Re: [PATCH v4 3/7] KVM: x86: Move enable_pml variable to common x86 code

[Sean Christopherson <seanjc@google.com>]

On Tue, Oct 14, 2025, Kai Huang wrote:
> On Mon, 2025-10-13 at 06:25 +0000, Nikunj A Dadhania wrote:
> > Move the enable_pml module parameter from VMX-specific code to common x86
> > KVM code. This allows both VMX and SVM implementations to access the same
> > PML enable/disable control.
> > 
> > No functional change, just code reorganization to support shared PML
> > infrastructure.
> > 
> > Suggested-by: Kai Huang <kai.huang@intel.com>
> 
> For the record :-)
> 
> When I moved the 'enable_pml' from VMX to x86 in the diff I attached to v6
> was purely because vmx_update_cpu_dirty_logging() checks 'enable_pml' and
> after it got moved to x86 the new kvm_vcpu_update_cpu_dirty_logging() also
> needed to use it (for the sake of just moving code).
> 
> I didn't mean to suggest to use a common boolean in x86 and let SVM/VMX
> code to access it, since the downside is we need to export it.  But I
> think it's not a bad idea either.

Ya.  At some point it might makes sense to define "struct kvm_params", a la
"kvm_caps" and "kvm_host_values", so that we don't need a pile of one-off exports.
I'm not sure I'm entirely in favor of that idea though, as I think it'd be a net
negative for overall code readability.  And with EXPORT_SYMBOL_FOR_KVM_INTERNAL,
exports feel a lot less gross :-)