feat: implement MlxResetFW to reset the FW on VF changes

Signed-off-by: Tobias Giese <tgiese@nvidia.com>

Dockerfile.sriov-network-config-daemon

@@ -5,7 +5,7 @@ RUN make _build-sriov-network-config-daemon BIN_PATH=build/_output/cmd
 
 FROM quay.io/centos/centos:stream9
 ARG MSTFLINT=mstflint
-RUN ARCH_DEP_PKGS=$(if [ "$(uname -m)" != "s390x" ]; then echo -n ${MSTFLINT} ; fi) && yum -y install hwdata $ARCH_DEP_PKGS && yum clean all
+RUN ARCH_DEP_PKGS=$(if [ "$(uname -m)" != "s390x" ]; then echo -n ${MSTFLINT} ; fi) && yum -y install hwdata pciutils $ARCH_DEP_PKGS && yum clean all
 LABEL io.k8s.display-name="sriov-network-config-daemon" \
       io.k8s.description="This is a daemon that manage and config sriov network devices in Kubernetes cluster"
 COPY --from=builder /go/src/github.com/k8snetworkplumbingwg/sriov-network-operator/build/_output/cmd/sriov-network-config-daemon /usr/bin/

pkg/plugins/mellanox/mellanox_plugin.go

@@ -19,6 +19,7 @@ type MellanoxPlugin struct {
 	helpers     helper.HostHelpersInterface
 }
 
+var pciAddressesToReset []string
 var attributesToChange map[string]mlx.MlxNic
 var mellanoxNicsStatus map[string]map[string]sriovnetworkv1.InterfaceExt
 var mellanoxNicsSpec map[string]sriovnetworkv1.Interface
@@ -51,6 +52,7 @@ func (p *MellanoxPlugin) OnNodeStateChange(new *sriovnetworkv1.SriovNetworkNodeS
 	needDrain = false
 	needReboot = false
 	err = nil
+	pciAddressesToReset = []string{}
 	attributesToChange = map[string]mlx.MlxNic{}
 	mellanoxNicsStatus = map[string]map[string]sriovnetworkv1.InterfaceExt{}
 	mellanoxNicsSpec = map[string]sriovnetworkv1.Interface{}
@@ -131,6 +133,10 @@ func (p *MellanoxPlugin) OnNodeStateChange(new *sriovnetworkv1.SriovNetworkNodeS
 		if needReboot || changeWithoutReboot {
 			attributesToChange[ifaceSpec.PciAddress] = *attrs
 		}
+
+		if needReboot {
+			pciAddressesToReset = append(pciAddressesToReset, ifaceSpec.PciAddress)
+		}
 	}
 
 	// Set total VFs to 0 for mellanox interfaces with no spec
@@ -189,7 +195,10 @@ func (p *MellanoxPlugin) Apply() error {
 		return nil
 	}
 	log.Log.Info("mellanox plugin Apply()")
-	return p.helpers.MlxConfigFW(attributesToChange)
+	if err := p.helpers.MlxConfigFW(attributesToChange); err != nil {
+		return err
+	}
+	return p.helpers.MlxResetFW(pciAddressesToReset)
 }
 
 // nicHasExternallyManagedPFs returns true if one of the ports(interface) of the NIC is marked as externally managed

pkg/vendors/mellanox/mellanox.go

@@ -60,6 +60,7 @@ type MellanoxInterface interface {
 	GetMlxNicFwData(pciAddress string) (current, next *MlxNic, err error)
 
 	MlxConfigFW(attributesToChange map[string]MlxNic) error
+	MlxResetFW(pciAddresses []string) error
 }
 
 type mellanoxHelper struct {
@@ -141,6 +142,20 @@ func (m *mellanoxHelper) GetMellanoxBlueFieldMode(PciAddress string) (BlueFieldM
 	return -1, fmt.Errorf("MellanoxBlueFieldMode(): unknown device status for %s", PciAddress)
 }
 
+func (m *mellanoxHelper) MlxResetFW(pciAddresses []string) error {
+	log.Log.Info("mellanox-plugin resetFW()")
+	for _, pciAddress := range pciAddresses {
+		cmdArgs := []string{"-d", pciAddress, "-l", "3", "-y", "reset"}
+		log.Log.Info("mellanox-plugin: resetFW()", "cmd-args", cmdArgs)
+		_, stderr, err := m.utils.RunCommand("mstfwreset", cmdArgs...)
+		if err != nil {
+			log.Log.Error(err, "mellanox-plugin resetFW(): failed", "stderr", stderr)
+			return err
+		}
+	}
+	return nil
+}
+
 func (m *mellanoxHelper) MlxConfigFW(attributesToChange map[string]MlxNic) error {
 	log.Log.Info("mellanox-plugin configFW()")
 	for pciAddr, fwArgs := range attributesToChange {