this now sorts the symbols properly (local, global defined, undefined; the latter two by name), marks the three ranges within LC_DYSYMTAB, generates a __got section (non-lazy pointers) and slots for relocations which need them, and the indirect symbol mapping for them. This doesn't yet deal with undefined symbols. But it means compared to last example now this also works, i.e. read access to _global_ data: % cat simple3.c int loc = 42; int _start(void) { return loc - 42; }
Unified · Split
| @@ -214,7 +214,9 @@ struct macho { · 213 unmodified lines | ||
| 214 | 214 | int machosect; |
| 215 | 215 | } sk_to_sect[sk_last]; |
| 216 | 216 | int *elfsectomacho; |
| 217 | − Section *linkedit, *symtab, *strtab, *wdata; | |
| 217 | + int *e2msym; | |
| 218 | + Section *linkedit, *symtab, *strtab, *wdata, *indirsyms; | |
| 219 | + uint32_t ilocal, iextdef, iundef; | |
| 218 | 220 | }; |
| 219 | 221 | |
| 220 | 222 | #define SHT_LINKEDIT (SHT_LOOS + 42) |
| @@ -289,25 +291,78 @@ static void * add_dylib(struct macho *mo, char *name) · 68 unmodified lines | ||
| 289 | 291 | return add_lc(mo, lc); |
| 290 | 292 | } |
| 291 | 293 | |
| 292 | −static int check_symbols(TCCState *s1) | |
| 294 | +static void check_relocs(TCCState *s1, struct macho *mo) | |
| 295 | +{ | |
| 296 | + Section *s; | |
| 297 | + ElfW_Rel *rel; | |
| 298 | + ElfW(Sym) *sym; | |
| 299 | + int i, type, gotplt_entry, sym_index; | |
| 300 | + struct sym_attr *attr; | |
| 301 | + | |
| 302 | + s1->got = new_section(s1, ".got", SHT_PROGBITS, SHF_ALLOC | SHF_WRITE); | |
| 303 | + mo->indirsyms = new_section(s1, "LEINDIR", SHT_LINKEDIT, SHF_ALLOC | SHF_WRITE); | |
| 304 | + for (i = 1; i < s1->nb_sections; i++) { | |
| 305 | + s = s1->sections[i]; | |
| 306 | + if (s->sh_type != SHT_RELX) | |
| 307 | + continue; | |
| 308 | + for_each_elem(s, 0, rel, ElfW_Rel) { | |
| 309 | + type = ELFW(R_TYPE)(rel->r_info); | |
| 310 | + gotplt_entry = gotplt_entry_type(type); | |
| 311 | + /* We generate a non-lazy pointer for used undefined symbols | |
| 312 | + and for defined symbols that must have a place for their | |
| 313 | + address due to codegen (i.e. a reloc requiring a got slot). */ | |
| 314 | + sym_index = ELFW(R_SYM)(rel->r_info); | |
| 315 | + sym = &((ElfW(Sym) *)symtab_section->data)[sym_index]; | |
| 316 | + if (sym->st_shndx == SHN_UNDEF | |
| 317 | + || gotplt_entry == ALWAYS_GOTPLT_ENTRY) { | |
| 318 | + attr = get_sym_attr(s1, sym_index, 1); | |
| 319 | + if (!attr->plt_offset) { | |
| 320 | + uint32_t *pi = section_ptr_add(mo->indirsyms, sizeof(*pi)); | |
| 321 | + attr->got_offset = s1->got->data_offset; | |
| 322 | + attr->plt_offset = 1; /* used as flag */ | |
| 323 | + section_ptr_add(s1->got, PTR_SIZE); | |
| 324 | + *pi = mo->e2msym[sym_index]; | |
| 325 | + } | |
| 326 | + } | |
| 327 | + } | |
| 328 | + } | |
| 329 | +} | |
| 330 | + | |
| 331 | +static int check_symbols(TCCState *s1, struct macho *mo) | |
| 293 | 332 | { |
| 294 | 333 | int sym_index, sym_end; |
| 295 | 334 | int ret = 0; |
| 296 | 335 | |
| 336 | + mo->ilocal = mo->iextdef = mo->iundef = -1; | |
| 297 | 337 | sym_end = symtab_section->data_offset / sizeof(ElfW(Sym)); |
| 298 | 338 | for (sym_index = 1; sym_index < sym_end; ++sym_index) { |
| 299 | − ElfW(Sym) *sym = (ElfW(Sym) *)symtab_section->data + sym_index; | |
| 339 | + int elf_index = ((struct nlist_64 *)mo->symtab->data + sym_index - 1)->n_value; | |
| 340 | + ElfW(Sym) *sym = (ElfW(Sym) *)symtab_section->data + elf_index; | |
| 300 | 341 | const char *name = (char*)symtab_section->link->data + sym->st_name; |
| 301 | 342 | unsigned type = ELFW(ST_TYPE)(sym->st_info); |
| 302 | 343 | unsigned bind = ELFW(ST_BIND)(sym->st_info); |
| 303 | 344 | unsigned vis = ELFW(ST_VISIBILITY)(sym->st_other); |
| 304 | 345 | |
| 305 | − printf("%4d: %09llx %4d %4d %4d %3d %s\n", | |
| 306 | − sym_index, sym->st_value, | |
| 346 | + printf("%4d (%4d): %09llx %4d %4d %4d %3d %s\n", | |
| 347 | + sym_index, elf_index, sym->st_value, | |
| 307 | 348 | type, bind, vis, sym->st_shndx, name); |
| 308 | − if (sym->st_shndx == SHN_UNDEF) { | |
| 349 | + if (bind == STB_LOCAL) { | |
| 350 | + if (mo->ilocal == -1) | |
| 351 | + mo->ilocal = sym_index - 1; | |
| 352 | + if (mo->iextdef != -1 || mo->iundef != -1) | |
| 353 | + tcc_error("local syms after global ones"); | |
| 354 | + } else if (sym->st_shndx != SHN_UNDEF) { | |
| 355 | + if (mo->iextdef == -1) | |
| 356 | + mo->iextdef = sym_index - 1; | |
| 357 | + if (mo->iundef != -1) | |
| 358 | + tcc_error("external defined symbol after undefined"); | |
| 359 | + } else if (sym->st_shndx == SHN_UNDEF) { | |
| 360 | + if (mo->iundef == -1) | |
| 361 | + mo->iundef = sym_index - 1; | |
| 309 | 362 | if (ELFW(ST_BIND)(sym->st_info) == STB_WEAK) |
| 310 | 363 | continue; |
| 364 | + if (get_sym_attr(s1, elf_index, 0)) | |
| 365 | + continue; | |
| 311 | 366 | tcc_error_noabort("undefined symbol '%s'", name); |
| 312 | 367 | ret = -1; |
| 313 | 368 | } |
| @@ -315,12 +370,11 @@ static int check_symbols(TCCState *s1) · 1 unmodified line | ||
| 315 | 370 | return ret; |
| 316 | 371 | } |
| 317 | 372 | |
| 318 | −static void convert_symbol(TCCState *s1, struct macho *mo, struct nlist_64 *pn, ElfSym *sym) | |
| 373 | +static void convert_symbol(TCCState *s1, struct macho *mo, struct nlist_64 *pn) | |
| 319 | 374 | { |
| 320 | 375 | struct nlist_64 n = *pn; |
| 376 | + ElfSym *sym = (ElfW(Sym) *)symtab_section->data + pn->n_value; | |
| 321 | 377 | const char *name = (char*)symtab_section->link->data + sym->st_name; |
| 322 | − if (sym != (ElfW(Sym) *)symtab_section->data + pn->n_value) | |
| 323 | − tcc_error("blaeh"); | |
| 324 | 378 | switch(ELFW(ST_TYPE)(sym->st_info)) { |
| 325 | 379 | case STT_NOTYPE: |
| 326 | 380 | case STT_OBJECT: |
| @@ -361,9 +415,33 @@ static void convert_symbols(TCCState *s1, struct macho *mo) · 34 unmodified lines | ||
| 361 | 415 | |
| 362 | 416 | sym_end = symtab_section->data_offset / sizeof(ElfW(Sym)); |
