Add csv cdi tests

go.mod

@@ -48,3 +48,5 @@ require (
 	gopkg.in/yaml.v3 v3.0.1 // indirect
 	sigs.k8s.io/yaml v1.4.0 // indirect
 )
+
+replace github.com/NVIDIA/go-nvml => ../go-nvml

go.sum

@@ -2,8 +2,6 @@ cyphar.com/go-pathrs v0.2.1 h1:9nx1vOgwVvX1mNBWDu93+vaceedpbsDqo+XuBGL40b8=
 cyphar.com/go-pathrs v0.2.1/go.mod h1:y8f1EMG7r+hCuFf/rXsKqMJrJAUoADZGNh5/vZPKcGc=
 github.com/NVIDIA/go-nvlib v0.9.1-0.20251202135446-d0f42ba016dd h1:MC1w/VYuo9Zt0se4SSx9BVid4a46ai+voN3knRvVWjE=
 github.com/NVIDIA/go-nvlib v0.9.1-0.20251202135446-d0f42ba016dd/go.mod h1:7mzx9FSdO9fXWP9NKuZmWkCwhkEcSWQFe2tmFwtLb9c=
-github.com/NVIDIA/go-nvml v0.13.0-1 h1:OLX8Jq3dONuPOQPC7rndB6+iDmDakw0XTYgzMxObkEw=
-github.com/NVIDIA/go-nvml v0.13.0-1/go.mod h1:+KNA7c7gIBH7SKSJ1ntlwkfN80zdx8ovl4hrK3LmPt4=
 github.com/blang/semver/v4 v4.0.0 h1:1PFHFE6yCCTv8C1TeyNNarDzntLi7wMI5i/pzqYIsAM=
 github.com/blang/semver/v4 v4.0.0/go.mod h1:IbckMUScFkM3pff0VJDNKRiT6TG/YpiHIM2yvyW5YoQ=
 github.com/containerd/log v0.1.0 h1:TCJt7ioM2cr/tfR8GPbGf9/VRAX8D2B4PjzCpfX540I=

internal/platform-support/tegra/filter.go

@@ -77,6 +77,9 @@ type matcherAsFilter struct {
 }
 
 type filterByMountSpecType map[csv.MountSpecType]filter
+type filterByMountSpecPathsByTyper struct {
+	MountSpecPathsByTyper
+}
 
 type pathPatterns []string
 type pathPattern string
@@ -125,6 +128,14 @@ func (p filterByMountSpecType) Apply(input MountSpecPathsByTyper) MountSpecPaths
 	return ms
 }
 
+func (p filterByMountSpecPathsByTyper) Apply(input MountSpecPathsByTyper) MountSpecPathsByTyper {
+	f := make(filterByMountSpecType)
+	for t, p := range p.MountSpecPathsByType() {
+		f[t] = &matcherAsFilter{pathPatterns(p)}
+	}
+	return f.Apply(input)
+}
+
 // apply uses a matcher to filter an input string.
 // Each element in the input that matches is skipped and the remaining elements
 // are returned.

internal/platform-support/tegra/mount_specs.go

@@ -130,6 +130,10 @@ func WithoutDeviceNodes() Transformer {
 	}
 }
 
+func Without(m MountSpecPathsByTyper) Transformer {
+	return filterByMountSpecPathsByTyper{m}
+}
+
 // WithoutRegularDeviceNodes creates a transfomer which removes
 // regular `/dev/nvidia[0-9]+` device nodes from the source.
 func WithoutRegularDeviceNodes() Transformer {

pkg/nvcdi/lib-csv.go

@@ -46,11 +46,7 @@ var _ deviceSpecGeneratorFactory = (*csvlib)(nil)
 // If NVML is not available or the disable-multiple-csv-devices feature flag is
 // enabled, a single device is assumed.
 func (l *csvlib) DeviceSpecGenerators(ids ...string) (DeviceSpecGenerator, error) {
-	if l.featureFlags[FeatureDisableMultipleCSVDevices] {
-		return l.purecsvDeviceSpecGenerators(ids...)
-	}
-	hasNVML, _ := l.infolib.HasNvml()
-	if !hasNVML {
+	if l.usePureCSVDeviceSpecGenerator() {
 		return l.purecsvDeviceSpecGenerators(ids...)
 	}
 	mixed, err := l.mixedDeviceSpecGenerators(ids...)
@@ -61,6 +57,29 @@ func (l *csvlib) DeviceSpecGenerators(ids ...string) (DeviceSpecGenerator, error
 	return mixed, nil
 }
 
