esp-idf/tools/mkuf2.py
2022-08-12 15:36:11 +02:00

198 lines
7.4 KiB
Python
Executable File

#!/usr/bin/env python
#
# SPDX-FileCopyrightText: 2020-2022 Espressif Systems (Shanghai) CO LTD
# SPDX-License-Identifier: Apache-2.0
from __future__ import division
import argparse
import hashlib
import json
import os
import struct
from functools import partial
from typing import Dict, List
from future.utils import iteritems
def round_up_int_div(n: int, d: int) -> int:
# equivalent to math.ceil(n / d)
return (n + d - 1) // d
class UF2Writer(object):
# The UF2 format is described here: https://github.com/microsoft/uf2
UF2_BLOCK_SIZE = 512
UF2_DATA_SIZE = 476 # max value of CHUNK_SIZE reduced by optional parts. Currently, MD5_PART only.
UF2_MD5_PART_SIZE = 24
UF2_FIRST_MAGIC = 0x0A324655
UF2_SECOND_MAGIC = 0x9E5D5157
UF2_FINAL_MAGIC = 0x0AB16F30
UF2_FLAG_FAMILYID_PRESENT = 0x00002000
UF2_FLAG_MD5_PRESENT = 0x00004000
def __init__(self, chip_id: int, output_file: os.PathLike, chunk_size: int) -> None:
self.chip_id = chip_id
self.CHUNK_SIZE = self.UF2_DATA_SIZE - self.UF2_MD5_PART_SIZE if chunk_size is None else chunk_size
self.f = open(output_file, 'wb')
def __enter__(self) -> 'UF2Writer':
return self
def __exit__(self, exc_type: str, exc_val: int, exc_tb: List) -> None:
if self.f:
self.f.close()
@staticmethod
def _to_uint32(num: int) -> bytes:
return struct.pack('<I', num)
def _write_block(self, addr: int, chunk: bytes, len_chunk: int, block_no: int, blocks: int) -> None:
assert len_chunk > 0
assert len_chunk <= self.CHUNK_SIZE
assert block_no < blocks
block = self._to_uint32(self.UF2_FIRST_MAGIC)
block += self._to_uint32(self.UF2_SECOND_MAGIC)
block += self._to_uint32(self.UF2_FLAG_FAMILYID_PRESENT | self.UF2_FLAG_MD5_PRESENT)
block += self._to_uint32(addr)
block += self._to_uint32(len_chunk)
block += self._to_uint32(block_no)
block += self._to_uint32(blocks)
block += self._to_uint32(self.chip_id)
block += chunk
md5_part = self._to_uint32(addr)
md5_part += self._to_uint32(len_chunk)
md5_part += hashlib.md5(chunk).digest()
assert len(md5_part) == self.UF2_MD5_PART_SIZE
block += md5_part
block += b'\x00' * (self.UF2_DATA_SIZE - self.UF2_MD5_PART_SIZE - len_chunk)
block += self._to_uint32(self.UF2_FINAL_MAGIC)
assert len(block) == self.UF2_BLOCK_SIZE
self.f.write(block)
def add_file(self, addr: int, f_path: os.PathLike) -> None:
blocks = round_up_int_div(os.path.getsize(f_path), self.CHUNK_SIZE)
with open(f_path, 'rb') as fin:
a = addr
for i, chunk in enumerate(iter(partial(fin.read, self.CHUNK_SIZE), b'')):
len_chunk = len(chunk)
self._write_block(a, chunk, len_chunk, i, blocks)
a += len_chunk
def action_write(args: Dict) -> None:
with UF2Writer(args['chip_id'], args['output_file'], args['chunk_size']) as writer:
for addr, f in args['files']:
print('Adding {} at {:#x}'.format(f, addr))
writer.add_file(addr, f)
print('"{}" has been written.'.format(args['output_file']))
def main() -> None:
parser = argparse.ArgumentParser()
def four_byte_aligned(integer: int) -> bool:
return integer & 3 == 0
def parse_chunk_size(string: str) -> int:
num = int(string, 0)
if not four_byte_aligned(num):
raise argparse.ArgumentTypeError('Chunk size should be a 4-byte aligned number')
return num
def parse_chip_id(string: str) -> int:
num = int(string, 16)
if num < 0 or num > 0xFFFFFFFF:
raise argparse.ArgumentTypeError('Chip ID should be a 4-byte unsigned integer')
return num
# Provision to add "info" command
subparsers = parser.add_subparsers(dest='command')
write_parser = subparsers.add_parser('write')
write_parser.add_argument('-o', '--output-file',
help='Filename for storing the output UF2 image',
required=True)
write_parser.add_argument('--chip-id',
required=True,
type=parse_chip_id,
help='Hexa-decimal chip identificator')
write_parser.add_argument('--chunk-size',
required=False,
type=parse_chunk_size,
default=None,
help='Specify the used data part of the 512 byte UF2 block. A common value is 256. By '
'default the largest possible value will be used.')
write_parser.add_argument('--json',
help='Optional file for loading "flash_files" dictionary with <address> <file> items')
write_parser.add_argument('--bin',
help='Use only a subset of binaries from the JSON file, e.g. "partition_table '
'bootloader app"',
nargs='*')
write_parser.add_argument('files',
metavar='<address> <file>', help='Add <file> at <address>',
nargs='*')
args = parser.parse_args()
def check_file(file_name: str) -> str:
if not os.path.isfile(file_name):
raise RuntimeError('{} is not a regular file!'.format(file_name))
return file_name
def parse_addr(string: str) -> int:
num = int(string, 0)
if not four_byte_aligned(num):
raise RuntimeError('{} is not a 4-byte aligned valid address'.format(string))
return num
files = []
if args.files:
files += [(parse_addr(addr), check_file(f_name)) for addr, f_name in zip(args.files[::2], args.files[1::2])]
if args.json:
json_dir = os.path.dirname(os.path.abspath(args.json))
def process_json_file(path: str) -> str:
'''
The input path is relative to json_dir. This function makes it relative to the current working
directory.
'''
return check_file(os.path.relpath(os.path.join(json_dir, path), start=os.curdir))
with open(args.json) as f:
json_content = json.load(f)
if args.bin:
try:
bin_selection = [json_content[b] for b in args.bin]
flash_dic = dict((x['offset'], x['file']) for x in bin_selection)
except KeyError:
print('Invalid binary was selected.')
valid = [k if all(x in v for x in ('offset', 'file')) else None for k, v in iteritems(json_content)]
print('Valid ones:', ' '.join(x for x in valid if x))
exit(1)
else:
flash_dic = json_content['flash_files']
files += [(parse_addr(addr), process_json_file(f_name)) for addr, f_name in iteritems(flash_dic)]
files = sorted([(addr, f_name) for addr, f_name in iteritems(dict(files))],
key=lambda x: x[0]) # remove possible duplicates and sort based on the address
cmd_args = {'output_file': args.output_file,
'files': files,
'chip_id': args.chip_id,
'chunk_size': args.chunk_size,
}
{'write': action_write
}[args.command](cmd_args)
if __name__ == '__main__':
main()