조건부 반복 명령 문서의 검증 코드다. if __name__ == "__main__": 아래가 자체 테스트다.
"""조건부 반복 명령 문서의 값 확인 (파이썬 흉내, 실제 MASM 실행 아님).
실행: python3 33_conditional-loop-instructions_verify.py"""
import random
import re
# ---------------------------------------------------------------------------
# x86 흉내기: 이 파일에서 쓰는 만큼만 흉내 낸다 (MASM 문법의 일부를 그대로 읽는다).
# 레지스터 8·16·32비트, ZF/SF/CF/OF/PF, 리틀 엔디언 메모리, 스택, CALL/RET.
# 코드 주소는 실제 기계어 길이가 아니다. 줄 앞에 8자리 16진수를 쓰면 그 주소를 쓰고,
# 안 쓰면 CALL은 5, 나머지는 2씩 늘린다(예: 00000020의 call 다음 줄은 00000025).
# ---------------------------------------------------------------------------
REG32 = ["eax", "ecx", "edx", "ebx", "esp", "ebp", "esi", "edi"]
REG16 = {"ax": "eax", "cx": "ecx", "dx": "edx", "bx": "ebx",
"sp": "esp", "bp": "ebp", "si": "esi", "di": "edi"}
REG8L = {"al": "eax", "cl": "ecx", "dl": "edx", "bl": "ebx"}
REG8H = {"ah": "eax", "ch": "ecx", "dh": "edx", "bh": "ebx"}
TYPES = {"byte": 1, "sbyte": 1, "word": 2, "sword": 2, "dword": 4, "sdword": 4}
STOP = 0xFFFFFFF0 # call_proc이 스택에 넣어 두는 가짜 복귀 주소
JCC = {
"jz": lambda f: f["ZF"], "je": lambda f: f["ZF"],
"jnz": lambda f: not f["ZF"], "jne": lambda f: not f["ZF"],
"jc": lambda f: f["CF"], "jb": lambda f: f["CF"], "jnae": lambda f: f["CF"],
"jnc": lambda f: not f["CF"], "jae": lambda f: not f["CF"], "jnb": lambda f: not f["CF"],
"jo": lambda f: f["OF"], "jno": lambda f: not f["OF"],
"js": lambda f: f["SF"], "jns": lambda f: not f["SF"],
"jp": lambda f: f["PF"], "jnp": lambda f: not f["PF"],
"ja": lambda f: not f["CF"] and not f["ZF"], "jnbe": lambda f: not f["CF"] and not f["ZF"],
"jbe": lambda f: f["CF"] or f["ZF"], "jna": lambda f: f["CF"] or f["ZF"],
"jg": lambda f: not f["ZF"] and f["SF"] == f["OF"],
"jnle": lambda f: not f["ZF"] and f["SF"] == f["OF"],
"jge": lambda f: f["SF"] == f["OF"], "jnl": lambda f: f["SF"] == f["OF"],
"jl": lambda f: f["SF"] != f["OF"], "jnge": lambda f: f["SF"] != f["OF"],
"jle": lambda f: f["ZF"] or f["SF"] != f["OF"],
"jng": lambda f: f["ZF"] or f["SF"] != f["OF"],
}
def signed(v, n):
"""n바이트 값 v를 부호 있는 정수로 읽는다."""
