tccgen: x86_64: fix garbage in the SValue upper bits
[tinycc.git] / tccelf.c
blob568b40fc89948465c928e285510e8b093b47cfae
1 /*
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
21 #include "tcc.h"
23 /* XXX: avoid static variable */
24 static int new_undef_sym = 0; /* Is there a new undefined sym since last new_undef_sym() */
26 ST_FUNC int put_elf_str(Section *s, const char *sym)
28 int offset, len;
29 char *ptr;
31 len = strlen(sym) + 1;
32 offset = s->data_offset;
33 ptr = section_ptr_add(s, len);
34 memcpy(ptr, sym, len);
35 return offset;
38 /* elf symbol hashing function */
39 static unsigned long elf_hash(const unsigned char *name)
41 unsigned long h = 0, g;
43 while (*name) {
44 h = (h << 4) + *name++;
45 g = h & 0xf0000000;
46 if (g)
47 h ^= g >> 24;
48 h &= ~g;
50 return h;
53 /* rebuild hash table of section s */
54 /* NOTE: we do factorize the hash table code to go faster */
55 static void rebuild_hash(Section *s, unsigned int nb_buckets)
57 ElfW(Sym) *sym;
58 int *ptr, *hash, nb_syms, sym_index, h;
59 unsigned char *strtab;
61 strtab = s->link->data;
62 nb_syms = s->data_offset / sizeof(ElfW(Sym));
64 s->hash->data_offset = 0;
65 ptr = section_ptr_add(s->hash, (2 + nb_buckets + nb_syms) * sizeof(int));
66 ptr[0] = nb_buckets;
67 ptr[1] = nb_syms;
68 ptr += 2;
69 hash = ptr;
70 memset(hash, 0, (nb_buckets + 1) * sizeof(int));
71 ptr += nb_buckets + 1;
73 sym = (ElfW(Sym) *)s->data + 1;
74 for(sym_index = 1; sym_index < nb_syms; sym_index++) {
75 if (ELFW(ST_BIND)(sym->st_info) != STB_LOCAL) {
76 h = elf_hash(strtab + sym->st_name) % nb_buckets;
77 *ptr = hash[h];
78 hash[h] = sym_index;
79 } else {
80 *ptr = 0;
82 ptr++;
83 sym++;
87 /* return the symbol number */
88 ST_FUNC int put_elf_sym(Section *s, addr_t value, unsigned long size,
89 int info, int other, int shndx, const char *name)
91 int name_offset, sym_index;
92 int nbuckets, h;
93 ElfW(Sym) *sym;
94 Section *hs;
96 sym = section_ptr_add(s, sizeof(ElfW(Sym)));
97 if (name)
98 name_offset = put_elf_str(s->link, name);
99 else
100 name_offset = 0;
101 /* XXX: endianness */
102 sym->st_name = name_offset;
103 sym->st_value = value;
104 sym->st_size = size;
105 sym->st_info = info;
106 sym->st_other = other;
107 sym->st_shndx = shndx;
108 sym_index = sym - (ElfW(Sym) *)s->data;
109 hs = s->hash;
110 if (hs) {
111 int *ptr, *base;
112 ptr = section_ptr_add(hs, sizeof(int));
113 base = (int *)hs->data;
114 /* only add global or weak symbols */
115 if (ELFW(ST_BIND)(info) != STB_LOCAL) {
116 /* add another hashing entry */
117 nbuckets = base[0];
118 h = elf_hash((unsigned char *) name) % nbuckets;
119 *ptr = base[2 + h];
120 base[2 + h] = sym_index;
121 base[1]++;
122 /* we resize the hash table */
123 hs->nb_hashed_syms++;
124 if (hs->nb_hashed_syms > 2 * nbuckets) {
125 rebuild_hash(s, 2 * nbuckets);
127 } else {
128 *ptr = 0;
129 base[1]++;
132 return sym_index;
135 /* find global ELF symbol 'name' and return its index. Return 0 if not
136 found. */
137 ST_FUNC int find_elf_sym(Section *s, const char *name)
139 ElfW(Sym) *sym;
140 Section *hs;
141 int nbuckets, sym_index, h;
142 const char *name1;
144 hs = s->hash;
145 if (!hs)
146 return 0;
147 nbuckets = ((int *)hs->data)[0];
148 h = elf_hash((unsigned char *) name) % nbuckets;
149 sym_index = ((int *)hs->data)[2 + h];
150 while (sym_index != 0) {
151 sym = &((ElfW(Sym) *)s->data)[sym_index];
152 name1 = (char *) s->link->data + sym->st_name;
153 if (!strcmp(name, name1))
154 return sym_index;
155 sym_index = ((int *)hs->data)[2 + nbuckets + sym_index];
157 return 0;
160 /* return elf symbol value, signal error if 'err' is nonzero */
161 ST_FUNC addr_t get_elf_sym_addr(TCCState *s, const char *name, int err)
163 int sym_index;
164 ElfW(Sym) *sym;
166 sym_index = find_elf_sym(s->symtab, name);
167 sym = &((ElfW(Sym) *)s->symtab->data)[sym_index];
168 if (!sym_index || sym->st_shndx == SHN_UNDEF) {
169 if (err)
170 tcc_error("%s not defined", name);
171 return 0;
173 return sym->st_value;
176 /* return elf symbol value */
177 LIBTCCAPI void *tcc_get_symbol(TCCState *s, const char *name)
179 return (void*)(uintptr_t)get_elf_sym_addr(s, name, 0);
182 #if defined TCC_IS_NATIVE || defined TCC_TARGET_PE
183 /* return elf symbol value or error */
184 ST_FUNC void* tcc_get_symbol_err(TCCState *s, const char *name)
186 return (void*)(uintptr_t)get_elf_sym_addr(s, name, 1);
188 #endif
190 /* add an elf symbol : check if it is already defined and patch
191 it. Return symbol index. NOTE that sh_num can be SHN_UNDEF. */
192 ST_FUNC int add_elf_sym(Section *s, addr_t value, unsigned long size,
193 int info, int other, int sh_num, const char *name)
195 ElfW(Sym) *esym;
196 int sym_bind, sym_index, sym_type, esym_bind;
197 unsigned char sym_vis, esym_vis, new_vis;
199 sym_bind = ELFW(ST_BIND)(info);
200 sym_type = ELFW(ST_TYPE)(info);
201 sym_vis = ELFW(ST_VISIBILITY)(other);
203 if (sym_bind != STB_LOCAL) {
204 /* we search global or weak symbols */
205 sym_index = find_elf_sym(s, name);
206 if (!sym_index)
207 goto do_def;
208 esym = &((ElfW(Sym) *)s->data)[sym_index];
209 if (esym->st_shndx != SHN_UNDEF) {
210 esym_bind = ELFW(ST_BIND)(esym->st_info);
211 /* propagate the most constraining visibility */
212 /* STV_DEFAULT(0)<STV_PROTECTED(3)<STV_HIDDEN(2)<STV_INTERNAL(1) */
213 esym_vis = ELFW(ST_VISIBILITY)(esym->st_other);
214 if (esym_vis == STV_DEFAULT) {
215 new_vis = sym_vis;
216 } else if (sym_vis == STV_DEFAULT) {
217 new_vis = esym_vis;
218 } else {
219 new_vis = (esym_vis < sym_vis) ? esym_vis : sym_vis;
221 esym->st_other = (esym->st_other & ~ELFW(ST_VISIBILITY)(-1))
222 | new_vis;
223 other = esym->st_other; /* in case we have to patch esym */
224 if (sh_num == SHN_UNDEF) {
225 /* ignore adding of undefined symbol if the
226 corresponding symbol is already defined */
227 } else if (sym_bind == STB_GLOBAL && esym_bind == STB_WEAK) {
228 /* global overrides weak, so patch */
229 goto do_patch;
230 } else if (sym_bind == STB_WEAK && esym_bind == STB_GLOBAL) {
231 /* weak is ignored if already global */
232 } else if (sym_bind == STB_WEAK && esym_bind == STB_WEAK) {
233 /* keep first-found weak definition, ignore subsequents */
234 } else if (sym_vis == STV_HIDDEN || sym_vis == STV_INTERNAL) {
235 /* ignore hidden symbols after */
236 } else if (esym->st_shndx == SHN_COMMON
237 && (sh_num < SHN_LORESERVE || sh_num == SHN_COMMON)) {
238 /* gr: Happens with 'tcc ... -static tcctest.c' on e.g. Ubuntu 6.01
239 No idea if this is the correct solution ... */
240 goto do_patch;
241 } else if (s == tcc_state->dynsymtab_section) {
242 /* we accept that two DLL define the same symbol */
243 } else {
244 #if 0
245 printf("new_bind=%x new_shndx=%x new_vis=%x old_bind=%x old_shndx=%x old_vis=%x\n",
246 sym_bind, sh_num, new_vis, esym_bind, esym->st_shndx, esym_vis);
247 #endif
248 tcc_error_noabort("'%s' defined twice", name);
250 } else {
251 do_patch:
252 esym->st_info = ELFW(ST_INFO)(sym_bind, sym_type);
253 esym->st_shndx = sh_num;
254 new_undef_sym = 1;
255 esym->st_value = value;
256 esym->st_size = size;
257 esym->st_other = other;
259 } else {
260 do_def:
261 sym_index = put_elf_sym(s, value, size,
262 ELFW(ST_INFO)(sym_bind, sym_type), other,
263 sh_num, name);
265 return sym_index;
268 /* put relocation */
269 ST_FUNC void put_elf_reloc(Section *symtab, Section *s, unsigned long offset,
270 int type, int symbol)
272 char buf[256];
273 Section *sr;
274 ElfW_Rel *rel;
276 sr = s->reloc;
277 if (!sr) {
278 /* if no relocation section, create it */
279 snprintf(buf, sizeof(buf), REL_SECTION_FMT, s->name);
280 /* if the symtab is allocated, then we consider the relocation
281 are also */
282 sr = new_section(tcc_state, buf, SHT_RELX, symtab->sh_flags);
283 sr->sh_entsize = sizeof(ElfW_Rel);
284 sr->link = symtab;
285 sr->sh_info = s->sh_num;
286 s->reloc = sr;
288 rel = section_ptr_add(sr, sizeof(ElfW_Rel));
289 rel->r_offset = offset;
290 rel->r_info = ELFW(R_INFO)(symbol, type);
291 #ifdef TCC_TARGET_X86_64
292 rel->r_addend = 0;
293 #endif
296 /* put stab debug information */
298 ST_FUNC void put_stabs(const char *str, int type, int other, int desc,
299 unsigned long value)
301 Stab_Sym *sym;
303 sym = section_ptr_add(stab_section, sizeof(Stab_Sym));
304 if (str) {
305 sym->n_strx = put_elf_str(stabstr_section, str);
306 } else {
307 sym->n_strx = 0;
309 sym->n_type = type;
310 sym->n_other = other;
311 sym->n_desc = desc;
312 sym->n_value = value;
315 ST_FUNC void put_stabs_r(const char *str, int type, int other, int desc,
316 unsigned long value, Section *sec, int sym_index)
318 put_stabs(str, type, other, desc, value);
319 put_elf_reloc(symtab_section, stab_section,
320 stab_section->data_offset - sizeof(unsigned int),
321 R_DATA_32, sym_index);
324 ST_FUNC void put_stabn(int type, int other, int desc, int value)
326 put_stabs(NULL, type, other, desc, value);
329 ST_FUNC void put_stabd(int type, int other, int desc)
331 put_stabs(NULL, type, other, desc, 0);
334 /* Browse each elem of type <type> in section <sec> starting at elem <startoff>
335 using variable <elem> */
336 #define for_each_elem(sec, startoff, elem, type) \
337 for (elem = (type *) sec->data + startoff; \
338 elem < (type *) (sec->data + sec->data_offset); elem++)
340 /* In an ELF file symbol table, the local symbols must appear below
341 the global and weak ones. Since TCC cannot sort it while generating
342 the code, we must do it after. All the relocation tables are also
343 modified to take into account the symbol table sorting */
344 static void sort_syms(TCCState *s1, Section *s)
346 int *old_to_new_syms;
347 ElfW(Sym) *new_syms;
348 int nb_syms, i;
349 ElfW(Sym) *p, *q;
350 ElfW_Rel *rel;
351 Section *sr;
352 int type, sym_index;
354 nb_syms = s->data_offset / sizeof(ElfW(Sym));
355 new_syms = tcc_malloc(nb_syms * sizeof(ElfW(Sym)));
356 old_to_new_syms = tcc_malloc(nb_syms * sizeof(int));
358 /* first pass for local symbols */
359 p = (ElfW(Sym) *)s->data;
360 q = new_syms;
361 for(i = 0; i < nb_syms; i++) {
362 if (ELFW(ST_BIND)(p->st_info) == STB_LOCAL) {
363 old_to_new_syms[i] = q - new_syms;
364 *q++ = *p;
366 p++;
368 /* save the number of local symbols in section header */
369 s->sh_info = q - new_syms;
371 /* then second pass for non local symbols */
372 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;
376 *q++ = *p;
378 p++;
381 /* we copy the new symbols to the old */
382 memcpy(s->data, new_syms, nb_syms * sizeof(ElfW(Sym)));
383 tcc_free(new_syms);
385 /* now we modify all the relocations */
386 for(i = 1; i < s1->nb_sections; i++) {
387 sr = s1->sections[i];
388 if (sr->sh_type == SHT_RELX && sr->link == s) {
389 for_each_elem(sr, 0, rel, ElfW_Rel) {
390 sym_index = ELFW(R_SYM)(rel->r_info);
391 type = ELFW(R_TYPE)(rel->r_info);
392 sym_index = old_to_new_syms[sym_index];
393 rel->r_info = ELFW(R_INFO)(sym_index, type);
398 tcc_free(old_to_new_syms);
401 /* relocate common symbols in the .bss section */
402 ST_FUNC void relocate_common_syms(void)
404 ElfW(Sym) *sym;
405 unsigned long offset, align;
407 for_each_elem(symtab_section, 1, sym, ElfW(Sym)) {
408 if (sym->st_shndx == SHN_COMMON) {
409 /* align symbol */
410 align = sym->st_value;
411 offset = bss_section->data_offset;
412 offset = (offset + align - 1) & -align;
413 sym->st_value = offset;
414 sym->st_shndx = bss_section->sh_num;
415 offset += sym->st_size;
416 bss_section->data_offset = offset;
421 /* relocate symbol table, resolve undefined symbols if do_resolve is
