Create VGPU changes with VFIO Framework

cmd/nvidia-vgpu-dm/apply/config.go

@@ -28,7 +28,10 @@ import (
 // VGPUConfig applies the selected vGPU config to the node
 func VGPUConfig(c *Context) error {
 	return assert.WalkSelectedVGPUConfigForEachGPU(c.VGPUConfig, func(vc *v1.VGPUConfigSpec, i int, d types.DeviceID) error {
-		configManager := vgpu.NewNvlibVGPUConfigManager()
+		configManager, err := vgpu.NewNvlibVGPUConfigManager()
+		if err != nil {
+			return fmt.Errorf("error creating vGPU config manager: %v", err)
+		}
 		current, err := configManager.GetVGPUConfig(i)
 		if err != nil {
 			return fmt.Errorf("error getting vGPU config: %v", err)

cmd/nvidia-vgpu-dm/assert/config.go

@@ -36,7 +36,10 @@ func VGPUConfig(c *Context) error {
 
 	matched := make([]bool, len(gpus))
 	err = WalkSelectedVGPUConfigForEachGPU(c.VGPUConfig, func(vc *v1.VGPUConfigSpec, i int, d types.DeviceID) error {
-		configManager := vgpu.NewNvlibVGPUConfigManager()
+		configManager, err := vgpu.NewNvlibVGPUConfigManager()
+		if err != nil {
+			return fmt.Errorf("error creating vGPU config manager: %v", err)
+		}
 		current, err := configManager.GetVGPUConfig(i)
 		if err != nil {
 			return fmt.Errorf("error getting vGPU config: %v", err)

deployments/container/Dockerfile.distroless

@@ -28,6 +28,8 @@ FROM nvcr.io/nvidia/cloud-native/k8s-mig-manager:v0.13.1 as mig-manager
 FROM nvcr.io/nvidia/distroless/go:v4.0.3-dev
 
 ENV NVIDIA_VISIBLE_DEVICES=void
+# Preload NVIDIA NVML library from the driver mounted at /driver-root
+ENV LD_PRELOAD=/driver-root/usr/lib/x86_64-linux-gnu/libnvidia-ml.so.1
 
 COPY --from=build /artifacts/nvidia-vgpu-dm /usr/bin/nvidia-vgpu-dm
 COPY --from=build /artifacts/nvidia-k8s-vgpu-dm /usr/bin/nvidia-k8s-vgpu-dm

internal/vgpu/api.go

@@ -0,0 +1,43 @@
+package vgpu
+
+import (
+	"fmt"
+
+	"github.com/NVIDIA/go-nvlib/pkg/nvpci"
+)
+
+// Interface is the exported vGPU manager API (implemented by *nvvfio and *nvmdev).
+type Interface interface {
+	GetAllDevices() ([]Device, error)
+	GetAllParentDevices() ([]ParentDevice, error)
+	CreateVGPUDevices(device *nvpci.NvidiaPCIDevice, vgpuType string, count int) error
+}
+
+// ParentDevice is implemented by *vfioParentDevice and *mdevParentDevice.
+type ParentDevice interface {
+	GetPhysicalFunction() *nvpci.NvidiaPCIDevice
+	IsVGPUTypeAvailable(string) (bool, error)
+	CreateVGPUDevice(string, string) error
+	GetAvailableVGPUInstances(string) (int, error)
+}
+
+// Device is implemented by *vfioDevice and *mdevDevice.
+type Device interface {
+	GetPhysicalFunction() *nvpci.NvidiaPCIDevice
+	Delete() error
+}
+
+// New returns an Interface, selecting *nvvfio when isVFIOEnabled(0) is true, otherwise *nvmdev from newNvmdev().
+func New() (Interface, error) {
+	vfioInstance := newNvvfio()
+
+	isVFIOMode, err := vfioInstance.isVFIOEnabled(0)
+	if err != nil {
+		return nil, fmt.Errorf("error checking VFIO mode: %w", err)
+	}
+
+	if isVFIOMode {
+		return vfioInstance, nil
+	}
+	return newNvmdev(), nil
+}