X-Git-Url: https://notaz.gp2x.de/cgi-bin/gitweb.cgi?a=blobdiff_plain;f=tools%2Ftranslate.c;h=de2da473ede18d721756b1cfa52d1c7d6766e146;hb=e627c4d0268fe9237839e68463382c2919ed59fd;hp=9922f32996c724ccd21f184db6ea1340dca71017;hpb=e7f5bc393577e0486826a9f913a2825d5c57a2f1;p=ia32rtools.git diff --git a/tools/translate.c b/tools/translate.c index 9922f32..de2da47 100644 --- a/tools/translate.c +++ b/tools/translate.c @@ -4,6 +4,15 @@ * * This work is licensed under the terms of 3-clause BSD license. * See COPYING file in the top-level directory. + * + * recognized asm hint comments: + * sctattr - function attributes (see code) + * sctend - force end of function/chunk + * sctpatch:
- replace current asm line with
+ * sctproto:
- prototype of ref'd function or struct
+ * sctref - variable is referenced, make global
+ * sctskip_start - start of skipped code chunk (inclusive)
+ * sctskip_end - end of skipped code chunk (inclusive)
*/
#define _GNU_SOURCE
@@ -60,6 +69,9 @@ enum op_flags {
OPF_DONE = (1 << 19), /* already fully handled by analysis */
OPF_PPUSH = (1 << 20), /* part of complex push-pop graph */
OPF_NOREGS = (1 << 21), /* don't track regs of this op */
+ OPF_FPUSH = (1 << 22), /* pushes x87 stack */
+ OPF_FPOP = (1 << 23), /* pops x87 stack */
+ OPF_FSHIFT = (1 << 24), /* x87 stack shift is actually needed */
};
enum op_op {
@@ -67,6 +79,8 @@ enum op_op {
OP_NOP,
OP_PUSH,
OP_POP,
+ OP_PUSHA,
+ OP_POPA,
OP_LEAVE,
OP_MOV,
OP_LEA,
@@ -76,6 +90,7 @@ enum op_op {
OP_NOT,
OP_XLAT,
OP_CDQ,
+ OP_BSWAP,
OP_LODS,
OP_STOS,
OP_MOVS,
@@ -116,11 +131,44 @@ enum op_op {
OP_LOOP,
OP_JCC,
OP_SCC,
- // x87
- // mmx
- OP_EMMS,
- // undefined
- OP_UD2,
+ // x87
+ OP_FLD,
+ OP_FILD,
+ OP_FLDc,
+ OP_FST,
+ OP_FIST,
+ OP_FADD,
+ OP_FDIV,
+ OP_FMUL,
+ OP_FSUB,
+ OP_FDIVR,
+ OP_FSUBR,
+ OP_FIADD,
+ OP_FIDIV,
+ OP_FIMUL,
+ OP_FISUB,
+ OP_FIDIVR,
+ OP_FISUBR,
+ OP_FCOM,
+ OP_FNSTSW,
+ OP_FCHS,
+ OP_FCOS,
+ OP_FPATAN,
+ OP_FPTAN,
+ OP_FSIN,
+ OP_FSQRT,
+ OP_FXCH,
+ OP_FYL2X,
+ // mmx
+ OP_EMMS,
+ // pseudo-ops for lib calls
+ OPP_ALLSHL,
+ OPP_ALLSHR,
+ OPP_FTOL,
+ OPP_CIPOW,
+ OPP_ABORT,
+ // undefined
+ OP_UD2,
};
enum opr_type {
@@ -187,10 +235,12 @@ struct parsed_op {
};
// datap:
-// OP_CALL - parser proto hint (str)
+// on start: function/data type hint (sctproto)
+// after analysis:
// (OPF_CC) - points to one of (OPF_FLAGS) that affects cc op
// OP_PUSH - points to OP_POP in complex push/pop graph
// OP_POP - points to OP_PUSH in simple push/pop pair
+// OP_FCOM - needed_status_word_bits | (is_z_check << 16)
struct parsed_equ {
char name[64];
@@ -227,6 +277,21 @@ enum ida_func_attr {
IDAFA_FPD = (1 << 5),
};
+enum sct_func_attr {
+ SCTFA_CLEAR_SF = (1 << 0), // clear stack frame
+ SCTFA_CLEAR_REGS = (1 << 1), // clear registers (mask)
+};
+
+enum x87_const {
+ X87_CONST_1 = 1,
+ X87_CONST_2T,
+ X87_CONST_2E,
+ X87_CONST_PI,
+ X87_CONST_LG2,
+ X87_CONST_LN2,
+ X87_CONST_Z,
+};
+
// note: limited to 32k due to p_argnext
#define MAX_OPS 4096
#define MAX_ARG_GRP 2
@@ -239,6 +304,7 @@ static struct label_ref g_label_refs[MAX_OPS];
static const struct parsed_proto *g_func_pp;
static struct parsed_data *g_func_pd;
static int g_func_pd_cnt;
+static int g_func_lmods;
static char g_func[256];
static char g_comment[256];
static int g_bp_frame;
@@ -246,8 +312,13 @@ static int g_sp_frame;
static int g_stack_frame_used;
static int g_stack_fsz;
static int g_ida_func_attr;
+static int g_sct_func_attr;
+static int g_stack_clear_start; // in dwords
+static int g_stack_clear_len;
+static int g_regmask_init;
static int g_skip_func;
static int g_allow_regfunc;
+static int g_allow_user_icall;
static int g_quiet_pp;
static int g_header_mode;
@@ -269,12 +340,30 @@ const char *regs_r32[] = {
"eax", "ebx", "ecx", "edx", "esi", "edi", "ebp", "esp",
// not r32, but list here for easy parsing and printing
"mm0", "mm1", "mm2", "mm3", "mm4", "mm5", "mm6", "mm7",
+ "st", "st(1)", "st(2)", "st(3)", "st(4)", "st(5)", "st(6)", "st(7)"
};
const char *regs_r16[] = { "ax", "bx", "cx", "dx", "si", "di", "bp", "sp" };
const char *regs_r8l[] = { "al", "bl", "cl", "dl" };
const char *regs_r8h[] = { "ah", "bh", "ch", "dh" };
-enum x86_regs { xUNSPEC = -1, xAX, xBX, xCX, xDX, xSI, xDI, xBP, xSP };
+enum x86_regs {
+ xUNSPEC = -1,
+ xAX, xBX, xCX, xDX,
+ xSI, xDI, xBP, xSP,
+ xMM0, xMM1, xMM2, xMM3, // mmx
+ xMM4, xMM5, xMM6, xMM7,
+ xST0, xST1, xST2, xST3, // x87
+ xST4, xST5, xST6, xST7,
+};
+
+#define mxAX (1 << xAX)
+#define mxCX (1 << xCX)
+#define mxDX (1 << xDX)
+#define mxST0 (1 << xST0)
+#define mxST1 (1 << xST1)
+#define mxST1_0 (mxST1 | mxST0)
+#define mxST7_2 (0xfc << xST0)
+#define mxSTa (0xff << xST0)
// possible basic comparison types (without inversion)
enum parsed_flag_op {
@@ -434,12 +523,12 @@ static int is_reg_in_str(const char *s)
}
static const char *parse_stack_el(const char *name, char *extra_reg,
- int early_try)
+ int *base_val, int early_try)
{
const char *p, *p2, *s;
char *endp = NULL;
char buf[32];
- long val;
+ long val = -1;
int len;
if (g_bp_frame || early_try)
@@ -512,6 +601,8 @@ static const char *parse_stack_el(const char *name, char *extra_reg,
if ('0' <= *p && *p <= '9')
return NULL;
+ if (base_val != NULL)
+ *base_val = val;
return p;
}
@@ -539,6 +630,9 @@ static int guess_lmod_from_name(struct parsed_opr *opr)
static int guess_lmod_from_c_type(enum opr_lenmod *lmod,
const struct parsed_type *c_type)
{
+ static const char *qword_types[] = {
+ "uint64_t", "int64_t", "__int64",
+ };
static const char *dword_types[] = {
"uint32_t", "int", "_DWORD", "UINT_PTR", "DWORD",
"WPARAM", "LPARAM", "UINT", "__int32",
@@ -587,6 +681,13 @@ static int guess_lmod_from_c_type(enum opr_lenmod *lmod,
}
}
+ for (i = 0; i < ARRAY_SIZE(qword_types); i++) {
+ if (IS(n, qword_types[i])) {
+ *lmod = OPLM_QWORD;
+ return 1;
+ }
+ }
+
return 0;
}
@@ -595,7 +696,7 @@ static char *default_cast_to(char *buf, size_t buf_size,
{
buf[0] = 0;
- if (!opr->is_ptr)
+ if (!opr->is_ptr || strchr(opr->name, '['))
return buf;
if (opr->pp == NULL || opr->pp->type.name == NULL
|| opr->pp->is_fptr)
@@ -746,13 +847,17 @@ static int parse_operand(struct parsed_opr *opr,
aerr("[] parse failure\n");
parse_indmode(opr->name, regmask_indirect, 1);
- if (opr->lmod == OPLM_UNSPEC && parse_stack_el(opr->name, NULL, 1))
+ if (opr->lmod == OPLM_UNSPEC
+ && parse_stack_el(opr->name, NULL, NULL, 1))
{
// might be an equ
struct parsed_equ *eq =
- equ_find(NULL, parse_stack_el(opr->name, NULL, 1), &i);
