guest: apply the DNS server and listener announcements of a policy replaced at an untiered restore by ppenna · Pull Request #1072 · microsoft/nvx

Pedro Henrique Penna (ppenna) · 2026-10-31T02:26:46Z

Fixes #943. Fixes #942. Pins microsoft/openvmm#4831 at f1eb78302a296f455a99f8ac622a46f4c0862908. That pin lets the restore of a NIC template without a tier install a network policy of its own, and hands that restore's guest the new policy's DNS server and listener announcements in version 5 of the restore packet. On the NVX side:

the guest reads version 5 and applies the settings before it acknowledges the restore;
nvx-rs run --restore-network-policy asks for such a restore;
a new microVM scenario restores one untiered NIC template with four policies on KVM, MSHV, and WHP.

Based on dev at 2d1edb3fd. microsoft/openvmm#4831 is two commits directly on 3d7512af6, the head of feature/microvm/dev and the revision that dev pins. Its CI is green.

Change

Commit	Files	Change
`fb4aef9a1` openvmm	`openvmm`	`3d7512af6` → `f1eb78302`: the two commits of microsoft/openvmm#4831 (gitlink only)
`77845c67b` doc	`doc/openvmm-upstream-roadmap.md`	Records them as positions 373 and 374, moves the NVX pin, and records that microsoft/openvmm#4816 merged positions 371 and 372
`9cc72095f` nvx-rs	`nvx-rs/src/guest/ingress/*`, `nvx-rs/tests/guest_tools.rs`, `nvx-rs/README.md`	`nvx-ingress start GATEWAY_IPV4` starts the announcer as `Announcer::start` does: it binds the privileged source port, runs `nvx-ingress inherit` on it, and exits once that runs. It exits 1 if the announcer does not start
`cd4e2512e` guest	`nvx-rs/src/guest/time/*`, `nvx-rs/src/contract.rs`, `guest/common/nvx-snapshot`, guest and vector tests, design docs 3.1, 3.5, 5.5	`nvx-time` decodes restore packet versions 4 and 5. `nvx-snapshot` applies the DNS server and listener announcements in guest repair step 4, before the acknowledgement
`a2cc8211b` nvx-rs	`nvx-rs/src/host/cli/run.rs`, `nvx-rs/src/testing/fake_openvmm.rs`, CLI tests, `doc/run.md`, `doc/usage.md`, `nvx-rs/README.md`	`nvx-rs run --restore-network-policy` passes OpenVMM's option. The fake OpenVMM mirrors OpenVMM's refusals
`82a9f9cab` scripts	`scripts/nvx_tools/microvm_tests.py`, `untiered-network-policy.sh.in`, `scripts/test_microvm_tests.py`, `microvm_shards.json`, `doc/ci.md`	New default-suite scenario `untiered-network-policy`; shards replanned with its measured durations

OpenVMM changes (microsoft/openvmm#4831)

An untiered NIC template (e4fd1655a, OpenVMM prerequisite: a new network policy at the restore of a NIC template without a tier #942). --restore-network-policy now admits a snapshot without a tier as well as a platform snapshot, and still refuses workload-start and instance-checkpoint snapshots.
- Such a snapshot may hold a workload that has already run. Under the template's complete deny-all, that workload had no connectivity, so nothing in the snapshot depends on the policy.
- Everything else of [DONT REVIEW] microVM: let a platform restore install a new network policy on a NIC template openvmm#4767's rule stays. The NIC must be a complete deny-all template. The policy is validated as at a cold boot, bound to the captured identity, and installed with its forwards before any vCPU runs.
Restore packet version 5 (f1eb78302, guest: apply the DNS server and listener announcements of a replaced policy after a non-sandbox restore #943). A platform clone learns its DNS server and listener announcements from its activation. The guest of an untiered snapshot gets no activation, and its captured command line names no DNS server and starts no nvx-ingress.
- At an untiered restore with --restore-network-policy, OpenVMM writes packet version 5. Byte 7, reserved in version 4, becomes the network byte: bits 0-1 give the DNS family (none, IPv4, or IPv6), bit 2 asks for listener announcements, and the other bits are zero. A 16-byte DNS server follows the entropy: IPv4 plus 12 zero bytes, IPv6, or zeros.
- The DNS server is the policy's gateway_dns_server(), what a cold boot would pass as virtnet_dns=. The announcement is set exactly when the policy publishes listeners.
- The restore always requires the guest's acknowledgement, so host input stays gated until the guest has applied them.
- Every other restore, platform clones and ttrpc restores included, keeps version 4. A guest that reads only version 4 fails just such a restore instead of ignoring its settings.

