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 write32le(ptr
, read32le(ptr
) +
945 (s1
->got
->sh_addr
- addr
+
946 s1
->sym_attrs
[sym_index
].got_offset
- 4));
948 case R_X86_64_GOTTPOFF
:
949 write32le(ptr
, read32le(ptr
) + val
- s1
->got
->sh_addr
);
952 /* we load the got offset */
953 write32le(ptr
, read32le(ptr
) + s1
->sym_attrs
[sym_index
].got_offset
);
956 #error unsupported processor
960 /* if the relocation is allocated, we change its symbol table */
961 if (sr
->sh_flags
& SHF_ALLOC
)
962 sr
->link
= s1
->dynsym
;
965 /* relocate relocation table in 'sr' */
966 static void relocate_rel(TCCState
*s1
, Section
*sr
)
971 s
= s1
->sections
[sr
->sh_info
];
972 for_each_elem(sr
, 0, rel
, ElfW_Rel
)
973 rel
->r_offset
+= s
->sh_addr
;
976 /* count the number of dynamic relocations so that we can reserve
978 static int prepare_dynamic_rel(TCCState
*s1
, Section
*sr
)
981 int sym_index
, esym_index
, type
, count
;
984 for_each_elem(sr
, 0, rel
, ElfW_Rel
) {
985 sym_index
= ELFW(R_SYM
)(rel
->r_info
);
986 type
= ELFW(R_TYPE
)(rel
->r_info
);
988 #if defined(TCC_TARGET_I386)
990 #elif defined(TCC_TARGET_X86_64)
997 #if defined(TCC_TARGET_I386)
999 #elif defined(TCC_TARGET_X86_64)
1002 esym_index
= s1
->symtab_to_dynsym
[sym_index
];
1011 /* allocate the section */
1012 sr
->sh_flags
|= SHF_ALLOC
;
1013 sr
->sh_size
= count
* sizeof(ElfW_Rel
);
1018 static struct sym_attr
*alloc_sym_attr(TCCState
*s1
, int index
)
1021 struct sym_attr
*tab
;
1023 if (index
>= s1
->nb_sym_attrs
) {
1024 /* find immediately bigger power of 2 and reallocate array */
1028 tab
= tcc_realloc(s1
->sym_attrs
, n
* sizeof(*s1
->sym_attrs
));
1029 s1
->sym_attrs
= tab
;
1030 memset(s1
->sym_attrs
+ s1
->nb_sym_attrs
, 0,
1031 (n
- s1
->nb_sym_attrs
) * sizeof(*s1
->sym_attrs
));
1032 s1
->nb_sym_attrs
= n
;
1034 return &s1
->sym_attrs
[index
];
1037 static void build_got(TCCState
*s1
)
1041 /* if no got, then create it */
1042 s1
->got
= new_section(s1
, ".got", SHT_PROGBITS
, SHF_ALLOC
| SHF_WRITE
);
1043 s1
->got
->sh_entsize
= 4;
1044 add_elf_sym(symtab_section
, 0, 4, ELFW(ST_INFO
)(STB_GLOBAL
, STT_OBJECT
),
1045 0, s1
->got
->sh_num
, "_GLOBAL_OFFSET_TABLE_");
1046 ptr
= section_ptr_add(s1
->got
, 3 * PTR_SIZE
);
1048 /* keep space for _DYNAMIC pointer, if present */
1050 /* two dummy got entries */
1051 write32le(ptr
+ 4, 0);
1052 write32le(ptr
+ 8, 0);
1054 /* keep space for _DYNAMIC pointer, if present */
1056 write32le(ptr
+ 4, 0);
1057 /* two dummy got entries */
1058 write32le(ptr
+ 8, 0);
1059 write32le(ptr
+ 12, 0);
1060 write32le(ptr
+ 16, 0);
1061 write32le(ptr
+ 20, 0);
1065 /* put a got or plt entry corresponding to a symbol in symtab_section. 'size'
1066 and 'info' can be modifed if more precise info comes from the DLL.
1067 Returns offset of GOT or PLT slot. */
1068 static unsigned long put_got_entry(TCCState
*s1
,
1069 int reloc_type
, unsigned long size
, int info
,
1072 int index
, need_plt_entry
;
1075 unsigned long offset
;
1077 struct sym_attr
*symattr
;
1083 #ifdef TCC_TARGET_X86_64
1084 (reloc_type
== R_X86_64_JUMP_SLOT
);
1085 #elif defined(TCC_TARGET_I386)
1086 (reloc_type
== R_386_JMP_SLOT
);
1087 #elif defined(TCC_TARGET_ARM)
1088 (reloc_type
== R_ARM_JUMP_SLOT
);
1089 #elif defined(TCC_TARGET_ARM64)
1090 (reloc_type
== R_AARCH64_JUMP_SLOT
);
1095 if (need_plt_entry
&& !s1
->plt
) {
1097 s1
->plt
= new_section(s1
, ".plt", SHT_PROGBITS
,
1098 SHF_ALLOC
| SHF_EXECINSTR
);
1099 s1
->plt
->sh_entsize
= 4;
1102 /* If a got/plt entry already exists for that symbol, no need to add one */
1103 if (sym_index
< s1
->nb_sym_attrs
) {
1104 if (need_plt_entry
&& s1
->sym_attrs
[sym_index
].plt_offset
)
1105 return s1
->sym_attrs
[sym_index
].plt_offset
;
1106 else if (!need_plt_entry
&& s1
->sym_attrs
[sym_index
].got_offset
)
1107 return s1
->sym_attrs
[sym_index
].got_offset
;
1110 symattr
= alloc_sym_attr(s1
, sym_index
);
1112 /* Only store the GOT offset if it's not generated for the PLT entry. */
1113 if (!need_plt_entry
)
1114 symattr
->got_offset
= s1
->got
->data_offset
;
1116 sym
= &((ElfW(Sym
) *)symtab_section
->data
)[sym_index
];
1117 name
= (char *) symtab_section
->link
->data
+ sym
->st_name
;
1118 offset
= sym
->st_value
;
1119 #if defined(TCC_TARGET_I386) || defined(TCC_TARGET_X86_64)
1120 if (need_plt_entry
) {
1124 unsigned long relofs
;
1126 #if defined(TCC_OUTPUT_DLL_WITH_PLT)
1129 /* if we build a DLL, we add a %ebx offset */
1130 if (s1
->output_type
== TCC_OUTPUT_DLL
)
1136 /* add a PLT entry */
1138 if (plt
->data_offset
== 0) {
1139 /* first plt entry */
1140 p
= section_ptr_add(plt
, 16);
1141 p
[0] = 0xff; /* pushl got + PTR_SIZE */
1142 p
[1] = modrm
+ 0x10;
1143 write32le(p
+ 2, PTR_SIZE
);
1144 p
[6] = 0xff; /* jmp *(got + PTR_SIZE * 2) */
1146 write32le(p
+ 8, PTR_SIZE
* 2);
1149 /* The PLT slot refers to the relocation entry it needs
1150 via offset. The reloc entry is created below, so its
1151 offset is the current data_offset. */
1152 relofs
= s1
->got
->reloc
? s1
->got
->reloc
->data_offset
: 0;
1153 symattr
->plt_offset
= plt
->data_offset
;
1154 p
= section_ptr_add(plt
, 16);
1155 p
[0] = 0xff; /* jmp *(got + x) */
1157 write32le(p
+ 2, s1
->got
->data_offset
);
1158 p
[6] = 0x68; /* push $xxx */
1159 #ifdef TCC_TARGET_X86_64
1160 /* On x86-64, the relocation is referred to by _index_. */
1161 write32le(p
+ 7, relofs
/ sizeof (ElfW_Rel
));
1163 write32le(p
+ 7, relofs
);
1165 p
[11] = 0xe9; /* jmp plt_start */
1166 write32le(p
+ 12, -(plt
->data_offset
));
1168 /* If this was an UNDEF symbol set the offset in the
1169 dynsymtab to the PLT slot, so that PC32 relocs to it
1171 if (sym
->st_shndx
== SHN_UNDEF
)
1172 offset
= plt
->data_offset
- 16;
1174 #elif defined(TCC_TARGET_ARM)
1175 if (need_plt_entry
) {
1179 /* if we build a DLL, we add a %ebx offset */
1180 if (s1
->output_type
== TCC_OUTPUT_DLL
)
1181 tcc_error("DLLs unimplemented!");
1183 /* add a PLT entry */
1185 if (plt
->data_offset
== 0) {
1186 /* first plt entry */
1187 p
= section_ptr_add(plt
, 16);
1188 write32le(p
, 0xe52de004); /* push {lr} */
1189 write32le(p
+4, 0xe59fe010); /* ldr lr, [pc, #16] */
1190 write32le(p
+8, 0xe08fe00e); /* add lr, pc, lr */
1191 write32le(p
+12, 0xe5bef008); /* ldr pc, [lr, #8]! */
1194 symattr
->plt_offset
= plt
->data_offset
;
1195 if (symattr
->plt_thumb_stub
) {
1196 p
= section_ptr_add(plt
, 20);
1197 write32le(p
, 0x4778); /* bx pc */
1198 write32le(p
+2, 0x46c0); /* nop */
1201 p
= section_ptr_add(plt
, 16);
1202 write32le(p
, 0xe59fc004); /* ldr ip, [pc, #4] ; GOT entry offset */
1203 write32le(p
+4, 0xe08fc00c); /* add ip, pc, ip ; addr of GOT entry */
1204 write32le(p
+8, 0xe59cf000); /* ldr pc, [ip] ; jump to GOT entry */
1205 write32le(p
+12, s1
->got
->data_offset
); /* GOT entry off once patched */
1207 /* the symbol is modified so that it will be relocated to
1209 if (sym
->st_shndx
== SHN_UNDEF
)
1210 offset
= plt
->data_offset
- 16;
1212 #elif defined(TCC_TARGET_ARM64)
1213 if (need_plt_entry
) {
1217 if (s1
->output_type
== TCC_OUTPUT_DLL
)
1218 tcc_error("DLLs unimplemented!");
1221 if (plt
->data_offset
== 0)
1222 section_ptr_add(plt
, 32);
1223 symattr
->plt_offset
= plt
->data_offset
;
1224 p
= section_ptr_add(plt
, 16);
1225 write32le(p
, s1
->got
->data_offset
);
1226 write32le(p
+ 4, (uint64_t)s1
->got
->data_offset
>> 32);
1228 if (sym
->st_shndx
== SHN_UNDEF
)
1229 offset
= plt
->data_offset
- 16;
1231 #elif defined(TCC_TARGET_C67)
1233 tcc_error("C67 got not implemented");
