RFR: 8048248: G1 Class Unloading after completing a concurrent mark cycle
stefan.karlsson at oracle.com
Tue Jul 1 13:44:48 UTC 2014
Please, review this patch to enable unloading of classes and other
metadata after a G1 concurrent cycle.
The patch includes the following changes:
1) Tracing through alive Klasses and CLDs during concurrent mark,
instead of marking all of them during the initial mark pause.
2) Making HeapRegions walkable in the presence of unparseable objects
due to their classes being unloaded.
3) The process roots code has been changed to allow G1's combined
initial mark and scavenge.
4) The CodeBlobClosures have been refactored to distinguish the marking
variant from the oop updating variants.
5) Calls to the G1 pre-barrier have been added to some places, such as
the StringTable, to guard against object resurrection, similar to how
j.l.ref.Reference#get is treated with a read barrier.
6) Parallelizing the cleaning of metadata and compiled methods during
the remark pause.
A number of patches to prepare for this RFE has already been pushed to
8047362: Add a version of CompiledIC_at that doesn't create a new
8047326: Consolidate all CompiledIC::CompiledIC implementations and move
it to compiledIC.cpp
8047323: Remove unused _copy_metadata_obj_cl in G1CopyingKeepAliveClosure
8047373: Clean the ExceptionCache in one pass
8046670: Make CMS metadata aware closures applicable for other collectors
8035746: Add missing Klass::oop_is_instanceClassLoader() function
8035648: Don't use Handle in java_lang_String::print
8035412: Cleanup ClassLoaderData::is_alive
8035393: Use CLDClosure instead of CLDToOopClosure in
8034764: Use process_strong_roots to adjust the StringTable
8034761: Remove the do_code_roots parameter from process_strong_roots
8033923: Use BufferingOopClosure for G1 code root scanning
8033764: Remove the usage of StarTask from BufferingOopClosure
8012687: Remove unused is_root checks and closures
8047818: G1 HeapRegions can no longer be ContiguousSpaces
8048214: Linker error when compiling G1SATBCardTableModRefBS after
include order changes
8047821: G1 Does not use the save_marks functionality as intended
8047820: G1 Block offset table does not need to support generic Space
8047819: G1 HeapRegionDCTOC does not need to inherit ContiguousSpaceDCTOC
8038405: Clean up some virtual fucntions in Space class hierarchy
8038412: Move object_iterate_careful down from Space to ContigousSpace
8038404: Move object_iterate_mem from Space to CMS since it is only ever
used by CMS
8038399: Remove dead oop_iterate MemRegion variants from SharedHeap,
Generation and Space classe
8037958: ConcurrentMark::cleanup leaks BitMaps if VerifyDuringGC is enabled
8032379: Remove the is_scavenging flag to process_strong_roots
We've been running Kitchensink, gc-test-suite, internal nightly testing
and test lists, and CRM FA benchmarks.
StefanK & Mikael Gerdin
More information about the hotspot-dev