| 1 | /* |
| 2 | * ia32rtools |
| 3 | * (C) notaz, 2013-2015 |
| 4 | * |
| 5 | * This work is licensed under the terms of 3-clause BSD license. |
| 6 | * See COPYING file in the top-level directory. |
| 7 | */ |
| 8 | |
| 9 | #define _GNU_SOURCE |
| 10 | #include <stdio.h> |
| 11 | #include <stdlib.h> |
| 12 | #include <string.h> |
| 13 | |
| 14 | #include "my_assert.h" |
| 15 | #include "my_str.h" |
| 16 | #include "common.h" |
| 17 | |
| 18 | #define ARRAY_SIZE(x) (sizeof(x) / sizeof(x[0])) |
| 19 | #define IS(w, y) !strcmp(w, y) |
| 20 | #define IS_START(w, y) !strncmp(w, y, strlen(y)) |
| 21 | |
| 22 | #include "protoparse.h" |
| 23 | |
| 24 | static const char *asmfn; |
| 25 | static int asmln; |
| 26 | static FILE *g_fhdr; |
| 27 | |
| 28 | #define anote(fmt, ...) \ |
| 29 | printf("%s:%d: note: " fmt, asmfn, asmln, ##__VA_ARGS__) |
| 30 | #define awarn(fmt, ...) \ |
| 31 | printf("%s:%d: warning: " fmt, asmfn, asmln, ##__VA_ARGS__) |
| 32 | #define aerr(fmt, ...) do { \ |
| 33 | printf("%s:%d: error: " fmt, asmfn, asmln, ##__VA_ARGS__); \ |
| 34 | fcloseall(); \ |
| 35 | exit(1); \ |
| 36 | } while (0) |
| 37 | |
| 38 | #include "masm_tools.h" |
| 39 | |
| 40 | enum op_flags { |
| 41 | OPF_RMD = (1 << 0), /* removed from code generation */ |
| 42 | OPF_DATA = (1 << 1), /* data processing - writes to dst opr */ |
| 43 | OPF_FLAGS = (1 << 2), /* sets flags */ |
| 44 | OPF_JMP = (1 << 3), /* branch, call */ |
| 45 | OPF_CJMP = (1 << 4), /* cond. branch (cc or jecxz/loop) */ |
| 46 | OPF_CC = (1 << 5), /* uses flags */ |
| 47 | OPF_TAIL = (1 << 6), /* ret or tail call */ |
| 48 | OPF_RSAVE = (1 << 7), /* push/pop is local reg save/load */ |
| 49 | OPF_REP = (1 << 8), /* prefixed by rep */ |
| 50 | OPF_REPZ = (1 << 9), /* rep is repe/repz */ |
| 51 | OPF_REPNZ = (1 << 10), /* rep is repne/repnz */ |
| 52 | OPF_FARG = (1 << 11), /* push collected as func arg */ |
| 53 | OPF_FARGNR = (1 << 12), /* push collected as func arg (no reuse) */ |
| 54 | OPF_EBP_S = (1 << 13), /* ebp used as scratch here, not BP */ |
| 55 | OPF_DF = (1 << 14), /* DF flag set */ |
| 56 | OPF_ATAIL = (1 << 15), /* tail call with reused arg frame */ |
| 57 | OPF_32BIT = (1 << 16), /* 32bit division */ |
| 58 | OPF_LOCK = (1 << 17), /* op has lock prefix */ |
| 59 | OPF_VAPUSH = (1 << 18), /* vararg ptr push (as call arg) */ |
| 60 | OPF_DONE = (1 << 19), /* already fully handled by analysis */ |
| 61 | OPF_PPUSH = (1 << 20), /* part of complex push-pop graph */ |
| 62 | OPF_NOREGS = (1 << 21), /* don't track regs of this op */ |
| 63 | }; |
| 64 | |
| 65 | enum op_op { |
| 66 | OP_INVAL, |
| 67 | OP_NOP, |
| 68 | OP_PUSH, |
| 69 | OP_POP, |
| 70 | OP_LEAVE, |
| 71 | OP_MOV, |
| 72 | OP_LEA, |
| 73 | OP_MOVZX, |
| 74 | OP_MOVSX, |
| 75 | OP_XCHG, |
| 76 | OP_NOT, |
| 77 | OP_XLAT, |
| 78 | OP_CDQ, |
| 79 | OP_LODS, |
| 80 | OP_STOS, |
| 81 | OP_MOVS, |
| 82 | OP_CMPS, |
| 83 | OP_SCAS, |
| 84 | OP_STD, |
| 85 | OP_CLD, |
| 86 | OP_RET, |
| 87 | OP_ADD, |
| 88 | OP_SUB, |
| 89 | OP_AND, |
| 90 | OP_OR, |
| 91 | OP_XOR, |
| 92 | OP_SHL, |
| 93 | OP_SHR, |
| 94 | OP_SAR, |
| 95 | OP_SHLD, |
| 96 | OP_SHRD, |
| 97 | OP_ROL, |
| 98 | OP_ROR, |
| 99 | OP_RCL, |
| 100 | OP_RCR, |
| 101 | OP_ADC, |
| 102 | OP_SBB, |
| 103 | OP_BSF, |
| 104 | OP_INC, |
| 105 | OP_DEC, |
| 106 | OP_NEG, |
| 107 | OP_MUL, |
| 108 | OP_IMUL, |
| 109 | OP_DIV, |
| 110 | OP_IDIV, |
| 111 | OP_TEST, |
| 112 | OP_CMP, |
| 113 | OP_CALL, |
| 114 | OP_JMP, |
| 115 | OP_JECXZ, |
| 116 | OP_LOOP, |
| 117 | OP_JCC, |
| 118 | OP_SCC, |
| 119 | // x87 |
| 120 | // mmx |
| 121 | OP_EMMS, |
| 122 | // undefined |
| 123 | OP_UD2, |
| 124 | }; |
| 125 | |
| 126 | enum opr_type { |
| 127 | OPT_UNSPEC, |
| 128 | OPT_REG, |
| 129 | OPT_REGMEM, |
| 130 | OPT_LABEL, |
| 131 | OPT_OFFSET, |
| 132 | OPT_CONST, |
| 133 | }; |
| 134 | |
| 135 | // must be sorted (larger len must be further in enum) |
| 136 | enum opr_lenmod { |
| 137 | OPLM_UNSPEC, |
| 138 | OPLM_BYTE, |
| 139 | OPLM_WORD, |
| 140 | OPLM_DWORD, |
| 141 | OPLM_QWORD, |
| 142 | }; |
| 143 | |
| 144 | #define MAX_EXITS 128 |
| 145 | |
| 146 | #define MAX_OPERANDS 3 |
| 147 | #define NAMELEN 112 |
| 148 | |
| 149 | #define OPR_INIT(type_, lmod_, reg_) \ |
| 150 | { type_, lmod_, reg_, } |
| 151 | |
| 152 | struct parsed_opr { |
| 153 | enum opr_type type; |
| 154 | enum opr_lenmod lmod; |
| 155 | int reg; |
| 156 | unsigned int is_ptr:1; // pointer in C |
| 157 | unsigned int is_array:1; // array in C |
| 158 | unsigned int type_from_var:1; // .. in header, sometimes wrong |
| 159 | unsigned int size_mismatch:1; // type override differs from C |
| 160 | unsigned int size_lt:1; // type override is larger than C |
| 161 | unsigned int had_ds:1; // had ds: prefix |
| 162 | const struct parsed_proto *pp; // for OPT_LABEL |
| 163 | unsigned int val; |
| 164 | char name[NAMELEN]; |
| 165 | }; |
| 166 | |
| 167 | struct parsed_op { |
| 168 | enum op_op op; |
| 169 | struct parsed_opr operand[MAX_OPERANDS]; |
| 170 | unsigned int flags; |
| 171 | unsigned char pfo; |
| 172 | unsigned char pfo_inv; |
| 173 | unsigned char operand_cnt; |
| 174 | unsigned char p_argnum; // arg push: altered before call arg # |
| 175 | unsigned char p_arggrp; // arg push: arg group # for above |
| 176 | unsigned char p_argpass;// arg push: arg of host func |
| 177 | short p_argnext;// arg push: same arg pushed elsewhere or -1 |
| 178 | int regmask_src; // all referensed regs |
| 179 | int regmask_dst; |
| 180 | int pfomask; // flagop: parsed_flag_op that can't be delayed |
| 181 | int cc_scratch; // scratch storage during analysis |
| 182 | int bt_i; // branch target for branches |
| 183 | struct parsed_data *btj;// branch targets for jumptables |
| 184 | struct parsed_proto *pp;// parsed_proto for OP_CALL |
| 185 | void *datap; |
| 186 | int asmln; |
| 187 | }; |
| 188 | |
| 189 | // datap: |
| 190 | // OP_CALL - parser proto hint (str) |
| 191 | // (OPF_CC) - points to one of (OPF_FLAGS) that affects cc op |
| 192 | // OP_PUSH - points to OP_POP in complex push/pop graph |
| 193 | // OP_POP - points to OP_PUSH in simple push/pop pair |
| 194 | |
| 195 | struct parsed_equ { |
| 196 | char name[64]; |
| 197 | enum opr_lenmod lmod; |
| 198 | int offset; |
| 199 | }; |
| 200 | |
| 201 | struct parsed_data { |
| 202 | char label[256]; |
| 203 | enum opr_type type; |
| 204 | enum opr_lenmod lmod; |
| 205 | int count; |
| 206 | int count_alloc; |
| 207 | struct { |
| 208 | union { |
| 209 | char *label; |
| 210 | unsigned int val; |
| 211 | } u; |
| 212 | int bt_i; |
| 213 | } *d; |
| 214 | }; |
| 215 | |
| 216 | struct label_ref { |
| 217 | int i; |
| 218 | struct label_ref *next; |
| 219 | }; |
| 220 | |
| 221 | enum ida_func_attr { |
| 222 | IDAFA_BP_FRAME = (1 << 0), |
| 223 | IDAFA_LIB_FUNC = (1 << 1), |
| 224 | IDAFA_STATIC = (1 << 2), |
| 225 | IDAFA_NORETURN = (1 << 3), |
| 226 | IDAFA_THUNK = (1 << 4), |
| 227 | IDAFA_FPD = (1 << 5), |
| 228 | }; |
| 229 | |
| 230 | // note: limited to 32k due to p_argnext |
| 231 | #define MAX_OPS 4096 |
| 232 | #define MAX_ARG_GRP 2 |
| 233 | |
| 234 | static struct parsed_op ops[MAX_OPS]; |
| 235 | static struct parsed_equ *g_eqs; |
| 236 | static int g_eqcnt; |
| 237 | static char *g_labels[MAX_OPS]; |
| 238 | static struct label_ref g_label_refs[MAX_OPS]; |
| 239 | static const struct parsed_proto *g_func_pp; |
| 240 | static struct parsed_data *g_func_pd; |
| 241 | static int g_func_pd_cnt; |
| 242 | static char g_func[256]; |
| 243 | static char g_comment[256]; |
| 244 | static int g_bp_frame; |
| 245 | static int g_sp_frame; |
| 246 | static int g_stack_frame_used; |
| 247 | static int g_stack_fsz; |
| 248 | static int g_ida_func_attr; |
| 249 | static int g_skip_func; |
| 250 | static int g_allow_regfunc; |
| 251 | static int g_quiet_pp; |
| 252 | static int g_header_mode; |
| 253 | |
| 254 | #define ferr(op_, fmt, ...) do { \ |
| 255 | printf("%s:%d: error %u: [%s] '%s': " fmt, asmfn, (op_)->asmln, \ |
| 256 | __LINE__, g_func, dump_op(op_), ##__VA_ARGS__); \ |
| 257 | fcloseall(); \ |
| 258 | exit(1); \ |
| 259 | } while (0) |
| 260 | #define fnote(op_, fmt, ...) \ |
| 261 | printf("%s:%d: note: [%s] '%s': " fmt, asmfn, (op_)->asmln, g_func, \ |
| 262 | dump_op(op_), ##__VA_ARGS__) |
| 263 | |
| 264 | #define ferr_assert(op_, cond) do { \ |
| 265 | if (!(cond)) ferr(op_, "assertion '%s' failed\n", #cond); \ |
| 266 | } while (0) |
| 267 | |
| 268 | const char *regs_r32[] = { |
| 269 | "eax", "ebx", "ecx", "edx", "esi", "edi", "ebp", "esp", |
| 270 | // not r32, but list here for easy parsing and printing |
| 271 | "mm0", "mm1", "mm2", "mm3", "mm4", "mm5", "mm6", "mm7", |
| 272 | }; |
| 273 | const char *regs_r16[] = { "ax", "bx", "cx", "dx", "si", "di", "bp", "sp" }; |
| 274 | const char *regs_r8l[] = { "al", "bl", "cl", "dl" }; |
| 275 | const char *regs_r8h[] = { "ah", "bh", "ch", "dh" }; |
| 276 | |
| 277 | enum x86_regs { xUNSPEC = -1, xAX, xBX, xCX, xDX, xSI, xDI, xBP, xSP }; |
| 278 | |
| 279 | // possible basic comparison types (without inversion) |
| 280 | enum parsed_flag_op { |
| 281 | PFO_O, // 0 OF=1 |
| 282 | PFO_C, // 2 CF=1 |
| 283 | PFO_Z, // 4 ZF=1 |
| 284 | PFO_BE, // 6 CF=1||ZF=1 |
| 285 | PFO_S, // 8 SF=1 |
| 286 | PFO_P, // a PF=1 |
| 287 | PFO_L, // c SF!=OF |
| 288 | PFO_LE, // e ZF=1||SF!=OF |
| 289 | }; |
| 290 | |
| 291 | #define PFOB_O (1 << PFO_O) |
| 292 | #define PFOB_C (1 << PFO_C) |
| 293 | #define PFOB_Z (1 << PFO_Z) |
| 294 | #define PFOB_S (1 << PFO_S) |
| 295 | |
| 296 | static const char *parsed_flag_op_names[] = { |
| 297 | "o", "c", "z", "be", "s", "p", "l", "le" |
| 298 | }; |
| 299 | |
| 300 | static int char_array_i(const char *array[], size_t len, const char *s) |
| 301 | { |
| 302 | int i; |
| 303 | |
| 304 | for (i = 0; i < len; i++) |
| 305 | if (IS(s, array[i])) |
| 306 | return i; |
| 307 | |
| 308 | return -1; |
| 309 | } |
| 310 | |
| 311 | static void printf_number(char *buf, size_t buf_size, |
| 312 | unsigned long number) |
| 313 | { |
| 314 | // output in C-friendly form |
| 315 | snprintf(buf, buf_size, number < 10 ? "%lu" : "0x%02lx", number); |
| 316 | } |
| 317 | |
| 318 | static int check_segment_prefix(const char *s) |
| 319 | { |
| 320 | if (s[0] == 0 || s[1] != 's' || s[2] != ':') |
| 321 | return 0; |
| 322 | |
| 323 | switch (s[0]) { |
| 324 | case 'c': return 1; |
| 325 | case 'd': return 2; |
| 326 | case 's': return 3; |
| 327 | case 'e': return 4; |
| 328 | case 'f': return 5; |
| 329 | case 'g': return 6; |
| 330 | default: return 0; |
| 331 | } |
| 332 | } |
| 333 | |
| 334 | static int parse_reg(enum opr_lenmod *reg_lmod, const char *s) |
| 335 | { |
| 336 | int reg; |
| 337 | |
| 338 | reg = char_array_i(regs_r32, ARRAY_SIZE(regs_r32), s); |
| 339 | if (reg >= 8) { |
| 340 | *reg_lmod = OPLM_QWORD; |
| 341 | return reg; |
| 342 | } |
| 343 | if (reg >= 0) { |
| 344 | *reg_lmod = OPLM_DWORD; |
| 345 | return reg; |
| 346 | } |
| 347 | reg = char_array_i(regs_r16, ARRAY_SIZE(regs_r16), s); |
| 348 | if (reg >= 0) { |
| 349 | *reg_lmod = OPLM_WORD; |
| 350 | return reg; |
| 351 | } |
| 352 | reg = char_array_i(regs_r8h, ARRAY_SIZE(regs_r8h), s); |
| 353 | if (reg >= 0) { |
| 354 | *reg_lmod = OPLM_BYTE; |
| 355 | return reg; |
| 356 | } |
| 357 | reg = char_array_i(regs_r8l, ARRAY_SIZE(regs_r8l), s); |
| 358 | if (reg >= 0) { |
| 359 | *reg_lmod = OPLM_BYTE; |
| 360 | return reg; |
| 361 | } |
| 362 | |
| 363 | return -1; |
| 364 | } |
| 365 | |
| 366 | static int parse_indmode(char *name, int *regmask, int need_c_cvt) |
| 367 | { |
| 368 | enum opr_lenmod lmod; |
| 369 | char cvtbuf[256]; |
| 370 | char *d = cvtbuf; |
| 371 | char *s = name; |
| 372 | char w[64]; |
| 373 | long number; |
| 374 | int reg; |
| 375 | int c = 0; |
| 376 | |
| 377 | *d = 0; |
| 378 | |
| 379 | while (*s != 0) { |
| 380 | d += strlen(d); |
| 381 | while (my_isblank(*s)) |
| 382 | s++; |
| 383 | for (; my_issep(*s); d++, s++) |
| 384 | *d = *s; |
| 385 | while (my_isblank(*s)) |
| 386 | s++; |
| 387 | *d = 0; |
| 388 | |
| 389 | // skip '?s:' prefixes |
| 390 | if (check_segment_prefix(s)) |
| 391 | s += 3; |
| 392 | |
| 393 | s = next_idt(w, sizeof(w), s); |
| 394 | if (w[0] == 0) |
| 395 | break; |
| 396 | c++; |
| 397 | |
| 398 | reg = parse_reg(&lmod, w); |
| 399 | if (reg >= 0) { |
| 400 | *regmask |= 1 << reg; |
| 401 | goto pass; |
| 402 | } |
| 403 | |
| 404 | if ('0' <= w[0] && w[0] <= '9') { |
| 405 | number = parse_number(w); |
| 406 | printf_number(d, sizeof(cvtbuf) - (d - cvtbuf), number); |
| 407 | continue; |
| 408 | } |
| 409 | |
| 410 | // probably some label/identifier - pass |
| 411 | |
| 412 | pass: |
| 413 | snprintf(d, sizeof(cvtbuf) - (d - cvtbuf), "%s", w); |
| 414 | } |
| 415 | |
| 416 | if (need_c_cvt) |
| 417 | strcpy(name, cvtbuf); |
| 418 | |
| 419 | return c; |
| 420 | } |
| 421 | |
| 422 | static int is_reg_in_str(const char *s) |
| 423 | { |
| 424 | int i; |
| 425 | |
| 426 | if (strlen(s) < 3 || (s[3] && !my_issep(s[3]) && !my_isblank(s[3]))) |
| 427 | return 0; |
| 428 | |
| 429 | for (i = 0; i < ARRAY_SIZE(regs_r32); i++) |
| 430 | if (!strncmp(s, regs_r32[i], 3)) |
| 431 | return 1; |
| 432 | |
| 433 | return 0; |
| 434 | } |
| 435 | |
| 436 | static const char *parse_stack_el(const char *name, char *extra_reg, |
| 437 | int early_try) |
| 438 | { |
| 439 | const char *p, *p2, *s; |
| 440 | char *endp = NULL; |
| 441 | char buf[32]; |
| 442 | long val; |
| 443 | int len; |
| 444 | |
| 445 | if (g_bp_frame || early_try) |
| 446 | { |
| 447 | p = name; |
| 448 | if (IS_START(p + 3, "+ebp+") && is_reg_in_str(p)) { |
| 449 | p += 4; |
| 450 | if (extra_reg != NULL) { |
| 451 | strncpy(extra_reg, name, 3); |
| 452 | extra_reg[4] = 0; |
| 453 | } |
| 454 | } |
| 455 | |
| 456 | if (IS_START(p, "ebp+")) { |
| 457 | p += 4; |
| 458 | |
| 459 | p2 = strchr(p, '+'); |
| 460 | if (p2 != NULL && is_reg_in_str(p)) { |
| 461 | if (extra_reg != NULL) { |
| 462 | strncpy(extra_reg, p, p2 - p); |
| 463 | extra_reg[p2 - p] = 0; |
| 464 | } |
| 465 | p = p2 + 1; |
| 466 | } |
| 467 | |
| 468 | if (!('0' <= *p && *p <= '9')) |
| 469 | return p; |
| 470 | |
| 471 | return NULL; |
| 472 | } |
| 473 | } |
| 474 | |
| 475 | if (!IS_START(name, "esp+")) |
| 476 | return NULL; |
| 477 | |
| 478 | s = name + 4; |
| 479 | p = strchr(s, '+'); |
| 480 | if (p) { |
| 481 | if (is_reg_in_str(s)) { |
| 482 | if (extra_reg != NULL) { |
| 483 | strncpy(extra_reg, s, p - s); |
| 484 | extra_reg[p - s] = 0; |
| 485 | } |
| 486 | s = p + 1; |
| 487 | p = strchr(s, '+'); |
| 488 | if (p == NULL) |
| 489 | aerr("%s IDA stackvar not set?\n", __func__); |
| 490 | } |
| 491 | if (!('0' <= *s && *s <= '9')) { |
| 492 | aerr("%s IDA stackvar offset not set?\n", __func__); |
| 493 | return NULL; |
| 494 | } |
| 495 | if (s[0] == '0' && s[1] == 'x') |
| 496 | s += 2; |
| 497 | len = p - s; |
| 498 | if (len < sizeof(buf) - 1) { |
| 499 | strncpy(buf, s, len); |
| 500 | buf[len] = 0; |
| 501 | val = strtol(buf, &endp, 16); |
| 502 | if (val == 0 || *endp != 0) { |
| 503 | aerr("%s num parse fail for '%s'\n", __func__, buf); |
| 504 | return NULL; |
| 505 | } |
| 506 | } |
| 507 | p++; |
| 508 | } |
| 509 | else |
| 510 | p = name + 4; |
| 511 | |
| 512 | if ('0' <= *p && *p <= '9') |
| 513 | return NULL; |
| 514 | |
| 515 | return p; |
| 516 | } |
| 517 | |
| 518 | static int guess_lmod_from_name(struct parsed_opr *opr) |
| 519 | { |
| 520 | if (IS_START(opr->name, "dword_") || IS_START(opr->name, "off_")) { |
| 521 | opr->lmod = OPLM_DWORD; |
| 522 | return 1; |
| 523 | } |
| 524 | if (IS_START(opr->name, "word_")) { |
| 525 | opr->lmod = OPLM_WORD; |
| 526 | return 1; |
| 527 | } |
| 528 | if (IS_START(opr->name, "byte_")) { |
| 529 | opr->lmod = OPLM_BYTE; |
| 530 | return 1; |
| 531 | } |
| 532 | if (IS_START(opr->name, "qword_")) { |
| 533 | opr->lmod = OPLM_QWORD; |
| 534 | return 1; |
| 535 | } |
| 536 | return 0; |
| 537 | } |
| 538 | |
| 539 | static int guess_lmod_from_c_type(enum opr_lenmod *lmod, |
| 540 | const struct parsed_type *c_type) |
| 541 | { |
| 542 | static const char *dword_types[] = { |
| 543 | "uint32_t", "int", "_DWORD", "UINT_PTR", "DWORD", |
| 544 | "WPARAM", "LPARAM", "UINT", "__int32", |
| 545 | "LONG", "HIMC", "BOOL", "size_t", |
| 546 | "float", |
| 547 | }; |
| 548 | static const char *word_types[] = { |
| 549 | "uint16_t", "int16_t", "_WORD", "WORD", |
| 550 | "unsigned __int16", "__int16", |
| 551 | }; |
| 552 | static const char *byte_types[] = { |
| 553 | "uint8_t", "int8_t", "char", |
| 554 | "unsigned __int8", "__int8", "BYTE", "_BYTE", |
| 555 | "CHAR", "_UNKNOWN", |
| 556 | // structures.. deal the same as with _UNKNOWN for now |
| 557 | "CRITICAL_SECTION", |
| 558 | }; |
| 559 | const char *n; |
| 560 | int i; |
| 561 | |
| 562 | if (c_type->is_ptr) { |
| 563 | *lmod = OPLM_DWORD; |
| 564 | return 1; |
| 565 | } |
| 566 | |
| 567 | n = skip_type_mod(c_type->name); |
| 568 | |
| 569 | for (i = 0; i < ARRAY_SIZE(dword_types); i++) { |
| 570 | if (IS(n, dword_types[i])) { |
| 571 | *lmod = OPLM_DWORD; |
| 572 | return 1; |
| 573 | } |
| 574 | } |
| 575 | |
| 576 | for (i = 0; i < ARRAY_SIZE(word_types); i++) { |
| 577 | if (IS(n, word_types[i])) { |
| 578 | *lmod = OPLM_WORD; |
| 579 | return 1; |
| 580 | } |
| 581 | } |
| 582 | |
| 583 | for (i = 0; i < ARRAY_SIZE(byte_types); i++) { |
| 584 | if (IS(n, byte_types[i])) { |
| 585 | *lmod = OPLM_BYTE; |
| 586 | return 1; |
| 587 | } |
| 588 | } |
| 589 | |
| 590 | return 0; |
| 591 | } |
| 592 | |
| 593 | static char *default_cast_to(char *buf, size_t buf_size, |
| 594 | struct parsed_opr *opr) |
| 595 | { |
| 596 | buf[0] = 0; |
| 597 | |
| 598 | if (!opr->is_ptr) |
| 599 | return buf; |
| 600 | if (opr->pp == NULL || opr->pp->type.name == NULL |
| 601 | || opr->pp->is_fptr) |
| 602 | { |
| 603 | snprintf(buf, buf_size, "%s", "(void *)"); |
| 604 | return buf; |
| 605 | } |
| 606 | |
| 607 | snprintf(buf, buf_size, "(%s)", opr->pp->type.name); |
| 608 | return buf; |
| 609 | } |
| 610 | |
| 611 | static enum opr_type lmod_from_directive(const char *d) |
| 612 | { |
| 613 | if (IS(d, "dd")) |
| 614 | return OPLM_DWORD; |
| 615 | else if (IS(d, "dw")) |
| 616 | return OPLM_WORD; |
| 617 | else if (IS(d, "db")) |
| 618 | return OPLM_BYTE; |
| 619 | |
| 620 | aerr("unhandled directive: '%s'\n", d); |
| 621 | return OPLM_UNSPEC; |
| 622 | } |
| 623 | |
| 624 | static void setup_reg_opr(struct parsed_opr *opr, int reg, enum opr_lenmod lmod, |
| 625 | int *regmask) |
| 626 | { |
| 627 | opr->type = OPT_REG; |
| 628 | opr->reg = reg; |
| 629 | opr->lmod = lmod; |
| 630 | *regmask |= 1 << reg; |
| 631 | } |
| 632 | |
| 633 | static struct parsed_equ *equ_find(struct parsed_op *po, const char *name, |
| 634 | int *extra_offs); |
| 635 | |
| 636 | static int parse_operand(struct parsed_opr *opr, |
| 637 | int *regmask, int *regmask_indirect, |
| 638 | char words[16][256], int wordc, int w, unsigned int op_flags) |
| 639 | { |
| 640 | const struct parsed_proto *pp = NULL; |
| 641 | enum opr_lenmod tmplmod; |
| 642 | unsigned long number; |
| 643 | char buf[256]; |
| 644 | int ret, len; |
| 645 | int wordc_in; |
| 646 | char *p; |
| 647 | int i; |
| 648 | |
| 649 | if (w >= wordc) |
| 650 | aerr("parse_operand w %d, wordc %d\n", w, wordc); |
| 651 | |
| 652 | opr->reg = xUNSPEC; |
| 653 | |
| 654 | for (i = w; i < wordc; i++) { |
| 655 | len = strlen(words[i]); |
| 656 | if (words[i][len - 1] == ',') { |
| 657 | words[i][len - 1] = 0; |
| 658 | wordc = i + 1; |
| 659 | break; |
| 660 | } |
| 661 | } |
| 662 | |
| 663 | wordc_in = wordc - w; |
| 664 | |
| 665 | if ((op_flags & OPF_JMP) && wordc_in > 0 |
| 666 | && !('0' <= words[w][0] && words[w][0] <= '9')) |
| 667 | { |
| 668 | const char *label = NULL; |
| 669 | |
| 670 | if (wordc_in == 3 && !strncmp(words[w], "near", 4) |
| 671 | && IS(words[w + 1], "ptr")) |
| 672 | label = words[w + 2]; |
| 673 | else if (wordc_in == 2 && IS(words[w], "short")) |
| 674 | label = words[w + 1]; |
| 675 | else if (wordc_in == 1 |
| 676 | && strchr(words[w], '[') == NULL |
| 677 | && parse_reg(&tmplmod, words[w]) < 0) |
| 678 | label = words[w]; |
| 679 | |
| 680 | if (label != NULL) { |
| 681 | opr->type = OPT_LABEL; |
| 682 | ret = check_segment_prefix(label); |
| 683 | if (ret != 0) { |
| 684 | if (ret >= 5) |
| 685 | aerr("fs/gs used\n"); |
| 686 | opr->had_ds = 1; |
| 687 | label += 3; |
| 688 | } |
| 689 | strcpy(opr->name, label); |
| 690 | return wordc; |
| 691 | } |
| 692 | } |
| 693 | |
| 694 | if (wordc_in >= 3) { |
| 695 | if (IS(words[w + 1], "ptr")) { |
| 696 | if (IS(words[w], "dword")) |
| 697 | opr->lmod = OPLM_DWORD; |
| 698 | else if (IS(words[w], "word")) |
| 699 | opr->lmod = OPLM_WORD; |
| 700 | else if (IS(words[w], "byte")) |
| 701 | opr->lmod = OPLM_BYTE; |
| 702 | else if (IS(words[w], "qword")) |
| 703 | opr->lmod = OPLM_QWORD; |
| 704 | else |
| 705 | aerr("type parsing failed\n"); |
| 706 | w += 2; |
| 707 | wordc_in = wordc - w; |
| 708 | } |
| 709 | } |
| 710 | |
| 711 | if (wordc_in == 2) { |
| 712 | if (IS(words[w], "offset")) { |
| 713 | opr->type = OPT_OFFSET; |
| 714 | opr->lmod = OPLM_DWORD; |
| 715 | strcpy(opr->name, words[w + 1]); |
| 716 | pp = proto_parse(g_fhdr, opr->name, 1); |
| 717 | goto do_label; |
| 718 | } |
| 719 | if (IS(words[w], "(offset")) { |
| 720 | p = strchr(words[w + 1], ')'); |
| 721 | if (p == NULL) |
| 722 | aerr("parse of bracketed offset failed\n"); |
| 723 | *p = 0; |
| 724 | opr->type = OPT_OFFSET; |
| 725 | strcpy(opr->name, words[w + 1]); |
| 726 | return wordc; |
| 727 | } |
| 728 | } |
| 729 | |
| 730 | if (wordc_in != 1) |
| 731 | aerr("parse_operand 1 word expected\n"); |
| 732 | |
| 733 | ret = check_segment_prefix(words[w]); |
| 734 | if (ret != 0) { |
| 735 | if (ret >= 5) |
| 736 | aerr("fs/gs used\n"); |
| 737 | opr->had_ds = 1; |
| 738 | memmove(words[w], words[w] + 3, strlen(words[w]) - 2); |
| 739 | } |
| 740 | strcpy(opr->name, words[w]); |
| 741 | |
| 742 | if (words[w][0] == '[') { |
| 743 | opr->type = OPT_REGMEM; |
| 744 | ret = sscanf(words[w], "[%[^]]]", opr->name); |
| 745 | if (ret != 1) |
| 746 | aerr("[] parse failure\n"); |
| 747 | |
| 748 | parse_indmode(opr->name, regmask_indirect, 1); |
| 749 | if (opr->lmod == OPLM_UNSPEC && parse_stack_el(opr->name, NULL, 1)) |
| 750 | { |
| 751 | // might be an equ |
| 752 | struct parsed_equ *eq = |
| 753 | equ_find(NULL, parse_stack_el(opr->name, NULL, 1), &i); |
| 754 | if (eq) |
| 755 | opr->lmod = eq->lmod; |
| 756 | } |
| 757 | return wordc; |
| 758 | } |
| 759 | else if (strchr(words[w], '[')) { |
| 760 | // label[reg] form |
| 761 | p = strchr(words[w], '['); |
| 762 | opr->type = OPT_REGMEM; |
| 763 | parse_indmode(p, regmask_indirect, 0); |
| 764 | strncpy(buf, words[w], p - words[w]); |
| 765 | buf[p - words[w]] = 0; |
| 766 | pp = proto_parse(g_fhdr, buf, 1); |
| 767 | goto do_label; |
| 768 | } |
| 769 | else if (('0' <= words[w][0] && words[w][0] <= '9') |
| 770 | || words[w][0] == '-') |
| 771 | { |
| 772 | number = parse_number(words[w]); |
| 773 | opr->type = OPT_CONST; |
| 774 | opr->val = number; |
| 775 | printf_number(opr->name, sizeof(opr->name), number); |
| 776 | return wordc; |
| 777 | } |
| 778 | |
| 779 | ret = parse_reg(&tmplmod, opr->name); |
| 780 | if (ret >= 0) { |
| 781 | setup_reg_opr(opr, ret, tmplmod, regmask); |
| 782 | return wordc; |
| 783 | } |
| 784 | |
| 785 | // most likely var in data segment |
| 786 | opr->type = OPT_LABEL; |
| 787 | pp = proto_parse(g_fhdr, opr->name, g_quiet_pp); |
| 788 | |
| 789 | do_label: |
| 790 | if (pp != NULL) { |
| 791 | if (pp->is_fptr || pp->is_func) { |
| 792 | opr->lmod = OPLM_DWORD; |
| 793 | opr->is_ptr = 1; |
| 794 | } |
| 795 | else { |
| 796 | tmplmod = OPLM_UNSPEC; |
| 797 | if (!guess_lmod_from_c_type(&tmplmod, &pp->type)) |
| 798 | anote("unhandled C type '%s' for '%s'\n", |
| 799 | pp->type.name, opr->name); |
| 800 | |
| 801 | if (opr->lmod == OPLM_UNSPEC) { |
| 802 | opr->lmod = tmplmod; |
| 803 | opr->type_from_var = 1; |
| 804 | } |
| 805 | else if (opr->lmod != tmplmod) { |
| 806 | opr->size_mismatch = 1; |
| 807 | if (tmplmod < opr->lmod) |
| 808 | opr->size_lt = 1; |
| 809 | } |
| 810 | opr->is_ptr = pp->type.is_ptr; |
| 811 | } |
| 812 | opr->is_array = pp->type.is_array; |
| 813 | } |
| 814 | opr->pp = pp; |
| 815 | |
| 816 | if (opr->lmod == OPLM_UNSPEC) |
| 817 | guess_lmod_from_name(opr); |
| 818 | return wordc; |
| 819 | } |
| 820 | |
| 821 | static const struct { |
| 822 | const char *name; |
| 823 | unsigned int flags; |
| 824 | } pref_table[] = { |
| 825 | { "rep", OPF_REP }, |
| 826 | { "repe", OPF_REP|OPF_REPZ }, |
| 827 | { "repz", OPF_REP|OPF_REPZ }, |
| 828 | { "repne", OPF_REP|OPF_REPNZ }, |
| 829 | { "repnz", OPF_REP|OPF_REPNZ }, |
| 830 | { "lock", OPF_LOCK }, // ignored for now.. |
| 831 | }; |
| 832 | |
| 833 | #define OPF_CJMP_CC (OPF_JMP|OPF_CJMP|OPF_CC) |
| 834 | |
| 835 | static const struct { |
| 836 | const char *name; |
| 837 | enum op_op op; |
| 838 | unsigned short minopr; |
| 839 | unsigned short maxopr; |
| 840 | unsigned int flags; |
| 841 | unsigned char pfo; |
| 842 | unsigned char pfo_inv; |
| 843 | } op_table[] = { |
| 844 | { "nop", OP_NOP, 0, 0, 0 }, |
| 845 | { "push", OP_PUSH, 1, 1, 0 }, |
| 846 | { "pop", OP_POP, 1, 1, OPF_DATA }, |
| 847 | { "leave",OP_LEAVE, 0, 0, OPF_DATA }, |
| 848 | { "mov" , OP_MOV, 2, 2, OPF_DATA }, |
| 849 | { "lea", OP_LEA, 2, 2, OPF_DATA }, |
| 850 | { "movzx",OP_MOVZX, 2, 2, OPF_DATA }, |
| 851 | { "movsx",OP_MOVSX, 2, 2, OPF_DATA }, |
| 852 | { "xchg", OP_XCHG, 2, 2, OPF_DATA }, |
| 853 | { "not", OP_NOT, 1, 1, OPF_DATA }, |
| 854 | { "xlat", OP_XLAT, 0, 0, OPF_DATA }, |
| 855 | { "cdq", OP_CDQ, 0, 0, OPF_DATA }, |
| 856 | { "lodsb",OP_LODS, 0, 0, OPF_DATA }, |
| 857 | { "lodsw",OP_LODS, 0, 0, OPF_DATA }, |
| 858 | { "lodsd",OP_LODS, 0, 0, OPF_DATA }, |
| 859 | { "stosb",OP_STOS, 0, 0, OPF_DATA }, |
| 860 | { "stosw",OP_STOS, 0, 0, OPF_DATA }, |
| 861 | { "stosd",OP_STOS, 0, 0, OPF_DATA }, |
| 862 | { "movsb",OP_MOVS, 0, 0, OPF_DATA }, |
| 863 | { "movsw",OP_MOVS, 0, 0, OPF_DATA }, |
| 864 | { "movsd",OP_MOVS, 0, 0, OPF_DATA }, |
| 865 | { "cmpsb",OP_CMPS, 0, 0, OPF_DATA|OPF_FLAGS }, |
| 866 | { "cmpsw",OP_CMPS, 0, 0, OPF_DATA|OPF_FLAGS }, |
| 867 | { "cmpsd",OP_CMPS, 0, 0, OPF_DATA|OPF_FLAGS }, |
| 868 | { "scasb",OP_SCAS, 0, 0, OPF_DATA|OPF_FLAGS }, |
| 869 | { "scasw",OP_SCAS, 0, 0, OPF_DATA|OPF_FLAGS }, |
| 870 | { "scasd",OP_SCAS, 0, 0, OPF_DATA|OPF_FLAGS }, |
| 871 | { "std", OP_STD, 0, 0, OPF_DATA }, // special flag |
| 872 | { "cld", OP_CLD, 0, 0, OPF_DATA }, |
| 873 | { "add", OP_ADD, 2, 2, OPF_DATA|OPF_FLAGS }, |
| 874 | { "sub", OP_SUB, 2, 2, OPF_DATA|OPF_FLAGS }, |
| 875 | { "and", OP_AND, 2, 2, OPF_DATA|OPF_FLAGS }, |
| 876 | { "or", OP_OR, 2, 2, OPF_DATA|OPF_FLAGS }, |
| 877 | { "xor", OP_XOR, 2, 2, OPF_DATA|OPF_FLAGS }, |
| 878 | { "shl", OP_SHL, 2, 2, OPF_DATA|OPF_FLAGS }, |
| 879 | { "shr", OP_SHR, 2, 2, OPF_DATA|OPF_FLAGS }, |
| 880 | { "sal", OP_SHL, 2, 2, OPF_DATA|OPF_FLAGS }, |
| 881 | { "sar", OP_SAR, 2, 2, OPF_DATA|OPF_FLAGS }, |
| 882 | { "shld", OP_SHLD, 3, 3, OPF_DATA|OPF_FLAGS }, |
| 883 | { "shrd", OP_SHRD, 3, 3, OPF_DATA|OPF_FLAGS }, |
| 884 | { "rol", OP_ROL, 2, 2, OPF_DATA|OPF_FLAGS }, |
| 885 | { "ror", OP_ROR, 2, 2, OPF_DATA|OPF_FLAGS }, |
| 886 | { "rcl", OP_RCL, 2, 2, OPF_DATA|OPF_FLAGS|OPF_CC, PFO_C }, |
| 887 | { "rcr", OP_RCR, 2, 2, OPF_DATA|OPF_FLAGS|OPF_CC, PFO_C }, |
| 888 | { "adc", OP_ADC, 2, 2, OPF_DATA|OPF_FLAGS|OPF_CC, PFO_C }, |
| 889 | { "sbb", OP_SBB, 2, 2, OPF_DATA|OPF_FLAGS|OPF_CC, PFO_C }, |
| 890 | { "bsf", OP_BSF, 2, 2, OPF_DATA|OPF_FLAGS }, |
| 891 | { "inc", OP_INC, 1, 1, OPF_DATA|OPF_FLAGS }, |
| 892 | { "dec", OP_DEC, 1, 1, OPF_DATA|OPF_FLAGS }, |
| 893 | { "neg", OP_NEG, 1, 1, OPF_DATA|OPF_FLAGS }, |
| 894 | { "mul", OP_MUL, 1, 1, OPF_DATA|OPF_FLAGS }, |
| 895 | { "imul", OP_IMUL, 1, 3, OPF_DATA|OPF_FLAGS }, |
| 896 | { "div", OP_DIV, 1, 1, OPF_DATA|OPF_FLAGS }, |
| 897 | { "idiv", OP_IDIV, 1, 1, OPF_DATA|OPF_FLAGS }, |
| 898 | { "test", OP_TEST, 2, 2, OPF_FLAGS }, |
| 899 | { "cmp", OP_CMP, 2, 2, OPF_FLAGS }, |
| 900 | { "retn", OP_RET, 0, 1, OPF_TAIL }, |
| 901 | { "call", OP_CALL, 1, 1, OPF_JMP|OPF_DATA|OPF_FLAGS }, |
| 902 | { "jmp", OP_JMP, 1, 1, OPF_JMP }, |
| 903 | { "jecxz",OP_JECXZ, 1, 1, OPF_JMP|OPF_CJMP }, |
| 904 | { "loop", OP_LOOP, 1, 1, OPF_JMP|OPF_CJMP|OPF_DATA }, |
| 905 | { "jo", OP_JCC, 1, 1, OPF_CJMP_CC, PFO_O, 0 }, // 70 OF=1 |
| 906 | { "jno", OP_JCC, 1, 1, OPF_CJMP_CC, PFO_O, 1 }, // 71 OF=0 |
| 907 | { "jc", OP_JCC, 1, 1, OPF_CJMP_CC, PFO_C, 0 }, // 72 CF=1 |
| 908 | { "jb", OP_JCC, 1, 1, OPF_CJMP_CC, PFO_C, 0 }, // 72 |
| 909 | { "jnc", OP_JCC, 1, 1, OPF_CJMP_CC, PFO_C, 1 }, // 73 CF=0 |
| 910 | { "jnb", OP_JCC, 1, 1, OPF_CJMP_CC, PFO_C, 1 }, // 73 |
| 911 | { "jae", OP_JCC, 1, 1, OPF_CJMP_CC, PFO_C, 1 }, // 73 |
| 912 | { "jz", OP_JCC, 1, 1, OPF_CJMP_CC, PFO_Z, 0 }, // 74 ZF=1 |
| 913 | { "je", OP_JCC, 1, 1, OPF_CJMP_CC, PFO_Z, 0 }, // 74 |
| 914 | { "jnz", OP_JCC, 1, 1, OPF_CJMP_CC, PFO_Z, 1 }, // 75 ZF=0 |
| 915 | { "jne", OP_JCC, 1, 1, OPF_CJMP_CC, PFO_Z, 1 }, // 75 |
| 916 | { "jbe", OP_JCC, 1, 1, OPF_CJMP_CC, PFO_BE, 0 }, // 76 CF=1||ZF=1 |
| 917 | { "jna", OP_JCC, 1, 1, OPF_CJMP_CC, PFO_BE, 0 }, // 76 |
| 918 | { "ja", OP_JCC, 1, 1, OPF_CJMP_CC, PFO_BE, 1 }, // 77 CF=0&&ZF=0 |
| 919 | { "jnbe", OP_JCC, 1, 1, OPF_CJMP_CC, PFO_BE, 1 }, // 77 |
| 920 | { "js", OP_JCC, 1, 1, OPF_CJMP_CC, PFO_S, 0 }, // 78 SF=1 |
| 921 | { "jns", OP_JCC, 1, 1, OPF_CJMP_CC, PFO_S, 1 }, // 79 SF=0 |
| 922 | { "jp", OP_JCC, 1, 1, OPF_CJMP_CC, PFO_P, 0 }, // 7a PF=1 |
| 923 | { "jpe", OP_JCC, 1, 1, OPF_CJMP_CC, PFO_P, 0 }, // 7a |
| 924 | { "jnp", OP_JCC, 1, 1, OPF_CJMP_CC, PFO_P, 1 }, // 7b PF=0 |
| 925 | { "jpo", OP_JCC, 1, 1, OPF_CJMP_CC, PFO_P, 1 }, // 7b |
| 926 | { "jl", OP_JCC, 1, 1, OPF_CJMP_CC, PFO_L, 0 }, // 7c SF!=OF |
| 927 | { "jnge", OP_JCC, 1, 1, OPF_CJMP_CC, PFO_L, 0 }, // 7c |
| 928 | { "jge", OP_JCC, 1, 1, OPF_CJMP_CC, PFO_L, 1 }, // 7d SF=OF |
| 929 | { "jnl", OP_JCC, 1, 1, OPF_CJMP_CC, PFO_L, 1 }, // 7d |
| 930 | { "jle", OP_JCC, 1, 1, OPF_CJMP_CC, PFO_LE, 0 }, // 7e ZF=1||SF!=OF |
| 931 | { "jng", OP_JCC, 1, 1, OPF_CJMP_CC, PFO_LE, 0 }, // 7e |
| 932 | { "jg", OP_JCC, 1, 1, OPF_CJMP_CC, PFO_LE, 1 }, // 7f ZF=0&&SF=OF |
| 933 | { "jnle", OP_JCC, 1, 1, OPF_CJMP_CC, PFO_LE, 1 }, // 7f |
| 934 | { "seto", OP_SCC, 1, 1, OPF_DATA|OPF_CC, PFO_O, 0 }, |
| 935 | { "setno", OP_SCC, 1, 1, OPF_DATA|OPF_CC, PFO_O, 1 }, |
| 936 | { "setc", OP_SCC, 1, 1, OPF_DATA|OPF_CC, PFO_C, 0 }, |
| 937 | { "setb", OP_SCC, 1, 1, OPF_DATA|OPF_CC, PFO_C, 0 }, |
| 938 | { "setnc", OP_SCC, 1, 1, OPF_DATA|OPF_CC, PFO_C, 1 }, |
| 939 | { "setae", OP_SCC, 1, 1, OPF_DATA|OPF_CC, PFO_C, 1 }, |
| 940 | { "setnb", OP_SCC, 1, 1, OPF_DATA|OPF_CC, PFO_C, 1 }, |
| 941 | { "setz", OP_SCC, 1, 1, OPF_DATA|OPF_CC, PFO_Z, 0 }, |
| 942 | { "sete", OP_SCC, 1, 1, OPF_DATA|OPF_CC, PFO_Z, 0 }, |
| 943 | { "setnz", OP_SCC, 1, 1, OPF_DATA|OPF_CC, PFO_Z, 1 }, |
| 944 | { "setne", OP_SCC, 1, 1, OPF_DATA|OPF_CC, PFO_Z, 1 }, |
| 945 | { "setbe", OP_SCC, 1, 1, OPF_DATA|OPF_CC, PFO_BE, 0 }, |
