Add support for GDRCOPY

api/config/v1/flags.go

@@ -60,6 +60,7 @@ type CommandLineFlags struct {
 	MpsRoot                 *string                 `json:"mpsRoot,omitempty"          yaml:"mpsRoot,omitempty"`
 	NvidiaDriverRoot        *string                 `json:"nvidiaDriverRoot,omitempty" yaml:"nvidiaDriverRoot,omitempty"`
 	NvidiaDevRoot           *string                 `json:"nvidiaDevRoot,omitempty"    yaml:"nvidiaDevRoot,omitempty"`
+	GDRCopyEnabled          *bool                   `json:"gdrcopyEnabled"             yaml:"gdrcopyEnabled"`
 	GDSEnabled              *bool                   `json:"gdsEnabled"                 yaml:"gdsEnabled"`
 	MOFEDEnabled            *bool                   `json:"mofedEnabled"               yaml:"mofedEnabled"`
 	UseNodeFeatureAPI       *bool                   `json:"useNodeFeatureAPI"          yaml:"useNodeFeatureAPI"`
@@ -126,6 +127,8 @@ func (f *Flags) UpdateFromCLIFlags(c *cli.Context, flags []cli.Flag) {
 				updateFromCLIFlag(&f.NvidiaDriverRoot, c, n)
 			case "dev-root", "nvidia-dev-root":
 				updateFromCLIFlag(&f.NvidiaDevRoot, c, n)
+			case "gdrcopy-enabled":
+				updateFromCLIFlag(&f.GDRCopyEnabled, c, n)
 			case "gds-enabled":
 				updateFromCLIFlag(&f.GDSEnabled, c, n)
 			case "mofed-enabled":

api/config/v1/flags_test.go

@@ -160,6 +160,7 @@ func TestMarshalFlags(t *testing.T) {
 			output: `{
 				"migStrategy": null,
 				"failOnInitError": null,
+				"gdrcopyEnabled": null,
 				"gdsEnabled": null,
 				"mofedEnabled": null,
 				"useNodeFeatureAPI": null,
@@ -177,6 +178,7 @@ func TestMarshalFlags(t *testing.T) {
 			output: `{
 				"migStrategy": null,
 				"failOnInitError": null,
+				"gdrcopyEnabled": null,
 				"gdsEnabled": null,
 				"mofedEnabled": null,
 				"useNodeFeatureAPI": null,
@@ -201,6 +203,7 @@ func TestMarshalFlags(t *testing.T) {
 			output: `{
 				"migStrategy": null,
 				"failOnInitError": null,
+				"gdrcopyEnabled": null,
 				"gdsEnabled": null,
 				"mofedEnabled": null,
 				"useNodeFeatureAPI": null,

cmd/nvidia-device-plugin/main.go

@@ -100,14 +100,19 @@ func main() {
 			Usage:   "the desired strategy for passing device IDs to the underlying runtime:\n\t\t[uuid | index]",
 			EnvVars: []string{"DEVICE_ID_STRATEGY"},
 		},
+		&cli.BoolFlag{
+			Name:    "gdrcopy-enabled",
+			Usage:   "ensure that containers that request NVIDIA GPU resources are started with GDRCopy support",
+			EnvVars: []string{"GDRCOPY_ENABLED"},
+		},
 		&cli.BoolFlag{
 			Name:    "gds-enabled",
-			Usage:   "ensure that containers are started with NVIDIA_GDS=enabled",
+			Usage:   "ensure that containers that request NVIDIA GPU resources are started with GPUDirect Storage support",
 			EnvVars: []string{"GDS_ENABLED"},
 		},
 		&cli.BoolFlag{
 			Name:    "mofed-enabled",
-			Usage:   "ensure that containers are started with NVIDIA_MOFED=enabled",
+			Usage:   "ensure that containers that request NVIDIA GPU resources are started with MOFED support",
 			EnvVars: []string{"MOFED_ENABLED"},
 		},
 		&cli.StringFlag{

