On Wed, Sep 13, 2023 at 12:56:03PM +0200, Amadeusz Sławiński wrote:
> On 9/12/2023 6:32 PM, Richard Fitzgerald wrote:
> > +#ifdef CONFIG_PCI
> > + /*
> > + * PCI does not define 0 as invalid, so pci_subsystem_set indicates
> > + * whether a value has been written to these fields.
> > + */
> > + unsigned short pci_subsystem_vendor;
> > + unsigned short pci_subsystem_device;
> > + bool pci_subsystem_set;
> > +#endif /* CONFIG_PCI */
> > +
> > char topology_shortname[32];
> > struct device *dev;
> This looks bit weird to me, snd_soc_card is _generic_ struct which is not
> device specific in any way, and now you add fields for PCI, can't this
> somehow be done using drvdata or something?
You're right that it's a bit messy but if we use driver data then it
becomes specific to a particular driver and there's a goal here to share
with subfunction drivers. I was thinking we could refactor to a union
or otherwise extend if we find other users with a similar need.
On 13/09/2023 13:58, Mark Brown wrote:
> On Wed, Sep 13, 2023 at 12:56:03PM +0200, Amadeusz Sławiński wrote:
>> On 9/12/2023 6:32 PM, Richard Fitzgerald wrote:
>
>>> +#ifdef CONFIG_PCI
>>> + /*
>>> + * PCI does not define 0 as invalid, so pci_subsystem_set indicates
>>> + * whether a value has been written to these fields.
>>> + */
>>> + unsigned short pci_subsystem_vendor;
>>> + unsigned short pci_subsystem_device;
>>> + bool pci_subsystem_set;
>>> +#endif /* CONFIG_PCI */
>>> +
>>> char topology_shortname[32];
>>> struct device *dev;
>
>> This looks bit weird to me, snd_soc_card is _generic_ struct which is not
>> device specific in any way, and now you add fields for PCI, can't this
>> somehow be done using drvdata or something?
>
> You're right that it's a bit messy but if we use driver data then it
> becomes specific to a particular driver and there's a goal here to share
> with subfunction drivers. I was thinking we could refactor to a union
> or otherwise extend if we find other users with a similar need.
Yes, I was trying to avoid multiple custom ways of passing the same
data around. A significant advantage of explicitly passing the SSID
(if it's available) rather than a more abstract identifier (such as a
char *) is that's it's very well defined exactly what a PCI SSID is so
we know we can use it verbatim. A more abstract identifier creates an
implied trust (or mistrust) between the machine driver and the component
receiving it whether it's unique and in a useful format.
I could de-ugly it a bit by moving it out into a separate struct/union
and having just a member of that struct type in snd_soc_card.
An alternative was to add a function like the existing
snd_soc_component_set_whatever() family but that means adding a callback
pointer to struct snd_soc_component_driver, which is creating more
space overhead than one value in the snd_soc_card.