On 08/11/2018 19:41, Liran Alon wrote:
> So what I plan to do is indeed to define first 4K of data as vmcs12 and next 4K as shadow_vmcs12.
> I will also define each of them in a separate VMState subsection that each will have it’s own .needed()
> method that will decide if it’s relevant to send it based on kvm_state.size.
> vmcs12 and shadow_vmcs12 will be put in a struct which is unioned with a struct
> of 2 pages buffer to clearly indicate that one well-defined struct is used for VMX and the other for SVM.
> (based on kvm_state.format)
And SVM will use other subsections.
> In addition, I will change KVM_{GET,SET}_NESTED_STATE documentation to indicate that
> future nested state fields should be added at the end of struct and weather they are valid should
> be determined by a flag in kvm_state.flags.
> QEMU will handle these new states in the future by adding more VMState subsections that have
> .needed() method based on appropriate flag in kvm_state.flags.
> The struct itself that userspace use against it’s local kernel will be determined based on KVM_CAPs.
>
> If there are no objections, I will write the needed patches for QEMU (and the documentation patch for KVM).
Sure. You can also place the flags and format in yet another subsection.
Paolo
