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 {
OP_LOOP,
OP_JCC,
OP_SCC,
- // x87
- // mmx
- OP_EMMS,
- // undefined
- OP_UD2,
+ // x87
+ OP_FLD,
+ OP_FILD,
+ OP_FLDc,
+ OP_FST,
+ 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_FCOS,
+ OP_FPATAN,
+ OP_FPTAN,
+ OP_FSIN,
+ OP_FSQRT,
+ // mmx
+ OP_EMMS,
+ // pseudo-ops for lib calls
+ OPP_FTOL,
+ // undefined
+ OP_UD2,
};
enum opr_type {
};
// 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
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
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;
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_quiet_pp;
"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)
// possible basic comparison types (without inversion)
enum parsed_flag_op {
{
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)
equ_find(NULL, parse_stack_el(opr->name, NULL, 1), &i);
if (eq)
opr->lmod = eq->lmod;
+
+ // might be unaligned access
+ g_func_lmods |= 1 << OPLM_BYTE;
}
return wordc;
}
{ "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 },
+ { "fldz", OP_FLDc, 0, 0, OPF_FPUSH },
+ { "fstp", OP_FST, 1, 1, OPF_FPOP },
+ { "fst", OP_FST, 1, 1, 0 },
+ { "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 },
+ { "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 },
// 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
+ { "_ftol", OPP_FTOL },
// must be last
{ "ud2", OP_UD2 },
};
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)
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, "z"))
+ op->operand[0].val = X87_CONST_Z;
+ else
+ aerr("TODO\n");
+ break;
+
+ case OP_FST:
+ 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_FCOS:
+ case OP_FSIN:
+ case OP_FSQRT:
+ op->regmask_src |= mxST0;
+ op->regmask_dst |= mxST0;
+ break;
+
+ case OP_FPATAN:
+ op->regmask_src |= mxST0 | mxST1;
+ op->regmask_dst |= mxST0;
+ break;
+
+ case OP_FPTAN:
+ aerr("TODO\n");
+ break;
+
default:
break;
}
return regs_r32[popr->reg];
}
+static int check_simple_cast(const char *cast, int *bits, int *is_signed)
+{
+ if (IS_START(cast, "(s8)") || IS_START(cast, "(u8)"))
+ *bits = 8;
+ else if (IS_START(cast, "(s16)") || IS_START(cast, "(u16)"))
+ *bits = 16;
+ else if (IS_START(cast, "(s32)") || IS_START(cast, "(u32)"))
+ *bits = 32;
+ else if (IS_START(cast, "(s64)") || IS_START(cast, "(u64)"))
+ *bits = 64;
+ else
+ return -1;
+
+ *is_signed = cast[1] == 's' ? 1 : 0;
+ return 0;
+}
+
+static int check_deref_cast(const char *cast, int *bits)
+{
+ if (IS_START(cast, "*(u8 *)"))
+ *bits = 8;
+ else if (IS_START(cast, "*(u16 *)"))
+ *bits = 16;
+ else if (IS_START(cast, "*(u32 *)"))
+ *bits = 32;
+ else if (IS_START(cast, "*(u64 *)"))
+ *bits = 64;
+ else
+ return -1;
+
+ return 0;
+}
+
// cast1 is the "final" cast
static const char *simplify_cast(const char *cast1, const char *cast2)
{
static char buf[256];
+ int bits1, bits2;
+ int s1, s2;
if (cast1[0] == 0)
return cast2;
return cast1;
if (IS(cast1, cast2))
return cast1;
- if (IS(cast1, "(s8)") && IS(cast2, "(u8)"))
- return cast1;
- if (IS(cast1, "(s16)") && IS(cast2, "(u16)"))
- return cast1;
- if (IS(cast1, "(u8)") && IS_START(cast2, "*(u8 *)"))
- return cast2;
- if (IS(cast1, "(u16)") && IS_START(cast2, "*(u16 *)"))
- return cast2;
+
+ if (check_simple_cast(cast1, &bits1, &s1) == 0
+ && check_simple_cast(cast2, &bits2, &s2) == 0)
+ {
+ if (bits1 <= bits2)
+ return cast1;
+ }
+ if (check_simple_cast(cast1, &bits1, &s1) == 0
+ && check_deref_cast(cast2, &bits2) == 0)
+ {
+ if (bits1 == bits2) {
+ snprintf(buf, sizeof(buf), "*(%c%d *)", s1 ? 's' : 'u', bits1);
+ return buf;
+ }
+ }
+
if (strchr(cast1, '*') && IS_START(cast2, "(u32)"))
return cast1;
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);
+ // float callers set is_lea
+ ferr_assert(po, is_lea);
+ snprintf(buf, buf_size, "%ssf.q[%d]", prefix, sf_ofs / 8);
+ break;
+
default:
ferr(po, "bp_stack bad lmod: %d\n", popr->lmod);
}
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)
+{
+ 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);
+
+ 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)
+{
+ // same?
