Merge pull request #1495 from elezar/expose-nvcdi-features

Allow CDI feature flags to be set

Author: elezar

cmd/nvidia-device-plugin/main.go

@@ -169,6 +169,11 @@ func main() {
 			Usage:   "The specified IMEX channels are required",
 			EnvVars: []string{"IMEX_REQUIRED"},
 		},
+		&cli.StringSliceFlag{
+			Name:    "cdi-feature-flags",
+			Usage:   "A set of feature flags to be passed to the CDI spec generation logic",
+			EnvVars: []string{"CDI_FEATURE_FLAGS"},
+		},
 	}
 	o.flags = c.Flags
 

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 "string" .Values.cdi.featureFlags }}
+          - name: CDI_FEATURE_FLAGS
+            value: {{ .Values.cdi.featureFlags }}
+        {{- end }}
         {{- if typeIs "bool" .Values.gdrcopyEnabled }}
           - name: GDRCOPY_ENABLED
             value: {{ .Values.gdrcopyEnabled | quote }}

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

@@ -163,3 +163,6 @@ cdi:
   # nvidiaHookPath specifies the path to the nvidia-cdi-hook or nvidia-ctk executables on the host.
   # This is required to ensure that the generated CDI specification refers to the correct CDI hooks.
   nvidiaHookPath: null
+  # featureFlags allows CDI feature flags to be specified for CDI specification generation.
+  # This is specified as a comma-separated list of strings.
+  featureFlags: null

internal/cdi/cdi.go

@@ -57,6 +57,9 @@ type cdiHandler struct {
 
 	deviceListStrategies spec.DeviceListStrategies
 
+	// nvcdiFeatureFlags allows finer control over CDI spec generation.
+	nvcdiFeatureFlags []string
+
 	gdsEnabled     bool
 	mofedEnabled   bool
 	gdrcopyEnabled bool
@@ -113,19 +116,26 @@ func New(infolib info.Interface, nvmllib nvml.Interface, devicelib device.Interf
 		return nil, err
 	}
 
-	c.cdilibs = make(map[string]nvcdi.SpecGenerator)
-
-	c.cdilibs["gpu"], err = nvcdi.New(
-		nvcdi.WithInfoLib(c.infolib),
-		nvcdi.WithNvmlLib(c.nvmllib),
+	commonOptions := []nvcdi.Option{
 		nvcdi.WithDeviceLib(c.devicelib),
+		nvcdi.WithDevRoot(c.devRoot),
+		nvcdi.WithDriverRoot(c.driverRoot),
+		nvcdi.WithFeatureFlags(c.nvcdiFeatureFlags...),
+		nvcdi.WithInfoLib(c.infolib),
 		nvcdi.WithLogger(c.logger),
 		nvcdi.WithNVIDIACDIHookPath(c.nvidiaCTKPath),
-		nvcdi.WithDriverRoot(c.driverRoot),
-		nvcdi.WithDevRoot(c.devRoot),
-		nvcdi.WithDeviceNamers(deviceNamer),
+		nvcdi.WithNvmlLib(c.nvmllib),
 		nvcdi.WithVendor(c.vendor),
-		nvcdi.WithClass("gpu"),
+	}
+
+	c.cdilibs = make(map[string]nvcdi.SpecGenerator)
+
+	c.cdilibs["gpu"], err = nvcdi.New(
+		append(
+			commonOptions,
+			nvcdi.WithDeviceNamers(deviceNamer),
+			nvcdi.WithClass("gpu"),
+		)...,
 	)
 	if err != nil {
 		return nil, fmt.Errorf("failed to create nvcdi library: %v", err)
@@ -147,13 +157,10 @@ func New(infolib info.Interface, nvmllib nvml.Interface, devicelib device.Interf
 
 	for _, mode := range c.additionalModes {
 		lib, err := nvcdi.New(
-			nvcdi.WithInfoLib(c.infolib),
-			nvcdi.WithLogger(c.logger),
-			nvcdi.WithNVIDIACDIHookPath(c.nvidiaCTKPath),
-			nvcdi.WithDriverRoot(c.driverRoot),
-			nvcdi.WithDevRoot(c.devRoot),
-			nvcdi.WithVendor(c.vendor),
-			nvcdi.WithMode(mode),
+			append(
+				commonOptions,
+				nvcdi.WithMode(mode),
+			)...,
 		)
 		if err != nil {
 			return nil, fmt.Errorf("failed to create nvcdi library: %v", err)

internal/cdi/options.go

@@ -24,6 +24,14 @@ import (
 // Option defines a function for passing options to the New() call
 type Option func(*cdiHandler)
 
+// WithFeatureFlags provides and option to set the feature flags for the nvcdi
+// spec generation instance.
+func WithFeatureFlags(featureFlags ...string) Option {
+	return func(c *cdiHandler) {
+		c.nvcdiFeatureFlags = featureFlags
+	}
+}
+
 // WithDeviceListStrategies provides an Option to set the enabled flag used by the 'cdi' interface
 func WithDeviceListStrategies(deviceListStrategies spec.DeviceListStrategies) Option {
 	return func(c *cdiHandler) {