RFR: 8202591: Remove usage of CMSEdenChunksRecordAlways in defNewGeneration.cpp

Kim Barrett kim.barrett at oracle.com
Fri May 4 19:04:42 UTC 2018


> On May 4, 2018, at 4:58 AM, Stefan Karlsson <stefan.karlsson at oracle.com> wrote:
> 
> Hi all,
> 
> Please review this patch to remove the usage of CMSEdenChunksRecordAlways in defNewGeneration.cpp.
> 
> http://cr.openjdk.java.net/~stefank/8202591/webrev.01/
> https://bugs.openjdk.java.net/browse/JDK-8202591

Looks good.  Thanks for the detailed description below.

> This is a step towards getting rid of CMS dependent code from defNewGeneration.cpp.
> 
> DefNewGeneration::allocate and DefNewGeneration::par_allocate both have this code:
> 
> #if INCLUDE_ALL_GCS
> if (CMSEdenChunksRecordAlways && _old_gen != NULL) {
>    if (_old_gen != NULL) {
>       _old_gen->sample_eden_chunk();
>     }
> #endif
> 
> _old_gen is TenuredGeneration when running with Serial, which uses the default sample_eden_chunk() implementation in Generation:
>  virtual void sample_eden_chunk() {}
> 
> _old_gen is ConcurrentMarkSweepGeneration when running with CMS, which uses:
>  void sample_eden_chunk() {
>    //Delegate to collector
>    return collector()->sample_eden_chunk();
>  }
> 
> and this is the code in CMSCollector's sample_eden_chunk:
> void CMSCollector::sample_eden_chunk() {
>  if (CMSEdenChunksRecordAlways && _eden_chunk_array != NULL) {
> 
> Note also that CMSEdenChunksRecordAlways is always true, even when we're not running with CMS.
> 
> I propose that we get rid of the CMSEdenChunksRecordAlways check in defNewGeneration.cpp.
> 
> Thanks,
> StefanK




More information about the hotspot-gc-dev mailing list