RFR/RFC: Passive should opt-in the barriers, not opt-out

[Roman Kennke]

Am 17.11.2017 um 16:07 schrieb Aleksey Shipilev:
> http://cr.openjdk.java.net/~shade/shenandoah/passive-opt-in/webrev.01/
>
> Current "passive" heuristics is diagnostic and only runs Full GCs. In that mode, barriers are not
> required, and our Wiki mentions that is the way to dissect the barriers cost. As Shenandoah adoption
> goes up, we might want to have this in more convenient form. That is, let's make barriers opt-in,
> not opt-out with "passive" heuristics. This allows users to make diagnostic runs with all barriers
> disabled at once, without knowing what barriers they need to disable. New code also prints the
> barrier names it implicitly disables, so that users can selectively enable them back without looking
> into the source or our wiki.
>
> We still allocate fwdptr slot, because Full GC code needs it for storing relocation data.
>
> Our existing testing covers all interesting tests with "passive" heuristics already.
>
> Testing: hotspot_gc_shenandoah {fastdebug|release}
>
> Thanks,
> -Aleksey
>
Good idea. Patch looks good too.