[CP 1171] [Feature] Add support to disable watch for KMM resources to totally disable KMM for users

README.md

@@ -78,6 +78,7 @@ helm install amd-gpu-operator rocm/gpu-operator-charts \
 Installation Options
   - Skip NFD installation: `--set node-feature-discovery.enabled=false`
   - Skip KMM installation: `--set kmm.enabled=false`
+  - Disable KMM watching/usage: `--set kmm.watch=false`
   - Skip Auto Node Remediation: `--set remediation.enabled=false`
   - Enable DRA driver (instead of device plugin): `--set deviceConfig.spec.draDriver.enable=true --set deviceConfig.spec.devicePlugin.enableDevicePlugin=false`
   - Disable DeviceClass creation: `--set draDriver.deviceClass.create=false`

cmd/main.go

@@ -34,6 +34,8 @@ package main
 
 import (
 	"flag"
+	"os"
+	"strconv"
 
 	workflowv1alpha1 "github.com/argoproj/argo-workflows/v3/pkg/apis/workflow/v1alpha1"
 	kmmv1beta1 "github.com/rh-ecosystem-edge/kernel-module-management/api/v1beta1"
@@ -98,7 +100,18 @@ func main() {
 
 	setupLogger := logger.WithName("setup")
 
-	setupLogger.Info("Creating manager", "version", Version, "git commit", GitCommit, "build tag", BuildTag)
+	// Read KMM watch configuration from environment variable
+	kmmWatchEnabled := true // default to true for backward compatibility
+	if kmmWatchEnv := os.Getenv("KMM_WATCH_ENABLED"); kmmWatchEnv != "" {
+		var err error
+		kmmWatchEnabled, err = strconv.ParseBool(kmmWatchEnv)
+		if err != nil {
+			setupLogger.Error(err, "invalid KMM_WATCH_ENABLED value, defaulting to true", "value", kmmWatchEnv)
+			kmmWatchEnabled = true
+		}
+	}
+
+	setupLogger.Info("Creating manager", "version", Version, "git commit", GitCommit, "build tag", BuildTag, "kmmWatchEnabled", kmmWatchEnabled)
 
 	setupLogger.Info("Parsing configuration file", "path", configFile)
 
@@ -118,7 +131,12 @@ func main() {
 	// Use manager's client, it may read from a cache.
 	client := mgr.GetClient()
 	isOpenShift := utils.IsOpenShift(setupLogger)
-	kmmHandler := kmmmodule.NewKMMModule(client, scheme, isOpenShift)
+	var kmmHandler kmmmodule.KMMModuleAPI
+	if kmmWatchEnabled {
+		kmmHandler = kmmmodule.NewKMMModule(client, scheme, isOpenShift)
+	} else {
+		kmmHandler = kmmmodule.NewNoOpKMMModule()
+	}
 	dpHandler := plugin.NewDevicePlugin(client, scheme, isOpenShift)
 	nlHandler := nodelabeller.NewNodeLabeller(scheme, isOpenShift)
 	metricsHandler := metricsexporter.NewMetricsExporter(scheme)
@@ -135,7 +153,8 @@ func main() {
 		testrunnerHandler,
 		configmanagerHandler,
 		workerMgr,
-		isOpenShift)
+		isOpenShift,
+		kmmWatchEnabled)
 	if err = dcr.SetupWithManager(mgr); err != nil {
 		cmd.FatalError(setupLogger, err, "unable to create controller", "name", controllers.DeviceConfigReconcilerName)
 	}

docs/installation/kubernetes-helm.md

@@ -126,8 +126,14 @@ helm install amd-gpu-operator rocm/gpu-operator-charts \
 Installation Options
   - Skip NFD installation: `--set node-feature-discovery.enabled=false`
   - Skip KMM installation: `--set kmm.enabled=false` <br> Although KMM is a [Kubernetes-SIGs](https://github.com/kubernetes-sigs) maintained project, it is strongly recommended to use AMD optimized and published KMM images included in each operator release.
+  - Disable GPU operator watching KMM resources: `--set kmm.watch=false`
   - Skip Auto Node Remediation: `--set remediation.enabled=false`
   - Disable default DeviceConfig installation: `--set crds.defaultCR.install=false`
+
+**KMM Configuration Examples:**
+  - Use existing KMM installation: `--set kmm.enabled=false --set kmm.watch=true`
+  - Skip KMM for alternative driver management: `--set kmm.enabled=false --set kmm.watch=false`
+  - Install KMM without GPU operator using it: `--set kmm.enabled=true --set kmm.watch=false`
 ```
 
