[8u] RFR 8252904: VM crashes when JFR is used and JFR event class is transformed

[Andrey Petushkov]

added jtreg test https://cr.openjdk.java.net/~apetushkov/8252904.jdk/
the test reliably fails on slowdebug VM without the bugfix and reliably
passes with a fix. note that the test will not reliably fail on a
product VM without fix, only occasionally

Andrey

On 09.09.2020 12:50, Andrew Dinn wrote:
> Hi Andrey,
>
> On 08/09/2020 17:39, Andrey Petushkov wrote:
>
>> updated webrev https://cr.openjdk.java.net/~apetushkov/8252904.01/
> Thanks. That looks better. I agree that aligning the resource mark with
> the lifetime of the ClassFileParser instance is the correct model.
>
> Of course, in SystemDictionary::resolve_from_stream that lifetime is
> longer than it needs to be. However, fixing that is not needed to deal
> with the bug so I'm more than happy to leave well alone, minimizing
> perturbation of the existing code.
>
>>>> Last but not least, what have you done to test this patch?
>>> We have a proprietary though simple test which instruments all the java
>>> classes loaded into VM. I've used this test in conjunction with the
>>> slowdebug VM to test the fix. As mentioned, the VM crashes reliably
>>> without a fix, while successfully working with it (given the nature of
>>> the test it in fact verifies not only that JFR event classes can now be
>>> instrumented successfully, but also serves some sanity check that all
>>> other class loads happened via other code paths are not affected, to the
>>> extent the slowdebug VM assertions can verify)
>>>
>>> I cannot publish the source code of a test but if necessary I can
>>> prepare a minimal reproducer
> Yes, please.
>
> regards,
>
>
> Andrew Dinn
> -----------
> Red Hat Distinguished Engineer
> Red Hat UK Ltd
> Registered in England and Wales under Company Registration No. 03798903
> Directors: Michael Cunningham, Michael ("Mike") O'Neill