On Sun, Nov 6, 2022, at 11:37, Arnd Bergmann wrote:
> On Sun, Nov 6, 2022, at 11:26, Russell King (Oracle) wrote:
>> On Sun, Nov 06, 2022 at 08:43:50AM +0100, Arnd Bergmann wrote:
>>> --- a/arch/arm/mach-omap1/Kconfig
>>> +++ b/arch/arm/mach-omap1/Kconfig
>>> @@ -49,7 +49,7 @@ config ARCH_OMAP1_ANY
>>> select ARCH_OMAP
>>> def_bool ARCH_OMAP730 || ARCH_OMAP850 || ARCH_OMAP15XX || ARCH_OMAP16XX
>>>
>>> -config ARCH_OMAP
>>> +config ARCH_OMAP1_ANY
>>
>> This patch can't be right - look at the first line of context above, you
>> have symbols that select ARCH_OMAP and you've just removed the
>> definition of ARCH_OMAP.
>
> Right, I misread this, I thought this was an 'if ARCH_OMAP' block
> instead of the Kconfig symbol. Part of the problem is now is
> possibly that there are two symbols with that name now, I'll have
> to try to reproduce the problem first and then see what caused it
> originally.
It seems the root cause was actually 804f7f19c2e2 ("fbdev: omap:
avoid using mach/*.h files"), where I started allowing building with
CONFIG_COMPILE_TEST, but this fails when ARCH_OMAP1_ANY is disabled,
as Randy correctly found.
We could tighten this gain by adding something like
depends on ARCH_OMAP1_ANY || !OMAP_MUX
to still allow compile-testing but specifically avoiding the
broken case. I think we have the same problem in the pcmcia and
spi-uwire drivers, so maybe something like this would be
a better approach:
--- a/include/linux/soc/ti/omap1-mux.h
+++ b/include/linux/soc/ti/omap1-mux.h
@@ -302,7 +302,7 @@ enum omap1xxx_index {
};
-#ifdef CONFIG_OMAP_MUX
+#if defined(CONFIG_OMAP_MUX) && defined(CONFIG_ARCH_OMAP1_ANY)
extern int omap_cfg_reg(unsigned long reg_cfg);
#else
static inline int omap_cfg_reg(unsigned long reg_cfg) { return 0; }
Arnd
