5 * This work is licensed under the terms of 3-clause BSD license.
6 * See COPYING file in the top-level directory.
14 #include "my_assert.h"
17 #define ARRAY_SIZE(x) (sizeof(x) / sizeof(x[0]))
18 #define IS(w, y) !strcmp(w, y)
19 #define IS_START(w, y) !strncmp(w, y, strlen(y))
21 #include "protoparse.h"
23 static const char *asmfn;
27 #define anote(fmt, ...) \
28 printf("%s:%d: note: " fmt, asmfn, asmln, ##__VA_ARGS__)
29 #define awarn(fmt, ...) \
30 printf("%s:%d: warning: " fmt, asmfn, asmln, ##__VA_ARGS__)
31 #define aerr(fmt, ...) do { \
32 printf("%s:%d: error: " fmt, asmfn, asmln, ##__VA_ARGS__); \
37 #include "masm_tools.h"
40 OPF_RMD = (1 << 0), /* removed from code generation */
41 OPF_DATA = (1 << 1), /* data processing - writes to dst opr */
42 OPF_FLAGS = (1 << 2), /* sets flags */
43 OPF_JMP = (1 << 3), /* branch, call */
44 OPF_CJMP = (1 << 4), /* cond. branch (cc or jecxz) */
45 OPF_CC = (1 << 5), /* uses flags */
46 OPF_TAIL = (1 << 6), /* ret or tail call */
47 OPF_RSAVE = (1 << 7), /* push/pop is local reg save/load */
48 OPF_REP = (1 << 8), /* prefixed by rep */
49 OPF_REPZ = (1 << 9), /* rep is repe/repz */
50 OPF_REPNZ = (1 << 10), /* rep is repne/repnz */
51 OPF_FARG = (1 << 11), /* push collected as func arg */
52 OPF_FARGNR = (1 << 12), /* push collected as func arg (no reuse) */
53 OPF_EBP_S = (1 << 13), /* ebp used as scratch here, not BP */
54 OPF_DF = (1 << 14), /* DF flag set */
55 OPF_ATAIL = (1 << 15), /* tail call with reused arg frame */
56 OPF_32BIT = (1 << 16), /* 32bit division */
57 OPF_LOCK = (1 << 17), /* op has lock prefix */
58 OPF_VAPUSH = (1 << 18), /* vararg ptr push (as call arg) */
59 OPF_DONE = (1 << 19), /* already fully handled by analysis */
129 // must be sorted (larger len must be further in enum)
138 #define MAX_OPERANDS 3
143 enum opr_lenmod lmod;
144 unsigned int is_ptr:1; // pointer in C
145 unsigned int is_array:1; // array in C
146 unsigned int type_from_var:1; // .. in header, sometimes wrong
147 unsigned int size_mismatch:1; // type override differs from C
148 unsigned int size_lt:1; // type override is larger than C
149 unsigned int had_ds:1; // had ds: prefix
150 const struct parsed_proto *pp; // for OPT_LABEL
158 struct parsed_opr operand[MAX_OPERANDS];
161 unsigned char pfo_inv;
162 unsigned char operand_cnt;
163 unsigned char p_argnum; // arg push: altered before call arg #
164 unsigned char p_arggrp; // arg push: arg group # for above
165 unsigned char p_argpass;// arg push: arg of host func
166 short p_argnext;// arg push: same arg pushed elsewhere or -1
167 int regmask_src; // all referensed regs
169 int pfomask; // flagop: parsed_flag_op that can't be delayed
170 int cc_scratch; // scratch storage during analysis
171 int bt_i; // branch target for branches
172 struct parsed_data *btj;// branch targets for jumptables
173 struct parsed_proto *pp;// parsed_proto for OP_CALL
179 // OP_CALL - parser proto hint (str)
180 // (OPF_CC) - points to one of (OPF_FLAGS) that affects cc op
181 // OP_POP - points to OP_PUSH in push/pop pair
185 enum opr_lenmod lmod;
192 enum opr_lenmod lmod;
206 struct label_ref *next;
210 IDAFA_BP_FRAME = (1 << 0),
211 IDAFA_LIB_FUNC = (1 << 1),
212 IDAFA_STATIC = (1 << 2),
213 IDAFA_NORETURN = (1 << 3),
214 IDAFA_THUNK = (1 << 4),
215 IDAFA_FPD = (1 << 5),
218 // note: limited to 32k due to p_argnext
220 #define MAX_ARG_GRP 2
222 static struct parsed_op ops[MAX_OPS];
223 static struct parsed_equ *g_eqs;
225 static char *g_labels[MAX_OPS];
226 static struct label_ref g_label_refs[MAX_OPS];
227 static const struct parsed_proto *g_func_pp;
228 static struct parsed_data *g_func_pd;
229 static int g_func_pd_cnt;
230 static char g_func[256];
231 static char g_comment[256];
232 static int g_bp_frame;
233 static int g_sp_frame;
234 static int g_stack_frame_used;
235 static int g_stack_fsz;
236 static int g_ida_func_attr;
237 static int g_allow_regfunc;
238 static int g_quiet_pp;
239 static int g_header_mode;
241 #define ferr(op_, fmt, ...) do { \
242 printf("%s:%d: error: [%s] '%s': " fmt, asmfn, (op_)->asmln, g_func, \
243 dump_op(op_), ##__VA_ARGS__); \
247 #define fnote(op_, fmt, ...) \
248 printf("%s:%d: note: [%s] '%s': " fmt, asmfn, (op_)->asmln, g_func, \
249 dump_op(op_), ##__VA_ARGS__)
251 #define ferr_assert(op_, cond) do { \
252 if (!(cond)) ferr(op_, "assertion '%s' failed on ln :%d\n", #cond, \
256 const char *regs_r32[] = {
257 "eax", "ebx", "ecx", "edx", "esi", "edi", "ebp", "esp",
258 // not r32, but list here for easy parsing and printing
259 "mm0", "mm1", "mm2", "mm3", "mm4", "mm5", "mm6", "mm7",
261 const char *regs_r16[] = { "ax", "bx", "cx", "dx", "si", "di", "bp", "sp" };
262 const char *regs_r8l[] = { "al", "bl", "cl", "dl" };
263 const char *regs_r8h[] = { "ah", "bh", "ch", "dh" };
265 enum x86_regs { xUNSPEC = -1, xAX, xBX, xCX, xDX, xSI, xDI, xBP, xSP };
267 // possible basic comparison types (without inversion)
268 enum parsed_flag_op {
272 PFO_BE, // 6 CF=1||ZF=1
276 PFO_LE, // e ZF=1||SF!=OF
279 #define PFOB_O (1 << PFO_O)
280 #define PFOB_C (1 << PFO_C)
281 #define PFOB_Z (1 << PFO_Z)
282 #define PFOB_S (1 << PFO_S)
284 static const char *parsed_flag_op_names[] = {
285 "o", "c", "z", "be", "s", "p", "l", "le"
288 static int char_array_i(const char *array[], size_t len, const char *s)
292 for (i = 0; i < len; i++)
299 static void printf_number(char *buf, size_t buf_size,
300 unsigned long number)
302 // output in C-friendly form
303 snprintf(buf, buf_size, number < 10 ? "%lu" : "0x%02lx", number);
306 static int check_segment_prefix(const char *s)
308 if (s[0] == 0 || s[1] != 's' || s[2] != ':')
322 static int parse_reg(enum opr_lenmod *reg_lmod, const char *s)
326 reg = char_array_i(regs_r32, ARRAY_SIZE(regs_r32), s);
328 *reg_lmod = OPLM_QWORD;
332 *reg_lmod = OPLM_DWORD;
335 reg = char_array_i(regs_r16, ARRAY_SIZE(regs_r16), s);
337 *reg_lmod = OPLM_WORD;
340 reg = char_array_i(regs_r8h, ARRAY_SIZE(regs_r8h), s);
342 *reg_lmod = OPLM_BYTE;
345 reg = char_array_i(regs_r8l, ARRAY_SIZE(regs_r8l), s);
347 *reg_lmod = OPLM_BYTE;
354 static int parse_indmode(char *name, int *regmask, int need_c_cvt)
356 enum opr_lenmod lmod;
369 while (my_isblank(*s))
371 for (; my_issep(*s); d++, s++)
373 while (my_isblank(*s))
377 // skip '?s:' prefixes
378 if (check_segment_prefix(s))
381 s = next_idt(w, sizeof(w), s);
386 reg = parse_reg(&lmod, w);
388 *regmask |= 1 << reg;
392 if ('0' <= w[0] && w[0] <= '9') {
393 number = parse_number(w);
394 printf_number(d, sizeof(cvtbuf) - (d - cvtbuf), number);
398 // probably some label/identifier - pass
401 snprintf(d, sizeof(cvtbuf) - (d - cvtbuf), "%s", w);
405 strcpy(name, cvtbuf);
410 static int is_reg_in_str(const char *s)
414 if (strlen(s) < 3 || (s[3] && !my_issep(s[3]) && !my_isblank(s[3])))
417 for (i = 0; i < ARRAY_SIZE(regs_r32); i++)
418 if (!strncmp(s, regs_r32[i], 3))
424 static const char *parse_stack_el(const char *name, char *extra_reg,
427 const char *p, *p2, *s;
433 if (g_bp_frame || early_try)
436 if (IS_START(p + 3, "+ebp+") && is_reg_in_str(p)) {
438 if (extra_reg != NULL) {
439 strncpy(extra_reg, name, 3);
444 if (IS_START(p, "ebp+")) {
448 if (p2 != NULL && is_reg_in_str(p)) {
449 if (extra_reg != NULL) {
450 strncpy(extra_reg, p, p2 - p);
451 extra_reg[p2 - p] = 0;
456 if (!('0' <= *p && *p <= '9'))
463 if (!IS_START(name, "esp+"))
469 if (is_reg_in_str(s)) {
470 if (extra_reg != NULL) {
471 strncpy(extra_reg, s, p - s);
472 extra_reg[p - s] = 0;
477 aerr("%s IDA stackvar not set?\n", __func__);
479 if (!('0' <= *s && *s <= '9')) {
480 aerr("%s IDA stackvar offset not set?\n", __func__);
483 if (s[0] == '0' && s[1] == 'x')
486 if (len < sizeof(buf) - 1) {
487 strncpy(buf, s, len);
489 val = strtol(buf, &endp, 16);
490 if (val == 0 || *endp != 0) {
491 aerr("%s num parse fail for '%s'\n", __func__, buf);
500 if ('0' <= *p && *p <= '9')
506 static int guess_lmod_from_name(struct parsed_opr *opr)
508 if (!strncmp(opr->name, "dword_", 6)) {
509 opr->lmod = OPLM_DWORD;
512 if (!strncmp(opr->name, "word_", 5)) {
513 opr->lmod = OPLM_WORD;
516 if (!strncmp(opr->name, "byte_", 5)) {
517 opr->lmod = OPLM_BYTE;
520 if (!strncmp(opr->name, "qword_", 6)) {
521 opr->lmod = OPLM_QWORD;
527 static int guess_lmod_from_c_type(enum opr_lenmod *lmod,
528 const struct parsed_type *c_type)
530 static const char *dword_types[] = {
531 "uint32_t", "int", "_DWORD", "UINT_PTR", "DWORD",
532 "WPARAM", "LPARAM", "UINT", "__int32",
533 "LONG", "HIMC", "BOOL", "size_t",
536 static const char *word_types[] = {
537 "uint16_t", "int16_t", "_WORD", "WORD",
538 "unsigned __int16", "__int16",
540 static const char *byte_types[] = {
541 "uint8_t", "int8_t", "char",
542 "unsigned __int8", "__int8", "BYTE", "_BYTE",
544 // structures.. deal the same as with _UNKNOWN for now
550 if (c_type->is_ptr) {
555 n = skip_type_mod(c_type->name);
557 for (i = 0; i < ARRAY_SIZE(dword_types); i++) {
558 if (IS(n, dword_types[i])) {
564 for (i = 0; i < ARRAY_SIZE(word_types); i++) {
565 if (IS(n, word_types[i])) {
571 for (i = 0; i < ARRAY_SIZE(byte_types); i++) {
572 if (IS(n, byte_types[i])) {
581 static char *default_cast_to(char *buf, size_t buf_size,
582 struct parsed_opr *opr)
588 if (opr->pp == NULL || opr->pp->type.name == NULL
591 snprintf(buf, buf_size, "%s", "(void *)");
595 snprintf(buf, buf_size, "(%s)", opr->pp->type.name);
599 static enum opr_type lmod_from_directive(const char *d)
603 else if (IS(d, "dw"))
605 else if (IS(d, "db"))
608 aerr("unhandled directive: '%s'\n", d);
612 static void setup_reg_opr(struct parsed_opr *opr, int reg, enum opr_lenmod lmod,
618 *regmask |= 1 << reg;
621 static struct parsed_equ *equ_find(struct parsed_op *po, const char *name,
624 static int parse_operand(struct parsed_opr *opr,
625 int *regmask, int *regmask_indirect,
626 char words[16][256], int wordc, int w, unsigned int op_flags)
628 const struct parsed_proto *pp = NULL;
629 enum opr_lenmod tmplmod;
630 unsigned long number;
638 aerr("parse_operand w %d, wordc %d\n", w, wordc);
642 for (i = w; i < wordc; i++) {
643 len = strlen(words[i]);
644 if (words[i][len - 1] == ',') {
645 words[i][len - 1] = 0;
651 wordc_in = wordc - w;
653 if ((op_flags & OPF_JMP) && wordc_in > 0
654 && !('0' <= words[w][0] && words[w][0] <= '9'))
656 const char *label = NULL;
658 if (wordc_in == 3 && !strncmp(words[w], "near", 4)
659 && IS(words[w + 1], "ptr"))
660 label = words[w + 2];
661 else if (wordc_in == 2 && IS(words[w], "short"))
662 label = words[w + 1];
663 else if (wordc_in == 1
664 && strchr(words[w], '[') == NULL
665 && parse_reg(&tmplmod, words[w]) < 0)
669 opr->type = OPT_LABEL;
670 ret = check_segment_prefix(label);
673 aerr("fs/gs used\n");
677 strcpy(opr->name, label);
683 if (IS(words[w + 1], "ptr")) {
684 if (IS(words[w], "dword"))
685 opr->lmod = OPLM_DWORD;
686 else if (IS(words[w], "word"))
687 opr->lmod = OPLM_WORD;
688 else if (IS(words[w], "byte"))
689 opr->lmod = OPLM_BYTE;
690 else if (IS(words[w], "qword"))
691 opr->lmod = OPLM_QWORD;
693 aerr("type parsing failed\n");
695 wordc_in = wordc - w;
700 if (IS(words[w], "offset")) {
701 opr->type = OPT_OFFSET;
702 opr->lmod = OPLM_DWORD;
703 strcpy(opr->name, words[w + 1]);
704 pp = proto_parse(g_fhdr, opr->name, 1);
707 if (IS(words[w], "(offset")) {
708 p = strchr(words[w + 1], ')');
710 aerr("parse of bracketed offset failed\n");
712 opr->type = OPT_OFFSET;
713 strcpy(opr->name, words[w + 1]);
719 aerr("parse_operand 1 word expected\n");
721 ret = check_segment_prefix(words[w]);
724 aerr("fs/gs used\n");
726 memmove(words[w], words[w] + 3, strlen(words[w]) - 2);
728 strcpy(opr->name, words[w]);
730 if (words[w][0] == '[') {
731 opr->type = OPT_REGMEM;
732 ret = sscanf(words[w], "[%[^]]]", opr->name);
734 aerr("[] parse failure\n");
736 parse_indmode(opr->name, regmask_indirect, 1);
737 if (opr->lmod == OPLM_UNSPEC && parse_stack_el(opr->name, NULL, 1))
740 struct parsed_equ *eq =
741 equ_find(NULL, parse_stack_el(opr->name, NULL, 1), &i);
743 opr->lmod = eq->lmod;
747 else if (strchr(words[w], '[')) {
749 p = strchr(words[w], '[');
750 opr->type = OPT_REGMEM;
751 parse_indmode(p, regmask_indirect, 0);
752 strncpy(buf, words[w], p - words[w]);
753 buf[p - words[w]] = 0;
754 pp = proto_parse(g_fhdr, buf, 1);
757 else if (('0' <= words[w][0] && words[w][0] <= '9')
758 || words[w][0] == '-')
760 number = parse_number(words[w]);
761 opr->type = OPT_CONST;
763 printf_number(opr->name, sizeof(opr->name), number);
767 ret = parse_reg(&tmplmod, opr->name);
769 setup_reg_opr(opr, ret, tmplmod, regmask);
773 // most likely var in data segment
774 opr->type = OPT_LABEL;
775 pp = proto_parse(g_fhdr, opr->name, g_quiet_pp);
779 if (pp->is_fptr || pp->is_func) {
780 opr->lmod = OPLM_DWORD;
784 tmplmod = OPLM_UNSPEC;
785 if (!guess_lmod_from_c_type(&tmplmod, &pp->type))
786 anote("unhandled C type '%s' for '%s'\n",
787 pp->type.name, opr->name);
789 if (opr->lmod == OPLM_UNSPEC) {
791 opr->type_from_var = 1;
793 else if (opr->lmod != tmplmod) {
794 opr->size_mismatch = 1;
795 if (tmplmod < opr->lmod)
798 opr->is_ptr = pp->type.is_ptr;
800 opr->is_array = pp->type.is_array;
804 if (opr->lmod == OPLM_UNSPEC)
805 guess_lmod_from_name(opr);
809 static const struct {
814 { "repe", OPF_REP|OPF_REPZ },
815 { "repz", OPF_REP|OPF_REPZ },
816 { "repne", OPF_REP|OPF_REPNZ },
817 { "repnz", OPF_REP|OPF_REPNZ },
818 { "lock", OPF_LOCK }, // ignored for now..