422 true and output error if undefined symbol. */
423 ST_FUNC void relocate_syms(TCCState *s1, int do_resolve)
425 ElfW(Sym) *sym, *esym;
426 int sym_bind, sh_num, sym_index;
427 const char *name;
429 for_each_elem(symtab_section, 1, sym, ElfW(Sym)) {
430 sh_num = sym->st_shndx;
431 if (sh_num == SHN_UNDEF) {
432 name = (char *) strtab_section->data + sym->st_name;
433 /* Use ld.so to resolve symbol for us (for tcc -run) */
434 if (do_resolve) {
435 #if defined TCC_IS_NATIVE && !defined _WIN32
436 void *addr;
437 name = (char *) symtab_section->link->data + sym->st_name;
438 addr = resolve_sym(s1, name);
439 if (addr) {
440 sym->st_value = (addr_t)addr;
441 goto found;
443 #endif
444 } else if (s1->dynsym) {
445 /* if dynamic symbol exist, then use it */
446 sym_index = find_elf_sym(s1->dynsym, name);
447 if (sym_index) {
448 esym = &((ElfW(Sym) *)s1->dynsym->data)[sym_index];
449 sym->st_value = esym->st_value;
450 goto found;
453 /* XXX: _fp_hw seems to be part of the ABI, so we ignore
454 it */
455 if (!strcmp(name, "_fp_hw"))
456 goto found;
457 /* only weak symbols are accepted to be undefined. Their
458 value is zero */
459 sym_bind = ELFW(ST_BIND)(sym->st_info);
460 if (sym_bind == STB_WEAK) {
461 sym->st_value = 0;
462 } else {
463 tcc_error_noabort("undefined symbol '%s'", name);
465 } else if (sh_num < SHN_LORESERVE) {
466 /* add section base */
467 sym->st_value += s1->sections[sym->st_shndx]->sh_addr;
469 found: ;
473 #ifdef TCC_HAS_RUNTIME_PLTGOT
474 #ifdef TCC_TARGET_X86_64
475 #define JMP_TABLE_ENTRY_SIZE 14
476 static addr_t add_jmp_table(TCCState *s1, addr_t val)
478 char *p = s1->runtime_plt_and_got + s1->runtime_plt_and_got_offset;
479 s1->runtime_plt_and_got_offset += JMP_TABLE_ENTRY_SIZE;
480 /* jmp *0x0(%rip) */
481 p[0] = 0xff;
482 p[1] = 0x25;
483 *(int *)(p + 2) = 0;
484 *(addr_t *)(p + 6) = val;
485 return (addr_t)p;
488 static addr_t add_got_table(TCCState *s1, addr_t val)
490 addr_t *p = (addr_t *)(s1->runtime_plt_and_got + s1->runtime_plt_and_got_offset);
491 s1->runtime_plt_and_got_offset += sizeof(addr_t);
492 *p = val;
493 return (addr_t)p;
495 #elif defined TCC_TARGET_ARM
496 #define JMP_TABLE_ENTRY_SIZE 8
497 static addr_t add_jmp_table(TCCState *s1, int val)
499 uint32_t *p = (uint32_t *)(s1->runtime_plt_and_got + s1->runtime_plt_and_got_offset);
500 s1->runtime_plt_and_got_offset += JMP_TABLE_ENTRY_SIZE;
501 /* ldr pc, [pc, #-4] */
502 p[0] = 0xE51FF004;
503 p[1] = val;
504 return (addr_t)p;
506 #endif
507 #endif /* def TCC_HAS_RUNTIME_PLTGOT */
509 /* relocate a given section (CPU dependent) by applying the relocations
510 in the associated relocation section */
511 ST_FUNC void relocate_section(TCCState *s1, Section *s)
513 Section *sr = s->reloc;
514 ElfW_Rel *rel;
515 ElfW(Sym) *sym;
516 int type, sym_index;
517 unsigned char *ptr;
518 addr_t val, addr;
519 #if defined TCC_TARGET_I386 || defined TCC_TARGET_X86_64
520 ElfW_Rel *qrel = (ElfW_Rel *) sr->data; /* ptr to next reloc entry reused */
521 int esym_index;
522 #endif
524 for_each_elem(sr, 0, rel, ElfW_Rel) {
525 ptr = s->data + rel->r_offset;
527 sym_index = ELFW(R_SYM)(rel->r_info);
528 sym = &((ElfW(Sym) *)symtab_section->data)[sym_index];
529 val = sym->st_value;
530 #ifdef TCC_TARGET_X86_64
531 val += rel->r_addend;
532 #endif
533 type = ELFW(R_TYPE)(rel->r_info);
534 addr = s->sh_addr + rel->r_offset;
536 /* CPU specific */
537 switch(type) {
538 #if defined(TCC_TARGET_I386)
539 case R_386_32:
540 if (s1->output_type == TCC_OUTPUT_DLL) {
541 esym_index = s1->symtab_to_dynsym[sym_index];
542 qrel->r_offset = rel->r_offset;
543 if (esym_index) {
544 qrel->r_info = ELFW(R_INFO)(esym_index, R_386_32);
545 qrel++;
546 break;
547 } else {
548 qrel->r_info = ELFW(R_INFO)(0, R_386_RELATIVE);
549 qrel++;
552 *(int *)ptr += val;
553 break;
554 case R_386_PC32:
555 if (s1->output_type == TCC_OUTPUT_DLL) {
556 /* DLL relocation */
557 esym_index = s1->symtab_to_dynsym[sym_index];
558 if (esym_index) {
559 qrel->r_offset = rel->r_offset;
560 qrel->r_info = ELFW(R_INFO)(esym_index, R_386_PC32);
561 qrel++;
562 break;
565 *(int *)ptr += val - addr;
566 break;
567 case R_386_PLT32:
568 *(int *)ptr += val - addr;
569 break;
570 case R_386_GLOB_DAT:
571 case R_386_JMP_SLOT:
572 *(int *)ptr = val;
573 break;
574 case R_386_GOTPC:
575 *(int *)ptr += s1->got->sh_addr - addr;
576 break;
577 case R_386_GOTOFF:
578 *(int *)ptr += val - s1->got->sh_addr;
579 break;
580 case R_386_GOT32:
581 /* we load the got offset */
582 *(int *)ptr += s1->sym_attrs[sym_index].got_offset;
583 break;
584 case R_386_16:
585 if (s1->output_format != TCC_OUTPUT_FORMAT_BINARY) {
586 output_file:
587 tcc_error("can only produce 16-bit binary files");
589 *(short *)ptr += val;
590 break;
591 case R_386_PC16:
592 if (s1->output_format != TCC_OUTPUT_FORMAT_BINARY)
593 goto output_file;
594 *(short *)ptr += val - addr;
595 break;
596 #elif defined(TCC_TARGET_ARM)
597 case R_ARM_PC24:
598 case R_ARM_CALL:
599 case R_ARM_JUMP24:
600 case R_ARM_PLT32:
602 int x, is_thumb, is_call, h, blx_avail, is_bl, th_ko;
603 x = (*(int *) ptr) & 0xffffff;
604 (*(int *)ptr) &= 0xff000000;
605 if (x & 0x800000)
606 x -= 0x1000000;
607 x <<= 2;
608 blx_avail = (TCC_ARM_VERSION >= 5);
609 is_thumb = val & 1;
610 is_bl = (*(unsigned *) ptr) >> 24 == 0xeb;
611 is_call = (type == R_ARM_CALL || (type == R_ARM_PC24 && is_bl));
612 x += val - addr;
613 h = x & 2;
614 th_ko = (x & 3) && (!blx_avail || !is_call);
615 #ifdef TCC_HAS_RUNTIME_PLTGOT
616 if (s1->output_type == TCC_OUTPUT_MEMORY) {
617 if (th_ko || x >= 0x2000000 || x < -0x2000000) {
618 x += add_jmp_table(s1, val) - val; /* add veneer */
619 th_ko = (x & 3) && (!blx_avail || !is_call);
620 is_thumb = 0; /* Veneer uses ARM instructions */
623 #endif
624 if (th_ko || x >= 0x2000000 || x < -0x2000000)
625 tcc_error("can't relocate value at %x",addr);
626 x >>= 2;
627 x &= 0xffffff;
628 /* Only reached if blx is avail and it is a call */
629 if (is_thumb) {
630 x |= h << 24;
631 (*(int *)ptr) = 0xfa << 24; /* bl -> blx */
633 (*(int *) ptr) |= x;
635 break;
636 /* Since these relocations only concern Thumb-2 and blx instruction was
637 introduced before Thumb-2, we can assume blx is available and not
638 guard its use */
639 case R_ARM_THM_PC22:
640 case R_ARM_THM_JUMP24:
642 int x, hi, lo, s, j1, j2, i1, i2, imm10, imm11;
643 int to_thumb, is_call, to_plt, blx_bit = 1 << 12;
644 Section *plt;
646 /* weak reference */
647 if (sym->st_shndx == SHN_UNDEF &&
648 ELFW(ST_BIND)(sym->st_info) == STB_WEAK)
649 break;
651 /* Get initial offset */
652 hi = (*(uint16_t *)ptr);
653 lo = (*(uint16_t *)(ptr+2));
654 s = (hi >> 10) & 1;
655 j1 = (lo >> 13) & 1;
656 j2 = (lo >> 11) & 1;
657 i1 = (j1 ^ s) ^ 1;
658 i2 = (j2 ^ s) ^ 1;
659 imm10 = hi & 0x3ff;
660 imm11 = lo & 0x7ff;
661 x = (s << 24) | (i1 << 23) | (i2 << 22) |
662 (imm10 << 12) | (imm11 << 1);
663 if (x & 0x01000000)
664 x -= 0x02000000;
666 /* Relocation infos */
667 to_thumb = val & 1;
668 plt = s1->plt;
669 to_plt = (val >= plt->sh_addr) &&
670 (val < plt->sh_addr + plt->data_offset);
671 is_call = (type == R_ARM_THM_PC22);
673 /* Compute final offset */
674 if (to_plt && !is_call) /* Point to 1st instr of Thumb stub */
675 x -= 4;
676 x += val - addr;
677 if (!to_thumb && is_call) {
678 blx_bit = 0; /* bl -> blx */
679 x = (x + 3) & -4; /* Compute offset from aligned PC */
682 /* Check that relocation is possible
683 * offset must not be out of range
684 * if target is to be entered in arm mode:
685 - bit 1 must not set
686 - instruction must be a call (bl) or a jump to PLT */
687 if (!to_thumb || x >= 0x1000000 || x < -0x1000000)
688 if (to_thumb || (val & 2) || (!is_call && !to_plt))
689 tcc_error("can't relocate value at %x",addr);
691 /* Compute and store final offset */
692 s = (x >> 24) & 1;
693 i1 = (x >> 23) & 1;
694 i2 = (x >> 22) & 1;
695 j1 = s ^ (i1 ^ 1);
696 j2 = s ^ (i2 ^ 1);
697 imm10 = (x >> 12) & 0x3ff;
698 imm11 = (x >> 1) & 0x7ff;
699 (*(uint16_t *)ptr) = (uint16_t) ((hi & 0xf800) |
700 (s << 10) | imm10);
701 (*(uint16_t *)(ptr+2)) = (uint16_t) ((lo & 0xc000) |
702 (j1 << 13) | blx_bit | (j2 << 11) |
703 imm11);
705 break;
706 case R_ARM_MOVT_ABS:
707 case R_ARM_MOVW_ABS_NC:
709 int x, imm4, imm12;
710 if (type == R_ARM_MOVT_ABS)
711 val >>= 16;
712 imm12 = val & 0xfff;
713 imm4 = (val >> 12) & 0xf;
714 x = (imm4 << 16) | imm12;
715 if (type == R_ARM_THM_MOVT_ABS)
716 *(int *)ptr |= x;
717 else
718 *(int *)ptr += x;
720 break;
721 case R_ARM_THM_MOVT_ABS:
722 case R_ARM_THM_MOVW_ABS_NC:
724 int x, i, imm4, imm3, imm8;
725 if (type == R_ARM_THM_MOVT_ABS)
726 val >>= 16;
727 imm8 = val & 0xff;
728 imm3 = (val >> 8) & 0x7;
729 i = (val >> 11) & 1;
730 imm4 = (val >> 12) & 0xf;
731 x = (imm3 << 28) | (imm8 << 16) | (i << 10) | imm4;
732 if (type == R_ARM_THM_MOVT_ABS)
733 *(int *)ptr |= x;
734 else
735 *(int *)ptr += x;
737 break;
738 case R_ARM_PREL31:
740 int x;
741 x = (*(int *)ptr) & 0x7fffffff;
742 (*(int *)ptr) &= 0x80000000;
743 x = (x * 2) / 2;
744 x += val - addr;
745 if((x^(x>>1))&0x40000000)
746 tcc_error("can't relocate value at %x",addr);
747 (*(int *)ptr) |= x & 0x7fffffff;
749 case R_ARM_ABS32:
750 *(int *)ptr += val;
751 break;
752 case R_ARM_REL32:
753 *(int *)ptr += val - addr;
754 break;
755 case R_ARM_GOTPC:
756 *(int *)ptr += s1->got->sh_addr - addr;
757 break;
758 case R_ARM_GOTOFF:
759 *(int *)ptr += val - s1->got->sh_addr;
760 break;
761 case R_ARM_GOT32:
762 /* we load the got offset */
763 *(int *)ptr += s1->sym_attrs[sym_index].got_offset;
764 break;
765 case R_ARM_COPY:
766 break;
767 case R_ARM_V4BX:
768 /* trade Thumb support for ARMv4 support */
769 if ((0x0ffffff0 & *(int*)ptr) == 0x012FFF10)
770 *(int*)ptr ^= 0xE12FFF10 ^ 0xE1A0F000; /* BX Rm -> MOV PC, Rm */
771 break;
772 default:
773 fprintf(stderr,"FIXME: handle reloc type %x at %x [%p] to %x\n",
774 type, (unsigned)addr, ptr, (unsigned)val);
775 break;
776 #elif defined(TCC_TARGET_C67)
777 case R_C60_32:
778 *(int *)ptr += val;
779 break;
780 case R_C60LO16:
782 uint32_t orig;
784 /* put the low 16 bits of the absolute address
785 add to what is already there */
787 orig = ((*(int *)(ptr )) >> 7) & 0xffff;
788 orig |= (((*(int *)(ptr+4)) >> 7) & 0xffff) << 16;
790 /* patch both at once - assumes always in pairs Low - High */
792 *(int *) ptr = (*(int *) ptr & (~(0xffff << 7)) ) | (((val+orig) & 0xffff) << 7);
793 *(int *)(ptr+4) = (*(int *)(ptr+4) & (~(0xffff << 7)) ) | ((((val+orig)>>16) & 0xffff) << 7);
795 break;
796 case R_C60HI16:
797 break;
798 default:
799 fprintf(stderr,"FIXME: handle reloc type %x at %x [%p] to %x\n",
800 type, (unsigned)addr, ptr, (unsigned)val);
801 break;
802 #elif defined(TCC_TARGET_X86_64)
803 case R_X86_64_64:
804 if (s1->output_type == TCC_OUTPUT_DLL) {
805 esym_index = s1->symtab_to_dynsym[sym_index];
806 qrel->r_offset = rel->r_offset;
807 if (esym_index) {
808 qrel->r_info = ELFW(R_INFO)(esym_index, R_X86_64_64);
809 qrel->r_addend = rel->r_addend;
810 qrel++;
811 break;
812 } else {
813 qrel->r_info = ELFW(R_INFO)(0, R_X86_64_RELATIVE);
814 qrel->r_addend = *(long long *)ptr + val;
815 qrel++;
818 *(long long *)ptr += val;
819 break;
820 case R_X86_64_32:
821 case R_X86_64_32S:
822 if (s1->output_type == TCC_OUTPUT_DLL) {
823 /* XXX: this logic may depend on TCC's codegen
824 now TCC uses R_X86_64_32 even for a 64bit pointer */
825 qrel->r_info = ELFW(R_INFO)(0, R_X86_64_RELATIVE);
826 qrel->r_addend = *(int *)ptr + val;
827 qrel++;
829 *(int *)ptr += val;
830 break;
832 case R_X86_64_PC32:
833 if (s1->output_type == TCC_OUTPUT_DLL) {
834 /* DLL relocation */
835 esym_index = s1->symtab_to_dynsym[sym_index];
836 if (esym_index) {
837 qrel->r_offset = rel->r_offset;
838 qrel->r_info = ELFW(R_INFO)(esym_index, R_X86_64_PC32);
839 qrel->r_addend = *(int *)ptr;
840 qrel++;
841 break;
844 goto plt32pc32;
846 case R_X86_64_PLT32:
847 /* We've put the PLT slot offset into r_addend when generating
848 it, and that's what we must use as relocation value (adjusted
849 by section offset of course). */
850 if (s1->output_type != TCC_OUTPUT_MEMORY)
851 val = s1->plt->sh_addr + rel->r_addend;
852 /* fallthrough. */
854 plt32pc32:
856 long long diff;
857 diff = (long long)val - addr;
858 if (diff <= -2147483647 || diff > 2147483647) {
859 #ifdef TCC_HAS_RUNTIME_PLTGOT
860 /* XXX: naive support for over 32bit jump */
861 if (s1->output_type == TCC_OUTPUT_MEMORY) {
862 val = (add_jmp_table(s1, val - rel->r_addend) +
863 rel->r_addend);
864 diff = val - addr;
866 #endif
867 if (diff <= -2147483647 || diff > 2147483647) {
868 tcc_error("internal error: relocation failed");
871 *(int *)ptr += diff;
873 break;
874 case R_X86_64_GLOB_DAT:
875 case R_X86_64_JUMP_SLOT:
876 /* They don't need addend */
877 *(int *)ptr = val - rel->r_addend;
878 break;
879 case R_X86_64_GOTPCREL:
880 #ifdef TCC_HAS_RUNTIME_PLTGOT
881 if (s1->output_type == TCC_OUTPUT_MEMORY) {
882 val = add_got_table(s1, val - rel->r_addend) + rel->r_addend;
883 *(int *)ptr += val - addr;
884 break;
886 #endif
887 *(int *)ptr += (s1->got->sh_addr - addr +
888 s1->sym_attrs[sym_index].got_offset - 4);
889 break;
890 case R_X86_64_GOTTPOFF:
891 *(int *)ptr += val - s1->got->sh_addr;
892 break;
893 case R_X86_64_GOT32:
894 /* we load the got offset */
895 *(int *)ptr += s1->sym_attrs[sym_index].got_offset;
896 break;
897 #else
898 #error unsupported processor
899 #endif
902 /* if the relocation is allocated, we change its symbol table */
903 if (sr->sh_flags & SHF_ALLOC)
904 sr->link = s1->dynsym;
907 /* relocate relocation table in 'sr' */
908 static void relocate_rel(TCCState *s1, Section *sr)
910 Section *s;
911 ElfW_Rel *rel;
913 s = s1->sections[sr->sh_info];
914 for_each_elem(sr, 0, rel, ElfW_Rel)
915 rel->r_offset += s->sh_addr;
918 /* count the number of dynamic relocations so that we can reserve
919 their space */
920 static int prepare_dynamic_rel(TCCState *s1, Section *sr)
922 ElfW_Rel *rel;
923 int sym_index, esym_index, type, count;
925 count = 0;
926 for_each_elem(sr, 0, rel, ElfW_Rel) {
927 sym_index = ELFW(R_SYM)(rel->r_info);
928 type = ELFW(R_TYPE)(rel->r_info);
929 switch(type) {
930 #if defined(TCC_TARGET_I386)
931 case R_386_32:
932 #elif defined(TCC_TARGET_X86_64)
933 case R_X86_64_32:
934 case R_X86_64_32S:
935 case R_X86_64_64:
936 #endif
937 count++;
938 break;
939 #if defined(TCC_TARGET_I386)
940 case R_386_PC32:
941 #elif defined(TCC_TARGET_X86_64)
942 case R_X86_64_PC32:
943 #endif
944 esym_index = s1->symtab_to_dynsym[sym_index];
945 if (esym_index)
946 count++;
947 break;
948 default:
949 break;
952 if (count) {
953 /* allocate the section */
954 sr->sh_flags |= SHF_ALLOC;
955 sr->sh_size = count * sizeof(ElfW_Rel);
957 return count;
960 static struct sym_attr *alloc_sym_attr(TCCState *s1, int index)
962 int n;
963 struct sym_attr *tab;
965 if (index >= s1->nb_sym_attrs) {
966 /* find immediately bigger power of 2 and reallocate array */
967 n = 1;
968 while (index >= n)
969 n *= 2;
970 tab = tcc_realloc(s1->sym_attrs, n * sizeof(*s1->sym_attrs));
971 s1->sym_attrs = tab;
972 memset(s1->sym_attrs + s1->nb_sym_attrs, 0,
973 (n - s1->nb_sym_attrs) * sizeof(*s1->sym_attrs));
974 s1->nb_sym_attrs = n;
976 return &s1->sym_attrs[index];
979 /* XXX: suppress that */
980 static void put32(unsigned char *p, uint32_t val)
982 p[0] = val;
983 p[1] = val >> 8;
984 p[2] = val >> 16;
985 p[3] = val >> 24;
988 #if defined(TCC_TARGET_I386) || defined(TCC_TARGET_ARM) || \
989 defined(TCC_TARGET_X86_64)
990 static uint32_t get32(unsigned char *p)
992 return p[0] | (p[1] << 8) | (p[2] << 16) | (p[3] << 24);
994 #endif
996 static void build_got(TCCState *s1)
998 unsigned char *ptr;
1000 /* if no got, then create it */
1001 s1->got = new_section(s1, ".got", SHT_PROGBITS, SHF_ALLOC | SHF_WRITE);
1002 s1->got->sh_entsize = 4;
1003 add_elf_sym(symtab_section, 0, 4, ELFW(ST_INFO)(STB_GLOBAL, STT_OBJECT),
1004 0, s1->got->sh_num, "_GLOBAL_OFFSET_TABLE_");
1005 ptr = section_ptr_add(s1->got, 3 * PTR_SIZE);
1006 #if PTR_SIZE == 4
1007 /* keep space for _DYNAMIC pointer, if present */
1008 put32(ptr, 0);
1009 /* two dummy got entries */
1010 put32(ptr + 4, 0);
1011 put32(ptr + 8, 0);
1012 #else
1013 /* keep space for _DYNAMIC pointer, if present */
1014 put32(ptr, 0);
1015 put32(ptr + 4, 0);
1016 /* two dummy got entries */
1017 put32(ptr + 8, 0);
1018 put32(ptr + 12, 0);
1019 put32(ptr + 16, 0);
1020 put32(ptr + 20, 0);
1021 #endif
1024 /* put a got or plt entry corresponding to a symbol in symtab_section. 'size'
1025 and 'info' can be modifed if more precise info comes from the DLL.
1026 Returns offset of GOT or PLT slot. */
1027 static unsigned long put_got_entry(TCCState *s1,
1028 int reloc_type, unsigned long size, int info,
1029 int sym_index)
1031 int index, need_plt_entry;
1032 const char *name;
1033 ElfW(Sym) *sym;
1034 unsigned long offset;
1035 int *ptr;
1036 struct sym_attr *symattr;
1038 if (!s1->got)
1039 build_got(s1);
1041 need_plt_entry = s1->dynsym &&
1042 #ifdef TCC_TARGET_X86_64
1043 (reloc_type == R_X86_64_JUMP_SLOT);
1044 #elif defined(TCC_TARGET_I386)
1045 (reloc_type == R_386_JMP_SLOT);
1046 #elif defined(TCC_TARGET_ARM)
1047 (reloc_type == R_ARM_JUMP_SLOT);
1048 #else
1050 #endif
1052 /* If a got/plt entry already exists for that symbol, no need to add one */
1053 if (sym_index < s1->nb_sym_attrs) {
1054 if (need_plt_entry && s1->sym_attrs[sym_index].plt_offset)
1055 return s1->sym_attrs[sym_index].plt_offset;
1056 else if (!need_plt_entry && s1->sym_attrs[sym_index].got_offset)
1057 return s1->sym_attrs[sym_index].got_offset;
1060 symattr = alloc_sym_attr(s1, sym_index);
1062 /* Only store the GOT offset if it's not generated for the PLT entry. */
1063 if (!need_plt_entry)
1064 symattr->got_offset = s1->got->data_offset;
1066 if (s1->dynsym) {
1067 sym = &((ElfW(Sym) *)symtab_section->data)[sym_index];
1068 name = (char *) symtab_section->link->data + sym->st_name;
1069 offset = sym->st_value;
1070 #if defined(TCC_TARGET_I386) || defined(TCC_TARGET_X86_64)
1071 if (need_plt_entry) {
1072 Section *plt;
1073 uint8_t *p;
1074 int modrm;
1075 unsigned long relofs;
1077 #if defined(TCC_OUTPUT_DLL_WITH_PLT)
1078 modrm = 0x25;
1079 #else
1080 /* if we build a DLL, we add a %ebx offset */
1081 if (s1->output_type == TCC_OUTPUT_DLL)
1082 modrm = 0xa3;
1083 else
1084 modrm = 0x25;
1085 #endif
1087 /* add a PLT entry */
1088 plt = s1->plt;
1089 if (plt->data_offset == 0) {
1090 /* first plt entry */
1091 p = section_ptr_add(plt, 16);
1092 p[0] = 0xff; /* pushl got + PTR_SIZE */
1093 p[1] = modrm + 0x10;
1094 put32(p + 2, PTR_SIZE);
1095 p[6] = 0xff; /* jmp *(got + PTR_SIZE * 2) */
1096 p[7] = modrm;
1097 put32(p + 8, PTR_SIZE * 2);
1100 /* The PLT slot refers to the relocation entry it needs
1101 via offset. The reloc entry is created below, so its
1102 offset is the current data_offset. */
1103 relofs = s1->got->reloc ? s1->got->reloc->data_offset : 0;
1104 symattr->plt_offset = plt->data_offset;
1105 p = section_ptr_add(plt, 16);
1106 p[0] = 0xff; /* jmp *(got + x) */
1107 p[1] = modrm;
1108 put32(p + 2, s1->got->data_offset);
1109 p[6] = 0x68; /* push $xxx */
1110 #ifdef TCC_TARGET_X86_64
1111 /* On x86-64, the relocation is referred to by _index_. */
1112 put32(p + 7, relofs / sizeof (ElfW_Rel));
1113 #else
1114 put32(p + 7, relofs);
1115 #endif
1116 p[11] = 0xe9; /* jmp plt_start */
1117 put32(p + 12, -(plt->data_offset));
1119 /* If this was an UNDEF symbol set the offset in the
1120 dynsymtab to the PLT slot, so that PC32 relocs to it
1121 can be resolved. */
1122 if (sym->st_shndx == SHN_UNDEF)
1123 offset = plt->data_offset - 16;
1125 #elif defined(TCC_TARGET_ARM)
1126 if (need_plt_entry) {
1127 Section *plt;
1128 uint8_t *p;
1130 /* if we build a DLL, we add a %ebx offset */
1131 if (s1->output_type == TCC_OUTPUT_DLL)
1132 tcc_error("DLLs unimplemented!");
1134 /* add a PLT entry */
1135 plt = s1->plt;
1136 if (plt->data_offset == 0) {
1137 /* first plt entry */
1138 p = section_ptr_add(plt, 16);
1139 put32(p, 0xe52de004); /* push {lr} */
1140 put32(p+4, 0xe59fe010); /* ldr lr, [pc, #16] */
1141 put32(p+8, 0xe08fe00e); /* add lr, pc, lr */
1142 put32(p+12, 0xe5bef008); /* ldr pc, [lr, #8]! */
1145 symattr->plt_offset = plt->data_offset;
1146 if (symattr->plt_thumb_stub) {
1147 p = section_ptr_add(plt, 20);
1148 put32(p, 0x4778); /* bx pc */
1149 put32(p+2, 0x46c0); /* nop */
1150 p += 4;
1151 } else
1152 p = section_ptr_add(plt, 16);
1153 put32(p, 0xe59fc004); /* ldr ip, [pc, #4] ; GOT entry offset */
1154 put32(p+4, 0xe08fc00c); /* add ip, pc, ip ; addr of GOT entry */
1155 put32(p+8, 0xe59cf000); /* ldr pc, [ip] ; jump to GOT entry */
1156 put32(p+12, s1->got->data_offset); /* GOT entry off once patched */
1158 /* the symbol is modified so that it will be relocated to
1159 the PLT */
1160 if (s1->output_type == TCC_OUTPUT_EXE)
1161 offset = plt->data_offset - 16;
1163 #elif defined(TCC_TARGET_C67)
1164 tcc_error("C67 got not implemented");
1165 #else
1166 #error unsupported CPU
1167 #endif
1168 /* XXX This might generate multiple syms for name. */
1169 index = put_elf_sym(s1->dynsym, offset,
1170 size, info, 0, sym->st_shndx, name);
1171 /* Create the relocation (it's against the GOT for PLT
1172 and GOT relocs). */
1173 put_elf_reloc(s1->dynsym, s1->got,
1174 s1->got->data_offset,
1175 reloc_type, index);
1176 } else {
1177 /* Without .dynsym (i.e. static link or memory output) we
1178 still need relocs against the generated got, so as to fill
1179 the entries with the symbol values (determined later). */
1180 put_elf_reloc(symtab_section, s1->got,
1181 s1->got->data_offset,
1182 reloc_type, sym_index);
1184 /* And now create the GOT slot itself. */
1185 ptr = section_ptr_add(s1->got, PTR_SIZE);
1186 *ptr = 0;
1187 if (need_plt_entry)
1188 return symattr->plt_offset;
1189 else
1190 return symattr->got_offset;
1193 /* build GOT and PLT entries */
1194 ST_FUNC void build_got_entries(TCCState *s1)
1196 Section *s;
1197 ElfW_Rel *rel;
1198 ElfW(Sym) *sym;
1199 int i, type, reloc_type, sym_index;
1201 for(i = 1; i < s1->nb_sections; i++) {
1202 s = s1->sections[i];
1203 if (s->sh_type != SHT_RELX)
1204 continue;
1205 /* no need to handle got relocations */
1206 if (s->link != symtab_section)
1207 continue;
1208 for_each_elem(s, 0, rel, ElfW_Rel) {
1209 type = ELFW(R_TYPE)(rel->r_info);
1210 switch(type) {
1211 #if defined(TCC_TARGET_I386)
1212 case R_386_GOT32:
1213 case R_386_GOTOFF:
1214 case R_386_GOTPC:
1215 case R_386_PLT32:
1216 if (!s1->got)
1217 build_got(s1);
1218 if (type == R_386_GOT32 || type == R_386_PLT32) {
1219 sym_index = ELFW(R_SYM)(rel->r_info);
1220 sym = &((ElfW(Sym) *)symtab_section->data)[sym_index];
1221 /* look at the symbol got offset. If none, then add one */
1222 if (type == R_386_GOT32)
1223 reloc_type = R_386_GLOB_DAT;
1224 else
1225 reloc_type = R_386_JMP_SLOT;
1226 put_got_entry(s1, reloc_type, sym->st_size, sym->st_info,
1227 sym_index);