bits = 8 * n
return v - (1 << bits) if v >> (bits - 1) else v
class CPU:
def __init__(self, esp=0x1000):
self.r = {k: 0 for k in REG32}
self.r["esp"] = esp
self.f = dict(ZF=0, SF=0, CF=0, OF=0, PF=0)
self.mem = {} # 주소 -> 바이트
self.sym = {} # 변수 이름 -> (주소, 원소 크기, 원소 수)
self.const = {} # 기호 상수 -> 값
self.here = 0x4000 # 데이터 위치 카운터
self.code = [] # (주소, 연산, [피연산자], 원문)
self.label = {} # 레이블 -> 코드 주소
self.at = {} # 코드 주소 -> code 인덱스
self.hook = {} # 라이브러리 프로시저 흉내: 이름 -> 함수(cpu)
self.steps = 0
self.trace = None # 리스트를 넣어 두면 (원문, 레지스터, 플래그)를 쌓는다
self.calls = [] # 호출한 프로시저 이름 기록
# ---- 레지스터와 메모리 ----
def get(self, name):
name = name.lower()
if name in self.r:
return self.r[name]
if name in REG16:
return self.r[REG16[name]] & 0xFFFF
if name in REG8L:
return self.r[REG8L[name]] & 0xFF
if name in REG8H:
return (self.r[REG8H[name]] >> 8) & 0xFF
raise KeyError(name)
def put(self, name, v):
name = name.lower()
if name in self.r:
self.r[name] = v & 0xFFFFFFFF
elif name in REG16:
k = REG16[name]
self.r[k] = (self.r[k] & 0xFFFF0000) | (v & 0xFFFF)
elif name in REG8L:
k = REG8L[name]
self.r[k] = (self.r[k] & 0xFFFFFF00) | (v & 0xFF)
elif name in REG8H:
k = REG8H[name]
self.r[k] = (self.r[k] & 0xFFFF00FF) | ((v & 0xFF) << 8)
else:
raise KeyError(name)
@staticmethod
def regsize(name):
name = name.lower()
if name in REG32:
return 4
if name in REG16:
return 2
if name in REG8L or name in REG8H:
return 1
return None
def rd(self, addr, n):
"""리틀 엔디언: 낮은 주소에 낮은 바이트."""
return sum(self.mem.get(addr + i, 0) << (8 * i) for i in range(n))
def wr(self, addr, n, v):
for i in range(n):
self.mem[addr + i] = (v >> (8 * i)) & 0xFF
# ---- 데이터 정의 ----
def data(self, name, typ, values):
"""name TYPE values. 문자열은 글자마다 한 바이트, 레이블 이름은 그 코드 주소."""
n = TYPES[typ.lower()]
flat = []
for v in values:
if isinstance(v, str) and v in self.label:
flat.append(self.label[v])
elif isinstance(v, str):
flat.extend(ord(c) for c in v)
else:
flat.append(v)
addr = self.here
if name:
self.sym[name.lower()] = (addr, n, len(flat))
for v in flat:
self.wr(self.here, n, v)
self.here += n
return addr
# ---- 식과 피연산자 ----
def num(self, s):
"""상수식: 10진, 16진(h), 2진(b), 글자, 기호 상수, OFFSET/LENGTHOF/TYPE/SIZEOF, + - * / ( )."""
s = s.strip()
s = re.sub(r"'(.)'", lambda m: str(ord(m.group(1))), s)
def op(m):
a, n, c = self.sym[m.group(2).lower()]
return str({"offset": a, "lengthof": c, "type": n, "sizeof": n * c}[m.group(1).lower()])
s = re.sub(r"(?i)\b(offset|lengthof|type|sizeof)\s+(\w+)", op, s)
def tok(m):
t = m.group(0)
low = t.lower()
if re.fullmatch(r"[0-9][0-9a-f]*h", low):
return str(int(low[:-1], 16))
if re.fullmatch(r"[01]+b", low):
return str(int(low[:-1], 2))
if re.fullmatch(r"[0-9]+d?", low):
return str(int(low.rstrip("d")))
if low in self.const:
return str(self.const[low])
if low in self.label:
return str(self.label[low])
if t in self.label:
return str(self.label[t])
raise ValueError("모르는 이름: " + t)
s = re.sub(r"\w+", tok, s)
if not re.fullmatch(r"[0-9+\-*/() ]+", s):
raise ValueError("식을 못 읽음: " + s)
return int(eval(s.replace("/", "//")))
def operand(self, s):
"""('reg', 이름, 크기) / ('imm', 값, None) / ('mem', 주소, 크기 또는 None)"""
s = s.strip()
size = None
m = re.match(r"(?i)(byte|sbyte|word|sword|dword|sdword|near)\s+ptr\s+(.*)", s)
if m:
size = 4 if m.group(1).lower() == "near" else TYPES[m.group(1).lower()]
s = m.group(2).strip()
if self.regsize(s):
return ("reg", s.lower(), self.regsize(s))
m = re.fullmatch(r"(\w+)?\s*\[([^\]]+)\]", s)
if m:
addr = 0
if m.group(1):
a, n, _ = self.sym[m.group(1).lower()]
addr, size = a, size or n
for sign, term in re.findall(r"([+-]?)\s*([^+-]+)", m.group(2)):
term = term.strip()
v = self.get(term) if self.regsize(term) else self.num(term)
addr += -v if sign == "-" else v
return ("mem", addr & 0xFFFFFFFF, size)
m = re.fullmatch(r"(\w+)\s*(?:([+-])\s*(\w+))?", s)
if m and m.group(1).lower() in self.sym and not re.match(r"(?i)(offset|lengthof|type|sizeof)\b", s):
a, n, _ = self.sym[m.group(1).lower()]
if m.group(2):
d = self.num(m.group(3))
a += -d if m.group(2) == "-" else d
return ("mem", a, size or n)
return ("imm", self.num(s), None)
def val(self, o, n):
if o[0] == "reg":
return self.get(o[1])
if o[0] == "imm":
return o[1] & ((1 << (8 * n)) - 1)
return self.rd(o[1], n)
def store(self, o, n, v):
v &= (1 << (8 * n)) - 1
if o[0] == "reg":
self.put(o[1], v)
else:
self.wr(o[1], n, v)
# ---- 플래그 ----
def szp(self, v, n):
self.f["ZF"] = int(v == 0)
self.f["SF"] = (v >> (8 * n - 1)) & 1
self.f["PF"] = int(bin(v & 0xFF).count("1") % 2 == 0)
def add(self, a, b, n):
mask = (1 << (8 * n)) - 1
r = (a + b) & mask
top = 8 * n - 1
self.f["CF"] = int(a + b > mask)
self.f["OF"] = int((a >> top) == (b >> top) and (r >> top) != (a >> top))
self.szp(r, n)
return r
def sub(self, a, b, n):
mask = (1 << (8 * n)) - 1
r = (a - b) & mask
top = 8 * n - 1
self.f["CF"] = int(a < b)
self.f["OF"] = int((a >> top) != (b >> top) and (r >> top) != (a >> top))
self.szp(r, n)
return r
def logic(self, r, n):
self.f["CF"] = self.f["OF"] = 0
self.szp(r, n)
return r
# ---- 스택 ----
def push(self, v, n=4):
self.r["esp"] = (self.r["esp"] - n) & 0xFFFFFFFF
self.wr(self.r["esp"], n, v)
def pop(self, n=4):
v = self.rd(self.r["esp"], n)
self.r["esp"] = (self.r["esp"] + n) & 0xFFFFFFFF
return v
def eflags(self):
f = self.f
return f["CF"] | 2 | f["PF"] << 2 | f["ZF"] << 6 | f["SF"] << 7 | f["OF"] << 11
# ---- 코드 읽기 ----
def load(self, text, base=0):
cur = base
pending, uses = [], []
for raw in text.splitlines():
line = raw.split(";")[0].strip()
if not line or line.lower() in (".code", ".data"):
continue
m = re.match(r"([0-9A-Fa-f]{8})\s+(.*)", line)
