RFR: 8142402: G1 should not redirty cards in free regions

Mikael Gerdin mikael.gerdin at oracle.com
Thu Nov 26 14:25:08 UTC 2015


Erik,

On 2015-11-26 15:24, Erik Helin wrote:
> On 2015-11-25, Erik Helin wrote:
>> On 2015-11-11, Thomas Schatzl wrote:
>>> Hi,
>>>
>>> On Tue, 2015-11-10 at 16:36 +0100, Erik Helin wrote:
>>>> And StefanK spotted that we no longer need to use
>>>> G1CollectedHeap::heap() in G1RedirtyLoggedCardsTasks since we store a
>>>> pointer to G1CollectedHeap in the field _g1h. Please see new webrevs:
>>>>
>>>> - incremental
>>>> http://cr.openjdk.java.net/~ehelin/8142402/webrev.01-02/
>>>> - full:
>>>> http://cr.openjdk.java.net/~ehelin/8142402/webrev.02/
>>>>
>>>> Thanks,
>>>> Erik
>>
>> Thomas, Stefan,
>>
>> I discovered (after some more testing) that the condition for when to
>> dirty cards is slightly wrong in this patch.
>>
>> Since free_collection_set is run *after* redirty_logged_cards
>> (which is being called from post_evacuate_collection_set), the regions
>> will not be free when they are being processed by redirty_logged_cards.
>> Therefore, we should only dirty cards if the region won't be freed by
>> free_collection_set. A region will be freed if it is in the
>> collection set and has not had an evacuation failure.
>>
>> Please see updated webrevs:
>> - inc: http://cr.openjdk.java.net/~ehelin/8142402/webrev.02-03/
>> - full: http://cr.openjdk.java.net/~ehelin/8142402/webrev.03/
>
> I discussed this patch with Thomas and Mikael offline and we decided to
> add some more documentation as to why this is needed. I have introduced
> two methods and also added some comments which hopefully make things
> clearer.
>
> Please see new webrevs:
> - inc: http://cr.openjdk.java.net/~ehelin/8142402/webrev.03-04/
> - full: http://cr.openjdk.java.net/~ehelin/8142402/webrev.04/

Ship it.
/Mikael

>
> Thanks,
> Erik
>
>> Testing:
>> - GC test suite will full verification
>>
>> Thanks,
>> Erik
>>
>>>
>>>    looks good to me.
>>>
>>> Thanks,
>>>    Thomas
>>>
>>>



More information about the hotspot-gc-dev mailing list