On Tue, Nov 25 2025, Peter Maydell <[email protected]> wrote:

> On Tue, 25 Nov 2025 at 16:09, Cornelia Huck <[email protected]> wrote:
>>
>> On Mon, Nov 24 2025, Peter Maydell <[email protected]> wrote:
>>
>> > On Wed, 19 Nov 2025 at 13:44, Cornelia Huck <[email protected]> wrote:
>> >>
>> >> This requires a bit of care, since we still have to handle the EL
>> >> specific part (DCZID_EL0.DZP). Callers can set/access dcz_blocksize
>> >> via a wrapper working on DCZID_EL.BS.
>> >>
>> >> KVM currently does not support DCZID_EL0 via ONE_REG, and actually
>> >> does not need to work with it, so provide a dummy value for now.
>> >
>> > That seems like an odd (unintended?) omission -- is it worth
>> > adding? (We would need to handle older kernels that don't
>> > expose it anyway, of course.)
>>
>> I'm not sure whether there's actually a usecase for KVM exposing this to
>> the VMM - AFAICS, KVM doesn't do anything special for DC ZVA and
>> friends, and doesn't tweak HCR_EL2.TDZ which would change behaviour.
>
> I guess the only one I can think of is to correctly fail
> migration from a source CPU with a DCZID_EL0.BS that doesn't
> match the one on the destination CPU. (We can't lie to the
> guest about the blocksize as part of "tell the guest it has
> a different CPU type from the actual host" unless we want to
> trap and emulate all the DC ZVA etc insns...)

Agreed.

Might actually be best to assert that kvm code is not doing anything
with the reg, rather than providing a dummy value. I'll respin.


Reply via email to