RFR (M): 8025813: G1: Code root marking unbalanced

Thomas Schatzl thomas.schatzl at oracle.com
Wed Apr 16 09:17:02 UTC 2014


Hi all,

On Mon, 2014-03-31 at 10:51 +0200, Thomas Schatzl wrote:
> Hi all,
> 
>   could I have reviews for the following change that parallelizes code
> root scan on a per region's code root chunk basis instead of on the code
> root set of an entire region basis.
> 
> This fixes some imbalances when scanning the code roots with some
> regions containing 90% of code roots.
> 
> The idea is to keep around a current code root chunk pointer for every
> region, that is claimed by the threads using a CAS.

  this change will get obsolete with class unloading changes, so I will
close it as won't fix unless somebody objects (and I get reviews :). 

Target version for this change moved to 8u40, and class unloading is
planned for 8u40 too.

In more detail, with class unloading enabled, this phase will completely
go away. With class unloading disabled, the CLDG scan during ext root
scan will mark all references from nmethods.

Thanks,
  Thomas





More information about the hotspot-gc-dev mailing list