RFR(XS): 8152185: ReferencePendingListLocker incorrectly assumes that the lock is never taken recursively

Per Liden per.liden at oracle.com
Fri Mar 18 14:40:40 UTC 2016


Thanks Stefan.

/Per

On 2016-03-18 15:39, Stefan Karlsson wrote:
> Hi Per,
>
> On 2016-03-18 15:29, Per Liden wrote:
>> Summary: The ReferencePendingListLocker asserts that the pending list
>> lock is never taken recursively. This is incorrect, as it can be taken
>> by a GC operation initiated by the ReferenceHandler thread, which
>> might already own the lock. The asserts were added as part of
>> JDK-8151601 in an attempt to strengthen the code, but they are
>> incorrect and should be removed.
>>
>> Bug: https://bugs.openjdk.java.net/browse/JDK-8152185
>> Webrev: http://cr.openjdk.java.net/~pliden/8152185/webrev.0/
>
> As we have discussed offline, it's quite unsafe to take this lock
> recursively from the ReferenceHandler thread, but your change didn't
> change that, it just found the problem. So, I'm fine with removing the
> assert and handle the real problem as a separate CR.
>
> StefanK
>
>>
>> thanks,
>> Per
>


More information about the hotspot-gc-dev mailing list