821 #define OPF_CJMP_CC (OPF_JMP|OPF_CJMP|OPF_CC)
823 static const struct {
826 unsigned short minopr;
827 unsigned short maxopr;
830 unsigned char pfo_inv;
832 { "nop", OP_NOP, 0, 0, 0 },
833 { "push", OP_PUSH, 1, 1, 0 },
834 { "pop", OP_POP, 1, 1, OPF_DATA },
835 { "leave",OP_LEAVE, 0, 0, OPF_DATA },
836 { "mov" , OP_MOV, 2, 2, OPF_DATA },
837 { "lea", OP_LEA, 2, 2, OPF_DATA },
838 { "movzx",OP_MOVZX, 2, 2, OPF_DATA },
839 { "movsx",OP_MOVSX, 2, 2, OPF_DATA },
840 { "xchg", OP_XCHG, 2, 2, OPF_DATA },
841 { "not", OP_NOT, 1, 1, OPF_DATA },
842 { "cdq", OP_CDQ, 0, 0, OPF_DATA },
843 { "lodsb",OP_LODS, 0, 0, OPF_DATA },
844 { "lodsw",OP_LODS, 0, 0, OPF_DATA },
845 { "lodsd",OP_LODS, 0, 0, OPF_DATA },
846 { "stosb",OP_STOS, 0, 0, OPF_DATA },
847 { "stosw",OP_STOS, 0, 0, OPF_DATA },
848 { "stosd",OP_STOS, 0, 0, OPF_DATA },
849 { "movsb",OP_MOVS, 0, 0, OPF_DATA },
850 { "movsw",OP_MOVS, 0, 0, OPF_DATA },
851 { "movsd",OP_MOVS, 0, 0, OPF_DATA },
852 { "cmpsb",OP_CMPS, 0, 0, OPF_DATA|OPF_FLAGS },
853 { "cmpsw",OP_CMPS, 0, 0, OPF_DATA|OPF_FLAGS },
854 { "cmpsd",OP_CMPS, 0, 0, OPF_DATA|OPF_FLAGS },
855 { "scasb",OP_SCAS, 0, 0, OPF_DATA|OPF_FLAGS },
856 { "scasw",OP_SCAS, 0, 0, OPF_DATA|OPF_FLAGS },
857 { "scasd",OP_SCAS, 0, 0, OPF_DATA|OPF_FLAGS },
858 { "std", OP_STD, 0, 0, OPF_DATA }, // special flag
859 { "cld", OP_CLD, 0, 0, OPF_DATA },
860 { "add", OP_ADD, 2, 2, OPF_DATA|OPF_FLAGS },
861 { "sub", OP_SUB, 2, 2, OPF_DATA|OPF_FLAGS },
862 { "and", OP_AND, 2, 2, OPF_DATA|OPF_FLAGS },
863 { "or", OP_OR, 2, 2, OPF_DATA|OPF_FLAGS },
864 { "xor", OP_XOR, 2, 2, OPF_DATA|OPF_FLAGS },
865 { "shl", OP_SHL, 2, 2, OPF_DATA|OPF_FLAGS },
866 { "shr", OP_SHR, 2, 2, OPF_DATA|OPF_FLAGS },
867 { "sal", OP_SHL, 2, 2, OPF_DATA|OPF_FLAGS },
868 { "sar", OP_SAR, 2, 2, OPF_DATA|OPF_FLAGS },
869 { "shrd", OP_SHRD, 3, 3, OPF_DATA|OPF_FLAGS },
870 { "rol", OP_ROL, 2, 2, OPF_DATA|OPF_FLAGS },
871 { "ror", OP_ROR, 2, 2, OPF_DATA|OPF_FLAGS },
872 { "rcl", OP_RCL, 2, 2, OPF_DATA|OPF_FLAGS|OPF_CC, PFO_C },
873 { "rcr", OP_RCR, 2, 2, OPF_DATA|OPF_FLAGS|OPF_CC, PFO_C },
874 { "adc", OP_ADC, 2, 2, OPF_DATA|OPF_FLAGS|OPF_CC, PFO_C },
875 { "sbb", OP_SBB, 2, 2, OPF_DATA|OPF_FLAGS|OPF_CC, PFO_C },
876 { "bsf", OP_BSF, 2, 2, OPF_DATA|OPF_FLAGS },
877 { "inc", OP_INC, 1, 1, OPF_DATA|OPF_FLAGS },
878 { "dec", OP_DEC, 1, 1, OPF_DATA|OPF_FLAGS },
879 { "neg", OP_NEG, 1, 1, OPF_DATA|OPF_FLAGS },
880 { "mul", OP_MUL, 1, 1, OPF_DATA|OPF_FLAGS },
881 { "imul", OP_IMUL, 1, 3, OPF_DATA|OPF_FLAGS },
882 { "div", OP_DIV, 1, 1, OPF_DATA|OPF_FLAGS },
883 { "idiv", OP_IDIV, 1, 1, OPF_DATA|OPF_FLAGS },
884 { "test", OP_TEST, 2, 2, OPF_FLAGS },
885 { "cmp", OP_CMP, 2, 2, OPF_FLAGS },
886 { "retn", OP_RET, 0, 1, OPF_TAIL },
887 { "call", OP_CALL, 1, 1, OPF_JMP|OPF_DATA|OPF_FLAGS },
888 { "jmp", OP_JMP, 1, 1, OPF_JMP },
889 { "jecxz",OP_JECXZ, 1, 1, OPF_JMP|OPF_CJMP },
890 { "jo", OP_JCC, 1, 1, OPF_CJMP_CC, PFO_O, 0 }, // 70 OF=1
891 { "jno", OP_JCC, 1, 1, OPF_CJMP_CC, PFO_O, 1 }, // 71 OF=0
892 { "jc", OP_JCC, 1, 1, OPF_CJMP_CC, PFO_C, 0 }, // 72 CF=1
893 { "jb", OP_JCC, 1, 1, OPF_CJMP_CC, PFO_C, 0 }, // 72
894 { "jnc", OP_JCC, 1, 1, OPF_CJMP_CC, PFO_C, 1 }, // 73 CF=0
895 { "jnb", OP_JCC, 1, 1, OPF_CJMP_CC, PFO_C, 1 }, // 73
896 { "jae", OP_JCC, 1, 1, OPF_CJMP_CC, PFO_C, 1 }, // 73
897 { "jz", OP_JCC, 1, 1, OPF_CJMP_CC, PFO_Z, 0 }, // 74 ZF=1
898 { "je", OP_JCC, 1, 1, OPF_CJMP_CC, PFO_Z, 0 }, // 74
899 { "jnz", OP_JCC, 1, 1, OPF_CJMP_CC, PFO_Z, 1 }, // 75 ZF=0
900 { "jne", OP_JCC, 1, 1, OPF_CJMP_CC, PFO_Z, 1 }, // 75
901 { "jbe", OP_JCC, 1, 1, OPF_CJMP_CC, PFO_BE, 0 }, // 76 CF=1||ZF=1
902 { "jna", OP_JCC, 1, 1, OPF_CJMP_CC, PFO_BE, 0 }, // 76
903 { "ja", OP_JCC, 1, 1, OPF_CJMP_CC, PFO_BE, 1 }, // 77 CF=0&&ZF=0
904 { "jnbe", OP_JCC, 1, 1, OPF_CJMP_CC, PFO_BE, 1 }, // 77
905 { "js", OP_JCC, 1, 1, OPF_CJMP_CC, PFO_S, 0 }, // 78 SF=1
906 { "jns", OP_JCC, 1, 1, OPF_CJMP_CC, PFO_S, 1 }, // 79 SF=0
907 { "jp", OP_JCC, 1, 1, OPF_CJMP_CC, PFO_P, 0 }, // 7a PF=1
908 { "jpe", OP_JCC, 1, 1, OPF_CJMP_CC, PFO_P, 0 }, // 7a
909 { "jnp", OP_JCC, 1, 1, OPF_CJMP_CC, PFO_P, 1 }, // 7b PF=0
910 { "jpo", OP_JCC, 1, 1, OPF_CJMP_CC, PFO_P, 1 }, // 7b
911 { "jl", OP_JCC, 1, 1, OPF_CJMP_CC, PFO_L, 0 }, // 7c SF!=OF
912 { "jnge", OP_JCC, 1, 1, OPF_CJMP_CC, PFO_L, 0 }, // 7c
913 { "jge", OP_JCC, 1, 1, OPF_CJMP_CC, PFO_L, 1 }, // 7d SF=OF
914 { "jnl", OP_JCC, 1, 1, OPF_CJMP_CC, PFO_L, 1 }, // 7d
915 { "jle", OP_JCC, 1, 1, OPF_CJMP_CC, PFO_LE, 0 }, // 7e ZF=1||SF!=OF
916 { "jng", OP_JCC, 1, 1, OPF_CJMP_CC, PFO_LE, 0 }, // 7e
917 { "jg", OP_JCC, 1, 1, OPF_CJMP_CC, PFO_LE, 1 }, // 7f ZF=0&&SF=OF
918 { "jnle", OP_JCC, 1, 1, OPF_CJMP_CC, PFO_LE, 1 }, // 7f
919 { "seto", OP_SCC, 1, 1, OPF_DATA|OPF_CC, PFO_O, 0 },
920 { "setno", OP_SCC, 1, 1, OPF_DATA|OPF_CC, PFO_O, 1 },
921 { "setc", OP_SCC, 1, 1, OPF_DATA|OPF_CC, PFO_C, 0 },
922 { "setb", OP_SCC, 1, 1, OPF_DATA|OPF_CC, PFO_C, 0 },
923 { "setnc", OP_SCC, 1, 1, OPF_DATA|OPF_CC, PFO_C, 1 },
924 { "setae", OP_SCC, 1, 1, OPF_DATA|OPF_CC, PFO_C, 1 },
925 { "setnb", OP_SCC, 1, 1, OPF_DATA|OPF_CC, PFO_C, 1 },
926 { "setz", OP_SCC, 1, 1, OPF_DATA|OPF_CC, PFO_Z, 0 },
927 { "sete", OP_SCC, 1, 1, OPF_DATA|OPF_CC, PFO_Z, 0 },
928 { "setnz", OP_SCC, 1, 1, OPF_DATA|OPF_CC, PFO_Z, 1 },
929 { "setne", OP_SCC, 1, 1, OPF_DATA|OPF_CC, PFO_Z, 1 },
930 { "setbe", OP_SCC, 1, 1, OPF_DATA|OPF_CC, PFO_BE, 0 },
931 { "setna", OP_SCC, 1, 1, OPF_DATA|OPF_CC, PFO_BE, 0 },
932 { "seta", OP_SCC, 1, 1, OPF_DATA|OPF_CC, PFO_BE, 1 },
933 { "setnbe", OP_SCC, 1, 1, OPF_DATA|OPF_CC, PFO_BE, 1 },
934 { "sets", OP_SCC, 1, 1, OPF_DATA|OPF_CC, PFO_S, 0 },
935 { "setns", OP_SCC, 1, 1, OPF_DATA|OPF_CC, PFO_S, 1 },
936 { "setp", OP_SCC, 1, 1, OPF_DATA|OPF_CC, PFO_P, 0 },
937 { "setpe", OP_SCC, 1, 1, OPF_DATA|OPF_CC, PFO_P, 0 },
938 { "setnp", OP_SCC, 1, 1, OPF_DATA|OPF_CC, PFO_P, 1 },
939 { "setpo", OP_SCC, 1, 1, OPF_DATA|OPF_CC, PFO_P, 1 },
940 { "setl", OP_SCC, 1, 1, OPF_DATA|OPF_CC, PFO_L, 0 },
941 { "setnge", OP_SCC, 1, 1, OPF_DATA|OPF_CC, PFO_L, 0 },
942 { "setge", OP_SCC, 1, 1, OPF_DATA|OPF_CC, PFO_L, 1 },
943 { "setnl", OP_SCC, 1, 1, OPF_DATA|OPF_CC, PFO_L, 1 },
944 { "setle", OP_SCC, 1, 1, OPF_DATA|OPF_CC, PFO_LE, 0 },
945 { "setng", OP_SCC, 1, 1, OPF_DATA|OPF_CC, PFO_LE, 0 },
946 { "setg", OP_SCC, 1, 1, OPF_DATA|OPF_CC, PFO_LE, 1 },
947 { "setnle", OP_SCC, 1, 1, OPF_DATA|OPF_CC, PFO_LE, 1 },
950 { "emms", OP_EMMS, 0, 0, OPF_DATA },
951 { "movq", OP_MOV, 2, 2, OPF_DATA },
956 static void parse_op(struct parsed_op *op, char words[16][256], int wordc)
958 enum opr_lenmod lmod = OPLM_UNSPEC;
959 int prefix_flags = 0;
967 for (i = 0; i < ARRAY_SIZE(pref_table); i++) {
968 if (IS(words[w], pref_table[i].name)) {
969 prefix_flags = pref_table[i].flags;
976 aerr("lone prefix: '%s'\n", words[0]);
981 for (i = 0; i < ARRAY_SIZE(op_table); i++) {
982 if (IS(words[w], op_table[i].name))
986 if (i == ARRAY_SIZE(op_table)) {
987 anote("unhandled op: '%s'\n", words[0]);
992 op->op = op_table[i].op;
993 op->flags = op_table[i].flags | prefix_flags;
994 op->pfo = op_table[i].pfo;
995 op->pfo_inv = op_table[i].pfo_inv;
996 op->regmask_src = op->regmask_dst = 0;
999 if (op->op == OP_UD2)
1002 for (opr = 0; opr < op_table[i].maxopr; opr++) {
1003 if (opr >= op_table[i].minopr && w >= wordc)
1006 regmask = regmask_ind = 0;
1007 w = parse_operand(&op->operand[opr], ®mask, ®mask_ind,
1008 words, wordc, w, op->flags);
1010 if (opr == 0 && (op->flags & OPF_DATA))
1011 op->regmask_dst = regmask;
1013 op->regmask_src |= regmask;
1014 op->regmask_src |= regmask_ind;
1018 aerr("parse_op %s incomplete: %d/%d\n",
1019 words[0], w, wordc);
1022 op->operand_cnt = opr;
1023 if (!strncmp(op_table[i].name, "set", 3))
1024 op->operand[0].lmod = OPLM_BYTE;
1027 // first operand is not dst
1030 op->regmask_src |= op->regmask_dst;
1031 op->regmask_dst = 0;
1034 // first operand is src too
1046 op->regmask_src |= op->regmask_dst;
1051 op->regmask_src |= op->regmask_dst;
1052 op->regmask_dst |= op->regmask_src;
1058 if (op->operand[0].type == OPT_REG && op->operand[1].type == OPT_REG
1059 && op->operand[0].lmod == op->operand[1].lmod
1060 && op->operand[0].reg == op->operand[1].reg
1061 && IS(op->operand[0].name, op->operand[1].name)) // ! ah, al..
1063 op->regmask_src = 0;
1066 op->regmask_src |= op->regmask_dst;
1069 // ops with implicit argumets
1071 op->operand_cnt = 2;
1072 setup_reg_opr(&op->operand[0], xDX, OPLM_DWORD, &op->regmask_dst);
1073 setup_reg_opr(&op->operand[1], xAX, OPLM_DWORD, &op->regmask_src);
1079 if (op->operand_cnt != 0)
1081 if (words[op_w][4] == 'b')
1083 else if (words[op_w][4] == 'w')
1085 else if (words[op_w][4] == 'd')
1087 op->operand_cnt = 3;
1088 setup_reg_opr(&op->operand[0], op->op == OP_LODS ? xSI : xDI,
1089 lmod, &op->regmask_src);
1090 setup_reg_opr(&op->operand[1], xCX, OPLM_DWORD, &op->regmask_src);
1091 op->regmask_dst = op->regmask_src;
1092 setup_reg_opr(&op->operand[2], xAX, OPLM_DWORD,
1093 op->op == OP_LODS ? &op->regmask_dst : &op->regmask_src);
1098 if (op->operand_cnt != 0)
1100 if (words[op_w][4] == 'b')
1102 else if (words[op_w][4] == 'w')
1104 else if (words[op_w][4] == 'd')
1106 op->operand_cnt = 3;
1107 setup_reg_opr(&op->operand[0], xDI, lmod, &op->regmask_src);
1108 setup_reg_opr(&op->operand[1], xSI, OPLM_DWORD, &op->regmask_src);
1109 setup_reg_opr(&op->operand[2], xCX, OPLM_DWORD, &op->regmask_src);
1110 op->regmask_dst = op->regmask_src;
1114 op->operand_cnt = 1;
1115 op->regmask_src = 1 << xCX;
1116 op->operand[0].type = OPT_REG;
1117 op->operand[0].reg = xCX;
1118 op->operand[0].lmod = OPLM_DWORD;
1122 if (op->operand_cnt != 1)
1127 op->regmask_src |= op->regmask_dst;
1128 op->regmask_dst = (1 << xDX) | (1 << xAX);
1129 if (op->operand[0].lmod == OPLM_UNSPEC)
1130 op->operand[0].lmod = OPLM_DWORD;
1135 // we could set up operands for edx:eax, but there is no real need to
1136 // (see is_opr_modified())
1137 op->regmask_src |= op->regmask_dst;
1138 op->regmask_dst = (1 << xDX) | (1 << xAX);
1139 if (op->operand[0].lmod == OPLM_UNSPEC)
1140 op->operand[0].lmod = OPLM_DWORD;
1148 op->regmask_src |= op->regmask_dst;
1149 if (op->operand[1].lmod == OPLM_UNSPEC)
1150 op->operand[1].lmod = OPLM_BYTE;
1154 op->regmask_src |= op->regmask_dst;
1155 if (op->operand[2].lmod == OPLM_UNSPEC)
1156 op->operand[2].lmod = OPLM_BYTE;
1160 op->regmask_src |= op->regmask_dst;
1161 op->regmask_dst = 0;
1162 if (op->operand[0].lmod == OPLM_UNSPEC
1163 && (op->operand[0].type == OPT_CONST
1164 || op->operand[0].type == OPT_OFFSET
1165 || op->operand[0].type == OPT_LABEL))
1166 op->operand[0].lmod = OPLM_DWORD;
1172 if (op->operand[0].type == OPT_REG && op->operand[1].type == OPT_REG
1173 && op->operand[0].lmod == op->operand[1].lmod
1174 && op->operand[0].reg == op->operand[1].reg
1175 && IS(op->operand[0].name, op->operand[1].name)) // ! ah, al..
1177 op->flags |= OPF_RMD | OPF_DONE;
1178 op->regmask_src = op->regmask_dst = 0;
1183 if (op->operand[0].type == OPT_REG
1184 && op->operand[1].type == OPT_REGMEM)
1187 snprintf(buf, sizeof(buf), "%s+0", op->operand[0].name);
1188 if (IS(buf, op->operand[1].name))
1189 op->flags |= OPF_RMD | OPF_DONE;
1194 // trashed regs must be explicitly detected later
1195 op->regmask_dst = 0;
1199 op->regmask_dst = (1 << xBP) | (1 << xSP);
1200 op->regmask_src = 1 << xBP;
1208 static const char *op_name(struct parsed_op *po)
1210 static char buf[16];
1214 if (po->op == OP_JCC || po->op == OP_SCC) {
1216 *p++ = (po->op == OP_JCC) ? 'j' : 's';
1219 strcpy(p, parsed_flag_op_names[po->pfo]);
1223 for (i = 0; i < ARRAY_SIZE(op_table); i++)
1224 if (op_table[i].op == po->op)
1225 return op_table[i].name;
1231 static const char *dump_op(struct parsed_op *po)
1233 static char out[128];
1240 snprintf(out, sizeof(out), "%s", op_name(po));
1241 for (i = 0; i < po->operand_cnt; i++) {
1245 snprintf(p, sizeof(out) - (p - out),
1246 po->operand[i].type == OPT_REGMEM ? " [%s]" : " %s",
1247 po->operand[i].name);
1253 static const char *lmod_type_u(struct parsed_op *po,
1254 enum opr_lenmod lmod)
1266 ferr(po, "invalid lmod: %d\n", lmod);
1267 return "(_invalid_)";
1271 static const char *lmod_cast_u(struct parsed_op *po,
1272 enum opr_lenmod lmod)
1284 ferr(po, "invalid lmod: %d\n", lmod);
1285 return "(_invalid_)";
1289 static const char *lmod_cast_u_ptr(struct parsed_op *po,
1290 enum opr_lenmod lmod)
1302 ferr(po, "invalid lmod: %d\n", lmod);
1303 return "(_invalid_)";
1307 static const char *lmod_cast_s(struct parsed_op *po,
1308 enum opr_lenmod lmod)
1320 ferr(po, "%s: invalid lmod: %d\n", __func__, lmod);
1321 return "(_invalid_)";
1325 static const char *lmod_cast(struct parsed_op *po,
1326 enum opr_lenmod lmod, int is_signed)
1329 lmod_cast_s(po, lmod) :
1330 lmod_cast_u(po, lmod);
1333 static int lmod_bytes(struct parsed_op *po, enum opr_lenmod lmod)
1345 ferr(po, "%s: invalid lmod: %d\n", __func__, lmod);
1350 static const char *opr_name(struct parsed_op *po, int opr_num)
1352 if (opr_num >= po->operand_cnt)
1353 ferr(po, "opr OOR: %d/%d\n", opr_num, po->operand_cnt);
1354 return po->operand[opr_num].name;
1357 static unsigned int opr_const(struct parsed_op *po, int opr_num)
1359 if (opr_num >= po->operand_cnt)
1360 ferr(po, "opr OOR: %d/%d\n", opr_num, po->operand_cnt);
1361 if (po->operand[opr_num].type != OPT_CONST)
1362 ferr(po, "opr %d: const expected\n", opr_num);
1363 return po->operand[opr_num].val;
1366 static const char *opr_reg_p(struct parsed_op *po, struct parsed_opr *popr)
1368 if ((unsigned int)popr->reg >= ARRAY_SIZE(regs_r32))
1369 ferr(po, "invalid reg: %d\n", popr->reg);
1370 return regs_r32[popr->reg];
1373 // cast1 is the "final" cast
1374 static const char *simplify_cast(const char *cast1, const char *cast2)
1376 static char buf[256];
1382 if (IS(cast1, cast2))
1384 if (IS(cast1, "(s8)") && IS(cast2, "(u8)"))
1386 if (IS(cast1, "(s16)") && IS(cast2, "(u16)"))
1388 if (IS(cast1, "(u8)") && IS_START(cast2, "*(u8 *)"))
1390 if (IS(cast1, "(u16)") && IS_START(cast2, "*(u16 *)"))
1392 if (strchr(cast1, '*') && IS_START(cast2, "(u32)"))
1395 snprintf(buf, sizeof(buf), "%s%s", cast1, cast2);
1399 static const char *simplify_cast_num(const char *cast, unsigned int val)
1401 if (IS(cast, "(u8)") && val < 0x100)
1403 if (IS(cast, "(s8)") && val < 0x80)
1405 if (IS(cast, "(u16)") && val < 0x10000)
1407 if (IS(cast, "(s16)") && val < 0x8000)
1409 if (IS(cast, "(s32)") && val < 0x80000000)
1415 static struct parsed_equ *equ_find(struct parsed_op *po, const char *name,
1424 namelen = strlen(name);
1426 p = strchr(name, '+');
1430 ferr(po, "equ parse failed for '%s'\n", name);
1432 if (IS_START(p, "0x"))
1434 *extra_offs = strtol(p, &endp, 16);
1436 ferr(po, "equ parse failed for '%s'\n", name);
1439 for (i = 0; i < g_eqcnt; i++)
1440 if (strncmp(g_eqs[i].name, name, namelen) == 0
1441 && g_eqs[i].name[namelen] == 0)
1445 ferr(po, "unresolved equ name: '%s'\n", name);
1452 static int is_stack_access(struct parsed_op *po,
1453 const struct parsed_opr *popr)
1455 return (parse_stack_el(popr->name, NULL, 0)
1456 || (g_bp_frame && !(po->flags & OPF_EBP_S)
1457 && IS_START(popr->name, "ebp")));
1460 static void parse_stack_access(struct parsed_op *po,
1461 const char *name, char *ofs_reg, int *offset_out,
1462 int *stack_ra_out, const char **bp_arg_out, int is_lea)
1464 const char *bp_arg = "";
1465 const char *p = NULL;
1466 struct parsed_equ *eq;
1473 if (IS_START(name, "ebp-")
1474 || (IS_START(name, "ebp+") && '0' <= name[4] && name[4] <= '9'))
1477 if (IS_START(p, "0x"))
1479 offset = strtoul(p, &endp, 16);
1483 ferr(po, "ebp- parse of '%s' failed\n", name);
1486 bp_arg = parse_stack_el(name, ofs_reg, 0);
1487 snprintf(g_comment, sizeof(g_comment), "%s", bp_arg);
1488 eq = equ_find(po, bp_arg, &offset);
1490 ferr(po, "detected but missing eq\n");
1491 offset += eq->offset;
1494 if (!strncmp(name, "ebp", 3))
1497 // yes it sometimes LEAs ra for compares..
1498 if (!is_lea && ofs_reg[0] == 0
1499 && stack_ra <= offset && offset < stack_ra + 4)
1501 ferr(po, "reference to ra? %d %d\n", offset, stack_ra);
1504 *offset_out = offset;
1505 *stack_ra_out = stack_ra;
1507 *bp_arg_out = bp_arg;
1510 static int stack_frame_access(struct parsed_op *po,
1511 struct parsed_opr *popr, char *buf, size_t buf_size,
1512 const char *name, const char *cast, int is_src, int is_lea)
1514 enum opr_lenmod tmp_lmod = OPLM_UNSPEC;
1515 const char *prefix = "";
1516 const char *bp_arg = NULL;
1517 char ofs_reg[16] = { 0, };
1518 int i, arg_i, arg_s;
1526 if (po->flags & OPF_EBP_S)
1527 ferr(po, "stack_frame_access while ebp is scratch\n");
1529 parse_stack_access(po, name, ofs_reg, &offset,
1530 &stack_ra, &bp_arg, is_lea);
1532 if (offset > stack_ra)
1534 arg_i = (offset - stack_ra - 4) / 4;
1535 if (arg_i < 0 || arg_i >= g_func_pp->argc_stack)
1537 if (g_func_pp->is_vararg
1538 && arg_i == g_func_pp->argc_stack && is_lea)
1540 // should be va_list
1543 snprintf(buf, buf_size, "%sap", cast);
1546 ferr(po, "offset %d (%s,%d) doesn't map to any arg\n",
1547 offset, bp_arg, arg_i);
1549 if (ofs_reg[0] != 0)
1550 ferr(po, "offset reg on arg access?\n");
1552 for (i = arg_s = 0; i < g_func_pp->argc; i++) {
1553 if (g_func_pp->arg[i].reg != NULL)
1559 if (i == g_func_pp->argc)
1560 ferr(po, "arg %d not in prototype?\n", arg_i);
1562 popr->is_ptr = g_func_pp->arg[i].type.is_ptr;
1569 ferr(po, "lea/byte to arg?\n");
1570 if (is_src && (offset & 3) == 0)
1571 snprintf(buf, buf_size, "%sa%d",
1572 simplify_cast(cast, "(u8)"), i + 1);
1574 snprintf(buf, buf_size, "%sBYTE%d(a%d)",
1575 cast, offset & 3, i + 1);
1580 ferr(po, "lea/word to arg?\n");
1585 ferr(po, "problematic arg store\n");
1586 snprintf(buf, buf_size, "%s((char *)&a%d + 1)",
1587 simplify_cast(cast, "*(u16 *)"), i + 1);
1590 ferr(po, "unaligned arg word load\n");
1592 else if (is_src && (offset & 2) == 0)
1593 snprintf(buf, buf_size, "%sa%d",
1594 simplify_cast(cast, "(u16)"), i + 1);
1596 snprintf(buf, buf_size, "%s%sWORD(a%d)",
1597 cast, (offset & 2) ? "HI" : "LO", i + 1);
1609 snprintf(buf, buf_size, "(u32)&a%d + %d",
1612 ferr(po, "unaligned arg store\n");
1614 // mov edx, [ebp+arg_4+2]; movsx ecx, dx
1615 snprintf(buf, buf_size, "%s(a%d >> %d)",
1616 prefix, i + 1, (offset & 3) * 8);
1620 snprintf(buf, buf_size, "%s%sa%d",
1621 prefix, is_lea ? "&" : "", i + 1);
1626 ferr(po, "bp_arg bad lmod: %d\n", popr->lmod);
1630 snprintf(g_comment, sizeof(g_comment), "%s unaligned", bp_arg);
1633 guess_lmod_from_c_type(&tmp_lmod, &g_func_pp->arg[i].type);
1634 if (tmp_lmod != OPLM_DWORD
1635 && (unaligned || (!is_src && lmod_bytes(po, tmp_lmod)
1636 < lmod_bytes(po, popr->lmod) + (offset & 3))))
1638 ferr(po, "bp_arg arg%d/w offset %d and type '%s' is too small\n",
1639 i + 1, offset, g_func_pp->arg[i].type.name);
1641 // can't check this because msvc likes to reuse
1642 // arg space for scratch..
1643 //if (popr->is_ptr && popr->lmod != OPLM_DWORD)
1644 // ferr(po, "bp_arg arg%d: non-dword ptr access\n", i + 1);
1648 if (g_stack_fsz == 0)
1649 ferr(po, "stack var access without stackframe\n");
1650 g_stack_frame_used = 1;
1652 sf_ofs = g_stack_fsz + offset;
1653 lim = (ofs_reg[0] != 0) ? -4 : 0;
1654 if (offset > 0 || sf_ofs < lim)
1655 ferr(po, "bp_stack offset %d/%d\n", offset, g_stack_fsz);
1665 snprintf(buf, buf_size, "%ssf.b[%d%s%s]",
1666 prefix, sf_ofs, ofs_reg[0] ? "+" : "", ofs_reg);
1670 if ((sf_ofs & 1) || ofs_reg[0] != 0) {
1671 // known unaligned or possibly unaligned
1672 strcat(g_comment, " unaligned");
1674 prefix = "*(u16 *)&";
1675 snprintf(buf, buf_size, "%ssf.b[%d%s%s]",
1676 prefix, sf_ofs, ofs_reg[0] ? "+" : "", ofs_reg);
1679 snprintf(buf, buf_size, "%ssf.w[%d]", prefix, sf_ofs / 2);
1683 if ((sf_ofs & 3) || ofs_reg[0] != 0) {
1684 // known unaligned or possibly unaligned
1685 strcat(g_comment, " unaligned");
1687 prefix = "*(u32 *)&";
1688 snprintf(buf, buf_size, "%ssf.b[%d%s%s]",
1689 prefix, sf_ofs, ofs_reg[0] ? "+" : "", ofs_reg);
1692 snprintf(buf, buf_size, "%ssf.d[%d]", prefix, sf_ofs / 4);
1696 ferr(po, "bp_stack bad lmod: %d\n", popr->lmod);
1703 static void check_func_pp(struct parsed_op *po,
1704 const struct parsed_proto *pp, const char *pfx)
1706 enum opr_lenmod tmp_lmod;
1710 if (pp->argc_reg != 0) {
1711 if (/*!g_allow_regfunc &&*/ !pp->is_fastcall) {
1712 pp_print(buf, sizeof(buf), pp);
1713 ferr(po, "%s: unexpected reg arg in icall: %s\n", pfx, buf);
1715 if (pp->argc_stack > 0 && pp->argc_reg != 2)
1716 ferr(po, "%s: %d reg arg(s) with %d stack arg(s)\n",
1717 pfx, pp->argc_reg, pp->argc_stack);
1720 // fptrs must use 32bit args, callsite might have no information and
1721 // lack a cast to smaller types, which results in incorrectly masked
1722 // args passed (callee may assume masked args, it does on ARM)
1723 if (!pp->is_osinc) {
1724 for (i = 0; i < pp->argc; i++) {
1725 ret = guess_lmod_from_c_type(&tmp_lmod, &pp->arg[i].type);
1726 if (ret && tmp_lmod != OPLM_DWORD)
1727 ferr(po, "reference to %s with arg%d '%s'\n", pp->name,
1728 i + 1, pp->arg[i].type.name);
1733 static const char *check_label_read_ref(struct parsed_op *po,
1736 const struct parsed_proto *pp;
1738 pp = proto_parse(g_fhdr, name, 0);
1740 ferr(po, "proto_parse failed for ref '%s'\n", name);
1743 check_func_pp(po, pp, "ref");
1748 static char *out_src_opr(char *buf, size_t buf_size,
1749 struct parsed_op *po, struct parsed_opr *popr, const char *cast,
1752 char tmp1[256], tmp2[256];
1761 switch (popr->type) {
1764 ferr(po, "lea from reg?\n");
1766 switch (popr->lmod) {
1768 snprintf(buf, buf_size, "%s%s.q", cast, opr_reg_p(po, popr));
1771 snprintf(buf, buf_size, "%s%s", cast, opr_reg_p(po, popr));
1774 snprintf(buf, buf_size, "%s%s",
1775 simplify_cast(cast, "(u16)"), opr_reg_p(po, popr));
1778 if (popr->name[1] == 'h') // XXX..