1236 #error unsupported CPU
1239 /* XXX This might generate multiple syms for name. */
1240 index
= put_elf_sym(s1
->dynsym
, offset
,
1241 size
, info
, 0, sym
->st_shndx
, name
);
1242 /* Create the relocation (it's against the GOT for PLT
1244 put_elf_reloc(s1
->dynsym
, s1
->got
,
1245 s1
->got
->data_offset
,
1248 /* Without .dynsym (i.e. static link or memory output) we
1249 still need relocs against the generated got, so as to fill
1250 the entries with the symbol values (determined later). */
1251 put_elf_reloc(symtab_section
, s1
->got
,
1252 s1
->got
->data_offset
,
1253 reloc_type
, sym_index
);
1255 /* And now create the GOT slot itself. */
1256 ptr
= section_ptr_add(s1
->got
, PTR_SIZE
);
1259 return symattr
->plt_offset
;
1261 return symattr
->got_offset
;
1264 /* build GOT and PLT entries */
1265 ST_FUNC
void build_got_entries(TCCState
*s1
)
1270 int i
, type
, reloc_type
, sym_index
;
1272 for(i
= 1; i
< s1
->nb_sections
; i
++) {
1273 s
= s1
->sections
[i
];
1274 if (s
->sh_type
!= SHT_RELX
)
1276 /* no need to handle got relocations */
1277 if (s
->link
!= symtab_section
)
1279 for_each_elem(s
, 0, rel
, ElfW_Rel
) {
1280 type
= ELFW(R_TYPE
)(rel
->r_info
);
1282 #if defined(TCC_TARGET_I386)
1289 if (type
== R_386_GOT32
|| type
== R_386_PLT32
) {
1290 sym_index
= ELFW(R_SYM
)(rel
->r_info
);
1291 sym
= &((ElfW(Sym
) *)symtab_section
->data
)[sym_index
];
1292 /* look at the symbol got offset. If none, then add one */
1293 if (type
== R_386_GOT32
)
1294 reloc_type
= R_386_GLOB_DAT
;
1296 reloc_type
= R_386_JMP_SLOT
;
1297 put_got_entry(s1
, reloc_type
, sym
->st_size
, sym
->st_info
,
1301 #elif defined(TCC_TARGET_ARM)
1311 sym_index
= ELFW(R_SYM
)(rel
->r_info
);
1312 sym
= &((ElfW(Sym
) *)symtab_section
->data
)[sym_index
];
1313 if (type
!= R_ARM_GOTOFF
&& type
!= R_ARM_GOTPC
1314 && sym
->st_shndx
== SHN_UNDEF
) {
1316 /* look at the symbol got offset. If none, then add one */
1317 if (type
== R_ARM_GOT32
)
1318 reloc_type
= R_ARM_GLOB_DAT
;
1320 reloc_type
= R_ARM_JUMP_SLOT
;
1321 ofs
= put_got_entry(s1
, reloc_type
, sym
->st_size
,
1322 sym
->st_info
, sym_index
);
1324 printf ("maybegot: %s, %d, %d --> ofs=0x%x\n",
1325 (char *) symtab_section
->link
->data
+ sym
->st_name
,
1326 type
, sym
->st_shndx
, ofs
);
1328 if (type
!= R_ARM_GOT32
) {
1329 addr_t
*ptr
= (addr_t
*)(s1
->sections
[s
->sh_info
]->data
1331 /* x must be signed! */
1332 int x
= *ptr
& 0xffffff;
1338 printf ("insn=0x%x --> 0x%x (x==0x%x)\n", *ptr
,
1339 (*ptr
& 0xff000000) | x
, x
);
1341 *ptr
= (*ptr
& 0xff000000) | x
;
1345 case R_ARM_THM_JUMP24
:
1346 sym_index
= ELFW(R_SYM
)(rel
->r_info
);
1347 sym
= &((ElfW(Sym
) *)symtab_section
->data
)[sym_index
];
1348 /* We are relocating a jump from thumb code to arm code */
1349 if (sym
->st_shndx
!= SHN_UNDEF
&& !(sym
->st_value
& 1)) {
1352 char *name
, buf
[1024];
1353 Section
*text_section
;
1355 name
= (char *) symtab_section
->link
->data
+ sym
->st_name
;
1356 text_section
= s1
->sections
[sym
->st_shndx
];
1357 /* Modify reloc to target a thumb stub to switch to ARM */
1358 snprintf(buf
, sizeof(buf
), "%s_from_thumb", name
);
1359 index
= put_elf_sym(symtab_section
,
1360 text_section
->data_offset
+ 1,
1361 sym
->st_size
, sym
->st_info
, 0,
1362 sym
->st_shndx
, buf
);
1363 rel
->r_info
= ELFW(R_INFO
)(index
, type
);
1364 /* Create a thumb stub fonction to switch to ARM mode */
1365 put_elf_reloc(symtab_section
, text_section
,
1366 text_section
->data_offset
+ 4, R_ARM_JUMP24
,
1368 p
= section_ptr_add(text_section
, 8);
1369 write32le(p
, 0x4778); /* bx pc */
1370 write32le(p
+2, 0x46c0); /* nop */
1371 write32le(p
+4, 0xeafffffe); /* b $sym */
1373 #elif defined(TCC_TARGET_ARM64)
1374 //xx Other cases may be required here:
1375 case R_AARCH64_ADR_GOT_PAGE
:
1376 case R_AARCH64_LD64_GOT_LO12_NC
:
1379 sym_index
= ELFW(R_SYM
)(rel
->r_info
);
1380 sym
= &((ElfW(Sym
) *)symtab_section
->data
)[sym_index
];
1381 reloc_type
= R_AARCH64_GLOB_DAT
;
1382 put_got_entry(s1
, reloc_type
, sym
->st_size
, sym
->st_info
,
1386 case R_AARCH64_JUMP26
:
1387 case R_AARCH64_CALL26
:
1390 sym_index
= ELFW(R_SYM
)(rel
->r_info
);
1391 sym
= &((ElfW(Sym
) *)symtab_section
->data
)[sym_index
];
1392 if (sym
->st_shndx
== SHN_UNDEF
) {
1394 reloc_type
= R_AARCH64_JUMP_SLOT
;
1395 ofs
= put_got_entry(s1
, reloc_type
, sym
->st_size
,
1396 sym
->st_info
, sym_index
);
1397 /* We store the place of the generated PLT slot
1399 rel
->r_addend
+= ofs
;
1402 #elif defined(TCC_TARGET_C67)
1409 if (type
== R_C60_GOT32
|| type
== R_C60_PLT32
) {
1410 sym_index
= ELFW(R_SYM
)(rel
->r_info
);
1411 sym
= &((ElfW(Sym
) *)symtab_section
->data
)[sym_index
];
1412 /* look at the symbol got offset. If none, then add one */
1413 if (type
== R_C60_GOT32
)
1414 reloc_type
= R_C60_GLOB_DAT
;
1416 reloc_type
= R_C60_JMP_SLOT
;
1417 put_got_entry(s1
, reloc_type
, sym
->st_size
, sym
->st_info
,
1421 #elif defined(TCC_TARGET_X86_64)
1422 case R_X86_64_GOT32
:
1423 case R_X86_64_GOTTPOFF
:
1424 case R_X86_64_GOTPCREL
:
1425 case R_X86_64_PLT32
:
1426 sym_index
= ELFW(R_SYM
)(rel
->r_info
);
1427 sym
= &((ElfW(Sym
) *)symtab_section
->data
)[sym_index
];
1428 if (type
== R_X86_64_PLT32
&&
1429 ELFW(ST_VISIBILITY
)(sym
->st_other
) != STV_DEFAULT
)
1431 rel
->r_info
= ELFW(R_INFO
)(sym_index
, R_X86_64_PC32
);
1437 sym
= &((ElfW(Sym
) *)symtab_section
->data
)[sym_index
];
1439 if (type
== R_X86_64_GOT32
|| type
== R_X86_64_GOTPCREL
||
1440 type
== R_X86_64_PLT32
) {
1442 /* look at the symbol got offset. If none, then add one */
1443 if (type
== R_X86_64_GOT32
|| type
== R_X86_64_GOTPCREL
)
1444 reloc_type
= R_X86_64_GLOB_DAT
;
1446 reloc_type
= R_X86_64_JUMP_SLOT
;
1447 ofs
= put_got_entry(s1
, reloc_type
, sym
->st_size
,
1448 sym
->st_info
, sym_index
);
1449 if (type
== R_X86_64_PLT32
)
1450 /* We store the place of the generated PLT slot
1452 rel
->r_addend
+= ofs
;
1456 #error unsupported CPU
1465 ST_FUNC Section
*new_symtab(TCCState
*s1
,
1466 const char *symtab_name
, int sh_type
, int sh_flags
,
1467 const char *strtab_name
,
1468 const char *hash_name
, int hash_sh_flags
)
1470 Section
*symtab
, *strtab
, *hash
;
1471 int *ptr
, nb_buckets
;
1473 symtab
= new_section(s1
, symtab_name
, sh_type
, sh_flags
);
1474 symtab
->sh_entsize
= sizeof(ElfW(Sym
));
1475 strtab
= new_section(s1
, strtab_name
, SHT_STRTAB
, sh_flags
);
1476 put_elf_str(strtab
, "");
1477 symtab
->link
= strtab
;
1478 put_elf_sym(symtab
, 0, 0, 0, 0, 0, NULL
);
1482 hash
= new_section(s1
, hash_name
, SHT_HASH
, hash_sh_flags
);
1483 hash
->sh_entsize
= sizeof(int);
1484 symtab
->hash
= hash
;
1485 hash
->link
= symtab
;
1487 ptr
= section_ptr_add(hash
, (2 + nb_buckets
+ 1) * sizeof(int));
1488 ptr
[0] = nb_buckets
;
1490 memset(ptr
+ 2, 0, (nb_buckets
+ 1) * sizeof(int));
1494 /* put dynamic tag */
1495 static void put_dt(Section
*dynamic
, int dt
, addr_t val
)
1498 dyn
= section_ptr_add(dynamic
, sizeof(ElfW(Dyn
)));
1500 dyn
->d_un
.d_val
= val
;
1503 static void add_init_array_defines(TCCState
*s1
, const char *section_name
)
1507 char sym_start
[1024];
1510 snprintf(sym_start
, sizeof(sym_start
), "__%s_start", section_name
+ 1);
1511 snprintf(sym_end
, sizeof(sym_end
), "__%s_end", section_name
+ 1);
1513 s
= find_section(s1
, section_name
);
1518 end_offset
= s
->data_offset
;
1521 add_elf_sym(symtab_section
,
1523 ELFW(ST_INFO
)(STB_GLOBAL
, STT_NOTYPE
), 0,
1524 s
->sh_num
, sym_start
);
1525 add_elf_sym(symtab_section
,
1527 ELFW(ST_INFO
)(STB_GLOBAL
, STT_NOTYPE
