tux3 doesn't compile on 3.2.0

Daniel Phillips daniel.raymond.phillips at gmail.com
Sun Jan 13 03:08:43 PST 2013


I would suggest using KVM. But if anybody wants to dig into UML and
get it working, I will support them. Any root filesystem image that
works with KVM will also work with UML.

Debugging with UML has some advantages over KVM and some drawbacks.
One big advantage is, UML can act a lot more like a command - you can
start it, get some output to the console and stop it all in one line,
whereas using KVM is more like booting and shutting down a real
machine - don't expect to get a stream of output to your terminal.
Drawbacks of UML include higher virtualization overhead which makes it
not great for benchmarking. On the whole, I would develop more with
UML than KVM if I had the choice.

Unfortunately, UML does not seem to be well maintained, and when it is
maintained, it seems that the maintainers do not care much about
preserving those features that make it nice as a kernel debugging
tool. Like integrating well with gdb. It used to be, you could just
start UML under gcc, and there was a special command option that
prevents it from dropping into the gdb command line on startup. Very
nice for automating tests, then just fall into the debugger on assert.
But I don't think it works like that any more, it's all a lot more
awkward now, to the point where I don't find it very pleasant to work
with. I hope I'm wrong. Maybe I just don't understand the "true
workflow" the UML devs have in mind, or maybe they just don't care
about the kernel debugging use case for UML. Which would be ironic,
because with KVM and Xen so advanced, that is now nearly the only
interesting use case remaining.

Daniel




More information about the Tux3 mailing list