Re: [RFC WIP PATCH] venus: add qcom,no-low-power property

[Vikash Garodia <quic_vgarodia@quicinc.com>]

On 2/26/2024 9:25 PM, Marc Gonzalez wrote:
> On 26/02/2024 15:30, Vikash Garodia wrote:
> 
>> On 2/26/2024 6:39 PM, Marc Gonzalez wrote:
>>
>>> On 23/02/2024 14:48, Vikash Garodia wrote:
>>>
>>>> On 2/20/2024 8:15 PM, Marc Gonzalez wrote:
>>>>
>>>>> On 20/02/2024 14:53, Vikash Garodia wrote:
>>>>>
>>>>>> msm8998 supports configuring the VCodec (venus core0) GDSC in HW power control
>>>>>> mode. Could you please check and confirm if the driver is configuring only the
>>>>>> VCodec GDSC and not the venus GDSC. Look for the attribute
>>>>>> "qcom,support-hw-trigger" in vendor dt file.
>>>>>
>>>>> [ Vendor DTS for easy reference: ]
>>>>> [ https://git.codelinaro.org/clo/la/kernel/msm-4.4/-/blob/caf_migration/kernel.lnx.4.4.r38-rel/arch/arm/boot/dts/qcom/msm8998.dtsi ]
>>>>>
>>>>> In the queue, we have a patch enabling the Venus Decoder (VDEC) in mainline.
>>>>> (It is using the previously proposed "qcom,no-low-power" mechanism, but that
>>>>> might not be necessary, if I understand correctly?)
>>>>>
>>>>>
>>>>> diff --git a/arch/arm64/boot/dts/qcom/msm8998.dtsi b/arch/arm64/boot/dts/qcom/msm8998.dtsi
>>>>> index 2793cc22d381a..5084191be1446 100644
>>>>> --- a/arch/arm64/boot/dts/qcom/msm8998.dtsi
>>>>> +++ b/arch/arm64/boot/dts/qcom/msm8998.dtsi
>>>>> @@ -3000,6 +3000,56 @@ mdss_dsi1_phy: phy@c996400 {
>>>>>  			};
>>>>>  		};
>>>>>  
>>>>> +		venus: video-codec@cc00000 {
>>>>> +			compatible = "qcom,msm8998-venus";
>>>>> +			reg = <0x0cc00000 0xff000>;
>>>>> +			interrupts = <GIC_SPI 287 IRQ_TYPE_LEVEL_HIGH>;
>>>>> +			power-domains = <&mmcc VIDEO_TOP_GDSC>;
>>>>> +			clocks = <&mmcc VIDEO_CORE_CLK>,
>>>>> +				 <&mmcc VIDEO_AHB_CLK>,
>>>>> +				 <&mmcc VIDEO_AXI_CLK>,
>>>>> +				 <&mmcc VIDEO_MAXI_CLK>;
>>>>> +			clock-names = "core", "iface", "bus", "mbus";
>>>>> +			iommus = <&mmss_smmu 0x400>,
>>>>> +				 <&mmss_smmu 0x401>,
>>>>> +				 <&mmss_smmu 0x40a>,
>>>>> +				 <&mmss_smmu 0x407>,
>>>>> +				 <&mmss_smmu 0x40e>,
>>>>> +				 <&mmss_smmu 0x40f>,
>>>>> +				 <&mmss_smmu 0x408>,
>>>>> +				 <&mmss_smmu 0x409>,
>>>>> +				 <&mmss_smmu 0x40b>,
>>>>> +				 <&mmss_smmu 0x40c>,
>>>>> +				 <&mmss_smmu 0x40d>,
>>>>> +				 <&mmss_smmu 0x410>,
>>>>> +				 <&mmss_smmu 0x411>,
>>>>> +				 <&mmss_smmu 0x421>,
>>>>> +				 <&mmss_smmu 0x428>,
>>>>> +				 <&mmss_smmu 0x429>,
>>>>> +				 <&mmss_smmu 0x42b>,
>>>>> +				 <&mmss_smmu 0x42c>,
>>>>> +				 <&mmss_smmu 0x42d>,
>>>>> +				 <&mmss_smmu 0x411>,
>>>>> +				 <&mmss_smmu 0x431>;
>>>>> +			memory-region = <&venus_mem>;
>>>>> +			status = "disabled";
>>>>> +			qcom,no-low-power; /*** WORK AROUND LOW-POWER ISSUE ***/
>>>>> +
>>>>> +			video-decoder {
>>>>> +				compatible = "venus-decoder";
>>>>> +				clocks = <&mmcc VIDEO_SUBCORE0_CLK>;
>>>>> +				clock-names = "core";
>>>>> +				power-domains = <&mmcc VIDEO_SUBCORE0_GDSC>;
>>>>> +			};
>>>>> +
>>>>> +			video-encoder {
>>>>> +				compatible = "venus-encoder";
>>>>> +				clocks = <&mmcc VIDEO_SUBCORE1_CLK>;
>>>>> +				clock-names = "core";
>>>>> +				power-domains = <&mmcc VIDEO_SUBCORE1_GDSC>;
>>>>> +			};
>>>>> +		};
>>>>> +
>>>>>  		mmss_smmu: iommu@cd00000 {
>>>>>  			compatible = "qcom,msm8998-smmu-v2", "qcom,smmu-v2";
>>>>>  			reg = <0x0cd00000 0x40000>;
>>>>> diff --git a/drivers/media/platform/qcom/venus/core.c b/drivers/media/platform/qcom/venus/core.c
>>>>> index a712dd4f02a5b..ad1705e510312 100644
>>>>> --- a/drivers/media/platform/qcom/venus/core.c
