While it'll take some time to know whether presumably the 2nd these
changes will help with the originally reported issue, there are
immediate positive effects even beyond that issue, which is why I
didn't want to wait with the submission. There's at least one more
issue to take care of (see the respective remark in patch 2), but
that's still under discussion / consideration.
1: use relative counts in calibration loops
2: improve TSC / CPU freq calibration accuracy