crash on ZGC try_mark_object

[Alan Bateman]

On 08/04/2020 20:03, Alen Vrečko wrote:
> Hello, everyone.
>
> Running non-loomed code with ZGC on JDK-14 = everything is fine.
> Running the same load on loomed code on EA build = consistent crash about
> once an hour.
>
> Does it ring any bells?
>
> I found this, but should be fixed already?
> https://bugs.openjdk.java.net/browse/JDK-8229221
>
> # A fatal error has been detected by the Java Runtime Environment:
> #
> #  SIGSEGV (0xb) at pc=0x00007fc0d5c49848, pid=8177, tid=8196
> #
> # JRE version: OpenJDK Runtime Environment (15.0+4) (build 15-loom+4-55)
> # Java VM: OpenJDK 64-Bit Server VM (15-loom+4-55, mixed mode, z gc,
> linux-amd64)
> # Problematic frame:
> # V  [libjvm.so+0xe2b848]  ZMark::try_mark_object(ZMarkCache*, unsigned
> long, bool)+0x1b8
>
One other point to mention is that this is 15-loom-4+55 is based on 
jdk-15+12 from 2020/2/22. If it's "non-loomed code" code that is 
SEGV'ing then it's possible you are running into something that could 
have happened with the main too (e.g. JDK-8237859 in fixed jdk-15+16). 
We recent sync'ed up loom to jdk-15+17 so it's closer to the main line 
and will hopefully refresh the EA build soon too.

-Alan