1229 break;
1230 #elif defined(TCC_TARGET_ARM)
1231 case R_ARM_GOT32:
1232 case R_ARM_GOTOFF:
1233 case R_ARM_GOTPC:
1234 case R_ARM_PLT32:
1235 if (!s1->got)
1236 build_got(s1);
1237 if (type == R_ARM_GOT32 || type == R_ARM_PLT32) {
1238 sym_index = ELFW(R_SYM)(rel->r_info);
1239 sym = &((ElfW(Sym) *)symtab_section->data)[sym_index];
1240 /* look at the symbol got offset. If none, then add one */
1241 if (type == R_ARM_GOT32)
1242 reloc_type = R_ARM_GLOB_DAT;
1243 else
1244 reloc_type = R_ARM_JUMP_SLOT;
1245 put_got_entry(s1, reloc_type, sym->st_size, sym->st_info,
1246 sym_index);
1248 break;
1249 case R_ARM_THM_JUMP24:
1250 sym_index = ELFW(R_SYM)(rel->r_info);
1251 sym = &((ElfW(Sym) *)symtab_section->data)[sym_index];
1252 /* We are relocating a jump from thumb code to arm code */
1253 if (sym->st_shndx != SHN_UNDEF && !(sym->st_value & 1)) {
1254 int index;
1255 uint8_t *p;
1256 char *name, buf[1024];
1257 Section *text_section;
1259 name = (char *) symtab_section->link->data + sym->st_name;
1260 text_section = s1->sections[sym->st_shndx];
1261 /* Modify reloc to target a thumb stub to switch to ARM */
1262 snprintf(buf, sizeof(buf), "%s_from_thumb", name);
1263 index = put_elf_sym(symtab_section,
1264 text_section->data_offset + 1,
1265 sym->st_size, sym->st_info, 0,
1266 sym->st_shndx, buf);
1267 rel->r_info = ELFW(R_INFO)(index, type);
1268 /* Create a thumb stub fonction to switch to ARM mode */
1269 put_elf_reloc(symtab_section, text_section,
1270 text_section->data_offset + 4, R_ARM_JUMP24,
1271 sym_index);
1272 p = section_ptr_add(text_section, 8);
1273 put32(p, 0x4778); /* bx pc */
1274 put32(p+2, 0x46c0); /* nop */
1275 put32(p+4, 0xeafffffe); /* b $sym */
1277 #elif defined(TCC_TARGET_C67)
1278 case R_C60_GOT32:
1279 case R_C60_GOTOFF:
1280 case R_C60_GOTPC:
1281 case R_C60_PLT32:
1282 if (!s1->got)
1283 build_got(s1);
1284 if (type == R_C60_GOT32 || type == R_C60_PLT32) {
1285 sym_index = ELFW(R_SYM)(rel->r_info);
1286 sym = &((ElfW(Sym) *)symtab_section->data)[sym_index];
1287 /* look at the symbol got offset. If none, then add one */
1288 if (type == R_C60_GOT32)
1289 reloc_type = R_C60_GLOB_DAT;
1290 else
1291 reloc_type = R_C60_JMP_SLOT;
1292 put_got_entry(s1, reloc_type, sym->st_size, sym->st_info,
1293 sym_index);
1295 break;
1296 #elif defined(TCC_TARGET_X86_64)
1297 case R_X86_64_GOT32:
1298 case R_X86_64_GOTTPOFF:
1299 case R_X86_64_GOTPCREL:
1300 case R_X86_64_PLT32:
1301 if (!s1->got)
1302 build_got(s1);
1303 if (type == R_X86_64_GOT32 || type == R_X86_64_GOTPCREL ||
1304 type == R_X86_64_PLT32) {
1305 unsigned long ofs;
1306 sym_index = ELFW(R_SYM)(rel->r_info);
1307 sym = &((ElfW(Sym) *)symtab_section->data)[sym_index];
1308 /* look at the symbol got offset. If none, then add one */
1309 if (type == R_X86_64_GOT32 || type == R_X86_64_GOTPCREL)
1310 reloc_type = R_X86_64_GLOB_DAT;
1311 else
1312 reloc_type = R_X86_64_JUMP_SLOT;
1313 ofs = put_got_entry(s1, reloc_type, sym->st_size,
1314 sym->st_info, sym_index);
1315 if (type == R_X86_64_PLT32
1316 && s1->output_type != TCC_OUTPUT_MEMORY)
1317 /* We store the place of the generated PLT slot
1318 in our addend. */
1319 rel->r_addend += ofs;
1321 break;
1322 #else
1323 #error unsupported CPU
1324 #endif
1325 default:
1326 break;
1332 ST_FUNC Section *new_symtab(TCCState *s1,
1333 const char *symtab_name, int sh_type, int sh_flags,
1334 const char *strtab_name,
1335 const char *hash_name, int hash_sh_flags)
1337 Section *symtab, *strtab, *hash;
1338 int *ptr, nb_buckets;
1340 symtab = new_section(s1, symtab_name, sh_type, sh_flags);
1341 symtab->sh_entsize = sizeof(ElfW(Sym));
1342 strtab = new_section(s1, strtab_name, SHT_STRTAB, sh_flags);
1343 put_elf_str(strtab, "");
1344 symtab->link = strtab;
1345 put_elf_sym(symtab, 0, 0, 0, 0, 0, NULL);
1347 nb_buckets = 1;
1349 hash = new_section(s1, hash_name, SHT_HASH, hash_sh_flags);
1350 hash->sh_entsize = sizeof(int);
1351 symtab->hash = hash;
1352 hash->link = symtab;
1354 ptr = section_ptr_add(hash, (2 + nb_buckets + 1) * sizeof(int));
1355 ptr[0] = nb_buckets;
1356 ptr[1] = 1;
1357 memset(ptr + 2, 0, (nb_buckets + 1) * sizeof(int));
1358 return symtab;
1361 /* put dynamic tag */
1362 static void put_dt(Section *dynamic, int dt, addr_t val)
1364 ElfW(Dyn) *dyn;
1365 dyn = section_ptr_add(dynamic, sizeof(ElfW(Dyn)));
1366 dyn->d_tag = dt;
1367 dyn->d_un.d_val = val;
1370 static void add_init_array_defines(TCCState *s1, const char *section_name)
1372 Section *s;
1373 long end_offset;
1374 char sym_start[1024];
1375 char sym_end[1024];
1377 snprintf(sym_start, sizeof(sym_start), "__%s_start", section_name + 1);
1378 snprintf(sym_end, sizeof(sym_end), "__%s_end", section_name + 1);
1380 s = find_section(s1, section_name);
1381 if (!s) {
1382 end_offset = 0;
1383 s = data_section;
1384 } else {
1385 end_offset = s->data_offset;
1388 add_elf_sym(symtab_section,
1389 0, 0,
1390 ELFW(ST_INFO)(STB_GLOBAL, STT_NOTYPE), 0,
1391 s->sh_num, sym_start);
1392 add_elf_sym(symtab_section,
1393 end_offset, 0,
1394 ELFW(ST_INFO)(STB_GLOBAL, STT_NOTYPE), 0,
1395 s->sh_num, sym_end);
1398 ST_FUNC void tcc_add_bcheck(TCCState *s1)
1400 #ifdef CONFIG_TCC_BCHECK
1401 unsigned long *ptr;
1402 Section *init_section;
1403 unsigned char *pinit;
1404 int sym_index;
1406 if (0 == s1->do_bounds_check)
1407 return;
1409 /* XXX: add an object file to do that */
1410 ptr = section_ptr_add(bounds_section, sizeof(unsigned long));
1411 *ptr = 0;
1412 add_elf_sym(symtab_section, 0, 0,
1413 ELFW(ST_INFO)(STB_GLOBAL, STT_NOTYPE), 0,
1414 bounds_section->sh_num, "__bounds_start");
1415 #ifdef TCC_TARGET_I386
1416 if (s1->output_type != TCC_OUTPUT_MEMORY) {
1417 /* add 'call __bound_init()' in .init section */
1418 init_section = find_section(s1, ".init");
1419 pinit = section_ptr_add(init_section, 5);
1420 pinit[0] = 0xe8;
1421 put32(pinit + 1, -4);
1422 sym_index = find_elf_sym(symtab_section, "__bound_init");
1423 put_elf_reloc(symtab_section, init_section,
1424 init_section->data_offset - 4, R_386_PC32, sym_index);
1426 #endif
1427 #endif
1430 static inline int tcc_add_support(TCCState *s1, const char *filename)
1432 char buf[1024];
1433 snprintf(buf, sizeof(buf), "%s/%s", s1->tcc_lib_path, filename);
1434 return tcc_add_file(s1, buf);
1437 /* add tcc runtime libraries */
1438 ST_FUNC void tcc_add_runtime(TCCState *s1)
1440 /* add libc */
1441 if (!s1->nostdlib) {
1442 tcc_add_library(s1, "c");
1443 #ifdef CONFIG_USE_LIBGCC
1444 if (!s1->static_link) {
1445 tcc_add_file(s1, TCC_LIBGCC);
1446 tcc_add_support(s1, "libtcc1.a");
1447 } else
1448 tcc_add_support(s1, "libtcc1.a");
1449 #else
1450 tcc_add_support(s1, "libtcc1.a");
1451 #endif
1454 /* tcc_add_bcheck tries to relocate a call to __bound_init in _init so
1455 libtcc1.a must be loaded before for __bound_init to be defined and
1456 crtn.o must be loaded after to not finalize _init too early. */
1457 tcc_add_bcheck(s1);
1459 if (!s1->nostdlib) {
1460 /* add crt end if not memory output */
1461 if (s1->output_type != TCC_OUTPUT_MEMORY)
1462 tcc_add_crt(s1, "crtn.o");
1466 /* add various standard linker symbols (must be done after the
1467 sections are filled (for example after allocating common
1468 symbols)) */
1469 ST_FUNC void tcc_add_linker_symbols(TCCState *s1)
1471 char buf[1024];
1472 int i;
1473 Section *s;
1475 add_elf_sym(symtab_section,
1476 text_section->data_offset, 0,
1477 ELFW(ST_INFO)(STB_GLOBAL, STT_NOTYPE), 0,
1478 text_section->sh_num, "_etext");
1479 add_elf_sym(symtab_section,
1480 data_section->data_offset, 0,
1481 ELFW(ST_INFO)(STB_GLOBAL, STT_NOTYPE), 0,
1482 data_section->sh_num, "_edata");
1483 add_elf_sym(symtab_section,
1484 bss_section->data_offset, 0,
1485 ELFW(ST_INFO)(STB_GLOBAL, STT_NOTYPE), 0,
1486 bss_section->sh_num, "_end");
1487 /* horrible new standard ldscript defines */
1488 add_init_array_defines(s1, ".preinit_array");
1489 add_init_array_defines(s1, ".init_array");
1490 add_init_array_defines(s1, ".fini_array");
1492 /* add start and stop symbols for sections whose name can be
1493 expressed in C */
1494 for(i = 1; i < s1->nb_sections; i++) {
1495 s = s1->sections[i];
1496 if (s->sh_type == SHT_PROGBITS &&
1497 (s->sh_flags & SHF_ALLOC)) {
1498 const char *p;
1499 int ch;
1501 /* check if section name can be expressed in C */
1502 p = s->name;
1503 for(;;) {
1504 ch = *p;
1505 if (!ch)
1506 break;
1507 if (!isid(ch) && !isnum(ch))
1508 goto next_sec;
1509 p++;
1511 snprintf(buf, sizeof(buf), "__start_%s", s->name);
1512 add_elf_sym(symtab_section,
1513 0, 0,
1514 ELFW(ST_INFO)(STB_GLOBAL, STT_NOTYPE), 0,
1515 s->sh_num, buf);
1516 snprintf(buf, sizeof(buf), "__stop_%s", s->name);
1517 add_elf_sym(symtab_section,
1518 s->data_offset, 0,
1519 ELFW(ST_INFO)(STB_GLOBAL, STT_NOTYPE), 0,
1520 s->sh_num, buf);
1522 next_sec: ;
1526 static void tcc_output_binary(TCCState *s1, FILE *f,
1527 const int *sec_order)
1529 Section *s;
1530 int i, offset, size;
1532 offset = 0;
1533 for(i=1;i<s1->nb_sections;i++) {
1534 s = s1->sections[sec_order[i]];
1535 if (s->sh_type != SHT_NOBITS &&
1536 (s->sh_flags & SHF_ALLOC)) {
1537 while (offset < s->sh_offset) {
1538 fputc(0, f);
1539 offset++;
1541 size = s->sh_size;
1542 fwrite(s->data, 1, size, f);
1543 offset += size;
1548 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
1549 #define HAVE_PHDR 1
1550 #define EXTRA_RELITEMS 14
1552 /* move the relocation value from .dynsym to .got */
1553 void patch_dynsym_undef(TCCState *s1, Section *s)
1555 uint32_t *gotd = (void *)s1->got->data;
1556 ElfW(Sym) *sym;
1558 gotd += 3; /* dummy entries in .got */
1559 /* relocate symbols in .dynsym */
1560 for_each_elem(s, 1, sym, ElfW(Sym)) {
1561 if (sym->st_shndx == SHN_UNDEF) {
1562 *gotd++ = sym->st_value + 6; /* XXX 6 is magic ? */
1563 sym->st_value = 0;
1567 #else
1568 #define HAVE_PHDR 1
1569 #define EXTRA_RELITEMS 9
1571 /* zero plt offsets of weak symbols in .dynsym */
1572 void patch_dynsym_undef(TCCState *s1, Section *s)
1574 ElfW(Sym) *sym;
1576 for_each_elem(s, 1, sym, ElfW(Sym))
1577 if (sym->st_shndx == SHN_UNDEF && ELFW(ST_BIND)(sym->st_info) == STB_WEAK)
1578 sym->st_value = 0;
1580 #endif
1582 ST_FUNC void fill_got_entry(TCCState *s1, ElfW_Rel *rel)
1584 int sym_index = ELFW(R_SYM) (rel->r_info);
1585 ElfW(Sym) *sym = &((ElfW(Sym) *) symtab_section->data)[sym_index];
1586 unsigned long offset;
1588 if (sym_index >= s1->nb_sym_attrs)
1589 return;
1590 offset = s1->sym_attrs[sym_index].got_offset;
1591 section_reserve(s1->got, offset + PTR_SIZE);
1592 #ifdef TCC_TARGET_X86_64
1593 /* only works for x86-64 */
1594 put32(s1->got->data + offset + 4, sym->st_value >> 32);
1595 #endif
1596 put32(s1->got->data + offset, sym->st_value & 0xffffffff);
1599 /* Perform relocation to GOT or PLT entries */
1600 ST_FUNC void fill_got(TCCState *s1)
1602 Section *s;
1603 ElfW_Rel *rel;
1604 int i;
1606 for(i = 1; i < s1->nb_sections; i++) {
1607 s = s1->sections[i];
1608 if (s->sh_type != SHT_RELX)
1609 continue;
1610 /* no need to handle got relocations */
1611 if (s->link != symtab_section)
1612 continue;
1613 for_each_elem(s, 0, rel, ElfW_Rel) {
1614 switch (ELFW(R_TYPE) (rel->r_info)) {
1615 case R_X86_64_GOT32:
1616 case R_X86_64_GOTPCREL:
1617 case R_X86_64_PLT32:
1618 fill_got_entry(s1, rel);
1619 break;
1625 /* Bind symbols of executable: resolve undefined symbols from exported symbols
1626 in shared libraries and export non local defined symbols to shared libraries
1627 if -rdynamic switch was given on command line */
