Add JFR logging structures and apis for GC events

[adpopescu]

This change adds all the necessary structures and apis to add some new GC JFR events.

[adpopescu]

@tajila please review

[adpopescu]

Needs eclipse-omr/omr#8065

[adpopescu]

@tajila @amicic I'd like to try to run through the JFR cmdLineTests for all the different GC policies to make sure that some errant failure does not occur. Let me know if you think that is not needed.

[amicic]

I think the name of both methods are not the best. How about (also @dmitripivkine's opinion):
j9gc_get_unique_GC_count -> j9gc_get_total_cycle_count
j9gc_get_unique_GC_ID -> j9gc_get_unique_cycle_ID

[adpopescu]

I like it, I'll change them.

[amicic]

If we go with that proposal we could/should at some point also match the OMR's name

getUniqueGCCycleCount -> getTotalCycleCount

That would require a few changes: starting with one to introduce new API, replace all invocation sites, and eventually retiring the old API.

[adpopescu]

Updated, will look into the name change on OMR side at a later time.

[amicic]

actually, proceed with the first step: introduce a new API in OMR and replace OMR call sites (but do not remove the old API, not update J9 sites till OMR change is promoted).

[amicic]

I want to at least comment here, but perhaps some of these should be also added into the change:

It makes sense to call this only during an active cycle (between cycle start and cycle end), otherwise 0 is returned. It's primarily meant to be called from cycle start and cycle end hook callback.

It should be called from a thread that it's involved in GC, which has cycle state initialized: typically a GC main or worker thread or in rare moments a mutator thread while paying tax.

I'm also wondering what this will return when we have 2 overlapping cycles (for example Scavenge and Concurrent Global in Gencon, but also PGC and GMP in Balanced). Depending on an exact moment (random one, between cycle start and cycle end) it's called, and/or the thread that's called from, it might return one or the other. But certainly, if called within a cycle start and cycle end hook callback, it should return the cycle ID of that cycle that started/ended, even if there is another cycle already in progress.

[adpopescu]

From the experimentation that I've tried with Gencon, it seems to return the correct ID event if there are overlapping cycles. It was one of the reasons why I had to introduce the API and variable.

[amicic]

consider separating this into 2 PRs, one pure GC change and the other for the rest (JFR)

[adpopescu]

I will be splitting into 2 PRs as suggested.

[adpopescu]

Split into

• GC part: Add apis for GC unique cycle ID #23074
• JFR part: Add JFR logging structures and for GC events #23076