Re: [Qemu-devel] Qemu start VM in huge page without '-mem-prealloc' will get memory leak?

["Daniel P. Berrange" <berrange@redhat.com>]

On Thu, Oct 26, 2017 at 11:09:49AM +0800, Sam wrote:
> For qemu-2.6.0, in huge page (1G) environment, after kill the qemu process,
> memory which is alloc for the vm could not be released. Detail is bellow.
> Or should I use some specific command to stop vm? I want to know if there
> someone has the same problem?
> 
> The start up command is:
> 
> CMD1="$QEMU_CMD -D qemu.log -trace events=qemu-events-all -enable-kvm -cpu
> qemu64,+vmx,+ssse3,+sse4.1,+sse4.2,+x2apic,+aes,+avx,+vme,
> +pat,+ss,+pclmulqdq,+xsave,level=13 -machine pc,accel=kvm -chardev
> socket,id=hmqmondev,port=55908,host=127.0.0.1,nodelay,server,nowait -mon
> chardev=hmqmondev,id=hmqmon,mode=readline -rtc
> base=utc,clock=host,driftfix=none
> -usb -device usb-tablet -daemonize -nodefaults -nodefconfig
> -no-kvm-pit-reinjection -global kvm-pit.lost_tick_policy=discard -vga std
> -k en-us -smp 8 -name gangyewei-qemutime-1 -m 40960 -boot order=cdn -vnc
> :8,password -drive file=$DISK_0,if=none,id=drive_
> 0,format=qcow2,cache=none,aio=native -device virtio-blk-pci,id=dev_drive_0,
> drive=drive_0,bus=pci.0,addr=0x5 -drive file=$DISK_1,if=none,id=drive_
> 1,format=qcow2,cache=none,aio=native -device virtio-blk-pci,id=dev_drive_1,
> drive=drive_1,bus=pci.0,addr=0x6 -drive file=$DISK_2,if=none,id=drive_
> 2,format=qcow2,cache=none,aio=native -device virtio-blk-pci,id=dev_drive_2,
> drive=drive_2,bus=pci.0,addr=0x7 -device ide-cd,drive=ide0-cd0,bus=ide.1,unit=1
> -drive id=ide0-cd0,media=cdrom,if=none -chardev socket,id=char-n-52b49b80,
> path=/usr/local/var/run/openvswitch/n-52b49b80,server -netdev
> type=vhost-user,id=n-52b49b80,chardev=char-n-52b49b80,vhostforce=on -device


Ok, here you have a vhost-user network device associated with a UNIX
socket

> virtio-net-pci,netdev=n-52b49b80,mac=00:22:52:b4:9b:
> 80,id=netdev-n-52b49b80,addr=0xf$(nic_speed 10000) -object
> memory-backend-file,id=mem,size=40960M,mem-path=/mnt/huge,share=on -numa

and here the QEMU RAM is marked shared.

> node,memdev=mem -pidfile $PID_FILE -chardev socket,path=/opt/cloud/
> workspace/servers/4511f52a-f450-40d3-9417-a1e0a27ed507/qga.sock,server,nowait,id=qga0
> -device virtio-serial -device virtserialport,chardev=qga0,
> name=org.qemu.guest_agent.0"
> 
> The stop script is just kill this process.
> 
> the result of `cat /proc/meminfo` show memory is still there.

I expect what has happened is that QEMU has connected to openvsiwtch via the
vhost-user netdev you have, and shared its guest RAM with openvswitch. Now
the openvswitch process has the 40G RAM page mapped.

Now you kill QEMU and QEMU exits and the kernel releases all its RAM mappings,
but the 40G guest RAM mapping is still used by openvswitch.

IOW, I suspect that openvswitch is not releasing the RAM mapping when QEMU
exits, and so it stays resident.

Take a look at the openvswitch processes to see if any of them have the
40GB RAM mapping still shown.


Regards,
Daniel
-- 
|: https://berrange.com      -o-    https://www.flickr.com/photos/dberrange :|
|: https://libvirt.org         -o-            https://fstop138.berrange.com :|
|: https://entangle-photo.org    -o-    https://www.instagram.com/dberrange :|