| 363 | 417 | for (sym_index = 1; sym_index < sym_end; ++sym_index) { |
| 364 | − ElfW(Sym) *sym = (ElfW(Sym) *)symtab_section->data + sym_index; | |
| 365 | − convert_symbol(s1, mo, pn + sym_index - 1, sym); | |
| 418 | + convert_symbol(s1, mo, pn + sym_index - 1); | |
| 419 | + } | |
| 420 | +} | |
| 421 | + | |
| 422 | +static int machosymcmp(const void *_a, const void *_b) | |
| 423 | +{ | |
| 424 | + TCCState *s1 = tcc_state; | |
| 425 | + int ea = ((struct nlist_64 *)_a)->n_value; | |
| 426 | + int eb = ((struct nlist_64 *)_b)->n_value; | |
| 427 | + ElfSym *sa = (ElfSym *)symtab_section->data + ea; | |
| 428 | + ElfSym *sb = (ElfSym *)symtab_section->data + eb; | |
| 429 | + int r; | |
| 430 | + r = (ELFW(ST_BIND)(sb->st_info) == STB_LOCAL) | |
| 431 | + - (ELFW(ST_BIND)(sa->st_info) == STB_LOCAL); | |
| 432 | + if (r) | |
| 433 | + return r; | |
| 434 | + r = (sb->st_shndx == SHN_UNDEF) - (sa->st_shndx == SHN_UNDEF); | |
| 435 | + if (r) | |
| 436 | + return r; | |
| 437 | + if (ELFW(ST_BIND)(sa->st_info) != STB_LOCAL) { | |
| 438 | + const char * na = (char*)symtab_section->link->data + sa->st_name; | |
| 439 | + const char * nb = (char*)symtab_section->link->data + sb->st_name; | |
| 440 | + r = strcmp(na, nb); | |
| 441 | + if (r) | |
| 442 | + return r; | |
| 366 | 443 | } |
| 444 | + return ea - eb; | |
| 367 | 445 | } |
| 368 | 446 | |
| 369 | 447 | static void create_symtab(TCCState *s1, struct macho *mo) |
| @@ -373,7 +451,7 @@ static void create_symtab(TCCState *s1, struct macho *mo) · 3 unmodified lines | ||
| 373 | 451 | |
| 374 | 452 | mo->symtab = new_section(s1, "LESYMTAB", SHT_LINKEDIT, SHF_ALLOC | SHF_WRITE); |
| 375 | 453 | mo->strtab = new_section(s1, "LESTRTAB", SHT_LINKEDIT, SHF_ALLOC | SHF_WRITE); |
| 376 | − put_elf_str(mo->strtab, " "); /* Mach-O start strtab with a space */ | |
| 454 | + put_elf_str(mo->strtab, " "); /* Mach-O starts strtab with a space */ | |
| 377 | 455 | sym_end = symtab_section->data_offset / sizeof(ElfW(Sym)); |
| 378 | 456 | pn = section_ptr_add(mo->symtab, sizeof(*pn) * (sym_end - 1)); |
| 379 | 457 | for (sym_index = 1; sym_index < sym_end; ++sym_index) { |
| @@ -382,6 +460,14 @@ static void create_symtab(TCCState *s1, struct macho *mo) · 2 unmodified lines | ||
| 382 | 460 | pn[sym_index - 1].n_strx = put_elf_str(mo->strtab, name); |
| 383 | 461 | pn[sym_index - 1].n_value = sym_index; |
| 384 | 462 | } |
| 463 | + tcc_enter_state(s1); /* qsort needs global state */ | |
| 464 | + qsort(pn, sym_end - 1, sizeof(*pn), machosymcmp); | |
| 465 | + tcc_exit_state(); | |
| 466 | + mo->e2msym = tcc_malloc(sym_end * sizeof(*mo->e2msym)); | |
| 467 | + mo->e2msym[0] = -1; | |
| 468 | + for (sym_index = 1; sym_index < sym_end; ++sym_index) { | |
| 469 | + mo->e2msym[pn[sym_index - 1].n_value] = sym_index - 1; | |
| 470 | + } | |
| 385 | 471 | } |
| 386 | 472 | |
| 387 | 473 | struct { |
| @@ -391,6 +477,7 @@ struct { · 3 unmodified lines | ||
