address comments (2)

Signed-off-by: Varun Ramachandra Sekar <[email protected]>

Author: varunrsekar

cmd/gpu-kubelet-plugin/cdi.go

@@ -31,6 +31,8 @@ import (
 	cdiapi "tags.cncf.io/container-device-interface/pkg/cdi"
 	cdiparser "tags.cncf.io/container-device-interface/pkg/parser"
 	cdispec "tags.cncf.io/container-device-interface/specs-go"
+
+	"github.com/NVIDIA/k8s-dra-driver-gpu/pkg/featuregates"
 )
 
 const (
@@ -168,9 +170,12 @@ func (cdi *CDIHandler) CreateStandardDeviceSpecFile(allocatable AllocatableDevic
 		klog.Errorf("failed to create standard nvidia device spec file: %v", err)
 		return err
 	}
-	if err := cdi.createStandardVfioDeviceSpecFile(allocatable); err != nil {
-		klog.Errorf("failed to create standard vfio device spec file: %v", err)
-		return err
+
+	if featuregates.Enabled(featuregates.PassthroughSupport) {
+		if err := cdi.createStandardVfioDeviceSpecFile(allocatable); err != nil {
+			klog.Errorf("failed to create standard vfio device spec file: %v", err)
+			return err
+		}
 	}
 	return nil
 }

cmd/gpu-kubelet-plugin/device_state.go

@@ -98,7 +98,7 @@ func NewDeviceState(ctx context.Context, config *Config) (*DeviceState, error) {
 
 	var vfioPciManager *VfioPciManager
 	if featuregates.Enabled(featuregates.PassthroughSupport) {
-		manager := NewVfioPciManager(string(containerDriverRoot), nvdevlib)
+		manager := NewVfioPciManager(string(containerDriverRoot), nvdevlib, true /* nvidiaEnabled */)
 		if err := manager.Prechecks(); err == nil {
 			vfioPciManager = manager
 		} else {
@@ -505,7 +505,7 @@ func (s *DeviceState) unprepareVfioDevices(ctx context.Context, devices Prepared
 		if err != nil {
 			return fmt.Errorf("error getting allocatable device for vfio device: %w", err)
 		}
-		if err := s.vfioPciManager.Unconfigure(vfioAllocatable.Vfio); err != nil {
+		if err := s.vfioPciManager.Unconfigure(ctx, vfioAllocatable.Vfio); err != nil {
 			return fmt.Errorf("error unconfiguring vfio device: %w", err)
 		}
 	}

cmd/gpu-kubelet-plugin/vfio-device.go

@@ -47,23 +47,26 @@ const (
 )
 
 type VfioPciManager struct {
-	driverRoot string
-	driver     string
-	nvlib      *deviceLib
+	driverRoot    string
+	driver        string
+	nvlib         *deviceLib
+	nvidiaEnabled bool
 }
 
-func NewVfioPciManager(driverRoot string, nvlib *deviceLib) *VfioPciManager {
+func NewVfioPciManager(driverRoot string, nvlib *deviceLib, nvidiaEnabled bool) *VfioPciManager {
 	vm := &VfioPciManager{
-		driverRoot: driverRoot,
-		driver:     vfioPciDriver,
-		nvlib:      nvlib,
+		driverRoot:    driverRoot,
+		driver:        vfioPciDriver,
+		nvlib:         nvlib,
+		nvidiaEnabled: nvidiaEnabled,
 	}
 	if !vm.isVfioPCIModuleLoaded() {
 		err := vm.loadVfioPciModule()
 		if err != nil {
 			klog.Fatalf("failed to load vfio_pci module: %v", err)
 		}
 	}
+
 	return vm
 }
 
@@ -124,7 +127,7 @@ func (vm *VfioPciManager) loadVfioPciModule() error {
 	return nil
 }
 
-func (vm *VfioPciManager) WaitForGPUFree(info *VfioDeviceInfo) error {
+func (vm *VfioPciManager) WaitForGPUFree(ctx context.Context, info *VfioDeviceInfo) error {
 	if info.parent == nil {
 		return nil
 	}
@@ -175,7 +178,11 @@ func (vm *VfioPciManager) Configure(ctx context.Context, info *VfioDeviceInfo) e
 	if driver == vm.driver {
 		return nil
 	}
-	err = vm.WaitForGPUFree(info)
+	// Only support vfio-pci or nvidia (if vm.nvidiaEnabled) driver.
+	if !vm.nvidiaEnabled || driver != nvidiaDriver {
+		return fmt.Errorf("gpu is bound to %q driver, expected %q or %q", driver, vm.driver, nvidiaDriver)
+	}
+	err = vm.WaitForGPUFree(ctx, info)
 	if err != nil {
 		return err
 	}
@@ -191,10 +198,15 @@ func (vm *VfioPciManager) Configure(ctx context.Context, info *VfioDeviceInfo) e
 }
 
 // Unconfigure binds the GPU to the nvidia driver.
-func (vm *VfioPciManager) Unconfigure(info *VfioDeviceInfo) error {
+func (vm *VfioPciManager) Unconfigure(ctx context.Context, info *VfioDeviceInfo) error {
 	perGpuLock.Get(info.pcieBusID).Lock()
 	defer perGpuLock.Get(info.pcieBusID).Unlock()
 
+	// Do nothing if we dont expect to switch to nvidia driver.
+	if !vm.nvidiaEnabled {
+		return nil
+	}
+
 	driver, err := getDriver(pciDevicesRoot, info.pcieBusID)
 	if err != nil {
 		return err