdr
));
2235 ph
->p_type
= PT_PHDR
;
2236 ph
->p_offset
= sizeof(ElfW(Ehdr
));
2237 ph
->p_vaddr
= interp
->sh_addr
- len
;
2238 ph
->p_paddr
= ph
->p_vaddr
;
2239 ph
->p_filesz
= ph
->p_memsz
= len
;
2240 ph
->p_flags
= PF_R
| PF_X
;
2241 ph
->p_align
= 4; /* interp->sh_addralign; */
2245 ph
->p_type
= PT_INTERP
;
2246 ph
->p_offset
= interp
->sh_offset
;
2247 ph
->p_vaddr
= interp
->sh_addr
;
2248 ph
->p_paddr
= ph
->p_vaddr
;
2249 ph
->p_filesz
= interp
->sh_size
;
2250 ph
->p_memsz
= interp
->sh_size
;
2252 ph
->p_align
= interp
->sh_addralign
;
2255 /* if dynamic section, then add corresponding program header */
2257 ph
= &phdr
[phnum
- 1];
2259 ph
->p_type
= PT_DYNAMIC
;
2260 ph
->p_offset
= dynamic
->sh_offset
;
2261 ph
->p_vaddr
= dynamic
->sh_addr
;
2262 ph
->p_paddr
= ph
->p_vaddr
;
2263 ph
->p_filesz
= dynamic
->sh_size
;
2264 ph
->p_memsz
= dynamic
->sh_size
;
2265 ph
->p_flags
= PF_R
| PF_W
;
2266 ph
->p_align
= dynamic
->sh_addralign
;
2270 /* Fill the dynamic section with tags describing the address and size of
2272 static void fill_dynamic(TCCState
*s1
, struct dyn_inf
*dyninf
)
2276 dynamic
= dyninf
->dynamic
;
2278 /* put dynamic section entries */
2279 dynamic
->data_offset
= dyninf
->dyn_rel_off
;
2280 put_dt(dynamic
, DT_HASH
, s1
->dynsym
->hash
->sh_addr
);
2281 put_dt(dynamic
, DT_STRTAB
, dyninf
->dynstr
->sh_addr
);
2282 put_dt(dynamic
, DT_SYMTAB
, s1
->dynsym
->sh_addr
);
2283 put_dt(dynamic
, DT_STRSZ
, dyninf
->dynstr
->data_offset
);
2284 put_dt(dynamic
, DT_SYMENT
, sizeof(ElfW(Sym
)));
2285 #if defined(TCC_TARGET_ARM64) || defined(TCC_TARGET_X86_64)
2286 put_dt(dynamic
, DT_RELA
, dyninf
->rel_addr
);
2287 put_dt(dynamic
, DT_RELASZ
, dyninf
->rel_size
);
2288 put_dt(dynamic
, DT_RELAENT
, sizeof(ElfW_Rel
));
2290 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
2291 put_dt(dynamic
, DT_PLTGOT
, s1
->got
->sh_addr
);
2292 put_dt(dynamic
, DT_PLTRELSZ
, dyninf
->rel_size
);
2293 put_dt(dynamic
, DT_JMPREL
, dyninf
->rel_addr
);
2294 put_dt(dynamic
, DT_PLTREL
, DT_REL
);
2295 put_dt(dynamic
, DT_REL
, dyninf
->bss_addr
);
2296 put_dt(dynamic
, DT_RELSZ
, dyninf
->bss_size
);
2298 put_dt(dynamic
, DT_REL
, dyninf
->rel_addr
);
2299 put_dt(dynamic
, DT_RELSZ
, dyninf
->rel_size
);
2300 put_dt(dynamic
, DT_RELENT
, sizeof(ElfW_Rel
));
2304 put_dt(dynamic
, DT_DEBUG
, 0);
2305 put_dt(dynamic
, DT_NULL
, 0);
2308 /* Relocate remaining sections and symbols (that is those not related to
2310 static int final_sections_reloc(TCCState
*s1
)
2315 relocate_syms(s1
, 0);
2317 if (s1
->nb_errors
!= 0)
2320 /* relocate sections */
2321 /* XXX: ignore sections with allocated relocations ? */
2322 for(i
= 1; i
< s1
->nb_sections
; i
++) {
2323 s
= s1
->sections
[i
];
2324 #ifdef TCC_TARGET_I386
2325 if (s
->reloc
&& s
!= s1
->got
&& (s
->sh_flags
& SHF_ALLOC
)) //gr
2326 /* On X86 gdb 7.3 works in any case but gdb 6.6 will crash if SHF_ALLOC
2327 checking is removed */
2329 if (s
->reloc
&& s
!= s1
->got
)
2330 /* On X86_64 gdb 7.3 will crash if SHF_ALLOC checking is present */
2332 relocate_section(s1
, s
);
2335 /* relocate relocation entries if the relocation tables are
2336 allocated in the executable */
2337 for(i
= 1; i
< s1
->nb_sections
; i
++) {
2338 s
= s1
->sections
[i
];
2339 if ((s
->sh_flags
& SHF_ALLOC
) &&
2340 s
->sh_type
== SHT_RELX
) {
2341 relocate_rel(s1
, s
);
2347 /* Create an ELF file on disk.
