Skip to content

openvmm: pin microVM v7.0, reorganized onto upstream main - #1668

Merged
Pedro Henrique Penna (ppenna) merged 2 commits into
devfrom
agents/reorganize-nvx-rebase-openvmm
Nov 16, 2026
Merged

Pedro Henrique Penna (ppenna) merged 2 commits into
devfrom
agents/reorganize-nvx-rebase-openvmm

Conversation

@ppenna

@ppenna Pedro Henrique Penna (ppenna) commented Nov 16, 2026 •

Copy link
Copy Markdown
Contributor

What and why

Status: The release landed on 2026-11-16: feature/microvm/release/v6.1 was first fast-forwarded without force from 73cf40357 to the old development head, fe61f9cbf; then feature/microvm/dev moved to 1b25c5bbc, matching feature/microvm/release/v7.0, with an exact-head lease on fe61f9cbf. All three remote refs were verified after the pushes. This NVX PR merged through the normal merge queue at 23:23:20 UTC as 13b4043bb, after merge-group CI passed; see Landing order. OpenVMM CI of the pin, run 43935939262, passes the six jobs that NVX requires; the SNP VMM test job, which NVX does not require, fails as it does on upstream main, and the TDX one passed on a rerun; see OpenVMM. NVX CI of this head passes (run 43942520775).

Pins openvmm to microVM v7.0: the microVM chain of feature/microvm/dev reorganized onto the latest upstream OpenVMM main as a major release. Upstream landed none of the chain's commits, so the reorganization drops none; three commits conflicted with upstream and were resolved in place, and the generated workflows of one more were regenerated. It then folds the 19 fixups of the v6.1 line into the commits that they fix, moves the two vmcore periodic-timer commits that PR10's RTC fixups need into PR10, and formats the virt_kvm x86-64 tests of six commits, so the 315 commits become 296. The second commit maps the reorganized chain in the OpenVMM upstream roadmap.

OpenVMM source

What changed in OpenVMM

No drops. openvmm_release.py drops finds no chain commit whose patch upstream has, and none of the roadmap's open upstream pull requests merged.

Three conflicts, resolved in place, and one regeneration:

Position Commit Upstream Resolution
046 a1726ac71 virtio_net: persist drained queue state for static-identity NICs microsoft/openvmm#4862 Petri's check for PCIe devices that cannot be saved takes upstream's pcie_device_supports_save_restore, which refuses NVMe, GDMA, and the USB xHCI controllers, and adds the commit's ports whose virtio-net NICs have no save/restore contract
059 d7ce3ef13 openvmm_entry: expose --machine microvm microsoft/openvmm#4862 pub mod microvm; precedes upstream's import of crate::usb::UsbBackend
122 49e738e3b ci: restrict privileged, self-hosted, and publishing jobs to microsoft/openvmm microsoft/openvmm#5070 The fork-gate helper gives every Patina nightly job upstream's needs-patina label condition, and its self-hosted jobs the repository gate too, instead of the per-job hook's condition replacing the fork gate; a new flowey test, patina_nightly_jobs_keep_the_patina_label_gate, checks it on every job; cargo xflowey regen reproduces the workflows

Position 123 (ci: run the checkin gates on the feature/microvm branches) replayed without a conflict, but its generated workflows lacked feature/microvm/** in the Patina nightly workflow's new pull-request trigger, so the reorganization regenerated them with cargo xflowey regen. No other upstream change broke the chain without a textual conflict.

19 fixups, folded into the commits that they fix, each re-expressed for the code that its anchor has:

v6.1 position Fixup v7.0 positions
234 dcc3065ad state_unit: handle a unit in a terminal state instead of panicking 011, 015, 213
235 b62b21163 openvmm_core: tear the VM down when a restart cannot resume it 011
236 e1433b346 state_unit: roll back a partial start without panicking on a lost unit 011, 067, 213
240 459c28110 virtio: keep a held input gate across a device reset 014, 038, 041, 046, 083, 090, 108, 167
241 066b40f14 microVM: refuse a decoded network prefix outside /1 to /30 078, 197
242 1ec79d60f microVM: refuse an unaligned or reversed decoded snapshot RAM range 093, 142
243 67000c096 openvmm_helpers: check paired scratch when opening a snapshot 088, 232
254 93c4f651e virt_kvm: build the LAPIC timer KVM tests for musl 248
259 3cad2668f chipset: keep the PIT restore tests within Linux's open file limit 252
262 55adb5c93 serial_socket, openvmm_entry: resolve user SIDs at upstream's path 105, 112
263 b9a8f7f6e openvmm_core: load microVMs through upstream's MP-table boot module 057, 178
264 7b98adc6d openvmm_entry, openvmm_helpers: read upstream's snapshots through OpenedSnapshot 033
265 5460c8e05 openvmm_entry, openvmm_helpers: record no snapshot ID in microVM captures 071, 121
266 649aeacf0 ttrpc: give the guest exit test's service upstream's SaveVM state 120
267 4a1d9b07a Guide, openvmm_entry: describe upstream's snapshots with the chain's publication 030, 034, 071
268 63688dd58 vmm_tests: save the ttrpc interface test's VM with a VMBus NIC 046
269 73cf40357 openvmm_helpers: drop an independent snapshot copy's page cache 030
271 67ac3e283 chipset: keep the RTC's time and timer phases across a snapshot 018, 019, 052, 195, 243
273 09cd771c0 chipset: bound the RTC's restored and advanced time 018, 019

Two moves. v6.1 positions 261 (vmcore: skip a periodic timer's missed periods at once, microsoft/openvmm#5073) and 270 (vmcore: start a periodic timer at a given deadline, microsoft/openvmm#5077) become positions 016 and 017, ahead of PR10's RTC commit at 018, whose fold of 271 restarts the RTC's periodic timer at its saved deadline through them. Each can still go upstream on its own, in that order.

Formatting. rustfmt rejected the layout of the tests in the virt_kvm x86-64 files of positions 260, 276, 277, 278, 282, and 289, which the v6.1 chain already had: since microsoft/openvmm#4883, vmm_core/virt_kvm/src/arch.rs declares those modules in a cfg_select!, so cargo xtask fmt, like cargo fmt, does not reach them. The six commits take rustfmt's layout.

One test adjustment. Upstream's decode_manifest_without_snapshot_id (microsoft/openvmm#4553) expects snapshot manifest version 6 to be unsupported, which position 131 makes supported for the time ABI, so it failed from v6.1 position 129 to 187. Position 131 has it decode version 7, and position 190, which makes the format's single version 2 and refuses version 6 again, restores version 6, as the head had it.

Positions. Positions 001-015 keep their numbers; v6.1 positions 016-233 are 018-235, 237-239 are 236-238, 244-253 are 239-248, 255-258 are 249-252, 260 is 253, 272 is 254, and 274-315 are 255-296. Later commits that build on a folded change adapt to it: positions 082, 085-087, 102, 110, 158-160, 163, 169, 186, 190, and 225.

What changed in NVX

  • Gitlink commit (d0451895d): moves openvmm from fe61f9cbf to 1b25c5bbc and changes nothing else. What nvx-rs copies from OpenVMM, its vectors, excerpts, contract versions, and CPU profiles, matches the pin, so no coordinated change is needed.
  • Roadmap commit (2deca46bc): remaps doc/openvmm-upstream-roadmap.md onto the new chain: the scope, upstream status, graph, branch DAG, the commits after the mapped tip, which now give each commit's v6.1 position, the PR summary, and the commit mapping. It adds The v7.0 reorganization, which says where each fold went and why, and Conflicts of the v7.0 rebase. The states of the open upstream pull requests, queried again on 2026-11-16, have not changed: three of the twelve that the roadmap names are ready for review, and none has an approving review. Upstream main is three commits past the fork base, none of which changes a file of the chain. The mapped tip is position 126, 227074dc1, and the pin is position 296, 170 commits past it.

