On Wed, May 23, 2012 at 05:28:39PM -0600, Stephen Warren wrote:

> I did briefly consider making this a property of regmap_config rather
> than regmap_bus, but as you say, it'd mean every MMIO user would have
> to specify the endianness value.

I didn't say that, that'd be appauling!

> Also, it doesn't seem right for a device to be able to specify
> register formatting endianness for MMIO; presumably we'd always want
> that native.

Depends on what's going on with bought in IP and register map standards
(and things like PCI) - you do get non-native IPs turning up often enough.

> I suppose we could avoid every device having to specify the endianness
> by introducing a fourth "DEFAULT" value == 0, and having the bus
> define what default means - that way, I wouldn't have to edit any
> drivers due to adding the regmap_bus field.

That's absolutely essential, there's already an explicit idea of this in
the current code's assumption that everything is big endian.