2348 This function handle ELF specific layout requirements */
2349 static void tcc_output_elf(TCCState
*s1
, FILE *f
, int phnum
, ElfW(Phdr
) *phdr
,
2350 int file_offset
, int *sec_order
)
2352 int i
, shnum
, offset
, size
, file_type
;
2355 ElfW(Shdr
) shdr
, *sh
;
2357 file_type
= s1
->output_type
;
2358 shnum
= s1
->nb_sections
;
2360 memset(&ehdr
, 0, sizeof(ehdr
));
2363 ehdr
.e_phentsize
= sizeof(ElfW(Phdr
));
2364 ehdr
.e_phnum
= phnum
;
2365 ehdr
.e_phoff
= sizeof(ElfW(Ehdr
));
2369 file_offset
= (file_offset
+ 3) & -4;
2372 ehdr
.e_ident
[0] = ELFMAG0
;
2373 ehdr
.e_ident
[1] = ELFMAG1
;
2374 ehdr
.e_ident
[2] = ELFMAG2
;
2375 ehdr
.e_ident
[3] = ELFMAG3
;
2376 ehdr
.e_ident
[4] = ELFCLASSW
;
2377 ehdr
.e_ident
[5] = ELFDATA2LSB
;
2378 ehdr
.e_ident
[6] = EV_CURRENT
;
2379 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
2380 ehdr
.e_ident
[EI_OSABI
] = ELFOSABI_FREEBSD
;
2382 #ifdef TCC_TARGET_ARM
2384 ehdr
.e_ident
[EI_OSABI
] = 0;
2385 ehdr
.e_flags
= EF_ARM_EABI_VER4
;
2386 if (file_type
== TCC_OUTPUT_EXE
|| file_type
== TCC_OUTPUT_DLL
)
2387 ehdr
.e_flags
|= EF_ARM_HASENTRY
;
2388 if (s1
->float_abi
== ARM_HARD_FLOAT
)
2389 ehdr
.e_flags
|= EF_ARM_VFP_FLOAT
;
2391 ehdr
.e_flags
|= EF_ARM_SOFT_FLOAT
;
2393 ehdr
.e_ident
[EI_OSABI
] = ELFOSABI_ARM
;
2398 case TCC_OUTPUT_EXE
:
2399 ehdr
.e_type
= ET_EXEC
;
2400 ehdr
.e_entry
= get_elf_sym_addr(s1
, "_start", 1);
2402 case TCC_OUTPUT_DLL
:
2403 ehdr
.e_type
= ET_DYN
;
2404 ehdr
.e_entry
= text_section
->sh_addr
; /* XXX: is it correct ? */
2406 case TCC_OUTPUT_OBJ
:
2407 ehdr
.e_type
= ET_REL
;
2410 ehdr
.e_machine
= EM_TCC_TARGET
;
2411 ehdr
.e_version
= EV_CURRENT
;
2412 ehdr
.e_shoff
= file_offset
;
2413 ehdr
.e_ehsize
= sizeof(ElfW(Ehdr
));
2414 ehdr
.e_shentsize
= sizeof(ElfW(Shdr
));
2415 ehdr
.e_shnum
= shnum
;
2416 ehdr
.e_shstrndx
= shnum
- 1;
2418 fwrite(&ehdr
, 1, sizeof(ElfW(Ehdr
)), f
);
2419 fwrite(phdr
, 1, phnum
* sizeof(ElfW(Phdr
)), f
);
2420 offset
= sizeof(ElfW(Ehdr
)) + phnum
* sizeof(ElfW(Phdr
));
2422 sort_syms(s1
, symtab_section
);
2423 for(i
= 1; i
< s1
->nb_sections
; i
++) {
2424 s
= s1
->sections
[sec_order
[i
]];
2425 if (s
->sh_type
!= SHT_NOBITS
) {
2426 if (s
->sh_type
== SHT_DYNSYM
)
2427 patch_dynsym_undef(s1
, s
);
2428 while (offset
< s
->sh_offset
) {
2434 fwrite(s
->data
, 1, size
, f
);
2439 /* output section headers */
2440 while (offset
< ehdr
.e_shoff
) {
2445 for(i
= 0; i
< s1
->nb_sections
; i
++) {
2447 memset(sh
, 0, sizeof(ElfW(Shdr
)));
2448 s
= s1
->sections
[i
];
2450 sh
->sh_name
= s
->sh_name
;
2451 sh
->sh_type
= s
->sh_type
;
2452 sh
->sh_flags
= s
->sh_flags
;
2453 sh
->sh_entsize
= s
->sh_entsize
;
2454 sh
->sh_info
= s
->sh_info
;
2456 sh
->sh_link
= s
->link
->sh_num
;
2457 sh
->sh_addralign
= s
->sh_addralign
;
2458 sh
->sh_addr
= s
->sh_addr
;
2459 sh
->sh_offset
= s
->sh_offset
;
2460 sh
->sh_size
= s
->sh_size
;
2462 fwrite(sh
, 1, sizeof(ElfW(Shdr
)), f
);
2466 /* Write an elf, coff or "binary" file */
2467 static int tcc_write_elf_file(TCCState
*s1
, const char *filename
, int phnum
,
2468 ElfW(Phdr
) *phdr
, int file_offset
, int *sec_order
)
2470 int fd
, mode
, file_type
;
2473 file_type
= s1
->output_type
;
2474 if (file_type
== TCC_OUTPUT_OBJ
)
2479 fd
= open(filename
, O_WRONLY
| O_CREAT
| O_TRUNC
| O_BINARY
, mode
);
2481 tcc_error_noabort("could not write '%s'", filename
);
2484 f
= fdopen(fd
, "wb");
2486 printf("<- %s\n", filename
);
2488 #ifdef TCC_TARGET_COFF
2489 if (s1
->output_format
== TCC_OUTPUT_FORMAT_COFF
)
2490 tcc_output_coff(s1
, f
);
2493 if (s1
->output_format
== TCC_OUTPUT_FORMAT_ELF
)
2494 tcc_output_elf(s1
, f
, phnum
, phdr
, file_offset
