On 14/09/21 17:48, Maxim Levitsky wrote:
commit adc2a23734ac ("KVM: nSVM: improve SYSENTER emulation on AMD"),
made init_vmcb set vmload/vmsave intercepts unconditionally,
and relied on svm_vcpu_after_set_cpuid to clear them when possible.
However init_vmcb is also called when the vCPU is reset, and it is
not followed by another call to svm_vcpu_after_set_cpuid because
the CPUID is already set.
This mistake makes the VMSAVE/VMLOAD intercept to be set when
it is not needed, and harms performance of the nested guest.
Fixes: adc2a23734ac ("KVM: nSVM: improve SYSENTER emulation on AMD")
Signed-off-by: Maxim Levitsky <mlevitsk@xxxxxxxxxx>
---
arch/x86/kvm/svm/svm.c | 6 ++++--
1 file changed, 4 insertions(+), 2 deletions(-)
diff --git a/arch/x86/kvm/svm/svm.c b/arch/x86/kvm/svm/svm.c
index 6645542df9bd..861ac9f74331 100644
--- a/arch/x86/kvm/svm/svm.c
+++ b/arch/x86/kvm/svm/svm.c
@@ -1199,8 +1199,6 @@ static void init_vmcb(struct kvm_vcpu *vcpu)
svm_set_intercept(svm, INTERCEPT_SHUTDOWN);
svm_set_intercept(svm, INTERCEPT_VMRUN);
svm_set_intercept(svm, INTERCEPT_VMMCALL);
- svm_set_intercept(svm, INTERCEPT_VMLOAD);
- svm_set_intercept(svm, INTERCEPT_VMSAVE);
svm_set_intercept(svm, INTERCEPT_STGI);
svm_set_intercept(svm, INTERCEPT_CLGI);
svm_set_intercept(svm, INTERCEPT_SKINIT);
@@ -1377,6 +1375,10 @@ static int svm_create_vcpu(struct kvm_vcpu *vcpu)
svm->guest_state_loaded = false;
svm_switch_vmcb(svm, &svm->vmcb01);
+
+ svm_set_intercept(svm, INTERCEPT_VMLOAD);
+ svm_set_intercept(svm, INTERCEPT_VMSAVE);
+
init_vmcb(vcpu);
svm_vcpu_init_msrpm(vcpu, svm->msrpm);
This needs to be redone after the latest refactoring of svm_vcpu_reset.
I'll send a patch myself.
Paolo