NVX changes

Guest decoding (nvx-rs/src/guest/time). nvx-time accepts versions 4 and 5. It refuses a network byte with family 3 or unknown bits, and a DNS record that does not match its family or is unspecified. It exposes RestoreNetwork on TimeAbi's Restore.
- nvx-time does not finish a packet whose settings ask for something itself, just as for processors or memory, so nvx-snapshot can apply them before the acknowledgement.
- nvx::contract::RESTORE_PACKET_NETWORK_VERSION (5) is checked against OpenVMM by tests/openvmm_vectors.rs. RESTORE_PACKET_VERSION and the template key stay at 4, since clones never see version 5.
Guest repair (guest/common/nvx-snapshot). Step 4 now also:
- writes nameserver ADDR to /etc/resolv.conf;
- runs /sbin/nvx-ingress start with virtnet_gw=, which returns only once the announcer's socket is bound to a privileged source port and its program runs.
A failure of either fails the restore, with status 1, before the acknowledgement. A restore whose settings ask for nothing stays quiet, and an untiered restore still leaves entropy and identity to its caller.
nvx-rs run --restore-network-policy. It requires --restore-snapshot and passes OpenVMM's option, so the restore installs the policy of the network options, ingress, explicit host loopback, and forwards included. The option stays explicit because OpenVMM refuses it for a snapshot that is not a NIC template.
Docs: doc/run.md (Networked snapshots), doc/usage.md (run table), design 3.1, 3.5.6, 5.5.5, and 5.5.7, nvx-rs/README.md, doc/ci.md.

Tests

untiered-network-policy captures an untiered NIC template in non-sandbox mode, on 192.0.2.0/24, and restores it four times with --restore-network-policy: the template's own deny-all; egress rules admitting DNS over IPv4 and one host service; explicit host loopback and a forward, with egress rules admitting DNS over IPv6 alone; and private-network ingress with allowed egress. For each restore:
- The guest's resolver must name exactly the policy's DNS server.
- nvx-ingress must run exactly when the policy publishes listeners.
- The guest must reach exactly the admitted host services.
- The host must reach the guest's server through the forward, and through the published port: on loopback alone for host-loopback access, which no other host address reaches, or through a host address other than loopback for private-network ingress.
OpenVMM must also refuse a restore that keeps the template's policy but passes another, and one that passes ingress without the option. No restore may change the template.
Unit tests:
- v5 decoding and malformed network settings;
- the nvx-time packet hook;
- nvx-ingress start's usage errors and startup failure;
- nvx-snapshot's network stage, ordering against the acknowledgement, and failures, including an announcer that does not start;
- the scenario's matrix, reports, simulated run, and failures, including loopback-only private ingress and host-loopback publication on another address;
- the CLI's and the fake OpenVMM's acceptance and refusals.
[DO NOT REVIEW] microVM: let an untiered restore install a new network policy on a NIC template openvmm#4831 adds unit and golden tests for the tier check, packet v5 encoding and decoding, and the network settings of ten policies.

Evidence

Hosts: nvx-kvm-test (Linux/KVM, Azure Linux 3, kernel 6.6), nvx-mshv-test (Linux/MSHV, Azure Linux 3, kernel 6.18.34.mshv3), and nvx-whp-test (Windows/WHP, build 26100), all virtual machines. Each ran in a separate worktree, with OpenVMM built there from a clean f1eb78302 and the other artifacts of development release v0.1.0-dev.bad440220c67. Each Linux host rebuilt the Alpine initramfs from this branch, and the WHP host used the KVM host's.

Check	KVM	MSHV	WHP
`untiered-network-policy` at `50d841d6c` (= head but the shard plan)	Passes (14.0 s)	Passes (14.1 s)	Passes (18.8 s)
Default microVM suite at `816d6bf47`¹	46/46 scenarios, 52 executions	46/46 scenarios, 52 executions	46/46 scenarios, 52 executions

¹ 816d6bf47 differs from the head only in the new scenario's host-side checks and template subnet, their tests and docs, and the shard plan. Guest, OpenVMM, and every other scenario are identical.

