Re: [PATCH v7 1/7] media: dt-bindings: venus: Add qcm2290 dt schema

[Jorge Ramirez <jorge.ramirez@oss.qualcomm.com>]

On 17/07/25 19:00:22, Jorge Ramirez wrote:
> On 17/07/25 13:16:31, Jorge Ramirez wrote:
> > On 17/07/25 08:45:17, Krzysztof Kozlowski wrote:
> > > On 17/07/2025 08:35, Jorge Ramirez wrote:
> > > > On 17/07/25 00:22:53, Bryan O'Donoghue wrote:
> > > >> On 15/07/2025 21:47, Jorge Ramirez-Ortiz wrote:
> > > >>> Add a schema for the venus video encoder/decoder on the qcm2290.
> > > >>>
> > > >>> Signed-off-by: Jorge Ramirez-Ortiz <jorge.ramirez@oss.qualcomm.com>
> > > >>> Reviewed-by: Krzysztof Kozlowski <krzysztof.kozlowski@linaro.org>
> > > >>> Reviewed-by: Bryan O'Donoghue <bryan.odonoghue@linaro.org>
> > > >>> ---
> > > >>>   .../bindings/media/qcom,qcm2290-venus.yaml    | 127 ++++++++++++++++++
> > > >>>   1 file changed, 127 insertions(+)
> > > >>>   create mode 100644 Documentation/devicetree/bindings/media/qcom,qcm2290-venus.yaml
> > > >>>
> > > >>> diff --git a/Documentation/devicetree/bindings/media/qcom,qcm2290-venus.yaml b/Documentation/devicetree/bindings/media/qcom,qcm2290-venus.yaml
> > > >>> new file mode 100644
> > > >>> index 000000000000..0371f8dd91a3
> > > >>> --- /dev/null
> > > >>> +++ b/Documentation/devicetree/bindings/media/qcom,qcm2290-venus.yaml
> > > >>> @@ -0,0 +1,127 @@
> > > >>> +# SPDX-License-Identifier: (GPL-2.0 OR BSD-2-Clause)
> > > >>> +%YAML 1.2
> > > >>> +---
> > > >>> +$id: http://devicetree.org/schemas/media/qcom,qcm2290-venus.yaml#
> > > >>> +$schema: http://devicetree.org/meta-schemas/core.yaml#
> > > >>> +
> > > >>> +title: Qualcomm QCM2290 Venus video encode and decode accelerators
> > > >>> +
> > > >>> +maintainers:
> > > >>> +  - Vikash Garodia <quic_vgarodia@quicinc.com>
> > > >>
> > > >> Shouldn't you be on this list ? If you upstream a file I think you should
> > > >> list yourself as responsible for its glory or its mess.
> > > > 
> > > > happy to do it. The MAINTAINER's file covered all the files named
> > > 
> > > This should be the person(s) interested and caring about this hardware,
> > > which means:
> > > 1. Subsystem maintainers: no
> > > 2. Driver maintainers: usually yes
> > > 3. Author(s) of new hardware support: usually yes
> > 
> > perfect, will do 
> > 
> > > 
> > > > schemas/media/*venus* so my understanding was that I shouldn't.
> > > 
> > > I cannot comment why people decided to go one way or another in other
> > > code, but it as well could be just incorrect choice thinking only people
> > > in MAINTAINERS care about hardware.
> > > 
> > > ...
> > > 
> > > >>> +
> > > >>> +        memory-region = <&pil_video_mem>;
> > > >>> +        iommus = <&apps_smmu 0x860 0x0>,
> > > >>> +                 <&apps_smmu 0x880 0x0>,
> > > >>> +                 <&apps_smmu 0x861 0x04>,
> > > >>> +                 <&apps_smmu 0x863 0x0>,
> > > >>> +                 <&apps_smmu 0x804 0xe0>;
> > > >>
> > > >> You're listing five iommus.
> > > >>
> > > >> I understand there's some disagreement about whether or not to list all of
> > > >> the potential use-cases but, TBH I don't think those are good arguments.
> > > >>
> > > >> Unless there's some technical prohibition I can't think of listing all five
> > > >> maxItems:5 .. let's just do that.
> > > > 
> > > > since the device tree should describe hardware and not policy, and the
> > > > driver seems to be able to ignore the unused SIDs I think this is the
> > > > right thing to do.
> > > 
> > > 
> > > It was never about the driver but about whether you should describe in
> > > DTS for non-secure world the entries which are secure world. The answer
> > > in general is that you can and there will be benefits (e.g. sharing DTS
> > > with secure world implementations).
> > 
> > all right, sounds good then, thanks
> 
> Not sure if I’ve shared this before, but following an internal
> discussion, I think it’s worth highlighting a functional dependency in
> the current kernel:
> 
>  - the driver only works if the first two IOMMUs in the list — the
> non-secure ones — are placed at the beginning. Reordering them breaks
> functionality, which introduces unexpected fragility.

by "the driver only works" I mean the firmware will fail to load
otherwise (should have been clear about the failure mode).

> 
> Regardless, this seems like a valid concern to me — a driver shouldn't
> rely on the order of phandles — and I just wanted to make sure you're
> aware of it before I post a v8 (likely sometime next week or the
> following, as I’ll be taking a short break soon).
> 
> Do you consider this serious enough to be called out in the commit
> message, or is this kind of behavior accepted as-is - ie, do you know if
> the DT binding for iommus rely on the order?
>