From c60143a808108bad115ae174f00a30adfdd75149 Mon Sep 17 00:00:00 2001
From: Peter Colledge <peter.colledge@amd.com>
Date: Wed, 20 Dec 2023 11:39:24 +0000
Subject: [PATCH] cleanup: Documentation on NFD

---
 README.md   |  4 ++++
 docs/nfd.md | 54 +++++++++++++++++++++++++++++++++++++++++++++++++++++
 2 files changed, 58 insertions(+)
 create mode 100644 docs/nfd.md

diff --git a/README.md b/README.md
index 71634b7..ce7917d 100644
--- a/README.md
+++ b/README.md
@@ -10,6 +10,7 @@ Use OpenOnload® or EnterpriseOnload® to accelerate your workloads in Kubernete
 * [AMD Solarflare](https://www.solarflare.com) hardware (`sfc`)
 * OpenShift Container Platform (OCP) 4.10+ with
   * [Kernel Module Management (KMM) Operator](https://kmm.sigs.k8s.io/) 1.1 ([OpenShift documentation](https://docs.openshift.com/container-platform/4.14/hardware_enablement/kmm-kernel-module-management.html))
+  * [Node Feature Discovery (NFD)](docs/nfd.md) Operator (optional)
 * Both restricted network or internet-connected clusters
 
 Deployment can also be performed on Kubernetes 1.23+ but full implementation details are not currently provided.
@@ -164,6 +165,9 @@ this recommended overlay further, see the variant steps below.
 The above overlay configures KMM to `modprobe onload` but `modprobe sfc` is also required.
 Please see [Out-of-tree `sfc` module](#out-of-tree-sfc-kernel-module) for options.
 
+The above overlay selects **all `worker` role nodes** in the cluster. To filter based on node hardware, you may wish
+to use the [recommended Node Feature Discovery configuration](docs/nfd.md).
+
 > [!IMPORTANT]
 > Due to Kubernetes limitations on label lengths, the combined length of the Name and Namespace of the Onload CR must be less than 32 characters.
 
diff --git a/docs/nfd.md b/docs/nfd.md
new file mode 100644
index 0000000..d759e4a
--- /dev/null
+++ b/docs/nfd.md
@@ -0,0 +1,54 @@
+
+# Selecting Nodes with AMD Solarflare hardware using Node Feature Discovery (NFD)
+
+## Cluster configuration
+
+[Node Feature Discovery (NFD)](https://kubernetes-sigs.github.io/node-feature-discovery)
+([Redhat documentation](https://docs.openshift.com/container-platform/4.14/hardware_enablement/psap-node-feature-discovery-operator.html#create-cd-cli_node-feature-discovery-operator))
+enables the selection of nodes based on hardware features and system configuration.
+NFD-Worker runs on each node to detect changes which are then used to label the node.
+
+A `NodeFeatureDiscovery` CR enables the detections you require. A full example is provided in the above documentation
+if you do not already have one configured.
+
+To enable detection of AMD Solarflare cards, identified by the PCIe Subsystem Vendor ID '1924',
+add the following configuration to your CR's `configData` section:
+
+```yaml
+kind: NodeFeatureDiscovery
+...
+spec:
+  ...
+  workerConfig:
+    configData: |
+      sources:
+        pci:
+          deviceClassWhitelist:
+            - "1924"
+          deviceLabelFields:
+            - "subsystem_vendor"
+```
+
+After NFD is deployed, configured, and its daemons have performed detections, verify with:
+
+```sh
+kubectl get nodes -l feature.node.kubernetes.io/pci-1924.present=true
+```
+
+## Onload Custom Resource (CR) & workload configuration
+
+Now the above is configured, automated build and loading of the out-of-tree `sfc` driver on all AMD Solarflare
+hardware nodes can be easily achieved through the addition the following node label selector in
+your Onload CR and/or workloads:
+
+```yaml
+  selector:
+    feature.node.kubernetes.io/pci-1924.present: "true"
+```
+
+## Footnotes
+
+```yaml
+SPDX-License-Identifier: MIT
+SPDX-FileCopyrightText: (c) Copyright 2023 Advanced Micro Devices, Inc.
+```