[PATCH] KVM: nVMX: Track vmx emulation errors

[PATCH] KVM: nVMX: Track vmx emulation errors

[Fred Griffoul]

From: Fred Griffoul <fgriffo@amazon.co.uk>

Add a new kvm_stat vcpu counter called "nested_errors" to track the
number of errors returned to an L1 hypervisor when emulated VMX
instructions fail.

This counter should help monitor nVMX health and troubleshoot issues
with L1 hypervisors.

Signed-off-by: Fred Griffoul <fgriffo@amazon.co.uk>
---
 arch/x86/include/asm/kvm_host.h | 1 +
 arch/x86/kvm/vmx/nested.c       | 2 ++
 arch/x86/kvm/x86.c              | 1 +
 3 files changed, 4 insertions(+)

diff --git a/arch/x86/include/asm/kvm_host.h b/arch/x86/include/asm/kvm_host.h
index 5a3bfa293e8b..a3aaccd6e6aa 100644
--- a/arch/x86/include/asm/kvm_host.h
+++ b/arch/x86/include/asm/kvm_host.h
@@ -1674,6 +1674,7 @@ struct kvm_vcpu_stat {
 	u64 preemption_other;
 	u64 guest_mode;
 	u64 notify_window_exits;
+	u64 nested_errors;
 };
 
 struct x86_instruction_info;
diff --git a/arch/x86/kvm/vmx/nested.c b/arch/x86/kvm/vmx/nested.c
index 6137e5307d0f..2e1394151945 100644
--- a/arch/x86/kvm/vmx/nested.c
+++ b/arch/x86/kvm/vmx/nested.c
@@ -162,6 +162,7 @@ static int nested_vmx_succeed(struct kvm_vcpu *vcpu)
 
 static int nested_vmx_failInvalid(struct kvm_vcpu *vcpu)
 {
+	++vcpu->stat.nested_errors;
 	vmx_set_rflags(vcpu, (vmx_get_rflags(vcpu)
 			& ~(X86_EFLAGS_PF | X86_EFLAGS_AF | X86_EFLAGS_ZF |
 			    X86_EFLAGS_SF | X86_EFLAGS_OF))
@@ -172,6 +173,7 @@ static int nested_vmx_failInvalid(struct kvm_vcpu *vcpu)
 static int nested_vmx_failValid(struct kvm_vcpu *vcpu,
 				u32 vm_instruction_error)
 {
+	++vcpu->stat.nested_errors;
 	vmx_set_rflags(vcpu, (vmx_get_rflags(vcpu)
 			& ~(X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF |
 			    X86_EFLAGS_SF | X86_EFLAGS_OF))
diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
index ff8812f3a129..475c8a2d704e 100644
--- a/arch/x86/kvm/x86.c
+++ b/arch/x86/kvm/x86.c
@@ -299,6 +299,7 @@ const struct _kvm_stats_desc kvm_vcpu_stats_desc[] = {
 	STATS_DESC_COUNTER(VCPU, preemption_other),
 	STATS_DESC_IBOOLEAN(VCPU, guest_mode),
 	STATS_DESC_COUNTER(VCPU, notify_window_exits),
+	STATS_DESC_COUNTER(VCPU, nested_errors),
 };
 
 const struct kvm_stats_header kvm_vcpu_stats_header = {

base-commit: 0499add8efd72456514c6218c062911ccc922a99
-- 
2.43.0

Re: [PATCH] KVM: nVMX: Track vmx emulation errors

[Sean Christopherson]

On Tue, Jan 20, 2026, Fred Griffoul wrote:
> From: Fred Griffoul <fgriffo@amazon.co.uk>
> 
> Add a new kvm_stat vcpu counter called "nested_errors" to track the
> number of errors returned to an L1 hypervisor when emulated VMX
> instructions fail.

This is too broad/vague, and very imperfect.  E.g. if a guest is tripping on a
specific VMREAD for whatever reason, it will pollute the count and make it hard
to detect more serious issues like VM-Entry failures.  And if VMCS shadowing is
in use, KVM could easily underreport meaningful errors.

I also dislike implementing this for nVMX but not nSVM.  nSVM doesn't have the
same ISA surface and so many of the errors simply don't exist, but at the very
least VMRUN failures should be accounted.

And counting VM-Fail but not faults is rather odd if the goal is to monitor L1
hypervisor health.

> This counter should help monitor nVMX health and troubleshoot issues
> with L1 hypervisors.

There are lots of ways to monitor the health of L1 hypervisors, it's not clear to
me why we need a KVM-provide stat for VM-Fail.