Re: [PATCH v4 6/7] KVM: SEV: Perform RMP optimizations on SNP guest shutdown

[Ackerley Tng]

Ashish Kalra <Ashish.Kalra@xxxxxxx> writes:

>
> [...snip...]
>
> diff --git a/arch/x86/kvm/svm/sev.c b/arch/x86/kvm/svm/sev.c
> index 3f9c1aa39a0a..e0f4f8ebef68 100644
> --- a/arch/x86/kvm/svm/sev.c
> +++ b/arch/x86/kvm/svm/sev.c
> @@ -2942,6 +2942,8 @@ void sev_vm_destroy(struct kvm *kvm)
>  	if (sev_snp_guest(kvm)) {
>  		snp_guest_req_cleanup(kvm);
>
> +		snp_rmpopt_all_physmem();
> +

I see this is what you suggested in [1]. The time-based batching you
suggeested works because adding to the workqueue when there's already a
job just does nothing. Thanks!

I think optimizing when the VM is destroyed makes sense, in most cases
for SNP VMs, we don't expect large 1G blocks of memory to be shared
anyway, so even if we try to RMPOPT on every conversion to private, most
of those tries would be optimizing nothing.

I guess the remaining optimization would be to update based on only the
range of pfns where guest_memfd has private memory, but that could be
done in another patch series.

Reviewed-by: Ackerley Tng <ackerleytng@xxxxxxxxxx>

[1] https://lore.kernel.org/all/31040bb7-653a-40f9-8899-40bc852f7e1f@xxxxxxx/

>  		/*
>  		 * Decomission handles unbinding of the ASID. If it fails for
>  		 * some unexpected reason, just leak the ASID.
> --
> 2.43.0