The cpu xml copy is incorrect: the memAccess field is not copied.
The lack of copy of this memAccess field can cause unexpected behaviour for live
migration when vhost user is used.
For example if guest has the following configuration:
....
<cpu>
<model>Westmere</model>
<topology sockets="1" cores="4" threads="1"/>
<numa>
<cell id='0' cpus='0-3' memory='2097152' memAccess='shared'/>
</numa>
</cpu>
....
The used configuration on the remote host in case of live migration is:
....
<cpu mode='custom' match='exact'>
<model fallback='allow'>Westmere</model>
<topology sockets='1' cores='4' threads='1'/>
<numa>
<cell id='0' cpus='0-3' memory='2097152' unit='KiB'/>
</numa>
</cpu>
....
On the remote host the lack of memAccess info can cause unexpected error on the
qemu backend vhost user driver.
Fixes: def6b3598 ("docs, conf, schema: add support for shared memory mapping")
This issue is present only for libvirt1.2.9 to libvirt1.2.12
With patch 181742d43 ("conf: Move all NUMA configuration to virDomainNuma")
present since libvirt1.2.13 the problem does not exist anymore as NUMA
information are no more in the CPU configuration.
Signed-off-by: Thibaut Collet <thibaut.collet@xxxxxxxxx>
---
src/conf/cpu_conf.c | 1 +
1 file changed, 1 insertion(+)
diff --git a/src/conf/cpu_conf.c b/src/conf/cpu_conf.c
index 4a367a1..7113d70 100644
--- a/src/conf/cpu_conf.c
+++ b/src/conf/cpu_conf.c
@@ -158,6 +158,7 @@ virCPUDefCopy(const virCPUDef *cpu)
for (i = 0; i < cpu->ncells; i++) {
copy->cells[i].mem = cpu->cells[i].mem;
+ copy->cells[i].memAccess = cpu->cells[i].memAccess;
copy->cells[i].cpumask = virBitmapNewCopy(cpu->cells[i].cpumask);
--
1.7.10.4
--
libvir-list mailing list
libvir-list@xxxxxxxxxx
https://www.redhat.com/mailman/listinfo/libvir-list