20 Nov
2012
20 Nov
'12
1:03 p.m.
On 19/11/12 18:46, Takashi Iwai wrote:
I find the idea is great. Looking through the patch, a few things came to my mind:
- Endianness and alignment
- Forward-compatibility
- Avoid the bitfield usages
About endianness: the firmware is usually provided as endian-independent. That is, the driver is supposed to convert to CPU endianness properly, or check and reject the invalid firmware, at least.
Yeah, that's something we don't convert atm, but we do check with a u32 magic number in each header (and will reject that object if the header is missing or incorrect).
I think we are good with the magic number check atm and I can add some explicit endianess conversion when we have more users. Atm, the only users will be ARM based.
Thanks
Liam