[libvirt PATCH] Revert "qemuDomainSetNumaParamsLive: set nodeset for root cgroup"

Pavel Hrdina phrdina at redhat.com
Fri May 21 13:40:26 UTC 2021

This reverts commit <1b22dd6dd44202094e0f78f887cbe790c00e9ebc>.

First of all, the reverted commit is incomplete. It only sets
cpuset.mems in the VM root cgroup when the API is used but there is no
code that would do the same when the VM is started.

Libvirt never places any process into the VM root cgroup directly. All
the supporting processes like slirp-helper or dbus-daemon are placed
into the emulator sub-cgroup and all the QEMU threads are distributed
between emulator, vcpu* and iothread* sub-cgroups. The scenario
described in the reverted commit can happen only if someone manually
adds any process there which we should not care about.

If we would like to set the limit in the VM root cgroup we need to
introduce better logic:

    - set both (old and new) numa group in the VM root cgroup
    - change the numa group in all sub-cgroups to new value
    - finally set only the new value in the VM root cgroup

The simplest fix now is to revert the commit.

Signed-off-by: Pavel Hrdina <phrdina at redhat.com>
 src/qemu/qemu_driver.c | 4 ----
 1 file changed, 4 deletions(-)

diff --git a/src/qemu/qemu_driver.c b/src/qemu/qemu_driver.c
index e10e699a1a..a972662c3f 100644
--- a/src/qemu/qemu_driver.c
+++ b/src/qemu/qemu_driver.c
@@ -8797,10 +8797,6 @@ qemuDomainSetNumaParamsLive(virDomainObj *vm,
             return -1;
-    /* set nodeset for root cgroup */
-    if (virCgroupSetCpusetMems(priv->cgroup, nodeset_str) < 0)
-        return -1;
     return 0;

More information about the libvir-list mailing list