[Vm-dev] Oddd image lock-up on Unix

[Andreas Raab]

Bert Freudenberg wrote:
> A backtrace from gdb might have been useful, too.

Damn, didn't think of that. We *did* poke it via SIGUSR before attaching 
GDB though and the result looked perfectly reasonable:

Received user signal, printing stack trace:

-1742550148 ProcessorScheduler class>idleProcess
-1743108016 [] in ProcessorScheduler class>startUp
-1743107924 [] in BlockContext>newProcess

Given that the VM responded properly I wouldn't suspect there to be a 
problem with nanosleep() or so.

> There was an increasing number of reports of Linux VMs freezing or 
> crashing lately. We suspect some possibly unrelated problems. One is due 
> to oops-as-signed, a symptom is that GC fails to find the end of memory. 
> The other seems related to async i/o. In yet others (and this may be 
> yours) we see nanosleeps that are called with huge negative microsecond 
> counts, yet higher up in the stack the value was 1000 microseconds.

It is the very first time we've seen this kind of problem - usually the 
symptoms are 100% CPU not 0% CPU consumption ;-) This is also why I 
suspected that the delay fix might have to do with it.

Cheers,
   - Andreas