+ equ_find(NULL, parse_stack_el(opr->name, NULL, NULL, 1), &i);
if (eq)
opr->lmod = eq->lmod;
+
+ // might be unaligned access
+ g_func_lmods |= 1 << OPLM_BYTE;
}
return wordc;
}
@@ -844,6 +949,8 @@ static const struct {
{ "nop", OP_NOP, 0, 0, 0 },
{ "push", OP_PUSH, 1, 1, 0 },
{ "pop", OP_POP, 1, 1, OPF_DATA },
+ { "pusha",OP_PUSHA, 0, 0, 0 },
+ { "popa", OP_POPA, 0, 0, OPF_DATA },
{ "leave",OP_LEAVE, 0, 0, OPF_DATA },
{ "mov" , OP_MOV, 2, 2, OPF_DATA },
{ "lea", OP_LEA, 2, 2, OPF_DATA },
@@ -853,6 +960,7 @@ static const struct {
{ "not", OP_NOT, 1, 1, OPF_DATA },
{ "xlat", OP_XLAT, 0, 0, OPF_DATA },
{ "cdq", OP_CDQ, 0, 0, OPF_DATA },
+ { "bswap",OP_BSWAP, 1, 1, OPF_DATA },
{ "lodsb",OP_LODS, 0, 0, OPF_DATA },
{ "lodsw",OP_LODS, 0, 0, OPF_DATA },
{ "lodsd",OP_LODS, 0, 0, OPF_DATA },
@@ -961,9 +1069,52 @@ static const struct {
{ "setg", OP_SCC, 1, 1, OPF_DATA|OPF_CC, PFO_LE, 1 },
{ "setnle", OP_SCC, 1, 1, OPF_DATA|OPF_CC, PFO_LE, 1 },
// x87
+ { "fld", OP_FLD, 1, 1, OPF_FPUSH },
+ { "fild", OP_FILD, 1, 1, OPF_FPUSH },
+ { "fld1", OP_FLDc, 0, 0, OPF_FPUSH },
+ { "fldln2", OP_FLDc, 0, 0, OPF_FPUSH },
+ { "fldz", OP_FLDc, 0, 0, OPF_FPUSH },
+ { "fst", OP_FST, 1, 1, 0 },
+ { "fstp", OP_FST, 1, 1, OPF_FPOP },
+ { "fist", OP_FIST, 1, 1, 0 },
+ { "fistp", OP_FIST, 1, 1, OPF_FPOP },
+ { "fadd", OP_FADD, 0, 2, 0 },
+ { "faddp", OP_FADD, 0, 2, OPF_FPOP },
+ { "fdiv", OP_FDIV, 0, 2, 0 },
+ { "fdivp", OP_FDIV, 0, 2, OPF_FPOP },
+ { "fmul", OP_FMUL, 0, 2, 0 },
+ { "fmulp", OP_FMUL, 0, 2, OPF_FPOP },
+ { "fsub", OP_FSUB, 0, 2, 0 },
+ { "fsubp", OP_FSUB, 0, 2, OPF_FPOP },
+ { "fdivr", OP_FDIVR, 0, 2, 0 },
+ { "fdivrp", OP_FDIVR, 0, 2, OPF_FPOP },
+ { "fsubr", OP_FSUBR, 0, 2, 0 },
+ { "fsubrp", OP_FSUBR, 0, 2, OPF_FPOP },
+ { "fiadd", OP_FIADD, 1, 1, 0 },
+ { "fidiv", OP_FIDIV, 1, 1, 0 },
+ { "fimul", OP_FIMUL, 1, 1, 0 },
+ { "fisub", OP_FISUB, 1, 1, 0 },
+ { "fidivr", OP_FIDIVR, 1, 1, 0 },
+ { "fisubr", OP_FISUBR, 1, 1, 0 },
+ { "fcom", OP_FCOM, 0, 1, 0 },
+ { "fcomp", OP_FCOM, 0, 1, OPF_FPOP },
+ { "fnstsw", OP_FNSTSW, 1, 1, OPF_DATA },
+ { "fchs", OP_FCHS, 0, 0, 0 },
+ { "fcos", OP_FCOS, 0, 0, 0 },
+ { "fpatan", OP_FPATAN, 0, 0, OPF_FPOP },
+ { "fptan", OP_FPTAN, 0, 0, OPF_FPUSH },
+ { "fsin", OP_FSIN, 0, 0, 0 },
+ { "fsqrt", OP_FSQRT, 0, 0, 0 },
+ { "fxch", OP_FXCH, 1, 1, 0 },
+ { "fyl2x", OP_FYL2X, 0, 0, OPF_FPOP },
// mmx
- { "emms", OP_EMMS, 0, 0, OPF_DATA },
- { "movq", OP_MOV, 2, 2, OPF_DATA },
+ { "emms", OP_EMMS, 0, 0, OPF_DATA },
+ { "movq", OP_MOV, 2, 2, OPF_DATA },
+ // pseudo-ops for lib calls
+ { "_allshl",OPP_ALLSHL },
+ { "_allshr",OPP_ALLSHR },
+ { "_ftol", OPP_FTOL },
+ { "_CIpow", OPP_CIPOW },
// must be last
{ "ud2", OP_UD2 },
};
@@ -1028,6 +1179,9 @@ static void parse_op(struct parsed_op *op, char words[16][256], int wordc)
else
op->regmask_src |= regmask;
op->regmask_src |= regmask_ind;
+
+ if (op->operand[opr].lmod != OPLM_UNSPEC)
+ g_func_lmods |= 1 << op->operand[opr].lmod;
}
if (w < wordc)
@@ -1058,6 +1212,7 @@ static void parse_op(struct parsed_op *op, char words[16][256], int wordc)
case OP_INC:
case OP_DEC:
case OP_NEG:
+ case OP_BSWAP:
// more below..
op->regmask_src |= op->regmask_dst;
break;
@@ -1237,6 +1392,78 @@ static void parse_op(struct parsed_op *op, char words[16][256], int wordc)
op->regmask_src = 1 << xBP;
break;
+ case OP_FLD:
+ case OP_FILD:
+ op->regmask_dst |= mxST0;
+ break;
+
+ case OP_FLDc:
+ op->regmask_dst |= mxST0;
+ if (IS(words[op_w] + 3, "1"))
+ op->operand[0].val = X87_CONST_1;
+ else if (IS(words[op_w] + 3, "ln2"))
+ op->operand[0].val = X87_CONST_LN2;
+ else if (IS(words[op_w] + 3, "z"))
+ op->operand[0].val = X87_CONST_Z;
+ else
+ aerr("TODO\n");
+ break;
+
+ case OP_FST:
+ case OP_FIST:
+ op->regmask_src |= mxST0;
+ break;
+
+ case OP_FADD:
+ case OP_FDIV:
+ case OP_FMUL:
+ case OP_FSUB:
+ case OP_FDIVR:
+ case OP_FSUBR:
+ op->regmask_src |= mxST0;
+ if (op->operand_cnt == 2)
+ op->regmask_src |= op->regmask_dst;
+ else if (op->operand_cnt == 1) {
+ memcpy(&op->operand[1], &op->operand[0], sizeof(op->operand[1]));
+ op->operand[0].type = OPT_REG;
+ op->operand[0].lmod = OPLM_QWORD;
+ op->operand[0].reg = xST0;
+ op->regmask_dst |= mxST0;
+ }
+ else
+ // IDA doesn't use this
+ aerr("no operands?\n");
+ break;
+
+ case OP_FIADD:
+ case OP_FIDIV:
+ case OP_FIMUL:
+ case OP_FISUB:
+ case OP_FIDIVR:
+ case OP_FISUBR:
+ case OP_FCHS:
+ case OP_FCOS:
+ case OP_FSIN:
+ case OP_FSQRT:
+ case OP_FXCH:
+ op->regmask_src |= mxST0;
+ op->regmask_dst |= mxST0;
+ break;
+
+ case OP_FPATAN:
+ case OP_FYL2X:
+ op->regmask_src |= mxST0 | mxST1;
+ op->regmask_dst |= mxST0;
+ break;
+
+ case OP_FPTAN:
+ aerr("TODO\n");
+ break;
+
+ case OP_FCOM:
+ op->regmask_src |= mxST0;
+ break;
+
default:
break;
}
@@ -1517,14 +1744,12 @@ static struct parsed_equ *equ_find(struct parsed_op *po, const char *name,
*extra_offs = 0;
namelen = strlen(name);
- p = strchr(name, '+');
+ p = strpbrk(name, "+-");
if (p != NULL) {
namelen = p - name;
if (namelen <= 0)
ferr(po, "equ parse failed for '%s'\n", name);
- if (IS_START(p, "0x"))
- p += 2;
*extra_offs = strtol(p, &endp, 16);
if (*endp != 0)
ferr(po, "equ parse failed for '%s'\n", name);
@@ -1546,7 +1771,7 @@ static struct parsed_equ *equ_find(struct parsed_op *po, const char *name,
static int is_stack_access(struct parsed_op *po,
const struct parsed_opr *popr)
{
- return (parse_stack_el(popr->name, NULL, 0)
+ return (parse_stack_el(popr->name, NULL, NULL, 0)
|| (g_bp_frame && !(po->flags & OPF_EBP_S)
&& IS_START(popr->name, "ebp")));
}
@@ -1577,8 +1802,7 @@ static void parse_stack_access(struct parsed_op *po,
ferr(po, "ebp- parse of '%s' failed\n", name);
}
else {
- bp_arg = parse_stack_el(name, ofs_reg, 0);
- snprintf(g_comment, sizeof(g_comment), "%s", bp_arg);
+ bp_arg = parse_stack_el(name, ofs_reg, NULL, 0);
eq = equ_find(po, bp_arg, &offset);
if (eq == NULL)
ferr(po, "detected but missing eq\n");
@@ -1596,11 +1820,46 @@ static void parse_stack_access(struct parsed_op *po,
}
*offset_out = offset;
- *stack_ra_out = stack_ra;
+ if (stack_ra_out)
+ *stack_ra_out = stack_ra;
if (bp_arg_out)
*bp_arg_out = bp_arg;
}
+static int parse_stack_esp_offset(struct parsed_op *po,
+ const char *name, int *offset_out)
+{
+ char ofs_reg[16] = { 0, };
+ struct parsed_equ *eq;
+ const char *bp_arg;
+ char *endp = NULL;
+ int base_val = 0;
+ int offset = 0;
+
+ if (strstr(name, "esp") == NULL)
+ return -1;
+ bp_arg = parse_stack_el(name, ofs_reg, &base_val, 0);
+ if (bp_arg == NULL) {
+ // just plain offset?