| 946 | { "setna", OP_SCC, 1, 1, OPF_DATA|OPF_CC, PFO_BE, 0 }, |
| 947 | { "seta", OP_SCC, 1, 1, OPF_DATA|OPF_CC, PFO_BE, 1 }, |
| 948 | { "setnbe", OP_SCC, 1, 1, OPF_DATA|OPF_CC, PFO_BE, 1 }, |
| 949 | { "sets", OP_SCC, 1, 1, OPF_DATA|OPF_CC, PFO_S, 0 }, |
| 950 | { "setns", OP_SCC, 1, 1, OPF_DATA|OPF_CC, PFO_S, 1 }, |
| 951 | { "setp", OP_SCC, 1, 1, OPF_DATA|OPF_CC, PFO_P, 0 }, |
| 952 | { "setpe", OP_SCC, 1, 1, OPF_DATA|OPF_CC, PFO_P, 0 }, |
| 953 | { "setnp", OP_SCC, 1, 1, OPF_DATA|OPF_CC, PFO_P, 1 }, |
| 954 | { "setpo", OP_SCC, 1, 1, OPF_DATA|OPF_CC, PFO_P, 1 }, |
| 955 | { "setl", OP_SCC, 1, 1, OPF_DATA|OPF_CC, PFO_L, 0 }, |
| 956 | { "setnge", OP_SCC, 1, 1, OPF_DATA|OPF_CC, PFO_L, 0 }, |
| 957 | { "setge", OP_SCC, 1, 1, OPF_DATA|OPF_CC, PFO_L, 1 }, |
| 958 | { "setnl", OP_SCC, 1, 1, OPF_DATA|OPF_CC, PFO_L, 1 }, |
| 959 | { "setle", OP_SCC, 1, 1, OPF_DATA|OPF_CC, PFO_LE, 0 }, |
| 960 | { "setng", OP_SCC, 1, 1, OPF_DATA|OPF_CC, PFO_LE, 0 }, |
| 961 | { "setg", OP_SCC, 1, 1, OPF_DATA|OPF_CC, PFO_LE, 1 }, |
| 962 | { "setnle", OP_SCC, 1, 1, OPF_DATA|OPF_CC, PFO_LE, 1 }, |
| 963 | // x87 |
| 964 | // mmx |
| 965 | { "emms", OP_EMMS, 0, 0, OPF_DATA }, |
| 966 | { "movq", OP_MOV, 2, 2, OPF_DATA }, |
| 967 | // must be last |
| 968 | { "ud2", OP_UD2 }, |
| 969 | }; |
| 970 | |
| 971 | static void parse_op(struct parsed_op *op, char words[16][256], int wordc) |
| 972 | { |
| 973 | enum opr_lenmod lmod = OPLM_UNSPEC; |
| 974 | int prefix_flags = 0; |
| 975 | int regmask_ind; |
| 976 | int regmask; |
| 977 | int op_w = 0; |
| 978 | int opr = 0; |
| 979 | int w = 0; |
| 980 | int i, j; |
| 981 | |
| 982 | for (i = 0; i < ARRAY_SIZE(pref_table); i++) { |
| 983 | if (IS(words[w], pref_table[i].name)) { |
| 984 | prefix_flags = pref_table[i].flags; |
| 985 | break; |
| 986 | } |
| 987 | } |
| 988 | |
| 989 | if (prefix_flags) { |
| 990 | if (wordc <= 1) |
| 991 | aerr("lone prefix: '%s'\n", words[0]); |
| 992 | w++; |
| 993 | } |
| 994 | |
| 995 | op_w = w; |
| 996 | for (i = 0; i < ARRAY_SIZE(op_table); i++) { |
| 997 | if (IS(words[w], op_table[i].name)) |
| 998 | break; |
| 999 | } |
| 1000 | |
| 1001 | if (i == ARRAY_SIZE(op_table)) { |
| 1002 | if (!g_skip_func) |
| 1003 | aerr("unhandled op: '%s'\n", words[0]); |
| 1004 | i--; // OP_UD2 |
| 1005 | } |
| 1006 | w++; |
| 1007 | |
| 1008 | op->op = op_table[i].op; |
| 1009 | op->flags = op_table[i].flags | prefix_flags; |
| 1010 | op->pfo = op_table[i].pfo; |
| 1011 | op->pfo_inv = op_table[i].pfo_inv; |
| 1012 | op->regmask_src = op->regmask_dst = 0; |
| 1013 | op->asmln = asmln; |
| 1014 | |
| 1015 | if (op->op == OP_UD2) |
| 1016 | return; |
| 1017 | |
| 1018 | for (opr = 0; opr < op_table[i].maxopr; opr++) { |
| 1019 | if (opr >= op_table[i].minopr && w >= wordc) |
| 1020 | break; |
| 1021 | |
| 1022 | regmask = regmask_ind = 0; |
| 1023 | w = parse_operand(&op->operand[opr], ®mask, ®mask_ind, |
| 1024 | words, wordc, w, op->flags); |
| 1025 | |
| 1026 | if (opr == 0 && (op->flags & OPF_DATA)) |
| 1027 | op->regmask_dst = regmask; |
| 1028 | else |
| 1029 | op->regmask_src |= regmask; |
| 1030 | op->regmask_src |= regmask_ind; |
| 1031 | } |
| 1032 | |
| 1033 | if (w < wordc) |
| 1034 | aerr("parse_op %s incomplete: %d/%d\n", |
| 1035 | words[0], w, wordc); |
| 1036 | |
| 1037 | // special cases |
| 1038 | op->operand_cnt = opr; |
| 1039 | if (!strncmp(op_table[i].name, "set", 3)) |
| 1040 | op->operand[0].lmod = OPLM_BYTE; |
| 1041 | |
| 1042 | switch (op->op) { |
| 1043 | // first operand is not dst |
| 1044 | case OP_CMP: |
| 1045 | case OP_TEST: |
| 1046 | op->regmask_src |= op->regmask_dst; |
| 1047 | op->regmask_dst = 0; |
| 1048 | break; |
| 1049 | |
| 1050 | // first operand is src too |
| 1051 | case OP_NOT: |
| 1052 | case OP_ADD: |
| 1053 | case OP_AND: |
| 1054 | case OP_OR: |
| 1055 | case OP_RCL: |
| 1056 | case OP_RCR: |
| 1057 | case OP_ADC: |
| 1058 | case OP_INC: |
| 1059 | case OP_DEC: |
| 1060 | case OP_NEG: |
| 1061 | // more below.. |
| 1062 | op->regmask_src |= op->regmask_dst; |
| 1063 | break; |
| 1064 | |
| 1065 | // special |
| 1066 | case OP_XCHG: |
| 1067 | op->regmask_src |= op->regmask_dst; |
| 1068 | op->regmask_dst |= op->regmask_src; |
| 1069 | goto check_align; |
| 1070 | |
| 1071 | case OP_SUB: |
| 1072 | case OP_SBB: |
| 1073 | case OP_XOR: |
| 1074 | if (op->operand[0].type == OPT_REG && op->operand[1].type == OPT_REG |
| 1075 | && op->operand[0].lmod == op->operand[1].lmod |
| 1076 | && op->operand[0].reg == op->operand[1].reg |
| 1077 | && IS(op->operand[0].name, op->operand[1].name)) // ! ah, al.. |
| 1078 | { |
| 1079 | op->regmask_src = 0; |
| 1080 | } |
| 1081 | else |
| 1082 | op->regmask_src |= op->regmask_dst; |
| 1083 | break; |
| 1084 | |
| 1085 | // ops with implicit argumets |
| 1086 | case OP_XLAT: |
| 1087 | op->operand_cnt = 2; |
| 1088 | setup_reg_opr(&op->operand[0], xAX, OPLM_BYTE, &op->regmask_src); |
| 1089 | op->regmask_dst = op->regmask_src; |
| 1090 | setup_reg_opr(&op->operand[1], xDX, OPLM_DWORD, &op->regmask_src); |
| 1091 | break; |
| 1092 | |
| 1093 | case OP_CDQ: |
| 1094 | op->operand_cnt = 2; |
| 1095 | setup_reg_opr(&op->operand[0], xDX, OPLM_DWORD, &op->regmask_dst); |
| 1096 | setup_reg_opr(&op->operand[1], xAX, OPLM_DWORD, &op->regmask_src); |
| 1097 | break; |
| 1098 | |
| 1099 | case OP_LODS: |
| 1100 | case OP_STOS: |
| 1101 | case OP_SCAS: |
| 1102 | if (op->operand_cnt != 0) |
| 1103 | break; |
| 1104 | if (words[op_w][4] == 'b') |
| 1105 | lmod = OPLM_BYTE; |
| 1106 | else if (words[op_w][4] == 'w') |
| 1107 | lmod = OPLM_WORD; |
| 1108 | else if (words[op_w][4] == 'd') |
| 1109 | lmod = OPLM_DWORD; |
| 1110 | j = 0; |
| 1111 | setup_reg_opr(&op->operand[j++], op->op == OP_LODS ? xSI : xDI, |
| 1112 | lmod, &op->regmask_src); |
| 1113 | if (op->flags & OPF_REP) |
| 1114 | setup_reg_opr(&op->operand[j++], xCX, OPLM_DWORD, &op->regmask_src); |
| 1115 | op->regmask_dst = op->regmask_src; |
| 1116 | setup_reg_opr(&op->operand[j++], xAX, OPLM_DWORD, |
| 1117 | op->op == OP_LODS ? &op->regmask_dst : &op->regmask_src); |
| 1118 | op->operand_cnt = j; |
| 1119 | break; |
| 1120 | |
| 1121 | case OP_MOVS: |
| 1122 | case OP_CMPS: |
| 1123 | if (op->operand_cnt != 0) |
| 1124 | break; |
| 1125 | if (words[op_w][4] == 'b') |
| 1126 | lmod = OPLM_BYTE; |
| 1127 | else if (words[op_w][4] == 'w') |
| 1128 | lmod = OPLM_WORD; |
| 1129 | else if (words[op_w][4] == 'd') |
| 1130 | lmod = OPLM_DWORD; |
| 1131 | j = 0; |
| 1132 | setup_reg_opr(&op->operand[j++], xDI, lmod, &op->regmask_src); |
| 1133 | setup_reg_opr(&op->operand[j++], xSI, OPLM_DWORD, &op->regmask_src); |
| 1134 | if (op->flags & OPF_REP) |
| 1135 | setup_reg_opr(&op->operand[j++], xCX, OPLM_DWORD, &op->regmask_src); |
| 1136 | op->operand_cnt = j; |
| 1137 | op->regmask_dst = op->regmask_src; |
| 1138 | break; |
| 1139 | |
| 1140 | case OP_LOOP: |
| 1141 | op->regmask_dst = 1 << xCX; |
| 1142 | // fallthrough |
| 1143 | case OP_JECXZ: |
| 1144 | op->operand_cnt = 2; |
| 1145 | op->regmask_src = 1 << xCX; |
| 1146 | op->operand[1].type = OPT_REG; |
| 1147 | op->operand[1].reg = xCX; |
| 1148 | op->operand[1].lmod = OPLM_DWORD; |
| 1149 | break; |
| 1150 | |
| 1151 | case OP_IMUL: |
| 1152 | if (op->operand_cnt == 2) { |
| 1153 | if (op->operand[0].type != OPT_REG) |
| 1154 | aerr("reg expected\n"); |
| 1155 | op->regmask_src |= 1 << op->operand[0].reg; |
| 1156 | } |
| 1157 | if (op->operand_cnt != 1) |
| 1158 | break; |
| 1159 | // fallthrough |
| 1160 | case OP_MUL: |
| 1161 | // singleop mul |
| 1162 | op->regmask_src |= op->regmask_dst; |
| 1163 | op->regmask_dst = (1 << xDX) | (1 << xAX); |
| 1164 | if (op->operand[0].lmod == OPLM_UNSPEC) |
| 1165 | op->operand[0].lmod = OPLM_DWORD; |
| 1166 | break; |
| 1167 | |
| 1168 | case OP_DIV: |
| 1169 | case OP_IDIV: |
| 1170 | // we could set up operands for edx:eax, but there is no real need to |
| 1171 | // (see is_opr_modified()) |
| 1172 | op->regmask_src |= op->regmask_dst; |
| 1173 | op->regmask_dst = (1 << xDX) | (1 << xAX); |
| 1174 | if (op->operand[0].lmod == OPLM_UNSPEC) |
| 1175 | op->operand[0].lmod = OPLM_DWORD; |
| 1176 | break; |
| 1177 | |
| 1178 | case OP_SHL: |
| 1179 | case OP_SHR: |
| 1180 | case OP_SAR: |
| 1181 | case OP_ROL: |
| 1182 | case OP_ROR: |
| 1183 | op->regmask_src |= op->regmask_dst; |
| 1184 | if (op->operand[1].lmod == OPLM_UNSPEC) |
| 1185 | op->operand[1].lmod = OPLM_BYTE; |
| 1186 | break; |
| 1187 | |
| 1188 | case OP_SHLD: |
| 1189 | case OP_SHRD: |
| 1190 | op->regmask_src |= op->regmask_dst; |
| 1191 | if (op->operand[2].lmod == OPLM_UNSPEC) |
| 1192 | op->operand[2].lmod = OPLM_BYTE; |
| 1193 | break; |
| 1194 | |
| 1195 | case OP_PUSH: |
| 1196 | op->regmask_src |= op->regmask_dst; |
| 1197 | op->regmask_dst = 0; |
| 1198 | if (op->operand[0].lmod == OPLM_UNSPEC |
| 1199 | && (op->operand[0].type == OPT_CONST |
| 1200 | || op->operand[0].type == OPT_OFFSET |
| 1201 | || op->operand[0].type == OPT_LABEL)) |
| 1202 | op->operand[0].lmod = OPLM_DWORD; |
| 1203 | break; |
| 1204 | |
| 1205 | // alignment |
| 1206 | case OP_MOV: |
| 1207 | check_align: |
| 1208 | if (op->operand[0].type == OPT_REG && op->operand[1].type == OPT_REG |
| 1209 | && op->operand[0].lmod == op->operand[1].lmod |
| 1210 | && op->operand[0].reg == op->operand[1].reg |
| 1211 | && IS(op->operand[0].name, op->operand[1].name)) // ! ah, al.. |
| 1212 | { |
| 1213 | op->flags |= OPF_RMD | OPF_DONE | OPF_NOREGS; |
| 1214 | op->regmask_src = op->regmask_dst = 0; |
| 1215 | } |
| 1216 | break; |
| 1217 | |
| 1218 | case OP_LEA: |
| 1219 | if (op->operand[0].type == OPT_REG |
| 1220 | && op->operand[1].type == OPT_REGMEM) |
| 1221 | { |
| 1222 | char buf[16]; |
| 1223 | snprintf(buf, sizeof(buf), "%s+0", op->operand[0].name); |
| 1224 | if (IS(buf, op->operand[1].name)) |
| 1225 | op->flags |= OPF_RMD | OPF_DONE | OPF_NOREGS; |
| 1226 | } |
| 1227 | break; |
| 1228 | |
| 1229 | case OP_CALL: |
| 1230 | // trashed regs must be explicitly detected later |
| 1231 | op->regmask_dst = 0; |
| 1232 | break; |
| 1233 | |
| 1234 | case OP_LEAVE: |
| 1235 | op->regmask_dst = (1 << xBP) | (1 << xSP); |
| 1236 | op->regmask_src = 1 << xBP; |
| 1237 | break; |
| 1238 | |
| 1239 | default: |
| 1240 | break; |
| 1241 | } |
| 1242 | |
| 1243 | if (op->operand[0].type == OPT_REG |
| 1244 | && op->operand[1].type == OPT_CONST) |
| 1245 | { |
| 1246 | struct parsed_opr *op1 = &op->operand[1]; |
| 1247 | if ((op->op == OP_AND && op1->val == 0) |
| 1248 | || (op->op == OP_OR |
| 1249 | && (op1->val == ~0 |
| 1250 | || (op->operand[0].lmod == OPLM_WORD && op1->val == 0xffff) |
| 1251 | || (op->operand[0].lmod == OPLM_BYTE && op1->val == 0xff)))) |
| 1252 | { |
| 1253 | op->regmask_src = 0; |
| 1254 | } |
| 1255 | } |
| 1256 | } |
| 1257 | |
| 1258 | static const char *op_name(struct parsed_op *po) |
| 1259 | { |
| 1260 | static char buf[16]; |
| 1261 | char *p; |
| 1262 | int i; |
| 1263 | |
| 1264 | if (po->op == OP_JCC || po->op == OP_SCC) { |
| 1265 | p = buf; |
| 1266 | *p++ = (po->op == OP_JCC) ? 'j' : 's'; |
| 1267 | if (po->pfo_inv) |
| 1268 | *p++ = 'n'; |
| 1269 | strcpy(p, parsed_flag_op_names[po->pfo]); |
| 1270 | return buf; |
| 1271 | } |
| 1272 | |
| 1273 | for (i = 0; i < ARRAY_SIZE(op_table); i++) |
| 1274 | if (op_table[i].op == po->op) |
| 1275 | return op_table[i].name; |
| 1276 | |
| 1277 | return "???"; |
| 1278 | } |
| 1279 | |
| 1280 | // debug |
| 1281 | static const char *dump_op(struct parsed_op *po) |
| 1282 | { |
| 1283 | static char out[128]; |
| 1284 | char *p = out; |
| 1285 | int i; |
| 1286 | |
| 1287 | if (po == NULL) |
| 1288 | return "???"; |
| 1289 | |
| 1290 | snprintf(out, sizeof(out), "%s", op_name(po)); |
| 1291 | for (i = 0; i < po->operand_cnt; i++) { |
| 1292 | p += strlen(p); |
| 1293 | if (i > 0) |
| 1294 | *p++ = ','; |
| 1295 | snprintf(p, sizeof(out) - (p - out), |
| 1296 | po->operand[i].type == OPT_REGMEM ? " [%s]" : " %s", |
| 1297 | po->operand[i].name); |
| 1298 | } |
| 1299 | |
| 1300 | return out; |
| 1301 | } |
| 1302 | |
| 1303 | static const char *lmod_type_u(struct parsed_op *po, |
| 1304 | enum opr_lenmod lmod) |
| 1305 | { |
| 1306 | switch (lmod) { |
| 1307 | case OPLM_QWORD: |
| 1308 | return "u64"; |
| 1309 | case OPLM_DWORD: |
| 1310 | return "u32"; |
| 1311 | case OPLM_WORD: |
| 1312 | return "u16"; |
| 1313 | case OPLM_BYTE: |
| 1314 | return "u8"; |
| 1315 | default: |
| 1316 | ferr(po, "invalid lmod: %d\n", lmod); |
| 1317 | return "(_invalid_)"; |
| 1318 | } |
| 1319 | } |
| 1320 | |
| 1321 | static const char *lmod_cast_u(struct parsed_op *po, |
| 1322 | enum opr_lenmod lmod) |
| 1323 | { |
| 1324 | switch (lmod) { |
| 1325 | case OPLM_QWORD: |
| 1326 | return ""; |
| 1327 | case OPLM_DWORD: |
| 1328 | return ""; |
| 1329 | case OPLM_WORD: |
| 1330 | return "(u16)"; |
| 1331 | case OPLM_BYTE: |
| 1332 | return "(u8)"; |
| 1333 | default: |
| 1334 | ferr(po, "invalid lmod: %d\n", lmod); |
| 1335 | return "(_invalid_)"; |
| 1336 | } |
| 1337 | } |
| 1338 | |
| 1339 | static const char *lmod_cast_u_ptr(struct parsed_op *po, |
| 1340 | enum opr_lenmod lmod) |
| 1341 | { |
| 1342 | switch (lmod) { |
| 1343 | case OPLM_QWORD: |
| 1344 | return "*(u64 *)"; |
| 1345 | case OPLM_DWORD: |
| 1346 | return "*(u32 *)"; |
| 1347 | case OPLM_WORD: |
| 1348 | return "*(u16 *)"; |
| 1349 | case OPLM_BYTE: |
| 1350 | return "*(u8 *)"; |
| 1351 | default: |
| 1352 | ferr(po, "invalid lmod: %d\n", lmod); |
| 1353 | return "(_invalid_)"; |
| 1354 | } |
| 1355 | } |
| 1356 | |
| 1357 | static const char *lmod_cast_s(struct parsed_op *po, |
| 1358 | enum opr_lenmod lmod) |
| 1359 | { |
| 1360 | switch (lmod) { |
| 1361 | case OPLM_QWORD: |
| 1362 | return "(s64)"; |
| 1363 | case OPLM_DWORD: |
| 1364 | return "(s32)"; |
| 1365 | case OPLM_WORD: |
| 1366 | return "(s16)"; |
| 1367 | case OPLM_BYTE: |
| 1368 | return "(s8)"; |
| 1369 | default: |
| 1370 | ferr(po, "%s: invalid lmod: %d\n", __func__, lmod); |
| 1371 | return "(_invalid_)"; |
| 1372 | } |
| 1373 | } |
| 1374 | |
| 1375 | static const char *lmod_cast(struct parsed_op *po, |
| 1376 | enum opr_lenmod lmod, int is_signed) |
| 1377 | { |
| 1378 | return is_signed ? |
| 1379 | lmod_cast_s(po, lmod) : |
| 1380 | lmod_cast_u(po, lmod); |
| 1381 | } |
| 1382 | |
| 1383 | static int lmod_bytes(struct parsed_op *po, enum opr_lenmod lmod) |
| 1384 | { |
| 1385 | switch (lmod) { |
| 1386 | case OPLM_QWORD: |
| 1387 | return 8; |
| 1388 | case OPLM_DWORD: |
| 1389 | return 4; |
| 1390 | case OPLM_WORD: |
| 1391 | return 2; |
| 1392 | case OPLM_BYTE: |
| 1393 | return 1; |
| 1394 | default: |
| 1395 | ferr(po, "%s: invalid lmod: %d\n", __func__, lmod); |
| 1396 | return 0; |
| 1397 | } |
| 1398 | } |
| 1399 | |
| 1400 | static const char *opr_name(struct parsed_op *po, int opr_num) |
| 1401 | { |
| 1402 | if (opr_num >= po->operand_cnt) |
| 1403 | ferr(po, "opr OOR: %d/%d\n", opr_num, po->operand_cnt); |
| 1404 | return po->operand[opr_num].name; |
| 1405 | } |
| 1406 | |
| 1407 | static unsigned int opr_const(struct parsed_op *po, int opr_num) |
| 1408 | { |
| 1409 | if (opr_num >= po->operand_cnt) |
| 1410 | ferr(po, "opr OOR: %d/%d\n", opr_num, po->operand_cnt); |
| 1411 | if (po->operand[opr_num].type != OPT_CONST) |
| 1412 | ferr(po, "opr %d: const expected\n", opr_num); |
| 1413 | return po->operand[opr_num].val; |
| 1414 | } |
| 1415 | |
| 1416 | static const char *opr_reg_p(struct parsed_op *po, struct parsed_opr *popr) |
| 1417 | { |
| 1418 | if ((unsigned int)popr->reg >= ARRAY_SIZE(regs_r32)) |
| 1419 | ferr(po, "invalid reg: %d\n", popr->reg); |
| 1420 | return regs_r32[popr->reg]; |
| 1421 | } |
| 1422 | |
| 1423 | // cast1 is the "final" cast |
| 1424 | static const char *simplify_cast(const char *cast1, const char *cast2) |
| 1425 | { |
| 1426 | static char buf[256]; |
| 1427 | |
| 1428 | if (cast1[0] == 0) |
| 1429 | return cast2; |
| 1430 | if (cast2[0] == 0) |
| 1431 | return cast1; |
| 1432 | if (IS(cast1, cast2)) |
| 1433 | return cast1; |
| 1434 | if (IS(cast1, "(s8)") && IS(cast2, "(u8)")) |
| 1435 | return cast1; |
| 1436 | if (IS(cast1, "(s16)") && IS(cast2, "(u16)")) |
| 1437 | return cast1; |
| 1438 | if (IS(cast1, "(u8)") && IS_START(cast2, "*(u8 *)")) |
| 1439 | return cast2; |
| 1440 | if (IS(cast1, "(u16)") && IS_START(cast2, "*(u16 *)")) |
| 1441 | return cast2; |
| 1442 | if (strchr(cast1, '*') && IS_START(cast2, "(u32)")) |
| 1443 | return cast1; |
| 1444 | |
| 1445 | snprintf(buf, sizeof(buf), "%s%s", cast1, cast2); |
| 1446 | return buf; |
| 1447 | } |
| 1448 | |
| 1449 | static const char *simplify_cast_num(const char *cast, unsigned int val) |
| 1450 | { |
| 1451 | if (IS(cast, "(u8)") && val < 0x100) |
| 1452 | return ""; |
| 1453 | if (IS(cast, "(s8)") && val < 0x80) |
| 1454 | return ""; |
| 1455 | if (IS(cast, "(u16)") && val < 0x10000) |
| 1456 | return ""; |
| 1457 | if (IS(cast, "(s16)") && val < 0x8000) |
| 1458 | return ""; |
| 1459 | if (IS(cast, "(s32)") && val < 0x80000000) |
| 1460 | return ""; |
| 1461 | |
| 1462 | return cast; |
| 1463 | } |
| 1464 | |
| 1465 | static struct parsed_equ *equ_find(struct parsed_op *po, const char *name, |
| 1466 | int *extra_offs) |
| 1467 | { |
| 1468 | const char *p; |
| 1469 | char *endp; |
| 1470 | int namelen; |
| 1471 | int i; |
| 1472 | |
| 1473 | *extra_offs = 0; |
| 1474 | namelen = strlen(name); |
| 1475 | |
| 1476 | p = strchr(name, '+'); |
| 1477 | if (p != NULL) { |
| 1478 | namelen = p - name; |
| 1479 | if (namelen <= 0) |
| 1480 | ferr(po, "equ parse failed for '%s'\n", name); |
| 1481 | |
| 1482 | if (IS_START(p, "0x")) |
| 1483 | p += 2; |
| 1484 | *extra_offs = strtol(p, &endp, 16); |
| 1485 | if (*endp != 0) |
| 1486 | ferr(po, "equ parse failed for '%s'\n", name); |
| 1487 | } |
| 1488 | |
| 1489 | for (i = 0; i < g_eqcnt; i++) |
| 1490 | if (strncmp(g_eqs[i].name, name, namelen) == 0 |
| 1491 | && g_eqs[i].name[namelen] == 0) |
| 1492 | break; |
| 1493 | if (i >= g_eqcnt) { |
| 1494 | if (po != NULL) |
| 1495 | ferr(po, "unresolved equ name: '%s'\n", name); |
| 1496 | return NULL; |
| 1497 | } |
| 1498 | |
| 1499 | return &g_eqs[i]; |
| 1500 | } |
| 1501 | |
| 1502 | static int is_stack_access(struct parsed_op *po, |
| 1503 | const struct parsed_opr *popr) |
| 1504 | { |
| 1505 | return (parse_stack_el(popr->name, NULL, 0) |
| 1506 | || (g_bp_frame && !(po->flags & OPF_EBP_S) |
| 1507 | && IS_START(popr->name, "ebp"))); |
| 1508 | } |
| 1509 | |
| 1510 | static void parse_stack_access(struct parsed_op *po, |
| 1511 | const char *name, char *ofs_reg, int *offset_out, |
| 1512 | int *stack_ra_out, const char **bp_arg_out, int is_lea) |
| 1513 | { |
| 1514 | const char *bp_arg = ""; |
| 1515 | const char *p = NULL; |
| 1516 | struct parsed_equ *eq; |
| 1517 | char *endp = NULL; |
| 1518 | int stack_ra = 0; |
| 1519 | int offset = 0; |
| 1520 | |
| 1521 | ofs_reg[0] = 0; |
| 1522 | |
| 1523 | if (IS_START(name, "ebp-") |
| 1524 | || (IS_START(name, "ebp+") && '0' <= name[4] && name[4] <= '9')) |
| 1525 | { |
| 1526 | p = name + 4; |
| 1527 | if (IS_START(p, "0x")) |
| 1528 | p += 2; |
| 1529 | offset = strtoul(p, &endp, 16); |
| 1530 | if (name[3] == '-') |
| 1531 | offset = -offset; |
| 1532 | if (*endp != 0) |
| 1533 | ferr(po, "ebp- parse of '%s' failed\n", name); |
| 1534 | } |
| 1535 | else { |
| 1536 | bp_arg = parse_stack_el(name, ofs_reg, 0); |
| 1537 | snprintf(g_comment, sizeof(g_comment), "%s", bp_arg); |
| 1538 | eq = equ_find(po, bp_arg, &offset); |
| 1539 | if (eq == NULL) |
| 1540 | ferr(po, "detected but missing eq\n"); |
| 1541 | offset += eq->offset; |
| 1542 | } |
| 1543 | |
| 1544 | if (!strncmp(name, "ebp", 3)) |
| 1545 | stack_ra = 4; |
| 1546 | |
| 1547 | // yes it sometimes LEAs ra for compares.. |
| 1548 | if (!is_lea && ofs_reg[0] == 0 |
| 1549 | && stack_ra <= offset && offset < stack_ra + 4) |
| 1550 | { |
| 1551 | ferr(po, "reference to ra? %d %d\n", offset, stack_ra); |
| 1552 | } |
| 1553 | |
| 1554 | *offset_out = offset; |
| 1555 | *stack_ra_out = stack_ra; |
| 1556 | if (bp_arg_out) |
| 1557 | *bp_arg_out = bp_arg; |
| 1558 | } |
| 1559 | |
| 1560 | static int stack_frame_access(struct parsed_op *po, |
| 1561 | struct parsed_opr *popr, char *buf, size_t buf_size, |
| 1562 | const char *name, const char *cast, int is_src, int is_lea) |
| 1563 | { |
| 1564 | enum opr_lenmod tmp_lmod = OPLM_UNSPEC; |
| 1565 | const char *prefix = ""; |
| 1566 | const char *bp_arg = NULL; |
| 1567 | char ofs_reg[16] = { 0, }; |
| 1568 | int i, arg_i, arg_s; |
| 1569 | int unaligned = 0; |
| 1570 | int stack_ra = 0; |
| 1571 | int offset = 0; |
| 1572 | int retval = -1; |
| 1573 | int sf_ofs; |
| 1574 | int lim; |
| 1575 | |
| 1576 | if (po->flags & OPF_EBP_S) |
| 1577 | ferr(po, "stack_frame_access while ebp is scratch\n"); |
| 1578 | |
| 1579 | parse_stack_access(po, name, ofs_reg, &offset, |
| 1580 | &stack_ra, &bp_arg, is_lea); |
| 1581 | |
| 1582 | if (offset > stack_ra) |
| 1583 | { |
| 1584 | arg_i = (offset - stack_ra - 4) / 4; |
| 1585 | if (arg_i < 0 || arg_i >= g_func_pp->argc_stack) |
| 1586 | { |
| 1587 | if (g_func_pp->is_vararg |
| 1588 | && arg_i == g_func_pp->argc_stack && is_lea) |
| 1589 | { |
| 1590 | // should be va_list |
| 1591 | if (cast[0] == 0) |
| 1592 | cast = "(u32)"; |
| 1593 | snprintf(buf, buf_size, "%sap", cast); |
| 1594 | return -1; |
| 1595 | } |
| 1596 | ferr(po, "offset %d (%s,%d) doesn't map to any arg\n", |
| 1597 | offset, bp_arg, arg_i); |
| 1598 | } |
| 1599 | if (ofs_reg[0] != 0) |
| 1600 | ferr(po, "offset reg on arg access?\n"); |
| 1601 | |
| 1602 | for (i = arg_s = 0; i < g_func_pp->argc; i++) { |
| 1603 | if (g_func_pp->arg[i].reg != NULL) |
| 1604 | continue; |
| 1605 | if (arg_s == arg_i) |
| 1606 | break; |
| 1607 | arg_s++; |
| 1608 | } |
| 1609 | if (i == g_func_pp->argc) |
| 1610 | ferr(po, "arg %d not in prototype?\n", arg_i); |
| 1611 | |
| 1612 | popr->is_ptr = g_func_pp->arg[i].type.is_ptr; |
| 1613 | retval = i; |
| 1614 | |
| 1615 | switch (popr->lmod) |
| 1616 | { |
| 1617 | case OPLM_BYTE: |
| 1618 | if (is_lea) |
| 1619 | ferr(po, "lea/byte to arg?\n"); |
| 1620 | if (is_src && (offset & 3) == 0) |
| 1621 | snprintf(buf, buf_size, "%sa%d", |
| 1622 | simplify_cast(cast, "(u8)"), i + 1); |
| 1623 | else |
| 1624 | snprintf(buf, buf_size, "%sBYTE%d(a%d)", |
| 1625 | cast, offset & 3, i + 1); |
| 1626 | break; |
| 1627 | |
| 1628 | case OPLM_WORD: |
| 1629 | if (is_lea) |
| 1630 | ferr(po, "lea/word to arg?\n"); |
| 1631 | if (offset & 1) { |
| 1632 | unaligned = 1; |
| 1633 | if (!is_src) { |
| 1634 | if (offset & 2) |
| 1635 | ferr(po, "problematic arg store\n"); |
| 1636 | snprintf(buf, buf_size, "%s((char *)&a%d + 1)", |
| 1637 | simplify_cast(cast, "*(u16 *)"), i + 1); |
| 1638 | } |
| 1639 | else |
| 1640 | ferr(po, "unaligned arg word load\n"); |
| 1641 | } |
| 1642 | else if (is_src && (offset & 2) == 0) |
| 1643 | snprintf(buf, buf_size, "%sa%d", |
| 1644 | simplify_cast(cast, "(u16)"), i + 1); |
| 1645 | else |
| 1646 | snprintf(buf, buf_size, "%s%sWORD(a%d)", |
| 1647 | cast, (offset & 2) ? "HI" : "LO", i + 1); |
| 1648 | break; |
| 1649 | |
| 1650 | case OPLM_DWORD: |
| 1651 | if (cast[0]) |
| 1652 | prefix = cast; |
| 1653 | else if (is_src) |
| 1654 | prefix = "(u32)"; |
| 1655 | |
| 1656 | if (offset & 3) { |
| 1657 | unaligned = 1; |
| 1658 | if (is_lea) |
| 1659 | snprintf(buf, buf_size, "(u32)&a%d + %d", |
| 1660 | i + 1, offset & 3); |
| 1661 | else if (!is_src) |
| 1662 | ferr(po, "unaligned arg store\n"); |
| 1663 | else { |
| 1664 | // mov edx, [ebp+arg_4+2]; movsx ecx, dx |
| 1665 | snprintf(buf, buf_size, "%s(a%d >> %d)", |
| 1666 | prefix, i + 1, (offset & 3) * 8); |
| 1667 | } |
| 1668 | } |
| 1669 | else { |
| 1670 | snprintf(buf, buf_size, "%s%sa%d", |
| 1671 | prefix, is_lea ? "&" : "", i + 1); |
| 1672 | } |
| 1673 | break; |
| 1674 | |
| 1675 | default: |
| 1676 | ferr(po, "bp_arg bad lmod: %d\n", popr->lmod); |
| 1677 | } |
| 1678 | |
| 1679 | if (unaligned) |
| 1680 | snprintf(g_comment, sizeof(g_comment), "%s unaligned", bp_arg); |
| 1681 | |
| 1682 | // common problem |
| 1683 | guess_lmod_from_c_type(&tmp_lmod, &g_func_pp->arg[i].type); |
| 1684 | if (tmp_lmod != OPLM_DWORD |
| 1685 | && (unaligned || (!is_src && lmod_bytes(po, tmp_lmod) |
| 1686 | < lmod_bytes(po, popr->lmod) + (offset & 3)))) |
| 1687 | { |
| 1688 | ferr(po, "bp_arg arg%d/w offset %d and type '%s' is too small\n", |
| 1689 | i + 1, offset, g_func_pp->arg[i].type.name); |
| 1690 | } |
| 1691 | // can't check this because msvc likes to reuse |
| 1692 | // arg space for scratch.. |
| 1693 | //if (popr->is_ptr && popr->lmod != OPLM_DWORD) |
| 1694 | // ferr(po, "bp_arg arg%d: non-dword ptr access\n", i + 1); |
| 1695 | } |
| 1696 | else |
| 1697 | { |
| 1698 | if (g_stack_fsz == 0) |
| 1699 | ferr(po, "stack var access without stackframe\n"); |
| 1700 | g_stack_frame_used = 1; |
| 1701 | |
| 1702 | sf_ofs = g_stack_fsz + offset; |
| 1703 | lim = (ofs_reg[0] != 0) ? -4 : 0; |
| 1704 | if (offset > 0 || sf_ofs < lim) |
| 1705 | ferr(po, "bp_stack offset %d/%d\n", offset, g_stack_fsz); |
| 1706 | |
| 1707 | if (is_lea) |
| 1708 | prefix = "(u32)&"; |
| 1709 | else |
| 1710 | prefix = cast; |
| 1711 | |
| 1712 | switch (popr->lmod) |
| 1713 | { |
| 1714 | case OPLM_BYTE: |
| 1715 | snprintf(buf, buf_size, "%ssf.b[%d%s%s]", |
| 1716 | prefix, sf_ofs, ofs_reg[0] ? "+" : "", ofs_reg); |
| 1717 | break; |
| 1718 | |
| 1719 | case OPLM_WORD: |
| 1720 | if ((sf_ofs & 1) || ofs_reg[0] != 0) { |
| 1721 | // known unaligned or possibly unaligned |
| 1722 | strcat(g_comment, " unaligned"); |
| 1723 | if (prefix[0] == 0) |
| 1724 | prefix = "*(u16 *)&"; |
| 1725 | snprintf(buf, buf_size, "%ssf.b[%d%s%s]", |
| 1726 | prefix, sf_ofs, ofs_reg[0] ? "+" : "", ofs_reg); |
| 1727 | break; |
| 1728 | } |
| 1729 | snprintf(buf, buf_size, "%ssf.w[%d]", prefix, sf_ofs / 2); |
| 1730 | break; |
| 1731 | |
| 1732 | case OPLM_DWORD: |
| 1733 | if ((sf_ofs & 3) || ofs_reg[0] != 0) { |
| 1734 | // known unaligned or possibly unaligned |
| 1735 | strcat(g_comment, " unaligned"); |
| 1736 | if (prefix[0] == 0) |
| 1737 | prefix = "*(u32 *)&"; |
| 1738 | snprintf(buf, buf_size, "%ssf.b[%d%s%s]", |
| 1739 | prefix, sf_ofs, ofs_reg[0] ? "+" : "", ofs_reg); |
| 1740 | break; |
| 1741 | } |
| 1742 | snprintf(buf, buf_size, "%ssf.d[%d]", prefix, sf_ofs / 4); |
| 1743 | break; |
| 1744 | |
| 1745 | default: |
| 1746 | ferr(po, "bp_stack bad lmod: %d\n", popr->lmod); |
| 1747 | } |
| 1748 | } |
| 1749 | |
| 1750 | return retval; |
| 1751 | } |
| 1752 | |
| 1753 | static void check_func_pp(struct parsed_op *po, |
| 1754 | const struct parsed_proto *pp, const char *pfx) |
| 1755 | { |
| 1756 | enum opr_lenmod tmp_lmod; |
| 1757 | char buf[256]; |
| 1758 | int ret, i; |
| 1759 | |
| 1760 | if (pp->argc_reg != 0) { |
| 1761 | if (/*!g_allow_regfunc &&*/ !pp->is_fastcall) { |
| 1762 | pp_print(buf, sizeof(buf), pp); |
| 1763 | ferr(po, "%s: unexpected reg arg in icall: %s\n", pfx, buf); |
| 1764 | } |
| 1765 | if (pp->argc_stack > 0 && pp->argc_reg != 2) |
| 1766 | ferr(po, "%s: %d reg arg(s) with %d stack arg(s)\n", |
| 1767 | pfx, pp->argc_reg, pp->argc_stack); |
| 1768 | } |
| 1769 | |
| 1770 | // fptrs must use 32bit args, callsite might have no information and |
| 1771 | // lack a cast to smaller types, which results in incorrectly masked |
| 1772 | // args passed (callee may assume masked args, it does on ARM) |
| 1773 | if (!pp->is_osinc) { |
| 1774 | for (i = 0; i < pp->argc; i++) { |
| 1775 | ret = guess_lmod_from_c_type(&tmp_lmod, &pp->arg[i].type); |
| 1776 | if (ret && tmp_lmod != OPLM_DWORD) |
| 1777 | ferr(po, "reference to %s with arg%d '%s'\n", pp->name, |
| 1778 | i + 1, pp->arg[i].type.name); |
| 1779 | } |
| 1780 | } |
| 1781 | } |
| 1782 | |
| 1783 | static const char *check_label_read_ref(struct parsed_op *po, |
| 1784 | const char *name) |
| 1785 | { |
| 1786 | const struct parsed_proto *pp; |
| 1787 | |
| 1788 | pp = proto_parse(g_fhdr, name, 0); |
| 1789 | if (pp == NULL) |
| 1790 | ferr(po, "proto_parse failed for ref '%s'\n", name); |
| 1791 | |
| 1792 | if (pp->is_func) |
| 1793 | check_func_pp(po, pp, "ref"); |
| 1794 | |
| 1795 | return pp->name; |
| 1796 | } |
| 1797 | |
| 1798 | static char *out_src_opr(char *buf, size_t buf_size, |
| 1799 | struct parsed_op *po, struct parsed_opr *popr, const char *cast, |
| 1800 | int is_lea) |
| 1801 | { |
| 1802 | char tmp1[256], tmp2[256]; |
| 1803 | char expr[256]; |
| 1804 | const char *name; |
| 1805 | char *p; |
| 1806 | int ret; |
| 1807 | |
| 1808 | if (cast == NULL) |
| 1809 | cast = ""; |
| 1810 | |
| 1811 | switch (popr->type) { |
| 1812 | case OPT_REG: |
| 1813 | if (is_lea) |
| 1814 | ferr(po, "lea from reg?\n"); |
| 1815 | |
| 1816 | switch (popr->lmod) { |
| 1817 | case OPLM_QWORD: |
| 1818 | snprintf(buf, buf_size, "%s%s.q", cast, opr_reg_p(po, popr)); |
| 1819 | break; |
| 1820 | case OPLM_DWORD: |
| 1821 | snprintf(buf, buf_size, "%s%s", cast, opr_reg_p(po, popr)); |
| 1822 | break; |
| 1823 | case OPLM_WORD: |
| 1824 | snprintf(buf, buf_size, "%s%s", |
| 1825 | simplify_cast(cast, "(u16)"), opr_reg_p(po, popr)); |
| 1826 | break; |
| 1827 | case OPLM_BYTE: |
| 1828 | if (popr->name[1] == 'h') // XXX.. |
| 1829 | snprintf(buf, buf_size, "%s(%s >> 8)", |
| 1830 | simplify_cast(cast, "(u8)"), opr_reg_p(po, popr)); |
| 1831 | else |
| 1832 | snprintf(buf, buf_size, "%s%s", |
| 1833 | simplify_cast(cast, "(u8)"), opr_reg_p(po, popr)); |
| 1834 | break; |
| 1835 | default: |
| 1836 | ferr(po, "invalid src lmod: %d\n", popr->lmod); |
| 1837 | } |
| 1838 | break; |
| 1839 | |
| 1840 | case OPT_REGMEM: |
| 1841 | if (is_stack_access(po, popr)) { |
| 1842 | stack_frame_access(po, popr, buf, buf_size, |
| 1843 | popr->name, cast, 1, is_lea); |
| 1844 | break; |
| 1845 | } |
| 1846 | |
| 1847 | strcpy(expr, popr->name); |
| 1848 | if (strchr(expr, '[')) { |
| 1849 | // special case: '[' can only be left for label[reg] form |
| 1850 | ret = sscanf(expr, "%[^[][%[^]]]", tmp1, tmp2); |
| 1851 | if (ret != 2) |
| 1852 | ferr(po, "parse failure for '%s'\n", expr); |
| 1853 | if (tmp1[0] == '(') { |
| 1854 | // (off_4FFF50+3)[eax] |
| 1855 | p = strchr(tmp1 + 1, ')'); |
| 1856 | if (p == NULL || p[1] != 0) |
| 1857 | ferr(po, "parse failure (2) for '%s'\n", expr); |
| 1858 | *p = 0; |
| 1859 | memmove(tmp1, tmp1 + 1, strlen(tmp1)); |
| 1860 | } |
| 1861 | snprintf(expr, sizeof(expr), "(u32)&%s + %s", tmp1, tmp2); |
| 1862 | } |
| 1863 | |
| 1864 | // XXX: do we need more parsing? |
| 1865 | if (is_lea) { |
| 1866 | snprintf(buf, buf_size, "%s", expr); |
| 1867 | break; |
| 1868 | } |
| 1869 | |
| 1870 | snprintf(buf, buf_size, "%s(%s)", |
| 1871 | simplify_cast(cast, lmod_cast_u_ptr(po, popr->lmod)), expr); |
| 1872 | break; |
| 1873 | |
| 1874 | case OPT_LABEL: |
| 1875 | name = check_label_read_ref(po, popr->name); |
| 1876 | if (cast[0] == 0 && popr->is_ptr) |
| 1877 | cast = "(u32)"; |
| 1878 | |
| 1879 | if (is_lea) |
| 1880 | snprintf(buf, buf_size, "(u32)&%s", name); |
| 1881 | else if (popr->size_lt) |
| 1882 | snprintf(buf, buf_size, "%s%s%s%s", cast, |
| 1883 | lmod_cast_u_ptr(po, popr->lmod), |
