Preliminary synopsis was: "PPC64 fails jcstress Unsafe (?) memory ordering tests due to C2 (?) bug"

$ wget https://builds.shipilev.net/jcstress/jcstress-latest.jar
$ curl https://builds.shipilev.net/openjdk-jdk/openjdk-jdk-latest-linux-ppc64le-fastdebug.tar.xz | tar xJf -
$ jdk/bin/java -jar jcstress-latest.jar -t varHandles.byteBuffer.direct.*.acqrel.IntBooleanTest -m quick
...
  [FAILED] o.o.j.t.acqrel.varHandles.byteBuffer.direct.big.acqrel.IntBooleanTest
    (JVM args: [-XX:-TieredCompilation])
    Observed state Occurrences Expectation Interpretation
  false, false, -1 210 FORBIDDEN Seeing previous writes, forbidden with proper guard.
   false, false, 0 156,172 ACCEPTABLE Default value for guard, allowed to see anything.
   false, true, -1 350 ACCEPTABLE Observing the value for guard, allowed to see latest value.
    false, true, 0 1,331 ACCEPTABLE Default value for guard, allowed to see anything.
   true, false, -1 0 FORBIDDEN Seeing previous writes, forbidden with proper guard.
    true, false, 0 0 ACCEPTABLE Default value for guard, allowed to see anything.
    true, true, -1 105,163 ACCEPTABLE Observing the value for guard, allowed to see latest value.
     true, true, 0 337 ACCEPTABLE Default value for guard, allowed to see anything.

Seems to only happen with VarHandles bytebuffer view tests, only with C2 enabled, and maybe only with big-endian views on ppc64le machine. Or maybe we are just "lucky".