+func (l *csvlib) usePureCSVDeviceSpecGenerator() bool {
+	if l.featureFlags[FeatureDisableMultipleCSVDevices] {
+		return true
+	}
+	hasNVML, _ := l.infolib.HasNvml()
+	if !hasNVML {
+		return true
+	}
+	asNvmlLib := (*nvmllib)(l)
+	err := asNvmlLib.init()
+	if err != nil {
+		return true
+	}
+	defer asNvmlLib.tryShutdown()
+
+	numDevices, ret := l.nvmllib.DeviceGetCount()
+	if ret != nvml.SUCCESS {
+		return true
+	}
+
+	return numDevices <= 1
+}
+
 func (l *csvlib) purecsvDeviceSpecGenerators(ids ...string) (DeviceSpecGenerator, error) {
 	for _, id := range ids {
 		switch id {
@@ -88,6 +107,7 @@ type csvDeviceGenerator struct {
 	*csvlib
 	index int
 	uuid  string
+	mode  string
 }
 
 func (l *csvDeviceGenerator) GetUUID() (string, error) {
@@ -131,30 +151,63 @@ func (l *csvDeviceGenerator) GetDeviceSpecs() ([]specs.Device, error) {
 //   - The device node (i.e. /dev/nvidia{{ .index }}) associated with this
 //     particular device is added to the set of device nodes to be discovered.
 func (l *csvDeviceGenerator) deviceNodeDiscoverer() (discover.Discover, error) {
-	mountSpecs := tegra.Transform(
-		tegra.Transform(
-			tegra.MountSpecsFromCSVFiles(l.logger, l.csvFiles...),
-			// We remove non-device nodes.
-			tegra.OnlyDeviceNodes(),
-		),
-		// We remove the regular (nvidia[0-9]+) device nodes.
-		tegra.WithoutRegularDeviceNodes(),
-	)
 	return tegra.New(
 		tegra.WithLogger(l.logger),
 		tegra.WithDriverRoot(l.driverRoot),
 		tegra.WithDevRoot(l.devRoot),
 		tegra.WithHookCreator(l.hookCreator),
 		tegra.WithLdconfigPath(l.ldconfigPath),
 		tegra.WithLibrarySearchPaths(l.librarySearchPaths...),
-		tegra.WithMountSpecs(
-			mountSpecs,
-			// We add the specific device node for this device.
-			tegra.DeviceNodes(fmt.Sprintf("/dev/nvidia%d", l.index)),
-		),
+		tegra.WithMountSpecs(l.deviceNodeMountSpecs()),
 	)
 }
 
+func (l *csvDeviceGenerator) deviceNodeMountSpecs() tegra.MountSpecPathsByTyper {
+	mountSpecs := tegra.Transform(
+		tegra.MountSpecsFromCSVFiles(l.logger, l.csvFiles...),
+		// We remove non-device nodes.
+		tegra.OnlyDeviceNodes(),
+	)
+	switch l.mode {
+	// For a dGPU we remove all regular device nodes from the list of device
+	// nodes that we detect and only look for the node associated with the
+	// index.
+	case "dgpu":
+		return tegra.Transform(
+			tegra.Transform(
+				mountSpecs,
+				// We remove the regular (nvidia[0-9]+) device nodes.
+				// The device nodes for the GPU are discovered for the full GPU.
+				tegra.WithoutRegularDeviceNodes(),
+			),
+			// We also ignore control device nodes since these are included in
+			// the full GPU spec generator.
+			tegra.Without(
+				tegra.DeviceNodes(
+					"/dev/nvidia-modeset",
+					"/dev/nvidia-uvm-tools",
+					"/dev/nvidia-uvm",
+					"/dev/nvidiactl",
+				),
+			),
+		)
+	case "igpu":
+		return tegra.Merge(
+			tegra.Transform(
+				mountSpecs,
+				// We remove the /dev/nvidia1 device node.
+				// TODO: This assumes that the dGPU has the index 1 and remove
+				// it from the set of device nodes.
+				tegra.Without(tegra.DeviceNodes("/dev/nvidia1")),
+			),
+			// We add the display device from the iGPU.
+			tegra.DeviceNodes("/dev/nvidia2"),
+		)
+	default:
+		return mountSpecs
+	}
+}
+
 // GetCommonEdits generates a CDI specification that can be used for ANY devices
 // These explicitly do not include any device nodes.
 func (l *csvlib) GetCommonEdits() (*cdi.ContainerEdits, error) {
@@ -246,35 +299,56 @@ func (l *mixedcsvlib) csvDeviceSpecGenerator(index int, uuid string, device nvml
 		return nil, fmt.Errorf("is-integrated check failed for device (index=%v,uuid=%v)", index, uuid)
 	}
 
+	if isIntegrated {
+		return l.iGPUDeviceSpecGenerator(index, uuid)
+	}
+
+	return l.dGPUDeviceSpecGenerator(index, uuid, device)
+}
+
+func (l *mixedcsvlib) dGPUDeviceSpecGenerator(index int, uuid string, device nvml.Device) (DeviceSpecGenerator, error) {
+	if index != 1 {
+		return nil, fmt.Errorf("unexpected device index for dGPU: %d", index)
+	}
 	g := &csvDeviceGenerator{
 		csvlib: (*csvlib)(l),
 		index:  index,
 		uuid:   uuid,
+		mode:   "dgpu",
 	}
 
-	if !isIntegrated {
-		csvDeviceNodeDiscoverer, err := g.deviceNodeDiscoverer()
-		if err != nil {
-			return nil, fmt.Errorf("failed to create discoverer for devices nodes: %w", err)
-		}
-
-		// If this is not an integrated GPU, we also create a spec generator for
-		// the full GPU.
-		dgpu := (*nvmllib)(l).withInit(&fullGPUDeviceSpecGenerator{
-			nvmllib: (*nvmllib)(l),
-			uuid:    uuid,
-			index:   index,
-			// For the CSV case, we include the control device nodes at a
-			// device level.
-			additionalDiscoverers: []discover.Discover{
-				(*nvmllib)(l).controlDeviceNodeDiscoverer(),
-				csvDeviceNodeDiscoverer,
-			},
-			featureFlags: l.featureFlags,
-		})
-		return dgpu, nil
+	csvDeviceNodeDiscoverer, err := g.deviceNodeDiscoverer()
+	if err != nil {
+		return nil, fmt.Errorf("failed to create discoverer for devices nodes: %w", err)
 	}
 
+	// If this is not an integrated GPU, we also create a spec generator for
+	// the full GPU.
+	dgpu := (*nvmllib)(l).withInit(&fullGPUDeviceSpecGenerator{
+		nvmllib: (*nvmllib)(l),
+		uuid:    uuid,
+		index:   index,
+		// For the CSV case, we include the control device nodes at a
+		// device level.
+		additionalDiscoverers: []discover.Discover{
+			(*nvmllib)(l).controlDeviceNodeDiscoverer(),
+			csvDeviceNodeDiscoverer,
+		},
+		featureFlags: l.featureFlags,
+	})
+	return dgpu, nil
+}
+
+func (l *mixedcsvlib) iGPUDeviceSpecGenerator(index int, uuid string) (DeviceSpecGenerator, error) {
+	if index != 0 {
+		return nil, fmt.Errorf("unexpected device index for iGPU: %d", index)
+	}
+	g := &csvDeviceGenerator{
+		csvlib: (*csvlib)(l),
+		index:  index,
+		uuid:   uuid,
+		mode:   "igpu",
+	}
 	return g, nil
 }