[Spice-devel] spice crashes

[Alon Levy]

----- Original Message -----
> 2013/2/20 Alon Levy <alevy at redhat.com>:
> 
> >> I haven't recompiled with debugging for the stack trace yet; is it
> >> still necessary, or does the information above suffice?
> >
> > Sorry for dropping the ball on this. I'm afraid it's hard to debug
> > even with this information. If you could run the vm under valgrind
> > (you need to have a valgrind that has randomization disabled - see
> > http://spice-space.org/wiki/index.php?title=Valgrind) it would
> > perhaps point to the culprit. Stack trace would help too, but it
> > won't point to the problem assuming it is memory corruption.
> 
> Hi Alon,
> 
> when trying to use valgrind I'm getting "Syscall param ioctl(generic)
> points to uninitialised byte(s)", so I guess, if I understand
> correctly, I need to compile libssl with -DPURIFY, or is there a flag
> I can use when compiling valgrind? I would rather not experiment too

You are correct, I was mistaken - recompile libssl, not valgrind.

> much, as the machine is serving approximately 40 VMs in production
> use. I'm a bloody beginner; if you can provide me with more explicit
> instructions, so that I don't risk overwriting original libs and
> messing up the system, I'd be more than willing to give it a try.

After recompiling, assuming the lib is under /tmp/purify/libssl.so.X then you could do
LD_LIBRARY_PATH=/tmp/purify valgrind qemu ...

> 
> Under non-debugging circumstances, the crashes can be provoked quite
> easily, by overzealous use of the delete-key in msword or outlook.
> When running under valgrind albeit without randomization disabled, I
> cannot provoke the crashes; I'm guessing, it may be a race condition
> that fails to trigger under the substantial slowdown caused by the
> debugging.

That's too bad. So a backtrace could still be useful. You can also enable more debugging at the server level which might point to the faulting command by adding the following to the command line:

-global qxl-vga.cmdlog=1

> 
> regards and thanks,
> TF
>