Skip to content

openvmm: pin microVM v7.1, rebased onto upstream main - #1748

Open
Pedro Henrique Penna (ppenna) wants to merge 2 commits into
devfrom
agents/nvx-rebase-openvmm-ready-pr
Open

Pedro Henrique Penna (ppenna) wants to merge 2 commits into
devfrom
agents/nvx-rebase-openvmm-ready-pr

Conversation

@ppenna

Copy link
Copy Markdown
Contributor

What does this pull request change?

This PR pins openvmm to microVM v7.1. That is the 315-commit microVM chain of feature/microvm/dev, rebased as a minor release onto the latest upstream OpenVMM main, plus one fixup at its tip:

NVX's code does not change. The vectors, excerpts, contract versions, and CPU profiles that nvx copies from OpenVMM match the pin.

  • Gitlink commit (9ef5801ac): moves openvmm from 6e5303f94 to 85c3698f9, and nothing else.
  • Roadmap commit (a720ecc40): maps doc/openvmm-upstream-roadmap.md onto the rebased commits. It records the conflicts and the fixup, and the open upstream pull requests as of 2026-11-18: none merged, and none that closed reopened.

Start with OpenVMM pin below, then the roadmap's Conflicts of the v7.1 rebase.

Left out: the release has not landed in OpenVMM yet. feature/microvm/dev still points to the previous pin, 6e5303f94, and feature/microvm/release/v7.0 to 1b25c5bbc; see Landing order. Both must move before this PR merges.

Why is this change needed?

The rebase keeps the chain on current upstream main, so its pull requests to upstream stay reviewable there. It also brings NVX the 14 upstream commits since the v7.0 base, among them:

A minor rebase fits: nothing dropped, the two conflicts are small, and the one break that no conflict exposed is a lock-file entry. A reorganization (v8.0) would fold that fixup into positions 069 and 133 and renumber the roadmap for no change in behavior, so the fold waits for the next one.

A release completes no issue, so this pull request carries the no-issue label.

How was this change validated?

OpenVMM

  • Chain: python scripts\openvmm_release.py check-chain --dir <v7.1 worktree> --new-base 9a875b563 --old-base e6b458cc0 --old-head 6e5303f94 passes. It shows 316 commits, 0 merges, and 316 good SSH signatures. Each replayed commit keeps its author, date, subject, and trailers.
  • Per-commit: these checks pass at the commits that a conflict or the fixup changed, positions 060, 277, and 316, and at position 315, the old head:
    • On x86_64 Linux (WSL2 Ubuntu), openvmm_release.py check-commits --commits ... passes at all four: rustfmt --check on the changed .rs files, cargo check --locked --workspace --all-targets, and cargo xtask fmt --pass lints.
    • On x86_64 Windows 11, cargo check --locked --workspace --all-targets --exclude vtpm_util passes at all four. vtpm_util's vendored OpenSSL needs a native Perl.
    • Also on Windows, cargo xtask fmt --pass lints passes at 060, 277, and 315, and the full cargo xtask fmt passes at 316.
  • At the head, 85c3698f9:
    • With Rust 1.97.1, OpenVMM CI's toolchain, cargo xtask fmt passes all of its passes, verify-flowey among them, on Linux and Windows.
    • Also with 1.97.1, cargo clippy --workspace --all-targets -- -D warnings passes on Linux, and on Windows without vtpm_util, guest_test_uefi, and opentmk_executor. The hosts' default Rust 1.98 fails Clippy only on its new chunks_exact_to_as_chunks lint, in upstream code that the chain does not touch (pal, device_emulators, ucs2).
    • cargo nextest run --profile agent passes 414 of 414 tests on Linux (13 skipped) and 317 of 317 on Windows. It covers the 11 packages that the conflicts and the fixup change, or that build on them: petri, petri-tool, incubator, prep_steps, burette, virt_mshv, openvmm_hypervisors, openvmm_resources, tmk_vmm, cpu_profile, openvmm_helpers.
    • cargo xflowey vmm-tests-run --filter "test(microvm)" passes 7 of 7 on KVM (WSL2 Ubuntu) and 7 of 7 on WHP (Windows 11).
    • Both worktrees stay clean after every command, so no cargo command rewrites Cargo.lock anymore.
  • OpenVMM CI: the release branch's push ran run 44358359786. It passes the six jobs that NVX requires: unit tests [x64-linux], unit tests [x64-windows], and run vmm-tests on x64-linux-amd-kvm, x64-linux-intel-mshv, x64-windows-intel, and x64-windows-amd.
    • It also passes the AArch64, musl, macOS Clippy, TDX, mi-secure, and SNP-on-MSHV jobs, xtask fmt on both platforms, cargo hack, and every build: 30 of its 36 jobs.
    • The five non-required run vmm-perf jobs were still running when this was written.
    • One non-required job failed: run vmm-tests [x64-windows-amd-snp]. 6 of its 88 tests hit petri's 600-second test timeout while booting a Hyper-V OpenHCL SNP guest, and nothing else failed.
    • That job fails the same way on upstream main and on the previous chain heads. Every sampled run fails it: 1 to 4 tests on each of seven upstream main pushes of 2026-11-17 and 2026-11-18, including the new base 9a875b563 (run 44331768548, 2 tests). It also failed 15 tests on the previous pin 6e5303f94 (run 44227888038) and 21 on v7.0's 1b25c5bbc (run 43935939262).
    • Four of the six tests fail in those runs too (snp_boot, snp_boot_single_proc, snp_ak_retry, snp_vmbus_relay_force_mnf). The other two, snp_reboot_into_guest_vsm and snp_vmbus_relay, time out the same way.