, sec_order
);
2496 tcc_output_binary(s1
, f
, sec_order
);
2502 /* Output an elf, coff or binary file */
2503 /* XXX: suppress unneeded sections */
2504 static int elf_output_file(TCCState
*s1
, const char *filename
)
2506 int i
, ret
, phnum
, shnum
, file_type
, file_offset
, *sec_order
;
2507 struct dyn_inf dyninf
;
2510 Section
*strsec
, *interp
, *dynamic
, *dynstr
;
2512 file_type
= s1
->output_type
;
2515 /* if linking, also link in runtime libraries (libc, libgcc, etc.) */
2516 if (file_type
!= TCC_OUTPUT_OBJ
) {
2517 tcc_add_runtime(s1
);
2522 interp
= dynamic
= dynstr
= NULL
; /* avoid warning */
2523 dyninf
.dyn_rel_off
= 0; /* avoid warning */
2525 if (file_type
!= TCC_OUTPUT_OBJ
) {
2526 relocate_common_syms();
2528 tcc_add_linker_symbols(s1
);
2530 if (!s1
->static_link
) {
2531 if (file_type
== TCC_OUTPUT_EXE
) {
2533 /* allow override the dynamic loader */
2534 const char *elfint
= getenv("LD_SO");
2536 elfint
= DEFAULT_ELFINTERP(s1
);
2537 /* add interpreter section only if executable */
2538 interp
= new_section(s1
, ".interp", SHT_PROGBITS
, SHF_ALLOC
);
2539 interp
->sh_addralign
= 1;
2540 ptr
= section_ptr_add(interp
, 1 + strlen(elfint
));
2541 strcpy(ptr
, elfint
);
2544 /* add dynamic symbol table */
2545 s1
->dynsym
= new_symtab(s1
, ".dynsym", SHT_DYNSYM
, SHF_ALLOC
,
2547 ".hash", SHF_ALLOC
);
2548 dynstr
= s1
->dynsym
->link
;
2550 /* add dynamic section */
2551 dynamic
= new_section(s1
, ".dynamic", SHT_DYNAMIC
,
2552 SHF_ALLOC
| SHF_WRITE
);
2553 dynamic
->link
= dynstr
;
2554 dynamic
->sh_entsize
= sizeof(ElfW(Dyn
));
2558 if (file_type
== TCC_OUTPUT_EXE
) {
2559 bind_exe_dynsyms(s1
);
2561 if (s1
->nb_errors
) {
2566 bind_libs_dynsyms(s1
);
2567 } else /* shared library case: simply export all global symbols */
2568 export_global_syms(s1
);
2570 build_got_entries(s1
);
2572 /* add a list of needed dlls */
2573 for(i
= 0; i
< s1
->nb_loaded_dlls
; i
++) {
2574 DLLReference
*dllref
= s1
->loaded_dlls
[i
];
2575 if (dllref
->level
== 0)
2576 put_dt(dynamic
, DT_NEEDED
, put_elf_str(dynstr
, dllref
->name
));
2580 put_dt(dynamic
, DT_RPATH
, put_elf_str(dynstr
, s1
->rpath
));
2582 /* XXX: currently, since we do not handle PIC code, we
2583 must relocate the readonly segments */
2584 if (file_type
== TCC_OUTPUT_DLL
) {
2586 put_dt(dynamic
, DT_SONAME
, put_elf_str(dynstr
, s1
->soname
));
2587 put_dt(dynamic
, DT_TEXTREL
, 0);
2591 put_dt(dynamic
, DT_SYMBOLIC
, 0);
2593 /* add necessary space for other entries */
2594 dyninf
.dyn_rel_off
= dynamic
->data_offset
;
2595 dynamic
->data_offset
+= sizeof(ElfW(Dyn
)) * EXTRA_RELITEMS
;
2597 /* still need to build got entries in case of static link */
2598 build_got_entries(s1
);
2602 /* we add a section for symbols */
2603 strsec
= new_section(s1
, ".shstrtab", SHT_STRTAB
, 0);
2604 put_elf_str(strsec
, "");
2606 /* compute number of sections */
2607 shnum
= s1
->nb_sections
;
2609 /* this array is used to reorder sections in the output file */
2610 sec_order
= tcc_malloc(sizeof(int) * shnum
);
2613 /* compute number of program headers */
2616 case TCC_OUTPUT_OBJ
:
2619 case TCC_OUTPUT_EXE
:
2620 if (!s1
->static_link
)
2621 phnum
= 4 + HAVE_PHDR
;
2625 case TCC_OUTPUT_DLL
:
2630 /* Allocate strings for section names */
2631 alloc_sec_names(s1
, file_type
, strsec
);
2633 /* allocate program segment headers */
2634 phdr
= tcc_mallocz(phnum
* sizeof(ElfW(Phdr
)));
2636 /* compute section to program header mapping */
2637 file_offset
= layout_sections(s1
, phdr
, phnum
, interp
, strsec
, &dyninf
,
2640 /* Fill remaining program header and finalize relocation related to dynamic
2643 fill_unloadable_phdr(phdr
, phnum
, interp
, dynamic
);