+ if (!IS_START(name, "esp+"))
+ return -1;
+ offset = strtol(name + 4, &endp, 0);
+ if (endp == NULL || *endp != 0)
+ return -1;
+ *offset_out = offset;
+ return 0;
+ }
+
+ if (ofs_reg[0] != 0)
+ return -1;
+ eq = equ_find(po, bp_arg, &offset);
+ if (eq == NULL)
+ ferr(po, "detected but missing eq\n");
+ offset += eq->offset;
+ *offset_out = base_val + offset;
+ return 0;
+}
+
static int stack_frame_access(struct parsed_op *po,
struct parsed_opr *popr, char *buf, size_t buf_size,
const char *name, const char *cast, int is_src, int is_lea)
@@ -1615,7 +1874,6 @@ static int stack_frame_access(struct parsed_op *po,
int offset = 0;
int retval = -1;
int sf_ofs;
- int lim;
if (po->flags & OPF_EBP_S)
ferr(po, "stack_frame_access while ebp is scratch\n");
@@ -1623,6 +1881,8 @@ static int stack_frame_access(struct parsed_op *po,
parse_stack_access(po, name, ofs_reg, &offset,
&stack_ra, &bp_arg, is_lea);
+ snprintf(g_comment, sizeof(g_comment), "%s", bp_arg);
+
if (offset > stack_ra)
{
arg_i = (offset - stack_ra - 4) / 4;
@@ -1721,7 +1981,7 @@ static int stack_frame_access(struct parsed_op *po,
}
if (unaligned)
- snprintf(g_comment, sizeof(g_comment), "%s unaligned", bp_arg);
+ strcat(g_comment, " unaligned");
// common problem
guess_lmod_from_c_type(&tmp_lmod, &g_func_pp->arg[i].type);
@@ -1744,8 +2004,7 @@ static int stack_frame_access(struct parsed_op *po,
g_stack_frame_used = 1;
sf_ofs = g_stack_fsz + offset;
- lim = (ofs_reg[0] != 0) ? -4 : 0;
- if (offset > 0 || sf_ofs < lim)
+ if (ofs_reg[0] == 0 && (offset > 0 || sf_ofs < 0))
ferr(po, "bp_stack offset %d/%d\n", offset, g_stack_fsz);
if (is_lea)
@@ -1786,6 +2045,16 @@ static int stack_frame_access(struct parsed_op *po,
snprintf(buf, buf_size, "%ssf.d[%d]", prefix, sf_ofs / 4);
break;
+ case OPLM_QWORD:
+ ferr_assert(po, !(sf_ofs & 7));
+ ferr_assert(po, ofs_reg[0] == 0);
+ // only used for x87 int64/float, float sets is_lea
+ if (is_lea)
+ snprintf(buf, buf_size, "%ssf.q[%d]", prefix, sf_ofs / 8);
+ else
+ snprintf(buf, buf_size, "*(s64 *)&sf.q[%d]", sf_ofs / 8);
+ break;
+
default:
ferr(po, "bp_stack bad lmod: %d\n", popr->lmod);
}
@@ -1802,7 +2071,7 @@ static void check_func_pp(struct parsed_op *po,
int ret, i;
if (pp->argc_reg != 0) {
- if (/*!g_allow_regfunc &&*/ !pp->is_fastcall) {
+ if (!g_allow_user_icall && !pp->is_fastcall) {
pp_print(buf, sizeof(buf), pp);
ferr(po, "%s: unexpected reg arg in icall: %s\n", pfx, buf);
}
@@ -2020,6 +2289,60 @@ static char *out_src_opr_u32(char *buf, size_t buf_size,
return out_src_opr(buf, buf_size, po, popr, NULL, 0);
}
+static char *out_src_opr_float(char *buf, size_t buf_size,
+ struct parsed_op *po, struct parsed_opr *popr, int need_float_stack)
+{
+ const char *cast = NULL;
+ char tmp[256];
+
+ switch (popr->type) {
+ case OPT_REG:
+ if (popr->reg < xST0 || popr->reg > xST7)
+ ferr(po, "bad reg: %d\n", popr->reg);
+
+ if (need_float_stack) {
+ if (popr->reg == xST0)
+ snprintf(buf, buf_size, "f_st[f_stp & 7]");
+ else
+ snprintf(buf, buf_size, "f_st[(f_stp + %d) & 7]",
+ popr->reg - xST0);
+ }
+ else
+ snprintf(buf, buf_size, "f_st%d", popr->reg - xST0);
+ break;
+
+ case OPT_REGMEM:
+ case OPT_LABEL:
+ case OPT_OFFSET:
+ switch (popr->lmod) {
+ case OPLM_QWORD:
+ cast = "double";
+ break;
+ case OPLM_DWORD:
+ cast = "float";
+ break;
+ default:
+ ferr(po, "unhandled lmod: %d\n", popr->lmod);
+ break;
+ }
+ out_src_opr(tmp, sizeof(tmp), po, popr, "", 1);
+ snprintf(buf, buf_size, "*(%s *)(%s)", cast, tmp);
+ break;
+
+ default:
+ ferr(po, "invalid float type: %d\n", popr->type);
+ }
+
+ return buf;
+}
+
+static char *out_dst_opr_float(char *buf, size_t buf_size,
+ struct parsed_op *po, struct parsed_opr *popr, int need_float_stack)
+{
+ // same?
+ return out_src_opr_float(buf, buf_size, po, popr, need_float_stack);
+}
+
static void out_test_for_cc(char *buf, size_t buf_size,
struct parsed_op *po, enum parsed_flag_op pfo, int is_inv,
enum opr_lenmod lmod, const char *expr)
@@ -2031,7 +2354,7 @@ static void out_test_for_cc(char *buf, size_t buf_size,
switch (pfo) {
case PFO_Z:
- case PFO_BE: // CF=1||ZF=1; CF=0
+ case PFO_BE: // CF==1||ZF==1; CF=0
snprintf(buf, buf_size, "(%s%s %s 0)",
cast, expr, is_inv ? "!=" : "==");
break;
@@ -2042,18 +2365,24 @@ static void out_test_for_cc(char *buf, size_t buf_size,
scast, expr, is_inv ? ">=" : "<");
break;
- case PFO_LE: // ZF=1||SF!=OF; OF=0
+ case PFO_LE: // ZF==1||SF!=OF; OF=0
snprintf(buf, buf_size, "(%s%s %s 0)",
scast, expr, is_inv ? ">" : "<=");
break;
+ case PFO_C: // CF=0
+ case PFO_O: // OF=0
+ snprintf(buf, buf_size, "(%d)", !!is_inv);
+ break;
+
default:
ferr(po, "%s: unhandled parsed_flag_op: %d\n", __func__, pfo);
}
}
static void out_cmp_for_cc(char *buf, size_t buf_size,
- struct parsed_op *po, enum parsed_flag_op pfo, int is_inv)
+ struct parsed_op *po, enum parsed_flag_op pfo, int is_inv,
+ int is_neg)
{
const char *cast, *scast, *cast_use;
char buf1[256], buf2[256];
@@ -2087,8 +2416,13 @@ static void out_cmp_for_cc(char *buf, size_t buf_size,
out_src_opr(buf1, sizeof(buf1), po, &po->operand[0], cast_use, 0);
if (po->op == OP_DEC)
snprintf(buf2, sizeof(buf2), "1");
- else
- out_src_opr(buf2, sizeof(buf2), po, &po->operand[1], cast_use, 0);
+ else {
+ char cast_op2[64];
+ snprintf(cast_op2, sizeof(cast_op2) - 1, "%s", cast_use);
+ if (is_neg)
+ strcat(cast_op2, "-");
+ out_src_opr(buf2, sizeof(buf2), po, &po->operand[1], cast_op2, 0);
+ }
switch (pfo) {
case PFO_C:
@@ -2156,7 +2490,7 @@ static void out_cmp_test(char *buf, size_t buf_size,
po->operand[0].lmod, buf3);
}
else if (po->op == OP_CMP) {
- out_cmp_for_cc(buf, buf_size, po, pfo, is_inv);
+ out_cmp_for_cc(buf, buf_size, po, pfo, is_inv, 0);
}
else
ferr(po, "%s: unhandled op: %d\n", __func__, po->op);
@@ -2231,7 +2565,7 @@ static const char *op_to_c(struct parsed_op *po)
// note: this skips over calls and rm'd stuff assuming they're handled
// so it's intended to use at one of final passes
static int scan_for_pop(int i, int opcnt, int magic, int reg,
- int depth, int flags_set)
+ int depth, int seen_noreturn, int flags_set)
{
struct parsed_op *po;
int relevant;
@@ -2247,10 +2581,12 @@ static int scan_for_pop(int i, int opcnt, int magic, int reg,
if (po->flags & OPF_TAIL) {
if (po->op == OP_CALL) {
if (po->pp != NULL && po->pp->is_noreturn)
- // assume no stack cleanup for noreturn
- return 1;
+ seen_noreturn = 1;
+ else
+ return -1;
}
- return -1; // deadend
+ else
+ return -1; // deadend
}
if (po->flags & (OPF_RMD|OPF_DONE|OPF_FARG))
@@ -2262,7 +2598,7 @@ static int scan_for_pop(int i, int opcnt, int magic, int reg,
for (j = 0; j < po->btj->count; j++) {
check_i(po, po->btj->d[j].bt_i);
ret |= scan_for_pop(po->btj->d[j].bt_i, opcnt, magic, reg,
- depth, flags_set);
+ depth, seen_noreturn, flags_set);
if (ret < 0)
return ret; // dead end
}
@@ -2272,7 +2608,7 @@ static int scan_for_pop(int i, int opcnt, int magic, int reg,
check_i(po, po->bt_i);
if (po->flags & OPF_CJMP) {
ret |= scan_for_pop(po->bt_i, opcnt, magic, reg,
- depth, flags_set);
+ depth, seen_noreturn, flags_set);
if (ret < 0)
return ret; // dead end
}
@@ -2301,7 +2637,8 @@ static int scan_for_pop(int i, int opcnt, int magic, int reg,