+ return out_src_opr_float(buf, buf_size, po, popr);
+}
+
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)
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 (1 << xAX);
+ return regmask | mxAX;
}
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[] = {
+ { "__ftol", OPP_FTOL, OPF_FPOP, mxST0, mxAX | mxDX },
+ };
const struct parsed_proto *pp_c;
struct parsed_proto *pp;
struct parsed_data *pd;
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");
+
+ // 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);
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)
}
}
-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)
{
return -1;
}
+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;
+ int offset = -1;
+ char name[256];
+ char s_reg[4];
+ int reg, len;
+ int ret;
+
+ *multi_src = 0;
+ *pp_i = -1;
+
+ switch (ops[i].operand[0].type) {
+ case OPT_REGMEM:
+ // try to resolve struct member calls
+ ret = sscanf(ops[i].operand[0].name, "%3s+%x%n",
+ s_reg, &offset, &len);
+ if (ret == 2 && len == strlen(ops[i].operand[0].name))
+ {
+ reg = char_array_i(regs_r32, ARRAY_SIZE(regs_r32), s_reg);
+ if (reg >= 0) {
+ struct parsed_opr opr = OPR_INIT(OPT_REG, OPLM_DWORD, reg);
+ int j = -1;
+ ret = resolve_origin(i, &opr, i + opcnt * 19, &j, NULL);
+ if (ret != 1)
+ break;
+ if (ops[j].op == OP_MOV && ops[j].operand[1].type == OPT_REGMEM
+ && ops[j].operand[0].lmod == OPLM_DWORD
+ && ops[j].pp == NULL) // no hint
+ {
+ // allow one simple dereference (directx)
+ reg = char_array_i(regs_r32, ARRAY_SIZE(regs_r32),
+ ops[j].operand[1].name);
+ if (reg < 0)
+ break;
+ struct parsed_opr opr2 = OPR_INIT(OPT_REG, OPLM_DWORD, reg);
+ int k = -1;
+ ret = resolve_origin(j, &opr2, j + opcnt * 19, &k, NULL);
+ if (ret != 1)
+ break;
+ j = k;
+ }
+ if (ops[j].op != OP_MOV)
+ break;
+ if (ops[j].operand[0].lmod != OPLM_DWORD)
+ break;
+ if (ops[j].pp != NULL) {
+ // type hint in asm
+ pp = ops[j].pp;
+ }
+ else if (ops[j].operand[1].type == OPT_REGMEM) {
+ // allow 'hello[ecx]' - assume array of same type items
+ ret = sscanf(ops[j].operand[1].name, "%[^[][e%2s]",
+ name, s_reg);
+ if (ret != 2)
+ break;
+ pp = proto_parse(g_fhdr, name, g_quiet_pp);
+ }
+ else if (ops[j].operand[1].type == OPT_LABEL)
+ pp = proto_parse(g_fhdr, ops[j].operand[1].name, g_quiet_pp);
+ else
+ break;
+ if (pp == NULL)
+ break;
+ if (pp->is_func || pp->is_fptr || !pp->type.is_struct) {
+ pp = NULL;
+ break;
+ }
+ pp = proto_lookup_struct(g_fhdr, pp->type.name, offset);
+ }
+ break;
+ }
+ // fallthrough
+ 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;
+}
+
static struct parsed_proto *process_call_early(int i, int opcnt,
int *adj_i)
{
int *regmask_init, int regmask_arg)
{
