On Mon, Jul 15, 2024 at 08:12:39AM +0200, Arnd Bergmann wrote:
> On Mon, Jul 15, 2024, at 05:24, Nathan Chancellor wrote:
> > On Mon, Jun 17, 2024 at 05:26:02PM +0200, Marek Behún wrote:
...
> >
> > I am seeing the following Kconfig warning from ARCH=s390 allmodconfig:
> >
> > WARNING: unmet direct dependencies detected for RTC_CLASS
> > Depends on [n]: !S390 [=y]
> > Selected by [m]:
> > - TURRIS_OMNIA_MCU [=m] && CZNIC_PLATFORMS [=y] && (MACH_ARMADA_38X
> > || COMPILE_TEST [=y]) && I2C [=m] && OF [=y] && WATCHDOG [=y]
> >
> > because of:
> >
> > menuconfig RTC_CLASS
> > bool "Real Time Clock"
> > default n
> > depends on !S390
> >
> > which appears to have ultimately come from commit 9556fb73edfc ("[S390]
> > Kconfig: unwanted menus for s390."). No other driver appears to
> > unconditionally select this (I only see it selected within
> > arch/*/Kconfig), so it does not look like this has come up before.
> > Should s390 be excluded from the COMPILE_TEST dependency?
>
> There is really no reason for a driver to select another subsystem,
> it not just causes problems like this one but also leads to
> circular dependencies and surprises when someone turns on
> a random driver and then turns it off again, leaving the
> the other subsystems accidentally enabled.
Makes sense.
> I've applied the fixup below now, leaving GPIOLIB_IRQCHIP
> as the only selected symbol since this is not user-visible.
Thanks.
> Marek, you could consider changing the driver so it doesn't
> actually require all those subsystems at build time but instead
> just leaves out the functionality. Some subsystems actually
> have a stub implementation that makes it work by just dropping
> the dependency, but I did not try that here.
I will look into that.
Marek