>>>>> +++ b/drivers/media/platform/qcom/venus/core.c
>>>>> @@ -585,6 +585,43 @@ static const struct venus_resources msm8996_res = {
>>>>>  	.fwname = "qcom/venus-4.2/venus.mbn",
>>>>>  };
>>>>>  
>>>>> +static const struct freq_tbl msm8998_freq_table[] = {
>>>>> +	{ 1944000, 520000000 },	/* 4k UHD @ 60 (decode only) */
>>>>> +	{  972000, 520000000 },	/* 4k UHD @ 30 */
>>>>> +	{  489600, 346666667 },	/* 1080p @ 60 */
>>>>> +	{  244800, 150000000 },	/* 1080p @ 30 */
>>>>> +	{  108000,  75000000 },	/* 720p @ 30 */
>>>>> +};
>>>>> +
>>>>> +static const struct reg_val msm8998_reg_preset[] = {
>>>>> +    { 0x80124, 0x00000003 },
>>>>> +    { 0x80550, 0x01111111 },
>>>>> +    { 0x80560, 0x01111111 },
>>>>> +    { 0x80568, 0x01111111 },
>>>>> +    { 0x80570, 0x01111111 },
>>>>> +    { 0x80580, 0x01111111 },
>>>>> +    { 0xe2010, 0x00000000 },
>>>>> +};
>>>>> +
>>>>> +static const struct venus_resources msm8998_res = {
>>>>> +	.freq_tbl = msm8998_freq_table,
>>>>> +	.freq_tbl_size = ARRAY_SIZE(msm8998_freq_table),
>>>>> +	.reg_tbl = msm8998_reg_preset,
>>>>> +	.reg_tbl_size = ARRAY_SIZE(msm8998_reg_preset),
>>>>> +	.clks = {"core", "iface", "bus", "mbus"},
>>>>> +	.clks_num = 4,
>>>>> +	.vcodec0_clks = { "core" },
>>>>> +	.vcodec1_clks = { "core" },
>>>>> +	.vcodec_clks_num = 1,
>>>>> +	.max_load = 2563200,
>>>>> +	.hfi_version = HFI_VERSION_3XX,
>>>>> +	.vmem_id = VIDC_RESOURCE_NONE,
>>>>> +	.vmem_size = 0,
>>>>> +	.vmem_addr = 0,
>>>>> +	.dma_mask = 0xddc00000 - 1,
>>>>> +	.fwname = "qcom/venus-4.4/venus.mbn",
>>>>> +};
>>>>> +
>>>>>  static const struct freq_tbl sdm660_freq_table[] = {
>>>>>  	{ 979200, 518400000 },
>>>>>  	{ 489600, 441600000 },
>>>>> @@ -891,6 +928,7 @@ static const struct venus_resources sc7280_res = {
>>>>>  static const struct of_device_id venus_dt_match[] = {
>>>>>  	{ .compatible = "qcom,msm8916-venus", .data = &msm8916_res, },
>>>>>  	{ .compatible = "qcom,msm8996-venus", .data = &msm8996_res, },
>>>>> +	{ .compatible = "qcom,msm8998-venus", .data = &msm8998_res, },
>>>>>  	{ .compatible = "qcom,sdm660-venus", .data = &sdm660_res, },
>>>>>  	{ .compatible = "qcom,sdm845-venus", .data = &sdm845_res, },
>>>>>  	{ .compatible = "qcom,sdm845-venus-v2", .data = &sdm845_res_v2, },
>>>>>
>>>>>
>>>>>
>>>>> @Vikash, are you saying that perhaps the DTS for video-codec@cc00000
>>>>> needs to be written slightly differently?
>>>>
>>>>
>>>> Certainly yes. For ex, in the clock list, i do not see the core clocks listed
>>>> i.e clk_mmss_video_subcore0_clk and clk_mmss_video_subcore1_clk. You can refer
>>>> the downstream video DT node [1] and then align it as per venus driver
>>>> [1]
>>>> https://git.codelinaro.org/clo/la/kernel/msm-4.4/-/blob/caf_migration/kernel.lnx.4.4.r38-rel/arch/arm/boot/dts/qcom/msm8998-vidc.dtsi
>>>
>>> If I understand correctly (which is far from certain),
>>> we should base the "qcom,msm8998-venus" DT node on
>>> "qcom,sdm845-venus-v2" rather than "qcom,msm8996-venus" ?
>>
>> That's correct, but that's just another way to do the configuration. With the
>> existing node, is video decode as well as encode working ?
> 
> Errr, there is currently no existing node for msm8998-venus?
My bad, i meant your initial node msm8998-venus, without migrating to v2.
> 
> With the proposed node above (based on msm8996-venus)
> AND the proposed work-around disabling low-power mode,
> decoding works correctly.
Nice, lets fix the work-around part before migrating to v2. Could you share the
configurations for VIDEO_SUBCORE0_GDSC and VIDEO_SUBCORE1_GDSC ?

