WITHDRAWN: CRR: 7042285: G1: native memory leak during humongous object allocation

Tony Printezis tony.printezis at oracle.com
Mon May 16 11:14:25 PDT 2011


Hi all,

I'm withdrawing this code review request to extend the change to also 
address a few other issues in the HeapRegionSeq class and its use. I'll 
open the new webrev for code review a bit later in the week (after a bit 
more testing).

Tony

Tony Printezis wrote:
> Hi all,
>
> Can I get a couple of code reviews for this change:
>
> http://cr.openjdk.java.net/~tonyp/7042285/webrev.0/
>
> When we grow the heap we create new HeapRegion instances for the 
> regions we are committing. Unfortunately, when we shrink the heap, we 
> are not reclaiming the HeapRegion instances for the regions we are 
> uncommitting. So, we are leaking HeapRegion instances.
>
> The fix is to keep track of all the HeapRegion instances we create, 
> even if when we shrink the heap (again we don't reclaim them but we 
> don't remove them from the _regions array either), to be able to 
> re-use them in the future (i.e., for a particular region we'll only 
> create its HeapRegion instance at most once, irrespective how many 
> times we'll commit it / uncommit it).
>
> Tony
>
>


More information about the hotspot-gc-dev mailing list