| 391 | 477 | } skinfo[sk_last] = { |
| 392 | 478 | [sk_text] = { 1, 0x80000400, "__text" }, |
| 393 | 479 | [sk_ro_data] = { 1, 0, "__rodata" }, |
| 480 | + [sk_nl_ptr] = { 2, 6, "__got" }, /* S_NON_LAZY_SYMBOL_POINTERS */ | |
| 394 | 481 | [sk_rw_data] = { 2, 0, "__data" }, |
| 395 | 482 | [sk_bss] = { 2, 1, "__bss" }, |
| 396 | 483 | [sk_linkedit] = { 3, 0, NULL }, |
| @@ -453,8 +540,6 @@ static void collect_sections(TCCState *s1, struct macho *mo) · 56 unmodified lines | ||
| 453 | 540 | /* LINKEDIT can't be empty (XXX remove once we have symbol table) */ |
| 454 | 541 | section_ptr_add(mo->linkedit, 256); |
| 455 | 542 | |
| 456 | − create_symtab(s1, mo); | |
| 457 | − | |
| 458 | 543 | /* dyld requires a writable segment, but ignores zero-sized segments |
| 459 | 544 | for this, so force to have some data. */ |
| 460 | 545 | mo->wdata = new_section(s1, " wdata", SHT_PROGBITS, SHF_ALLOC | SHF_WRITE); |
| @@ -475,7 +560,9 @@ static void collect_sections(TCCState *s1, struct macho *mo) · 14 unmodified lines | ||
| 475 | 560 | case SHT_RELX: sk = sk_discard; break; |
| 476 | 561 | case SHT_LINKEDIT: sk = sk_linkedit; break; |
| 477 | 562 | case SHT_PROGBITS: |
| 478 | − if (flags & SHF_EXECINSTR) | |
| 563 | + if (s == s1->got) | |
| 564 | + sk = sk_nl_ptr; | |
| 565 | + else if (flags & SHF_EXECINSTR) | |
| 479 | 566 | sk = sk_text; |
| 480 | 567 | else if (flags & SHF_WRITE) |
| 481 | 568 | sk = sk_rw_data; |
| @@ -584,6 +671,15 @@ static void collect_sections(TCCState *s1, struct macho *mo) · 102 unmodified lines | ||
| 584 | 671 | symlc->nsyms = mo->symtab->data_offset / sizeof(struct nlist_64); |
| 585 | 672 | symlc->stroff = mo->strtab->sh_offset; |
| 586 | 673 | symlc->strsize = mo->strtab->data_offset; |
| 674 | + | |
| 675 | + dysymlc->iundefsym = mo->iundef == -1 ? symlc->nsyms : mo->iundef; | |
| 676 | + dysymlc->iextdefsym = mo->iextdef == -1 ? dysymlc->iundefsym : mo->iextdef; | |
| 677 | + dysymlc->ilocalsym = mo->ilocal == -1 ? dysymlc->iextdefsym : mo->ilocal; | |
| 678 | + dysymlc->nlocalsym = dysymlc->iextdefsym - dysymlc->ilocalsym; | |
| 679 | + dysymlc->nextdefsym = dysymlc->iundefsym - dysymlc->iextdefsym; | |
| 680 | + dysymlc->nundefsym = symlc->nsyms - dysymlc->iundefsym; | |
| 681 | + dysymlc->indirectsymoff = mo->indirsyms->sh_offset; | |
| 682 | + dysymlc->nindirectsyms = mo->indirsyms->data_offset / sizeof(uint32_t); | |
| 587 | 683 | } |
| 588 | 684 | |
| 589 | 685 | static void macho_write(TCCState *s1, struct macho *mo, FILE *fp) |
| @@ -659,7 +755,9 @@ ST_FUNC int macho_output_file(TCCState *s1, const char *filename) · 69 unmodified lines | ||
| 659 | 755 | printf("<- %s\n", filename); |
| 660 | 756 | |
| 661 | 757 | resolve_common_syms(s1); |
| 662 | − ret = check_symbols(s1); | |
| 758 | + create_symtab(s1, &mo); | |
| 759 | + check_relocs(s1, &mo); | |
| 760 | + ret = check_symbols(s1, &mo); | |
| 663 | 761 | if (!ret) { |
| 664 | 762 | int i; |
| 665 | 763 | Section *s; |