RFR/RFC [8u]: Enable weak JNI handling with Shenandoah after JDK-8176100

[Roman Kennke]

Looks like we've got a clash between rscratch1 used in
g1_write_barrier_pre() and the rscratch that's passed as tmp from
sharedRuntime_aarch64.cpp. You need this fix:

diff -r 9f523f185755 src/cpu/aarch64/vm/sharedRuntime_aarch64.cpp
--- a/src/cpu/aarch64/vm/sharedRuntime_aarch64.cpp	Mon May 27 12:23:24
2019 +0200
+++ b/src/cpu/aarch64/vm/sharedRuntime_aarch64.cpp	Mon May 27 09:27:22
2019 -0400
@@ -2066,7 +2066,7 @@
       __ g1_write_barrier_pre(noreg /* obj */,
                               r0 /* pre_val */,
                               rthread /* thread */,
-                              rscratch1 /* tmp */,
+                              rscratch2 /* tmp */,
                               true /* tosca_live */,
                               true /* expand_call */);
     }

This makes the failing test pass for me.

It probably warrants a more careful fix, extra asserts in
g1_write_barrier_pre() and checking if newer versions have similar problems.

Roman

> In Shenandoah/jdk8u code, we have a workaround for weak JNI handling:
>   http://hg.openjdk.java.net/shenandoah/jdk8/hotspot/rev/119e9a5b24d5
>   http://hg.openjdk.java.net/shenandoah/jdk8/hotspot/rev/d305c31da9f5
> 
> This workaround is not necessary anymore, after we got the backports of JDK-8176100:
>   http://hg.openjdk.java.net/shenandoah/jdk8/hotspot/rev/32bc598624bd
>   http://hg.openjdk.java.net/shenandoah/jdk8/hotspot/rev/327f9583546e
> 
> So, this changeset rolls back the workaround and injects Shenandoah calls where appropriate.
> 
> 8u candidate fix:
>   http://cr.openjdk.java.net/~shade/shenandoah/8u-enable-weak-jni/webrev.01/
> 
> Note aarch64 fails! This might mean the upstream patch is incomplete. And I cannot debug it properly
> without having working aarch64 machine.
> 
> Testing: hotspot_gc_shenandoah (includes gc/shenandoah/jni/TestJNIGlobalRefs.sh
> test that fails without the workaround, and passes again without the workaround _and_ the fixes in
> platform-specific code); tested with x86_64
>