Exynos build failure in -next allmodconfig

[Tomasz Figa]

On 16.09.2014 13:44, Krzysztof Kozłowski wrote:
> On 15.09.2014 19:57, Russell King - ARM Linux wrote:
>> On Mon, Sep 15, 2014 at 09:34:58AM -0700, Mark Brown wrote:
>>> On Mon, Sep 15, 2014 at 11:57:09AM +0100, Build bot for Mark Brown
>>> wrote:
>>>
>>> Today's -next got a build failure in ARM allmodconfig due to platsmp.c:
>>>
>>> | arch/arm/mach-exynos/platsmp.c:198:31: warning: incorrect type in
>>> return expression (different address spaces)
>>> | arch/arm/mach-exynos/platsmp.c:198:31:    expected void [noderef]
>>> <asn:2>*
>>> | arch/arm/mach-exynos/platsmp.c:198:31:    got void *
>>> | arch/arm/mach-exynos/platsmp.c:198:31: warning: incorrect type in
>>> return expression (different address spaces)
>>> | arch/arm/mach-exynos/platsmp.c:198:31:    expected void [noderef]
>>> <asn:2>*
>>> | arch/arm/mach-exynos/platsmp.c:198:31:    got void *
>>> |   CC      arch/arm/mach-exynos/platsmp.o
>>> | /tmp/ccC9fkwF.s: Assembler messages:
>>> | /tmp/ccC9fkwF.s:423: Error: selected processor does not support ARM
>>> mode `isb '
>>> | /tmp/ccC9fkwF.s:428: Error: selected processor does not support ARM
>>> mode `isb '
>>> | /tmp/ccC9fkwF.s:429: Error: selected processor does not support ARM
>>> mode `dsb '
>>> | scripts/Makefile.build:257: recipe for target
>>> 'arch/arm/mach-exynos/platsmp.o' failed
>>>
>>> Looks like we need a compiler flags override for that file.
>>
>> Or.. the question is why a .c file is not using the proper macros.
> 
> Actually I am the one to blame for build failure (commit: "ARM: EXYNOS:
> Move code from hotplug.c to platsmp.c"). The problem is
> v7_exit_coherency_flush() which I think does not make sense on ARMv6.
> 
> I'll replace the ISB and DSB commands with macros but the real question
> is whether the mach-exynos/platsmp.c file and mach-exynos directory
> should be compiled when CONFIG_ARCH_EXYNOS is not defined?

I think the problematic case here is v6+v7 multiplatform, where even
though CONFIG_ARCH_EXYNOS is defined, compiler flags for lowest common
denominator (v6) must be used. Using appropriate macros should fix the
problem indeed.

Best regards,
Tomasz