2645 dyninf
.dynamic
= dynamic
;
2646 dyninf
.dynstr
= dynstr
;
2648 fill_dynamic(s1
, &dyninf
);
2650 /* put in GOT the dynamic section address and relocate PLT */
2651 write32le(s1
->got
->data
, dynamic
->sh_addr
);
2652 if (file_type
== TCC_OUTPUT_EXE
2653 #if defined(TCC_OUTPUT_DLL_WITH_PLT)
2654 || file_type
== TCC_OUTPUT_DLL
2659 /* relocate symbols in .dynsym now that final addresses are known */
2660 for_each_elem(s1
->dynsym
, 1, sym
, ElfW(Sym
)) {
2661 if (sym
->st_shndx
== SHN_UNDEF
) {
2662 /* relocate to PLT if symbol corresponds to a PLT entry,
2663 but not if it's a weak symbol */
2664 if (ELFW(ST_BIND
)(sym
->st_info
) == STB_WEAK
)
2666 else if (sym
->st_value
)
2667 sym
->st_value
+= s1
->plt
->sh_addr
;
2668 } else if (sym
->st_shndx
< SHN_LORESERVE
) {
2669 /* do symbol relocation */
2670 sym
->st_value
+= s1
->sections
[sym
->st_shndx
]->sh_addr
;
2676 /* if building executable or DLL, then relocate each section
2677 except the GOT which is already relocated */
2678 if (file_type
!= TCC_OUTPUT_OBJ
) {
2679 ret
= final_sections_reloc(s1
);
2684 /* Perform relocation to GOT or PLT entries */
2685 if (file_type
== TCC_OUTPUT_EXE
&& s1
->static_link
)
2688 /* Create the ELF file with name 'filename' */
2689 ret
= tcc_write_elf_file(s1
, filename
, phnum
, phdr
, file_offset
, sec_order
);
2692 const char *strip_cmd
= "sstrip "; // super strip utility from ELFkickers
2693 const char *null_dev
= " 2> /dev/null";
2695 snprintf(buf
, sizeof(buf
), "%s%s%s", strip_cmd
, filename
, null_dev
);
2698 system(buf
+1); // call a strip utility from binutils
2701 tcc_free(s1
->symtab_to_dynsym
);
2702 tcc_free(sec_order
);
2704 tcc_free(s1
->sym_attrs
);
2705 s1
->sym_attrs
= NULL
;
2709 LIBTCCAPI
int tcc_output_file(TCCState
*s
, const char *filename
)
2712 #ifdef TCC_TARGET_PE
2713 if (s
->output_type
!= TCC_OUTPUT_OBJ
) {
2714 ret
= pe_output_file(s
, filename
);
2717 ret
= elf_output_file(s
, filename
);
2721 static void *load_data(int fd
, unsigned long file_offset
, unsigned long size
)
2725 data
= tcc_malloc(size
);
2726 lseek(fd
, file_offset
, SEEK_SET
);
2727 read(fd
, data
, size
);
2731 typedef struct SectionMergeInfo
{
2732 Section
*s
; /* corresponding existing section */
2733 unsigned long offset
; /* offset of the new section in the existing section */
2734 uint8_t new_section
; /* true if section 's' was added */
2735 uint8_t link_once
; /* true if link once section */
2738 /* load an object file and merge it with current files */
2739 /* XXX: handle correctly stab (debug) info */
2740 ST_FUNC
int tcc_load_object_file(TCCState
*s1
,
2741 int fd
, unsigned long file_offset
)
2744 ElfW(Shdr
) *shdr
, *sh
;
2745 int size
, i
, j
, offset
, offseti
, nb_syms
, sym_index
, ret
;
2746 unsigned char *strsec
, *strtab
;
2747 int *old_to_new_syms
;
2748 char *sh_name
, *name
;
2749 SectionMergeInfo
*sm_table
, *sm
;
2750 ElfW(Sym
) *sym
, *symtab
;
2757 stab_index
= stabstr_index
= 0;
2759 if (read(fd
, &ehdr
, sizeof(ehdr
)) != sizeof(ehdr
))
2761 if (ehdr
.e_ident
[0] != ELFMAG0
||
2762 ehdr
.e_ident
[1] != ELFMAG1
||
2763 ehdr
.e_ident
[2] != ELFMAG2
||
2764 ehdr
.e_ident
[3] != ELFMAG3
)
2766 /* test if object file */
2767 if (ehdr
.e_type
!= ET_REL
)
2769 /* test CPU specific stuff */
2770 if (ehdr
.e_ident
[5] != ELFDATA2LSB
||
2771 ehdr
.e_machine
!= EM_TCC_TARGET
) {
2773 tcc_error_noabort("invalid object file");
2777 shdr
= load_data(fd
, file_offset
+ ehdr
.e_shoff
,
2778 sizeof(ElfW(Shdr
)) * ehdr
.e_shnum
);
2779 sm_table
= tcc_mallocz(sizeof(SectionMergeInfo
) * ehdr
.e_shnum
);
2781 /* load section names */
2782 sh
= &shdr
[ehdr
.e_shstrndx
];
2783 strsec
= load_data(fd
