OPF_REP = (1 << 7), /* prefixed by rep */
OPF_REPZ = (1 << 8), /* rep is repe/repz */
OPF_REPNZ = (1 << 9), /* rep is repne/repnz */
+ OPF_FARG = (1 << 10), /* push collected as func arg (no reuse) */
};
enum op_op {
OPT_CONST,
};
+// must be sorted (larger len must be further in enum)
enum opr_lenmod {
OPLM_UNSPEC,
OPLM_BYTE,
enum opr_lenmod lmod;
unsigned int is_ptr:1; // pointer in C
unsigned int is_array:1; // array in C
+ unsigned int size_mismatch:1; // type override differs from C
+ unsigned int size_lt:1; // type override is larger than C
int reg;
unsigned int val;
char name[256];
int regmask_src; // all referensed regs
int regmask_dst;
int pfomask; // flagop: parsed_flag_op that can't be delayed
- int argmask; // push: args that are altered before call
+ int argnum; // push: altered before call arg #
int cc_scratch; // scratch storage during analysis
int bt_i; // branch target for branches
struct parsed_data *btj;// branch targets for jumptables
// datap:
// OP_CALL - ptr to parsed_proto
-// (OPF_CC) - point to corresponding (OPF_FLAGS)
+// (OPF_CC) - point to one of (OPF_FLAGS) that affects cc op
struct parsed_equ {
char name[64];
static int g_eqcnt;
static char g_labels[MAX_OPS][32];
static struct label_ref g_label_refs[MAX_OPS];
-static struct parsed_proto g_func_pp;
+static const struct parsed_proto *g_func_pp;
static struct parsed_data *g_func_pd;
static int g_func_pd_cnt;
static char g_func[256];
fcloseall(); \
exit(1); \
} while (0)
+#define fnote(op_, fmt, ...) \
+ printf("error:%s:#%zd: '%s': " fmt, g_func, (op_) - ops, \
+ dump_op(op_), ##__VA_ARGS__)
#define MAX_REGS 8
{
static const char *dword_types[] = {
"int", "_DWORD", "DWORD", "HANDLE", "HWND", "HMODULE",
+ "WPARAM", "UINT",
};
static const char *word_types[] = {
- "__int16", "unsigned __int16",
+ "uint16_t", "int16_t",
+ "unsigned __int16", "__int16",
};
static const char *byte_types[] = {
- "char", "__int8", "unsigned __int8", "BYTE",
+ "uint8_t", "int8_t", "char",
+ "unsigned __int8", "__int8", "BYTE",
};
const char *n;
int i;
int *regmask, int *regmask_indirect,
char words[16][256], int wordc, int w, unsigned int op_flags)
{
- struct parsed_proto pp;
+ const struct parsed_proto *pp;
enum opr_lenmod tmplmod;
unsigned long number;
int ret, len;
// most likely var in data segment
opr->type = OPT_LABEL;
- ret = proto_parse(g_fhdr, opr->name, &pp);
- if (ret == 0) {
- if (pp.is_fptr) {
+ pp = proto_parse(g_fhdr, opr->name);
+ if (pp != NULL) {
+ if (pp->is_fptr) {
opr->lmod = OPLM_DWORD;
opr->is_ptr = 1;
}
- else if (opr->lmod == OPLM_UNSPEC) {
- if (!guess_lmod_from_c_type(&opr->lmod, &pp.type))
- anote("unhandled C type '%s' for '%s'\n", pp.type.name, opr->name);
+ else {
+ if (!guess_lmod_from_c_type(&tmplmod, &pp->type))
+ anote("unhandled C type '%s' for '%s'\n",
+ pp->type.name, opr->name);
+
+ if (opr->lmod == OPLM_UNSPEC)
+ opr->lmod = tmplmod;
+ else if (opr->lmod != tmplmod) {
+ opr->size_mismatch = 1;
+ if (tmplmod < opr->lmod)
+ opr->size_lt = 1;
+ }
+ opr->is_ptr = pp->type.is_ptr;
}
- opr->is_ptr = pp.type.is_ptr;
- opr->is_array = pp.type.is_array;
+ opr->is_array = pp->type.is_array;
}
- proto_release(&pp);
if (opr->lmod == OPLM_UNSPEC)
guess_lmod_from_name(opr);
{ "test", OP_TEST, 2, 2, OPF_FLAGS },
{ "cmp", OP_CMP, 2, 2, OPF_FLAGS },
{ "retn", OP_RET, 0, 1, OPF_JMP|OPF_TAIL },
- { "call", OP_CALL, 1, 1, OPF_JMP|OPF_FLAGS },
+ { "call", OP_CALL, 1, 1, OPF_JMP|OPF_DATA|OPF_FLAGS },
{ "jmp", OP_JMP, 1, 1, OPF_JMP },
{ "jo", OP_JO, 1, 1, OPF_JMP|OPF_CC }, // 70 OF=1
{ "jno", OP_JNO, 1, 1, OPF_JMP|OPF_CC }, // 71 OF=0
}
break;
+ case OP_LEA:
+ if (op->operand[0].type == OPT_REG
+ && op->operand[1].type == OPT_REGMEM)
+ {
+ char buf[16];
+ snprintf(buf, sizeof(buf), "%s+0", op->operand[0].name);
+ if (IS(buf, op->operand[1].name))
+ op->flags |= OPF_RMD;
+ }
+ break;
+
default:
break;
}
}
static void stack_frame_access(struct parsed_op *po,
- enum opr_lenmod lmod, char *buf, size_t buf_size,
+ struct parsed_opr *popr, char *buf, size_t buf_size,
const char *name, const char *cast, int is_src, int is_lea)
{
enum opr_lenmod tmp_lmod = OPLM_UNSPEC;
const char *prefix = "";
+ const char *bp_arg = NULL;
char ofs_reg[16] = { 0, };
struct parsed_equ *eq;
+ const char *p;
+ char *endp = NULL;
int i, arg_i, arg_s;
- const char *bp_arg;
int stack_ra = 0;
int offset = 0;
int sf_ofs;
+ int lim;
- bp_arg = parse_stack_el(name, ofs_reg);
