I believe this is a QEMU bug. I'm able to reproduce and the cmd line that's generated on the first start of the VM contains the following:

-object '{"qom-type":"memory-backend-file","id":"memdimm0","mem-path":"/var/lib/libvirt/qemu/ram/3-fedora/dimm0","discard-data":false,"share":false,"prealloc":true,"prealloc-threads":16,"size":268435456}' \
-device '{"driver":"pc-dimm","node":0,"memdev":"memdimm0","id":"dimm0","slot":0}' \
-object '{"qom-type":"memory-backend-file","id":"memvirtiomem0","mem-path":"/hugepages2M/libvirt/qemu/3-fedora","discard-data":true,"share":false,"reserve":false,"size":2147483648}' \
-device '{"driver":"virtio-mem-pci","node":0,"block-size":2097152,"requested-size":1073741824,"memdev":"memvirtiomem0","prealloc":true,"id":"virtiomem0","bus":"pci.0","addr":"0xa"}' \

and after the dimm0 is unplugged, saved and restored this is the cmd line that libvirt tries to run:

-object '{"qom-type":"memory-backend-file","id":"memvirtiomem0","mem-path":"/hugepages2M/libvirt/qemu/4-fedora","discard-data":true,"share":false,"reserve":false,"size":2147483648}' \
-device '{"driver":"virtio-mem-pci","node":0,"block-size":2097152,"requested-size":1073741824,"memdev":"memvirtiomem0","prealloc":true,"id":"virtiomem0","bus":"pci.0","addr":"0xa"}' \

IOW - the pc-dimm device is gone (which is expected - it was unplugged). Let me switch over to QEMU for further investigation.

QEMU will auto-assign memory addresses for memory devices, however these addresses cannot really get migrated, because they have to be known when plugging such a device. So upper layers have to provide these properties when creating the devices (e.g., read property on source and configure for destination). Similar to, DIMM slots or PCI addresses.

That's nothing special about virtio-mem (it's the only one that actually checks instead of crashing the guest later :) ), it's the same for all memory devices, just the "address property" name differs.

For DIMMs/NVDIMMS it's the "addr" property. For virtio-mem/virtio-pmem, it's the "memaddr" property.

I would assume that such handling is already in place for DIMMs/NVDIMMs? Otherwise unplugging some DIMMs and migrating the VM would crash the VM on the destination when the location of some DIMMs changes in guest address space.

@Michal, is such handling for DIMMs already in place? Then we need similar handling for virtio-mem/virtio-pmem in libvirt. If it's not around for DIMMs, then we need it there as well ...

Yeah, it's implemented for DIMMs and indeed virtio-mem was the missing piece. After I made memaddr stable I can restore from a savefile. So this is indeed a libvirt bug. Let me switch back to libvirt and post patches.

Patches posted on the list:

https://listman.redhat.com/archives/libvir-list/2023-March/239051.html

Merged upstream as:

a1bdffdd96 qemu_command: Generate .memaddr for virtio-mem and virtio-pmem
2c15506254 qemu: Fill virtio-mem/virtio-pmem .memaddr at runtime
677156f662 conf: Introduce <address/> for virtio-mem and virtio-pmem

v9.4.0-rc1-5-ga1bdffdd96

Pre-verified on upstream build libvirt v9.4.0-12-gf26923fb2e 

Test steps:
1. In guest set kernel parameter memhp_default_state to online_movable
# grubby --update-kernel=ALL --remove-args=memhp_default_state --args=memhp_default_state=online_movable
# reboot



2. Define and start the guest with dimm, virtio-mem memory devices related config xml as below:
   <memory model="dimm" discard="no">
  <source>
    <pagesize unit="KiB">4</pagesize>
  </source>
  <target>
    <size unit="KiB">131072</size>
    <node>0</node>
   </target>
 </memory>
<memory model="dimm" discard="no">
  <source>
    <pagesize unit="KiB">4</pagesize>
  </source>
  <target>
    <size unit="KiB">131072</size>
    <node>0</node>
   </target>
 </memory>
<memory model="dimm" discard="no">
  <source>
    <pagesize unit="KiB">4</pagesize>
  </source>
  <target>
    <size unit="KiB">131072</size>
    <node>0</node>
   </target>
 </memory>
<memory model="dimm" discard="no">
  <source>
    <pagesize unit="KiB">4</pagesize>
  </source>
  <target>
    <size unit="KiB">131072</size>
    <node>0</node>
   </target>
 </memory>
<memory model='virtio-mem'>
    <target>
      <size unit='KiB'>2097152</size>
      <node>0</node>
      <block unit='KiB'>2048</block>
      <requested unit='KiB'>1048576</requested>
    </target>
</memory>
...


