Please review fix for 7196103: NPG: Unable to allocate bit map for parallel garbage collection for the requested heap size

Jon Masamitsu jon.masamitsu at oracle.com
Fri Sep 7 09:47:49 PDT 2012


Thanks.  Ship it.

On 9/7/2012 9:44 AM, Coleen Phillimore wrote:
>
>
> On 9/7/2012 10:52 AM, Jon Masamitsu wrote:
>> Coleen,
>>
>> Could you just enforce this with code instead of an assertion?
>> I don't want to see that assertion in tomorrow's nightlies :-)
>
> Me neither.  I fixed this as suggested.  The minimal 
> ClassMetaspaceSize is 1*M which is always greater than 5 MediumChunks, 
> but now it's safer if someone changes something.
>>
>> *+     assert(class_word_size<  ClassMetaspaceSize/2, "should be no 
>> more then half class metaspace size");*
>>
>> class_word_size = 
>> MIN2(**size_t)SpaceManager::MediumChunk*5,****ClassMetaspaceSize/2**)
>>
>> ClassMetaspaceSize is used to reserve space?  I see that it's not used
>> for the initial Metaspace size.
>
> No, these numbers are independent.
>
> Thanks,
> Coleen
>>
>> Otherwise, looks good.
>>
>> Jon
>>
>> On 9/7/2012 6:33 AM, Coleen Phillimore wrote:
>>>
>>> Please reload.  I incorporated some suggestions and corrections from 
>>> stefank and did some tuning of the class metaspace size.
>>>
>>> Thanks,
>>> Coleen
>>>
>>> On 9/7/2012 7:22 AM, Coleen Phillimore wrote:
>>>> Summary: Don't allocate huge class metaspace size by default on x64
>>>>
>>>> This was preventing other mmap regions from being initialized on 
>>>> windows 64.  I also added some checking around the 
>>>> ClassMetaspaceSize option.  1*M seems to be minimal for startup size
>>>>
>>>> open webrev at http://cr.openjdk.java.net/~coleenp/7196103/
>>>>
>>>> Thanks,
>>>> Coleen
>>>>
>>>>


More information about the hotspot-dev mailing list