}
}
- return -1;
+ // for noreturn, assume msvc skipped stack cleanup
+ return seen_noreturn ? 1 : -1;
}
// scan for 'reg' pop backwards starting from i
@@ -2788,8 +3125,11 @@ static int scan_for_mod_opr0(struct parsed_op *po_test,
return -1;
}
-static int scan_for_flag_set(int i, int magic, int *branched,
- int *setters, int *setter_cnt)
+static int try_resolve_const(int i, const struct parsed_opr *opr,
+ int magic, unsigned int *val);
+
+static int scan_for_flag_set(int i, int opcnt, int magic,
+ int *branched, int *setters, int *setter_cnt)
{
struct label_ref *lr;
int ret;
@@ -2808,7 +3148,7 @@ static int scan_for_flag_set(int i, int magic, int *branched,
lr = &g_label_refs[i];
for (; lr->next; lr = lr->next) {
check_i(&ops[i], lr->i);
- ret = scan_for_flag_set(lr->i, magic,
+ ret = scan_for_flag_set(lr->i, opcnt, magic,
branched, setters, setter_cnt);
if (ret < 0)
return ret;
@@ -2819,7 +3159,7 @@ static int scan_for_flag_set(int i, int magic, int *branched,
i = lr->i;
continue;
}
- ret = scan_for_flag_set(lr->i, magic,
+ ret = scan_for_flag_set(lr->i, opcnt, magic,
branched, setters, setter_cnt);
if (ret < 0)
return ret;
@@ -2829,6 +3169,20 @@ static int scan_for_flag_set(int i, int magic, int *branched,
if (ops[i].flags & OPF_FLAGS) {
setters[*setter_cnt] = i;
(*setter_cnt)++;
+
+ if (ops[i].flags & OPF_REP) {
+ struct parsed_opr opr = OPR_INIT(OPT_REG, OPLM_DWORD, xCX);
+ unsigned int uval;
+
+ ret = try_resolve_const(i, &opr, i + opcnt * 7, &uval);
+ if (ret != 1 || uval == 0) {
+ // can't treat it as full setter because of ecx=0 case,
+ // also disallow delayed compare
+ *branched = 1;
+ continue;
+ }
+ }
+
return 0;
}
@@ -3028,7 +3382,8 @@ static void scan_fwd_set_flags(int i, int opcnt, int magic, int flags)
}
static const struct parsed_proto *try_recover_pp(
- struct parsed_op *po, const struct parsed_opr *opr, int *search_instead)
+ struct parsed_op *po, const struct parsed_opr *opr,
+ int is_call, int *search_instead)
{
const struct parsed_proto *pp = NULL;
char buf[256];
@@ -3067,10 +3422,12 @@ static const struct parsed_proto *try_recover_pp(
if (arg == g_func_pp->argc)
ferr(po, "stack arg %d not in prototype?\n", arg_i);
- pp = g_func_pp->arg[arg].fptr;
- if (pp == NULL)
- ferr(po, "icall sa: arg%d is not a fptr?\n", arg + 1);
- check_func_pp(po, pp, "icall arg");
+ pp = g_func_pp->arg[arg].pp;
+ if (is_call) {
+ if (pp == NULL)
+ ferr(po, "icall arg: arg%d has no pp\n", arg + 1);
+ check_func_pp(po, pp, "icall arg");
+ }
}
else if (opr->type == OPT_REGMEM && strchr(opr->name + 1, '[')) {
// label[index]
@@ -3151,14 +3508,14 @@ static void scan_for_call_type(int i, const struct parsed_opr *opr,
}
if (i == g_func_pp->argc)
return;
- pp = g_func_pp->arg[i].fptr;
+ pp = g_func_pp->arg[i].pp;
if (pp == NULL)
ferr(po, "icall: arg%d (%s) is not a fptr?\n",
i + 1, g_func_pp->arg[i].reg);
check_func_pp(po, pp, "icall reg-arg");
}
else
- pp = try_recover_pp(po, opr, NULL);
+ pp = try_recover_pp(po, opr, 1, NULL);
if (*pp_found != NULL && pp != NULL && *pp_found != pp) {
if (!IS((*pp_found)->ret_type.name, pp->ret_type.name)
@@ -3268,16 +3625,54 @@ static int get_pp_arg_regmask_src(const struct parsed_proto *pp)
static int get_pp_arg_regmask_dst(const struct parsed_proto *pp)
{
+ int regmask = 0;
+ int i, reg;
+
+ if (pp->has_retreg) {
+ for (i = 0; i < pp->argc; i++) {
+ if (pp->arg[i].type.is_retreg) {
+ reg = char_array_i(regs_r32,
+ ARRAY_SIZE(regs_r32), pp->arg[i].reg);
+ ferr_assert(ops, reg >= 0);
+ regmask |= 1 << reg;
+ }
+ }
+ }
+
if (strstr(pp->ret_type.name, "int64"))
- return (1 << xAX) | (1 << xDX);
+ return regmask | (1 << xAX) | (1 << xDX);
+ if (IS(pp->ret_type.name, "float")
+ || IS(pp->ret_type.name, "double"))
+ {
+ return regmask | mxST0;
+ }
if (strcasecmp(pp->ret_type.name, "void") == 0)
- return 0;
+ return regmask;
+
+ return regmask | mxAX;
+}
- return (1 << xAX);
+static int are_ops_same(struct parsed_op *po1, struct parsed_op *po2)
+{
+ return po1->op == po2->op && po1->operand_cnt == po2->operand_cnt
+ && memcmp(po1->operand, po2->operand,
+ sizeof(po1->operand[0]) * po1->operand_cnt) == 0;
}
static void resolve_branches_parse_calls(int opcnt)
{
+ static const struct {
+ const char *name;
+ enum op_op op;
+ unsigned int flags;
+ unsigned int regmask_src;
+ unsigned int regmask_dst;
+ } pseudo_ops[] = {
+ { "__allshl", OPP_ALLSHL, OPF_DATA, mxAX|mxDX|mxCX, mxAX|mxDX },
+ { "__allshr", OPP_ALLSHR, OPF_DATA, mxAX|mxDX|mxCX, mxAX|mxDX },
+ { "__ftol", OPP_FTOL, OPF_FPOP, mxST0, mxAX | mxDX },
+ { "__CIpow", OPP_CIPOW, OPF_FPOP, mxST0|mxST1, mxST0 },
+ };
const struct parsed_proto *pp_c;
struct parsed_proto *pp;
struct parsed_data *pd;
@@ -3292,13 +3687,47 @@ static void resolve_branches_parse_calls(int opcnt)
po->bt_i = -1;
po->btj = NULL;
+ if (po->datap != NULL) {
+ pp = calloc(1, sizeof(*pp));
+ my_assert_not(pp, NULL);
+
+ ret = parse_protostr(po->datap, pp);
+ if (ret < 0)
+ ferr(po, "bad protostr supplied: %s\n", (char *)po->datap);
+ free(po->datap);
+ po->datap = NULL;
+ po->pp = pp;
+ }
+
if (po->op == OP_CALL) {
pp = NULL;
- if (po->operand[0].type == OPT_LABEL) {
+ if (po->pp != NULL)
+ pp = po->pp;
+ else if (po->operand[0].type == OPT_LABEL)
+ {
tmpname = opr_name(po, 0);
if (IS_START(tmpname, "loc_"))
ferr(po, "call to loc_*\n");
+ if (IS(tmpname, "__alloca_probe"))
+ continue;
+
+ // convert some calls to pseudo-ops
+ for (l = 0; l < ARRAY_SIZE(pseudo_ops); l++) {
+ if (!IS(tmpname, pseudo_ops[l].name))
+ continue;
+
+ po->op = pseudo_ops[l].op;
+ po->operand_cnt = 0;
+ po->regmask_src = pseudo_ops[l].regmask_src;
+ po->regmask_dst = pseudo_ops[l].regmask_dst;
+ po->flags = pseudo_ops[l].flags;
+ po->flags |= po->regmask_dst ? OPF_DATA : 0;
+ break;
+ }
+ if (l < ARRAY_SIZE(pseudo_ops))
+ continue;
+
pp_c = proto_parse(g_fhdr, tmpname, g_header_mode);
if (!g_header_mode && pp_c == NULL)
ferr(po, "proto_parse failed for call '%s'\n", tmpname);
@@ -3308,16 +3737,6 @@ static void resolve_branches_parse_calls(int opcnt)
my_assert_not(pp, NULL);
}
}
- else if (po->datap != NULL) {
- pp = calloc(1, sizeof(*pp));
- my_assert_not(pp, NULL);
-
- ret = parse_protostr(po->datap, pp);
- if (ret < 0)
- ferr(po, "bad protostr supplied: %s\n", (char *)po->datap);
- free(po->datap);
- po->datap = NULL;
- }
if (pp != NULL) {
if (pp->is_fptr)
@@ -3376,7 +3795,8 @@ tailcall:
static void scan_prologue_epilogue(int opcnt)
{
- int ecx_push = 0, esp_sub = 0;
+ int ecx_push = 0, esp_sub = 0, pusha = 0;
+ int sandard_epilogue;
int found;
int i, j, l;
@@ -3390,14 +3810,19 @@ static void scan_prologue_epilogue(int opcnt)
ops[1].flags |= OPF_RMD | OPF_DONE | OPF_NOREGS;
i = 2;
- if (ops[2].op == OP_SUB && IS(opr_name(&ops[2], 0), "esp")) {
- g_stack_fsz = opr_const(&ops[2], 1);
- ops[2].flags |= OPF_RMD | OPF_DONE | OPF_NOREGS;
+ if (ops[i].op == OP_PUSHA) {
+ ops[i].flags |= OPF_RMD | OPF_DONE | OPF_NOREGS;
+ pusha = 1;
+ i++;
+ }
+
+ if (ops[i].op == OP_SUB && IS(opr_name(&ops[i], 0), "esp")) {
+ g_stack_fsz = opr_const(&ops[i], 1);
+ ops[i].flags |= OPF_RMD | OPF_DONE | OPF_NOREGS;
i++;
}
else {
// another way msvc builds stack frame..
