writable ID_AA64MMFR0_EL1.TGRAN*_2 ?

writable ID_AA64MMFR0_EL1.TGRAN*_2 ?

[Sebastian Ott]

Hi,

I'm looking into the possibility of migrating between 2 (Graviton) hosts
and the only ID register diffs that are not yet writable are TGRAN*_2.
Both hosts support the same sizes for S2 but express that differently
(2==supported vs 0=="look at TGRAN*").

One idea would be to either set the guest view for these to NI (for the
"normal" / not nested guest) or allow to convert between the 2 values
(and maybe also allow to write NI for good measure).

What do you think?

Sebastian

Re: writable ID_AA64MMFR0_EL1.TGRAN*_2 ?

[Marc Zyngier]

On Fri, 28 Feb 2025 15:19:06 +0000,
Sebastian Ott <sebott@redhat.com> wrote:
> 
> Hi,
> 
> I'm looking into the possibility of migrating between 2 (Graviton) hosts
> and the only ID register diffs that are not yet writable are TGRAN*_2.
> Both hosts support the same sizes for S2 but express that differently
> (2==supported vs 0=="look at TGRAN*").
> 
> One idea would be to either set the guest view for these to NI (for the
> "normal" / not nested guest) or allow to convert between the 2 values
> (and maybe also allow to write NI for good measure).

I think we can allow any value to be written to any of the TGRAN*_2 in
a non-NV context. For NV, we must be fairly strict, by enforcing that
NI is the floor limit for the feature (we cannot let 0 be written for
any of them).

Thanks,

	M.

-- 
Without deviation from the norm, progress is not possible.