struct parsed_op *po;
+ unsigned int mask;
int already_saved;
int regmask_new;
int regmask_op;
if (po->flags & OPF_NOREGS)
continue;
+ if (po->flags & OPF_FPUSH) {
+ if (regmask_now & mxST1)
+ ferr(po, "TODO: FPUSH on active ST1\n");
+ if (regmask_now & mxST0)
+ po->flags |= OPF_FSHIFT;
+ mask = mxST0 | mxST1;
+ regmask_now = (regmask_now & ~mask) | ((regmask_now & mxST0) << 1);
+ }
+
// if incomplete register is used, clear it on init to avoid
// later use of uninitialized upper part in some situations
if ((po->flags & OPF_DATA) && po->operand[0].type == OPT_REG
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);
regmask_now |= regmask_op;
*regmask |= regmask_now;
- if (po->flags & OPF_TAIL)
- return;
+ // released regs
+ if (po->flags & OPF_FPOP) {
+ mask = mxST0 | mxST1;
+ if (!(regmask_now & mask))
+ ferr(po, "float pop on empty stack?\n");
+ if (regmask_now & mxST1)
+ po->flags |= OPF_FSHIFT;
+ regmask_now = (regmask_now & ~mask) | ((regmask_now & mxST1) >> 1);
+ }
+
+ if (po->flags & OPF_TAIL) {
+ if (regmask_now & (mxST0 | mxST1))
+ ferr(po, "float regs on tail: %x\n", regmask_now);
+
+ // there is support for "conditional tailcall", sort of
+ if (!(po->flags & OPF_CC))
+ return;
+ }
}
}
unsigned int uval;
int save_arg_vars[MAX_ARG_GRP] = { 0, };
unsigned char cbits[MAX_OPS / 8];
+ const char *float_type;
int cond_vars = 0;
int need_tmp_var = 0;
int need_tmp64 = 0;
int had_decl = 0;
int label_pending = 0;
+ int need_double = 0;
int regmask_save = 0; // regs saved/restored in this func
int regmask_arg; // regs from this function args (fastcall, etc)
int regmask_ret; // regs needed on ret
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)
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
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;
// - 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);
// pass7:
}
else if (po->op == OP_CLD)
po->flags |= OPF_RMD | OPF_DONE;
+ else if (po->op == OPP_FTOL) {
+ struct parsed_opr opr = OPR_INIT(OPT_REG, OPLM_DWORD, xDX);
+ j = -1;
+ find_next_read(i + 1, opcnt, &opr, i + opcnt * 18, &j);
+ if (j == -1)
+ po->flags |= OPF_32BIT;
+ }
+ else if (po->op == OP_FLD && po->operand[0].lmod == OPLM_QWORD)
+ need_double = 1;
if (po->op == OP_RCL || po->op == OP_RCR || po->op == OP_XCHG)
need_tmp_var = 1;
}
+ float_type = need_double ? "double" : "float";
+
// output starts here
// define userstack size
// declare stack frame, va_arg
if (g_stack_fsz) {
- fprintf(fout, " union { u32 d[%d]; u16 w[%d]; u8 b[%d]; } sf;\n",
- (g_stack_fsz + 3) / 4, (g_stack_fsz + 1) / 2, g_stack_fsz);
+ fprintf(fout, " union { u32 d[%d];", (g_stack_fsz + 3) / 4);
+ if (g_func_lmods & (1 << OPLM_WORD))
+ fprintf(fout, " u16 w[%d];", (g_stack_fsz + 1) / 2);
