Re: [PATCH v6 2/9] doc: DT: venus: binding document for Qualcomm video driver

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 





On 02/22/2017 04:17 PM, Rob Herring wrote:
> On Wed, Feb 22, 2017 at 3:25 AM, Stanimir Varbanov
> <stanimir.varbanov@xxxxxxxxxx> wrote:
>> Hi Rob,
>>
>> On 02/22/2017 02:09 AM, Rob Herring wrote:
>>> On Tue, Feb 07, 2017 at 03:10:17PM +0200, Stanimir Varbanov wrote:
>>>> Add binding document for Venus video encoder/decoder driver
>>>>
>>>> Cc: Rob Herring <robh+dt@xxxxxxxxxx>
>>>> Cc: devicetree@xxxxxxxxxxxxxxx
>>>> Signed-off-by: Stanimir Varbanov <stanimir.varbanov@xxxxxxxxxx>
>>>> ---
>>>> Changes since previous v5:
>>>>  * dropped rproc phandle (remoteproc is not used anymore)
>>>>  * added subnodes paragraph with descrition of three subnodes:
>>>>     - video-decoder and video-encoder - describes decoder (core0) and
>>>>     encoder (core1) power-domains and clocks (applicable for msm8996
>>>>     Venus core).
>>>>     - video-firmware - needed to get reserved memory region where the
>>>>     firmware is stored.
>>>>
>>>>  .../devicetree/bindings/media/qcom,venus.txt       | 112 +++++++++++++++++++++
>>>>  1 file changed, 112 insertions(+)
>>>>  create mode 100644 Documentation/devicetree/bindings/media/qcom,venus.txt
>>>>
>>>> diff --git a/Documentation/devicetree/bindings/media/qcom,venus.txt b/Documentation/devicetree/bindings/media/qcom,venus.txt
>>>> new file mode 100644
>>>> index 000000000000..4427af3ca5a5
>>>> --- /dev/null
>>>> +++ b/Documentation/devicetree/bindings/media/qcom,venus.txt
>>>> @@ -0,0 +1,112 @@
>>>
>>> [...]
>>>
>>>> +* Subnodes
>>>> +The Venus node must contain three subnodes representing video-decoder,
>>>> +video-encoder and video-firmware.
>>>
>>> [...]
>>>
>>>> +The video-firmware subnode should contain:
>>>> +
>>>> +- memory-region:
>>>> +    Usage: required
>>>> +    Value type: <phandle>
>>>> +    Definition: reference to the reserved-memory for the memory region
>>>> +
>>>> +* An Example
>>>> +    video-codec@1d00000 {
>>>> +            compatible = "qcom,msm8916-venus";
>>>> +            reg = <0x01d00000 0xff000>;
>>>> +            interrupts = <GIC_SPI 44 IRQ_TYPE_LEVEL_HIGH>;
>>>> +            clocks = <&gcc GCC_VENUS0_VCODEC0_CLK>,
>>>> +                     <&gcc GCC_VENUS0_AHB_CLK>,
>>>> +                     <&gcc GCC_VENUS0_AXI_CLK>;
>>>> +            clock-names = "core", "iface", "bus";
>>>> +            power-domains = <&gcc VENUS_GDSC>;
>>>> +            iommus = <&apps_iommu 5>;
>>>> +
>>>> +            video-decoder {
>>>> +                    compatible = "venus-decoder";
>>>> +                    clocks = <&mmcc VIDEO_SUBCORE0_CLK>;
>>>> +                    clock-names = "core";
>>>> +                    power-domains = <&mmcc VENUS_CORE0_GDSC>;
>>>> +            };
>>>> +
>>>> +            video-encoder {
>>>> +                    compatible = "venus-encoder";
>>>> +                    clocks = <&mmcc VIDEO_SUBCORE1_CLK>;
>>>> +                    clock-names = "core";
>>>> +                    power-domains = <&mmcc VENUS_CORE1_GDSC>;
>>>> +            };
>>>> +
>>>> +            video-firmware {
>>>> +                    memory-region = <&venus_mem>;
>>>
>>> Why does this need to be a sub node?
>>
>> Because firmware reserved memory region must have separate struct
>> device, otherwise allocating video buffers (and map them through iommu)
>> for video-codec will fail because dma_alloc_coherent trying to allocate
>> from per-device coherent area.
> 
> Why can't the struct device be the video-codec device? Looking at the
> code, I don't see why you need the 2nd struct device.

The issue when using one device (video-codec device) is coming from
dma_alloc_from_coherent() which is called by dma_alloc_attrs(). The
firmware memory-region is parsed and initialize coherent memory such a
region with flag DMA_MEMORY_EXCLUSIVE which means that following
invocations of dma_alloc_coherent (for example to allocate video buffers
and map them through IOMMU) will no fallback to generic memory allocator
i.e. dma_ops. Because Venus is behind IOMMU the dma_ops are pointing to
iommu_dma_ops for ARM64.

> 
> In any case, this is letting the driver design the binding which is
> wrong. From a binding perspective, there's no reason to have this
> node.

OK, thanks for the comments. I have few changes in the firmware memory
allocation part of the driver which makes possible to remove
video-firmware subnode and move memory-region to video-codec DT node.

Is that fine for you?

-- 
regards,
Stan
--
To unsubscribe from this list: send the line "unsubscribe devicetree" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html



[Index of Archives]     [Device Tree Compilter]     [Device Tree Spec]     [Linux Driver Backports]     [Video for Linux]     [Linux USB Devel]     [Linux PCI Devel]     [Linux Audio Users]     [Linux Kernel]     [Linux SCSI]     [XFree86]     [Yosemite Backpacking]
  Powered by Linux