Re: [PATCH v13 21/48] arm64: RMI: Handle RMI_EXIT_RIPAS_CHANGE

[Steven Price]

On 20/03/2026 11:15, Suzuki K Poulose wrote:
> Hi Steven
> 
> On 18/03/2026 15:53, Steven Price wrote:
>> The guest can request that a region of it's protected address space is
>> switched between RIPAS_RAM and RIPAS_EMPTY (and back) using
>> RSI_IPA_STATE_SET. This causes a guest exit with the
>> RMI_EXIT_RIPAS_CHANGE code. We treat this as a request to convert a
>> protected region to unprotected (or back), exiting to the VMM to make
>> the necessary changes to the guest_memfd and memslot mappings. On the
>> next entry the RIPAS changes are committed by making RMI_RTT_SET_RIPAS
>> calls.
>>
>> The VMM may wish to reject the RIPAS change requested by the guest. For
>> now it can only do this by no longer scheduling the VCPU as we don't
>> currently have a usecase for returning that rejection to the guest, but
>> by postponing the RMI_RTT_SET_RIPAS changes to entry we leave the door
>> open for adding a new ioctl in the future for this purpose.
> 
> I have been thinking about this. Today we do a KVM_MEMORY_FAULT_EXIT
> to the VMM to handle the request. The other option is to make this
> a KVM_EXIT_HYPERCALL with SMC_RSI_SET_RIPAS. But this would leak RSI
> implementation to the VMM. The advantage is that the VMM can provide
> a clear response RSI_ACCEPT vs RSI_REJECT (including accepting a partial
> range) and KVM can satisfy the RMI_RTT_SET_RIPAS.

Faking up hypercalls based on a very narrow interface seems like a bad
API design. The guest is of course perfectly allowed to have another
interface directly to the VMM to communicate intent.

> We may end up doing something similar for Device assignment too, where
> the VMM gets a chance to reject any inconsistent device mappings.
> 
> Like you mentioned, the VMM can stop the Realm today as an alternate
> approach.
The intention here is that we start with a very basic interface and can
then extend it when we work out how to handle rejection. The ordering in
the kernel is explicitly to allow that extension. I get the impression
that device assignment has a much better justification for handling the
reject flow so it's probably best to leave that until device assignment
is implemented.

Thanks,
Steve