Re: [PATCH 1/2] dt-bindings: usb: dwc3: Add system bus request info

From: Krzysztof Kozlowski
Date: Fri Feb 09 2024 - 02:32:02 EST

On 08/02/2024 22:20, Frank Li wrote:
> On Tue, Jan 30, 2024 at 05:10:54PM +0100, Krzysztof Kozlowski wrote:
>> On 30/01/2024 16:20, Frank Li wrote:
>>> On Tue, Jan 30, 2024 at 08:40:29AM +0100, Krzysztof Kozlowski wrote:
>>>> On 29/01/2024 18:41, Frank Li wrote:
>>>>> On Mon, Jan 29, 2024 at 04:49:21PM +0000, Conor Dooley wrote:
>>>>>> On Mon, Jan 29, 2024 at 10:19:24AM -0500, Frank Li wrote:
>>>>>>> On Tue, Jan 23, 2024 at 05:23:53PM -0500, Frank Li wrote:
>>>>>>>> On Tue, Jan 23, 2024 at 10:46:39PM +0100, Krzysztof Kozlowski wrote:
>>>>>>>>> On 23/01/2024 20:22, Frank Li wrote:
>>>>>>>>>> On Tue, Jan 23, 2024 at 06:42:27PM +0000, Conor Dooley wrote:
>>>>>>>>>>> On Tue, Jan 23, 2024 at 01:02:21PM -0500, Frank Li wrote:
>>>>>>>>>>>> On Tue, Jan 23, 2024 at 05:51:48PM +0000, Conor Dooley wrote:
>>>>>>>>>>>>> On Tue, Jan 23, 2024 at 12:49:27PM -0500, Frank Li wrote:
>>>>>>>>>>>>>> On Tue, Jan 23, 2024 at 05:27:13PM +0000, Conor Dooley wrote:
>>>>>>>>>>>>>>> On Tue, Jan 23, 2024 at 12:02:05PM -0500, Frank Li wrote:
>>>>>>>>>>>>>>>> Add device tree binding allow platform overwrite default value of *REQIN in
>>>>>>>>>>>>>>>> GSBUSCFG0.
>>>>>>>>>>>>>>> Why might a platform actually want to do this? Why does this need to be
>>>>>>>>>>>>>>> set at the board level and being aware of which SoC is in use is not
>>>>>>>>>>>>>>> sufficient for the driver to set the correct values?
>>>>>>>>>>>>>> In snps,dwc3.yaml, there are already similary proptery, such as
>>>>>>>>>>>>>> snps,incr-burst-type-adjustment. Use this method can keep whole dwc3 usb
>>>>>>>>>>>>>> driver keep consistent. And not all platform try enable hardware
>>>>>>>>>>>>>> dma_cohenrence. It is configable for difference platform.
>>>>>>>>>>>>> When you say "platform", what do you mean? I understand that term to
>>>>>>>>>>>>> mean a combination of board, soc and firmware.
>>>>>>>>>>>> In my company's environment, "platform" is "board". I will use "board" in
>>>>>>>>>>>> future. Is it big difference here?
>>>>>>>>>>> Nah, that's close enough that it makes no difference here.
>>>>>>>>>>> I'd still like an explanation for why a platform would need to actually
>>>>>>>>>>> set these properties though, and why information about coherency cannot
>>>>>>>>>>> be determined from whether or not the boss the usb controller is on is
>>>>>>>>>>> communicated to be dma coherent via the existing devicetree properties
>>>>>>>>>>> for that purpose.
>>>>>>>>>> Actually, I am not very clear about reason. I guest maybe treat off power
>>>>>>>>>> consumption and performance.
>>>>>>>>>> What's your judgement about proptery, which should be in dts. Such as
>>>>>>>>>> reg, clk, reset, dma and irq, which is tighted with SOC. It is the fixed
>>>>>>>>>> value for every SOC. The board dts never change these.
>>>>>>>>> Then it can be deduced from the compatible and there is no need for new
>>>>>>>>> properties.
>>>>>>>> Okay, I think "*reqinfo" match this. When new Soc(using compatible dwc usb
>>>>>>>> controller) appear regardless dma-cohorence or not, connect by AXI3 or
>>>>>>>> AXI4, needn't add new propterties.
>>>>>>> Anyone have objection? I will prepare v2 to fix rob's bot error.
>>>>>> I'm not sure what you want me to object to/not object to.
>>>>>> Your last message said "needn't add new propterties", seemingly in
>>>>>> agreement with Krzysztoff saying that it can be deduced from the
>>>>>> compatible. That seems like a good way forward for me.
>>>>> Okay, let me clear it again. dwc usb is quite common IP. The below is
>>>>> what reason why need "*reginfo* instead of using compatible string.
>>>>> 1. *reginfo* property is decscript hardware behevior, which will be changed
>>>>> at difference SOC.
>>>>> 2. it may change at board level according to if enable dma coherence.
>>>> dma coherence is not a board property. Anyway, you said it will never
>>>> change in the board.
>>> Sorry, let's correct what my previous said. There are two kinds bus in
>>> system, one is dma_coherence, the other is none dma_coherence. There are
>>> some dwc usb core ip, which is the exact the same. Some connect to
>>> dma_coherence bus, some connect to non-dma_coherence bus.
>>> So dma_coherence will be varible for this case. we need set *reginfo* for
>>> dwc usb core, which connnect to dma_coherence bus. not set "reginfo* for
>>> the dwc usb core, which connect to none dma_coherence bus.
>> OK, that makes sense. Please provide link to upstream DTS
>> (mainline/next/lore link/other upstream projects) showing this.
> Look like I still have not found the real case yet. But I think it is
> existed (maybe not upstream yet). Can I forward using this method?

No, because downstream code does things wrong and does not follow basic

Best regards,