[CP 1658] fix(docs,ci): Radeon device IDs in OpenShift NFD rules + drift guard (GPUOP-1062)

[ci-penbot-01]

cp of pensando/gpu-operator#1658

---

Source PR Description (pensando/gpu-operator#1658):

Summary

Deploying via OLM on OpenShift with Radeon GPUs detects zero GPUs. On an 8x Radeon AI PRO R9700 (0x7551) SNO cluster this failed ~300 tests with No nodes with AMD/GPU found in the cluster (jobd 33207089).

On OpenShift the NodeFeatureRule is created by hand from docs/installation/openshift-olm.md. Both copy-paste YAML blocks in that doc — a NodeFeatureDiscovery CR and a NodeFeatureRule CR — listed Instinct device IDs only, so no node received feature.node.kubernetes.io/amd-gpu and every test bailed at setup in k8_get_gpu_nodes() (tests/pytests/lib/k8_util.py:176).

Fixes GPUOP-1062.

Root cause

The device-ID list is duplicated with nothing enforcing agreement. Commit 9c5ef17 added 0x7551 to the two gpu-nfd-default-rule.yaml copies and left the docs untouched. GPUOP-990 (the 1.5.1 Radeon docs pass) updated framing and the version matrix but never the device lists.

Drift measured: 16 missing amd-gpu IDs and 2 missing amd-vgpu IDs.

Changes

File	Change
docs/installation/openshift-olm.md	+16 amd-gpu / +2 amd-vgpu IDs, both YAML blocks
docs/troubleshooting.md	New section: AMD PCI device visible but no amd-gpu label
hack/check-nfd-device-ids.py	Drift checker (stdlib only)
Makefile	make check-nfd-device-ids
.claude/hooks/check-nfd-drift.sh + .claude/settings.json	PostToolUse hook reporting drift in-session
.github/workflows/nfd-device-id-check.yml	CI, calls the make target

Source of truth is hack/k8s-patch/template-patch/gpu-nfd-default-rule.yaml. make helm-k8s runs rm -rf helm-charts-k8s and repopulates from template-patch, so the helm copy is a build output — drift there means the wrong file was edited.

The workflow carries no branch or path filter: a path-filtered workflow reports no status on unrelated PRs, which stalls merges once it is a required check.

Validation

• Checker run before the doc fix failed with exactly the 16+2 IDs, independently reproducing the manual analysis. It drove the doc edit, so the fix cannot be partial. Passes after.
• Both doc YAML blocks parse via pyyaml, yielding 30 amd-gpu and 11 amd-vgpu entries — the added standalone comments inside the flow sequences do not break copy-paste.
• Hook verified: silent on unrelated edits; reports on drift; catches the extra-ID direction; trips its parse-failure and min_blocks guards; fails open with python3 absent or a malformed payload.
• make check-nfd-device-ids green; settings.json valid JSON; workflow valid YAML; make target unaffected by a missing docker socket.

Docs lint: not runnable locally (markdownlint-cli2 / pyspelling not installed). The identical Markdown passes both docs-lint jobs in CI on the ROCm counterpart PR (#632).

Cherrypick triggered by: ACP-Automation

[ci-penbot-01]

AI-Assisted Cherry-Pick

Source PR: #1658
Target Branch: main

The cherry-pick operation encountered merge conflicts which were resolved automatically using AI assistance.

Files with conflicts (resolved by AI):

• .claude/settings.json:56-61

Original conflict in .claude/settings.json

<<<<<<< HEAD (deleted)
=======
      {
        "matcher": "Write|Edit",
        "hooks": [
          { "type": "command", "command": "${CLAUDE_PROJECT_DIR}/.claude/hooks/gofmt-on-edit.sh" },
          { "type": "command", "command": "${CLAUDE_PROJECT_DIR}/.claude/hooks/check-nfd-drift.sh" }
        ]
      }
>>>>>>> 2aa0eaaa (incoming commit wanted to keep file with new hook)

Cherry-pick triggered by: ACP-Automation