- snprintf(g_comment, sizeof(g_comment), "%s", bp_arg);
- eq = equ_find(po, bp_arg, &offset);
- if (eq == NULL)
- ferr(po, "detected but missing eq\n");
-
- offset += eq->offset;
+ if (!IS_START(name, "ebp-")) {
+ bp_arg = parse_stack_el(name, ofs_reg);
+ snprintf(g_comment, sizeof(g_comment), "%s", bp_arg);
+ eq = equ_find(po, bp_arg, &offset);
+ if (eq == NULL)
+ ferr(po, "detected but missing eq\n");
+ offset += eq->offset;
+ }
+ else {
+ p = name + 4;
+ if (IS_START(p, "0x"))
+ p += 2;
+ offset = -strtoul(p, &endp, 16);
+ if (*endp != 0)
+ ferr(po, "ebp- parse of '%s' failed\n", name);
+ }
if (!strncmp(name, "ebp", 3))
stack_ra = 4;
if (stack_ra <= offset && offset < stack_ra + 4)
ferr(po, "reference to ra? %d %d\n", offset, stack_ra);
- if (offset > stack_ra) {
+ if (offset > stack_ra)
+ {
arg_i = (offset - stack_ra - 4) / 4;
- if (arg_i < 0 || arg_i >= g_func_pp.argc_stack)
- ferr(po, "offset %d (%s) doesn't map to any arg\n",
- offset, bp_arg);
+ if (arg_i < 0 || arg_i >= g_func_pp->argc_stack)
+ {
+ if (g_func_pp->is_vararg
+ && arg_i == g_func_pp->argc_stack && is_lea)
+ {
+ // should be va_list
+ if (cast[0] == 0)
+ cast = "(u32)";
+ snprintf(buf, buf_size, "%sap", cast);
+ return;
+ }
+ ferr(po, "offset %d (%s,%d) doesn't map to any arg\n",
+ offset, bp_arg, arg_i);
+ }
if (ofs_reg[0] != 0)
ferr(po, "offset reg on arg access?\n");
- for (i = arg_s = 0; i < g_func_pp.argc; i++) {
- if (g_func_pp.arg[i].reg != NULL)
+ for (i = arg_s = 0; i < g_func_pp->argc; i++) {
+ if (g_func_pp->arg[i].reg != NULL)
continue;
if (arg_s == arg_i)
break;
arg_s++;
}
- if (i == g_func_pp.argc)
+ if (i == g_func_pp->argc)
ferr(po, "arg %d not in prototype?\n", arg_i);
- if (is_lea)
- ferr(po, "lea to arg?\n");
- switch (lmod)
+ popr->is_ptr = g_func_pp->arg[i].type.is_ptr;
+
+ switch (popr->lmod)
{
case OPLM_BYTE:
+ if (is_lea)
+ ferr(po, "lea/byte to arg?\n");
if (is_src && (offset & 3) == 0)
snprintf(buf, buf_size, "(u8)a%d", i + 1);
else
break;
case OPLM_WORD:
+ if (is_lea)
+ ferr(po, "lea/word to arg?\n");
if (offset & 1)
ferr(po, "unaligned arg access\n");
if (is_src && (offset & 2) == 0)
break;
case OPLM_DWORD:
- if (offset & 3)
- ferr(po, "unaligned arg access\n");
if (cast[0])
prefix = cast;
else if (is_src)
prefix = "(u32)";
- snprintf(buf, buf_size, "%sa%d", prefix, i + 1);
+ if (offset & 3) {
+ snprintf(g_comment, sizeof(g_comment), "%s unaligned", bp_arg);
+ if (is_lea)
+ snprintf(buf, buf_size, "(u32)&a%d + %d",
+ i + 1, offset & 3);
+ else
+ snprintf(buf, buf_size, "%s(a%d >> %d)",
+ prefix, i + 1, (offset & 3) * 8);
+ }
+ else {
+ snprintf(buf, buf_size, "%s%sa%d",
+ prefix, is_lea ? "&" : "", i + 1);
+ }
break;
default:
- ferr(po, "bp_arg bad lmod: %d\n", lmod);
+ ferr(po, "bp_arg bad lmod: %d\n", popr->lmod);
}
// common problem
- guess_lmod_from_c_type(&tmp_lmod, &g_func_pp.arg[i].type);
+ guess_lmod_from_c_type(&tmp_lmod, &g_func_pp->arg[i].type);
if ((offset & 3) && tmp_lmod != OPLM_DWORD)
ferr(po, "bp_arg arg/w offset %d and type '%s'\n",
- offset, g_func_pp.arg[i].type.name);
+ offset, g_func_pp->arg[i].type.name);
}
- else {
+ else
+ {
if (g_stack_fsz == 0)
ferr(po, "stack var access without stackframe\n");
sf_ofs = g_stack_fsz + offset;
- if (sf_ofs < 0)
+ lim = (ofs_reg[0] != 0) ? -4 : 0;
+ if (offset > 0 || sf_ofs < lim)
ferr(po, "bp_stack offset %d/%d\n", offset, g_stack_fsz);
if (is_lea)
else
prefix = cast;
- switch (lmod)
+ switch (popr->lmod)
{
case OPLM_BYTE:
snprintf(buf, buf_size, "%ssf.b[%d%s%s]",
break;
default:
- ferr(po, "bp_stack bad lmod: %d\n", lmod);
+ ferr(po, "bp_stack bad lmod: %d\n", popr->lmod);
}
}
}
break;
case OPT_REGMEM:
- if (parse_stack_el(popr->name, NULL)) {
- stack_frame_access(po, popr->lmod, buf, buf_size,
+ if (parse_stack_el(popr->name, NULL)
+ || (g_bp_frame && IS_START(popr->name, "ebp-")))
+ {
+ stack_frame_access(po, popr, buf, buf_size,
popr->name, cast, 1, is_lea);
break;
}
check_label_read_ref(po, popr->name);
if (cast[0] == 0 && popr->is_ptr)
cast = "(u32)";
+
if (is_lea)
snprintf(buf, buf_size, "(u32)&%s", popr->name);
+ else if (popr->size_lt)
+ snprintf(buf, buf_size, "%s%s%s%s", cast,
+ lmod_cast_u_ptr(po, popr->lmod),
+ popr->is_array ? "" : "&",
+ popr->name);
else
snprintf(buf, buf_size, "%s%s%s", cast, popr->name,
popr->is_array ? "[0]" : "");
return buf;
}
+// note: may set is_ptr (we find that out late for ebp frame..)
