Re: [PATCH v11 07/12] cxl: Validate HDM ranges before CXL reset

[Srirangan Madhavan]

On 9/4/26 2:19 AM, Richard Cheng wrote:
> > +					"cxl_reset");
>
> How does this function provide us the required quiescing ?
> I do not see it revoking existing PFN/VFIO mapping or preventing a driver
> from issueing a new access after cpu_cache_invalidate_memregion()
>
> Want to know what prevents new CXL.mem requests between cache invalidation
> and reset completion ?
> It does not require the range to be offline or unmapped first ?
>
> Best regards,
> Richard Cheng.
Hi Richard,

You are right that request_mem_region() does not revoke PFN or VFIO 
mappings, and this series does not offline memory itself.

It temporarily reserves every enabled HDM range and rejects the reset 
with -EBUSY if a conflicting resource owns one of those ranges. I 
documented this limitation explicitly in v12.

--
Regards,
Srirangan