Re: [RFC PATCH] cxl: Allow reprogramming misconfigured hdm decoders

[Alejandro Lucero Palau <alucerop@amd.com>]

On 5/7/25 07:42, Zhijian Li (Fujitsu) wrote:
>
> On 30/04/2025 16:42, Alejandro Lucero Palau wrote:
>>> Hi Gregory,
>>>
>>> Thank you for the feedback. Based on current CXL driver behavior, user-space tools
>>> can indeed reprogram firmware-provisioned HDM decoders in practice.
>>>
>>> For example, after a successful boot, one may destroy the auto-constructed region
>>> via cxl destroy-region and create a new different region.
>>> This indicates that the kernel does not inherently lock down these decoders.
>>>
>>> As for the locking rationale you mentioned, platform vendors might enforce their policies
>>> through mechanisms like the *Lock-On-Commit* in CXL HDM Decoder n Control Register
>>>
>>> While platform vendors may have valid considerations (as you noted), from a driver and
>>> end-user perspective, depending solely on firmware updates to fix transient failures
>>> is not smooth sometimes 🙂.
>>>
>> Hi Zhijan,
>>
>>
>>   From my current effort trying to get a Type2 device properly initialized by the kernel after the BIOS/platform firmware doing whatever it needs to do, I really think we should have a wider discussion regarding this sync, and maybe to have first something from the kernel expectation of what the BIOS should and should not do.
>>
>>
>> If this makes sense, I could work on a initial draft about the outline or points to discuss about this.
>>
> Hi Alejandro,
>
> Thanks for sharing this concrete pain point. Your experience highlights a critical gap in
> "defining clear handoff protocols between firmware and the kernel" for CXL device initialization.
> I agree that we need a community-driven effort to establish these expectations.
>
>
> I’m happy to see your draft or thread for deeper discussion.


Hi Zhijan,


Great. I'll work on this and hopefully something to share in a couple of 
weeks.


Thanks


>
> Best,
> Zhijian
>
>
>> Thank you