]>
Commit | Line | Data |
---|---|---|
7d13299d FB |
1 | /* |
2 | * Generic Dynamic compiler generator | |
3 | * | |
4 | * Copyright (c) 2003 Fabrice Bellard | |
5 | * | |
6 | * This program is free software; you can redistribute it and/or modify | |
7 | * it under the terms of the GNU General Public License as published by | |
8 | * the Free Software Foundation; either version 2 of the License, or | |
9 | * (at your option) any later version. | |
10 | * | |
11 | * This program 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 | |
14 | * GNU General Public License for more details. | |
15 | * | |
16 | * You should have received a copy of the GNU General Public License | |
17 | * along with this program; if not, write to the Free Software | |
18 | * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. | |
19 | */ | |
367e86e8 FB |
20 | #include <stdlib.h> |
21 | #include <stdio.h> | |
22 | #include <stdarg.h> | |
23 | #include <inttypes.h> | |
24 | #include <elf.h> | |
25 | #include <unistd.h> | |
26 | #include <fcntl.h> | |
27 | ||
28 | #include "thunk.h" | |
29 | ||
30 | /* all dynamically generated functions begin with this code */ | |
31 | #define OP_PREFIX "op" | |
32 | ||
33 | int elf_must_swap(Elf32_Ehdr *h) | |
34 | { | |
35 | union { | |
36 | uint32_t i; | |
37 | uint8_t b[4]; | |
38 | } swaptest; | |
39 | ||
40 | swaptest.i = 1; | |
41 | return (h->e_ident[EI_DATA] == ELFDATA2MSB) != | |
42 | (swaptest.b[0] == 0); | |
43 | } | |
44 | ||
45 | void swab16s(uint16_t *p) | |
46 | { | |
47 | *p = bswap16(*p); | |
48 | } | |
49 | ||
50 | void swab32s(uint32_t *p) | |
51 | { | |
52 | *p = bswap32(*p); | |
53 | } | |
54 | ||
55 | void swab64s(uint32_t *p) | |
56 | { | |
57 | *p = bswap64(*p); | |
58 | } | |
59 | ||
60 | void elf_swap_ehdr(Elf32_Ehdr *h) | |
61 | { | |
62 | swab16s(&h->e_type); /* Object file type */ | |
63 | swab16s(&h-> e_machine); /* Architecture */ | |
64 | swab32s(&h-> e_version); /* Object file version */ | |
65 | swab32s(&h-> e_entry); /* Entry point virtual address */ | |
66 | swab32s(&h-> e_phoff); /* Program header table file offset */ | |
67 | swab32s(&h-> e_shoff); /* Section header table file offset */ | |
68 | swab32s(&h-> e_flags); /* Processor-specific flags */ | |
69 | swab16s(&h-> e_ehsize); /* ELF header size in bytes */ | |
70 | swab16s(&h-> e_phentsize); /* Program header table entry size */ | |
71 | swab16s(&h-> e_phnum); /* Program header table entry count */ | |
72 | swab16s(&h-> e_shentsize); /* Section header table entry size */ | |
73 | swab16s(&h-> e_shnum); /* Section header table entry count */ | |
74 | swab16s(&h-> e_shstrndx); /* Section header string table index */ | |
75 | } | |
76 | ||
77 | void elf_swap_shdr(Elf32_Shdr *h) | |
78 | { | |
79 | swab32s(&h-> sh_name); /* Section name (string tbl index) */ | |
80 | swab32s(&h-> sh_type); /* Section type */ | |
81 | swab32s(&h-> sh_flags); /* Section flags */ | |