- i = 2;
while (ops[i].op == OP_PUSH && IS(opr_name(&ops[i], 0), "ecx")) {
g_stack_fsz += 4;
ops[i].flags |= OPF_RMD | OPF_DONE | OPF_NOREGS;
@@ -3431,10 +3856,20 @@ static void scan_prologue_epilogue(int opcnt)
j--;
}
- if ((ops[j].op == OP_POP && IS(opr_name(&ops[j], 0), "ebp"))
- || ops[j].op == OP_LEAVE)
+ sandard_epilogue = 0;
+ if (ops[j].op == OP_POP && IS(opr_name(&ops[j], 0), "ebp"))
+ {
+ ops[j].flags |= OPF_RMD | OPF_DONE | OPF_NOREGS;
+ // the standard epilogue is sometimes even used without a sf
+ if (ops[j - 1].op == OP_MOV
+ && IS(opr_name(&ops[j - 1], 0), "esp")
+ && IS(opr_name(&ops[j - 1], 1), "ebp"))
+ sandard_epilogue = 1;
+ }
+ else if (ops[j].op == OP_LEAVE)
{
ops[j].flags |= OPF_RMD | OPF_DONE | OPF_NOREGS;
+ sandard_epilogue = 1;
}
else if (ops[i].op == OP_CALL && ops[i].pp != NULL
&& ops[i].pp->is_noreturn)
@@ -3447,13 +3882,10 @@ static void scan_prologue_epilogue(int opcnt)
else if (!(g_ida_func_attr & IDAFA_NORETURN))
ferr(&ops[j], "'pop ebp' expected\n");
- if (g_stack_fsz != 0) {
+ if (g_stack_fsz != 0 || sandard_epilogue) {
if (ops[j].op == OP_LEAVE)
j--;
- else if (ops[j].op == OP_POP
- && ops[j - 1].op == OP_MOV
- && IS(opr_name(&ops[j - 1], 0), "esp")
- && IS(opr_name(&ops[j - 1], 1), "ebp"))
+ else if (sandard_epilogue) // mov esp, ebp
{
ops[j - 1].flags |= OPF_RMD | OPF_DONE | OPF_NOREGS;
j -= 2;
@@ -3470,6 +3902,13 @@ static void scan_prologue_epilogue(int opcnt)
}
}
+ if (pusha) {
+ if (ops[j].op == OP_POPA)
+ ops[j].flags |= OPF_RMD | OPF_DONE | OPF_NOREGS;
+ else
+ ferr(&ops[j], "popa expected\n");
+ }
+
found = 1;
i++;
} while (i < opcnt);
@@ -3496,6 +3935,20 @@ static void scan_prologue_epilogue(int opcnt)
{
g_stack_fsz = ops[i].operand[1].val;
ops[i].flags |= OPF_RMD | OPF_DONE | OPF_NOREGS;
+ i++;
+ esp_sub = 1;
+ break;
+ }
+ else if (ops[i].op == OP_MOV && ops[i].operand[0].reg == xAX
+ && ops[i].operand[1].type == OPT_CONST
+ && ops[i + 1].op == OP_CALL
+ && IS(opr_name(&ops[i + 1], 0), "__alloca_probe"))
+ {
+ g_stack_fsz += ops[i].operand[1].val;
+ ops[i].flags |= OPF_RMD | OPF_DONE | OPF_NOREGS;
+ i++;
+ ops[i].flags |= OPF_RMD | OPF_DONE | OPF_NOREGS;
+ i++;
esp_sub = 1;
break;
}
@@ -3538,11 +3991,11 @@ static void scan_prologue_epilogue(int opcnt)
{
g_sp_frame = 1;
- i++;
do {
for (; i < opcnt; i++)
if (ops[i].flags & OPF_TAIL)
break;
+
j = i - 1;
if (i == opcnt && (ops[j].flags & OPF_JMP)) {
if (ops[j].bt_i != -1 || ops[j].btj != NULL)
@@ -3579,7 +4032,16 @@ static void scan_prologue_epilogue(int opcnt)
|| !IS(opr_name(&ops[j], 0), "esp")
|| ops[j].operand[1].type != OPT_CONST
|| ops[j].operand[1].val != g_stack_fsz)
+ {
+ if (ops[i].op == OP_CALL && ops[i].pp != NULL
+ && ops[i].pp->is_noreturn)
+ {
+ // noreturn tailcall with no epilogue
+ i++;
+ continue;
+ }
ferr(&ops[j], "'add esp' expected\n");
+ }
ops[j].flags |= OPF_RMD | OPF_DONE | OPF_NOREGS;
ops[j].operand[1].val = 0; // hack for stack arg scanner
@@ -3594,46 +4056,17 @@ static void scan_prologue_epilogue(int opcnt)
}
}
-static const struct parsed_proto *resolve_icall(int i, int opcnt,
- int *pp_i, int *multi_src)
-{
- const struct parsed_proto *pp = NULL;
- int search_advice = 0;
-
- *multi_src = 0;
- *pp_i = -1;
-
- switch (ops[i].operand[0].type) {
- case OPT_REGMEM:
- case OPT_LABEL:
- case OPT_OFFSET:
- pp = try_recover_pp(&ops[i], &ops[i].operand[0], &search_advice);
- if (!search_advice)
- break;
- // fallthrough
- default:
- scan_for_call_type(i, &ops[i].operand[0], i + opcnt * 9, &pp,
- pp_i, multi_src);
- break;
- }
-
- return pp;
-}
-
// find an instruction that changed opr before i op
// *op_i must be set to -1 by the caller
-// *entry is set to 1 if one source is determined to be the caller
+// *is_caller is set to 1 if one source is determined to be g_func arg
// returns 1 if found, *op_i is then set to origin
+// returns -1 if multiple origins are found
static int resolve_origin(int i, const struct parsed_opr *opr,
int magic, int *op_i, int *is_caller)
{
struct label_ref *lr;
int ret = 0;
- if (ops[i].cc_scratch == magic)
- return 0;
- ops[i].cc_scratch = magic;
-
while (1) {
if (g_labels[i] != NULL) {
lr = &g_label_refs[i];
@@ -3662,10 +4095,9 @@ static int resolve_origin(int i, const struct parsed_opr *opr,
continue;
if (*op_i >= 0) {
- if (*op_i == i)
+ if (*op_i == i || are_ops_same(&ops[*op_i], &ops[i]))
return ret | 1;
- // XXX: could check if the other op does the same
return -1;
}
@@ -3684,10 +4116,6 @@ static int resolve_last_ref(int i, const struct parsed_opr *opr,
struct label_ref *lr;
int ret = 0;
- if (ops[i].cc_scratch == magic)
- return 0;
- ops[i].cc_scratch = magic;
-
while (1) {
if (g_labels[i] != NULL) {
lr = &g_label_refs[i];
@@ -3718,6 +4146,45 @@ static int resolve_last_ref(int i, const struct parsed_opr *opr,
}
}
+// adjust datap of all reachable 'op' insns when moving back
+// returns 1 if at least 1 op was found
+// returns -1 if path without an op was found
+static int adjust_prev_op(int i, enum op_op op, int magic, void *datap)
+{
+ struct label_ref *lr;
+ int ret = 0;
+
+ if (ops[i].cc_scratch == magic)
+ return 0;
+ ops[i].cc_scratch = magic;
+
+ while (1) {
+ if (g_labels[i] != NULL) {
+ lr = &g_label_refs[i];
+ for (; lr != NULL; lr = lr->next) {
+ check_i(&ops[i], lr->i);
+ ret |= adjust_prev_op(lr->i, op, magic, datap);
+ }
+ if (i > 0 && LAST_OP(i - 1))
+ return ret;
+ }
+
+ i--;
+ if (i < 0)
+ return -1;
+
+ if (ops[i].cc_scratch == magic)
+ return 0;
+ ops[i].cc_scratch = magic;
+
+ if (ops[i].op != op)
+ continue;
+
+ ops[i].datap = datap;
+ return 1;
+ }
+}
+
// find next instruction that reads opr
// *op_i must be set to -1 by the caller
// on return, *op_i is set to first referencer insn
@@ -3760,11 +4227,13 @@ static int find_next_read(int i, int opcnt,
}
if (!is_opr_read(opr, po)) {
- if (is_opr_modified(opr, po)
- && (po->op == OP_CALL
- || ((po->flags & OPF_DATA)
- && po->operand[0].lmod == OPLM_DWORD)))
+ int full_opr = 1;
+ if (opr->type == OPT_REG && po->operand[0].type == OPT_REG
+ && opr->reg == po->operand[0].reg && (po->flags & OPF_DATA))
{
+ full_opr = po->operand[0].lmod >= opr->lmod;
+ }
+ if (is_opr_modified(opr, po) && full_opr) {
// it's overwritten
return ret;
}
@@ -3783,8 +4252,67 @@ static int find_next_read(int i, int opcnt,
return 0;
}
-static int try_resolve_const(int i, const struct parsed_opr *opr,
- int magic, unsigned int *val)
+// find next instruction that reads opr
+// *op_i must be set to -1 by the caller
+// on return, *op_i is set to first flag user insn
+// returns 1 if exactly 1 flag user is found
+static int find_next_flag_use(int i, int opcnt, int magic, int *op_i)
+{
+ struct parsed_op *po;
+ int j, ret = 0;
+
+ for (; i < opcnt; i++)
+ {
+ if (ops[i].cc_scratch == magic)
+ return ret;
+ ops[i].cc_scratch = magic;
+
+ po = &ops[i];
+ if (po->op == OP_CALL)
+ return -1;
+ if (po->flags & OPF_JMP) {
+ if (po->btj != NULL) {
+ // jumptable
+ for (j = 0; j < po->btj->count; j++) {
+ check_i(po, po->btj->d[j].bt_i);
+ ret |= find_next_flag_use(po->btj->d[j].bt_i, opcnt,
+ magic, op_i);
+ }
+ return ret;
+ }
+
+ if (po->flags & OPF_RMD)
+ continue;
+ check_i(po, po->bt_i);
+ if (po->flags & OPF_CJMP)
+ goto found;
+ else
+ i = po->bt_i - 1;
+ continue;
+ }
+
+ if (!(po->flags & OPF_CC)) {
+ if (po->flags & OPF_FLAGS)
+ // flags changed
+ return ret;
+ if (po->flags & OPF_TAIL)
+ return ret;
+ continue;
+ }
+
+found:
+ if (*op_i >= 0)
+ return -1;
+
+ *op_i = i;
+ return 1;
+ }
+
+ return 0;
+}
+
+static int try_resolve_const(int i, const struct parsed_opr *opr,
+ int magic, unsigned int *val)
{
int s_i = -1;
int ret;
@@ -3802,6 +4330,169 @@ static int try_resolve_const(int i, const struct parsed_opr *opr,
return -1;
}
+static int resolve_used_bits(int i, int opcnt, int reg,
+ int *mask, int *is_z_check)
+{
+ struct parsed_opr opr = OPR_INIT(OPT_REG, OPLM_WORD, reg);
+ int j = -1, k = -1;
+ int ret;
+
+ ret = find_next_read(i, opcnt, &opr, i + opcnt * 20, &j);
+ if (ret != 1)
+ return -1;
+
+ find_next_read(j + 1, opcnt, &opr, i + opcnt * 20 + 1, &k);
+ if (k != -1) {
+ fnote(&ops[j], "(first read)\n");
+ ferr(&ops[k], "TODO: bit resolve: multiple readers\n");
+ }
+
+ if (ops[j].op != OP_TEST || ops[j].operand[1].type != OPT_CONST)
+ ferr(&ops[j], "TODO: bit resolve: not a const test\n");
+
+ ferr_assert(&ops[j], ops[j].operand[0].type == OPT_REG);
+ ferr_assert(&ops[j], ops[j].operand[0].reg == reg);
+
+ *mask = ops[j].operand[1].val;
+ if (ops[j].operand[0].lmod == OPLM_BYTE
+ && ops[j].operand[0].name[1] == 'h')
+ {
+ *mask <<= 8;
+ }
+ ferr_assert(&ops[j], (*mask & ~0xffff) == 0);
+
+ *is_z_check = 0;
+ ret = find_next_flag_use(j + 1, opcnt, i + opcnt * 20 + 2, &k);
+ if (ret == 1)
+ *is_z_check = ops[k].pfo == PFO_Z;
+
+ return 0;
+}
+
+static const struct parsed_proto *resolve_deref(int i, int magic,
+ struct parsed_opr *opr, int level)
+{
+ struct parsed_opr opr_s = OPR_INIT(OPT_REG, OPLM_DWORD, 0);
+ const struct parsed_proto *pp = NULL;
+ int from_caller = 0;
+ char s_reg[4];
+ int offset = 0;
+ int len = 0;
+ int j = -1;
+ int k = -1;
+ int reg;
+ int ret;
+
+ ret = sscanf(opr->name, "%3s+%x%n", s_reg, &offset, &len);
+ if (ret != 2 || len != strlen(opr->name)) {
+ ret = sscanf(opr->name, "%3s%n", s_reg, &len);
+ if (ret != 1 || len != strlen(opr->name))
+ return NULL;
+ }
+
+ reg = char_array_i(regs_r32, ARRAY_SIZE(regs_r32), s_reg);
+ if (reg < 0)
+ return NULL;
+
+ opr_s.reg = reg;
+ ret = resolve_origin(i, &opr_s, i + magic, &j, NULL);
+ if (ret != 1)
+ return NULL;
+
+ if (ops[j].op == OP_MOV && ops[j].operand[1].type == OPT_REGMEM
+ && strlen(ops[j].operand[1].name) == 3
+ && ops[j].operand[0].lmod == OPLM_DWORD
+ && ops[j].pp == NULL // no hint
+ && level == 0)
+ {
+ // allow one simple dereference (com/directx)
+ reg = char_array_i(regs_r32, ARRAY_SIZE(regs_r32),
+ ops[j].operand[1].name);
+ if (reg < 0)
+ return NULL;
+ opr_s.reg = reg;
+ ret = resolve_origin(j, &opr_s, j + magic, &k, NULL);
+ if (ret != 1)
+ return NULL;
+ j = k;
+ }
+ if (ops[j].op != OP_MOV || ops[j].operand[0].lmod != OPLM_DWORD)
+ return NULL;
+
+ if (ops[j].pp != NULL) {
+ // type hint in asm
+ pp = ops[j].pp;
+ }
+ else if (ops[j].operand[1].type == OPT_REGMEM) {
+ pp = try_recover_pp(&ops[j], &ops[j].operand[1], 0, NULL);
+ if (pp == NULL) {
+ // maybe structure ptr in structure
+ pp = resolve_deref(j, magic, &ops[j].operand[1], level + 1);
+ }
+ }
+ else if (ops[j].operand[1].type == OPT_LABEL)
+ pp = proto_parse(g_fhdr, ops[j].operand[1].name, g_quiet_pp);
+ else if (ops[j].operand[1].type == OPT_REG) {
+ // maybe arg reg?
