On 2012年12月13日 15:01, Hu Tao wrote:
Add a flag VIR_CGROUP_DISABLE_CPUSET to disable cgroup cpuset.
This flag inhibits making of directory under cpuset.
---
src/util/cgroup.c | 7 +++++++
1 file changed, 7 insertions(+)
diff --git a/src/util/cgroup.c b/src/util/cgroup.c
index f867fb7..e955a22 100644
--- a/src/util/cgroup.c
+++ b/src/util/cgroup.c
@@ -72,6 +72,7 @@ typedef enum {
*/
VIR_CGROUP_VCPU = 1<< 1, /* create subdir only under the cgroup cpu,
* cpuacct and cpuset if possible. */
+ VIR_CGROUP_DISABLE_CPUSET = 1<< 2,
} virCgroupFlags;
/**
@@ -542,6 +543,12 @@ static int virCgroupMakeGroup(virCgroupPtr parent,
if (!group->controllers[i].mountPoint)
continue;
+ if ((flags& VIR_CGROUP_DISABLE_CPUSET)&&
+ i == VIR_CGROUP_CONTROLLER_CPUSET) {
+ group->controllers[i].mountPoint = NULL;
+ continue;
+ }
This actually exposes an existed bug, regardless of whether the user
will configure qemu.conf without the controller or not. It will always
be created.
As it iterates over the all the internal-defined controllers without
honoring the configuration in qemu.conf. (see the 'for' loop)
However, It's expected to not create the cgroup as long as it's not
configured in qemu.conf.
The right way is to fix virCgroupMakeCgroup to honor the configuration
in qemu.conf instead, a new flag here only fixes the case "cpuset" is
not configured, it doesn't fix the root cause, and the new flags is
not needed if the root cause is fixed.
I see this set passes driver->cgroupContollers to virCgroupMakeCgroup,
which is good, but it still iterates over the internal-defined all
controllers.
Correct me if I'm wrong.
Regards,
Osier
--
libvir-list mailing list
libvir-list@xxxxxxxxxx
https://www.redhat.com/mailman/listinfo/libvir-list