On Wed, Sep 14, 2022 at 03:19:01PM +0300, Mikko Perttunen wrote: > On 9/14/22 15:08, Rob Herring wrote: > > On Tue, Sep 13, 2022 at 04:14:41PM +0300, Mikko Perttunen wrote: > > > From: Mikko Perttunen <mperttunen@xxxxxxxxxx> > > > > > > Update NVDEC bindings for Tegra234. This new engine version only has > > > two memory clients, but now requires three clocks, and as a bigger > > > change the engine loads firmware from a secure carveout configured by > > > the bootloader. > > > > > > For the latter, we need to add a phandle to the memory controller > > > to query the location of this carveout, and several other properties > > > containing offsets into the firmware inside the carveout. These > > > properties are intended to be populated through a device tree overlay > > > configured at flashing time, so that the values correspond to the > > > flashed NVDEC firmware. > > > > > > As the binding was getting large with many conditional properties, > > > also split the Tegra234 version out into a separate file. > > > > > > Signed-off-by: Mikko Perttunen <mperttunen@xxxxxxxxxx> > > > --- > > > v2: > > > - Split out into separate file to avoid complexity with > > > conditionals etc. > > > --- > > > .../gpu/host1x/nvidia,tegra234-nvdec.yaml | 154 ++++++++++++++++++ > > > 1 file changed, 154 insertions(+) > > > create mode 100644 Documentation/devicetree/bindings/gpu/host1x/nvidia,tegra234-nvdec.yaml > > > > > > diff --git a/Documentation/devicetree/bindings/gpu/host1x/nvidia,tegra234-nvdec.yaml b/Documentation/devicetree/bindings/gpu/host1x/nvidia,tegra234-nvdec.yaml > > > new file mode 100644 > > > index 000000000000..eab0475ca983 > > > --- /dev/null > > > +++ b/Documentation/devicetree/bindings/gpu/host1x/nvidia,tegra234-nvdec.yaml > > > @@ -0,0 +1,154 @@ > > > +# SPDX-License-Identifier: (GPL-2.0 OR BSD-2-Clause) > > > +%YAML 1.2 > > > +--- > > > +$id: "http://devicetree.org/schemas/gpu/host1x/nvidia,tegra234-nvdec.yaml#"; > > > +$schema: "http://devicetree.org/meta-schemas/core.yaml#"; > > > + > > > +title: Device tree binding for NVIDIA Tegra234 NVDEC > > > + > > > +description: | > > > + NVDEC is the hardware video decoder present on NVIDIA Tegra210 > > > + and newer chips. It is located on the Host1x bus and typically > > > + programmed through Host1x channels. > > > + > > > +maintainers: > > > + - Thierry Reding <treding@xxxxxxxxx> > > > + - Mikko Perttunen <mperttunen@xxxxxxxxxx> > > > + > > > +properties: > > > + $nodename: > > > + pattern: "^nvdec@[0-9a-f]*$" > > > + > > > + compatible: > > > + enum: > > > + - nvidia,tegra234-nvdec > > > + > > > + reg: > > > + maxItems: 1 > > > + > > > + clocks: > > > + maxItems: 3 > > > + > > > + clock-names: > > > + items: > > > + - const: nvdec > > > + - const: fuse > > > + - const: tsec_pka > > > + > > > + resets: > > > + maxItems: 1 > > > + > > > + reset-names: > > > + items: > > > + - const: nvdec > > > + > > > + power-domains: > > > + maxItems: 1 > > > + > > > + iommus: > > > + maxItems: 1 > > > + > > > + dma-coherent: true > > > + > > > + interconnects: > > > + items: > > > + - description: DMA read memory client > > > + - description: DMA write memory client > > > + > > > + interconnect-names: > > > + items: > > > + - const: dma-mem > > > + - const: write > > > + > > > + nvidia,memory-controller: > > > + $ref: /schemas/types.yaml#/definitions/phandle > > > + description: > > > + phandle to the memory controller for determining carveout information. > > > + > > > + nvidia,bl-manifest-offset: > > > + $ref: /schemas/types.yaml#/definitions/uint32 > > > + description: > > > + Offset to bootloader manifest from beginning of firmware. Typically set as > > > + part of a device tree overlay corresponding to flashed firmware. > > > + > > > + nvidia,bl-code-offset: > > > + $ref: /schemas/types.yaml#/definitions/uint32 > > > + description: > > > + Offset to bootloader code section from beginning of firmware. Typically set as > > > + part of a device tree overlay corresponding to flashed firmware. > > > + > > > + nvidia,bl-data-offset: > > > + $ref: /schemas/types.yaml#/definitions/uint32 > > > + description: > > > + Offset to bootloader data section from beginning of firmware. Typically set as > > > + part of a device tree overlay corresponding to flashed firmware. > > > + > > > + nvidia,os-manifest-offset: > > > + $ref: /schemas/types.yaml#/definitions/uint32 > > > + description: > > > + Offset to operating system manifest from beginning of firmware. Typically set as > > > + part of a device tree overlay corresponding to flashed firmware. > > > + > > > + nvidia,os-code-offset: > > > + $ref: /schemas/types.yaml#/definitions/uint32 > > > + description: > > > + Offset to operating system code section from beginning of firmware. Typically set as > > > + part of a device tree overlay corresponding to flashed firmware. > > > + > > > + nvidia,os-data-offset: > > > + $ref: /schemas/types.yaml#/definitions/uint32 > > > + description: > > > + Offset to operating system data section from beginning of firmware. Typically set as > > > + part of a device tree overlay corresponding to flashed firmware. > > > > I don't think DT is the place for describing your runtime loaded > > firmware layout. > > > > Rob > > The way I see it, from the kernel's point of view it's not runtime loaded > but a contract with the bootloader. Bootloader sets up hardware in a certain > way the kernel doesn't otherwise know so the bootloader needs to tell the > kernel how the hardware is set up. > > The fact that the information is supplied through an overlay is accidental > -- equivalently the bootloader that sets up the firmware could adjust the > device tree like we do in other situations, but in this case an overlay is > an easier implementation method. I think the key bit of information to know here is that the kernel is not permitted to load this firmware. It's a bootloader early in the boot process that sets this up in a secure context and then needs to convey that information to the kernel. Perhaps a slightly more idiomatic way to pass this information would be using a reserved memory node? That could span the entirety of the secure carveout (therefore removing the need to query the memory controller for that information) and be identified with a compatible string that would allow custom properties for these various offsets. Yet another way would be to have each of the bootloader and OS regions (manifest, code and data) be their own reserved memory regions. But given that this is one chunk with different regions inside makes that seem excessive. Rob, do you have any other ideas how this information could be passed to the kernel if not via DT? Thierry
Attachment:
signature.asc
Description: PGP signature