+ k = -1;
+ ret = resolve_origin(j, &ops[j].operand[1], i + magic,
+ &k, &from_caller);
+ if (ret != 1 && from_caller && k == -1 && g_func_pp != NULL) {
+ for (k = 0; k < g_func_pp->argc; k++) {
+ if (g_func_pp->arg[k].reg == NULL)
+ continue;
+ if (IS(g_func_pp->arg[k].reg, ops[j].operand[1].name)) {
+ pp = g_func_pp->arg[k].pp;
+ break;
+ }
+ }
+ }
+ }
+
+ if (pp == NULL)
+ return NULL;
+ if (pp->is_func || pp->is_fptr || !pp->type.is_struct) {
+ if (offset != 0)
+ ferr(&ops[j], "expected struct, got '%s %s'\n",
+ pp->type.name, pp->name);
+ return NULL;
+ }
+
+ return proto_lookup_struct(g_fhdr, pp->type.name, offset);
+}
+
+static const struct parsed_proto *resolve_icall(int i, int opcnt,
+ int *pp_i, int *multi_src)
+{
+ const struct parsed_proto *pp = NULL;
+ int search_advice = 0;
+
+ *multi_src = 0;
+ *pp_i = -1;
+
+ switch (ops[i].operand[0].type) {
+ case OPT_REGMEM:
+ // try to resolve struct member calls
+ pp = resolve_deref(i, i + opcnt * 19, &ops[i].operand[0], 0);
+ if (pp != NULL)
+ break;
+ // fallthrough
+ case OPT_LABEL:
+ case OPT_OFFSET:
+ pp = try_recover_pp(&ops[i], &ops[i].operand[0],
+ 1, &search_advice);
+ if (!search_advice)
+ break;
+ // fallthrough
+ default:
+ scan_for_call_type(i, &ops[i].operand[0], i + opcnt * 9, &pp,
+ pp_i, multi_src);
+ break;
+ }
+
+ return pp;
+}
+
static struct parsed_proto *process_call_early(int i, int opcnt,
int *adj_i)
{
@@ -3960,9 +4651,69 @@ static struct parsed_proto *process_call(int i, int opcnt)
return pp;
}
-static int collect_call_args_early(struct parsed_op *po, int i,
- struct parsed_proto *pp, int *regmask)
+static int collect_call_args_no_push(int i, struct parsed_proto *pp,
+ int *regmask_ffca)
+{
+ struct parsed_op *po;
+ int offset = 0;
+ int base_arg;
+ int j, arg;
+ int ret;
+
+ for (base_arg = 0; base_arg < pp->argc; base_arg++)
+ if (pp->arg[base_arg].reg == NULL)
+ break;
+
+ for (j = i; j > 0; )
+ {
+ ferr_assert(&ops[j], g_labels[j] == NULL);
+ j--;
+
+ po = &ops[j];
+ ferr_assert(po, po->op != OP_PUSH);
+ if (po->op == OP_FST)
+ {
+ if (po->operand[0].type != OPT_REGMEM)
+ continue;
+ ret = parse_stack_esp_offset(po, po->operand[0].name, &offset);
+ if (ret != 0)
+ continue;
+ if (offset < 0 || offset >= pp->argc_stack * 4 || (offset & 3))
+ ferr(po, "bad offset %d (%d args)\n", offset, pp->argc_stack);
+
+ arg = base_arg + offset / 4;
+ po->p_argnext = -1;
+ po->p_argnum = arg + 1;
+ ferr_assert(po, pp->arg[arg].datap == NULL);
+ pp->arg[arg].datap = po;
+ po->flags |= OPF_DONE | OPF_FARGNR | OPF_FARG;
+ if (regmask_ffca != NULL)
+ *regmask_ffca |= 1 << arg;
+ }
+ else if (po->op == OP_SUB && po->operand[0].reg == xSP
+ && po->operand[1].type == OPT_CONST)
+ {
+ po->flags |= OPF_RMD | OPF_DONE | OPF_FARGNR | OPF_FARG;
+ break;
+ }
+ }
+
+ for (arg = base_arg; arg < pp->argc; arg++) {
+ ferr_assert(&ops[i], pp->arg[arg].reg == NULL);
+ po = pp->arg[arg].datap;
+ if (po == NULL)
+ ferr(&ops[i], "arg %d/%d not found\n", arg, pp->argc);
+ if (po->operand[0].lmod == OPLM_QWORD)
+ arg++;
+ }
+
+ return 0;
+}
+
+static int collect_call_args_early(int i, struct parsed_proto *pp,
+ int *regmask, int *regmask_ffca)
{
+ struct parsed_op *po;
int arg, ret;
int j;
@@ -3977,20 +4728,23 @@ static int collect_call_args_early(struct parsed_op *po, int i,
return -1;
j--;
- if (ops[j].op == OP_CALL)
+ po = &ops[j];
+ if (po->op == OP_CALL)
return -1;
- else if (ops[j].op == OP_ADD && ops[j].operand[0].reg == xSP)
+ else if (po->op == OP_ADD && po->operand[0].reg == xSP)
return -1;
- else if (ops[j].op == OP_POP)
+ else if (po->op == OP_POP)
return -1;
- else if (ops[j].flags & OPF_CJMP)
+ else if (po->flags & OPF_CJMP)
return -1;
- else if (ops[j].op == OP_PUSH) {
- if (ops[j].flags & (OPF_FARG|OPF_FARGNR))
- return -1;
- ret = scan_for_mod(&ops[j], j + 1, i, 1);
- if (ret >= 0)
+ else if (po->op == OP_PUSH) {
+ if (po->flags & (OPF_FARG|OPF_FARGNR))
return -1;
+ if (!g_header_mode) {
+ ret = scan_for_mod(po, j + 1, i, 1);
+ if (ret >= 0)
+ return -1;
+ }
if (pp->arg[arg].type.is_va_list)
return -1;
@@ -4000,6 +4754,17 @@ static int collect_call_args_early(struct parsed_op *po, int i,
if (pp->arg[arg].reg == NULL)
break;
}
+ else if (po->op == OP_SUB && po->operand[0].reg == xSP
+ && po->operand[1].type == OPT_CONST)
+ {
+ if (po->flags & (OPF_RMD|OPF_DONE))
+ return -1;
+ if (po->operand[1].val != pp->argc_stack * 4)
+ ferr(po, "unexpected esp adjust: %d\n",
+ po->operand[1].val * 4);
+ ferr_assert(po, pp->argc - arg == pp->argc_stack);
+ return collect_call_args_no_push(i, pp, regmask_ffca);
+ }
}
if (arg < pp->argc)
@@ -4020,7 +4785,7 @@ static int collect_call_args_early(struct parsed_op *po, int i,
ferr_assert(&ops[j], pp->arg[arg].datap == NULL);
pp->arg[arg].datap = &ops[j];
- if (ops[j].operand[0].type == OPT_REG)
+ if (regmask != NULL && ops[j].operand[0].type == OPT_REG)
*regmask |= 1 << ops[j].operand[0].reg;
ops[j].flags |= OPF_RMD | OPF_DONE | OPF_FARGNR | OPF_FARG;
@@ -4036,9 +4801,31 @@ static int collect_call_args_early(struct parsed_op *po, int i,
return 0;
}
+static int sync_argnum(struct parsed_op *po, int argnum)
+{
+ struct parsed_op *po_tmp;
+
+ // see if other branches don't have higher argnum
+ for (po_tmp = po; po_tmp != NULL; ) {
+ if (argnum < po_tmp->p_argnum)
+ argnum = po_tmp->p_argnum;
+ // note: p_argnext is active on current collect_call_args only
+ po_tmp = po_tmp->p_argnext >= 0 ? &ops[po_tmp->p_argnext] : NULL;
+ }
+
+ // make all argnums consistent
+ for (po_tmp = po; po_tmp != NULL; ) {
+ if (po_tmp->p_argnum != 0)
+ po_tmp->p_argnum = argnum;
+ po_tmp = po_tmp->p_argnext >= 0 ? &ops[po_tmp->p_argnext] : NULL;
+ }
+
+ return argnum;
+}
+
static int collect_call_args_r(struct parsed_op *po, int i,
- struct parsed_proto *pp, int *regmask, int *save_arg_vars,
- int *arg_grp, int arg, int magic, int need_op_saving, int may_reuse)
+ struct parsed_proto *pp, int *regmask, int *arg_grp,
+ int arg, int argnum, int magic, int need_op_saving, int may_reuse)
{
struct parsed_proto *pp_tmp;
struct parsed_op *po_tmp;
@@ -4046,7 +4833,6 @@ static int collect_call_args_r(struct parsed_op *po, int i,
int need_to_save_current;
int arg_grp_current = 0;
int save_args_seen = 0;
- int save_args;
int ret = 0;
int reg;
char buf[32];
@@ -4057,7 +4843,7 @@ static int collect_call_args_r(struct parsed_op *po, int i,
return -1;
}
- for (; arg < pp->argc; arg++)
+ for (; arg < pp->argc; arg++, argnum++)
if (pp->arg[arg].reg == NULL)
break;
magic = (magic & 0xffffff) | (arg << 24);
@@ -4083,8 +4869,8 @@ static int collect_call_args_r(struct parsed_op *po, int i,
check_i(&ops[j], lr->i);
if ((ops[lr->i].flags & (OPF_JMP|OPF_CJMP)) != OPF_JMP)
may_reuse = 1;
- ret = collect_call_args_r(po, lr->i, pp, regmask, save_arg_vars,
- arg_grp, arg, magic, need_op_saving, may_reuse);
+ ret = collect_call_args_r(po, lr->i, pp, regmask, arg_grp,
+ arg, argnum, magic, need_op_saving, may_reuse);
if (ret < 0)
