RFR: Consistenly order STS scope outside of OOM scope to avoid deadlocks

[Aleksey Shipilev]

On 10/17/2018 12:31 PM, Roman Kennke wrote:
> All implemented here:
> http://cr.openjdk.java.net/~rkennke/sts-oom-deadlock/webrev.01/

Looks much better, thanks.

So, this thing can deadlock before hitting the assert, right? This is probably okay if we just want
to catch unlucky placements that do not deadlock most of the time.

*) I think assert messages should say something about the ordering:

 159     assert(!ShenandoahThreadLocalData::is_evac_allowed(Thread::current()), "sanity");

For example:

  assert(!ShenandoahThreadLocalData::is_evac_allowed(Thread::current()),
         "STS should be joined before evac scope");

  assert(!ShenandoahThreadLocalData::is_evac_allowed(Thread::current()),
         "STS should be left after evac scope");


Thanks,
-Aleksey