cmd/nvidia-device-plugin/plugin-manager.go

@@ -54,6 +54,7 @@ func GetPlugins(ctx context.Context, infolib info.Interface, nvmllib nvml.Interf
 		cdi.WithNvidiaCTKPath(*config.Flags.Plugin.NvidiaCTKPath),
 		cdi.WithDeviceIDStrategy(*config.Flags.Plugin.DeviceIDStrategy),
 		cdi.WithVendor("k8s.device-plugin.nvidia.com"),
+		cdi.WithGdrcopyEnabled(*config.Flags.GDRCopyEnabled),
 		cdi.WithGdsEnabled(*config.Flags.GDSEnabled),
 		cdi.WithMofedEnabled(*config.Flags.MOFEDEnabled),
 		cdi.WithImexChannels(imexChannels),

deployments/helm/nvidia-device-plugin/templates/daemonset-device-plugin.yml

@@ -169,6 +169,10 @@ spec:
           - name: NVIDIA_CDI_HOOK_PATH
             value: {{ .Values.cdi.nvidiaHookPath }}
         {{- end }}
+        {{- if typeIs "bool" .Values.gdrcopyEnabled }}
+          - name: GDRCOPY_ENABLED
+            value: {{ .Values.gdrcopyEnabled | quote }}
+        {{- end }}
         {{- if typeIs "bool" .Values.gdsEnabled }}
           - name: GDS_ENABLED
             value: {{ .Values.gdsEnabled | quote }}

deployments/helm/nvidia-device-plugin/values.yaml

@@ -33,6 +33,7 @@ failOnInitError: null
 deviceListStrategy: null
 deviceIDStrategy: null
 nvidiaDriverRoot: null
+gdrcopyEnabled: null
 gdsEnabled: null
 mofedEnabled: null
 deviceDiscoveryStrategy: null

go.mod

@@ -2,13 +2,13 @@ module github.com/NVIDIA/k8s-device-plugin
 
 go 1.24.0
 
-toolchain go1.24.1
+toolchain go1.24.5
 
 require (
 	github.com/NVIDIA/go-gpuallocator v0.6.0
-	github.com/NVIDIA/go-nvlib v0.8.0
+	github.com/NVIDIA/go-nvlib v0.8.1
 	github.com/NVIDIA/go-nvml v0.13.0-1
-	github.com/NVIDIA/nvidia-container-toolkit v1.17.8
+	github.com/NVIDIA/nvidia-container-toolkit v1.18.0-rc.5
 	github.com/fsnotify/fsnotify v1.9.0
 	github.com/google/renameio v1.0.1
 	github.com/google/uuid v1.6.0
@@ -28,8 +28,8 @@ require (
 	sigs.k8s.io/node-feature-discovery v0.17.3
 	sigs.k8s.io/node-feature-discovery/api/nfd v0.17.3
 	sigs.k8s.io/yaml v1.4.0
-	tags.cncf.io/container-device-interface v0.8.1
-	tags.cncf.io/container-device-interface/specs-go v0.8.0
+	tags.cncf.io/container-device-interface v1.0.1
+	tags.cncf.io/container-device-interface/specs-go v1.0.0
 )
 
 require (
@@ -65,7 +65,7 @@ require (
 	github.com/xrash/smetrics v0.0.0-20240521201337-686a1a2994c1 // indirect
 	golang.org/x/net v0.38.0 // indirect
 	golang.org/x/oauth2 v0.27.0 // indirect
-	golang.org/x/sys v0.33.0 // indirect
+	golang.org/x/sys v0.36.0 // indirect
 	golang.org/x/term v0.30.0 // indirect
 	golang.org/x/text v0.23.0 // indirect
 	golang.org/x/time v0.8.0 // indirect

