RFR: 8326820: Metadata artificially kept alive

[Axel Boldt-Christmas]

On Tue, 18 Jun 2024 13:41:05 GMT, Erik Österlund <eosterlund at openjdk.org> wrote:

>> src/hotspot/share/prims/jvmtiGetLoadedClasses.cpp line 108:
>> 
>>> 106:     // and collect them using the LoadedClassesClosure
>>> 107:     MutexLocker mcld(ClassLoaderDataGraph_lock);
>>> 108:     ClassLoaderDataGraph::loaded_classes_do_keepalive(&closure);
>> 
>> This one looks like it might not be safe to exposes without keeping the classes alive.
>
> Looks like we fetch the mirror with Klass::java_mirror() and not Klass::java_mirror_no_keepalive(). Its tempting to think that java_mirror will, unlike its evil twin function, keep the mirror alive. However, this maps to _java_mirror.resolve() vs _java_mirror.peek(). The difference between these was only a thing in single generation ZGC as the _java_mirror is an OopHandle with strong references. Single generation ZGC was the only collector that needed to keep oops alive with strong reference loads - no other collector does that.
> 
> In summary, unless you use single generation ZGC, we don't seem to keep the mirrors alive that we expose from here.

`ClassLoaderDataGraph::loaded_classes_do_keepalive` is a `keepalive` iteration. That is it will load the holder of the CLD.
https://github.com/openjdk/jdk/blob/08366b1244775e5892bbbb184660821e8774f37a/src/hotspot/share/classfile/classLoaderDataGraph.cpp#L328-L335

-------------

PR Review Comment: https://git.openjdk.org/jdk/pull/19769#discussion_r1644738502