), 0,
1528 s
->sh_num
, sym_end
);
1531 static int tcc_add_support(TCCState
*s1
, const char *filename
)
1534 snprintf(buf
, sizeof(buf
), "%s/%s/%s", s1
->tcc_lib_path
,
1535 /* an cpu specific path inside tcc_lib_path, mainly for keeping libtcc1.a */
1536 #ifdef TCC_TARGET_I386
1539 #ifdef TCC_TARGET_X86_64
1542 #ifdef TCC_TARGET_ARM
1545 #ifdef TCC_TARGET_ARM64
1548 #ifdef TCC_TARGET_C67
1553 return tcc_add_file(s1
, buf
, TCC_FILETYPE_BINARY
);
1556 ST_FUNC
void tcc_add_bcheck(TCCState
*s1
)
1558 #ifdef CONFIG_TCC_BCHECK
1561 if (0 == s1
->do_bounds_check
)
1564 /* XXX: add an object file to do that */
1565 ptr
= section_ptr_add(bounds_section
, sizeof(*ptr
));
1567 add_elf_sym(symtab_section
, 0, 0,
1568 ELFW(ST_INFO
)(STB_GLOBAL
, STT_NOTYPE
), 0,
1569 bounds_section
->sh_num
, "__bounds_start");
1570 if (s1
->output_type
!= TCC_OUTPUT_MEMORY
) {
1571 /* add 'call __bound_init()' in .init section */
1573 /* XXX not called on MSYS, reason is unknown. For this
1574 case a call to __bound_init is performed in bcheck.c
1575 when __bound_ptr_add, __bound_new_region,
1576 __bound_delete_region called */
1578 int sym_index
= find_elf_sym(symtab_section
, "__bound_init");
1580 Section
*init_section
= find_section(s1
, ".init");
1581 unsigned char *pinit
= section_ptr_add(init_section
, 5);
1583 write32le(pinit
+ 1, -4);
1584 put_elf_reloc(symtab_section
, init_section
,
1585 init_section
->data_offset
- 4, R_386_PC32
, sym_index
);
1588 tcc_warning("__bound_init not defined");
1593 /* add tcc runtime libraries */
1594 ST_FUNC
void tcc_add_runtime(TCCState
*s1
)
1596 tcc_add_pragma_libs(s1
);
1599 if (!s1
->nostdlib
) {
1600 tcc_add_library(s1
, "c");
1601 #ifdef CONFIG_USE_LIBGCC
1602 if (!s1
->static_link
) {
1603 tcc_add_file(s1
, TCC_LIBGCC
, TCC_FILETYPE_BINARY
);
1606 tcc_add_support(s1
, "libtcc1.a");
1609 /* tcc_add_bcheck tries to relocate a call to __bound_init in _init so
1610 libtcc1.a must be loaded before for __bound_init to be defined and
1611 crtn.o must be loaded after to not finalize _init too early. */
1614 if (!s1
->nostdlib
) {
1615 /* add crt end if not memory output */
1616 if (s1
->output_type
!= TCC_OUTPUT_MEMORY
)
1617 tcc_add_crt(s1
, "crtn.o");
1621 /* add various standard linker symbols (must be done after the
1622 sections are filled (for example after allocating common
1624 ST_FUNC
void tcc_add_linker_symbols(TCCState
*s1
)
1630 add_elf_sym(symtab_section
,
1631 text_section
->data_offset
, 0,
1632 ELFW(ST_INFO
)(STB_GLOBAL
, STT_NOTYPE
), 0,
1633 text_section
->sh_num
, "_etext");
1634 add_elf_sym(symtab_section
,
1635 data_section
->data_offset
, 0,
1636 ELFW(ST_INFO
)(STB_GLOBAL
, STT_NOTYPE
), 0,
1637 data_section
->sh_num
, "_edata");
1638 add_elf_sym(symtab_section
,
1639 bss_section
->data_offset
, 0,
1640 ELFW(ST_INFO
)(STB_GLOBAL
, STT_NOTYPE
), 0,
1641 bss_section
->sh_num
, "_end");
1642 /* horrible new standard ldscript defines */
1643 add_init_array_defines(s1
, ".preinit_array");
1644 add_init_array_defines(s1
, ".init_array");
1645 add_init_array_defines(s1
, ".fini_array");
1647 /* add start and stop symbols for sections whose name can be
1649 for(i
= 1; i
< s1
->nb_sections
; i
++) {
1650 s
= s1
->sections
[i
];
1651 if (s
->sh_type
== SHT_PROGBITS
&&
1652 (s
->sh_flags
& SHF_ALLOC
)) {
1656 /* check if section name can be expressed in C */
1662 if (!isid(ch
) && !isnum(ch
))
1666 snprintf(buf
, sizeof(buf
), "__start_%s", s
->name
);
1667 add_elf_sym(symtab_section
,
1669 ELFW(ST_INFO
)(STB_GLOBAL
, STT_NOTYPE
), 0,
1671 snprintf(buf
, sizeof(buf
), "__stop_%s", s
->name
);
1672 add_elf_sym(symtab_section
,
1674 ELFW(ST_INFO
)(STB_GLOBAL
, STT_NOTYPE
), 0,
1681 static void tcc_output_binary(TCCState
*s1
, FILE *f
,
1682 const int *sec_order
)
1685 int i
, offset
, size
;
1688 for(i
=1;i
<s1
->nb_sections
;i
++) {
1689 s
= s1
->sections
[sec_order
[i
]];
1690 if (s
->sh_type
!= SHT_NOBITS
&&
1691 (s
->sh_flags
& SHF_ALLOC
)) {
1692 while (offset
< s
->sh_offset
) {
1697 fwrite(s
->data
, 1, size
, f
);
1703 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
1705 #define EXTRA_RELITEMS 14
1707 /* move the relocation value from .dynsym to .got */
1708 void patch_dynsym_undef(TCCState
*s1
, Section
*s
)
1710 uint32_t *gotd
= (void *)s1
->got
->data
;
1713 gotd
+= 3; /* dummy entries in .got */
1714 /* relocate symbols in .dynsym */
1715 for_each_elem(s
, 1, sym
, ElfW(Sym
)) {
1716 if (sym
->st_shndx
== SHN_UNDEF
) {
1717 *gotd
++ = sym
->st_value
+ 6; /* XXX 6 is magic ? */
1724 #define EXTRA_RELITEMS 9
1726 /* zero plt offsets of weak symbols in .dynsym */
1727 void patch_dynsym_undef(TCCState
*s1
, Section
*s
)
1731 for_each_elem(s
, 1, sym
, ElfW(Sym
))
1732 if (sym
->st_shndx
== SHN_UNDEF
&& ELFW(ST_BIND
)(sym
->st_info
) == STB_WEAK
)
1737 ST_FUNC
void fill_got_entry(TCCState
*s1
, ElfW_Rel
*rel
)
1739 int sym_index
= ELFW(R_SYM
) (rel
->r_info
);
1740 ElfW(Sym
) *sym
= &((ElfW(Sym
) *) symtab_section
->data
)[sym_index
];
1741 unsigned long offset
;
1743 if (sym_index
>= s1
->nb_sym_attrs
)
1745 offset
= s1
->sym_attrs
[sym_index
].got_offset
;
1746 section_reserve(s1
->got
, offset
+ PTR_SIZE
);
1747 #ifdef TCC_TARGET_X86_64
1748 /* only works for x86-64 */
1749 write32le(s1
->got
->data
+ offset
+ 4, sym
->st_value
>> 32);
1751 write32le(s1
->got
->data
+ offset
, sym
->st_value
& 0xffffffff);
1754 /* Perform relocation to GOT or PLT entries */
1755 ST_FUNC
void fill_got(TCCState
*s1
)
1761 for(i
= 1; i
< s1
->nb_sections
; i
++) {
1762 s
= s1
->sections
[i
];
1763 if (s
->sh_type
!= SHT_RELX
)
1765 /* no need to handle got relocations */
1766 if (s
->link
!= symtab_section
)
1768 for_each_elem(s
, 0, rel
, ElfW_Rel
) {
1769 switch (ELFW(R_TYPE
) (rel
->r_info
)) {
1770 case R_X86_64_GOT32
:
1771 case R_X86_64_GOTPCREL
:
1772 case R_X86_64_PLT32
:
1773 fill_got_entry(s1
, rel
);
1780 /* Bind symbols of executable: resolve undefined symbols from exported symbols
1781 in shared libraries and export non local defined symbols to shared libraries
1782 if -rdynamic switch was given on command line */
1783 static void bind_exe_dynsyms(TCCState
*s1
)
1786 int sym_index
, index
;
1787 ElfW(Sym
) *sym
, *esym
;
1790 /* Resolve undefined symbols from dynamic symbols. When there is a match:
1791 - if STT_FUNC or STT_GNU_IFUNC symbol -> add it in PLT
1792 - if STT_OBJECT symbol -> add it in .bss section with suitable reloc */
1793 for_each_elem(symtab_section
, 1, sym
, ElfW(Sym
)) {
1794 if (sym
->st_shndx
== SHN_UNDEF
) {
1795 name
= (char *) symtab_section
->link
->data
+ sym
->st_name
;
1796 sym_index
= find_elf_sym(s1
->dynsymtab_section
, name
);
1798 esym
= &((ElfW(Sym
) *)s1
->dynsymtab_section
->data
)[sym_index
];
1799 type
= ELFW(ST_TYPE
)(esym
->st_info
);
1800 if ((type
== STT_FUNC
) || (type
== STT_GNU_IFUNC
)) {
1801 /* Indirect functions shall have STT_FUNC type in executable
1802 * dynsym section. Indeed, a dlsym call following a lazy
1803 * resolution would pick the symbol value from the
1804 * executable dynsym entry which would contain the address
1805 * of the function wanted by the caller of dlsym instead of
1806 * the address of the function that would return that
1808 put_got_entry(s1
, R_JMP_SLOT
, esym
->st_size
,
1809 ELFW(ST_INFO
)(STB_GLOBAL
,STT_FUNC
),
1810 sym
- (ElfW(Sym
) *)symtab_section
->data
);
1811 } else if (type
== STT_OBJECT
) {
1812 unsigned long offset
;
1814 offset
= bss_section
->data_offset
;
1815 /* XXX: which alignment ? */
1816 offset
= (offset
+ 16 - 1) & -16;