1628 static void bind_exe_dynsyms(TCCState *s1)
1630 const char *name;
1631 int sym_index, index;
1632 ElfW(Sym) *sym, *esym;
1633 int type;
1635 /* Resolve undefined symbols from dynamic symbols. When there is a match:
1636 - if STT_FUNC or STT_GNU_IFUNC symbol -> add it in PLT
1637 - if STT_OBJECT symbol -> add it in .bss section with suitable reloc */
1638 for_each_elem(symtab_section, 1, sym, ElfW(Sym)) {
1639 if (sym->st_shndx == SHN_UNDEF) {
1640 name = (char *) symtab_section->link->data + sym->st_name;
1641 sym_index = find_elf_sym(s1->dynsymtab_section, name);
1642 if (sym_index) {
1643 esym = &((ElfW(Sym) *)s1->dynsymtab_section->data)[sym_index];
1644 type = ELFW(ST_TYPE)(esym->st_info);
1645 if ((type == STT_FUNC) || (type == STT_GNU_IFUNC)) {
1646 /* Indirect functions shall have STT_FUNC type in executable
1647 * dynsym section. Indeed, a dlsym call following a lazy
1648 * resolution would pick the symbol value from the
1649 * executable dynsym entry which would contain the address
1650 * of the function wanted by the caller of dlsym instead of
1651 * the address of the function that would return that
1652 * address */
1653 put_got_entry(s1, R_JMP_SLOT, esym->st_size,
1654 ELFW(ST_INFO)(STB_GLOBAL,STT_FUNC),
1655 sym - (ElfW(Sym) *)symtab_section->data);
1656 } else if (type == STT_OBJECT) {
1657 unsigned long offset;
1658 ElfW(Sym) *dynsym;
1659 offset = bss_section->data_offset;
1660 /* XXX: which alignment ? */
1661 offset = (offset + 16 - 1) & -16;
1662 index = put_elf_sym(s1->dynsym, offset, esym->st_size,
1663 esym->st_info, 0, bss_section->sh_num,
1664 name);
1665 /* Ensure R_COPY works for weak symbol aliases */
1666 if (ELFW(ST_BIND)(esym->st_info) == STB_WEAK) {
1667 for_each_elem(s1->dynsymtab_section, 1, dynsym, ElfW(Sym)) {
1668 if ((dynsym->st_value == esym->st_value)
1669 && (ELFW(ST_BIND)(dynsym->st_info) == STB_GLOBAL)) {
1670 char *dynname = (char *) s1->dynsymtab_section->link->data
1671 + dynsym->st_name;
1672 put_elf_sym(s1->dynsym, offset, dynsym->st_size,
1673 dynsym->st_info, 0,
1674 bss_section->sh_num, dynname);
1675 break;
1679 put_elf_reloc(s1->dynsym, bss_section,
1680 offset, R_COPY, index);
1681 offset += esym->st_size;
1682 bss_section->data_offset = offset;
1684 } else {
1685 /* STB_WEAK undefined symbols are accepted */
1686 /* XXX: _fp_hw seems to be part of the ABI, so we ignore it */
1687 if (ELFW(ST_BIND)(sym->st_info) == STB_WEAK ||
1688 !strcmp(name, "_fp_hw")) {
1689 } else {
1690 tcc_error_noabort("undefined symbol '%s'", name);
1693 } else if (s1->rdynamic && ELFW(ST_BIND)(sym->st_info) != STB_LOCAL) {
1694 /* if -rdynamic option, then export all non local symbols */
1695 name = (char *) symtab_section->link->data + sym->st_name;
1696 put_elf_sym(s1->dynsym, sym->st_value, sym->st_size, sym->st_info,
1697 0, sym->st_shndx, name);
1702 /* Bind symbols of libraries: export non local symbols of executable that
1703 resolve undefined symbols of shared libraries */
1704 static void bind_libs_dynsyms(TCCState *s1)
1706 const char *name;
1707 int sym_index;
1708 ElfW(Sym) *sym, *esym;
1710 /* now look at unresolved dynamic symbols and export
1711 corresponding symbol */
1712 for_each_elem(s1->dynsymtab_section, 1, esym, ElfW(Sym)) {
1713 if (esym->st_shndx == SHN_UNDEF) {
1714 name = (char *) s1->dynsymtab_section->link->data + esym->st_name;
1715 sym_index = find_elf_sym(symtab_section, name);
1716 if (sym_index) {
1717 /* XXX: avoid adding a symbol if already present because of
1718 -rdynamic ? */
1719 sym = &((ElfW(Sym) *)symtab_section->data)[sym_index];
1720 put_elf_sym(s1->dynsym, sym->st_value, sym->st_size,
1721 sym->st_info, 0, sym->st_shndx, name);
1722 } else {
1723 /* weak symbols can stay undefined */
1724 if (ELFW(ST_BIND)(esym->st_info) != STB_WEAK)
1725 tcc_warning("undefined dynamic symbol '%s'", name);
1731 /* Export all non local symbols (for shared libraries) */
1732 static void export_global_syms(TCCState *s1)
1734 int nb_syms, dynindex, index;
1735 const char *name;
1736 ElfW(Sym) *sym;
1738 nb_syms = symtab_section->data_offset / sizeof(ElfW(Sym));
1739 s1->symtab_to_dynsym = tcc_mallocz(sizeof(int) * nb_syms);
1740 for_each_elem(symtab_section, 1, sym, ElfW(Sym)) {
1741 if (ELFW(ST_BIND)(sym->st_info) != STB_LOCAL) {
1742 name = (char *) symtab_section->link->data + sym->st_name;
1743 dynindex = put_elf_sym(s1->dynsym, sym->st_value, sym->st_size,
1744 sym->st_info, 0, sym->st_shndx, name);
1745 index = sym - (ElfW(Sym) *) symtab_section->data;
1746 s1->symtab_to_dynsym[index] = dynindex;
1751 /* relocate the PLT: compute addresses and offsets in the PLT now that final
1752 address for PLT and GOT are known (see fill_program_header) */
1753 static void relocate_plt(TCCState *s1)
1755 uint8_t *p, *p_end;
1757 p = s1->plt->data;
1758 p_end = p + s1->plt->data_offset;
1759 if (p < p_end) {
1760 #if defined(TCC_TARGET_I386)
1761 put32(p + 2, get32(p + 2) + s1->got->sh_addr);
1762 put32(p + 8, get32(p + 8) + s1->got->sh_addr);
1763 p += 16;
1764 while (p < p_end) {
1765 put32(p + 2, get32(p + 2) + s1->got->sh_addr);
1766 p += 16;
1768 #elif defined(TCC_TARGET_X86_64)
1769 int x = s1->got->sh_addr - s1->plt->sh_addr - 6;
1770 put32(p + 2, get32(p + 2) + x);
1771 put32(p + 8, get32(p + 8) + x - 6);
1772 p += 16;
1773 while (p < p_end) {
1774 put32(p + 2, get32(p + 2) + x + s1->plt->data - p);
1775 p += 16;
1777 #elif defined(TCC_TARGET_ARM)
1778 int x;
1779 x=s1->got->sh_addr - s1->plt->sh_addr - 12;
1780 p += 16;
1781 while (p < p_end) {
1782 if (get32(p) == 0x46c04778) /* PLT Thumb stub present */
1783 p += 4;
1784 put32(p + 12, x + get32(p + 12) + s1->plt->data - p);
1785 p += 16;
1787 #elif defined(TCC_TARGET_C67)
1788 /* XXX: TODO */
1789 #else
1790 #error unsupported CPU
1791 #endif
1795 /* Allocate strings for section names and decide if an unallocated section
1796 should be output.
1798 NOTE: the strsec section comes last, so its size is also correct ! */
1799 static void alloc_sec_names(TCCState *s1, int file_type, Section *strsec)
1801 int i;
1802 Section *s;
1804 /* Allocate strings for section names */
1805 for(i = 1; i < s1->nb_sections; i++) {
1806 s = s1->sections[i];
1807 s->sh_name = put_elf_str(strsec, s->name);
1808 /* when generating a DLL, we include relocations but we may
1809 patch them */
1810 if (file_type == TCC_OUTPUT_DLL &&
1811 s->sh_type == SHT_RELX &&
1812 !(s->sh_flags & SHF_ALLOC)) {
1813 /* gr: avoid bogus relocs for empty (debug) sections */
1814 if (s1->sections[s->sh_info]->sh_flags & SHF_ALLOC)
1815 prepare_dynamic_rel(s1, s);
1816 else if (s1->do_debug)
1817 s->sh_size = s->data_offset;
1818 } else if (s1->do_debug ||
1819 file_type == TCC_OUTPUT_OBJ ||
1820 (s->sh_flags & SHF_ALLOC) ||
1821 i == (s1->nb_sections - 1)) {
1822 /* we output all sections if debug or object file */
1823 s->sh_size = s->data_offset;
1828 /* Info to be copied in dynamic section */
1829 struct dyn_inf {
1830 Section *dynamic;
1831 Section *dynstr;
1832 unsigned long dyn_rel_off;
1833 addr_t rel_addr;
1834 addr_t rel_size;
1835 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
1836 addr_t bss_addr;
1837 addr_t bss_size;
1838 #endif
1841 /* Assign sections to segments and decide how are sections laid out when loaded
1842 in memory. This function also fills corresponding program headers. */
1843 static int layout_sections(TCCState *s1, ElfW(Phdr) *phdr, int phnum,
1844 Section *interp, struct dyn_inf *dyninf,
1845 int *sec_order)
1847 int i, j, k, file_type, sh_order_index, file_offset;
1848 long long tmp;
1849 addr_t addr;
1850 ElfW(Phdr) *ph;
1851 Section *s;
1853 file_type = s1->output_type;
1854 sh_order_index = 1;
1855 if (s1->output_format == TCC_OUTPUT_FORMAT_ELF)
1856 file_offset = sizeof(ElfW(Ehdr)) + phnum * sizeof(ElfW(Phdr));
1857 else
1858 file_offset = 0;
1860 if (phnum > 0) {
1861 if (s1->has_text_addr) {
1862 int a_offset, p_offset;
1863 addr = s1->text_addr;
1864 /* we ensure that (addr % ELF_PAGE_SIZE) == file_offset %
1865 ELF_PAGE_SIZE */
1866 a_offset = (int) (addr & (s1->section_align - 1));
1867 p_offset = file_offset & (s1->section_align - 1);
1868 if (a_offset < p_offset)
1869 a_offset += s1->section_align;
1870 file_offset += (a_offset - p_offset);
1871 } else {
1872 if (file_type == TCC_OUTPUT_DLL)
1873 addr = 0;
1874 else
1875 addr = ELF_START_ADDR;
1876 /* compute address after headers */
1877 addr += (file_offset & (s1->section_align - 1));
1880 ph = &phdr[0];
1881 /* Leave one program headers for the program interpreter and one for
1882 the program header table itself if needed. These are done later as
1883 they require section layout to be done first. */
1884 if (interp)
1885 ph += 1 + HAVE_PHDR;
1887 /* dynamic relocation table information, for .dynamic section */
1888 dyninf->rel_addr = dyninf->rel_size = 0;
1889 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
1890 dyninf->bss_addr = dyninf->bss_size = 0;
1891 #endif
1893 for(j = 0; j < 2; j++) {
1894 ph->p_type = PT_LOAD;
1895 if (j == 0)
1896 ph->p_flags = PF_R | PF_X;
1897 else
1898 ph->p_flags = PF_R | PF_W;
1899 ph->p_align = s1->section_align;
1901 /* Decide the layout of sections loaded in memory. This must
1902 be done before program headers are filled since they contain
1903 info about the layout. We do the following ordering: interp,
1904 symbol tables, relocations, progbits, nobits */
1905 /* XXX: do faster and simpler sorting */
1906 for(k = 0; k < 5; k++) {
1907 for(i = 1; i < s1->nb_sections; i++) {
1908 s = s1->sections[i];
1909 /* compute if section should be included */
1910 if (j == 0) {
1911 if ((s->sh_flags & (SHF_ALLOC | SHF_WRITE)) !=
1912 SHF_ALLOC)
1913 continue;
1914 } else {
1915 if ((s->sh_flags & (SHF_ALLOC | SHF_WRITE)) !=
1916 (SHF_ALLOC | SHF_WRITE))
1917 continue;
1919 if (s == interp) {
1920 if (k != 0)
1921 continue;
1922 } else if (s->sh_type == SHT_DYNSYM ||
1923 s->sh_type == SHT_STRTAB ||
1924 s->sh_type == SHT_HASH) {
1925 if (k != 1)
1926 continue;
1927 } else if (s->sh_type == SHT_RELX) {
1928 if (k != 2)
1929 continue;
1930 } else if (s->sh_type == SHT_NOBITS) {
1931 if (k != 4)
1932 continue;
1933 } else {
1934 if (k != 3)
1935 continue;
1937 sec_order[sh_order_index++] = i;
1939 /* section matches: we align it and add its size */
1940 tmp = addr;
1941 addr = (addr + s->sh_addralign - 1) &
1942 ~(s->sh_addralign - 1);
1943 file_offset += (int) ( addr - tmp );
1944 s->sh_offset = file_offset;
1945 s->sh_addr = addr;
1947 /* update program header infos */
1948 if (ph->p_offset == 0) {
1949 ph->p_offset = file_offset;
1950 ph->p_vaddr = addr;
1951 ph->p_paddr = ph->p_vaddr;
1953 /* update dynamic relocation infos */
1954 if (s->sh_type == SHT_RELX) {
1955 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
1956 if (!strcmp(strsec->data + s->sh_name, ".rel.got")) {
1957 dyninf->rel_addr = addr;
1958 dyninf->rel_size += s->sh_size; /* XXX only first rel. */
1960 if (!strcmp(strsec->data + s->sh_name, ".rel.bss")) {
1961 dyninf->bss_addr = addr;
1962 dyninf->bss_size = s->sh_size; /* XXX only first rel. */
1964 #else
1965 if (dyninf->rel_size == 0)
1966 dyninf->rel_addr = addr;
1967 dyninf->rel_size += s->sh_size;
1968 #endif
1970 addr += s->sh_size;
1971 if (s->sh_type != SHT_NOBITS)
1972 file_offset += s->sh_size;
1975 if (j == 0) {
1976 /* Make the first PT_LOAD segment include the program
1977 headers itself (and the ELF header as well), it'll
1978 come out with same memory use but will make various
