Re: [PATCH v14 12/16] cxl: Validate and synchronize HDM ranges around reset

[Srirangan Madhavan]

On 10/2/26 1:06 AM, Richard Cheng wrote:
> Hi Sririangan,
>
> I tested your series on some machine with CXL type 2 device, those device were
> unbound and no VFs enabled.
>
> After executing cxl_reset, all type 2 device error out with
> """
> [Errno 25] Inappropriate ioctl for device
> """
>
> Execution stops at the above CPU cache-invalidation check. Since the registered
> cache-maintenance provider list is empty. The earlier check
> "if (probe) return 0" succeeds before checking this one, which explains why
> the method is advertised.
>
> Is CPU cache-invalidation provider expected to be on standalone driver or
> something else ?
>
>
> Best regards,
> Richard Cheng.
> Hi Richard,

Thanks for testing. Yes, CPU-cache invalidation is provided separately 
through the cache-maintenance framework. Apologies for missing this on v14.

On arm64, the SMCCC provide series is here:

https://lore.kernel.org/all/20260910082142.1522865-1-smadhavan@xxxxxxxxxx/T/

It also requires matching firmware support. I've documented this 
dependency in the v15 cover letter: 
https://lore.kernel.org/linux-cxl/20261011021422.3428136-1-smadhavan@xxxxxxxxxx/

--
Regards,
Srirangan