+ if (g_func_lmods & (1 << OPLM_BYTE))
+ fprintf(fout, " u8 b[%d];", g_stack_fsz);
+ if (g_func_lmods & (1 << OPLM_QWORD))
+ fprintf(fout, " double q[%d];", (g_stack_fsz + 7) / 8);
+ fprintf(fout, " } sf;\n");
had_decl = 1;
}
}
}
}
+ // ... mmx
if (regmask_now & 0xff00) {
for (reg = 8; reg < 16; reg++) {
if (regmask_now & (1 << reg)) {
}
}
}
+ // ... x87
+ if (regmask_now & 0xff0000) {
+ for (reg = 16; reg < 24; reg++) {
+ if (regmask_now & (1 << reg)) {
+ fprintf(fout, " %s f_st%d", float_type, reg - 16);
+ if (regmask_init & (1 << reg))
+ fprintf(fout, " = 0");
+ fprintf(fout, ";\n");
+ had_decl = 1;
+ }
+ }
+ }
if (regmask_save) {
for (reg = 0; reg < 8; reg++) {
if (had_decl)
fprintf(fout, "\n");
+ // do stack clear, if needed
+ if (g_sct_func_attr & SCTFA_CLEAR_SF) {
+ fprintf(fout, " ");
+ if (g_stack_clear_len != 0) {
+ if (g_stack_clear_len <= 4) {
+ for (i = 0; i < g_stack_clear_len; i++)
+ fprintf(fout, "sf.d[%d] = ", g_stack_clear_start + i);
+ fprintf(fout, "0;\n");
+ }
+ else {
+ fprintf(fout, "memset(&sf[%d], 0, %d);\n",
+ g_stack_clear_start, g_stack_clear_len * 4);
+ }
+ }
+ else
+ fprintf(fout, "memset(&sf, 0, sizeof(sf));\n");
+ }
+
if (g_func_pp->is_vararg) {
if (g_func_pp->argc_stack == 0)
ferr(ops, "vararg func without stack args?\n");
fprintf(fout, " cond_c = tmp64 >> 32;\n");
fprintf(fout, " %s = (u32)tmp64;",
out_dst_opr(buf1, sizeof(buf1), po, &po->operand[0]));
- strcat(g_comment, "add64");
+ strcat(g_comment, " add64");
}
else {
fprintf(fout, " cond_c = ((u32)%s + %s) >> %d;\n",
output_std_flags(fout, po, &pfomask, buf1);
last_arith_dst = &po->operand[0];
delayed_flag_op = NULL;
- strcat(g_comment, "bsf");
+ strcat(g_comment, " bsf");
break;
case OP_DEC:
fprintf(fout, " LOWORD(eax) = %s / %s%s;",
buf2, cast, buf1);
}
- strcat(g_comment, "div16");
+ strcat(g_comment, " div16");
break;
default:
ferr(po, "unhandled div lmod %d\n", po->operand[0].lmod);
case OP_JECXZ:
fprintf(fout, " if (ecx == 0)\n");
fprintf(fout, " goto %s;", po->operand[0].name);
- strcat(g_comment, "jecxz");
+ strcat(g_comment, " jecxz");
break;
case OP_LOOP:
fprintf(fout, " if (--ecx != 0)\n");
fprintf(fout, " goto %s;", po->operand[0].name);
- strcat(g_comment, "loop");
+ strcat(g_comment, " loop");
break;
case OP_JMP:
}
else if (!IS(pp->ret_type.name, "void")) {
if (po->flags & OPF_TAIL) {
- if (regmask_ret & (1 << xAX)) {
+ if (regmask_ret & mxAX) {
fprintf(fout, "return ");
if (g_func_pp->ret_type.is_ptr != pp->ret_type.is_ptr)
fprintf(fout, "(%s)", g_func_pp->ret_type.name);
}
+ else if (regmask_ret & mxST0)
+ ferr(po, "float tailcall\n");
}
- else if (po->regmask_dst & (1 << xAX)) {
+ else if (po->regmask_dst & mxAX) {
fprintf(fout, "eax = ");
