[PATCH] KVM: x86: nSVM: fix switch to guest mmu

[PATCH] KVM: x86: nSVM: fix switch to guest mmu

[Vitaly Kuznetsov]

Recent optimizations in MMU code broke nested SVM with NPT in L1
completely: when we do nested_svm_{,un}init_mmu_context() we want
to switch from TDP MMU to shadow MMU, both init_kvm_tdp_mmu() and
kvm_init_shadow_mmu() check if re-configuration is needed by looking
at cache source data. The data, however, doesn't change - it's only
the type of the MMU which changes. We end up not re-initializing
guest MMU as shadow and everything goes off the rails.

The issue could have been fixed by putting MMU type into extended MMU
role but this is not really needed. We can just split root and guest MMUs
the exact same way we did for nVMX, their types never change in the
lifetime of a vCPU.

There is still room for improvement: currently, we reset all MMU roots
when switching from L1 to L2 and back and this is not needed.

Fixes: 7dcd57552008 ("x86/kvm/mmu: check if tdp/shadow MMU reconfiguration is needed")
Signed-off-by: Vitaly Kuznetsov <vkuznets@redhat.com>
---
 arch/x86/kvm/svm.c | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)

diff --git a/arch/x86/kvm/svm.c b/arch/x86/kvm/svm.c
index b78c691adea2..38821b0b78da 100644
--- a/arch/x86/kvm/svm.c
+++ b/arch/x86/kvm/svm.c
@@ -2923,6 +2923,8 @@ static void nested_svm_inject_npf_exit(struct kvm_vcpu *vcpu,
 static void nested_svm_init_mmu_context(struct kvm_vcpu *vcpu)
 {
 	WARN_ON(mmu_is_nested(vcpu));
+
+	vcpu->arch.mmu = &vcpu->arch.guest_mmu;
 	kvm_init_shadow_mmu(vcpu);
 	vcpu->arch.mmu->set_cr3           = nested_svm_set_tdp_cr3;
 	vcpu->arch.mmu->get_cr3           = nested_svm_get_tdp_cr3;
@@ -2935,6 +2937,7 @@ static void nested_svm_init_mmu_context(struct kvm_vcpu *vcpu)
 
 static void nested_svm_uninit_mmu_context(struct kvm_vcpu *vcpu)
 {
+	vcpu->arch.mmu = &vcpu->arch.root_mmu;
 	vcpu->arch.walk_mmu = &vcpu->arch.root_mmu;
 }
 
@@ -3444,7 +3447,6 @@ static void enter_svm_guest_mode(struct vcpu_svm *svm, u64 vmcb_gpa,
 		svm->vcpu.arch.hflags &= ~HF_HIF_MASK;
 
 	if (nested_vmcb->control.nested_ctl & SVM_NESTED_CTL_NP_ENABLE) {
-		kvm_mmu_unload(&svm->vcpu);
 		svm->nested.nested_cr3 = nested_vmcb->control.nested_cr3;
 		nested_svm_init_mmu_context(&svm->vcpu);
 	}
-- 
2.19.2

Re: [PATCH] KVM: x86: nSVM: fix switch to guest mmu

[Paolo Bonzini]

On 19/12/18 17:25, Vitaly Kuznetsov wrote:
> Recent optimizations in MMU code broke nested SVM with NPT in L1
> completely: when we do nested_svm_{,un}init_mmu_context() we want
> to switch from TDP MMU to shadow MMU, both init_kvm_tdp_mmu() and
> kvm_init_shadow_mmu() check if re-configuration is needed by looking
> at cache source data. The data, however, doesn't change - it's only
> the type of the MMU which changes. We end up not re-initializing
> guest MMU as shadow and everything goes off the rails.
> 
> The issue could have been fixed by putting MMU type into extended MMU
> role but this is not really needed. We can just split root and guest MMUs
> the exact same way we did for nVMX, their types never change in the
> lifetime of a vCPU.
> 
> There is still room for improvement: currently, we reset all MMU roots
> when switching from L1 to L2 and back and this is not needed.
> 
> Fixes: 7dcd57552008 ("x86/kvm/mmu: check if tdp/shadow MMU reconfiguration is needed")
> Signed-off-by: Vitaly Kuznetsov <vkuznets@redhat.com>

I'm queueing this to kvm-master.  Radim, please pick the other patches
from him to kvm/next and let me know when you're done---the tree is
yours; I'll handle the merge window pull request.

Paolo