Validation

OpenVMM

With Rust 1.97.1, the toolchain of OpenVMM's CI:

  • Per-commit, at each of the 296 commits, on x86_64 Linux and x86_64 Windows: cargo check --locked --workspace --all-targets passes, on Windows without vtpm_util, whose TPM needs OpenSSL and which upstream's Windows jobs exclude too, as do cargo xtask fmt --pass lints, rustfmt --check --edition 2024 on each changed .rs file, and cargo xtask fmt --pass verify-flowey at the four commits that change the pipelines, positions 002, 122, 123, and 181. Linux ran on nvx-kvm-test (Azure Linux 3) for positions 001-099 and 123-173, on nvx-mshv-test (Azure Linux 3) for 100-122 and 174-198, and on Ubuntu under WSL2 for 199-296; Windows ran on Windows 11 for 001-148 and 223-259 and on nvx-whp-test for 149-296.
  • At position 296: cargo xtask fmt passes all of its passes. cargo clippy --locked --workspace --all-targets -- -D warnings passes on Linux (Ubuntu under WSL2) and on Windows (nvx-whp-test), there without vtpm_util and without guest_test_uefi and opentmk_executor, which upstream's Clippy job checks for their UEFI targets alone. cargo nextest run --locked of the 26 packages that the conflicts, folds, and formatting change or build on, each hypervisor's package on the host that has that hypervisor (virt_kvm, virt_mshv, and virt_mshv_vtl on Linux, virt_whp on Windows), passes 1780 of 1780 tests on Linux and 1574 of 1574 on Windows, 22 skipped.
  • VMM tests: cargo xflowey vmm-tests-run --filter "test(microvm)" passes 6 of 6 at position 296 on KVM (Ubuntu under WSL2) and on WHP (nvx-whp-test). MSHV's VMM tests are left to OpenVMM CI.
  • Chain: python scripts\openvmm_release.py check-chain reports 296 commits, no merges, 296 good signatures, and the range diff's counts under Equivalence.
  • OpenVMM CI: run 43935939262, started by the release branch's push, passes all six jobs that NVX requires: the unit tests on x64 Linux and Windows and the VMM tests on Linux KVM, Linux MSHV, Windows Intel, and Windows AMD. Two jobs that NVX does not require failed. run vmm-tests [x64-windows-amd-snp] fails 21 Hyper-V OpenHCL SNP tests of 87; it fails 19-21 of them, a different set each time, on upstream main at the base e6b458cc0 (run 43521036432, 19) and at 5a2fdbbd4 (run 43913518108, 21), and on the previous pin fe61f9cbf (run 43863283839, 20). run vmm-tests [x64-windows-intel-tdx] failed one test of 120 in its first attempt, multiarch::tpm::hyperv_openhcl_uefi_x64_ubuntu_2504_server_x64_tdx_ak_cert_cache: the guest boots, but its pipette never connects to the host (Operation timed out (os error 110)) until the test times out after 722 s. Upstream main at 5b2d16b03 (run 43515746990) failed a sibling test, multiarch::tpm138::hyperv_openhcl_uefi_x64_ubuntu_2504_server_x64_tdx_ak_refresh, the same way, while the job passed on the base and on the previous pin. A rerun of the job on the pin passes, so the failure is intermittent. The run's vmm-perf jobs, which NVX does not require either, pass.

NVX

  • python scripts\nvx.py verify, python scripts\openvmm_pin.py check-roadmap --worktree (296 positions, 298 references), python scripts\openvmm_pin.py check-branch --target origin/dev (standalone, no coordinated files, two verified commits), python scripts\check_markdown_links.py, python scripts\check_test_changes.py --base origin/dev (no changes to existing tests), and git diff --check pass at this head.
  • python scripts\local_gates.py --base origin/dev --run --keep-going, with NVX_RS set, Git's usr\bin on PATH, and XDG_CACHE_HOME unset: test-changes, markdown-links, test-layout, openvmm-vectors (4 passed), openvmm-cpu-profiles (1 passed), and nvx-cli (2246 and 84 Python tests, OK, 88 and 11 skipped) pass at 799f5bfcf, which differs from this head only in the roadmap's upstream status. openvmm-ci is left to CI. The change touches no nvx-rs input, so no nvx-rs gate applies.
  • test-microvm: python scripts/nvx.py test-microvm --backend BACKEND, with OpenVMM built at the pin, passes 57 of 57 scenario runs on WHP (Windows 11, locally, with the gitlink staged) and on KVM (nvx-kvm-test, an Azure Linux 3 virtual machine, in a temporary worktree at d0451895d). On each host, nvx-rs download first installed release v0.1.0-dev.32508f76abc2, and python scripts/nvx.py build-openvmm then replaced its OpenVMM, as the changed SHA-256 shows (WHP 45ea2d1c… to 17a17a24…, KVM 29ebd7e6… to 99b19600…). The time-ABI evidence, which never gates, reports no backward warp and no over-budget CPU on WHP, and one restore over its CPU budget on KVM (restore_cpu_over_budget=1), on a shared virtual machine.
  • Left to NVX CI: the MSHV scenarios, since nvx-mshv-test was running another session's microVM; NVX CI passes them.
  • NVX CI: run 43942520775, of this head, passes: 35 jobs succeed, among them OpenVMM CI of the pinned revision, Required status check, and the microVM tests on KVM, MSHV (both shards, guests and sandboxes, and the debug kernel), and WHP, and the 8 that a pull request's run skips (publication, performance recording, the crate's packaging and publication, the merge-group cancellation, and the KVM and WHP debug-kernel microVM tests) are skipped.
  • Merge-group CI: run 43975731622, at 13b4043bb46551c2faae48fad4c40ce5dc17c3e4, passes with 34 jobs succeeding, 8 skipped, and none failed, including a successful Required status check. The queue merged this exact commit into NVX dev.
  • Not run locally: the AArch64, macOS, TDX, and SEV-SNP configurations, which only OpenVMM CI covers.

