RFR(S) 8054546: NMT2 leaks memory

Yumin Qi yumin.qi at oracle.com
Tue Aug 19 20:33:05 UTC 2014


Looks good to me.

Yumin

On 8/13/2014 8:34 AM, Zhengyu Gu wrote:
> New native memory tracking implementation tries to baseline memory 
> into an arena. Unfortunately, the class "ReservedMemoryRegion" 
> contains "CommittedMemoryRegion" which is C-heap object. When the 
> baseline is discarded, reset arena memory does not invoke 
> ReservedMemoryRegion's dtor, as the result, embedded committed memory 
> regions are leaked.
>
> The solution is to move memory baseline back to C-heap memory, and 
> ensure ReservedMemoryRegion's dtor does get invoked.
>
> Bug:https://bugs.openjdk.java.net/browse/JDK-8054546 
> <https://bugs.openjdk.java.net/browse/JDK-8054546>
> Webrev: http://cr.openjdk.java.net/~zgu/8054546/webrev.00/ 
> <http://cr.openjdk.java.net/%7Ezgu/8054546/webrev.00/>
>
> Test:
>   Ran SwingSet2.jar with "-XX:NativeMemoryTracking=detail". Used jcmd 
> to baseline memory and do detail comparison. No memory leak was found.
>
>
> Thanks,
>
> -Zhengyu



More information about the hotspot-runtime-dev mailing list