if m:
cur, line = int(m.group(1), 16), m.group(2).strip()
m = re.fullmatch(r"(?i)(\w+)\s+PROC(?:\s+USES\s+(.*))?", line)
if m:
pending.append(m.group(1))
uses = m.group(2).split() if m.group(2) else []
for reg in uses: # USES가 만드는 push
cur = self._emit(cur, "push", [reg], "push " + reg, pending)
pending = []
continue
if re.fullmatch(r"(?i)\w+\s+ENDP", line):
uses = []
continue
m = re.match(r"(\w+)::?\s*(.*)", line)
if m and ":" in line.split()[0]:
pending.append(m.group(1))
line = m.group(2).strip()
if not line:
continue
parts = line.split(None, 1)
op = parts[0].lower()
args = [a.strip() for a in parts[1].split(",")] if len(parts) > 1 else []
if op == "ret":
for reg in reversed(uses): # USES가 만드는 pop
cur = self._emit(cur, "pop", [reg], "pop " + reg, pending)
pending = []
cur = self._emit(cur, op, args, line, pending)
pending = []
for name in pending: # 맨 끝의 레이블(next: 같은 것)
self.label[name] = self.label[name.lower()] = cur
self.end = cur
def _emit(self, cur, op, args, line, pending):
for name in pending:
self.label[name] = self.label[name.lower()] = cur
self.at[cur] = len(self.code)
self.code.append((cur, op, args, line))
return cur + (5 if op == "call" else 2)
def target(self, s):
if s in self.label:
return self.label[s]
if s.lower() in self.label:
return self.label[s.lower()]
o = self.operand(s)
return self.val(o, 4)
# ---- 실행 ----
def run(self, start=None, max_steps=10 ** 6):
eip = self.target(start) if isinstance(start, str) else (start if start is not None else self.code[0][0])
while eip in self.at:
self.steps += 1
if self.steps > max_steps:
raise RuntimeError("너무 오래 돈다")
i = self.at[eip]
addr, op, a, line = self.code[i]
nxt = self.code[i + 1][0] if i + 1 < len(self.code) else self.end
eip = self.step(op, a, nxt)
if self.trace is not None:
self.trace.append((line, dict(self.r), dict(self.f)))
if eip is None:
break
self.eip = eip
return self
def call_proc(self, name):
self.push(STOP)
return self.run(name)
def step(self, op, a, nxt):
f = self.f
if op in ("exit", "hlt"):
return None
if op == "nop":
return nxt
if op == "jmp":
return self.target(a[0])
if op in JCC:
return self.target(a[0]) if JCC[op](f) else nxt
if op == "jecxz":
return self.target(a[0]) if self.r["ecx"] == 0 else nxt
if op == "jcxz":
return self.target(a[0]) if self.get("cx") == 0 else nxt
if op in ("loop", "loope", "loopz", "loopne", "loopnz"):
self.r["ecx"] = (self.r["ecx"] - 1) & 0xFFFFFFFF # 플래그는 안 바뀐다
go = self.r["ecx"] != 0
if op in ("loope", "loopz"):
go = go and f["ZF"] == 1
if op in ("loopne", "loopnz"):
go = go and f["ZF"] == 0
return self.target(a[0]) if go else nxt
if op == "call":
if a[0] in self.hook:
self.calls.append(a[0])
self.hook[a[0]](self)
return nxt
t = self.target(a[0])
self.calls.append(a[0])
self.push(nxt)
return t
