On Fri, Feb 10, 2012 at 01:33:16PM +0100, Lars-Peter Clausen wrote:

> regmap_write doesn't always go the raw_write path. Also with this approach

We can ignore the non-byte formats as a wart for this; they're cut out
of this area of the subsystem functionality and just get single register
write operatons with single register cache access.

> we end up formatting the value into the raw format, only to parse it again
> in the next step.

Yes, like I say there's room for optimisation.  For 8 bit values there
would be win from working out that we don't need to do anything at all, 
though I doubt it's ever going to be performance critical.

> Furthermore I don't think it makes sense to cache raw values as the cache
> operates on a register level, not on a byte level.

The entire API operates on registers really.  The only reason
raw_write() takes a length is because it's what most of the users are
actually likely to have to hand, trying to write to less than a register
isn't particularly sensible.