if (pp->ret_type.is_ptr)
fprintf(fout, "(u32)");
}
+ else if (po->regmask_dst & mxST0) {
+ fprintf(fout, "f_st0 = ");
+ }
}
if (pp->name[0] == 0)
no_output = 1;
break;
+ // x87
+ case OP_FLD:
+ if (po->flags & OPF_FSHIFT)
+ fprintf(fout, " f_st1 = f_st0;\n");
+ if (po->operand[0].type == OPT_REG
+ && po->operand[0].reg == xST0)
+ {
+ strcat(g_comment, " fld st");
+ break;
+ }
+ fprintf(fout, " f_st0 = %s;",
+ out_src_opr_float(buf1, sizeof(buf1), po, &po->operand[0]));
+ strcat(g_comment, " fld");
+ break;
+
+ case OP_FILD:
+ if (po->flags & OPF_FSHIFT)
+ fprintf(fout, " f_st1 = f_st0;\n");
+ fprintf(fout, " f_st0 = (%s)%s;", float_type,
+ out_src_opr(buf1, sizeof(buf1), po, &po->operand[0],
+ lmod_cast(po, po->operand[0].lmod, 1), 0));
+ strcat(g_comment, " fild");
+ break;
+
+ case OP_FLDc:
+ if (po->flags & OPF_FSHIFT)
+ fprintf(fout, " f_st1 = f_st0;\n");
+ fprintf(fout, " f_st0 = ");
+ switch (po->operand[0].val) {
+ case X87_CONST_1: fprintf(fout, "1.0;"); break;
+ case X87_CONST_Z: fprintf(fout, "0.0;"); break;
+ default: ferr(po, "TODO\n"); break;
+ }
+ break;
+
+ case OP_FST:
+ if ((po->flags & OPF_FPOP) && po->operand[0].type == OPT_REG
+ && po->operand[0].reg == xST0)
+ {
+ no_output = 1;
+ break;
+ }
+ fprintf(fout, " %s = f_st0;",
+ out_dst_opr_float(buf1, sizeof(buf1), po, &po->operand[0]));
+ if (po->flags & OPF_FSHIFT)
+ fprintf(fout, "\n f_st0 = f_st1;");
+ strcat(g_comment, " fst");
+ break;
+
+ case OP_FADD:
+ case OP_FDIV:
+ case OP_FMUL:
+ case OP_FSUB:
+ switch (po->op) {
+ case OP_FADD: j = '+'; break;
+ case OP_FDIV: j = '/'; break;
+ case OP_FMUL: j = '*'; break;
+ case OP_FSUB: j = '-'; break;
+ default: j = 'x'; break;
+ }
+ if (po->flags & OPF_FSHIFT) {
+ fprintf(fout, " f_st0 = f_st1 %c f_st0;", j);
+ }
+ else {
+ fprintf(fout, " %s %c= %s;",
+ out_dst_opr_float(buf1, sizeof(buf1), po, &po->operand[0]),
+ j,
+ out_src_opr_float(buf2, sizeof(buf2), po, &po->operand[1]));
+ }
+ break;
+
+ case OP_FDIVR:
+ case OP_FSUBR:
+ if (po->flags & OPF_FSHIFT)
+ snprintf(buf1, sizeof(buf1), "f_st0");
+ else
+ out_dst_opr_float(buf1, sizeof(buf1), po, &po->operand[0]);
+ fprintf(fout, " %s = %s %c %s;", buf1,
+ out_src_opr_float(buf2, sizeof(buf2), po, &po->operand[1]),
+ po->op == OP_FDIVR ? '/' : '-',
+ out_src_opr_float(buf3, sizeof(buf3), po, &po->operand[0]));
+ break;
+
+ case OP_FIADD:
+ case OP_FIDIV:
+ case OP_FIMUL:
+ case OP_FISUB:
+ switch (po->op) {
+ case OP_FIADD: j = '+'; break;
+ case OP_FIDIV: j = '/'; break;
+ case OP_FIMUL: j = '*'; break;
+ case OP_FISUB: j = '-'; break;
+ default: j = 'x'; break;
+ }
+ fprintf(fout, " f_st0 %c= (%s)%s;", j, float_type,
+ out_src_opr(buf1, sizeof(buf1), po, &po->operand[0],
