Re: [PATCH 3/3] venus: core: add sc7180 DT compatible and resource struct

[Stanimir Varbanov <stanimir.varbanov@linaro.org>]

Hi Dikshita,

On 12/23/19 12:04 PM, dikshita@codeaurora.org wrote:
> Hi Stan,
> 
> Thanks for the review!
> 
> On 2019-12-20 15:03, Stanimir Varbanov wrote:
>> Hi Dikshita,
>>
>> Thanks for the patch!
>>
>> On 12/20/19 9:59 AM, Dikshita Agarwal wrote:
>>> This add DT compatible string and resource structure for sc7180.
>>>
>>> Signed-off-by: Dikshita Agarwal <dikshita@codeaurora.org>
>>> ---
>>>  drivers/media/platform/qcom/venus/core.c | 58
>>> +++++++++++++++++++++++++++++++-
>>>  1 file changed, 57 insertions(+), 1 deletion(-)
>>>
>>> diff --git a/drivers/media/platform/qcom/venus/core.c
>>> b/drivers/media/platform/qcom/venus/core.c
>>> index c7525d9..e8c8b28 100644
>>> --- a/drivers/media/platform/qcom/venus/core.c
>>> +++ b/drivers/media/platform/qcom/venus/core.c
>>> @@ -469,7 +469,7 @@ static __maybe_unused int
>>> venus_runtime_resume(struct device *dev)
>>>      { 2073600, 3929000, 0, 5551000, 0 },    /* 4096x2160@60 */
>>>      { 1036800, 1987000, 0, 2797000, 0 },    /* 4096x2160@30 */
>>>      {  489600, 1040000, 0, 1298000, 0 },    /* 1920x1080@60 */
>>> -    {  244800,  530000, 0,  659000, 0 },    /* 1920x1080@30 */
>>> +    {  244800,  442000, 0,  659000, 0 },    /* 1920x1080@30 */
>>
>> unrelated change, please drop it
> Sure, I will address this in next version.
>>
>>>  };
>>>
>>>  static const struct venus_resources sdm845_res = {
>>> @@ -521,11 +521,67 @@ static __maybe_unused int
>>> venus_runtime_resume(struct device *dev)
>>>      .fwname = "qcom/venus-5.2/venus.mdt",
>>>  };
>>>
>>> +static const struct freq_tbl sc7180_freq_table[] = {
>>> +    {  0, 380000000 },
>>> +    {  0, 340000000 },
>>> +    {  0, 270000000 },
>>> +    {  0, 150000000 },
>>
>> why .load is zero?
> .load is not being used any longer to calculate load and is a dummy value.
> So keeping it 0.

Hmm, ok I forgot about that fact. I suppose it is fine then.

I found some other issue. Looking into [1], for sc7180 we have two more
clock-controller frequencies, could you add them too in the table. And
last, in the same patch we have 19.2MHz do you think such frequency
makes any sense?

[1] https://lkml.org/lkml/2019/11/15/361
> 
>>
>>> +};
>>> +
>>> +static struct codec_freq_data sc7180_codec_freq_data[] =  {
>>> +    { V4L2_PIX_FMT_H264, VIDC_SESSION_TYPE_ENC, 675, 10 },
>>> +    { V4L2_PIX_FMT_HEVC, VIDC_SESSION_TYPE_ENC, 675, 10 },
>>> +    { V4L2_PIX_FMT_VP8, VIDC_SESSION_TYPE_ENC, 675, 10 },
>>> +    { V4L2_PIX_FMT_MPEG2, VIDC_SESSION_TYPE_DEC, 200, 10 },
>>> +    { V4L2_PIX_FMT_H264, VIDC_SESSION_TYPE_DEC, 200, 10 },
>>> +    { V4L2_PIX_FMT_HEVC, VIDC_SESSION_TYPE_DEC, 200, 10 },
>>> +    { V4L2_PIX_FMT_VP8, VIDC_SESSION_TYPE_DEC, 200, 10 },
>>> +    { V4L2_PIX_FMT_VP9, VIDC_SESSION_TYPE_DEC, 200, 10 },
>>> +};
>>
>> the table is exactly the same as sdm845 one, please reuse it.
> Sure, I will address this in next version.
>>
>>> +
>>> +static const struct bw_tbl sc7180_bw_table_enc[] = {
>>> +    {  972000,  750000, 0, 0, 0 },    /* 3840x2160@30 */
>>> +    {  489600,  451000, 0, 0, 0 },    /* 1920x1080@60 */
>>> +    {  244800,  234000, 0, 0, 0 },    /* 1920x1080@30 */
>>> +};
>>> +
>>> +static const struct bw_tbl sc7180_bw_table_dec[] = {
>>> +    { 1036800, 1386000, 0, 1875000, 0 },    /* 4096x2160@30 */
>>> +    {  489600,  865000, 0, 1146000, 0 },    /* 1920x1080@60 */
>>> +    {  244800,  530000, 0,  583000, 0 },    /* 1920x1080@30 */
>>> +};
>>> +
>>> +static const struct venus_resources sc7180_res = {
>>> +    .freq_tbl = sc7180_freq_table,
>>> +    .freq_tbl_size = ARRAY_SIZE(sc7180_freq_table),
>>> +    .bw_tbl_enc = sc7180_bw_table_enc,
>>> +    .bw_tbl_enc_size = ARRAY_SIZE(sc7180_bw_table_enc),
>>> +    .bw_tbl_dec = sc7180_bw_table_dec,
>>> +    .bw_tbl_dec_size = ARRAY_SIZE(sc7180_bw_table_dec),
>>> +    .codec_freq_data = sc7180_codec_freq_data,
>>> +    .codec_freq_data_size = ARRAY_SIZE(sc7180_codec_freq_data),
>>> +    .clks = {"core", "iface", "bus" },
>>> +    .clks_num = 3,
>>> +    .vcodec0_clks = { "vcodec0_core", "vcodec0_bus" },
>>> +    .vcodec_clks_num = 2,
>>> +    .vcodec_pmdomains = { "venus", "vcodec0" },
>>> +    .vcodec_pmdomains_num = 2,
>>> +    .vcodec_num = 1,
>>> +    .max_load = 3110400,    /* 4096x2160@90 */
>>
>> Looking into above bandwidth tables I can guess that the maximimum load
>> is reached at 4096x2160@30? If so you have to change it here.
> 
> After checking further on this I see that max_load can be removed since
> it is not being used now to determine if H/W is overloaded or not.
> What do you suggest?

Lets have it just for informational reasons.

-- 
regards,
Stan