From mboxrd@z Thu Jan  1 00:00:00 1970
From: jonathan.austin@arm.com (Jonathan Austin)
Date: Tue, 21 May 2013 17:25:11 +0100
Subject: [PATCH v2 1/3] ARM: allow platforms to use generic headers without
 ARCH_MULTIPLATFORM
In-Reply-To: <20130515233335.GO5600@atomide.com>
References: <1368530845-7415-1-git-send-email-jonathan.austin@arm.com>
 <1368530845-7415-2-git-send-email-jonathan.austin@arm.com>
 <5192532B.1070705@gmail.com> <20130515233335.GO5600@atomide.com>
Message-ID: <519B9FE7.3030907@arm.com>
To: linux-arm-kernel@lists.infradead.org
List-Id: linux-arm-kernel.lists.infradead.org

Hi Tony,

On 16/05/13 00:33, Tony Lindgren wrote:
> * Rob Herring <robherring2@gmail.com> [130514 08:12]:
>> On 05/14/2013 06:27 AM, Jonathan Austin wrote:
>>> Currently the use of generic mach headers is only is controlled by the same
>>> config option as selection of a multiplatform kernel (i.e.
>>> ARCH_MULTIPLATFORM). However, there are cases where it is valid to use the
>>> generic headers but still build a single platform kernel.
>>>
>>> One such example is when using the vexpress platform with a core that does
>>> not have an MMU; a multiplatform kernel does not make sense, but use of the
>>> generic headers is required by the vexpress platform since 617276307cd4c
>>> ("ARM: vexpress: convert to multi-platform").
>>>
>>> This patch untangles these two options by introducing a new
>>> ARM_GENERIC_HEADERS config option that ARCH_MULTIPLATFORM selects. Other
>>> platforms can now select ARM_GENERIC_HEADERS independently.
>>
>> I find this name confusing. It is no headers you are selecting and
>> generic is a bit overused IMHO. So how about something like
>> NEED_NO_MACH_HEADERS to somewhat align to the other header related
>> config options (e.g. NEED_MACH_GPIO_H).
>
> Hmm it _looks_ like this won't allow the drivers again to
> #include <plat/*.h> and <mach/*.h>. Have you checked that?
>
> It might be worth checking as we really don't want to go back there
> again.. Currently arch/arm/Makefile has:
>
> ifneq ($(CONFIG_ARCH_MULTIPLATFORM),y)
> ifeq ($(KBUILD_SRC),)
> KBUILD_CPPFLAGS += $(patsubst %,-I%include,$(machdirs) $(platdirs))
> else
> KBUILD_CPPFLAGS += $(patsubst %,-I$(srctree)/%include,$(machdirs) $(platdirs))
> endif
> endif
>
> So you may need to do something there to prevent drivers from again
> including machdirs and platdirs.
>
If I've understood you correctly, then this patch already deals with 
this case (though it isn't hugely obvious from the diff...) by replacing 
CONFIG_ARCH_MULTIPLATFORM in your example above with 
CONFIG_ARM_GENERIC_HEADERS (soon to be renamed, as per Rob's suggestion).

Because ARCH_MULTIPLATFORM selects 
CONFIG_(FORMERLY_KNOWN_AS_)ARM_GENERIC_HEADERS it shouldn't be possible 
for anything multi-platform to start including <plat/*.h> and <mach/*.h> 
again.

This seems to be the right thing to do because it also makes it 
impossible to include <plat/*.h> and <mach/*.h> in !MULTIPLATFORM 
situations that don't need mach/plat headers, like the VEXPRESS_NOMMU case.

Is that what you were getting at? Please let me know if not...

Thanks,

Jonny

> Regards,
>
> Tony
>