 ```{tip}
@@ -167,6 +173,7 @@ The following parameters are able to be configued when using the Helm Chart. In
 | controllerManager.nodeSelector | object | `{}` | Node selector for AMD GPU operator controller manager deployment |
 | installdefaultNFDRule | bool | `true` | Set to true to install default NFD rule for detecting AMD GPU hardware based on pci vendor ID and device ID |
 | kmm.enabled | bool | `true` | Set to true/false to enable/disable the installation of kernel module management (KMM) operator |
+| kmm.watch | bool | `true` | Set to true/false to enable/disable GPU operator watching and using KMM resources |
 | node-feature-discovery.enabled | bool | `true` | Set to true/false to enable/disable the installation of node feature discovery (NFD) operator |
 | upgradeCRD | bool | `true` | CRD will be patched as pre-upgrade/pre-rollback hook when doing helm upgrade/rollback to current helm chart |
 | kmm.controller.affinity | object | `{"nodeAffinity":{"preferredDuringSchedulingIgnoredDuringExecution":[{"preference":{"matchExpressions":[{"key":"node-role.kubernetes.io/control-plane","operator":"Exists"}]},"weight":1}]}}` | Affinity for the KMM controller manager deployment |
@@ -206,7 +213,6 @@ The following parameters are able to be configued when using the Helm Chart. In
 | kmm.kubernetesClusterDomain | string | `"cluster.local"` |  |
 | kmm.managerConfig.controllerConfigYaml | string | `"healthProbeBindAddress: :8081\nwebhookPort: 9443\nleaderElection:\n  enabled: true\n  resourceID: kmm.sigs.x-k8s.io\nmetrics:\n  enableAuthnAuthz: true\n  bindAddress: 0.0.0.0:8443\n  secureServing: true\nworker:\n  runAsUser: 0\n  seLinuxType: spc_t\n  firmwareHostPath: /var/lib/firmware"` |  |
 | kmm.webhookServer.affinity | object | `{"nodeAffinity":{"preferredDuringSchedulingIgnoredDuringExecution":[{"preference":{"matchExpressions":[{"key":"node-role.kubernetes.io/control-plane","operator":"Exists"}]},"weight":1}]}}` | KMM webhook's deployment affinity configs |
-| kmm.enabled | bool | `true` | Set to true/false to enable/disable the installation of kernel module management (KMM) operator |
 | kmm.webhookServer.nodeAffinity.nodeSelectorTerms | list | `[{"key":"node-role.kubernetes.io/control-plane","operator":"Exists"},{"key":"node-role.kubernetes.io/master","operator":"Exists"}]` | Node affinity selector terms config for the KMM webhook deployment, set it to [] if you want to make affinity config empty |
 | kmm.webhookServer.nodeSelector | object | `{}` | KMM webhook's deployment node selector |
 | kmm.webhookServer.replicas | int | `1` |  |

docs/releasenotes.md

@@ -1,5 +1,17 @@
 # Release Notes
 
+## GPU Operator v1.5.0 Release Notes
+
+### Release Highlights
+
+- **Enhanced KMM (Kernel Module Management) Configuration Control**
+  - **Independent Control of KMM Installation and Usage**
+    - New helm parameters provide separate control over KMM installation and resource watching:
+      - `kmm.enabled`: Controls KMM subchart installation (default: `true`)
+      - `kmm.watch`: Controls GPU operator watching KMM resources (default: `true`)
+    - Supports multiple deployment scenarios: use existing KMM installations (`enabled=false, watch=true`), skip KMM entirely for alternative driver solutions (`enabled=false, watch=false`), or install KMM without asking for GPU Operator to use it (`enabled=true, watch=false`)
+    - Fully backward compatible: existing configurations with `kmm.enabled=false` continue to work without changes
+
 ## GPU Operator v1.4.1 Release Notes
 
 The AMD GPU Operator v1.4.1 release extends platform support to OpenShift v4.20 and Debian 12, and introduces the ability to build `amdgpu` kernel modules directly within air-gapped OpenShift clusters.

