Re: irq-loongson-pch-pic.c:undefined reference to `of_iomap'

[Krzysztof Kozlowski <krzk@kernel.org>]

On Mon, Nov 16, 2020 at 10:21:26AM +0100, Arnd Bergmann wrote:
> On Mon, Nov 16, 2020 at 9:37 AM Geert Uytterhoeven <geert@linux-m68k.org> wrote:
> > On Mon, Nov 16, 2020 at 9:33 AM Arnd Bergmann <arnd@kernel.org> wrote:
> > > On Mon, Nov 16, 2020 at 5:33 AM Randy Dunlap <rdunlap@infradead.org> wrote:
> > > > On 11/13/20 3:27 PM, kernel test robot wrote:
> > > (adding s390 folks to cc)
> > >
> > > I think fixing this requires a larger-scale effort. I tried building
> > > an s390 allmodconfig
> > > with CONFIG_PCI disabled, and got warnings and failures in many other places,
> > > see full log at the end of this mail.
> > >
> > > While in theory, all of those should depend on 'HAS_IOMEM' or some other symbol,
> > > keeping these dependencies sounds like an uphill battle, and there is not much
> > > to be gained from building the drivers for s390 on top of compile-testing them
> > > on more conventional architectures.
> >
> > Don't we need the dependencies on HAS_IOMEM for the CONFIG_UML=y
> > case, too?
> 
> I would have expected that as well, but I don't see the problem when building
> an arch/um kernel, all I get is
> 
> ERROR: modpost: "devm_platform_ioremap_resource"
> [drivers/iio/adc/adi-axi-adc.ko] undefined!
> ERROR: modpost: "devm_platform_ioremap_resource"
> [drivers/ptp/ptp_ines.ko] undefined!
> ERROR: modpost: "devm_ioremap_resource"
> [drivers/net/ethernet/xilinx/xilinx_emac.ko] undefined!
> ERROR: modpost: "devm_platform_ioremap_resource_byname"
> [drivers/net/ethernet/xilinx/ll_temac.ko] undefined!
> ERROR: modpost: "devm_ioremap"
> [drivers/net/ethernet/xilinx/ll_temac.ko] undefined!
> ERROR: modpost: "devm_of_iomap"
> [drivers/net/ethernet/xilinx/ll_temac.ko] undefined!
> ERROR: modpost: "__open64_2" [fs/hostfs/hostfs.ko] undefined!
> 
> If I disable those five drivers, I can build and link a uml kernel without
> warnings. I could not find the difference compared to s390 here.
> 
> Looking a bit further, I now find that we ended up disabling CONFIG_COMPILE_TEST
> entirely for arch/um, which is clearly an option that would also work for s390.

Yes, that was the easier solution than to spread "depends on HAS_IOMEM"
all over Kconfigs.

+Cc Greg KH,

I got similar report around phy drivers:
https://lore.kernel.org/lkml/202011140335.tceVqHmN-lkp@intel.com/

When reproducing this, I saw multiple unmet dependencies on s390 for
MFD_SYSCON and MFD_STM32_TIMERS.

I suppose there is no point to fix them all because this will be
basically UML case, so HAS_IOMEM all over the tree.

Best regards,
Krzysztof