sup39
709f283e53
codehandler.bin: - Use Nintendont's codehandler - Clearing cache for 04 code type kernel.py: - Fix codelength - Fix code optimization (TODO 28, E2) dolreader.py: - Fix verbose 3 logging bug
398 lines
15 KiB
Python
398 lines
15 KiB
Python
from io import BytesIO
|
|
|
|
import tools
|
|
from fileutils import align_byte_size, read_uint32, write_uint32
|
|
|
|
class UnmappedAddressError(Exception): pass
|
|
class SectionCountFullError(Exception): pass
|
|
class AddressOutOfRangeError(Exception): pass
|
|
|
|
class DolFile(object):
|
|
|
|
class SectionType:
|
|
Text = 0
|
|
Data = 1
|
|
|
|
maxTextSections = 7
|
|
maxDataSections = 11
|
|
offsetInfoLoc = 0
|
|
addressInfoLoc = 0x48
|
|
sizeInfoLoc = 0x90
|
|
bssInfoLoc = 0xD8
|
|
entryInfoLoc = 0xE0
|
|
|
|
def __init__(self, f=None):
|
|
|
|
self.textSections = []
|
|
self.dataSections = []
|
|
|
|
self.bssAddress = 0
|
|
self.bssSize = 0
|
|
self.entryPoint = 0x80003000
|
|
|
|
if f is None: return
|
|
|
|
# Read text and data section addresses and sizes
|
|
for i in range(DolFile.maxTextSections + DolFile.maxDataSections):
|
|
f.seek(DolFile.offsetInfoLoc + (i << 2))
|
|
offset = read_uint32(f)
|
|
f.seek(DolFile.addressInfoLoc + (i << 2))
|
|
address = read_uint32(f)
|
|
f.seek(DolFile.sizeInfoLoc + (i << 2))
|
|
size = read_uint32(f)
|
|
|
|
if offset >= 0x100:
|
|
f.seek(offset)
|
|
data = BytesIO(f.read(size))
|
|
if i < DolFile.maxTextSections:
|
|
self.textSections.append({"offset": offset, "address": address, "size": size, "data": data, "type": DolFile.SectionType.Text})
|
|
else:
|
|
self.dataSections.append({"offset": offset, "address": address, "size": size, "data": data, "type": DolFile.SectionType.Data})
|
|
|
|
f.seek(DolFile.bssInfoLoc)
|
|
self.bssAddress = read_uint32(f)
|
|
self.bssSize = read_uint32(f)
|
|
|
|
f.seek(DolFile.entryInfoLoc)
|
|
self.entryPoint = read_uint32(f)
|
|
|
|
self._currLogicAddr = self.first_section["address"]
|
|
self.seek(self._currLogicAddr)
|
|
f.seek(0)
|
|
|
|
def __repr__(self) -> str:
|
|
return f"repr={vars(self)}"
|
|
|
|
def __str__(self) -> str:
|
|
return f"Nintendo DOL executable {self.__repr__()}"
|
|
|
|
def resolve_address(self, gcAddr: int) -> tuple:
|
|
""" Returns the data of the section that houses the given address\n
|
|
UnmappedAddressError is raised when the address is unmapped """
|
|
|
|
for section in self.sections:
|
|
if section["address"] <= gcAddr < (section["address"] + section["size"]):
|
|
return section
|
|
|
|
raise UnmappedAddressError(f"Unmapped address: 0x{gcAddr:X}")
|
|
|
|
def seek_nearest_unmapped(self, gcAddr: int, buffer=0) -> int:
|
|
'''Returns the nearest unmapped address (greater) if the given address is already taken by data'''
|
|
|
|
for section in self.sections:
|
|
if section["address"] > (gcAddr + buffer) or (section["address"] + section["size"]) < gcAddr:
|
|
continue
|
|
gcAddr = section["address"] + section["size"]
|
|
|
|
try:
|
|
self.resolve_address(gcAddr)
|
|
except UnmappedAddressError:
|
|
break
|
|
return gcAddr
|
|
|
|
@property
|
|
def sections(self) -> tuple:
|
|
""" Generator that yields each section's data """
|
|
|
|
for i in self.textSections:
|
|
yield i
|
|
for i in self.dataSections:
|
|
yield i
|
|
|
|
@property
|
|
def last_section(self) -> tuple:
|
|
""" Returns the last section in the dol file as sorted by internal offset """
|
|