1779 snprintf(buf, buf_size, "%s(%s >> 8)",
1780 simplify_cast(cast, "(u8)"), opr_reg_p(po, popr));
1782 snprintf(buf, buf_size, "%s%s",
1783 simplify_cast(cast, "(u8)"), opr_reg_p(po, popr));
1786 ferr(po, "invalid src lmod: %d\n", popr->lmod);
1791 if (is_stack_access(po, popr)) {
1792 stack_frame_access(po, popr, buf, buf_size,
1793 popr->name, cast, 1, is_lea);
1797 strcpy(expr, popr->name);
1798 if (strchr(expr, '[')) {
1799 // special case: '[' can only be left for label[reg] form
1800 ret = sscanf(expr, "%[^[][%[^]]]", tmp1, tmp2);
1802 ferr(po, "parse failure for '%s'\n", expr);
1803 if (tmp1[0] == '(') {
1804 // (off_4FFF50+3)[eax]
1805 p = strchr(tmp1 + 1, ')');
1806 if (p == NULL || p[1] != 0)
1807 ferr(po, "parse failure (2) for '%s'\n", expr);
1809 memmove(tmp1, tmp1 + 1, strlen(tmp1));
1811 snprintf(expr, sizeof(expr), "(u32)&%s + %s", tmp1, tmp2);
1814 // XXX: do we need more parsing?
1816 snprintf(buf, buf_size, "%s", expr);
1820 snprintf(buf, buf_size, "%s(%s)",
1821 simplify_cast(cast, lmod_cast_u_ptr(po, popr->lmod)), expr);
1825 name = check_label_read_ref(po, popr->name);
1826 if (cast[0] == 0 && popr->is_ptr)
1830 snprintf(buf, buf_size, "(u32)&%s", name);
1831 else if (popr->size_lt)
1832 snprintf(buf, buf_size, "%s%s%s%s", cast,
1833 lmod_cast_u_ptr(po, popr->lmod),
1834 popr->is_array ? "" : "&", name);
1836 snprintf(buf, buf_size, "%s%s%s", cast, name,
1837 popr->is_array ? "[0]" : "");
1841 name = check_label_read_ref(po, popr->name);
1845 ferr(po, "lea an offset?\n");
1846 snprintf(buf, buf_size, "%s&%s", cast, name);
1851 ferr(po, "lea from const?\n");
1853 printf_number(tmp1, sizeof(tmp1), popr->val);
1854 if (popr->val == 0 && strchr(cast, '*'))
1855 snprintf(buf, buf_size, "NULL");
1857 snprintf(buf, buf_size, "%s%s",
1858 simplify_cast_num(cast, popr->val), tmp1);
1862 ferr(po, "invalid src type: %d\n", popr->type);
1868 // note: may set is_ptr (we find that out late for ebp frame..)
1869 static char *out_dst_opr(char *buf, size_t buf_size,
1870 struct parsed_op *po, struct parsed_opr *popr)
1872 switch (popr->type) {
1874 switch (popr->lmod) {
1876 snprintf(buf, buf_size, "%s.q", opr_reg_p(po, popr));
1879 snprintf(buf, buf_size, "%s", opr_reg_p(po, popr));
1883 snprintf(buf, buf_size, "LOWORD(%s)", opr_reg_p(po, popr));
1887 if (popr->name[1] == 'h') // XXX..
1888 snprintf(buf, buf_size, "BYTE1(%s)", opr_reg_p(po, popr));
1890 snprintf(buf, buf_size, "LOBYTE(%s)", opr_reg_p(po, popr));
1893 ferr(po, "invalid dst lmod: %d\n", popr->lmod);
1898 if (is_stack_access(po, popr)) {
1899 stack_frame_access(po, popr, buf, buf_size,
1900 popr->name, "", 0, 0);
1904 return out_src_opr(buf, buf_size, po, popr, NULL, 0);
1907 if (popr->size_mismatch)
1908 snprintf(buf, buf_size, "%s%s%s",
1909 lmod_cast_u_ptr(po, popr->lmod),
1910 popr->is_array ? "" : "&", popr->name);
1912 snprintf(buf, buf_size, "%s%s", popr->name,
1913 popr->is_array ? "[0]" : "");
1917 ferr(po, "invalid dst type: %d\n", popr->type);
1923 static char *out_src_opr_u32(char *buf, size_t buf_size,
1924 struct parsed_op *po, struct parsed_opr *popr)
1926 return out_src_opr(buf, buf_size, po, popr, NULL, 0);
1929 static void out_test_for_cc(char *buf, size_t buf_size,
1930 struct parsed_op *po, enum parsed_flag_op pfo, int is_inv,
1931 enum opr_lenmod lmod, const char *expr)
1933 const char *cast, *scast;
1935 cast = lmod_cast_u(po, lmod);
1936 scast = lmod_cast_s(po, lmod);
1940 case PFO_BE: // CF=1||ZF=1; CF=0
1941 snprintf(buf, buf_size, "(%s%s %s 0)",
1942 cast, expr, is_inv ? "!=" : "==");
1946 case PFO_L: // SF!=OF; OF=0
1947 snprintf(buf, buf_size, "(%s%s %s 0)",
1948 scast, expr, is_inv ? ">=" : "<");
1951 case PFO_LE: // ZF=1||SF!=OF; OF=0
1952 snprintf(buf, buf_size, "(%s%s %s 0)",
1953 scast, expr, is_inv ? ">" : "<=");
1957 ferr(po, "%s: unhandled parsed_flag_op: %d\n", __func__, pfo);
1961 static void out_cmp_for_cc(char *buf, size_t buf_size,
1962 struct parsed_op *po, enum parsed_flag_op pfo, int is_inv)
1964 const char *cast, *scast, *cast_use;
1965 char buf1[256], buf2[256];
1966 enum opr_lenmod lmod;
1968 if (po->op != OP_DEC && po->operand[0].lmod != po->operand[1].lmod)
1969 ferr(po, "%s: lmod mismatch: %d %d\n", __func__,
1970 po->operand[0].lmod, po->operand[1].lmod);
1971 lmod = po->operand[0].lmod;
1973 cast = lmod_cast_u(po, lmod);
1974 scast = lmod_cast_s(po, lmod);
1990 ferr(po, "%s: unhandled parsed_flag_op: %d\n", __func__, pfo);
1993 out_src_opr(buf1, sizeof(buf1), po, &po->operand[0], cast_use, 0);
1994 if (po->op == OP_DEC)
1995 snprintf(buf2, sizeof(buf2), "1");
1997 out_src_opr(buf2, sizeof(buf2), po, &po->operand[1], cast_use, 0);
2001 // note: must be unsigned compare
2002 snprintf(buf, buf_size, "(%s %s %s)",
2003 buf1, is_inv ? ">=" : "<", buf2);
2007 snprintf(buf, buf_size, "(%s %s %s)",
2008 buf1, is_inv ? "!=" : "==", buf2);
2012 // note: must be unsigned compare
2013 snprintf(buf, buf_size, "(%s %s %s)",
2014 buf1, is_inv ? ">" : "<=", buf2);
2017 if (is_inv && lmod == OPLM_BYTE
2018 && po->operand[1].type == OPT_CONST
2019 && po->operand[1].val == 0xff)
2021 snprintf(g_comment, sizeof(g_comment), "if %s", buf);
2022 snprintf(buf, buf_size, "(0)");
2026 // note: must be signed compare
2028 snprintf(buf, buf_size, "(%s(%s - %s) %s 0)",
2029 scast, buf1, buf2, is_inv ? ">=" : "<");
2033 snprintf(buf, buf_size, "(%s %s %s)",
2034 buf1, is_inv ? ">=" : "<", buf2);
2038 snprintf(buf, buf_size, "(%s %s %s)",
2039 buf1, is_inv ? ">" : "<=", buf2);
2047 static void out_cmp_test(char *buf, size_t buf_size,
2048 struct parsed_op *po, enum parsed_flag_op pfo, int is_inv)
2050 char buf1[256], buf2[256], buf3[256];
2052 if (po->op == OP_TEST) {
2053 if (IS(opr_name(po, 0), opr_name(po, 1))) {
2054 out_src_opr_u32(buf3, sizeof(buf3), po, &po->operand[0]);
2057 out_src_opr_u32(buf1, sizeof(buf1), po, &po->operand[0]);
2058 out_src_opr_u32(buf2, sizeof(buf2), po, &po->operand[1]);
2059 snprintf(buf3, sizeof(buf3), "(%s & %s)", buf1, buf2);
2061 out_test_for_cc(buf, buf_size, po, pfo, is_inv,
2062 po->operand[0].lmod, buf3);
2064 else if (po->op == OP_CMP) {
2065 out_cmp_for_cc(buf, buf_size, po, pfo, is_inv);
2068 ferr(po, "%s: unhandled op: %d\n", __func__, po->op);
2071 static void propagate_lmod(struct parsed_op *po, struct parsed_opr *popr1,
2072 struct parsed_opr *popr2)
2074 if (popr1->lmod == OPLM_UNSPEC && popr2->lmod == OPLM_UNSPEC)
2075 ferr(po, "missing lmod for both operands\n");
2077 if (popr1->lmod == OPLM_UNSPEC)
2078 popr1->lmod = popr2->lmod;
2079 else if (popr2->lmod == OPLM_UNSPEC)
2080 popr2->lmod = popr1->lmod;
2081 else if (popr1->lmod != popr2->lmod) {
2082 if (popr1->type_from_var) {
2083 popr1->size_mismatch = 1;
2084 if (popr1->lmod < popr2->lmod)
2086 popr1->lmod = popr2->lmod;
2088 else if (popr2->type_from_var) {
2089 popr2->size_mismatch = 1;
2090 if (popr2->lmod < popr1->lmod)
2092 popr2->lmod = popr1->lmod;
2095 ferr(po, "conflicting lmods: %d vs %d\n",
2096 popr1->lmod, popr2->lmod);
2100 static const char *op_to_c(struct parsed_op *po)
2124 ferr(po, "op_to_c was supplied with %d\n", po->op);
2128 static void op_set_clear_flag(struct parsed_op *po,
2129 enum op_flags flag_set, enum op_flags flag_clear)
2131 po->flags |= flag_set;
2132 po->flags &= ~flag_clear;
2135 // last op in stream - unconditional branch or ret
2136 #define LAST_OP(_i) ((ops[_i].flags & OPF_TAIL) \
2137 || ((ops[_i].flags & (OPF_JMP|OPF_CJMP|OPF_RMD)) == OPF_JMP \
2138 && ops[_i].op != OP_CALL))
2140 static int scan_for_pop(int i, int opcnt, const char *reg,
2141 int magic, int depth, int *maxdepth, int do_flags)
2143 const struct parsed_proto *pp;
2144 struct parsed_op *po;
2148 for (; i < opcnt; i++) {
2150 if (po->cc_scratch == magic)
2151 break; // already checked
2152 po->cc_scratch = magic;
2154 if (po->flags & OPF_TAIL) {
2155 if (po->op == OP_CALL) {
2156 pp = proto_parse(g_fhdr, po->operand[0].name, g_quiet_pp);
2157 if (pp != NULL && pp->is_noreturn)
2158 // no stack cleanup for noreturn
2161 return -1; // deadend
2164 if ((po->flags & (OPF_RMD|OPF_DONE))
2165 || (po->op == OP_PUSH && po->p_argnum != 0)) // arg push
2168 if ((po->flags & OPF_JMP) && po->op != OP_CALL) {
2169 if (po->btj != NULL) {
2171 for (j = 0; j < po->btj->count; j++) {
2172 ret |= scan_for_pop(po->btj->d[j].bt_i, opcnt, reg, magic,
2173 depth, maxdepth, do_flags);
2175 return ret; // dead end
2181 ferr(po, "dead branch\n");
2185 if (po->flags & OPF_CJMP) {
2186 ret |= scan_for_pop(po->bt_i, opcnt, reg, magic,
2187 depth, maxdepth, do_flags);
2189 return ret; // dead end
2197 if ((po->op == OP_POP || po->op == OP_PUSH)
2198 && po->operand[0].type == OPT_REG
2199 && IS(po->operand[0].name, reg))
2201 if (po->op == OP_PUSH && !(po->flags & OPF_FARGNR)) {
2203 if (depth > *maxdepth)
2206 op_set_clear_flag(po, OPF_RSAVE, OPF_RMD);
2208 else if (po->op == OP_POP) {
2211 op_set_clear_flag(po, OPF_RMD, OPF_RSAVE);
2216 if (depth < 0) // should not happen
2217 ferr(po, "fail with depth\n");
2219 op_set_clear_flag(po, OPF_RSAVE, OPF_RMD);
2228 // scan for pop starting from 'ret' op (all paths)
2229 static int scan_for_pop_ret(int i, int opcnt, const char *reg,
2235 for (; i < opcnt; i++) {
2236 if (!(ops[i].flags & OPF_TAIL))
2239 for (j = i - 1; j >= 0; j--) {
2240 if (ops[j].flags & (OPF_RMD|OPF_DONE))
2242 if (ops[j].flags & OPF_JMP)
2245 if (ops[j].op == OP_POP && ops[j].datap == NULL
2246 && ops[j].operand[0].type == OPT_REG
2247 && IS(ops[j].operand[0].name, reg))
2250 ops[j].flags |= flag_set;
2254 if (g_labels[j] != NULL)
2259 return found ? 0 : -1;
2262 static void scan_for_pop_const(int i, int opcnt)
2266 for (j = i + 1; j < opcnt; j++) {
2267 if ((ops[j].flags & (OPF_JMP|OPF_TAIL|OPF_RSAVE))
2268 || ops[j].op == OP_PUSH || g_labels[i] != NULL)
2273 if (ops[j].op == OP_POP && !(ops[j].flags & (OPF_RMD|OPF_DONE)))
2275 ops[i].flags |= OPF_RMD | OPF_DONE;
2276 ops[j].flags |= OPF_DONE;
2277 ops[j].datap = &ops[i];
2283 static void scan_propagate_df(int i, int opcnt)
2285 struct parsed_op *po = &ops[i];
2288 for (; i < opcnt; i++) {
2290 if (po->flags & OPF_DF)
2291 return; // already resolved
2292 po->flags |= OPF_DF;
2294 if (po->op == OP_CALL)
2295 ferr(po, "call with DF set?\n");
2297 if (po->flags & OPF_JMP) {
2298 if (po->btj != NULL) {
2300 for (j = 0; j < po->btj->count; j++)
2301 scan_propagate_df(po->btj->d[j].bt_i, opcnt);
2306 ferr(po, "dead branch\n");
2310 if (po->flags & OPF_CJMP)
2311 scan_propagate_df(po->bt_i, opcnt);
2317 if (po->flags & OPF_TAIL)
2320 if (po->op == OP_CLD) {
2321 po->flags |= OPF_RMD | OPF_DONE;
2326 ferr(po, "missing DF clear?\n");
2329 // is operand 'opr' modified by parsed_op 'po'?
2330 static int is_opr_modified(const struct parsed_opr *opr,
2331 const struct parsed_op *po)
2335 if ((po->flags & OPF_RMD) || !(po->flags & OPF_DATA))
2338 if (opr->type == OPT_REG) {
2339 if (po->op == OP_CALL) {
2340 mask = (1 << xAX) | (1 << xCX) | (1 << xDX);
2341 if ((1 << opr->reg) & mask)
2347 if (po->operand[0].type == OPT_REG) {
2348 if (po->regmask_dst & (1 << opr->reg))
2355 return IS(po->operand[0].name, opr->name);
2358 // is any operand of parsed_op 'po_test' modified by parsed_op 'po'?
2359 static int is_any_opr_modified(const struct parsed_op *po_test,
2360 const struct parsed_op *po, int c_mode)
2365 if ((po->flags & OPF_RMD) || !(po->flags & OPF_DATA))
2368 if (po_test->operand_cnt == 1 && po_test->operand[0].type == OPT_CONST)
2371 if ((po_test->regmask_src | po_test->regmask_dst) & po->regmask_dst)
2374 // in reality, it can wreck any register, but in decompiled C
2375 // version it can only overwrite eax or edx:eax
2376 mask = (1 << xAX) | (1 << xDX);
2380 if (po->op == OP_CALL
2381 && ((po_test->regmask_src | po_test->regmask_dst) & mask))
2384 for (i = 0; i < po_test->operand_cnt; i++)
2385 if (IS(po_test->operand[i].name, po->operand[0].name))
2391 // scan for any po_test operand modification in range given
2392 static int scan_for_mod(struct parsed_op *po_test, int i, int opcnt,
2395 if (po_test->operand_cnt == 1 && po_test->operand[0].type == OPT_CONST)
2398 for (; i < opcnt; i++) {
2399 if (is_any_opr_modified(po_test, &ops[i], c_mode))
2406 // scan for po_test operand[0] modification in range given
2407 static int scan_for_mod_opr0(struct parsed_op *po_test,
2410 for (; i < opcnt; i++) {
2411 if (is_opr_modified(&po_test->operand[0], &ops[i]))
2418 #define check_i(po, i) \
2420 ferr(po, "bad " #i ": %d\n", i)
2422 static int scan_for_flag_set(int i, int magic, int *branched,
2423 int *setters, int *setter_cnt)
2425 struct label_ref *lr;
2429 if (ops[i].cc_scratch == magic) {
2430 ferr(&ops[i], "%s looped\n", __func__);
2433 ops[i].cc_scratch = magic;
2435 if (g_labels[i] != NULL) {
2438 lr = &g_label_refs[i];
2439 for (; lr->next; lr = lr->next) {
2440 check_i(&ops[i], lr->i);
2441 ret = scan_for_flag_set(lr->i, magic,
2442 branched, setters, setter_cnt);
2447 check_i(&ops[i], lr->i);
2448 if (i > 0 && LAST_OP(i - 1)) {
2452 ret = scan_for_flag_set(lr->i, magic,
2453 branched, setters, setter_cnt);
2459 if (ops[i].flags & OPF_FLAGS) {
2460 setters[*setter_cnt] = i;
2465 if ((ops[i].flags & (OPF_JMP|OPF_CJMP)) == OPF_JMP)
2472 // scan back for cdq, if anything modifies edx, fail
2473 static int scan_for_cdq_edx(int i)
2476 if (g_labels[i] != NULL) {
2477 if (g_label_refs[i].next != NULL)
2479 if (i > 0 && LAST_OP(i - 1)) {
2480 i = g_label_refs[i].i;
2487 if (ops[i].op == OP_CDQ)
2490 if (ops[i].regmask_dst & (1 << xDX))
2497 static int scan_for_reg_clear(int i, int reg)
2500 if (g_labels[i] != NULL) {
2501 if (g_label_refs[i].next != NULL)
2503 if (i > 0 && LAST_OP(i - 1)) {
2504 i = g_label_refs[i].i;
2511 if (ops[i].op == OP_XOR
2512 && ops[i].operand[0].lmod == OPLM_DWORD
2513 && ops[i].operand[0].reg == ops[i].operand[1].reg
2514 && ops[i].operand[0].reg == reg)
2517 if (ops[i].regmask_dst & (1 << reg))
2524 static void patch_esp_adjust(struct parsed_op *po, int adj)
2526 ferr_assert(po, po->op == OP_ADD);
2527 ferr_assert(po, IS(opr_name(po, 0), "esp"));
2528 ferr_assert(po, po->operand[1].type == OPT_CONST);
2530 // this is a bit of a hack, but deals with use of
2531 // single adj for multiple calls
2532 po->operand[1].val -= adj;
2533 po->flags |= OPF_RMD;
2534 if (po->operand[1].val == 0)
2535 po->flags |= OPF_DONE;
2536 ferr_assert(po, (int)po->operand[1].val >= 0);
2539 // scan for positive, constant esp adjust
2540 // multipath case is preliminary
2541 static int scan_for_esp_adjust(int i, int opcnt,
2542 int adj_expect, int *adj, int *is_multipath, int do_update)
2544 struct parsed_op *po;
2547 *adj = *is_multipath = 0;
2549 for (; i < opcnt && *adj < adj_expect; i++) {
2550 if (g_labels[i] != NULL)
2554 if (po->flags & OPF_DONE)
2557 if (po->op == OP_ADD && po->operand[0].reg == xSP) {
2558 if (po->operand[1].type != OPT_CONST)
2559 ferr(&ops[i], "non-const esp adjust?\n");
2560 *adj += po->operand[1].val;
2562 ferr(&ops[i], "unaligned esp adjust: %x\n", *adj);
2565 patch_esp_adjust(po, adj_expect);
2567 po->flags |= OPF_RMD;
2571 else if (po->op == OP_PUSH) {
2572 //if (first_pop == -1)
2573 // first_pop = -2; // none
2574 *adj -= lmod_bytes(po, po->operand[0].lmod);
2576 else if (po->op == OP_POP) {
2577 if (!(po->flags & OPF_DONE)) {
2578 // seems like msvc only uses 'pop ecx' for stack realignment..
2579 if (po->operand[0].type != OPT_REG || po->operand[0].reg != xCX)
2581 if (first_pop == -1 && *adj >= 0)
2584 if (do_update && *adj >= 0) {
2585 po->flags |= OPF_RMD;
2587 po->flags |= OPF_DONE;
2590 *adj += lmod_bytes(po, po->operand[0].lmod);
2592 else if (po->flags & (OPF_JMP|OPF_TAIL)) {
2593 if (po->op == OP_JMP && po->btj == NULL) {
2599 if (po->op != OP_CALL)
2601 if (po->operand[0].type != OPT_LABEL)
2603 if (po->pp != NULL && po->pp->is_stdcall)
2605 // assume it's another cdecl call
2609 if (first_pop >= 0) {
2610 // probably 'pop ecx' was used..
2617 static void scan_fwd_set_flags(int i, int opcnt, int magic, int flags)
2619 struct parsed_op *po;
2623 ferr(ops, "%s: followed bad branch?\n", __func__);
2625 for (; i < opcnt; i++) {
2627 if (po->cc_scratch == magic)
2629 po->cc_scratch = magic;
2632 if ((po->flags & OPF_JMP) && po->op != OP_CALL) {
2633 if (po->btj != NULL) {
2635 for (j = 0; j < po->btj->count; j++)
2636 scan_fwd_set_flags(po->btj->d[j].bt_i, opcnt, magic, flags);
2640 scan_fwd_set_flags(po->bt_i, opcnt, magic, flags);
2641 if (!(po->flags & OPF_CJMP))
2644 if (po->flags & OPF_TAIL)
2649 static const struct parsed_proto *try_recover_pp(
2650 struct parsed_op *po, const struct parsed_opr *opr, int *search_instead)
2652 const struct parsed_proto *pp = NULL;
2656 // maybe an arg of g_func?