82 | swab32s(&h-> sh_addr); /* Section virtual addr at execution */ | |
83 | swab32s(&h-> sh_offset); /* Section file offset */ | |
84 | swab32s(&h-> sh_size); /* Section size in bytes */ | |
85 | swab32s(&h-> sh_link); /* Link to another section */ | |
86 | swab32s(&h-> sh_info); /* Additional section information */ | |
87 | swab32s(&h-> sh_addralign); /* Section alignment */ | |
88 | swab32s(&h-> sh_entsize); /* Entry size if section holds table */ | |
89 | } | |
90 | ||
91 | void elf_swap_phdr(Elf32_Phdr *h) | |
92 | { | |
93 | swab32s(&h->p_type); /* Segment type */ | |
94 | swab32s(&h->p_offset); /* Segment file offset */ | |
95 | swab32s(&h->p_vaddr); /* Segment virtual address */ | |
96 | swab32s(&h->p_paddr); /* Segment physical address */ | |
97 | swab32s(&h->p_filesz); /* Segment size in file */ | |
98 | swab32s(&h->p_memsz); /* Segment size in memory */ | |
99 | swab32s(&h->p_flags); /* Segment flags */ | |
100 | swab32s(&h->p_align); /* Segment alignment */ | |
101 | } | |
102 | ||
103 | int do_swap; | |
104 | int e_machine; | |
105 | ||
106 | uint16_t get16(uint16_t *p) | |
107 | { | |
108 | uint16_t val; | |
109 | val = *p; | |
110 | if (do_swap) | |
111 | val = bswap16(val); | |
112 | return val; | |
113 | } | |
114 | ||
115 | uint32_t get32(uint32_t *p) | |
116 | { | |
117 | uint32_t val; | |
118 | val = *p; | |
119 | if (do_swap) | |
120 | val = bswap32(val); | |
121 | return val; | |
122 | } | |
123 | ||
124 | void put16(uint16_t *p, uint16_t val) | |
125 | { | |
126 | if (do_swap) | |
127 | val = bswap16(val); | |
128 | *p = val; | |
129 | } | |
130 | ||
131 | void put32(uint32_t *p, uint32_t val) | |
132 | { | |
133 | if (do_swap) | |
134 | val = bswap32(val); | |
135 | *p = val; | |
136 | } | |
137 | ||
138 | void __attribute__((noreturn)) error(const char *fmt, ...) | |
139 | { | |
140 | va_list ap; | |
141 | va_start(ap, fmt); | |
142 | fprintf(stderr, "dyngen: "); | |
143 | vfprintf(stderr, fmt, ap); | |
144 | fprintf(stderr, "\n"); | |
145 | va_end(ap); | |
146 | exit(1); | |
147 | } | |
148 | ||
149 | ||
150 | Elf32_Shdr *find_elf_section(Elf32_Shdr *shdr, int shnum, const char *shstr, | |
151 | const char *name) | |
152 | { | |
153 | int i; | |
154 | const char *shname; | |
155 | Elf32_Shdr *sec; | |
156 | ||
157 | for(i = 0; i < shnum; i++) { | |
158 | sec = &shdr[i]; | |
159 | if (!sec->sh_name) | |
160 | continue; | |
161 | shname = shstr + sec->sh_name; | |
162 | if (!strcmp(shname, name)) | |
163 | return sec; | |
164 | } | |
165 | return NULL; | |
166 | } | |
167 | ||
168 | void *load_data(int fd, long offset, unsigned int size) | |
169 | { | |
170 | char *data; | |
171 | ||
172 | data = malloc(size); | |
173 | if (!data) | |
174 | return NULL; | |
175 | lseek(fd, offset, SEEK_SET); | |
176 | if (read(fd, data, size) != size) { | |
177 | free(data); | |
178 | return NULL; | |
179 | } | |
180 | return data; | |
181 | } | |
182 | ||
183 | int strstart(const char *str, const char *val, const char **ptr) | |
