RFR (XS): 8005108: NPG: Add specialized Metachunk sizes for reflection and anonymous classloaders

Stefan Karlsson stefan.karlsson at oracle.com
Tue Dec 18 04:16:46 PST 2012


On 12/18/2012 11:47 AM, Bengt Rutisson wrote:
>
> Hi Stefan,
>
> Looks good to me.
>
> I think this comment in metaspace.cpp can also be removed:
>
> 2220 // The total words available for metadata allocation.  This
> 2221 // uses Metaspace capacity_words() which is the total words
> 2222 // in chunks allocated for a Metaspace.

My change just reverses some of the changes in 3854. This comment was 
already wrong before that change.

Jon, do you think I should remove or reword the comment.

thanks,
StefanK

> Bengt
>
> On 12/18/12 11:02 AM, Stefan Karlsson wrote:
>> http://cr.openjdk.java.net/~stefank/8005108/webrev.00/
>>
>> Description from the bug:
>> ---
>> The implementation of MetaspaceAux::used_in_bytes(), 
>> capacity_in_bytes() and reserved_in_bytes() changed as part of this 
>> changeset:
>>
>> changeset: 3854:90273fc0a981
>> user: coleenp
>> date: Thu Nov 29 16:50:29 2012 -0500
>> summary: 8000662: NPG: nashorn ant clean test262 out-of-memory with 
>> Java heap
>>
>> Before this change it seems like data + classes = total. After this 
>> does not seem to hold.
>>
>> This can be verified by running a simple test that does System.gc() 
>> with -XX:+PrintHeapAtGC:
>>
>> Change 3853:
>>
>>  Metaspace total 3400K, used 2760K, reserved 108544K
>>   data space 3076K, used 2477K, reserved 6144K
>>   class space 324K, used 283K, reserved 102400
>>
>>
>> Change 3854:
>>
>>  Metaspace total 3400K, used 3400K, reserved 108544K
>>   data space 3076K, used 2476K, reserved 6144K
>>   class space 324K, used 283K, reserved 102400K
>>
>> In this example capacity and reserved add up, but used does not. I 
>> have seen cases where at least capacity also did not add up.
>> ---
>>
>> This patch reverts back to the old implementation. We probably want a 
>> better way to count these values, but right now we need consistent 
>> numbers.
>>
>> StefanK
>



More information about the hotspot-gc-dev mailing list