Re: [PATCH 0/5] Add missing OPP tables for Venus on qcom/arm64

[Stephan Gerhold <stephan.gerhold@linaro.org>]

On Wed, Jun 11, 2025 at 10:46:17PM +0200, Konrad Dybcio wrote:
> On 6/11/25 10:43 PM, Konrad Dybcio wrote:
> > On 6/2/25 10:09 AM, Stephan Gerhold wrote:
> >> On Sat, May 31, 2025 at 02:27:18PM +0200, Konrad Dybcio wrote:
> >>> Sparked by <20250530-add-venus-for-qcs615-v8-0-c0092ac616d0@quicinc.com>
> >>>
> >>> No external dependencies
> >>>
> >>
> >> Are you sure?
> >>
> >>> Signed-off-by: Konrad Dybcio <konrad.dybcio@oss.qualcomm.com>
> >>> ---
> >>> Konrad Dybcio (5):
> >>>       arm64: dts: qcom: msm8916: Add Venus OPP table
> >>>       arm64: dts: qcom: msm8996: Add Venus OPP table
> >>>       arm64: dts: qcom: msm8998: Add Venus OPP table
> >>>       arm64: dts: qcom: sdm630: Add Venus OPP table
> >>
> >> None of these platforms has a power domain that supports performance
> >> states specified in the venus node of the DT, and the venus GDSC does
> >> not have any parent either. I think you will need to update the venus
> >> bindings and add
> >>
> >> 	.opp_pmdomain = (const char *[]) { "cx" /*???*/ },
> >>
> >> for all these in the venus driver (plus backwards compat if not already
> >> there). And then add that power domain additionally in the DT.
> > 
> > Making use of these tables would certainly be welcome.. This patchset
> > was aimed at pushing them to fdt, so that we can debate dropping the
> > hardcoded values in the driver in the future.
> 
> I don't think we can just plug them in to the driver right now, as
> it would also happen to break backwards compat (since
> devm_pm_domain_attach_list() seems to not be particularly happy about
> missing resources) - though arguments can be made both ways: "it
> could have never *really* worked" vs "don't poke at the old driver for
> old hardware too much, as it's gonna break"
> 

You could just make it ignore the returned error for the call to

	ret = devm_pm_domain_attach_list(dev, &opp_pd_data, &core->opp_pmdomain);

? Perhaps with some extra flag that only enables that for the older
platforms.

I suppose having the DT complete is also helpful, even without the
driver changes. But at the end this would be untested "dead" properties
that may turn out being subtly wrong when someone tries to actually
enable it. So having it properly enabled and tested would be good,
especially since the required changes don't sound too massive and
intrusive to me.

Thanks,
Stephan