제로 확장과 부호 확장 문서의 검증 코드다. if __name__ == "__main__": 아래가 자체 테스트다.
"""「제로 확장과 부호 확장」: MOVZX·MOVSX 예시 값, 8비트 전수 검사, 카드 C1·C3
실행: python3 17_zero-sign-extension_verify.py (표준 라이브러리만 쓴다)
MASM으로 돌린 결과가 아니다. 아래의 작은 x86 흉내기(8·16·32비트 레지스터, ZF·SF·CF·OF,
리틀 엔디언 메모리)로 문서에 적힌 값을 확인한다.
주소와 배열 번호는 0부터 센다.
"""
# ---- x86 흉내기 (이 파일 안에서만 쓴다. 다른 파일을 import하지 않는다) ----
MASK = {1: 0xFF, 2: 0xFFFF, 4: 0xFFFFFFFF}
R32 = ["eax", "ebx", "ecx", "edx", "esi", "edi", "ebp", "esp"]
PART = {}
for _r in "abcd":
PART[_r + "l"] = ("e" + _r + "x", 0, 1)
PART[_r + "h"] = ("e" + _r + "x", 8, 1)
PART[_r + "x"] = ("e" + _r + "x", 0, 2)
for _r in ("si", "di", "bp", "sp"):
PART[_r] = ("e" + _r, 0, 2)
for _r in R32:
PART[_r] = (_r, 0, 4)
SREG = {"cs", "ds", "ss", "es", "fs", "gs"}
class AsmError(Exception):
"""어셈블러가 받아 주지 않는 문장."""
def signed(v, size):
"""size바이트 값 v를 부호 있는 수로 읽는다."""
bits = 8 * size
v &= MASK[size]
return v - (1 << bits) if v >> (bits - 1) else v
class CPU:
def __init__(self):
self.r = dict.fromkeys(R32, 0)
self.mem = {}
self.ZF = self.SF = self.CF = self.OF = 0
# --- 메모리: 리틀 엔디언 ---
def rd(self, addr, size):
return sum(self.mem.get(addr + i, 0) << (8 * i) for i in range(size))
def wr(self, addr, size, v):
for i in range(size):
self.mem[addr + i] = (v >> (8 * i)) & 0xFF
def dump(self, addr, n):
return [self.mem.get(addr + i, 0) for i in range(n)]
# --- 피연산자: "al" 같은 레지스터 이름, 정수(즉시값), ("m", 주소, 크기) ---
def size(self, op):
if isinstance(op, str):
if op in SREG or op in ("ip",):
return 2
if op == "eip":
return 4
return PART[op][2]
if isinstance(op, tuple):
return op[2]
return None
def get(self, op):
if isinstance(op, int):
return op
if isinstance(op, tuple):
return self.rd(op[1], op[2])
full, shift, size = PART[op]
return (self.r[full] >> shift) & MASK[size]
def set(self, op, v):
if isinstance(op, tuple):
self.wr(op[1], op[2], v & MASK[op[2]])
return
full, shift, size = PART[op]
keep = self.r[full] & ~(MASK[size] << shift) & 0xFFFFFFFF
self.r[full] = keep | ((v & MASK[size]) << shift)
def _check(self, d, s):
if isinstance(d, int):
raise AsmError("immediate value cannot be destination")
if d in ("eip", "ip", "cs"):
raise AsmError("EIP/IP/CS cannot be the destination")
if isinstance(d, tuple) and isinstance(s, tuple):
raise AsmError("memory-to-memory not permitted")
if isinstance(d, str) and d in SREG and isinstance(s, int):
raise AsmError("immediate move to segment register not permitted")
ds, ss = self.size(d), self.size(s)
if ds is None:
raise AsmError("ambiguous operand size")
if ss is not None and ds != ss:
raise AsmError("size mismatch")
if ss is None and not -(1 << (8 * ds - 1)) <= s <= MASK[ds]:
raise AsmError("immediate too large")
return ds
def flags(self):
return (self.ZF, self.SF, self.CF, self.OF)
def _zs(self, r, n):
self.ZF = int(r == 0)
self.SF = (r >> (8 * n - 1)) & 1
# --- 명령 ---
def mov(self, d, s):
self._check(d, s)