return ret;
}
@@ -4098,8 +4884,8 @@ static int collect_call_args_r(struct parsed_op *po, int i,
continue;
}
need_op_saving = 1;
- ret = collect_call_args_r(po, lr->i, pp, regmask, save_arg_vars,
- arg_grp, arg, magic, need_op_saving, may_reuse);
+ ret = collect_call_args_r(po, lr->i, pp, regmask, arg_grp,
+ arg, argnum, magic, need_op_saving, may_reuse);
if (ret < 0)
return ret;
}
@@ -4157,8 +4943,9 @@ static int collect_call_args_r(struct parsed_op *po, int i,
ops[j].p_argnext = po_tmp - ops;
pp->arg[arg].datap = &ops[j];
+ argnum = sync_argnum(&ops[j], argnum);
+
need_to_save_current = 0;
- save_args = 0;
reg = -1;
if (ops[j].operand[0].type == OPT_REG)
reg = ops[j].operand[0].reg;
@@ -4168,25 +4955,15 @@ static int collect_call_args_r(struct parsed_op *po, int i,
need_to_save_current = (ret >= 0);
}
if (need_op_saving || need_to_save_current) {
- // mark this push as one that needs operand saving
- ops[j].flags &= ~OPF_RMD;
- if (ops[j].p_argnum == 0) {
- ops[j].p_argnum = arg + 1;
- save_args |= 1 << arg;
- }
- else if (ops[j].p_argnum < arg + 1) {
- // XXX: might kill valid var..
- //*save_arg_vars &= ~(1 << (ops[j].p_argnum - 1));
- ops[j].p_argnum = arg + 1;
- save_args |= 1 << arg;
- }
+ // mark this arg as one that needs operand saving
+ pp->arg[arg].is_saved = 1;
- if (save_args_seen & (1 << (ops[j].p_argnum - 1))) {
+ if (save_args_seen & (1 << (argnum - 1))) {
save_args_seen = 0;
arg_grp_current++;
if (arg_grp_current >= MAX_ARG_GRP)
ferr(&ops[j], "out of arg groups (arg%d), f %s\n",
- ops[j].p_argnum, pp->name);
+ argnum, pp->name);
}
}
else if (ops[j].p_argnum == 0)
@@ -4222,7 +4999,7 @@ static int collect_call_args_r(struct parsed_op *po, int i,
{
ops[k].flags |= OPF_RMD | OPF_NOREGS | OPF_DONE;
ops[j].flags |= OPF_RMD | OPF_NOREGS | OPF_VAPUSH;
- save_args &= ~(1 << arg);
+ pp->arg[arg].is_saved = 0;
reg = -1;
}
else
@@ -4238,23 +5015,27 @@ static int collect_call_args_r(struct parsed_op *po, int i,
ops[k].flags |= OPF_RMD | OPF_DONE;
ops[j].flags |= OPF_RMD;
ops[j].p_argpass = ret + 1;
- save_args &= ~(1 << arg);
+ pp->arg[arg].is_saved = 0;
reg = -1;
}
}
}
}
- *save_arg_vars |= save_args;
+ if (pp->arg[arg].is_saved) {
+ ops[j].flags &= ~OPF_RMD;
+ ops[j].p_argnum = argnum;
+ }
// tracking reg usage
if (reg >= 0)
*regmask |= 1 << reg;
arg++;
+ argnum++;
if (!pp->is_unresolved) {
// next arg
- for (; arg < pp->argc; arg++)
+ for (; arg < pp->argc; arg++, argnum++)
if (pp->arg[arg].reg == NULL)
break;
}
@@ -4282,19 +5063,17 @@ static int collect_call_args_r(struct parsed_op *po, int i,
}
static int collect_call_args(struct parsed_op *po, int i,
- struct parsed_proto *pp, int *regmask, int *save_arg_vars,
- int magic)
+ struct parsed_proto *pp, int *regmask, int magic)
{
// arg group is for cases when pushes for
// multiple funcs are going on
struct parsed_op *po_tmp;
- int save_arg_vars_current = 0;
int arg_grp = 0;
int ret;
int a;
- ret = collect_call_args_r(po, i, pp, regmask,
- &save_arg_vars_current, &arg_grp, 0, magic, 0, 0);
+ ret = collect_call_args_r(po, i, pp, regmask, &arg_grp,
+ 0, 1, magic, 0, 0);
if (ret < 0)
return ret;
@@ -4307,14 +5086,10 @@ static int collect_call_args(struct parsed_op *po, int i,
po_tmp = pp->arg[a].datap;
while (po_tmp != NULL) {
po_tmp->p_arggrp = arg_grp;
- if (po_tmp->p_argnext > 0)
- po_tmp = &ops[po_tmp->p_argnext];
- else
- po_tmp = NULL;
+ po_tmp = po_tmp->p_argnext >= 0 ? &ops[po_tmp->p_argnext] : NULL;
}
}
}
- save_arg_vars[arg_grp] |= save_arg_vars_current;
if (pp->is_unresolved) {
pp->argc += ret;
@@ -4384,9 +5159,10 @@ static void reg_use_pass(int i, int opcnt, unsigned char *cbits,
flags_set = OPF_RSAVE | OPF_DONE;
}
- ret = scan_for_pop(i + 1, opcnt, i + opcnt * 3, reg, 0, 0);
+ ret = scan_for_pop(i + 1, opcnt, i + opcnt * 3, reg, 0, 0, 0);
if (ret == 1) {
- scan_for_pop(i + 1, opcnt, i + opcnt * 4, reg, 0, flags_set);
+ scan_for_pop(i + 1, opcnt, i + opcnt * 4,
+ reg, 0, 0, flags_set);
}
else {
ret = scan_for_pop_ret(i + 1, opcnt, po->operand[0].reg, 0);
@@ -4432,6 +5208,10 @@ static void reg_use_pass(int i, int opcnt, unsigned char *cbits,
po->regmask_dst &= ~(1 << xAX);
}
}
+
+ // not "full stack" mode and have something in stack
+ if (!(regmask_now & mxST7_2) && (regmask_now & mxST1_0))
+ ferr(po, "float stack is not empty on func call\n");
}
if (po->flags & OPF_NOREGS)
@@ -4456,21 +5236,6 @@ static void reg_use_pass(int i, int opcnt, unsigned char *cbits,
if (g_bp_frame && !(po->flags & OPF_EBP_S))
regmask_new &= ~(1 << xBP);
- if (po->op == OP_CALL) {
- // allow fastcall calls from anywhere, calee may be also sitting
- // in some fastcall table even when it's not using reg args
- if (regmask_new & po->regmask_src & (1 << xCX)) {
- *regmask_init |= (1 << xCX);
- regmask_now |= (1 << xCX);
- regmask_new &= ~(1 << xCX);
- }
- if (regmask_new & po->regmask_src & (1 << xDX)) {
- *regmask_init |= (1 << xDX);
- regmask_now |= (1 << xDX);
- regmask_new &= ~(1 << xDX);
- }
- }
-
if (regmask_new != 0)
fnote(po, "uninitialized reg mask: %x\n", regmask_new);
@@ -4484,11 +5249,47 @@ static void reg_use_pass(int i, int opcnt, unsigned char *cbits,
}
}
+ if (po->flags & OPF_FPUSH) {
+ if (regmask_now & mxST1)
+ regmask_now |= mxSTa; // switch to "full stack" mode
+ if (regmask_now & mxSTa)
+ po->flags |= OPF_FSHIFT;
+ if (!(regmask_now & mxST7_2)) {
+ regmask_now =
+ (regmask_now & ~mxST1_0) | ((regmask_now & mxST0) << 1);
+ }
+ }
+
regmask_now |= regmask_op;
*regmask |= regmask_now;
- if (po->flags & OPF_TAIL)
- return;
+ // released regs
+ if (po->flags & OPF_FPOP) {
+ if ((regmask_now & mxSTa) == 0)
+ ferr(po, "float pop on empty stack?\n");
+ if (regmask_now & (mxST7_2 | mxST1))
+ po->flags |= OPF_FSHIFT;
+ if (!(regmask_now & mxST7_2)) {
+ regmask_now =
+ (regmask_now & ~mxST1_0) | ((regmask_now & mxST1) >> 1);
+ }
+ }
+
+ if (po->flags & OPF_TAIL) {
+ if (!(regmask_now & mxST7_2)) {
+ if (get_pp_arg_regmask_dst(g_func_pp) & mxST0) {
+ if (!(regmask_now & mxST0))
+ ferr(po, "no st0 on float return, mask: %x\n",
+ regmask_now);
+ }
+ else if (regmask_now & mxST1_0)
+ ferr(po, "float regs on tail: %x\n", regmask_now);
+ }
+
+ // there is support for "conditional tailcall", sort of
+ if (!(po->flags & OPF_CC))
+ return;
+ }
}
}
@@ -4567,9 +5368,11 @@ static void output_pp(FILE *fout, const struct parsed_proto *pp,
for (i = 0; i < pp->argc; i++) {
if (i > 0)
fprintf(fout, ", ");
- if (pp->arg[i].fptr != NULL && !(flags & OPP_SIMPLE_ARGS)) {
+ if (pp->arg[i].pp != NULL && pp->arg[i].pp->is_func
+ && !(flags & OPP_SIMPLE_ARGS))
+ {
// func pointer
- output_pp(fout, pp->arg[i].fptr, 0);
+ output_pp(fout, pp->arg[i].pp, 0);
}
else if (pp->arg[i].type.is_retreg) {
fprintf(fout, "u32 *r_%s", pp->arg[i].reg);
@@ -4609,23 +5412,30 @@ static void gen_func(FILE *fout, FILE *fhdr, const char *funcn, int opcnt)
char buf1[256], buf2[256], buf3[256], cast[64];
struct parsed_proto *pp, *pp_tmp;
struct parsed_data *pd;