| 1884 | popr->is_array ? "" : "&", name); |
| 1885 | else |
| 1886 | snprintf(buf, buf_size, "%s%s%s", cast, name, |
| 1887 | popr->is_array ? "[0]" : ""); |
| 1888 | break; |
| 1889 | |
| 1890 | case OPT_OFFSET: |
| 1891 | name = check_label_read_ref(po, popr->name); |
| 1892 | if (cast[0] == 0) |
| 1893 | cast = "(u32)"; |
| 1894 | if (is_lea) |
| 1895 | ferr(po, "lea an offset?\n"); |
| 1896 | snprintf(buf, buf_size, "%s&%s", cast, name); |
| 1897 | break; |
| 1898 | |
| 1899 | case OPT_CONST: |
| 1900 | if (is_lea) |
| 1901 | ferr(po, "lea from const?\n"); |
| 1902 | |
| 1903 | printf_number(tmp1, sizeof(tmp1), popr->val); |
| 1904 | if (popr->val == 0 && strchr(cast, '*')) |
| 1905 | snprintf(buf, buf_size, "NULL"); |
| 1906 | else |
| 1907 | snprintf(buf, buf_size, "%s%s", |
| 1908 | simplify_cast_num(cast, popr->val), tmp1); |
| 1909 | break; |
| 1910 | |
| 1911 | default: |
| 1912 | ferr(po, "invalid src type: %d\n", popr->type); |
| 1913 | } |
| 1914 | |
| 1915 | return buf; |
| 1916 | } |
| 1917 | |
| 1918 | // note: may set is_ptr (we find that out late for ebp frame..) |
| 1919 | static char *out_dst_opr(char *buf, size_t buf_size, |
| 1920 | struct parsed_op *po, struct parsed_opr *popr) |
| 1921 | { |
| 1922 | switch (popr->type) { |
| 1923 | case OPT_REG: |
| 1924 | switch (popr->lmod) { |
| 1925 | case OPLM_QWORD: |
| 1926 | snprintf(buf, buf_size, "%s.q", opr_reg_p(po, popr)); |
| 1927 | break; |
| 1928 | case OPLM_DWORD: |
| 1929 | snprintf(buf, buf_size, "%s", opr_reg_p(po, popr)); |
| 1930 | break; |
| 1931 | case OPLM_WORD: |
| 1932 | // ugh.. |
| 1933 | snprintf(buf, buf_size, "LOWORD(%s)", opr_reg_p(po, popr)); |
| 1934 | break; |
| 1935 | case OPLM_BYTE: |
| 1936 | // ugh.. |
| 1937 | if (popr->name[1] == 'h') // XXX.. |
| 1938 | snprintf(buf, buf_size, "BYTE1(%s)", opr_reg_p(po, popr)); |
| 1939 | else |
| 1940 | snprintf(buf, buf_size, "LOBYTE(%s)", opr_reg_p(po, popr)); |
| 1941 | break; |
| 1942 | default: |
| 1943 | ferr(po, "invalid dst lmod: %d\n", popr->lmod); |
| 1944 | } |
| 1945 | break; |
| 1946 | |
| 1947 | case OPT_REGMEM: |
| 1948 | if (is_stack_access(po, popr)) { |
| 1949 | stack_frame_access(po, popr, buf, buf_size, |
| 1950 | popr->name, "", 0, 0); |
| 1951 | break; |
| 1952 | } |
| 1953 | |
| 1954 | return out_src_opr(buf, buf_size, po, popr, NULL, 0); |
| 1955 | |
| 1956 | case OPT_LABEL: |
| 1957 | if (popr->size_mismatch) |
| 1958 | snprintf(buf, buf_size, "%s%s%s", |
| 1959 | lmod_cast_u_ptr(po, popr->lmod), |
| 1960 | popr->is_array ? "" : "&", popr->name); |
| 1961 | else |
| 1962 | snprintf(buf, buf_size, "%s%s", popr->name, |
| 1963 | popr->is_array ? "[0]" : ""); |
| 1964 | break; |
| 1965 | |
| 1966 | default: |
| 1967 | ferr(po, "invalid dst type: %d\n", popr->type); |
| 1968 | } |
| 1969 | |
| 1970 | return buf; |
| 1971 | } |
| 1972 | |
| 1973 | static char *out_src_opr_u32(char *buf, size_t buf_size, |
| 1974 | struct parsed_op *po, struct parsed_opr *popr) |
| 1975 | { |
| 1976 | return out_src_opr(buf, buf_size, po, popr, NULL, 0); |
| 1977 | } |
| 1978 | |
| 1979 | static void out_test_for_cc(char *buf, size_t buf_size, |
| 1980 | struct parsed_op *po, enum parsed_flag_op pfo, int is_inv, |
| 1981 | enum opr_lenmod lmod, const char *expr) |
| 1982 | { |
| 1983 | const char *cast, *scast; |
| 1984 | |
| 1985 | cast = lmod_cast_u(po, lmod); |
| 1986 | scast = lmod_cast_s(po, lmod); |
| 1987 | |
| 1988 | switch (pfo) { |
| 1989 | case PFO_Z: |
| 1990 | case PFO_BE: // CF=1||ZF=1; CF=0 |
| 1991 | snprintf(buf, buf_size, "(%s%s %s 0)", |
| 1992 | cast, expr, is_inv ? "!=" : "=="); |
| 1993 | break; |
| 1994 | |
| 1995 | case PFO_S: |
| 1996 | case PFO_L: // SF!=OF; OF=0 |
| 1997 | snprintf(buf, buf_size, "(%s%s %s 0)", |
| 1998 | scast, expr, is_inv ? ">=" : "<"); |
| 1999 | break; |
| 2000 | |
| 2001 | case PFO_LE: // ZF=1||SF!=OF; OF=0 |
| 2002 | snprintf(buf, buf_size, "(%s%s %s 0)", |
| 2003 | scast, expr, is_inv ? ">" : "<="); |
| 2004 | break; |
| 2005 | |
| 2006 | default: |
| 2007 | ferr(po, "%s: unhandled parsed_flag_op: %d\n", __func__, pfo); |
| 2008 | } |
| 2009 | } |
| 2010 | |
| 2011 | static void out_cmp_for_cc(char *buf, size_t buf_size, |
| 2012 | struct parsed_op *po, enum parsed_flag_op pfo, int is_inv) |
| 2013 | { |
| 2014 | const char *cast, *scast, *cast_use; |
| 2015 | char buf1[256], buf2[256]; |
| 2016 | enum opr_lenmod lmod; |
| 2017 | |
| 2018 | if (po->op != OP_DEC && po->operand[0].lmod != po->operand[1].lmod) |
| 2019 | ferr(po, "%s: lmod mismatch: %d %d\n", __func__, |
| 2020 | po->operand[0].lmod, po->operand[1].lmod); |
| 2021 | lmod = po->operand[0].lmod; |
| 2022 | |
| 2023 | cast = lmod_cast_u(po, lmod); |
| 2024 | scast = lmod_cast_s(po, lmod); |
| 2025 | |
| 2026 | switch (pfo) { |
| 2027 | case PFO_C: |
| 2028 | case PFO_Z: |
| 2029 | case PFO_BE: // !a |
| 2030 | cast_use = cast; |
| 2031 | break; |
| 2032 | |
| 2033 | case PFO_S: |
| 2034 | case PFO_L: // !ge |
| 2035 | case PFO_LE: |
| 2036 | cast_use = scast; |
| 2037 | break; |
| 2038 | |
| 2039 | default: |
| 2040 | ferr(po, "%s: unhandled parsed_flag_op: %d\n", __func__, pfo); |
| 2041 | } |
| 2042 | |
| 2043 | out_src_opr(buf1, sizeof(buf1), po, &po->operand[0], cast_use, 0); |
| 2044 | if (po->op == OP_DEC) |
| 2045 | snprintf(buf2, sizeof(buf2), "1"); |
| 2046 | else |
| 2047 | out_src_opr(buf2, sizeof(buf2), po, &po->operand[1], cast_use, 0); |
| 2048 | |
| 2049 | switch (pfo) { |
| 2050 | case PFO_C: |
| 2051 | // note: must be unsigned compare |
| 2052 | snprintf(buf, buf_size, "(%s %s %s)", |
| 2053 | buf1, is_inv ? ">=" : "<", buf2); |
| 2054 | break; |
| 2055 | |
| 2056 | case PFO_Z: |
| 2057 | snprintf(buf, buf_size, "(%s %s %s)", |
| 2058 | buf1, is_inv ? "!=" : "==", buf2); |
| 2059 | break; |
| 2060 | |
| 2061 | case PFO_BE: // !a |
| 2062 | // note: must be unsigned compare |
| 2063 | snprintf(buf, buf_size, "(%s %s %s)", |
| 2064 | buf1, is_inv ? ">" : "<=", buf2); |
| 2065 | |
| 2066 | // annoying case |
| 2067 | if (is_inv && lmod == OPLM_BYTE |
| 2068 | && po->operand[1].type == OPT_CONST |
| 2069 | && po->operand[1].val == 0xff) |
| 2070 | { |
| 2071 | snprintf(g_comment, sizeof(g_comment), "if %s", buf); |
| 2072 | snprintf(buf, buf_size, "(0)"); |
| 2073 | } |
| 2074 | break; |
| 2075 | |
| 2076 | // note: must be signed compare |
| 2077 | case PFO_S: |
| 2078 | snprintf(buf, buf_size, "(%s(%s - %s) %s 0)", |
| 2079 | scast, buf1, buf2, is_inv ? ">=" : "<"); |
| 2080 | break; |
| 2081 | |
| 2082 | case PFO_L: // !ge |
| 2083 | snprintf(buf, buf_size, "(%s %s %s)", |
| 2084 | buf1, is_inv ? ">=" : "<", buf2); |
| 2085 | break; |
| 2086 | |
| 2087 | case PFO_LE: // !g |
| 2088 | snprintf(buf, buf_size, "(%s %s %s)", |
| 2089 | buf1, is_inv ? ">" : "<=", buf2); |
| 2090 | break; |
| 2091 | |
| 2092 | default: |
| 2093 | break; |
| 2094 | } |
| 2095 | } |
| 2096 | |
| 2097 | static void out_cmp_test(char *buf, size_t buf_size, |
| 2098 | struct parsed_op *po, enum parsed_flag_op pfo, int is_inv) |
| 2099 | { |
| 2100 | char buf1[256], buf2[256], buf3[256]; |
| 2101 | |
| 2102 | if (po->op == OP_TEST) { |
| 2103 | if (IS(opr_name(po, 0), opr_name(po, 1))) { |
| 2104 | out_src_opr_u32(buf3, sizeof(buf3), po, &po->operand[0]); |
| 2105 | } |
| 2106 | else { |
| 2107 | out_src_opr_u32(buf1, sizeof(buf1), po, &po->operand[0]); |
| 2108 | out_src_opr_u32(buf2, sizeof(buf2), po, &po->operand[1]); |
| 2109 | snprintf(buf3, sizeof(buf3), "(%s & %s)", buf1, buf2); |
| 2110 | } |
| 2111 | out_test_for_cc(buf, buf_size, po, pfo, is_inv, |
| 2112 | po->operand[0].lmod, buf3); |
| 2113 | } |
| 2114 | else if (po->op == OP_CMP) { |
| 2115 | out_cmp_for_cc(buf, buf_size, po, pfo, is_inv); |
| 2116 | } |
| 2117 | else |
| 2118 | ferr(po, "%s: unhandled op: %d\n", __func__, po->op); |
| 2119 | } |
| 2120 | |
| 2121 | static void propagate_lmod(struct parsed_op *po, struct parsed_opr *popr1, |
| 2122 | struct parsed_opr *popr2) |
| 2123 | { |
| 2124 | if (popr1->lmod == OPLM_UNSPEC && popr2->lmod == OPLM_UNSPEC) |
| 2125 | ferr(po, "missing lmod for both operands\n"); |
| 2126 | |
| 2127 | if (popr1->lmod == OPLM_UNSPEC) |
| 2128 | popr1->lmod = popr2->lmod; |
| 2129 | else if (popr2->lmod == OPLM_UNSPEC) |
| 2130 | popr2->lmod = popr1->lmod; |
| 2131 | else if (popr1->lmod != popr2->lmod) { |
| 2132 | if (popr1->type_from_var) { |
| 2133 | popr1->size_mismatch = 1; |
| 2134 | if (popr1->lmod < popr2->lmod) |
| 2135 | popr1->size_lt = 1; |
| 2136 | popr1->lmod = popr2->lmod; |
| 2137 | } |
| 2138 | else if (popr2->type_from_var) { |
| 2139 | popr2->size_mismatch = 1; |
| 2140 | if (popr2->lmod < popr1->lmod) |
| 2141 | popr2->size_lt = 1; |
| 2142 | popr2->lmod = popr1->lmod; |
| 2143 | } |
| 2144 | else |
| 2145 | ferr(po, "conflicting lmods: %d vs %d\n", |
| 2146 | popr1->lmod, popr2->lmod); |
| 2147 | } |
| 2148 | } |
| 2149 | |
| 2150 | static const char *op_to_c(struct parsed_op *po) |
| 2151 | { |
| 2152 | switch (po->op) |
| 2153 | { |
| 2154 | case OP_ADD: |
| 2155 | case OP_ADC: |
| 2156 | return "+"; |
| 2157 | case OP_SUB: |
| 2158 | case OP_SBB: |
| 2159 | return "-"; |
| 2160 | case OP_AND: |
| 2161 | return "&"; |
| 2162 | case OP_OR: |
| 2163 | return "|"; |
| 2164 | case OP_XOR: |
| 2165 | return "^"; |
| 2166 | case OP_SHL: |
| 2167 | return "<<"; |
| 2168 | case OP_SHR: |
| 2169 | return ">>"; |
| 2170 | case OP_MUL: |
| 2171 | case OP_IMUL: |
| 2172 | return "*"; |
| 2173 | default: |
| 2174 | ferr(po, "op_to_c was supplied with %d\n", po->op); |
| 2175 | } |
| 2176 | } |
| 2177 | |
| 2178 | // last op in stream - unconditional branch or ret |
| 2179 | #define LAST_OP(_i) ((ops[_i].flags & OPF_TAIL) \ |
| 2180 | || ((ops[_i].flags & (OPF_JMP|OPF_CJMP|OPF_RMD)) == OPF_JMP \ |
| 2181 | && ops[_i].op != OP_CALL)) |
| 2182 | |
| 2183 | #define check_i(po, i) \ |
| 2184 | if ((i) < 0) \ |
| 2185 | ferr(po, "bad " #i ": %d\n", i) |
| 2186 | |
| 2187 | // note: this skips over calls and rm'd stuff assuming they're handled |
| 2188 | // so it's intended to use at one of final passes |
| 2189 | static int scan_for_pop(int i, int opcnt, int magic, int reg, |
| 2190 | int depth, int flags_set) |
| 2191 | { |
| 2192 | struct parsed_op *po; |
| 2193 | int relevant; |
| 2194 | int ret = 0; |
| 2195 | int j; |
| 2196 | |
| 2197 | for (; i < opcnt; i++) { |
| 2198 | po = &ops[i]; |
| 2199 | if (po->cc_scratch == magic) |
| 2200 | return ret; // already checked |
| 2201 | po->cc_scratch = magic; |
| 2202 | |
| 2203 | if (po->flags & OPF_TAIL) { |
| 2204 | if (po->op == OP_CALL) { |
| 2205 | if (po->pp != NULL && po->pp->is_noreturn) |
| 2206 | // assume no stack cleanup for noreturn |
| 2207 | return 1; |
| 2208 | } |
| 2209 | return -1; // deadend |
| 2210 | } |
| 2211 | |
| 2212 | if (po->flags & (OPF_RMD|OPF_DONE|OPF_FARG)) |
| 2213 | continue; |
| 2214 | |
| 2215 | if ((po->flags & OPF_JMP) && po->op != OP_CALL) { |
| 2216 | if (po->btj != NULL) { |
| 2217 | // jumptable |
| 2218 | for (j = 0; j < po->btj->count; j++) { |
| 2219 | check_i(po, po->btj->d[j].bt_i); |
| 2220 | ret |= scan_for_pop(po->btj->d[j].bt_i, opcnt, magic, reg, |
| 2221 | depth, flags_set); |
| 2222 | if (ret < 0) |
| 2223 | return ret; // dead end |
| 2224 | } |
| 2225 | return ret; |
| 2226 | } |
| 2227 | |
| 2228 | check_i(po, po->bt_i); |
| 2229 | if (po->flags & OPF_CJMP) { |
| 2230 | ret |= scan_for_pop(po->bt_i, opcnt, magic, reg, |
| 2231 | depth, flags_set); |
| 2232 | if (ret < 0) |
| 2233 | return ret; // dead end |
| 2234 | } |
| 2235 | else { |
| 2236 | i = po->bt_i - 1; |
| 2237 | } |
| 2238 | continue; |
| 2239 | } |
| 2240 | |
| 2241 | relevant = 0; |
| 2242 | if ((po->op == OP_POP || po->op == OP_PUSH) |
| 2243 | && po->operand[0].type == OPT_REG && po->operand[0].reg == reg) |
| 2244 | { |
| 2245 | relevant = 1; |
| 2246 | } |
| 2247 | |
| 2248 | if (po->op == OP_PUSH) { |
| 2249 | depth++; |
| 2250 | } |
| 2251 | else if (po->op == OP_POP) { |
| 2252 | if (relevant && depth == 0) { |
| 2253 | po->flags |= flags_set; |
| 2254 | return 1; |
| 2255 | } |
| 2256 | depth--; |
| 2257 | } |
| 2258 | } |
| 2259 | |
| 2260 | return -1; |
| 2261 | } |
| 2262 | |
| 2263 | // scan for 'reg' pop backwards starting from i |
| 2264 | // intended to use for register restore search, so other reg |
| 2265 | // references are considered an error |
| 2266 | static int scan_for_rsave_pop_reg(int i, int magic, int reg, int set_flags) |
| 2267 | { |
| 2268 | struct parsed_op *po; |
| 2269 | struct label_ref *lr; |
| 2270 | int ret = 0; |
| 2271 | |
| 2272 | ops[i].cc_scratch = magic; |
| 2273 | |
| 2274 | while (1) |
| 2275 | { |
| 2276 | if (g_labels[i] != NULL) { |
| 2277 | lr = &g_label_refs[i]; |
| 2278 | for (; lr != NULL; lr = lr->next) { |
| 2279 | check_i(&ops[i], lr->i); |
| 2280 | ret |= scan_for_rsave_pop_reg(lr->i, magic, reg, set_flags); |
| 2281 | if (ret < 0) |
| 2282 | return ret; |
| 2283 | } |
| 2284 | if (i > 0 && LAST_OP(i - 1)) |
| 2285 | return ret; |
| 2286 | } |
| 2287 | |
| 2288 | i--; |
| 2289 | if (i < 0) |
| 2290 | break; |
| 2291 | |
| 2292 | if (ops[i].cc_scratch == magic) |
| 2293 | return ret; |
| 2294 | ops[i].cc_scratch = magic; |
| 2295 | |
| 2296 | po = &ops[i]; |
| 2297 | if (po->op == OP_POP && po->operand[0].reg == reg) { |
| 2298 | if (po->flags & (OPF_RMD|OPF_DONE)) |
| 2299 | return -1; |
| 2300 | |
| 2301 | po->flags |= set_flags; |
| 2302 | return 1; |
| 2303 | } |
| 2304 | |
| 2305 | // this also covers the case where we reach corresponding push |
| 2306 | if ((po->regmask_dst | po->regmask_src) & (1 << reg)) |
| 2307 | return -1; |
| 2308 | } |
| 2309 | |
| 2310 | // nothing interesting on this path |
| 2311 | return 0; |
| 2312 | } |
| 2313 | |
| 2314 | static void find_reachable_exits(int i, int opcnt, int magic, |
| 2315 | int *exits, int *exit_count) |
| 2316 | { |
| 2317 | struct parsed_op *po; |
| 2318 | int j; |
| 2319 | |
| 2320 | for (; i < opcnt; i++) |
| 2321 | { |
| 2322 | po = &ops[i]; |
| 2323 | if (po->cc_scratch == magic) |
| 2324 | return; |
| 2325 | po->cc_scratch = magic; |
| 2326 | |
| 2327 | if (po->flags & OPF_TAIL) { |
| 2328 | ferr_assert(po, *exit_count < MAX_EXITS); |
| 2329 | exits[*exit_count] = i; |
| 2330 | (*exit_count)++; |
| 2331 | return; |
| 2332 | } |
| 2333 | |
| 2334 | if ((po->flags & OPF_JMP) && po->op != OP_CALL) { |
| 2335 | if (po->flags & OPF_RMD) |
| 2336 | continue; |
| 2337 | |
| 2338 | if (po->btj != NULL) { |
| 2339 | for (j = 0; j < po->btj->count; j++) { |
| 2340 | check_i(po, po->btj->d[j].bt_i); |
| 2341 | find_reachable_exits(po->btj->d[j].bt_i, opcnt, magic, |
| 2342 | exits, exit_count); |
| 2343 | } |
| 2344 | return; |
| 2345 | } |
| 2346 | |
| 2347 | check_i(po, po->bt_i); |
| 2348 | if (po->flags & OPF_CJMP) |
| 2349 | find_reachable_exits(po->bt_i, opcnt, magic, exits, exit_count); |
| 2350 | else |
| 2351 | i = po->bt_i - 1; |
| 2352 | continue; |
| 2353 | } |
| 2354 | } |
| 2355 | } |
| 2356 | |
| 2357 | // scan for 'reg' pop backwards starting from exits (all paths) |
| 2358 | static int scan_for_pop_ret(int i, int opcnt, int reg, int set_flags) |
| 2359 | { |
| 2360 | static int exits[MAX_EXITS]; |
| 2361 | static int exit_count; |
| 2362 | int j, ret; |
| 2363 | |
| 2364 | if (!set_flags) { |
| 2365 | exit_count = 0; |
| 2366 | find_reachable_exits(i, opcnt, i + opcnt * 15, exits, |
| 2367 | &exit_count); |
| 2368 | ferr_assert(&ops[i], exit_count > 0); |
| 2369 | } |
| 2370 | |
| 2371 | for (j = 0; j < exit_count; j++) { |
| 2372 | ret = scan_for_rsave_pop_reg(exits[j], i + opcnt * 16 + set_flags, |
| 2373 | reg, set_flags); |
| 2374 | if (ret == -1) |
| 2375 | return -1; |
| 2376 | } |
| 2377 | |
| 2378 | return 1; |
| 2379 | } |
| 2380 | |
| 2381 | // scan for one or more pop of push <const> |
| 2382 | static int scan_for_pop_const_r(int i, int opcnt, int magic, |
| 2383 | int push_i, int is_probe) |
| 2384 | { |
| 2385 | struct parsed_op *po; |
| 2386 | struct label_ref *lr; |
| 2387 | int ret = 0; |
| 2388 | int j; |
| 2389 | |
| 2390 | for (; i < opcnt; i++) |
| 2391 | { |
| 2392 | po = &ops[i]; |
| 2393 | if (po->cc_scratch == magic) |
| 2394 | return ret; // already checked |
| 2395 | po->cc_scratch = magic; |
| 2396 | |
| 2397 | if (po->flags & OPF_JMP) { |
| 2398 | if (po->flags & OPF_RMD) |
| 2399 | continue; |
| 2400 | if (po->op == OP_CALL) |
| 2401 | return -1; |
| 2402 | |
| 2403 | if (po->btj != NULL) { |
| 2404 | for (j = 0; j < po->btj->count; j++) { |
| 2405 | check_i(po, po->btj->d[j].bt_i); |
| 2406 | ret |= scan_for_pop_const_r(po->btj->d[j].bt_i, opcnt, magic, |
| 2407 | push_i, is_probe); |
| 2408 | if (ret < 0) |
| 2409 | return ret; |
| 2410 | } |
| 2411 | return ret; |
| 2412 | } |
| 2413 | |
| 2414 | check_i(po, po->bt_i); |
| 2415 | if (po->flags & OPF_CJMP) { |
| 2416 | ret |= scan_for_pop_const_r(po->bt_i, opcnt, magic, push_i, |
| 2417 | is_probe); |
| 2418 | if (ret < 0) |
| 2419 | return ret; |
| 2420 | } |
| 2421 | else { |
| 2422 | i = po->bt_i - 1; |
| 2423 | } |
| 2424 | continue; |
| 2425 | } |
| 2426 | |
| 2427 | if ((po->flags & (OPF_TAIL|OPF_RSAVE)) || po->op == OP_PUSH) |
| 2428 | return -1; |
| 2429 | |
| 2430 | if (g_labels[i] != NULL) { |
| 2431 | // all refs must be visited |
| 2432 | lr = &g_label_refs[i]; |
| 2433 | for (; lr != NULL; lr = lr->next) { |
| 2434 | check_i(po, lr->i); |
| 2435 | if (ops[lr->i].cc_scratch != magic) |
| 2436 | return -1; |
| 2437 | } |
| 2438 | if (i > 0 && !LAST_OP(i - 1) && ops[i - 1].cc_scratch != magic) |
| 2439 | return -1; |
| 2440 | } |
| 2441 | |
| 2442 | if (po->op == OP_POP) |
| 2443 | { |
| 2444 | if (po->flags & (OPF_RMD|OPF_DONE)) |
| 2445 | return -1; |
| 2446 | |
| 2447 | if (!is_probe) { |
| 2448 | po->flags |= OPF_DONE; |
| 2449 | po->datap = &ops[push_i]; |
| 2450 | } |
| 2451 | return 1; |
| 2452 | } |
| 2453 | } |
| 2454 | |
| 2455 | return -1; |
| 2456 | } |
| 2457 | |
| 2458 | static void scan_for_pop_const(int i, int opcnt, int magic) |
| 2459 | { |
| 2460 | int ret; |
| 2461 | |
| 2462 | ret = scan_for_pop_const_r(i + 1, opcnt, magic, i, 1); |
| 2463 | if (ret == 1) { |
| 2464 | ops[i].flags |= OPF_RMD | OPF_DONE; |
| 2465 | scan_for_pop_const_r(i + 1, opcnt, magic + 1, i, 0); |
| 2466 | } |
| 2467 | } |
| 2468 | |
| 2469 | // check if all branch targets within a marked path are also marked |
| 2470 | // note: the path checked must not be empty or end with a branch |
| 2471 | static int check_path_branches(int opcnt, int magic) |
| 2472 | { |
| 2473 | struct parsed_op *po; |
| 2474 | int i, j; |
| 2475 | |
| 2476 | for (i = 0; i < opcnt; i++) { |
| 2477 | po = &ops[i]; |
| 2478 | if (po->cc_scratch != magic) |
| 2479 | continue; |
| 2480 | |
| 2481 | if (po->flags & OPF_JMP) { |
| 2482 | if ((po->flags & OPF_RMD) || po->op == OP_CALL) |
| 2483 | continue; |
| 2484 | |
| 2485 | if (po->btj != NULL) { |
| 2486 | for (j = 0; j < po->btj->count; j++) { |
| 2487 | check_i(po, po->btj->d[j].bt_i); |
| 2488 | if (ops[po->btj->d[j].bt_i].cc_scratch != magic) |
| 2489 | return 0; |
| 2490 | } |
| 2491 | } |
| 2492 | |
| 2493 | check_i(po, po->bt_i); |
| 2494 | if (ops[po->bt_i].cc_scratch != magic) |
| 2495 | return 0; |
| 2496 | if ((po->flags & OPF_CJMP) && ops[i + 1].cc_scratch != magic) |
| 2497 | return 0; |
| 2498 | } |
| 2499 | } |
| 2500 | |
| 2501 | return 1; |
| 2502 | } |
| 2503 | |
| 2504 | // scan for multiple pushes for given pop |
| 2505 | static int scan_pushes_for_pop_r(int i, int magic, int pop_i, |
| 2506 | int is_probe) |
| 2507 | { |
| 2508 | int reg = ops[pop_i].operand[0].reg; |
| 2509 | struct parsed_op *po; |
| 2510 | struct label_ref *lr; |
| 2511 | int ret = 0; |
| 2512 | |
| 2513 | ops[i].cc_scratch = magic; |
| 2514 | |
| 2515 | while (1) |
| 2516 | { |
| 2517 | if (g_labels[i] != NULL) { |
| 2518 | lr = &g_label_refs[i]; |
| 2519 | for (; lr != NULL; lr = lr->next) { |
| 2520 | check_i(&ops[i], lr->i); |
| 2521 | ret |= scan_pushes_for_pop_r(lr->i, magic, pop_i, is_probe); |
| 2522 | if (ret < 0) |
| 2523 | return ret; |
| 2524 | } |
| 2525 | if (i > 0 && LAST_OP(i - 1)) |
| 2526 | return ret; |
| 2527 | } |
| 2528 | |
| 2529 | i--; |
| 2530 | if (i < 0) |
| 2531 | break; |
| 2532 | |
| 2533 | if (ops[i].cc_scratch == magic) |
| 2534 | return ret; |
| 2535 | ops[i].cc_scratch = magic; |
| 2536 | |
| 2537 | po = &ops[i]; |
| 2538 | if (po->op == OP_CALL) |
| 2539 | return -1; |
| 2540 | if ((po->flags & (OPF_TAIL|OPF_RSAVE)) || po->op == OP_POP) |
| 2541 | return -1; |
| 2542 | |
| 2543 | if (po->op == OP_PUSH) |
| 2544 | { |
| 2545 | if (po->datap != NULL) |
| 2546 | return -1; |
| 2547 | if (po->operand[0].type == OPT_REG && po->operand[0].reg == reg) |
| 2548 | // leave this case for reg save/restore handlers |
| 2549 | return -1; |
| 2550 | |
| 2551 | if (!is_probe) { |
| 2552 | po->flags |= OPF_PPUSH | OPF_DONE; |
| 2553 | po->datap = &ops[pop_i]; |
| 2554 | } |
| 2555 | return 1; |
| 2556 | } |
| 2557 | } |
| 2558 | |
| 2559 | return -1; |
| 2560 | } |
| 2561 | |
| 2562 | static void scan_pushes_for_pop(int i, int opcnt, int *regmask_pp) |
| 2563 | { |
| 2564 | int magic = i + opcnt * 14; |
| 2565 | int ret; |
| 2566 | |
| 2567 | ret = scan_pushes_for_pop_r(i, magic, i, 1); |
| 2568 | if (ret == 1) { |
| 2569 | ret = check_path_branches(opcnt, magic); |
| 2570 | if (ret == 1) { |
| 2571 | ops[i].flags |= OPF_PPUSH | OPF_DONE; |
| 2572 | *regmask_pp |= 1 << ops[i].operand[0].reg; |
| 2573 | scan_pushes_for_pop_r(i, magic + 1, i, 0); |
| 2574 | } |
| 2575 | } |
| 2576 | } |
| 2577 | |
| 2578 | static void scan_propagate_df(int i, int opcnt) |
| 2579 | { |
| 2580 | struct parsed_op *po = &ops[i]; |
| 2581 | int j; |
| 2582 | |
| 2583 | for (; i < opcnt; i++) { |
| 2584 | po = &ops[i]; |
| 2585 | if (po->flags & OPF_DF) |
| 2586 | return; // already resolved |
| 2587 | po->flags |= OPF_DF; |
| 2588 | |
| 2589 | if (po->op == OP_CALL) |
| 2590 | ferr(po, "call with DF set?\n"); |
| 2591 | |
| 2592 | if (po->flags & OPF_JMP) { |
| 2593 | if (po->btj != NULL) { |
| 2594 | // jumptable |
| 2595 | for (j = 0; j < po->btj->count; j++) { |
| 2596 | check_i(po, po->btj->d[j].bt_i); |
| 2597 | scan_propagate_df(po->btj->d[j].bt_i, opcnt); |
| 2598 | } |
| 2599 | return; |
| 2600 | } |
| 2601 | |
| 2602 | if (po->flags & OPF_RMD) |
| 2603 | continue; |
| 2604 | check_i(po, po->bt_i); |
| 2605 | if (po->flags & OPF_CJMP) |
| 2606 | scan_propagate_df(po->bt_i, opcnt); |
| 2607 | else |
| 2608 | i = po->bt_i - 1; |
| 2609 | continue; |
| 2610 | } |
| 2611 | |
| 2612 | if (po->flags & OPF_TAIL) |
| 2613 | break; |
| 2614 | |
| 2615 | if (po->op == OP_CLD) { |
| 2616 | po->flags |= OPF_RMD | OPF_DONE; |
| 2617 | return; |
| 2618 | } |
| 2619 | } |
| 2620 | |
| 2621 | ferr(po, "missing DF clear?\n"); |
| 2622 | } |
| 2623 | |
| 2624 | // is operand 'opr' referenced by parsed_op 'po'? |
| 2625 | static int is_opr_referenced(const struct parsed_opr *opr, |
| 2626 | const struct parsed_op *po) |
| 2627 | { |
| 2628 | int i, mask; |
| 2629 | |
| 2630 | if (opr->type == OPT_REG) { |
| 2631 | mask = po->regmask_dst | po->regmask_src; |
| 2632 | if (po->op == OP_CALL) |
| 2633 | mask |= (1 << xAX) | (1 << xCX) | (1 << xDX); |
| 2634 | if ((1 << opr->reg) & mask) |
| 2635 | return 1; |
| 2636 | else |
| 2637 | return 0; |
| 2638 | } |
| 2639 | |
| 2640 | for (i = 0; i < po->operand_cnt; i++) |
| 2641 | if (IS(po->operand[0].name, opr->name)) |
| 2642 | return 1; |
| 2643 | |
| 2644 | return 0; |
| 2645 | } |
| 2646 | |
| 2647 | // is operand 'opr' read by parsed_op 'po'? |
| 2648 | static int is_opr_read(const struct parsed_opr *opr, |
| 2649 | const struct parsed_op *po) |
| 2650 | { |
| 2651 | if (opr->type == OPT_REG) { |
| 2652 | if (po->regmask_src & (1 << opr->reg)) |
| 2653 | return 1; |
| 2654 | else |
| 2655 | return 0; |
| 2656 | } |
| 2657 | |
| 2658 | // yes I'm lazy |
| 2659 | return 0; |
| 2660 | } |
| 2661 | |
| 2662 | // is operand 'opr' modified by parsed_op 'po'? |
| 2663 | static int is_opr_modified(const struct parsed_opr *opr, |
| 2664 | const struct parsed_op *po) |
| 2665 | { |
| 2666 | int mask; |
| 2667 | |
| 2668 | if (opr->type == OPT_REG) { |
| 2669 | if (po->op == OP_CALL) { |
| 2670 | mask = po->regmask_dst; |
| 2671 | mask |= (1 << xAX) | (1 << xCX) | (1 << xDX); // ? |
| 2672 | if (mask & (1 << opr->reg)) |
| 2673 | return 1; |
| 2674 | else |
| 2675 | return 0; |
| 2676 | } |
| 2677 | |
| 2678 | if (po->regmask_dst & (1 << opr->reg)) |
| 2679 | return 1; |
| 2680 | else |
| 2681 | return 0; |
| 2682 | } |
| 2683 | |
| 2684 | return IS(po->operand[0].name, opr->name); |
| 2685 | } |
| 2686 | |
| 2687 | // is any operand of parsed_op 'po_test' modified by parsed_op 'po'? |
| 2688 | static int is_any_opr_modified(const struct parsed_op *po_test, |
| 2689 | const struct parsed_op *po, int c_mode) |
| 2690 | { |
| 2691 | int mask; |
| 2692 | int i; |
| 2693 | |
| 2694 | if ((po->flags & OPF_RMD) || !(po->flags & OPF_DATA)) |
| 2695 | return 0; |
| 2696 | |
| 2697 | if (po_test->operand_cnt == 1 && po_test->operand[0].type == OPT_CONST) |
| 2698 | return 0; |
| 2699 | |
| 2700 | if ((po_test->regmask_src | po_test->regmask_dst) & po->regmask_dst) |
| 2701 | return 1; |
| 2702 | |
| 2703 | // in reality, it can wreck any register, but in decompiled C |
| 2704 | // version it can only overwrite eax or edx:eax |
| 2705 | mask = (1 << xAX) | (1 << xDX); |
| 2706 | if (!c_mode) |
| 2707 | mask |= 1 << xCX; |
| 2708 | |
| 2709 | if (po->op == OP_CALL |
| 2710 | && ((po_test->regmask_src | po_test->regmask_dst) & mask)) |
| 2711 | return 1; |
| 2712 | |
| 2713 | for (i = 0; i < po_test->operand_cnt; i++) |
| 2714 | if (IS(po_test->operand[i].name, po->operand[0].name)) |
| 2715 | return 1; |
| 2716 | |
| 2717 | return 0; |
| 2718 | } |
| 2719 | |
| 2720 | // scan for any po_test operand modification in range given |
| 2721 | static int scan_for_mod(struct parsed_op *po_test, int i, int opcnt, |
| 2722 | int c_mode) |
| 2723 | { |
| 2724 | if (po_test->operand_cnt == 1 && po_test->operand[0].type == OPT_CONST) |
| 2725 | return -1; |
| 2726 | |
| 2727 | for (; i < opcnt; i++) { |
| 2728 | if (is_any_opr_modified(po_test, &ops[i], c_mode)) |
| 2729 | return i; |
| 2730 | } |
| 2731 | |
| 2732 | return -1; |
| 2733 | } |
| 2734 | |
| 2735 | // scan for po_test operand[0] modification in range given |
| 2736 | static int scan_for_mod_opr0(struct parsed_op *po_test, |
| 2737 | int i, int opcnt) |
| 2738 | { |
| 2739 | for (; i < opcnt; i++) { |
| 2740 | if (is_opr_modified(&po_test->operand[0], &ops[i])) |
| 2741 | return i; |
| 2742 | } |
| 2743 | |
| 2744 | return -1; |
| 2745 | } |
| 2746 | |
| 2747 | static int scan_for_flag_set(int i, int magic, int *branched, |
| 2748 | int *setters, int *setter_cnt) |
| 2749 | { |
| 2750 | struct label_ref *lr; |
| 2751 | int ret; |
| 2752 | |
| 2753 | while (i >= 0) { |
| 2754 | if (ops[i].cc_scratch == magic) { |
| 2755 | // is this a problem? |
| 2756 | //ferr(&ops[i], "%s looped\n", __func__); |
| 2757 | return 0; |
| 2758 | } |
| 2759 | ops[i].cc_scratch = magic; |
| 2760 | |
| 2761 | if (g_labels[i] != NULL) { |
| 2762 | *branched = 1; |
| 2763 | |
| 2764 | lr = &g_label_refs[i]; |
| 2765 | for (; lr->next; lr = lr->next) { |
| 2766 | check_i(&ops[i], lr->i); |
| 2767 | ret = scan_for_flag_set(lr->i, magic, |
| 2768 | branched, setters, setter_cnt); |
| 2769 | if (ret < 0) |
| 2770 | return ret; |
| 2771 | } |
| 2772 | |
| 2773 | check_i(&ops[i], lr->i); |
| 2774 | if (i > 0 && LAST_OP(i - 1)) { |
| 2775 | i = lr->i; |
| 2776 | continue; |
| 2777 | } |
| 2778 | ret = scan_for_flag_set(lr->i, magic, |
| 2779 | branched, setters, setter_cnt); |
| 2780 | if (ret < 0) |
| 2781 | return ret; |
| 2782 | } |
| 2783 | i--; |
| 2784 | |
| 2785 | if (ops[i].flags & OPF_FLAGS) { |
| 2786 | setters[*setter_cnt] = i; |
| 2787 | (*setter_cnt)++; |
| 2788 | return 0; |
| 2789 | } |
| 2790 | |
| 2791 | if ((ops[i].flags & (OPF_JMP|OPF_CJMP)) == OPF_JMP) |
| 2792 | return -1; |
| 2793 | } |
| 2794 | |
| 2795 | return -1; |
| 2796 | } |
| 2797 | |
| 2798 | // scan back for cdq, if anything modifies edx, fail |
| 2799 | static int scan_for_cdq_edx(int i) |
| 2800 | { |
| 2801 | while (i >= 0) { |
| 2802 | if (g_labels[i] != NULL) { |
| 2803 | if (g_label_refs[i].next != NULL) |
| 2804 | return -1; |
| 2805 | if (i > 0 && LAST_OP(i - 1)) { |
| 2806 | i = g_label_refs[i].i; |
| 2807 | continue; |
| 2808 | } |
| 2809 | return -1; |
| 2810 | } |
| 2811 | i--; |
| 2812 | |
| 2813 | if (ops[i].op == OP_CDQ) |
| 2814 | return i; |
| 2815 | |
| 2816 | if (ops[i].regmask_dst & (1 << xDX)) |
| 2817 | return -1; |
| 2818 | } |
| 2819 | |
| 2820 | return -1; |
| 2821 | } |
| 2822 | |
| 2823 | static int scan_for_reg_clear(int i, int reg) |
| 2824 | { |
| 2825 | while (i >= 0) { |
| 2826 | if (g_labels[i] != NULL) { |
| 2827 | if (g_label_refs[i].next != NULL) |
| 2828 | return -1; |
| 2829 | if (i > 0 && LAST_OP(i - 1)) { |
| 2830 | i = g_label_refs[i].i; |
| 2831 | continue; |
| 2832 | } |
| 2833 | return -1; |
| 2834 | } |
| 2835 | i--; |
| 2836 | |
| 2837 | if (ops[i].op == OP_XOR |
| 2838 | && ops[i].operand[0].lmod == OPLM_DWORD |
| 2839 | && ops[i].operand[0].reg == ops[i].operand[1].reg |
| 2840 | && ops[i].operand[0].reg == reg) |
| 2841 | return i; |
| 2842 | |
| 2843 | if (ops[i].regmask_dst & (1 << reg)) |
| 2844 | return -1; |
| 2845 | } |
| 2846 | |
| 2847 | return -1; |
| 2848 | } |
| 2849 | |
| 2850 | static void patch_esp_adjust(struct parsed_op *po, int adj) |
| 2851 | { |
| 2852 | ferr_assert(po, po->op == OP_ADD); |
| 2853 | ferr_assert(po, IS(opr_name(po, 0), "esp")); |
| 2854 | ferr_assert(po, po->operand[1].type == OPT_CONST); |
| 2855 | |
| 2856 | // this is a bit of a hack, but deals with use of |
| 2857 | // single adj for multiple calls |
| 2858 | po->operand[1].val -= adj; |
| 2859 | po->flags |= OPF_RMD; |
| 2860 | if (po->operand[1].val == 0) |
| 2861 | po->flags |= OPF_DONE; |
| 2862 | ferr_assert(po, (int)po->operand[1].val >= 0); |
| 2863 | } |
| 2864 | |
| 2865 | // scan for positive, constant esp adjust |
| 2866 | // multipath case is preliminary |
| 2867 | static int scan_for_esp_adjust(int i, int opcnt, |
| 2868 | int adj_expect, int *adj, int *is_multipath, int do_update) |
| 2869 | { |
| 2870 | int adj_expect_unknown = 0; |
| 2871 | struct parsed_op *po; |
| 2872 | int first_pop = -1; |
| 2873 | int adj_best = 0; |
| 2874 | |
| 2875 | *adj = *is_multipath = 0; |
| 2876 | if (adj_expect < 0) { |
| 2877 | adj_expect_unknown = 1; |
| 2878 | adj_expect = 32 * 4; // enough? |
| 2879 | } |
| 2880 | |
| 2881 | for (; i < opcnt && *adj < adj_expect; i++) { |
| 2882 | if (g_labels[i] != NULL) |
| 2883 | *is_multipath = 1; |
| 2884 | |
| 2885 | po = &ops[i]; |
| 2886 | if (po->flags & OPF_DONE) |
| 2887 | continue; |
| 2888 | |
| 2889 | if (po->op == OP_ADD && po->operand[0].reg == xSP) { |
| 2890 | if (po->operand[1].type != OPT_CONST) |
| 2891 | ferr(&ops[i], "non-const esp adjust?\n"); |
| 2892 | *adj += po->operand[1].val; |
| 2893 | if (*adj & 3) |
| 2894 | ferr(&ops[i], "unaligned esp adjust: %x\n", *adj); |
| 2895 | if (do_update) { |
| 2896 | if (!*is_multipath) |
| 2897 | patch_esp_adjust(po, adj_expect); |
