metrics: add metric for KVM_KVMCLOCK_CTRL ioctl failures

Add a metric that counts KVM_KVMCLOCK_CTRL ioctl call failures.
These failures might happen because the guest simply doesn't use the KVM
clock. In these cases, the metric will increase expectedly. Otherwise,
non-zero values will indicate something actually going wrong.

Signed-off-by: Babis Chalios <[email protected]>

Author: bchalios

src/vmm/src/logger/metrics.rs

@@ -779,6 +779,8 @@ pub struct VcpuMetrics {
     pub exit_mmio_write: SharedIncMetric,
     /// Number of errors during this VCPU's run.
     pub failures: SharedIncMetric,
+    /// Number of times that the `KVM_KVMCLOCK_CTRL` ioctl failed.
+    pub kvmclock_ctrl_fails: SharedIncMetric,
     /// Provides Min/max/sum for KVM exits handling input IO.
     pub exit_io_in_agg: LatencyAggregateMetrics,
     /// Provides Min/max/sum for KVM exits handling output IO.
@@ -797,6 +799,7 @@ impl VcpuMetrics {
             exit_mmio_read: SharedIncMetric::new(),
             exit_mmio_write: SharedIncMetric::new(),
             failures: SharedIncMetric::new(),
+            kvmclock_ctrl_fails: SharedIncMetric::new(),
             exit_io_in_agg: LatencyAggregateMetrics::new(),
             exit_io_out_agg: LatencyAggregateMetrics::new(),
             exit_mmio_read_agg: LatencyAggregateMetrics::new(),

src/vmm/src/vstate/vcpu/mod.rs

@@ -340,9 +340,9 @@ impl Vcpu {
                 // does not panic on resume, see https://docs.kernel.org/virt/kvm/api.html .
                 // We do not want to fail if the call is not successful, because depending
                 // that may be acceptable depending on the workload.
-                // TODO: add a metric
                 #[cfg(target_arch = "x86_64")]
                 if let Err(err) = self.kvm_vcpu.fd.kvmclock_ctrl() {
+                    METRICS.vcpu.kvmclock_ctrl_fails.inc();
                     warn!("KVM_KVMCLOCK_CTRL call failed {}", err);
                 }
 

tests/host_tools/fcmetrics.py

@@ -234,6 +234,7 @@ def validate_fc_metrics(metrics):
             "exit_mmio_read",
             "exit_mmio_write",
             "failures",
+            "kvmclock_ctrl_fails",
             {"exit_io_in_agg": latency_agg_metrics_fields},
             {"exit_io_out_agg": latency_agg_metrics_fields},
             {"exit_mmio_read_agg": latency_agg_metrics_fields},