Trouble with KVM after upgrade to bookworm
Hello.
I use Debian as KVM host for the various VMs (Ubuntu, FreeBSD,
no Debian) which serve my home network. I've been doing this
since Debian stretch. Until bookworm, each release upgrade
has been tedious, but boringly unremarkable.
This upgrade seems to have broken KVM so as to cause one or other
guest VM to enter what seems to be a busy wait, and to do so
unpredictably and intermittently. According to virt-top, the affected guest seizes the full capacity of the CPU core where it is currently
running; on my 4-core unit, virt-top shows this as 25% CPU utilization.
When this happens, top shows zero idle time for the corresponding core,
and an aggregate of system and user utilization (the only non-zero
utilization values) closer to 50% than to 100%. In normal operation,
I see each core busy for about 3% of the time, idle for 95% or more,
and an the remaining 2% or so unaccounted for.
When a VM goes busy like this, I find that it can be made unbusy
by use of the command, `virsh inject-nmi $GUEST`.
I think I have read the release notes for bookworm carefully, but
may have missed something. I've repeatedly re-read chapters 4 and 5.
I'm baffled, and would appreciate any clue.
Thanks in advance.
Niall O'Reilly
Reply to: