Actually, migration isn't required to reproduce this issue at all, it is
the stop/cont involved in the migration here that triggers the bug. It
is significantly easier to reproduce the bug with the following steps:
1) on host, adjtimex -f 1000
2) start guest
3) wait 20 minutes
4) stop and
As a further test, I disabled ntpd on the host and ran ntpdate via cron
every 12 hours, so that the clock would be relatively accurate, but no
clock skew would be involved. This also reproduced the failure as
initially described.
This is interesting as it means that a much simpler and faster
Two important findings:
1) If I disable ntpd on the host, this issue goes away.
2) If I forcefully induce substantial clock skew on the host (with adjtimex -f
1000), it becomes much less time intensive to reproduce this issue.
Using the attached reproducer but replacing the 18h sleep