go.sum

@@ -1,11 +1,11 @@
 github.com/NVIDIA/go-gpuallocator v0.6.0 h1:2PA2swx59gJYREPkZNTGtyCP6Pnz3WEgnYsXlRkyvkk=
 github.com/NVIDIA/go-gpuallocator v0.6.0/go.mod h1:c+Yspg+/QxWOmoSQeuI48Z/7nS+mMPtxyj1NYUTwewY=
-github.com/NVIDIA/go-nvlib v0.8.0 h1:vorMvnsJYvZaxiluSXFd+fIFeQFPWSiSjNPiJyvDs0c=
-github.com/NVIDIA/go-nvlib v0.8.0/go.mod h1:bV+OEgjJCbFXf5T8c082mVPFuiF+gKwf9CMT7DWGUBI=
+github.com/NVIDIA/go-nvlib v0.8.1 h1:OPEHVvn3zcV5OXB68A7WRpeCnYMRSPl7LdeJH/d3gZI=
+github.com/NVIDIA/go-nvlib v0.8.1/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/NVIDIA/nvidia-container-toolkit v1.17.8 h1:ndE23TKvQBicsZT88mzZudygn6JNOe6+UsIgqk6gGvw=
-github.com/NVIDIA/nvidia-container-toolkit v1.17.8/go.mod h1:khOgMW80+g8eX/1zPlO4demLShHht9I0YEm8ngcPgwk=
+github.com/NVIDIA/nvidia-container-toolkit v1.18.0-rc.5 h1:ft4S4nyT2jI1tV6CAFRMMZdrtd8HAfmuE9X9ieoDz+Y=
+github.com/NVIDIA/nvidia-container-toolkit v1.18.0-rc.5/go.mod h1:t/awbHrDkz8ec0vecKo82Cn/11YkuD2ngE5RT9wuAgU=
 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/cpuguy83/go-md2man/v2 v2.0.7 h1:zbFlGlXEAKlwXpmvle3d8Oe3YnkKIK4xSRTd3sHPnBo=
@@ -169,8 +169,8 @@ golang.org/x/sys v0.0.0-20191026070338-33540a1f6037/go.mod h1:h1NjWce9XRLGQEsW7w
 golang.org/x/sys v0.0.0-20191115151921-52ab43148777/go.mod h1:h1NjWce9XRLGQEsW7wpKNCjG9DtNlClVuFLEZdDNbEs=
 golang.org/x/sys v0.0.0-20200930185726-fdedc70b468f/go.mod h1:h1NjWce9XRLGQEsW7wpKNCjG9DtNlClVuFLEZdDNbEs=
 golang.org/x/sys v0.0.0-20220715151400-c0bba94af5f8/go.mod h1:oPkhp1MJrh7nUepCBck5+mAzfO9JrbApNNgaTdGDITg=
-golang.org/x/sys v0.33.0 h1:q3i8TbbEz+JRD9ywIRlyRAQbM0qF7hu24q3teo2hbuw=
-golang.org/x/sys v0.33.0/go.mod h1:BJP2sWEmIv4KK5OTEluFJCKSidICx8ciO85XgH3Ak8k=
+golang.org/x/sys v0.36.0 h1:KVRy2GtZBrk1cBYA7MKu5bEZFxQk4NIDV6RLVcC8o0k=
+golang.org/x/sys v0.36.0/go.mod h1:OgkHotnGiDImocRcuBABYBEXf8A9a87e/uXjp9XT3ks=
 golang.org/x/term v0.30.0 h1:PQ39fJZ+mfadBm0y5WlL4vlM7Sx1Hgf13sMIY2+QS9Y=
 golang.org/x/term v0.30.0/go.mod h1:NYYFdzHoI5wRh/h5tDMdMqCqPJZEuNqVR5xJLd/n67g=
 golang.org/x/text v0.3.0/go.mod h1:NqM8EUOU14njkJ3fqMW+pc6Ldnwhi/IjpwHt7yyuwOQ=
@@ -232,7 +232,7 @@ sigs.k8s.io/structured-merge-diff/v4 v4.4.2 h1:MdmvkGuXi/8io6ixD5wud3vOLwc1rj0aN
 sigs.k8s.io/structured-merge-diff/v4 v4.4.2/go.mod h1:N8f93tFZh9U6vpxwRArLiikrE5/2tiu1w1AGfACIGE4=
 sigs.k8s.io/yaml v1.4.0 h1:Mk1wCc2gy/F0THH0TAp1QYyJNzRm2KCLy3o5ASXVI5E=
 sigs.k8s.io/yaml v1.4.0/go.mod h1:Ejl7/uTz7PSA4eKMyQCUTnhZYNmLIl+5c2lQPGR2BPY=
-tags.cncf.io/container-device-interface v0.8.1 h1:c0jN4Mt6781jD67NdPajmZlD1qrqQyov/Xfoab37lj0=
-tags.cncf.io/container-device-interface v0.8.1/go.mod h1:Apb7N4VdILW0EVdEMRYXIDVRZfNJZ+kmEUss2kRRQ6Y=
-tags.cncf.io/container-device-interface/specs-go v0.8.0 h1:QYGFzGxvYK/ZLMrjhvY0RjpUavIn4KcmRmVP/JjdBTA=
-tags.cncf.io/container-device-interface/specs-go v0.8.0/go.mod h1:BhJIkjjPh4qpys+qm4DAYtUyryaTDg9zris+AczXyws=
+tags.cncf.io/container-device-interface v1.0.1 h1:KqQDr4vIlxwfYh0Ed/uJGVgX+CHAkahrgabg6Q8GYxc=
+tags.cncf.io/container-device-interface v1.0.1/go.mod h1:JojJIOeW3hNbcnOH2q0NrWNha/JuHoDZcmYxAZwb2i0=
+tags.cncf.io/container-device-interface/specs-go v1.0.0 h1:8gLw29hH1ZQP9K1YtAzpvkHCjjyIxHZYzBAvlQ+0vD8=
+tags.cncf.io/container-device-interface/specs-go v1.0.0/go.mod h1:u86hoFWqnh3hWz3esofRFKbI261bUlvUfLKGrDhJkgQ=

