Re: [Hypervisor Live Update] Notes from June 15 and June 29, 2026

[Pasha Tatashin <pasha.tatashin@soleen.com>]

Adding missing CCs to the mailing lists.

On 07-03 20:43, Pasha Tatashin wrote:
> Hi everybody,
> 
> Here are the combined notes from our last two Hypervisor Live Update
> calls, held on June 15 and June 29. These notes are intended to bring
> those who could not attend up to speed and to keep the momentum going
> between meetings.
> 
> ----->o-----
> LPC 2026 Call for Proposals
> 
> The Call for Proposals for the Live Update Microconference at LPC 2026
> in Prague is open. Please make sure to submit your topics and abstracts
> before the upcoming deadline on July 24th.
> 
> https://lore.kernel.org/all/ahcc3Qyuy7Oy03Iq@plex
> 
> ----->o-----
> VFIO & PCIe Persistence Roadmap
> 
> * Roadmap Recap (June 15): Vipin Sharma outlined the five-milestone
>   roadmap for preserving PCIe endpoint functions (and upstream bridges)
>   across live updates without disrupting registers or ongoing DMA.
> 
> * Status Update (June 29): Vipin has addressed feedback on the v4
>   submission. Revision v5 is currently undergoing internal validation
>   and is on track for the mailing list before our next sync.
> 
> PCIe Core Updates
> 
> * v6/v7 Evolution: David Matlack is preparing the v7 PCI series. The
>   group aligned on deploying an FLB reference-counting model to maintain
>   symmetry for incoming/outgoing sessions across kexec.
> 
> * Pratyush Yadav suggested that drivers should explicitly isolate
>   structural "probe" routines from "retrieve" logic and separate
>   "shutdown" from "finish" callbacks. David is evaluating the
>   feasibility of deferring probes until the retrieve phase to improve
>   driver cleanliness.
> 
> * David is refactoring the PCI series to utilize KHO blocks.
> 
> IOMMU & Dependency Management
> 
> * v3 Status: Samiullah has posted v3 of the IOMMU persistence series,
>   addressing hot-plug handling and context entry preservation.
> 
> * Dependency Policy: The group agreed that the Live Update framework
>   will not enforce a hardcoded preservation sequence. Instead,
>   individual subsystem file handlers will verify dependency states
>   locally during their freeze/finish callbacks.
> 
> LUO, KHO
> 
> * Terminology Updates: "Scratch" memory is now designated as
>   "unpreserved memory" (memblock perspective). KHO "Scratch" is renamed
>   to "KHO boot memory".
> 
> * LUO Limits: Limits on the number of sessions and preserved files have
>   been removed and landed in upstream Linux.
> 
> Guest Memory (guest_memfd)
> 
> * Naming Conventions: Regarding the KVM file binding debate, the team is
>   moving away from the debugfs-based incremented value approach (due to
>   production compilation concerns). We are exploring using unique names
>   derived from user-space file-handler compatibility strings. Tarun will
>   escalate this to the KVM maintainers.
> 
> ----->o-----
> Action Items & Next Steps
> 
> Pasha: Consult with Mike regarding KHO component splitting without
> versioning support.
> 
> David M: Send next version of the FLB ref-count patch; refactor PCI
> series to use KHO blocks; share recording link from June 29.
> 
> Pratyush: Implement rename of KHO scratch/boot mem.
> 
> Vipin: Submit VFIO v5.
> 
> Group: Review Guest Memory Preservation V3 series (Pasha/David).
> 
> ----->o-----
> 
> Next meeting will be on Monday, July 13 at 8am PDT (UTC-7), everybody is
> welcome!
> 
> Join Here: https://meet.google.com/rjn-dmzu-hgq
> 
> Please let me know if you'd like to propose additional topics for
> discussion, thank you!
> 
> Best regards,
> Pasha