Intel-Gpu-Levelzero: Failed to get Indices from levelZero on WSL2 on Intel Lunar Lake Core Ultra 9 processor

[shailesh837]

Describe the bug
We have Customer called CelebalTech based in Dubai, Finland, Bangalore, Targeting there Customer contact Call center Framework running on AIPC running single node rke2 kuberenetes cluster using OPEA-ERAG based RAG solution.
WSL2 : Ubuntu – 24.04
Document I used to reproduce the issue.

Currently we have the Latest the whole OPEA-Based RAG running usign docker-compose on LLM Model : llama3.2 running on iGPU (/dev/dri) exposure for iGPU (Windows driver : version):
After applying side-car (WSL2 : Ubuntu 24-04 gpu device plugin) there is level zero crash or failure inside pod:

To Reproduce
Steps to reproduce the behavior:

1. RKE2 version 1.31.4 Installtion
    sudo curl -sfL https://get.rke2.io | sudo INSTALL_RKE2_VERSION=v1.31.4+rke2r1 sh -
2. Enable the rke2-server service
    systemctl enable rke2-server.service
3. Start the service
    systemctl start rke2-server.service
4. Follow the logs, if you like
     journalctl -u rke2-server –f

5. Replace canal cni with calico
   Step 1: Edit this file /etc/rancher/rke2/config.yaml and add cni: calico
   Step 2: restart daemonsets and rke2 service
   sudo systemctl daemon-reload
   sudo systemctl restart rke2-service
   WSL layer enables Intel GPU detection with WSL (Windows Subsystem for Linux) Kubernetes clusters.
   kubectl apply -k 'https://github.com/intel/intel-device-plugins-for-kubernetes/deployments/gpu_plugin/overlays/wsl?ref=v0.32.0'

Expected behavior
A clear and concise description of what you expected to happen.
Intel-gpu-plugin should detect levelzero indices and have gpu pod running under rke2 cluster

Screenshots
If applicable, add screenshots to help explain your problem.

System (please complete the following information):

• OS version: WSL2 : Ubuntu 24.04
• Kernel version: microsoft Linux 5.15
• Device plugins version: v0.32.0
• Hardware info: Core Ultra 7 288V

[eero-t]

Potentially related: intel/compute-runtime#721 (driver segfault fix merged, but not yet in any release)

[tkatila]

@eero-t it's not. LunarLake seems to require the igc packages that we have dropped by force in the containers. I have debugged the issue with @shailesh837 today.

[shailesh837]

@tkatila : : Please let me know when we can have new intel-gpu-levelzero container with this changes in intel docker hub, As its urgently required, we have tested it in UAT with customer the local version we created , But now we need to test on multiple customer machines.

[tkatila]

@shailesh837 it's in progress. We ran into some issues with our containers which we need to fix first.

[tkatila]

@shailesh837 the fix was merged yesterday. There is a container available here:
docker.io/intel/intel-gpu-levelzero@sha256:2326d1fb56c93388239786ef8bb7152c3d37dd9ed04a165131211d98d7128831.

Could you verify that it still works with the LunarLake device? If it does, we'll include the fix in our 0.32.1 release.

[tkatila]

Noting it also here that @shailesh837 tested the container with the LunarLake device and the detection succeeded.