static char *out_dst_opr(char *buf, size_t buf_size,
struct parsed_op *po, struct parsed_opr *popr)
{
break;
case OPT_REGMEM:
- if (parse_stack_el(popr->name, NULL)) {
- stack_frame_access(po, popr->lmod, buf, buf_size,
+ if (parse_stack_el(popr->name, NULL)
+ || (g_bp_frame && IS_START(popr->name, "ebp-")))
+ {
+ stack_frame_access(po, popr, buf, buf_size,
popr->name, "", 0, 0);
break;
}
return out_src_opr(buf, buf_size, po, popr, NULL, 0);
case OPT_LABEL:
- snprintf(buf, buf_size, "%s%s", popr->name,
- popr->is_array ? "[0]" : "");
+ if (popr->size_mismatch)
+ snprintf(buf, buf_size, "%s%s%s",
+ lmod_cast_u_ptr(po, popr->lmod),
+ popr->is_array ? "" : "&", popr->name);
+ else
+ snprintf(buf, buf_size, "%s%s", popr->name,
+ popr->is_array ? "[0]" : "");
break;
default:
po->flags |= flag;
}
+// last op in stream - unconditional branch or ret
+#define LAST_OP(_i) ((ops[_i].flags & OPF_TAIL) \
+ || (ops[_i].flags & (OPF_JMP|OPF_CC)) == OPF_JMP)
+
static int scan_for_pop(int i, int opcnt, const char *reg,
int magic, int depth, int *maxdepth, int do_flags)
{
return -1; // deadend
if ((po->flags & OPF_RMD)
- || (po->op == OP_PUSH && po->argmask)) // arg push
+ || (po->op == OP_PUSH && po->argnum != 0)) // arg push
continue;
if ((po->flags & OPF_JMP) && po->op != OP_CALL) {
if ((po->flags & OPF_RMD) || !(po->flags & OPF_DATA))
return 0;
- if (po_test->regmask_src & po->regmask_dst)
+ if (po_test->operand_cnt == 1 && po_test->operand[0].type == OPT_CONST)
+ return 0;
+
+ if ((po_test->regmask_src | po_test->regmask_dst) & po->regmask_dst)
+ return 1;
+
+ // in reality, it can wreck any register, but in decompiled C
+ // version it can only overwrite eax or edx:eax
+ if (po->op == OP_CALL
+ && ((po_test->regmask_src | po_test->regmask_dst)
+ & ((1 << xAX)|(1 << xDX))))
return 1;
for (i = 0; i < po_test->operand_cnt; i++)
// scan for any po_test operand modification in range given
static int scan_for_mod(struct parsed_op *po_test, int i, int opcnt)
{
+ if (po_test->operand_cnt == 1 && po_test->operand[0].type == OPT_CONST)
+ return -1;
+
for (; i < opcnt; i++) {
if (is_any_opr_modified(po_test, &ops[i]))
return i;
return -1;
}
-static int scan_for_flag_set(int i)
+static int scan_for_flag_set(int i, int *branched, int *setters,
+ int *setter_cnt)
{
- for (; i >= 0; i--) {
- if (ops[i].flags & OPF_FLAGS)
- return i;
+ int ret;
+
+ while (i >= 0) {
+ if (g_labels[i][0] != 0) {
+ *branched = 1;
+ if (g_label_refs[i].next != NULL)
+ return -1;
+ if (i > 0 && LAST_OP(i - 1)) {
+ i = g_label_refs[i].i;
+ continue;
+ }
+ ret = scan_for_flag_set(g_label_refs[i].i, branched,
+ setters, setter_cnt);
+ if (ret < 0)
+ return ret;
+ }
+ i--;
+
+ if (ops[i].flags & OPF_FLAGS) {
+ setters[*setter_cnt] = i;
+ (*setter_cnt)++;
+ return 0;
+ }
if ((ops[i].flags & OPF_JMP) && !(ops[i].flags & OPF_CC))
return -1;
- if (g_labels[i][0] != 0)
- return -1;
}
return -1;
*adj -= lmod_bytes(po, po->operand[0].lmod);
else if (po->op == OP_POP)
*adj += lmod_bytes(po, po->operand[0].lmod);
- else if ((po->flags & (OPF_JMP|OPF_TAIL)) && po->op != OP_CALL)
- return -1;
+ else if (po->flags & (OPF_JMP|OPF_TAIL)) {
+ if (po->op != OP_CALL)
+ return -1;
+ if (po->operand[0].type != OPT_LABEL)
+ return -1;
+ // TODO: should only allow combining __cdecl calls..
