tests/ocp/sriov: fix Mellanox CX6-DX switchdev and netdev-to-vfiopci test failures

tests/ocp/sriov/internal/sriovenv/sriovenv.go

@@ -248,6 +248,8 @@ func initVFWithDevType(name, deviceID, interfaceName, vendor, devType string, vf
 			policy.Definition.Spec.NicSelector.DeviceID = deviceID
 		}
 
+		policy.Definition.Spec.EswitchMode = "legacy"
+
 		if _, err := policy.Create(); err != nil {
 			klog.V(90).Infof("Failed to create policy on node %q: %v", nodeName, err)
 
@@ -632,7 +634,7 @@ func VerifyLinkStateConfiguration(networkName, namespace, description string,
 // ============================================================================
 
 func verifySpoofCheck(clientPod *pod.Builder, interfaceName, expectedState string) error {
-	// Get node name and MAC
+	// Get node name
 	refreshedPod, err := pod.Pull(APIClient, clientPod.Definition.Name, clientPod.Definition.Namespace)
 	if err != nil {
 		return fmt.Errorf("failed to refresh pod: %w", err)
@@ -643,11 +645,6 @@ func verifySpoofCheck(clientPod *pod.Builder, interfaceName, expectedState strin
 		return fmt.Errorf("pod node name is empty")
 	}
 
-	mac, err := ExtractPodInterfaceMAC(clientPod, "net1")
-	if err != nil {
-		return fmt.Errorf("failed to extract MAC: %w", err)
-	}
-
 	// Execute on node via cluster helper
 	outputMap, err := cluster.ExecCmdWithStdout(APIClient, fmt.Sprintf("ip link show %s", interfaceName),
 		metav1.ListOptions{LabelSelector: fmt.Sprintf("kubernetes.io/hostname=%s", nodeName)})
@@ -671,19 +668,33 @@ func verifySpoofCheck(clientPod *pod.Builder, interfaceName, expectedState strin
 		}
 	}
 
-	// Find line with MAC and check spoof state
+	// Search VF lines for the expected spoof check state.
+	// Note: In switchdev mode (e.g., Mellanox CX6-DX), `ip link show <PF>` reports VF MACs as
+	// 00:00:00:00:00:00 regardless of the MAC assigned inside the pod, so matching by MAC address
+	// is unreliable. Instead, we check any VF line for the expected state, since the SR-IOV policy
+	// applies a uniform spoof checking configuration to all VFs it creates.
+	vfLinesFound := 0
+
 	for _, line := range strings.Split(output, "\n") {
-		if strings.Contains(line, mac) {
-			if strings.Contains(line, fmt.Sprintf("spoof checking %s", expectedState)) ||
-				strings.Contains(line, fmt.Sprintf("spoofchk %s", expectedState)) {
-				klog.V(90).Infof("Spoof check verified: %s for MAC %s", expectedState, mac)
+		if !strings.Contains(line, "vf ") {
+			continue
+		}
 
-				return nil
-			}
+		vfLinesFound++
+
+		if strings.Contains(line, fmt.Sprintf("spoof checking %s", expectedState)) ||
+			strings.Contains(line, fmt.Sprintf("spoofchk %s", expectedState)) {
+			klog.V(90).Infof("Spoof check verified: %s on interface %s", expectedState, interfaceName)
+
+			return nil
 		}
 	}
 
-	return fmt.Errorf("spoof check %s not found for MAC %s", expectedState, mac)
+	if vfLinesFound == 0 {
+		return fmt.Errorf("no VF information found in ip link show output for interface %s", interfaceName)
+	}
+
+	return fmt.Errorf("spoof check %s not found for interface %s", expectedState, interfaceName)
 }
 
