[2/2] KVM: x86: fix maintaining of kvm_clock stability on guest CPU hotplug

Commit Message

Denis Plotnikov April 7, 2017, 9:06 a.m. UTC

VCPU TSC synchronization is perfromed in kvm_write_tsc() when the TSC
value being set is within 1 second from the expected, as obtained by
extrapolating of the TSC in already synchronized VCPUs.

This is naturally achieved on all VCPUs at VM start and resume;
however on VCPU hotplug it is not: the newly added VCPU is created
with TSC == 0 while others are well ahead.

To compensate for that, consider host-initiated kvm_write_tsc() with
TSC == 0 a special case requiring synchronization regardless of the
current TSC on other VCPUs.

Signed-off-by: Denis Plotnikov <[e-mail]>
Reviewed-by: Roman Kagan <[e-mail]>
---
 arch/x86/kvm/x86.c | 29 +++++++++++++++++++----------
 1 file changed, 19 insertions(+), 10 deletions(-)

Patch

diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
index c59e38f..cf67091 100644
--- a/arch/x86/kvm/x86.c
+++ b/arch/x86/kvm/x86.c
@@ -1455,16 +1455,25 @@  void kvm_write_tsc(struct kvm_vcpu *vcpu, struct msr_data *msr)
 	elapsed = ns - kvm->arch.last_tsc_nsec;
 
 	if (vcpu->arch.virtual_tsc_khz) {
-		u64 tsc_exp = kvm->arch.last_tsc_write +
-					nsec_to_cycles(vcpu, elapsed);
-		u64 tsc_hz = vcpu->arch.virtual_tsc_khz * 1000LL;
-		/*
-		 * Special case: TSC write with a small delta (1 second)
-		 * of virtual cycle time against real time is
-		 * interpreted as an attempt to synchronize the CPU.
-		 */
-		synchronizing = data < tsc_exp + tsc_hz &&
-				data + tsc_hz > tsc_exp;
+		if (data == 0 && msr->host_initiated) {
+			/*
+			 * detection of vcpu initialization -- need to sync
+			 * with other vCPUs. This particularly helps to keep
+			 * kvm_clock stable after CPU hotplug
+			 */
+			synchronizing = true;
+		} else {
+			u64 tsc_exp = kvm->arch.last_tsc_write +
+						nsec_to_cycles(vcpu, elapsed);
+			u64 tsc_hz = vcpu->arch.virtual_tsc_khz * 1000LL;
+			/*
+			 * Special case: TSC write with a small delta (1 second)
+			 * of virtual cycle time against real time is
+			 * interpreted as an attempt to synchronize the CPU.
+			 */
+			synchronizing = data < tsc_exp + tsc_hz &&
+					data + tsc_hz > tsc_exp;
+		}
 	}
 
 	/*