Hi Kbuild/config people,

Could you please take a look at this select problem.

I attempted to solve this kconfig select logic problem a few years back 
but IIRC, my solution leaked memory or had some other problem I could not 
find solution to (and now the code has changed enough I couldn't even get 
my buggy solution easily forward-ported so I just dropped the solution and 
left just the test case into the patch).

As the Kconfig problem back then got resolved through other means, I never 
mentioned this back then but seems the same problem likely happens here
in some other form (bool selecting tristate that ends up getting only =m).

On Mon, 10 Mar 2025, Kurt Borja wrote:
> On Mon Mar 10, 2025 at 9:29 AM -05, Ilpo Järvinen wrote:
> > On Sun, 9 Mar 2025, Kurt Borja wrote:
> >
> >> If ACPI_PLATFORM_PROFILE is selected by ALIENWARE_WMI_WMAX, the former
> >> is forced to be at least =m, because the latter is a bool.
> >> 
> >> This allows the following config:
> >> 
> >> 	CONFIG_ALIENWARE_WMI=y
> >> 	CONFIG_ACPI_PLATFORM_PROFILE=m
> >
> > Hi,
> >
> > selecting from =y should not result in =m for the other symbol. This is 
> > a bug in Kconfig infrastructure.
> >
> > I ran across this a few years back and even had a test case to prove the 
> > select bug but back then the original problem eventually was solved in a 
> > different way which no longer hit the problem. I never could figure out
> > how to fix the kconfig logic though without breaking something and it 
> > ended up into low priority bin and never got solved.
> >
> > Sadly, it seems I've lost the test case patch that exhibits the bug 
> > somewhere... I'll try to look for it from my archived files.
> 
> That's funny.
> 
> I thought this was a Kconfig quirk, that resulted from the following
> hierarchy:
> 
> Type		0	1	2
> =============== ======= ======= =======
> Bool		n	y

I think y should be 2 in both cases so select should cause the selected 
symbol to becomes =y but there's a logic problem deep in select logic in 
the kconfig code.

I've attached the multi-select based reproducer test case. make 
testconfig runs kconfig selftests and results in CONFIG_A3=m despite 
CONFIG_C3=y selecting it.

Please do not apply the test case before the problem is fixed.

> Tristate	n	m	y
> 
> So a <bool> selecting the <tristate> would force it to be at least =m.
> 
> The same thing happens with depend, where a dependecy would be fulfilled
> for a <bool> if a <tristate> was at least =m. That's why in the kernel
> robot report the linking error was also due to the HWMON dependency.
> 
> Anyway, this patch could serve as a workaround if you feel it's
> necessary. I'm going to put the HWMON dependecy in the ALIENWARE_WMI
> symbol for my other series.
> 
> 

-- 
 i.