184 | { | |
185 | const char *p, *q; | |
186 | p = str; | |
187 | q = val; | |
188 | while (*q != '\0') { | |
189 | if (*p != *q) | |
190 | return 0; | |
191 | p++; | |
192 | q++; | |
193 | } | |
194 | if (ptr) | |
195 | *ptr = p; | |
196 | return 1; | |
197 | } | |
198 | ||
199 | #define MAX_ARGS 3 | |
200 | ||
201 | /* generate op code */ | |
202 | void gen_code(const char *name, unsigned long offset, unsigned long size, | |
203 | FILE *outfile, uint8_t *text, void *relocs, int nb_relocs, int reloc_sh_type, | |
204 | Elf32_Sym *symtab, char *strtab) | |
205 | { | |
206 | int copy_size = 0; | |
207 | uint8_t *p_start, *p_end; | |
208 | int nb_args, i; | |
209 | uint8_t args_present[MAX_ARGS]; | |
210 | const char *sym_name, *p; | |
211 | ||
212 | /* compute exact size excluding return instruction */ | |
213 | p_start = text + offset; | |
214 | p_end = p_start + size; | |
215 | switch(e_machine) { | |
216 | case EM_386: | |
217 | { | |
218 | uint8_t *p; | |
219 | p = p_end - 1; | |
367e86e8 FB |
220 | if (p == p_start) |
221 | error("empty code for %s", name); | |
4b74fe1f FB |
222 | if (p[0] != 0xc3) |
223 | error("ret expected at the end of %s", name); | |
367e86e8 FB |
224 | copy_size = p - p_start; |
225 | } | |
226 | break; | |
227 | case EM_PPC: | |
228 | { | |
229 | uint8_t *p; | |
230 | p = (void *)(p_end - 4); | |
231 | /* find ret */ | |
232 | while (p > p_start && get32((uint32_t *)p) != 0x4e800020) | |
233 | p -= 4; | |
234 | /* skip double ret */ | |
235 | if (p > p_start && get32((uint32_t *)(p - 4)) == 0x4e800020) | |
236 | p -= 4; | |
237 | if (p == p_start) | |
238 | error("empty code for %s", name); | |
239 | copy_size = p - p_start; | |
240 | } | |
241 | break; | |
242 | default: | |
243 | error("unsupported CPU (%d)", e_machine); | |
244 | } | |
245 | ||
246 | /* compute the number of arguments by looking at the relocations */ | |
247 | for(i = 0;i < MAX_ARGS; i++) | |
248 | args_present[i] = 0; | |
249 | ||
250 | if (reloc_sh_type == SHT_REL) { | |
251 | Elf32_Rel *rel; | |
252 | int n; | |
253 | for(i = 0, rel = relocs;i < nb_relocs; i++, rel++) { | |
254 | if (rel->r_offset >= offset && rel->r_offset < offset + copy_size) { | |
255 | sym_name = strtab + symtab[ELF32_R_SYM(rel->r_info)].st_name; | |
256 | if (strstart(sym_name, "__op_param", &p)) { | |
257 | n = strtoul(p, NULL, 10); | |
258 | if (n >= MAX_ARGS) | |
259 | error("too many arguments in %s", name); | |
260 | args_present[n - 1] = 1; | |
927f621e FB |
261 | } else { |
262 | fprintf(outfile, "extern char %s;\n", sym_name); | |
367e86e8 FB |
263 | } |
264 | } | |
265 | } | |
266 | } else { | |
267 | Elf32_Rela *rel; | |
268 | int n; | |
269 | for(i = 0, rel = relocs;i < nb_relocs; i++, rel++) { | |
270 | if (rel->r_offset >= offset && rel->r_offset < offset + copy_size) { | |
271 | sym_name = strtab + symtab[ELF32_R_SYM(rel->r_info)].st_name; | |
272 | if (strstart(sym_name, "__op_param", &p)) { | |
