RFR: 8252294: Remove OopsInGenClosure usage from younger_refs_iterate
stefan.karlsson at oracle.com
Tue Aug 25 08:12:24 UTC 2020
Please review this patch to remove the usage of OopsInGenClosure from
younger_refs_iterate and related functions.
Before the patch we had this call sequence:
Generation::younger_refs_iterate (CardGeneration as the dynamic type)
and we passed the OopsInGenClosure down the entire way.
After CMS, when we only have one concrete implementation of
GenCollectedHeap, we can simplify this. I turned the code inside out and
we now have:
SerialHeap::young_process_roots (I moved to SerialHeap)
- Because of this we don't need the virtual
Generation::younger_refs_iterate, nor the unimplemented version in
DefNewGeneration. It suffices to have a concrete
- Inside CardTableRS::younger_refs_in_space_iterate we only used
OopsInGenClosure to extract the gen_boundary() value. (The value set in
OopsInGenClosure::set_generation()). Now that I've straightened out the
code we can now directly extract the "gen boundary" value out of
CardGeneration (old_gen) instead of passing this information down
through the closure.
- I moved young_process_roots to SerialHeap, since I need to have access
to CardGeneration and GenCollectedHeap only provides a Generation* field.
- Note: SerialHeap::young_process_roots still use OopsInGenClosure. I
have a follow-up patch that changes that. I'd like to keep this patch
- Note: There's a lot of dead code and potential cleanups that can be
made to CardTableRS and GenCollectedHeap. I don't intend to fix that
with this RFE.
Local testing and currently running through tier1-7 on Linux.
More information about the hotspot-gc-dev