, file_offset
+ sh
->sh_offset
, sh
->sh_size
);
2785 /* load symtab and strtab */
2786 old_to_new_syms
= NULL
;
2790 for(i
= 1; i
< ehdr
.e_shnum
; i
++) {
2792 if (sh
->sh_type
== SHT_SYMTAB
) {
2794 tcc_error_noabort("object must contain only one symtab");
2799 nb_syms
= sh
->sh_size
/ sizeof(ElfW(Sym
));
2800 symtab
= load_data(fd
, file_offset
+ sh
->sh_offset
, sh
->sh_size
);
2801 sm_table
[i
].s
= symtab_section
;
2803 /* now load strtab */
2804 sh
= &shdr
[sh
->sh_link
];
2805 strtab
= load_data(fd
, file_offset
+ sh
->sh_offset
, sh
->sh_size
);
2809 /* now examine each section and try to merge its content with the
2811 for(i
= 1; i
< ehdr
.e_shnum
; i
++) {
2812 /* no need to examine section name strtab */
2813 if (i
== ehdr
.e_shstrndx
)
2816 sh_name
= (char *) strsec
+ sh
->sh_name
;
2817 /* ignore sections types we do not handle */
2818 if (sh
->sh_type
!= SHT_PROGBITS
&&
2819 sh
->sh_type
!= SHT_RELX
&&
2821 sh
->sh_type
!= SHT_ARM_EXIDX
&&
2823 sh
->sh_type
!= SHT_NOBITS
&&
2824 sh
->sh_type
!= SHT_PREINIT_ARRAY
&&
2825 sh
->sh_type
!= SHT_INIT_ARRAY
&&
2826 sh
->sh_type
!= SHT_FINI_ARRAY
&&
2827 strcmp(sh_name
, ".stabstr")
2830 if (sh
->sh_addralign
< 1)
2831 sh
->sh_addralign
= 1;
2832 /* find corresponding section, if any */
2833 for(j
= 1; j
< s1
->nb_sections
;j
++) {
2834 s
= s1
->sections
[j
];
2835 if (!strcmp(s
->name
, sh_name
)) {
2836 if (!strncmp(sh_name
, ".gnu.linkonce",
2837 sizeof(".gnu.linkonce") - 1)) {
2838 /* if a 'linkonce' section is already present, we
2839 do not add it again. It is a little tricky as
2840 symbols can still be defined in
2842 sm_table
[i
].link_once
= 1;
2849 /* not found: create new section */
2850 s
= new_section(s1
, sh_name
, sh
->sh_type
, sh
->sh_flags
);
2851 /* take as much info as possible from the section. sh_link and
2852 sh_info will be updated later */
2853 s
->sh_addralign
= sh
->sh_addralign
;
2854 s
->sh_entsize
= sh
->sh_entsize
;
2855 sm_table
[i
].new_section
= 1;
2857 if (sh
->sh_type
!= s
->sh_type
) {
2858 tcc_error_noabort("invalid section type");
2862 /* align start of section */
2863 offset
= s
->data_offset
;
2865 if (0 == strcmp(sh_name
, ".stab")) {
2869 if (0 == strcmp(sh_name
, ".stabstr")) {
2874 size
= sh
->sh_addralign
- 1;
2875 offset
= (offset
+ size
) & ~size
;
2876 if (sh
->sh_addralign
> s
->sh_addralign
)
2877 s
->sh_addralign
= sh
->sh_addralign
;
2878 s
->data_offset
= offset
;
2880 sm_table
[i
].offset
= offset
;
2882 /* concatenate sections */
2884 if (sh
->sh_type
!= SHT_NOBITS
) {
2886 lseek(fd
, file_offset
+ sh
->sh_offset
, SEEK_SET
);
2887 ptr
= section_ptr_add(s
, size
);
2888 read(fd
, ptr
, size
);
2890 s
->data_offset
+= size
;
2895 /* gr relocate stab strings */
2896 if (stab_index
&& stabstr_index
) {
2899 s
= sm_table
[stab_index
].s
;
2900 a
= (Stab_Sym
*)(s
->data
+ sm_table
[stab_index
].offset
);
2901 b
= (Stab_Sym
*)(s
->data
+ s
->data_offset
);
2902 o
= sm_table
[stabstr_index
].offset
;
2904 a
->n_strx
+= o
, a
++;
2907 /* second short pass to update sh_link and sh_info fields of new
2909 for(i
= 1; i
< ehdr
.e_shnum
; i
++) {
2911 if (!s
|| !sm_table
[i
].new_section
)
2914 if (sh
->sh_link
> 0)
2915 s
->link
= sm_table
[sh
->sh_link
].s
;
2916 if (sh
->sh_type
== SHT_RELX
) {
2917 s
->sh_info
= sm_table
[sh
->sh_info
].s
->sh_num
;
2918 /* update backward link */
2919 s1
->sections
[s
->sh_info
]->reloc
= s
;
2924 /* resolve symbols */
2925 old_to_new_syms
= tcc_mallocz(nb_syms
* sizeof(int));
2928 for(i
= 1; i
< nb_syms
; i
++, sym
++) {
2929 if (sym
->st_shndx
!= SHN_UNDEF
&&
2930 sym
->st_shndx
< SHN_LORESERVE
) {
2931 sm
= &sm_table
[sym
->st_shndx
];
2932 if (sm
->link_once
) {
