On Wed, Feb 26, 2025 at 1:27 AM Sean Christopherson [off-list ref] wrote:

On Wed, Feb 26, 2025, Paolo Bonzini wrote:

quoted

On 2/25/25 00:55, Sean Christopherson wrote:

quoted

Free vCPUs before freeing any VM state, as both SVM and VMX may access
VM state when "freeing" a vCPU that is currently "in" L2, i.e. that needs
to be kicked out of nested guest mode.

Commit 6fcee03df6a1 ("KVM: x86: avoid loading a vCPU after .vm_destroy was
called") partially fixed the issue, but for unknown reasons only moved the
MMU unloading before VM destruction.  Complete the change, and free all
vCPU state prior to destroying VM state, as nVMX accesses even more state
than nSVM.

I applied this to kvm-coco-queue, I will place it in kvm/master too unless
you shout.

Depends on what "this" is :-)

My plan/hope is to land patches 1 and 2 in 6.14, i.e. in kvm/master

I meant only 1, but if you want to have 2 as well then that's fine too.

As to kvm-coco-queue, based on Yan's reply I have 1 (of course), 4 and
an extra patch to move kvm_x86_call(vm_destroy) at the very end of
kvm_arch_destroy_vm; I'll post everything as soon as I finish building
and testing.

Paolo

rest are firmly 6.15 IMO.  And based on Yan's feedback, I'm planning on adding a
few more cleanups (though I think they're fully additive, i.e. can go on top).