Re: iomap API and big endian interactivity

[Paul Mundt <lethal@linux-sh.org>]

Hi Phil,

On Mon, Apr 02, 2012 at 01:20:33PM +0100, phil.edworthy@renesas.com wrote:
> > On Wed, Oct 12, 2011 at 09:03:24AM +0100, Phil Edworthy wrote:
> > > Commit 37b7a97884ba64bf7d403351ac2a9476ab4f1bba makes all ioread/iowrite
> > > functions treat registers as little endian. For SH2 this is not the case,
> > > they are big endian.
> > >
> > Ok, it looks like we're finally going to have to bite the bullet and
> > support I/O swabbing for the platforms that need it. I dislike this for
> > the reasons that the I/O accessors are supposed to be little endian and
> > in the PCI case we can deal with CPU vs host controller endianness
> > mismatches just fine, ensuring a consistent little endian space.
> ...
> >
> > Hopefully this will be all that's necessary for fixing up the SH-2(A) and
> > SH-4A big endian use cases. Boards can of course override the swabbing
> > behaviour where necessary, though I don't forsee much need to deviate
> > from the generic implementation any time soon given that we've ripped out
> > the address swizzling (in which case we could probably just inline it in
> > io.h directly).
> 
> Works for me on sh7264 (sh2a).
> 
Great, thanks for testing.

> Would you like me to add SWAP_IO_SPACE to all the sh2 defconfigs?
> 
> One minor comment, this option won't show up in a config menu. Is that ok
> for devices that can be used as either big or little endian?
> 
We can conditionalize that in the Kconfig language. So, something like:

	select SWAP_IO_SPACE if CPU_BIG_ENDIAN

should do what you want.