1817 index
= put_elf_sym(s1
->dynsym
, offset
, esym
->st_size
,
1818 esym
->st_info
, 0, bss_section
->sh_num
,
1820 /* Ensure R_COPY works for weak symbol aliases */
1821 if (ELFW(ST_BIND
)(esym
->st_info
) == STB_WEAK
) {
1822 for_each_elem(s1
->dynsymtab_section
, 1, dynsym
, ElfW(Sym
)) {
1823 if ((dynsym
->st_value
== esym
->st_value
)
1824 && (ELFW(ST_BIND
)(dynsym
->st_info
) == STB_GLOBAL
)) {
1825 char *dynname
= (char *) s1
->dynsymtab_section
->link
->data
1827 put_elf_sym(s1
->dynsym
, offset
, dynsym
->st_size
,
1829 bss_section
->sh_num
, dynname
);
1834 put_elf_reloc(s1
->dynsym
, bss_section
,
1835 offset
, R_COPY
, index
);
1836 offset
+= esym
->st_size
;
1837 bss_section
->data_offset
= offset
;
1840 /* STB_WEAK undefined symbols are accepted */
1841 /* XXX: _fp_hw seems to be part of the ABI, so we ignore it */
1842 if (ELFW(ST_BIND
)(sym
->st_info
) == STB_WEAK
||
1843 !strcmp(name
, "_fp_hw")) {
1845 tcc_error_noabort("undefined symbol '%s'", name
);
1848 } else if (s1
->rdynamic
&& ELFW(ST_BIND
)(sym
->st_info
) != STB_LOCAL
) {
1849 /* if -rdynamic option, then export all non local symbols */
1850 name
= (char *) symtab_section
->link
->data
+ sym
->st_name
;
1851 put_elf_sym(s1
->dynsym
, sym
->st_value
, sym
->st_size
, sym
->st_info
,
1852 0, sym
->st_shndx
, name
);
1857 /* Bind symbols of libraries: export non local symbols of executable that
1858 resolve undefined symbols of shared libraries */
1859 static void bind_libs_dynsyms(TCCState
*s1
)
1863 ElfW(Sym
) *sym
, *esym
;
1865 /* now look at unresolved dynamic symbols and export
1866 corresponding symbol */
1867 for_each_elem(s1
->dynsymtab_section
, 1, esym
, ElfW(Sym
)) {
1868 name
= (char *) s1
->dynsymtab_section
->link
->data
+ esym
->st_name
;
1869 sym_index
= find_elf_sym(symtab_section
, name
);
1871 /* XXX: avoid adding a symbol if already present because of
1873 sym
= &((ElfW(Sym
) *)symtab_section
->data
)[sym_index
];
1874 if (sym
->st_shndx
!= SHN_UNDEF
)
1875 put_elf_sym(s1
->dynsym
, sym
->st_value
, sym
->st_size
,
1876 sym
->st_info
, 0, sym
->st_shndx
, name
);
1877 } else if (esym
->st_shndx
== SHN_UNDEF
) {
1878 /* weak symbols can stay undefined */
1879 if (ELFW(ST_BIND
)(esym
->st_info
) != STB_WEAK
)
1880 tcc_warning("undefined dynamic symbol '%s'", name
);
1885 /* Export all non local symbols (for shared libraries) */
1886 static void export_global_syms(TCCState
*s1
)
1888 int nb_syms
, dynindex
, index
;
1892 nb_syms
= symtab_section
->data_offset
/ sizeof(ElfW(Sym
));
1893 s1
->symtab_to_dynsym
= tcc_mallocz(sizeof(int) * nb_syms
);
1894 for_each_elem(symtab_section
, 1, sym
, ElfW(Sym
)) {
1895 if (ELFW(ST_BIND
)(sym
->st_info
) != STB_LOCAL
) {
1896 name
= (char *) symtab_section
->link
->data
+ sym
->st_name
;
1897 dynindex
= put_elf_sym(s1
->dynsym
, sym
->st_value
, sym
->st_size
,
1898 sym
->st_info
, 0, sym
->st_shndx
, name
);
1899 index
= sym
- (ElfW(Sym
) *) symtab_section
->data
;
1900 s1
->symtab_to_dynsym
[index
] = dynindex
;
1905 /* relocate the PLT: compute addresses and offsets in the PLT now that final
1906 address for PLT and GOT are known (see fill_program_header) */
1907 ST_FUNC
void relocate_plt(TCCState
*s1
)
1915 p_end
= p
+ s1
->plt
->data_offset
;
1917 #if defined(TCC_TARGET_I386)
1918 write32le(p
+ 2, read32le(p
+ 2) + s1
->got
->sh_addr
);
1919 write32le(p
+ 8, read32le(p
+ 8) + s1
->got
->sh_addr
);
1922 write32le(p
+ 2, read32le(p
+ 2) + s1
->got
->sh_addr
);
1925 #elif defined(TCC_TARGET_X86_64)
1926 int x
= s1
->got
->sh_addr
- s1
->plt
->sh_addr
- 6;
1927 write32le(p
+ 2, read32le(p
+ 2) + x
);
1928 write32le(p
+ 8, read32le(p
+ 8) + x
- 6);
1931 write32le(p
+ 2, read32le(p
+ 2) + x
+ s1
->plt
->data
- p
);
1934 #elif defined(TCC_TARGET_ARM)
1936 x
=s1
->got
->sh_addr
- s1
->plt
->sh_addr
- 12;
1939 if (read32le(p
) == 0x46c04778) /* PLT Thumb stub present */
1941 write32le(p
+ 12, x
+ read32le(p
+ 12) + s1
->plt
->data
- p
);
1944 #elif defined(TCC_TARGET_ARM64)
1945 uint64_t plt
= s1
->plt
->sh_addr
;
1946 uint64_t got
= s1
->got
->sh_addr
;
1947 uint64_t off
= (got
>> 12) - (plt
>> 12);
1948 if ((off
+ ((uint32_t)1 << 20)) >> 21)
1949 tcc_error("Failed relocating PLT (off=0x%lx, got=0x%lx, plt=0x%lx)", off
, got
, plt
);
1950 write32le(p
, 0xa9bf7bf0); // stp x16,x30,[sp,#-16]!
1951 write32le(p
+ 4, (0x90000010 | // adrp x16,...
1952 (off
& 0x1ffffc) << 3 | (off
& 3) << 29));
1953 write32le(p
+ 8, (0xf9400211 | // ldr x17,[x16,#...]
1954 (got
& 0xff8) << 7));
1955 write32le(p
+ 12, (0x91000210 | // add x16,x16,#...
1956 (got
& 0xfff) << 10));
1957 write32le(p
+ 16, 0xd61f0220); // br x17
1958 write32le(p
+ 20, 0xd503201f); // nop
1959 write32le(p
+ 24, 0xd503201f); // nop
1960 write32le(p
+ 28, 0xd503201f); // nop
1963 uint64_t pc
= plt
+ (p
- s1
->plt
->data
);
1964 uint64_t addr
= got
+ read64le(p
);
1965 uint64_t off
= (addr
>> 12) - (pc
>> 12);
1966 if ((off
+ ((uint32_t)1 << 20)) >> 21)
1967 tcc_error("Failed relocating PLT (off=0x%lx, addr=0x%lx, pc=0x%lx)", off
, addr
, pc
);
1968 write32le(p
, (0x90000010 | // adrp x16,...
1969 (off
& 0x1ffffc) << 3 | (off
& 3) << 29));
1970 write32le(p
+ 4, (0xf9400211 | // ldr x17,[x16,#...]
1971 (addr
& 0xff8) << 7));
1972 write32le(p
+ 8, (0x91000210 | // add x16,x16,#...
1973 (addr
& 0xfff) << 10));
1974 write32le(p
+ 12, 0xd61f0220); // br x17
1977 #elif defined(TCC_TARGET_C67)
1980 #error unsupported CPU
1985 /* Allocate strings for section names and decide if an unallocated section
1988 NOTE: the strsec section comes last, so its size is also correct ! */
1989 static void alloc_sec_names(TCCState
*s1
, int file_type
, Section
*strsec
)
1994 /* Allocate strings for section names */
1995 for(i
= 1; i
< s1
->nb_sections
; i
++) {
1996 s
= s1
->sections
[i
];
1997 s
->sh_name
= put_elf_str(strsec
, s
->name
);
1998 /* when generating a DLL, we include relocations but we may
2000 if (file_type
== TCC_OUTPUT_DLL
&&
2001 s
->sh_type
== SHT_RELX
&&
2002 !(s
->sh_flags
& SHF_ALLOC
)) {
2003 /* gr: avoid bogus relocs for empty (debug) sections */
2004 if (s1
->sections
[s
->sh_info
]->sh_flags
& SHF_ALLOC
)
2005 prepare_dynamic_rel(s1
, s
);
2006 else if (s1
->do_debug
)
2007 s
->sh_size
= s
->data_offset
;
2008 } else if (s1
->do_debug
||
2009 file_type
== TCC_OUTPUT_OBJ
||
2010 file_type
== TCC_OUTPUT_EXE
||
2011 (s
->sh_flags
& SHF_ALLOC
) ||
2012 i
== (s1
->nb_sections
- 1)) {
2013 /* we output all sections if debug or object file */
2014 s
->sh_size
= s
->data_offset
;
2019 /* Info to be copied in dynamic section */
2023 unsigned long dyn_rel_off
;
2026 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
2032 /* Assign sections to segments and decide how are sections laid out when loaded
2033 in memory. This function also fills corresponding program headers. */
2034 static int layout_sections(TCCState
*s1
, ElfW(Phdr
) *phdr
, int phnum
,
2035 Section
*interp
, Section
* strsec
,
2036 struct dyn_inf
*dyninf
, int *sec_order
)
2038 int i
, j
, k
, file_type
, sh_order_index
, file_offset
;
2039 unsigned long s_align
;
2045 file_type
= s1
->output_type
;
2048 if (s1
->output_format
== TCC_OUTPUT_FORMAT_ELF
)
2049 file_offset
= sizeof(ElfW(Ehdr
)) + phnum
* sizeof(ElfW(Phdr
));
2050 s_align
= ELF_PAGE_SIZE
;
2051 if (s1
->section_align
)
2052 s_align
= s1
->section_align
;
2055 if (s1
->has_text_addr
) {
2056 int a_offset
, p_offset
;
2057 addr
= s1
->text_addr
;
2058 /* we ensure that (addr % ELF_PAGE_SIZE) == file_offset %
2060 a_offset
= (int) (addr
& (s_align
