Re: [PATCH 1/2] asm-generic/io.h: provide default ioremap/iounmap for !HAS_IOMEM

From: Rob Herring
Date: Wed Mar 30 2016 - 09:30:17 EST

On Wed, Mar 30, 2016 at 5:03 AM, Arnd Bergmann <arnd@xxxxxxxx> wrote:
> On Wednesday 30 March 2016 10:13:53 Richard Weinberger wrote:
>> I fully understand your point of view. COMPILE_TEST is a monster that
>> can do the heavy lifting for you, but monsters also have claws and fangs.
>> Having COMPILE_TEST having depend on !UML works for me. But don't
>> we have other archs without io mem? At least a few years ago while
>> porting nandsim to UML I found s390 that lacks of io mem too.
> s390 gained IOMEM support when they started having PCI attachments.

I'm confused how s390 defines HAS_IOMEM and includes it from lib/Kconfig though.

> arch/score selects NO_IOMEM, though they do in fact use MMIO, and
> I'm sure their architecture has lots of other problems with build
> testing that nobody cares about.
> arch/tile can select NO_IOMEM when PCI is disabled, they might
> care about this, though I think they also have other build-time
> issues.
>> Maybe we depend COMPILE_TEST on HAS_IOMEM?
> That sounds fine with me as well.

That or !UML is fine, but I still think we should rid drivers from
depending on HAS_IOMEM. We probably still have cases where the only
dependency besides the driver's subsystem is HAS_IOMEM.