1979 tools like binutils strip work better. */
1980 ph->p_offset &= ~(ph->p_align - 1);
1981 ph->p_vaddr &= ~(ph->p_align - 1);
1982 ph->p_paddr &= ~(ph->p_align - 1);
1984 ph->p_filesz = file_offset - ph->p_offset;
1985 ph->p_memsz = addr - ph->p_vaddr;
1986 ph++;
1987 if (j == 0) {
1988 if (s1->output_format == TCC_OUTPUT_FORMAT_ELF) {
1989 /* if in the middle of a page, we duplicate the page in
1990 memory so that one copy is RX and the other is RW */
1991 if ((addr & (s1->section_align - 1)) != 0)
1992 addr += s1->section_align;
1993 } else {
1994 addr = (addr + s1->section_align - 1) & ~(s1->section_align - 1);
1995 file_offset = (file_offset + s1->section_align - 1) &
1996 ~(s1->section_align - 1);
2002 /* all other sections come after */
2003 for(i = 1; i < s1->nb_sections; i++) {
2004 s = s1->sections[i];
2005 if (phnum > 0 && (s->sh_flags & SHF_ALLOC))
2006 continue;
2007 sec_order[sh_order_index++] = i;
2009 file_offset = (file_offset + s->sh_addralign - 1) &
2010 ~(s->sh_addralign - 1);
2011 s->sh_offset = file_offset;
2012 if (s->sh_type != SHT_NOBITS)
2013 file_offset += s->sh_size;
2016 return file_offset;
2019 static void fill_unloadable_phdr(ElfW(Phdr) *phdr, int phnum, Section *interp,
2020 Section *dynamic)
2022 ElfW(Phdr) *ph;
2024 /* if interpreter, then add corresponding program header */
2025 if (interp) {
2026 ph = &phdr[0];
2028 if (HAVE_PHDR)
2030 int len = phnum * sizeof(ElfW(Phdr));
2032 ph->p_type = PT_PHDR;
2033 ph->p_offset = sizeof(ElfW(Ehdr));
2034 ph->p_vaddr = interp->sh_addr - len;
2035 ph->p_paddr = ph->p_vaddr;
2036 ph->p_filesz = ph->p_memsz = len;
2037 ph->p_flags = PF_R | PF_X;
2038 ph->p_align = 4; /* interp->sh_addralign; */
2039 ph++;
2042 ph->p_type = PT_INTERP;
2043 ph->p_offset = interp->sh_offset;
2044 ph->p_vaddr = interp->sh_addr;
2045 ph->p_paddr = ph->p_vaddr;
2046 ph->p_filesz = interp->sh_size;
2047 ph->p_memsz = interp->sh_size;
2048 ph->p_flags = PF_R;
2049 ph->p_align = interp->sh_addralign;
2052 /* if dynamic section, then add corresponding program header */
2053 if (dynamic) {
2054 ph = &phdr[phnum - 1];
2056 ph->p_type = PT_DYNAMIC;
2057 ph->p_offset = dynamic->sh_offset;
2058 ph->p_vaddr = dynamic->sh_addr;
2059 ph->p_paddr = ph->p_vaddr;
2060 ph->p_filesz = dynamic->sh_size;
2061 ph->p_memsz = dynamic->sh_size;
2062 ph->p_flags = PF_R | PF_W;
2063 ph->p_align = dynamic->sh_addralign;
2067 /* Fill the dynamic section with tags describing the address and size of
2068 sections */
2069 static void fill_dynamic(TCCState *s1, struct dyn_inf *dyninf)
2071 Section *dynamic;
2073 dynamic = dyninf->dynamic;
2075 /* put dynamic section entries */
2076 dynamic->data_offset = dyninf->dyn_rel_off;
2077 put_dt(dynamic, DT_HASH, s1->dynsym->hash->sh_addr);
2078 put_dt(dynamic, DT_STRTAB, dyninf->dynstr->sh_addr);
2079 put_dt(dynamic, DT_SYMTAB, s1->dynsym->sh_addr);
2080 put_dt(dynamic, DT_STRSZ, dyninf->dynstr->data_offset);
2081 put_dt(dynamic, DT_SYMENT, sizeof(ElfW(Sym)));
2082 #ifdef TCC_TARGET_X86_64
2083 put_dt(dynamic, DT_RELA, dyninf->rel_addr);
2084 put_dt(dynamic, DT_RELASZ, dyninf->rel_size);
2085 put_dt(dynamic, DT_RELAENT, sizeof(ElfW_Rel));
2086 #else
2087 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
2088 put_dt(dynamic, DT_PLTGOT, s1->got->sh_addr);
2089 put_dt(dynamic, DT_PLTRELSZ, dyninf->rel_size);
2090 put_dt(dynamic, DT_JMPREL, dyninf->rel_addr);
2091 put_dt(dynamic, DT_PLTREL, DT_REL);
2092 put_dt(dynamic, DT_REL, dyninf->bss_addr);
2093 put_dt(dynamic, DT_RELSZ, dyninf->bss_size);
2094 #else
2095 put_dt(dynamic, DT_REL, dyninf->rel_addr);
2096 put_dt(dynamic, DT_RELSZ, dyninf->rel_size);
2097 put_dt(dynamic, DT_RELENT, sizeof(ElfW_Rel));
2098 #endif
2099 #endif
2100 if (s1->do_debug)
2101 put_dt(dynamic, DT_DEBUG, 0);
2102 put_dt(dynamic, DT_NULL, 0);
2105 /* Relocate remaining sections and symbols (that is those not related to
2106 dynamic linking) */
2107 static int final_sections_reloc(TCCState *s1)
2109 int i;
2110 Section *s;
2112 relocate_syms(s1, 0);
2114 if (s1->nb_errors != 0)
2115 return -1;
2117 /* relocate sections */
2118 /* XXX: ignore sections with allocated relocations ? */
2119 for(i = 1; i < s1->nb_sections; i++) {
2120 s = s1->sections[i];
2121 if (s->reloc && s != s1->got)
2122 relocate_section(s1, s);
2125 /* relocate relocation entries if the relocation tables are
2126 allocated in the executable */
2127 for(i = 1; i < s1->nb_sections; i++) {
2128 s = s1->sections[i];
2129 if ((s->sh_flags & SHF_ALLOC) &&
2130 s->sh_type == SHT_RELX) {
2131 relocate_rel(s1, s);
2134 return 0;
2137 /* Create an ELF file on disk.
2138 This function handle ELF specific layout requirements */
2139 static void tcc_output_elf(TCCState *s1, FILE *f, int phnum, ElfW(Phdr) *phdr,
2140 int file_offset, int *sec_order)
2142 int i, shnum, offset, size, file_type;
2143 Section *s;
2144 ElfW(Ehdr) ehdr;
2145 ElfW(Shdr) shdr, *sh;
2147 file_type = s1->output_type;
2148 shnum = s1->nb_sections;
2150 memset(&ehdr, 0, sizeof(ehdr));
2152 if (phnum > 0) {
2153 ehdr.e_phentsize = sizeof(ElfW(Phdr));
2154 ehdr.e_phnum = phnum;
2155 ehdr.e_phoff = sizeof(ElfW(Ehdr));
2158 /* align to 4 */
2159 file_offset = (file_offset + 3) & -4;
2161 /* fill header */
2162 ehdr.e_ident[0] = ELFMAG0;
2163 ehdr.e_ident[1] = ELFMAG1;
2164 ehdr.e_ident[2] = ELFMAG2;
2165 ehdr.e_ident[3] = ELFMAG3;
2166 ehdr.e_ident[4] = ELFCLASSW;
2167 ehdr.e_ident[5] = ELFDATA2LSB;
2168 ehdr.e_ident[6] = EV_CURRENT;
2169 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
2170 ehdr.e_ident[EI_OSABI] = ELFOSABI_FREEBSD;
2171 #endif
2172 #ifdef TCC_TARGET_ARM
2173 #ifdef TCC_ARM_EABI
2174 ehdr.e_ident[EI_OSABI] = 0;
2175 ehdr.e_flags = EF_ARM_EABI_VER4;
2176 if (file_type == TCC_OUTPUT_EXE || file_type == TCC_OUTPUT_DLL)
2177 ehdr.e_flags |= EF_ARM_HASENTRY;
2178 if (s1->float_abi == ARM_HARD_FLOAT)
2179 ehdr.e_flags |= EF_ARM_VFP_FLOAT;
2180 else
2181 ehdr.e_flags |= EF_ARM_SOFT_FLOAT;
2182 #else
2183 ehdr.e_ident[EI_OSABI] = ELFOSABI_ARM;
2184 #endif
2185 #endif
2186 switch(file_type) {
2187 default:
2188 case TCC_OUTPUT_EXE:
2189 ehdr.e_type = ET_EXEC;
2190 ehdr.e_entry = get_elf_sym_addr(s1, "_start", 1);
2191 break;
2192 case TCC_OUTPUT_DLL:
2193 ehdr.e_type = ET_DYN;
2194 ehdr.e_entry = text_section->sh_addr; /* XXX: is it correct ? */
2195 break;
2196 case TCC_OUTPUT_OBJ:
2197 ehdr.e_type = ET_REL;
2198 break;
2200 ehdr.e_machine = EM_TCC_TARGET;
2201 ehdr.e_version = EV_CURRENT;
2202 ehdr.e_shoff = file_offset;
2203 ehdr.e_ehsize = sizeof(ElfW(Ehdr));
2204 ehdr.e_shentsize = sizeof(ElfW(Shdr));
2205 ehdr.e_shnum = shnum;
2206 ehdr.e_shstrndx = shnum - 1;
2208 fwrite(&ehdr, 1, sizeof(ElfW(Ehdr)), f);
2209 fwrite(phdr, 1, phnum * sizeof(ElfW(Phdr)), f);
2210 offset = sizeof(ElfW(Ehdr)) + phnum * sizeof(ElfW(Phdr));
2212 sort_syms(s1, symtab_section);
2213 for(i = 1; i < s1->nb_sections; i++) {
2214 s = s1->sections[sec_order[i]];
2215 if (s->sh_type != SHT_NOBITS) {
2216 if (s->sh_type == SHT_DYNSYM)
2217 patch_dynsym_undef(s1, s);
2218 while (offset < s->sh_offset) {
2219 fputc(0, f);
2220 offset++;
2222 size = s->sh_size;
2223 fwrite(s->data, 1, size, f);
2224 offset += size;
2228 /* output section headers */
2229 while (offset < ehdr.e_shoff) {
2230 fputc(0, f);
2231 offset++;
2234 for(i = 0; i < s1->nb_sections; i++) {
2235 sh = &shdr;
2236 memset(sh, 0, sizeof(ElfW(Shdr)));
2237 s = s1->sections[i];
2238 if (s) {
2239 sh->sh_name = s->sh_name;
2240 sh->sh_type = s->sh_type;
2241 sh->sh_flags = s->sh_flags;
2242 sh->sh_entsize = s->sh_entsize;
2243 sh->sh_info = s->sh_info;
2244 if (s->link)
2245 sh->sh_link = s->link->sh_num;
2246 sh->sh_addralign = s->sh_addralign;
2247 sh->sh_addr = s->sh_addr;
2248 sh->sh_offset = s->sh_offset;
2249 sh->sh_size = s->sh_size;
2251 fwrite(sh, 1, sizeof(ElfW(Shdr)), f);
2255 /* Write an elf, coff or "binary" file */
2256 static int tcc_write_elf_file(TCCState *s1, const char *filename, int phnum,
2257 ElfW(Phdr) *phdr, int file_offset, int *sec_order)
2259 int fd, mode, file_type;
2260 FILE *f;
2262 file_type = s1->output_type;
2263 if (file_type == TCC_OUTPUT_OBJ)
2264 mode = 0666;
2265 else
2266 mode = 0777;
2267 unlink(filename);
2268 fd = open(filename, O_WRONLY | O_CREAT | O_TRUNC | O_BINARY, mode);
2269 if (fd < 0) {
2270 tcc_error_noabort("could not write '%s'", filename);
2271 return -1;
2273 f = fdopen(fd, "wb");
2274 if (s1->verbose)
2275 printf("<- %s\n", filename);
2277 #ifdef TCC_TARGET_COFF
2278 if (s1->output_format == TCC_OUTPUT_FORMAT_COFF)
2279 tcc_output_coff(s1, f);
2280 else
2281 #endif
2282 if (s1->output_format == TCC_OUTPUT_FORMAT_ELF)
2283 tcc_output_elf(s1, f, phnum, phdr, file_offset, sec_order);
2284 else
2285 tcc_output_binary(s1, f, sec_order);
2286 fclose(f);
2288 return 0;
2291 /* Output an elf, coff or binary file */
2292 /* XXX: suppress unneeded sections */
2293 static int elf_output_file(TCCState *s1, const char *filename)
2295 int i, ret, phnum, shnum, file_type, file_offset, *sec_order;
2296 struct dyn_inf dyninf;
2297 ElfW(Phdr) *phdr;
2298 ElfW(Sym) *sym;
2299 Section *strsec, *interp, *dynamic, *dynstr;
2301 file_type = s1->output_type;
2302 s1->nb_errors = 0;
2304 /* if linking, also link in runtime libraries (libc, libgcc, etc.) */
2305 if (file_type != TCC_OUTPUT_OBJ) {
2306 tcc_add_runtime(s1);
2309 phdr = NULL;
2310 sec_order = NULL;
2311 interp = dynamic = dynstr = NULL; /* avoid warning */
2312 dyninf.dyn_rel_off = 0; /* avoid warning */
2314 if (file_type != TCC_OUTPUT_OBJ) {
2315 relocate_common_syms();
2317 tcc_add_linker_symbols(s1);
2319 if (!s1->static_link) {
2320 if (file_type == TCC_OUTPUT_EXE) {
2321 char *ptr;
2322 /* allow override the dynamic loader */
2323 const char *elfint = getenv("LD_SO");
2324 if (elfint == NULL)
2325 elfint = DEFAULT_ELFINTERP(s1);
2326 /* add interpreter section only if executable */
2327 interp = new_section(s1, ".interp", SHT_PROGBITS, SHF_ALLOC);
2328 interp->sh_addralign = 1;
2329 ptr = section_ptr_add(interp, 1 + strlen(elfint));
2330 strcpy(ptr, elfint);
2333 /* add dynamic symbol table */
2334 s1->dynsym = new_symtab(s1, ".dynsym", SHT_DYNSYM, SHF_ALLOC,
2335 ".dynstr",
2336 ".hash", SHF_ALLOC);
2337 dynstr = s1->dynsym->link;
2339 /* add dynamic section */
2340 dynamic = new_section(s1, ".dynamic", SHT_DYNAMIC,
2341 SHF_ALLOC | SHF_WRITE);
2342 dynamic->link = dynstr;
2343 dynamic->sh_entsize = sizeof(ElfW(Dyn));
2345 /* add PLT */
2346 s1->plt = new_section(s1, ".plt", SHT_PROGBITS,
2347 SHF_ALLOC | SHF_EXECINSTR);
2348 s1->plt->sh_entsize = 4;
2350 build_got(s1);
2352 if (file_type == TCC_OUTPUT_EXE) {
2353 bind_exe_dynsyms(s1);
2355 if (s1->nb_errors) {
2356 ret = -1;
2357 goto the_end;
2360 bind_libs_dynsyms(s1);
2361 } else /* shared library case: simply export all global symbols */
2362 export_global_syms(s1);
2364 build_got_entries(s1);
2366 /* add a list of needed dlls */
2367 for(i = 0; i < s1->nb_loaded_dlls; i++) {
2368 DLLReference *dllref = s1->loaded_dlls[i];
2369 if (dllref->level == 0)
2370 put_dt(dynamic, DT_NEEDED, put_elf_str(dynstr, dllref->name));
2373 if (s1->rpath)
2374 put_dt(dynamic, DT_RPATH, put_elf_str(dynstr, s1->rpath));
2376 /* XXX: currently, since we do not handle PIC code, we
2377 must relocate the readonly segments */
2378 if (file_type == TCC_OUTPUT_DLL) {
2379 if (s1->soname)
2380 put_dt(dynamic, DT_SONAME, put_elf_str(dynstr, s1->soname));
2381 put_dt(dynamic, DT_TEXTREL, 0);
2384 if (s1->symbolic)
