RFR (S) 8232112: MDO extra_data_lock leaks during class unloading

[David Holmes]

On 17/10/2019 12:18 pm, coleen.phillimore at oracle.com wrote:
> On 10/16/19 8:25 PM, David Holmes wrote:
>> Hi Coleen,
>>
>> On 16/10/2019 11:48 pm, coleen.phillimore at oracle.com wrote:
>>> Summary: call the MDO destructor during class unloading.
>>>
>>> Also moved the other C heap deallocation from method_data to 
>>> release_C_heap_structures.  Methods won't have method_data unless 
>>> they're deallocated from the InstanceKlass.  The other paths to 
>>> deallocate_contents are before the method is called.
>>
>> I'm not understanding why delete is not used to run the destructor and 
>> free the memory. Is this memory that is being specially managed?
> 
> The _extra_data_lock Mutex is embedded in the MethodData, and MethodData 
> is allocated in Metaspace which is deleted en-masse when a class is 
> unloaded.  It also may be deleted during redefinition, when the 
> InstanceKlass is deleted (all old methods are not running).

Okay - thanks for clarifying.

The change seems fine to me.

Thanks,
David
-----

> I actually tested a version where I changed _extra_data_lock to a Mutex* 
> and used new/delete for it.  This caused a slight degredation in one of 
> our performance benchmarks, so I changed it to call the destructor 
> explicitly (with Kim's help with syntax).
> 
> Thanks,
> Coleen
>>
>> Thanks,
>> David
>>
>>> Tested with tier1 all Oracle platforms, and tier2-6 on linux-x64.
>>>
>>> open webrev at 
>>> http://cr.openjdk.java.net/~coleenp/2019/8232112.01/webrev
>>> bug link https://bugs.openjdk.java.net/browse/JDK-8232112
>>>
>>> Thanks,
>>> Coleen
>>>
>>>
>