cpuset for libvirt set to 0 after suspend/resume
Affects | Status | Importance | Assigned to | Milestone | |
---|---|---|---|---|---|
libvirt (Fedora) |
Fix Released
|
Undecided
|
|||
libvirt (Ubuntu) |
Fix Released
|
Low
|
Unassigned | ||
linux (Ubuntu) |
Fix Released
|
Medium
|
Unassigned |
Bug Description
Starting 12.04 (afaict), guest KVM processes are put into their own cgroup with the cpuset of that cgroup at /sys/fs/
On a clean boot, the cpuset.cpus for libvirt contain all the cpus. However, after a suspend/resume, cpuset.cpus for libvirt and all children is 0, which effectively pins all vcpus in the guest to a cpu 0 in the host causing massive performance problems.
There is already a bug that RedHat is tracking for this issue:
https:/
Expected:
The cpuset for libvirt and children remains consistent across a suspend/resume
What happens:
The cpuset for libvirt and children gets set to 0
ProblemType: Bug
DistroRelease: Ubuntu 12.04
Package: libvirt-bin 0.9.8-2ubuntu17
ProcVersionSign
Uname: Linux 3.2.0-23-generic x86_64
NonfreeKernelMo
ApportVersion: 2.0.1-0ubuntu5
Architecture: amd64
Date: Wed May 2 10:22:31 2012
InstallationMedia: Xubuntu 12.04 LTS "Precise Pangolin" - Release amd64 (20120425)
ProcEnviron:
LANGUAGE=en_US:en
TERM=xterm
PATH=(custom, user)
LANG=en_US.UTF-8
SHELL=/bin/bash
SourcePackage: libvirt
UpgradeStatus: No upgrade log present (probably fresh install)
modified.
Changed in linux (Ubuntu): | |
status: | Incomplete → Confirmed |
Changed in libvirt (Fedora): | |
importance: | Unknown → Undecided |
status: | Unknown → Fix Released |
Thanks for reporting this bug. I've linked the redhat bug, and marked it confirmed based on that. I've marked it as affecting the kernel, since there has been talk of a fix in the kernel (per the redhat bug).
I'm going to mark this low priority because there is a workaround.