internal/cdi/api.go

@@ -16,16 +16,11 @@
 
 package cdi
 
-import "github.com/NVIDIA/nvidia-container-toolkit/pkg/nvcdi/spec"
-
 // Interface provides the API to the 'cdi' package
 //
-//go:generate moq -stub -out api_mock.go . Interface
+//go:generate moq -rm -fmt=goimports -stub -out api_mock.go . Interface
 type Interface interface {
 	CreateSpecFile() error
 	QualifiedName(string, string) string
-}
-
-type cdiSpecGenerator interface {
-	GetSpec() (spec.Interface, error)
+	AdditionalDevices() []string
 }

internal/cdi/cdi.go

@@ -57,12 +57,14 @@ type cdiHandler struct {
 
 	deviceListStrategies spec.DeviceListStrategies
 
-	gdsEnabled   bool
-	mofedEnabled bool
+	gdsEnabled     bool
+	mofedEnabled   bool
+	gdrcopyEnabled bool
 
 	imexChannels imex.Channels
 
-	cdilibs map[string]cdiSpecGenerator
+	cdilibs         map[string]nvcdi.SpecGenerator
+	additionalModes []string
 }
 
 var _ Interface = &cdiHandler{}
@@ -111,7 +113,7 @@ func New(infolib info.Interface, nvmllib nvml.Interface, devicelib device.Interf
 		return nil, err
 	}
 
-	c.cdilibs = make(map[string]cdiSpecGenerator)
+	c.cdilibs = make(map[string]nvcdi.SpecGenerator)
 
 	c.cdilibs["gpu"], err = nvcdi.New(
 		nvcdi.WithInfoLib(c.infolib),
@@ -133,15 +135,17 @@ func New(infolib info.Interface, nvmllib nvml.Interface, devicelib device.Interf
 		c.cdilibs["imex-channel"] = c.newImexChannelSpecGenerator()
 	}
 
-	var additionalModes []string
+	if c.gdrcopyEnabled {
+		c.additionalModes = append(c.additionalModes, "gdrcopy")
+	}
 	if c.gdsEnabled {
-		additionalModes = append(additionalModes, "gds")
+		c.additionalModes = append(c.additionalModes, "gds")
 	}
 	if c.mofedEnabled {
-		additionalModes = append(additionalModes, "mofed")
+		c.additionalModes = append(c.additionalModes, "mofed")
 	}
 
