Re: [PATCH v3 1/6] dt-bindings: crypto: Document support for SPAcc

[Krzysztof Kozlowski <krzk@kernel.org>]

On 06/06/2025 14:58, Pavitrakumar Managutte wrote:
> On Fri, Jun 6, 2025 at 4:55 PM Krzysztof Kozlowski <krzk@kernel.org> wrote:
>>
>> On 06/06/2025 13:02, Pavitrakumar Managutte wrote:
>>> Hi Krzysztof,
>>>   Appreciate your inputs and feedback. My comments are embedded below.
>>>
>>> Warm regards,
>>> PK
>>>
>>> On Wed, Jun 4, 2025 at 7:37 PM Krzysztof Kozlowski <krzk@kernel.org> wrote:
>>>>
>>>> On 04/06/2025 14:20, Pavitrakumar Managutte wrote:
>>>>>>
>>>>>>>>> +
>>>>>>>>> +  snps,vspacc-id:
>>>>>>>>> +    $ref: /schemas/types.yaml#/definitions/uint32
>>>>>>>>> +    description: |
>>>>>>>>> +      Virtual SPAcc instance identifier.
>>>>>>>>> +      The SPAcc hardware supports multiple virtual instances (determined by
>>>>>>>>> +      ELP_SPACC_CONFIG_VSPACC_CNT parameter), and this ID is used to identify
>>>>>>>>> +      which virtual instance this node represents.
>>>>>>>>
>>>>>>>> No, IDs are not accepted.
>>>>>>>
>>>>>>> PK: This represents the specific virtual SPAcc that is being used in
>>>>>>> the current configuration. It is used to index into the register banks
>>>>>>> and the context memories of the virtual SPAcc that is being used. The
>>>>>>> SPAcc IP can be configured as dedicated virtual SPAccs in
>>>>>>> heterogeneous environments.
>>>>>>
>>>>>> OK. Why registers are not narrowed to only this instance? It feels like
>>>>>> you provide here full register space for multiple devices and then
>>>>>> select the bank with above ID.
>>>>>
>>>>> PK: No, we cant narrow the registers to only this instance since its
>>>>> is just a single SPAcc with multiple virtual SPAcc instances. The same
>>>>> set of registers(aka register banks) and context memories are
>>>>> repeated, but sit at different offset addresses (i*4000 +
>>>>> register-offsets). The crypto hardware engine inside is shared by all
>>>>> the virtual SPAccs. This is very much for a heterogeneous computing
>>>>> scenario.
>>>>
>>>> Then maybe you have one crypto engine? You ask us to guess all of this,
>>>> also because you do not upstream the DTS for real product. Any
>>>> mentioning of "virtual" already raises concerns...
>>>
>>> PK: Yes this is a single crypto engine, maybe I should have detailed
>>> that in the cover letter. I will fix that. And what I have pushed in
>>
>> So one node, thus no need for this entire virtual device split.
> 
> PK: Agreed, its one node for our test case.

We do not talk about test case. We talk about this device.
> 
>>
>>> the patch is my complete DTS. It might need updating depending on the
>>
>> If this is complete, then obviously "snps,vspacc-id" is not necessary.
> 
> PK: Yes, its one node, to keep things simple. So we pick a virtual
> spacc with its vspacc-id for testing. That way we could test all the
> virtual spaccs with a single node, on a need basis.
> 
> On the other hand we could create 'n' nodes for 'n' virtual spaccs and

You said it is complete, now you said you have 'n' more.

> register 'n' vspacc devices with the crypto subsystem. And bind the
> individual nodes with unique vspacc-ids. That might depend on the

I don't understand what is "binding" here. Use Linux or DT terminology.

> vendor use case, for which we will add incremental support.

You did not get the point but you keep saying "yes". This discussion is
getting meaningless and you really do not want to listen. You have
either incomplete picture here or you have only one node. In both cases
virtual ID is not necessary. If you claim virtual ID is necessary, I
claim you have here incomplete picture and you are trying to represent
one device in multiple nodes. No.

Typically one device, one node.

NOT one device and 10 virtual nodes representing virtual devices.

Amount of ping pongs here is way beyond my patience, so before you
respond read that carefully and come with full and accurate hardware
description, so we will not have to ping pong trying to get any sort of
details.

Best regards,
Krzysztof