2385 put_dt(dynamic, DT_SYMBOLIC, 0);
2387 /* add necessary space for other entries */
2388 dyninf.dyn_rel_off = dynamic->data_offset;
2389 dynamic->data_offset += sizeof(ElfW(Dyn)) * EXTRA_RELITEMS;
2390 } else {
2391 /* still need to build got entries in case of static link */
2392 build_got_entries(s1);
2396 /* we add a section for symbols */
2397 strsec = new_section(s1, ".shstrtab", SHT_STRTAB, 0);
2398 put_elf_str(strsec, "");
2400 /* compute number of sections */
2401 shnum = s1->nb_sections;
2403 /* this array is used to reorder sections in the output file */
2404 sec_order = tcc_malloc(sizeof(int) * shnum);
2405 sec_order[0] = 0;
2407 /* compute number of program headers */
2408 switch(file_type) {
2409 default:
2410 case TCC_OUTPUT_OBJ:
2411 phnum = 0;
2412 break;
2413 case TCC_OUTPUT_EXE:
2414 if (!s1->static_link)
2415 phnum = 4 + HAVE_PHDR;
2416 else
2417 phnum = 2;
2418 break;
2419 case TCC_OUTPUT_DLL:
2420 phnum = 3;
2421 break;
2424 /* Allocate strings for section names */
2425 alloc_sec_names(s1, file_type, strsec);
2427 /* allocate program segment headers */
2428 phdr = tcc_mallocz(phnum * sizeof(ElfW(Phdr)));
2430 /* compute section to program header mapping */
2431 file_offset = layout_sections(s1, phdr, phnum, interp, &dyninf, sec_order);
2433 /* Fill remaining program header and finalize relocation related to dynamic
2434 linking. */
2435 if (phnum > 0) {
2436 fill_unloadable_phdr(phdr, phnum, interp, dynamic);
2437 if (dynamic) {
2438 dyninf.dynamic = dynamic;
2439 dyninf.dynstr = dynstr;
2441 fill_dynamic(s1, &dyninf);
2443 /* put in GOT the dynamic section address and relocate PLT */
2444 put32(s1->got->data, dynamic->sh_addr);
2445 if (file_type == TCC_OUTPUT_EXE
2446 #if defined(TCC_OUTPUT_DLL_WITH_PLT)
2447 || file_type == TCC_OUTPUT_DLL
2448 #endif
2450 relocate_plt(s1);
2452 /* relocate symbols in .dynsym now that final addresses are known */
2453 for_each_elem(s1->dynsym, 1, sym, ElfW(Sym)) {
2454 /* relocate to PLT if symbol corresponds to a PLT entry */
2455 if (sym->st_shndx == SHN_UNDEF) {
2456 if (sym->st_value)
2457 sym->st_value += s1->plt->sh_addr;
2458 } else if (sym->st_shndx < SHN_LORESERVE) {
2459 /* do symbol relocation */
2460 sym->st_value += s1->sections[sym->st_shndx]->sh_addr;
2466 /* if building executable or DLL, then relocate each section
2467 except the GOT which is already relocated */
2468 if (file_type != TCC_OUTPUT_OBJ) {
2469 ret = final_sections_reloc(s1);
2470 if (ret)
2471 goto the_end;
2474 /* Perform relocation to GOT or PLT entries */
2475 if (file_type == TCC_OUTPUT_EXE && s1->static_link)
2476 fill_got(s1);
2478 /* Create the ELF file with name 'filename' */
2479 ret = tcc_write_elf_file(s1, filename, phnum, phdr, file_offset, sec_order);
2480 the_end:
2481 tcc_free(s1->symtab_to_dynsym);
2482 tcc_free(sec_order);
2483 tcc_free(phdr);
2484 tcc_free(s1->sym_attrs);
2485 s1->sym_attrs = NULL;
2486 return ret;
2489 LIBTCCAPI int tcc_output_file(TCCState *s, const char *filename)
2491 int ret;
2492 #ifdef TCC_TARGET_PE
2493 if (s->output_type != TCC_OUTPUT_OBJ) {
2494 ret = pe_output_file(s, filename);
2495 } else
2496 #endif
2497 ret = elf_output_file(s, filename);
2498 return ret;
2501 static void *load_data(int fd, unsigned long file_offset, unsigned long size)
2503 void *data;
2505 data = tcc_malloc(size);
2506 lseek(fd, file_offset, SEEK_SET);
2507 read(fd, data, size);
2508 return data;
2511 typedef struct SectionMergeInfo {
2512 Section *s; /* corresponding existing section */
2513 unsigned long offset; /* offset of the new section in the existing section */
2514 uint8_t new_section; /* true if section 's' was added */
2515 uint8_t link_once; /* true if link once section */
2516 } SectionMergeInfo;
2518 /* load an object file and merge it with current files */
2519 /* XXX: handle correctly stab (debug) info */
2520 ST_FUNC int tcc_load_object_file(TCCState *s1,
2521 int fd, unsigned long file_offset)
2523 ElfW(Ehdr) ehdr;
2524 ElfW(Shdr) *shdr, *sh;
2525 int size, i, j, offset, offseti, nb_syms, sym_index, ret;
2526 unsigned char *strsec, *strtab;
2527 int *old_to_new_syms;
2528 char *sh_name, *name;
2529 SectionMergeInfo *sm_table, *sm;
2530 ElfW(Sym) *sym, *symtab;
2531 ElfW_Rel *rel;
2532 Section *s;
2534 int stab_index;
2535 int stabstr_index;
2537 stab_index = stabstr_index = 0;
2539 if (read(fd, &ehdr, sizeof(ehdr)) != sizeof(ehdr))
2540 goto fail1;
2541 if (ehdr.e_ident[0] != ELFMAG0 ||
2542 ehdr.e_ident[1] != ELFMAG1 ||
2543 ehdr.e_ident[2] != ELFMAG2 ||
2544 ehdr.e_ident[3] != ELFMAG3)
2545 goto fail1;
2546 /* test if object file */
2547 if (ehdr.e_type != ET_REL)
2548 goto fail1;
2549 /* test CPU specific stuff */
2550 if (ehdr.e_ident[5] != ELFDATA2LSB ||
2551 ehdr.e_machine != EM_TCC_TARGET) {
2552 fail1:
2553 tcc_error_noabort("invalid object file");
2554 return -1;
2556 /* read sections */
2557 shdr = load_data(fd, file_offset + ehdr.e_shoff,
2558 sizeof(ElfW(Shdr)) * ehdr.e_shnum);
2559 sm_table = tcc_mallocz(sizeof(SectionMergeInfo) * ehdr.e_shnum);
2561 /* load section names */
2562 sh = &shdr[ehdr.e_shstrndx];
2563 strsec = load_data(fd, file_offset + sh->sh_offset, sh->sh_size);
2565 /* load symtab and strtab */
2566 old_to_new_syms = NULL;
2567 symtab = NULL;
2568 strtab = NULL;
2569 nb_syms = 0;
2570 for(i = 1; i < ehdr.e_shnum; i++) {
2571 sh = &shdr[i];
2572 if (sh->sh_type == SHT_SYMTAB) {
2573 if (symtab) {
2574 tcc_error_noabort("object must contain only one symtab");
2575 fail:
2576 ret = -1;
2577 goto the_end;
2579 nb_syms = sh->sh_size / sizeof(ElfW(Sym));
2580 symtab = load_data(fd, file_offset + sh->sh_offset, sh->sh_size);
2581 sm_table[i].s = symtab_section;
2583 /* now load strtab */
2584 sh = &shdr[sh->sh_link];
2585 strtab = load_data(fd, file_offset + sh->sh_offset, sh->sh_size);
2589 /* now examine each section and try to merge its content with the
2590 ones in memory */
2591 for(i = 1; i < ehdr.e_shnum; i++) {
2592 /* no need to examine section name strtab */
2593 if (i == ehdr.e_shstrndx)
2594 continue;
2595 sh = &shdr[i];
2596 sh_name = (char *) strsec + sh->sh_name;
2597 /* ignore sections types we do not handle */
2598 if (sh->sh_type != SHT_PROGBITS &&
2599 sh->sh_type != SHT_RELX &&
2600 #ifdef TCC_ARM_EABI
2601 sh->sh_type != SHT_ARM_EXIDX &&
2602 #endif
2603 sh->sh_type != SHT_NOBITS &&
2604 sh->sh_type != SHT_PREINIT_ARRAY &&
2605 sh->sh_type != SHT_INIT_ARRAY &&
2606 sh->sh_type != SHT_FINI_ARRAY &&
2607 strcmp(sh_name, ".stabstr")
2609 continue;
2610 if (sh->sh_addralign < 1)
2611 sh->sh_addralign = 1;
2612 /* find corresponding section, if any */
2613 for(j = 1; j < s1->nb_sections;j++) {
2614 s = s1->sections[j];
2615 if (!strcmp(s->name, sh_name)) {
2616 if (!strncmp(sh_name, ".gnu.linkonce",
2617 sizeof(".gnu.linkonce") - 1)) {
2618 /* if a 'linkonce' section is already present, we
2619 do not add it again. It is a little tricky as
2620 symbols can still be defined in
2621 it. */
2622 sm_table[i].link_once = 1;
2623 goto next;
2624 } else {
2625 goto found;
2629 /* not found: create new section */
2630 s = new_section(s1, sh_name, sh->sh_type, sh->sh_flags);
2631 /* take as much info as possible from the section. sh_link and
2632 sh_info will be updated later */
2633 s->sh_addralign = sh->sh_addralign;
2634 s->sh_entsize = sh->sh_entsize;
2635 sm_table[i].new_section = 1;
2636 found:
2637 if (sh->sh_type != s->sh_type) {
2638 tcc_error_noabort("invalid section type");
2639 goto fail;
2642 /* align start of section */
2643 offset = s->data_offset;
2645 if (0 == strcmp(sh_name, ".stab")) {
2646 stab_index = i;
2647 goto no_align;
2649 if (0 == strcmp(sh_name, ".stabstr")) {
2650 stabstr_index = i;
2651 goto no_align;
2654 size = sh->sh_addralign - 1;
2655 offset = (offset + size) & ~size;
2656 if (sh->sh_addralign > s->sh_addralign)
2657 s->sh_addralign = sh->sh_addralign;
2658 s->data_offset = offset;
2659 no_align:
2660 sm_table[i].offset = offset;
2661 sm_table[i].s = s;
2662 /* concatenate sections */
2663 size = sh->sh_size;
2664 if (sh->sh_type != SHT_NOBITS) {
2665 unsigned char *ptr;
2666 lseek(fd, file_offset + sh->sh_offset, SEEK_SET);
2667 ptr = section_ptr_add(s, size);
2668 read(fd, ptr, size);
2669 } else {
2670 s->data_offset += size;
2672 next: ;
2675 /* gr relocate stab strings */
2676 if (stab_index && stabstr_index) {
2677 Stab_Sym *a, *b;
2678 unsigned o;
2679 s = sm_table[stab_index].s;
2680 a = (Stab_Sym *)(s->data + sm_table[stab_index].offset);
2681 b = (Stab_Sym *)(s->data + s->data_offset);
2682 o = sm_table[stabstr_index].offset;
2683 while (a < b)
2684 a->n_strx += o, a++;
2687 /* second short pass to update sh_link and sh_info fields of new
2688 sections */
2689 for(i = 1; i < ehdr.e_shnum; i++) {
2690 s = sm_table[i].s;
2691 if (!s || !sm_table[i].new_section)
2692 continue;
2693 sh = &shdr[i];
2694 if (sh->sh_link > 0)
2695 s->link = sm_table[sh->sh_link].s;
2696 if (sh->sh_type == SHT_RELX) {
2697 s->sh_info = sm_table[sh->sh_info].s->sh_num;
2698 /* update backward link */
2699 s1->sections[s->sh_info]->reloc = s;
2702 sm = sm_table;
2704 /* resolve symbols */
2705 old_to_new_syms = tcc_mallocz(nb_syms * sizeof(int));
2707 sym = symtab + 1;
2708 for(i = 1; i < nb_syms; i++, sym++) {
2709 if (sym->st_shndx != SHN_UNDEF &&
2710 sym->st_shndx < SHN_LORESERVE) {
2711 sm = &sm_table[sym->st_shndx];
2712 if (sm->link_once) {
2713 /* if a symbol is in a link once section, we use the
2714 already defined symbol. It is very important to get
2715 correct relocations */
2716 if (ELFW(ST_BIND)(sym->st_info) != STB_LOCAL) {
2717 name = (char *) strtab + sym->st_name;
2718 sym_index = find_elf_sym(symtab_section, name);
2719 if (sym_index)
2720 old_to_new_syms[i] = sym_index;
2722 continue;
2724 /* if no corresponding section added, no need to add symbol */
2725 if (!sm->s)
2726 continue;
2727 /* convert section number */
2728 sym->st_shndx = sm->s->sh_num;
2729 /* offset value */
2730 sym->st_value += sm->offset;
2732 /* add symbol */
2733 name = (char *) strtab + sym->st_name;
2734 sym_index = add_elf_sym(symtab_section, sym->st_value, sym->st_size,
2735 sym->st_info, sym->st_other,
2736 sym->st_shndx, name);
2737 old_to_new_syms[i] = sym_index;
2740 /* third pass to patch relocation entries */
2741 for(i = 1; i < ehdr.e_shnum; i++) {
2742 s = sm_table[i].s;
2743 if (!s)
2744 continue;
2745 sh = &shdr[i];
2746 offset = sm_table[i].offset;
2747 switch(s->sh_type) {
2748 case SHT_RELX:
2749 /* take relocation offset information */
2750 offseti = sm_table[sh->sh_info].offset;
2751 for_each_elem(s, (offset / sizeof(*rel)), rel, ElfW_Rel) {
2752 int type;
2753 unsigned sym_index;
2754 /* convert symbol index */
2755 type = ELFW(R_TYPE)(rel->r_info);
2756 sym_index = ELFW(R_SYM)(rel->r_info);
2757 /* NOTE: only one symtab assumed */
2758 if (sym_index >= nb_syms)
2759 goto invalid_reloc;
2760 sym_index = old_to_new_syms[sym_index];
2761 /* ignore link_once in rel section. */
2762 if (!sym_index && !sm->link_once
2763 #ifdef TCC_TARGET_ARM
2764 && type != R_ARM_V4BX
2765 #endif
2767 invalid_reloc:
2768 tcc_error_noabort("Invalid relocation entry [%2d] '%s' @ %.8x",
2769 i, strsec + sh->sh_name, rel->r_offset);
2770 goto fail;
2772 rel->r_info = ELFW(R_INFO)(sym_index, type);
2773 /* offset the relocation offset */
2774 rel->r_offset += offseti;
2775 #ifdef TCC_TARGET_ARM
2776 /* Jumps and branches from a Thumb code to a PLT entry need
2777 special handling since PLT entries are ARM code.