+ }
if (g_labels[i][0] != 0)
return -1;
return -1;
}
+static int collect_call_args(struct parsed_op *po, int i,
+ struct parsed_proto *pp, int *save_arg_vars, int arg,
+ int need_op_saving, int may_reuse)
+{
+ struct parsed_proto *pp_tmp;
+ struct label_ref *lr;
+ int need_to_save_current;
+ int ret = 0;
+ int j;
+
+ if (i < 0)
+ ferr(po, "no refs for '%s'?\n", g_labels[i]);
+
+ for (; arg < pp->argc; arg++)
+ if (pp->arg[arg].reg == NULL)
+ break;
+
+ for (j = i; j >= 0 && arg < pp->argc; )
+ {
+ if (g_labels[j][0] != 0) {
+ lr = &g_label_refs[j];
+ if (lr->next != NULL)
+ need_op_saving = 1;
+ for (; lr->next; lr = lr->next) {
+ if ((ops[lr->i].flags & (OPF_JMP|OPF_CC)) != OPF_JMP)
+ may_reuse = 1;
+ ret |= collect_call_args(po, lr->i, pp, save_arg_vars,
+ arg, need_op_saving, may_reuse);
+ }
+
+ if ((ops[lr->i].flags & (OPF_JMP|OPF_CC)) != OPF_JMP)
+ may_reuse = 1;
+ if (j > 0 && LAST_OP(j - 1)) {
+ // follow last branch in reverse
+ j = lr->i;
+ continue;
+ }
+ need_op_saving = 1;
+ ret |= collect_call_args(po, lr->i, pp, save_arg_vars,
+ arg, need_op_saving, may_reuse);
+ }
+ j--;
+
+ if (ops[j].op == OP_CALL)
+ {
+ pp_tmp = ops[j].datap;
+ if (pp_tmp == NULL)
+ ferr(po, "arg collect hit unparsed call\n");
+ if (may_reuse && pp_tmp->argc_stack > 0)
+ ferr(po, "arg collect %d/%d hit '%s' with %d stack args\n",
+ arg, pp->argc, opr_name(&ops[j], 0), pp_tmp->argc_stack);
+ }
+ else if (ops[j].op == OP_ADD && ops[j].operand[0].reg == xSP) {
+ ferr(po, "arg collect %d/%d hit esp adjust\n",
+ arg, pp->argc);
+ }
+ else if (ops[j].op == OP_POP) {
+ ferr(po, "arg collect %d/%d hit pop\n", arg, pp->argc);
+ }
+ else if ((ops[j].flags & (OPF_JMP|OPF_CC)) == (OPF_JMP|OPF_CC))
+ {
+ may_reuse = 1;
+ }
+ else if (ops[j].op == OP_PUSH && !(ops[j].flags & OPF_FARG))
+ {
+ pp->arg[arg].datap = &ops[j];
+ need_to_save_current = 0;
+ if (!need_op_saving) {
+ ret = scan_for_mod(&ops[j], j + 1, 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].argnum == 0) {
+ ops[j].argnum = arg + 1;
+ *save_arg_vars |= 1 << arg;
+ }
+ else if (ops[j].argnum < arg + 1)
+ ferr(&ops[j], "argnum conflict (%d<%d) for '%s'\n",
+ ops[j].argnum, arg + 1, pp->name);
+ }
+ else if (ops[j].argnum == 0)
+ ops[j].flags |= OPF_RMD;
+
+ // some PUSHes are reused by different calls on other branches,
+ // but that can't happen if we didn't branch, so they
+ // can be removed from future searches (handles nested calls)
+ if (!may_reuse)
+ ops[j].flags |= OPF_FARG;
+
+ // next arg
+ for (arg++; arg < pp->argc; arg++)
+ if (pp->arg[arg].reg == NULL)
+ break;
+ }
+ }
+
+ if (arg < pp->argc) {
+ ferr(po, "arg collect failed for '%s': %d/%d\n",
+ pp->name, arg, pp->argc);
+ ret = -1;
+ }
+ return ret;
+}
+
static void add_label_ref(struct label_ref *lr, int op_i)
{
struct label_ref *lr_new;
lr_new = calloc(1, sizeof(*lr_new));
lr_new->i = op_i;
+ lr_new->next = lr->next;
lr->next = lr_new;
}
struct parsed_op *po, *delayed_flag_op = NULL, *tmp_op;
struct parsed_opr *last_arith_dst = NULL;
char buf1[256], buf2[256], buf3[256], cast[64];
+ const struct parsed_proto *pp_c;
struct parsed_proto *pp, *pp_tmp;
struct parsed_data *pd;
const char *tmpname;
g_bp_frame = g_sp_frame = g_stack_fsz = 0;
- ret = proto_parse(fhdr, funcn, &g_func_pp);
- if (ret)
+ g_func_pp = proto_parse(fhdr, funcn);
+ if (g_func_pp == NULL)
ferr(ops, "proto_parse failed for '%s'\n", funcn);
- fprintf(fout, "%s %s(", g_func_pp.ret_type.name, funcn);
- for (i = 0; i < g_func_pp.argc; i++) {
+ fprintf(fout, "%s ", g_func_pp->ret_type.name);
+ if (g_func_pp->is_stdcall && g_func_pp->argc_reg == 0)
+ fprintf(fout, "__stdcall ");
+ if (g_ida_func_attr & IDAFA_NORETURN)
+ fprintf(fout, "noreturn ");
+ fprintf(fout, "%s(", funcn);
+
+ for (i = 0; i < g_func_pp->argc; i++) {
+ if (i > 0)
+ fprintf(fout, ", ");
+ if (g_func_pp->arg[i].fptr != NULL) {
+ // func pointer..
