On Fri, Mar 28, 2014 at 03:52:16PM -0400, Nicolas Pitre wrote:

I see it might be hard to make the last two common, unless it is defined 
up front and the odd case does a #undef virt_to_pfn before redefining 
it.  Which way is best I'm not sure.

Given the problems we've had in this area with changes ending up with
build failures, I think it's far better to have simplicity even if it
means duplicating definitions, rather than trying to invent some clever
way to avoid duplications.

We've had soo many build failures due to trying to be too clever that
I'm not playing the "try to be clever" game here anymore.  We can be
more verbose instead.

Also this needs to take care of those machines overriding PHYS_OFFSET at 
run time (see commit a77e0c7b2774f).

Platforms don't override PHYS_OFFSET.  They override PLAT_PHYS_OFFSET
instead.

However it looks like no code 
relying on the LPAE version of early_paging_init() has been merged in 
mainline yet.

I believe that's fully reliant on CONFIG_ARM_PATCH_PHYS_VIRT being set,
and the difference for that case would be setting __pv_phys_pfn_offset
instead of __pv_phys_offset, which is a simple modification.

However, that's something which should be done by core code, not
individually by platforms.  Since we don't have any view of code doing
this yet, it's not relevant here.  We can only deal with what is in
mainline and not in external trees.

-- 
FTTC broadband for 0.8mile line: now at 9.7Mbps down 460kbps up... slowly
improving, and getting towards what was expected from it.