Re: [PATCH 2/2] ASoC: intel: fix soundwire dependencies

[Arnd Bergmann <arnd@kernel.org>]

On Tue, Jan 12, 2021 at 10:03 PM Pierre-Louis Bossart
<pierre-louis.bossart@linux.intel.com> wrote:
> On 1/12/21 2:32 PM, Arnd Bergmann wrote:
> > From: Arnd Bergmann <arnd@arndb.de>
> >
> > The Kconfig logic around SND_SOC_SOF_INTEL_SOUNDWIRE tries to
> > ensure that all sound modules can be built with the minimal
> > dependencies, but this fails in some configurations:
> >
> > x86_64-linux-ld: sound/hda/intel-dsp-config.o: in function `snd_intel_dsp_driver_probe':
> > intel-dsp-config.c:(.text+0x134): undefined reference to `sdw_intel_acpi_scan'
> >
> > Specifically, this happens if the dsp-config driver is built-in but does
> > not need to use soundwire, while CONFIG_SOUNDWIRE_INTEL is enabled as
> > a loadable module.
> >
> > An easier and more correct way to do this is to remove
> > CONFIG_SND_SOC_SOF_INTEL_SOUNDWIRE_LINK and instead have
> > the two drivers that can link against SOUNDWIRE_INTEL,
> > i.e. DSP_CONFIG and SND_SOC_SOF_HDA, select that driver whenever
> > SND_SOC_SOF_INTEL_SOUNDWIRE_LINK is set.
> >
> > This however means that SND_SOC_SOF_INTEL_SOUNDWIRE cannot be selected
> > by users when SOUNDWIRE is only usable by loadable modules and one or
> > more drivers using SND_SOC_SOF_INTEL_SOUNDWIRE_LINK_BASELINE is built-in.
>
> The problem is real, but the proposal isn't completely right, there is
> absolutely no logical link or functional dependency between
> INTEL_DSP_CONFIG and SOUNDWIRE.

If that is true, would it be possible to move the call to
sdw_intel_acpi_scan() out of these drivers and one layer
higher where the dependency actually is?

I was indeed wondering whether the intel-dsp-config.c is just
another layering violation: this is another generic piece
of code that seems to contain too much knowledge about
specific hardware implementations.

> We have a similar case with HDaudio, the two buses can be selected as
> tristates, but the SOF configuration needs to match.
>
> In both cases, either we add a 'depends' and users need to make sure the
> configurations match on the two sides. Or we use select but one of the
> selections will be overridden and becomes meaningless.

Maybe something like this:

config SND_SOC_SOF_INTEL_SOUNDWIRE
-        bool "SOF support for SoundWire"
+       tristate "SOF support for SoundWire"
-       depends on SOUNDWIRE && ACPI
+       depends on SND_SOC_SOF_INTEL_SOUNDWIRE_LINK_BASELINE
+       depends on SOUNDWIRE
+       depends on ACPI
+       depends on !(SOUNDWIRE=m && SND_SOC_SOF_INTEL_SOUNDWIRE_LINK_BASELINE=y)
+       select SOUNDWIRE_INTEL

I have not tried it, but that should keep it all in one place.

> Arnd, do you mind if I give it a try on my side?

I have no specific attachment to my patch, this was just what I came up
with to fix the build regression locally.

       Arnd