On Thu, 2022-06-30 at 12:58 +0100, David Woodhouse wrote:
>  The TSC of each vCPU#n was <Tₙ> at a given value of kvmclock <Kₙ>

That one isn't even particularly hard to do. At its core, it looks
something like this (thanks to Simon for fixing the arithmetic. My
first proof-of-concept was only a *few* orders of magnitude out...)

Simon then has a follow-on patch to wrap an ioctl around it and
actually allow the kvm_ns to be passed in.

diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
index 1910e1e78b15..ddb67c9566e4 100644
--- a/arch/x86/kvm/x86.c
+++ b/arch/x86/kvm/x86.c
@@ -2629,7 +2629,7 @@ static void __kvm_synchronize_tsc(struct kvm_vcpu *vcpu, u64 offset, u64 tsc,
 	kvm_track_tsc_matching(vcpu);
 }
 
-static void kvm_synchronize_tsc(struct kvm_vcpu *vcpu, u64 data)
+static void kvm_synchronize_tsc(struct kvm_vcpu *vcpu, u64 data, u64 *kvm_ns)
 {
 	struct kvm *kvm = vcpu->kvm;
 	u64 offset, ns, elapsed;
@@ -2638,8 +2638,23 @@ static void kvm_synchronize_tsc(struct kvm_vcpu *vcpu, u64 data)
 	bool synchronizing = false;
 
 	raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
-	offset = kvm_compute_l1_tsc_offset(vcpu, data);
 	ns = get_kvmclock_base_ns();
+	if (kvm_ns) {
+ 		/*
+ 		 * We have been provided a KVM clock reference time point at
+ 		 * which this TSC value was correct. Use this time point to
+		 * compensate for any delays that were incurred since that
+		 * TSC value was computed. Note that time is still passing;
+		 * in an ideal world the get_kvmclock_base_ns() above would
+		 * be based on the *same* TSC value that's hidden in the call
+		 * to kvm_compute_l1_tsc_offset() below.
+ 		 */
+ 		u32 tsc_khz = vcpu->arch.virtual_tsc_khz;
+ 		s64 delta_ns = ns + vcpu->kvm->arch.kvmclock_offset - *kvm_ns;
+ 		s64 delta_counts = delta_ns * tsc_khz / NSEC_PER_MSEC;
+ 		data += delta_counts;
+	}
+	offset = kvm_compute_l1_tsc_offset(vcpu, data);
 	elapsed = ns - kvm->arch.last_tsc_nsec;
 
 	if (vcpu->arch.virtual_tsc_khz) {
@@ -3581,7 +3596,7 @@ int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
 		break;
 	case MSR_IA32_TSC:
 		if (msr_info->host_initiated) {
-			kvm_synchronize_tsc(vcpu, data);
+			kvm_synchronize_tsc(vcpu, data, NULL);
 		} else {
 			u64 adj = kvm_compute_l1_tsc_offset(vcpu, data) - vcpu->arch.l1_tsc_offset;
 			adjust_tsc_offset_guest(vcpu, adj);
@@ -11392,7 +11407,7 @@ void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
 	if (mutex_lock_killable(&vcpu->mutex))
 		return;
 	vcpu_load(vcpu);
-	kvm_synchronize_tsc(vcpu, 0);
+	kvm_synchronize_tsc(vcpu, 0, NULL);
 	vcpu_put(vcpu);
 
 	/* poll control enabled by default */