hyperv_ic: implement SaveRestore for KVP & timesync

[Steven Malis (smalis-msft)]

We've discovered a Linux kernel bug that results in crashes when vmbus channels are rapidly offered and rescinded, a pattern we exercise in our vmm_tests. Work around this by implementing SaveRestore for the last two relevant devices that didn't already support it. Mimic Hyper-V's behavior by regenerating the devices on restore, rather than actually saving state.

A future change may make SaveRestore support non-optional, that's still under discussion.

This doesn't restabilize the broken tests yet, I'll let this bake for a bit first.

[Copilot]

Pull request overview

Implements save/restore support for the Hyper-V IC KVP and timesync devices so they can be restored without revoking/reoffering their VMBus channels, working around a Linux kernel crash triggered by rapid offer/rescind patterns.

Changes:

• Add saved-state wire types for negotiated IC versions (SavedVersions) and per-device channel state.
• Enable supports_save_restore() for TimesyncIc and KvpIc, with corresponding save_open / restore_open implementations.
• On restore, handle in-flight states safely (e.g., timesync resends a sync message; KVP drains an outstanding response before accepting new requests).

Reviewed changes

Copilot reviewed 3 out of 3 changed files in this pull request and generated no comments.

File	Description
vm/devices/hyperv_ic/src/timesync.rs	Adds save/restore state capture and restore behavior for timesync channel state machine.
vm/devices/hyperv_ic/src/kvp.rs	Adds save/restore for KVP channel state machine, including response-drain handling after restore.
vm/devices/hyperv_ic/src/common.rs	Introduces SavedVersions and conversions to serialize negotiated version pairs in saved state.

[Copilot]

Pull request overview

Copilot reviewed 3 out of 3 changed files in this pull request and generated no new comments.

[Copilot]

Pull request overview

Copilot reviewed 4 out of 5 changed files in this pull request and generated no new comments.

Comments suppressed due to low confidence (2)

vm/devices/hyperv_ic/src/common.rs:196

• test_with_tracing::test is a sync-only attribute macro (it rejects async fns) and it’s not used here (tests use #[async_test]). This import will be unused and the dev-dependency becomes unnecessary unless you explicitly call test_with_tracing::init() in the async tests.

    use pal_async::async_test;
    use test_with_tracing::test;
    use vmbus_channel::connected_async_channels;

vm/devices/hyperv_ic/Cargo.toml:38

• test_with_tracing is only referenced via an unused import in src/common.rs tests. Once that import is removed, this dev-dependency should be dropped to avoid unnecessary workspace churn (and it will likely get removed by fmt anyway).

[dev-dependencies]
test_with_tracing.workspace = true

[Copilot]

Pull request overview

Copilot reviewed 4 out of 5 changed files in this pull request and generated no new comments.

Comments suppressed due to low confidence (1)

vm/devices/hyperv_ic/src/common.rs:109

• During version negotiation you currently ignore the response status returned by read_message (_status). If the guest reports a failure status for the VERSION_NEGOTIATION transaction, we should treat negotiation as failed instead of proceeding to parse the body as if it succeeded.

                    let (message_type, _status, buf) = self.read_message().await?;

[Ben Hillis (benhillis)]

Backported to release/1.8.2607 in #4127