RFR jck test fails with C2: vm/jvmti/FollowReferences/fref001/fref00113/fref00113.html
vladimir.kozlov at oracle.com
Mon Jul 2 18:43:07 UTC 2018
On 7/2/18 10:39 AM, Paul Sandoz wrote:
> Please review this fix for 11:
Seems good to me. I will run testing with it.
> This is a regression introduced by constant dynamic that causes a JCK test to fail when using "-Xcomp -XX:-TieredCompilation”.
> The problem is due to a lossy mapping between a constant tag and basic type.
> A constant tag of say JVM_CONSTANT_Class is mapped to a basic type of T_OBJECT. T_OBJECT is mapped to a constant tag of JVM_CONSTANT_String. So the round trip is lossy, and this for some reasons results in the JVMTI test failure under certain conditions. This also includes mapping JVM_CONSTANT_UnresolvedClass to T_OBJECT, which might be the source of the problem.
> I verified all the JCK JVMTI tests pass with "-Xcomp -XX:-TieredCompilation”.
> I am not familiar with the interactions between JVMTI, -Xcomp, and C2. So i don’t fully understand why the failure occurred under certain conditions and not others.
-Xcomp -XX:-TieredCompilation combination means is compiled all called
methods and only C2 JIT is used.
I think we may not running JCK testing with this combination of flags.
The problem was reported by Goetz from SAP.
More information about the hotspot-dev