- 1));
2061 p_offset
= file_offset
& (s_align
- 1);
2062 if (a_offset
< p_offset
)
2063 a_offset
+= s_align
;
2064 file_offset
+= (a_offset
- p_offset
);
2066 if (file_type
== TCC_OUTPUT_DLL
)
2069 addr
= ELF_START_ADDR
;
2070 /* compute address after headers */
2071 addr
+= (file_offset
& (s_align
- 1));
2075 /* Leave one program headers for the program interpreter and one for
2076 the program header table itself if needed. These are done later as
2077 they require section layout to be done first. */
2079 ph
+= 1 + HAVE_PHDR
;
2081 /* dynamic relocation table information, for .dynamic section */
2082 dyninf
->rel_addr
= dyninf
->rel_size
= 0;
2083 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
2084 dyninf
->bss_addr
= dyninf
->bss_size
= 0;
2087 for(j
= 0; j
< 2; j
++) {
2088 ph
->p_type
= PT_LOAD
;
2090 ph
->p_flags
= PF_R
| PF_X
;
2092 ph
->p_flags
= PF_R
| PF_W
;
2093 ph
->p_align
= s_align
;
2095 /* Decide the layout of sections loaded in memory. This must
2096 be done before program headers are filled since they contain
2097 info about the layout. We do the following ordering: interp,
2098 symbol tables, relocations, progbits, nobits */
2099 /* XXX: do faster and simpler sorting */
2100 for(k
= 0; k
< 5; k
++) {
2101 for(i
= 1; i
< s1
->nb_sections
; i
++) {
2102 s
= s1
->sections
[i
];
2103 /* compute if section should be included */
2105 if ((s
->sh_flags
& (SHF_ALLOC
| SHF_WRITE
)) !=
2109 if ((s
->sh_flags
& (SHF_ALLOC
| SHF_WRITE
)) !=
2110 (SHF_ALLOC
| SHF_WRITE
))
2116 } else if (s
->sh_type
== SHT_DYNSYM
||
2117 s
->sh_type
== SHT_STRTAB
||
2118 s
->sh_type
== SHT_HASH
) {
2121 } else if (s
->sh_type
== SHT_RELX
) {
2124 } else if (s
->sh_type
== SHT_NOBITS
) {
2131 sec_order
[sh_order_index
++] = i
;
2133 /* section matches: we align it and add its size */
2135 addr
= (addr
+ s
->sh_addralign
- 1) &
2136 ~(s
->sh_addralign
- 1);
2137 file_offset
+= (int) ( addr
- tmp
);
2138 s
->sh_offset
= file_offset
;
2141 /* update program header infos */
2142 if (ph
->p_offset
== 0) {
2143 ph
->p_offset
= file_offset
;
2145 ph
->p_paddr
= ph
->p_vaddr
;
2147 /* update dynamic relocation infos */
2148 if (s
->sh_type
== SHT_RELX
) {
2149 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
2150 if (!strcmp(strsec
->data
+ s
->sh_name
, ".rel.got")) {
2151 dyninf
->rel_addr
= addr
;
2152 dyninf
->rel_size
+= s
->sh_size
; /* XXX only first rel. */
2154 if (!strcmp(strsec
->data
+ s
->sh_name
, ".rel.bss")) {
2155 dyninf
->bss_addr
= addr
;
2156 dyninf
->bss_size
= s
->sh_size
; /* XXX only first rel. */
2159 if (dyninf
->rel_size
== 0)
2160 dyninf
->rel_addr
= addr
;
2161 dyninf
->rel_size
+= s
->sh_size
;
2165 if (s
->sh_type
!= SHT_NOBITS
)
2166 file_offset
+= s
->sh_size
;
2170 /* Make the first PT_LOAD segment include the program
2171 headers itself (and the ELF header as well), it'll
2172 come out with same memory use but will make various
2173 tools like binutils strip work better. */
2174 ph
->p_offset
&= ~(ph
->p_align
- 1);
2175 ph
->p_vaddr
&= ~(ph
->p_align
- 1);
2176 ph
->p_paddr
&= ~(ph
->p_align
- 1);
2178 ph
->p_filesz
= file_offset
- ph
->p_offset
;
2179 ph
->p_memsz
= addr
- ph
->p_vaddr
;
2182 if (s1
->output_format
== TCC_OUTPUT_FORMAT_ELF
) {
2183 /* if in the middle of a page, we duplicate the page in
2184 memory so that one copy is RX and the other is RW */
2185 if ((addr
& (s_align
- 1)) != 0)
2188 addr
= (addr
+ s_align
- 1) & ~(s_align
- 1);
2189 file_offset
= (file_offset
+ s_align
- 1) & ~(s_align
- 1);
2195 /* all other sections come after */
2196 for(i
= 1; i
< s1
->nb_sections
; i
++) {
2197 s
= s1
->sections
[i
];
2198 if (phnum
> 0 && (s
->sh_flags
& SHF_ALLOC
))
2200 sec_order
[sh_order_index
++] = i
;
2202 file_offset
= (file_offset
+ s
->sh_addralign
- 1) &
2203 ~(s
->sh_addralign
- 1);
2204 s
->sh_offset
= file_offset
;
2205 if (s
->sh_type
!= SHT_NOBITS
)
2206 file_offset
+= s
->sh_size
;
2212 static void fill_unloadable_phdr(ElfW(Phdr
) *phdr
, int phnum
, Section
*interp
,
2217 /* if interpreter, then add corresponding program header */
2223 int len
= phnum
* sizeof(ElfW(Phdr
));
2225 ph
->p_type
= PT_PHDR
;
2226 ph
->p_offset
= sizeof(ElfW(Ehdr
));
2227 ph
->p_vaddr
= interp
->sh_addr
- len
;
2228 ph
->p_paddr
= ph
->p_vaddr
;
2229 ph
->p_filesz
= ph
->p_memsz
= len
;
2230 ph
->p_flags
= PF_R
| PF_X
;
2231 ph
->p_align
= 4; /* interp->sh_addralign; */
2235 ph
->p_type
= PT_INTERP
;
2236 ph
->p_offset
= interp
->sh_offset
;
2237 ph
->p_vaddr
= interp
->sh_addr
;
2238 ph
->p_paddr
= ph
->p_vaddr
;
2239 ph
->p_filesz
= interp
->sh_size
;
2240 ph
->p_memsz
= interp
->sh_size
;
2242 ph
->p_align
= interp
->sh_addralign
;
2245 /* if dynamic section, then add corresponding program header */
2247 ph
= &phdr
[phnum
- 1];
2249 ph
->p_type
= PT_DYNAMIC
;
2250 ph
->p_offset
= dynamic
->sh_offset
;
2251 ph
->p_vaddr
= dynamic
->sh_addr
;
2252 ph
->p_paddr
= ph
->p_vaddr
;
2253 ph
->p_filesz
= dynamic
->sh_size
;
2254 ph
->p_memsz
= dynamic
->sh_size
;
2255 ph
->p_flags
= PF_R
| PF_W
;
2256 ph
->p_align
= dynamic
->sh_addralign
;
2260 /* Fill the dynamic section with tags describing the address and size of
2262 static void fill_dynamic(TCCState
*s1
, struct dyn_inf
*dyninf
)
2266 dynamic
= dyninf
->dynamic
;
2268 /* put dynamic section entries */
2269 dynamic
->data_offset
= dyninf
->dyn_rel_off
;
2270 put_dt(dynamic
, DT_HASH
, s1
->dynsym
->hash
->sh_addr
);
2271 put_dt(dynamic
, DT_STRTAB
, dyninf
->dynstr
->sh_addr
);
2272 put_dt(dynamic
, DT_SYMTAB
, s1
->dynsym
->sh_addr
);
2273 put_dt(dynamic
, DT_STRSZ
, dyninf
->dynstr
->data_offset
);
2274 put_dt(dynamic
, DT_SYMENT
, sizeof(ElfW(Sym
)));
2275 #if defined(TCC_TARGET_ARM64) || defined(TCC_TARGET_X86_64)
2276 put_dt(dynamic
, DT_RELA
, dyninf
->rel_addr
);
2277 put_dt(dynamic
, DT_RELASZ
, dyninf
->rel_size
);
2278 put_dt(dynamic
, DT_RELAENT
, sizeof(ElfW_Rel
));
2280 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
2281 put_dt(dynamic
, DT_PLTGOT
, s1
->got
->sh_addr
);
2282 put_dt(dynamic
, DT_PLTRELSZ
, dyninf
->rel_size
);
2283 put_dt(dynamic
, DT_JMPREL
, dyninf
->rel_addr
);
2284 put_dt(dynamic
, DT_PLTREL
, DT_REL
);
2285 put_dt(dynamic
, DT_REL
, dyninf
->bss_addr
);
2286 put_dt(dynamic
, DT_RELSZ
, dyninf
->bss_size
);
2288 put_dt(dynamic
, DT_REL
, dyninf
->rel_addr
);
2289 put_dt(dynamic
, DT_RELSZ
, dyninf
->rel_size
);
2290 put_dt(dynamic
, DT_RELENT
, sizeof(ElfW_Rel
));
2294 put_dt(dynamic
, DT_DEBUG
, 0);
2295 put_dt(dynamic
, DT_NULL
, 0);
2298 /* Relocate remaining sections and symbols (that is those not related to
2300 static int final_sections_reloc(TCCState
*s1
)
2305 relocate_syms(s1
, 0);
2307 if (s1
->nb_errors
!= 0)
2310 /* relocate sections */
2311 /* XXX: ignore sections with allocated relocations ? */
2312 for(i
= 1; i
< s1
->nb_sections
; i
++) {
2313 s
= s1
->sections
[i
];
2314 #ifdef TCC_TARGET_I386
2315 if (s
->reloc
&& s
!= s1
->got
&& (s
->sh_flags
& SHF_ALLOC
)) //gr
2316 /* On X86 gdb 7.3 works in any case but gdb 6.6 will crash if SHF_ALLOC
2317 checking is removed */
2319 if (s
->reloc
&& s
!= s1
->got
)
2320 /* On X86_64 gdb 7.3 will crash if SHF_ALLOC checking is present */
2322 relocate_section(s1
, s
);
2325 /* relocate relocation entries if the relocation tables are
2326 allocated in the executable */
2327 for(i
= 1; i
< s1
->nb_sections
; i
++) {
2328 s
= s1
->sections
[i
];
2329 if ((s
->sh_flags
& SHF_ALLOC
) &&
2330 s
->sh_type
== SHT_RELX
) {
2331 relocate_rel(s1
, s
);
2337 /* Create an ELF file on disk.
