RFR: 8198509: Move satisfy_failed_metadata_allocation out from CollectorPolicy

Stefan Johansson stefan.johansson at oracle.com
Thu Feb 22 13:58:04 UTC 2018


Looks ok,

I don't really care where you move it to, I think both alternatives are 
better than the current.

StefanJ

On 2018-02-22 14:50, Per Liden wrote:
> I'm not too sure about moving this into MetaspaceGC, since Metaspace 
> would then make assumptions about how the GC works. I think I'd rather 
> keep this in CollectedHeap for now. That's also far from perfect, but 
> I'd rather contain the "uglyness" in CollectedHeap for now, and work 
> on getting the right abstraction in place later.
>
> cheers,
> Per
>
> On 02/21/2018 02:34 PM, Stefan Karlsson wrote:
>> Hi all,
>>
>> Please review this patch to move satisfy_failed_metadata_allocation 
>> out from CollectorPolicy.
>>
>>   http://cr.openjdk.java.net/~stefank/8198509/webrev.01/
>>   https://bugs.openjdk.java.net/browse/JDK-8198509
>>
>> This is done in preparation for the work to remove CollectorPolicy:
>>   https://bugs.openjdk.java.net/browse/JDK-8198505
>>
>> The function and the code moves to MetaspaceGC, and CollectedHeap 
>> gets a virtual function that calls 
>> MetaspaceGC::satisfy_failed_metadata_allocation. MetaspaceGC seems 
>> like a good place to put this function, but I'm open for suggestions.
>>
>> The current allocation code would then be:
>>   Metaspace::allocte
>>    virtual CollectedHeap::satisfy_failed_metadata_allocation
>>     MetaspaceGC::satisfy_failed_metadata_allocation
>>
>> And GCs that want to provide alternative implementations to 
>> CollectedHeap::satisfy_failed_metadata_allocation (ZGC for example), 
>> could still easily do so.
>>
>> Thanks,
>> StefanK
>>



More information about the hotspot-gc-dev mailing list