Review request: 8025059: Metspace::should_expand mixes bytes and words in check against MaxMetaspaceSize
Bengt Rutisson
bengt.rutisson at oracle.com
Thu Sep 19 13:55:41 PDT 2013
On 9/19/13 7:01 PM, Stefan Karlsson wrote:
> On 9/19/13 5:10 PM, Bengt Rutisson wrote:
>>
>> Hi Stefan,
>>
>> Given that we just want to get the byte/word conversion right I think
>> this change is fine. The bug JDK-8024547 contains a long list of
>> improvements so it is a bit difficult to tell, but the asymmetry with
>> using reserved memory for NonClassType and allocated memory for
>> ClassType will go away with some of the changes for JDK-8024547, right?
>>
>> 1345 size_t nonclass_allocated =
>> MetaspaceAux::reserved_bytes(Metaspace::NonClassType);
>> 1346 size_t class_allocated =
>> MetaspaceAux::allocated_capacity_bytes(Metaspace::ClassType);
>> 1347 size_t real_allocated = nonclass_allocated +
>> real_allocated;
>
> Yes. The current code we're working on calls
> MetaspaceAux::committed_bytes() directly.
OK. Sounds good. Then I'm fine with this change.
Bengt
>
> StefanK
>>
>> Thanks,
>> Bengt
>>
>>
>> On 9/19/13 4:59 PM, Stefan Karlsson wrote:
>>> http://cr.openjdk.java.net/~stefank/8025059/webrev.00/
>>> https://bugs.openjdk.java.net/browse/JDK-8025059
>>>
>>> The following code in Metaspace::should_expand:
>>>
>>> if (!FLAG_IS_DEFAULT(MaxMetaspaceSize)) {
>>> size_t real_allocated = Metaspace::space_list()->reserved_words() +
>>> MetaspaceAux::allocated_capacity_bytes(Metaspace::ClassType);
>>> if (real_allocated >= MaxMetaspaceSize) {
>>> return false;
>>> }
>>>
>>> doesn't convert Metaspace::space_list()->reserved_words() to bytes,
>>> as it should.
>>>
>>> Note that when JDK-8024547 gets fixed, this check will be rewritten
>>> to limit the amount of committed memory in both the Class and
>>> NonClass metaspaces. See:
>>> https://bugs.openjdk.java.net/browse/JDK-8024547: MaxMetaspaceSize
>>> should limit the committed memory used by the metaspaces
>>>
>>> Testing: I've run our internal testing that sets the MaxMetaspaceSize.
>>>
>>> thanks,
>>> StefanK
>>
>
More information about the hotspot-dev
mailing list