|
|
largestOffset = 0
|
|
indexToTarget = 0
|
|
targetType = DolFile.SectionType.Text
|
|
|
|
for i, section in enumerate(self.textSections):
|
|
if section["offset"] > largestOffset:
|
|
largestOffset = section["offset"]
|
|
indexToTarget = i
|
|
targetType = DolFile.SectionType.Text
|
|
for i, section in enumerate(self.dataSections):
|
|
if section["offset"] > largestOffset:
|
|
largestOffset = section["offset"]
|
|
indexToTarget = i
|
|
targetType = DolFile.SectionType.Data
|
|
|
|
if targetType == DolFile.SectionType.Text:
|
|
return self.textSections[indexToTarget]
|
|
else:
|
|
return self.dataSections[indexToTarget]
|
|
|
|
@property
|
|
def first_section(self) -> tuple:
|
|
""" Returns the first section in the dol file as sorted by internal offset """
|
|
|
|
smallestOffset = 0xFFFFFFFF
|
|
indexToTarget = 0
|
|
targetType = DolFile.SectionType.Text
|
|
|
|
for i, section in enumerate(self.textSections):
|
|
if section["offset"] < smallestOffset:
|
|
smallestOffset = section["offset"]
|
|
indexToTarget = i
|
|
targetType = DolFile.SectionType.Text
|
|
for i, section in enumerate(self.dataSections):
|
|
if section["offset"] < smallestOffset:
|
|
smallestOffset = section["offset"]
|
|
indexToTarget = i
|
|
targetType = DolFile.SectionType.Data
|
|
|
|
if targetType == DolFile.SectionType.Text:
|
|
return self.textSections[indexToTarget]
|
|
else:
|
|
return self.dataSections[indexToTarget]
|
|
|
|
# Unsupported: Reading an entire dol file
|
|
# Assumption: A read should not go beyond the current section
|
|
def read(self, _size: int) -> bytes:
|
|
section = self.resolve_address(self._currLogicAddr)
|
|
if self._currLogicAddr + _size > (section["address"] + section["size"]):
|
|
raise UnmappedAddressError("Read goes over current section")
|
|
|
|
self._currLogicAddr += _size
|
|
return section["data"].read(_size)
|
|
|
|
# Assumption: A write should not go beyond the current section
|
|
def write(self, _data: bytes):
|
|
section = self.resolve_address(self._currLogicAddr)
|
|
if self._currLogicAddr + len(_data) > (section["address"] + section["size"]):
|
|
raise UnmappedAddressError("Write goes over current section")
|
|
|
|
section["data"].write(_data)
|
|
self._currLogicAddr += len(_data)
|
|
|
|
def seek(self, where: int, whence: int = 0):
|
|
if whence == 0:
|
|
section = self.resolve_address(where)
|
|
section["data"].seek(where - section["address"])
|
|
|
|
self._currLogicAddr = where
|
|
elif whence == 1:
|
|
section = self.resolve_address(self._currLogicAddr + where)
|
|
section["data"].seek((self._currLogicAddr + where) - section["address"])
|
|
|
|
self._currLogicAddr += where
|
|
else:
|
|
raise NotImplementedError(f"Unsupported whence type '{whence}'")
|
|
|
|
def tell(self) -> int:
|
|
return self._currLogicAddr
|
|
|
|
def save(self, f):
|
|
f.seek(0)
|
|
f.write(b"\x00" * self.size)
|
|
|
|
for i, section in enumerate(self.sections):
|
|
if section["type"] == DolFile.SectionType.Data:
|
|
entry = i + (DolFile.maxTextSections - len(self.textSections))
|
|
else:
|
|
entry = i
|
|
|
|
f.seek(DolFile.offsetInfoLoc + (entry << 2))
|
|
write_uint32(f, section["offset"]) #offset in file
|
|
f.seek(DolFile.addressInfoLoc + (entry << 2))
|
|
write_uint32(f, section["address"]) #game address
|
|
f.seek(DolFile.sizeInfoLoc + (entry << 2))
|
|
write_uint32(f, section["size"]) #size in file
|
|
|
|