- unsigned int uval;
int save_arg_vars[MAX_ARG_GRP] = { 0, };
unsigned char cbits[MAX_OPS / 8];
- int cond_vars = 0;
+ const char *float_type;
+ const char *float_st0;
+ const char *float_st1;
+ int need_float_stack = 0;
+ int need_float_sw = 0; // status word
int need_tmp_var = 0;
int need_tmp64 = 0;
+ int cond_vars = 0;
int had_decl = 0;
int label_pending = 0;
- int regmask_save = 0; // regs saved/restored in this func
+ int need_double = 0;
+ int regmask_save = 0; // used regs saved/restored in this func
int regmask_arg; // regs from this function args (fastcall, etc)
int regmask_ret; // regs needed on ret
int regmask_now; // temp
int regmask_init = 0; // regs that need zero initialization
int regmask_pp = 0; // regs used in complex push-pop graph
+ int regmask_ffca = 0; // float function call args
int regmask = 0; // used regs
int pfomask = 0;
int found = 0;
+ int dead_dst;
int no_output;
int i, j, l;
int arg;
@@ -4634,6 +5444,8 @@ static void gen_func(FILE *fout, FILE *fhdr, const char *funcn, int opcnt)
g_bp_frame = g_sp_frame = g_stack_fsz = 0;
g_stack_frame_used = 0;
+ if (g_sct_func_attr & SCTFA_CLEAR_REGS)
+ regmask_init = g_regmask_init;
g_func_pp = proto_parse(fhdr, funcn, 0);
if (g_func_pp == NULL)
@@ -4642,17 +5454,6 @@ static void gen_func(FILE *fout, FILE *fhdr, const char *funcn, int opcnt)
regmask_arg = get_pp_arg_regmask_src(g_func_pp);
regmask_ret = get_pp_arg_regmask_dst(g_func_pp);
- if (g_func_pp->has_retreg) {
- for (arg = 0; arg < g_func_pp->argc; arg++) {
- if (g_func_pp->arg[arg].type.is_retreg) {
- reg = char_array_i(regs_r32,
- ARRAY_SIZE(regs_r32), g_func_pp->arg[arg].reg);
- ferr_assert(ops, reg >= 0);
- regmask_ret |= 1 << reg;
- }
- }
- }
-
// pass1:
// - resolve all branches
// - parse calls with labels
@@ -4688,10 +5489,12 @@ static void gen_func(FILE *fout, FILE *fhdr, const char *funcn, int opcnt)
{
pp = process_call_early(i, opcnt, &j);
if (pp != NULL) {
- if (!(po->flags & OPF_ATAIL))
+ if (!(po->flags & OPF_ATAIL)) {
// since we know the args, try to collect them
- if (collect_call_args_early(po, i, pp, ®mask) != 0)
+ ret = collect_call_args_early(i, pp, ®mask, ®mask_ffca);
+ if (ret != 0)
pp = NULL;
+ }
}
if (pp != NULL) {
@@ -4717,8 +5520,11 @@ static void gen_func(FILE *fout, FILE *fhdr, const char *funcn, int opcnt)
// - process calls, stage 2
// - handle some push/pop pairs
// - scan for STD/CLD, propagate DF
+ // - try to resolve needed x87 status word bits
for (i = 0; i < opcnt; i++)
{
+ int mask, z_check;
+
po = &ops[i];
if (po->flags & OPF_RMD)
continue;
@@ -4730,8 +5536,7 @@ static void gen_func(FILE *fout, FILE *fhdr, const char *funcn, int opcnt)
if (!pp->is_unresolved && !(po->flags & OPF_ATAIL)) {
// since we know the args, collect them
- collect_call_args(po, i, pp, ®mask, save_arg_vars,
- i + opcnt * 2);
+ collect_call_args(po, i, pp, ®mask, i + opcnt * 2);
}
// for unresolved, collect after other passes
}
@@ -4742,6 +5547,9 @@ static void gen_func(FILE *fout, FILE *fhdr, const char *funcn, int opcnt)
po->regmask_src |= get_pp_arg_regmask_src(pp);
po->regmask_dst |= get_pp_arg_regmask_dst(pp);
+ if (po->regmask_dst & mxST0)
+ po->flags |= OPF_FPUSH;
+
if (strstr(pp->ret_type.name, "int64"))
need_tmp64 = 1;
@@ -4751,16 +5559,39 @@ static void gen_func(FILE *fout, FILE *fhdr, const char *funcn, int opcnt)
if (po->flags & OPF_DONE)
continue;
- if (po->op == OP_PUSH && !(po->flags & OPF_FARG)
- && !(po->flags & OPF_RSAVE) && po->operand[0].type == OPT_CONST)
- {
- scan_for_pop_const(i, opcnt, i + opcnt * 12);
- }
- else if (po->op == OP_POP)
+ switch (po->op) {
+ case OP_PUSH:
+ if (!(po->flags & OPF_FARG) && !(po->flags & OPF_RSAVE)
+ && po->operand[0].type == OPT_CONST)
+ {
+ scan_for_pop_const(i, opcnt, i + opcnt * 12);
+ }
+ break;
+
+ case OP_POP:
scan_pushes_for_pop(i, opcnt, ®mask_pp);
- else if (po->op == OP_STD) {
+ break;
+
+ case OP_STD:
po->flags |= OPF_DF | OPF_RMD | OPF_DONE;
scan_propagate_df(i + 1, opcnt);
+ break;
+
+ case OP_FNSTSW:
+ need_float_sw = 1;
+ if (po->operand[0].type != OPT_REG || po->operand[0].reg != xAX)
+ ferr(po, "TODO: fnstsw to mem\n");
+ ret = resolve_used_bits(i + 1, opcnt, xAX, &mask, &z_check);
+ if (ret != 0)
+ ferr(po, "fnstsw resolve failed\n");
+ ret = adjust_prev_op(i, OP_FCOM, i + opcnt * 21,
+ (void *)(long)(mask | (z_check << 16)));
+ if (ret != 1)
+ ferr(po, "failed to find fcom: %d\n", ret);
+ break;
+
+ default:
+ break;
}
}
@@ -4768,13 +5599,16 @@ static void gen_func(FILE *fout, FILE *fhdr, const char *funcn, int opcnt)
// - find POPs for PUSHes, rm both
// - scan for all used registers
memset(cbits, 0, sizeof(cbits));
- reg_use_pass(0, opcnt, cbits, 0, ®mask,
+ reg_use_pass(0, opcnt, cbits, regmask_init, ®mask,
0, ®mask_save, ®mask_init, regmask_arg);
+ need_float_stack = !!(regmask & mxST7_2);
+
// pass7:
// - find flag set ops for their users
// - do unresolved calls
// - declare indirect functions
+ // - other op specific processing
for (i = 0; i < opcnt; i++)
{
po = &ops[i];
@@ -4785,7 +5619,7 @@ static void gen_func(FILE *fout, FILE *fhdr, const char *funcn, int opcnt)
{
int setters[16], cnt = 0, branched = 0;
- ret = scan_for_flag_set(i, i + opcnt * 6,
+ ret = scan_for_flag_set(i, opcnt, i + opcnt * 6,
&branched, setters, &cnt);
if (ret < 0 || cnt <= 0)
ferr(po, "unable to trace flag setter(s)\n");
@@ -4839,35 +5673,41 @@ static void gen_func(FILE *fout, FILE *fhdr, const char *funcn, int opcnt)
cond_vars |= 1 << PFO_C;
}
- if (po->op == OP_CMPS || po->op == OP_SCAS) {
+ switch (po->op) {
+ case OP_CMPS:
+ case OP_SCAS:
cond_vars |= 1 << PFO_Z;
- }
- else if (po->op == OP_MUL
- || (po->op == OP_IMUL && po->operand_cnt == 1))
- {
+ break;
+
+ case OP_MUL:
if (po->operand[0].lmod == OPLM_DWORD)
need_tmp64 = 1;
- }
- else if (po->op == OP_CALL) {
+ break;
+
+ case OP_IMUL:
+ if (po->operand_cnt == 1 && po->operand[0].lmod == OPLM_DWORD)
+ need_tmp64 = 1;
+ break;
+
+ case OP_CALL:
// note: resolved non-reg calls are OPF_DONE already
pp = po->pp;
ferr_assert(po, pp != NULL);
if (pp->is_unresolved) {
int regmask_stack = 0;
- collect_call_args(po, i, pp, ®mask, save_arg_vars,
- i + opcnt * 2);
+ collect_call_args(po, i, pp, ®mask, i + opcnt * 2);
// this is pretty rough guess:
// see ecx and edx were pushed (and not their saved versions)
for (arg = 0; arg < pp->argc; arg++) {
- if (pp->arg[arg].reg != NULL)
+ if (pp->arg[arg].reg != NULL && !pp->arg[arg].is_saved)
continue;
tmp_op = pp->arg[arg].datap;
if (tmp_op == NULL)
ferr(po, "parsed_op missing for arg%d\n", arg);
- if (tmp_op->p_argnum == 0 && tmp_op->operand[0].type == OPT_REG)
+ if (tmp_op->operand[0].type == OPT_REG)
regmask_stack |= 1 << tmp_op->operand[0].reg;
}
@@ -4895,27 +5735,31 @@ static void gen_func(FILE *fout, FILE *fhdr, const char *funcn, int opcnt)
if (pp->argc_stack > 0)
pp->is_stdcall = 1;
}
- }
- else if (po->op == OP_MOV && po->operand[0].pp != NULL
- && po->operand[1].pp != NULL)
- {
- // = offset