2657 if (opr->type == OPT_REGMEM && is_stack_access(po, opr))
2659 char ofs_reg[16] = { 0, };
2660 int arg, arg_s, arg_i;
2667 parse_stack_access(po, opr->name, ofs_reg,
2668 &offset, &stack_ra, NULL, 0);
2669 if (ofs_reg[0] != 0)
2670 ferr(po, "offset reg on arg access?\n");
2671 if (offset <= stack_ra) {
2672 // search who set the stack var instead
2673 if (search_instead != NULL)
2674 *search_instead = 1;
2678 arg_i = (offset - stack_ra - 4) / 4;
2679 for (arg = arg_s = 0; arg < g_func_pp->argc; arg++) {
2680 if (g_func_pp->arg[arg].reg != NULL)
2686 if (arg == g_func_pp->argc)
2687 ferr(po, "stack arg %d not in prototype?\n", arg_i);
2689 pp = g_func_pp->arg[arg].fptr;
2691 ferr(po, "icall sa: arg%d is not a fptr?\n", arg + 1);
2692 check_func_pp(po, pp, "icall arg");
2694 else if (opr->type == OPT_REGMEM && strchr(opr->name + 1, '[')) {
2696 p = strchr(opr->name + 1, '[');
2697 memcpy(buf, opr->name, p - opr->name);
2698 buf[p - opr->name] = 0;
2699 pp = proto_parse(g_fhdr, buf, g_quiet_pp);
2701 else if (opr->type == OPT_OFFSET || opr->type == OPT_LABEL) {
2702 pp = proto_parse(g_fhdr, opr->name, g_quiet_pp);
2705 ferr(po, "proto_parse failed for icall to '%s'\n", opr->name);
2708 check_func_pp(po, pp, "reg-fptr ref");
2714 static void scan_for_call_type(int i, const struct parsed_opr *opr,
2715 int magic, const struct parsed_proto **pp_found, int *multi)
2717 const struct parsed_proto *pp = NULL;
2718 struct parsed_op *po;
2719 struct label_ref *lr;
2721 ops[i].cc_scratch = magic;
2724 if (g_labels[i] != NULL) {
2725 lr = &g_label_refs[i];
2726 for (; lr != NULL; lr = lr->next) {
2727 check_i(&ops[i], lr->i);
2728 scan_for_call_type(lr->i, opr, magic, pp_found, multi);
2730 if (i > 0 && LAST_OP(i - 1))
2738 if (ops[i].cc_scratch == magic)
2740 ops[i].cc_scratch = magic;
2742 if (!(ops[i].flags & OPF_DATA))
2744 if (!is_opr_modified(opr, &ops[i]))
2746 if (ops[i].op != OP_MOV && ops[i].op != OP_LEA) {
2747 // most probably trashed by some processing
2752 opr = &ops[i].operand[1];
2753 if (opr->type != OPT_REG)
2757 po = (i >= 0) ? &ops[i] : ops;
2760 // reached the top - can only be an arg-reg
2761 if (opr->type != OPT_REG)
2764 for (i = 0; i < g_func_pp->argc; i++) {
2765 if (g_func_pp->arg[i].reg == NULL)
2767 if (IS(opr->name, g_func_pp->arg[i].reg))
2770 if (i == g_func_pp->argc)
2772 pp = g_func_pp->arg[i].fptr;
2774 ferr(po, "icall: arg%d (%s) is not a fptr?\n",
2775 i + 1, g_func_pp->arg[i].reg);
2776 check_func_pp(po, pp, "icall reg-arg");
2779 pp = try_recover_pp(po, opr, NULL);
2781 if (*pp_found != NULL && pp != NULL && *pp_found != pp) {
2782 if (!IS((*pp_found)->ret_type.name, pp->ret_type.name)
2783 || (*pp_found)->is_stdcall != pp->is_stdcall
2784 || (*pp_found)->is_fptr != pp->is_fptr
2785 || (*pp_found)->argc != pp->argc
2786 || (*pp_found)->argc_reg != pp->argc_reg
2787 || (*pp_found)->argc_stack != pp->argc_stack)
2789 ferr(po, "icall: parsed_proto mismatch\n");
2797 // early check for tail call or branch back
2798 static int is_like_tailjmp(int j)
2800 if (!(ops[j].flags & OPF_JMP))
2803 if (ops[j].op == OP_JMP && !ops[j].operand[0].had_ds)
2804 // probably local branch back..
2806 if (ops[j].op == OP_CALL)
2807 // probably noreturn call..
2813 static void scan_prologue_epilogue(int opcnt)
2815 int ecx_push = 0, esp_sub = 0;
2819 if (ops[0].op == OP_PUSH && IS(opr_name(&ops[0], 0), "ebp")
2820 && ops[1].op == OP_MOV
2821 && IS(opr_name(&ops[1], 0), "ebp")
2822 && IS(opr_name(&ops[1], 1), "esp"))
2825 ops[0].flags |= OPF_RMD | OPF_DONE;
2826 ops[1].flags |= OPF_RMD | OPF_DONE;
2829 if (ops[2].op == OP_SUB && IS(opr_name(&ops[2], 0), "esp")) {
2830 g_stack_fsz = opr_const(&ops[2], 1);
2831 ops[2].flags |= OPF_RMD | OPF_DONE;
2835 // another way msvc builds stack frame..
2837 while (ops[i].op == OP_PUSH && IS(opr_name(&ops[i], 0), "ecx")) {
2839 ops[i].flags |= OPF_RMD | OPF_DONE;
2843 // and another way..
2844 if (i == 2 && ops[i].op == OP_MOV && ops[i].operand[0].reg == xAX
2845 && ops[i].operand[1].type == OPT_CONST
2846 && ops[i + 1].op == OP_CALL
2847 && IS(opr_name(&ops[i + 1], 0), "__alloca_probe"))
2849 g_stack_fsz += ops[i].operand[1].val;
2850 ops[i].flags |= OPF_RMD | OPF_DONE;
2852 ops[i].flags |= OPF_RMD | OPF_DONE;
2859 for (; i < opcnt; i++)
2860 if (ops[i].op == OP_RET)
2863 if (i == opcnt && (ops[j].flags & OPF_JMP)) {
2864 if (found && is_like_tailjmp(j))
2869 if ((ops[j].op == OP_POP && IS(opr_name(&ops[j], 0), "ebp"))
2870 || ops[j].op == OP_LEAVE)
2872 ops[j].flags |= OPF_RMD | OPF_DONE;
2874 else if (!(g_ida_func_attr & IDAFA_NORETURN))
2875 ferr(&ops[j], "'pop ebp' expected\n");
2877 if (g_stack_fsz != 0) {
2878 if (ops[j - 1].op == OP_MOV
2879 && IS(opr_name(&ops[j - 1], 0), "esp")
2880 && IS(opr_name(&ops[j - 1], 1), "ebp"))
2882 ops[j - 1].flags |= OPF_RMD | OPF_DONE;
2884 else if (ops[j].op != OP_LEAVE
2885 && !(g_ida_func_attr & IDAFA_NORETURN))
2887 ferr(&ops[j - 1], "esp restore expected\n");
2890 if (ecx_push && ops[j - 2].op == OP_POP
2891 && IS(opr_name(&ops[j - 2], 0), "ecx"))
2893 ferr(&ops[j - 2], "unexpected ecx pop\n");
2899 } while (i < opcnt);
2906 while (ops[i].op == OP_PUSH && IS(opr_name(&ops[i], 0), "ecx")) {
2907 ops[i].flags |= OPF_RMD | OPF_DONE;
2913 for (; i < opcnt; i++) {
2914 if (ops[i].op == OP_PUSH || (ops[i].flags & (OPF_JMP|OPF_TAIL)))
2916 if (ops[i].op == OP_SUB && ops[i].operand[0].reg == xSP
2917 && ops[i].operand[1].type == OPT_CONST)
2919 g_stack_fsz = ops[i].operand[1].val;
2920 ops[i].flags |= OPF_RMD | OPF_DONE;
2926 if (ecx_push && !esp_sub) {
2927 // could actually be args for a call..
2928 for (; i < opcnt; i++)
2929 if (ops[i].op != OP_PUSH)
2932 if (ops[i].op == OP_CALL && ops[i].operand[0].type == OPT_LABEL) {
2933 const struct parsed_proto *pp;
2934 pp = proto_parse(g_fhdr, opr_name(&ops[i], 0), 1);
2935 j = pp ? pp->argc_stack : 0;
2936 while (i > 0 && j > 0) {
2938 if (ops[i].op == OP_PUSH) {
2939 ops[i].flags &= ~(OPF_RMD | OPF_DONE);
2944 ferr(&ops[i], "unhandled prologue\n");
2947 i = g_stack_fsz = ecx_push = 0;
2948 while (ops[i].op == OP_PUSH && IS(opr_name(&ops[i], 0), "ecx")) {
2949 if (!(ops[i].flags & OPF_RMD))
2959 if (ecx_push || esp_sub)
2965 for (; i < opcnt; i++)
2966 if (ops[i].op == OP_RET)
2969 if (i == opcnt && (ops[j].flags & OPF_JMP)) {
2970 if (found && is_like_tailjmp(j))
2976 for (l = 0; l < ecx_push; l++) {
2977 if (ops[j].op == OP_POP && IS(opr_name(&ops[j], 0), "ecx"))
2979 else if (ops[j].op == OP_ADD
2980 && IS(opr_name(&ops[j], 0), "esp")
2981 && ops[j].operand[1].type == OPT_CONST)
2984 l += ops[j].operand[1].val / 4 - 1;
2987 ferr(&ops[j], "'pop ecx' expected\n");
2989 ops[j].flags |= OPF_RMD | OPF_DONE;
2993 ferr(&ops[j], "epilogue scan failed\n");
2999 if (ops[j].op != OP_ADD
3000 || !IS(opr_name(&ops[j], 0), "esp")
3001 || ops[j].operand[1].type != OPT_CONST
3002 || ops[j].operand[1].val != g_stack_fsz)
3003 ferr(&ops[j], "'add esp' expected\n");
3005 ops[j].flags |= OPF_RMD | OPF_DONE;
3006 ops[j].operand[1].val = 0; // hack for stack arg scanner
3011 } while (i < opcnt);
3015 static const struct parsed_proto *resolve_icall(int i, int opcnt,
3018 const struct parsed_proto *pp = NULL;
3019 int search_advice = 0;
3023 switch (ops[i].operand[0].type) {
3027 pp = try_recover_pp(&ops[i], &ops[i].operand[0], &search_advice);
3032 scan_for_call_type(i, &ops[i].operand[0], i + opcnt * 9, &pp,
3040 // find an instruction that changed opr before i op
3041 // *op_i must be set to -1 by caller
3042 // *entry is set to 1 if one source is determined to be the caller
3043 // returns 1 if found, *op_i is then set to origin
3044 static int resolve_origin(int i, const struct parsed_opr *opr,
3045 int magic, int *op_i, int *is_caller)
3047 struct label_ref *lr;
3050 if (ops[i].cc_scratch == magic)
3052 ops[i].cc_scratch = magic;
3055 if (g_labels[i] != NULL) {
3056 lr = &g_label_refs[i];
3057 for (; lr != NULL; lr = lr->next) {
3058 check_i(&ops[i], lr->i);
3059 ret |= resolve_origin(lr->i, opr, magic, op_i, is_caller);
3061 if (i > 0 && LAST_OP(i - 1))
3067 if (is_caller != NULL)
3072 if (ops[i].cc_scratch == magic)
3074 ops[i].cc_scratch = magic;
3076 if (!(ops[i].flags & OPF_DATA))
3078 if (!is_opr_modified(opr, &ops[i]))
3084 // XXX: could check if the other op does the same
3093 static int try_resolve_const(int i, const struct parsed_opr *opr,
3094 int magic, unsigned int *val)
3099 ret = resolve_origin(i, opr, magic, &s_i, NULL);
3102 if (ops[i].op != OP_MOV && ops[i].operand[1].type != OPT_CONST)
3105 *val = ops[i].operand[1].val;
3112 static struct parsed_proto *process_call_early(int i, int opcnt,
3115 struct parsed_op *po = &ops[i];
3116 struct parsed_proto *pp;
3122 if (pp == NULL || pp->is_vararg || pp->argc_reg != 0)
3126 // look for and make use of esp adjust
3128 if (!pp->is_stdcall && pp->argc_stack > 0)
3129 ret = scan_for_esp_adjust(i + 1, opcnt,
3130 pp->argc_stack * 4, &adj, &multipath, 0);
3132 if (pp->argc_stack > adj / 4)
3136 if (ops[ret].op == OP_POP && adj != 4)
3144 static struct parsed_proto *process_call(int i, int opcnt)
3146 struct parsed_op *po = &ops[i];
3147 const struct parsed_proto *pp_c;
3148 struct parsed_proto *pp;
3149 const char *tmpname;
3150 int adj = 0, multipath = 0;
3153 tmpname = opr_name(po, 0);
3158 pp_c = resolve_icall(i, opcnt, &multipath);
3160 if (!pp_c->is_func && !pp_c->is_fptr)
3161 ferr(po, "call to non-func: %s\n", pp_c->name);
3162 pp = proto_clone(pp_c);
3163 my_assert_not(pp, NULL);
3165 // not resolved just to single func
3168 switch (po->operand[0].type) {
3170 // we resolved this call and no longer need the register
3171 po->regmask_src &= ~(1 << po->operand[0].reg);
3181 pp = calloc(1, sizeof(*pp));
3182 my_assert_not(pp, NULL);
3185 ret = scan_for_esp_adjust(i + 1, opcnt,
3186 32*4, &adj, &multipath, 0);
3187 if (ret < 0 || adj < 0) {
3188 if (!g_allow_regfunc)
3189 ferr(po, "non-__cdecl indirect call unhandled yet\n");
3190 pp->is_unresolved = 1;
3194 if (adj > ARRAY_SIZE(pp->arg))
3195 ferr(po, "esp adjust too large: %d\n", adj);
3196 pp->ret_type.name = strdup("int");
3197 pp->argc = pp->argc_stack = adj;
3198 for (arg = 0; arg < pp->argc; arg++)
3199 pp->arg[arg].type.name = strdup("int");
3204 // look for and make use of esp adjust
3207 if (!pp->is_stdcall && pp->argc_stack > 0)
3208 ret = scan_for_esp_adjust(i + 1, opcnt,
3209 pp->argc_stack * 4, &adj, &multipath, 0);
3211 if (pp->is_vararg) {
3212 if (adj / 4 < pp->argc_stack) {
3213 fnote(po, "(this call)\n");
3214 ferr(&ops[ret], "esp adjust is too small: %x < %x\n",
3215 adj, pp->argc_stack * 4);
3217 // modify pp to make it have varargs as normal args
3219 pp->argc += adj / 4 - pp->argc_stack;
3220 for (; arg < pp->argc; arg++) {
3221 pp->arg[arg].type.name = strdup("int");
3224 if (pp->argc > ARRAY_SIZE(pp->arg))
3225 ferr(po, "too many args for '%s'\n", tmpname);
3227 if (pp->argc_stack > adj / 4) {
3228 fnote(po, "(this call)\n");
3229 ferr(&ops[ret], "stack tracking failed for '%s': %x %x\n",
3230 tmpname, pp->argc_stack * 4, adj);
3233 scan_for_esp_adjust(i + 1, opcnt,
3234 pp->argc_stack * 4, &adj, &multipath, 1);
3236 else if (pp->is_vararg)
3237 ferr(po, "missing esp_adjust for vararg func '%s'\n",
3243 static int collect_call_args_early(struct parsed_op *po, int i,
3244 struct parsed_proto *pp, int *regmask)
3249 for (arg = 0; arg < pp->argc; arg++)
3250 if (pp->arg[arg].reg == NULL)
3253 // first see if it can be easily done
3254 for (j = i; j > 0 && arg < pp->argc; )
3256 if (g_labels[j] != NULL)
3260 if (ops[j].op == OP_CALL)
3262 else if (ops[j].op == OP_ADD && ops[j].operand[0].reg == xSP)
3264 else if (ops[j].op == OP_POP)
3266 else if (ops[j].flags & OPF_CJMP)
3268 else if (ops[j].op == OP_PUSH) {
3269 if (ops[j].flags & (OPF_FARG|OPF_FARGNR))
3271 ret = scan_for_mod(&ops[j], j + 1, i, 1);
3275 if (pp->arg[arg].type.is_va_list)
3279 for (arg++; arg < pp->argc; arg++)
3280 if (pp->arg[arg].reg == NULL)
3289 for (arg = 0; arg < pp->argc; arg++)
3290 if (pp->arg[arg].reg == NULL)
3293 for (j = i; j > 0 && arg < pp->argc; )
3297 if (ops[j].op == OP_PUSH)
3299 ops[j].p_argnext = -1;
3300 ferr_assert(&ops[j], pp->arg[arg].datap == NULL);
3301 pp->arg[arg].datap = &ops[j];
3303 if (ops[j].operand[0].type == OPT_REG)
3304 *regmask |= 1 << ops[j].operand[0].reg;
3306 ops[j].flags |= OPF_RMD | OPF_DONE | OPF_FARGNR | OPF_FARG;
3307 ops[j].flags &= ~OPF_RSAVE;
3310 for (arg++; arg < pp->argc; arg++)
3311 if (pp->arg[arg].reg == NULL)
3319 static int collect_call_args_r(struct parsed_op *po, int i,
3320 struct parsed_proto *pp, int *regmask, int *save_arg_vars,
3321 int *arg_grp, int arg, int magic, int need_op_saving, int may_reuse)
3323 struct parsed_proto *pp_tmp;
3324 struct parsed_op *po_tmp;
3325 struct label_ref *lr;
3326 int need_to_save_current;
3327 int arg_grp_current = 0;
3328 int save_args_seen = 0;
3336 ferr(po, "dead label encountered\n");
3340 for (; arg < pp->argc; arg++)
3341 if (pp->arg[arg].reg == NULL)
3343 magic = (magic & 0xffffff) | (arg << 24);
3345 for (j = i; j >= 0 && (arg < pp->argc || pp->is_unresolved); )
3347 if (((ops[j].cc_scratch ^ magic) & 0xffffff) == 0) {
3348 if (ops[j].cc_scratch != magic) {
3349 ferr(&ops[j], "arg collect hit same path with diff args for %s\n",
3353 // ok: have already been here
3356 ops[j].cc_scratch = magic;
3358 if (g_labels[j] != NULL && g_label_refs[j].i != -1) {
3359 lr = &g_label_refs[j];
3360 if (lr->next != NULL)
3362 for (; lr->next; lr = lr->next) {
3363 check_i(&ops[j], lr->i);
3364 if ((ops[lr->i].flags & (OPF_JMP|OPF_CJMP)) != OPF_JMP)
3366 ret = collect_call_args_r(po, lr->i, pp, regmask, save_arg_vars,
3367 arg_grp, arg, magic, need_op_saving, may_reuse);
3372 check_i(&ops[j], lr->i);
3373 if ((ops[lr->i].flags & (OPF_JMP|OPF_CJMP)) != OPF_JMP)
3375 if (j > 0 && LAST_OP(j - 1)) {
3376 // follow last branch in reverse
3381 ret = collect_call_args_r(po, lr->i, pp, regmask, save_arg_vars,
3382 arg_grp, arg, magic, need_op_saving, may_reuse);
3388 if (ops[j].op == OP_CALL)
3390 if (pp->is_unresolved)
3395 ferr(po, "arg collect hit unparsed call '%s'\n",
3396 ops[j].operand[0].name);
3397 if (may_reuse && pp_tmp->argc_stack > 0)
3398 ferr(po, "arg collect %d/%d hit '%s' with %d stack args\n",
3399 arg, pp->argc, opr_name(&ops[j], 0), pp_tmp->argc_stack);
3401 // esp adjust of 0 means we collected it before
3402 else if (ops[j].op == OP_ADD && ops[j].operand[0].reg == xSP
3403 && (ops[j].operand[1].type != OPT_CONST
3404 || ops[j].operand[1].val != 0))
3406 if (pp->is_unresolved)
3409 ferr(po, "arg collect %d/%d hit esp adjust of %d\n",
3410 arg, pp->argc, ops[j].operand[1].val);
3412 else if (ops[j].op == OP_POP && !(ops[j].flags & OPF_DONE))
3414 if (pp->is_unresolved)
3417 ferr(po, "arg collect %d/%d hit pop\n", arg, pp->argc);
3419 else if (ops[j].flags & OPF_CJMP)
3421 if (pp->is_unresolved)
3426 else if (ops[j].op == OP_PUSH
3427 && !(ops[j].flags & (OPF_FARGNR|OPF_DONE)))
3429 if (pp->is_unresolved && (ops[j].flags & OPF_RMD))
3432 ops[j].p_argnext = -1;
3433 po_tmp = pp->arg[arg].datap;
3435 ops[j].p_argnext = po_tmp - ops;
3436 pp->arg[arg].datap = &ops[j];
3438 need_to_save_current = 0;
3441 if (ops[j].operand[0].type == OPT_REG)
3442 reg = ops[j].operand[0].reg;
3444 if (!need_op_saving) {
3445 ret = scan_for_mod(&ops[j], j + 1, i, 1);
3446 need_to_save_current = (ret >= 0);
3448 if (need_op_saving || need_to_save_current) {
3449 // mark this push as one that needs operand saving
3450 ops[j].flags &= ~OPF_RMD;
3451 if (ops[j].p_argnum == 0) {
3452 ops[j].p_argnum = arg + 1;
3453 save_args |= 1 << arg;
3455 else if (ops[j].p_argnum < arg + 1) {
3456 // XXX: might kill valid var..