+ lmod_cast(po, po->operand[0].lmod, 1), 0));
+ break;
+
+ case OP_FIDIVR:
+ case OP_FISUBR:
+ fprintf(fout, " f_st0 = %s %c f_st0;",
+ out_src_opr_float(buf2, sizeof(buf2), po, &po->operand[1]),
+ po->op == OP_FIDIVR ? '/' : '-');
+ break;
+
+ case OP_FCOS:
+ fprintf(fout, " f_st0 = cos%s(f_st0);",
+ need_double ? "" : "f");
+ break;
+
+ case OP_FPATAN:
+ fprintf(fout, " f_st0 = atan%s(f_st1 / f_st0);",
+ need_double ? "" : "f");
+ break;
+
+ case OP_FSIN:
+ fprintf(fout, " f_st0 = sin%s(f_st0);",
+ need_double ? "" : "f");
+ break;
+
+ case OP_FSQRT:
+ fprintf(fout, " f_st0 = sqrt%s(f_st0);",
+ need_double ? "" : "f");
+ break;
+
+ case OPP_FTOL:
+ ferr_assert(po, po->flags & OPF_32BIT);
+ fprintf(fout, " eax = (s32)f_st0;");
+ if (po->flags & OPF_FSHIFT)
+ fprintf(fout, "\n f_st0 = f_st1;");
+ strcat(g_comment, " ftol");
+ break;
+
// mmx
case OP_EMMS:
- strcpy(g_comment, "(emms)");
+ strcpy(g_comment, " (emms)");
break;
default:
} *hg_vars;
static int hg_var_cnt;
+static char **hg_refs;
+static int hg_ref_cnt;
+
static void output_hdr_fp(FILE *fout, const struct func_prototype *fp,
int count);
-struct func_prototype *hg_fp_add(const char *funcn)
+static struct func_prototype *hg_fp_add(const char *funcn)
{
struct func_prototype *fp;
}
#endif
+static void hg_ref_add(const char *name)
+{
+ if ((hg_ref_cnt & 0xff) == 0) {
+ hg_refs = realloc(hg_refs, sizeof(hg_refs[0]) * (hg_ref_cnt + 0x100));
+ my_assert_not(hg_refs, NULL);
+ memset(hg_refs + hg_ref_cnt, 0, sizeof(hg_refs[0]) * 0x100);
+ }
+
+ hg_refs[hg_ref_cnt] = strdup(name);
+ my_assert_not(hg_refs[hg_ref_cnt], NULL);
+ hg_ref_cnt++;
+}
+
// recursive register dep pass
// - track saved regs (part 2)
// - try to figure out arg-regs
}
}
+// make all thiscall/edx arg functions referenced from .data fastcall
+static void do_func_refs_from_data(void)
+{
+ struct func_prototype *fp, fp_s;
+ int i;
+
+ for (i = 0; i < hg_ref_cnt; i++) {
+ strcpy(fp_s.name, hg_refs[i]);
+ fp = bsearch(&fp_s, hg_fp, hg_fp_cnt,
+ sizeof(hg_fp[0]), hg_fp_cmp_name);
+ if (fp == NULL)
+ continue;
+
+ if (fp->argc_stack != 0 && (fp->regmask_dep & (mxCX | mxDX)))
+ fp->regmask_dep |= mxCX | mxDX;
+ }
+}
+
static void output_hdr_fp(FILE *fout, const struct func_prototype *fp,
int count)
{
char *p, namebuf[NAMELEN];
const char *name;
int regmask_dep;
- int argc_stack;
+ int argc_normal;
int j, arg;
for (; count > 0; count--, fp++) {
}
regmask_dep = fp->regmask_dep;
- argc_stack = fp->argc_stack;
+ argc_normal = fp->argc_stack;
fprintf(fout, "%-5s", fp->pp ? fp->pp->ret_type.name :
(fp->has_ret ? "int" : "void"));
- if (regmask_dep && (fp->is_stdcall || argc_stack == 0)
- && (regmask_dep & ~((1 << xCX) | (1 << xDX))) == 0)
+ if (regmask_dep && (fp->is_stdcall || fp->argc_stack > 0)
