Re: [PATCH 04/17] Fix deep C-state TSC desynchronization

From: Zachary Amsden
Date: Tue Jun 15 2010 - 04:14:15 EST

On 06/14/2010 10:09 PM, Avi Kivity wrote:
On 06/15/2010 10:34 AM, Zachary Amsden wrote:
When CPUs with unstable TSCs enter deep C-state, TSC may stop
running. This causes us to require resynchronization. Since
we can't tell when this may potentially happen, we assume the
worst by forcing re-compensation for it at every point the VCPU
task is descheduled.

Signed-off-by: Zachary Amsden<zamsden@xxxxxxxxxx>
arch/x86/kvm/x86.c | 11 +++++++++++
1 files changed, 11 insertions(+), 0 deletions(-)

diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
index c8289d0..618c435 100644
--- a/arch/x86/kvm/x86.c
+++ b/arch/x86/kvm/x86.c
@@ -1822,7 +1822,18 @@ void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
vcpu->arch.last_host_tsc = native_read_tsc();
+ /*
+ * When potentially leaving a CPU with unstable TSCs, we risk
+ * that the CPU enters deep C-state. If it does, the TSC may
+ * go out of sync but we will not recalibrate because the test
+ * vcpu->cpu != cpu can not detect this condition. So set
+ * vcpu->cpu = -1 to force the recalibration above.
+ */
+ if (check_tsc_unstable())
+ vcpu->cpu = -1;

That will cause us to miss a vmclear later on. Also it invalidates an invariant that the per-cpu list vcpus_on_cpu has all the vcpus with vcpu->cpu == cpu on this cpu (try saything that fast).

Blasted SVM / VMX differences! There are other, less elegant ways to accomplish this task however.

Good catch, BTW.
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at
Please read the FAQ at