3. Wait for guest booting up and virtio-mem current size is not 0. 
# virsh dumpxml vm1 --xpath "//memory[@model='virtio-mem']"
<memory model="virtio-mem">
  <target>
    <size unit="KiB">2097152</size>
    <node>0</node>
    <block unit="KiB">2048</block>
    <requested unit="KiB">1048576</requested>
    <current unit="KiB">0</current>
    <address base="0x120000000"/>
  </target>
  <alias name="virtiomem0"/>
  <address type="pci" domain="0x0000" bus="0x07" slot="0x00" function="0x0"/>
</memory>


4. Prepare a dimm memory device with config xml:
# cat memory1.xml
<memory model="dimm" discard="no">
  <source>
    <pagesize unit="KiB">4</pagesize>
  </source>
  <target>
    <size unit="KiB">131072</size>
    <node>0</node>
  </target>
  <alias name="dimm2"/>
  <address type="dimm" slot="2" base="0x110000000"/>
</memory>


5. Hot unplug the dimm memory device with config xml in step4
# virsh detach-device vm1 memory1.xml
Device detached successfully


6. Save and restore the domain.
# virsh save vm1 vm1.save
Domain 'vm1' saved to vm1.save

# virsh restore vm1.save

Verified on build:
# rpm -q libvirt qemu-kvm
libvirt-9.5.0-3.el9.x86_64
qemu-kvm-8.0.0-9.el9.x86_64

Test steps:
1. In guest set kernel parameter memhp_default_state to online_movable
# grubby --update-kernel=ALL --remove-args=memhp_default_state --args=memhp_default_state=online_movable
# reboot



2. Define and start the guest with dimm, virtio-mem memory devices related config xml as below:
   <memory model="dimm" discard="no">
  <source>
    <pagesize unit="KiB">4</pagesize>
  </source>
  <target>
    <size unit="KiB">131072</size>
    <node>0</node>
   </target>
 </memory>
<memory model="dimm" discard="no">
  <source>
    <pagesize unit="KiB">4</pagesize>
  </source>
  <target>
    <size unit="KiB">131072</size>
    <node>0</node>
   </target>
 </memory>
<memory model="dimm" discard="no">
  <source>
    <pagesize unit="KiB">4</pagesize>
  </source>
  <target>
    <size unit="KiB">131072</size>
    <node>0</node>
   </target>
 </memory>
<memory model="dimm" discard="no">
  <source>
    <pagesize unit="KiB">4</pagesize>
  </source>
  <target>
    <size unit="KiB">131072</size>
    <node>0</node>
   </target>
 </memory>
<memory model='virtio-mem'>
    <target>
      <size unit='KiB'>2097152</size>
      <node>0</node>
      <block unit='KiB'>2048</block>
      <requested unit='KiB'>1048576</requested>
    </target>
</memory>
...


3. Wait for guest booting up and virtio-mem current size is not 0. 
# virsh dumpxml vm1 --xpath "//memory[@model='virtio-mem']"
<memory model="virtio-mem">
  <target>
    <size unit="KiB">2097152</size>
    <node>0</node>
    <block unit="KiB">2048</block>
    <requested unit="KiB">1048576</requested>
    <current unit="KiB">1048576</current>
    <address base="0x120000000"/>
  </target>
  <alias name="virtiomem0"/>
  <address type="pci" domain="0x0000" bus="0x07" slot="0x00" function="0x0"/>
</memory>



4. Prepare a dimm memory device with config xml:
# cat mem.xml
<memory model="dimm" discard="no">
  <source>
    <pagesize unit="KiB">4</pagesize>
  </source>
  <target>
    <size unit="KiB">131072</size>
    <node>0</node>
  </target>
  <alias name="dimm3"/>
  <address type="dimm" slot="3" base="0x118000000"/>
</memory>



5. Hot unplug the dimm memory device with config xml in step4
# virsh detach-device vm1 mem.xml
Device detached successfully


6. Save and restore the domain.
# virsh save vm1 vm1.save
Domain 'vm1' saved to vm1.save

# virsh restore vm1.save

7. Check memory device config by virsh dumpxml
# virsh dumpxml vm1 --xpath "//memory"
<memory unit="KiB">4587520</memory>
<memory model="dimm" discard="no">
  <source>
    <pagesize unit="KiB">4</pagesize>
  </source>
  <target>
    <size unit="KiB">131072</size>
    <node>0</node>
  </target>
  <alias name="dimm0"/>
  <address type="dimm" slot="0" base="0x100000000"/>
</memory>
<memory model="dimm" discard="no">
  <source>
    <pagesize unit="KiB">4</pagesize>
  </source>
  <target>
    <size unit="KiB">131072</size>
    <node>0</node>
  </target>
  <alias name="dimm1"/>
  <address type="dimm" slot="1" base="0x108000000"/>
</memory>
<memory model="dimm" discard="no">
  <source>
    <pagesize unit="KiB">4</pagesize>
  </source>
  <target>
    <size unit="KiB">131072</size>
    <node>0</node>
  </target>
  <alias name="dimm2"/>
  <address type="dimm" slot="2" base="0x110000000"/>
</memory>
<memory model="virtio-mem">
  <target>
    <size unit="KiB">2097152</size>
    <node>0</node>
    <block unit="KiB">2048</block>
    <requested unit="KiB">1048576</requested>
    <current unit="KiB">1048576</current>
    <address base="0x120000000"/>
  </target>
  <alias name="virtiomem0"/>
  <address type="pci" domain="0x0000" bus="0x07" slot="0x00" function="0x0"/>
</memory>

mark it verified for comment 7

Since the problem described in this bug report should be
resolved in a recent advisory, it has been closed with a
resolution of ERRATA.

For information on the advisory (Moderate: libvirt security, bug fix, and enhancement update), and where to find the updated
files, follow the link below.

If the solution does not work for you, open a new bug report.

https://access.redhat.com/errata/RHSA-2023:6409