273 | n = strtoul(p, NULL, 10); | |
274 | if (n >= MAX_ARGS) | |
275 | error("too many arguments in %s", name); | |
276 | args_present[n - 1] = 1; | |
927f621e FB |
277 | } else { |
278 | fprintf(outfile, "extern char %s;\n", sym_name); | |
367e86e8 FB |
279 | } |
280 | } | |
281 | } | |
282 | } | |
283 | ||
284 | nb_args = 0; | |
285 | while (nb_args < MAX_ARGS && args_present[nb_args]) | |
286 | nb_args++; | |
287 | for(i = nb_args; i < MAX_ARGS; i++) { | |
288 | if (args_present[i]) | |
289 | error("inconsistent argument numbering in %s", name); | |
290 | } | |
291 | ||
292 | /* output C code */ | |
293 | fprintf(outfile, "extern void %s();\n", name); | |
294 | fprintf(outfile, "static inline void gen_%s(", name); | |
295 | if (nb_args == 0) { | |
296 | fprintf(outfile, "void"); | |
297 | } else { | |
298 | for(i = 0; i < nb_args; i++) { | |
299 | if (i != 0) | |
300 | fprintf(outfile, ", "); | |
301 | fprintf(outfile, "long param%d", i + 1); | |
302 | } | |
303 | } | |
304 | fprintf(outfile, ")\n"); | |
305 | fprintf(outfile, "{\n"); | |
306 | fprintf(outfile, " memcpy(gen_code_ptr, &%s, %d);\n", name, copy_size); | |
307 | ||
308 | /* patch relocations */ | |
309 | switch(e_machine) { | |
310 | case EM_386: | |
311 | { | |
312 | Elf32_Rel *rel; | |
313 | char name[256]; | |
314 | int type; | |
315 | long addend; | |
316 | for(i = 0, rel = relocs;i < nb_relocs; i++, rel++) { | |
317 | if (rel->r_offset >= offset && rel->r_offset < offset + copy_size) { | |
318 | sym_name = strtab + symtab[ELF32_R_SYM(rel->r_info)].st_name; | |
319 | if (strstart(sym_name, "__op_param", &p)) { | |
320 | snprintf(name, sizeof(name), "param%s", p); | |
321 | } else { | |
322 | snprintf(name, sizeof(name), "(long)(&%s)", sym_name); | |
323 | } | |
324 | type = ELF32_R_TYPE(rel->r_info); | |
325 | addend = get32((uint32_t *)(text + rel->r_offset)); | |
326 | switch(type) { | |
327 | case R_386_32: | |
328 | fprintf(outfile, " *(uint32_t *)(gen_code_ptr + %ld) = %s + %ld;\n", | |
329 | rel->r_offset - offset, name, addend); | |
330 | break; | |
331 | case R_386_PC32: | |
332 | fprintf(outfile, " *(uint32_t *)(gen_code_ptr + %ld) = %s - (long)(gen_code_ptr + %ld) + %ld;\n", | |
333 | rel->r_offset - offset, name, rel->r_offset - offset, addend); | |
334 | break; | |
335 | default: | |
336 | error("unsupported i386 relocation (%d)", type); | |
337 | } | |
338 | } | |
339 | } | |
340 | } | |
341 | break; | |
342 | default: | |
343 | error("unsupported CPU for relocations (%d)", e_machine); | |
344 | } | |
345 | ||
346 | ||
347 | fprintf(outfile, " gen_code_ptr += %d;\n", copy_size); | |
348 | fprintf(outfile, "}\n\n"); | |
349 | } | |
350 | ||
351 | /* load an elf object file */ | |
352 | int load_elf(const char *filename, FILE *outfile) | |
353 | { | |
354 | int fd; | |
355 | Elf32_Ehdr ehdr; | |
356 | Elf32_Shdr *sec, *shdr, *symtab_sec, *strtab_sec, *text_sec; | |
357 | int i, j, nb_syms; | |
