HAMi-core (libvgpu.so) is the in-container enforcement component of HAMi and follows the HAMi security policy.

Please do not open a public issue for a security report.

Report privately through HAMi's GitHub Security Advisories and mention HAMi-core in the title. Reports are triaged by the maintainers listed in OWNERS.

Please include:

• A clear description of the vulnerability.
• Steps to reproduce, ideally the CUDA or NVML calls involved.
• The impact, and which tenant or process is affected.
• A suggested fix, if you have one.

HAMi-core carries no tags of its own. It is consumed as the libvgpu submodule of HAMi and released on HAMi's version line, as the projecthami/hamicore image and inside projecthami/hami. Fixes land on main and reach users with the next HAMi release, so the HAMi security policy is what lists which versions receive security fixes.

HAMi-core limits GPU memory and compute for cooperative multi-tenant sharing on a trusted cluster. It is not a hard security boundary against a workload with enough privilege to bypass its own hook, for example by unsetting LD_PRELOAD, using a statically linked binary, or ptrace.

• A workload exceeding its own quota, without affecting another tenant, is a correctness bug. Please open a normal issue for it.
• A workload reaching another tenant's memory, device or namespace that it was not granted is a vulnerability. Please report it privately.

Maintainers aim to acknowledge a report within 5 working days. Weekends, holidays and time zone differences can affect that.