-	for _, mode := range additionalModes {
+	for _, mode := range c.additionalModes {
 		lib, err := nvcdi.New(
 			nvcdi.WithInfoLib(c.infolib),
 			nvcdi.WithLogger(c.logger),
@@ -162,6 +166,7 @@ func New(infolib info.Interface, nvmllib nvml.Interface, devicelib device.Interf
 
 // CreateSpecFile creates a CDI spec file for the specified devices.
 func (cdi *cdiHandler) CreateSpecFile() error {
+	var emptySpecs []string
 	for class, cdilib := range cdi.cdilibs {
 		cdi.logger.Infof("Generating CDI spec for resource: %s/%s", cdi.vendor, class)
 
@@ -193,10 +198,22 @@ func (cdi *cdiHandler) CreateSpecFile() error {
 
 		err = spec.Save(filepath.Join(cdiRoot, specName+".json"))
 		if err != nil {
+			// TODO: This is a brittle check since it relies on exact string matches.
+			// We should pull this functionality into the CDI tooling instead.
+			if strings.Contains(err.Error(), "invalid device, empty device edits") {
+				klog.ErrorS(err, "Ignoring empty CDI specs", "vendor", cdi.vendor, "class", class)
+				emptySpecs = append(emptySpecs, class)
+				continue
+			}
 			return fmt.Errorf("failed to save CDI spec: %v", err)
 		}
 	}
 
+	// Remove the classes with empty specs from the supported types.
+	for _, emptySpec := range emptySpecs {
+		delete(cdi.cdilibs, emptySpec)
+	}
+
 	return nil
 }
 
@@ -229,3 +246,18 @@ func (cdi *cdiHandler) getRootTransformer() transform.Transformer {
 func (cdi *cdiHandler) QualifiedName(class string, id string) string {
 	return cdiparser.QualifiedName(cdi.vendor, class, id)
 }
+
+// AdditionalDevices returns the optional CDI devices based on the device plugin
+// configuration.
+// Here we check for requested modes as well as whether the modes have a valid
+// CDI spec associated with them.
+func (cdi *cdiHandler) AdditionalDevices() []string {
+	var devices []string
+	for _, mode := range cdi.additionalModes {
+		if cdi.cdilibs[mode] == nil {
+			continue
+		}
+		devices = append(devices, cdi.QualifiedName(mode, "all"))
+	}
+	return devices
+}

internal/cdi/imex.go

@@ -19,6 +19,7 @@ package cdi
 import (
 	"tags.cncf.io/container-device-interface/specs-go"
 
+	"github.com/NVIDIA/nvidia-container-toolkit/pkg/nvcdi"
 	"github.com/NVIDIA/nvidia-container-toolkit/pkg/nvcdi/spec"
 
 	"github.com/NVIDIA/k8s-device-plugin/internal/imex"
@@ -29,7 +30,7 @@ type imexChannelCDILib struct {
 	imexChannels imex.Channels
 }
 
-func (cdi *cdiHandler) newImexChannelSpecGenerator() cdiSpecGenerator {
+func (cdi *cdiHandler) newImexChannelSpecGenerator() nvcdi.SpecGenerator {
 	lib := &imexChannelCDILib{
 		vendor:       cdi.vendor,
 		imexChannels: cdi.imexChannels,
@@ -39,7 +40,7 @@ func (cdi *cdiHandler) newImexChannelSpecGenerator() cdiSpecGenerator {
 }
 
 // GetSpec returns the CDI specs for IMEX channels.
-func (l *imexChannelCDILib) GetSpec() (spec.Interface, error) {
+func (l *imexChannelCDILib) GetSpec(...string) (spec.Interface, error) {
 	var deviceSpecs []specs.Device
 	for _, channel := range l.imexChannels {
 		deviceSpec := specs.Device{

internal/cdi/null.go

@@ -30,6 +30,10 @@ func NewNullHandler() Interface {
 	return &null{}
 }
 
+func (n *null) AdditionalDevices() []string {
+	return nil
+}
+
 // CreateSpecFile is a no-op for the null handler.
 func (n *null) CreateSpecFile() error {
 	return nil