Re: [PATCH 19/24] sh: use __iomem pointers for MMIO

[Arnd Bergmann <arnd@arndb.de>]

On Tuesday 18 September 2012, Paul Mundt wrote:
> On Fri, Sep 14, 2012 at 11:34:47PM +0200, Arnd Bergmann wrote:
> > ARM is moving to stricter checks on readl/write functions,
> > so we need to use the correct types everywhere.
> > 
> > I'm not completely sure about this patch, and it will
> > probably require some arch/sh changes to go along with it,
> > but it's clear that something has to be done to avoid
> > getting hundreds of new warnings on each shmobile build
> > in v3.7.
> > Please see this as a prototype.
> > 
> I have no intention of making this change for arch/sh.
> 
> The __raw variants already accept both __iomem pointers and integer
> addresses, which was largely intentional. New code could use the __iomem
> annotations while older code could continue to use the integer addresses
> without issue.

Ok, I'm dropping this patch from the series then.

> If you wish to go through the kernel and audit every
> single __raw user, you're certainly welcome to, but until then such a
> change is premature.

I've done it for all the defconfig files for now, which probably covers
most of the drivers that are relevant on ARM. I still have a backlog
of unrelated warning fixes from that. Once I'm done with those, I'm planning
to do another round of randconfig builds, which hopefully catches the
rest. I'll just exclude drivers/sh from those builds for now.

	Arnd