+ && (regmask_dep & ~mxCX) == 0)
+ {
+ fprintf(fout, "/*__thiscall*/ ");
+ argc_normal++;
+ regmask_dep = 0;
+ }
+ else if (regmask_dep && (fp->is_stdcall || fp->argc_stack == 0)
+ && (regmask_dep & ~(mxCX | mxDX)) == 0)
{
fprintf(fout, " __fastcall ");
- if (!(regmask_dep & (1 << xDX)) && argc_stack == 0)
- argc_stack = 1;
+ if (!(regmask_dep & (1 << xDX)) && fp->argc_stack == 0)
+ argc_normal = 1;
else
- argc_stack += 2;
+ argc_normal += 2;
regmask_dep = 0;
}
else if (regmask_dep && !fp->is_stdcall) {
}
}
- for (j = 0; j < argc_stack; j++) {
+ for (j = 0; j < argc_normal; j++) {
arg++;
if (arg != 1)
fprintf(fout, ", ");
for (i = 0; i < hg_fp_cnt; i++)
hg_fp_resolve_deps(&hg_fp[i]);
+ // adjust functions referenced from data segment
+ do_func_refs_from_data();
+
// note: messes up .proto ptr, don't use
//qsort(hg_fp, hg_fp_cnt, sizeof(hg_fp[0]), hg_fp_cmp_id);
s = sskip(s);
i = 0;
- if (*s == '\'') {
+ if (*s == '\'' && s[1] != '\r' && s[1] != '\n') {
w[0] = s[0];
for (i = 1; i < wsize - 1; i++) {
if (s[i] == 0) {
return s + i;
}
-static void scan_variables(FILE *fasm)
+static int cmpstringp(const void *p1, const void *p2)
+{
+ return strcmp(*(char * const *)p1, *(char * const *)p2);
+}
+
+static int is_xref_needed(char *p, char **rlist, int rlist_len)
+{
+ char *p2;
+
+ p = sskip(p);
+ if (strstr(p, "..."))
+ // unable to determine, assume needed
+ return 1;
+
+ if (*p == '.') // .text, .data, ...
+ // ref from other data or non-function -> no
+ return 0;
+
+ p2 = strpbrk(p, "+:\r\n\x18");
+ if (p2 != NULL)
+ *p2 = 0;
+ if (bsearch(&p, rlist, rlist_len, sizeof(rlist[0]), cmpstringp))
+ // referenced from removed code
+ return 0;
+
+ return 1;
+}
+
+static int ida_xrefs_show_need(FILE *fasm, char *p,
+ char **rlist, int rlist_len)
+{
+ int found_need = 0;
+ char line[256];
+ long pos;
+
+ p = strrchr(p, ';');
+ if (p != NULL && *p == ';' && IS_START(p + 2, "DATA XREF: ")) {
+ p += 13;
+ if (is_xref_needed(p, rlist, rlist_len))
+ return 1;
+ }
+
+ pos = ftell(fasm);
+ while (1)
+ {
+ if (!my_fgets(line, sizeof(line), fasm))
+ break;
+ // non-first line is always indented
+ if (!my_isblank(line[0]))
+ break;
+
+ // should be no content, just comment
+ p = sskip(line);
+ if (*p != ';')
+ break;
+
+ p = strrchr(p, ';');
+ p += 2;
+ // it's printed once, but no harm to check again
+ if (IS_START(p, "DATA XREF: "))
+ p += 11;
+
+ if (is_xref_needed(p, rlist, rlist_len)) {
+ found_need = 1;
+ break;
+ }
+ }
+ fseek(fasm, pos, SEEK_SET);
+ return found_need;
+}
+
+static void scan_variables(FILE *fasm, char **rlist, int rlist_len)
{
struct scanned_var *var;
char line[256] = { 0, };
- char words[3][256];
+ char words[4][256];
+ int no_identifier;