if op == "ret":
t = self.pop()
return None if t == STOP else t
if op == "push":
o = self.operand(a[0])
n = 2 if (o[0] == "reg" and o[2] == 2) or (o[0] == "mem" and o[2] == 2) else 4
self.push(self.val(o, n), n)
return nxt
if op == "pop":
o = self.operand(a[0])
n = 2 if o[2] == 2 else 4
self.store(o, n, self.pop(n))
return nxt
if op == "pushfd":
self.push(self.eflags())
return nxt
if op == "popfd":
v = self.pop()
f.update(CF=v & 1, PF=(v >> 2) & 1, ZF=(v >> 6) & 1, SF=(v >> 7) & 1, OF=(v >> 11) & 1)
return nxt
if op == "pushad":
old = self.r["esp"]
for k in ["eax", "ecx", "edx", "ebx"]:
self.push(self.r[k])
self.push(old)
for k in ["ebp", "esi", "edi"]:
self.push(self.r[k])
return nxt
if op == "popad":
for k in ["edi", "esi", "ebp"]:
self.r[k] = self.pop()
self.pop() # ESP 칸은 버린다
for k in ["ebx", "edx", "ecx", "eax"]:
self.r[k] = self.pop()
return nxt
# ---- 피연산자가 있는 명령 ----
d = self.operand(a[0])
s = self.operand(a[1]) if len(a) > 1 else None
n = d[2] or (s[2] if s else None)
if n is None:
raise ValueError("크기를 알 수 없다: " + " ".join(a))
if op in ("movzx", "movsx"):
sn = s[2]
v = self.val(s, sn)
self.store(d, d[2], signed(v, sn) if op == "movsx" else v)
return nxt
if s is not None and s[0] != "imm" and d[2] and s[2] and d[2] != s[2] and op not in ("shl", "sal", "shr", "sar"):
raise ValueError("크기가 다르다: " + " ".join(a))
x = self.val(d, n)
y = self.val(s, n) if s else None
top = 8 * n - 1
if op == "mov":
self.store(d, n, y)
elif op == "xchg":
self.store(d, n, y)
self.store(s, n, x)
elif op == "add":
self.store(d, n, self.add(x, y, n))
elif op == "sub":
self.store(d, n, self.sub(x, y, n))
elif op == "cmp":
self.sub(x, y, n) # 결과는 버린다
elif op == "inc":
c = f["CF"]
self.store(d, n, self.add(x, 1, n))
f["CF"] = c # INC는 CF를 안 건드린다
elif op == "dec":
c = f["CF"]
self.store(d, n, self.sub(x, 1, n))
f["CF"] = c
elif op == "neg":
self.store(d, n, self.sub(0, x, n))
elif op == "and":
self.store(d, n, self.logic(x & y, n))
elif op == "test":
self.logic(x & y, n) # 결과는 버린다
elif op == "or":
self.store(d, n, self.logic(x | y, n))
elif op == "xor":
self.store(d, n, self.logic(x ^ y, n))
elif op == "not":
self.store(d, n, ~x) # 플래그는 안 바뀐다
elif op in ("shl", "sal", "shr", "sar"):
c = self.val(s, 1) & 31
for _ in range(c):
if op in ("shl", "sal"):
f["CF"] = (x >> top) & 1
x = (x << 1) & ((1 << (8 * n)) - 1)
else:
f["CF"] = x & 1
x = (x >> 1) | ((x >> top) << top if op == "sar" else 0)
if c:
self.szp(x, n)
self.store(d, n, x)
else:
raise ValueError("모르는 명령: " + op)
return nxt
def machine(code, esp=0x1000, base=0):
c = CPU(esp)
c.load(code, base)
return c
FIRST_POS = """mov esi,OFFSET array
mov ecx,LENGTHOF array
next:
test WORD PTR [esi],8000h
pushfd
add esi,TYPE array
popfd
loopnz next
jnz quit
sub esi,TYPE array
quit:"""
FIRST_NONZERO = """mov esi,OFFSET array
