Re: [PATCH] mmc: dw_mmc-rockchip: add dependency to ROCKCHIP_PM_DOMAINS

[Mikko Rapeli <mikko.rapeli@linaro.org>]

Hi,

On Fri, Sep 12, 2025 at 10:47:29AM +0200, Arnd Bergmann wrote:
> On Thu, Sep 11, 2025, at 18:05, Heiko Stübner wrote:
> >
> > Am Donnerstag, 11. September 2025, 17:03:14 Mitteleuropäische 
> > Sommerzeit schrieb Ulf Hansson:
> >> On Thu, 11 Sept 2025 at 16:43, Mikko Rapeli <mikko.rapeli@linaro.org> wrote:
> >> > @@ -866,7 +866,7 @@ config MMC_DW_PCI
> >> >
> >> >  config MMC_DW_ROCKCHIP
> >> >         tristate "Rockchip specific extensions for Synopsys DW Memory Card Interface"
> >> > -       depends on MMC_DW && ARCH_ROCKCHIP
> >> > +       depends on MMC_DW && ARCH_ROCKCHIP && ROCKCHIP_PM_DOMAINS
> 
> The hard dependencies are usually only for compile-time requirements.
> 
> Ideally this should go the other way and use
> 
>       depends on (ARCH_ROCKCHIP || COMPILE_TEST)
> 
> after you check that this actually builds on x86 with COMPILE_TEST
> enabled, as there may be other compile-time dependencies.

Ok so a lot of mmc driver set this correctly but few don't:

 * MMC_PXA
 * MMC_OMAP
 * MMC_ATMELMCI
 * MMC_MXC
 * MMC_MXS
 * MMC_DW_ROCKCHIP
 * MMC_WMT

Maybe there are good reasons why these are not part of COMPILE_TEST.
I can try adding MMC_DW_ROCKCHIP. Did not yet find out how to start
the COMPILE_TEST build. Hints welcome. Which top level Makefile
target or script to run?

> >> Rather than "depends on", I think a "select" is better to be added
> >> from the platform's Kconfig. Probably drivers/soc/rockchip/Kconfig is
> >> where to put this.
> >> 
> >> Assuming that ROCKCHIP_PM_DOMAINS is a critical piece for most
> >> Rockchip platforms to work.
> >
> > I'd think
> > - arch/arm64/Kconfig.platforms
> > - arch/arm/mach-rockchip/Kconfig
> > would be the correct positions.
> >
> > And as Ulf suggested, this should be a "select"
> 
> I think in this case a 'default ARCH_ROCKCHIP' in the
> ROCKCHIP_PM_DOMAINS definition is sufficient to have it
> normally enabled, and still allows someone to try turning
> it into a loadable module later, which would be a requirement
> e.g. for Android GKI.

Ok I can test this out. kernel.org defconfigs have ROCKCHIP_PM_DOMAINS
enabled so they don't see any issues but when users/distros configure
kernels, then ROCKCHIP_PM_DOMAINS is needed for MMC_DW_ROCKCHIP to work.
To me a warning is sufficient from kernel config tooling since yocto
side tooling picks this up but enabling by default is even better.
These runtime dependencies hard to track down. Kernel modules would
be sufficient if support was there since in my case there is always
an initrd with udev and modules.

Cheers,

-Mikko