if isinstance(d, str) and d in SREG:
return
self.set(d, self.get(s)) # 플래그는 건드리지 않는다
def xchg(self, a, b):
if isinstance(a, int) or isinstance(b, int):
raise AsmError("no immediate operands")
self._check(a, b)
va, vb = self.get(a), self.get(b)
self.set(a, vb)
self.set(b, va)
def movzx(self, d, s):
if not isinstance(d, str):
raise AsmError("destination must be a register")
if self.size(s) >= self.size(d):
raise AsmError("source must be smaller")
self.set(d, self.get(s))
def movsx(self, d, s):
if not isinstance(d, str):
raise AsmError("destination must be a register")
if self.size(s) >= self.size(d):
raise AsmError("source must be smaller")
self.set(d, signed(self.get(s), self.size(s)))
def _add(self, a, b, n, keep_cf=False):
m, top = MASK[n], 8 * n - 1
full = a + b
r = full & m
if not keep_cf:
self.CF = (full >> (top + 1)) & 1
self.OF = (((a ^ r) & (b ^ r)) >> top) & 1
self._zs(r, n)
return r
def _sub(self, a, b, n, keep_cf=False):
m, top = MASK[n], 8 * n - 1
r = (a - b) & m
if not keep_cf:
self.CF = int(a < b)
self.OF = (((a ^ b) & (a ^ r)) >> top) & 1
self._zs(r, n)
return r
def add(self, d, s):
n = self._check(d, s)
self.set(d, self._add(self.get(d), self.get(s) & MASK[n], n))
def sub(self, d, s):
n = self._check(d, s)
self.set(d, self._sub(self.get(d), self.get(s) & MASK[n], n))
def inc(self, d):
n = self.size(d)
if n is None:
raise AsmError("ambiguous operand size")
self.set(d, self._add(self.get(d), 1, n, keep_cf=True)) # CF는 그대로
def dec(self, d):
n = self.size(d)
if n is None:
raise AsmError("ambiguous operand size")
self.set(d, self._sub(self.get(d), 1, n, keep_cf=True)) # CF는 그대로
def neg(self, d):
n = self.size(d)
self.set(d, self._sub(0, self.get(d), n)) # 0 - 값
def dup(count, *items):
return list(items) * count
class Data:
"""데이터 세그먼트. 정의한 순서대로 바이트를 채우고 이름마다 주소·TYPE·LENGTHOF·SIZEOF를 적어 둔다."""
def __init__(self, cpu, base=0x00404000):
self.cpu, self.base, self.loc, self.sym = cpu, base, base, {}
def _flat(self, items):
for it in items:
if isinstance(it, list):
yield from self._flat(it)
else:
yield it
def define(self, name, tsize, *items):
start, count = self.loc, 0
for it in self._flat(items):
if isinstance(it, str) and tsize == 1:
for ch in it:
self.cpu.wr(self.loc, 1, ord(ch))
self.loc += 1
count += 1
else:
self.cpu.wr(self.loc, tsize, (it or 0) & MASK.get(tsize, (1 << 8 * tsize) - 1))
self.loc += tsize
count += 1
if name:
self.sym[name] = {"addr": start, "type": tsize, "length": count, "size": count * tsize}
def byte(self, name, *items):
self.define(name, 1, *items)
def word(self, name, *items):
self.define(name, 2, *items)
def dword(self, name, *items):
self.define(name, 4, *items)
def label(self, name, tsize):
self.sym[name] = {"addr": self.loc, "type": tsize, "length": 1, "size": tsize}
def offset(self, name):
return self.sym[name]["addr"]
def m(self, name, off=0, size=None):
"""이름+오프셋 메모리 피연산자. size를 주면 PTR로 크기를 바꾼 것이다."""
