Re: [PATCH v4 2/6] KVM: x86: Introduce CPUID_8000_0007_EDX 'scattered' leaf

[Sean Christopherson]

On Thu, Sep 22, 2022, Jim Mattson wrote:
> Why do we need a new 'scattered' leaf? We are only using two bits in
> the first one, right? And now we're only going to use one bit in the
> second one?
> 
> I thought the point of the 'scattered' leaves was to collect a bunch
> of feature bits from different CPUID leaves in a single feature word.
> 
> Allocating a new feature word for one or two bits seems extravagant.

Ah, these leafs aren't scattered from KVM's perspective.

The scattered terminology comes from the kernel side, where the KVM-only leafs
_may_ be used to deal with features that are scattered by the kernel.  But there
is no requirement that KVM-only leafs _must_ be scattered by the kernel, e.g. we
can and should use this for leafs that KVM wants to expose to the guest, but are
completely ignored by the kernel.  Intel's PSFD feature flag is a good example.

A better shortlog would be:

  KVM: x86: Add a KVM-only leaf for CPUID_8000_0007_EDX