[Qemu-discuss] Behaviour of the qemu-monitor info mem/tlb command

[Daniel Seybold]

Hi everyone,

I'm playing a bit with the qemu-monitor commands, especially with the 
"mem" and "tlb" commands with the intent to get a deeper look into the 
actually memory usage of the guest.

My test setup consists of a host with the following specifications:

CentOS release 6.6 (Final)
Linux version 2.6.32-431.29.2.el6.x86_64
Compiled against library: libvirt 0.10.2
Using library: libvirt 0.10.2
Using API: QEMU 0.10.2
Running hypervisor: QEMU 0.12.1

The guest system is a Ubuntu Server 14.04 (Cloud Image), 4 Cores, 4 GB RAM
On the guest the the load generation tool stress is running and 
allocates 100MB of RAM.

Now I'm trying to get the virtual memory mapping inside the guest system 
with through this command:
virsh qemu-monitor-command --hmp INSTANCEID 'info mem'

Running this aforementioned setup I observed different behaviours of the 
"info mem"/"info tlb" commands.
The follwowing two points describe the possible behaviours when running 
stress. But I wasn't able to determine in which case behaviour 1 or 2 
occurs.

1. I don't get any results when executing info mem/tlb but there is 2Gb 
of RAM allocated inside the VM
2. I get the the virtual memory mapping and physical memory mapping from 
the insode of the VM. But if I sum up all the virtual memory pages I get 
a clearly higher amout of memory which is allocated (nearly 1GB) .  And 
I get the same results  summing up the tlb pages.
Or is this the wrong approach to get a deeper look into the actual 
memory allocation inside a VM? (I'm aware of the virsh dommemstat 
command and  possible agent based solutions inside the VM but I'm trying 
to get a the memory information on the host level)

Thanks in advance for your help!

Cheers,
Daniel