RFR (S) 8208677: Move inner metaspace cleaning out of class unloading

[Erik Österlund]

Hi Coleen,

Thank you for moving this stuff out of the class unloading path.

One tiny question... would it be possible to have 
ClassLoaderDataGraph::should_clean_metaspaces() not have side effects, 
and make the resetting explicit? I always get confused when getters have 
side effects. Or maybe change the name to 
should_clean_metaspaces_and_reset() to make it clear that there are side 
effects, similar to has_previous_versions() vs 
has_previous_versions_and_reset().

Thanks,
/Erik

On 2018-08-06 20:47, coleen.phillimore at oracle.com wrote:
> Summary: move to safepoint cleanup actions to do if needed.
>
> This is in support of concurrent class unloading.   See bug 
> description for more details.
>
> open webrev at http://cr.openjdk.java.net/~coleenp/8208677.01/webrev
> bug link https://bugs.openjdk.java.net/browse/JDK-8208677
>
> Ran mach5 hs-tier1-5, with no failures on Friday.  hs-tier1,2 with 
> expected failures today.
>
> The test/hotspot/jtreg/runtime/RedefineRunningMethods.java and 
> RedefineLeak.java test this code.  Fixed and verified with logging in 
> the first test.
>
> thanks,
> Coleen