On Wed, May 23, 2012 at 04:33:53PM -0600, Stephen Warren wrote:

> I'm not 100% sure if this is the approach you had in mind. However, anything
> more all-encompassing would require separating the concepts of endianness and

I wouldn't do this on the bus in the first instance, I'd do it on the
device - it's pretty much orthogonal to the bus what the device wants
and it's perfectly plausible that a device on another bus might've made
unusual endiannness choices.  Otherwise it's pretty much what I was
thinking of.

For MMIO I'd expect that a large proportion of devices on platform buses
would pick native endianness.

> serialization that are coupled together in functions like
> regmap_format_4_12_write, and I'm not sure how that would work conceptually.

Anything using format_write() can be ignored, it's already lost large
chunks of functionality just from that.  Non-integer byte sizes cause
issues.