hack/k8s-patch/metadata-patch/values.yaml

@@ -17,8 +17,10 @@ node-feature-discovery:
     nodeSelector: {}
 # KMM related configs
 kmm:
-  # -- Set to true/false to enable/disable the installation of kernel module management (KMM) operator
+  # -- Set to true/false to enable/disable the installation of kernel module management (KMM) operator subchart
   enabled: true
+  # -- Set to true/false to enable/disable GPU operator watching and using KMM resources
+  watch: true
 # Remediation related configs
 remediation:
   # -- Set to true/false to enable/disable the installation of remediation workflow controller

hack/k8s-patch/template-patch/deployment.yaml

@@ -41,6 +41,8 @@ spec:
           value: {{ quote .Values.kubernetesClusterDomain }}
         - name: SIM_ENABLE
           value: {{ quote .Values.controllerManager.env.simEnable }}
+        - name: KMM_WATCH_ENABLED
+          value: {{ quote .Values.kmm.watch }}
         image: {{ .Values.controllerManager.manager.image.repository }}:{{ .Values.controllerManager.manager.image.tag
           | default .Chart.AppVersion }}
         imagePullPolicy: {{ .Values.controllerManager.manager.imagePullPolicy }}

helm-charts-k8s/README.md

@@ -78,6 +78,7 @@ helm install amd-gpu-operator rocm/gpu-operator-charts \
 Installation Options
   - Skip NFD installation: `--set node-feature-discovery.enabled=false`
   - Skip KMM installation: `--set kmm.enabled=false`
+  - Disable KMM watching/usage: `--set kmm.watch=false`
   - Skip Auto Node Remediation: `--set remediation.enabled=false`
   - Enable DRA driver (instead of device plugin): `--set deviceConfig.spec.draDriver.enable=true --set deviceConfig.spec.devicePlugin.enableDevicePlugin=false`
   - Disable DeviceClass creation: `--set draDriver.deviceClass.create=false`
@@ -266,7 +267,8 @@ Kubernetes: `>= 1.29.0-0`
 | deviceConfig.spec.testRunner.upgradePolicy.upgradeStrategy | string | `"RollingUpdate"` | the type of daemonset upgrade, RollingUpdate or OnDelete |
 | draDriver.deviceClass.create | bool | `true` | Create the gpu.amd.com DeviceClass resource. Set to false if managing the DRA driver independently. |
 | installdefaultNFDRule | bool | `true` | Default NFD rule will detect amd gpu based on pci vendor ID |
-| kmm.enabled | bool | `true` | Set to true/false to enable/disable the installation of kernel module management (KMM) operator |
+| kmm.enabled | bool | `true` | Set to true/false to enable/disable the installation of kernel module management (KMM) operator subchart |
+| kmm.watch | bool | `true` | Set to true/false to enable/disable GPU operator watching and using KMM resources |
 | node-feature-discovery.enabled | bool | `true` | Set to true/false to enable/disable the installation of node feature discovery (NFD) operator |
 | node-feature-discovery.worker.nodeSelector | object | `{}` | Set nodeSelector for NFD worker daemonset |
 | node-feature-discovery.worker.tolerations | list | `[{"effect":"NoExecute","key":"amd-dcm","operator":"Equal","value":"up"},{"effect":"NoSchedule","key":"amd-gpu-unhealthy","operator":"Exists"}]` | Set tolerations for NFD worker daemonset |

helm-charts-k8s/templates/deployment.yaml

@@ -41,6 +41,8 @@ spec:
           value: {{ quote .Values.kubernetesClusterDomain }}
         - name: SIM_ENABLE
           value: {{ quote .Values.controllerManager.env.simEnable }}
+        - name: KMM_WATCH_ENABLED
+          value: {{ quote .Values.kmm.watch }}
         image: {{ .Values.controllerManager.manager.image.repository }}:{{ .Values.controllerManager.manager.image.tag
           | default .Chart.AppVersion }}
         imagePullPolicy: {{ .Values.controllerManager.manager.imagePullPolicy }}

helm-charts-k8s/values.yaml

@@ -17,8 +17,10 @@ node-feature-discovery:
     nodeSelector: {}
 # KMM related configs
 kmm:
-  # -- Set to true/false to enable/disable the installation of kernel module management (KMM) operator
+  # -- Set to true/false to enable/disable the installation of kernel module management (KMM) operator subchart
   enabled: true
+  # -- Set to true/false to enable/disable GPU operator watching and using KMM resources
+  watch: true
 # Remediation related configs
 remediation:
   # -- Set to true/false to enable/disable the installation of remediation workflow controller