| 2898 | else |
| 2899 | po->flags |= OPF_RMD; |
| 2900 | } |
| 2901 | return i; |
| 2902 | } |
| 2903 | else if (po->op == OP_PUSH) { |
| 2904 | //if (first_pop == -1) |
| 2905 | // first_pop = -2; // none |
| 2906 | *adj -= lmod_bytes(po, po->operand[0].lmod); |
| 2907 | } |
| 2908 | else if (po->op == OP_POP) { |
| 2909 | if (!(po->flags & OPF_DONE)) { |
| 2910 | // seems like msvc only uses 'pop ecx' for stack realignment.. |
| 2911 | if (po->operand[0].type != OPT_REG || po->operand[0].reg != xCX) |
| 2912 | break; |
| 2913 | if (first_pop == -1 && *adj >= 0) |
| 2914 | first_pop = i; |
| 2915 | } |
| 2916 | if (do_update && *adj >= 0) { |
| 2917 | po->flags |= OPF_RMD; |
| 2918 | if (!*is_multipath) |
| 2919 | po->flags |= OPF_DONE | OPF_NOREGS; |
| 2920 | } |
| 2921 | |
| 2922 | *adj += lmod_bytes(po, po->operand[0].lmod); |
| 2923 | if (*adj > adj_best) |
| 2924 | adj_best = *adj; |
| 2925 | } |
| 2926 | else if (po->flags & (OPF_JMP|OPF_TAIL)) { |
| 2927 | if (po->op == OP_JMP && po->btj == NULL) { |
| 2928 | if (po->bt_i <= i) |
| 2929 | break; |
| 2930 | i = po->bt_i - 1; |
| 2931 | continue; |
| 2932 | } |
| 2933 | if (po->op != OP_CALL) |
| 2934 | break; |
| 2935 | if (po->operand[0].type != OPT_LABEL) |
| 2936 | break; |
| 2937 | if (po->pp != NULL && po->pp->is_stdcall) |
| 2938 | break; |
| 2939 | if (adj_expect_unknown && first_pop >= 0) |
| 2940 | break; |
| 2941 | // assume it's another cdecl call |
| 2942 | } |
| 2943 | } |
| 2944 | |
| 2945 | if (first_pop >= 0) { |
| 2946 | // probably only 'pop ecx' was used |
| 2947 | *adj = adj_best; |
| 2948 | return first_pop; |
| 2949 | } |
| 2950 | |
| 2951 | return -1; |
| 2952 | } |
| 2953 | |
| 2954 | static void scan_fwd_set_flags(int i, int opcnt, int magic, int flags) |
| 2955 | { |
| 2956 | struct parsed_op *po; |
| 2957 | int j; |
| 2958 | |
| 2959 | if (i < 0) |
| 2960 | ferr(ops, "%s: followed bad branch?\n", __func__); |
| 2961 | |
| 2962 | for (; i < opcnt; i++) { |
| 2963 | po = &ops[i]; |
| 2964 | if (po->cc_scratch == magic) |
| 2965 | return; |
| 2966 | po->cc_scratch = magic; |
| 2967 | po->flags |= flags; |
| 2968 | |
| 2969 | if ((po->flags & OPF_JMP) && po->op != OP_CALL) { |
| 2970 | if (po->btj != NULL) { |
| 2971 | // jumptable |
| 2972 | for (j = 0; j < po->btj->count; j++) |
| 2973 | scan_fwd_set_flags(po->btj->d[j].bt_i, opcnt, magic, flags); |
| 2974 | return; |
| 2975 | } |
| 2976 | |
| 2977 | scan_fwd_set_flags(po->bt_i, opcnt, magic, flags); |
| 2978 | if (!(po->flags & OPF_CJMP)) |
| 2979 | return; |
| 2980 | } |
| 2981 | if (po->flags & OPF_TAIL) |
| 2982 | return; |
| 2983 | } |
| 2984 | } |
| 2985 | |
| 2986 | static const struct parsed_proto *try_recover_pp( |
| 2987 | struct parsed_op *po, const struct parsed_opr *opr, int *search_instead) |
| 2988 | { |
| 2989 | const struct parsed_proto *pp = NULL; |
| 2990 | char buf[256]; |
| 2991 | char *p; |
| 2992 | |
| 2993 | // maybe an arg of g_func? |
| 2994 | if (opr->type == OPT_REGMEM && is_stack_access(po, opr)) |
| 2995 | { |
| 2996 | char ofs_reg[16] = { 0, }; |
| 2997 | int arg, arg_s, arg_i; |
| 2998 | int stack_ra = 0; |
| 2999 | int offset = 0; |
| 3000 | |
| 3001 | if (g_header_mode) |
| 3002 | return NULL; |
| 3003 | |
| 3004 | parse_stack_access(po, opr->name, ofs_reg, |
| 3005 | &offset, &stack_ra, NULL, 0); |
| 3006 | if (ofs_reg[0] != 0) |
| 3007 | ferr(po, "offset reg on arg access?\n"); |
| 3008 | if (offset <= stack_ra) { |
| 3009 | // search who set the stack var instead |
| 3010 | if (search_instead != NULL) |
| 3011 | *search_instead = 1; |
| 3012 | return NULL; |
| 3013 | } |
| 3014 | |
| 3015 | arg_i = (offset - stack_ra - 4) / 4; |
| 3016 | for (arg = arg_s = 0; arg < g_func_pp->argc; arg++) { |
| 3017 | if (g_func_pp->arg[arg].reg != NULL) |
| 3018 | continue; |
| 3019 | if (arg_s == arg_i) |
| 3020 | break; |
| 3021 | arg_s++; |
| 3022 | } |
| 3023 | if (arg == g_func_pp->argc) |
| 3024 | ferr(po, "stack arg %d not in prototype?\n", arg_i); |
| 3025 | |
| 3026 | pp = g_func_pp->arg[arg].fptr; |
| 3027 | if (pp == NULL) |
| 3028 | ferr(po, "icall sa: arg%d is not a fptr?\n", arg + 1); |
| 3029 | check_func_pp(po, pp, "icall arg"); |
| 3030 | } |
| 3031 | else if (opr->type == OPT_REGMEM && strchr(opr->name + 1, '[')) { |
| 3032 | // label[index] |
| 3033 | p = strchr(opr->name + 1, '['); |
| 3034 | memcpy(buf, opr->name, p - opr->name); |
| 3035 | buf[p - opr->name] = 0; |
| 3036 | pp = proto_parse(g_fhdr, buf, g_quiet_pp); |
| 3037 | } |
| 3038 | else if (opr->type == OPT_OFFSET || opr->type == OPT_LABEL) { |
| 3039 | pp = proto_parse(g_fhdr, opr->name, g_quiet_pp); |
| 3040 | if (pp == NULL) { |
| 3041 | if (!g_header_mode) |
| 3042 | ferr(po, "proto_parse failed for icall to '%s'\n", opr->name); |
| 3043 | } |
| 3044 | else |
| 3045 | check_func_pp(po, pp, "reg-fptr ref"); |
| 3046 | } |
| 3047 | |
| 3048 | return pp; |
| 3049 | } |
| 3050 | |
| 3051 | static void scan_for_call_type(int i, const struct parsed_opr *opr, |
| 3052 | int magic, const struct parsed_proto **pp_found, int *pp_i, |
| 3053 | int *multi) |
| 3054 | { |
| 3055 | const struct parsed_proto *pp = NULL; |
| 3056 | struct parsed_op *po; |
| 3057 | struct label_ref *lr; |
| 3058 | |
| 3059 | ops[i].cc_scratch = magic; |
| 3060 | |
| 3061 | while (1) { |
| 3062 | if (g_labels[i] != NULL) { |
| 3063 | lr = &g_label_refs[i]; |
| 3064 | for (; lr != NULL; lr = lr->next) { |
| 3065 | check_i(&ops[i], lr->i); |
| 3066 | scan_for_call_type(lr->i, opr, magic, pp_found, pp_i, multi); |
| 3067 | } |
| 3068 | if (i > 0 && LAST_OP(i - 1)) |
| 3069 | return; |
| 3070 | } |
| 3071 | |
| 3072 | i--; |
| 3073 | if (i < 0) |
| 3074 | break; |
| 3075 | |
| 3076 | if (ops[i].cc_scratch == magic) |
| 3077 | return; |
| 3078 | ops[i].cc_scratch = magic; |
| 3079 | |
| 3080 | if (!(ops[i].flags & OPF_DATA)) |
| 3081 | continue; |
| 3082 | if (!is_opr_modified(opr, &ops[i])) |
| 3083 | continue; |
| 3084 | if (ops[i].op != OP_MOV && ops[i].op != OP_LEA) { |
| 3085 | // most probably trashed by some processing |
| 3086 | *pp_found = NULL; |
| 3087 | return; |
| 3088 | } |
| 3089 | |
| 3090 | opr = &ops[i].operand[1]; |
| 3091 | if (opr->type != OPT_REG) |
| 3092 | break; |
| 3093 | } |
| 3094 | |
| 3095 | po = (i >= 0) ? &ops[i] : ops; |
| 3096 | |
| 3097 | if (i < 0) { |
| 3098 | // reached the top - can only be an arg-reg |
| 3099 | if (opr->type != OPT_REG || g_func_pp == NULL) |
| 3100 | return; |
| 3101 | |
| 3102 | for (i = 0; i < g_func_pp->argc; i++) { |
| 3103 | if (g_func_pp->arg[i].reg == NULL) |
| 3104 | continue; |
| 3105 | if (IS(opr->name, g_func_pp->arg[i].reg)) |
| 3106 | break; |
| 3107 | } |
| 3108 | if (i == g_func_pp->argc) |
| 3109 | return; |
| 3110 | pp = g_func_pp->arg[i].fptr; |
| 3111 | if (pp == NULL) |
| 3112 | ferr(po, "icall: arg%d (%s) is not a fptr?\n", |
| 3113 | i + 1, g_func_pp->arg[i].reg); |
| 3114 | check_func_pp(po, pp, "icall reg-arg"); |
| 3115 | } |
| 3116 | else |
| 3117 | pp = try_recover_pp(po, opr, NULL); |
| 3118 | |
| 3119 | if (*pp_found != NULL && pp != NULL && *pp_found != pp) { |
| 3120 | if (!IS((*pp_found)->ret_type.name, pp->ret_type.name) |
| 3121 | || (*pp_found)->is_stdcall != pp->is_stdcall |
| 3122 | || (*pp_found)->is_fptr != pp->is_fptr |
| 3123 | || (*pp_found)->argc != pp->argc |
| 3124 | || (*pp_found)->argc_reg != pp->argc_reg |
| 3125 | || (*pp_found)->argc_stack != pp->argc_stack) |
| 3126 | { |
| 3127 | ferr(po, "icall: parsed_proto mismatch\n"); |
| 3128 | } |
| 3129 | *multi = 1; |
| 3130 | } |
| 3131 | if (pp != NULL) { |
| 3132 | *pp_found = pp; |
| 3133 | *pp_i = po - ops; |
| 3134 | } |
| 3135 | } |
| 3136 | |
| 3137 | static void add_label_ref(struct label_ref *lr, int op_i) |
| 3138 | { |
| 3139 | struct label_ref *lr_new; |
| 3140 | |
| 3141 | if (lr->i == -1) { |
| 3142 | lr->i = op_i; |
| 3143 | return; |
| 3144 | } |
| 3145 | |
| 3146 | lr_new = calloc(1, sizeof(*lr_new)); |
| 3147 | lr_new->i = op_i; |
| 3148 | lr_new->next = lr->next; |
| 3149 | lr->next = lr_new; |
| 3150 | } |
| 3151 | |
| 3152 | static struct parsed_data *try_resolve_jumptab(int i, int opcnt) |
| 3153 | { |
| 3154 | struct parsed_op *po = &ops[i]; |
| 3155 | struct parsed_data *pd; |
| 3156 | char label[NAMELEN], *p; |
| 3157 | int len, j, l; |
| 3158 | |
| 3159 | p = strchr(po->operand[0].name, '['); |
| 3160 | if (p == NULL) |
| 3161 | return NULL; |
| 3162 | |
| 3163 | len = p - po->operand[0].name; |
| 3164 | strncpy(label, po->operand[0].name, len); |
| 3165 | label[len] = 0; |
| 3166 | |
| 3167 | for (j = 0, pd = NULL; j < g_func_pd_cnt; j++) { |
| 3168 | if (IS(g_func_pd[j].label, label)) { |
| 3169 | pd = &g_func_pd[j]; |
| 3170 | break; |
| 3171 | } |
| 3172 | } |
| 3173 | if (pd == NULL) |
| 3174 | //ferr(po, "label '%s' not parsed?\n", label); |
| 3175 | return NULL; |
| 3176 | |
| 3177 | if (pd->type != OPT_OFFSET) |
| 3178 | ferr(po, "label '%s' with non-offset data?\n", label); |
| 3179 | |
| 3180 | // find all labels, link |
| 3181 | for (j = 0; j < pd->count; j++) { |
| 3182 | for (l = 0; l < opcnt; l++) { |
| 3183 | if (g_labels[l] != NULL && IS(g_labels[l], pd->d[j].u.label)) { |
| 3184 | add_label_ref(&g_label_refs[l], i); |
| 3185 | pd->d[j].bt_i = l; |
| 3186 | break; |
| 3187 | } |
| 3188 | } |
| 3189 | } |
| 3190 | |
| 3191 | return pd; |
| 3192 | } |
| 3193 | |
| 3194 | static void clear_labels(int count) |
| 3195 | { |
| 3196 | int i; |
| 3197 | |
| 3198 | for (i = 0; i < count; i++) { |
| 3199 | if (g_labels[i] != NULL) { |
| 3200 | free(g_labels[i]); |
| 3201 | g_labels[i] = NULL; |
| 3202 | } |
| 3203 | } |
| 3204 | } |
| 3205 | |
| 3206 | static int get_pp_arg_regmask_src(const struct parsed_proto *pp) |
| 3207 | { |
| 3208 | int regmask = 0; |
| 3209 | int i, reg; |
| 3210 | |
| 3211 | for (i = 0; i < pp->argc; i++) { |
| 3212 | if (pp->arg[i].reg != NULL) { |
| 3213 | reg = char_array_i(regs_r32, |
| 3214 | ARRAY_SIZE(regs_r32), pp->arg[i].reg); |
| 3215 | if (reg < 0) |
| 3216 | ferr(ops, "arg '%s' of func '%s' is not a reg?\n", |
| 3217 | pp->arg[i].reg, pp->name); |
| 3218 | regmask |= 1 << reg; |
| 3219 | } |
| 3220 | } |
| 3221 | |
| 3222 | return regmask; |
| 3223 | } |
| 3224 | |
| 3225 | static int get_pp_arg_regmask_dst(const struct parsed_proto *pp) |
| 3226 | { |
| 3227 | if (strstr(pp->ret_type.name, "int64")) |
| 3228 | return (1 << xAX) | (1 << xDX); |
| 3229 | if (strcasecmp(pp->ret_type.name, "void") == 0) |
| 3230 | return 0; |
| 3231 | |
| 3232 | return (1 << xAX); |
| 3233 | } |
| 3234 | |
| 3235 | static void resolve_branches_parse_calls(int opcnt) |
| 3236 | { |
| 3237 | const struct parsed_proto *pp_c; |
| 3238 | struct parsed_proto *pp; |
| 3239 | struct parsed_data *pd; |
| 3240 | struct parsed_op *po; |
| 3241 | const char *tmpname; |
| 3242 | int i, l; |
| 3243 | int ret; |
| 3244 | |
| 3245 | for (i = 0; i < opcnt; i++) |
| 3246 | { |
| 3247 | po = &ops[i]; |
| 3248 | po->bt_i = -1; |
| 3249 | po->btj = NULL; |
| 3250 | |
| 3251 | if (po->op == OP_CALL) { |
| 3252 | pp = NULL; |
| 3253 | |
| 3254 | if (po->operand[0].type == OPT_LABEL) { |
| 3255 | tmpname = opr_name(po, 0); |
| 3256 | if (IS_START(tmpname, "loc_")) |
| 3257 | ferr(po, "call to loc_*\n"); |
| 3258 | pp_c = proto_parse(g_fhdr, tmpname, g_header_mode); |
| 3259 | if (!g_header_mode && pp_c == NULL) |
| 3260 | ferr(po, "proto_parse failed for call '%s'\n", tmpname); |
| 3261 | |
| 3262 | if (pp_c != NULL) { |
| 3263 | pp = proto_clone(pp_c); |
| 3264 | my_assert_not(pp, NULL); |
| 3265 | } |
| 3266 | } |
| 3267 | else if (po->datap != NULL) { |
| 3268 | pp = calloc(1, sizeof(*pp)); |
| 3269 | my_assert_not(pp, NULL); |
| 3270 | |
| 3271 | ret = parse_protostr(po->datap, pp); |
| 3272 | if (ret < 0) |
| 3273 | ferr(po, "bad protostr supplied: %s\n", (char *)po->datap); |
| 3274 | free(po->datap); |
| 3275 | po->datap = NULL; |
| 3276 | } |
| 3277 | |
| 3278 | if (pp != NULL) { |
| 3279 | if (pp->is_fptr) |
| 3280 | check_func_pp(po, pp, "fptr var call"); |
| 3281 | if (pp->is_noreturn) |
| 3282 | po->flags |= OPF_TAIL; |
| 3283 | } |
| 3284 | po->pp = pp; |
| 3285 | continue; |
| 3286 | } |
| 3287 | |
| 3288 | if (!(po->flags & OPF_JMP) || po->op == OP_RET) |
| 3289 | continue; |
| 3290 | |
| 3291 | if (po->operand[0].type == OPT_REGMEM) { |
| 3292 | pd = try_resolve_jumptab(i, opcnt); |
| 3293 | if (pd == NULL) |
| 3294 | goto tailcall; |
| 3295 | |
| 3296 | po->btj = pd; |
| 3297 | continue; |
| 3298 | } |
| 3299 | |
| 3300 | for (l = 0; l < opcnt; l++) { |
| 3301 | if (g_labels[l] != NULL |
| 3302 | && IS(po->operand[0].name, g_labels[l])) |
| 3303 | { |
| 3304 | if (l == i + 1 && po->op == OP_JMP) { |
| 3305 | // yet another alignment type.. |
| 3306 | po->flags |= OPF_RMD|OPF_DONE; |
| 3307 | break; |
| 3308 | } |
| 3309 | add_label_ref(&g_label_refs[l], i); |
| 3310 | po->bt_i = l; |
| 3311 | break; |
| 3312 | } |
| 3313 | } |
| 3314 | |
| 3315 | if (po->bt_i != -1 || (po->flags & OPF_RMD)) |
| 3316 | continue; |
| 3317 | |
| 3318 | if (po->operand[0].type == OPT_LABEL) |
| 3319 | // assume tail call |
| 3320 | goto tailcall; |
| 3321 | |
| 3322 | ferr(po, "unhandled branch\n"); |
| 3323 | |
| 3324 | tailcall: |
| 3325 | po->op = OP_CALL; |
| 3326 | po->flags |= OPF_TAIL; |
| 3327 | if (i > 0 && ops[i - 1].op == OP_POP) |
| 3328 | po->flags |= OPF_ATAIL; |
| 3329 | i--; // reprocess |
| 3330 | } |
| 3331 | } |
| 3332 | |
| 3333 | static void scan_prologue_epilogue(int opcnt) |
| 3334 | { |
| 3335 | int ecx_push = 0, esp_sub = 0; |
| 3336 | int found; |
| 3337 | int i, j, l; |
| 3338 | |
| 3339 | if (ops[0].op == OP_PUSH && IS(opr_name(&ops[0], 0), "ebp") |
| 3340 | && ops[1].op == OP_MOV |
| 3341 | && IS(opr_name(&ops[1], 0), "ebp") |
| 3342 | && IS(opr_name(&ops[1], 1), "esp")) |
| 3343 | { |
| 3344 | g_bp_frame = 1; |
| 3345 | ops[0].flags |= OPF_RMD | OPF_DONE | OPF_NOREGS; |
| 3346 | ops[1].flags |= OPF_RMD | OPF_DONE | OPF_NOREGS; |
| 3347 | i = 2; |
| 3348 | |
| 3349 | if (ops[2].op == OP_SUB && IS(opr_name(&ops[2], 0), "esp")) { |
| 3350 | g_stack_fsz = opr_const(&ops[2], 1); |
| 3351 | ops[2].flags |= OPF_RMD | OPF_DONE | OPF_NOREGS; |
| 3352 | i++; |
| 3353 | } |
| 3354 | else { |
| 3355 | // another way msvc builds stack frame.. |
| 3356 | i = 2; |
| 3357 | while (ops[i].op == OP_PUSH && IS(opr_name(&ops[i], 0), "ecx")) { |
| 3358 | g_stack_fsz += 4; |
| 3359 | ops[i].flags |= OPF_RMD | OPF_DONE | OPF_NOREGS; |
| 3360 | ecx_push++; |
| 3361 | i++; |
| 3362 | } |
| 3363 | // and another way.. |
| 3364 | if (i == 2 && ops[i].op == OP_MOV && ops[i].operand[0].reg == xAX |
| 3365 | && ops[i].operand[1].type == OPT_CONST |
| 3366 | && ops[i + 1].op == OP_CALL |
| 3367 | && IS(opr_name(&ops[i + 1], 0), "__alloca_probe")) |
| 3368 | { |
| 3369 | g_stack_fsz += ops[i].operand[1].val; |
| 3370 | ops[i].flags |= OPF_RMD | OPF_DONE | OPF_NOREGS; |
| 3371 | i++; |
| 3372 | ops[i].flags |= OPF_RMD | OPF_DONE | OPF_NOREGS; |
| 3373 | i++; |
| 3374 | } |
| 3375 | } |
| 3376 | |
| 3377 | found = 0; |
| 3378 | do { |
| 3379 | for (; i < opcnt; i++) |
| 3380 | if (ops[i].flags & OPF_TAIL) |
| 3381 | break; |
| 3382 | j = i - 1; |
| 3383 | if (i == opcnt && (ops[j].flags & OPF_JMP)) { |
| 3384 | if (ops[j].bt_i != -1 || ops[j].btj != NULL) |
| 3385 | break; |
| 3386 | i--; |
| 3387 | j--; |
| 3388 | } |
| 3389 | |
| 3390 | if ((ops[j].op == OP_POP && IS(opr_name(&ops[j], 0), "ebp")) |
| 3391 | || ops[j].op == OP_LEAVE) |
| 3392 | { |
| 3393 | ops[j].flags |= OPF_RMD | OPF_DONE | OPF_NOREGS; |
| 3394 | } |
| 3395 | else if (ops[i].op == OP_CALL && ops[i].pp != NULL |
| 3396 | && ops[i].pp->is_noreturn) |
| 3397 | { |
| 3398 | // on noreturn, msvc sometimes cleans stack, sometimes not |
| 3399 | i++; |
| 3400 | found = 1; |
| 3401 | continue; |
| 3402 | } |
| 3403 | else if (!(g_ida_func_attr & IDAFA_NORETURN)) |
| 3404 | ferr(&ops[j], "'pop ebp' expected\n"); |
| 3405 | |
| 3406 | if (g_stack_fsz != 0) { |
| 3407 | if (ops[j].op == OP_LEAVE) |
| 3408 | j--; |
| 3409 | else if (ops[j].op == OP_POP |
| 3410 | && ops[j - 1].op == OP_MOV |
| 3411 | && IS(opr_name(&ops[j - 1], 0), "esp") |
| 3412 | && IS(opr_name(&ops[j - 1], 1), "ebp")) |
| 3413 | { |
| 3414 | ops[j - 1].flags |= OPF_RMD | OPF_DONE | OPF_NOREGS; |
| 3415 | j -= 2; |
| 3416 | } |
| 3417 | else if (!(g_ida_func_attr & IDAFA_NORETURN)) |
| 3418 | { |
| 3419 | ferr(&ops[j], "esp restore expected\n"); |
| 3420 | } |
| 3421 | |
| 3422 | if (ecx_push && j >= 0 && ops[j].op == OP_POP |
| 3423 | && IS(opr_name(&ops[j], 0), "ecx")) |
| 3424 | { |
| 3425 | ferr(&ops[j], "unexpected ecx pop\n"); |
| 3426 | } |
| 3427 | } |
| 3428 | |
| 3429 | found = 1; |
| 3430 | i++; |
| 3431 | } while (i < opcnt); |
| 3432 | |
| 3433 | if (!found) |
| 3434 | ferr(ops, "missing ebp epilogue\n"); |
| 3435 | return; |
| 3436 | } |
| 3437 | |
| 3438 | // non-bp frame |
| 3439 | i = 0; |
| 3440 | while (ops[i].op == OP_PUSH && IS(opr_name(&ops[i], 0), "ecx")) { |
| 3441 | ops[i].flags |= OPF_RMD | OPF_DONE | OPF_NOREGS; |
| 3442 | g_stack_fsz += 4; |
| 3443 | ecx_push++; |
| 3444 | i++; |
| 3445 | } |
| 3446 | |
| 3447 | for (; i < opcnt; i++) { |
| 3448 | if (ops[i].op == OP_PUSH || (ops[i].flags & (OPF_JMP|OPF_TAIL))) |
| 3449 | break; |
| 3450 | if (ops[i].op == OP_SUB && ops[i].operand[0].reg == xSP |
| 3451 | && ops[i].operand[1].type == OPT_CONST) |
| 3452 | { |
| 3453 | g_stack_fsz = ops[i].operand[1].val; |
| 3454 | ops[i].flags |= OPF_RMD | OPF_DONE | OPF_NOREGS; |
| 3455 | esp_sub = 1; |
| 3456 | break; |
| 3457 | } |
| 3458 | } |
| 3459 | |
| 3460 | if (ecx_push && !esp_sub) { |
| 3461 | // could actually be args for a call.. |
| 3462 | for (; i < opcnt; i++) |
| 3463 | if (ops[i].op != OP_PUSH) |
| 3464 | break; |
| 3465 | |
| 3466 | if (ops[i].op == OP_CALL && ops[i].operand[0].type == OPT_LABEL) { |
| 3467 | const struct parsed_proto *pp; |
| 3468 | pp = proto_parse(g_fhdr, opr_name(&ops[i], 0), 1); |
| 3469 | j = pp ? pp->argc_stack : 0; |
| 3470 | while (i > 0 && j > 0) { |
| 3471 | i--; |
| 3472 | if (ops[i].op == OP_PUSH) { |
| 3473 | ops[i].flags &= ~(OPF_RMD | OPF_DONE | OPF_NOREGS); |
| 3474 | j--; |
| 3475 | } |
| 3476 | } |
| 3477 | if (j != 0) |
| 3478 | ferr(&ops[i], "unhandled prologue\n"); |
| 3479 | |
| 3480 | // recheck |
| 3481 | i = g_stack_fsz = ecx_push = 0; |
| 3482 | while (ops[i].op == OP_PUSH && IS(opr_name(&ops[i], 0), "ecx")) { |
| 3483 | if (!(ops[i].flags & OPF_RMD)) |
| 3484 | break; |
| 3485 | g_stack_fsz += 4; |
| 3486 | ecx_push++; |
| 3487 | i++; |
| 3488 | } |
| 3489 | } |
| 3490 | } |
| 3491 | |
| 3492 | found = 0; |
| 3493 | if (ecx_push || esp_sub) |
| 3494 | { |
| 3495 | g_sp_frame = 1; |
| 3496 | |
| 3497 | i++; |
| 3498 | do { |
| 3499 | for (; i < opcnt; i++) |
| 3500 | if (ops[i].flags & OPF_TAIL) |
| 3501 | break; |
| 3502 | j = i - 1; |
| 3503 | if (i == opcnt && (ops[j].flags & OPF_JMP)) { |
| 3504 | if (ops[j].bt_i != -1 || ops[j].btj != NULL) |
| 3505 | break; |
| 3506 | i--; |
| 3507 | j--; |
| 3508 | } |
| 3509 | |
| 3510 | if (ecx_push > 0) { |
| 3511 | for (l = 0; l < ecx_push; l++) { |
| 3512 | if (ops[j].op == OP_POP && IS(opr_name(&ops[j], 0), "ecx")) |
| 3513 | /* pop ecx */; |
| 3514 | else if (ops[j].op == OP_ADD |
| 3515 | && IS(opr_name(&ops[j], 0), "esp") |
| 3516 | && ops[j].operand[1].type == OPT_CONST) |
| 3517 | { |
| 3518 | /* add esp, N */ |
| 3519 | l += ops[j].operand[1].val / 4 - 1; |
| 3520 | } |
| 3521 | else |
| 3522 | ferr(&ops[j], "'pop ecx' expected\n"); |
| 3523 | |
| 3524 | ops[j].flags |= OPF_RMD | OPF_DONE | OPF_NOREGS; |
| 3525 | j--; |
| 3526 | } |
| 3527 | if (l != ecx_push) |
| 3528 | ferr(&ops[j], "epilogue scan failed\n"); |
| 3529 | |
| 3530 | found = 1; |
| 3531 | } |
| 3532 | |
| 3533 | if (esp_sub) { |
| 3534 | if (ops[j].op != OP_ADD |
| 3535 | || !IS(opr_name(&ops[j], 0), "esp") |
| 3536 | || ops[j].operand[1].type != OPT_CONST |
| 3537 | || ops[j].operand[1].val != g_stack_fsz) |
| 3538 | ferr(&ops[j], "'add esp' expected\n"); |
| 3539 | |
| 3540 | ops[j].flags |= OPF_RMD | OPF_DONE | OPF_NOREGS; |
| 3541 | ops[j].operand[1].val = 0; // hack for stack arg scanner |
| 3542 | found = 1; |
| 3543 | } |
| 3544 | |
| 3545 | i++; |
| 3546 | } while (i < opcnt); |
| 3547 | |
| 3548 | if (!found) |
| 3549 | ferr(ops, "missing esp epilogue\n"); |
| 3550 | } |
| 3551 | } |
| 3552 | |
| 3553 | static const struct parsed_proto *resolve_icall(int i, int opcnt, |
| 3554 | int *pp_i, int *multi_src) |
| 3555 | { |
| 3556 | const struct parsed_proto *pp = NULL; |
| 3557 | int search_advice = 0; |
| 3558 | |
| 3559 | *multi_src = 0; |
| 3560 | *pp_i = -1; |
| 3561 | |
| 3562 | switch (ops[i].operand[0].type) { |
| 3563 | case OPT_REGMEM: |
| 3564 | case OPT_LABEL: |
| 3565 | case OPT_OFFSET: |
| 3566 | pp = try_recover_pp(&ops[i], &ops[i].operand[0], &search_advice); |
| 3567 | if (!search_advice) |
| 3568 | break; |
| 3569 | // fallthrough |
| 3570 | default: |
| 3571 | scan_for_call_type(i, &ops[i].operand[0], i + opcnt * 9, &pp, |
| 3572 | pp_i, multi_src); |
| 3573 | break; |
| 3574 | } |
| 3575 | |
| 3576 | return pp; |
| 3577 | } |
| 3578 | |
| 3579 | // find an instruction that changed opr before i op |
| 3580 | // *op_i must be set to -1 by the caller |
| 3581 | // *entry is set to 1 if one source is determined to be the caller |
| 3582 | // returns 1 if found, *op_i is then set to origin |
| 3583 | static int resolve_origin(int i, const struct parsed_opr *opr, |
| 3584 | int magic, int *op_i, int *is_caller) |
| 3585 | { |
| 3586 | struct label_ref *lr; |
| 3587 | int ret = 0; |
| 3588 | |
| 3589 | if (ops[i].cc_scratch == magic) |
| 3590 | return 0; |
| 3591 | ops[i].cc_scratch = magic; |
| 3592 | |
| 3593 | while (1) { |
| 3594 | if (g_labels[i] != NULL) { |
| 3595 | lr = &g_label_refs[i]; |
| 3596 | for (; lr != NULL; lr = lr->next) { |
| 3597 | check_i(&ops[i], lr->i); |
| 3598 | ret |= resolve_origin(lr->i, opr, magic, op_i, is_caller); |
| 3599 | } |
| 3600 | if (i > 0 && LAST_OP(i - 1)) |
| 3601 | return ret; |
| 3602 | } |
| 3603 | |
| 3604 | i--; |
| 3605 | if (i < 0) { |
| 3606 | if (is_caller != NULL) |
| 3607 | *is_caller = 1; |
| 3608 | return -1; |
| 3609 | } |
| 3610 | |
| 3611 | if (ops[i].cc_scratch == magic) |
| 3612 | return ret; |
| 3613 | ops[i].cc_scratch = magic; |
| 3614 | |
| 3615 | if (!(ops[i].flags & OPF_DATA)) |
| 3616 | continue; |
| 3617 | if (!is_opr_modified(opr, &ops[i])) |
| 3618 | continue; |
| 3619 | |
| 3620 | if (*op_i >= 0) { |
| 3621 | if (*op_i == i) |
| 3622 | return ret | 1; |
| 3623 | |
| 3624 | // XXX: could check if the other op does the same |
| 3625 | return -1; |
| 3626 | } |
| 3627 | |
| 3628 | *op_i = i; |
| 3629 | return ret | 1; |
| 3630 | } |
| 3631 | } |
| 3632 | |
| 3633 | // find an instruction that previously referenced opr |
| 3634 | // if multiple results are found - fail |
| 3635 | // *op_i must be set to -1 by the caller |
| 3636 | // returns 1 if found, *op_i is then set to referencer insn |
| 3637 | static int resolve_last_ref(int i, const struct parsed_opr *opr, |
| 3638 | int magic, int *op_i) |
| 3639 | { |
| 3640 | struct label_ref *lr; |
| 3641 | int ret = 0; |
| 3642 | |
| 3643 | if (ops[i].cc_scratch == magic) |
| 3644 | return 0; |
| 3645 | ops[i].cc_scratch = magic; |
| 3646 | |
| 3647 | while (1) { |
| 3648 | if (g_labels[i] != NULL) { |
| 3649 | lr = &g_label_refs[i]; |
| 3650 | for (; lr != NULL; lr = lr->next) { |
| 3651 | check_i(&ops[i], lr->i); |
| 3652 | ret |= resolve_last_ref(lr->i, opr, magic, op_i); |
| 3653 | } |
| 3654 | if (i > 0 && LAST_OP(i - 1)) |
| 3655 | return ret; |
| 3656 | } |
| 3657 | |
| 3658 | i--; |
| 3659 | if (i < 0) |
| 3660 | return -1; |
| 3661 | |
| 3662 | if (ops[i].cc_scratch == magic) |
| 3663 | return 0; |
| 3664 | ops[i].cc_scratch = magic; |
| 3665 | |
| 3666 | if (!is_opr_referenced(opr, &ops[i])) |
| 3667 | continue; |
| 3668 | |
| 3669 | if (*op_i >= 0) |
| 3670 | return -1; |
| 3671 | |
| 3672 | *op_i = i; |
| 3673 | return 1; |
| 3674 | } |
| 3675 | } |
| 3676 | |
| 3677 | // find next instruction that reads opr |
| 3678 | // *op_i must be set to -1 by the caller |
| 3679 | // on return, *op_i is set to first referencer insn |
| 3680 | // returns 1 if exactly 1 referencer is found |
| 3681 | static int find_next_read(int i, int opcnt, |
| 3682 | const struct parsed_opr *opr, int magic, int *op_i) |
| 3683 | { |
| 3684 | struct parsed_op *po; |
| 3685 | int j, ret = 0; |
| 3686 | |
| 3687 | for (; i < opcnt; i++) |
| 3688 | { |
| 3689 | if (ops[i].cc_scratch == magic) |
| 3690 | return ret; |
| 3691 | ops[i].cc_scratch = magic; |
| 3692 | |
| 3693 | po = &ops[i]; |
| 3694 | if ((po->flags & OPF_JMP) && po->op != OP_CALL) { |
| 3695 | if (po->btj != NULL) { |
| 3696 | // jumptable |
| 3697 | for (j = 0; j < po->btj->count; j++) { |
| 3698 | check_i(po, po->btj->d[j].bt_i); |
| 3699 | ret |= find_next_read(po->btj->d[j].bt_i, opcnt, opr, |
| 3700 | magic, op_i); |
| 3701 | } |
| 3702 | return ret; |
| 3703 | } |
| 3704 | |
| 3705 | if (po->flags & OPF_RMD) |
| 3706 | continue; |
| 3707 | check_i(po, po->bt_i); |
| 3708 | if (po->flags & OPF_CJMP) { |
| 3709 | ret |= find_next_read(po->bt_i, opcnt, opr, magic, op_i); |
| 3710 | if (ret < 0) |
| 3711 | return ret; |
| 3712 | } |
| 3713 | else |
| 3714 | i = po->bt_i - 1; |
| 3715 | continue; |
| 3716 | } |
| 3717 | |
| 3718 | if (!is_opr_read(opr, po)) { |
| 3719 | if (is_opr_modified(opr, po) |
| 3720 | && (po->op == OP_CALL |
| 3721 | || ((po->flags & OPF_DATA) |
| 3722 | && po->operand[0].lmod == OPLM_DWORD))) |
| 3723 | { |
| 3724 | // it's overwritten |
| 3725 | return ret; |
| 3726 | } |
| 3727 | if (po->flags & OPF_TAIL) |
| 3728 | return ret; |
| 3729 | continue; |
| 3730 | } |
| 3731 | |
| 3732 | if (*op_i >= 0) |
| 3733 | return -1; |
| 3734 | |
| 3735 | *op_i = i; |
| 3736 | return 1; |
| 3737 | } |
| 3738 | |
| 3739 | return 0; |
| 3740 | } |
| 3741 | |
| 3742 | static int try_resolve_const(int i, const struct parsed_opr *opr, |
| 3743 | int magic, unsigned int *val) |
| 3744 | { |
| 3745 | int s_i = -1; |
| 3746 | int ret; |
| 3747 | |
| 3748 | ret = resolve_origin(i, opr, magic, &s_i, NULL); |
| 3749 | if (ret == 1) { |
| 3750 | i = s_i; |
| 3751 | if (ops[i].op != OP_MOV && ops[i].operand[1].type != OPT_CONST) |
| 3752 | return -1; |
| 3753 | |
| 3754 | *val = ops[i].operand[1].val; |
| 3755 | return 1; |
| 3756 | } |
| 3757 | |
| 3758 | return -1; |
| 3759 | } |
| 3760 | |
| 3761 | static struct parsed_proto *process_call_early(int i, int opcnt, |
| 3762 | int *adj_i) |
| 3763 | { |
| 3764 | struct parsed_op *po = &ops[i]; |
| 3765 | struct parsed_proto *pp; |
| 3766 | int multipath = 0; |
| 3767 | int adj = 0; |
| 3768 | int j, ret; |
| 3769 | |
| 3770 | pp = po->pp; |
| 3771 | if (pp == NULL || pp->is_vararg || pp->argc_reg != 0) |
| 3772 | // leave for later |
| 3773 | return NULL; |
| 3774 | |
| 3775 | // look for and make use of esp adjust |
| 3776 | *adj_i = ret = -1; |
| 3777 | if (!pp->is_stdcall && pp->argc_stack > 0) |
| 3778 | ret = scan_for_esp_adjust(i + 1, opcnt, |
| 3779 | pp->argc_stack * 4, &adj, &multipath, 0); |
| 3780 | if (ret >= 0) { |
| 3781 | if (pp->argc_stack > adj / 4) |
| 3782 | return NULL; |
| 3783 | if (multipath) |
| 3784 | return NULL; |
| 3785 | if (ops[ret].op == OP_POP) { |
| 3786 | for (j = 1; j < adj / 4; j++) { |
| 3787 | if (ops[ret + j].op != OP_POP |
| 3788 | || ops[ret + j].operand[0].reg != xCX) |
| 3789 | { |
| 3790 | return NULL; |
| 3791 | } |
| 3792 | } |
| 3793 | } |
| 3794 | } |
| 3795 | |
| 3796 | *adj_i = ret; |
| 3797 | return pp; |
| 3798 | } |
| 3799 | |
| 3800 | static struct parsed_proto *process_call(int i, int opcnt) |
| 3801 | { |
| 3802 | struct parsed_op *po = &ops[i]; |
| 3803 | const struct parsed_proto *pp_c; |
| 3804 | struct parsed_proto *pp; |
| 3805 | const char *tmpname; |
| 3806 | int call_i = -1, ref_i = -1; |
| 3807 | int adj = 0, multipath = 0; |
| 3808 | int ret, arg; |
| 3809 | |
| 3810 | tmpname = opr_name(po, 0); |
| 3811 | pp = po->pp; |
| 3812 | if (pp == NULL) |
| 3813 | { |
| 3814 | // indirect call |
| 3815 | pp_c = resolve_icall(i, opcnt, &call_i, &multipath); |
| 3816 | if (pp_c != NULL) { |
| 3817 | if (!pp_c->is_func && !pp_c->is_fptr) |
| 3818 | ferr(po, "call to non-func: %s\n", pp_c->name); |
| 3819 | pp = proto_clone(pp_c); |
| 3820 | my_assert_not(pp, NULL); |
| 3821 | if (multipath) |
| 3822 | // not resolved just to single func |
| 3823 | pp->is_fptr = 1; |
| 3824 | |
| 3825 | switch (po->operand[0].type) { |
| 3826 | case OPT_REG: |
| 3827 | // we resolved this call and no longer need the register |
| 3828 | po->regmask_src &= ~(1 << po->operand[0].reg); |
| 3829 | |
| 3830 | if (!multipath && i != call_i && ops[call_i].op == OP_MOV |
| 3831 | && ops[call_i].operand[1].type == OPT_LABEL) |
| 3832 | { |
| 3833 | // no other source users? |
| 3834 | ret = resolve_last_ref(i, &po->operand[0], i + opcnt * 10, |
| 3835 | &ref_i); |
| 3836 | if (ret == 1 && call_i == ref_i) { |
| 3837 | // and nothing uses it after us? |
| 3838 | ref_i = -1; |
| 3839 | find_next_read(i + 1, opcnt, &po->operand[0], |
| 3840 | i + opcnt * 11, &ref_i); |
| 3841 | if (ref_i == -1) |
| 3842 | // then also don't need the source mov |
| 3843 | ops[call_i].flags |= OPF_RMD | OPF_NOREGS; |
| 3844 | } |
| 3845 | } |
| 3846 | break; |
| 3847 | case OPT_REGMEM: |
| 3848 | pp->is_fptr = 1; |
| 3849 | break; |
| 3850 | default: |
| 3851 | break; |
| 3852 | } |
| 3853 | } |
| 3854 | if (pp == NULL) { |
| 3855 | pp = calloc(1, sizeof(*pp)); |
| 3856 | my_assert_not(pp, NULL); |
| 3857 | |
| 3858 | pp->is_fptr = 1; |
| 3859 | ret = scan_for_esp_adjust(i + 1, opcnt, |
| 3860 | -1, &adj, &multipath, 0); |
| 3861 | if (ret < 0 || adj < 0) { |
| 3862 | if (!g_allow_regfunc) |
| 3863 | ferr(po, "non-__cdecl indirect call unhandled yet\n"); |
| 3864 | pp->is_unresolved = 1; |
| 3865 | adj = 0; |
| 3866 | } |
| 3867 | adj /= 4; |
| 3868 | if (adj > ARRAY_SIZE(pp->arg)) |
| 3869 | ferr(po, "esp adjust too large: %d\n", adj); |
| 3870 | pp->ret_type.name = strdup("int"); |
| 3871 | pp->argc = pp->argc_stack = adj; |
| 3872 | for (arg = 0; arg < pp->argc; arg++) |
| 3873 | pp->arg[arg].type.name = strdup("int"); |
| 3874 | } |
| 3875 | po->pp = pp; |
| 3876 | } |
| 3877 | |
| 3878 | // look for and make use of esp adjust |
| 3879 | multipath = 0; |
| 3880 | ret = -1; |
| 3881 | if (!pp->is_stdcall && pp->argc_stack > 0) { |
| 3882 | int adj_expect = pp->is_vararg ? -1 : pp->argc_stack * 4; |
| 3883 | ret = scan_for_esp_adjust(i + 1, opcnt, |
| 3884 | adj_expect, &adj, &multipath, 0); |
| 3885 | } |
| 3886 | if (ret >= 0) { |
| 3887 | if (pp->is_vararg) { |
| 3888 | if (adj / 4 < pp->argc_stack) { |
| 3889 | fnote(po, "(this call)\n"); |
| 3890 | ferr(&ops[ret], "esp adjust is too small: %x < %x\n", |
| 3891 | adj, pp->argc_stack * 4); |
| 3892 | } |
| 3893 | // modify pp to make it have varargs as normal args |
| 3894 | arg = pp->argc; |
| 3895 | pp->argc += adj / 4 - pp->argc_stack; |
| 3896 | for (; arg < pp->argc; arg++) { |
| 3897 | pp->arg[arg].type.name = strdup("int"); |
| 3898 | pp->argc_stack++; |
| 3899 | } |