2933 /* if a symbol is in a link once section, we use the
2934 already defined symbol. It is very important to get
2935 correct relocations */
2936 if (ELFW(ST_BIND
)(sym
->st_info
) != STB_LOCAL
) {
2937 name
= (char *) strtab
+ sym
->st_name
;
2938 sym_index
= find_elf_sym(symtab_section
, name
);
2940 old_to_new_syms
[i
] = sym_index
;
2944 /* if no corresponding section added, no need to add symbol */
2947 /* convert section number */
2948 sym
->st_shndx
= sm
->s
->sh_num
;
2950 sym
->st_value
+= sm
->offset
;
2953 name
= (char *) strtab
+ sym
->st_name
;
2954 sym_index
= add_elf_sym(symtab_section
, sym
->st_value
, sym
->st_size
,
2955 sym
->st_info
, sym
->st_other
,
2956 sym
->st_shndx
, name
);
2957 old_to_new_syms
[i
] = sym_index
;
2960 /* third pass to patch relocation entries */
2961 for(i
= 1; i
< ehdr
.e_shnum
; i
++) {
2966 offset
= sm_table
[i
].offset
;
2967 switch(s
->sh_type
) {
2969 /* take relocation offset information */
2970 offseti
= sm_table
[sh
->sh_info
].offset
;
2971 for_each_elem(s
, (offset
/ sizeof(*rel
)), rel
, ElfW_Rel
) {
2974 /* convert symbol index */
2975 type
= ELFW(R_TYPE
)(rel
->r_info
);
2976 sym_index
= ELFW(R_SYM
)(rel
->r_info
);
2977 /* NOTE: only one symtab assumed */
2978 if (sym_index
>= nb_syms
)
2980 sym_index
= old_to_new_syms
[sym_index
];
2981 /* ignore link_once in rel section. */
2982 if (!sym_index
&& !sm
->link_once
2983 #ifdef TCC_TARGET_ARM
2984 && type
!= R_ARM_V4BX
2988 tcc_error_noabort("Invalid relocation entry [%2d] '%s' @ %.8x",
2989 i
, strsec
+ sh
->sh_name
, rel
->r_offset
);
2992 rel
->r_info
= ELFW(R_INFO
)(sym_index
, type
);
2993 /* offset the relocation offset */
2994 rel
->r_offset
+= offseti
;
2995 #ifdef TCC_TARGET_ARM
2996 /* Jumps and branches from a Thumb code to a PLT entry need
2997 special handling since PLT entries are ARM code.
2998 Unconditional bl instructions referencing PLT entries are
2999 handled by converting these instructions into blx
3000 instructions. Other case of instructions referencing a PLT
3001 entry require to add a Thumb stub before the PLT entry to
3002 switch to ARM mode. We set bit plt_thumb_stub of the
3003 attribute of a symbol to indicate such a case. */
3004 if (type
== R_ARM_THM_JUMP24
)
3005 alloc_sym_attr(s1
, sym_index
)->plt_thumb_stub
= 1;
3018 tcc_free(old_to_new_syms
);
3025 typedef struct ArchiveHeader
{
3026 char ar_name
[16]; /* name of this member */
3027 char ar_date
[12]; /* file mtime */
3028 char ar_uid
[6]; /* owner uid; printed as decimal */
3029 char ar_gid
[6]; /* owner gid; printed as decimal */
3030 char ar_mode
[8]; /* file mode, printed as octal */
3031 char ar_size
[10]; /* file size, printed as decimal */
3032 char ar_fmag
[2]; /* should contain ARFMAG */
3035 static int get_be32(const uint8_t *b
)
3037 return b
[3] | (b
[2] << 8) | (b
[1] << 16) | (b
[0] << 24);
3040 /* load only the objects which resolve undefined symbols */
3041 static int tcc_load_alacarte(TCCState
*s1
, int fd
, int size
)
3043 int i
, bound
, nsyms
, sym_index
, off
, ret
;
3045 const char *ar_names
, *p
;
3046 const uint8_t *ar_index
;
3049 data
= tcc_malloc(size
);
3050 if (read(fd
, data
, size
) != size
)
3052 nsyms
= get_be32(data
);
3053 ar_index
= data
+ 4;
3054 ar_names
= (char *) ar_index
+ nsyms
* 4;
3058 for(p
= ar_names
, i
= 0; i
< nsyms
; i
++, p
+= strlen(p
)+1) {
3059 sym_index
= find_elf_sym(symtab_section
, p
);
3061 sym
= &((ElfW(Sym
) *)symtab_section
->data
)[sym_index
];
3062 if(sym
->st_shndx
== SHN_UNDEF
) {
3063 off
= get_be32(ar_index
+ i
* 4) + sizeof(ArchiveHeader
);
3065 lseek(fd
, off
, SEEK_SET
);
3066 if(tcc_load_object_file(s1
, fd
, off
) < 0) {
3081 /* load a '.a' file */
