Re: [RFC 18/21] arm/cpu: Introduce a customizable kvm host cpu model

[Peter Maydell]

On Mon, 11 Nov 2024 at 14:29, Cornelia Huck <cohuck@redhat.com> wrote:
> That leaves some headscratchers (at least for me.) E.g. FEAT_UINJ, which
> is optional from v9.5, and mandatory from v9.6, but where I'm unsure how
> we'd discover it, especially as we do not have a way to discover the
> architecture extensions.

I think that's just an accidental omission from the FEAT_UINJ
documentation. There's a UINJ field in ID_AA64PFR2_EL1 which
tells you whether FEAT_UINJ is implemented:

https://developer.arm.com/documentation/109697/2024_09/Feature-descriptions/The-Armv9-6-architecture-extension

> I'm also not sure if I understand FEAT_CHK, which is listed as
> optional from v8.0 and mandatory from v9.4, but every aarch64 cpu is
> supposed to be compliant, because CHKFEAT might be a NOP (and this is
> only supposed to check for FEAT_GCS? Yes, I'm confused.)

CHKFEAT is there for cases where software (especially EL0 software)
might need a fast cross-OS-portable way to detect whether a feature
is present and enabled -- for instance if you want the compiler
to emit a function prologue sequence that includes "do XYZ if
GCS is on". CHKFEAT is in the hint area of the encoding space,
so for any CPU that pre-dates FEAT_CHK it will do nothing (which is
how "no features detectable with CHKFEAT are present" is indicated:
if you look at what CHKFEAT actually does it has a definition that
isn't how you'd naturally implement a feature-detect unless you
needed to make NOP mean no-features-present.)

Currently FEAT_GCS is the only feature where it has been deemed
necessary to provide this kind of is-it-enabled detection, but some
future features might also use the CHKFEAT mechanism.

If you're a piece of system software wanting to see if FEAT_GCS
is present you would use ID_AA64PFR1_EL1.GCS. (In particular, if
you're an OS then until you have enabled GCS for yourself via
GCSCR_EL1, CHKFEAT will tell you "not enabled", so you first need
to look at the ID register to see if the GCSCR_EL1 register is
there at all...)

thanks
-- PMM