Re: [alsa-devel] [PATCH v5 08/11] ASoC: Intel: atom: Make PCI dependency explicit

[Rafael J. Wysocki]

On Wed, Jan 2, 2019 at 9:33 PM Pierre-Louis Bossart
<pierre-louis.bossart@xxxxxxxxxxxxxxx> wrote:
>
> I have three opens with this ACPI/PCI change
>
> 1. the baseline change fails on my cross-compilation checks, see below
> the result of the attached script (simplification of the one I use to
> avoid 0day reports).

What baseline change?

That failure is not related to PCI if I'm not missing anything.

> 2. there are different patterns to express the dependency on PCI e.g.
>
>   config MMC_SDHCI_ACPI
>       tristate "SDHCI support for ACPI enumerated SDHCI controllers"
>       depends on MMC_SDHCI && ACPI
> -    select IOSF_MBI if X86
> +    select IOSF_MBI if (X86 && PCI)
>
> but
>
> config SND_SST_ATOM_HIFI2_PLATFORM_ACPI
>       tristate "ACPI HiFi2 (Baytrail, Cherrytrail) Platforms"
>       default ACPI
> -    depends on X86 && ACPI
> +    depends on X86 && ACPI && PCI
>       select SND_SST_IPC_ACPI
>       select SND_SST_ATOM_HIFI2_PLATFORM
>       select SND_SOC_ACPI_INTEL_MATCH
>
> IOSF is only needed for Baytrail-CR detection, and the code will compile
> fine without it, so maybe it'd be a better model if you used the
> following diff?
>
> diff --git a/sound/soc/intel/Kconfig b/sound/soc/intel/Kconfig
> index 2fd1b61e8331..68af0ea5c96c 100644
> --- a/sound/soc/intel/Kconfig
> +++ b/sound/soc/intel/Kconfig
> @@ -95,7 +95,7 @@ config SND_SST_ATOM_HIFI2_PLATFORM_ACPI
>          select SND_SST_IPC_ACPI
>          select SND_SST_ATOM_HIFI2_PLATFORM
>          select SND_SOC_ACPI_INTEL_MATCH
> -       select IOSF_MBI
> +       select IOSF_MBI if PCI

Well, does it actually make sense to ever set
SND_SST_ATOM_HIFI2_PLATFORM_ACPI without PCI?

> 3. All the Intel machine drivers depend on X86_INTEL_LPSS which depends
> on PCI. But for Baytrail/Haswell/Broadwell we have only a dependency on
> ACPI, so we expose drivers that can be selected but fail on probe since
> there are no machine drivers. I am not sure if we want to be strict and
> only expose meaningful configurations, or allow for more compilations
> tests and corner cases?

I would only expose meaningful configurations to start with and then
*maybe* relax that going forward as long as the benefit is worth it.

Cheers,
Rafael