NVX

  • python scripts\nvx.py verify passes.
  • python scripts\openvmm_pin.py check-roadmap --worktree passes (316 positions, 319 references).
  • python scripts\openvmm_pin.py check-branch --target origin/dev passes: two verified commits, standalone, no coordinated files.
  • python scripts\check_markdown_links.py and git diff --check pass.
  • python scripts\local_gates.py --base origin/dev --run --keep-going --only test-changes markdown-links test-layout openvmm-vectors openvmm-cpu-profiles nvx-cli passes all six gates. It ran with NVX set to a local release build of nvx and Git's usr\bin on PATH:
    • test-changes: 0 changes to existing tests;
    • openvmm-vectors: 4 passed;
    • openvmm-cpu-profiles: 1 passed;
    • nvx-cli: 2309 and 84 Python tests OK, 89 and 11 skipped.
  • python scripts\nvx.py check-openvmm-ci, the openvmm-ci gate that local_gates.py also selects, passes: it finds the pin's six required jobs passed in run 44358359786.
  • test-microvm: each run used OpenVMM built at the pin with python scripts\nvx.py build-openvmm, over the guest artifacts of release v0.1.0-dev.41b10586a5b0, which nvx download installed. build/openvmm.provenance.json names 85c3698f9 with a clean source, and the binary's SHA-256 changed from the release's.
    • WHP, Windows 11, locally: python scripts\nvx.py test-microvm --backend whp passes 60 of 60 scenario runs.
    • KVM, WSL2 Ubuntu on the same workstation: python3 scripts/nvx.py test-microvm --backend kvm cannot run there, at this pin or at the previous one.
      • --cpu-profile auto selects the built-in intel.alderlake.v1 profile for this CPU, which WSL2's nested KVM cannot boot. The first scenario fails with [E_PROFILE_UNSUPPORTED] the backend does not support CPU profile intel.alderlake.v1: MSR 0x10a bits 8, 17 are not supported.
      • The release's OpenVMM, built from the previous pin 6e5303f94, fails the same cold boot with the same error. vmm_core/cpu_profile, vmm_core/virt_kvm, and vmm_core/virt/src/x86 are identical at both pins. So the failure is the host's limit, which doc/usage.md's CPU profiles section describes, not a regression.
      • On that host, nvx run --hypervisor kvm --cpu-profile host at the pin boots the Alpine guest (ALPINE-MICROVM-BOOT-OK, NVX-GUEST-BOOT-OK), runs a command on its console, and exits 0 through /sbin/nvx-exit 0.

Backend, host OS, host type, and guest:

  • WHP on the local Windows 11 workstation, a physical machine with an Intel Core i9-12900H. Every scenario ran on the built-in intel.alderlake.v1 profile: no run printed the NVX-CPU-PROFILE-FALLBACK warning.
  • KVM in WSL2 Ubuntu, a virtual machine on that workstation, where only the host profile boots.
  • The guests are NVX's Alpine and Ubuntu guests of that release, and OpenVMM's own test guests for the VMM tests.

Left to CI:

  • NVX's test-microvm on KVM and MSHV, on the CI runners, whose CPUs built-in profiles serve. The MSHV test host was unreachable; OpenVMM CI's MSHV jobs pass.
  • Benchmarks.