2338 This function handle ELF specific layout requirements */
2339 static void tcc_output_elf(TCCState
*s1
, FILE *f
, int phnum
, ElfW(Phdr
) *phdr
,
2340 int file_offset
, int *sec_order
)
2342 int i
, shnum
, offset
, size
, file_type
;
2345 ElfW(Shdr
) shdr
, *sh
;
2347 file_type
= s1
->output_type
;
2348 shnum
= s1
->nb_sections
;
2350 memset(&ehdr
, 0, sizeof(ehdr
));
2353 ehdr
.e_phentsize
= sizeof(ElfW(Phdr
));
2354 ehdr
.e_phnum
= phnum
;
2355 ehdr
.e_phoff
= sizeof(ElfW(Ehdr
));
2359 file_offset
= (file_offset
+ 3) & -4;
2362 ehdr
.e_ident
[0] = ELFMAG0
;
2363 ehdr
.e_ident
[1] = ELFMAG1
;
2364 ehdr
.e_ident
[2] = ELFMAG2
;
2365 ehdr
.e_ident
[3] = ELFMAG3
;
2366 ehdr
.e_ident
[4] = ELFCLASSW
;
2367 ehdr
.e_ident
[5] = ELFDATA2LSB
;
2368 ehdr
.e_ident
[6] = EV_CURRENT
;
2369 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
2370 ehdr
.e_ident
[EI_OSABI
] = ELFOSABI_FREEBSD
;
2372 #ifdef TCC_TARGET_ARM
2374 ehdr
.e_ident
[EI_OSABI
] = 0;
2375 ehdr
.e_flags
= EF_ARM_EABI_VER4
;
2376 if (file_type
== TCC_OUTPUT_EXE
|| file_type
== TCC_OUTPUT_DLL
)
2377 ehdr
.e_flags
|= EF_ARM_HASENTRY
;
2378 if (s1
->float_abi
== ARM_HARD_FLOAT
)
2379 ehdr
.e_flags
|= EF_ARM_VFP_FLOAT
;
2381 ehdr
.e_flags
|= EF_ARM_SOFT_FLOAT
;
2383 ehdr
.e_ident
[EI_OSABI
] = ELFOSABI_ARM
;
2388 case TCC_OUTPUT_EXE
:
2389 ehdr
.e_type
= ET_EXEC
;
2390 ehdr
.e_entry
= get_elf_sym_addr(s1
, "_start", 1);
2392 case TCC_OUTPUT_DLL
:
2393 ehdr
.e_type
= ET_DYN
;
2394 ehdr
.e_entry
= text_section
->sh_addr
; /* XXX: is it correct ? */
2396 case TCC_OUTPUT_OBJ
:
2397 ehdr
.e_type
= ET_REL
;
2400 ehdr
.e_machine
= EM_TCC_TARGET
;
2401 ehdr
.e_version
= EV_CURRENT
;
2402 ehdr
.e_shoff
= file_offset
;
2403 ehdr
.e_ehsize
= sizeof(ElfW(Ehdr
));
2404 ehdr
.e_shentsize
= sizeof(ElfW(Shdr
));
2405 ehdr
.e_shnum
= shnum
;
2406 ehdr
.e_shstrndx
= shnum
- 1;
2408 fwrite(&ehdr
, 1, sizeof(ElfW(Ehdr
)), f
);
2409 fwrite(phdr
, 1, phnum
* sizeof(ElfW(Phdr
)), f
);
2410 offset
= sizeof(ElfW(Ehdr
)) + phnum
* sizeof(ElfW(Phdr
));
2412 sort_syms(s1
, symtab_section
);
2413 for(i
= 1; i
< s1
->nb_sections
; i
++) {
2414 s
= s1
->sections
[sec_order
[i
]];
2415 if (s
->sh_type
!= SHT_NOBITS
) {
2416 if (s
->sh_type
== SHT_DYNSYM
)
2417 patch_dynsym_undef(s1
, s
);
2418 while (offset
< s
->sh_offset
) {
2424 fwrite(s
->data
, 1, size
, f
);
2429 /* output section headers */
2430 while (offset
< ehdr
.e_shoff
) {
2435 for(i
= 0; i
< s1
->nb_sections
; i
++) {
2437 memset(sh
, 0, sizeof(ElfW(Shdr
)));
2438 s
= s1
->sections
[i
];
2440 sh
->sh_name
= s
->sh_name
;
2441 sh
->sh_type
= s
->sh_type
;
2442 sh
->sh_flags
= s
->sh_flags
;
2443 sh
->sh_entsize
= s
->sh_entsize
;
2444 sh
->sh_info
= s
->sh_info
;
2446 sh
->sh_link
= s
->link
->sh_num
;
2447 sh
->sh_addralign
= s
->sh_addralign
;
2448 sh
->sh_addr
= s
->sh_addr
;
2449 sh
->sh_offset
= s
->sh_offset
;
2450 sh
->sh_size
= s
->sh_size
;
2452 fwrite(sh
, 1, sizeof(ElfW(Shdr
)), f
);
2456 /* Write an elf, coff or "binary" file */
2457 static int tcc_write_elf_file(TCCState
*s1
, const char *filename
, int phnum
,
2458 ElfW(Phdr
) *phdr
, int file_offset
, int *sec_order
)
2460 int fd
, mode
, file_type
;
2463 file_type
= s1
->output_type
;
2464 if (file_type
== TCC_OUTPUT_OBJ
)
2469 fd
= open(filename
, O_WRONLY
| O_CREAT
| O_TRUNC
| O_BINARY
, mode
);
2471 tcc_error_noabort("could not write '%s'", filename
);
2474 f
= fdopen(fd
, "wb");
2476 printf("<- %s\n", filename
);
2478 #ifdef TCC_TARGET_COFF
2479 if (s1
->output_format
== TCC_OUTPUT_FORMAT_COFF
)
2480 tcc_output_coff(s1
, f
);
2483 if (s1
->output_format
== TCC_OUTPUT_FORMAT_ELF
)
2484 tcc_output_elf(s1
, f
, phnum
, phdr
, file_offset
, sec_order
);
2486 tcc_output_binary(s1
, f
, sec_order
);
2492 /* Output an elf, coff or binary file */
2493 /* XXX: suppress unneeded sections */
2494 static int elf_output_file(TCCState
*s1
, const char *filename
)
2496 int i
, ret
, phnum
, shnum
, file_type
, file_offset
, *sec_order
;
2497 struct dyn_inf dyninf
;
2500 Section
*strsec
, *interp
, *dynamic
, *dynstr
;
2502 file_type
= s1
->output_type
;
2505 /* if linking, also link in runtime libraries (libc, libgcc, etc.) */
2506 if (file_type
!= TCC_OUTPUT_OBJ
) {
2507 tcc_add_runtime(s1
);
2512 interp
= dynamic
= dynstr
= NULL
; /* avoid warning */
2513 dyninf
.dyn_rel_off
= 0; /* avoid warning */
2515 if (file_type
!= TCC_OUTPUT_OBJ
) {
2516 relocate_common_syms();
2518 tcc_add_linker_symbols(s1
);
2520 if (!s1
->static_link
) {
2521 if (file_type
== TCC_OUTPUT_EXE
) {
2523 /* allow override the dynamic loader */
2524 const char *elfint
= getenv("LD_SO");
2526 elfint
= DEFAULT_ELFINTERP(s1
);
2527 /* add interpreter section only if executable */
2528 interp
= new_section(s1
, ".interp", SHT_PROGBITS
, SHF_ALLOC
);
2529 interp
->sh_addralign
= 1;
2530 ptr
= section_ptr_add(interp
, 1 + strlen(elfint
));
2531 strcpy(ptr
, elfint
);
2534 /* add dynamic symbol table */
2535 s1
->dynsym
= new_symtab(s1
, ".dynsym", SHT_DYNSYM
, SHF_ALLOC
,
2537 ".hash", SHF_ALLOC
);
2538 dynstr
= s1
->dynsym
->link
;
2540 /* add dynamic section */
2541 dynamic
= new_section(s1
, ".dynamic", SHT_DYNAMIC
,
2542 SHF_ALLOC
| SHF_WRITE
);
2543 dynamic
->link
= dynstr
;
2544 dynamic
->sh_entsize
= sizeof(ElfW(Dyn
));
2548 if (file_type
== TCC_OUTPUT_EXE
) {
2549 bind_exe_dynsyms(s1
);
2551 if (s1
->nb_errors
) {
2556 bind_libs_dynsyms(s1
);
2557 } else /* shared library case: simply export all global symbols */
2558 export_global_syms(s1
);
2560 build_got_entries(s1
);
2562 /* add a list of needed dlls */
2563 for(i
= 0; i
< s1
->nb_loaded_dlls
; i
++) {
2564 DLLReference
*dllref
= s1
->loaded_dlls
[i
];
2565 if (dllref
->level
== 0)
2566 put_dt(dynamic
, DT_NEEDED
, put_elf_str(dynstr
, dllref
->name
));
2570 put_dt(dynamic
, DT_RPATH
, put_elf_str(dynstr
, s1
->rpath
));
2572 /* XXX: currently, since we do not handle PIC code, we
2573 must relocate the readonly segments */
2574 if (file_type
== TCC_OUTPUT_DLL
) {
2576 put_dt(dynamic
, DT_SONAME
, put_elf_str(dynstr
, s1
->soname
));
2577 put_dt(dynamic
, DT_TEXTREL
, 0);
2581 put_dt(dynamic
, DT_SYMBOLIC
, 0);
2583 /* add necessary space for other entries */
2584 dyninf
.dyn_rel_off
= dynamic
->data_offset
;
2585 dynamic
->data_offset
+= sizeof(ElfW(Dyn
)) * EXTRA_RELITEMS
;
2587 /* still need to build got entries in case of static link */
2588 build_got_entries(s1
);
2592 /* we add a section for symbols */
2593 strsec
= new_section(s1
, ".shstrtab", SHT_STRTAB
, 0);
2594 put_elf_str(strsec
, "");
2596 /* compute number of sections */
2597 shnum
= s1
->nb_sections
;
2599 /* this array is used to reorder sections in the output file */
2600 sec_order
= tcc_malloc(sizeof(int) * shnum
);
2603 /* compute number of program headers */
2606 case TCC_OUTPUT_OBJ
:
2609 case TCC_OUTPUT_EXE
:
2610 if (!s1
->static_link
)
2611 phnum
= 4 + HAVE_PHDR
;
2615 case TCC_OUTPUT_DLL
:
2620 /* Allocate strings for section names */
2621 alloc_sec_names(s1
, file_type
, strsec
);
2623 /* allocate program segment headers */
2624 phdr
= tcc_mallocz(phnum
* sizeof(ElfW(Phdr
)));
2626 /* compute section to program header mapping */
2627 file_offset
= layout_sections(s1
, phdr
, phnum
, interp
, strsec
, &dyninf
,
2630 /* Fill remaining program header and finalize relocation related to dynamic
2633 fill_unloadable_phdr(phdr
, phnum
, interp
, dynamic
);
2635 dyninf
.dynamic
= dynamic
;
2636 dyninf
.dynstr
= dynstr
;
2638 fill_dynamic(s1
, &dyninf
);
2640 /* put in GOT the dynamic section address and relocate PLT */
2641 write32le(s1
->got
->data
, dynamic
->sh_addr
);
2642 if (file_type
== TCC_OUTPUT_EXE
2643 #if defined(TCC_OUTPUT_DLL_WITH_PLT)
2644 || file_type
== TCC_OUTPUT_DLL
2649 /* relocate symbols in .dynsym now that final addresses are known */
2650 for_each_elem(s1
->dynsym
, 1, sym
, ElfW(Sym
)) {
2651 if (sym
->st_shndx
== SHN_UNDEF
) {
2652 /* relocate to PLT if symbol corresponds to a PLT entry,
2653 but not if it's a weak symbol */
2654 if (ELFW(ST_BIND
)(sym
->st_info
) == STB_WEAK
)
2656 else if (sym
->st_value
)
2657 sym
->st_value
+= s1
->plt
->sh_addr
;
2658 } else if (sym
->st_shndx
< SHN_LORESERVE
) {
2659 /* do symbol relocation */
2660 sym
->st_value
+= s1
->sections
[sym
->st_shndx
]->sh_addr
;
2666 /* if building executable or DLL, then relocate each section