+ pp = g_func_pp->arg[i].fptr;
+ fprintf(fout, "%s (", pp->ret_type.name);
+ if (pp->is_stdcall && pp->argc_reg == 0)
+ fprintf(fout, "__stdcall ");
+ fprintf(fout, "*a%d)(", i + 1);
+ for (j = 0; j < pp->argc; j++) {
+ if (j > 0)
+ fprintf(fout, ", ");
+ if (pp->arg[j].fptr)
+ ferr(ops, "nested fptr\n");
+ fprintf(fout, "%s", pp->arg[j].type.name);
+ }
+ fprintf(fout, ")");
+ }
+ else {
+ fprintf(fout, "%s a%d", g_func_pp->arg[i].type.name, i + 1);
+ }
+ }
+ if (g_func_pp->is_vararg) {
if (i > 0)
fprintf(fout, ", ");
- fprintf(fout, "%s a%d", g_func_pp.arg[i].type.name, i + 1);
+ fprintf(fout, "...");
}
+
fprintf(fout, ")\n{\n");
// pass1:
if (i == opcnt && (ops[i - 1].flags & OPF_JMP) && found)
break;
- if (ops[i - 1].op != OP_POP || !IS(opr_name(&ops[i - 1], 0), "ebp"))
+ if (ops[i - 1].op == OP_POP && IS(opr_name(&ops[i - 1], 0), "ebp"))
+ ops[i - 1].flags |= OPF_RMD;
+ else if (!(g_ida_func_attr & IDAFA_NORETURN))
ferr(&ops[i - 1], "'pop ebp' expected\n");
- ops[i - 1].flags |= OPF_RMD;
if (g_stack_fsz != 0) {
- if (ops[i - 2].op != OP_MOV
- || !IS(opr_name(&ops[i - 2], 0), "esp")
- || !IS(opr_name(&ops[i - 2], 1), "ebp"))
+ if (ops[i - 2].op == OP_MOV
+ && IS(opr_name(&ops[i - 2], 0), "esp")
+ && IS(opr_name(&ops[i - 2], 1), "ebp"))
{
- ferr(&ops[i - 2], "esp restore expected\n");
+ ops[i - 2].flags |= OPF_RMD;
}
- ops[i - 2].flags |= OPF_RMD;
+ else if (!(g_ida_func_attr & IDAFA_NORETURN))
+ ferr(&ops[i - 2], "esp restore expected\n");
if (ecx_push && ops[i - 3].op == OP_POP
&& IS(opr_name(&ops[i - 3], 0), "ecx"))
// pass2:
// - resolve all branches
- for (i = 0; i < opcnt; i++) {
+ for (i = 0; i < opcnt; i++)
+ {
po = &ops[i];
po->bt_i = -1;
po->btj = NULL;
pp->arg[arg].type.name = strdup("int");
}
else {
- ret = proto_parse(fhdr, tmpname, pp);
- if (ret)
+ pp_c = proto_parse(fhdr, tmpname);
+ if (pp_c == NULL)
ferr(po, "proto_parse failed for call '%s'\n", tmpname);
+ pp = proto_clone(pp_c);
}
// look for and make use of esp adjust
ret = -1;
- if (!pp->is_stdcall)
+ if (!pp->is_stdcall && pp->argc_stack > 0)
ret = scan_for_esp_adjust(i + 1, opcnt, &j);
if (ret >= 0) {
if (pp->is_vararg) {
pp->argc_stack++;
}
if (pp->argc > ARRAY_SIZE(pp->arg))
- ferr(po, "too many args\n");
+ ferr(po, "too many args for '%s'\n", tmpname);
}
if (pp->argc_stack != j / 4)
- ferr(po, "stack tracking failed: %x %x\n",
- pp->argc_stack * 4, j);
+ ferr(po, "stack tracking failed for '%s': %x %x\n",
+ tmpname, pp->argc_stack * 4, j);
ops[ret].flags |= OPF_RMD;
// a bit of a hack, but deals with use of
}
}
- // collect all stack args
- for (arg = 0; arg < pp->argc; arg++)
- if (pp->arg[arg].reg == NULL)
- break;
+ collect_call_args(po, i, pp, &save_arg_vars, 0, 0, 0);
- for (j = i; j >= 0 && arg < pp->argc; )
- {
- if (g_labels[j][0] != 0) {
- if (j > 0 && ((ops[j - 1].flags & OPF_TAIL)
- || (ops[j - 1].flags & (OPF_JMP|OPF_CC)) == OPF_JMP))
- {
- // follow the branch in reverse
- if (g_label_refs[j].i == -1)
- ferr(po, "no refs for '%s'?\n", g_labels[j]);
- if (g_label_refs[j].next != NULL)
- ferr(po, "unhandled multiple refs to '%s'\n", g_labels[j]);
- j = g_label_refs[j].i + 1;
- continue;
- }
- break;
- }
- j--;
-
- if (ops[j].op == OP_CALL)
- {
- pp_tmp = ops[j].datap;
- if (pp_tmp == NULL)
- ferr(po, "arg collect hit unparsed call\n");
- if (pp_tmp->argc_stack > 0)
- ferr(po, "arg collect hit '%s' with %d stack args\n",
- opr_name(&ops[j], 0), pp_tmp->argc_stack);
- }
- else if ((ops[j].flags & OPF_TAIL)
- || (ops[j].flags & (OPF_JMP|OPF_CC)) == OPF_JMP)
- {
- break;
- }
- else if (ops[j].op == OP_PUSH)
- {
- pp->arg[arg].datap = &ops[j];
- ret = scan_for_mod(&ops[j], j + 1, i);
- if (ret >= 0) {
- // mark this push as one that needs operand saving
- ops[j].flags &= ~OPF_RMD;
- ops[j].argmask |= 1 << arg;
- save_arg_vars |= 1 << arg;
- }
- else
- ops[j].flags |= OPF_RMD;
-
- // next arg
- for (arg++; arg < pp->argc; arg++)
- if (pp->arg[arg].reg == NULL)
- break;
- }
- }
- if (arg < pp->argc)
- ferr(po, "arg collect failed for '%s': %d/%d\n",
- tmpname, arg, pp->argc);
+ if (strstr(pp->ret_type.name, "int64"))
+ need_mul_var = 1;
po->datap = pp;
}
}
continue;
if (po->op == OP_PUSH
- && po->argmask == 0 && !(po->flags & OPF_RSAVE)
+ && po->argnum == 0 && !(po->flags & OPF_RSAVE)
&& po->operand[0].type == OPT_REG)
{
reg = po->operand[0].reg;
if (po->flags & OPF_CC)
{
- ret = scan_for_flag_set(i - 1);
- if (ret < 0)
- ferr(po, "unable to trace flag setter\n");
+ int setters[16], cnt = 0, branched = 0;
+
+ ret = scan_for_flag_set(i, &branched, setters, &cnt);
+ if (ret < 0 || cnt <= 0)
+ ferr(po, "unable to trace flag setter(s)\n");
+ if (cnt > ARRAY_SIZE(setters))
+ ferr(po, "too many flag setters\n");
- tmp_op = &ops[ret]; // flag setter
pfo = split_cond(po, po->op, &dummy);
- pfomask = 0;
-
- // to get nicer code, we try to delay test and cmp;
- // if we can't because of operand modification, or if we
- // have math op, make it calculate flags explicitly
- if (tmp_op->op == OP_TEST || tmp_op->op == OP_CMP) {
- if (scan_for_mod(tmp_op, ret + 1, i) >= 0)
- pfomask = 1 << pfo;
- }
- else if (tmp_op->op == OP_CMPS) {
- pfomask = 1 << PFO_Z;
- }
- else {
- if ((pfo != PFO_Z && pfo != PFO_S && pfo != PFO_P)
- || scan_for_mod_opr0(tmp_op, ret + 1, i) >= 0)
- pfomask = 1 << pfo;
- }
- if (pfomask) {
- tmp_op->pfomask |= pfomask;
- cmp_result_vars |= pfomask;
- po->datap = tmp_op;
+ for (j = 0; j < cnt; j++)
+ {
+ tmp_op = &ops[setters[j]]; // flag setter
+ pfomask = 0;
+
+ // to get nicer code, we try to delay test and cmp;
+ // if we can't because of operand modification, or if we
+ // have math op, or branch, make it calculate flags explicitly
+ if (tmp_op->op == OP_TEST || tmp_op->op == OP_CMP) {
+ if (branched || scan_for_mod(tmp_op, setters[j] + 1, i) >= 0)
+ pfomask = 1 << pfo;
+ }
+ else if (tmp_op->op == OP_CMPS) {
+ pfomask = 1 << PFO_Z;
+ }
+ else {
+ if ((pfo != PFO_Z && pfo != PFO_S && pfo != PFO_P)
+ || scan_for_mod_opr0(tmp_op, setters[j] + 1, i) >= 0)
+ pfomask = 1 << pfo;
+ }
+ if (pfomask) {
+ tmp_op->pfomask |= pfomask;
+ cmp_result_vars |= pfomask;
+ // note: may overwrite, currently not a problem
+ po->datap = tmp_op;
+ }
}
if (po->op == OP_ADC || po->op == OP_SBB)
fprintf(fout, " };\n");
}
- // declare stack frame
+ // 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);
+ if (g_func_pp->is_vararg)
+ fprintf(fout, " va_list ap;\n");
+
// declare arg-registers
- for (i = 0; i < g_func_pp.argc; i++) {
- if (g_func_pp.arg[i].reg != NULL) {
+ for (i = 0; i < g_func_pp->argc; i++) {
+ if (g_func_pp->arg[i].reg != NULL) {
reg = char_array_i(regs_r32,
- ARRAY_SIZE(regs_r32), g_func_pp.arg[i].reg);
+ ARRAY_SIZE(regs_r32), g_func_pp->arg[i].reg);
if (reg < 0)
- ferr(ops, "arg '%s' is not a reg?\n", g_func_pp.arg[i].reg);
+ ferr(ops, "arg '%s' is not a reg?\n", g_func_pp->arg[i].reg);
regmask_arg |= 1 << reg;
fprintf(fout, " u32 %s = (u32)a%d;\n",
- g_func_pp.arg[i].reg, i + 1);
+ g_func_pp->arg[i].reg, i + 1);
had_decl = 1;
}
}
// declare other regs - special case for eax
if (!((regmask | regmask_arg) & 1)
- && !IS(g_func_pp.ret_type.name, "void"))
+ && !IS(g_func_pp->ret_type.name, "void"))
{
fprintf(fout, " u32 eax = 0;\n");
had_decl = 1;
if (had_decl)
fprintf(fout, "\n");
+ if (g_func_pp->is_vararg) {
+ if (g_func_pp->argc_stack == 0)
+ ferr(ops, "vararg func without stack args?\n");
+ fprintf(fout, " va_start(ap, a%d);\n", g_func_pp->argc);
+ }
+
// output ops
for (i = 0; i < opcnt; i++)
{
if (g_labels[i][0] != 0 && g_label_refs[i].i != -1) {
fprintf(fout, "\n%s:\n", g_labels[i]);
label_pending = 1;
+
+ delayed_flag_op = NULL;
+ last_arith_dst = NULL;
}
po = &ops[i];
case OP_MOV:
assert_operand_cnt(2);
propagate_lmod(po, &po->operand[0], &po->operand[1]);
- fprintf(fout, " %s = %s;",
- out_dst_opr(buf1, sizeof(buf1), po, &po->operand[0]),
+ out_dst_opr(buf1, sizeof(buf1), po, &po->operand[0]);
+ fprintf(fout, " %s = %s;", buf1,
out_src_opr(buf2, sizeof(buf2), po, &po->operand[1],
po->operand[0].is_ptr ? "(void *)" : "", 0));
break;
break;
case OP_IMUL:
- if (po->operand_cnt == 2)
+ if (po->operand_cnt == 2) {
+ propagate_lmod(po, &po->operand[0], &po->operand[1]);
goto dualop_arith;
+ }
if (po->operand_cnt == 3)
ferr(po, "TODO imul3\n");
// fallthrough
case OP_JMP:
assert_operand_cnt(1);
+ last_arith_dst = NULL;
+ delayed_flag_op = NULL;
+
if (po->operand[0].type == OPT_REGMEM) {
ret = sscanf(po->operand[0].name, "%[^[][%[^*]*4]",
buf1, buf2);
out_src_opr_u32(buf1, sizeof(buf1), po, &po->operand[0]));
fprintf(fout, " ");
- if (!IS(pp->ret_type.name, "void")) {
+ if (strstr(pp->ret_type.name, "int64")) {
if (po->flags & OPF_TAIL)
+ ferr(po, "int64 and tail?\n");
+ fprintf(fout, "mul_tmp = ");
+ }
+ else if (!IS(pp->ret_type.name, "void")) {
+ if (po->flags & OPF_TAIL) {
fprintf(fout, "return ");
- else
+ if (g_func_pp->ret_type.is_ptr != pp->ret_type.is_ptr)
+ fprintf(fout, "(%s)", g_func_pp->ret_type.name);
+ }
+ else {
fprintf(fout, "eax = ");
- if (pp->ret_type.is_ptr)
- fprintf(fout, "(u32)");
+ if (pp->ret_type.is_ptr)
+ fprintf(fout, "(u32)");
+ }
}
if (po->operand[0].type != OPT_LABEL) {
else {
if (pp->name[0] == 0)
ferr(po, "missing pp->name\n");
- fprintf(fout, "%s(", pp->name);
+ fprintf(fout, "%s%s(", pp->name,
+ pp->has_structarg ? "_sa" : "");
}
for (arg = 0; arg < pp->argc; arg++) {
tmp_op = pp->arg[arg].datap;
if (tmp_op == NULL)
ferr(po, "parsed_op missing for arg%d\n", arg);
- if (tmp_op->argmask) {
- fprintf(fout, "%ss_a%d", cast, arg + 1);
+ if (tmp_op->argnum != 0) {
+ fprintf(fout, "%ss_a%d", cast, tmp_op->argnum);
}
else {
fprintf(fout, "%s",
}
fprintf(fout, ");");
+ if (strstr(pp->ret_type.name, "int64")) {
+ fprintf(fout, "\n");
+ fprintf(fout, " edx = mul_tmp >> 32;\n");
+ fprintf(fout, " eax = mul_tmp;");
+ }
+
if (po->flags & OPF_TAIL) {
strcpy(g_comment, "tailcall");
- if (IS(pp->ret_type.name, "void")) {
+ if (IS(pp->ret_type.name, "void")
+ && !(g_ida_func_attr & IDAFA_NORETURN))
+ {
fprintf(fout, "\n return;");
strcpy(g_comment, "^ tailcall");
}
break;
case OP_RET:
- if (IS(g_func_pp.ret_type.name, "void")) {
+ if (g_func_pp->is_vararg)
+ fprintf(fout, " va_end(ap);\n");
+
+ if (IS(g_func_pp->ret_type.name, "void")) {
if (i != opcnt - 1 || label_pending)
fprintf(fout, " return;");
}
- else if (g_func_pp.ret_type.is_ptr) {
+ else if (g_func_pp->ret_type.is_ptr) {
fprintf(fout, " return (%s)eax;",
- g_func_pp.ret_type.name);
+ g_func_pp->ret_type.name);
}
else
fprintf(fout, " return eax;");
+
+ last_arith_dst = NULL;
+ delayed_flag_op = NULL;
break;
case OP_PUSH:
- if (po->argmask) {
+ if (po->argnum != 0) {
// special case - saved func arg
out_src_opr_u32(buf1, sizeof(buf1), po, &po->operand[0]);
- for (j = 0; j < 32; j++) {
- if (po->argmask & (1 << j))
- fprintf(fout, " s_a%d = %s;", j + 1, buf1);
- }
+ fprintf(fout, " s_a%d = %s;", po->argnum, buf1);
break;
}
else if (po->flags & OPF_RSAVE) {
fprintf(fout, " s_%s = %s;", buf1, buf1);
break;
}
- ferr(po, "stray push encountered\n");
+ if (!(g_ida_func_attr & IDAFA_NORETURN))
+ ferr(po, "stray push encountered\n");
+ no_output = 1;
break;
case OP_POP:
break;
case OP_NOP:
+ no_output = 1;
break;
default:
break;
}
- // some sanity checking
- if ((po->flags & OPF_REP) && po->op != OP_STOS
- && po->op != OP_MOVS && po->op != OP_CMPS)
- ferr(po, "unexpected rep\n");
- if ((po->flags & (OPF_REPZ|OPF_REPNZ)) && po->op != OP_CMPS)
- ferr(po, "unexpected repz/repnz\n");
-
if (g_comment[0] != 0) {
fprintf(fout, " // %s", g_comment);
g_comment[0] = 0;
if (!no_output)
fprintf(fout, "\n");
+ // some sanity checking
+ if ((po->flags & OPF_REP) && po->op != OP_STOS
+ && po->op != OP_MOVS && po->op != OP_CMPS)
+ ferr(po, "unexpected rep\n");
+ if ((po->flags & (OPF_REPZ|OPF_REPNZ)) && po->op != OP_CMPS)
+ ferr(po, "unexpected repz/repnz\n");
+
if (pfomask != 0)
ferr(po, "missed flag calc, pfomask=%x\n", pfomask);
if (ops[i].op == OP_CALL) {
pp = ops[i].datap;
- if (pp) {
+ if (pp)
proto_release(pp);
- free(pp);
- }
}
}
- proto_release(&g_func_pp);
+ g_func_pp = NULL;
}
static void set_label(int i, const char *name)
if (len > sizeof(g_labels[0]) - 1)
aerr("label '%s' too long: %d\n", name, len);
- if (g_labels[i][0] != 0)
- aerr("dupe label '%s'?\n", name);
+ if (g_labels[i][0] != 0 && !IS_START(g_labels[i], "algn_"))
+ aerr("dupe label '%s' vs '%s'?\n", name, g_labels[i]);
memcpy(g_labels[i], name, len);
g_labels[i][len] = 0;
}
return s + i;
}
+struct chunk_item {
+ char *name;
+ long fptr;
+ int asmln;
+};
+
+static int cmp_chunks(const void *p1, const void *p2)
+{
+ const struct chunk_item *c1 = p1, *c2 = p2;
+ return strcmp(c1->name, c2->name);
+}
+
static int cmpstringp(const void *p1, const void *p2)
{
return strcmp(*(char * const *)p1, *(char * const *)p2);
char **rlist = NULL;
int rlist_len = 0;
int rlist_alloc = 0;
+ struct chunk_item *func_chunks;
+ int func_chunks_used = 0;
+ int func_chunks_sorted = 0;
+ int func_chunk_cnt = 0;
+ int func_chunk_alloc;
+ int func_chunk_i = -1;
+ long func_chunk_ret = 0;
+ int func_chunk_ret_ln = 0;
char line[256];
char words[16][256];
enum opr_lenmod lmod;
int arg_out;
int arg = 1;
int pi = 0;
- int i, j, len;
+ int i, j;
+ int ret, len;
char *p;
int wordc;
// needs special handling..
rlist[rlist_len++] = "__alloca_probe";
+ func_chunk_alloc = 32;
+ func_chunks = malloc(func_chunk_alloc * sizeof(func_chunks[0]));
+ my_assert_not(func_chunks, NULL);
+
for (; arg < argc; arg++) {
frlist = fopen(argv[arg], "r");
my_assert_not(frlist, NULL);
if (*p == 0)
continue;
- if (*p == ';') {
- static const char *attrs[] = {
- "bp-based frame",
- "library function",
- "static",
- "noreturn",
- "thunk",
- "fpd=",
- };
+ // get rid of random tabs
+ for (i = 0; line[i] != 0; i++)
+ if (line[i] == '\t')
+ line[i] = ' ';
+
+ if (*p == ';')
+ {
if (p[2] == '=' && IS_START(p, "; =============== S U B"))
goto do_pending_endp; // eww..
- if (p[2] != 'A' || !IS_START(p, "; Attributes:"))
- continue;
-
- // parse IDA's attribute-list comment
- g_ida_func_attr = 0;
- p = sskip(p + 13);
- // get rid of random tabs
- for (i = 0; p[i] != 0; i++)
- if (p[i] == '\t')
- p[i] = ' ';
-
- for (; *p != 0; p = sskip(p)) {
- for (i = 0; i < ARRAY_SIZE(attrs); i++) {
- if (!strncmp(p, attrs[i], strlen(attrs[i]))) {
- g_ida_func_attr |= 1 << i;
- p += strlen(attrs[i]);
+ if (p[2] == 'A' && IS_START(p, "; Attributes:"))
+ {
+ static const char *attrs[] = {
+ "bp-based frame",
+ "library function",
+ "static",
+ "noreturn",
+ "thunk",
+ "fpd=",
+ };
+
+ // parse IDA's attribute-list comment
+ g_ida_func_attr = 0;
+ p = sskip(p + 13);
+
+ for (; *p != 0; p = sskip(p)) {
+ for (i = 0; i < ARRAY_SIZE(attrs); i++) {
+ if (!strncmp(p, attrs[i], strlen(attrs[i]))) {
+ g_ida_func_attr |= 1 << i;
+ p += strlen(attrs[i]);
+ break;
+ }
+ }
+ if (i == ARRAY_SIZE(attrs)) {
+ anote("unparsed IDA attr: %s\n", p);
break;
}
+ if (IS(attrs[i], "fpd=")) {
+ p = next_word(words[0], sizeof(words[0]), p);
+ // ignore for now..
+ }
}
- if (i == ARRAY_SIZE(attrs)) {
- anote("unparsed IDA attr: %s\n", p);
- break;
+ }
+ else if (p[2] == 'S' && IS_START(p, "; START OF FUNCTION CHUNK FOR "))
+ {
+ p += 30;
+ next_word(words[0], sizeof(words[0]), p);
+ if (words[0][0] == 0)
+ aerr("missing nam for func chunk?\n");
+ if (func_chunk_cnt >= func_chunk_alloc) {
+ func_chunk_alloc *= 2;
+ func_chunks = realloc(func_chunks,
+ func_chunk_alloc * sizeof(func_chunks[0]));
+ my_assert_not(func_chunks, NULL);
+ }
+ func_chunks[func_chunk_cnt].fptr = ftell(fasm);
+ func_chunks[func_chunk_cnt].name = strdup(words[0]);
+ func_chunks[func_chunk_cnt].asmln = asmln;
+ func_chunk_cnt++;
+ func_chunks_sorted = 0;
+ }
+ else if (p[2] == 'E' && IS_START(p, "; END OF FUNCTION CHUNK"))
+ {
+ if (func_chunk_i >= 0) {
+ if (func_chunk_i < func_chunk_cnt
+ && IS(func_chunks[func_chunk_i].name, g_func))
+ {
+ // move on to next chunk
+ ret = fseek(fasm, func_chunks[func_chunk_i].fptr, SEEK_SET);
+ if (ret)
+ aerr("seek failed for '%s' chunk #%d\n",
+ g_func, func_chunk_i);
+ asmln = func_chunks[func_chunk_i].asmln;
+ func_chunk_i++;
+ }
+ else {
+ if (func_chunk_ret == 0)
+ aerr("no return from chunk?\n");
+ fseek(fasm, func_chunk_ret, SEEK_SET);
+ asmln = func_chunk_ret_ln;
+ func_chunk_ret = 0;
+ pending_endp = 1;
+ }
}
- if (IS(attrs[i], "fpd=")) {
- p = next_word(words[0], sizeof(words[0]), p);
- // ignore for now..
+ }
+ else if (p[2] == 'F' && IS_START(p, "; FUNCTION CHUNK AT ")) {
+ func_chunks_used = 1;
+ p += 20;
+ if (IS_START(g_func, "sub_")) {
+ unsigned long addr = strtoul(p, NULL, 16);
+ unsigned long f_addr = strtoul(g_func + 4, NULL, 16);
+ if (addr > f_addr)
+ aerr("need a chunk %lX that is after %s\n", addr, g_func);
}
}
continue;
- }
+ } // *p == ';'
parse_words:
memset(words, 0, sizeof(words));
skip_warned = 0;
skip_func = 0;
g_func[0] = 0;
+ func_chunks_used = 0;
+ func_chunk_i = -1;
if (pi != 0) {
memset(&ops, 0, pi * sizeof(ops[0]));
memset(g_labels, 0, pi * sizeof(g_labels[0]));
continue;
}
- if (IS(words[1], "endp")) {
+ if (IS(words[1], "endp"))
+ {
if (!in_func)
aerr("endp '%s' while not in_func?\n", words[0]);
if (!IS(g_func, words[0]))
aerr("endp '%s' while in_func '%s'?\n",
words[0], g_func);
+ if (!skip_func && func_chunks_used) {
+ // start processing chunks
+ struct chunk_item *ci, key = { g_func, 0 };
+
+ func_chunk_ret = ftell(fasm);
+ func_chunk_ret_ln = asmln;
+ if (!func_chunks_sorted) {
+ qsort(func_chunks, func_chunk_cnt,
+ sizeof(func_chunks[0]), cmp_chunks);
+ func_chunks_sorted = 1;
+ }
+ ci = bsearch(&key, func_chunks, func_chunk_cnt,
+ sizeof(func_chunks[0]), cmp_chunks);
+ if (ci == NULL)
+ aerr("'%s' needs chunks, but none found\n", g_func);
+ func_chunk_i = ci - func_chunks;
+ for (; func_chunk_i > 0; func_chunk_i--)
+ if (!IS(func_chunks[func_chunk_i - 1].name, g_func))
+ break;
+
+ ret = fseek(fasm, func_chunks[func_chunk_i].fptr, SEEK_SET);
+ if (ret)
+ aerr("seek failed for '%s' chunk #%d\n", g_func, func_chunk_i);
+ asmln = func_chunks[func_chunk_i].asmln;
+ func_chunk_i++;
+ continue;
+ }
pending_endp = 1;
continue;
}