Tests modified

None: python scripts\check_test_changes.py --base origin/dev lists no changes to existing tests.

Equivalence

A reorganization that changes commits claims neither tree nor patch equivalence with the previous pin; it rests on the per-commit builds and the head's tests below.

  • Tree: the pin's tree, 52504d232, is the tree of the replay of the 315 commits onto e6b458cc0 without the folds, 07e36f824, apart from the five virt_kvm x86-64 files that the reorganization formats and the feature/microvm/** branch that position 123's regeneration adds to the Patina nightly workflow's pull-request trigger.
  • Patches: git range-diff 7d90c4104..fe61f9cbf e6b458cc0..1b25c5bbc finds the patches of 225 commits identical, the 19 fixups folded, the patches of 58 commits changed by the conflicts, folds, moves, formatting, and adjustment above, and those of 13 more changed only in their context.

Landing order

  1. Done: OpenVMM CI passes NVX's six required jobs on the pin, 1b25c5bbc (run 43935939262).
  2. Done on 2026-11-16: python scripts\openvmm_release.py land --remote origin --old-head fe61f9cbf04ffa323694e8bc798386670842fe92 --new-head 1b25c5bbc550943472bf500efd84bd1378fa3da9 --prev-release feature/microvm/release/v6.1 --push passed. After python scripts\nvx.py verify and python scripts\nvx.py check-openvmm-ci --revision 1b25c5bbc550943472bf500efd84bd1378fa3da9 passed, the landing command fast-forwarded feature/microvm/release/v6.1 from 73cf40357 to fe61f9cbf without force and verified that remote ref, preserving all 46 commits added after the release. Only then did it move feature/microvm/dev from fe61f9cbf to 1b25c5bbc with an exact-head lease. Independent git ls-remote checks confirmed release/v6.1 at fe61f9cbf and both dev and the unchanged release/v7.0 at 1b25c5bbc. [DO NOT REVIEW] cpu_profile: add NVIDIA N1X WHP profile openvmm#5096 and [DO NOT REVIEW] microvm: enable governed ARM64 cold boot on WHP openvmm#5045 now need rebases onto it.
  3. Done on 2026-11-16: gh pr merge 1668 -R microsoft/nvx --match-head-commit 2deca46bcc471aa7bff07e8442a43b56fa8dcf2d queued the PR at 23:02:32 UTC, after step 2, without an admin bypass. Merge-group CI passed, and the normal queue merged it at 23:23:20 UTC as 13b4043bb46551c2faae48fad4c40ce5dc17c3e4. After fetching origin/dev, git merge-base --is-ancestor confirmed that both this merge and the PR head are on dev, and git ls-tree confirmed the pin at both the merge and the current development head.

NVX head: 2deca46bcc471aa7bff07e8442a43b56fa8dcf2d, on dev at 28a0c326f5e.

Pin the openvmm submodule to 1b25c5bbc, the head of the new
feature/microvm/release/v7.0 branch of microsoft/openvmm. That branch
reorganizes the 315-commit microVM chain of feature/microvm/dev,
through the previous pin, fe61f9cbf, into 296 commits on
microsoft/openvmm main at e6b458cc0, eight upstream commits past the
v6.1 base, 7d90c4104.