Every restore reported the same outcomes on every backend. On KVM:

Restore	Policy at restore	Guest report	Repair	Inbound exchanges
`deny-all`	`--network-egress deny --host-loopback deny`	`resolver=none announcer=none allowed=denied other=denied`	quiet	0
`egress-dns`	deny + `192.0.2.1/32:udp:53`, `192.0.2.1/32:tcp:ALLOWED`	`resolver=192.0.2.1 announcer=none allowed=reached other=denied`	`NVX-POST-RESTORE-OK: tier=legacy`	0
`host-loopback`	deny + `fd00::c000:201/128:udp:53`, `--host-loopback allow`, forward	`resolver=fd00::c000:201 announcer=running allowed=denied other=denied`	`NVX-POST-RESTORE-OK: tier=legacy`	2 (forward, loopback publication)
`private-ingress`	`--network-egress allow --network-ingress allow --host-loopback allow`	`resolver=192.0.2.1 announcer=running allowed=reached other=reached`	`NVX-POST-RESTORE-OK: tier=legacy`	1 (through the host's `10.0.0.14`)

The refusals failed before the guest ran:

restore-time egress policy does not match the snapshot contract (another policy without the option);
microVM snapshots do not support --network-ingress allow, ..., except at a restore with --restore-network-policy.

Negative checks on KVM:

OpenVMM b92daa350 (before OpenVMM prerequisite: a new network policy at the restore of a NIC template without a tier #942): the first restore fails with --restore-network-policy requires a platform snapshot.
The guest of the release, which reads version 4 alone, on the pinned OpenVMM: the first restore fails closed with NVX-TIME-ABI-VIOLATION: ... code=G_REPAIR_PACKET ... detail="restore packet version is 5", so the guest never ignores the new settings.

Validation

Linux, on nvx-kvm-test:
- nvx-rs guest_tools (38 passed), and guest::ingress plus guest::time lib tests (161 passed);
- the pinned shellcheck and shfmt images on nvx-snapshot and init.
Windows: repository checks:
- nvx.py verify, ruff check and format, pyright (Linux and Windows), and check_markdown_links.py;
- test_nvx_tools.py, test_markdown_links.py, test_time_abi.py, test_module_diagrams.py, and test_guest_binaries.py;
- test_microvm_tests.py passes, except 7 subtests of the existing test_host_loopback_allow_script_checks_publication_and_traffic and test_private_network_ingress_script_checks_policy_and_traffic. Those time out under Git sh.exe on this machine, identically at the unchanged base.
Windows: nvx-rs: fmt; clippy with all features, lib only, no default features, and the musl guest target; tests; the ignored openvmm_vectors and openvmm_cpu_profiles; docs; MSRV; the runtime adapter.

Review

Copilot's review raised two points, both fixed in this branch:

Announcer startup. nvx-snapshot checked only that nvx-ingress was spawned, not that it started. nvx-ingress start now confirms that the announcer's socket is bound and its program runs before the acknowledgement.
Private ingress check. The scenario's private-ingress check could not tell loopback-only publication from private-network publication. It now exchanges through a non-loopback host address, and moves the template to 192.0.2.0/24, since Consomme refuses ingress peers inside the guest's subnets, and the hosts sit on 10.0.0.0/24.