return ("m", self.sym[name]["addr"] + off, size or self.sym[name]["type"])
def bytes_of(self, name):
return self.cpu.dump(self.sym[name]["addr"], self.sym[name]["size"])
def must_fail(fn, *args):
try:
fn(*args)
except AsmError as e:
return str(e)
raise AssertionError("오류가 나야 하는데 통과했다")
# ---- 흉내기 끝 ----
if __name__ == "__main__":
c = CPU()
# movzx·movsx ax,bl (BL = 8Fh)
c.mov("bl", 0b10001111); c.movzx("ax", "bl"); assert c.get("ax") == 0b0000000010001111 == 0x008F
c.mov("bl", 0b10001111); c.movsx("ax", "bl"); assert c.get("ax") == 0b1111111110001111 == 0xFF8F
assert signed(0x8F, 1) == -113 and signed(0xFF8F, 2) == -113 and 0x8F == 143
# 크기가 다르면 MOV는 안 된다
assert must_fail(c.mov, "ax", "bl") == "size mismatch"
# 목적지는 레지스터여야 한다
assert must_fail(c.movzx, ("m", 0x100, 2), "bl") == "destination must be a register"
assert must_fail(c.movsx, ("m", 0x100, 2), "bl") == "destination must be a register"
# 전수: 8비트 -> 16비트, 8비트 -> 32비트, 16비트 -> 32비트(표본)
for v in range(256):
c.mov("bl", v)
c.mov("eax", 0xFFFFFFFF); c.movzx("ax", "bl"); assert c.get("ax") == v and c.get("eax") >> 16 == 0xFFFF
c.movzx("eax", "bl"); assert c.get("eax") == v
c.movsx("ax", "bl"); assert signed(c.get("ax"), 2) == signed(v, 1)
c.movsx("eax", "bl"); assert signed(c.get("eax"), 4) == signed(v, 1)
# 윗부분은 모두 부호 비트의 복사
assert c.get("eax") >> 8 == (0xFFFFFF if v >> 7 else 0)
# 양수(부호 비트 0)에서는 둘이 같다
c.movzx("ecx", "bl")
assert (c.get("ecx") == c.get("eax")) == (v < 128)
for v in range(0, 0x10000, 3):
c.mov("bx", v); c.movzx("eax", "bx"); assert c.get("eax") == v
c.movsx("eax", "bx"); assert signed(c.get("eax"), 4) == signed(v, 2)
# 손으로 하는 제로 확장
c.mov("bl", 0x8F); c.mov("eax", 0xDEADBEEF); c.mov("eax", 0); c.mov("al", "bl"); assert c.get("eax") == 0x8F
# 값이 그대로인 까닭: -2^7 = -2^15 + (2^14 + ... + 2^7)
assert -2**7 == -2**15 + sum(2**k for k in range(7, 15))
# 카드 C1
c.mov("bl", 0x7F); c.movzx("ax", "bl"); a = c.get("ax"); c.movsx("ax", "bl"); assert a == c.get("ax") == 0x007F
c.mov("bl", 0xF0); c.movzx("ax", "bl"); assert c.get("ax") == 0x00F0
c.movsx("ax", "bl"); assert c.get("ax") == 0xFFF0
# 카드 C3: 나이 200을 MOVSX로 넓히면 -56, 기온 -5를 MOVZX로 넓히면 251
c.mov("bl", 200); c.movsx("eax", "bl"); assert signed(c.get("eax"), 4) == -56
c.mov("bl", -5 & 0xFF); c.movzx("eax", "bl"); assert c.get("eax") == 251
c.movsx("eax", "bl"); assert signed(c.get("eax"), 4) == -5 and c.get("eax") == 0xFFFFFFFB
print("ALL CHECKS PASSED")