| 3900 | if (pp->argc > ARRAY_SIZE(pp->arg)) |
| 3901 | ferr(po, "too many args for '%s'\n", tmpname); |
| 3902 | } |
| 3903 | if (pp->argc_stack > adj / 4) { |
| 3904 | fnote(po, "(this call)\n"); |
| 3905 | ferr(&ops[ret], "stack tracking failed for '%s': %x %x\n", |
| 3906 | tmpname, pp->argc_stack * 4, adj); |
| 3907 | } |
| 3908 | |
| 3909 | scan_for_esp_adjust(i + 1, opcnt, |
| 3910 | pp->argc_stack * 4, &adj, &multipath, 1); |
| 3911 | } |
| 3912 | else if (pp->is_vararg) |
| 3913 | ferr(po, "missing esp_adjust for vararg func '%s'\n", |
| 3914 | pp->name); |
| 3915 | |
| 3916 | return pp; |
| 3917 | } |
| 3918 | |
| 3919 | static int collect_call_args_early(struct parsed_op *po, int i, |
| 3920 | struct parsed_proto *pp, int *regmask) |
| 3921 | { |
| 3922 | int arg, ret; |
| 3923 | int j; |
| 3924 | |
| 3925 | for (arg = 0; arg < pp->argc; arg++) |
| 3926 | if (pp->arg[arg].reg == NULL) |
| 3927 | break; |
| 3928 | |
| 3929 | // first see if it can be easily done |
| 3930 | for (j = i; j > 0 && arg < pp->argc; ) |
| 3931 | { |
| 3932 | if (g_labels[j] != NULL) |
| 3933 | return -1; |
| 3934 | j--; |
| 3935 | |
| 3936 | if (ops[j].op == OP_CALL) |
| 3937 | return -1; |
| 3938 | else if (ops[j].op == OP_ADD && ops[j].operand[0].reg == xSP) |
| 3939 | return -1; |
| 3940 | else if (ops[j].op == OP_POP) |
| 3941 | return -1; |
| 3942 | else if (ops[j].flags & OPF_CJMP) |
| 3943 | return -1; |
| 3944 | else if (ops[j].op == OP_PUSH) { |
| 3945 | if (ops[j].flags & (OPF_FARG|OPF_FARGNR)) |
| 3946 | return -1; |
| 3947 | ret = scan_for_mod(&ops[j], j + 1, i, 1); |
| 3948 | if (ret >= 0) |
| 3949 | return -1; |
| 3950 | |
| 3951 | if (pp->arg[arg].type.is_va_list) |
| 3952 | return -1; |
| 3953 | |
| 3954 | // next arg |
| 3955 | for (arg++; arg < pp->argc; arg++) |
| 3956 | if (pp->arg[arg].reg == NULL) |
| 3957 | break; |
| 3958 | } |
| 3959 | } |
| 3960 | |
| 3961 | if (arg < pp->argc) |
| 3962 | return -1; |
| 3963 | |
| 3964 | // now do it |
| 3965 | for (arg = 0; arg < pp->argc; arg++) |
| 3966 | if (pp->arg[arg].reg == NULL) |
| 3967 | break; |
| 3968 | |
| 3969 | for (j = i; j > 0 && arg < pp->argc; ) |
| 3970 | { |
| 3971 | j--; |
| 3972 | |
| 3973 | if (ops[j].op == OP_PUSH) |
| 3974 | { |
| 3975 | ops[j].p_argnext = -1; |
| 3976 | ferr_assert(&ops[j], pp->arg[arg].datap == NULL); |
| 3977 | pp->arg[arg].datap = &ops[j]; |
| 3978 | |
| 3979 | if (ops[j].operand[0].type == OPT_REG) |
| 3980 | *regmask |= 1 << ops[j].operand[0].reg; |
| 3981 | |
| 3982 | ops[j].flags |= OPF_RMD | OPF_DONE | OPF_FARGNR | OPF_FARG; |
| 3983 | ops[j].flags &= ~OPF_RSAVE; |
| 3984 | |
| 3985 | // next arg |
| 3986 | for (arg++; arg < pp->argc; arg++) |
| 3987 | if (pp->arg[arg].reg == NULL) |
| 3988 | break; |
| 3989 | } |
| 3990 | } |
| 3991 | |
| 3992 | return 0; |
| 3993 | } |
| 3994 | |
| 3995 | static int collect_call_args_r(struct parsed_op *po, int i, |
| 3996 | struct parsed_proto *pp, int *regmask, int *save_arg_vars, |
| 3997 | int *arg_grp, int arg, int magic, int need_op_saving, int may_reuse) |
| 3998 | { |
| 3999 | struct parsed_proto *pp_tmp; |
| 4000 | struct parsed_op *po_tmp; |
| 4001 | struct label_ref *lr; |
| 4002 | int need_to_save_current; |
| 4003 | int arg_grp_current = 0; |
| 4004 | int save_args_seen = 0; |
| 4005 | int save_args; |
| 4006 | int ret = 0; |
| 4007 | int reg; |
| 4008 | char buf[32]; |
| 4009 | int j, k; |
| 4010 | |
| 4011 | if (i < 0) { |
| 4012 | ferr(po, "dead label encountered\n"); |
| 4013 | return -1; |
| 4014 | } |
| 4015 | |
| 4016 | for (; arg < pp->argc; arg++) |
| 4017 | if (pp->arg[arg].reg == NULL) |
| 4018 | break; |
| 4019 | magic = (magic & 0xffffff) | (arg << 24); |
| 4020 | |
| 4021 | for (j = i; j >= 0 && (arg < pp->argc || pp->is_unresolved); ) |
| 4022 | { |
| 4023 | if (((ops[j].cc_scratch ^ magic) & 0xffffff) == 0) { |
| 4024 | if (ops[j].cc_scratch != magic) { |
| 4025 | ferr(&ops[j], "arg collect hit same path with diff args for %s\n", |
| 4026 | pp->name); |
| 4027 | return -1; |
| 4028 | } |
| 4029 | // ok: have already been here |
| 4030 | return 0; |
| 4031 | } |
| 4032 | ops[j].cc_scratch = magic; |
| 4033 | |
| 4034 | if (g_labels[j] != NULL && g_label_refs[j].i != -1) { |
| 4035 | lr = &g_label_refs[j]; |
| 4036 | if (lr->next != NULL) |
| 4037 | need_op_saving = 1; |
| 4038 | for (; lr->next; lr = lr->next) { |
| 4039 | check_i(&ops[j], lr->i); |
| 4040 | if ((ops[lr->i].flags & (OPF_JMP|OPF_CJMP)) != OPF_JMP) |
| 4041 | may_reuse = 1; |
| 4042 | ret = collect_call_args_r(po, lr->i, pp, regmask, save_arg_vars, |
| 4043 | arg_grp, arg, magic, need_op_saving, may_reuse); |
| 4044 | if (ret < 0) |
| 4045 | return ret; |
| 4046 | } |
| 4047 | |
| 4048 | check_i(&ops[j], lr->i); |
| 4049 | if ((ops[lr->i].flags & (OPF_JMP|OPF_CJMP)) != OPF_JMP) |
| 4050 | may_reuse = 1; |
| 4051 | if (j > 0 && LAST_OP(j - 1)) { |
| 4052 | // follow last branch in reverse |
| 4053 | j = lr->i; |
| 4054 | continue; |
| 4055 | } |
| 4056 | need_op_saving = 1; |
| 4057 | ret = collect_call_args_r(po, lr->i, pp, regmask, save_arg_vars, |
| 4058 | arg_grp, arg, magic, need_op_saving, may_reuse); |
| 4059 | if (ret < 0) |
| 4060 | return ret; |
| 4061 | } |
| 4062 | j--; |
| 4063 | |
| 4064 | if (ops[j].op == OP_CALL) |
| 4065 | { |
| 4066 | if (pp->is_unresolved) |
| 4067 | break; |
| 4068 | |
| 4069 | pp_tmp = ops[j].pp; |
| 4070 | if (pp_tmp == NULL) |
| 4071 | ferr(po, "arg collect hit unparsed call '%s'\n", |
| 4072 | ops[j].operand[0].name); |
| 4073 | if (may_reuse && pp_tmp->argc_stack > 0) |
| 4074 | ferr(po, "arg collect %d/%d hit '%s' with %d stack args\n", |
| 4075 | arg, pp->argc, opr_name(&ops[j], 0), pp_tmp->argc_stack); |
| 4076 | } |
| 4077 | // esp adjust of 0 means we collected it before |
| 4078 | else if (ops[j].op == OP_ADD && ops[j].operand[0].reg == xSP |
| 4079 | && (ops[j].operand[1].type != OPT_CONST |
| 4080 | || ops[j].operand[1].val != 0)) |
| 4081 | { |
| 4082 | if (pp->is_unresolved) |
| 4083 | break; |
| 4084 | |
| 4085 | fnote(po, "(this call)\n"); |
| 4086 | ferr(&ops[j], "arg collect %d/%d hit esp adjust of %d\n", |
| 4087 | arg, pp->argc, ops[j].operand[1].val); |
| 4088 | } |
| 4089 | else if (ops[j].op == OP_POP && !(ops[j].flags & OPF_DONE)) |
| 4090 | { |
| 4091 | if (pp->is_unresolved) |
| 4092 | break; |
| 4093 | |
| 4094 | fnote(po, "(this call)\n"); |
| 4095 | ferr(&ops[j], "arg collect %d/%d hit pop\n", arg, pp->argc); |
| 4096 | } |
| 4097 | else if (ops[j].flags & OPF_CJMP) |
| 4098 | { |
| 4099 | if (pp->is_unresolved) |
| 4100 | break; |
| 4101 | |
| 4102 | may_reuse = 1; |
| 4103 | } |
| 4104 | else if (ops[j].op == OP_PUSH |
| 4105 | && !(ops[j].flags & (OPF_FARGNR|OPF_DONE))) |
| 4106 | { |
| 4107 | if (pp->is_unresolved && (ops[j].flags & OPF_RMD)) |
| 4108 | break; |
| 4109 | |
| 4110 | ops[j].p_argnext = -1; |
| 4111 | po_tmp = pp->arg[arg].datap; |
| 4112 | if (po_tmp != NULL) |
| 4113 | ops[j].p_argnext = po_tmp - ops; |
| 4114 | pp->arg[arg].datap = &ops[j]; |
| 4115 | |
| 4116 | need_to_save_current = 0; |
| 4117 | save_args = 0; |
| 4118 | reg = -1; |
| 4119 | if (ops[j].operand[0].type == OPT_REG) |
| 4120 | reg = ops[j].operand[0].reg; |
| 4121 | |
| 4122 | if (!need_op_saving) { |
| 4123 | ret = scan_for_mod(&ops[j], j + 1, i, 1); |
| 4124 | need_to_save_current = (ret >= 0); |
| 4125 | } |
| 4126 | if (need_op_saving || need_to_save_current) { |
| 4127 | // mark this push as one that needs operand saving |
| 4128 | ops[j].flags &= ~OPF_RMD; |
| 4129 | if (ops[j].p_argnum == 0) { |
| 4130 | ops[j].p_argnum = arg + 1; |
| 4131 | save_args |= 1 << arg; |
| 4132 | } |
| 4133 | else if (ops[j].p_argnum < arg + 1) { |
| 4134 | // XXX: might kill valid var.. |
| 4135 | //*save_arg_vars &= ~(1 << (ops[j].p_argnum - 1)); |
| 4136 | ops[j].p_argnum = arg + 1; |
| 4137 | save_args |= 1 << arg; |
| 4138 | } |
| 4139 | |
| 4140 | if (save_args_seen & (1 << (ops[j].p_argnum - 1))) { |
| 4141 | save_args_seen = 0; |
| 4142 | arg_grp_current++; |
| 4143 | if (arg_grp_current >= MAX_ARG_GRP) |
| 4144 | ferr(&ops[j], "out of arg groups (arg%d), f %s\n", |
| 4145 | ops[j].p_argnum, pp->name); |
| 4146 | } |
| 4147 | } |
| 4148 | else if (ops[j].p_argnum == 0) |
| 4149 | ops[j].flags |= OPF_RMD; |
| 4150 | |
| 4151 | // some PUSHes are reused by different calls on other branches, |
| 4152 | // but that can't happen if we didn't branch, so they |
| 4153 | // can be removed from future searches (handles nested calls) |
| 4154 | if (!may_reuse) |
| 4155 | ops[j].flags |= OPF_FARGNR; |
| 4156 | |
| 4157 | ops[j].flags |= OPF_FARG; |
| 4158 | ops[j].flags &= ~OPF_RSAVE; |
| 4159 | |
| 4160 | // check for __VALIST |
| 4161 | if (!pp->is_unresolved && g_func_pp != NULL |
| 4162 | && pp->arg[arg].type.is_va_list) |
| 4163 | { |
| 4164 | k = -1; |
| 4165 | ret = resolve_origin(j, &ops[j].operand[0], |
| 4166 | magic + 1, &k, NULL); |
| 4167 | if (ret == 1 && k >= 0) |
| 4168 | { |
| 4169 | if (ops[k].op == OP_LEA) { |
| 4170 | if (!g_func_pp->is_vararg) |
| 4171 | ferr(&ops[k], "lea <arg> used, but %s is not vararg?\n", |
| 4172 | g_func_pp->name); |
| 4173 | |
| 4174 | snprintf(buf, sizeof(buf), "arg_%X", |
| 4175 | g_func_pp->argc_stack * 4); |
| 4176 | if (strstr(ops[k].operand[1].name, buf) |
| 4177 | || strstr(ops[k].operand[1].name, "arglist")) |
| 4178 | { |
| 4179 | ops[k].flags |= OPF_RMD | OPF_NOREGS | OPF_DONE; |
| 4180 | ops[j].flags |= OPF_RMD | OPF_NOREGS | OPF_VAPUSH; |
| 4181 | save_args &= ~(1 << arg); |
| 4182 | reg = -1; |
| 4183 | } |
| 4184 | else |
| 4185 | ferr(&ops[k], "va_list arg detection failed\n"); |
| 4186 | } |
| 4187 | // check for va_list from g_func_pp arg too |
| 4188 | else if (ops[k].op == OP_MOV |
| 4189 | && is_stack_access(&ops[k], &ops[k].operand[1])) |
| 4190 | { |
| 4191 | ret = stack_frame_access(&ops[k], &ops[k].operand[1], |
| 4192 | buf, sizeof(buf), ops[k].operand[1].name, "", 1, 0); |
| 4193 | if (ret >= 0) { |
| 4194 | ops[k].flags |= OPF_RMD | OPF_DONE; |
| 4195 | ops[j].flags |= OPF_RMD; |
| 4196 | ops[j].p_argpass = ret + 1; |
| 4197 | save_args &= ~(1 << arg); |
| 4198 | reg = -1; |
| 4199 | } |
| 4200 | } |
| 4201 | } |
| 4202 | } |
| 4203 | |
| 4204 | *save_arg_vars |= save_args; |
| 4205 | |
| 4206 | // tracking reg usage |
| 4207 | if (reg >= 0) |
| 4208 | *regmask |= 1 << reg; |
| 4209 | |
| 4210 | arg++; |
| 4211 | if (!pp->is_unresolved) { |
| 4212 | // next arg |
| 4213 | for (; arg < pp->argc; arg++) |
| 4214 | if (pp->arg[arg].reg == NULL) |
| 4215 | break; |
| 4216 | } |
| 4217 | magic = (magic & 0xffffff) | (arg << 24); |
| 4218 | } |
| 4219 | |
| 4220 | if (ops[j].p_arggrp > arg_grp_current) { |
| 4221 | save_args_seen = 0; |
| 4222 | arg_grp_current = ops[j].p_arggrp; |
| 4223 | } |
| 4224 | if (ops[j].p_argnum > 0) |
| 4225 | save_args_seen |= 1 << (ops[j].p_argnum - 1); |
| 4226 | } |
| 4227 | |
| 4228 | if (arg < pp->argc) { |
| 4229 | ferr(po, "arg collect failed for '%s': %d/%d\n", |
| 4230 | pp->name, arg, pp->argc); |
| 4231 | return -1; |
| 4232 | } |
| 4233 | |
| 4234 | if (arg_grp_current > *arg_grp) |
| 4235 | *arg_grp = arg_grp_current; |
| 4236 | |
| 4237 | return arg; |
| 4238 | } |
| 4239 | |
| 4240 | static int collect_call_args(struct parsed_op *po, int i, |
| 4241 | struct parsed_proto *pp, int *regmask, int *save_arg_vars, |
| 4242 | int magic) |
| 4243 | { |
| 4244 | // arg group is for cases when pushes for |
| 4245 | // multiple funcs are going on |
| 4246 | struct parsed_op *po_tmp; |
| 4247 | int save_arg_vars_current = 0; |
| 4248 | int arg_grp = 0; |
| 4249 | int ret; |
| 4250 | int a; |
| 4251 | |
| 4252 | ret = collect_call_args_r(po, i, pp, regmask, |
| 4253 | &save_arg_vars_current, &arg_grp, 0, magic, 0, 0); |
| 4254 | if (ret < 0) |
| 4255 | return ret; |
| 4256 | |
| 4257 | if (arg_grp != 0) { |
| 4258 | // propagate arg_grp |
| 4259 | for (a = 0; a < pp->argc; a++) { |
| 4260 | if (pp->arg[a].reg != NULL) |
| 4261 | continue; |
| 4262 | |
| 4263 | po_tmp = pp->arg[a].datap; |
| 4264 | while (po_tmp != NULL) { |
| 4265 | po_tmp->p_arggrp = arg_grp; |
| 4266 | if (po_tmp->p_argnext > 0) |
| 4267 | po_tmp = &ops[po_tmp->p_argnext]; |
| 4268 | else |
| 4269 | po_tmp = NULL; |
| 4270 | } |
| 4271 | } |
| 4272 | } |
| 4273 | save_arg_vars[arg_grp] |= save_arg_vars_current; |
| 4274 | |
| 4275 | if (pp->is_unresolved) { |
| 4276 | pp->argc += ret; |
| 4277 | pp->argc_stack += ret; |
| 4278 | for (a = 0; a < pp->argc; a++) |
| 4279 | if (pp->arg[a].type.name == NULL) |
| 4280 | pp->arg[a].type.name = strdup("int"); |
| 4281 | } |
| 4282 | |
| 4283 | return ret; |
| 4284 | } |
| 4285 | |
| 4286 | static void reg_use_pass(int i, int opcnt, unsigned char *cbits, |
| 4287 | int regmask_now, int *regmask, |
| 4288 | int regmask_save_now, int *regmask_save, |
| 4289 | int *regmask_init, int regmask_arg) |
| 4290 | { |
| 4291 | struct parsed_op *po; |
| 4292 | int already_saved; |
| 4293 | int regmask_new; |
| 4294 | int regmask_op; |
| 4295 | int flags_set; |
| 4296 | int ret, reg; |
| 4297 | int j; |
| 4298 | |
| 4299 | for (; i < opcnt; i++) |
| 4300 | { |
| 4301 | po = &ops[i]; |
| 4302 | if (cbits[i >> 3] & (1 << (i & 7))) |
| 4303 | return; |
| 4304 | cbits[i >> 3] |= (1 << (i & 7)); |
| 4305 | |
| 4306 | if ((po->flags & OPF_JMP) && po->op != OP_CALL) { |
| 4307 | if (po->flags & (OPF_RMD|OPF_DONE)) |
| 4308 | continue; |
| 4309 | if (po->btj != NULL) { |
| 4310 | for (j = 0; j < po->btj->count; j++) { |
| 4311 | check_i(po, po->btj->d[j].bt_i); |
| 4312 | reg_use_pass(po->btj->d[j].bt_i, opcnt, cbits, |
| 4313 | regmask_now, regmask, regmask_save_now, regmask_save, |
| 4314 | regmask_init, regmask_arg); |
| 4315 | } |
| 4316 | return; |
| 4317 | } |
| 4318 | |
| 4319 | check_i(po, po->bt_i); |
| 4320 | if (po->flags & OPF_CJMP) |
| 4321 | reg_use_pass(po->bt_i, opcnt, cbits, |
| 4322 | regmask_now, regmask, regmask_save_now, regmask_save, |
| 4323 | regmask_init, regmask_arg); |
| 4324 | else |
| 4325 | i = po->bt_i - 1; |
| 4326 | continue; |
| 4327 | } |
| 4328 | |
| 4329 | if (po->op == OP_PUSH && !(po->flags & (OPF_FARG|OPF_DONE)) |
| 4330 | && !g_func_pp->is_userstack |
| 4331 | && po->operand[0].type == OPT_REG) |
| 4332 | { |
| 4333 | reg = po->operand[0].reg; |
| 4334 | ferr_assert(po, reg >= 0); |
| 4335 | |
| 4336 | already_saved = 0; |
| 4337 | flags_set = OPF_RSAVE | OPF_RMD | OPF_DONE; |
| 4338 | if (regmask_now & (1 << reg)) { |
| 4339 | already_saved = regmask_save_now & (1 << reg); |
| 4340 | flags_set = OPF_RSAVE | OPF_DONE; |
| 4341 | } |
| 4342 | |
| 4343 | ret = scan_for_pop(i + 1, opcnt, i + opcnt * 3, reg, 0, 0); |
| 4344 | if (ret == 1) { |
| 4345 | scan_for_pop(i + 1, opcnt, i + opcnt * 4, reg, 0, flags_set); |
| 4346 | } |
| 4347 | else { |
| 4348 | ret = scan_for_pop_ret(i + 1, opcnt, po->operand[0].reg, 0); |
| 4349 | if (ret == 1) { |
| 4350 | scan_for_pop_ret(i + 1, opcnt, po->operand[0].reg, |
| 4351 | flags_set); |
| 4352 | } |
| 4353 | } |
| 4354 | if (ret == 1) { |
| 4355 | ferr_assert(po, !already_saved); |
| 4356 | po->flags |= flags_set; |
| 4357 | |
| 4358 | if (regmask_now & (1 << reg)) { |
| 4359 | regmask_save_now |= (1 << reg); |
| 4360 | *regmask_save |= regmask_save_now; |
| 4361 | } |
| 4362 | continue; |
| 4363 | } |
| 4364 | } |
| 4365 | else if (po->op == OP_POP && (po->flags & OPF_RSAVE)) { |
| 4366 | reg = po->operand[0].reg; |
| 4367 | ferr_assert(po, reg >= 0); |
| 4368 | |
| 4369 | if (regmask_save_now & (1 << reg)) |
| 4370 | regmask_save_now &= ~(1 << reg); |
| 4371 | else |
| 4372 | regmask_now &= ~(1 << reg); |
| 4373 | continue; |
| 4374 | } |
| 4375 | else if (po->op == OP_CALL) { |
| 4376 | if ((po->regmask_dst & (1 << xAX)) |
| 4377 | && !(po->regmask_dst & (1 << xDX))) |
| 4378 | { |
| 4379 | if (po->flags & OPF_TAIL) |
| 4380 | // don't need eax, will do "return f();" or "f(); return;" |
| 4381 | po->regmask_dst &= ~(1 << xAX); |
| 4382 | else { |
| 4383 | struct parsed_opr opr = OPR_INIT(OPT_REG, OPLM_DWORD, xAX); |
| 4384 | j = -1; |
| 4385 | find_next_read(i + 1, opcnt, &opr, i + opcnt * 17, &j); |
| 4386 | if (j == -1) |
| 4387 | // not used |
| 4388 | po->regmask_dst &= ~(1 << xAX); |
| 4389 | } |
| 4390 | } |
| 4391 | } |
| 4392 | |
| 4393 | if (po->flags & OPF_NOREGS) |
| 4394 | continue; |
| 4395 | |
| 4396 | // if incomplete register is used, clear it on init to avoid |
| 4397 | // later use of uninitialized upper part in some situations |
| 4398 | if ((po->flags & OPF_DATA) && po->operand[0].type == OPT_REG |
| 4399 | && po->operand[0].lmod != OPLM_DWORD) |
| 4400 | { |
| 4401 | reg = po->operand[0].reg; |
| 4402 | ferr_assert(po, reg >= 0); |
| 4403 | |
| 4404 | if (!(regmask_now & (1 << reg))) |
| 4405 | *regmask_init |= 1 << reg; |
| 4406 | } |
| 4407 | |
| 4408 | regmask_op = po->regmask_src | po->regmask_dst; |
| 4409 | |
| 4410 | regmask_new = po->regmask_src & ~regmask_now & ~regmask_arg; |
| 4411 | regmask_new &= ~(1 << xSP); |
| 4412 | if (g_bp_frame && !(po->flags & OPF_EBP_S)) |
| 4413 | regmask_new &= ~(1 << xBP); |
| 4414 | |
| 4415 | if (po->op == OP_CALL) { |
| 4416 | // allow fastcall calls from anywhere, calee may be also sitting |
| 4417 | // in some fastcall table even when it's not using reg args |
| 4418 | if (regmask_new & po->regmask_src & (1 << xCX)) { |
| 4419 | *regmask_init |= (1 << xCX); |
| 4420 | regmask_now |= (1 << xCX); |
| 4421 | regmask_new &= ~(1 << xCX); |
| 4422 | } |
| 4423 | if (regmask_new & po->regmask_src & (1 << xDX)) { |
| 4424 | *regmask_init |= (1 << xDX); |
| 4425 | regmask_now |= (1 << xDX); |
| 4426 | regmask_new &= ~(1 << xDX); |
| 4427 | } |
| 4428 | } |
| 4429 | |
| 4430 | if (regmask_new != 0) |
| 4431 | fnote(po, "uninitialized reg mask: %x\n", regmask_new); |
| 4432 | |
| 4433 | if (regmask_op & (1 << xBP)) { |
| 4434 | if (g_bp_frame && !(po->flags & OPF_EBP_S)) { |
| 4435 | if (po->regmask_dst & (1 << xBP)) |
| 4436 | // compiler decided to drop bp frame and use ebp as scratch |
| 4437 | scan_fwd_set_flags(i + 1, opcnt, i + opcnt * 5, OPF_EBP_S); |
| 4438 | else |
| 4439 | regmask_op &= ~(1 << xBP); |
| 4440 | } |
| 4441 | } |
| 4442 | |
| 4443 | regmask_now |= regmask_op; |
| 4444 | *regmask |= regmask_now; |
| 4445 | |
| 4446 | if (po->flags & OPF_TAIL) |
| 4447 | return; |
| 4448 | } |
| 4449 | } |
| 4450 | |
| 4451 | static void pp_insert_reg_arg(struct parsed_proto *pp, const char *reg) |
| 4452 | { |
| 4453 | int i; |
| 4454 | |
| 4455 | for (i = 0; i < pp->argc; i++) |
| 4456 | if (pp->arg[i].reg == NULL) |
| 4457 | break; |
| 4458 | |
| 4459 | if (pp->argc_stack) |
| 4460 | memmove(&pp->arg[i + 1], &pp->arg[i], |
| 4461 | sizeof(pp->arg[0]) * pp->argc_stack); |
| 4462 | memset(&pp->arg[i], 0, sizeof(pp->arg[i])); |
| 4463 | pp->arg[i].reg = strdup(reg); |
| 4464 | pp->arg[i].type.name = strdup("int"); |
| 4465 | pp->argc++; |
| 4466 | pp->argc_reg++; |
| 4467 | } |
| 4468 | |
| 4469 | static void output_std_flags(FILE *fout, struct parsed_op *po, |
| 4470 | int *pfomask, const char *dst_opr_text) |
| 4471 | { |
| 4472 | if (*pfomask & (1 << PFO_Z)) { |
| 4473 | fprintf(fout, "\n cond_z = (%s%s == 0);", |
| 4474 | lmod_cast_u(po, po->operand[0].lmod), dst_opr_text); |
| 4475 | *pfomask &= ~(1 << PFO_Z); |
| 4476 | } |
| 4477 | if (*pfomask & (1 << PFO_S)) { |
| 4478 | fprintf(fout, "\n cond_s = (%s%s < 0);", |
| 4479 | lmod_cast_s(po, po->operand[0].lmod), dst_opr_text); |
| 4480 | *pfomask &= ~(1 << PFO_S); |
| 4481 | } |
| 4482 | } |
| 4483 | |
| 4484 | enum { |
| 4485 | OPP_FORCE_NORETURN = (1 << 0), |
| 4486 | OPP_SIMPLE_ARGS = (1 << 1), |
| 4487 | OPP_ALIGN = (1 << 2), |
| 4488 | }; |
| 4489 | |
| 4490 | static void output_pp_attrs(FILE *fout, const struct parsed_proto *pp, |
| 4491 | int flags) |
| 4492 | { |
| 4493 | const char *cconv = ""; |
| 4494 | |
| 4495 | if (pp->is_fastcall) |
| 4496 | cconv = "__fastcall "; |
| 4497 | else if (pp->is_stdcall && pp->argc_reg == 0) |
| 4498 | cconv = "__stdcall "; |
| 4499 | |
| 4500 | fprintf(fout, (flags & OPP_ALIGN) ? "%-16s" : "%s", cconv); |
| 4501 | |
| 4502 | if (pp->is_noreturn || (flags & OPP_FORCE_NORETURN)) |
| 4503 | fprintf(fout, "noreturn "); |
| 4504 | } |
| 4505 | |
| 4506 | static void output_pp(FILE *fout, const struct parsed_proto *pp, |
| 4507 | int flags) |
| 4508 | { |
| 4509 | int i; |
| 4510 | |
| 4511 | fprintf(fout, (flags & OPP_ALIGN) ? "%-5s" : "%s ", |
| 4512 | pp->ret_type.name); |
| 4513 | if (pp->is_fptr) |
| 4514 | fprintf(fout, "("); |
| 4515 | output_pp_attrs(fout, pp, flags); |
| 4516 | if (pp->is_fptr) |
| 4517 | fprintf(fout, "*"); |
| 4518 | fprintf(fout, "%s", pp->name); |
| 4519 | if (pp->is_fptr) |
| 4520 | fprintf(fout, ")"); |
| 4521 | |
| 4522 | fprintf(fout, "("); |
| 4523 | for (i = 0; i < pp->argc; i++) { |
| 4524 | if (i > 0) |
| 4525 | fprintf(fout, ", "); |
| 4526 | if (pp->arg[i].fptr != NULL && !(flags & OPP_SIMPLE_ARGS)) { |
| 4527 | // func pointer |
| 4528 | output_pp(fout, pp->arg[i].fptr, 0); |
| 4529 | } |
| 4530 | else if (pp->arg[i].type.is_retreg) { |
| 4531 | fprintf(fout, "u32 *r_%s", pp->arg[i].reg); |
| 4532 | } |
| 4533 | else { |
| 4534 | fprintf(fout, "%s", pp->arg[i].type.name); |
| 4535 | if (!pp->is_fptr) |
| 4536 | fprintf(fout, " a%d", i + 1); |
| 4537 | } |
| 4538 | } |
| 4539 | if (pp->is_vararg) { |
| 4540 | if (i > 0) |
| 4541 | fprintf(fout, ", "); |
| 4542 | fprintf(fout, "..."); |
| 4543 | } |
| 4544 | fprintf(fout, ")"); |
| 4545 | } |
| 4546 | |
| 4547 | static char *saved_arg_name(char *buf, size_t buf_size, int grp, int num) |
| 4548 | { |
| 4549 | char buf1[16]; |
| 4550 | |
| 4551 | buf1[0] = 0; |
| 4552 | if (grp > 0) |
| 4553 | snprintf(buf1, sizeof(buf1), "%d", grp); |
| 4554 | snprintf(buf, buf_size, "s%s_a%d", buf1, num); |
| 4555 | |
| 4556 | return buf; |
| 4557 | } |
| 4558 | |
| 4559 | static void gen_x_cleanup(int opcnt); |
| 4560 | |
| 4561 | static void gen_func(FILE *fout, FILE *fhdr, const char *funcn, int opcnt) |
| 4562 | { |
| 4563 | struct parsed_op *po, *delayed_flag_op = NULL, *tmp_op; |
| 4564 | struct parsed_opr *last_arith_dst = NULL; |
| 4565 | char buf1[256], buf2[256], buf3[256], cast[64]; |
| 4566 | struct parsed_proto *pp, *pp_tmp; |
| 4567 | struct parsed_data *pd; |
| 4568 | unsigned int uval; |
| 4569 | int save_arg_vars[MAX_ARG_GRP] = { 0, }; |
| 4570 | unsigned char cbits[MAX_OPS / 8]; |
| 4571 | int cond_vars = 0; |
| 4572 | int need_tmp_var = 0; |
| 4573 | int need_tmp64 = 0; |
| 4574 | int had_decl = 0; |
| 4575 | int label_pending = 0; |
| 4576 | int regmask_save = 0; // regs saved/restored in this func |
| 4577 | int regmask_arg; // regs from this function args (fastcall, etc) |
| 4578 | int regmask_ret; // regs needed on ret |
| 4579 | int regmask_now; // temp |
| 4580 | int regmask_init = 0; // regs that need zero initialization |
| 4581 | int regmask_pp = 0; // regs used in complex push-pop graph |
| 4582 | int regmask = 0; // used regs |
| 4583 | int pfomask = 0; |
| 4584 | int found = 0; |
| 4585 | int no_output; |
| 4586 | int i, j, l; |
| 4587 | int arg; |
| 4588 | int reg; |
| 4589 | int ret; |
| 4590 | |
| 4591 | g_bp_frame = g_sp_frame = g_stack_fsz = 0; |
| 4592 | g_stack_frame_used = 0; |
| 4593 | |
| 4594 | g_func_pp = proto_parse(fhdr, funcn, 0); |
| 4595 | if (g_func_pp == NULL) |
| 4596 | ferr(ops, "proto_parse failed for '%s'\n", funcn); |
| 4597 | |
| 4598 | regmask_arg = get_pp_arg_regmask_src(g_func_pp); |
| 4599 | regmask_ret = get_pp_arg_regmask_dst(g_func_pp); |
| 4600 | |
| 4601 | if (g_func_pp->has_retreg) { |
| 4602 | for (arg = 0; arg < g_func_pp->argc; arg++) { |
| 4603 | if (g_func_pp->arg[arg].type.is_retreg) { |
| 4604 | reg = char_array_i(regs_r32, |
| 4605 | ARRAY_SIZE(regs_r32), g_func_pp->arg[arg].reg); |
| 4606 | ferr_assert(ops, reg >= 0); |
| 4607 | regmask_ret |= 1 << reg; |
| 4608 | } |
| 4609 | } |
| 4610 | } |
| 4611 | |
| 4612 | // pass1: |
| 4613 | // - resolve all branches |
| 4614 | // - parse calls with labels |
| 4615 | resolve_branches_parse_calls(opcnt); |
| 4616 | |
| 4617 | // pass2: |
| 4618 | // - handle ebp/esp frame, remove ops related to it |
| 4619 | scan_prologue_epilogue(opcnt); |
| 4620 | |
| 4621 | // pass3: |
| 4622 | // - remove dead labels |
| 4623 | // - set regs needed at ret |
| 4624 | for (i = 0; i < opcnt; i++) |
| 4625 | { |
| 4626 | if (g_labels[i] != NULL && g_label_refs[i].i == -1) { |
| 4627 | free(g_labels[i]); |
| 4628 | g_labels[i] = NULL; |
| 4629 | } |
| 4630 | |
| 4631 | if (ops[i].op == OP_RET) |
| 4632 | ops[i].regmask_src |= regmask_ret; |
| 4633 | } |
| 4634 | |
| 4635 | // pass4: |
| 4636 | // - process trivial calls |
| 4637 | for (i = 0; i < opcnt; i++) |
| 4638 | { |
| 4639 | po = &ops[i]; |
| 4640 | if (po->flags & (OPF_RMD|OPF_DONE)) |
| 4641 | continue; |
| 4642 | |
| 4643 | if (po->op == OP_CALL) |
| 4644 | { |
| 4645 | pp = process_call_early(i, opcnt, &j); |
| 4646 | if (pp != NULL) { |
| 4647 | if (!(po->flags & OPF_ATAIL)) |
| 4648 | // since we know the args, try to collect them |
| 4649 | if (collect_call_args_early(po, i, pp, ®mask) != 0) |
| 4650 | pp = NULL; |
| 4651 | } |
| 4652 | |
| 4653 | if (pp != NULL) { |
| 4654 | if (j >= 0) { |
| 4655 | // commit esp adjust |
| 4656 | if (ops[j].op != OP_POP) |
| 4657 | patch_esp_adjust(&ops[j], pp->argc_stack * 4); |
| 4658 | else { |
| 4659 | for (l = 0; l < pp->argc_stack; l++) |
| 4660 | ops[j + l].flags |= OPF_DONE | OPF_RMD | OPF_NOREGS; |
| 4661 | } |
| 4662 | } |
| 4663 | |
| 4664 | if (strstr(pp->ret_type.name, "int64")) |
| 4665 | need_tmp64 = 1; |
| 4666 | |
| 4667 | po->flags |= OPF_DONE; |
| 4668 | } |
| 4669 | } |
| 4670 | } |
| 4671 | |
| 4672 | // pass5: |
| 4673 | // - process calls, stage 2 |
| 4674 | // - handle some push/pop pairs |
| 4675 | // - scan for STD/CLD, propagate DF |
| 4676 | for (i = 0; i < opcnt; i++) |
| 4677 | { |
| 4678 | po = &ops[i]; |
| 4679 | if (po->flags & OPF_RMD) |
| 4680 | continue; |
| 4681 | |
| 4682 | if (po->op == OP_CALL) |
| 4683 | { |
| 4684 | if (!(po->flags & OPF_DONE)) { |
| 4685 | pp = process_call(i, opcnt); |
| 4686 | |
| 4687 | if (!pp->is_unresolved && !(po->flags & OPF_ATAIL)) { |
| 4688 | // since we know the args, collect them |
| 4689 | collect_call_args(po, i, pp, ®mask, save_arg_vars, |
| 4690 | i + opcnt * 2); |
| 4691 | } |
| 4692 | // for unresolved, collect after other passes |
| 4693 | } |
| 4694 | |
| 4695 | pp = po->pp; |
| 4696 | ferr_assert(po, pp != NULL); |
| 4697 | |
| 4698 | po->regmask_src |= get_pp_arg_regmask_src(pp); |
| 4699 | po->regmask_dst |= get_pp_arg_regmask_dst(pp); |
| 4700 | |
| 4701 | if (strstr(pp->ret_type.name, "int64")) |
| 4702 | need_tmp64 = 1; |
| 4703 | |
| 4704 | continue; |
| 4705 | } |
| 4706 | |
| 4707 | if (po->flags & OPF_DONE) |
| 4708 | continue; |
| 4709 | |
| 4710 | if (po->op == OP_PUSH && !(po->flags & OPF_FARG) |
| 4711 | && !(po->flags & OPF_RSAVE) && po->operand[0].type == OPT_CONST) |
| 4712 | { |
| 4713 | scan_for_pop_const(i, opcnt, i + opcnt * 12); |
| 4714 | } |
| 4715 | else if (po->op == OP_POP) |
| 4716 | scan_pushes_for_pop(i, opcnt, ®mask_pp); |
| 4717 | else if (po->op == OP_STD) { |
| 4718 | po->flags |= OPF_DF | OPF_RMD | OPF_DONE; |
| 4719 | scan_propagate_df(i + 1, opcnt); |
| 4720 | } |
| 4721 | } |
| 4722 | |
| 4723 | // pass6: |
| 4724 | // - find POPs for PUSHes, rm both |
| 4725 | // - scan for all used registers |
| 4726 | memset(cbits, 0, sizeof(cbits)); |
| 4727 | reg_use_pass(0, opcnt, cbits, 0, ®mask, |
| 4728 | 0, ®mask_save, ®mask_init, regmask_arg); |
| 4729 | |
| 4730 | // pass7: |
| 4731 | // - find flag set ops for their users |
| 4732 | // - do unresolved calls |
| 4733 | // - declare indirect functions |
| 4734 | for (i = 0; i < opcnt; i++) |
| 4735 | { |
| 4736 | po = &ops[i]; |
| 4737 | if (po->flags & (OPF_RMD|OPF_DONE)) |
| 4738 | continue; |
| 4739 | |
| 4740 | if (po->flags & OPF_CC) |
| 4741 | { |
| 4742 | int setters[16], cnt = 0, branched = 0; |
| 4743 | |
| 4744 | ret = scan_for_flag_set(i, i + opcnt * 6, |
| 4745 | &branched, setters, &cnt); |
| 4746 | if (ret < 0 || cnt <= 0) |
| 4747 | ferr(po, "unable to trace flag setter(s)\n"); |
| 4748 | if (cnt > ARRAY_SIZE(setters)) |
| 4749 | ferr(po, "too many flag setters\n"); |
| 4750 | |
| 4751 | for (j = 0; j < cnt; j++) |
| 4752 | { |
| 4753 | tmp_op = &ops[setters[j]]; // flag setter |
| 4754 | pfomask = 0; |
| 4755 | |
| 4756 | // to get nicer code, we try to delay test and cmp; |
| 4757 | // if we can't because of operand modification, or if we |
| 4758 | // have arith op, or branch, make it calculate flags explicitly |
| 4759 | if (tmp_op->op == OP_TEST || tmp_op->op == OP_CMP) |
| 4760 | { |
| 4761 | if (branched || scan_for_mod(tmp_op, setters[j] + 1, i, 0) >= 0) |
| 4762 | pfomask = 1 << po->pfo; |
| 4763 | } |
| 4764 | else if (tmp_op->op == OP_CMPS || tmp_op->op == OP_SCAS) { |
| 4765 | pfomask = 1 << po->pfo; |
| 4766 | } |
| 4767 | else { |
| 4768 | // see if we'll be able to handle based on op result |
| 4769 | if ((tmp_op->op != OP_AND && tmp_op->op != OP_OR |
| 4770 | && po->pfo != PFO_Z && po->pfo != PFO_S |
| 4771 | && po->pfo != PFO_P) |
| 4772 | || branched |
| 4773 | || scan_for_mod_opr0(tmp_op, setters[j] + 1, i) >= 0) |
| 4774 | { |
| 4775 | pfomask = 1 << po->pfo; |
| 4776 | } |
| 4777 | |
| 4778 | if (tmp_op->op == OP_ADD && po->pfo == PFO_C) { |
| 4779 | propagate_lmod(tmp_op, &tmp_op->operand[0], |
| 4780 | &tmp_op->operand[1]); |
| 4781 | if (tmp_op->operand[0].lmod == OPLM_DWORD) |
| 4782 | need_tmp64 = 1; |
| 4783 | } |
| 4784 | } |
| 4785 | if (pfomask) { |
| 4786 | tmp_op->pfomask |= pfomask; |
| 4787 | cond_vars |= pfomask; |
| 4788 | } |
| 4789 | // note: may overwrite, currently not a problem |
| 4790 | po->datap = tmp_op; |
| 4791 | } |
| 4792 | |
| 4793 | if (po->op == OP_RCL || po->op == OP_RCR |
| 4794 | || po->op == OP_ADC || po->op == OP_SBB) |
| 4795 | cond_vars |= 1 << PFO_C; |
| 4796 | } |
| 4797 | |
| 4798 | if (po->op == OP_CMPS || po->op == OP_SCAS) { |
| 4799 | cond_vars |= 1 << PFO_Z; |
| 4800 | } |
| 4801 | else if (po->op == OP_MUL |
| 4802 | || (po->op == OP_IMUL && po->operand_cnt == 1)) |
| 4803 | { |
| 4804 | if (po->operand[0].lmod == OPLM_DWORD) |
| 4805 | need_tmp64 = 1; |
| 4806 | } |
| 4807 | else if (po->op == OP_CALL) { |
| 4808 | // note: resolved non-reg calls are OPF_DONE already |
| 4809 | pp = po->pp; |
| 4810 | ferr_assert(po, pp != NULL); |
| 4811 | |
| 4812 | if (pp->is_unresolved) { |
| 4813 | int regmask_stack = 0; |
| 4814 | collect_call_args(po, i, pp, ®mask, save_arg_vars, |
| 4815 | i + opcnt * 2); |
| 4816 | |
| 4817 | // this is pretty rough guess: |
| 4818 | // see ecx and edx were pushed (and not their saved versions) |
| 4819 | for (arg = 0; arg < pp->argc; arg++) { |
| 4820 | if (pp->arg[arg].reg != NULL) |
| 4821 | continue; |
| 4822 | |
| 4823 | tmp_op = pp->arg[arg].datap; |
| 4824 | if (tmp_op == NULL) |
| 4825 | ferr(po, "parsed_op missing for arg%d\n", arg); |
| 4826 | if (tmp_op->p_argnum == 0 && tmp_op->operand[0].type == OPT_REG) |
| 4827 | regmask_stack |= 1 << tmp_op->operand[0].reg; |
| 4828 | } |
| 4829 | |
| 4830 | if (!((regmask_stack & (1 << xCX)) |
| 4831 | && (regmask_stack & (1 << xDX)))) |
| 4832 | { |
| 4833 | if (pp->argc_stack != 0 |
| 4834 | || ((regmask | regmask_arg) & ((1 << xCX)|(1 << xDX)))) |
| 4835 | { |
| 4836 | pp_insert_reg_arg(pp, "ecx"); |
| 4837 | pp->is_fastcall = 1; |
| 4838 | regmask_init |= 1 << xCX; |
| 4839 | regmask |= 1 << xCX; |
| 4840 | } |
| 4841 | if (pp->argc_stack != 0 |
| 4842 | || ((regmask | regmask_arg) & (1 << xDX))) |
| 4843 | { |
| 4844 | pp_insert_reg_arg(pp, "edx"); |
| 4845 | regmask_init |= 1 << xDX; |
| 4846 | regmask |= 1 << xDX; |
| 4847 | } |
| 4848 | } |
| 4849 | |
| 4850 | // note: __cdecl doesn't fall into is_unresolved category |
| 4851 | if (pp->argc_stack > 0) |
| 4852 | pp->is_stdcall = 1; |
| 4853 | } |
| 4854 | } |
| 4855 | else if (po->op == OP_MOV && po->operand[0].pp != NULL |
| 4856 | && po->operand[1].pp != NULL) |
| 4857 | { |
| 4858 | // <var> = offset <something> |
| 4859 | if ((po->operand[1].pp->is_func || po->operand[1].pp->is_fptr) |
| 4860 | && !IS_START(po->operand[1].name, "off_")) |
| 4861 | { |
| 4862 | if (!po->operand[0].pp->is_fptr) |
| 4863 | ferr(po, "%s not declared as fptr when it should be\n", |
| 4864 | po->operand[0].name); |
| 4865 | if (pp_cmp_func(po->operand[0].pp, po->operand[1].pp)) { |
| 4866 | pp_print(buf1, sizeof(buf1), po->operand[0].pp); |
| 4867 | pp_print(buf2, sizeof(buf2), po->operand[1].pp); |
| 4868 | fnote(po, "var: %s\n", buf1); |
| 4869 | fnote(po, "func: %s\n", buf2); |
| 4870 | ferr(po, "^ mismatch\n"); |
| 4871 | } |
| 4872 | } |
| 4873 | } |
| 4874 | else if (po->op == OP_DIV || po->op == OP_IDIV) { |
| 4875 | if (po->operand[0].lmod == OPLM_DWORD) { |
| 4876 | // 32bit division is common, look for it |
| 4877 | if (po->op == OP_DIV) |
| 4878 | ret = scan_for_reg_clear(i, xDX); |
| 4879 | else |
| 4880 | ret = scan_for_cdq_edx(i); |
| 4881 | if (ret >= 0) |
| 4882 | po->flags |= OPF_32BIT; |
| 4883 | else |
| 4884 | need_tmp64 = 1; |
| 4885 | } |
| 4886 | else |
| 4887 | need_tmp_var = 1; |
| 4888 | } |
| 4889 | else if (po->op == OP_CLD) |
| 4890 | po->flags |= OPF_RMD | OPF_DONE; |
| 4891 | |
| 4892 | if (po->op == OP_RCL || po->op == OP_RCR || po->op == OP_XCHG) |
| 4893 | need_tmp_var = 1; |
| 4894 | } |
| 4895 | |
| 4896 | // output starts here |
| 4897 | |
| 4898 | // define userstack size |
| 4899 | if (g_func_pp->is_userstack) { |
| 4900 | fprintf(fout, "#ifndef US_SZ_%s\n", g_func_pp->name); |
| 4901 | fprintf(fout, "#define US_SZ_%s USERSTACK_SIZE\n", g_func_pp->name); |
| 4902 | fprintf(fout, "#endif\n"); |
| 4903 | } |
| 4904 | |
| 4905 | // the function itself |
| 4906 | ferr_assert(ops, !g_func_pp->is_fptr); |
| 4907 | output_pp(fout, g_func_pp, |
| 4908 | (g_ida_func_attr & IDAFA_NORETURN) ? OPP_FORCE_NORETURN : 0); |
| 4909 | fprintf(fout, "\n{\n"); |
| 4910 | |
| 4911 | // declare indirect functions |
| 4912 | for (i = 0; i < opcnt; i++) { |
| 4913 | po = &ops[i]; |
| 4914 | if (po->flags & OPF_RMD) |
| 4915 | continue; |
| 4916 | |
| 4917 | if (po->op == OP_CALL) { |
| 4918 | pp = po->pp; |
| 4919 | if (pp == NULL) |
| 4920 | ferr(po, "NULL pp\n"); |
| 4921 | |
| 4922 | if (pp->is_fptr && !(pp->name[0] != 0 && pp->is_arg)) { |
| 4923 | if (pp->name[0] != 0) { |
| 4924 | memmove(pp->name + 2, pp->name, strlen(pp->name) + 1); |
| 4925 | memcpy(pp->name, "i_", 2); |
| 4926 | |
| 4927 | // might be declared already |
| 4928 | found = 0; |
| 4929 | for (j = 0; j < i; j++) { |
| 4930 | if (ops[j].op == OP_CALL && (pp_tmp = ops[j].pp)) { |
| 4931 | if (pp_tmp->is_fptr && IS(pp->name, pp_tmp->name)) { |
| 4932 | found = 1; |
| 4933 | break; |
| 4934 | } |
| 4935 | } |
| 4936 | } |
| 4937 | if (found) |
| 4938 | continue; |
| 4939 | } |
| 4940 | else |
| 4941 | snprintf(pp->name, sizeof(pp->name), "icall%d", i); |
| 4942 | |
| 4943 | fprintf(fout, " "); |
| 4944 | output_pp(fout, pp, OPP_SIMPLE_ARGS); |
| 4945 | fprintf(fout, ";\n"); |
| 4946 | } |
| 4947 | } |
| 4948 | } |
| 4949 | |
| 4950 | // output LUTs/jumptables |
| 4951 | for (i = 0; i < g_func_pd_cnt; i++) { |
| 4952 | pd = &g_func_pd[i]; |
| 4953 | fprintf(fout, " static const "); |
| 4954 | if (pd->type == OPT_OFFSET) { |
| 4955 | fprintf(fout, "void *jt_%s[] =\n { ", pd->label); |
| 4956 | |
| 4957 | for (j = 0; j < pd->count; j++) { |
| 4958 | if (j > 0) |
| 4959 | fprintf(fout, ", "); |
| 4960 | fprintf(fout, "&&%s", pd->d[j].u.label); |
| 4961 | } |
| 4962 | } |
| 4963 | else { |
| 4964 | fprintf(fout, "%s %s[] =\n { ", |
| 4965 | lmod_type_u(ops, pd->lmod), pd->label); |
| 4966 | |
| 4967 | for (j = 0; j < pd->count; j++) { |
| 4968 | if (j > 0) |
| 4969 | fprintf(fout, ", "); |
| 4970 | fprintf(fout, "%u", pd->d[j].u.val); |
| 4971 | } |
| 4972 | } |
| 4973 | fprintf(fout, " };\n"); |
| 4974 | had_decl = 1; |
| 4975 | } |
| 4976 | |
| 4977 | // declare stack frame, va_arg |
| 4978 | if (g_stack_fsz) { |
| 4979 | fprintf(fout, " union { u32 d[%d]; u16 w[%d]; u8 b[%d]; } sf;\n", |
| 4980 | (g_stack_fsz + 3) / 4, (g_stack_fsz + 1) / 2, g_stack_fsz); |
| 4981 | had_decl = 1; |
| 4982 | } |
| 4983 | |
| 4984 | if (g_func_pp->is_userstack) { |
| 4985 | fprintf(fout, " u32 fake_sf[US_SZ_%s / 4];\n", g_func_pp->name); |
| 4986 | fprintf(fout, " u32 *esp = &fake_sf[sizeof(fake_sf) / 4];\n"); |
| 4987 | had_decl = 1; |
| 4988 | } |
| 4989 | |
| 4990 | if (g_func_pp->is_vararg) { |
| 4991 | fprintf(fout, " va_list ap;\n"); |
| 4992 | had_decl = 1; |
| 4993 | } |
| 4994 | |
| 4995 | // declare arg-registers |
| 4996 | for (i = 0; i < g_func_pp->argc; i++) { |
| 4997 | if (g_func_pp->arg[i].reg != NULL) { |
| 4998 | reg = char_array_i(regs_r32, |
| 4999 | ARRAY_SIZE(regs_r32), g_func_pp->arg[i].reg); |
| 5000 | if (regmask & (1 << reg)) { |
| 5001 | if (g_func_pp->arg[i].type.is_retreg) |
| 5002 | fprintf(fout, " u32 %s = *r_%s;\n", |
| 5003 | g_func_pp->arg[i].reg, g_func_pp->arg[i].reg); |
| 5004 | else |
| 5005 | fprintf(fout, " u32 %s = (u32)a%d;\n", |
| 5006 | g_func_pp->arg[i].reg, i + 1); |
| 5007 | } |
| 5008 | else { |
| 5009 | if (g_func_pp->arg[i].type.is_retreg) |
| 5010 | ferr(ops, "retreg '%s' is unused?\n", |
| 5011 | g_func_pp->arg[i].reg); |
| 5012 | fprintf(fout, " // %s = a%d; // unused\n", |
| 5013 | g_func_pp->arg[i].reg, i + 1); |
| 5014 | } |
| 5015 | had_decl = 1; |
| 5016 | } |
| 5017 | } |
| 5018 | |
| 5019 | // declare normal registers |
| 5020 | regmask_now = regmask & ~regmask_arg; |
| 5021 | regmask_now &= ~(1 << xSP); |
| 5022 | if (regmask_now & 0x00ff) { |
| 5023 | for (reg = 0; reg < 8; reg++) { |
| 5024 | if (regmask_now & (1 << reg)) { |
| 5025 | fprintf(fout, " u32 %s", regs_r32[reg]); |
| 5026 | if (regmask_init & (1 << reg)) |
| 5027 | fprintf(fout, " = 0"); |
| 5028 | fprintf(fout, ";\n"); |
| 5029 | had_decl = 1; |
| 5030 | } |
| 5031 | } |
| 5032 | } |
| 5033 | if (regmask_now & 0xff00) { |
| 5034 | for (reg = 8; reg < 16; reg++) { |
| 5035 | if (regmask_now & (1 << reg)) { |
| 5036 | fprintf(fout, " mmxr %s", regs_r32[reg]); |
| 5037 | if (regmask_init & (1 << reg)) |
| 5038 | fprintf(fout, " = { 0, }"); |
| 5039 | fprintf(fout, ";\n"); |
| 5040 | had_decl = 1; |
| 5041 | } |
| 5042 | } |
| 5043 | } |
| 5044 | |
| 5045 | if (regmask_save) { |
| 5046 | for (reg = 0; reg < 8; reg++) { |
| 5047 | if (regmask_save & (1 << reg)) { |
| 5048 | fprintf(fout, " u32 s_%s;\n", regs_r32[reg]); |
| 5049 | had_decl = 1; |
| 5050 | } |
| 5051 | } |
| 5052 | } |
| 5053 | |
| 5054 | for (i = 0; i < ARRAY_SIZE(save_arg_vars); i++) { |
| 5055 | if (save_arg_vars[i] == 0) |
| 5056 | continue; |
| 5057 | for (reg = 0; reg < 32; reg++) { |
| 5058 | if (save_arg_vars[i] & (1 << reg)) { |
| 5059 | fprintf(fout, " u32 %s;\n", |
| 5060 | saved_arg_name(buf1, sizeof(buf1), i, reg + 1)); |
| 5061 | had_decl = 1; |
| 5062 | } |
| 5063 | } |
| 5064 | } |
| 5065 | |
| 5066 | // declare push-pop temporaries |
| 5067 | if (regmask_pp) { |
| 5068 | for (reg = 0; reg < 8; reg++) { |
| 5069 | if (regmask_pp & (1 << reg)) { |
| 5070 | fprintf(fout, " u32 pp_%s;\n", regs_r32[reg]); |
| 5071 | had_decl = 1; |
| 5072 | } |
| 5073 | } |
| 5074 | } |
| 5075 | |
| 5076 | if (cond_vars) { |
| 5077 | for (i = 0; i < 8; i++) { |
| 5078 | if (cond_vars & (1 << i)) { |
| 5079 | fprintf(fout, " u32 cond_%s;\n", parsed_flag_op_names[i]); |
| 5080 | had_decl = 1; |
| 5081 | } |
| 5082 | } |
| 5083 | } |
| 5084 | |
| 5085 | if (need_tmp_var) { |
| 5086 | fprintf(fout, " u32 tmp;\n"); |
| 5087 | had_decl = 1; |
| 5088 | } |
| 5089 | |
| 5090 | if (need_tmp64) { |
| 5091 | fprintf(fout, " u64 tmp64;\n"); |
| 5092 | had_decl = 1; |
| 5093 | } |
| 5094 | |
| 5095 | if (had_decl) |
| 5096 | fprintf(fout, "\n"); |
| 5097 | |
| 5098 | if (g_func_pp->is_vararg) { |
| 5099 | if (g_func_pp->argc_stack == 0) |
| 5100 | ferr(ops, "vararg func without stack args?\n"); |
| 5101 | fprintf(fout, " va_start(ap, a%d);\n", g_func_pp->argc); |
| 5102 | } |
| 5103 | |
| 5104 | // output ops |
| 5105 | for (i = 0; i < opcnt; i++) |
| 5106 | { |
| 5107 | if (g_labels[i] != NULL) { |
| 5108 | fprintf(fout, "\n%s:\n", g_labels[i]); |
| 5109 | label_pending = 1; |
| 5110 | |
| 5111 | delayed_flag_op = NULL; |
| 5112 | last_arith_dst = NULL; |
| 5113 | } |
| 5114 | |
| 5115 | po = &ops[i]; |
| 5116 | if (po->flags & OPF_RMD) |
| 5117 | continue; |
| 5118 | |
| 5119 | no_output = 0; |
| 5120 | |
| 5121 | #define assert_operand_cnt(n_) \ |
| 5122 | if (po->operand_cnt != n_) \ |
| 5123 | ferr(po, "operand_cnt is %d/%d\n", po->operand_cnt, n_) |
| 5124 | |
| 5125 | // conditional/flag using op? |
| 5126 | if (po->flags & OPF_CC) |
| 5127 | { |
| 5128 | int is_delayed = 0; |
| 5129 | |
| 5130 | tmp_op = po->datap; |
| 5131 | |
| 5132 | // we go through all this trouble to avoid using parsed_flag_op, |
| 5133 | // which makes generated code much nicer |
| 5134 | if (delayed_flag_op != NULL) |
| 5135 | { |
| 5136 | out_cmp_test(buf1, sizeof(buf1), delayed_flag_op, |
| 5137 | po->pfo, po->pfo_inv); |
| 5138 | is_delayed = 1; |
| 5139 | } |
| 5140 | else if (last_arith_dst != NULL |
| 5141 | && (po->pfo == PFO_Z || po->pfo == PFO_S || po->pfo == PFO_P |
| 5142 | || (tmp_op && (tmp_op->op == OP_AND || tmp_op->op == OP_OR)) |
| 5143 | )) |
| 5144 | { |
| 5145 | out_src_opr_u32(buf3, sizeof(buf3), po, last_arith_dst); |
| 5146 | out_test_for_cc(buf1, sizeof(buf1), po, po->pfo, po->pfo_inv, |
| 5147 | last_arith_dst->lmod, buf3); |
| 5148 | is_delayed = 1; |
| 5149 | } |
| 5150 | else if (tmp_op != NULL) { |
| 5151 | // use preprocessed flag calc results |
| 5152 | if (!(tmp_op->pfomask & (1 << po->pfo))) |
| 5153 | ferr(po, "not prepared for pfo %d\n", po->pfo); |
| 5154 | |
| 5155 | // note: pfo_inv was not yet applied |
| 5156 | snprintf(buf1, sizeof(buf1), "(%scond_%s)", |
| 5157 | po->pfo_inv ? "!" : "", parsed_flag_op_names[po->pfo]); |
| 5158 | } |
| 5159 | else { |
| 5160 | ferr(po, "all methods of finding comparison failed\n"); |
| 5161 | } |
| 5162 | |
| 5163 | if (po->flags & OPF_JMP) { |
| 5164 | fprintf(fout, " if %s", buf1); |
| 5165 | } |
| 5166 | else if (po->op == OP_RCL || po->op == OP_RCR |
| 5167 | || po->op == OP_ADC || po->op == OP_SBB) |
| 5168 | { |
| 5169 | if (is_delayed) |
| 5170 | fprintf(fout, " cond_%s = %s;\n", |
| 5171 | parsed_flag_op_names[po->pfo], buf1); |
| 5172 | } |
| 5173 | else if (po->flags & OPF_DATA) { // SETcc |
| 5174 | out_dst_opr(buf2, sizeof(buf2), po, &po->operand[0]); |
| 5175 | fprintf(fout, " %s = %s;", buf2, buf1); |
| 5176 | } |
| 5177 | else { |
| 5178 | ferr(po, "unhandled conditional op\n"); |
| 5179 | } |
| 5180 | } |
| 5181 | |
| 5182 | pfomask = po->pfomask; |
| 5183 | |
| 5184 | if (po->flags & (OPF_REPZ|OPF_REPNZ)) { |
| 5185 | struct parsed_opr opr = OPR_INIT(OPT_REG, OPLM_DWORD, xCX); |
| 5186 | ret = try_resolve_const(i, &opr, opcnt * 7 + i, &uval); |
| 5187 | |
| 5188 | if (ret != 1 || uval == 0) { |
| 5189 | // we need initial flags for ecx=0 case.. |
| 5190 | if (i > 0 && ops[i - 1].op == OP_XOR |
| 5191 | && IS(ops[i - 1].operand[0].name, |
| 5192 | ops[i - 1].operand[1].name)) |
| 5193 | { |
| 5194 | fprintf(fout, " cond_z = "); |
| 5195 | if (pfomask & (1 << PFO_C)) |
| 5196 | fprintf(fout, "cond_c = "); |
| 5197 | fprintf(fout, "0;\n"); |
| 5198 | } |
| 5199 | else if (last_arith_dst != NULL) { |
| 5200 | out_src_opr_u32(buf3, sizeof(buf3), po, last_arith_dst); |
| 5201 | out_test_for_cc(buf1, sizeof(buf1), po, PFO_Z, 0, |
| 5202 | last_arith_dst->lmod, buf3); |
| 5203 | fprintf(fout, " cond_z = %s;\n", buf1); |
| 5204 | } |
| 5205 | else |
| 5206 | ferr(po, "missing initial ZF\n"); |
| 5207 | } |
| 5208 | } |
| 5209 | |
| 5210 | switch (po->op) |
| 5211 | { |
| 5212 | case OP_MOV: |
| 5213 | assert_operand_cnt(2); |
| 5214 | propagate_lmod(po, &po->operand[0], &po->operand[1]); |
| 5215 | out_dst_opr(buf1, sizeof(buf1), po, &po->operand[0]); |
| 5216 | default_cast_to(buf3, sizeof(buf3), &po->operand[0]); |
| 5217 | fprintf(fout, " %s = %s;", buf1, |
| 5218 | out_src_opr(buf2, sizeof(buf2), po, &po->operand[1], |
| 5219 | buf3, 0)); |
| 5220 | break; |
| 5221 | |
| 5222 | case OP_LEA: |
| 5223 | assert_operand_cnt(2); |
| 5224 | po->operand[1].lmod = OPLM_DWORD; // always |
| 5225 | fprintf(fout, " %s = %s;", |
| 5226 | out_dst_opr(buf1, sizeof(buf1), po, &po->operand[0]), |
| 5227 | out_src_opr(buf2, sizeof(buf2), po, &po->operand[1], |
| 5228 | NULL, 1)); |
| 5229 | break; |
| 5230 | |
| 5231 | case OP_MOVZX: |
| 5232 | assert_operand_cnt(2); |
| 5233 | fprintf(fout, " %s = %s;", |
| 5234 | out_dst_opr(buf1, sizeof(buf1), po, &po->operand[0]), |
| 5235 | out_src_opr_u32(buf2, sizeof(buf2), po, &po->operand[1])); |
| 5236 | break; |
| 5237 | |
| 5238 | case OP_MOVSX: |
| 5239 | assert_operand_cnt(2); |
| 5240 | switch (po->operand[1].lmod) { |
| 5241 | case OPLM_BYTE: |
| 5242 | strcpy(buf3, "(s8)"); |
| 5243 | break; |
| 5244 | case OPLM_WORD: |
| 5245 | strcpy(buf3, "(s16)"); |
| 5246 | break; |
| 5247 | default: |
| 5248 | ferr(po, "invalid src lmod: %d\n", po->operand[1].lmod); |
| 5249 | } |
| 5250 | fprintf(fout, " %s = %s;", |
| 5251 | out_dst_opr(buf1, sizeof(buf1), po, &po->operand[0]), |
| 5252 | out_src_opr(buf2, sizeof(buf2), po, &po->operand[1], |
| 5253 | buf3, 0)); |
| 5254 | break; |
| 5255 | |
| 5256 | case OP_XCHG: |
| 5257 | assert_operand_cnt(2); |
| 5258 | propagate_lmod(po, &po->operand[0], &po->operand[1]); |
| 5259 | fprintf(fout, " tmp = %s;", |
| 5260 | out_src_opr(buf1, sizeof(buf1), po, &po->operand[0], "", 0)); |
| 5261 | fprintf(fout, " %s = %s;", |
| 5262 | out_dst_opr(buf1, sizeof(buf1), po, &po->operand[0]), |
| 5263 | out_src_opr(buf2, sizeof(buf2), po, &po->operand[1], |
| 5264 | default_cast_to(buf3, sizeof(buf3), &po->operand[0]), 0)); |
| 5265 | fprintf(fout, " %s = %stmp;", |
| 5266 | out_dst_opr(buf1, sizeof(buf1), po, &po->operand[1]), |
| 5267 | default_cast_to(buf3, sizeof(buf3), &po->operand[1])); |
| 5268 | snprintf(g_comment, sizeof(g_comment), "xchg"); |
| 5269 | break; |
| 5270 | |
| 5271 | case OP_NOT: |
| 5272 | assert_operand_cnt(1); |
| 5273 | out_dst_opr(buf1, sizeof(buf1), po, &po->operand[0]); |
| 5274 | fprintf(fout, " %s = ~%s;", buf1, buf1); |
| 5275 | break; |
| 5276 | |
| 5277 | case OP_XLAT: |
| 5278 | assert_operand_cnt(2); |
| 5279 | out_dst_opr(buf1, sizeof(buf1), po, &po->operand[0]); |
| 5280 | out_src_opr_u32(buf2, sizeof(buf2), po, &po->operand[1]); |
| 5281 | fprintf(fout, " %s = *(u8 *)(%s + %s);", buf1, buf2, buf1); |
| 5282 | strcpy(g_comment, "xlat"); |
| 5283 | break; |
| 5284 | |
| 5285 | case OP_CDQ: |
| 5286 | assert_operand_cnt(2); |
| 5287 | fprintf(fout, " %s = (s32)%s >> 31;", |
| 5288 | out_dst_opr(buf1, sizeof(buf1), po, &po->operand[0]), |
| 5289 | out_src_opr_u32(buf2, sizeof(buf2), po, &po->operand[1])); |
| 5290 | strcpy(g_comment, "cdq"); |
| 5291 | break; |
| 5292 | |
| 5293 | case OP_LODS: |
| 5294 | if (po->flags & OPF_REP) { |
| 5295 | assert_operand_cnt(3); |
| 5296 | // hmh.. |
| 5297 | ferr(po, "TODO\n"); |
| 5298 | } |
| 5299 | else { |
| 5300 | assert_operand_cnt(2); |
| 5301 | fprintf(fout, " eax = %sesi; esi %c= %d;", |
| 5302 | lmod_cast_u_ptr(po, po->operand[0].lmod), |
| 5303 | (po->flags & OPF_DF) ? '-' : '+', |
| 5304 | lmod_bytes(po, po->operand[0].lmod)); |
| 5305 | strcpy(g_comment, "lods"); |
| 5306 | } |
| 5307 | break; |
| 5308 | |
| 5309 | case OP_STOS: |
| 5310 | if (po->flags & OPF_REP) { |
| 5311 | assert_operand_cnt(3); |
| 5312 | fprintf(fout, " for (; ecx != 0; ecx--, edi %c= %d)\n", |
| 5313 | (po->flags & OPF_DF) ? '-' : '+', |
| 5314 | lmod_bytes(po, po->operand[0].lmod)); |
| 5315 | fprintf(fout, " %sedi = eax;", |
| 5316 | lmod_cast_u_ptr(po, po->operand[0].lmod)); |
| 5317 | strcpy(g_comment, "rep stos"); |
| 5318 | } |
| 5319 | else { |
| 5320 | assert_operand_cnt(2); |
| 5321 | fprintf(fout, " %sedi = eax; edi %c= %d;", |
| 5322 | lmod_cast_u_ptr(po, po->operand[0].lmod), |
| 5323 | (po->flags & OPF_DF) ? '-' : '+', |
| 5324 | lmod_bytes(po, po->operand[0].lmod)); |
| 5325 | strcpy(g_comment, "stos"); |
| 5326 | } |
| 5327 | break; |
| 5328 | |
| 5329 | case OP_MOVS: |
| 5330 | j = lmod_bytes(po, po->operand[0].lmod); |
| 5331 | strcpy(buf1, lmod_cast_u_ptr(po, po->operand[0].lmod)); |
| 5332 | l = (po->flags & OPF_DF) ? '-' : '+'; |
| 5333 | if (po->flags & OPF_REP) { |
| 5334 | assert_operand_cnt(3); |
| 5335 | fprintf(fout, |
| 5336 | " for (; ecx != 0; ecx--, edi %c= %d, esi %c= %d)\n", |
| 5337 | l, j, l, j); |
| 5338 | fprintf(fout, |
| 5339 | " %sedi = %sesi;", buf1, buf1); |
| 5340 | strcpy(g_comment, "rep movs"); |
| 5341 | } |
| 5342 | else { |
| 5343 | assert_operand_cnt(2); |
| 5344 | fprintf(fout, " %sedi = %sesi; edi %c= %d; esi %c= %d;", |
| 5345 | buf1, buf1, l, j, l, j); |
| 5346 | strcpy(g_comment, "movs"); |
| 5347 | } |
| 5348 | break; |
| 5349 | |
| 5350 | case OP_CMPS: |
| 5351 | // repe ~ repeat while ZF=1 |
| 5352 | j = lmod_bytes(po, po->operand[0].lmod); |
| 5353 | strcpy(buf1, lmod_cast_u_ptr(po, po->operand[0].lmod)); |
| 5354 | l = (po->flags & OPF_DF) ? '-' : '+'; |
| 5355 | if (po->flags & OPF_REP) { |
| 5356 | assert_operand_cnt(3); |
| 5357 | fprintf(fout, |
| 5358 | " for (; ecx != 0; ecx--) {\n"); |
| 5359 | if (pfomask & (1 << PFO_C)) { |
| 5360 | // ugh.. |
| 5361 | fprintf(fout, |
| 5362 | " cond_c = %sesi < %sedi;\n", buf1, buf1); |
| 5363 | pfomask &= ~(1 << PFO_C); |
| 5364 | } |
| 5365 | fprintf(fout, |
| 5366 | " cond_z = (%sesi == %sedi); esi %c= %d, edi %c= %d;\n", |
| 5367 | buf1, buf1, l, j, l, j); |
| 5368 | fprintf(fout, |
| 5369 | " if (cond_z %s 0) break;\n", |
| 5370 | (po->flags & OPF_REPZ) ? "==" : "!="); |
| 5371 | fprintf(fout, |
| 5372 | " }"); |
| 5373 | snprintf(g_comment, sizeof(g_comment), "rep%s cmps", |
| 5374 | (po->flags & OPF_REPZ) ? "e" : "ne"); |
| 5375 | } |
| 5376 | else { |
| 5377 | assert_operand_cnt(2); |
| 5378 | fprintf(fout, |
| 5379 | " cond_z = (%sesi == %sedi); esi %c= %d; edi %c= %d;", |
| 5380 | buf1, buf1, l, j, l, j); |
| 5381 | strcpy(g_comment, "cmps"); |
| 5382 | } |
| 5383 | pfomask &= ~(1 << PFO_Z); |
| 5384 | last_arith_dst = NULL; |
| 5385 | delayed_flag_op = NULL; |
| 5386 | break; |
| 5387 | |
| 5388 | case OP_SCAS: |
| 5389 | // only does ZF (for now) |
| 5390 | // repe ~ repeat while ZF=1 |
| 5391 | j = lmod_bytes(po, po->operand[0].lmod); |
| 5392 | l = (po->flags & OPF_DF) ? '-' : '+'; |
| 5393 | if (po->flags & OPF_REP) { |
| 5394 | assert_operand_cnt(3); |
| 5395 | fprintf(fout, |
| 5396 | " for (; ecx != 0; ecx--) {\n"); |
| 5397 | fprintf(fout, |
| 5398 | " cond_z = (%seax == %sedi); edi %c= %d;\n", |
| 5399 | lmod_cast_u(po, po->operand[0].lmod), |
| 5400 | lmod_cast_u_ptr(po, po->operand[0].lmod), l, j); |
| 5401 | fprintf(fout, |
| 5402 | " if (cond_z %s 0) break;\n", |
| 5403 | (po->flags & OPF_REPZ) ? "==" : "!="); |
| 5404 | fprintf(fout, |
| 5405 | " }"); |
| 5406 | snprintf(g_comment, sizeof(g_comment), "rep%s scas", |
| 5407 | (po->flags & OPF_REPZ) ? "e" : "ne"); |
| 5408 | } |
| 5409 | else { |
| 5410 | assert_operand_cnt(2); |
| 5411 | fprintf(fout, " cond_z = (%seax == %sedi); edi %c= %d;", |
| 5412 | lmod_cast_u(po, po->operand[0].lmod), |
| 5413 | lmod_cast_u_ptr(po, po->operand[0].lmod), l, j); |
| 5414 | strcpy(g_comment, "scas"); |
| 5415 | } |
| 5416 | pfomask &= ~(1 << PFO_Z); |
| 5417 | last_arith_dst = NULL; |
| 5418 | delayed_flag_op = NULL; |
| 5419 | break; |
| 5420 | |
| 5421 | // arithmetic w/flags |
| 5422 | case OP_AND: |
| 5423 | if (po->operand[1].type == OPT_CONST && !po->operand[1].val) |
| 5424 | goto dualop_arith_const; |
| 5425 | propagate_lmod(po, &po->operand[0], &po->operand[1]); |
| 5426 | goto dualop_arith; |
| 5427 | |
| 5428 | case OP_OR: |
| 5429 | propagate_lmod(po, &po->operand[0], &po->operand[1]); |
| 5430 | if (po->operand[1].type == OPT_CONST) { |
| 5431 | j = lmod_bytes(po, po->operand[0].lmod); |
| 5432 | if (((1ull << j * 8) - 1) == po->operand[1].val) |
| 5433 | goto dualop_arith_const; |
| 5434 | } |
| 5435 | goto dualop_arith; |
| 5436 | |
| 5437 | dualop_arith: |
| 5438 | assert_operand_cnt(2); |
| 5439 | fprintf(fout, " %s %s= %s;", |
| 5440 | out_dst_opr(buf1, sizeof(buf1), po, &po->operand[0]), |
| 5441 | op_to_c(po), |
| 5442 | out_src_opr_u32(buf2, sizeof(buf2), po, &po->operand[1])); |
| 5443 | output_std_flags(fout, po, &pfomask, buf1); |
| 5444 | last_arith_dst = &po->operand[0]; |
| 5445 | delayed_flag_op = NULL; |
| 5446 | break; |
| 5447 | |
| 5448 | dualop_arith_const: |
| 5449 | // and 0, or ~0 used instead mov |
| 5450 | assert_operand_cnt(2); |
| 5451 | fprintf(fout, " %s = %s;", |
| 5452 | out_dst_opr(buf1, sizeof(buf1), po, &po->operand[0]), |
| 5453 | out_src_opr(buf2, sizeof(buf2), po, &po->operand[1], |
| 5454 | default_cast_to(buf3, sizeof(buf3), &po->operand[0]), 0)); |
| 5455 | output_std_flags(fout, po, &pfomask, buf1); |
| 5456 | last_arith_dst = &po->operand[0]; |
| 5457 | delayed_flag_op = NULL; |
| 5458 | break; |
| 5459 | |
| 5460 | case OP_SHL: |
| 5461 | case OP_SHR: |
| 5462 | assert_operand_cnt(2); |
| 5463 | out_dst_opr(buf1, sizeof(buf1), po, &po->operand[0]); |
| 5464 | if (pfomask & (1 << PFO_C)) { |
| 5465 | if (po->operand[1].type == OPT_CONST) { |
| 5466 | l = lmod_bytes(po, po->operand[0].lmod) * 8; |
| 5467 | j = po->operand[1].val; |
| 5468 | j %= l; |
| 5469 | if (j != 0) { |
| 5470 | if (po->op == OP_SHL) |
| 5471 | j = l - j; |
| 5472 | else |
| 5473 | j -= 1; |
| 5474 | fprintf(fout, " cond_c = (%s >> %d) & 1;\n", |
| 5475 | buf1, j); |
| 5476 | } |
| 5477 | else |
| 5478 | ferr(po, "zero shift?\n"); |
| 5479 | } |
| 5480 | else |
| 5481 | ferr(po, "TODO\n"); |
| 5482 | pfomask &= ~(1 << PFO_C); |
| 5483 | } |
| 5484 | fprintf(fout, " %s %s= %s", buf1, op_to_c(po), |
| 5485 | out_src_opr_u32(buf2, sizeof(buf2), po, &po->operand[1])); |
| 5486 | if (po->operand[1].type != OPT_CONST) |
| 5487 | fprintf(fout, " & 0x1f"); |
| 5488 | fprintf(fout, ";"); |
| 5489 | output_std_flags(fout, po, &pfomask, buf1); |
| 5490 | last_arith_dst = &po->operand[0]; |
| 5491 | delayed_flag_op = NULL; |
| 5492 | break; |
| 5493 | |
| 5494 | case OP_SAR: |
| 5495 | assert_operand_cnt(2); |
| 5496 | out_dst_opr(buf1, sizeof(buf1), po, &po->operand[0]); |
| 5497 | fprintf(fout, " %s = %s%s >> %s;", buf1, |
| 5498 | lmod_cast_s(po, po->operand[0].lmod), buf1, |
| 5499 | out_src_opr_u32(buf2, sizeof(buf2), po, &po->operand[1])); |
| 5500 | output_std_flags(fout, po, &pfomask, buf1); |
| 5501 | last_arith_dst = &po->operand[0]; |
| 5502 | delayed_flag_op = NULL; |
| 5503 | break; |
| 5504 | |
| 5505 | case OP_SHLD: |
| 5506 | case OP_SHRD: |
| 5507 | assert_operand_cnt(3); |
| 5508 | propagate_lmod(po, &po->operand[0], &po->operand[1]); |
| 5509 | l = lmod_bytes(po, po->operand[0].lmod) * 8; |
| 5510 | out_src_opr_u32(buf3, sizeof(buf3), po, &po->operand[2]); |
| 5511 | if (po->operand[2].type != OPT_CONST) { |
| 5512 | // no handling for "undefined" case, hopefully not needed |
| 5513 | snprintf(buf2, sizeof(buf2), "(%s & 0x1f)", buf3); |
| 5514 | strcpy(buf3, buf2); |
| 5515 | } |
| 5516 | out_src_opr_u32(buf2, sizeof(buf2), po, &po->operand[1]); |
| 5517 | out_dst_opr(buf1, sizeof(buf1), po, &po->operand[0]); |
| 5518 | if (po->op == OP_SHLD) { |
| 5519 | fprintf(fout, " %s <<= %s; %s |= %s >> (%d - %s);", |
| 5520 | buf1, buf3, buf1, buf2, l, buf3); |
| 5521 | strcpy(g_comment, "shld"); |
| 5522 | } |
| 5523 | else { |
| 5524 | fprintf(fout, " %s >>= %s; %s |= %s << (%d - %s);", |
| 5525 | buf1, buf3, buf1, buf2, l, buf3); |
| 5526 | strcpy(g_comment, "shrd"); |
| 5527 | } |
| 5528 | output_std_flags(fout, po, &pfomask, buf1); |
| 5529 | last_arith_dst = &po->operand[0]; |
| 5530 | delayed_flag_op = NULL; |
| 5531 | break; |
| 5532 | |
| 5533 | case OP_ROL: |
| 5534 | case OP_ROR: |
| 5535 | assert_operand_cnt(2); |
| 5536 | out_dst_opr(buf1, sizeof(buf1), po, &po->operand[0]); |
| 5537 | if (po->operand[1].type == OPT_CONST) { |
| 5538 | j = po->operand[1].val; |
| 5539 | j %= lmod_bytes(po, po->operand[0].lmod) * 8; |
| 5540 | fprintf(fout, po->op == OP_ROL ? |
| 5541 | " %s = (%s << %d) | (%s >> %d);" : |
| 5542 | " %s = (%s >> %d) | (%s << %d);", |
| 5543 | buf1, buf1, j, buf1, |
| 5544 | lmod_bytes(po, po->operand[0].lmod) * 8 - j); |
| 5545 | } |
| 5546 | else |
| 5547 | ferr(po, "TODO\n"); |
| 5548 | output_std_flags(fout, po, &pfomask, buf1); |
| 5549 | last_arith_dst = &po->operand[0]; |
| 5550 | delayed_flag_op = NULL; |
| 5551 | break; |
| 5552 | |
| 5553 | case OP_RCL: |
| 5554 | case OP_RCR: |
| 5555 | assert_operand_cnt(2); |
| 5556 | out_dst_opr(buf1, sizeof(buf1), po, &po->operand[0]); |
| 5557 | l = lmod_bytes(po, po->operand[0].lmod) * 8; |
| 5558 | if (po->operand[1].type == OPT_CONST) { |
| 5559 | j = po->operand[1].val % l; |
| 5560 | if (j == 0) |
| 5561 | ferr(po, "zero rotate\n"); |
| 5562 | fprintf(fout, " tmp = (%s >> %d) & 1;\n", |
| 5563 | buf1, (po->op == OP_RCL) ? (l - j) : (j - 1)); |
| 5564 | if (po->op == OP_RCL) { |
| 5565 | fprintf(fout, |
| 5566 | " %s = (%s << %d) | (cond_c << %d)", |
| 5567 | buf1, buf1, j, j - 1); |
| 5568 | if (j != 1) |
| 5569 | fprintf(fout, " | (%s >> %d)", buf1, l + 1 - j); |
| 5570 | } |
| 5571 | else { |
| 5572 | fprintf(fout, |
| 5573 | " %s = (%s >> %d) | (cond_c << %d)", |
| 5574 | buf1, buf1, j, l - j); |
| 5575 | if (j != 1) |
| 5576 | fprintf(fout, " | (%s << %d)", buf1, l + 1 - j); |
| 5577 | } |
| 5578 | fprintf(fout, ";\n"); |
| 5579 | fprintf(fout, " cond_c = tmp;"); |
| 5580 | } |
| 5581 | else |
| 5582 | ferr(po, "TODO\n"); |
| 5583 | strcpy(g_comment, (po->op == OP_RCL) ? "rcl" : "rcr"); |
| 5584 | output_std_flags(fout, po, &pfomask, buf1); |
| 5585 | last_arith_dst = &po->operand[0]; |
| 5586 | delayed_flag_op = NULL; |
| 5587 | break; |
| 5588 | |
| 5589 | case OP_XOR: |
| 5590 | assert_operand_cnt(2); |
| 5591 | propagate_lmod(po, &po->operand[0], &po->operand[1]); |
| 5592 | if (IS(opr_name(po, 0), opr_name(po, 1))) { |
| 5593 | // special case for XOR |
| 5594 | if (pfomask & (1 << PFO_BE)) { // weird, but it happens.. |
| 5595 | fprintf(fout, " cond_be = 1;\n"); |
| 5596 | pfomask &= ~(1 << PFO_BE); |
| 5597 | } |
| 5598 | fprintf(fout, " %s = 0;", |
| 5599 | out_dst_opr(buf1, sizeof(buf1), po, &po->operand[0])); |
| 5600 | last_arith_dst = &po->operand[0]; |
| 5601 | delayed_flag_op = NULL; |
| 5602 | break; |
| 5603 | } |
| 5604 | goto dualop_arith; |
| 5605 | |
| 5606 | case OP_ADD: |
| 5607 | assert_operand_cnt(2); |
| 5608 | propagate_lmod(po, &po->operand[0], &po->operand[1]); |
| 5609 | if (pfomask & (1 << PFO_C)) { |
| 5610 | out_src_opr_u32(buf1, sizeof(buf1), po, &po->operand[0]); |
| 5611 | out_src_opr_u32(buf2, sizeof(buf2), po, &po->operand[1]); |
| 5612 | if (po->operand[0].lmod == OPLM_DWORD) { |
| 5613 | fprintf(fout, " tmp64 = (u64)%s + %s;\n", buf1, buf2); |
| 5614 | fprintf(fout, " cond_c = tmp64 >> 32;\n"); |
| 5615 | fprintf(fout, " %s = (u32)tmp64;", |
| 5616 | out_dst_opr(buf1, sizeof(buf1), po, &po->operand[0])); |
| 5617 | strcat(g_comment, "add64"); |
| 5618 | } |
| 5619 | else { |
| 5620 | fprintf(fout, " cond_c = ((u32)%s + %s) >> %d;\n", |
| 5621 | buf1, buf2, lmod_bytes(po, po->operand[0].lmod) * 8); |
| 5622 | fprintf(fout, " %s += %s;", |
| 5623 | out_dst_opr(buf1, sizeof(buf1), po, &po->operand[0]), |
| 5624 | buf2); |
| 5625 | } |
| 5626 | pfomask &= ~(1 << PFO_C); |
| 5627 | output_std_flags(fout, po, &pfomask, buf1); |
| 5628 | last_arith_dst = &po->operand[0]; |
| 5629 | delayed_flag_op = NULL; |
| 5630 | break; |
| 5631 | } |
| 5632 | goto dualop_arith; |
| 5633 | |
| 5634 | case OP_SUB: |
| 5635 | assert_operand_cnt(2); |
| 5636 | propagate_lmod(po, &po->operand[0], &po->operand[1]); |
| 5637 | if (pfomask & ~((1 << PFO_Z) | (1 << PFO_S))) { |
| 5638 | for (j = 0; j <= PFO_LE; j++) { |
| 5639 | if (!(pfomask & (1 << j))) |
| 5640 | continue; |
| 5641 | if (j == PFO_Z || j == PFO_S) |
| 5642 | continue; |
| 5643 | |
| 5644 | out_cmp_for_cc(buf1, sizeof(buf1), po, j, 0); |
| 5645 | fprintf(fout, " cond_%s = %s;\n", |
| 5646 | parsed_flag_op_names[j], buf1); |
| 5647 | pfomask &= ~(1 << j); |
| 5648 | } |
| 5649 | } |
| 5650 | goto dualop_arith; |
| 5651 | |
| 5652 | case OP_ADC: |
| 5653 | case OP_SBB: |
| 5654 | assert_operand_cnt(2); |
| 5655 | propagate_lmod(po, &po->operand[0], &po->operand[1]); |
| 5656 | out_dst_opr(buf1, sizeof(buf1), po, &po->operand[0]); |
| 5657 | if (po->op == OP_SBB |
| 5658 | && IS(po->operand[0].name, po->operand[1].name)) |
| 5659 | { |
| 5660 | // avoid use of unitialized var |
| 5661 | fprintf(fout, " %s = -cond_c;", buf1); |
| 5662 | // carry remains what it was |
| 5663 | pfomask &= ~(1 << PFO_C); |
| 5664 | } |
| 5665 | else { |
| 5666 | fprintf(fout, " %s %s= %s + cond_c;", buf1, op_to_c(po), |
| 5667 | out_src_opr_u32(buf2, sizeof(buf2), po, &po->operand[1])); |
| 5668 | } |
| 5669 | output_std_flags(fout, po, &pfomask, buf1); |
| 5670 | last_arith_dst = &po->operand[0]; |
| 5671 | delayed_flag_op = NULL; |
| 5672 | break; |
| 5673 | |
| 5674 | case OP_BSF: |
| 5675 | assert_operand_cnt(2); |
| 5676 | out_src_opr_u32(buf2, sizeof(buf2), po, &po->operand[1]); |
| 5677 | fprintf(fout, " %s = %s ? __builtin_ffs(%s) - 1 : 0;", |
| 5678 | out_dst_opr(buf1, sizeof(buf1), po, &po->operand[0]), |
| 5679 | buf2, buf2); |
| 5680 | output_std_flags(fout, po, &pfomask, buf1); |
| 5681 | last_arith_dst = &po->operand[0]; |
| 5682 | delayed_flag_op = NULL; |
| 5683 | strcat(g_comment, "bsf"); |
| 5684 | break; |
| 5685 | |
| 5686 | case OP_DEC: |
| 5687 | if (pfomask & ~(PFOB_S | PFOB_S | PFOB_C)) { |
| 5688 | for (j = 0; j <= PFO_LE; j++) { |
| 5689 | if (!(pfomask & (1 << j))) |
| 5690 | continue; |
| 5691 | if (j == PFO_Z || j == PFO_S || j == PFO_C) |
| 5692 | continue; |
| 5693 | |
| 5694 | out_cmp_for_cc(buf1, sizeof(buf1), po, j, 0); |
| 5695 | fprintf(fout, " cond_%s = %s;\n", |
| 5696 | parsed_flag_op_names[j], buf1); |
| 5697 | pfomask &= ~(1 << j); |
| 5698 | } |
| 5699 | } |
| 5700 | // fallthrough |
| 5701 | |
| 5702 | case OP_INC: |
| 5703 | if (pfomask & (1 << PFO_C)) |
| 5704 | // carry is unaffected by inc/dec.. wtf? |
| 5705 | ferr(po, "carry propagation needed\n"); |
| 5706 | |
| 5707 | out_dst_opr(buf1, sizeof(buf1), po, &po->operand[0]); |
| 5708 | if (po->operand[0].type == OPT_REG) { |
| 5709 | strcpy(buf2, po->op == OP_INC ? "++" : "--"); |
| 5710 | fprintf(fout, " %s%s;", buf1, buf2); |
| 5711 | } |
| 5712 | else { |
| 5713 | strcpy(buf2, po->op == OP_INC ? "+" : "-"); |
| 5714 | fprintf(fout, " %s %s= 1;", buf1, buf2); |
| 5715 | } |
| 5716 | output_std_flags(fout, po, &pfomask, buf1); |
| 5717 | last_arith_dst = &po->operand[0]; |
| 5718 | delayed_flag_op = NULL; |
| 5719 | break; |
| 5720 | |
| 5721 | case OP_NEG: |
| 5722 | out_dst_opr(buf1, sizeof(buf1), po, &po->operand[0]); |
| 5723 | out_src_opr_u32(buf2, sizeof(buf2), po, &po->operand[0]); |
| 5724 | fprintf(fout, " %s = -%s%s;", buf1, |
| 5725 | lmod_cast_s(po, po->operand[0].lmod), buf2); |
| 5726 | last_arith_dst = &po->operand[0]; |
| 5727 | delayed_flag_op = NULL; |
| 5728 | if (pfomask & (1 << PFO_C)) { |
| 5729 | fprintf(fout, "\n cond_c = (%s != 0);", buf1); |
| 5730 | pfomask &= ~(1 << PFO_C); |
| 5731 | } |
| 5732 | break; |
| 5733 | |
| 5734 | case OP_IMUL: |
| 5735 | if (po->operand_cnt == 2) { |
| 5736 | propagate_lmod(po, &po->operand[0], &po->operand[1]); |
| 5737 | goto dualop_arith; |
| 5738 | } |
| 5739 | if (po->operand_cnt == 3) |
| 5740 | ferr(po, "TODO imul3\n"); |
| 5741 | // fallthrough |
| 5742 | case OP_MUL: |
| 5743 | assert_operand_cnt(1); |
| 5744 | switch (po->operand[0].lmod) { |
| 5745 | case OPLM_DWORD: |
| 5746 | strcpy(buf1, po->op == OP_IMUL ? "(s64)(s32)" : "(u64)"); |
| 5747 | fprintf(fout, " tmp64 = %seax * %s%s;\n", buf1, buf1, |
| 5748 | out_src_opr_u32(buf2, sizeof(buf2), po, &po->operand[0])); |
| 5749 | fprintf(fout, " edx = tmp64 >> 32;\n"); |
| 5750 | fprintf(fout, " eax = tmp64;"); |
| 5751 | break; |
| 5752 | case OPLM_BYTE: |
| 5753 | strcpy(buf1, po->op == OP_IMUL ? "(s16)(s8)" : "(u16)(u8)"); |
| 5754 | fprintf(fout, " LOWORD(eax) = %seax * %s;", buf1, |
| 5755 | out_src_opr(buf2, sizeof(buf2), po, &po->operand[0], |
| 5756 | buf1, 0)); |
| 5757 | break; |
| 5758 | default: |
| 5759 | ferr(po, "TODO: unhandled mul type\n"); |
| 5760 | break; |
| 5761 | } |
| 5762 | last_arith_dst = NULL; |
| 5763 | delayed_flag_op = NULL; |
| 5764 | break; |
| 5765 | |
| 5766 | case OP_DIV: |
| 5767 | case OP_IDIV: |
| 5768 | assert_operand_cnt(1); |
| 5769 | out_src_opr_u32(buf1, sizeof(buf1), po, &po->operand[0]); |
| 5770 | strcpy(cast, lmod_cast(po, po->operand[0].lmod, |
| 5771 | po->op == OP_IDIV)); |
| 5772 | switch (po->operand[0].lmod) { |
| 5773 | case OPLM_DWORD: |
| 5774 | if (po->flags & OPF_32BIT) |
| 5775 | snprintf(buf2, sizeof(buf2), "%seax", cast); |
| 5776 | else { |
| 5777 | fprintf(fout, " tmp64 = ((u64)edx << 32) | eax;\n"); |
| 5778 | snprintf(buf2, sizeof(buf2), "%stmp64", |
| 5779 | (po->op == OP_IDIV) ? "(s64)" : ""); |
| 5780 | } |
| 5781 | if (po->operand[0].type == OPT_REG |
| 5782 | && po->operand[0].reg == xDX) |
| 5783 | { |
| 5784 | fprintf(fout, " eax = %s / %s%s;\n", buf2, cast, buf1); |
| 5785 | fprintf(fout, " edx = %s %% %s%s;", buf2, cast, buf1); |
| 5786 | } |
| 5787 | else { |
| 5788 | fprintf(fout, " edx = %s %% %s%s;\n", buf2, cast, buf1); |
| 5789 | fprintf(fout, " eax = %s / %s%s;", buf2, cast, buf1); |
| 5790 | } |
| 5791 | break; |
| 5792 | case OPLM_WORD: |
| 5793 | fprintf(fout, " tmp = (edx << 16) | (eax & 0xffff);\n"); |
| 5794 | snprintf(buf2, sizeof(buf2), "%stmp", |
| 5795 | (po->op == OP_IDIV) ? "(s32)" : ""); |
| 5796 | if (po->operand[0].type == OPT_REG |
| 5797 | && po->operand[0].reg == xDX) |
| 5798 | { |
| 5799 | fprintf(fout, " LOWORD(eax) = %s / %s%s;\n", |
| 5800 | buf2, cast, buf1); |
| 5801 | fprintf(fout, " LOWORD(edx) = %s %% %s%s;", |
| 5802 | buf2, cast, buf1); |
| 5803 | } |
| 5804 | else { |
| 5805 | fprintf(fout, " LOWORD(edx) = %s %% %s%s;\n", |
| 5806 | buf2, cast, buf1); |
| 5807 | fprintf(fout, " LOWORD(eax) = %s / %s%s;", |
| 5808 | buf2, cast, buf1); |
| 5809 | } |
| 5810 | strcat(g_comment, "div16"); |
| 5811 | break; |
| 5812 | default: |
| 5813 | ferr(po, "unhandled div lmod %d\n", po->operand[0].lmod); |
| 5814 | } |
| 5815 | last_arith_dst = NULL; |
| 5816 | delayed_flag_op = NULL; |
| 5817 | break; |
| 5818 | |
| 5819 | case OP_TEST: |
| 5820 | case OP_CMP: |
| 5821 | propagate_lmod(po, &po->operand[0], &po->operand[1]); |
| 5822 | if (pfomask != 0) { |
| 5823 | for (j = 0; j < 8; j++) { |
| 5824 | if (pfomask & (1 << j)) { |
| 5825 | out_cmp_test(buf1, sizeof(buf1), po, j, 0); |
| 5826 | fprintf(fout, " cond_%s = %s;", |
| 5827 | parsed_flag_op_names[j], buf1); |
| 5828 | } |
| 5829 | } |
| 5830 | pfomask = 0; |
| 5831 | } |
| 5832 | else |
| 5833 | no_output = 1; |
| 5834 | last_arith_dst = NULL; |
| 5835 | delayed_flag_op = po; |
| 5836 | break; |
| 5837 | |
| 5838 | case OP_SCC: |
| 5839 | // SETcc - should already be handled |
| 5840 | break; |
| 5841 | |
| 5842 | // note: we reuse OP_Jcc for SETcc, only flags differ |
| 5843 | case OP_JCC: |
| 5844 | fprintf(fout, "\n goto %s;", po->operand[0].name); |
| 5845 | break; |
| 5846 | |
| 5847 | case OP_JECXZ: |
| 5848 | fprintf(fout, " if (ecx == 0)\n"); |
| 5849 | fprintf(fout, " goto %s;", po->operand[0].name); |
| 5850 | strcat(g_comment, "jecxz"); |
| 5851 | break; |
| 5852 | |
| 5853 | case OP_LOOP: |
| 5854 | fprintf(fout, " if (--ecx == 0)\n"); |
| 5855 | fprintf(fout, " goto %s;", po->operand[0].name); |
| 5856 | strcat(g_comment, "loop"); |
| 5857 | break; |
| 5858 | |
| 5859 | case OP_JMP: |
| 5860 | assert_operand_cnt(1); |
| 5861 | last_arith_dst = NULL; |
| 5862 | delayed_flag_op = NULL; |
| 5863 | |
| 5864 | if (po->operand[0].type == OPT_REGMEM) { |
| 5865 | ret = sscanf(po->operand[0].name, "%[^[][%[^*]*4]", |
| 5866 | buf1, buf2); |
| 5867 | if (ret != 2) |
| 5868 | ferr(po, "parse failure for jmp '%s'\n", |
| 5869 | po->operand[0].name); |
| 5870 | fprintf(fout, " goto *jt_%s[%s];", buf1, buf2); |
| 5871 | break; |
| 5872 | } |
| 5873 | else if (po->operand[0].type != OPT_LABEL) |
| 5874 | ferr(po, "unhandled jmp type\n"); |
| 5875 | |
| 5876 | fprintf(fout, " goto %s;", po->operand[0].name); |
| 5877 | break; |
| 5878 | |
| 5879 | case OP_CALL: |
| 5880 | assert_operand_cnt(1); |
| 5881 | pp = po->pp; |
| 5882 | my_assert_not(pp, NULL); |
| 5883 | |
| 5884 | strcpy(buf3, " "); |
| 5885 | if (po->flags & OPF_CC) { |
| 5886 | // we treat conditional branch to another func |
| 5887 | // (yes such code exists..) as conditional tailcall |
| 5888 | strcat(buf3, " "); |
| 5889 | fprintf(fout, " {\n"); |
| 5890 | } |
| 5891 | |
| 5892 | if (pp->is_fptr && !pp->is_arg) { |
| 5893 | fprintf(fout, "%s%s = %s;\n", buf3, pp->name, |
| 5894 | out_src_opr(buf1, sizeof(buf1), po, &po->operand[0], |
| 5895 | "(void *)", 0)); |
| 5896 | if (pp->is_unresolved) |
| 5897 | fprintf(fout, "%sunresolved_call(\"%s:%d\", %s);\n", |
| 5898 | buf3, asmfn, po->asmln, pp->name); |
| 5899 | } |
| 5900 | |
| 5901 | fprintf(fout, "%s", buf3); |
| 5902 | if (strstr(pp->ret_type.name, "int64")) { |
| 5903 | if (po->flags & OPF_TAIL) |
| 5904 | ferr(po, "int64 and tail?\n"); |
| 5905 | fprintf(fout, "tmp64 = "); |
| 5906 | } |
| 5907 | else if (!IS(pp->ret_type.name, "void")) { |
| 5908 | if (po->flags & OPF_TAIL) { |
| 5909 | if (regmask_ret & (1 << xAX)) { |
| 5910 | fprintf(fout, "return "); |
| 5911 | if (g_func_pp->ret_type.is_ptr != pp->ret_type.is_ptr) |
| 5912 | fprintf(fout, "(%s)", g_func_pp->ret_type.name); |
| 5913 | } |
| 5914 | } |
| 5915 | else if (po->regmask_dst & (1 << xAX)) { |
| 5916 | fprintf(fout, "eax = "); |
| 5917 | if (pp->ret_type.is_ptr) |
| 5918 | fprintf(fout, "(u32)"); |
| 5919 | } |
| 5920 | } |
| 5921 | |
| 5922 | if (pp->name[0] == 0) |
| 5923 | ferr(po, "missing pp->name\n"); |
| 5924 | fprintf(fout, "%s%s(", pp->name, |
| 5925 | pp->has_structarg ? "_sa" : ""); |
| 5926 | |
| 5927 | if (po->flags & OPF_ATAIL) { |
| 5928 | if (pp->argc_stack != g_func_pp->argc_stack |
| 5929 | || (pp->argc_stack > 0 |
| 5930 | && pp->is_stdcall != g_func_pp->is_stdcall)) |
| 5931 | ferr(po, "incompatible tailcall\n"); |
| 5932 | if (g_func_pp->has_retreg) |
| 5933 | ferr(po, "TODO: retreg+tailcall\n"); |
| 5934 | |
| 5935 | for (arg = j = 0; arg < pp->argc; arg++) { |
| 5936 | if (arg > 0) |
| 5937 | fprintf(fout, ", "); |
| 5938 | |
| 5939 | cast[0] = 0; |
| 5940 | if (pp->arg[arg].type.is_ptr) |
| 5941 | snprintf(cast, sizeof(cast), "(%s)", |
| 5942 | pp->arg[arg].type.name); |
| 5943 | |
| 5944 | if (pp->arg[arg].reg != NULL) { |
| 5945 | fprintf(fout, "%s%s", cast, pp->arg[arg].reg); |
| 5946 | continue; |
| 5947 | } |
| 5948 | // stack arg |
| 5949 | for (; j < g_func_pp->argc; j++) |
| 5950 | if (g_func_pp->arg[j].reg == NULL) |
| 5951 | break; |
| 5952 | fprintf(fout, "%sa%d", cast, j + 1); |
| 5953 | j++; |
| 5954 | } |
| 5955 | } |
| 5956 | else { |
| 5957 | for (arg = 0; arg < pp->argc; arg++) { |
| 5958 | if (arg > 0) |
| 5959 | fprintf(fout, ", "); |
| 5960 | |
| 5961 | cast[0] = 0; |
| 5962 | if (pp->arg[arg].type.is_ptr) |
| 5963 | snprintf(cast, sizeof(cast), "(%s)", |
| 5964 | pp->arg[arg].type.name); |
| 5965 | |
| 5966 | if (pp->arg[arg].reg != NULL) { |
| 5967 | if (pp->arg[arg].type.is_retreg) |
| 5968 | fprintf(fout, "&%s", pp->arg[arg].reg); |
| 5969 | else |
| 5970 | fprintf(fout, "%s%s", cast, pp->arg[arg].reg); |
| 5971 | continue; |
| 5972 | } |
| 5973 | |
| 5974 | // stack arg |
| 5975 | tmp_op = pp->arg[arg].datap; |
| 5976 | if (tmp_op == NULL) |
| 5977 | ferr(po, "parsed_op missing for arg%d\n", arg); |
| 5978 | |
| 5979 | if (tmp_op->flags & OPF_VAPUSH) { |
| 5980 | fprintf(fout, "ap"); |
| 5981 | } |
| 5982 | else if (tmp_op->p_argpass != 0) { |
| 5983 | fprintf(fout, "a%d", tmp_op->p_argpass); |
| 5984 | } |
| 5985 | else if (tmp_op->p_argnum != 0) { |
| 5986 | fprintf(fout, "%s%s", cast, |
| 5987 | saved_arg_name(buf1, sizeof(buf1), |
| 5988 | tmp_op->p_arggrp, tmp_op->p_argnum)); |
| 5989 | } |
| 5990 | else { |
| 5991 | fprintf(fout, "%s", |
| 5992 | out_src_opr(buf1, sizeof(buf1), |
| 5993 | tmp_op, &tmp_op->operand[0], cast, 0)); |
| 5994 | } |
| 5995 | } |
| 5996 | } |
| 5997 | fprintf(fout, ");"); |
| 5998 | |
| 5999 | if (strstr(pp->ret_type.name, "int64")) { |
| 6000 | fprintf(fout, "\n"); |
| 6001 | fprintf(fout, "%sedx = tmp64 >> 32;\n", buf3); |
| 6002 | fprintf(fout, "%seax = tmp64;", buf3); |
| 6003 | } |
| 6004 | |
| 6005 | if (pp->is_unresolved) { |
| 6006 | snprintf(buf2, sizeof(buf2), " unresolved %dreg", |
| 6007 | pp->argc_reg); |
| 6008 | strcat(g_comment, buf2); |
| 6009 | } |
| 6010 | |
| 6011 | if (po->flags & OPF_TAIL) { |
| 6012 | ret = 0; |
| 6013 | if (i == opcnt - 1 || pp->is_noreturn) |
| 6014 | ret = 0; |
| 6015 | else if (IS(pp->ret_type.name, "void")) |
| 6016 | ret = 1; |
| 6017 | else if (!(regmask_ret & (1 << xAX))) |
| 6018 | ret = 1; |
| 6019 | // else already handled as 'return f()' |
| 6020 | |
| 6021 | if (ret) { |
| 6022 | fprintf(fout, "\n%sreturn;", buf3); |
| 6023 | strcat(g_comment, " ^ tailcall"); |
| 6024 | } |
| 6025 | else |
| 6026 | strcat(g_comment, " tailcall"); |
| 6027 | |
| 6028 | if ((regmask_ret & (1 << xAX)) |
| 6029 | && IS(pp->ret_type.name, "void") && !pp->is_noreturn) |
| 6030 | { |
| 6031 | ferr(po, "int func -> void func tailcall?\n"); |
| 6032 | } |
| 6033 | } |
| 6034 | if (pp->is_noreturn) |
| 6035 | strcat(g_comment, " noreturn"); |
| 6036 | if ((po->flags & OPF_ATAIL) && pp->argc_stack > 0) |
| 6037 | strcat(g_comment, " argframe"); |
| 6038 | if (po->flags & OPF_CC) |
| 6039 | strcat(g_comment, " cond"); |
| 6040 | |
| 6041 | if (po->flags & OPF_CC) |
| 6042 | fprintf(fout, "\n }"); |
| 6043 | |
| 6044 | delayed_flag_op = NULL; |
| 6045 | last_arith_dst = NULL; |
| 6046 | break; |
| 6047 | |
| 6048 | case OP_RET: |
| 6049 | if (g_func_pp->is_vararg) |
| 6050 | fprintf(fout, " va_end(ap);\n"); |
| 6051 | if (g_func_pp->has_retreg) { |
| 6052 | for (arg = 0; arg < g_func_pp->argc; arg++) |
| 6053 | if (g_func_pp->arg[arg].type.is_retreg) |
| 6054 | fprintf(fout, " *r_%s = %s;\n", |
| 6055 | g_func_pp->arg[arg].reg, g_func_pp->arg[arg].reg); |
| 6056 | } |
| 6057 | |
| 6058 | if (!(regmask_ret & (1 << xAX))) { |
| 6059 | if (i != opcnt - 1 || label_pending) |
| 6060 | fprintf(fout, " return;"); |
| 6061 | } |
| 6062 | else if (g_func_pp->ret_type.is_ptr) { |
| 6063 | fprintf(fout, " return (%s)eax;", |
| 6064 | g_func_pp->ret_type.name); |
| 6065 | } |
| 6066 | else if (IS(g_func_pp->ret_type.name, "__int64")) |
| 6067 | fprintf(fout, " return ((u64)edx << 32) | eax;"); |
| 6068 | else |
| 6069 | fprintf(fout, " return eax;"); |
| 6070 | |
| 6071 | last_arith_dst = NULL; |
| 6072 | delayed_flag_op = NULL; |
| 6073 | break; |
| 6074 | |
| 6075 | case OP_PUSH: |
| 6076 | out_src_opr_u32(buf1, sizeof(buf1), po, &po->operand[0]); |
| 6077 | if (po->p_argnum != 0) { |
| 6078 | // special case - saved func arg |
| 6079 | fprintf(fout, " %s = %s;", |
| 6080 | saved_arg_name(buf2, sizeof(buf2), |
| 6081 | po->p_arggrp, po->p_argnum), buf1); |
| 6082 | break; |
| 6083 | } |
| 6084 | else if (po->flags & OPF_RSAVE) { |
| 6085 | fprintf(fout, " s_%s = %s;", buf1, buf1); |
| 6086 | break; |
| 6087 | } |
| 6088 | else if (po->flags & OPF_PPUSH) { |
| 6089 | tmp_op = po->datap; |
| 6090 | ferr_assert(po, tmp_op != NULL); |
| 6091 | out_dst_opr(buf2, sizeof(buf2), po, &tmp_op->operand[0]); |
| 6092 | fprintf(fout, " pp_%s = %s;", buf2, buf1); |
| 6093 | break; |
| 6094 | } |
| 6095 | else if (g_func_pp->is_userstack) { |
| 6096 | fprintf(fout, " *(--esp) = %s;", buf1); |
| 6097 | break; |
| 6098 | } |
| 6099 | if (!(g_ida_func_attr & IDAFA_NORETURN)) |
| 6100 | ferr(po, "stray push encountered\n"); |
| 6101 | no_output = 1; |
| 6102 | break; |
| 6103 | |
| 6104 | case OP_POP: |
| 6105 | out_dst_opr(buf1, sizeof(buf1), po, &po->operand[0]); |
| 6106 | if (po->flags & OPF_RSAVE) { |
| 6107 | fprintf(fout, " %s = s_%s;", buf1, buf1); |
| 6108 | break; |
| 6109 | } |
| 6110 | else if (po->flags & OPF_PPUSH) { |
| 6111 | // push/pop graph / non-const |
| 6112 | ferr_assert(po, po->datap == NULL); |
| 6113 | fprintf(fout, " %s = pp_%s;", buf1, buf1); |
| 6114 | break; |
| 6115 | } |
| 6116 | else if (po->datap != NULL) { |
| 6117 | // push/pop pair |
| 6118 | tmp_op = po->datap; |
| 6119 | fprintf(fout, " %s = %s;", buf1, |
| 6120 | out_src_opr(buf2, sizeof(buf2), |
| 6121 | tmp_op, &tmp_op->operand[0], |
| 6122 | default_cast_to(buf3, sizeof(buf3), &po->operand[0]), 0)); |
| 6123 | break; |
| 6124 | } |
| 6125 | else if (g_func_pp->is_userstack) { |
| 6126 | fprintf(fout, " %s = *esp++;", buf1); |
| 6127 | break; |
| 6128 | } |
| 6129 | else |
| 6130 | ferr(po, "stray pop encountered\n"); |
| 6131 | break; |
| 6132 | |
| 6133 | case OP_NOP: |
| 6134 | no_output = 1; |
| 6135 | break; |
| 6136 | |
| 6137 | // mmx |
| 6138 | case OP_EMMS: |
| 6139 | strcpy(g_comment, "(emms)"); |
| 6140 | break; |
| 6141 | |
| 6142 | default: |
| 6143 | no_output = 1; |
| 6144 | ferr(po, "unhandled op type %d, flags %x\n", |
| 6145 | po->op, po->flags); |
| 6146 | break; |
| 6147 | } |
| 6148 | |
| 6149 | if (g_comment[0] != 0) { |
| 6150 | char *p = g_comment; |
| 6151 | while (my_isblank(*p)) |
| 6152 | p++; |
| 6153 | fprintf(fout, " // %s", p); |
| 6154 | g_comment[0] = 0; |
| 6155 | no_output = 0; |
| 6156 | } |
| 6157 | if (!no_output) |
| 6158 | fprintf(fout, "\n"); |
| 6159 | |
| 6160 | // some sanity checking |
| 6161 | if (po->flags & OPF_REP) { |
| 6162 | if (po->op != OP_STOS && po->op != OP_MOVS |
| 6163 | && po->op != OP_CMPS && po->op != OP_SCAS) |
| 6164 | ferr(po, "unexpected rep\n"); |
| 6165 | if (!(po->flags & (OPF_REPZ|OPF_REPNZ)) |
| 6166 | && (po->op == OP_CMPS || po->op == OP_SCAS)) |
| 6167 | ferr(po, "cmps/scas with plain rep\n"); |
| 6168 | } |
| 6169 | if ((po->flags & (OPF_REPZ|OPF_REPNZ)) |
| 6170 | && po->op != OP_CMPS && po->op != OP_SCAS) |
| 6171 | ferr(po, "unexpected repz/repnz\n"); |
| 6172 | |
| 6173 | if (pfomask != 0) |
| 6174 | ferr(po, "missed flag calc, pfomask=%x\n", pfomask); |
| 6175 | |
| 6176 | // see is delayed flag stuff is still valid |
| 6177 | if (delayed_flag_op != NULL && delayed_flag_op != po) { |
| 6178 | if (is_any_opr_modified(delayed_flag_op, po, 0)) |
| 6179 | delayed_flag_op = NULL; |
| 6180 | } |
| 6181 | |
| 6182 | if (last_arith_dst != NULL && last_arith_dst != &po->operand[0]) { |
| 6183 | if (is_opr_modified(last_arith_dst, po)) |
| 6184 | last_arith_dst = NULL; |
| 6185 | } |
| 6186 | |
| 6187 | label_pending = 0; |
| 6188 | } |
| 6189 | |
| 6190 | if (g_stack_fsz && !g_stack_frame_used) |
| 6191 | fprintf(fout, " (void)sf;\n"); |
| 6192 | |
| 6193 | fprintf(fout, "}\n\n"); |
| 6194 | |
| 6195 | gen_x_cleanup(opcnt); |
| 6196 | } |
| 6197 | |
| 6198 | static void gen_x_cleanup(int opcnt) |
| 6199 | { |
| 6200 | int i; |
| 6201 | |
| 6202 | for (i = 0; i < opcnt; i++) { |
| 6203 | struct label_ref *lr, *lr_del; |
| 6204 | |
| 6205 | lr = g_label_refs[i].next; |
| 6206 | while (lr != NULL) { |
| 6207 | lr_del = lr; |
| 6208 | lr = lr->next; |
| 6209 | free(lr_del); |
| 6210 | } |
| 6211 | g_label_refs[i].i = -1; |
| 6212 | g_label_refs[i].next = NULL; |
| 6213 | |
| 6214 | if (ops[i].op == OP_CALL) { |
| 6215 | if (ops[i].pp) |
| 6216 | proto_release(ops[i].pp); |
| 6217 | } |
| 6218 | } |
| 6219 | g_func_pp = NULL; |
| 6220 | } |
| 6221 | |
| 6222 | struct func_proto_dep; |
| 6223 | |
| 6224 | struct func_prototype { |
| 6225 | char name[NAMELEN]; |
| 6226 | int id; |
| 6227 | int argc_stack; |
| 6228 | int regmask_dep; |
| 6229 | int has_ret:3; // -1, 0, 1: unresolved, no, yes |
| 6230 | unsigned int dep_resolved:1; |
| 6231 | unsigned int is_stdcall:1; |
| 6232 | struct func_proto_dep *dep_func; |
| 6233 | int dep_func_cnt; |
| 6234 | const struct parsed_proto *pp; // seed pp, if any |
| 6235 | }; |
| 6236 | |
| 6237 | struct func_proto_dep { |
| 6238 | char *name; |
| 6239 | struct func_prototype *proto; |
| 6240 | int regmask_live; // .. at the time of call |
| 6241 | unsigned int ret_dep:1; // return from this is caller's return |
| 6242 | }; |
| 6243 | |
| 6244 | static struct func_prototype *hg_fp; |
| 6245 | static int hg_fp_cnt; |
| 6246 | |
| 6247 | static struct scanned_var { |
| 6248 | char name[NAMELEN]; |
| 6249 | enum opr_lenmod lmod; |
| 6250 | unsigned int is_seeded:1; |
| 6251 | unsigned int is_c_str:1; |
| 6252 | const struct parsed_proto *pp; // seed pp, if any |
| 6253 | } *hg_vars; |
| 6254 | static int hg_var_cnt; |
| 6255 | |
| 6256 | static void output_hdr_fp(FILE *fout, const struct func_prototype *fp, |
| 6257 | int count); |
| 6258 | |
| 6259 | struct func_prototype *hg_fp_add(const char *funcn) |
| 6260 | { |
| 6261 | struct func_prototype *fp; |
| 6262 | |
| 6263 | if ((hg_fp_cnt & 0xff) == 0) { |
| 6264 | hg_fp = realloc(hg_fp, sizeof(hg_fp[0]) * (hg_fp_cnt + 0x100)); |
| 6265 | my_assert_not(hg_fp, NULL); |
| 6266 | memset(hg_fp + hg_fp_cnt, 0, sizeof(hg_fp[0]) * 0x100); |
| 6267 | } |
| 6268 | |
| 6269 | fp = &hg_fp[hg_fp_cnt]; |
| 6270 | snprintf(fp->name, sizeof(fp->name), "%s", funcn); |
| 6271 | fp->id = hg_fp_cnt; |
| 6272 | fp->argc_stack = -1; |
| 6273 | hg_fp_cnt++; |
| 6274 | |
| 6275 | return fp; |
| 6276 | } |
| 6277 | |
| 6278 | static struct func_proto_dep *hg_fp_find_dep(struct func_prototype *fp, |
| 6279 | const char *name) |
| 6280 | { |
| 6281 | int i; |
| 6282 | |
| 6283 | for (i = 0; i < fp->dep_func_cnt; i++) |
| 6284 | if (IS(fp->dep_func[i].name, name)) |
| 6285 | return &fp->dep_func[i]; |
| 6286 | |
| 6287 | return NULL; |
| 6288 | } |
| 6289 | |
| 6290 | static void hg_fp_add_dep(struct func_prototype *fp, const char *name) |
| 6291 | { |
| 6292 | // is it a dupe? |
| 6293 | if (hg_fp_find_dep(fp, name)) |
| 6294 | return; |
| 6295 | |
| 6296 | if ((fp->dep_func_cnt & 0xff) == 0) { |
| 6297 | fp->dep_func = realloc(fp->dep_func, |
| 6298 | sizeof(fp->dep_func[0]) * (fp->dep_func_cnt + 0x100)); |
| 6299 | my_assert_not(fp->dep_func, NULL); |
| 6300 | memset(&fp->dep_func[fp->dep_func_cnt], 0, |
| 6301 | sizeof(fp->dep_func[0]) * 0x100); |
| 6302 | } |
| 6303 | fp->dep_func[fp->dep_func_cnt].name = strdup(name); |
| 6304 | fp->dep_func_cnt++; |
| 6305 | } |
| 6306 | |
| 6307 | static int hg_fp_cmp_name(const void *p1_, const void *p2_) |
| 6308 | { |
| 6309 | const struct func_prototype *p1 = p1_, *p2 = p2_; |
| 6310 | return strcmp(p1->name, p2->name); |
| 6311 | } |
| 6312 | |
| 6313 | #if 0 |
| 6314 | static int hg_fp_cmp_id(const void *p1_, const void *p2_) |
| 6315 | { |
| 6316 | const struct func_prototype *p1 = p1_, *p2 = p2_; |
| 6317 | return p1->id - p2->id; |
| 6318 | } |
| 6319 | #endif |
| 6320 | |
| 6321 | // recursive register dep pass |
| 6322 | // - track saved regs (part 2) |
| 6323 | // - try to figure out arg-regs |
| 6324 | // - calculate reg deps |
| 6325 | static void gen_hdr_dep_pass(int i, int opcnt, unsigned char *cbits, |
| 6326 | struct func_prototype *fp, int regmask_save, int regmask_dst, |
| 6327 | int *regmask_dep, int *has_ret) |
| 6328 | { |
| 6329 | struct func_proto_dep *dep; |
| 6330 | struct parsed_op *po; |
| 6331 | int from_caller = 0; |
| 6332 | int j, l; |
| 6333 | int reg; |
| 6334 | int ret; |
| 6335 | |
| 6336 | for (; i < opcnt; i++) |
| 6337 | { |
| 6338 | if (cbits[i >> 3] & (1 << (i & 7))) |
| 6339 | return; |
| 6340 | cbits[i >> 3] |= (1 << (i & 7)); |
| 6341 | |
| 6342 | po = &ops[i]; |
| 6343 | |
| 6344 | if ((po->flags & OPF_JMP) && po->op != OP_CALL) { |
| 6345 | if (po->flags & OPF_RMD) |
| 6346 | continue; |
| 6347 | |
| 6348 | if (po->btj != NULL) { |
| 6349 | // jumptable |
| 6350 | for (j = 0; j < po->btj->count; j++) { |
| 6351 | check_i(po, po->btj->d[j].bt_i); |
| 6352 | gen_hdr_dep_pass(po->btj->d[j].bt_i, opcnt, cbits, fp, |
| 6353 | regmask_save, regmask_dst, regmask_dep, has_ret); |
| 6354 | } |
| 6355 | return; |
| 6356 | } |
| 6357 | |
| 6358 | check_i(po, po->bt_i); |
| 6359 | if (po->flags & OPF_CJMP) { |
| 6360 | gen_hdr_dep_pass(po->bt_i, opcnt, cbits, fp, |
| 6361 | regmask_save, regmask_dst, regmask_dep, has_ret); |
| 6362 | } |
| 6363 | else { |
| 6364 | i = po->bt_i - 1; |
| 6365 | } |
| 6366 | continue; |
| 6367 | } |
| 6368 | |
| 6369 | if (po->flags & OPF_FARG) |
| 6370 | /* (just calculate register deps) */; |
| 6371 | else if (po->op == OP_PUSH && po->operand[0].type == OPT_REG) |
| 6372 | { |
| 6373 | reg = po->operand[0].reg; |
| 6374 | ferr_assert(po, reg >= 0); |
| 6375 | |
| 6376 | if (po->flags & OPF_RSAVE) { |
| 6377 | regmask_save |= 1 << reg; |
| 6378 | continue; |
| 6379 | } |
| 6380 | if (po->flags & OPF_DONE) |
| 6381 | continue; |
| 6382 | |
| 6383 | ret = scan_for_pop(i + 1, opcnt, i + opcnt * 2, reg, 0, 0); |
| 6384 | if (ret == 1) { |
| 6385 | regmask_save |= 1 << reg; |
| 6386 | po->flags |= OPF_RMD; |
| 6387 | scan_for_pop(i + 1, opcnt, i + opcnt * 3, reg, 0, OPF_RMD); |
| 6388 | continue; |
| 6389 | } |
| 6390 | } |
| 6391 | else if (po->flags & OPF_RMD) |
| 6392 | continue; |
| 6393 | else if (po->op == OP_CALL) { |
| 6394 | po->regmask_dst |= 1 << xAX; |
| 6395 | |
| 6396 | dep = hg_fp_find_dep(fp, po->operand[0].name); |
| 6397 | if (dep != NULL) |
| 6398 | dep->regmask_live = regmask_save | regmask_dst; |
| 6399 | } |
| 6400 | else if (po->op == OP_RET) { |
| 6401 | if (po->operand_cnt > 0) { |
| 6402 | fp->is_stdcall = 1; |
| 6403 | if (fp->argc_stack >= 0 |
| 6404 | && fp->argc_stack != po->operand[0].val / 4) |
| 6405 | ferr(po, "ret mismatch? (%d)\n", fp->argc_stack * 4); |
| 6406 | fp->argc_stack = po->operand[0].val / 4; |
| 6407 | } |
| 6408 | } |
| 6409 | |
| 6410 | // if has_ret is 0, there is uninitialized eax path, |
| 6411 | // which means it's most likely void func |
| 6412 | if (*has_ret != 0 && (po->flags & OPF_TAIL)) { |
| 6413 | if (po->op == OP_CALL) { |
| 6414 | j = i; |
| 6415 | ret = 1; |
| 6416 | } |
| 6417 | else { |
| 6418 | struct parsed_opr opr = OPR_INIT(OPT_REG, OPLM_DWORD, xAX); |
| 6419 | j = -1; |
| 6420 | from_caller = 0; |
| 6421 | ret = resolve_origin(i, &opr, i + opcnt * 4, &j, &from_caller); |
| 6422 | } |
| 6423 | |
| 6424 | if (ret != 1 && from_caller) { |
| 6425 | // unresolved eax - probably void func |
| 6426 | *has_ret = 0; |
| 6427 | } |
| 6428 | else { |
| 6429 | if (j >= 0 && ops[j].op == OP_CALL) { |
| 6430 | dep = hg_fp_find_dep(fp, ops[j].operand[0].name); |
| 6431 | if (dep != NULL) |
| 6432 | dep->ret_dep = 1; |
| 6433 | else |
| 6434 | *has_ret = 1; |
| 6435 | } |
| 6436 | else |
| 6437 | *has_ret = 1; |
| 6438 | } |
| 6439 | } |
| 6440 | |
| 6441 | l = regmask_save | regmask_dst; |
| 6442 | if (g_bp_frame && !(po->flags & OPF_EBP_S)) |
| 6443 | l |= 1 << xBP; |
| 6444 | |
| 6445 | l = po->regmask_src & ~l; |
| 6446 | #if 0 |
| 6447 | if (l) |
| 6448 | fnote(po, "dep |= %04x, dst %04x, save %04x (f %x)\n", |
| 6449 | l, regmask_dst, regmask_save, po->flags); |
| 6450 | #endif |
| 6451 | *regmask_dep |= l; |
| 6452 | regmask_dst |= po->regmask_dst; |
| 6453 | |
| 6454 | if (po->flags & OPF_TAIL) |
| 6455 | return; |
| 6456 | } |
| 6457 | } |
| 6458 | |
| 6459 | static void gen_hdr(const char *funcn, int opcnt) |
| 6460 | { |
| 6461 | int save_arg_vars[MAX_ARG_GRP] = { 0, }; |
| 6462 | unsigned char cbits[MAX_OPS / 8]; |
| 6463 | const struct parsed_proto *pp_c; |
| 6464 | struct parsed_proto *pp; |
| 6465 | struct func_prototype *fp; |
| 6466 | struct parsed_op *po; |
| 6467 | int regmask_dummy = 0; |
| 6468 | int regmask_dep; |
| 6469 | int max_bp_offset = 0; |
| 6470 | int has_ret; |
| 6471 | int i, j, l; |
| 6472 | int ret; |
| 6473 | |
| 6474 | pp_c = proto_parse(g_fhdr, funcn, 1); |
| 6475 | if (pp_c != NULL) |
| 6476 | // already in seed, will add to hg_fp later |
| 6477 | return; |
| 6478 | |
| 6479 | fp = hg_fp_add(funcn); |
| 6480 | |
| 6481 | g_bp_frame = g_sp_frame = g_stack_fsz = 0; |
| 6482 | g_stack_frame_used = 0; |
| 6483 | |
| 6484 | // pass1: |
| 6485 | // - resolve all branches |
| 6486 | // - parse calls with labels |
| 6487 | resolve_branches_parse_calls(opcnt); |
| 6488 | |
| 6489 | // pass2: |
| 6490 | // - handle ebp/esp frame, remove ops related to it |
| 6491 | scan_prologue_epilogue(opcnt); |
| 6492 | |
| 6493 | // pass3: |
| 6494 | // - remove dead labels |
| 6495 | // - collect calls |
| 6496 | for (i = 0; i < opcnt; i++) |
| 6497 | { |
| 6498 | if (g_labels[i] != NULL && g_label_refs[i].i == -1) { |
| 6499 | free(g_labels[i]); |
| 6500 | g_labels[i] = NULL; |
| 6501 | } |
| 6502 | |
| 6503 | po = &ops[i]; |
| 6504 | if (po->flags & (OPF_RMD|OPF_DONE)) |
| 6505 | continue; |
| 6506 | |
| 6507 | if (po->op == OP_CALL) { |
| 6508 | if (po->operand[0].type == OPT_LABEL) |
| 6509 | hg_fp_add_dep(fp, opr_name(po, 0)); |
| 6510 | else if (po->pp != NULL) |
| 6511 | hg_fp_add_dep(fp, po->pp->name); |
| 6512 | } |
| 6513 | } |
| 6514 | |
| 6515 | // pass4: |
| 6516 | // - remove dead labels |
| 6517 | // - handle push <const>/pop pairs |
| 6518 | for (i = 0; i < opcnt; i++) |
| 6519 | { |
| 6520 | if (g_labels[i] != NULL && g_label_refs[i].i == -1) { |
| 6521 | free(g_labels[i]); |
| 6522 | g_labels[i] = NULL; |
| 6523 | } |
| 6524 | |
| 6525 | po = &ops[i]; |
| 6526 | if (po->flags & (OPF_RMD|OPF_DONE)) |
| 6527 | continue; |
| 6528 | |
| 6529 | if (po->op == OP_PUSH && po->operand[0].type == OPT_CONST) |
| 6530 | scan_for_pop_const(i, opcnt, i + opcnt * 13); |
| 6531 | } |
| 6532 | |
| 6533 | // pass5: |
| 6534 | // - process trivial calls |
| 6535 | for (i = 0; i < opcnt; i++) |
| 6536 | { |
| 6537 | po = &ops[i]; |
| 6538 | if (po->flags & (OPF_RMD|OPF_DONE)) |
| 6539 | continue; |
| 6540 | |
| 6541 | if (po->op == OP_CALL) |
| 6542 | { |
| 6543 | pp = process_call_early(i, opcnt, &j); |
| 6544 | if (pp != NULL) { |
| 6545 | if (!(po->flags & OPF_ATAIL)) |
| 6546 | // since we know the args, try to collect them |
| 6547 | if (collect_call_args_early(po, i, pp, ®mask_dummy) != 0) |
| 6548 | pp = NULL; |
| 6549 | } |
| 6550 | |
| 6551 | if (pp != NULL) { |
| 6552 | if (j >= 0) { |
| 6553 | // commit esp adjust |
| 6554 | if (ops[j].op != OP_POP) |
| 6555 | patch_esp_adjust(&ops[j], pp->argc_stack * 4); |
| 6556 | else { |
| 6557 | for (l = 0; l < pp->argc_stack; l++) |
| 6558 | ops[j + l].flags |= OPF_DONE | OPF_RMD | OPF_NOREGS; |
| 6559 | } |
| 6560 | } |
| 6561 | |
| 6562 | po->flags |= OPF_DONE; |
| 6563 | } |
| 6564 | } |
| 6565 | } |
| 6566 | |
| 6567 | // pass6: |
| 6568 | // - track saved regs (simple) |
| 6569 | // - process calls |
| 6570 | for (i = 0; i < opcnt; i++) |
| 6571 | { |
| 6572 | po = &ops[i]; |
| 6573 | if (po->flags & (OPF_RMD|OPF_DONE)) |
| 6574 | continue; |
| 6575 | |
| 6576 | if (po->op == OP_PUSH && po->operand[0].type == OPT_REG |
| 6577 | && po->operand[0].reg != xCX) |
| 6578 | { |
| 6579 | ret = scan_for_pop_ret(i + 1, opcnt, po->operand[0].reg, 0); |
| 6580 | if (ret == 1) { |
| 6581 | // regmask_save |= 1 << po->operand[0].reg; // do it later |
| 6582 | po->flags |= OPF_RSAVE | OPF_RMD | OPF_DONE; |
| 6583 | scan_for_pop_ret(i + 1, opcnt, po->operand[0].reg, OPF_RMD); |
| 6584 | } |
| 6585 | } |
| 6586 | else if (po->op == OP_CALL) |
| 6587 | { |
| 6588 | pp = process_call(i, opcnt); |
| 6589 | |
| 6590 | if (!pp->is_unresolved && !(po->flags & OPF_ATAIL)) { |
| 6591 | // since we know the args, collect them |
| 6592 | ret = collect_call_args(po, i, pp, ®mask_dummy, save_arg_vars, |
| 6593 | i + opcnt * 1); |
| 6594 | } |
| 6595 | } |
| 6596 | } |
| 6597 | |
| 6598 | // pass7 |
| 6599 | memset(cbits, 0, sizeof(cbits)); |
| 6600 | regmask_dep = 0; |
| 6601 | has_ret = -1; |
| 6602 | |
| 6603 | gen_hdr_dep_pass(0, opcnt, cbits, fp, 0, 0, ®mask_dep, &has_ret); |
| 6604 | |
| 6605 | // find unreachable code - must be fixed in IDA |
| 6606 | for (i = 0; i < opcnt; i++) |
| 6607 | { |
| 6608 | if (cbits[i >> 3] & (1 << (i & 7))) |
| 6609 | continue; |
| 6610 | |
| 6611 | if (g_labels[i] == NULL && i > 0 && ops[i - 1].op == OP_CALL |
| 6612 | && ops[i - 1].pp != NULL && ops[i - 1].pp->is_osinc) |
| 6613 | { |
| 6614 | // the compiler sometimes still generates code after |
| 6615 | // noreturn OS functions |
| 6616 | break; |
| 6617 | } |
| 6618 | if (ops[i].op != OP_NOP) |
| 6619 | ferr(&ops[i], "unreachable code\n"); |
| 6620 | } |
| 6621 | |
| 6622 | for (i = 0; i < g_eqcnt; i++) { |
| 6623 | if (g_eqs[i].offset > max_bp_offset && g_eqs[i].offset < 4*32) |
| 6624 | max_bp_offset = g_eqs[i].offset; |
| 6625 | } |
| 6626 | |
| 6627 | if (fp->argc_stack < 0) { |
| 6628 | max_bp_offset = (max_bp_offset + 3) & ~3; |
| 6629 | fp->argc_stack = max_bp_offset / 4; |
| 6630 | if ((g_ida_func_attr & IDAFA_BP_FRAME) && fp->argc_stack > 0) |
| 6631 | fp->argc_stack--; |
| 6632 | } |
| 6633 | |
| 6634 | fp->regmask_dep = regmask_dep & ~(1 << xSP); |
| 6635 | fp->has_ret = has_ret; |
| 6636 | #if 0 |
| 6637 | printf("// has_ret %d, regmask_dep %x\n", |
| 6638 | fp->has_ret, fp->regmask_dep); |
| 6639 | output_hdr_fp(stdout, fp, 1); |
| 6640 | if (IS(funcn, "sub_10007F72")) exit(1); |
| 6641 | #endif |
| 6642 | |
| 6643 | gen_x_cleanup(opcnt); |
| 6644 | } |
| 6645 | |
| 6646 | static void hg_fp_resolve_deps(struct func_prototype *fp) |
| 6647 | { |
| 6648 | struct func_prototype fp_s; |
| 6649 | int dep; |
| 6650 | int i; |
| 6651 | |
| 6652 | // this thing is recursive, so mark first.. |
| 6653 | fp->dep_resolved = 1; |
| 6654 | |
| 6655 | for (i = 0; i < fp->dep_func_cnt; i++) { |
