RFR(S): JDK-8019845 NPG: Memory leak during class redefinition

Mikael Gerdin mikael.gerdin at oracle.com
Fri Jul 26 15:29:28 UTC 2013

On 07/26/2013 05:11 PM, Coleen Phillmore wrote:
> On 7/26/2013 10:04 AM, Daniel D. Daugherty wrote:
>> On 7/26/13 5:50 AM, Coleen Phillmore wrote:
>>> On 7/26/2013 6:12 AM, frederic parain wrote:
>>>>>     If the allocate/deallocate subsystem doesn't support partial
>>>>> deallocates
>>>>>      then why does deallocate() have a size parameter? If you always
>>>>> have to
>>>>>      deallocate() exactly what you allocate()'ed, then deallocate()
>>>>> only
>>>>> needs
>>>>>      the pointer originally returned by allocate() and doesn't need
>>>>> the
>>>>> size
>>>>>      parameter.
>>>> I'm not familiar enough with Metaspace to answer this questions. I'll
>>>> let the NPG experts respond to this one.
>>> We pass the size because the block that is allocated does  not have
>>> the size stored in it (to save 1 word of space per allocation).   So
>>> you can't just free the pointer.
>> Isn't there some sanity check info in the non-product version?
>> Can the allocate() size be saved there and verified at deallocate()
>> time in non-product mode? I wouldn't be surprised if such a check
>> revealed other smaller leaks.
>> The {Redefine,Retransform}BigClass tests are rather abusive in their
>> hunt for a specific style of memory leak and we were lucky that they
>> caught this leak also. We're also fortunate that Ivan G. improved
>> the tests to catch smaller memory leaks.
> Yes, this was good work to improve this test and detect this!
> And yes, we should save size to compare against in debug mode (I thought
> we did in an earlier version of this code).

It appears that we save the size in a Metablock before it's returned to 
the operator new(), see the exit paths from Metaspace::allocate calling 
The problem seems to be that initialize just returns the raw pointer 
"result" instead of returning "result + overhead()" (since 
get_raw_word_size accounts for Metablock::overhead)

It should be pretty simple to do, and it would be good if more people 
get familiar with this code, so I suggest that someone else does it :)


> Coleen
>> Dan
>>> Coleen
>>>> Thanks,
>>>> Fred

More information about the hotspot-gc-dev mailing list