Pin microsoft/openvmm#4831 at f1eb78302a296f455a99f8ac622a46f4c0862908, two commits on 3d7512af6, the previous pin, which is the head of feature/microvm/dev. A snapshot whose NIC was captured with the complete deny-all policy is a NIC template, and --restore-network-policy lets a restore install a new policy on it, but OpenVMM admitted the option only for a platform snapshot and refused it for a snapshot without a tier (#942). A platform clone also learns its DNS server, and whether to announce its listeners, from its activation after the restore, which the guest of a snapshot without a tier never receives, so under a new policy that guest would keep the deny-all's configuration: no DNS server and no nvx-ingress (#943). The first commit admits the option for a snapshot without a tier, and still refuses it for a workload-start or instance-checkpoint snapshot. Such a snapshot may hold a workload that has already run, but under the deny-all that workload had no connectivity, so nothing in the snapshot depends on the policy. The second hands the guest of such a restore the new policy's DNS server, the gateway that a cold boot under the policy names, and whether to announce its listeners, in version 5 of the restore packet: its network byte, which version 4 reserves, gives the server's family and the announcement, and 16 bytes after the entropy give the server. Such a restore requires the guest's acknowledgement, so the guest applies the settings before host input reaches it. Every other restore, a platform clone's included, keeps version 4, so the pin changes nothing for NVX's guests until a restore asks for version 5, which a later commit teaches them to read. Neither OpenVMM's control-session protocol vectors nor its built-in CPU profiles change, so their copies in nvx-rs stay as they are. Part of #943 Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Record the commits that the previous commit pins as positions 373 and 374 of the microVM chain, and move the NVX pin to f1eb78302. Position 373 lets --restore-network-policy install a new policy on the NIC template of a snapshot without a tier, and position 374 hands the guest of such a restore the new policy's network settings in version 5 of the restore packet. They change only the tier check of position 347's restore that installs a new policy and the time ABI's restore packet, which positions 159-269 add, so they build on those fork-only commits; say so in the table's remarks. The pin is now 217 commits past the mapped tip. Positions 371 and 372 are no longer only a proposal: microsoft/openvmm#4816 merged them, so position 372, 3d7512af6, is the head of feature/microvm/dev. Say so in the scope, and name microsoft/openvmm#4831 as the proposal of positions 373 and 374. Part of #943 Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Copilot

🟡 Changes recommended

The immediate process-existence check does not confirm announcer socket initialization, allowing restore acknowledgment and host input release after startup failure; a bounded startup handshake, fail-closed handling, and regression coverage are needed.

1 open issue

Verify announcer startup before acknowledging restore · New

What changed in this PR

This PR enables untiered NIC template restores to apply a replacement network policy’s DNS server and listener announcements before acknowledging the restore, addressing #942 and #943.

Changes:

Adds restore packet v5 decoding and guest-side network repair while preserving v4 support.
Exposes nvx-rs run --restore-network-policy, with CLI validation and fake OpenVMM coverage.
Adds a four-policy microVM scenario, updates test shards, and documents the restore behavior and OpenVMM integration.

File	Description
`scripts/test_microvm_tests.py`	Tests network repair, failures, and scenario behavior.
`scripts/nvx_tools/microvm_tests.py`	Adds the four-policy untiered restore scenario.
`scripts/nvx_tools/microvm_test_scripts/untiered-network-policy.sh.in`	Checks guest settings and network traffic.
`scripts/nvx_tools/microvm_shards.json`	Adds the scenario and rebalances shards.
`nvx-rs/tests/vectors/README.md`	Documents checks for both restore packet versions.
`nvx-rs/tests/openvmm_vectors.rs`	Checks the v5 constant against OpenVMM.
`nvx-rs/tests/guest_tools.rs`	Tests v5 decoding and malformed settings.
`nvx-rs/tests/fake_openvmm.rs`	Tests restore-policy ingress acceptance and refusal.
`nvx-rs/tests/cli/validation.rs`	Tests rejection without a restore snapshot.
`nvx-rs/tests/cli/openvmm.rs`	Recognizes the new valueless option.
`nvx-rs/tests/cli/one_shot.rs`	Tests launching a policy-replacing restore.
`nvx-rs/tests/cli/dry_run.rs`	Tests generated restore-policy arguments.
`nvx-rs/src/testing/fake_openvmm.rs`	Models restore-policy option validation.
`nvx-rs/src/host/cli/run.rs`	Parses, validates, and forwards the new option.
`nvx-rs/src/guest/time/test_mode.rs`	Prints decoded network settings in test mode.
`nvx-rs/src/guest/time/mod.rs`	Exports `RestoreNetwork`.
`nvx-rs/src/guest/time/decoders.rs`	Decodes and validates v5 network records.
`nvx-rs/src/guest/time/commands.rs`	Includes network settings in capture metadata.
`nvx-rs/src/guest/time/api.rs`	Exposes settings and defers completion for network repair.
`nvx-rs/src/contract.rs`	Defines the restore packet v5 constant.
`nvx-rs/README.md`	Documents CLI behavior and compatibility checks.
`guest/common/nvx-snapshot`	Applies DNS and starts listener announcements before acknowledgement.
`doc/usage.md`	Updates restore and network option references.
`doc/run.md`	Explains NIC template policy replacement.
`doc/openvmm-upstream-roadmap.md`	Records the new pin and its two commits.
`doc/design/5.5-restore-protocol.md`	Specifies v5 layout and network repair ordering.
`doc/design/3.5-network.md`	Describes replacement policies and guest configuration.
`doc/design/3.1-platform-control-device.md`	Documents restore packet network settings.
`doc/ci.md`	Describes the new scenario and version checks.
`doc/build.md`	Documents v4 and v5 parser support.
`.github/skills/nvx-openvmm-promote/SKILL.md`	Updates guidance to reference multiple restore packet versions.