f.seek(section["offset"])
|
|
f.write(section["data"].getbuffer())
|
|
|
|
f.seek(DolFile.bssInfoLoc)
|
|
write_uint32(f, self.bssAddress)
|
|
write_uint32(f, self.bssSize)
|
|
|
|
f.seek(DolFile.entryInfoLoc)
|
|
write_uint32(f, self.entryPoint)
|
|
align_byte_size(f, 256)
|
|
|
|
@property
|
|
def size(self) -> int:
|
|
try:
|
|
section = self.last_section
|
|
return (section["offset"] + section["size"] + 255) & -256
|
|
except IndexError:
|
|
return 0x100
|
|
|
|
def get_section_size(self, index: int, section: SectionType) -> int:
|
|
""" Return the current size of the specified section\n
|
|
section: DolFile.SectionType """
|
|
|
|
if section == DolFile.SectionType.Text:
|
|
return self.textSections[index]["size"]
|
|
else:
|
|
return self.dataSections[index]["size"]
|
|
|
|
|
|
def append_text_sections(self, sectionsList: list):
|
|
""" Follows the list format: [tuple(<Bytes>Data, <Int>GameAddress or None), tuple(<Bytes>Data... """
|
|
|
|
for i, dataSet in enumerate(sectionsList):
|
|
if len(self.textSections) >= DolFile.maxTextSections:
|
|
raise SectionCountFullError(f"Exceeded max text section limit of {DolFile.maxTextSections}")
|
|
|
|
finalSection = self.last_section
|
|
lastSection = self.textSections[len(self.textSections) - 1]
|
|
data, address = dataSet
|
|
|
|
if not hasattr(data, "getbuffer"):
|
|
if hasattr(data, "read"):
|
|
data.seek(0)
|
|
data = BytesIO(data.read())
|
|
else:
|
|
data = BytesIO(data)
|
|
|
|
offset = finalSection["offset"] + finalSection["size"]
|
|
|
|
if i < len(sectionsList) - 1:
|
|
size = (len(data.getbuffer()) + 31) & -32
|
|
else:
|
|
size = (len(data.getbuffer()) + 255) & -256
|
|
|
|
if address is None:
|
|
address = self.seek_nearest_unmapped(lastSection["address"] + lastSection["size"], size)
|
|
|
|
if address < 0x80000000 or address >= 0x81200000:
|
|
raise AddressOutOfRangeError(f"Address '{address:08X}' of text section {i} is beyond scope (0x80000000 <-> 0x81200000)")
|
|
|
|
self.textSections.append({"offset": offset, "address": address, "size": size, "data": data, "type": DolFile.SectionType.Text})
|
|
|
|
def append_data_sections(self, sectionsList: list):
|
|
""" Follows the list format: [tuple(<Bytes>Data, <Int>GameAddress or None), tuple(<Bytes>Data... """
|
|
|
|
for i, dataSet in enumerate(sectionsList):
|
|
if len(self.dataSections) >= DolFile.maxDataSections:
|
|
raise SectionCountFullError(f"Exceeded max data section limit of {DolFile.maxDataSections}")
|
|
|
|
finalSection = self.last_section
|
|
lastSection = self.dataSections[len(self.dataSections) - 1]
|
|
data, address = dataSet
|
|
|
|
if not hasattr(data, "getbuffer"):
|
|
if hasattr(data, "read"):
|
|
data.seek(0)
|
|
data = BytesIO(data.read())
|
|
else:
|
|
data = BytesIO(data)
|
|
|
|
offset = finalSection["offset"] + finalSection["size"]
|
|
|
|
if i < len(sectionsList) - 1:
|
|
size = (len(data.getbuffer()) + 31) & -32
|
|
else:
|
|
size = (len(data.getbuffer()) + 255) & -256
|
|
|
|
if address is None:
|
|
address = self.seek_nearest_unmapped(lastSection["address"] + lastSection["size"], size)
|
|
|
|
if address < 0x80000000 or address >= 0x81200000:
|
|
raise AddressOutOfRangeError(f"Address '{address:08X}' of data section {i} is beyond scope (0x80000000 <-> 0x81200000)")
|
|
|
|
self.dataSections.append({"offset": offset, "address": address, "size": size, "data": data, "type": DolFile.SectionType.Data})
|
|
|
|
def insert_branch(self, to: int, _from: int, lk=0):
|
|
""" Insert a branch instruction at _from\n
|
|
to: address to branch to\n
|
|
_from: address to branch from\n
|
|
lk: 0 | 1, is branch linking? """
|
|
|
|
_from &= 0xFFFFFFFC
|
|
to &= 0xFFFFFFFC
|
|
self.seek(_from)
|
|
write_uint32(self, (to - _from) & 0x3FFFFFD | 0x48000000 | lk)
|
|
|
|
def extract_branch_addr(self, bAddr: int) -> tuple:
|
|
""" Returns the destination of the given branch,
|
|
and if the branch is conditional """
|
|
|
|
self.seek(bAddr)
|
|
|
|
ppc = read_uint32(self)
|
|
conditional = (ppc >> 24) & 0xFF < 0x48
|
|
|
|
if conditional is True:
|
|
if (ppc & 0x8000):
|
|
offset = (ppc & 0xFFFD) - 0x10000
|
|
else:
|
|
offset = ppc & 0xFFFD
|
|
else:
|
|
if (ppc & 0x2000000):
|
|
offset = (ppc & 0x3FFFFFD) - 0x4000000
|
|
else:
|
|
offset = ppc & 0x3FFFFFD
|
|
|
|
return (bAddr + offset, conditional)
|
|
|
|
def read_string(self, addr: int = None, maxlen: int = 0, encoding: str = "utf-8") -> str:
|
|
""" Reads a null terminated string from the specified address """
|
|
|
|
if addr != None:
|
|
self.seek(addr)
|
|
|
|
length = 0
|
|
string = ""
|
|
while (char := self.read(1)) != b"\x00":
|
|
try:
|
|
string += char.decode(encoding)
|
|
length += 1
|
|
except UnicodeDecodeError:
|
|
print(f"{char} at pos {length}, (address 0x{addr + length:08X}) is not a valid utf-8 character")
|
|
return ""
|
|
if length > (maxlen-1) and maxlen != 0:
|
|
return string
|
|
|
|
return string
|
|
|
|
def print_info(self):
|
|
print("")
|
|
print("|-- DOL INFO --|".center(20, " "))
|
|
print("")
|
|
|
|
attrs = ("offset", "address", "size")
|
|
for i, (offset, addr, size) in enumerate((sec[attr] for attr in attrs) for sec in self.textSections):
|
|
header = f"| Text section {i} |"
|
|
info = [ "-"*len(header) + "\n" + header + "\n" + "-"*len(header),
|
|
"File Offset:".ljust(16, " ") + f"0x{offset:X}",
|
|
"Virtual addr:".ljust(16, " ") + f"0x{addr:X}",
|
|
"Size:".ljust(16, " ") + f"0x{size:X}" ]
|
|
|
|
print("\n".join(info) + "\n")
|
|
|
|
for i, (offset, addr, size) in enumerate((sec[attr] for attr in attrs) for sec in self.dataSections):
|
|
header = f"| Data section {i} |"
|
|
info = [ "-"*len(header) + "\n" + header + "\n" + "-"*len(header),
|
|
"File Offset:".ljust(16, " ") + f"0x{offset:X}",
|
|
"Virtual addr:".ljust(16, " ") + f"0x{addr:X}",
|
|
"Size:".ljust(16, " ") + f"0x{size:X}" ]
|
|
|
|
print("\n".join(info) + "\n")
|
|
|
|
header = "| BSS section |"
|
|
info = [ "-"*len(header) + "\n" + header + "\n" + "-"*len(header),
|
|
"Virtual addr:".ljust(16, " ") + f"0x{self.bssAddress:X}",
|
|
"Size:".ljust(16, " ") + f"0x{self.bssSize:X}",
|
|
"End:".ljust(16, " ") + f"0x{self.bssAddress+self.bssSize:X}" ]
|
|
|
|
print("\n".join(info) + "\n")
|
|
|
|
header = "| Miscellaneous Info |"
|
|
info = [ "-"*len(header) + "\n" + header + "\n" + "-"*len(header),
|
|
"Text sections:".ljust(16, " ") + f"0x{len(self.textSections):X}",
|
|
"Data sections:".ljust(16, " ") + f"0x{len(self.dataSections):X}",
|
|
"File length:".ljust(16, " ") + f"0x{self.size:X}" ]
|
|
|
|
print("\n".join(info) + "\n")
|
|
|
|
if __name__ == "__main__":
|
|
# Example usage (Reading global string "mario" from Super Mario Sunshine (NTSC-U))
|
|
|
|
with open("Start.dol", "rb") as f:
|
|
dol = DolFile(f)
|
|
|
|
name = dol.read_string(addr=0x804165A0)
|
|
print(name)
|