RFR: 8286066: FillerObject_klass should be loaded as early as possible [v5]

[Thomas Schatzl]

On Wed, 4 May 2022 08:20:05 GMT, Jie Fu <jiefu at openjdk.org> wrote:

> Then part of the unused memory would be filled with `Object` and the others would be filled with `FillerObject`. I'm not sure if this kind of change is safe and if it would complicate the design/implementation/optimization of HotSpot GC algorithms. What do you think of my worries? Thanks.

The gc filler objects only provide additional verification and easier debugging (i.e. any reference to such a filler object is simply an error somewhere). We don't (and can't) verify that every non-FillerObject is being referenced at all times.

Also, the "wrong" filler objects would also only persist until the first garbage collection, as
* all Universe initialization needs to happen before that (the VM will fail otherwise)
* (g1 only) all objects allocated during initialization are (for g1) non-humongous
* evacuation will reclaim these wrong fillers (and a potential evacuation failure will re-fill the holes with the correct fillers)

This can certainly be combined with early as possible initialization of that global variable with the correct `j.i.v.FillerObject` to reduce the number of these objects.

So I believe this is safe (and actually necessary to avoid specific TLAB sizes to fail anyway) to do.

Thanks,
  Thomas

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

PR: https://git.openjdk.java.net/jdk/pull/8519