358 | Elf32_Sym *symtab, *sym; | |
359 | const char *cpu_name; | |
360 | char *shstr, *strtab; | |
361 | uint8_t *text; | |
362 | void *relocs; | |
363 | int nb_relocs, reloc_sh_type; | |
364 | ||
365 | fd = open(filename, O_RDONLY); | |
366 | if (fd < 0) | |
367 | error("can't open file '%s'", filename); | |
368 | ||
369 | /* Read ELF header. */ | |
370 | if (read(fd, &ehdr, sizeof (ehdr)) != sizeof (ehdr)) | |
371 | error("unable to read file header"); | |
372 | ||
373 | /* Check ELF identification. */ | |
374 | if (ehdr.e_ident[EI_MAG0] != ELFMAG0 | |
375 | || ehdr.e_ident[EI_MAG1] != ELFMAG1 | |
376 | || ehdr.e_ident[EI_MAG2] != ELFMAG2 | |
377 | || ehdr.e_ident[EI_MAG3] != ELFMAG3 | |
378 | || ehdr.e_ident[EI_CLASS] != ELFCLASS32 | |
379 | || ehdr.e_ident[EI_VERSION] != EV_CURRENT) { | |
380 | error("bad ELF header"); | |
381 | } | |
382 | ||
383 | do_swap = elf_must_swap(&ehdr); | |
384 | if (do_swap) | |
385 | elf_swap_ehdr(&ehdr); | |
386 | if (ehdr.e_type != ET_REL) | |
387 | error("ELF object file expected"); | |
388 | if (ehdr.e_version != EV_CURRENT) | |
389 | error("Invalid ELF version"); | |
390 | e_machine = ehdr.e_machine; | |
391 | ||
392 | /* read section headers */ | |
393 | shdr = load_data(fd, ehdr.e_shoff, ehdr.e_shnum * sizeof(Elf32_Shdr)); | |
394 | if (do_swap) { | |
395 | for(i = 0; i < ehdr.e_shnum; i++) { | |
396 | elf_swap_shdr(&shdr[i]); | |
397 | } | |
398 | } | |
399 | ||
400 | sec = &shdr[ehdr.e_shstrndx]; | |
401 | shstr = load_data(fd, sec->sh_offset, sec->sh_size); | |
402 | ||
403 | /* text section */ | |
404 | ||
405 | text_sec = find_elf_section(shdr, ehdr.e_shnum, shstr, ".text"); | |
406 | if (!text_sec) | |
407 | error("could not find .text section"); | |
408 | text = load_data(fd, text_sec->sh_offset, text_sec->sh_size); | |
409 | ||
410 | /* find text relocations, if any */ | |
411 | nb_relocs = 0; | |
412 | relocs = NULL; | |
413 | reloc_sh_type = 0; | |
414 | for(i = 0; i < ehdr.e_shnum; i++) { | |
415 | sec = &shdr[i]; | |
416 | if ((sec->sh_type == SHT_REL || sec->sh_type == SHT_RELA) && | |
417 | sec->sh_info == (text_sec - shdr)) { | |
418 | reloc_sh_type = sec->sh_type; | |
419 | relocs = load_data(fd, sec->sh_offset, sec->sh_size); | |
420 | nb_relocs = sec->sh_size / sec->sh_entsize; | |
421 | if (do_swap) { | |
422 | if (sec->sh_type == SHT_REL) { | |
423 | Elf32_Rel *rel = relocs; | |
424 | for(j = 0, rel = relocs; j < nb_relocs; j++, rel++) { | |
425 | swab32s(&rel->r_offset); | |
426 | swab32s(&rel->r_info); | |
427 | } | |
428 | } else { | |
429 | Elf32_Rela *rel = relocs; | |
430 | for(j = 0, rel = relocs; j < nb_relocs; j++, rel++) { | |
431 | swab32s(&rel->r_offset); | |
432 | swab32s(&rel->r_info); | |
433 | swab32s(&rel->r_addend); | |
434 | } | |
435 | } | |
436 | } | |
437 | break; | |
438 | } | |
439 | } | |
440 | ||
441 | symtab_sec = find_elf_section(shdr, ehdr.e_shnum, shstr, ".symtab"); | |
442 | if (!symtab_sec) | |