2667 except the GOT which is already relocated */
2668 if (file_type
!= TCC_OUTPUT_OBJ
) {
2669 ret
= final_sections_reloc(s1
);
2674 /* Perform relocation to GOT or PLT entries */
2675 if (file_type
== TCC_OUTPUT_EXE
&& s1
->static_link
)
2678 /* Create the ELF file with name 'filename' */
2679 ret
= tcc_write_elf_file(s1
, filename
, phnum
, phdr
, file_offset
, sec_order
);
2682 const char *strip_cmd
= "sstrip "; // super strip utility from ELFkickers
2683 const char *null_dev
= " 2> /dev/null";
2685 snprintf(buf
, sizeof(buf
), "%s%s%s", strip_cmd
, filename
, null_dev
);
2688 system(buf
+1); // call a strip utility from binutils
2691 tcc_free(s1
->symtab_to_dynsym
);
2692 tcc_free(sec_order
);
2694 tcc_free(s1
->sym_attrs
);
2695 s1
->sym_attrs
= NULL
;
2699 LIBTCCAPI
int tcc_output_file(TCCState
*s
, const char *filename
)
2702 #ifdef TCC_TARGET_PE
2703 if (s
->output_type
!= TCC_OUTPUT_OBJ
) {
2704 ret
= pe_output_file(s
, filename
);
2707 ret
= elf_output_file(s
, filename
);
2711 static void *load_data(int fd
, unsigned long file_offset
, unsigned long size
)
2715 data
= tcc_malloc(size
);
2716 lseek(fd
, file_offset
, SEEK_SET
);
2717 read(fd
, data
, size
);
2721 typedef struct SectionMergeInfo
{
2722 Section
*s
; /* corresponding existing section */
2723 unsigned long offset
; /* offset of the new section in the existing section */
2724 uint8_t new_section
; /* true if section 's' was added */
2725 uint8_t link_once
; /* true if link once section */
2728 /* load an object file and merge it with current files */
2729 /* XXX: handle correctly stab (debug) info */
2730 ST_FUNC
int tcc_load_object_file(TCCState
*s1
,
2731 int fd
, unsigned long file_offset
)
2734 ElfW(Shdr
) *shdr
, *sh
;
2735 int size
, i
, j
, offset
, offseti
, nb_syms
, sym_index
, ret
;
2736 unsigned char *strsec
, *strtab
;
2737 int *old_to_new_syms
;
2738 char *sh_name
, *name
;
2739 SectionMergeInfo
*sm_table
, *sm
;
2740 ElfW(Sym
) *sym
, *symtab
;
2747 stab_index
= stabstr_index
= 0;
2749 if (read(fd
, &ehdr
, sizeof(ehdr
)) != sizeof(ehdr
))
2751 if (ehdr
.e_ident
[0] != ELFMAG0
||
2752 ehdr
.e_ident
[1] != ELFMAG1
||
2753 ehdr
.e_ident
[2] != ELFMAG2
||
2754 ehdr
.e_ident
[3] != ELFMAG3
)
2756 /* test if object file */
2757 if (ehdr
.e_type
!= ET_REL
)
2759 /* test CPU specific stuff */
2760 if (ehdr
.e_ident
[5] != ELFDATA2LSB
||
2761 ehdr
.e_machine
!= EM_TCC_TARGET
) {
2763 tcc_error_noabort("invalid object file");
2767 shdr
= load_data(fd
, file_offset
+ ehdr
.e_shoff
,
2768 sizeof(ElfW(Shdr
)) * ehdr
.e_shnum
);
2769 sm_table
= tcc_mallocz(sizeof(SectionMergeInfo
) * ehdr
.e_shnum
);
2771 /* load section names */
2772 sh
= &shdr
[ehdr
.e_shstrndx
];
2773 strsec
= load_data(fd
, file_offset
+ sh
->sh_offset
, sh
->sh_size
);
2775 /* load symtab and strtab */
2776 old_to_new_syms
= NULL
;
2780 for(i
= 1; i
< ehdr
.e_shnum
; i
++) {
2782 if (sh
->sh_type
== SHT_SYMTAB
) {
2784 tcc_error_noabort("object must contain only one symtab");
2789 nb_syms
= sh
->sh_size
/ sizeof(ElfW(Sym
));
2790 symtab
= load_data(fd
, file_offset
+ sh
->sh_offset
, sh
->sh_size
);
2791 sm_table
[i
].s
= symtab_section
;
2793 /* now load strtab */
2794 sh
= &shdr
[sh
->sh_link
];
2795 strtab
= load_data(fd
, file_offset
+ sh
->sh_offset
, sh
->sh_size
);
2799 /* now examine each section and try to merge its content with the
2801 for(i
= 1; i
< ehdr
.e_shnum
; i
++) {
2802 /* no need to examine section name strtab */
2803 if (i
== ehdr
.e_shstrndx
)
2806 sh_name
= (char *) strsec
+ sh
->sh_name
;
2807 /* ignore sections types we do not handle */
2808 if (sh
->sh_type
!= SHT_PROGBITS
&&
2809 sh
->sh_type
!= SHT_RELX
&&
2811 sh
->sh_type
!= SHT_ARM_EXIDX
&&
2813 sh
->sh_type
!= SHT_NOBITS
&&
2814 sh
->sh_type
!= SHT_PREINIT_ARRAY
&&
2815 sh
->sh_type
!= SHT_INIT_ARRAY
&&
2816 sh
->sh_type
!= SHT_FINI_ARRAY
&&
2817 strcmp(sh_name
, ".stabstr")
2820 if (sh
->sh_addralign
< 1)
2821 sh
->sh_addralign
= 1;
2822 /* find corresponding section, if any */
2823 for(j
= 1; j
< s1
->nb_sections
;j
++) {
2824 s
= s1
->sections
[j
];
2825 if (!strcmp(s
->name
, sh_name
)) {
2826 if (!strncmp(sh_name
, ".gnu.linkonce",
2827 sizeof(".gnu.linkonce") - 1)) {
2828 /* if a 'linkonce' section is already present, we
2829 do not add it again. It is a little tricky as
2830 symbols can still be defined in
2832 sm_table
[i
].link_once
= 1;
2839 /* not found: create new section */
2840 s
= new_section(s1
, sh_name
, sh
->sh_type
, sh
->sh_flags
);
2841 /* take as much info as possible from the section. sh_link and
2842 sh_info will be updated later */
2843 s
->sh_addralign
= sh
->sh_addralign
;
2844 s
->sh_entsize
= sh
->sh_entsize
;
2845 sm_table
[i
].new_section
= 1;
2847 if (sh
->sh_type
!= s
->sh_type
) {
2848 tcc_error_noabort("invalid section type");
2852 /* align start of section */
2853 offset
= s
->data_offset
;
2855 if (0 == strcmp(sh_name
, ".stab")) {
2859 if (0 == strcmp(sh_name
, ".stabstr")) {
2864 size
= sh
->sh_addralign
- 1;
2865 offset
= (offset
+ size
) & ~size
;
2866 if (sh
->sh_addralign
> s
->sh_addralign
)
2867 s
->sh_addralign
= sh
->sh_addralign
;
2868 s
->data_offset
= offset
;
2870 sm_table
[i
].offset
= offset
;
2872 /* concatenate sections */
2874 if (sh
->sh_type
!= SHT_NOBITS
) {
2876 lseek(fd
, file_offset
+ sh
->sh_offset
, SEEK_SET
);
2877 ptr
= section_ptr_add(s
, size
);
2878 read(fd
, ptr
, size
);
2880 s
->data_offset
+= size
;
2885 /* gr relocate stab strings */
2886 if (stab_index
&& stabstr_index
) {
2889 s
= sm_table
[stab_index
].s
;
2890 a
= (Stab_Sym
*)(s
->data
+ sm_table
[stab_index
].offset
);
2891 b
= (Stab_Sym
*)(s
->data
+ s
->data_offset
);
2892 o
= sm_table
[stabstr_index
].offset
;
2894 a
->n_strx
+= o
, a
++;
2897 /* second short pass to update sh_link and sh_info fields of new
2899 for(i
= 1; i
< ehdr
.e_shnum
; i
++) {
2901 if (!s
|| !sm_table
[i
].new_section
)
2904 if (sh
->sh_link
> 0)
2905 s
->link
= sm_table
[sh
->sh_link
].s
;
2906 if (sh
->sh_type
== SHT_RELX
) {
2907 s
->sh_info
= sm_table
[sh
->sh_info
].s
->sh_num
;
2908 /* update backward link */
2909 s1
->sections
[s
->sh_info
]->reloc
= s
;
2914 /* resolve symbols */
2915 old_to_new_syms
= tcc_mallocz(nb_syms
* sizeof(int));
2918 for(i
= 1; i
< nb_syms
; i
++, sym
++) {
2919 if (sym
->st_shndx
!= SHN_UNDEF
&&
2920 sym
->st_shndx
< SHN_LORESERVE
) {
2921 sm
= &sm_table
[sym
->st_shndx
];
2922 if (sm
->link_once
) {
2923 /* if a symbol is in a link once section, we use the
2924 already defined symbol. It is very important to get
2925 correct relocations */
2926 if (ELFW(ST_BIND
)(sym
->st_info
) != STB_LOCAL
) {
2927 name
= (char *) strtab
+ sym
->st_name
;
2928 sym_index
= find_elf_sym(symtab_section
, name
);
2930 old_to_new_syms
[i
] = sym_index
;
2934 /* if no corresponding section added, no need to add symbol */
2937 /* convert section number */
2938 sym
->st_shndx
= sm
->s
->sh_num
;
2940 sym
->st_value
+= sm
->offset
;
2943 name
= (char *) strtab
+ sym
->st_name
;
2944 sym_index
= add_elf_sym(symtab_section
, sym
->st_value
, sym
->st_size
,
2945 sym
->st_info
, sym
->st_other
,
2946 sym
->st_shndx
, name
);
2947 old_to_new_syms
[i
] = sym_index
;
2950 /* third pass to patch relocation entries */
2951 for(i
= 1; i
< ehdr
.e_shnum
; i
++) {
2956 offset
= sm_table
[i
].offset
;
2957 switch(s
->sh_type
) {
2959 /* take relocation offset information */
2960 offseti
= sm_table
[sh
->sh_info
].offset
;
2961 for_each_elem(s
, (offset
/ sizeof(*rel
)), rel
, ElfW_Rel
) {
2964 /* convert symbol index */
2965 type
= ELFW(R_TYPE
)(rel
->r_info
);
2966 sym_index
= ELFW(R_SYM
)(rel
->r_info
);
2967 /* NOTE: only one symtab assumed */
2968 if (sym_index
>= nb_syms
)
2970 sym_index
= old_to_new_syms
[sym_index
];
2971 /* ignore link_once in rel section. */
2972 if (!sym_index
&& !sm
->link_once
2973 #ifdef TCC_TARGET_ARM
2974 && type
!= R_ARM_V4BX
2978 tcc_error_noabort("Invalid relocation entry [%2d] '%s' @ %.8x",
2979 i
, strsec
+ sh
->sh_name
, rel
->r_offset
);
2982 rel
->r_info
= ELFW(R_INFO
)(sym_index
, type
);
2983 /* offset the relocation offset */
2984 rel
->r_offset
+= offseti
;
2985 #ifdef TCC_TARGET_ARM
2986 /* Jumps and branches from a Thumb code to a PLT entry need
2987 special handling since PLT entries are ARM code.
