Request for review: JDK-8012052: java/lang/invoke/6987555/Test6987555.java crashes with assert(mcs != NULL) failed: MethodCounters cannot be NULL

Jiangli Zhou jiangli.zhou at oracle.com
Mon Apr 15 10:03:06 PDT 2013


Hi David,

On 04/14/2013 08:01 PM, David Holmes wrote:
>>
>> The race is rare. If it does happen, the overwritten one would just
>> become "orphan".
>
> Does "orphan" == leak? Do we need to fix this - either now or in a 
> follow up CR? Are there other non-thread-safe operations on 
> MethodHandles?

If it does happen, then it is a leak. If we want to completely eliminate 
the possibility, a lock is needed. I've experimented using the 
MethodData_lock. There were some issues with using the MethodData_lock. 
A separate lock probably is needed. I will file a follow up CR. There 
are no other non-thread-safe operations on MethodHandles.

Thanks,
Jiangli

>
> Thanks,
> David
>



More information about the hotspot-runtime-dev mailing list