Add metric for erroneous vCPU kicks from stale cancellations

.squashed-commit-marker

@@ -0,0 +1 @@
+Squashed commit: a39633dd5d0cdd1727690af873752b2162021d67

src/hyperlight_host/src/hypervisor/mod.rs

@@ -23,7 +23,7 @@ use crate::hypervisor::regs::{
     CommonFpu, CommonRegisters, CommonSegmentRegister, CommonSpecialRegisters,
 };
 use crate::mem::memory_region::{MemoryRegion, MemoryRegionFlags};
-use crate::metrics::METRIC_GUEST_CANCELLATION;
+use crate::metrics::{METRIC_ERRONEOUS_VCPU_KICKS, METRIC_GUEST_CANCELLATION};
 #[cfg(feature = "mem_profile")]
 use crate::sandbox::trace::MemTraceInfo;
 use crate::{HyperlightError, Result, log_then_return};
@@ -469,8 +469,12 @@ 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);
                         // treat this the same as a HyperlightExit::Retry, the cancel was not meant for this call
                         continue;
                     }

src/hyperlight_host/src/metrics/mod.rs

@@ -21,6 +21,12 @@ 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 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
 #[cfg(feature = "function_call_metrics")]
 pub(crate) static METRIC_GUEST_FUNC_DURATION: &str = "guest_call_duration_seconds";