char *p = NULL;
int wordc;
int l;
asmln++;
p = line;
- if (my_isblank(*p))
- continue;
+ no_identifier = my_isblank(*p);
p = sskip(p);
if (*p == 0 || *p == ';')
if (wordc < 2)
continue;
+ if (no_identifier) {
+ if (wordc >= 3 && IS(words[0], "dd") && IS(words[1], "offset"))
+ hg_ref_add(words[2]);
+ continue;
+ }
+
if (IS_START(words[0], "__IMPORT_DESCRIPTOR_")) {
// when this starts, we don't need anything from this section
break;
}
+ // check refs comment(s)
+ if (!ida_xrefs_show_need(fasm, p, rlist, rlist_len))
+ continue;
+
if ((hg_var_cnt & 0xff) == 0) {
hg_vars = realloc(hg_vars, sizeof(hg_vars[0])
* (hg_var_cnt + 0x100));
continue;
}
- if (IS(words[1], "dd"))
+ if (IS(words[1], "dd")) {
var->lmod = OPLM_DWORD;
+ if (wordc >= 4 && IS(words[2], "offset"))
+ hg_ref_add(words[3]);
+ }
else if (IS(words[1], "dw"))
var->lmod = OPLM_WORD;
else if (IS(words[1], "db")) {
return strcmp(c1->name, c2->name);
}
-static int cmpstringp(const void *p1, const void *p2)
-{
- return strcmp(*(char * const *)p1, *(char * const *)p2);
-}
-
static void scan_ahead(FILE *fasm)
{
char words[2][256];
}
if (g_header_mode)
- scan_variables(fasm);
+ scan_variables(fasm, rlist, rlist_len);
while (my_fgets(line, sizeof(line), fasm))
{
}
}
}
+ else if (p[2] == 's' && IS_START(p, "; sctattr:"))
+ {
+ static const char *attrs[] = {
+ "clear_sf",
+ "clear_regmask",
+ };
+
+ // parse manual attribute-list comment
+ g_sct_func_attr = 0;
+ p = sskip(p + 10);
+
+ for (; *p != 0; p = sskip(p)) {
+ for (i = 0; i < ARRAY_SIZE(attrs); i++) {
+ if (!strncmp(p, attrs[i], strlen(attrs[i]))) {
+ g_sct_func_attr |= 1 << i;
+ p += strlen(attrs[i]);
+ break;
+ }
+ }
+ if (*p == '=') {
+ j = ret = 0;
+ if (i == 0)
+ // clear_sf=start,len (in dwords)
+ ret = sscanf(p, "=%d,%d%n", &g_stack_clear_start,
+ &g_stack_clear_len, &j);
+ else if (i == 1)
+ // clear_regmask=<mask>
+ ret = sscanf(p, "=%x%n", &g_regmask_init, &j) + 1;
+ if (ret < 2) {
+ anote("unparsed attr value: %s\n", p);
+ break;
+ }
+ p += j;
+ }
+ else if (i == ARRAY_SIZE(attrs)) {
+ anote("unparsed sct attr: %s\n", p);
+ break;
+ }
+ }
+ }
else if (p[2] == 'S' && IS_START(p, "; START OF FUNCTION CHUNK FOR "))
{
p += 30;
pending_endp = 0;
in_func = 0;
g_ida_func_attr = 0;
+ g_sct_func_attr = 0;
+ g_stack_clear_start = 0;
+ g_stack_clear_len = 0;
+ g_regmask_init = 0;
skip_warned = 0;
g_skip_func = 0;
g_func[0] = 0;
pd->d = NULL;
}
g_func_pd_cnt = 0;
+ g_func_lmods = 0;
pd = NULL;
if (end)
parse_op(&ops[pi], words, wordc);
- if (sctproto != NULL) {
- if (ops[pi].op == OP_CALL || ops[pi].op == OP_JMP)
- ops[pi].datap = sctproto;
- sctproto = NULL;
- }
+ ops[pi].datap = sctproto;
+ sctproto = NULL;
pi++;
}
notaz.gp2x.de / ia32rtools.git / blobdiff