3082 ST_FUNC
int tcc_load_archive(TCCState
*s1
, int fd
)
3089 unsigned long file_offset
;
3091 /* skip magic which was already checked */
3092 read(fd
, magic
, sizeof(magic
));
3095 len
= read(fd
, &hdr
, sizeof(hdr
));
3098 if (len
!= sizeof(hdr
)) {
3099 tcc_error_noabort("invalid archive");
3102 memcpy(ar_size
, hdr
.ar_size
, sizeof(hdr
.ar_size
));
3103 ar_size
[sizeof(hdr
.ar_size
)] = '\0';
3104 size
= strtol(ar_size
, NULL
, 0);
3105 memcpy(ar_name
, hdr
.ar_name
, sizeof(hdr
.ar_name
));
3106 for(i
= sizeof(hdr
.ar_name
) - 1; i
>= 0; i
--) {
3107 if (ar_name
[i
] != ' ')
3110 ar_name
[i
+ 1] = '\0';
3111 file_offset
= lseek(fd
, 0, SEEK_CUR
);
3113 size
= (size
+ 1) & ~1;
3114 if (!strcmp(ar_name
, "/")) {
3115 /* coff symbol table : we handle it */
3116 if(s1
->alacarte_link
)
3117 return tcc_load_alacarte(s1
, fd
, size
);
3118 } else if (!strcmp(ar_name
, "//") ||
3119 !strcmp(ar_name
, "__.SYMDEF") ||
3120 !strcmp(ar_name
, "__.SYMDEF/") ||
3121 !strcmp(ar_name
, "ARFILENAMES/")) {
3122 /* skip symbol table or archive names */
3124 if (tcc_load_object_file(s1
, fd
, file_offset
) < 0)
3127 lseek(fd
, file_offset
+ size
, SEEK_SET
);
3132 #ifndef TCC_TARGET_PE
3133 /* load a DLL and all referenced DLLs. 'level = 0' means that the DLL
3134 is referenced by the user (so it should be added as DT_NEEDED in
3135 the generated ELF file) */
3136 ST_FUNC
int tcc_load_dll(TCCState
*s1
, int fd
, const char *filename
, int level
)
3139 ElfW(Shdr
) *shdr
, *sh
, *sh1
;
3140 int i
, j
, nb_syms
, nb_dts
, sym_bind
, ret
;
3141 ElfW(Sym
) *sym
, *dynsym
;
3142 ElfW(Dyn
) *dt
, *dynamic
;
3143 unsigned char *dynstr
;
3144 const char *name
, *soname
;
3145 DLLReference
*dllref
;
3147 read(fd
, &ehdr
, sizeof(ehdr
));
3149 /* test CPU specific stuff */
3150 if (ehdr
.e_ident
[5] != ELFDATA2LSB
||
3151 ehdr
.e_machine
!= EM_TCC_TARGET
) {
3152 tcc_error_noabort("bad architecture");
3157 shdr
= load_data(fd
, ehdr
.e_shoff
, sizeof(ElfW(Shdr
)) * ehdr
.e_shnum
);
3159 /* load dynamic section and dynamic symbols */
3163 dynsym
= NULL
; /* avoid warning */
3164 dynstr
= NULL
; /* avoid warning */
3165 for(i
= 0, sh
= shdr
; i
< ehdr
.e_shnum
; i
++, sh
++) {
3166 switch(sh
->sh_type
) {
3168 nb_dts
= sh
->sh_size
/ sizeof(ElfW(Dyn
));
3169 dynamic
= load_data(fd
, sh
->sh_offset
, sh
->sh_size
);
3172 nb_syms
= sh
->sh_size
/ sizeof(ElfW(Sym
));
3173 dynsym
= load_data(fd
, sh
->sh_offset
, sh
->sh_size
);
3174 sh1
= &shdr
[sh
->sh_link
];
3175 dynstr
= load_data(fd
, sh1
->sh_offset
, sh1
->sh_size
);
3182 /* compute the real library name */
3183 soname
= tcc_basename(filename
);
3185 for(i
= 0, dt
= dynamic
; i
< nb_dts
; i
++, dt
++) {
3186 if (dt
->d_tag
== DT_SONAME
) {
3187 soname
= (char *) dynstr
+ dt
->d_un
.d_val
;
3191 /* if the dll is already loaded, do not load it */
3192 for(i
= 0; i
< s1
->nb_loaded_dlls
; i
++) {
3193 dllref
= s1
->loaded_dlls
[i
];
3194 if (!strcmp(soname
, dllref
->name
)) {
3195 /* but update level if needed */
3196 if (level
< dllref
->level
)
3197 dllref
->level
= level
;
3203 /* add the dll and its level */
3204 dllref
= tcc_mallocz(sizeof(DLLReference
) + strlen(soname
));
3205 dllref
->level
= level
;
3206 strcpy(dllref
->name
, soname
);
3207 dynarray_add((void ***)&s1
->loaded_dlls
, &s1
->nb_loaded_dlls
, dllref
);
3209 /* add dynamic symbols in dynsym_section */
3210 for(i
= 1, sym
= dynsym
+ 1; i
< nb_syms
; i
++, sym
++) {
3211 sym_bind
= ELFW(ST_BIND
)(sym
->st_info
);
3212 if (sym_bind
== STB_LOCAL
)
3214 name
= (char *) dynstr
+ sym
->st_name
;
3215 add_elf_sym(s1
->dynsymtab_section
, sym
->st_value
, sym
->st_size
,
3216 sym
->st_info
, sym