443 | error("could not find .symtab section"); | |
444 | strtab_sec = &shdr[symtab_sec->sh_link]; | |
445 | ||
446 | symtab = load_data(fd, symtab_sec->sh_offset, symtab_sec->sh_size); | |
447 | strtab = load_data(fd, strtab_sec->sh_offset, strtab_sec->sh_size); | |
448 | ||
449 | nb_syms = symtab_sec->sh_size / sizeof(Elf32_Sym); | |
450 | if (do_swap) { | |
451 | for(i = 0, sym = symtab; i < nb_syms; i++, sym++) { | |
452 | swab32s(&sym->st_name); | |
453 | swab32s(&sym->st_value); | |
454 | swab32s(&sym->st_size); | |
455 | swab16s(&sym->st_shndx); | |
456 | } | |
457 | } | |
458 | ||
459 | switch(e_machine) { | |
460 | case EM_386: | |
461 | cpu_name = "i386"; | |
462 | break; | |
463 | case EM_PPC: | |
464 | cpu_name = "ppc"; | |
465 | break; | |
466 | case EM_MIPS: | |
467 | cpu_name = "mips"; | |
468 | break; | |
469 | case EM_ARM: | |
470 | cpu_name = "arm"; | |
471 | break; | |
472 | case EM_SPARC: | |
473 | cpu_name = "sparc"; | |
474 | break; | |
475 | default: | |
476 | error("unsupported CPU (e_machine=%d)", e_machine); | |
477 | } | |
478 | ||
479 | fprintf(outfile, "#include \"gen-%s.h\"\n\n", cpu_name); | |
480 | ||
481 | for(i = 0, sym = symtab; i < nb_syms; i++, sym++) { | |
482 | const char *name; | |
483 | name = strtab + sym->st_name; | |
484 | if (strstart(name, "op_", NULL) || | |
485 | strstart(name, "op1_", NULL) || | |
486 | strstart(name, "op2_", NULL) || | |
487 | strstart(name, "op3_", NULL)) { | |
488 | #if 0 | |
489 | printf("%4d: %s pos=0x%08x len=%d\n", | |
490 | i, name, sym->st_value, sym->st_size); | |
491 | #endif | |
492 | if (sym->st_shndx != (text_sec - shdr)) | |
493 | error("invalid section for opcode (0x%x)", sym->st_shndx); | |
494 | gen_code(name, sym->st_value, sym->st_size, outfile, | |
495 | text, relocs, nb_relocs, reloc_sh_type, symtab, strtab); | |
496 | } | |
497 | } | |
498 | ||
499 | close(fd); | |
500 | return 0; | |
501 | } | |
502 | ||
503 | void usage(void) | |
504 | { | |
505 | printf("dyngen (c) 2003 Fabrice Bellard\n" | |
506 | "usage: dyngen [-o outfile] objfile\n" | |
507 | "Generate a dynamic code generator from an object file\n"); | |
508 | exit(1); | |
509 | } | |
510 | ||
511 | int main(int argc, char **argv) | |
512 | { | |
513 | int c; | |
514 | const char *filename, *outfilename; | |
515 | FILE *outfile; | |
516 | ||
517 | outfilename = "out.c"; | |
518 | for(;;) { | |
519 | c = getopt(argc, argv, "ho:"); | |
520 | if (c == -1) | |
521 | break; | |
522 | switch(c) { | |
523 | case 'h': | |
524 | usage(); | |
525 | break; | |
526 | case 'o': | |
527 | outfilename = optarg; | |
528 | break; | |
529 | } | |
530 | } | |
531 | if (optind >= argc) | |
532 | usage(); | |
533 | filename = argv[optind]; | |
534 | outfile = fopen(outfilename, "w"); | |
535 | if (!outfile) | |
536 | error("could not open '%s'", outfilename); | |
537 | load_elf(filename, outfile); | |
538 | fclose(outfile); | |
539 | return 0; | |
540 | } |