Re: [PATCH] clk: Allow drivers to build if COMPILE_TEST is enabled

[Javier Martinez Canillas <javier@osg.samsung.com>]

Hello Krzysztof,

On 10/15/2015 04:04 AM, Krzysztof Kozlowski wrote:
> 2015-10-15 4:08 GMT+09:00 Javier Martinez Canillas <javier@osg.samsung.com>:
>> Hello Stephen,
>>
>> On 10/14/2015 08:38 PM, Stephen Boyd wrote:
>>> On 10/13, Javier Martinez Canillas wrote:
>>>> diff --git a/drivers/clk/versatile/Kconfig b/drivers/clk/versatile/Kconfig
>>>> index 1530c9352a76..fc50b6264bed 100644
>>>> --- a/drivers/clk/versatile/Kconfig
>>>> +++ b/drivers/clk/versatile/Kconfig
>>>> @@ -1,6 +1,6 @@
>>>>  config COMMON_CLK_VERSATILE
>>>>      bool "Clock driver for ARM Reference designs"
>>>> -    depends on ARCH_INTEGRATOR || ARCH_REALVIEW || ARCH_VEXPRESS || ARM64
>>>> +    depends on ARCH_INTEGRATOR || ARCH_REALVIEW || ARCH_VEXPRESS || ARM64 || COMPILE_TEST
>>>
>>> Have you compiled these drivers on an architecture that doesn't
>>> have IOMEM? Perhaps tile or um? I'm all for more build coverage,
>>> but it's not always as simple as just sprinkling some
>>> COMPILE_TEST around the Kconfigs.
>>>
>>
>> No, I only build tested on arm32 and x86. The 0-day bot haven't reported a
>> build error yet and I didn't see any platform dependent code in the drivers.
> 
> I see you guys with Luis are adding a lot of COMPILE_TEST. But

Yes, the motivation for this was that I've been helping Mauro with a big
rework in the media subsystem [0] and was annoying to audit that all the
drivers were converted to the new APIs and no compile regressions were
introduced in drivers that could not be built with COMPILE_TEST enabled.

Most media drivers are able to be build though so I thought it would be
a good idea to extend the build coverage in all the other subsystems.

> building only on these two architectures *is not enough*. Run at least
> armv8, PPC and the x86_64. MIPS would be nice as well (I use the
> CodeSourcery's MIPS). All of these (ARM64, X86_64, PPC, MIPS) can be
> easily installed on typical debian-like Linux distro. Really easily.
>

Thanks, Stephen also pointed out to the toolchains in kernel.org [1].
 
> By adding this non-tested build coverage you can actually fail some
> other architecture's allyesconfig/allmodconfig builds.
>

Agreed, unfortunately having more build coverage is not as trivial as I
originally thought. Not only because it can break the build in obscure
archs that I don't have a toolchain to test but also exposes more build
warnings (as reported by the 0-day bot) that I've the bandwidth to fix.

So personally I'll stop trying to enabled COMPILE_TEST just to be safe.

> Best regards,
> Krzysztof
> 

[0]: http://lists.infradead.org/pipermail/linux-arm-kernel/2015-August/367109.html
[1]: https://www.kernel.org/pub/tools/crosstool/

Best regards,
-- 
Javier Martinez Canillas
Open Source Group
Samsung Research America