Notes: 2026-07-01 CC Virtualization Simulation subgroup
33 views
Skip to first unread message
Julianus Larson
Jul 2, 2026, 1:27:31 PMJul 2
to CoreCollective Virtualization
Hi,
The meeting focused on discussing QEMU virtualization features and the SplitAccel project, where Alex presented the current state of ARM emulation roadmap and proposed approaches for supporting EL3/EL2 guests under acceleration. The group explored two main technical approaches: nested hypervisor support and the introduction of a new "Qvisor" component to handle EL2-EL3 guest operations, with potential performance benefits including reduced SMMU overhead. Mark expressed strong interest in the Qvisor concept, particularly for Qualcomm's needs, while Arve from Google raised concerns about QEMU's memory model accuracy and cache emulation gaps that could hide architectural bugs. The discussion included considerations about sponsorship structures through Core Collective, with Grant explaining different levels of support available and the process for creating technical proposals with associated funding requirements. The group agreed to take several weeks to further develop the technical specifications and reach out to other potential stakeholders like AWS before presenting a more comprehensive proposal at the next meeting.
Next steps
Alex
KVM Forum Meeting Management
Mark expressed confusion about weekly KVM forum meetings and discussed challenges with calendar management. Alex suggested using the Enable Arm Architecture homepage calendar subscription to keep track of meeting dates. The conversation briefly touched on the distinction between public and internal sprint meetings, and Mark mentioned Qualcomm's use of MS Teams.
Virtualization Working Group Discussion
Mark and Grant agreed to meet on Tuesday at 10 o'clock to discuss the virtualization working group and its current status. Mark acknowledged being more to blame for not following up on their previous discussion, as he had mentioned being away but didn't follow through. Grant expressed concerns about the virtualization working group potentially splitting focus across different topics, though both agreed that coordinating between related initiatives makes sense. The conversation ended with Randy asking Grant about proposed changes to realign within the broader virtualization effort, which Grant had not yet reviewed.
Split accel Platform Proposal Discussion
The group discussed a proposal that gained traction in a previous call, particularly focusing on the concept of SplitAccel and a new platform creation approach. Mark expressed interest in moving forward with this topic, especially since Qualcomm would likely be interested. Alex suggested postponing the detailed presentation to a later meeting due to low attendance, but Mark recommended creating slides and supporting documentation to share internally, suggesting that Qualcomm might be more interested in providing manpower rather than funding.
Core Collective Sponsorship Structure Overview
Grant explained Core Collective's approach to project collaboration and sponsorship structure. He outlined that the organization offers flexible support including marketing, hosting, project management, and engineering resources at different sponsorship levels based on project needs. Grant cautioned against using assigned engineers from member companies due to management challenges, and emphasized that the sponsorship structure would need to cover at least basic costs for marketing and project management.
AWS Working Group Proposal Discussion
The group discussed engaging AWS in a working group, with Grant confirming that while AWS is not yet a member, they would be open to joining if there was a compelling reason. Mark suggested reaching out to both Alex and Mohammed from AWS, as well as involving Phil who is now working with Mark's organization. Grant outlined a process for developing a technical and business proposal, starting with defining the initiative's scope and requirements, followed by Core Collective providing cost estimates and recommended sponsorship structures for membership consideration.
SplitAccel ARM Emulation Project Update
Alex presented an outline of the current status and potential approaches for SplitAccel, a project aimed at improving ARM emulation in QEMU by splitting acceleration between TCG and hypervisor. The discussion covered roadmap updates, including upcoming features like machine learning instructions and permission overlay extensions, as well as challenges with hypervisor support and maintenance needs for HVF. Mark expressed interest in the QVisor approach as it could potentially improve TCG performance and avoid complications with hypervisor integration, suggesting Qualcomm's potential involvement. The group agreed to take several weeks to further develop the SplitAccel concept and reach out to other interested parties like AWS before the next meeting.
Thanks for attending yesterday's CoreCollective Simulation meeting. Here are the notes. Alex'x slides are attached.
Best regards,
Julianus Larson
Attendees:
Bertrand Marquis
Alex Bennée
Mark Burton
Grant Likely
Randy Linnell
Sumit Semwal
ilias Apalodimas
Arve Hjønnevåg
Julianus Larson
Alex Bennée
Mark Burton
Grant Likely
Randy Linnell
Sumit Semwal
ilias Apalodimas
Arve Hjønnevåg
Julianus Larson
Quick recap
The meeting focused on discussing QEMU virtualization features and the SplitAccel project, where Alex presented the current state of ARM emulation roadmap and proposed approaches for supporting EL3/EL2 guests under acceleration. The group explored two main technical approaches: nested hypervisor support and the introduction of a new "Qvisor" component to handle EL2-EL3 guest operations, with potential performance benefits including reduced SMMU overhead. Mark expressed strong interest in the Qvisor concept, particularly for Qualcomm's needs, while Arve from Google raised concerns about QEMU's memory model accuracy and cache emulation gaps that could hide architectural bugs. The discussion included considerations about sponsorship structures through Core Collective, with Grant explaining different levels of support available and the process for creating technical proposals with associated funding requirements. The group agreed to take several weeks to further develop the technical specifications and reach out to other potential stakeholders like AWS before presenting a more comprehensive proposal at the next meeting.
Next steps
Alex
- Share the SplitAccel presentation slides with the group and interested parties.
- Reach out to Mohammed (AWS) and Alexander Graf (AWS) to discuss SplitAccel and QVisor.
- Investigate and follow up with Arve regarding ARM emulation issues related to memory types and cache behavior.
- Work on flashing out the technical proposal for the QVisor/SplitAccel initiative, including effort estimation and potential benefits.
- Prepare a refined version of the SplitAccel presentation for the next meetup.
- Send a calendar invite to Grant for a meeting on Tuesday, 14th at 10:00 (UK time).
- All members: Identify and share any ARM emulation feature gaps or interests for potential collective funding.
KVM Forum Meeting Management
Mark expressed confusion about weekly KVM forum meetings and discussed challenges with calendar management. Alex suggested using the Enable Arm Architecture homepage calendar subscription to keep track of meeting dates. The conversation briefly touched on the distinction between public and internal sprint meetings, and Mark mentioned Qualcomm's use of MS Teams.
Virtualization Working Group Discussion
Mark and Grant agreed to meet on Tuesday at 10 o'clock to discuss the virtualization working group and its current status. Mark acknowledged being more to blame for not following up on their previous discussion, as he had mentioned being away but didn't follow through. Grant expressed concerns about the virtualization working group potentially splitting focus across different topics, though both agreed that coordinating between related initiatives makes sense. The conversation ended with Randy asking Grant about proposed changes to realign within the broader virtualization effort, which Grant had not yet reviewed.
Split accel Platform Proposal Discussion
The group discussed a proposal that gained traction in a previous call, particularly focusing on the concept of SplitAccel and a new platform creation approach. Mark expressed interest in moving forward with this topic, especially since Qualcomm would likely be interested. Alex suggested postponing the detailed presentation to a later meeting due to low attendance, but Mark recommended creating slides and supporting documentation to share internally, suggesting that Qualcomm might be more interested in providing manpower rather than funding.
Core Collective Sponsorship Structure Overview
Grant explained Core Collective's approach to project collaboration and sponsorship structure. He outlined that the organization offers flexible support including marketing, hosting, project management, and engineering resources at different sponsorship levels based on project needs. Grant cautioned against using assigned engineers from member companies due to management challenges, and emphasized that the sponsorship structure would need to cover at least basic costs for marketing and project management.
AWS Working Group Proposal Discussion
The group discussed engaging AWS in a working group, with Grant confirming that while AWS is not yet a member, they would be open to joining if there was a compelling reason. Mark suggested reaching out to both Alex and Mohammed from AWS, as well as involving Phil who is now working with Mark's organization. Grant outlined a process for developing a technical and business proposal, starting with defining the initiative's scope and requirements, followed by Core Collective providing cost estimates and recommended sponsorship structures for membership consideration.
SplitAccel ARM Emulation Project Update
Alex presented an outline of the current status and potential approaches for SplitAccel, a project aimed at improving ARM emulation in QEMU by splitting acceleration between TCG and hypervisor. The discussion covered roadmap updates, including upcoming features like machine learning instructions and permission overlay extensions, as well as challenges with hypervisor support and maintenance needs for HVF. Mark expressed interest in the QVisor approach as it could potentially improve TCG performance and avoid complications with hypervisor integration, suggesting Qualcomm's potential involvement. The group agreed to take several weeks to further develop the SplitAccel concept and reach out to other interested parties like AWS before the next meeting.
Reply all
Reply to author
Forward
0 new messages