2778 Unconditional bl instructions referencing PLT entries are
2779 handled by converting these instructions into blx
2780 instructions. Other case of instructions referencing a PLT
2781 entry require to add a Thumb stub before the PLT entry to
2782 switch to ARM mode. We set bit plt_thumb_stub of the
2783 attribute of a symbol to indicate such a case. */
2784 if (type == R_ARM_THM_JUMP24)
2785 alloc_sym_attr(s1, sym_index)->plt_thumb_stub = 1;
2786 #endif
2788 break;
2789 default:
2790 break;
2794 ret = 0;
2795 the_end:
2796 tcc_free(symtab);
2797 tcc_free(strtab);
2798 tcc_free(old_to_new_syms);
2799 tcc_free(sm_table);
2800 tcc_free(strsec);
2801 tcc_free(shdr);
2802 return ret;
2805 typedef struct ArchiveHeader {
2806 char ar_name[16]; /* name of this member */
2807 char ar_date[12]; /* file mtime */
2808 char ar_uid[6]; /* owner uid; printed as decimal */
2809 char ar_gid[6]; /* owner gid; printed as decimal */
2810 char ar_mode[8]; /* file mode, printed as octal */
2811 char ar_size[10]; /* file size, printed as decimal */
2812 char ar_fmag[2]; /* should contain ARFMAG */
2813 } ArchiveHeader;
2815 static int get_be32(const uint8_t *b)
2817 return b[3] | (b[2] << 8) | (b[1] << 16) | (b[0] << 24);
2820 /* load only the objects which resolve undefined symbols */
2821 static int tcc_load_alacarte(TCCState *s1, int fd, int size)
2823 int i, bound, nsyms, sym_index, off, ret;
2824 uint8_t *data;
2825 const char *ar_names, *p;
2826 const uint8_t *ar_index;
2827 ElfW(Sym) *sym;
2829 data = tcc_malloc(size);
2830 if (read(fd, data, size) != size)
2831 goto fail;
2832 nsyms = get_be32(data);
2833 ar_index = data + 4;
2834 ar_names = (char *) ar_index + nsyms * 4;
2836 do {
2837 bound = 0;
2838 for(p = ar_names, i = 0; i < nsyms; i++, p += strlen(p)+1) {
2839 sym_index = find_elf_sym(symtab_section, p);
2840 if(sym_index) {
2841 sym = &((ElfW(Sym) *)symtab_section->data)[sym_index];
2842 if(sym->st_shndx == SHN_UNDEF) {
2843 off = get_be32(ar_index + i * 4) + sizeof(ArchiveHeader);
2844 ++bound;
2845 lseek(fd, off, SEEK_SET);
2846 if(tcc_load_object_file(s1, fd, off) < 0) {
2847 fail:
2848 ret = -1;
2849 goto the_end;
2854 } while(bound);
2855 ret = 0;
2856 the_end:
2857 tcc_free(data);
2858 return ret;
2861 /* load a '.a' file */
2862 ST_FUNC int tcc_load_archive(TCCState *s1, int fd)
2864 ArchiveHeader hdr;
2865 char ar_size[11];
2866 char ar_name[17];
2867 char magic[8];
2868 int size, len, i;
2869 unsigned long file_offset;
2871 /* skip magic which was already checked */
2872 read(fd, magic, sizeof(magic));
2874 for(;;) {
2875 len = read(fd, &hdr, sizeof(hdr));
2876 if (len == 0)
2877 break;
2878 if (len != sizeof(hdr)) {
2879 tcc_error_noabort("invalid archive");
2880 return -1;
2882 memcpy(ar_size, hdr.ar_size, sizeof(hdr.ar_size));
2883 ar_size[sizeof(hdr.ar_size)] = '\0';
2884 size = strtol(ar_size, NULL, 0);
2885 memcpy(ar_name, hdr.ar_name, sizeof(hdr.ar_name));
2886 for(i = sizeof(hdr.ar_name) - 1; i >= 0; i--) {
2887 if (ar_name[i] != ' ')
2888 break;
2890 ar_name[i + 1] = '\0';
2891 file_offset = lseek(fd, 0, SEEK_CUR);
2892 /* align to even */
2893 size = (size + 1) & ~1;
2894 if (!strcmp(ar_name, "/")) {
2895 /* coff symbol table : we handle it */
2896 if(s1->alacarte_link)
2897 return tcc_load_alacarte(s1, fd, size);
2898 } else if (!strcmp(ar_name, "//") ||
2899 !strcmp(ar_name, "__.SYMDEF") ||
2900 !strcmp(ar_name, "__.SYMDEF/") ||
2901 !strcmp(ar_name, "ARFILENAMES/")) {
2902 /* skip symbol table or archive names */
2903 } else {
2904 if (tcc_load_object_file(s1, fd, file_offset) < 0)
2905 return -1;
2907 lseek(fd, file_offset + size, SEEK_SET);
2909 return 0;
2912 #ifndef TCC_TARGET_PE
2913 /* load a DLL and all referenced DLLs. 'level = 0' means that the DLL
2914 is referenced by the user (so it should be added as DT_NEEDED in
2915 the generated ELF file) */
2916 ST_FUNC int tcc_load_dll(TCCState *s1, int fd, const char *filename, int level)
2918 ElfW(Ehdr) ehdr;
2919 ElfW(Shdr) *shdr, *sh, *sh1;
2920 int i, j, nb_syms, nb_dts, sym_bind, ret;
2921 ElfW(Sym) *sym, *dynsym;
2922 ElfW(Dyn) *dt, *dynamic;
2923 unsigned char *dynstr;
2924 const char *name, *soname;
2925 DLLReference *dllref;
2927 read(fd, &ehdr, sizeof(ehdr));
2929 /* test CPU specific stuff */
2930 if (ehdr.e_ident[5] != ELFDATA2LSB ||
2931 ehdr.e_machine != EM_TCC_TARGET) {
2932 tcc_error_noabort("bad architecture");
2933 return -1;
2936 /* read sections */
2937 shdr = load_data(fd, ehdr.e_shoff, sizeof(ElfW(Shdr)) * ehdr.e_shnum);
2939 /* load dynamic section and dynamic symbols */
2940 nb_syms = 0;
2941 nb_dts = 0;
2942 dynamic = NULL;
2943 dynsym = NULL; /* avoid warning */
2944 dynstr = NULL; /* avoid warning */
2945 for(i = 0, sh = shdr; i < ehdr.e_shnum; i++, sh++) {
2946 switch(sh->sh_type) {
2947 case SHT_DYNAMIC:
2948 nb_dts = sh->sh_size / sizeof(ElfW(Dyn));
2949 dynamic = load_data(fd, sh->sh_offset, sh->sh_size);
2950 break;
2951 case SHT_DYNSYM:
2952 nb_syms = sh->sh_size / sizeof(ElfW(Sym));
2953 dynsym = load_data(fd, sh->sh_offset, sh->sh_size);
2954 sh1 = &shdr[sh->sh_link];
2955 dynstr = load_data(fd, sh1->sh_offset, sh1->sh_size);
2956 break;
2957 default:
2958 break;
2962 /* compute the real library name */
2963 soname = tcc_basename(filename);
2965 for(i = 0, dt = dynamic; i < nb_dts; i++, dt++) {
2966 if (dt->d_tag == DT_SONAME) {
2967 soname = (char *) dynstr + dt->d_un.d_val;
2971 /* if the dll is already loaded, do not load it */
2972 for(i = 0; i < s1->nb_loaded_dlls; i++) {
2973 dllref = s1->loaded_dlls[i];
2974 if (!strcmp(soname, dllref->name)) {
2975 /* but update level if needed */
2976 if (level < dllref->level)
2977 dllref->level = level;
2978 ret = 0;
2979 goto the_end;
2983 /* add the dll and its level */
2984 dllref = tcc_mallocz(sizeof(DLLReference) + strlen(soname));
2985 dllref->level = level;
2986 strcpy(dllref->name, soname);
2987 dynarray_add((void ***)&s1->loaded_dlls, &s1->nb_loaded_dlls, dllref);
2989 /* add dynamic symbols in dynsym_section */
2990 for(i = 1, sym = dynsym + 1; i < nb_syms; i++, sym++) {
2991 sym_bind = ELFW(ST_BIND)(sym->st_info);
2992 if (sym_bind == STB_LOCAL)
2993 continue;
2994 name = (char *) dynstr + sym->st_name;
2995 add_elf_sym(s1->dynsymtab_section, sym->st_value, sym->st_size,
2996 sym->st_info, sym->st_other, sym->st_shndx, name);
2999 /* load all referenced DLLs */
3000 for(i = 0, dt = dynamic; i < nb_dts; i++, dt++) {
3001 switch(dt->d_tag) {
3002 case DT_NEEDED:
3003 name = (char *) dynstr + dt->d_un.d_val;
3004 for(j = 0; j < s1->nb_loaded_dlls; j++) {
3005 dllref = s1->loaded_dlls[j];
3006 if (!strcmp(name, dllref->name))
3007 goto already_loaded;
3009 if (tcc_add_dll(s1, name, AFF_REFERENCED_DLL) < 0) {
3010 tcc_error_noabort("referenced dll '%s' not found", name);
3011 ret = -1;
3012 goto the_end;
3014 already_loaded:
3015 break;
3018 ret = 0;
3019 the_end:
3020 tcc_free(dynstr);
3021 tcc_free(dynsym);
3022 tcc_free(dynamic);
3023 tcc_free(shdr);
3024 return ret;
3027 #define LD_TOK_NAME 256
3028 #define LD_TOK_EOF (-1)
3030 /* return next ld script token */
3031 static int ld_next(TCCState *s1, char *name, int name_size)
3033 int c;
3034 char *q;
3036 redo:
3037 switch(ch) {
3038 case ' ':
3039 case '\t':
3040 case '\f':
3041 case '\v':
3042 case '\r':
3043 case '\n':
3044 inp();
3045 goto redo;
3046 case '/':
3047 minp();
3048 if (ch == '*') {
3049 file->buf_ptr = parse_comment(file->buf_ptr);
3050 ch = file->buf_ptr[0];
3051 goto redo;
3052 } else {
3053 q = name;
3054 *q++ = '/';
3055 goto parse_name;
3057 break;
3058 /* case 'a' ... 'z': */
3059 case 'a':
3060 case 'b':
3061 case 'c':
3062 case 'd':
3063 case 'e':
3064 case 'f':
3065 case 'g':
3066 case 'h':
3067 case 'i':
3068 case 'j':
3069 case 'k':
3070 case 'l':
3071 case 'm':
3072 case 'n':
3073 case 'o':
3074 case 'p':
3075 case 'q':
3076 case 'r':
3077 case 's':
3078 case 't':
3079 case 'u':
3080 case 'v':
3081 case 'w':
3082 case 'x':
3083 case 'y':
3084 case 'z':
3085 /* case 'A' ... 'z': */
3086 case 'A':
3087 case 'B':
3088 case 'C':
3089 case 'D':
3090 case 'E':
3091 case 'F':
3092 case 'G':
3093 case 'H':
3094 case 'I':
3095 case 'J':
3096 case 'K':
3097 case 'L':
3098 case 'M':
3099 case 'N':
3100 case 'O':
3101 case 'P':
3102 case 'Q':
3103 case 'R':
3104 case 'S':
3105 case 'T':
3106 case 'U':
3107 case 'V':
3108 case 'W':
3109 case 'X':
3110 case 'Y':
3111 case 'Z':
3112 case '_':
3113 case '\\':
3114 case '.':
3115 case '$':
3116 case '~':
3117 q = name;
3118 parse_name:
3119 for(;;) {
3120 if (!((ch >= 'a' && ch <= 'z') ||
3121 (ch >= 'A' && ch <= 'Z') ||
3122 (ch >= '0' && ch <= '9') ||
3123 strchr("/.-_+=$:\\,~", ch)))
3124 break;
3125 if ((q - name) < name_size - 1) {
3126 *q++ = ch;
3128 minp();
3130 *q = '\0';
3131 c = LD_TOK_NAME;
3132 break;
3133 case CH_EOF:
3134 c = LD_TOK_EOF;
3135 break;
3136 default:
3137 c = ch;
3138 inp();
3139 break;
3141 return c;
3144 static int ld_add_file(TCCState *s1, const char filename[])
3146 int ret;
3148 ret = tcc_add_file_internal(s1, filename, 0);
3149 if (ret)
3150 ret = tcc_add_dll(s1, filename, 0);
3151 return ret;
3154 static inline int new_undef_syms(void)
3156 int ret = 0;
3157 ret = new_undef_sym;
3158 new_undef_sym = 0;
3159 return ret;
3162 static int ld_add_file_list(TCCState *s1, const char *cmd, int as_needed)
3164 char filename[1024], libname[1024];
3165 int t, group, nblibs = 0, ret = 0;
3166 char **libs = NULL;
3168 group = !strcmp(cmd, "GROUP");
3169 if (!as_needed)
3170 new_undef_syms();
3171 t = ld_next(s1, filename, sizeof(filename));
3172 if (t != '(')
3173 expect("(");
3174 t = ld_next(s1, filename, sizeof(filename));
3175 for(;;) {
3176 libname[0] = '\0';
3177 if (t == LD_TOK_EOF) {
3178 tcc_error_noabort("unexpected end of file");
3179 ret = -1;
3180 goto lib_parse_error;
3181 } else if (t == ')') {
3182 break;
3183 } else if (t == '-') {
3184 t = ld_next(s1, filename, sizeof(filename));
3185 if ((t != LD_TOK_NAME) || (filename[0] != 'l')) {
3186 tcc_error_noabort("library name expected");
3187 ret = -1;
3188 goto lib_parse_error;
3190 pstrcpy(libname, sizeof libname, &filename[1]);
3191 if (s1->static_link) {
3192 snprintf(filename, sizeof filename, "lib%s.a", libname);
3193 } else {
3194 snprintf(filename, sizeof filename, "lib%s.so", libname);
3196 } else if (t != LD_TOK_NAME) {
3197 tcc_error_noabort("filename expected");
3198 ret = -1;
3199 goto lib_parse_error;
3201 if (!strcmp(filename, "AS_NEEDED")) {
3202 ret = ld_add_file_list(s1, cmd, 1);
3203 if (ret)
3204 goto lib_parse_error;
3205 } else {
3206 /* TODO: Implement AS_NEEDED support. Ignore it for now */
3207 if (!as_needed) {
3208 ret = ld_add_file(s1, filename);
3209 if (ret)
3210 goto lib_parse_error;
3211 if (group) {
3212 /* Add the filename *and* the libname to avoid future conversions */
3213 dynarray_add((void ***) &libs, &nblibs, tcc_strdup(filename));
3214 if (libname[0] != '\0')
3215 dynarray_add((void ***) &libs, &nblibs, tcc_strdup(libname));
3219 t = ld_next(s1, filename, sizeof(filename));
3220 if (t == ',') {
3221 t = ld_next(s1, filename, sizeof(filename));
3224 if (group && !as_needed) {
3225 while (new_undef_syms()) {
3226 int i;
3228 for (i = 0; i < nblibs; i ++)
3229 ld_add_file(s1, libs[i]);
3232 lib_parse_error:
3233 dynarray_reset(&libs, &nblibs);
3234 return ret;
3237 /* interpret a subset of GNU ldscripts to handle the dummy libc.so
3238 files */
3239 ST_FUNC int tcc_load_ldscript(TCCState *s1)
3241 char cmd[64];
3242 char filename[1024];
3243 int t, ret;
3245 ch = file->buf_ptr[0];
3246 ch = handle_eob();
3247 for(;;) {
3248 t = ld_next(s1, cmd, sizeof(cmd));
3249 if (t == LD_TOK_EOF)
3250 return 0;
3251 else if (t != LD_TOK_NAME)
3252 return -1;
3253 if (!strcmp(cmd, "INPUT") ||
3254 !strcmp(cmd, "GROUP")) {
3255 ret = ld_add_file_list(s1, cmd, 0);
3256 if (ret)
3257 return ret;
3258 } else if (!strcmp(cmd, "OUTPUT_FORMAT") ||
3259 !strcmp(cmd, "TARGET")) {
3260 /* ignore some commands */
3261 t = ld_next(s1, cmd, sizeof(cmd));
3262 if (t != '(')
3263 expect("(");
3264 for(;;) {
3265 t = ld_next(s1, filename, sizeof(filename));
3266 if (t == LD_TOK_EOF) {
3267 tcc_error_noabort("unexpected end of file");
3268 return -1;
3269 } else if (t == ')') {
3270 break;
3273 } else {
3274 return -1;
3277 return 0;
3279 #endif /* !TCC_TARGET_PE */