Re: [PATCH v3 1/3] dt-bindings: usb: mtk-xhci: add a property for Gen1 isoc-in transfer issue

[AngeloGioacchino Del Regno <angelogioacchino.delregno@collabora.com>]

Il 20/12/23 08:38, Krzysztof Kozlowski ha scritto:
> On 20/12/2023 03:58, Chunfeng Yun wrote:
>> For Gen1 isoc-in endpoint on controller before about SSUSB IPM v1.6.0, it
>> still send out unexpected ACK after receiving a short packet in burst
>> transfer, this will cause an exception on connected device, specially for
>> a 4k camera.
>> Add a quirk property "rx-fifo-depth" to work around this hardware issue,
>> prefer to use 3k bytes;
>> The side-effect is that may cause performance drop about 10%, including
>> bulk transfer.
>>
>> Signed-off-by: Chunfeng Yun <chunfeng.yun@mediatek.com>
>> ---
>> v3: add fifo depth unit, change the value range from 0-3 to 1-4
>> v2: change 'mediatek,rxfifo-depth' to 'rx-fifo-depth'
>> ---
>>   .../devicetree/bindings/usb/mediatek,mtk-xhci.yaml   | 12 ++++++++++++
>>   1 file changed, 12 insertions(+)
>>
>> diff --git a/Documentation/devicetree/bindings/usb/mediatek,mtk-xhci.yaml b/Documentation/devicetree/bindings/usb/mediatek,mtk-xhci.yaml
>> index e9644e333d78..9478b7031796 100644
>> --- a/Documentation/devicetree/bindings/usb/mediatek,mtk-xhci.yaml
>> +++ b/Documentation/devicetree/bindings/usb/mediatek,mtk-xhci.yaml
>> @@ -124,6 +124,18 @@ properties:
>>         defined in the xHCI spec on MTK's controller.
>>       default: 5000
>>   
>> +  rx-fifo-depth:
>> +    $ref: /schemas/types.yaml#/definitions/uint32
>> +    description:
>> +      It is a quirk used to work around Gen1 isoc-in endpoint transfer issue
>> +      that still send out unexpected ACK after device finish the burst transfer
>> +      with a short packet and cause an exception, specially on a 4K camera
>> +      device, it happens on controller before about IPM v1.6.0; the side-effect
>> +      is that may cause performance drop about 10%, include bulk transfer,
>> +      prefer to use 3 here. The unit is 1K bytes.
> 
> NAK. Read comments on previous submission.
> 
> Best regards,
> Krzysztof
> 

Chunfeng, I think the discussion was not clear for you, so I will try to give
you a different explanation: this should be expressed in bytes, so 1000, or 1024,
2048, 4096, etc, and not 1/2/3/4/5/n.

The driver shall then validate and map your bytes number to hardware register
value and subsequently write to the registers.

Cheers,
Angleo