Turmvilla scenario

[Norman Feske]

Hi Vincent,

> Thanks for sharing your experiences on Wi Fi driver and NIC driver with
> NIC Bridge. With renewed confidence, I've persisted with a new GIT
> version of Genode and now I am happy to share that I'm able to get NIC
> Bridge working with Wi Fi driver as well for 2 x sessions of Virtualbox,
> one with an Ubuntu Linux guest and the other with a Windows 7 guest. The
> funny thing is that I used the same configuration file as previous so I
> am not sure why it works now while previously it could not.

great that you got it working! :-)

> As I proceed to use Turmvilla as regular, I would like to share 2 issues
> and see if there are any suggestions to improve.
> 
> a. Shutdown of VMM - Fixes #1687. I would attempt to shutdown my Windows
> or Linux session and the VMM window would hang. I thought it was fixed
> by Fixes #1687 but it seems the problem is still there. I have checked
> and double-checked that Fixes #1687 is in. Any idea ? I've not manage to
> capture the logs for this issue.

I can confirm this issue. The mechanism should work like this:

* At the end of the shutdown procedure of the guest OS, the guest OS is
  expected to power off the machine.

* VirtualBox detects the attempt of the guest OS to power off the
  machine and responds by telling its parent that it want to exit
  (via the 'exit' operation on the parent interface).

* The init instance of the 'linux' subsystem looks into the
  <start> node of the virtualbox component. If the node is
  configured with <exit propagate="yes"/>, init responds to
  the exit of the child by exiting.

* The parent of the 'linux' init instance is the CLI monitor.
  It handles the exit request by killing the corresponding
  subsystem.

In my setup, the shutdown and automated killing of the VirtualBox
subsystem works sometimes. But in most cases, the shutdown of the guest
seem to remain undetected by the VMM. I guess that the power-down
procedure of the guest OS is not properly recognized by VirtualBox, but
the information could be lost somewhere else on the way to CLI monitor.

However, this behavior is merely an inconvenience. It is still possible
to kill the subsystem manually via the CLI monitor to regain the memory
resources used by the subsystem.

> b. Memory leak or stability issue. I have encountered a couple of times
> leaving the Turmvilla scenario with both sessions running (just web
> browsers) for 1-2 hours. When I returned to the machine, the entire
> computer hanged up. I could not even access CLI terminal to terminate
> anything. As usual, besides the logs, any suggestions to troubleshoot
> the issue.

I had this issue twice since the recent update of my turmvilla branch.
When it happened, the LOG messages indicated repeated attempts by
rump_fs to allocate more memory until it eventually got stuck. Once
rump_fs stops working, all users of the Genode partition will block
forever. This includes the VMs and also CLI monitor (since it obtains
the subsystem configurations from the Genode partition).

Emery recently discovered an allocation issue in rump_fs that may be
related to the problem. But I haven't tried the fix yet:

  https://github.com/genodelabs/genode/issues/1752#issuecomment-152488824

Cheers
Norman

-- 
Dr.-Ing. Norman Feske
Genode Labs

http://www.genode-labs.com · http://genode.org

Genode Labs GmbH · Amtsgericht Dresden · HRB 28424 · Sitz Dresden
Geschäftsführer: Dr.-Ing. Norman Feske, Christian Helmuth