RFC 7038914: VM could throw uncaught OOME in ReferenceHandler thread

[David Holmes]

On 16/05/2013 8:44 PM, Thomas Schatzl wrote:
> On Mon, 2013-05-13 at 13:55 +0200, Thomas Schatzl wrote:
>>    I updated the test program and the patch in java.lang.ref.Reference
>> accordingly.
>>
>> As for the problem of reproducibility, in my tests I had a 100%
>> reproduction rate with the previous version of the test.
>>
>> However, now I also set -XX:-UseTLAB and -Xmx16M in the test program as
>> suggested in some other emails.
>>
>> I will report back with a new webrev after some testing on more
>> platforms as suggested by David.
>
>    a new webrev for the patch is at
>
> http://cr.openjdk.java.net/~tschatzl/7038914/webrev.2/

I think the comment is somewhat confusing, but then the details here are 
quite confusing. I guess the key part of this is that if OOME is thrown 
we don't want to try and load InterruptedException - though I'm unclear, 
based on normal exception processing semantics, when that might occur.

> Testing:
> jprt; manual tests with the test program on a jdk with and without the
> patch on linux 32/64 bit, windows 32/64 bit, and sparc 32/64 bit. The
> test program showed 100% reproduction of the error without the patch,
> and 100% fix success with a jdk containing the patch (each multiple
> tries at that).
> Note that jprt seems to run all jdk unit tests always anyway.

Thanks for the detailed testing and validation.

> I also fixed the copyright date in java/lang/ref/Reference.java.

Not necessary for JDK code as they rely on periodic updates (in contrast 
to some hotspot teams).

> If this patch is considered acceptable, I need two reviewers as usual
> (possibly from GC/hsx team and one from jdk), and one sponsor pushing
> the change. (I am still intent on making Peter the author of the patch
> as both the change and the test program came from him)
>
> I am not even author for the jdk, but can act as a non-Reviewer reviewer
> for the gc/hsx team.

You can count me as a Reviewer and sponsor. I think only a second JDK/TL 
Reviewer is needed here as no impact on hotspot.

Thanks for your efforts with this.

David
-----

> Thanks a lot,
> Thomas
>
>