->st_other
, sym
->st_shndx
, name
);
3219 /* load all referenced DLLs */
3220 for(i
= 0, dt
= dynamic
; i
< nb_dts
; i
++, dt
++) {
3223 name
= (char *) dynstr
+ dt
->d_un
.d_val
;
3224 for(j
= 0; j
< s1
->nb_loaded_dlls
; j
++) {
3225 dllref
= s1
->loaded_dlls
[j
];
3226 if (!strcmp(name
, dllref
->name
))
3227 goto already_loaded
;
3229 if (tcc_add_dll(s1
, name
, AFF_REFERENCED_DLL
) < 0) {
3230 tcc_error_noabort("referenced dll '%s' not found", name
);
3247 #define LD_TOK_NAME 256
3248 #define LD_TOK_EOF (-1)
3250 /* return next ld script token */
3251 static int ld_next(TCCState
*s1
, char *name
, int name_size
)
3269 file
->buf_ptr
= parse_comment(file
->buf_ptr
);
3270 ch
= file
->buf_ptr
[0];
3283 /* case 'a' ... 'z': */
3310 /* case 'A' ... 'z': */
3344 if (!((ch
>= 'a' && ch
<= 'z') ||
3345 (ch
>= 'A' && ch
<= 'Z') ||
3346 (ch
>= '0' && ch
<= '9') ||
3347 strchr("/.-_+=$:\\,~", ch
)))
3349 if ((q
- name
) < name_size
- 1) {
3368 static int ld_add_file(TCCState
*s1
, const char filename
[])
3372 ret
= tcc_add_file_internal(s1
, filename
, 0, TCC_FILETYPE_BINARY
);
3374 ret
= tcc_add_dll(s1
, filename
, 0);
3378 static inline int new_undef_syms(void)
3381 ret
= new_undef_sym
;
3386 static int ld_add_file_list(TCCState
*s1
, const char *cmd
, int as_needed
)
3388 char filename
[1024], libname
[1024];
3389 int t
, group
, nblibs
= 0, ret
= 0;
3392 group
= !strcmp(cmd
, "GROUP");
3395 t
= ld_next(s1
, filename
, sizeof(filename
));
3398 t
= ld_next(s1
, filename
, sizeof(filename
));
3401 if (t
== LD_TOK_EOF
) {
3402 tcc_error_noabort("unexpected end of file");
3404 goto lib_parse_error
;
3405 } else if (t
== ')') {
3407 } else if (t
== '-') {
3408 t
= ld_next(s1
, filename
, sizeof(filename
));
3409 if ((t
!= LD_TOK_NAME
) || (filename
[0] != 'l')) {
3410 tcc_error_noabort("library name expected");
3412 goto lib_parse_error
;
3414 pstrcpy(libname
, sizeof libname
, &filename
[1]);
3415 if (s1
->static_link
) {
3416 snprintf(filename
, sizeof filename
, "lib%s.a", libname
);
3418 snprintf(filename
, sizeof filename
, "lib%s.so", libname
);
3420 } else if (t
!= LD_TOK_NAME
) {
3421 tcc_error_noabort("filename expected");
3423 goto lib_parse_error
;
3425 if (!strcmp(filename
, "AS_NEEDED")) {
3426 ret
= ld_add_file_list(s1
, cmd
, 1);
3428 goto lib_parse_error
;
3430 /* TODO: Implement AS_NEEDED support. Ignore it for now */
3432 ret
= ld_add_file(s1
, filename
);
3434 goto lib_parse_error
;
3436 /* Add the filename *and* the libname to avoid future conversions */
3437 dynarray_add((void ***) &libs
, &nblibs
, tcc_strdup(filename
));
3438 if (libname
[0] != '\0')
3439 dynarray_add((void ***) &libs
, &nblibs
, tcc_strdup(libname
));
3443 t
= ld_next(s1
, filename
, sizeof(filename
));
3445 t
= ld_next(s1
, filename
, sizeof(filename
));
3448 if (group
&& !as_needed
) {
3449 while (new_undef_syms()) {
3452 for (i
= 0; i
< nblibs
; i
++)
3453 ld_add_file(s1
, libs
[i
]);
3457 dynarray_reset(&libs
, &nblibs
);
3461 /* interpret a subset of GNU ldscripts to handle the dummy libc.so
3463 ST_FUNC
int tcc_load_ldscript(TCCState
*s1
)
3466 char filename
[1024];
3471 t
= ld_next(s1
, cmd
, sizeof(cmd
));
3472 if (t
== LD_TOK_EOF
)
3474 else if (t
!= LD_TOK_NAME
)
3476 if (!strcmp(cmd
, "INPUT") ||
3477 !strcmp(cmd
, "GROUP")) {
3478 ret
= ld_add_file_list(s1
, cmd
, 0);
3481 } else if (!strcmp(cmd
, "OUTPUT_FORMAT") ||
3482 !strcmp(cmd
, "TARGET")) {
3483 /* ignore some commands */
3484 t
= ld_next(s1
, cmd
, sizeof(cmd
));
3488 t
= ld_next(s1
, filename
, sizeof(filename
));
3489 if (t
== LD_TOK_EOF
) {
3490 tcc_error_noabort("unexpected end of file");
3492 } else if (t
== ')') {
3502 #endif /* !TCC_TARGET_PE */