3457 //*save_arg_vars &= ~(1 << (ops[j].p_argnum - 1));
3458 ops[j].p_argnum = arg + 1;
3459 save_args |= 1 << arg;
3462 if (save_args_seen & (1 << (ops[j].p_argnum - 1))) {
3465 if (arg_grp_current >= MAX_ARG_GRP)
3466 ferr(&ops[j], "out of arg groups (arg%d), f %s\n",
3467 ops[j].p_argnum, pp->name);
3470 else if (ops[j].p_argnum == 0)
3471 ops[j].flags |= OPF_RMD;
3473 // some PUSHes are reused by different calls on other branches,
3474 // but that can't happen if we didn't branch, so they
3475 // can be removed from future searches (handles nested calls)
3477 ops[j].flags |= OPF_FARGNR;
3479 ops[j].flags |= OPF_FARG;
3480 ops[j].flags &= ~OPF_RSAVE;
3482 // check for __VALIST
3483 if (!pp->is_unresolved && pp->arg[arg].type.is_va_list) {
3485 ret = resolve_origin(j, &ops[j].operand[0],
3486 magic + 1, &k, NULL);
3487 if (ret == 1 && k >= 0)
3489 if (ops[k].op == OP_LEA) {
3490 snprintf(buf, sizeof(buf), "arg_%X",
3491 g_func_pp->argc_stack * 4);
3492 if (!g_func_pp->is_vararg
3493 || strstr(ops[k].operand[1].name, buf))
3495 ops[k].flags |= OPF_RMD | OPF_DONE;
3496 ops[j].flags |= OPF_RMD | OPF_VAPUSH;
3497 save_args &= ~(1 << arg);
3501 ferr(&ops[j], "lea va_list used, but no vararg?\n");
3503 // check for va_list from g_func_pp arg too
3504 else if (ops[k].op == OP_MOV
3505 && is_stack_access(&ops[k], &ops[k].operand[1]))
3507 ret = stack_frame_access(&ops[k], &ops[k].operand[1],
3508 buf, sizeof(buf), ops[k].operand[1].name, "", 1, 0);
3510 ops[k].flags |= OPF_RMD | OPF_DONE;
3511 ops[j].flags |= OPF_RMD;
3512 ops[j].p_argpass = ret + 1;
3513 save_args &= ~(1 << arg);
3520 *save_arg_vars |= save_args;
3522 // tracking reg usage
3524 *regmask |= 1 << reg;
3527 if (!pp->is_unresolved) {
3529 for (; arg < pp->argc; arg++)
3530 if (pp->arg[arg].reg == NULL)
3533 magic = (magic & 0xffffff) | (arg << 24);
3536 if (ops[j].p_arggrp > arg_grp_current) {
3538 arg_grp_current = ops[j].p_arggrp;
3540 if (ops[j].p_argnum > 0)
3541 save_args_seen |= 1 << (ops[j].p_argnum - 1);
3544 if (arg < pp->argc) {
3545 ferr(po, "arg collect failed for '%s': %d/%d\n",
3546 pp->name, arg, pp->argc);
3550 if (arg_grp_current > *arg_grp)
3551 *arg_grp = arg_grp_current;
3556 static int collect_call_args(struct parsed_op *po, int i,
3557 struct parsed_proto *pp, int *regmask, int *save_arg_vars,
3560 // arg group is for cases when pushes for
3561 // multiple funcs are going on
3562 struct parsed_op *po_tmp;
3563 int save_arg_vars_current = 0;
3568 ret = collect_call_args_r(po, i, pp, regmask,
3569 &save_arg_vars_current, &arg_grp, 0, magic, 0, 0);
3574 // propagate arg_grp
3575 for (a = 0; a < pp->argc; a++) {
3576 if (pp->arg[a].reg != NULL)
3579 po_tmp = pp->arg[a].datap;
3580 while (po_tmp != NULL) {
3581 po_tmp->p_arggrp = arg_grp;
3582 if (po_tmp->p_argnext > 0)
3583 po_tmp = &ops[po_tmp->p_argnext];
3589 save_arg_vars[arg_grp] |= save_arg_vars_current;
3591 if (pp->is_unresolved) {
3593 pp->argc_stack += ret;
3594 for (a = 0; a < pp->argc; a++)
3595 if (pp->arg[a].type.name == NULL)
3596 pp->arg[a].type.name = strdup("int");
3602 static void pp_insert_reg_arg(struct parsed_proto *pp, const char *reg)
3606 for (i = 0; i < pp->argc; i++)
3607 if (pp->arg[i].reg == NULL)
3611 memmove(&pp->arg[i + 1], &pp->arg[i],
3612 sizeof(pp->arg[0]) * pp->argc_stack);
3613 memset(&pp->arg[i], 0, sizeof(pp->arg[i]));
3614 pp->arg[i].reg = strdup(reg);
3615 pp->arg[i].type.name = strdup("int");
3620 static void add_label_ref(struct label_ref *lr, int op_i)
3622 struct label_ref *lr_new;
3629 lr_new = calloc(1, sizeof(*lr_new));
3631 lr_new->next = lr->next;
3635 static struct parsed_data *try_resolve_jumptab(int i, int opcnt)
3637 struct parsed_op *po = &ops[i];
3638 struct parsed_data *pd;
3639 char label[NAMELEN], *p;
3642 p = strchr(po->operand[0].name, '[');
3646 len = p - po->operand[0].name;
3647 strncpy(label, po->operand[0].name, len);
3650 for (j = 0, pd = NULL; j < g_func_pd_cnt; j++) {
3651 if (IS(g_func_pd[j].label, label)) {
3657 //ferr(po, "label '%s' not parsed?\n", label);
3660 if (pd->type != OPT_OFFSET)
3661 ferr(po, "label '%s' with non-offset data?\n", label);
3663 // find all labels, link
3664 for (j = 0; j < pd->count; j++) {
3665 for (l = 0; l < opcnt; l++) {
3666 if (g_labels[l] != NULL && IS(g_labels[l], pd->d[j].u.label)) {
3667 add_label_ref(&g_label_refs[l], i);
3677 static void clear_labels(int count)
3681 for (i = 0; i < count; i++) {
3682 if (g_labels[i] != NULL) {
3689 static void output_std_flags(FILE *fout, struct parsed_op *po,
3690 int *pfomask, const char *dst_opr_text)
3692 if (*pfomask & (1 << PFO_Z)) {
3693 fprintf(fout, "\n cond_z = (%s%s == 0);",
3694 lmod_cast_u(po, po->operand[0].lmod), dst_opr_text);
3695 *pfomask &= ~(1 << PFO_Z);
3697 if (*pfomask & (1 << PFO_S)) {
3698 fprintf(fout, "\n cond_s = (%s%s < 0);",
3699 lmod_cast_s(po, po->operand[0].lmod), dst_opr_text);
3700 *pfomask &= ~(1 << PFO_S);
3705 OPP_FORCE_NORETURN = (1 << 0),
3706 OPP_SIMPLE_ARGS = (1 << 1),
3707 OPP_ALIGN = (1 << 2),
3710 static void output_pp_attrs(FILE *fout, const struct parsed_proto *pp,
3713 const char *cconv = "";
3715 if (pp->is_fastcall)
3716 cconv = "__fastcall ";
3717 else if (pp->is_stdcall && pp->argc_reg == 0)
3718 cconv = "__stdcall ";
3720 fprintf(fout, (flags & OPP_ALIGN) ? "%-16s" : "%s", cconv);
3722 if (pp->is_noreturn || (flags & OPP_FORCE_NORETURN))
3723 fprintf(fout, "noreturn ");
3726 static void output_pp(FILE *fout, const struct parsed_proto *pp,
3731 fprintf(fout, (flags & OPP_ALIGN) ? "%-5s" : "%s ",
3735 output_pp_attrs(fout, pp, flags);
3738 fprintf(fout, "%s", pp->name);
3743 for (i = 0; i < pp->argc; i++) {
3745 fprintf(fout, ", ");
3746 if (pp->arg[i].fptr != NULL && !(flags & OPP_SIMPLE_ARGS)) {
3748 output_pp(fout, pp->arg[i].fptr, 0);
3750 else if (pp->arg[i].type.is_retreg) {
3751 fprintf(fout, "u32 *r_%s", pp->arg[i].reg);
3754 fprintf(fout, "%s", pp->arg[i].type.name);
3756 fprintf(fout, " a%d", i + 1);
3759 if (pp->is_vararg) {
3761 fprintf(fout, ", ");
3762 fprintf(fout, "...");
3767 static int get_pp_arg_regmask(const struct parsed_proto *pp)
3772 for (i = 0; i < pp->argc; i++) {
3773 if (pp->arg[i].reg != NULL) {
3774 reg = char_array_i(regs_r32,
3775 ARRAY_SIZE(regs_r32), pp->arg[i].reg);
3777 ferr(ops, "arg '%s' of func '%s' is not a reg?\n",
3778 pp->arg[i].reg, pp->name);
3779 regmask |= 1 << reg;
3786 static char *saved_arg_name(char *buf, size_t buf_size, int grp, int num)
3792 snprintf(buf1, sizeof(buf1), "%d", grp);
3793 snprintf(buf, buf_size, "s%s_a%d", buf1, num);
3798 static void gen_x_cleanup(int opcnt);
3800 static void gen_func(FILE *fout, FILE *fhdr, const char *funcn, int opcnt)
3802 struct parsed_op *po, *delayed_flag_op = NULL, *tmp_op;
3803 struct parsed_opr *last_arith_dst = NULL;
3804 char buf1[256], buf2[256], buf3[256], cast[64];
3805 const struct parsed_proto *pp_c;
3806 struct parsed_proto *pp, *pp_tmp;
3807 struct parsed_data *pd;
3808 const char *tmpname;
3810 int save_arg_vars[MAX_ARG_GRP] = { 0, };
3812 int need_tmp_var = 0;
3815 int label_pending = 0;
3816 int regmask_save = 0;
3817 int regmask_arg = 0;
3818 int regmask_now = 0;
3819 int regmask_init = 0;
3830 g_bp_frame = g_sp_frame = g_stack_fsz = 0;
3831 g_stack_frame_used = 0;
3833 g_func_pp = proto_parse(fhdr, funcn, 0);
3834 if (g_func_pp == NULL)
3835 ferr(ops, "proto_parse failed for '%s'\n", funcn);
3837 regmask_arg = get_pp_arg_regmask(g_func_pp);
3840 // - handle ebp/esp frame, remove ops related to it
3841 scan_prologue_epilogue(opcnt);
3844 // - parse calls with labels
3845 // - resolve all branches
3846 for (i = 0; i < opcnt; i++)
3852 if (po->flags & (OPF_RMD|OPF_DONE))
3855 if (po->op == OP_CALL) {
3858 if (po->operand[0].type == OPT_LABEL) {
3859 tmpname = opr_name(po, 0);
3860 if (IS_START(tmpname, "loc_"))
3861 ferr(po, "call to loc_*\n");
3862 pp_c = proto_parse(fhdr, tmpname, 0);
3864 ferr(po, "proto_parse failed for call '%s'\n", tmpname);
3866 pp = proto_clone(pp_c);
3867 my_assert_not(pp, NULL);
3869 else if (po->datap != NULL) {
3870 pp = calloc(1, sizeof(*pp));
3871 my_assert_not(pp, NULL);
3873 ret = parse_protostr(po->datap, pp);
3875 ferr(po, "bad protostr supplied: %s\n", (char *)po->datap);
3882 check_func_pp(po, pp, "fptr var call");
3883 if (pp->is_noreturn)
3884 po->flags |= OPF_TAIL;
3890 if (!(po->flags & OPF_JMP) || po->op == OP_RET)
3893 if (po->operand[0].type == OPT_REGMEM) {
3894 pd = try_resolve_jumptab(i, opcnt);
3902 for (l = 0; l < opcnt; l++) {
3903 if (g_labels[l] != NULL
3904 && IS(po->operand[0].name, g_labels[l]))
3906 if (l == i + 1 && po->op == OP_JMP) {
3907 // yet another alignment type..
3908 po->flags |= OPF_RMD|OPF_DONE;
3911 add_label_ref(&g_label_refs[l], i);
3917 if (po->bt_i != -1 || (po->flags & OPF_RMD))
3920 if (po->operand[0].type == OPT_LABEL)
3924 ferr(po, "unhandled branch\n");
3928 po->flags |= OPF_TAIL;
3929 if (i > 0 && ops[i - 1].op == OP_POP)
3930 po->flags |= OPF_ATAIL;
3935 // - remove dead labels
3936 // - process trivial calls
3937 for (i = 0; i < opcnt; i++)
3939 if (g_labels[i] != NULL && g_label_refs[i].i == -1) {
3945 if (po->flags & (OPF_RMD|OPF_DONE))
3948 if (po->op == OP_CALL)
3950 pp = process_call_early(i, opcnt, &j);
3952 if (!(po->flags & OPF_ATAIL))
3953 // since we know the args, try to collect them
3954 if (collect_call_args_early(po, i, pp, ®mask) != 0)
3960 // commit esp adjust
3961 ops[j].flags |= OPF_RMD;
3962 if (ops[j].op != OP_POP)
3963 patch_esp_adjust(&ops[j], pp->argc_stack * 4);
3965 ops[j].flags |= OPF_DONE;
3968 if (strstr(pp->ret_type.name, "int64"))
3971 po->flags |= OPF_DONE;
3978 // - handle push <const>/pop pairs
3979 for (i = 0; i < opcnt; i++)
3982 if (po->flags & (OPF_RMD|OPF_DONE))
3985 if (po->op == OP_CALL && !(po->flags & OPF_DONE))
3987 pp = process_call(i, opcnt);
3989 if (!pp->is_unresolved && !(po->flags & OPF_ATAIL)) {
3990 // since we know the args, collect them
3991 collect_call_args(po, i, pp, ®mask, save_arg_vars,
3995 if (strstr(pp->ret_type.name, "int64"))
3998 else if (po->op == OP_PUSH && !(po->flags & OPF_FARG)
3999 && !(po->flags & OPF_RSAVE) && po->operand[0].type == OPT_CONST)
4000 scan_for_pop_const(i, opcnt);
4004 // - find POPs for PUSHes, rm both
4005 // - scan for STD/CLD, propagate DF
4006 // - scan for all used registers
4007 // - find flag set ops for their users
4008 // - do unreselved calls
4009 // - declare indirect functions
4010 for (i = 0; i < opcnt; i++)
4013 if (po->flags & (OPF_RMD|OPF_DONE))
4016 if (po->op == OP_PUSH && (po->flags & OPF_RSAVE)) {
4017 reg = po->operand[0].reg;
4018 if (!(regmask & (1 << reg)))
4019 // not a reg save after all, rerun scan_for_pop
4020 po->flags &= ~OPF_RSAVE;
4022 regmask_save |= 1 << reg;
4025 if (po->op == OP_PUSH && !(po->flags & OPF_FARG)
4026 && !(po->flags & OPF_RSAVE) && !g_func_pp->is_userstack)
4028 if (po->operand[0].type == OPT_REG)
4030 reg = po->operand[0].reg;
4032 ferr(po, "reg not set for push?\n");
4035 ret = scan_for_pop(i + 1, opcnt,
4036 po->operand[0].name, i + opcnt * 3, 0, &depth, 0);
4039 ferr(po, "too much depth: %d\n", depth);
4041 po->flags |= OPF_RMD;
4042 scan_for_pop(i + 1, opcnt, po->operand[0].name,
4043 i + opcnt * 4, 0, &depth, 1);
4046 ret = scan_for_pop_ret(i + 1, opcnt, po->operand[0].name, 0);
4049 if (regmask & (1 << reg)) {
4050 if (regmask_save & (1 << reg))
4051 ferr(po, "%s already saved?\n", po->operand[0].name);
4055 scan_for_pop_ret(i + 1, opcnt, po->operand[0].name, arg);
4061 if (po->op == OP_STD) {
4062 po->flags |= OPF_DF | OPF_RMD | OPF_DONE;
4063 scan_propagate_df(i + 1, opcnt);
4066 regmask_now = po->regmask_src | po->regmask_dst;
4067 if (regmask_now & (1 << xBP)) {
4068 if (g_bp_frame && !(po->flags & OPF_EBP_S)) {
4069 if (po->regmask_dst & (1 << xBP))
4070 // compiler decided to drop bp frame and use ebp as scratch
4071 scan_fwd_set_flags(i + 1, opcnt, i + opcnt * 5, OPF_EBP_S);
4073 regmask_now &= ~(1 << xBP);
4077 regmask |= regmask_now;
4079 if (po->flags & OPF_CC)
4081 int setters[16], cnt = 0, branched = 0;
4083 ret = scan_for_flag_set(i, i + opcnt * 6,
4084 &branched, setters, &cnt);
4085 if (ret < 0 || cnt <= 0)
4086 ferr(po, "unable to trace flag setter(s)\n");
4087 if (cnt > ARRAY_SIZE(setters))
4088 ferr(po, "too many flag setters\n");
4090 for (j = 0; j < cnt; j++)
4092 tmp_op = &ops[setters[j]]; // flag setter
4095 // to get nicer code, we try to delay test and cmp;
4096 // if we can't because of operand modification, or if we
4097 // have arith op, or branch, make it calculate flags explicitly
4098 if (tmp_op->op == OP_TEST || tmp_op->op == OP_CMP)
4100 if (branched || scan_for_mod(tmp_op, setters[j] + 1, i, 0) >= 0)
4101 pfomask = 1 << po->pfo;
4103 else if (tmp_op->op == OP_CMPS || tmp_op->op == OP_SCAS) {
4104 pfomask = 1 << po->pfo;
4107 // see if we'll be able to handle based on op result
4108 if ((tmp_op->op != OP_AND && tmp_op->op != OP_OR
4109 && po->pfo != PFO_Z && po->pfo != PFO_S
4110 && po->pfo != PFO_P)
4112 || scan_for_mod_opr0(tmp_op, setters[j] + 1, i) >= 0)
4114 pfomask = 1 << po->pfo;
4117 if (tmp_op->op == OP_ADD && po->pfo == PFO_C) {
4118 propagate_lmod(tmp_op, &tmp_op->operand[0],
4119 &tmp_op->operand[1]);
4120 if (tmp_op->operand[0].lmod == OPLM_DWORD)
4125 tmp_op->pfomask |= pfomask;
4126 cond_vars |= pfomask;
4128 // note: may overwrite, currently not a problem
4132 if (po->op == OP_RCL || po->op == OP_RCR
4133 || po->op == OP_ADC || po->op == OP_SBB)
4134 cond_vars |= 1 << PFO_C;
4137 if (po->op == OP_CMPS || po->op == OP_SCAS) {
4138 cond_vars |= 1 << PFO_Z;
4140 else if (po->op == OP_MUL
4141 || (po->op == OP_IMUL && po->operand_cnt == 1))
4143 if (po->operand[0].lmod == OPLM_DWORD)
4146 else if (po->op == OP_CALL) {
4147 // note: resolved non-reg calls are OPF_DONE already
4150 ferr(po, "NULL pp\n");
4152 if (pp->is_unresolved) {
4153 int regmask_stack = 0;
4154 collect_call_args(po, i, pp, ®mask, save_arg_vars,
4157 // this is pretty rough guess:
4158 // see ecx and edx were pushed (and not their saved versions)
4159 for (arg = 0; arg < pp->argc; arg++) {
4160 if (pp->arg[arg].reg != NULL)
4163 tmp_op = pp->arg[arg].datap;
4165 ferr(po, "parsed_op missing for arg%d\n", arg);
4166 if (tmp_op->p_argnum == 0 && tmp_op->operand[0].type == OPT_REG)
4167 regmask_stack |= 1 << tmp_op->operand[0].reg;
4170 if (!((regmask_stack & (1 << xCX))
4171 && (regmask_stack & (1 << xDX))))
4173 if (pp->argc_stack != 0
4174 || ((regmask | regmask_arg) & ((1 << xCX)|(1 << xDX))))
4176 pp_insert_reg_arg(pp, "ecx");
4177 pp->is_fastcall = 1;
4178 regmask_init |= 1 << xCX;
4179 regmask |= 1 << xCX;
4181 if (pp->argc_stack != 0
4182 || ((regmask | regmask_arg) & (1 << xDX)))
4184 pp_insert_reg_arg(pp, "edx");
4185 regmask_init |= 1 << xDX;
4186 regmask |= 1 << xDX;
4190 // note: __cdecl doesn't fall into is_unresolved category
4191 if (pp->argc_stack > 0)
4195 for (arg = 0; arg < pp->argc; arg++) {
4196 if (pp->arg[arg].reg != NULL) {
4197 reg = char_array_i(regs_r32,
4198 ARRAY_SIZE(regs_r32), pp->arg[arg].reg);
4200 ferr(ops, "arg '%s' is not a reg?\n", pp->arg[arg].reg);
4201 if (!(regmask & (1 << reg))) {
4202 regmask_init |= 1 << reg;
4203 regmask |= 1 << reg;
4208 else if (po->op == OP_MOV && po->operand[0].pp != NULL
4209 && po->operand[1].pp != NULL)
4211 // <var> = offset <something>
4212 if ((po->operand[1].pp->is_func || po->operand[1].pp->is_fptr)
4213 && !IS_START(po->operand[1].name, "off_"))
4215 if (!po->operand[0].pp->is_fptr)
4216 ferr(po, "%s not declared as fptr when it should be\n",
4217 po->operand[0].name);
4218 if (pp_cmp_func(po->operand[0].pp, po->operand[1].pp)) {
4219 pp_print(buf1, sizeof(buf1), po->operand[0].pp);
4220 pp_print(buf2, sizeof(buf2), po->operand[1].pp);
4221 fnote(po, "var: %s\n", buf1);
4222 fnote(po, "func: %s\n", buf2);
4223 ferr(po, "^ mismatch\n");
4227 else if (po->op == OP_RET && !IS(g_func_pp->ret_type.name, "void"))
4228 regmask |= 1 << xAX;
4229 else if (po->op == OP_DIV || po->op == OP_IDIV) {
4230 // 32bit division is common, look for it
4231 if (po->op == OP_DIV)
4232 ret = scan_for_reg_clear(i, xDX);
4234 ret = scan_for_cdq_edx(i);
4236 po->flags |= OPF_32BIT;
4240 else if (po->op == OP_CLD)
4241 po->flags |= OPF_RMD | OPF_DONE;
4243 if (po->op == OP_RCL || po->op == OP_RCR || po->op == OP_XCHG) {
4249 // - confirm regmask_save, it might have been reduced
4250 if (regmask_save != 0)
4253 for (i = 0; i < opcnt; i++) {
4255 if (po->flags & OPF_RMD)
4258 if (po->op == OP_PUSH && (po->flags & OPF_RSAVE))
4259 regmask_save |= 1 << po->operand[0].reg;
4263 // output starts here
4265 // define userstack size
4266 if (g_func_pp->is_userstack) {
4267 fprintf(fout, "#ifndef US_SZ_%s\n", g_func_pp->name);
4268 fprintf(fout, "#define US_SZ_%s USERSTACK_SIZE\n", g_func_pp->name);
4269 fprintf(fout, "#endif\n");
4272 // the function itself
4273 ferr_assert(ops, !g_func_pp->is_fptr);
4274 output_pp(fout, g_func_pp,
4275 (g_ida_func_attr & IDAFA_NORETURN) ? OPP_FORCE_NORETURN : 0);
4276 fprintf(fout, "\n{\n");
4278 // declare indirect functions
4279 for (i = 0; i < opcnt; i++) {
4281 if (po->flags & OPF_RMD)
4284 if (po->op == OP_CALL) {
4287 ferr(po, "NULL pp\n");
4289 if (pp->is_fptr && !(pp->name[0] != 0 && pp->is_arg)) {
4290 if (pp->name[0] != 0) {
4291 memmove(pp->name + 2, pp->name, strlen(pp->name) + 1);
4292 memcpy(pp->name, "i_", 2);
4294 // might be declared already
4296 for (j = 0; j < i; j++) {
4297 if (ops[j].op == OP_CALL && (pp_tmp = ops[j].pp)) {
4298 if (pp_tmp->is_fptr && IS(pp->name, pp_tmp->name)) {
4308 snprintf(pp->name, sizeof(pp->name), "icall%d", i);
4311 output_pp(fout, pp, OPP_SIMPLE_ARGS);
4312 fprintf(fout, ";\n");
4317 // output LUTs/jumptables
4318 for (i = 0; i < g_func_pd_cnt; i++) {
4320 fprintf(fout, " static const ");
4321 if (pd->type == OPT_OFFSET) {
4322 fprintf(fout, "void *jt_%s[] =\n { ", pd->label);
4324 for (j = 0; j < pd->count; j++) {
4326 fprintf(fout, ", ");
4327 fprintf(fout, "&&%s", pd->d[j].u.label);
4331 fprintf(fout, "%s %s[] =\n { ",
4332 lmod_type_u(ops, pd->lmod), pd->label);
4334 for (j = 0; j < pd->count; j++) {
4336 fprintf(fout, ", ");
4337 fprintf(fout, "%u", pd->d[j].u.val);
4340 fprintf(fout, " };\n");
4344 // declare stack frame, va_arg
4346 fprintf(fout, " union { u32 d[%d]; u16 w[%d]; u8 b[%d]; } sf;\n",
4347 (g_stack_fsz + 3) / 4, (g_stack_fsz + 1) / 2, g_stack_fsz);
4351 if (g_func_pp->is_userstack) {
4352 fprintf(fout, " u32 fake_sf[US_SZ_%s / 4];\n", g_func_pp->name);
4353 fprintf(fout, " u32 *esp = &fake_sf[sizeof(fake_sf) / 4];\n");
4357 if (g_func_pp->is_vararg) {
4358 fprintf(fout, " va_list ap;\n");
4362 // declare arg-registers
4363 for (i = 0; i < g_func_pp->argc; i++) {
4364 if (g_func_pp->arg[i].reg != NULL) {
4365 reg = char_array_i(regs_r32,
4366 ARRAY_SIZE(regs_r32), g_func_pp->arg[i].reg);
4367 if (regmask & (1 << reg)) {
4368 if (g_func_pp->arg[i].type.is_retreg)
4369 fprintf(fout, " u32 %s = *r_%s;\n",
4370 g_func_pp->arg[i].reg, g_func_pp->arg[i].reg);
4372 fprintf(fout, " u32 %s = (u32)a%d;\n",
4373 g_func_pp->arg[i].reg, i + 1);
4376 if (g_func_pp->arg[i].type.is_retreg)
4377 ferr(ops, "retreg '%s' is unused?\n",
4378 g_func_pp->arg[i].reg);
4379 fprintf(fout, " // %s = a%d; // unused\n",
4380 g_func_pp->arg[i].reg, i + 1);
4386 regmask_now = regmask & ~regmask_arg;
4387 regmask_now &= ~(1 << xSP);
4388 if (regmask_now & 0x00ff) {
4389 for (reg = 0; reg < 8; reg++) {
4390 if (regmask_now & (1 << reg)) {
4391 fprintf(fout, " u32 %s", regs_r32[reg]);
4392 if (regmask_init & (1 << reg))
4393 fprintf(fout, " = 0");
4394 fprintf(fout, ";\n");
4399 if (regmask_now & 0xff00) {
4400 for (reg = 8; reg < 16; reg++) {
4401 if (regmask_now & (1 << reg)) {
4402 fprintf(fout, " mmxr %s", regs_r32[reg]);
4403 if (regmask_init & (1 << reg))
4404 fprintf(fout, " = { 0, }");
4405 fprintf(fout, ";\n");
4412 for (reg = 0; reg < 8; reg++) {
4413 if (regmask_save & (1 << reg)) {
4414 fprintf(fout, " u32 s_%s;\n", regs_r32[reg]);
4420 for (i = 0; i < ARRAY_SIZE(save_arg_vars); i++) {
4421 if (save_arg_vars[i] == 0)
4423 for (reg = 0; reg < 32; reg++) {
4424 if (save_arg_vars[i] & (1 << reg)) {
4425 fprintf(fout, " u32 %s;\n",
4426 saved_arg_name(buf1, sizeof(buf1), i, reg + 1));
4433 for (i = 0; i < 8; i++) {
4434 if (cond_vars & (1 << i)) {
4435 fprintf(fout, " u32 cond_%s;\n", parsed_flag_op_names[i]);
4442 fprintf(fout, " u32 tmp;\n");
4447 fprintf(fout, " u64 tmp64;\n");
4452 fprintf(fout, "\n");
4454 if (g_func_pp->is_vararg) {
4455 if (g_func_pp->argc_stack == 0)
4456 ferr(ops, "vararg func without stack args?\n");
4457 fprintf(fout, " va_start(ap, a%d);\n", g_func_pp->argc);
4461 for (i = 0; i < opcnt; i++)
4463 if (g_labels[i] != NULL) {
4464 fprintf(fout, "\n%s:\n", g_labels[i]);
4467 delayed_flag_op = NULL;
4468 last_arith_dst = NULL;
4472 if (po->flags & OPF_RMD)
4477 #define assert_operand_cnt(n_) \
4478 if (po->operand_cnt != n_) \
4479 ferr(po, "operand_cnt is %d/%d\n", po->operand_cnt, n_)
4481 // conditional/flag using op?