Checklist

  • Tests cover the new or fixed behavior, in the layout that
    python3 scripts/check_test_layout.py checks.
  • python3 scripts/check_test_changes.py --base origin/dev lists no change
    to an existing test, or a Changes to existing tests section below
    explains each one.
  • The documentation that the change affects is updated, including
    .github/copilot-instructions.md and the skills when a convention or
    workflow changes.
  • A change to the public API of nvx updates nvx-rs/public-api/ with
    python3 scripts/public_api.py --update.
  • Snapshots, protocols, and on-disk formats stay compatible, or the
    description says how the change versions or migrates them.

The pin changes no NVX code, tests, or public API. The roadmap is the documentation it affects. The snapshot manifest, contract versions, and vectors that nvx checks match the pin.

OpenVMM pin

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.

Two conflicts, resolved in place:

Position Commit Upstream Resolution
060 65258e052 petri, vmm_tests: add a microVM builder, portb harness, and cold-boot test microsoft/openvmm#4941 uses_pipette_as_init keeps upstream's condition (non-OpenHCL Linux direct boot, or an MSHV base initrd). It applies the chain's microvm::uses_pipette_as_init gate, which admits only the standard machine profile, to both, before it requires a Pipette binary.
277 eaf039eed virt: save a guest's user-wait, debug, and feature MSRs microsoft/openvmm#4942 virt_mshv's ErrorInner keeps upstream's InvalidVpRegisterValue and InvalidVpRegisterMessage, followed by the chain's x86-64 UnsupportedState.

One fixup at the tip, to fold into the commits that it names at the next reorganization:

Position Fixup Upstream Fold into
316 85c3698f9 cpu_profile, openvmm_helpers: name sha2 as upstream's lock file does microsoft/openvmm#4941 069, 133

Here is why the fixup is needed:

  • vmm_tests: boot an x64 Linux child through a native MSHV root openvmm#4941 removed sha2 0.10 from Cargo.lock, which leaves 0.11.0 the only version, so upstream's lock file now names sha2 without a version.
  • Positions 069 (openvmm_helpers: define the authoritative microVM machine contract) and 133 (cpu_profile: fingerprint backend CPU surfaces and pin CPU profiles) add sha2 to their crates' dependencies. Their entries still name it sha2 0.11.0.
  • cargo check --locked accepts that, because it compares the resolves rather than the text. But any cargo command that may update the lock file rewrites the two lines and leaves the checkout dirty. The replay showed this in both release worktrees.

Upstream's other 12 commits apply without conflicts.

What changed in NVX

The gitlink and the roadmap, as What does this pull request change? describes. The comparisons of what nvx copies from OpenVMM pass at the pin: cargo test --test openvmm_vectors -- --ignored (4 passed) and --test openvmm_cpu_profiles (1 passed). So the gitlink commit carries no coordinated change.

Equivalence

  • Patches: git range-diff e6b458cc0..6e5303f94 9a875b563..85c3698f9 pairs all 315 replayed commits in order:

    • 312 have identical patches;
    • two differ by their resolutions (060, 277);
    • one differs only in context (007, virt_mshv: make RUN_VP cancellation race-free).

    The fixup, 316, is new. No commit is dropped or reordered.

  • Tree: git merge-tree --write-tree 6e5303f94 9a875b563 is a three-way textual merge of upstream's 14 commits into the previous pin. It conflicts in the two files of the resolutions. The pin's tree differs from it in seven files:

    • those two files;
    • Cargo.lock, from the fixup;
    • the four generated openvmm-* workflows. .gitattributes merges them with merge=flowey-theirs, so merge-tree takes upstream's copies and drops the chain's changes. The pin's copies carry upstream's changes and the chain's together, and the verify-flowey pass of cargo xtask fmt accepts them.

Roadmap positions

Every commit keeps its v7.0 position. The mapped tip is still position 126, c928e8dc1. The pin is position 316, 190 commits past it, and positions 127-301 and 303-316 are not yet assigned to roadmap PRs.

Landing order

  1. Done: OpenVMM CI passed NVX's six required jobs on the pin (run 44358359786), and python scripts\nvx.py check-openvmm-ci passes.

  2. Pending: fast-forward feature/microvm/release/v7.0 from 1b25c5bbc to the previous pin, 6e5303f94, without force. Every revision of the old line that NVX pinned then stays on a branch that git submodule update of an older NVX commit can fetch.

  3. Pending: move feature/microvm/dev from 6e5303f94 to the pin, 85c3698f9, with a lease on 6e5303f94, because .gitmodules tracks that branch. Steps 2 and 3 are one command, which waits for a maintainer's go-ahead:

    python scripts/openvmm_release.py land --remote origin --old-head 6e5303f9457b1d1f2dc45ced689ab09311325f1f --new-head 85c3698f94e8c8308ac1b09e99b31b17ae50a037 --prev-release feature/microvm/release/v7.0 --push

    It stops if feature/microvm/dev moved in the meantime. Then the new commits go onto the release, and the release is validated, published, and pinned again.

  4. Pending: this PR merges through the merge queue once steps 2 and 3 are done and a maintainer approves it.

NVX head: a720ecc40d6e9b3a70b34a549cf78837725de719, on dev at 4c1bfa09ed9.

Pin the openvmm submodule to 85c3698f9, the head of the new
feature/microvm/release/v7.1 branch of microsoft/openvmm. That branch
rebases the 315-commit microVM chain of feature/microvm/dev, through
the previous pin, 6e5303f94, from e6b458cc0 onto microsoft/openvmm main
at 9a875b563, 14 upstream commits later.

Upstream landed none of the chain's commits, so the rebase drops none,
and every commit keeps its position. Replaying them conflicts at two.
At position 060, microsoft/openvmm#4941 lets Petri boot an x64 Linux
child through a native MSHV root with Pipette as its init, and the
chain's microVM builder still admits Pipette as init only for the
standard machine profile. At position 277, the invalid VP register
errors of microsoft/openvmm#4942 join the chain's unsupported-state
error in virt_mshv. A fixup at the tip, position 316, names sha2
without a version in the lock file entries that positions 069 and 133
add, since microsoft/openvmm#4941 left sha2 one version, and any cargo
command that may update Cargo.lock otherwise rewrites them. The patches
of the other 313 replayed commits are unchanged, apart from the context
of position 007.

The vectors, excerpts, contract versions, and CPU profiles that nvx
copies from OpenVMM match the pin.

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
The feature/microvm/release/v7.1 rebase, which the previous commit
pins, moved the 315-commit microVM chain from e6b458cc0 onto upstream
main at 9a875b563 without dropping any commit, so every commit keeps
its v7.0 position. Remap the scope, upstream status, graph, branch
DAG, the commits after the mapped tip, PR summary, and commit mapping
onto the rebased commits. Record the conflicts that the rebase
resolved at positions 060 and 277, with microsoft/openvmm#4941 and
microsoft/openvmm#4942, and the fixup at position 316, which folds
into positions 069 and 133, and say that upstream main has no commits
past the new base.

As of 2026-11-18, none of the open upstream pull requests has merged,
and none that was closed has reopened. The remarks on whether commits
apply to upstream main record what was found on earlier bases; the
rebase did not check them again.

The pin, 85c3698f9, is position 316, 190 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 18, 2026 18:22
@ppenna Pedro Henrique Penna (ppenna) added the no-issue Needs no issue reference; applied by a maintainer label Nov 18, 2026

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.

🟡 Changes recommended

The roadmap omits deferred upstream commit 715927303 and incorrectly reports no post-base work, leaving the remaining rebase work undocumented.

1 open issue
What changed in this PR

This PR updates NVX’s OpenVMM pin to microVM v7.1, bringing in 14 upstream commits while preserving the existing microVM chain. NVX application code and public APIs remain unchanged.

Changes:

  • Rebases all 315 chain commits without drops or reordering, resolving two conflicts and adding one lock-file fixup.
  • Remaps the roadmap’s commit hashes, base, tip, and pin while preserving existing positions.
  • Documents the conflict resolutions and where the new fixup should be folded in a future reorganization.
File Description
doc/​openvmm-upstream-roadmap.md Updates v7.1 commit mappings and records rebase conflicts and the lock-file fixup.

Review effort: Balanced

Comment on lines +1207 to +1212
Upstream main has no commits past the fork base: as of 2026-11-18, its head is the fork base,
`9a875b563`. Of the 14 commits that upstream main had past the v7.0 base, `e6b458cc0`, the v7.1
rebase resolved `9b05d16fd` (microsoft/openvmm#4941) at position 060 and `20582cf41`
(microsoft/openvmm#4942) at position 277; see
[Conflicts of the v7.1 rebase](#conflicts-of-the-v71-rebase). The other 12 merged with the chain
without conflicts.

This branch has not been deployed

No deployments
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

no-issue Needs no issue reference; applied by a maintainer

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants