What were you trying to do that didn't work?

A guest with overall guest numa policy with below:
  <numatune>
    <memory mode="preferred" nodeset="1" />
  </numatune>
And there is a memory device which define the source like below:
    <source>
      <pagesize unit='KiB'>1048576</pagesize>
      <nodemask>0</nodemask>
    </source>
After the guest start up, the memory device numa policy is override to "preferred" not "bind" (strict), and I also observe the hugepage memory allocation drift to other numa node sometimes, but not always. 

Please provide the package NVR for which the bug is seen:

rhel9.7:

# rpm -q libvirt qemu-kvm
libvirt-10.10.0-15.el9.x86_64
qemu-kvm-9.1.0-26.el9.x86_64

rhel10.1:

# rpm -q libvirt qemu-kvm
libvirt-11.5.0-4.el10.x86_64
qemu-kvm-10.0.0-12.el10.x86_64

How reproducible is this bug?: 100%

Steps to reproduce

1. on a multi numa node host:

# numactl --hard
available: 2 nodes (0-1)
node 0 cpus: 0 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30 32 34 36 38
node 0 size: 31560 MB
node 0 free: 27664 MB
node 1 cpus: 1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39
node 1 size: 32199 MB
node 1 free: 26162 MB
node distances:
node     0    1 
   0:   10   21 
   1:   21   10 

2. prepare the hugepage env.

# echo 2 > /sys/devices/system/node/node1/hugepages/hugepages-1048576kB/nr_hugepages
# echo 2 > /sys/devices/system/node/node0/hugepages/hugepages-1048576kB/nr_hugepages
# mkdir /dev/hugepages1048576
# mount -t hugetlbfs -o pagesize=1048576K hugetlbfs /dev/hugepages1048576
# systemctl restart virtqemud

2. Start the guest with below config xml:
  <maxMemory slots="16" unit="KiB">15243264</maxMemory>
  <memory unit="KiB">3145728</memory>
  <currentMemory unit="KiB">3145728</currentMemory>
   <vcpu placement="static">4</vcpu>
  <numatune>
    <memory mode="preferred" nodeset="1" />
  </numatune>
  <cpu mode='host-model' check='partial'>
    <numa>
      <cell id="0" cpus="0-1" memory="1048576" unit="KiB" />
      <cell id="1" cpus="2-3" memory="1048576" unit="KiB" />
    </numa>
  </cpu>
  ...
  <memory model='dimm'>
    <source>
      <pagesize unit='KiB'>1048576</pagesize>
      <nodemask>0</nodemask>
    </source>
    <target>
      <size unit='KiB'>1048576</size>
      <node>0</node>
  </target>
  </memory>
  ...

# virsh start avocado-vt-vm1
Domain 'avocado-vt-vm1' started

3. Check the qemu cmd line, memory device numa policy is same with guest "preferred".

# ps -ef | grep qemu-kvm
-object {"qom-type":"thread-context","id":"tc-memdimm0","node-affinity":[0]} -object {"qom-type":"memory-backend-file","id":"memdimm0","mem-path":"/dev/hugepages1048576/libvirt/qemu/5-avocado-vt-vm1","prealloc":true,"size":1073741824,"host-nodes":[0],"policy":"preferred","prealloc-context":"tc-memdimm0"} -device {"driver":"pc-dimm","node":0,"memdev":"memdimm0","id":"dimm0","slot":0}

3. Check the hugepage allocation, memory device 1G hugepage memory are allocated on node0.

# virsh freepages --all
Node 0:
4KiB: 6936642
2048KiB: 0
1048576KiB: 1
Node 1:
4KiB: 7526007
2048KiB: 0
1048576KiB: 2

4. After the guest boot up and wait for some time (around 2mins) check hugepage allocation, the memory device hugepage memory drift to node 1 (not always)

# virsh freepages --all
Node 0:
4KiB: 6937610
2048KiB: 0
1048576KiB: 2
Node 1:
4KiB: 7326549
2048KiB: 0
1048576KiB: 1

Expected results:

1. memory device numa policy should be "bind" (strict) not overall policy "preferred";
2. the hugepage used by memory device should always be allocated on node 0;

Actual results:

1. memory device numa policy is overide by overall policy "preferred";
2. the node 0 hugepage used by memory device drift to node 1 after the guest boot up;