Upstream landed none of the chain's commits, so the reorganization
drops none. Replaying them conflicted at three positions, which it
resolved in place: 046 and 059 with microsoft/openvmm#4862, upstream's
USB command line, and 122, the fork gates of the CI jobs, with
microsoft/openvmm#5070, which gates the Patina nightly jobs on a label.
Position 123 regenerates its workflows so that the Patina nightly
workflow, which microsoft/openvmm#5070 triggers on pull requests, runs
on feature/microvm/** too. The reorganization folds the 19 fixups of
the v6.1 line into the commits that they fix, each re-expressed for
the code of those commits: v6.1 positions 234-236, 240-243, 254, and
259, which pull requests added after v6.0, the v6.1 rebase's fixups at
262-269, and v6.1 positions 271 and 273, which pull requests added
after v6.1. It moves v6.1 positions 261 and 270, which make a periodic
timer skip its missed periods at once and start at a given deadline,
ahead of the RTC commit as positions 016 and 017, since the RTC's fold
restarts its timers at their saved deadlines through them. Six commits
take rustfmt's layout of their tests in five virt_kvm x86-64 files,
which cargo xtask fmt does not reach since microsoft/openvmm#4883
declares those modules in a cfg_select!.

The pin's tree is that of the replay that keeps the fixups as commits,
07e36f824, apart from those five files and the feature/microvm/**
trigger of the Patina nightly workflow. Every commit of the chain
builds and passes the lints on Linux and Windows.

What nvx-rs copies from OpenVMM, its vectors, excerpts, contract
versions, and CPU profiles, matches the pin.

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
The feature/microvm/release/v7.0 reorganization, which the previous
commit pins, moved the 315 commits of the microVM chain, the 269 of
v6.1 and the 46 that microsoft/openvmm#5077 to microsoft/openvmm#5126
added after it, from 7d90c4104 onto upstream main at e6b458cc0, and
regrouped them into 296: it folded the 19 fixups of the v6.1 line into
the commits that they fix and moved v6.1 positions 261 and 270 into
PR10 as positions 016 and 017. Positions 001-015 keep their numbers,
and the others move: v6.1 positions 016-233 are 018-235, 237-239 are
236-238, 244-253 are 239-248, 255-258 are 249-252, 260 is 253, 272 is
254, and 274-315 are 255-296.

Remap the scope, upstream status, graph, branch DAG, the commits after
the mapped tip, PR summary, and commit mapping onto the new chain, and
name each commit's v6.1 position in the table of commits after the
mapped tip and in the mapping. Describe the reorganization: where each
fixup went, the moved timer commits, the conflicts that the folds
caused and their resolution, the adjustment of upstream's manifest
version test at positions 131 and 190, and the rustfmt layout of the
virt_kvm x86-64 tests, which cargo xtask fmt does not reach. Record
the three conflicts that the replay resolved, with
microsoft/openvmm#4862 and microsoft/openvmm#5070, and the workflows
that it regenerated at position 123, and say which release's positions
the sections on earlier releases count in. Record the per-commit
builds and lints on Linux and Windows and the gates at the head.

As of 2026-11-16, the states of the ten open upstream pull requests,
queried again, have not changed, and upstream main is three commits
past the fork base, none of which changes a file of the chain. The remarks on whether commits apply to upstream main
record what was found on earlier bases; the reorganization did not
check them again.

The pin, 1b25c5bbc, is position 296, 170 commits past the mapped tip.

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Copilot AI balanced review requested due to automatic review settings November 16, 2026 19:44

Copilot AI left a comment

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Copilot wasn't able to review this pull request because it exceeds the maximum diff size. Try reducing the number of changed files and lines, and requesting a review from Copilot again.

@ppenna
Pedro Henrique Penna (ppenna) added this pull request to the merge queue Nov 16, 2026
Merged via the queue into dev with commit 13b4043 Nov 16, 2026
54 of 55 checks passed
@ppenna
Pedro Henrique Penna (ppenna) deleted the agents/reorganize-nvx-rebase-openvmm branch November 16, 2026 23:23
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants