Re: Booting Nested KVM

[Kashyap Chamarthy <kchamart@redhat.com>]

On Thu, Mar 24, 2016 at 11:21:32AM -0400, Jacob Abraham Graff wrote:
> Hello,
> 
> I am a student at Columbia University. My research partner Andy (CC'd)
> and I are working on booting a nested KVM virtual machine on x86.
> However, we have run into a problem that we have not been able to fix.
> 
> The host machine, L0, has the 'kvm-intel.nested=1'  flag set in it's
> grub config file, and the guest operating system, L1, does not
> complain when attempting to start a VM using virt-install.
> 
> When we attempt to actually install, we run into a couple of issues.
> When we run virt-install, using the command:
> 
> virt-install -r 1024 --accelerate -n Ubu \
> -f /var/lib/libvirt/images/guest.img \
> --cdrom /root/ubuntu-14.04.2-server-amd64.iso

> 
> We receive the error message
> 
> Could not start virtual network default

It seems like you don't have the default libvirt network active.  If it
were active, you'd see something like:

    $ virsh net-list
     Name                 State      Autostart     Persistent
    ----------------------------------------------------------
     default              active     yes           yes

If you don't, you can start the 'default' libvirt network manually:

    $ virsh net-start default
    $ virsh net-autostart default
    $ virsh net-list # Here the 'default' network should be active

Then, re-run your `virt-install` invocation.

Once you are inside L1, check if you have KVM character device exposed
in it, by doing `file /dev/kvm`.  If it is not present, then you should
expose the virtualization extensions to your L1 guest, by doing:

    $ virt-xml L1VM --edit --cpu host-passthrough,clearxml=yes

Reboot the guest, and check again if the KVM char device is present.

* * *

Speaking of installing VMs, for what it's worth, I prefer this  approach
for installing VMs quickly (I'm using Fedora-based distribution as an
example) using the 'virt-builder'[1] utility.

  (1) Build the VM template.
  
    # Install the tools
    $ sudo dnf install libguestfs-tools-c

    # Build the template
    $ virt-builder fedora-23 --size 40G

  (2) Import the image into libvirt

    $ virt-install --name f21vm --ram 2048 \
        --disk path=./fedora-23.img,format=raw \
        --nographics --import --os-variant fedora22

Then you can manipulate it using `virt-manager` (or from command-line,
if you prefer that way).


[1] http://libguestfs.org/virt-builder.1.html

Some related notes:

https://kashyapc.fedorapeople.org/virt/procedure-to-enable-nested-virt-on-intel-machines.txt

> In order to get around this error message, we run pass the nonetworks flag:
> 
> virt-install -r 1024 --accelerate --nonetworks -n Ubu \
> -f /var/lib/libvirt/images/guest.img \
> --cdrom /root/ubuntu-14.04.2-server-amd64.iso
> 
> At this point, the install seems to succeed, and the nested VM starts
> to boot. However, we cannot see any output at all - the console hangs,
> and we cannot type anything into the console at this point. If we open
> a new console, and run virt-viewer, we get what appears to be an
> ubuntu login screen, but the text is unreadable (blurry and pixelated)
> and again, we cannot type anything into this window. The virt-manager
> gui seems to indicate that this guest does have activity, which makes
> it seem strange that we are unable to communicate with it in any way.
> 
> Is there something we have missed? Do we have to specify IO in some
> special way?
> 

-- 
/kashyap