Notes: 2026-06-08 CC Virtualization Hypervisor/Kernel subgroup

[Julianus Larson]

Hi, 

Here are the notes from yesterday's meeting. Thanks to everybody who participated.

Best regards,

Julianus

Attendance:

Mark Burton
Bertrand Marquis
Manos Pitsidianakis
Julianus Larson
Edgar Iglesias
Lukas Jünger
Alex Benée 
Bill Mills 
Stefano Stabellini
Grant Likely 
Don Harbin 
Michael
Sumit Semwal
Armelle Laine
Randy Linell 
Ilias Apalodimas 

Quick recap

This was the kickoff meeting for the Core Collective Virtualization Group, where participants discussed their interests and areas of focus across different hypervisors and related technologies. The group included representatives from various organizations including Google, Qualcomm, Machineware, and others, with participants sharing their specific interests in hypervisors like KVM, Xen, and QEMU, as well as related technologies including real-time operating systems, safety certification, and simulation acceleration. Key topics discussed included KVM and PKVM for Android virtualization, Xen architecture support for ARM including VHE and pointer authentication features, simulation acceleration use cases, real-time and safety certification requirements, and the need for better test coverage across different hypervisors. The group agreed to hold future meetings bi-weekly, with the next meeting scheduled for June 29th at the same time slot, and decided to focus future discussions on specific technical topics rather than trying to cover everything in one meeting.

Next steps

Bertrand

• Share meeting notes and a list of proposed discussion topics before the next meeting
• Organize dedicated sessions for specific technical topics (e.g., simulation acceleration, partitioning, etc.) to allow focused discussions outside main meetings

Grant

• Continue work on setting up Causeway infrastructure for the group, including mailing lists and calendaring
• Consider registering/reserving "Core Collective" on GitHub as suggested by Stefano

Julianus

• Send out the invitation for the next meeting (29th June, same time slot) and set up the recurring meeting series

Collaboration

• All interesged parties: Review the proposed topics list (to be sent by Bertrand) and prepare for a live vote at the next meeting to select discussion subjects
• All group members: Identify any hypervisor features or areas that are under-maintained and could be candidates for collective maintenance work
• All group members: Review the upcoming announcement about the new safety committee under the Xen project and consider participation if interested in safety certification topics
• All group members: Consider and provide feedback on the need for public-facing file sharing solutions in addition to Causeway's internal features
• All group members: Raise any needs for additional infrastructure (e.g., GitLab, GitHub) to Grant for consideration

Summary

Hypervisor Technologies and Acceleration Discussion

The group discussed their interests in various hypervisors, with Mark noting his role as chair and Qualcomm's interest in acceleration for simulations, particularly KVM. Edgar highlighted their work on Xen and related projects including virtio, Zephyr, and Android, while also mentioning their use of KVM for test acceleration. Bill expressed interest in real-time safety features and RTOS as a client, specifically for Xen, though noted these features wouldn't apply to KVM or PKVM. Armelle clarified that real-time functionality in KVM and PKVM is handled by the host DOM0 rather than the hypervisor itself, and mentioned Google's focus on PKVM and AVF with emphasis on the Ontario Message transport.

Technical Virtualization Architecture Discussion

The team discussed several technical topics including simulation acceleration use cases, virtualization, and ARM architecture support. Stefano highlighted real-time and safety as key areas of interest, particularly mentioning Zephyr and Unikraft as examples. Michael expressed interest in improving architectural support for Xen on ARM, specifically mentioning VHE and point authentication as initial focus areas, with potential future work on MTE and GCS. The discussion also touched on Android virtualization, which Michael noted might require nested virtualization approaches.

Hypervisor I/O Standardization Discussion

Sumit expressed interest in exploring how different hypervisors handle standardizing I/O interfaces and RTOSes, particularly mentioning Android PKVM. Armelle discussed current enablement work on FFA notifications and message transport, noting specific impacts on the FFA driver. Stefano highlighted the importance of Virtio Message for ARM-specific real-time work and mentioned a new safety committee being formed to address certification artifacts, suggesting potential collaboration opportunities.

Android Virtualization Pathways Discussion

The group discussed potential pathways for Android virtualization, with micpet01 proposing the use of an existing runtime based on KVM and Xen as a potential approach, though this is still under investigation and not a near-term solution. Stefano confirmed that the virtualization part is feasible based on previous architecture studies, though implementation details were not fully explored. The discussion touched on topics including pointer authentication, VHE, and nested virtualization as potential candidates for future development.

VHE Implementation for Xen/KVM

The group discussed VHE implementation for Xen and KVM, with Alex asking about VHE's use case for Xen. Bertrand explained that Xen's model is more natural for ARM virtualization compared to KVM's EL2 execution model. The discussion then moved to architectural enablements like pointer authentication and MTE, where Bertrand indicated that exposing these features to guests is a priority, though internal hypervisor usage remains an open question. Alex explained simulation acceleration concepts, describing how a "trap-me-harder" mode could allow KVM to report all traps back to a VMM for faster emulation, though KVM maintainers have been reluctant to make changes outside server virtualization use cases. Mark suggested exploring a nested case where a new object would sit as a guest inside KVM to enable QEMU simulation.

Hypervisor and Virtualization Technical Discussion

The group discussed several technical topics related to hypervisors and virtualization. Alex presented work on nested hypervisor approaches using TCG emulation and HVF, while Bill raised interest in hard partitioning virtual CPUs for real-time applications. The team agreed to have dedicated sessions for specific technical discussions rather than covering everything in one meeting. They also discussed the need for better test coverage of various hypervisors, particularly for hardware that isn't well supported in CI systems. Grant announced that Causeway would be used for infrastructure including mailing lists and calendaring, though specific details about file sharing and Git/GitHub integration are still being determined. The next meeting was scheduled for June 29th at the same time slot.