 // ============================================================================

tests/ocp/sriov/tests/exposemtu.go

@@ -24,6 +24,7 @@ var _ = Describe("SRIOV: Expose MTU:", Ordered, Label(tsparams.LabelExposeMTUTes
 		workerNodeList           []*nodes.Builder
 		err                      error
 		sriovInterfacesUnderTest []string
+		vfNum                    int
 	)
 
 	BeforeAll(func() {
@@ -38,6 +39,11 @@ var _ = Describe("SRIOV: Expose MTU:", Ordered, Label(tsparams.LabelExposeMTUTes
 		sriovInterfacesUnderTest, err = SriovOcpConfig.GetSriovInterfaces(1)
 		Expect(err).ToNot(HaveOccurred(), "Failed to retrieve SR-IOV interfaces for testing")
 
+		By("Loading VF configuration")
+
+		vfNum, err = SriovOcpConfig.GetVFNum()
+		Expect(err).ToNot(HaveOccurred(), "Failed to get VF number")
+
 		By("Verifying if expose MTU tests can be executed on given cluster")
 
 		err = sriovocpenv.DoesClusterHaveEnoughNodes(1, 1)
@@ -65,19 +71,19 @@ var _ = Describe("SRIOV: Expose MTU:", Ordered, Label(tsparams.LabelExposeMTUTes
 	})
 
 	It("netdev 1500", reportxml.ID("73786"), func() {
-		testExposeMTU(1500, sriovInterfacesUnderTest, "netdevice")
+		testExposeMTU(1500, sriovInterfacesUnderTest, "netdevice", vfNum)
 	})
 
 	It("netdev 9000", reportxml.ID("73787"), func() {
-		testExposeMTU(9000, sriovInterfacesUnderTest, "netdevice")
+		testExposeMTU(9000, sriovInterfacesUnderTest, "netdevice", vfNum)
 	})
 
 	It("vfio 1500", reportxml.ID("73789"), func() {
-		testExposeMTU(1500, sriovInterfacesUnderTest, "vfio-pci")
+		testExposeMTU(1500, sriovInterfacesUnderTest, "vfio-pci", vfNum)
 	})
 
 	It("vfio 9000", reportxml.ID("73790"), func() {
-		testExposeMTU(9000, sriovInterfacesUnderTest, "vfio-pci")
+		testExposeMTU(9000, sriovInterfacesUnderTest, "vfio-pci", vfNum)
 	})
 
 	It("netdev 2 Policies with different MTU", reportxml.ID("73788"), func() {
@@ -93,7 +99,7 @@ var _ = Describe("SRIOV: Expose MTU:", Ordered, Label(tsparams.LabelExposeMTUTes
 			sriovAndResourceName5000,
 			SriovOcpConfig.SriovOperatorNamespace,
 			sriovAndResourceName5000,
-			5,
+			vfNum,
 			[]string{fmt.Sprintf("%s#0-1", sriovInterfacesUnderTest[0])}, SriovOcpConfig.WorkerLabelMap).
 			WithDevType("netdevice").WithMTU(5000).Create()
 		Expect(err).ToNot(HaveOccurred(), "Failed to configure SR-IOV policy with mtu 5000")
@@ -103,7 +109,7 @@ var _ = Describe("SRIOV: Expose MTU:", Ordered, Label(tsparams.LabelExposeMTUTes
 			sriovAndResourceName9000,
 			SriovOcpConfig.SriovOperatorNamespace,
 			sriovAndResourceName9000,
-			5,
+			vfNum,
 			[]string{fmt.Sprintf("%s#2-3", sriovInterfacesUnderTest[0])}, SriovOcpConfig.WorkerLabelMap).
 			WithDevType("netdevice").WithMTU(9000).Create()
 		Expect(err).ToNot(HaveOccurred(), "Failed to configure SR-IOV policy with mtu 9000")
@@ -166,7 +172,7 @@ var _ = Describe("SRIOV: Expose MTU:", Ordered, Label(tsparams.LabelExposeMTUTes
 	})
 })
 
