* Tony Lindgren <tony@atomide.com> [080516 14:49]:
> * David Brownell <david-b@pacbell.net> [080516 14:26]:
> > On Friday 16 May 2008, Tony Lindgren wrote:
> > > @@ -38,19 +38,19 @@
> > >  #define        CF_BASE 0xfffe2800
> > >  
> > >  /* status; read after IRQ */
> > > -#define CF_STATUS_REG          __REG16(CF_BASE + 0x00)
> > > +#define CF_STATUS                      (CF_BASE + 0x00)
> > >  #      define  CF_STATUS_BAD_READ      (1 << 2)
> > >  #      define  CF_STATUS_BAD_WRITE     (1 << 1)
> > >  #      define  CF_STATUS_CARD_DETECT   (1 << 0)
> > >  
> > >  /* which chipselect (CS0..CS3) is used for CF (active low) */
> > > -#define CF_CFG_REG             __REG16(CF_BASE + 0x02)
> > > +#define CF_CFG                         (CF_BASE + 0x02)
> > >  
> > >	...
> > 
> > Trying to understand the plan here.  This first patches
> > are to remove __REG*() access, we hillater patches will 
> > be needed to convert things to omap_readl(BASE + OFFSET)
> > style accessors?  (BASE being SOC-specific, and passed
> > down from system init code.)
> > 
> > Not that CF is a good example of that.  I don't think
> > it exists on current chips.  ;)
> 
> Well ideally we would set the base offset during driver init, then
> just use  __raw_read/write().
> 
> But that's lot of work, so it's easier first to convert __REG access
> to use omap_read/write(). I'll post multi-omap series as soon as I have
> it booting.. But to give you and idea, we can have something like following
> work for multi-omap.

...

Turns out there were bunch of issues with the patches above..
Here's a patch fixing them. I will repost this series soon, just FYI.

Tony