Hi Stefan,
On Oct 03 2015 19:54, Stefan Richter wrote:
On Oct 03 Takashi Sakamoto wrote:
On 2015年10月03日 18:14, Clemens Ladisch wrote:
Most data in the actual configuration ROM is organized as big-endian 32- bit values, so the FireWire core converts _all_ words from big endian to CPU endian. The words that are not actually 32-bit values but strings have to be converted back; the correct function for this is cpu_to_be32().
Yes. And I forgot a pair of cpu_to_be32/be32_to_cpu is symmetric each other! (I'm really a stupid guy, sigh...)
Nah, it's easy to miss that struct fw_device.config_rom ist not identical with the on-the-wire format (on little endian hosts, that is). This has irritated others before. I think this CPU-endian ROM buffer caters to simpler ROM parsing code in the overall picture. Text leaf parsing seems to be the only exception where a big-endian buffer would be preferable, but before your TASCAM driver all of the text leaf parsing was concentrated in core-device.c::textual_leaf_to_string().
Yep. I didn't consider about the pair of macros are preprocessed according to CPU-endianness. Anyway this driver require enough implementation to parse the string.
Thanks
Takashi Sakamoto