-func testExposeMTU(mtu int, interfacesUnderTest []string, devType string) {
+func testExposeMTU(mtu int, interfacesUnderTest []string, devType string, vfsNumber int) {
 	By("Creating SR-IOV policy")
 
 	const sriovAndResourceNameExposeMTU = "exposemtu"
@@ -176,7 +182,7 @@ func testExposeMTU(mtu int, interfacesUnderTest []string, devType string) {
 		sriovAndResourceNameExposeMTU,
 		SriovOcpConfig.SriovOperatorNamespace,
 		sriovAndResourceNameExposeMTU,
-		5,
+		vfsNumber,
 		interfacesUnderTest, SriovOcpConfig.WorkerLabelMap).WithDevType(devType).WithMTU(mtu)
 
 	err := sriovoperator.CreateSriovPolicyAndWaitUntilItsApplied(

tests/ocp/sriov/tests/metricsExporter.go

@@ -304,9 +304,22 @@
 		"Failed to add server mac address in client pod mac table. Output: %s", outputbuf.String()))
 
 	By("ICMP check between client and server pods")
-	Eventually(func() error {
-		return sriovocpenv.ICMPConnectivityCheck(cPod, []string{tsparams.ServerIPv4IPAddress}, "net1")
-	}, 1*time.Minute, 2*time.Second).Should(HaveOccurred(), "ICMP fail scenario could not be executed")
+
+	// Derive the expected ICMP outcome from the device type that defineMetricsPolicy() actually
+	// configured on the server policy, rather than re-checking the vendor ID here.
+	// With true vfio-pci (Intel), the VF is exclusively owned by DPDK and ICMP fails.
+	// With netdevice+RDMA (Mellanox "vfiopci" mode), the kernel network stack remains active
+	// on the VF and ICMP succeeds.
+	if serverResources.sriovPolicy.Definition.Spec.DeviceType == "vfio-pci" {
+		Eventually(func() error {
+			return sriovocpenv.ICMPConnectivityCheck(cPod, []string{tsparams.ServerIPv4IPAddress}, "net1")
+		}, 1*time.Minute, 2*time.Second).Should(HaveOccurred(), "ICMP fail scenario could not be executed")
+	} else {
+		Eventually(func() error {
+			return sriovocpenv.ICMPConnectivityCheck(cPod, []string{tsparams.ServerIPv4IPAddress}, "net1")
+		}, 1*time.Minute, 2*time.Second).ShouldNot(HaveOccurred(),
+			"ICMP connectivity check failed for netdevice+RDMA server")
+	}
 
 	checkMetricsWithPromQL()
 }
@@ -472,16 +485,22 @@
 }
 
 func createMetricsTestResources(cRes, sRes metricsTestResource) *pod.Builder {
+	// Create all policies first so the SR-IOV daemon sees them in the same reconcile generation.
+	// Creating policies sequentially with network/NAD waits in between introduces a gap that
+	// causes the daemon to process them in separate generations, leading to premature
+	// WaitForSriovStable returns before all device plugin resources are registered.
 	for _, res := range []metricsTestResource{cRes, sRes} {
-		By("Create SriovNetworkNodePolicy")
+		By(fmt.Sprintf("Create SriovNetworkNodePolicy %s", res.policy.Definition.Name))
 
 		_, err := res.policy.Create()
 		Expect(err).ToNot(HaveOccurred(),
 			fmt.Sprintf("Failed to Create SriovNetworkNodePolicy %s", res.policy.Definition.Name))
+	}
 
-		By("Create SriovNetwork")
+	for _, res := range []metricsTestResource{cRes, sRes} {
+		By(fmt.Sprintf("Create SriovNetwork %s", res.network.Definition.Name))
 
-		_, err = res.network.Create()
+		_, err := res.network.Create()
 		Expect(err).ToNot(HaveOccurred(),
 			fmt.Sprintf("Failed to create SriovNetwork %s", res.network.Definition.Name))