lib: implement reference TSC enlightenment

[gjcolombo]

Add a Hyper-V reference time enlightenment to the Hyper-V stack. See the new tsc module's doc comment for details about how the enlightenment operates.

Add PHD tests that try to verify both that the enlightenment is present (by querying the current clocksource) and that it keeps time roughly correctly (by comparing host and guest time readings). The latter test also takes readings over a migration to help get some coverage of the time data phase of live migration (though the migration is not currently cross-machine, so this is a little less interesting than it might otherwise be).

Tests: cargo test; PHD w/Alpine, Debian 11, WS2022 guests.

Fixes #328.

[jordanhendricks]

initial pass; want to learn a bit more about this feature in depth and I'll take another one

[gjcolombo]

I've rebased this to the latest master. This was unfortunately a bit of a mess, because #861 obsoleted a fair amount of the migration logic that was present previously. It may help to look at just the last five commits to get a sense of what's actually changed (beyond merge conflict resolutions during the rebase).

Now that the Hyper-V stack doesn't migrate overlay page contents directly, it needs to be more careful to ensure that it generates TSC overlays that are based on the correct guest frequency. The challenges here are

• we can't read the guest frequency during enlightenment creation because we don't have a VmmHdl yet
• during import, we do have a VmmHdl, but we'd like to avoid assuming that the time-data phase of migration has already happened (which has to happen for us to read the correct guest frequency out of the kernel VM)

Most of the new code involves a new ReferenceTsc helper type that keeps track of whether the enlightenment is on, whether it's initialized, and what guest frequency it's supposed to use when constructing the reference page. A lot of the newest code is meant to work with this type and to make sure its configuration data doesn't get blown away during VM reset.

EDIT: Things that blessedly did not change are the reference TSC page's contents (and the math we use to compute them) and the new PHD tests.

[hawkw]

Hm, the introduction of the Uninitialized state etc is not my favorite thing, but I agree that this is necessary due to the changes to how we migrate overlays, and I can't think up a solution that avoids it. This seems good to me!

[iximeow]

more nits about the panics :) seems good, very excited for this!

[gjcolombo]

Thanks for re-reviewing! ae5cfac should take care of the latest batch of comments.

[hawkw]

new commentary looks lovely, thank you!

[hawkw]

mmmmmaybe this ought to have a big leading "WARNING:" on it? not a big deal though.

[hawkw]

good call with these notes!

[jordanhendricks]

🚢