mov ecx,LENGTHOF array
L1:cmp WORD PTR [esi],0
pushfd
add esi,TYPE array
popfd
loope L1
jz quit
sub esi,TYPE array
quit:"""
def scan(code, values, sentinel):
c = machine(code)
a = c.data("array", "SWORD", [v & 0xFFFF for v in values])
c.data("sentinel", "SWORD", [sentinel])
c.trace = []
c.run()
return c, (c.r["esi"] - a) // 2
def main():
# 1. LOOPNZ 예: 첫 양수는 10, 인덱스 4
vals = [-3, -6, -1, -10, 10, 30, 40, 4]
c, i = scan(FIRST_POS, vals, 0)
assert i == 4 and signed(c.rd(c.r["esi"], 2), 2) == 10 and c.r["ecx"] == 3
assert c.r["esp"] == 0x1000
assert len([1 for l, r, f in c.trace if l == "loopnz next"]) == 5
# 모두 음수면 ESI는 보초값(배열 바로 뒤)을 가리킨다
c, i = scan(FIRST_POS, [-1] * 8, 0)
assert i == 8 and c.rd(c.r["esi"], 2) == 0 and c.r["ecx"] == 0
# 마지막 원소가 첫 양수: ECX = 0으로 끝나도 LOOPNZ 직후 ZF = 1이라 찾은 것으로 가린다
c, i = scan(FIRST_POS, [-1] * 7 + [5], 0)
assert i == 7 and c.r["ecx"] == 0
assert [f["ZF"] for l, r, f in c.trace if l == "loopnz next"][-1] == 1
# 무작위 비교
rnd = random.Random(5)
for _ in range(300):
n = rnd.randint(1, 12)
vals = [rnd.choice([-5, -1, -32768, 0, 1, 7, 32767]) for _ in range(n)]
want = next((k for k, v in enumerate(vals) if v >= 0), n) # 부호 비트가 0인 첫 원소
assert scan(FIRST_POS, vals, 0)[1] == want
want = next((k for k, v in enumerate(vals) if v != 0), n)
assert scan(FIRST_NONZERO, vals, 0xFFFF)[1] == want
# 2. 연습: 50칸에서 0이 아닌 첫 값
for pos in (0, 1, 17, 49):
vals = [0] * 50
vals[pos] = 9
c, i = scan(FIRST_NONZERO, vals, 0xFFFF)
assert i == pos and c.r["ecx"] == 49 - pos
c, i = scan(FIRST_NONZERO, [0] * 50, 0xFFFF)
assert i == 50 and c.rd(c.r["esi"], 2) == 0xFFFF # 보초값
# 3. PUSHFD/POPFD를 빼면: ADD가 ZF를 0으로 바꿔 LOOPNZ가 끝까지 돈다
broken = FIRST_POS.replace("pushfd\n", "").replace("popfd\n", "")
c, i = scan(broken, [-3, -6, -1, -10, 10, 30, 40, 4], 0)
assert i == 8
# LOOP 계열은 플래그를 바꾸지 않는다. ECX만 줄인다
c = machine("mov ecx,2\nmov al,1\ncmp al,1\nL: loope L").run()
assert c.f["ZF"] == 1 and c.r["ecx"] == 0
# 4. 동작 규칙표: (ECX가 줄어든 뒤 0인가, ZF) -> 점프하나
def go(op, ecx, zf):
c = machine("%s T\nmov bl,1\nT:" % op)
c.r["ecx"] = ecx
c.f["ZF"] = zf
c.run()
return c.get("bl") == 0
for op in ("loope", "loopz"):
assert [go(op, 2, 1), go(op, 2, 0), go(op, 1, 1), go(op, 1, 0)] == [True, False, False, False]
for op in ("loopne", "loopnz"):
assert [go(op, 2, 1), go(op, 2, 0), go(op, 1, 1), go(op, 1, 0)] == [False, True, False, False]
assert [go("loop", 2, 1), go("loop", 2, 0), go("loop", 1, 1)] == [True, True, False]
# 5. 카드 C2: 3, 3, 7, 3에서 3이 아닌 첫 값
code = """mov esi,OFFSET array
mov ecx,LENGTHOF array
L1:cmp WORD PTR [esi],3
pushfd
add esi,TYPE array
popfd
loope L1
jz quit
sub esi,TYPE array
quit:"""
c, i = scan(code, [3, 3, 7, 3], 0)
assert i == 2 and c.r["ecx"] == 1 and c.rd(c.r["esi"], 2) == 7
if __name__ == "__main__":
main()
print("ALL CHECKS PASSED")