5.15.0-76: Qemu live migration regression due to PKRU leakage
Affects | Status | Importance | Assigned to | Milestone | |
---|---|---|---|---|---|
linux (Ubuntu) |
Expired
|
Undecided
|
Unassigned | ||
qemu (Ubuntu) |
Expired
|
Undecided
|
Unassigned |
Bug Description
I am trying to live migrate Qemu VMs between Amd Zen3 to Zen2 servers.
What we expect:
VMs continue to run after live-migrate
What happened:
VMs instantly get stuck after migration with 100% CPU usage.
All Nodes are running Ubuntu 22.04.
We have tested Kernel releases back to 5.15.0-70.
Nodes running Ubuntu 5.19.0-
Zen2 -> Zen3 migration: OK
Zen3 -> Zen2 migration: OK
Nodes running Ubuntu 5.15.0-
Zen2 -> Zen3 migration: OK
Zen3 -> Zen2 migration: NOT OK
Mixed Kernel versions:
Zen2 5.15.0-
Zen3 5.19.0-
Zen2 5.19.0-
Zen3 5.15.0-
We've tested it with fresh started and with live-migrated VMs, getting same results.
Its probably related to
KVM: SVM: fix tsc scaling cache logic
https:/
Marking the Qemu task as Incomplete, given that the problem goes away by using a different kernel, and there's nothing specific pointing to a Qemu bug.