Re: [PATCH 1/3] dt-bindings: wireless: add ath11k pcie bindings

Re: [PATCH 1/3] dt-bindings: wireless: add ath11k pcie bindings

[Kalle Valo]

+ ath11k list

Johan Hovold <johan@kernel.org> writes:

> On Mon, Mar 20, 2023 at 02:22:12PM +0200, Kalle Valo wrote:
>> Johan Hovold <johan+linaro@kernel.org> writes:
>> 
>> > Add devicetree bindings for Qualcomm ath11k PCIe devices such as WCN6856
>> > for which the calibration data variant may need to be described.
>> >
>> > Signed-off-by: Johan Hovold <johan+linaro@kernel.org>
>> > ---
>> >  .../bindings/net/wireless/pci17cb,1103.yaml   | 56 +++++++++++++++++++
>> >  1 file changed, 56 insertions(+)
>> >  create mode 100644
>> > Documentation/devicetree/bindings/net/wireless/pci17cb,1103.yaml
>> 
>> I'm confused (as usual), how does this differ from
>> bindings/net/wireless/qcom,ath11k.yaml? Why we need two .yaml files?
>
> Almost none of bindings/net/wireless/qcom,ath11k.yaml applies to WCN6856
> when using PCIe (e.g. as most properties are then discoverable).
>
> We could try to encode everything in one file, but that would likely
> just result in a big mess of a schema with conditionals all over.

Ah, so the current qcom,ath11k.yaml would be only for ath11k AHB devices
and this new file is only for ath11k PCI devices? But why still the odd
name pci17cb,1103.yaml? It's not really descriptive and I'm for sure
will not remember that pci17cb,1103.yaml is for ath11k :)

Also it doesn't look good that we have qcom,ath11k-calibration-variant
documented twice now. I'm no DT expert but isn't there any other way? Is
it possible to include other files? For example, if we would have three
files:

qcom,ath11k.yaml
qcom,ath11k-ahb.yaml
qcom,ath11k-pci.yaml

Then have the common properties like ath11k-calibration-variant in the
first file and ahb/pci files would include that.

-- 
https://patchwork.kernel.org/project/linux-wireless/list/

https://wireless.wiki.kernel.org/en/developers/documentation/submittingpatches

-- 
ath11k mailing list
ath11k@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/ath11k

Re: [PATCH 1/3] dt-bindings: wireless: add ath11k pcie bindings

[Johan Hovold]

On Mon, Mar 20, 2023 at 08:41:21PM +0200, Kalle Valo wrote:
> + ath11k list
> 
> Johan Hovold <johan@kernel.org> writes:
> 
> > On Mon, Mar 20, 2023 at 02:22:12PM +0200, Kalle Valo wrote:
> >> Johan Hovold <johan+linaro@kernel.org> writes:
> >> 
> >> > Add devicetree bindings for Qualcomm ath11k PCIe devices such as WCN6856
> >> > for which the calibration data variant may need to be described.
> >> >
> >> > Signed-off-by: Johan Hovold <johan+linaro@kernel.org>
> >> > ---
> >> >  .../bindings/net/wireless/pci17cb,1103.yaml   | 56 +++++++++++++++++++
> >> >  1 file changed, 56 insertions(+)
> >> >  create mode 100644
> >> > Documentation/devicetree/bindings/net/wireless/pci17cb,1103.yaml
> >> 
> >> I'm confused (as usual), how does this differ from
> >> bindings/net/wireless/qcom,ath11k.yaml? Why we need two .yaml files?
> >
> > Almost none of bindings/net/wireless/qcom,ath11k.yaml applies to WCN6856
> > when using PCIe (e.g. as most properties are then discoverable).
> >
> > We could try to encode everything in one file, but that would likely
> > just result in a big mess of a schema with conditionals all over.
> 
> Ah, so the current qcom,ath11k.yaml would be only for ath11k AHB devices
> and this new file is only for ath11k PCI devices?

Right, there would two separate schema files for the two device classes.

> But why still the odd
> name pci17cb,1103.yaml? It's not really descriptive and I'm for sure
> will not remember that pci17cb,1103.yaml is for ath11k :)

Yeah, it's not the best name from that perspective, but it follows the
current convention of naming the schema files after the first compatible
added.

That said, we don't have many schemas for PCI devices so perhaps we can
establish a new convention for those. Perhaps by replacing the numerical
ids with what we'd use if these were platform devices (e.g.
'qcom,wcn6855.yaml').

As long as the DT maintainers are OK with it, I'd also be happy with
something like you suggest below:

	qcom,ath11k-ahb.yaml
	qcom,ath11k-pci.yaml

(or simply not renaming the current file 'qcom,ath11k.yaml') but I have
gotten push back on that in the past.

> Also it doesn't look good that we have qcom,ath11k-calibration-variant
> documented twice now. I'm no DT expert but isn't there any other way? Is
> it possible to include other files? For example, if we would have three
> files:
> 
> qcom,ath11k.yaml
> qcom,ath11k-ahb.yaml
> qcom,ath11k-pci.yaml
> 
> Then have the common properties like ath11k-calibration-variant in the
> first file and ahb/pci files would include that.

That should be possible, but it's not necessarily better as you'd then
have to look up two files to see the bindings for either device class
(and as far as I can tell there would not be much sharing beyond this
single property).

Note that the property could just have well have been named
'qcom,calibration-variant' and then it would be shared also with the
ath10k set of devices which currently holds another definition of what
is essentially the same property ('qcom,ath10k-calibration-variant').

Johan

-- 
ath11k mailing list
ath11k@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/ath11k

Re: [PATCH 1/3] dt-bindings: wireless: add ath11k pcie bindings

[Kalle Valo]

Johan Hovold <johan@kernel.org> writes:

> On Mon, Mar 20, 2023 at 08:41:21PM +0200, Kalle Valo wrote:
>
>> + ath11k list
>> 
>> Johan Hovold <johan@kernel.org> writes:
>> 
>> > On Mon, Mar 20, 2023 at 02:22:12PM +0200, Kalle Valo wrote:
>> >> Johan Hovold <johan+linaro@kernel.org> writes:
>> >> 
>> >> > Add devicetree bindings for Qualcomm ath11k PCIe devices such as WCN6856
>> >> > for which the calibration data variant may need to be described.
>> >> >
>> >> > Signed-off-by: Johan Hovold <johan+linaro@kernel.org>
>> >> > ---
>> >> >  .../bindings/net/wireless/pci17cb,1103.yaml   | 56 +++++++++++++++++++
>> >> >  1 file changed, 56 insertions(+)
>> >> >  create mode 100644
>> >> > Documentation/devicetree/bindings/net/wireless/pci17cb,1103.yaml
>> >> 
>> >> I'm confused (as usual), how does this differ from
>> >> bindings/net/wireless/qcom,ath11k.yaml? Why we need two .yaml files?
>> >
>> > Almost none of bindings/net/wireless/qcom,ath11k.yaml applies to WCN6856
>> > when using PCIe (e.g. as most properties are then discoverable).
>> >
>> > We could try to encode everything in one file, but that would likely
>> > just result in a big mess of a schema with conditionals all over.
>> 
>> Ah, so the current qcom,ath11k.yaml would be only for ath11k AHB devices
>> and this new file is only for ath11k PCI devices?
>
> Right, there would two separate schema files for the two device classes.
>
>> But why still the odd
>> name pci17cb,1103.yaml? It's not really descriptive and I'm for sure
>> will not remember that pci17cb,1103.yaml is for ath11k :)
>
> Yeah, it's not the best name from that perspective, but it follows the
> current convention of naming the schema files after the first compatible
> added.
>
> That said, we don't have many schemas for PCI devices so perhaps we can
> establish a new convention for those. Perhaps by replacing the numerical
> ids with what we'd use if these were platform devices (e.g.
> 'qcom,wcn6855.yaml').
>
> As long as the DT maintainers are OK with it, I'd also be happy with
> something like you suggest below:
>
> 	qcom,ath11k-ahb.yaml
> 	qcom,ath11k-pci.yaml
>
> (or simply not renaming the current file 'qcom,ath11k.yaml') but I have
> gotten push back on that in the past.

Ok, maybe it's then better not to try renaming qcom,ath11k.yaml and keep
it as is.

>> Also it doesn't look good that we have qcom,ath11k-calibration-variant
>> documented twice now. I'm no DT expert but isn't there any other way? Is
>> it possible to include other files? For example, if we would have three
>> files:
>> 
>> qcom,ath11k.yaml
>> qcom,ath11k-ahb.yaml
>> qcom,ath11k-pci.yaml
>> 
>> Then have the common properties like ath11k-calibration-variant in the
>> first file and ahb/pci files would include that.
>
> That should be possible, but it's not necessarily better as you'd then
> have to look up two files to see the bindings for either device class
> (and as far as I can tell there would not be much sharing beyond this
> single property).
>
> Note that the property could just have well have been named
> 'qcom,calibration-variant' and then it would be shared also with the
> ath10k set of devices which currently holds another definition of what
> is essentially the same property ('qcom,ath10k-calibration-variant').

Oh man, having it as 'qcom,calibration-variant' would have been so much
better. Oh well, too late now :(

Thanks for explaining all this.

-- 
https://patchwork.kernel.org/project/linux-wireless/list/

https://wireless.wiki.kernel.org/en/developers/documentation/submittingpatches

-- 
ath11k mailing list
ath11k@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/ath11k