Re: [PATCH 09/13] ASoC: amd: acp: add machine driver support for pdm use case

[Mario Limonciello]

On 10/27/2023 10:51, Krzysztof Kozlowski wrote:
> On 27/10/2023 17:28, Mario Limonciello wrote:
> > On 10/27/2023 03:49, Krzysztof Kozlowski wrote:
> > > On 21/10/2023 16:50, Syed Saba Kareem wrote:
> > > > add pdm use case machine driver support
> > > >
> > > > Signed-off-by: Syed Saba Kareem <Syed.SabaKareem@xxxxxxx>
> > > > ---
> > >
> > >
> > > >    	dmi_id = dmi_first_match(acp_quirk_table);
> > > >    	if (dmi_id && dmi_id->driver_data)
> > > > @@ -214,6 +221,10 @@ static const struct platform_device_id board_ids[] = {
> > > >    		.name = "rmb-rt5682s-rt1019",
> > > >    		.driver_data = (kernel_ulong_t)&rt5682s_rt1019_rmb_data,
> > > >    	},
> > > > +	{
> > > > +		.name = "acp-pdm-mach",
> > > > +		.driver_data = (kernel_ulong_t)&acp_dmic_data,
> > > > +	},
> > > >    	{ }
> > > >    };
> > > >    static struct platform_driver acp_asoc_audio = {
> > > > @@ -235,4 +246,5 @@ MODULE_ALIAS("platform:acp3xalc5682s1019");
> > > >    MODULE_ALIAS("platform:acp3x-es83xx");
> > > >    MODULE_ALIAS("platform:rmb-nau8825-max");
> > > >    MODULE_ALIAS("platform:rmb-rt5682s-rt1019");
> > > > +MODULE_ALIAS("platform:acp-pdm-mach");
> > >
> > > Please stop growing the aliases. Module alias is not a substitute for
> > > missing MODULE_DEVICE_TABLE.
> > >
> > > Best regards,
> > > Krzysztof
> >
> > I thought the way that this works is that top level ACP driver (IE
> > acp-pci.c) will have MODULE_DEVICE_TABLE.  This is how that module gets
> > loaded.
> >
> > Then it creates platform devices based on the detected needs for the
> > situation and the creation of those platform devices triggers a uevent
> > which due to MODULE_ALIAS will get appropriate other platform drivers
> > like this one loaded.
>
> And why you cannot use MODULE_DEVICE_TABLE here? IOW, why do you need to
> manually duplicate entire table and re-invent MODULE_DEVICE_TABLE with
> MODULE_ALIAS?

What would actually go into MODULE_DEVICE_TABLE?

The platform devices created are contingent upon what was found during 
the top level ACP driver probe.  You don't want all the "child" platform 
drivers to load unless they're needed.