[Hypervisor Live Update] Notes from July 13, 2026

[Hypervisor Live Update] Notes from July 13, 2026

[Pasha Tatashin]

Hi everybody,

Here are the notes from our Hypervisor Live Update call held on July 13.
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 & Registration

The Call for Proposals for the Live Update Microconference at LPC 2026
in Prague remains open. Please submit your topics and abstracts before
the July 24th deadline. Pre-registration is currently open for content
submitters. General registration will open on July 24th; early
registration is highly encouraged as capacity is limited.

https://lore.kernel.org/all/ahcc3Qyuy7Oy03Iq@plex

----->o-----
VFIO Persistence

* VFIO v5 Status: Vipin Sharma's v5 series is ready, rebased on top of
  David Matlack's PCI v7. It is expected to be posted to the mailing
  list shortly.

* Finish Call Cleanup: The group discussed Vipin's proposal to clean up
  KHO state and reset the device if a finish call fails. The consensus
  is to keep it simple for the initial version: if retrieve/finish
  fails, return false and leak the resources. This is safer than
  attempting a partial or unsafe cleanup.

----->o-----
PCIe Persistence

* PCI v7: David Matlack sent out the v7 PCI series. Key changes include
  moving to a KHO block allocation model (instead of guessing max
  devices), which allows dynamic allocation. Additionally, retrieve
  failures are now treated as fatal (triggering kernel panic) to avoid
  memory corruption in dependent subsystems.
* Landing Strategy: The initial series will likely go through the PCI
  tree (pending discussion with Bjorn) to establish maintainership.
* Next Steps: David's PCI series will be applied to the next branch once
  the FLB fixes land, allowing him to rebase (v8).

----->o-----
IOMMU Persistence

* IOMMU v4 Status: Samiullah Khawaja is working on v4, incorporating
  feedback and KUnit tests. Landing is blocked on the PCIe and VFIO
  series.
* Retry Logic: Samiullah proposed allowing retries for file handlers
  (e.g., under memory pressure). The group agreed retries should be
  allowed on a case-by-case basis for specific recoverable errors,
  rather than as a generic mechanism. Samiullah will propose a patch
  series for this.

----->o-----
KHO, LUO & Selftests

* LoongArch KHO Support: v3 was posted by George Guo using device tree.
  The implementation needs refinement. The team is waiting for LoongArch
  maintainers to clarify their preferred boot protocol (command line vs.
  device tree) and confirm real use cases.
* KHO Scratch Extension & Rename: Pratyush Yadav sent v3 for scratch
  extension. He is working on v4 to address feedback and simplify
  conflict resolution with the MM tree. The rename of scratch/boot
  memory will follow the extension landing.
* LUO Fixes: David Matlack's FLB fixes (PATCH 0/2) have been applied to
  liveupdate/next.
* Selftests: Mike Rapoport's end-to-end test VM infrastructure (v4) has
  been applied.
* KHO Component Splitting: Pasha still plans to split KHO into
  components, pending discussion with Mike.
* Directory Location: The group agreed KHO Red X3 will remain in the KHO
  directory to maintain binary format stabilization guarantees.

----->o-----
Guest Memory (guest_memfd)

* guest_memfd Preservation: Tarun Sahu posted v3 and is working on v4.
* In-Host Migration: The team discussed implications of supporting
  multiple files per inode (needed for in-host migration, where a new VM
  starts before the old one dies). Currently guest_memfd operations are
  per-inode, posing corruption risks if accessed simultaneously. For
  now, this scenario might be marked as incompatible with live update if
  resolution is too complex.
* Submission Standard: Pasha reminded contributors to include cover
  letters with their patch series.

----->o-----
HugeTLB

* Status: On hold. Pratyush needs to land the extended scratch and KHO
  Xarray before resuming work on HugeTLB.

----->o-----
Action Items & Next Steps

Pasha:
* Review PCI v7 series.
* Consult with Mike on KHO component splitting.

David M:
* Rebase PCI series on top of liveupdate/next (v8).
* Consult with Bjorn on PCI landing tree strategy.

Pratyush:
* Follow up with LoongArch maintainers regarding boot protocol and use
  cases.
* Work on KHO scratch extension v4 and subsequent rename.

Vipin:
* Submit VFIO v5.

Samiullah:
* Work on IOMMU v4.

Tarun:
* Post Guestmemfd v4 (include cover letters).

Jork:
* Discuss resource accounting and session handover (systemd integration)
  with Luke.

----->o-----

Next meeting will be on Monday, July 27 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