If we see vendor code[1], sys power plane control is very much supported, so
ideally we should get it working without the workaround
[1]
https://git.codelinaro.org/clo/la/kernel/msm-4.4/-/blob/caf_migration/kernel.lnx.4.4.r38-rel/drivers/media/platform/msm/vidc/venus_hfi.c#L2223

> Encoding does not work, but it has never been used/tested on our device.
> 
> [h264_v4l2m2m @ 0xaaaaec9c44a0] Using device /dev/video1
> [h264_v4l2m2m @ 0xaaaaec9c44a0] driver 'qcom-venus' on card 'Qualcomm Venus video encoder' in mplane mode
> [h264_v4l2m2m @ 0xaaaaec9c44a0] requesting formats: output=NV12/yuv420p capture=H264/none
> [h264_v4l2m2m @ 0xaaaaec9c44a0] output VIDIOC_REQBUFS failed: Invalid argument
> [h264_v4l2m2m @ 0xaaaaec9c44a0] no v4l2 output context's buffers
> [h264_v4l2m2m @ 0xaaaaec9c44a0] can't configure encoder
> [vost#0:0/h264_v4l2m2m @ 0xaaaaec9c4160] Error while opening encoder - maybe incorrect parameters such as bit_rate, rate, width or height.

We can revisit the encoder failure once we have decode fixed without any workaround.

Regards,
Vikash