Problem description
In a specific Kylin V10 4.19 kernel, if an AMD GPU has been loaded by the host amdgpu driver, the host may become abnormal in certain environments when the device is unbound for GPU passthrough.
This behavior is related to a specific combination of the Kylin kernel and AMDGPU driver. It is recommended to avoid deploying AMD GPU passthrough workloads in this environment and to use a kernel and driver combination that has passed compatibility validation.
Symptoms
- When an AMD GPU is passed through to a VM instance or the device is unbound, the host suddenly restarts or becomes disconnected.
- The confirmed reproduction environment is Kylin V10 with kernel version
4.19.90-52.15.v2207.ky10.x86_64.
Impact
A host abnormality may affect VM instances running on the host and may trigger high availability or service recovery processes. It is recommended to complete compatibility confirmation before deployment and avoid directly validating this scenario in a production environment.
Workaround
- Before deploying AMD GPU passthrough, check the compatibility matrix for ZStack Cloud, the Kylin kernel, and the AMDGPU driver.
- For a host that matches the Kylin 4.19 kernel combination described above, it is recommended to avoid deploying AMD GPU passthrough workloads and testing GPU driver unbinding in a production environment.
- Use only a kernel and driver combination that the Kylin vendor confirms supports unbinding and passthrough for the target AMD GPU.
- If AMD GPU passthrough is required, contact ZStack technical support and the Kylin vendor to jointly confirm a kernel upgrade plan. Complete stress testing and host restart validation in an isolated environment first.
It is not recommended to independently install a third-party kernel reset module, manually unload the AMDGPU driver, or modify VFIO boot parameters. These operations involve the host kernel and boot configuration and must be implemented only after assessment by technical support.
Affected products
- Environments running ZStack Cloud 4.8.x, 5.4.x, or 5.5.x that use the same combination of the Kylin kernel and AMDGPU driver.
- Confirmed environment: Kylin V10, kernel
4.19.90-52.15.v2207.ky10.x86_64, and AMD GPU passthrough.
Fix status
- Current status: Confirmed; a workaround is available.
- Fixed version: A compatible kernel and driver version has not been identified.
- Related patch/fix package: No separate fix package has been confirmed for this kernel combination.