2988 Unconditional bl instructions referencing PLT entries are
2989 handled by converting these instructions into blx
2990 instructions. Other case of instructions referencing a PLT
2991 entry require to add a Thumb stub before the PLT entry to
2992 switch to ARM mode. We set bit plt_thumb_stub of the
2993 attribute of a symbol to indicate such a case. */
2994 if (type
== R_ARM_THM_JUMP24
)
2995 alloc_sym_attr(s1
, sym_index
)->plt_thumb_stub
= 1;
3008 tcc_free(old_to_new_syms
);
3015 typedef struct ArchiveHeader
{
3016 char ar_name
[16]; /* name of this member */
3017 char ar_date
[12]; /* file mtime */
3018 char ar_uid
[6]; /* owner uid; printed as decimal */
3019 char ar_gid
[6]; /* owner gid; printed as decimal */
3020 char ar_mode
[8]; /* file mode, printed as octal */
3021 char ar_size
[10]; /* file size, printed as decimal */
3022 char ar_fmag
[2]; /* should contain ARFMAG */
3025 static int get_be32(const uint8_t *b
)
3027 return b
[3] | (b
[2] << 8) | (b
[1] << 16) | (b
[0] << 24);
3030 /* load only the objects which resolve undefined symbols */
3031 static int tcc_load_alacarte(TCCState
*s1
, int fd
, int size
)
3033 int i
, bound
, nsyms
, sym_index
, off
, ret
;
3035 const char *ar_names
, *p
;
3036 const uint8_t *ar_index
;
3039 data
= tcc_malloc(size
);
3040 if (read(fd
, data
, size
) != size
)
3042 nsyms
= get_be32(data
);
3043 ar_index
= data
+ 4;
3044 ar_names
= (char *) ar_index
+ nsyms
* 4;
3048 for(p
= ar_names
, i
= 0; i
< nsyms
; i
++, p
+= strlen(p
)+1) {
3049 sym_index
= find_elf_sym(symtab_section
, p
);
3051 sym
= &((ElfW(Sym
) *)symtab_section
->data
)[sym_index
];
3052 if(sym
->st_shndx
== SHN_UNDEF
) {
3053 off
= get_be32(ar_index
+ i
* 4) + sizeof(ArchiveHeader
);
3055 lseek(fd
, off
, SEEK_SET
);
3056 if(tcc_load_object_file(s1
, fd
, off
) < 0) {
3071 /* load a '.a' file */
3072 ST_FUNC
int tcc_load_archive(TCCState
*s1
, int fd
)
3079 unsigned long file_offset
;
3081 /* skip magic which was already checked */
3082 read(fd
, magic
, sizeof(magic
));
3085 len
= read(fd
, &hdr
, sizeof(hdr
));
3088 if (len
!= sizeof(hdr
)) {
3089 tcc_error_noabort("invalid archive");
3092 memcpy(ar_size
, hdr
.ar_size
, sizeof(hdr
.ar_size
));
3093 ar_size
[sizeof(hdr
.ar_size
)] = '\0';
3094 size
= strtol(ar_size
, NULL
, 0);
3095 memcpy(ar_name
, hdr
.ar_name
, sizeof(hdr
.ar_name
));
3096 for(i
= sizeof(hdr
.ar_name
) - 1; i
>= 0; i
--) {
3097 if (ar_name
[i
] != ' ')
3100 ar_name
[i
+ 1] = '\0';
3101 file_offset
= lseek(fd
, 0, SEEK_CUR
);
3103 size
= (size
+ 1) & ~1;
3104 if (!strcmp(ar_name
, "/")) {
3105 /* coff symbol table : we handle it */
3106 if(s1
->alacarte_link
)
3107 return tcc_load_alacarte(s1
, fd
, size
);
3108 } else if (!strcmp(ar_name
, "//") ||
3109 !strcmp(ar_name
, "__.SYMDEF") ||
3110 !strcmp(ar_name
, "__.SYMDEF/") ||
3111 !strcmp(ar_name
, "ARFILENAMES/")) {
3112 /* skip symbol table or archive names */
3114 if (tcc_load_object_file(s1
, fd
, file_offset
) < 0)
3117 lseek(fd
, file_offset
+ size
, SEEK_SET
);
3122 #ifndef TCC_TARGET_PE
3123 /* load a DLL and all referenced DLLs. 'level = 0' means that the DLL
3124 is referenced by the user (so it should be added as DT_NEEDED in
3125 the generated ELF file) */
3126 ST_FUNC
int tcc_load_dll(TCCState
*s1
, int fd
, const char *filename
, int level
)
3129 ElfW(Shdr
) *shdr
, *sh
, *sh1
;
3130 int i
, j
, nb_syms
, nb_dts
, sym_bind
, ret
;
3131 ElfW(Sym
) *sym
, *dynsym
;
3132 ElfW(Dyn
) *dt
, *dynamic
;
3133 unsigned char *dynstr
;
3134 const char *name
, *soname
;
3135 DLLReference
*dllref
;
3137 read(fd
, &ehdr
, sizeof(ehdr
));
3139 /* test CPU specific stuff */
3140 if (ehdr
.e_ident
[5] != ELFDATA2LSB
||
3141 ehdr
.e_machine
!= EM_TCC_TARGET
) {
3142 tcc_error_noabort("bad architecture");
3147 shdr
= load_data(fd
, ehdr
.e_shoff
, sizeof(ElfW(Shdr
)) * ehdr
.e_shnum
);
3149 /* load dynamic section and dynamic symbols */
3153 dynsym
= NULL
; /* avoid warning */
3154 dynstr
= NULL
; /* avoid warning */
3155 for(i
= 0, sh
= shdr
; i
< ehdr
.e_shnum
; i
++, sh
++) {
3156 switch(sh
->sh_type
) {
3158 nb_dts
= sh
->sh_size
/ sizeof(ElfW(Dyn
));
3159 dynamic
= load_data(fd
, sh
->sh_offset
, sh
->sh_size
);
3162 nb_syms
= sh
->sh_size
/ sizeof(ElfW(Sym
));
3163 dynsym
= load_data(fd
, sh
->sh_offset
, sh
->sh_size
);
3164 sh1
= &shdr
[sh
->sh_link
];
3165 dynstr
= load_data(fd
, sh1
->sh_offset
, sh1
->sh_size
);
3172 /* compute the real library name */
3173 soname
= tcc_basename(filename
);
3175 for(i
= 0, dt
= dynamic
; i
< nb_dts
; i
++, dt
++) {
3176 if (dt
->d_tag
== DT_SONAME
) {
3177 soname
= (char *) dynstr
+ dt
->d_un
.d_val
;
3181 /* if the dll is already loaded, do not load it */
3182 for(i
= 0; i
< s1
->nb_loaded_dlls
; i
++) {
3183 dllref
= s1
->loaded_dlls
[i
];
3184 if (!strcmp(soname
, dllref
->name
)) {
3185 /* but update level if needed */
3186 if (level
< dllref
->level
)
3187 dllref
->level
= level
;
3193 /* add the dll and its level */
3194 dllref
= tcc_mallocz(sizeof(DLLReference
) + strlen(soname
));
3195 dllref
->level
= level
;
3196 strcpy(dllref
->name
, soname
);
3197 dynarray_add((void ***)&s1
->loaded_dlls
, &s1
->nb_loaded_dlls
, dllref
);
3199 /* add dynamic symbols in dynsym_section */
3200 for(i
= 1, sym
= dynsym
+ 1; i
< nb_syms
; i
++, sym
++) {
3201 sym_bind
= ELFW(ST_BIND
)(sym
->st_info
);
3202 if (sym_bind
== STB_LOCAL
)
3204 name
= (char *) dynstr
+ sym
->st_name
;
3205 add_elf_sym(s1
->dynsymtab_section
, sym
->st_value
, sym
->st_size
,
3206 sym
->st_info
, sym
->st_other
, sym
->st_shndx
, name
);
3209 /* load all referenced DLLs */
3210 for(i
= 0, dt
= dynamic
; i
< nb_dts
; i
++, dt
++) {
3213 name
= (char *) dynstr
+ dt
->d_un
.d_val
;
3214 for(j
= 0; j
< s1
->nb_loaded_dlls
; j
++) {
3215 dllref
= s1
->loaded_dlls
[j
];
3216 if (!strcmp(name
, dllref
->name
))
3217 goto already_loaded
;
3219 if (tcc_add_dll(s1
, name
, AFF_REFERENCED_DLL
) < 0) {
3220 tcc_error_noabort("referenced dll '%s' not found", name
);
3237 #define LD_TOK_NAME 256
3238 #define LD_TOK_EOF (-1)
3240 /* return next ld script token */
3241 static int ld_next(TCCState
*s1
, char *name
, int name_size
)
3259 file
->buf_ptr
= parse_comment(file
->buf_ptr
);
3260 ch
= file
->buf_ptr
[0];
3273 /* case 'a' ... 'z': */
3300 /* case 'A' ... 'z': */
3334 if (!((ch
>= 'a' && ch
<= 'z') ||
3335 (ch
>= 'A' && ch
<= 'Z') ||
3336 (ch
>= '0' && ch
<= '9') ||
3337 strchr("/.-_+=$:\\,~", ch
)))
3339 if ((q
- name
) < name_size
- 1) {
3358 static int ld_add_file(TCCState
*s1
, const char filename
[])
3362 ret
= tcc_add_file_internal(s1
, filename
, 0, TCC_FILETYPE_BINARY
);
3364 ret
= tcc_add_dll(s1
, filename
, 0);
3368 static inline int new_undef_syms(void)
3371 ret
= new_undef_sym
;
3376 static int ld_add_file_list(TCCState
*s1
, const char *cmd
, int as_needed
)
3378 char filename
[1024], libname
[1024];
3379 int t
, group
, nblibs
= 0, ret
= 0;
3382 group
= !strcmp(cmd
, "GROUP");
3385 t
= ld_next(s1
, filename
, sizeof(filename
));
3388 t
= ld_next(s1
, filename
, sizeof(filename
));
3391 if (t
== LD_TOK_EOF
) {
3392 tcc_error_noabort("unexpected end of file");
3394 goto lib_parse_error
;
3395 } else if (t
== ')') {
3397 } else if (t
== '-') {
3398 t
= ld_next(s1
, filename
, sizeof(filename
));
3399 if ((t
!= LD_TOK_NAME
) || (filename
[0] != 'l')) {
3400 tcc_error_noabort("library name expected");
3402 goto lib_parse_error
;
3404 pstrcpy(libname
, sizeof libname
, &filename
[1]);
3405 if (s1
->static_link
) {
3406 snprintf(filename
, sizeof filename
, "lib%s.a", libname
);
3408 snprintf(filename
, sizeof filename
, "lib%s.so", libname
);
3410 } else if (t
!= LD_TOK_NAME
) {
3411 tcc_error_noabort("filename expected");
3413 goto lib_parse_error
;
3415 if (!strcmp(filename
, "AS_NEEDED")) {
3416 ret
= ld_add_file_list(s1
, cmd
, 1);
3418 goto lib_parse_error
;
3420 /* TODO: Implement AS_NEEDED support. Ignore it for now */
3422 ret
= ld_add_file(s1
, filename
);
3424 goto lib_parse_error
;
3426 /* Add the filename *and* the libname to avoid future conversions */
3427 dynarray_add((void ***) &libs
, &nblibs
, tcc_strdup(filename
));
3428 if (libname
[0] != '\0')
3429 dynarray_add((void ***) &libs
, &nblibs
, tcc_strdup(libname
));
3433 t
= ld_next(s1
, filename
, sizeof(filename
));
3435 t
= ld_next(s1
, filename
, sizeof(filename
));
3438 if (group
&& !as_needed
) {
3439 while (new_undef_syms()) {
3442 for (i
= 0; i
< nblibs
; i
++)
3443 ld_add_file(s1
, libs
[i
]);
3447 dynarray_reset(&libs
, &nblibs
);
3451 /* interpret a subset of GNU ldscripts to handle the dummy libc.so
3453 ST_FUNC
int tcc_load_ldscript(TCCState
*s1
)
3456 char filename
[1024];
3461 t
= ld_next(s1
, cmd
, sizeof(cmd
));
3462 if (t
== LD_TOK_EOF
)
3464 else if (t
!= LD_TOK_NAME
)
3466 if (!strcmp(cmd
, "INPUT") ||
3467 !strcmp(cmd
, "GROUP")) {
3468 ret
= ld_add_file_list(s1
, cmd
, 0);
3471 } else if (!strcmp(cmd
, "OUTPUT_FORMAT") ||
3472 !strcmp(cmd
, "TARGET")) {
3473 /* ignore some commands */
3474 t
= ld_next(s1
, cmd
, sizeof(cmd
));
3478 t
= ld_next(s1
, filename
, sizeof(filename
));
3479 if (t
== LD_TOK_EOF
) {
3480 tcc_error_noabort("unexpected end of file");
3482 } else if (t
== ')') {
3492 #endif /* !TCC_TARGET_PE */