_{Review effort: Balanced}

A clone's application starts the announcer of the guest's listeners with Announcer::start, which binds the announcer's socket to a privileged source port and runs nvx-ingress inherit on it, so a failure to start it is the start's own, before the clone is ready. A guest agent that cannot hold the announcer, such as a shell script, could only run nvx-ingress in the background, as init does at a cold boot, and could not tell whether the announcer started: whether it found a privileged source port free, or ran at all. nvx-ingress start GATEWAY_IPV4 starts the announcer as Announcer::start does, and exits once the announcer's socket is bound and its program runs, which leaves the announcer to init, or to the nearest subreaper, or exits with status 1, naming why, when the announcer does not start. Part of #943 Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

When a restore installs a new network policy on the NIC template of a snapshot without a tier, OpenVMM hands the guest the policy's DNS server, and whether to announce its listeners, in version 5 of the restore packet, and gates host input until the guest acknowledges the restore. NVX's guest read version 4 alone, so it failed such a restore, and nothing would have applied the settings: the template's guest booted under the deny-all policy, whose command line names no DNS server and starts no nvx-ingress, and only a platform clone takes them from its activation. nvx-time now reads version 5 as well. It refuses a network byte with DNS family 3 or another bit set, and a DNS server that does not match its family or is unspecified, and the capture's metadata for nvx-snapshot gains the DNS server, or "-", and the announcement. As for a packet with processors or memory to activate, nvx-time leaves a packet whose settings ask for something to nvx-snapshot, which acknowledges the restore once it applied them. TimeAbi's Restore exposes the settings as a RestoreNetwork, and nvx-time's packet test hook prints them. nvx-snapshot applies them in guest repair step 4, after processors and memory, as init applies those of a cold boot under such a policy: it names the DNS server in /etc/resolv.conf, and starts the announcer of the guest's listeners with nvx-ingress start and the gateway of virtnet_gw=, which returns once the announcer's socket is bound to a privileged source port and its program runs. A failure of either fails the restore, with status 1, before its acknowledgement. A restore whose settings ask for nothing stays quiet, and an untiered restore still leaves entropy and identity to its caller. The restore-protocol, network, and byte-channel design documents describe version 5 and the repair. Part of #943 Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

OpenVMM lets a restore install a new network policy on a NIC template, a snapshot whose NIC was captured with the complete deny-all policy, with --restore-network-policy, for a platform snapshot and now for one without a tier, whose guest then applies the policy's DNS server and listener announcements before the host releases its input. nvx-rs run could not ask for it: every restore passed the network options to OpenVMM as the policy that the snapshot was captured with, so run could restore a NIC template only under its deny-all. run --restore-network-policy, which needs --restore-snapshot, passes the option to OpenVMM after the restore's other options, so the restore installs the policy of its network options, private-network ingress, explicit host loopback, and forwards included. run leaves the checks of the snapshot and the policy to OpenVMM, as it does for every other restore, rather than read the snapshot itself. The option stays explicit because OpenVMM refuses it for a snapshot that is no NIC template, so run cannot pass it to every restore. nvx-fake-openvmm accepts the option at a restore and refuses it at a cold boot, and, like OpenVMM, refuses a restore without it that asks for private-network ingress or explicit host-loopback access, with OpenVMM's messages. The usage and run guides and the nvx-rs README describe the option. Part of #943 Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Copilot

🟡 Changes recommended

The private-ingress test can pass with loopback-only publication, leaving incorrect restoration of private-network ingress undetected.

1 open issue

The private-ingress case passes even when the restored policy publishes only on host loopback: a… · New

1 resolved since last review

Verify announcer startup before acknowledging restore

_{Review effort: Balanced}

Copilot

🔵 Needs a closer look

The cross-repository restore ABI and input-release ordering require human confirmation of exact-pin OpenVMM integration and pending Windows/WHP end-to-end validation.