4482 if (po->flags & OPF_CC)
4488 // we go through all this trouble to avoid using parsed_flag_op,
4489 // which makes generated code much nicer
4490 if (delayed_flag_op != NULL)
4492 out_cmp_test(buf1, sizeof(buf1), delayed_flag_op,
4493 po->pfo, po->pfo_inv);
4496 else if (last_arith_dst != NULL
4497 && (po->pfo == PFO_Z || po->pfo == PFO_S || po->pfo == PFO_P
4498 || (tmp_op && (tmp_op->op == OP_AND || tmp_op->op == OP_OR))
4501 out_src_opr_u32(buf3, sizeof(buf3), po, last_arith_dst);
4502 out_test_for_cc(buf1, sizeof(buf1), po, po->pfo, po->pfo_inv,
4503 last_arith_dst->lmod, buf3);
4506 else if (tmp_op != NULL) {
4507 // use preprocessed flag calc results
4508 if (!(tmp_op->pfomask & (1 << po->pfo)))
4509 ferr(po, "not prepared for pfo %d\n", po->pfo);
4511 // note: pfo_inv was not yet applied
4512 snprintf(buf1, sizeof(buf1), "(%scond_%s)",
4513 po->pfo_inv ? "!" : "", parsed_flag_op_names[po->pfo]);
4516 ferr(po, "all methods of finding comparison failed\n");
4519 if (po->flags & OPF_JMP) {
4520 fprintf(fout, " if %s", buf1);
4522 else if (po->op == OP_RCL || po->op == OP_RCR
4523 || po->op == OP_ADC || po->op == OP_SBB)
4526 fprintf(fout, " cond_%s = %s;\n",
4527 parsed_flag_op_names[po->pfo], buf1);
4529 else if (po->flags & OPF_DATA) { // SETcc
4530 out_dst_opr(buf2, sizeof(buf2), po, &po->operand[0]);
4531 fprintf(fout, " %s = %s;", buf2, buf1);
4534 ferr(po, "unhandled conditional op\n");
4538 pfomask = po->pfomask;
4540 if (po->flags & (OPF_REPZ|OPF_REPNZ)) {
4541 struct parsed_opr opr = {0,};
4544 opr.lmod = OPLM_DWORD;
4545 ret = try_resolve_const(i, &opr, opcnt * 7 + i, &uval);
4547 if (ret != 1 || uval == 0) {
4548 // we need initial flags for ecx=0 case..
4549 if (i > 0 && ops[i - 1].op == OP_XOR
4550 && IS(ops[i - 1].operand[0].name,
4551 ops[i - 1].operand[1].name))
4553 fprintf(fout, " cond_z = ");
4554 if (pfomask & (1 << PFO_C))
4555 fprintf(fout, "cond_c = ");
4556 fprintf(fout, "0;\n");
4558 else if (last_arith_dst != NULL) {
4559 out_src_opr_u32(buf3, sizeof(buf3), po, last_arith_dst);
4560 out_test_for_cc(buf1, sizeof(buf1), po, PFO_Z, 0,
4561 last_arith_dst->lmod, buf3);
4562 fprintf(fout, " cond_z = %s;\n", buf1);
4565 ferr(po, "missing initial ZF\n");
4572 assert_operand_cnt(2);
4573 propagate_lmod(po, &po->operand[0], &po->operand[1]);
4574 out_dst_opr(buf1, sizeof(buf1), po, &po->operand[0]);
4575 default_cast_to(buf3, sizeof(buf3), &po->operand[0]);
4576 fprintf(fout, " %s = %s;", buf1,
4577 out_src_opr(buf2, sizeof(buf2), po, &po->operand[1],
4582 assert_operand_cnt(2);
4583 po->operand[1].lmod = OPLM_DWORD; // always
4584 fprintf(fout, " %s = %s;",
4585 out_dst_opr(buf1, sizeof(buf1), po, &po->operand[0]),
4586 out_src_opr(buf2, sizeof(buf2), po, &po->operand[1],
4591 assert_operand_cnt(2);
4592 fprintf(fout, " %s = %s;",
4593 out_dst_opr(buf1, sizeof(buf1), po, &po->operand[0]),
4594 out_src_opr_u32(buf2, sizeof(buf2), po, &po->operand[1]));
4598 assert_operand_cnt(2);
4599 switch (po->operand[1].lmod) {
4601 strcpy(buf3, "(s8)");
4604 strcpy(buf3, "(s16)");
4607 ferr(po, "invalid src lmod: %d\n", po->operand[1].lmod);
4609 fprintf(fout, " %s = %s;",
4610 out_dst_opr(buf1, sizeof(buf1), po, &po->operand[0]),
4611 out_src_opr(buf2, sizeof(buf2), po, &po->operand[1],
4616 assert_operand_cnt(2);
4617 propagate_lmod(po, &po->operand[0], &po->operand[1]);
4618 fprintf(fout, " tmp = %s;",
4619 out_src_opr(buf1, sizeof(buf1), po, &po->operand[0], "", 0));
4620 fprintf(fout, " %s = %s;",
4621 out_dst_opr(buf1, sizeof(buf1), po, &po->operand[0]),
4622 out_src_opr(buf2, sizeof(buf2), po, &po->operand[1],
4623 default_cast_to(buf3, sizeof(buf3), &po->operand[0]), 0));
4624 fprintf(fout, " %s = %stmp;",
4625 out_dst_opr(buf1, sizeof(buf1), po, &po->operand[1]),
4626 default_cast_to(buf3, sizeof(buf3), &po->operand[1]));
4627 snprintf(g_comment, sizeof(g_comment), "xchg");
4631 assert_operand_cnt(1);
4632 out_dst_opr(buf1, sizeof(buf1), po, &po->operand[0]);
4633 fprintf(fout, " %s = ~%s;", buf1, buf1);
4637 assert_operand_cnt(2);
4638 fprintf(fout, " %s = (s32)%s >> 31;",
4639 out_dst_opr(buf1, sizeof(buf1), po, &po->operand[0]),
4640 out_src_opr_u32(buf2, sizeof(buf2), po, &po->operand[1]));
4641 strcpy(g_comment, "cdq");
4645 assert_operand_cnt(3);
4646 if (po->flags & OPF_REP) {
4651 fprintf(fout, " eax = %sesi; esi %c= %d;",
4652 lmod_cast_u_ptr(po, po->operand[0].lmod),
4653 (po->flags & OPF_DF) ? '-' : '+',
4654 lmod_bytes(po, po->operand[0].lmod));
4655 strcpy(g_comment, "lods");
4660 assert_operand_cnt(3);
4661 if (po->flags & OPF_REP) {
4662 fprintf(fout, " for (; ecx != 0; ecx--, edi %c= %d)\n",
4663 (po->flags & OPF_DF) ? '-' : '+',
4664 lmod_bytes(po, po->operand[0].lmod));
4665 fprintf(fout, " %sedi = eax;",
4666 lmod_cast_u_ptr(po, po->operand[0].lmod));
4667 strcpy(g_comment, "rep stos");
4670 fprintf(fout, " %sedi = eax; edi %c= %d;",
4671 lmod_cast_u_ptr(po, po->operand[0].lmod),
4672 (po->flags & OPF_DF) ? '-' : '+',
4673 lmod_bytes(po, po->operand[0].lmod));
4674 strcpy(g_comment, "stos");
4679 assert_operand_cnt(3);
4680 j = lmod_bytes(po, po->operand[0].lmod);
4681 strcpy(buf1, lmod_cast_u_ptr(po, po->operand[0].lmod));
4682 l = (po->flags & OPF_DF) ? '-' : '+';
4683 if (po->flags & OPF_REP) {
4685 " for (; ecx != 0; ecx--, edi %c= %d, esi %c= %d)\n",
4688 " %sedi = %sesi;", buf1, buf1);
4689 strcpy(g_comment, "rep movs");
4692 fprintf(fout, " %sedi = %sesi; edi %c= %d; esi %c= %d;",
4693 buf1, buf1, l, j, l, j);
4694 strcpy(g_comment, "movs");
4699 // repe ~ repeat while ZF=1
4700 assert_operand_cnt(3);
4701 j = lmod_bytes(po, po->operand[0].lmod);
4702 strcpy(buf1, lmod_cast_u_ptr(po, po->operand[0].lmod));
4703 l = (po->flags & OPF_DF) ? '-' : '+';
4704 if (po->flags & OPF_REP) {
4706 " for (; ecx != 0; ecx--) {\n");
4707 if (pfomask & (1 << PFO_C)) {
4710 " cond_c = %sesi < %sedi;\n", buf1, buf1);
4711 pfomask &= ~(1 << PFO_C);
4714 " cond_z = (%sesi == %sedi); esi %c= %d, edi %c= %d;\n",
4715 buf1, buf1, l, j, l, j);
4717 " if (cond_z %s 0) break;\n",
4718 (po->flags & OPF_REPZ) ? "==" : "!=");
4721 snprintf(g_comment, sizeof(g_comment), "rep%s cmps",
4722 (po->flags & OPF_REPZ) ? "e" : "ne");
4726 " cond_z = (%sesi == %sedi); esi %c= %d; edi %c= %d;",
4727 buf1, buf1, l, j, l, j);
4728 strcpy(g_comment, "cmps");
4730 pfomask &= ~(1 << PFO_Z);
4731 last_arith_dst = NULL;
4732 delayed_flag_op = NULL;
4736 // only does ZF (for now)
4737 // repe ~ repeat while ZF=1
4738 assert_operand_cnt(3);
4739 j = lmod_bytes(po, po->operand[0].lmod);
4740 l = (po->flags & OPF_DF) ? '-' : '+';
4741 if (po->flags & OPF_REP) {
4743 " for (; ecx != 0; ecx--) {\n");
4745 " cond_z = (%seax == %sedi); edi %c= %d;\n",
4746 lmod_cast_u(po, po->operand[0].lmod),
4747 lmod_cast_u_ptr(po, po->operand[0].lmod), l, j);
4749 " if (cond_z %s 0) break;\n",
4750 (po->flags & OPF_REPZ) ? "==" : "!=");
4753 snprintf(g_comment, sizeof(g_comment), "rep%s scas",
4754 (po->flags & OPF_REPZ) ? "e" : "ne");
4757 fprintf(fout, " cond_z = (%seax == %sedi); edi %c= %d;",
4758 lmod_cast_u(po, po->operand[0].lmod),
4759 lmod_cast_u_ptr(po, po->operand[0].lmod), l, j);
4760 strcpy(g_comment, "scas");
4762 pfomask &= ~(1 << PFO_Z);
4763 last_arith_dst = NULL;
4764 delayed_flag_op = NULL;
4767 // arithmetic w/flags
4770 propagate_lmod(po, &po->operand[0], &po->operand[1]);
4773 assert_operand_cnt(2);
4774 fprintf(fout, " %s %s= %s;",
4775 out_dst_opr(buf1, sizeof(buf1), po, &po->operand[0]),
4777 out_src_opr_u32(buf2, sizeof(buf2), po, &po->operand[1]));
4778 output_std_flags(fout, po, &pfomask, buf1);
4779 last_arith_dst = &po->operand[0];
4780 delayed_flag_op = NULL;
4785 assert_operand_cnt(2);
4786 out_dst_opr(buf1, sizeof(buf1), po, &po->operand[0]);
4787 if (pfomask & (1 << PFO_C)) {
4788 if (po->operand[1].type == OPT_CONST) {
4789 l = lmod_bytes(po, po->operand[0].lmod) * 8;
4790 j = po->operand[1].val;
4793 if (po->op == OP_SHL)
4797 fprintf(fout, " cond_c = (%s >> %d) & 1;\n",
4801 ferr(po, "zero shift?\n");
4805 pfomask &= ~(1 << PFO_C);
4807 fprintf(fout, " %s %s= %s;", buf1, op_to_c(po),
4808 out_src_opr_u32(buf2, sizeof(buf2), po, &po->operand[1]));
4809 output_std_flags(fout, po, &pfomask, buf1);
4810 last_arith_dst = &po->operand[0];
4811 delayed_flag_op = NULL;
4815 assert_operand_cnt(2);
4816 out_dst_opr(buf1, sizeof(buf1), po, &po->operand[0]);
4817 fprintf(fout, " %s = %s%s >> %s;", buf1,
4818 lmod_cast_s(po, po->operand[0].lmod), buf1,
4819 out_src_opr_u32(buf2, sizeof(buf2), po, &po->operand[1]));
4820 output_std_flags(fout, po, &pfomask, buf1);
4821 last_arith_dst = &po->operand[0];
4822 delayed_flag_op = NULL;
4826 assert_operand_cnt(3);
4827 propagate_lmod(po, &po->operand[0], &po->operand[1]);
4828 l = lmod_bytes(po, po->operand[0].lmod) * 8;
4829 out_dst_opr(buf1, sizeof(buf1), po, &po->operand[0]);
4830 out_src_opr_u32(buf2, sizeof(buf2), po, &po->operand[1]);
4831 out_src_opr_u32(buf3, sizeof(buf3), po, &po->operand[2]);
4832 fprintf(fout, " %s >>= %s; %s |= %s << (%d - %s);",
4833 buf1, buf3, buf1, buf2, l, buf3);
4834 strcpy(g_comment, "shrd");
4835 output_std_flags(fout, po, &pfomask, buf1);
4836 last_arith_dst = &po->operand[0];
4837 delayed_flag_op = NULL;
4842 assert_operand_cnt(2);
4843 out_dst_opr(buf1, sizeof(buf1), po, &po->operand[0]);
4844 if (po->operand[1].type == OPT_CONST) {
4845 j = po->operand[1].val;
4846 j %= lmod_bytes(po, po->operand[0].lmod) * 8;
4847 fprintf(fout, po->op == OP_ROL ?
4848 " %s = (%s << %d) | (%s >> %d);" :
4849 " %s = (%s >> %d) | (%s << %d);",
4850 buf1, buf1, j, buf1,
4851 lmod_bytes(po, po->operand[0].lmod) * 8 - j);
4855 output_std_flags(fout, po, &pfomask, buf1);
4856 last_arith_dst = &po->operand[0];
4857 delayed_flag_op = NULL;
4862 assert_operand_cnt(2);
4863 out_dst_opr(buf1, sizeof(buf1), po, &po->operand[0]);
4864 l = lmod_bytes(po, po->operand[0].lmod) * 8;
4865 if (po->operand[1].type == OPT_CONST) {
4866 j = po->operand[1].val % l;
4868 ferr(po, "zero rotate\n");
4869 fprintf(fout, " tmp = (%s >> %d) & 1;\n",
4870 buf1, (po->op == OP_RCL) ? (l - j) : (j - 1));
4871 if (po->op == OP_RCL) {
4873 " %s = (%s << %d) | (cond_c << %d)",
4874 buf1, buf1, j, j - 1);
4876 fprintf(fout, " | (%s >> %d)", buf1, l + 1 - j);
4880 " %s = (%s >> %d) | (cond_c << %d)",
4881 buf1, buf1, j, l - j);
4883 fprintf(fout, " | (%s << %d)", buf1, l + 1 - j);
4885 fprintf(fout, ";\n");
4886 fprintf(fout, " cond_c = tmp;");
4890 strcpy(g_comment, (po->op == OP_RCL) ? "rcl" : "rcr");
4891 output_std_flags(fout, po, &pfomask, buf1);
4892 last_arith_dst = &po->operand[0];
4893 delayed_flag_op = NULL;
4897 assert_operand_cnt(2);
4898 propagate_lmod(po, &po->operand[0], &po->operand[1]);
4899 if (IS(opr_name(po, 0), opr_name(po, 1))) {
4900 // special case for XOR
4901 if (pfomask & (1 << PFO_BE)) { // weird, but it happens..
4902 fprintf(fout, " cond_be = 1;\n");
4903 pfomask &= ~(1 << PFO_BE);
4905 fprintf(fout, " %s = 0;",
4906 out_dst_opr(buf1, sizeof(buf1), po, &po->operand[0]));
4907 last_arith_dst = &po->operand[0];
4908 delayed_flag_op = NULL;
4914 assert_operand_cnt(2);
4915 propagate_lmod(po, &po->operand[0], &po->operand[1]);
4916 if (pfomask & (1 << PFO_C)) {
4917 out_src_opr_u32(buf1, sizeof(buf1), po, &po->operand[0]);
4918 out_src_opr_u32(buf2, sizeof(buf2), po, &po->operand[1]);
4919 if (po->operand[0].lmod == OPLM_DWORD) {
4920 fprintf(fout, " tmp64 = (u64)%s + %s;\n", buf1, buf2);
4921 fprintf(fout, " cond_c = tmp64 >> 32;\n");
4922 fprintf(fout, " %s = (u32)tmp64;",
4923 out_dst_opr(buf1, sizeof(buf1), po, &po->operand[0]));
4924 strcat(g_comment, "add64");
4927 fprintf(fout, " cond_c = ((u32)%s + %s) >> %d;\n",
4928 buf1, buf2, lmod_bytes(po, po->operand[0].lmod) * 8);
4929 fprintf(fout, " %s += %s;",
4930 out_dst_opr(buf1, sizeof(buf1), po, &po->operand[0]),
4933 pfomask &= ~(1 << PFO_C);
4934 output_std_flags(fout, po, &pfomask, buf1);
4935 last_arith_dst = &po->operand[0];
4936 delayed_flag_op = NULL;
4942 assert_operand_cnt(2);
4943 propagate_lmod(po, &po->operand[0], &po->operand[1]);
4944 if (pfomask & ~((1 << PFO_Z) | (1 << PFO_S))) {
4945 for (j = 0; j <= PFO_LE; j++) {
4946 if (!(pfomask & (1 << j)))
4948 if (j == PFO_Z || j == PFO_S)
4951 out_cmp_for_cc(buf1, sizeof(buf1), po, j, 0);
4952 fprintf(fout, " cond_%s = %s;\n",
4953 parsed_flag_op_names[j], buf1);
4954 pfomask &= ~(1 << j);
4961 assert_operand_cnt(2);
4962 propagate_lmod(po, &po->operand[0], &po->operand[1]);
4963 out_dst_opr(buf1, sizeof(buf1), po, &po->operand[0]);
4964 if (po->op == OP_SBB
4965 && IS(po->operand[0].name, po->operand[1].name))
4967 // avoid use of unitialized var
4968 fprintf(fout, " %s = -cond_c;", buf1);
4969 // carry remains what it was
4970 pfomask &= ~(1 << PFO_C);
4973 fprintf(fout, " %s %s= %s + cond_c;", buf1, op_to_c(po),
4974 out_src_opr_u32(buf2, sizeof(buf2), po, &po->operand[1]));
4976 output_std_flags(fout, po, &pfomask, buf1);
4977 last_arith_dst = &po->operand[0];
4978 delayed_flag_op = NULL;
4982 assert_operand_cnt(2);
4983 out_src_opr_u32(buf2, sizeof(buf2), po, &po->operand[1]);
4984 fprintf(fout, " %s = %s ? __builtin_ffs(%s) - 1 : 0;",
4985 out_dst_opr(buf1, sizeof(buf1), po, &po->operand[0]),
4987 output_std_flags(fout, po, &pfomask, buf1);
4988 last_arith_dst = &po->operand[0];
4989 delayed_flag_op = NULL;
4990 strcat(g_comment, "bsf");
4994 if (pfomask & ~(PFOB_S | PFOB_S | PFOB_C)) {
4995 for (j = 0; j <= PFO_LE; j++) {
4996 if (!(pfomask & (1 << j)))
4998 if (j == PFO_Z || j == PFO_S || j == PFO_C)
5001 out_cmp_for_cc(buf1, sizeof(buf1), po, j, 0);
5002 fprintf(fout, " cond_%s = %s;\n",
5003 parsed_flag_op_names[j], buf1);
5004 pfomask &= ~(1 << j);
5010 if (pfomask & (1 << PFO_C))
5011 // carry is unaffected by inc/dec.. wtf?
5012 ferr(po, "carry propagation needed\n");
5014 out_dst_opr(buf1, sizeof(buf1), po, &po->operand[0]);
5015 if (po->operand[0].type == OPT_REG) {
5016 strcpy(buf2, po->op == OP_INC ? "++" : "--");
5017 fprintf(fout, " %s%s;", buf1, buf2);
5020 strcpy(buf2, po->op == OP_INC ? "+" : "-");
5021 fprintf(fout, " %s %s= 1;", buf1, buf2);
5023 output_std_flags(fout, po, &pfomask, buf1);
5024 last_arith_dst = &po->operand[0];
5025 delayed_flag_op = NULL;
5029 out_dst_opr(buf1, sizeof(buf1), po, &po->operand[0]);
5030 out_src_opr_u32(buf2, sizeof(buf2), po, &po->operand[0]);
5031 fprintf(fout, " %s = -%s%s;", buf1,
5032 lmod_cast_s(po, po->operand[0].lmod), buf2);
5033 last_arith_dst = &po->operand[0];
5034 delayed_flag_op = NULL;
5035 if (pfomask & (1 << PFO_C)) {
5036 fprintf(fout, "\n cond_c = (%s != 0);", buf1);
5037 pfomask &= ~(1 << PFO_C);
5042 if (po->operand_cnt == 2) {
5043 propagate_lmod(po, &po->operand[0], &po->operand[1]);
5046 if (po->operand_cnt == 3)
5047 ferr(po, "TODO imul3\n");
5050 assert_operand_cnt(1);
5051 switch (po->operand[0].lmod) {
5053 strcpy(buf1, po->op == OP_IMUL ? "(s64)(s32)" : "(u64)");
5054 fprintf(fout, " tmp64 = %seax * %s%s;\n", buf1, buf1,
5055 out_src_opr_u32(buf2, sizeof(buf2), po, &po->operand[0]));
5056 fprintf(fout, " edx = tmp64 >> 32;\n");
5057 fprintf(fout, " eax = tmp64;");
5060 strcpy(buf1, po->op == OP_IMUL ? "(s16)(s8)" : "(u16)(u8)");
5061 fprintf(fout, " LOWORD(eax) = %seax * %s;", buf1,
5062 out_src_opr(buf2, sizeof(buf2), po, &po->operand[0],
5066 ferr(po, "TODO: unhandled mul type\n");
5069 last_arith_dst = NULL;
5070 delayed_flag_op = NULL;
5075 assert_operand_cnt(1);
5076 if (po->operand[0].lmod != OPLM_DWORD)
5077 ferr(po, "unhandled lmod %d\n", po->operand[0].lmod);
5079 out_src_opr_u32(buf1, sizeof(buf1), po, &po->operand[0]);
5080 strcpy(buf2, lmod_cast(po, po->operand[0].lmod,
5081 po->op == OP_IDIV));
5082 switch (po->operand[0].lmod) {
5084 if (po->flags & OPF_32BIT)
5085 snprintf(buf3, sizeof(buf3), "%seax", buf2);
5087 fprintf(fout, " tmp64 = ((u64)edx << 32) | eax;\n");
5088 snprintf(buf3, sizeof(buf3), "%stmp64",
5089 (po->op == OP_IDIV) ? "(s64)" : "");
5091 if (po->operand[0].type == OPT_REG
5092 && po->operand[0].reg == xDX)
5094 fprintf(fout, " eax = %s / %s%s;", buf3, buf2, buf1);
5095 fprintf(fout, " edx = %s %% %s%s;\n", buf3, buf2, buf1);
5098 fprintf(fout, " edx = %s %% %s%s;\n", buf3, buf2, buf1);
5099 fprintf(fout, " eax = %s / %s%s;", buf3, buf2, buf1);
5103 ferr(po, "unhandled division type\n");
5105 last_arith_dst = NULL;
5106 delayed_flag_op = NULL;
5111 propagate_lmod(po, &po->operand[0], &po->operand[1]);
5113 for (j = 0; j < 8; j++) {
5114 if (pfomask & (1 << j)) {
5115 out_cmp_test(buf1, sizeof(buf1), po, j, 0);
5116 fprintf(fout, " cond_%s = %s;",
5117 parsed_flag_op_names[j], buf1);
5124 last_arith_dst = NULL;
5125 delayed_flag_op = po;
5129 // SETcc - should already be handled
5132 // note: we reuse OP_Jcc for SETcc, only flags differ
5134 fprintf(fout, "\n goto %s;", po->operand[0].name);
5138 fprintf(fout, " if (ecx == 0)\n");
5139 fprintf(fout, " goto %s;", po->operand[0].name);
5140 strcat(g_comment, "jecxz");
5144 assert_operand_cnt(1);
5145 last_arith_dst = NULL;
5146 delayed_flag_op = NULL;
5148 if (po->operand[0].type == OPT_REGMEM) {
5149 ret = sscanf(po->operand[0].name, "%[^[][%[^*]*4]",
5152 ferr(po, "parse failure for jmp '%s'\n",
5153 po->operand[0].name);
5154 fprintf(fout, " goto *jt_%s[%s];", buf1, buf2);
5157 else if (po->operand[0].type != OPT_LABEL)
5158 ferr(po, "unhandled jmp type\n");
5160 fprintf(fout, " goto %s;", po->operand[0].name);
5164 assert_operand_cnt(1);
5166 my_assert_not(pp, NULL);
5169 if (po->flags & OPF_CC) {
5170 // we treat conditional branch to another func
5171 // (yes such code exists..) as conditional tailcall
5173 fprintf(fout, " {\n");
5176 if (pp->is_fptr && !pp->is_arg) {
5177 fprintf(fout, "%s%s = %s;\n", buf3, pp->name,
5178 out_src_opr(buf1, sizeof(buf1), po, &po->operand[0],
5180 if (pp->is_unresolved)
5181 fprintf(fout, "%sunresolved_call(\"%s:%d\", %s);\n",
5182 buf3, asmfn, po->asmln, pp->name);
5185 fprintf(fout, "%s", buf3);
5186 if (strstr(pp->ret_type.name, "int64")) {
5187 if (po->flags & OPF_TAIL)
5188 ferr(po, "int64 and tail?\n");
5189 fprintf(fout, "tmp64 = ");
5191 else if (!IS(pp->ret_type.name, "void")) {
5192 if (po->flags & OPF_TAIL) {
5193 if (!IS(g_func_pp->ret_type.name, "void")) {
5194 fprintf(fout, "return ");
5195 if (g_func_pp->ret_type.is_ptr != pp->ret_type.is_ptr)
5196 fprintf(fout, "(%s)", g_func_pp->ret_type.name);
5199 else if (regmask & (1 << xAX)) {
5200 fprintf(fout, "eax = ");
5201 if (pp->ret_type.is_ptr)
5202 fprintf(fout, "(u32)");
5206 if (pp->name[0] == 0)
5207 ferr(po, "missing pp->name\n");
5208 fprintf(fout, "%s%s(", pp->name,
5209 pp->has_structarg ? "_sa" : "");
5211 if (po->flags & OPF_ATAIL) {
5212 if (pp->argc_stack != g_func_pp->argc_stack
5213 || (pp->argc_stack > 0
5214 && pp->is_stdcall != g_func_pp->is_stdcall))
5215 ferr(po, "incompatible tailcall\n");
5216 if (g_func_pp->has_retreg)
5217 ferr(po, "TODO: retreg+tailcall\n");
5219 for (arg = j = 0; arg < pp->argc; arg++) {
5221 fprintf(fout, ", ");
5224 if (pp->arg[arg].type.is_ptr)
5225 snprintf(cast, sizeof(cast), "(%s)",
5226 pp->arg[arg].type.name);
5228 if (pp->arg[arg].reg != NULL) {
5229 fprintf(fout, "%s%s", cast, pp->arg[arg].reg);
5233 for (; j < g_func_pp->argc; j++)
5234 if (g_func_pp->arg[j].reg == NULL)
5236 fprintf(fout, "%sa%d", cast, j + 1);
5241 for (arg = 0; arg < pp->argc; arg++) {
5243 fprintf(fout, ", ");
5246 if (pp->arg[arg].type.is_ptr)
5247 snprintf(cast, sizeof(cast), "(%s)",
5248 pp->arg[arg].type.name);
5250 if (pp->arg[arg].reg != NULL) {
5251 if (pp->arg[arg].type.is_retreg)
5252 fprintf(fout, "&%s", pp->arg[arg].reg);
5254 fprintf(fout, "%s%s", cast, pp->arg[arg].reg);
5259 tmp_op = pp->arg[arg].datap;
5261 ferr(po, "parsed_op missing for arg%d\n", arg);
5263 if (tmp_op->flags & OPF_VAPUSH) {
5264 fprintf(fout, "ap");
5266 else if (tmp_op->p_argpass != 0) {
5267 fprintf(fout, "a%d", tmp_op->p_argpass);
5269 else if (tmp_op->p_argnum != 0) {
5270 fprintf(fout, "%s%s", cast,
5271 saved_arg_name(buf1, sizeof(buf1),
5272 tmp_op->p_arggrp, tmp_op->p_argnum));
5276 out_src_opr(buf1, sizeof(buf1),
5277 tmp_op, &tmp_op->operand[0], cast, 0));
5281 fprintf(fout, ");");
5283 if (strstr(pp->ret_type.name, "int64")) {
5284 fprintf(fout, "\n");
5285 fprintf(fout, "%sedx = tmp64 >> 32;\n", buf3);
5286 fprintf(fout, "%seax = tmp64;", buf3);
5289 if (pp->is_unresolved) {
5290 snprintf(buf2, sizeof(buf2), " unresolved %dreg",
5292 strcat(g_comment, buf2);
5295 if (po->flags & OPF_TAIL) {
5297 if (i == opcnt - 1 || pp->is_noreturn)
5299 else if (IS(pp->ret_type.name, "void"))
5301 else if (IS(g_func_pp->ret_type.name, "void"))
5303 // else already handled as 'return f()'
5306 if (!IS(g_func_pp->ret_type.name, "void")) {
5307 ferr(po, "int func -> void func tailcall?\n");
5310 fprintf(fout, "\n%sreturn;", buf3);
5311 strcat(g_comment, " ^ tailcall");
5315 strcat(g_comment, " tailcall");
5317 if (pp->is_noreturn)
5318 strcat(g_comment, " noreturn");
5319 if ((po->flags & OPF_ATAIL) && pp->argc_stack > 0)
5320 strcat(g_comment, " argframe");
5321 if (po->flags & OPF_CC)
5322 strcat(g_comment, " cond");
5324 if (po->flags & OPF_CC)
5325 fprintf(fout, "\n }");
5327 delayed_flag_op = NULL;
5328 last_arith_dst = NULL;
5332 if (g_func_pp->is_vararg)
5333 fprintf(fout, " va_end(ap);\n");
5334 if (g_func_pp->has_retreg) {
5335 for (arg = 0; arg < g_func_pp->argc; arg++)
5336 if (g_func_pp->arg[arg].type.is_retreg)
5337 fprintf(fout, " *r_%s = %s;\n",
5338 g_func_pp->arg[arg].reg, g_func_pp->arg[arg].reg);
5341 if (IS(g_func_pp->ret_type.name, "void")) {
5342 if (i != opcnt - 1 || label_pending)
5343 fprintf(fout, " return;");
5345 else if (g_func_pp->ret_type.is_ptr) {
5346 fprintf(fout, " return (%s)eax;",
5347 g_func_pp->ret_type.name);
5349 else if (IS(g_func_pp->ret_type.name, "__int64"))
5350 fprintf(fout, " return ((u64)edx << 32) | eax;");
5352 fprintf(fout, " return eax;");
5354 last_arith_dst = NULL;
5355 delayed_flag_op = NULL;
5359 out_src_opr_u32(buf1, sizeof(buf1), po, &po->operand[0]);
5360 if (po->p_argnum != 0) {
5361 // special case - saved func arg
5362 fprintf(fout, " %s = %s;",
5363 saved_arg_name(buf2, sizeof(buf2),
5364 po->p_arggrp, po->p_argnum), buf1);
5367 else if (po->flags & OPF_RSAVE) {
5368 fprintf(fout, " s_%s = %s;", buf1, buf1);
5371 else if (g_func_pp->is_userstack) {
5372 fprintf(fout, " *(--esp) = %s;", buf1);
5375 if (!(g_ida_func_attr & IDAFA_NORETURN))
5376 ferr(po, "stray push encountered\n");
5381 if (po->flags & OPF_RSAVE) {
5382 out_dst_opr(buf1, sizeof(buf1), po, &po->operand[0]);
5383 fprintf(fout, " %s = s_%s;", buf1, buf1);
5386 else if (po->datap != NULL) {
5389 out_dst_opr(buf1, sizeof(buf1), po, &po->operand[0]);
5390 fprintf(fout, " %s = %s;", buf1,
5391 out_src_opr(buf2, sizeof(buf2),
5392 tmp_op, &tmp_op->operand[0],
5393 default_cast_to(buf3, sizeof(buf3), &po->operand[0]), 0));
5396 else if (g_func_pp->is_userstack) {
5397 fprintf(fout, " %s = *esp++;",
5398 out_dst_opr(buf1, sizeof(buf1), po, &po->operand[0]));
5402 ferr(po, "stray pop encountered\n");
5411 strcpy(g_comment, "(emms)");
5416 ferr(po, "unhandled op type %d, flags %x\n",
5421 if (g_comment[0] != 0) {
5422 char *p = g_comment;
5423 while (my_isblank(*p))
5425 fprintf(fout, " // %s", p);
5430 fprintf(fout, "\n");
5432 // some sanity checking
5433 if (po->flags & OPF_REP) {
5434 if (po->op != OP_STOS && po->op != OP_MOVS
5435 && po->op != OP_CMPS && po->op != OP_SCAS)
5436 ferr(po, "unexpected rep\n");
5437 if (!(po->flags & (OPF_REPZ|OPF_REPNZ))
5438 && (po->op == OP_CMPS || po->op == OP_SCAS))
5439 ferr(po, "cmps/scas with plain rep\n");
5441 if ((po->flags & (OPF_REPZ|OPF_REPNZ))
5442 && po->op != OP_CMPS && po->op != OP_SCAS)
5443 ferr(po, "unexpected repz/repnz\n");
5446 ferr(po, "missed flag calc, pfomask=%x\n", pfomask);
5448 // see is delayed flag stuff is still valid
5449 if (delayed_flag_op != NULL && delayed_flag_op != po) {
5450 if (is_any_opr_modified(delayed_flag_op, po, 0))
5451 delayed_flag_op = NULL;
5454 if (last_arith_dst != NULL && last_arith_dst != &po->operand[0]) {
5455 if (is_opr_modified(last_arith_dst, po))
5456 last_arith_dst = NULL;
5462 if (g_stack_fsz && !g_stack_frame_used)
5463 fprintf(fout, " (void)sf;\n");
5465 fprintf(fout, "}\n\n");
5467 gen_x_cleanup(opcnt);
5470 static void gen_x_cleanup(int opcnt)
5474 for (i = 0; i < opcnt; i++) {
5475 struct label_ref *lr, *lr_del;
5477 lr = g_label_refs[i].next;
5478 while (lr != NULL) {
5483 g_label_refs[i].i = -1;
5484 g_label_refs[i].next = NULL;
5486 if (ops[i].op == OP_CALL) {
5488 proto_release(ops[i].pp);
5494 struct func_proto_dep;
5496 struct func_prototype {
5501 int has_ret:3; // -1, 0, 1: unresolved, no, yes
5502 unsigned int dep_resolved:1;
5503 unsigned int is_stdcall:1;
5504 struct func_proto_dep *dep_func;
5506 const struct parsed_proto *pp; // seed pp, if any
5509 struct func_proto_dep {
5511 struct func_prototype *proto;
5512 int regmask_live; // .. at the time of call
5513 unsigned int ret_dep:1; // return from this is caller's return
5516 static struct func_prototype *hg_fp;
5517 static int hg_fp_cnt;
5519 static struct scanned_var {
5521 enum opr_lenmod lmod;
5522 unsigned int is_seeded:1;
5523 unsigned int is_c_str:1;
5524 const struct parsed_proto *pp; // seed pp, if any
5526 static int hg_var_cnt;
5528 static void output_hdr_fp(FILE *fout, const struct func_prototype *fp,
5531 static struct func_proto_dep *hg_fp_find_dep(struct func_prototype *fp,
5536 for (i = 0; i < fp->dep_func_cnt; i++)
5537 if (IS(fp->dep_func[i].name, name))
5538 return &fp->dep_func[i];
5543 static void hg_fp_add_dep(struct func_prototype *fp, const char *name)
5546 if (hg_fp_find_dep(fp, name))
5549 if ((fp->dep_func_cnt & 0xff) == 0) {
5550 fp->dep_func = realloc(fp->dep_func,
5551 sizeof(fp->dep_func[0]) * (fp->dep_func_cnt + 0x100));
5552 my_assert_not(fp->dep_func, NULL);
5553 memset(&fp->dep_func[fp->dep_func_cnt], 0,
5554 sizeof(fp->dep_func[0]) * 0x100);
5556 fp->dep_func[fp->dep_func_cnt].name = strdup(name);
5560 static int hg_fp_cmp_name(const void *p1_, const void *p2_)
5562 const struct func_prototype *p1 = p1_, *p2 = p2_;
5563 return strcmp(p1->name, p2->name);
5567 static int hg_fp_cmp_id(const void *p1_, const void *p2_)
5569 const struct func_prototype *p1 = p1_, *p2 = p2_;
5570 return p1->id - p2->id;
5574 // recursive register dep pass
5575 // - track saved regs (part 2)
5576 // - try to figure out arg-regs
5577 // - calculate reg deps
5578 static void gen_hdr_dep_pass(int i, int opcnt, unsigned char *cbits,
5579 struct func_prototype *fp, int regmask_save, int regmask_dst,
5580 int *regmask_dep, int *has_ret)
5582 struct func_proto_dep *dep;
5583 struct parsed_op *po;
5584 int from_caller = 0;
5590 for (; i < opcnt; i++)
5592 if (cbits[i >> 3] & (1 << (i & 7)))
5594 cbits[i >> 3] |= (1 << (i & 7));
5598 if ((po->flags & OPF_JMP) && po->op != OP_CALL) {
5599 if (po->btj != NULL) {
5601 for (j = 0; j < po->btj->count; j++) {
5602 gen_hdr_dep_pass(po->btj->d[j].bt_i, opcnt, cbits, fp,
5603 regmask_save, regmask_dst, regmask_dep, has_ret);
5609 ferr(po, "dead branch\n");
5613 if (po->flags & OPF_CJMP) {
5614 gen_hdr_dep_pass(po->bt_i, opcnt, cbits, fp,
5615 regmask_save, regmask_dst, regmask_dep, has_ret);
5623 if (po->flags & OPF_FARG)
5624 /* (just calculate register deps) */;
5625 else if (po->op == OP_PUSH && po->operand[0].type == OPT_REG)
5627 reg = po->operand[0].reg;
5629 ferr(po, "reg not set for push?\n");
5631 if (po->flags & OPF_RSAVE) {
5632 regmask_save |= 1 << reg;
5635 if (po->flags & OPF_DONE)
5639 ret = scan_for_pop(i + 1, opcnt,
5640 po->operand[0].name, i + opcnt * 2, 0, &depth, 0);
5642 regmask_save |= 1 << reg;
5643 po->flags |= OPF_RMD;
5644 scan_for_pop(i + 1, opcnt,
5645 po->operand[0].name, i + opcnt * 3, 0, &depth, 1);
5649 else if (po->flags & OPF_RMD)
5651 else if (po->op == OP_CALL) {
5652 po->regmask_dst |= 1 << xAX;
5654 dep = hg_fp_find_dep(fp, po->operand[0].name);
5656 dep->regmask_live = regmask_save | regmask_dst;
5658 else if (po->op == OP_RET) {
5659 if (po->operand_cnt > 0) {
5661 if (fp->argc_stack >= 0
5662 && fp->argc_stack != po->operand[0].val / 4)
5663 ferr(po, "ret mismatch? (%d)\n", fp->argc_stack * 4);
5664 fp->argc_stack = po->operand[0].val / 4;
5668 if (*has_ret != 0 && (po->flags & OPF_TAIL)) {
5669 if (po->op == OP_CALL) {
5674 struct parsed_opr opr = { 0, };
5679 ret = resolve_origin(i, &opr, i + opcnt * 4, &j, &from_caller);
5682 if (ret == -1 && from_caller) {
5683 // unresolved eax - probably void func
5687 if (ops[j].op == OP_CALL) {
5688 dep = hg_fp_find_dep(fp, po->operand[0].name);
5699 l = regmask_save | regmask_dst;
5700 if (g_bp_frame && !(po->flags & OPF_EBP_S))
5703 l = po->regmask_src & ~l;
5706 fnote(po, "dep |= %04x, dst %04x, save %04x (f %x)\n",
5707 l, regmask_dst, regmask_save, po->flags);
5710 regmask_dst |= po->regmask_dst;
5712 if (po->flags & OPF_TAIL)
5717 static void gen_hdr(const char *funcn, int opcnt)
5719 int save_arg_vars[MAX_ARG_GRP] = { 0, };
5720 unsigned char cbits[MAX_OPS / 8];
5721 const struct parsed_proto *pp_c;
5722 struct parsed_proto *pp;
5723 struct func_prototype *fp;
5724 struct parsed_data *pd;
5725 struct parsed_op *po;
5726 const char *tmpname;
5727 int regmask_dummy = 0;
5729 int max_bp_offset = 0;
5733 if ((hg_fp_cnt & 0xff) == 0) {
5734 hg_fp = realloc(hg_fp, sizeof(hg_fp[0]) * (hg_fp_cnt + 0x100));
5735 my_assert_not(hg_fp, NULL);
5736 memset(hg_fp + hg_fp_cnt, 0, sizeof(hg_fp[0]) * 0x100);
5739 fp = &hg_fp[hg_fp_cnt];
5740 snprintf(fp->name, sizeof(fp->name), "%s", funcn);
5742 fp->argc_stack = -1;
5745 // perhaps already in seed header?
5746 fp->pp = proto_parse(g_fhdr, funcn, 1);
5747 if (fp->pp != NULL) {
5748 fp->argc_stack = fp->pp->argc_stack;
5749 fp->is_stdcall = fp->pp->is_stdcall;
5750 fp->regmask_dep = get_pp_arg_regmask(fp->pp);
5751 fp->has_ret = !IS(fp->pp->ret_type.name, "void");
5755 g_bp_frame = g_sp_frame = g_stack_fsz = 0;
5756 g_stack_frame_used = 0;
5759 // - handle ebp/esp frame, remove ops related to it
5760 scan_prologue_epilogue(opcnt);
5764 // - resolve all branches
5765 for (i = 0; i < opcnt; i++)
5771 if (po->flags & (OPF_RMD|OPF_DONE))
5774 if (po->op == OP_CALL) {
5775 tmpname = opr_name(po, 0);
5777 if (po->operand[0].type == OPT_LABEL) {
5778 hg_fp_add_dep(fp, tmpname);
5780 // perhaps a call to already known func?
5781 pp_c = proto_parse(g_fhdr, tmpname, 1);
5783 pp = proto_clone(pp_c);
5785 else if (po->datap != NULL) {
5786 pp = calloc(1, sizeof(*pp));
5787 my_assert_not(pp, NULL);
5789 ret = parse_protostr(po->datap, pp);
5791 ferr(po, "bad protostr supplied: %s\n", (char *)po->datap);
5795 if (pp != NULL && pp->is_noreturn)
5796 po->flags |= OPF_TAIL;
5802 if (!(po->flags & OPF_JMP) || po->op == OP_RET)
5805 if (po->operand[0].type == OPT_REGMEM) {
5806 pd = try_resolve_jumptab(i, opcnt);
5814 for (l = 0; l < opcnt; l++) {
5815 if (g_labels[l] != NULL
5816 && IS(po->operand[0].name, g_labels[l]))
5818 add_label_ref(&g_label_refs[l], i);
5824 if (po->bt_i != -1 || (po->flags & OPF_RMD))
5827 if (po->operand[0].type == OPT_LABEL)
5831 ferr(po, "unhandled branch\n");
5835 po->flags |= OPF_TAIL;
5836 if (i > 0 && ops[i - 1].op == OP_POP)
5837 po->flags |= OPF_ATAIL;
5842 // - remove dead labels
5843 // - handle push <const>/pop pairs
5844 for (i = 0; i < opcnt; i++)
5846 if (g_labels[i] != NULL && g_label_refs[i].i == -1) {
5852 if (po->flags & (OPF_RMD|OPF_DONE))
5855 if (po->op == OP_PUSH && po->operand[0].type == OPT_CONST)
5856 scan_for_pop_const(i, opcnt);
5860 // - process trivial calls
5861 for (i = 0; i < opcnt; i++)
5864 if (po->flags & (OPF_RMD|OPF_DONE))
5867 if (po->op == OP_CALL)
5869 pp = process_call_early(i, opcnt, &j);
5871 if (!(po->flags & OPF_ATAIL))
5872 // since we know the args, try to collect them
5873 if (collect_call_args_early(po, i, pp, ®mask_dummy) != 0)
5879 // commit esp adjust
5880 ops[j].flags |= OPF_RMD;
5881 if (ops[j].op != OP_POP)
5882 patch_esp_adjust(&ops[j], pp->argc_stack * 4);
5884 ops[j].flags |= OPF_DONE;
5887 po->flags |= OPF_DONE;
5893 // - track saved regs (simple)
5895 for (i = 0; i < opcnt; i++)
5898 if (po->flags & (OPF_RMD|OPF_DONE))
5901 if (po->op == OP_PUSH && po->operand[0].type == OPT_REG)
5903 ret = scan_for_pop_ret(i + 1, opcnt, po->operand[0].name, 0);
5905 // regmask_save |= 1 << po->operand[0].reg; // do it later
5906 po->flags |= OPF_RSAVE | OPF_RMD | OPF_DONE;
5907 scan_for_pop_ret(i + 1, opcnt, po->operand[0].name, OPF_RMD);
5910 else if (po->op == OP_CALL && !(po->flags & OPF_DONE))
5912 pp = process_call(i, opcnt);
5914 if (!pp->is_unresolved && !(po->flags & OPF_ATAIL)) {
5915 // since we know the args, collect them
5916 ret = collect_call_args(po, i, pp, ®mask_dummy, save_arg_vars,
5923 memset(cbits, 0, sizeof(cbits));
5927 gen_hdr_dep_pass(0, opcnt, cbits, fp, 0, 0, ®mask_dep, &has_ret);
5929 // find unreachable code - must be fixed in IDA
5930 for (i = 0; i < opcnt; i++)
5932 if (cbits[i >> 3] & (1 << (i & 7)))
5935 if (ops[i].op != OP_NOP)
5936 ferr(&ops[i], "unreachable code\n");
5939 if (has_ret == -1 && (regmask_dep & (1 << xAX)))
5942 for (i = 0; i < g_eqcnt; i++) {
5943 if (g_eqs[i].offset > max_bp_offset && g_eqs[i].offset < 4*32)
5944 max_bp_offset = g_eqs[i].offset;
5947 if (fp->argc_stack < 0) {
5948 max_bp_offset = (max_bp_offset + 3) & ~3;
5949 fp->argc_stack = max_bp_offset / 4;
5950 if ((g_ida_func_attr & IDAFA_BP_FRAME) && fp->argc_stack > 0)
5954 fp->regmask_dep = regmask_dep & ~(1 << xSP);
5955 fp->has_ret = has_ret;
5957 printf("// has_ret %d, regmask_dep %x\n",
5958 fp->has_ret, fp->regmask_dep);
5959 output_hdr_fp(stdout, fp, 1);
5960 if (IS(funcn, "sub_100073FD")) exit(1);
5963 gen_x_cleanup(opcnt);
5966 static void hg_fp_resolve_deps(struct func_prototype *fp)
5968 struct func_prototype fp_s;
5972 // this thing is recursive, so mark first..
5973 fp->dep_resolved = 1;
5975 for (i = 0; i < fp->dep_func_cnt; i++) {
5976 strcpy(fp_s.name, fp->dep_func[i].name);
5977 fp->dep_func[i].proto = bsearch(&fp_s, hg_fp, hg_fp_cnt,
5978 sizeof(hg_fp[0]), hg_fp_cmp_name);
5979 if (fp->dep_func[i].proto != NULL) {
5980 if (!fp->dep_func[i].proto->dep_resolved)
5981 hg_fp_resolve_deps(fp->dep_func[i].proto);
5983 dep = ~fp->dep_func[i].regmask_live
5984 & fp->dep_func[i].proto->regmask_dep;
5985 fp->regmask_dep |= dep;
5986 // printf("dep %s %s |= %x\n", fp->name,
5987 // fp->dep_func[i].name, dep);
5989 if (fp->has_ret == -1)
5990 fp->has_ret = fp->dep_func[i].proto->has_ret;
5995 static void output_hdr_fp(FILE *fout, const struct func_prototype *fp,
5998 const struct parsed_proto *pp;
5999 char *p, namebuf[NAMELEN];
6005 for (; count > 0; count--, fp++) {
6006 if (fp->has_ret == -1)
6007 fprintf(fout, "// ret unresolved\n");
6009 fprintf(fout, "// dep:");
6010 for (j = 0; j < fp->dep_func_cnt; j++) {
6011 fprintf(fout, " %s/", fp->dep_func[j].name);
6012 if (fp->dep_func[j].proto != NULL)
6013 fprintf(fout, "%04x/%d", fp->dep_func[j].proto->regmask_dep,
6014 fp->dep_func[j].proto->has_ret);
6016 fprintf(fout, "\n");
6019 p = strchr(fp->name, '@');
6021 memcpy(namebuf, fp->name, p - fp->name);
6022 namebuf[p - fp->name] = 0;
6030 pp = proto_parse(g_fhdr, name, 1);
6031 if (pp != NULL && pp->is_include)
6034 if (fp->pp != NULL) {
6035 // prefer fp for common style,
6036 // only use output_pp if args are complex
6037 for (j = 0; j < fp->pp->argc; j++) {
6038 if (fp->pp->arg[j].fptr != NULL)
6041 if (j != fp->pp->argc) {
6042 output_pp(fout, fp->pp, OPP_ALIGN);
6043 fprintf(fout, ";\n");
6048 regmask_dep = fp->regmask_dep;
6049 argc_stack = fp->argc_stack;
6051 fprintf(fout, "%-5s", fp->pp ? fp->pp->ret_type.name :
6052 (fp->has_ret ? "int" : "void"));
6053 if (regmask_dep && (fp->is_stdcall || argc_stack == 0)
6054 && (regmask_dep & ~((1 << xCX) | (1 << xDX))) == 0)
6056 fprintf(fout, " __fastcall ");
6057 if (!(regmask_dep & (1 << xDX)) && argc_stack == 0)
6063 else if (regmask_dep && !fp->is_stdcall) {
6064 fprintf(fout, "/*__usercall*/ ");
6066 else if (regmask_dep) {
6067 fprintf(fout, "/*__userpurge*/ ");
6069 else if (fp->is_stdcall)
6070 fprintf(fout, " __stdcall ");
6072 fprintf(fout, " __cdecl ");
6074 fprintf(fout, "%s(", name);
6077 for (j = 0; j < xSP; j++) {
6078 if (regmask_dep & (1 << j)) {
6081 fprintf(fout, ", ");
6083 fprintf(fout, "%s", fp->pp->arg[arg - 1].type.name);
6085 fprintf(fout, "int");
6086 fprintf(fout, " a%d/*<%s>*/", arg, regs_r32[j]);
6090 for (j = 0; j < argc_stack; j++) {
6093 fprintf(fout, ", ");
6094 if (fp->pp != NULL) {
6095 fprintf(fout, "%s", fp->pp->arg[arg - 1].type.name);
6096 if (!fp->pp->arg[arg - 1].type.is_ptr)
6100 fprintf(fout, "int ");
6101 fprintf(fout, "a%d", arg);
6104 fprintf(fout, ");\n");
6108 static void output_hdr(FILE *fout)
6110 static const char *lmod_c_names[] = {
6111 [OPLM_UNSPEC] = "???",
6112 [OPLM_BYTE] = "uint8_t",
6113 [OPLM_WORD] = "uint16_t",
6114 [OPLM_DWORD] = "uint32_t",
6115 [OPLM_QWORD] = "uint64_t",
6117 const struct scanned_var *var;
6118 char line[256] = { 0, };
6122 qsort(hg_fp, hg_fp_cnt, sizeof(hg_fp[0]), hg_fp_cmp_name);
6123 for (i = 0; i < hg_fp_cnt; i++)
6124 hg_fp_resolve_deps(&hg_fp[i]);
6126 // note: messes up .proto ptr, don't use
6127 //qsort(hg_fp, hg_fp_cnt, sizeof(hg_fp[0]), hg_fp_cmp_id);
6130 for (i = 0; i < hg_var_cnt; i++) {
6133 if (var->pp != NULL && var->pp->is_fptr) {
6134 fprintf(fout, "extern ");
6135 output_pp(fout, var->pp, 0);
6138 else if (var->is_c_str)
6139 fprintf(fout, "extern %-8s %s[];", "char", var->name);
6141 fprintf(fout, "extern %-8s %s;",
6142 lmod_c_names[var->lmod], var->name);
6145 fprintf(fout, " // seeded");
6146 fprintf(fout, "\n");
6149 fprintf(fout, "\n");
6151 // output function prototypes
6152 output_hdr_fp(fout, hg_fp, hg_fp_cnt);
6154 // include passthrough
6155 fprintf(fout, "\n// for translate\n");
6158 while (fgets(line, sizeof(line), g_fhdr)) {
6159 if (IS_START(line, "//#"))
6160 fwrite(line, 1, strlen(line), fout);
6164 // read a line, truncating it if it doesn't fit
6165 static char *my_fgets(char *s, size_t size, FILE *stream)
6175 ret = fgets(s, size, stream);
6176 if (ret != NULL && p >= 0 && s[p] != 0 && s[p] != '\n') {
6177 p = sizeof(buf) - 2;
6180 ret2 = fgets(buf, sizeof(buf), stream);
6182 while (ret2 != NULL && buf[p] != 0 && buf[p] != '\n');
6188 // '=' needs special treatment
6190 static char *next_word_s(char *w, size_t wsize, char *s)
6199 for (i = 1; i < wsize - 1; i++) {
6201 printf("warning: missing closing quote: \"%s\"\n", s);
6210 for (; i < wsize - 1; i++) {
6211 if (s[i] == 0 || my_isblank(s[i]) || (s[i] == '=' && i > 0))
6217 if (s[i] != 0 && !my_isblank(s[i]) && s[i] != '=')
6218 printf("warning: '%s' truncated\n", w);
6223 static void scan_variables(FILE *fasm)
6225 struct scanned_var *var;
6226 char line[256] = { 0, };
6234 // skip to next data section
6235 while (my_fgets(line, sizeof(line), fasm))
6240 if (*p == 0 || *p == ';')
6243 p = sskip(next_word_s(words[0], sizeof(words[0]), p));
6244 if (*p == 0 || *p == ';')
6247 if (*p != 's' || !IS_START(p, "segment para public"))
6253 if (p == NULL || !IS_START(p, "segment para public"))
6257 if (!IS_START(p, "'DATA'"))
6261 while (my_fgets(line, sizeof(line), fasm))
6270 if (*p == 0 || *p == ';')
6273 for (wordc = 0; wordc < ARRAY_SIZE(words); wordc++) {
6274 words[wordc][0] = 0;
6275 p = sskip(next_word_s(words[wordc], sizeof(words[0]), p));
6276 if (*p == 0 || *p == ';') {
6282 if (wordc == 2 && IS(words[1], "ends"))
6287 if ((hg_var_cnt & 0xff) == 0) {
6288 hg_vars = realloc(hg_vars, sizeof(hg_vars[0])
6289 * (hg_var_cnt + 0x100));
6290 my_assert_not(hg_vars, NULL);
6291 memset(hg_vars + hg_var_cnt, 0, sizeof(hg_vars[0]) * 0x100);
6294 var = &hg_vars[hg_var_cnt++];
6295 snprintf(var->name, sizeof(var->name), "%s", words[0]);
6297 // maybe already in seed header?
6298 var->pp = proto_parse(g_fhdr, var->name, 1);
6299 if (var->pp != NULL) {
6300 if (var->pp->is_fptr) {
6301 var->lmod = OPLM_DWORD;
6304 else if (var->pp->is_func)
6306 else if (!guess_lmod_from_c_type(&var->lmod, &var->pp->type))
6307 aerr("unhandled C type '%s' for '%s'\n",
6308 var->pp->type.name, var->name);
6314 if (IS(words[1], "dd"))
6315 var->lmod = OPLM_DWORD;
6316 else if (IS(words[1], "dw"))
6317 var->lmod = OPLM_WORD;
6318 else if (IS(words[1], "db")) {
6319 var->lmod = OPLM_BYTE;
6320 if (wordc >= 3 && (l = strlen(words[2])) > 4) {
6321 if (words[2][0] == '\'' && IS(words[2] + l - 2, ",0"))
6325 else if (IS(words[1], "dq"))
6326 var->lmod = OPLM_QWORD;
6327 //else if (IS(words[1], "dt"))
6329 aerr("type '%s' not known\n", words[1]);
6337 static void set_label(int i, const char *name)
6343 p = strchr(name, ':');
6347 if (g_labels[i] != NULL && !IS_START(g_labels[i], "algn_"))
6348 aerr("dupe label '%s' vs '%s'?\n", name, g_labels[i]);
6349 g_labels[i] = realloc(g_labels[i], len + 1);
6350 my_assert_not(g_labels[i], NULL);
6351 memcpy(g_labels[i], name, len);
6352 g_labels[i][len] = 0;
6361 static struct chunk_item *func_chunks;
6362 static int func_chunk_cnt;
6363 static int func_chunk_alloc;
6365 static void add_func_chunk(FILE *fasm, const char *name, int line)
6367 if (func_chunk_cnt >= func_chunk_alloc) {
6368 func_chunk_alloc *= 2;
6369 func_chunks = realloc(func_chunks,
6370 func_chunk_alloc * sizeof(func_chunks[0]));
6371 my_assert_not(func_chunks, NULL);
6373 func_chunks[func_chunk_cnt].fptr = ftell(fasm);
6374 func_chunks[func_chunk_cnt].name = strdup(name);
6375 func_chunks[func_chunk_cnt].asmln = line;
6379 static int cmp_chunks(const void *p1, const void *p2)
6381 const struct chunk_item *c1 = p1, *c2 = p2;
6382 return strcmp(c1->name, c2->name);
6385 static int cmpstringp(const void *p1, const void *p2)
6387 return strcmp(*(char * const *)p1, *(char * const *)p2);
6390 static void scan_ahead(FILE *fasm)
6400 oldpos = ftell(fasm);
6403 while (my_fgets(line, sizeof(line), fasm))
6414 // get rid of random tabs
6415 for (i = 0; line[i] != 0; i++)
6416 if (line[i] == '\t')
6419 if (p[2] == 'S' && IS_START(p, "; START OF FUNCTION CHUNK FOR "))
6422 next_word(words[0], sizeof(words[0]), p);
6423 if (words[0][0] == 0)
6424 aerr("missing name for func chunk?\n");
6426 add_func_chunk(fasm, words[0], asmln);
6428 else if (IS_START(p, "; sctend"))
6434 for (wordc = 0; wordc < ARRAY_SIZE(words); wordc++) {
6435 words[wordc][0] = 0;
6436 p = sskip(next_word_s(words[wordc], sizeof(words[0]), p));
6437 if (*p == 0 || *p == ';') {
6443 if (wordc == 2 && IS(words[1], "ends"))
6447 fseek(fasm, oldpos, SEEK_SET);
6451 int main(int argc, char *argv[])
6453 FILE *fout, *fasm, *frlist;
6454 struct parsed_data *pd = NULL;
6456 char **rlist = NULL;
6458 int rlist_alloc = 0;
6459 int func_chunks_used = 0;
6460 int func_chunks_sorted = 0;
6461 int func_chunk_i = -1;
6462 long func_chunk_ret = 0;
6463 int func_chunk_ret_ln = 0;
6464 int scanned_ahead = 0;
6466 char words[20][256];
6467 enum opr_lenmod lmod;
6468 char *sctproto = NULL;
6470 int pending_endp = 0;
6472 int skip_warned = 0;
6485 for (arg = 1; arg < argc; arg++) {
6486 if (IS(argv[arg], "-v"))
6488 else if (IS(argv[arg], "-rf"))
6489 g_allow_regfunc = 1;
6490 else if (IS(argv[arg], "-m"))
6492 else if (IS(argv[arg], "-hdr"))
6493 g_header_mode = g_quiet_pp = g_allow_regfunc = 1;
6498 if (argc < arg + 3) {
6499 printf("usage:\n%s [-v] [-rf] [-m] <.c> <.asm> <hdr.h> [rlist]*\n"
6500 "%s -hdr <out.h> <.asm> <seed.h> [rlist]*\n"
6502 " -hdr - header generation mode\n"
6503 " -rf - allow unannotated indirect calls\n"
6504 " -m - allow multiple .text sections\n"
6505 "[rlist] is a file with function names to skip,"
6513 asmfn = argv[arg++];
6514 fasm = fopen(asmfn, "r");
6515 my_assert_not(fasm, NULL);
6517 hdrfn = argv[arg++];
6518 g_fhdr = fopen(hdrfn, "r");
6519 my_assert_not(g_fhdr, NULL);
6522 rlist = malloc(rlist_alloc * sizeof(rlist[0]));
6523 my_assert_not(rlist, NULL);
6524 // needs special handling..
6525 rlist[rlist_len++] = "__alloca_probe";
6527 func_chunk_alloc = 32;
6528 func_chunks = malloc(func_chunk_alloc * sizeof(func_chunks[0]));
6529 my_assert_not(func_chunks, NULL);
6531 memset(words, 0, sizeof(words));
6533 for (; arg < argc; arg++) {
6534 frlist = fopen(argv[arg], "r");
6535 my_assert_not(frlist, NULL);
6537 while (my_fgets(line, sizeof(line), frlist)) {
6539 if (*p == 0 || *p == ';')
6542 if (IS_START(p, "#if 0")
6543 || (g_allow_regfunc && IS_START(p, "#if NO_REGFUNC")))
6547 else if (IS_START(p, "#endif"))
6554 p = next_word(words[0], sizeof(words[0]), p);
6555 if (words[0][0] == 0)
6558 if (rlist_len >= rlist_alloc) {
6559 rlist_alloc = rlist_alloc * 2 + 64;
6560 rlist = realloc(rlist, rlist_alloc * sizeof(rlist[0]));
6561 my_assert_not(rlist, NULL);
6563 rlist[rlist_len++] = strdup(words[0]);
6572 qsort(rlist, rlist_len, sizeof(rlist[0]), cmpstringp);
6574 fout = fopen(argv[arg_out], "w");
6575 my_assert_not(fout, NULL);
6578 g_eqs = malloc(eq_alloc * sizeof(g_eqs[0]));
6579 my_assert_not(g_eqs, NULL);
6581 for (i = 0; i < ARRAY_SIZE(g_label_refs); i++) {
6582 g_label_refs[i].i = -1;
6583 g_label_refs[i].next = NULL;
6587 scan_variables(fasm);
6589 while (my_fgets(line, sizeof(line), fasm))
6598 // get rid of random tabs
6599 for (i = 0; line[i] != 0; i++)
6600 if (line[i] == '\t')
6605 if (p[2] == '=' && IS_START(p, "; =============== S U B"))
6606 goto do_pending_endp; // eww..
6608 if (p[2] == 'A' && IS_START(p, "; Attributes:"))
6610 static const char *attrs[] = {
6619 // parse IDA's attribute-list comment
6620 g_ida_func_attr = 0;
6623 for (; *p != 0; p = sskip(p)) {
6624 for (i = 0; i < ARRAY_SIZE(attrs); i++) {
6625 if (!strncmp(p, attrs[i], strlen(attrs[i]))) {
6626 g_ida_func_attr |= 1 << i;
6627 p += strlen(attrs[i]);
6631 if (i == ARRAY_SIZE(attrs)) {
6632 anote("unparsed IDA attr: %s\n", p);
6635 if (IS(attrs[i], "fpd=")) {
6636 p = next_word(words[0], sizeof(words[0]), p);
6641 else if (p[2] == 'S' && IS_START(p, "; START OF FUNCTION CHUNK FOR "))
6644 next_word(words[0], sizeof(words[0]), p);
6645 if (words[0][0] == 0)
6646 aerr("missing name for func chunk?\n");
6648 if (!scanned_ahead) {
6649 add_func_chunk(fasm, words[0], asmln);
6650 func_chunks_sorted = 0;
6653 else if (p[2] == 'E' && IS_START(p, "; END OF FUNCTION CHUNK"))
6655 if (func_chunk_i >= 0) {
6656 if (func_chunk_i < func_chunk_cnt
6657 && IS(func_chunks[func_chunk_i].name, g_func))
6659 // move on to next chunk
6660 ret = fseek(fasm, func_chunks[func_chunk_i].fptr, SEEK_SET);
6662 aerr("seek failed for '%s' chunk #%d\n",
6663 g_func, func_chunk_i);
6664 asmln = func_chunks[func_chunk_i].asmln;
6668 if (func_chunk_ret == 0)
6669 aerr("no return from chunk?\n");
6670 fseek(fasm, func_chunk_ret, SEEK_SET);
6671 asmln = func_chunk_ret_ln;
6677 else if (p[2] == 'F' && IS_START(p, "; FUNCTION CHUNK AT ")) {
6678 func_chunks_used = 1;
6680 if (IS_START(g_func, "sub_")) {
6681 unsigned long addr = strtoul(p, NULL, 16);
6682 unsigned long f_addr = strtoul(g_func + 4, NULL, 16);
6683 if (addr > f_addr && !scanned_ahead) {
6684 anote("scan_ahead caused by '%s', addr %lx\n",
6688 func_chunks_sorted = 0;
6696 for (i = wordc; i < ARRAY_SIZE(words); i++)
6698 for (wordc = 0; wordc < ARRAY_SIZE(words); wordc++) {
6699 p = sskip(next_word_s(words[wordc], sizeof(words[0]), p));
6700 if (*p == 0 || *p == ';') {
6705 if (*p != 0 && *p != ';')
6706 aerr("too many words\n");
6708 // alow asm patches in comments
6710 if (IS_START(p, "; sctpatch:")) {
6712 if (*p == 0 || *p == ';')
6714 goto parse_words; // lame
6716 if (IS_START(p, "; sctproto:")) {
6717 sctproto = strdup(p + 11);
6719 else if (IS_START(p, "; sctend")) {
6728 awarn("wordc == 0?\n");
6732 // don't care about this:
6733 if (words[0][0] == '.'
6734 || IS(words[0], "include")
6735 || IS(words[0], "assume") || IS(words[1], "segment")
6736 || IS(words[0], "align"))
6742 // do delayed endp processing to collect switch jumptables
6744 if (in_func && !skip_func && !end && wordc >= 2
6745 && ((words[0][0] == 'd' && words[0][2] == 0)
6746 || (words[1][0] == 'd' && words[1][2] == 0)))
6749 if (words[1][0] == 'd' && words[1][2] == 0) {
6751 if (g_func_pd_cnt >= pd_alloc) {
6752 pd_alloc = pd_alloc * 2 + 16;
6753 g_func_pd = realloc(g_func_pd,
6754 sizeof(g_func_pd[0]) * pd_alloc);
6755 my_assert_not(g_func_pd, NULL);
6757 pd = &g_func_pd[g_func_pd_cnt];
6759 memset(pd, 0, sizeof(*pd));
6760 strcpy(pd->label, words[0]);
6761 pd->type = OPT_CONST;
6762 pd->lmod = lmod_from_directive(words[1]);
6768 anote("skipping alignment byte?\n");
6771 lmod = lmod_from_directive(words[0]);
6772 if (lmod != pd->lmod)
6773 aerr("lmod change? %d->%d\n", pd->lmod, lmod);
6776 if (pd->count_alloc < pd->count + wordc) {
6777 pd->count_alloc = pd->count_alloc * 2 + 14 + wordc;
6778 pd->d = realloc(pd->d, sizeof(pd->d[0]) * pd->count_alloc);
6779 my_assert_not(pd->d, NULL);
6781 for (; i < wordc; i++) {
6782 if (IS(words[i], "offset")) {
6783 pd->type = OPT_OFFSET;
6786 p = strchr(words[i], ',');
6789 if (pd->type == OPT_OFFSET)
6790 pd->d[pd->count].u.label = strdup(words[i]);
6792 pd->d[pd->count].u.val = parse_number(words[i]);
6793 pd->d[pd->count].bt_i = -1;
6799 if (in_func && !skip_func) {
6801 gen_hdr(g_func, pi);
6803 gen_func(fout, g_fhdr, g_func, pi);
6808 g_ida_func_attr = 0;
6812 func_chunks_used = 0;
6815 memset(&ops, 0, pi * sizeof(ops[0]));
6820 for (i = 0; i < g_func_pd_cnt; i++) {
6822 if (pd->type == OPT_OFFSET) {
6823 for (j = 0; j < pd->count; j++)
6824 free(pd->d[j].u.label);
6838 if (IS(words[1], "proc")) {
6840 aerr("proc '%s' while in_func '%s'?\n",
6843 if (bsearch(&p, rlist, rlist_len, sizeof(rlist[0]), cmpstringp))
6845 strcpy(g_func, words[0]);
6846 set_label(0, words[0]);
6851 if (IS(words[1], "endp"))
6854 aerr("endp '%s' while not in_func?\n", words[0]);
6855 if (!IS(g_func, words[0]))
6856 aerr("endp '%s' while in_func '%s'?\n",
6859 if ((g_ida_func_attr & IDAFA_THUNK) && pi == 1
6860 && ops[0].op == OP_JMP && ops[0].operand[0].had_ds)
6866 if (!skip_func && func_chunks_used) {
6867 // start processing chunks
6868 struct chunk_item *ci, key = { g_func, 0 };
6870 func_chunk_ret = ftell(fasm);
6871 func_chunk_ret_ln = asmln;
6872 if (!func_chunks_sorted) {
6873 qsort(func_chunks, func_chunk_cnt,
6874 sizeof(func_chunks[0]), cmp_chunks);
6875 func_chunks_sorted = 1;
6877 ci = bsearch(&key, func_chunks, func_chunk_cnt,
6878 sizeof(func_chunks[0]), cmp_chunks);
6880 aerr("'%s' needs chunks, but none found\n", g_func);
6881 func_chunk_i = ci - func_chunks;
6882 for (; func_chunk_i > 0; func_chunk_i--)
6883 if (!IS(func_chunks[func_chunk_i - 1].name, g_func))
6886 ret = fseek(fasm, func_chunks[func_chunk_i].fptr, SEEK_SET);
6888 aerr("seek failed for '%s' chunk #%d\n", g_func, func_chunk_i);
6889 asmln = func_chunks[func_chunk_i].asmln;
6897 if (wordc == 2 && IS(words[1], "ends")) {
6901 goto do_pending_endp;
6905 // scan for next text segment
6906 while (my_fgets(line, sizeof(line), fasm)) {
6909 if (*p == 0 || *p == ';')
6912 if (strstr(p, "segment para public 'CODE' use32"))
6919 p = strchr(words[0], ':');
6921 set_label(pi, words[0]);
6925 if (!in_func || skip_func) {
6926 if (!skip_warned && !skip_func && g_labels[pi] != NULL) {
6928 anote("skipping from '%s'\n", g_labels[pi]);
6932 g_labels[pi] = NULL;
6936 if (wordc > 1 && IS(words[1], "="))
6939 aerr("unhandled equ, wc=%d\n", wordc);
6940 if (g_eqcnt >= eq_alloc) {
6942 g_eqs = realloc(g_eqs, eq_alloc * sizeof(g_eqs[0]));
6943 my_assert_not(g_eqs, NULL);
6946 len = strlen(words[0]);
6947 if (len > sizeof(g_eqs[0].name) - 1)
6948 aerr("equ name too long: %d\n", len);
6949 strcpy(g_eqs[g_eqcnt].name, words[0]);
6951 if (!IS(words[3], "ptr"))
6952 aerr("unhandled equ\n");
6953 if (IS(words[2], "dword"))
6954 g_eqs[g_eqcnt].lmod = OPLM_DWORD;
6955 else if (IS(words[2], "word"))
6956 g_eqs[g_eqcnt].lmod = OPLM_WORD;
6957 else if (IS(words[2], "byte"))
6958 g_eqs[g_eqcnt].lmod = OPLM_BYTE;
6959 else if (IS(words[2], "qword"))
6960 g_eqs[g_eqcnt].lmod = OPLM_QWORD;
6962 aerr("bad lmod: '%s'\n", words[2]);
6964 g_eqs[g_eqcnt].offset = parse_number(words[4]);
6969 if (pi >= ARRAY_SIZE(ops))
6970 aerr("too many ops\n");
6972 parse_op(&ops[pi], words, wordc);
6974 if (sctproto != NULL) {
6975 if (ops[pi].op == OP_CALL || ops[pi].op == OP_JMP)
6976 ops[pi].datap = sctproto;
6992 // vim:ts=2:shiftwidth=2:expandtab