Re: [PATCH v3 0/5] Introduce "non-pixel" sub node within iris video node

[Krzysztof Kozlowski]

On 09/10/2025 20:06, Bryan O'Donoghue wrote:
> On 09/10/2025 11:45, Krzysztof Kozlowski wrote:
>> On 09/10/2025 19:40, Vikash Garodia wrote:
>>>
>>> On 10/9/2025 2:41 PM, Krzysztof Kozlowski wrote:
>>>> On 09/10/2025 17:38, Bryan O'Donoghue wrote:
>>>>> On 09/10/2025 02:04, Krzysztof Kozlowski wrote:
>>>>>>> The iommu description for this platform basically lacks the data that
>>>>>>> _should_ be there -> FUNCTION_ID.
>>>>>> No. The index tells that already.
>>>>>
>>>>> Hmm.
>>>>>>> The rule is that the DT should really describe the hardware right ?
>>>>>> It already does. Same as I wrote on IRC, DT already has all the
>>>>>> information. Entry 0 has function ID-foo. Entry 1 has function ID-bar.
>>>>>> Entry 2 has function ID-bar or whatever.
>>>>>
>>>>> That's the part I don't believe is true its a 1:Many relationship
>>>>> between FUNCTION_ID:SIDs
>>>>>
>>>>> Let me check the docs...
>>>>>
>>>>> Here's the example I gave on IRC for lore
>>>>>
>>>>> SID 0x1940 maps to AC_VM_HLOS (Linux)
>>>>> SID 0x1941 maps to AC_VM_CP_BITSTREAM - protected bitstream
>>>>> SID 0x1945 maps to AC_WM_CP_BITSTREAM
>>>>>
>>>>
>>>> I responded to this on IRC... Nothing proves here that 1:many cannot be
>>>> done.
>>>
>>> Kaanapali already has 1:Many relationship for FUNCTION_ID:SIDs.
>>
>> Sun is a star. How is that related? I am not going to duplicate
>> arguments from IRC, especially to that pointless argument. Read again
>> discussion on IRC.
>>
>> Best regards,
>> Krzysztof
> 
> But Krzysztof is it not the case DT should be a representation of the 
> real hardware and that this takes priority over established schema.
> 
> There seems to be no other reason to keep SID in the DT and FUNCTION_ID 
> in driver meta-data except the schema already says X.
> 
> There are as I count it, 189 TCU SID mappings for Hamoa.


That could be an argument in favor of different representation, so
present that exact case - comparison of bindings and driver in two cases.

> 
> So in the extreme case that means we have an iommu = <> for each of 
> those but then need a corresponding entry in a driver to map each SID to 
> its relevant FUNCTION_ID.
> 
> And do that over and over again for each new SoC.
> 
> OTOH if we extend the iommu to include FUNCTION_ID then only the DT 
> changes - with the iommu setup code changing once to accommodate it.


Existing patch does not support this so far. Existing
patch/code/proposal clearly points to mapping that index of entry is
sufficient enough. Happy to see different code - make your case, please.


Best regards,
Krzysztof