| 6656 | strcpy(fp_s.name, fp->dep_func[i].name); |
| 6657 | fp->dep_func[i].proto = bsearch(&fp_s, hg_fp, hg_fp_cnt, |
| 6658 | sizeof(hg_fp[0]), hg_fp_cmp_name); |
| 6659 | if (fp->dep_func[i].proto != NULL) { |
| 6660 | if (!fp->dep_func[i].proto->dep_resolved) |
| 6661 | hg_fp_resolve_deps(fp->dep_func[i].proto); |
| 6662 | |
| 6663 | dep = ~fp->dep_func[i].regmask_live |
| 6664 | & fp->dep_func[i].proto->regmask_dep; |
| 6665 | fp->regmask_dep |= dep; |
| 6666 | // printf("dep %s %s |= %x\n", fp->name, |
| 6667 | // fp->dep_func[i].name, dep); |
| 6668 | |
| 6669 | if (fp->has_ret == -1 && fp->dep_func[i].ret_dep) |
| 6670 | fp->has_ret = fp->dep_func[i].proto->has_ret; |
| 6671 | } |
| 6672 | } |
| 6673 | } |
| 6674 | |
| 6675 | static void output_hdr_fp(FILE *fout, const struct func_prototype *fp, |
| 6676 | int count) |
| 6677 | { |
| 6678 | const struct parsed_proto *pp; |
| 6679 | char *p, namebuf[NAMELEN]; |
| 6680 | const char *name; |
| 6681 | int regmask_dep; |
| 6682 | int argc_stack; |
| 6683 | int j, arg; |
| 6684 | |
| 6685 | for (; count > 0; count--, fp++) { |
| 6686 | if (fp->has_ret == -1) |
| 6687 | fprintf(fout, "// ret unresolved\n"); |
| 6688 | #if 0 |
| 6689 | fprintf(fout, "// dep:"); |
| 6690 | for (j = 0; j < fp->dep_func_cnt; j++) { |
| 6691 | fprintf(fout, " %s/", fp->dep_func[j].name); |
| 6692 | if (fp->dep_func[j].proto != NULL) |
| 6693 | fprintf(fout, "%04x/%d", fp->dep_func[j].proto->regmask_dep, |
| 6694 | fp->dep_func[j].proto->has_ret); |
| 6695 | } |
| 6696 | fprintf(fout, "\n"); |
| 6697 | #endif |
| 6698 | |
| 6699 | p = strchr(fp->name, '@'); |
| 6700 | if (p != NULL) { |
| 6701 | memcpy(namebuf, fp->name, p - fp->name); |
| 6702 | namebuf[p - fp->name] = 0; |
| 6703 | name = namebuf; |
| 6704 | } |
| 6705 | else |
| 6706 | name = fp->name; |
| 6707 | if (name[0] == '_') |
| 6708 | name++; |
| 6709 | |
| 6710 | pp = proto_parse(g_fhdr, name, 1); |
| 6711 | if (pp != NULL && pp->is_include) |
| 6712 | continue; |
| 6713 | |
| 6714 | if (fp->pp != NULL) { |
| 6715 | // part of seed, output later |
| 6716 | continue; |
| 6717 | } |
| 6718 | |
| 6719 | regmask_dep = fp->regmask_dep; |
| 6720 | argc_stack = fp->argc_stack; |
| 6721 | |
| 6722 | fprintf(fout, "%-5s", fp->pp ? fp->pp->ret_type.name : |
| 6723 | (fp->has_ret ? "int" : "void")); |
| 6724 | if (regmask_dep && (fp->is_stdcall || argc_stack == 0) |
| 6725 | && (regmask_dep & ~((1 << xCX) | (1 << xDX))) == 0) |
| 6726 | { |
| 6727 | fprintf(fout, " __fastcall "); |
| 6728 | if (!(regmask_dep & (1 << xDX)) && argc_stack == 0) |
| 6729 | argc_stack = 1; |
| 6730 | else |
| 6731 | argc_stack += 2; |
| 6732 | regmask_dep = 0; |
| 6733 | } |
| 6734 | else if (regmask_dep && !fp->is_stdcall) { |
| 6735 | fprintf(fout, "/*__usercall*/ "); |
| 6736 | } |
| 6737 | else if (regmask_dep) { |
| 6738 | fprintf(fout, "/*__userpurge*/ "); |
| 6739 | } |
| 6740 | else if (fp->is_stdcall) |
| 6741 | fprintf(fout, " __stdcall "); |
| 6742 | else |
| 6743 | fprintf(fout, " __cdecl "); |
| 6744 | |
| 6745 | fprintf(fout, "%s(", name); |
| 6746 | |
| 6747 | arg = 0; |
| 6748 | for (j = 0; j < xSP; j++) { |
| 6749 | if (regmask_dep & (1 << j)) { |
| 6750 | arg++; |
| 6751 | if (arg != 1) |
| 6752 | fprintf(fout, ", "); |
| 6753 | if (fp->pp != NULL) |
| 6754 | fprintf(fout, "%s", fp->pp->arg[arg - 1].type.name); |
| 6755 | else |
| 6756 | fprintf(fout, "int"); |
| 6757 | fprintf(fout, " a%d/*<%s>*/", arg, regs_r32[j]); |
| 6758 | } |
| 6759 | } |
| 6760 | |
| 6761 | for (j = 0; j < argc_stack; j++) { |
| 6762 | arg++; |
| 6763 | if (arg != 1) |
| 6764 | fprintf(fout, ", "); |
| 6765 | if (fp->pp != NULL) { |
| 6766 | fprintf(fout, "%s", fp->pp->arg[arg - 1].type.name); |
| 6767 | if (!fp->pp->arg[arg - 1].type.is_ptr) |
| 6768 | fprintf(fout, " "); |
| 6769 | } |
| 6770 | else |
| 6771 | fprintf(fout, "int "); |
| 6772 | fprintf(fout, "a%d", arg); |
| 6773 | } |
| 6774 | |
| 6775 | fprintf(fout, ");\n"); |
| 6776 | } |
| 6777 | } |
| 6778 | |
| 6779 | static void output_hdr(FILE *fout) |
| 6780 | { |
| 6781 | static const char *lmod_c_names[] = { |
| 6782 | [OPLM_UNSPEC] = "???", |
| 6783 | [OPLM_BYTE] = "uint8_t", |
| 6784 | [OPLM_WORD] = "uint16_t", |
| 6785 | [OPLM_DWORD] = "uint32_t", |
| 6786 | [OPLM_QWORD] = "uint64_t", |
| 6787 | }; |
| 6788 | const struct scanned_var *var; |
| 6789 | struct func_prototype *fp; |
| 6790 | char line[256] = { 0, }; |
| 6791 | char name[256]; |
| 6792 | int i; |
| 6793 | |
| 6794 | // add stuff from headers |
| 6795 | for (i = 0; i < pp_cache_size; i++) { |
| 6796 | if (pp_cache[i].is_cinc && !pp_cache[i].is_stdcall) |
| 6797 | snprintf(name, sizeof(name), "_%s", pp_cache[i].name); |
| 6798 | else |
| 6799 | snprintf(name, sizeof(name), "%s", pp_cache[i].name); |
| 6800 | fp = hg_fp_add(name); |
| 6801 | fp->pp = &pp_cache[i]; |
| 6802 | fp->argc_stack = fp->pp->argc_stack; |
| 6803 | fp->is_stdcall = fp->pp->is_stdcall; |
| 6804 | fp->regmask_dep = get_pp_arg_regmask_src(fp->pp); |
| 6805 | fp->has_ret = !IS(fp->pp->ret_type.name, "void"); |
| 6806 | } |
| 6807 | |
| 6808 | // resolve deps |
| 6809 | qsort(hg_fp, hg_fp_cnt, sizeof(hg_fp[0]), hg_fp_cmp_name); |
| 6810 | for (i = 0; i < hg_fp_cnt; i++) |
| 6811 | hg_fp_resolve_deps(&hg_fp[i]); |
| 6812 | |
| 6813 | // note: messes up .proto ptr, don't use |
| 6814 | //qsort(hg_fp, hg_fp_cnt, sizeof(hg_fp[0]), hg_fp_cmp_id); |
| 6815 | |
| 6816 | // output variables |
| 6817 | for (i = 0; i < hg_var_cnt; i++) { |
| 6818 | var = &hg_vars[i]; |
| 6819 | |
| 6820 | if (var->pp != NULL) |
| 6821 | // part of seed |
| 6822 | continue; |
| 6823 | else if (var->is_c_str) |
| 6824 | fprintf(fout, "extern %-8s %s[];", "char", var->name); |
| 6825 | else |
| 6826 | fprintf(fout, "extern %-8s %s;", |
| 6827 | lmod_c_names[var->lmod], var->name); |
| 6828 | |
| 6829 | if (var->is_seeded) |
| 6830 | fprintf(fout, " // seeded"); |
| 6831 | fprintf(fout, "\n"); |
| 6832 | } |
| 6833 | |
| 6834 | fprintf(fout, "\n"); |
| 6835 | |
| 6836 | // output function prototypes |
| 6837 | output_hdr_fp(fout, hg_fp, hg_fp_cnt); |
| 6838 | |
| 6839 | // seed passthrough |
| 6840 | fprintf(fout, "\n// - seed -\n"); |
| 6841 | |
| 6842 | rewind(g_fhdr); |
| 6843 | while (fgets(line, sizeof(line), g_fhdr)) |
| 6844 | fwrite(line, 1, strlen(line), fout); |
| 6845 | } |
| 6846 | |
| 6847 | // '=' needs special treatment |
| 6848 | // also ' quote |
| 6849 | static char *next_word_s(char *w, size_t wsize, char *s) |
| 6850 | { |
| 6851 | size_t i; |
| 6852 | |
| 6853 | s = sskip(s); |
| 6854 | |
| 6855 | i = 0; |
| 6856 | if (*s == '\'') { |
| 6857 | w[0] = s[0]; |
| 6858 | for (i = 1; i < wsize - 1; i++) { |
| 6859 | if (s[i] == 0) { |
| 6860 | printf("warning: missing closing quote: \"%s\"\n", s); |
| 6861 | break; |
| 6862 | } |
| 6863 | if (s[i] == '\'') |
| 6864 | break; |
| 6865 | w[i] = s[i]; |
| 6866 | } |
| 6867 | } |
| 6868 | |
| 6869 | for (; i < wsize - 1; i++) { |
| 6870 | if (s[i] == 0 || my_isblank(s[i]) || (s[i] == '=' && i > 0)) |
| 6871 | break; |
| 6872 | w[i] = s[i]; |
| 6873 | } |
| 6874 | w[i] = 0; |
| 6875 | |
| 6876 | if (s[i] != 0 && !my_isblank(s[i]) && s[i] != '=') |
| 6877 | printf("warning: '%s' truncated\n", w); |
| 6878 | |
| 6879 | return s + i; |
| 6880 | } |
| 6881 | |
| 6882 | static void scan_variables(FILE *fasm) |
| 6883 | { |
| 6884 | struct scanned_var *var; |
| 6885 | char line[256] = { 0, }; |
| 6886 | char words[3][256]; |
| 6887 | char *p = NULL; |
| 6888 | int wordc; |
| 6889 | int l; |
| 6890 | |
| 6891 | while (!feof(fasm)) |
| 6892 | { |
| 6893 | // skip to next data section |
| 6894 | while (my_fgets(line, sizeof(line), fasm)) |
| 6895 | { |
| 6896 | asmln++; |
| 6897 | |
| 6898 | p = sskip(line); |
| 6899 | if (*p == 0 || *p == ';') |
| 6900 | continue; |
| 6901 | |
| 6902 | p = sskip(next_word_s(words[0], sizeof(words[0]), p)); |
| 6903 | if (*p == 0 || *p == ';') |
| 6904 | continue; |
| 6905 | |
| 6906 | if (*p != 's' || !IS_START(p, "segment para public")) |
| 6907 | continue; |
| 6908 | |
| 6909 | break; |
| 6910 | } |
| 6911 | |
| 6912 | if (p == NULL || !IS_START(p, "segment para public")) |
| 6913 | break; |
| 6914 | p = sskip(p + 19); |
| 6915 | |
| 6916 | if (!IS_START(p, "'DATA'")) |
| 6917 | continue; |
| 6918 | |
| 6919 | // now process it |
| 6920 | while (my_fgets(line, sizeof(line), fasm)) |
| 6921 | { |
| 6922 | asmln++; |
| 6923 | |
| 6924 | p = line; |
| 6925 | if (my_isblank(*p)) |
| 6926 | continue; |
| 6927 | |
| 6928 | p = sskip(p); |
| 6929 | if (*p == 0 || *p == ';') |
| 6930 | continue; |
| 6931 | |
| 6932 | for (wordc = 0; wordc < ARRAY_SIZE(words); wordc++) { |
| 6933 | words[wordc][0] = 0; |
| 6934 | p = sskip(next_word_s(words[wordc], sizeof(words[0]), p)); |
| 6935 | if (*p == 0 || *p == ';') { |
| 6936 | wordc++; |
| 6937 | break; |
| 6938 | } |
| 6939 | } |
| 6940 | |
| 6941 | if (wordc == 2 && IS(words[1], "ends")) |
| 6942 | break; |
| 6943 | if (wordc < 2) |
| 6944 | continue; |
| 6945 | |
| 6946 | if (IS_START(words[0], "__IMPORT_DESCRIPTOR_")) { |
| 6947 | // when this starts, we don't need anything from this section |
| 6948 | break; |
| 6949 | } |
| 6950 | |
| 6951 | if ((hg_var_cnt & 0xff) == 0) { |
| 6952 | hg_vars = realloc(hg_vars, sizeof(hg_vars[0]) |
| 6953 | * (hg_var_cnt + 0x100)); |
| 6954 | my_assert_not(hg_vars, NULL); |
| 6955 | memset(hg_vars + hg_var_cnt, 0, sizeof(hg_vars[0]) * 0x100); |
| 6956 | } |
| 6957 | |
| 6958 | var = &hg_vars[hg_var_cnt++]; |
| 6959 | snprintf(var->name, sizeof(var->name), "%s", words[0]); |
| 6960 | |
| 6961 | // maybe already in seed header? |
| 6962 | var->pp = proto_parse(g_fhdr, var->name, 1); |
| 6963 | if (var->pp != NULL) { |
| 6964 | if (var->pp->is_fptr) { |
| 6965 | var->lmod = OPLM_DWORD; |
| 6966 | //var->is_ptr = 1; |
| 6967 | } |
| 6968 | else if (var->pp->is_func) |
| 6969 | aerr("func?\n"); |
| 6970 | else if (!guess_lmod_from_c_type(&var->lmod, &var->pp->type)) |
| 6971 | aerr("unhandled C type '%s' for '%s'\n", |
| 6972 | var->pp->type.name, var->name); |
| 6973 | |
| 6974 | var->is_seeded = 1; |
| 6975 | continue; |
| 6976 | } |
| 6977 | |
| 6978 | if (IS(words[1], "dd")) |
| 6979 | var->lmod = OPLM_DWORD; |
| 6980 | else if (IS(words[1], "dw")) |
| 6981 | var->lmod = OPLM_WORD; |
| 6982 | else if (IS(words[1], "db")) { |
| 6983 | var->lmod = OPLM_BYTE; |
| 6984 | if (wordc >= 3 && (l = strlen(words[2])) > 4) { |
| 6985 | if (words[2][0] == '\'' && IS(words[2] + l - 2, ",0")) |
| 6986 | var->is_c_str = 1; |
| 6987 | } |
| 6988 | } |
| 6989 | else if (IS(words[1], "dq")) |
| 6990 | var->lmod = OPLM_QWORD; |
| 6991 | //else if (IS(words[1], "dt")) |
| 6992 | else |
| 6993 | aerr("type '%s' not known\n", words[1]); |
| 6994 | } |
| 6995 | } |
| 6996 | |
| 6997 | rewind(fasm); |
| 6998 | asmln = 0; |
| 6999 | } |
| 7000 | |
| 7001 | static void set_label(int i, const char *name) |
| 7002 | { |
| 7003 | const char *p; |
| 7004 | int len; |
| 7005 | |
| 7006 | len = strlen(name); |
| 7007 | p = strchr(name, ':'); |
| 7008 | if (p != NULL) |
| 7009 | len = p - name; |
| 7010 | |
| 7011 | if (g_labels[i] != NULL && !IS_START(g_labels[i], "algn_")) |
| 7012 | aerr("dupe label '%s' vs '%s'?\n", name, g_labels[i]); |
| 7013 | g_labels[i] = realloc(g_labels[i], len + 1); |
| 7014 | my_assert_not(g_labels[i], NULL); |
| 7015 | memcpy(g_labels[i], name, len); |
| 7016 | g_labels[i][len] = 0; |
| 7017 | } |
| 7018 | |
| 7019 | struct chunk_item { |
| 7020 | char *name; |
| 7021 | long fptr; |
| 7022 | int asmln; |
| 7023 | }; |
| 7024 | |
| 7025 | static struct chunk_item *func_chunks; |
| 7026 | static int func_chunk_cnt; |
| 7027 | static int func_chunk_alloc; |
| 7028 | |
| 7029 | static void add_func_chunk(FILE *fasm, const char *name, int line) |
| 7030 | { |
| 7031 | if (func_chunk_cnt >= func_chunk_alloc) { |
| 7032 | func_chunk_alloc *= 2; |
| 7033 | func_chunks = realloc(func_chunks, |
| 7034 | func_chunk_alloc * sizeof(func_chunks[0])); |
| 7035 | my_assert_not(func_chunks, NULL); |
| 7036 | } |
| 7037 | func_chunks[func_chunk_cnt].fptr = ftell(fasm); |
| 7038 | func_chunks[func_chunk_cnt].name = strdup(name); |
| 7039 | func_chunks[func_chunk_cnt].asmln = line; |
| 7040 | func_chunk_cnt++; |
| 7041 | } |
| 7042 | |
| 7043 | static int cmp_chunks(const void *p1, const void *p2) |
| 7044 | { |
| 7045 | const struct chunk_item *c1 = p1, *c2 = p2; |
| 7046 | return strcmp(c1->name, c2->name); |
| 7047 | } |
| 7048 | |
| 7049 | static int cmpstringp(const void *p1, const void *p2) |
| 7050 | { |
| 7051 | return strcmp(*(char * const *)p1, *(char * const *)p2); |
| 7052 | } |
| 7053 | |
| 7054 | static void scan_ahead(FILE *fasm) |
| 7055 | { |
| 7056 | char words[2][256]; |
| 7057 | char line[256]; |
| 7058 | long oldpos; |
| 7059 | int oldasmln; |
| 7060 | int wordc; |
| 7061 | char *p; |
| 7062 | int i; |
| 7063 | |
| 7064 | oldpos = ftell(fasm); |
| 7065 | oldasmln = asmln; |
| 7066 | |
| 7067 | while (my_fgets(line, sizeof(line), fasm)) |
| 7068 | { |
| 7069 | wordc = 0; |
| 7070 | asmln++; |
| 7071 | |
| 7072 | p = sskip(line); |
| 7073 | if (*p == 0) |
| 7074 | continue; |
| 7075 | |
| 7076 | if (*p == ';') |
| 7077 | { |
| 7078 | // get rid of random tabs |
| 7079 | for (i = 0; line[i] != 0; i++) |
| 7080 | if (line[i] == '\t') |
| 7081 | line[i] = ' '; |
| 7082 | |
| 7083 | if (p[2] == 'S' && IS_START(p, "; START OF FUNCTION CHUNK FOR ")) |
| 7084 | { |
| 7085 | p += 30; |
| 7086 | next_word(words[0], sizeof(words[0]), p); |
| 7087 | if (words[0][0] == 0) |
| 7088 | aerr("missing name for func chunk?\n"); |
| 7089 | |
| 7090 | add_func_chunk(fasm, words[0], asmln); |
| 7091 | } |
| 7092 | else if (IS_START(p, "; sctend")) |
| 7093 | break; |
| 7094 | |
| 7095 | continue; |
| 7096 | } // *p == ';' |
| 7097 | |
| 7098 | for (wordc = 0; wordc < ARRAY_SIZE(words); wordc++) { |
| 7099 | words[wordc][0] = 0; |
| 7100 | p = sskip(next_word_s(words[wordc], sizeof(words[0]), p)); |
| 7101 | if (*p == 0 || *p == ';') { |
| 7102 | wordc++; |
| 7103 | break; |
| 7104 | } |
| 7105 | } |
| 7106 | |
| 7107 | if (wordc == 2 && IS(words[1], "ends")) |
| 7108 | break; |
| 7109 | } |
| 7110 | |
| 7111 | fseek(fasm, oldpos, SEEK_SET); |
| 7112 | asmln = oldasmln; |
| 7113 | } |
| 7114 | |
| 7115 | int main(int argc, char *argv[]) |
| 7116 | { |
| 7117 | FILE *fout, *fasm, *frlist; |
| 7118 | struct parsed_data *pd = NULL; |
| 7119 | int pd_alloc = 0; |
| 7120 | char **rlist = NULL; |
| 7121 | int rlist_len = 0; |
| 7122 | int rlist_alloc = 0; |
| 7123 | int func_chunks_used = 0; |
| 7124 | int func_chunks_sorted = 0; |
| 7125 | int func_chunk_i = -1; |
| 7126 | long func_chunk_ret = 0; |
| 7127 | int func_chunk_ret_ln = 0; |
| 7128 | int scanned_ahead = 0; |
| 7129 | char line[256]; |
| 7130 | char words[20][256]; |
| 7131 | enum opr_lenmod lmod; |
| 7132 | char *sctproto = NULL; |
| 7133 | int in_func = 0; |
| 7134 | int pending_endp = 0; |
| 7135 | int skip_func = 0; |
| 7136 | int skip_warned = 0; |
| 7137 | int eq_alloc; |
| 7138 | int verbose = 0; |
| 7139 | int multi_seg = 0; |
| 7140 | int end = 0; |
| 7141 | int arg_out; |
| 7142 | int arg; |
| 7143 | int pi = 0; |
| 7144 | int i, j; |
| 7145 | int ret, len; |
| 7146 | char *p; |
| 7147 | int wordc; |
| 7148 | |
| 7149 | for (arg = 1; arg < argc; arg++) { |
| 7150 | if (IS(argv[arg], "-v")) |
| 7151 | verbose = 1; |
| 7152 | else if (IS(argv[arg], "-rf")) |
| 7153 | g_allow_regfunc = 1; |
| 7154 | else if (IS(argv[arg], "-m")) |
| 7155 | multi_seg = 1; |
| 7156 | else if (IS(argv[arg], "-hdr")) |
| 7157 | g_header_mode = g_quiet_pp = g_allow_regfunc = 1; |
| 7158 | else |
| 7159 | break; |
| 7160 | } |
| 7161 | |
| 7162 | if (argc < arg + 3) { |
| 7163 | printf("usage:\n%s [-v] [-rf] [-m] <.c> <.asm> <hdr.h> [rlist]*\n" |
| 7164 | "%s -hdr <out.h> <.asm> <seed.h> [rlist]*\n" |
| 7165 | "options:\n" |
| 7166 | " -hdr - header generation mode\n" |
| 7167 | " -rf - allow unannotated indirect calls\n" |
| 7168 | " -m - allow multiple .text sections\n" |
| 7169 | "[rlist] is a file with function names to skip," |
| 7170 | " one per line\n", |
| 7171 | argv[0], argv[0]); |
| 7172 | return 1; |
| 7173 | } |
| 7174 | |
| 7175 | arg_out = arg++; |
| 7176 | |
| 7177 | asmfn = argv[arg++]; |
| 7178 | fasm = fopen(asmfn, "r"); |
| 7179 | my_assert_not(fasm, NULL); |
| 7180 | |
| 7181 | hdrfn = argv[arg++]; |
| 7182 | g_fhdr = fopen(hdrfn, "r"); |
| 7183 | my_assert_not(g_fhdr, NULL); |
| 7184 | |
| 7185 | rlist_alloc = 64; |
| 7186 | rlist = malloc(rlist_alloc * sizeof(rlist[0])); |
| 7187 | my_assert_not(rlist, NULL); |
| 7188 | // needs special handling.. |
| 7189 | rlist[rlist_len++] = "__alloca_probe"; |
| 7190 | |
| 7191 | func_chunk_alloc = 32; |
| 7192 | func_chunks = malloc(func_chunk_alloc * sizeof(func_chunks[0])); |
| 7193 | my_assert_not(func_chunks, NULL); |
| 7194 | |
| 7195 | memset(words, 0, sizeof(words)); |
| 7196 | |
| 7197 | for (; arg < argc; arg++) { |
| 7198 | frlist = fopen(argv[arg], "r"); |
| 7199 | my_assert_not(frlist, NULL); |
| 7200 | |
| 7201 | while (my_fgets(line, sizeof(line), frlist)) { |
| 7202 | p = sskip(line); |
| 7203 | if (*p == 0 || *p == ';') |
| 7204 | continue; |
| 7205 | if (*p == '#') { |
| 7206 | if (IS_START(p, "#if 0") |
| 7207 | || (g_allow_regfunc && IS_START(p, "#if NO_REGFUNC"))) |
| 7208 | { |
| 7209 | skip_func = 1; |
| 7210 | } |
| 7211 | else if (IS_START(p, "#endif")) |
| 7212 | skip_func = 0; |
| 7213 | continue; |
| 7214 | } |
| 7215 | if (skip_func) |
| 7216 | continue; |
| 7217 | |
| 7218 | p = next_word(words[0], sizeof(words[0]), p); |
| 7219 | if (words[0][0] == 0) |
| 7220 | continue; |
| 7221 | |
| 7222 | if (rlist_len >= rlist_alloc) { |
| 7223 | rlist_alloc = rlist_alloc * 2 + 64; |
| 7224 | rlist = realloc(rlist, rlist_alloc * sizeof(rlist[0])); |
| 7225 | my_assert_not(rlist, NULL); |
| 7226 | } |
| 7227 | rlist[rlist_len++] = strdup(words[0]); |
| 7228 | } |
| 7229 | skip_func = 0; |
| 7230 | |
| 7231 | fclose(frlist); |
| 7232 | frlist = NULL; |
| 7233 | } |
| 7234 | |
| 7235 | if (rlist_len > 0) |
| 7236 | qsort(rlist, rlist_len, sizeof(rlist[0]), cmpstringp); |
| 7237 | |
| 7238 | fout = fopen(argv[arg_out], "w"); |
| 7239 | my_assert_not(fout, NULL); |
| 7240 | |
| 7241 | eq_alloc = 128; |
| 7242 | g_eqs = malloc(eq_alloc * sizeof(g_eqs[0])); |
| 7243 | my_assert_not(g_eqs, NULL); |
| 7244 | |
| 7245 | for (i = 0; i < ARRAY_SIZE(g_label_refs); i++) { |
| 7246 | g_label_refs[i].i = -1; |
| 7247 | g_label_refs[i].next = NULL; |
| 7248 | } |
| 7249 | |
| 7250 | if (g_header_mode) |
| 7251 | scan_variables(fasm); |
| 7252 | |
| 7253 | while (my_fgets(line, sizeof(line), fasm)) |
| 7254 | { |
| 7255 | wordc = 0; |
| 7256 | asmln++; |
| 7257 | |
| 7258 | p = sskip(line); |
| 7259 | if (*p == 0) |
| 7260 | continue; |
| 7261 | |
| 7262 | // get rid of random tabs |
| 7263 | for (i = 0; line[i] != 0; i++) |
| 7264 | if (line[i] == '\t') |
| 7265 | line[i] = ' '; |
| 7266 | |
| 7267 | if (*p == ';') |
| 7268 | { |
| 7269 | if (p[2] == '=' && IS_START(p, "; =============== S U B")) |
| 7270 | goto do_pending_endp; // eww.. |
| 7271 | |
| 7272 | if (p[2] == 'A' && IS_START(p, "; Attributes:")) |
| 7273 | { |
| 7274 | static const char *attrs[] = { |
| 7275 | "bp-based frame", |
| 7276 | "library function", |
| 7277 | "static", |
| 7278 | "noreturn", |
| 7279 | "thunk", |
| 7280 | "fpd=", |
| 7281 | }; |
| 7282 | |
| 7283 | // parse IDA's attribute-list comment |
| 7284 | g_ida_func_attr = 0; |
| 7285 | p = sskip(p + 13); |
| 7286 | |
| 7287 | for (; *p != 0; p = sskip(p)) { |
| 7288 | for (i = 0; i < ARRAY_SIZE(attrs); i++) { |
| 7289 | if (!strncmp(p, attrs[i], strlen(attrs[i]))) { |
| 7290 | g_ida_func_attr |= 1 << i; |
| 7291 | p += strlen(attrs[i]); |
| 7292 | break; |
| 7293 | } |
| 7294 | } |
| 7295 | if (i == ARRAY_SIZE(attrs)) { |
| 7296 | anote("unparsed IDA attr: %s\n", p); |
| 7297 | break; |
| 7298 | } |
| 7299 | if (IS(attrs[i], "fpd=")) { |
| 7300 | p = next_word(words[0], sizeof(words[0]), p); |
| 7301 | // ignore for now.. |
| 7302 | } |
| 7303 | } |
| 7304 | } |
| 7305 | else if (p[2] == 'S' && IS_START(p, "; START OF FUNCTION CHUNK FOR ")) |
| 7306 | { |
| 7307 | p += 30; |
| 7308 | next_word(words[0], sizeof(words[0]), p); |
| 7309 | if (words[0][0] == 0) |
| 7310 | aerr("missing name for func chunk?\n"); |
| 7311 | |
| 7312 | if (!scanned_ahead) { |
| 7313 | add_func_chunk(fasm, words[0], asmln); |
| 7314 | func_chunks_sorted = 0; |
| 7315 | } |
| 7316 | } |
| 7317 | else if (p[2] == 'E' && IS_START(p, "; END OF FUNCTION CHUNK")) |
| 7318 | { |
| 7319 | if (func_chunk_i >= 0) { |
| 7320 | if (func_chunk_i < func_chunk_cnt |
| 7321 | && IS(func_chunks[func_chunk_i].name, g_func)) |
| 7322 | { |
| 7323 | // move on to next chunk |
| 7324 | ret = fseek(fasm, func_chunks[func_chunk_i].fptr, SEEK_SET); |
| 7325 | if (ret) |
| 7326 | aerr("seek failed for '%s' chunk #%d\n", |
| 7327 | g_func, func_chunk_i); |
| 7328 | asmln = func_chunks[func_chunk_i].asmln; |
| 7329 | func_chunk_i++; |
| 7330 | } |
| 7331 | else { |
| 7332 | if (func_chunk_ret == 0) |
| 7333 | aerr("no return from chunk?\n"); |
| 7334 | fseek(fasm, func_chunk_ret, SEEK_SET); |
| 7335 | asmln = func_chunk_ret_ln; |
| 7336 | func_chunk_ret = 0; |
| 7337 | pending_endp = 1; |
| 7338 | } |
| 7339 | } |
| 7340 | } |
| 7341 | else if (p[2] == 'F' && IS_START(p, "; FUNCTION CHUNK AT ")) { |
| 7342 | func_chunks_used = 1; |
| 7343 | p += 20; |
| 7344 | if (IS_START(g_func, "sub_")) { |
| 7345 | unsigned long addr = strtoul(p, NULL, 16); |
| 7346 | unsigned long f_addr = strtoul(g_func + 4, NULL, 16); |
| 7347 | if (addr > f_addr && !scanned_ahead) { |
| 7348 | //anote("scan_ahead caused by '%s', addr %lx\n", |
| 7349 | // g_func, addr); |
| 7350 | scan_ahead(fasm); |
| 7351 | scanned_ahead = 1; |
| 7352 | func_chunks_sorted = 0; |
| 7353 | } |
| 7354 | } |
| 7355 | } |
| 7356 | continue; |
| 7357 | } // *p == ';' |
| 7358 | |
| 7359 | parse_words: |
| 7360 | for (i = wordc; i < ARRAY_SIZE(words); i++) |
| 7361 | words[i][0] = 0; |
| 7362 | for (wordc = 0; wordc < ARRAY_SIZE(words); wordc++) { |
| 7363 | p = sskip(next_word_s(words[wordc], sizeof(words[0]), p)); |
| 7364 | if (*p == 0 || *p == ';') { |
| 7365 | wordc++; |
| 7366 | break; |
| 7367 | } |
| 7368 | } |
| 7369 | if (*p != 0 && *p != ';') |
| 7370 | aerr("too many words\n"); |
| 7371 | |
| 7372 | // alow asm patches in comments |
| 7373 | if (*p == ';') { |
| 7374 | if (IS_START(p, "; sctpatch:")) { |
| 7375 | p = sskip(p + 11); |
| 7376 | if (*p == 0 || *p == ';') |
| 7377 | continue; |
| 7378 | goto parse_words; // lame |
| 7379 | } |
| 7380 | if (IS_START(p, "; sctproto:")) { |
| 7381 | sctproto = strdup(p + 11); |
| 7382 | } |
| 7383 | else if (IS_START(p, "; sctend")) { |
| 7384 | end = 1; |
| 7385 | if (!pending_endp) |
| 7386 | break; |
| 7387 | } |
| 7388 | } |
| 7389 | |
| 7390 | if (wordc == 0) { |
| 7391 | // shouldn't happen |
| 7392 | awarn("wordc == 0?\n"); |
| 7393 | continue; |
| 7394 | } |
| 7395 | |
| 7396 | // don't care about this: |
| 7397 | if (words[0][0] == '.' |
| 7398 | || IS(words[0], "include") |
| 7399 | || IS(words[0], "assume") || IS(words[1], "segment") |
| 7400 | || IS(words[0], "align")) |
| 7401 | { |
| 7402 | continue; |
| 7403 | } |
| 7404 | |
| 7405 | do_pending_endp: |
| 7406 | // do delayed endp processing to collect switch jumptables |
| 7407 | if (pending_endp) { |
| 7408 | if (in_func && !g_skip_func && !end && wordc >= 2 |
| 7409 | && ((words[0][0] == 'd' && words[0][2] == 0) |
| 7410 | || (words[1][0] == 'd' && words[1][2] == 0))) |
| 7411 | { |
| 7412 | i = 1; |
| 7413 | if (words[1][0] == 'd' && words[1][2] == 0) { |
| 7414 | // label |
| 7415 | if (g_func_pd_cnt >= pd_alloc) { |
| 7416 | pd_alloc = pd_alloc * 2 + 16; |
| 7417 | g_func_pd = realloc(g_func_pd, |
| 7418 | sizeof(g_func_pd[0]) * pd_alloc); |
| 7419 | my_assert_not(g_func_pd, NULL); |
| 7420 | } |
| 7421 | pd = &g_func_pd[g_func_pd_cnt]; |
| 7422 | g_func_pd_cnt++; |
| 7423 | memset(pd, 0, sizeof(*pd)); |
| 7424 | strcpy(pd->label, words[0]); |
| 7425 | pd->type = OPT_CONST; |
| 7426 | pd->lmod = lmod_from_directive(words[1]); |
| 7427 | i = 2; |
| 7428 | } |
| 7429 | else { |
| 7430 | if (pd == NULL) { |
| 7431 | if (verbose) |
| 7432 | anote("skipping alignment byte?\n"); |
| 7433 | continue; |
| 7434 | } |
| 7435 | lmod = lmod_from_directive(words[0]); |
| 7436 | if (lmod != pd->lmod) |
| 7437 | aerr("lmod change? %d->%d\n", pd->lmod, lmod); |
| 7438 | } |
| 7439 | |
| 7440 | if (pd->count_alloc < pd->count + wordc) { |
| 7441 | pd->count_alloc = pd->count_alloc * 2 + 14 + wordc; |
| 7442 | pd->d = realloc(pd->d, sizeof(pd->d[0]) * pd->count_alloc); |
| 7443 | my_assert_not(pd->d, NULL); |
| 7444 | } |
| 7445 | for (; i < wordc; i++) { |
| 7446 | if (IS(words[i], "offset")) { |
| 7447 | pd->type = OPT_OFFSET; |
| 7448 | i++; |
| 7449 | } |
| 7450 | p = strchr(words[i], ','); |
| 7451 | if (p != NULL) |
| 7452 | *p = 0; |
| 7453 | if (pd->type == OPT_OFFSET) |
| 7454 | pd->d[pd->count].u.label = strdup(words[i]); |
| 7455 | else |
| 7456 | pd->d[pd->count].u.val = parse_number(words[i]); |
| 7457 | pd->d[pd->count].bt_i = -1; |
| 7458 | pd->count++; |
| 7459 | } |
| 7460 | continue; |
| 7461 | } |
| 7462 | |
| 7463 | if (in_func && !g_skip_func) { |
| 7464 | if (g_header_mode) |
| 7465 | gen_hdr(g_func, pi); |
| 7466 | else |
| 7467 | gen_func(fout, g_fhdr, g_func, pi); |
| 7468 | } |
| 7469 | |
| 7470 | pending_endp = 0; |
| 7471 | in_func = 0; |
| 7472 | g_ida_func_attr = 0; |
| 7473 | skip_warned = 0; |
| 7474 | g_skip_func = 0; |
| 7475 | g_func[0] = 0; |
| 7476 | func_chunks_used = 0; |
| 7477 | func_chunk_i = -1; |
| 7478 | if (pi != 0) { |
| 7479 | memset(&ops, 0, pi * sizeof(ops[0])); |
| 7480 | clear_labels(pi); |
| 7481 | pi = 0; |
| 7482 | } |
| 7483 | g_eqcnt = 0; |
| 7484 | for (i = 0; i < g_func_pd_cnt; i++) { |
| 7485 | pd = &g_func_pd[i]; |
| 7486 | if (pd->type == OPT_OFFSET) { |
| 7487 | for (j = 0; j < pd->count; j++) |
| 7488 | free(pd->d[j].u.label); |
| 7489 | } |
| 7490 | free(pd->d); |
| 7491 | pd->d = NULL; |
| 7492 | } |
| 7493 | g_func_pd_cnt = 0; |
| 7494 | pd = NULL; |
| 7495 | |
| 7496 | if (end) |
| 7497 | break; |
| 7498 | if (wordc == 0) |
| 7499 | continue; |
| 7500 | } |
| 7501 | |
| 7502 | if (IS(words[1], "proc")) { |
| 7503 | if (in_func) |
| 7504 | aerr("proc '%s' while in_func '%s'?\n", |
| 7505 | words[0], g_func); |
| 7506 | p = words[0]; |
| 7507 | if (bsearch(&p, rlist, rlist_len, sizeof(rlist[0]), cmpstringp)) |
| 7508 | g_skip_func = 1; |
| 7509 | strcpy(g_func, words[0]); |
| 7510 | set_label(0, words[0]); |
| 7511 | in_func = 1; |
| 7512 | continue; |
| 7513 | } |
| 7514 | |
| 7515 | if (IS(words[1], "endp")) |
| 7516 | { |
| 7517 | if (!in_func) |
| 7518 | aerr("endp '%s' while not in_func?\n", words[0]); |
| 7519 | if (!IS(g_func, words[0])) |
| 7520 | aerr("endp '%s' while in_func '%s'?\n", |
| 7521 | words[0], g_func); |
| 7522 | |
| 7523 | if ((g_ida_func_attr & IDAFA_THUNK) && pi == 1 |
| 7524 | && ops[0].op == OP_JMP && ops[0].operand[0].had_ds) |
| 7525 | { |
| 7526 | // import jump |
| 7527 | g_skip_func = 1; |
| 7528 | } |
| 7529 | |
| 7530 | if (!g_skip_func && func_chunks_used) { |
| 7531 | // start processing chunks |
| 7532 | struct chunk_item *ci, key = { g_func, 0 }; |
| 7533 | |
| 7534 | func_chunk_ret = ftell(fasm); |
| 7535 | func_chunk_ret_ln = asmln; |
| 7536 | if (!func_chunks_sorted) { |
| 7537 | qsort(func_chunks, func_chunk_cnt, |
| 7538 | sizeof(func_chunks[0]), cmp_chunks); |
| 7539 | func_chunks_sorted = 1; |
| 7540 | } |
| 7541 | ci = bsearch(&key, func_chunks, func_chunk_cnt, |
| 7542 | sizeof(func_chunks[0]), cmp_chunks); |
| 7543 | if (ci == NULL) |
| 7544 | aerr("'%s' needs chunks, but none found\n", g_func); |
| 7545 | func_chunk_i = ci - func_chunks; |
| 7546 | for (; func_chunk_i > 0; func_chunk_i--) |
| 7547 | if (!IS(func_chunks[func_chunk_i - 1].name, g_func)) |
| 7548 | break; |
| 7549 | |
| 7550 | ret = fseek(fasm, func_chunks[func_chunk_i].fptr, SEEK_SET); |
| 7551 | if (ret) |
| 7552 | aerr("seek failed for '%s' chunk #%d\n", g_func, func_chunk_i); |
| 7553 | asmln = func_chunks[func_chunk_i].asmln; |
| 7554 | func_chunk_i++; |
| 7555 | continue; |
| 7556 | } |
| 7557 | pending_endp = 1; |
| 7558 | continue; |
| 7559 | } |
| 7560 | |
| 7561 | if (wordc == 2 && IS(words[1], "ends")) { |
| 7562 | if (!multi_seg) { |
| 7563 | end = 1; |
| 7564 | if (pending_endp) |
| 7565 | goto do_pending_endp; |
| 7566 | break; |
| 7567 | } |
| 7568 | |
| 7569 | // scan for next text segment |
| 7570 | while (my_fgets(line, sizeof(line), fasm)) { |
| 7571 | asmln++; |
| 7572 | p = sskip(line); |
| 7573 | if (*p == 0 || *p == ';') |
| 7574 | continue; |
| 7575 | |
| 7576 | if (strstr(p, "segment para public 'CODE' use32")) |
| 7577 | break; |
| 7578 | } |
| 7579 | |
| 7580 | continue; |
| 7581 | } |
| 7582 | |
| 7583 | p = strchr(words[0], ':'); |
| 7584 | if (p != NULL) { |
| 7585 | set_label(pi, words[0]); |
| 7586 | continue; |
| 7587 | } |
| 7588 | |
| 7589 | if (!in_func || g_skip_func) { |
| 7590 | if (!skip_warned && !g_skip_func && g_labels[pi] != NULL) { |
| 7591 | if (verbose) |
| 7592 | anote("skipping from '%s'\n", g_labels[pi]); |
| 7593 | skip_warned = 1; |
| 7594 | } |
| 7595 | free(g_labels[pi]); |
| 7596 | g_labels[pi] = NULL; |
| 7597 | continue; |
| 7598 | } |
| 7599 | |
| 7600 | if (wordc > 1 && IS(words[1], "=")) |
| 7601 | { |
| 7602 | if (wordc != 5) |
| 7603 | aerr("unhandled equ, wc=%d\n", wordc); |
| 7604 | if (g_eqcnt >= eq_alloc) { |
| 7605 | eq_alloc *= 2; |
| 7606 | g_eqs = realloc(g_eqs, eq_alloc * sizeof(g_eqs[0])); |
| 7607 | my_assert_not(g_eqs, NULL); |
| 7608 | } |
| 7609 | |
| 7610 | len = strlen(words[0]); |
| 7611 | if (len > sizeof(g_eqs[0].name) - 1) |
| 7612 | aerr("equ name too long: %d\n", len); |
| 7613 | strcpy(g_eqs[g_eqcnt].name, words[0]); |
| 7614 | |
| 7615 | if (!IS(words[3], "ptr")) |
| 7616 | aerr("unhandled equ\n"); |
| 7617 | if (IS(words[2], "dword")) |
| 7618 | g_eqs[g_eqcnt].lmod = OPLM_DWORD; |
| 7619 | else if (IS(words[2], "word")) |
| 7620 | g_eqs[g_eqcnt].lmod = OPLM_WORD; |
| 7621 | else if (IS(words[2], "byte")) |
| 7622 | g_eqs[g_eqcnt].lmod = OPLM_BYTE; |
| 7623 | else if (IS(words[2], "qword")) |
| 7624 | g_eqs[g_eqcnt].lmod = OPLM_QWORD; |
| 7625 | else |
| 7626 | aerr("bad lmod: '%s'\n", words[2]); |
| 7627 | |
| 7628 | g_eqs[g_eqcnt].offset = parse_number(words[4]); |
| 7629 | g_eqcnt++; |
| 7630 | continue; |
| 7631 | } |
| 7632 | |
| 7633 | if (pi >= ARRAY_SIZE(ops)) |
| 7634 | aerr("too many ops\n"); |
| 7635 | |
| 7636 | parse_op(&ops[pi], words, wordc); |
| 7637 | |
| 7638 | if (sctproto != NULL) { |
| 7639 | if (ops[pi].op == OP_CALL || ops[pi].op == OP_JMP) |
| 7640 | ops[pi].datap = sctproto; |
| 7641 | sctproto = NULL; |
| 7642 | } |
| 7643 | pi++; |
| 7644 | } |
| 7645 | |
| 7646 | if (g_header_mode) |
| 7647 | output_hdr(fout); |
| 7648 | |
| 7649 | fclose(fout); |
| 7650 | fclose(fasm); |
| 7651 | fclose(g_fhdr); |
| 7652 | |
| 7653 | return 0; |
| 7654 | } |
| 7655 | |
| 7656 | // vim:ts=2:shiftwidth=2:expandtab |