Update comments to include Windows cancellation scenario

Co-authored-by: ludfjig <4257730+ludfjig@users.noreply.github.com>

Author: Copilot

src/hyperlight_host/src/hypervisor/mod.rs

@@ -469,7 +469,9 @@ impl VirtualCPU {
                 }
                 Ok(HyperlightExit::Cancelled()) => {
                     // If cancellation was not requested for this specific guest function call,
-                    // the vcpu was interrupted by a stale cancellation from a previous call
+                    // the vcpu was interrupted by a stale cancellation. This can occur when:
+                    // - Linux: A signal from a previous call arrives late
+                    // - Windows: WHvCancelRunVirtualProcessor called right after vcpu exits but RUNNING_BIT is still true
                     if !cancel_requested && !debug_interrupted {
                         // Track that an erroneous vCPU kick occurred
                         metrics::counter!(METRIC_ERRONEOUS_VCPU_KICKS).increment(1);

src/hyperlight_host/src/metrics/mod.rs

@@ -21,8 +21,10 @@ pub(crate) static METRIC_GUEST_ERROR_LABEL_CODE: &str = "code";
 // Counter metric that counts the number of times a guest function was called due to timing out
 pub(crate) static METRIC_GUEST_CANCELLATION: &str = "guest_cancellations_total";
 
-// Counter metric that counts the number of times a vCPU was erroneously kicked by a stale signal
-// This happens when a signal from a previous guest call arrives late and interrupts a new call
+// Counter metric that counts the number of times a vCPU was erroneously kicked by a stale cancellation
+// This can happen in two scenarios:
+// 1. Linux: A signal from a previous guest call arrives late and interrupts a new call
+// 2. Windows: WHvCancelRunVirtualProcessor is called right after vCPU exits but RUNNING_BIT is still true
 pub(crate) static METRIC_ERRONEOUS_VCPU_KICKS: &str = "erroneous_vcpu_kicks_total";
 
 // Histogram metric that measures the duration of guest function calls