No open issues 🎉

1 resolved since last review

The private-ingress case passes even when the restored policy publishes only on host loopback: a…

_{Review effort: Balanced}

Add the untiered-network-policy microVM scenario, which covers the restore of a NIC template without a tier, a snapshot without sandbox blocks, as the non-sandbox mode captures one, whose NIC was captured with the complete deny-all policy, with --restore-network-policy. It restores the template four times: under the template's own deny-all, under egress rules that admit DNS over IPv4 and one host service, with explicit host loopback and a forward under egress rules that admit DNS over IPv6 alone, and with private-network ingress and allowed egress. The template's guest learned no DNS server and runs no announcer of its listeners, so each restore's repair applies those of the restore's policy, which restore packet version 5 carries, before it acknowledges the restore. The guest's resolver must name exactly the policy's DNS server, and nvx-ingress must run exactly when the policy publishes the guest's listeners. Each restore must reach through its gateway exactly the host services that its policy admits, which receive no other request. The host must reach the guest's server through the forward, and through the port that the policy publishes: on host loopback alone, which no other host address reaches, or, for private-network ingress, through a host address other than loopback, since a probe that binds every host address also fails over a loopback-only listener. It must find the server on no host loopback port without a publication. OpenVMM must refuse, before the guest runs, a restore that keeps the template's policy but passes another, and one that passes private-network ingress without the option, and no restore may change the template. The template's NIC is on the directional network, 192.0.2.0/24, a documentation range that no host address shares: an ingress listener refuses peers inside the guest's subnets, so a host address there, as on hosts whose network is 10.0.0.0/24, would not reach the port that private-network ingress publishes. The scenario joins the default suite, so plan the shards again, with its time in local runs on virtual-machine hosts of each backend, 14.0 s on KVM, 14.1 s on MSHV, and 18.8 s on WHP, until CI's reports measure it. Part of #943 Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Copilot

🔵 Needs a closer look

The cross-repository restore ABI and host-input gating changes need maintainer confirmation of Windows/WHP end-to-end validation and that the exact pinned OpenVMM revision lands before integration.

No open issues 🎉

_{Review effort: Balanced}

Copilot

🔵 Needs a closer look

The pinned dependency microsoft/openvmm#4831 is still open, and its unchanged landing on feature/microvm/dev must be confirmed before merging the NVX integration.

No open issues 🎉

_{Review effort: Balanced}

Pedro Henrique Penna (ppenna) · 2026-10-31T05:39:58Z

Removed from the merge queue so that #1018 could land first; requeue this once it's rebased.

microsoft/openvmm#4830 (time ABI 2) is now merged: feature/microvm/dev was fast-forwarded from 3d7512af6 to 858b2f93a. microsoft/openvmm#4831, which this pull request pins at f1eb78302, is two commits on 3d7512af6, so it can no longer fast-forward the branch, and GitHub reports it conflicting.

The two OpenVMM changes also collide on restore packet version 5:

[DO NOT REVIEW] microVM: publish the host's clock correction with each time sample openvmm#4830 makes the regular restore packet version 5, with no host UTC and a 24-byte header.
[DO NOT REVIEW] microVM: let an untiered restore install a new network policy on a NIC template openvmm#4831 adds RESTORE_PACKET_NETWORK_VERSION = 5 for the untiered network-policy restore, beside the regular packet 4.

To land this pull request:

Rebase [DO NOT REVIEW] microVM: let an untiered restore install a new network policy on a NIC template openvmm#4831 onto 858b2f93a, renumber its network packet (for example to 6) on top of packet 5's layout, and rerun its OpenVMM CI.
Once time ABI: publish the host's clock correction with each time sample #1018 has merged, rebuild this pull request on dev with the new pin. Against time ABI: publish the host's clock correction with each time sample #1018 it conflicts in the openvmm gitlink, nvx-rs/src/contract.rs, nvx-rs/src/guest/time/decoders.rs, doc/design/5.5-restore-protocol.md, doc/build.md, doc/openvmm-upstream-roadmap.md, guest/common/nvx-snapshot, nvx-rs/README.md, scripts/nvx_tools/microvm_tests.py, and scripts/nvx_tools/microvm_shards.json.

Pedro Henrique Penna (ppenna) and others added 2 commits October 30, 2026 19:20

Copilot AI balanced review requested due to automatic review settings October 31, 2026 02:26

Copilot started reviewing on behalf of Pedro Henrique Penna (ppenna) October 31, 2026 02:27

View session