RFR 8028275: Metaspace ShrinkGrowTest causes fatal error if run with JFR

[David Holmes]

On 17/01/2014 11:07 PM, Coleen Phillimore wrote:
> On 1/16/2014 11:52 PM, David Holmes wrote:
>> Hi Coleen,
>>
>> On 17/01/2014 8:46 AM, Coleen Phillimore wrote:
>>> Summary: Clean up initialization from Threads::create_vm() so that
>>> exceptions cause vm_exit_during_initialization without an exception
>>> mark.
>>>
>>> My attempt at cleaning up initialization code paths that can throw an
>>> exception.  I only cleaned up the code in Threads::create_vm(). There
>>> are other places where EXCEPTION_MARK can occur after
>>> Universe::is_fully_initialized() and before is_initialization_complete()
>>> deeply embedded in the code.   But I did some refactoring.
>>
>> Why do you check
>>
>> 5210     if (Universe::is_fully_initialized()) {
>
> I can't get the Thread::current() unless the universe is initialized (or
> rather that's the check to prevent me from getting it too soon).

Ah and I was overlooking what HAS_PENDING_EXCEPTION expands into.

>>
>> before calling vm_exit_during_initialization? If we have a pending
>> exception shouldn't we be exiting that way regardless. Otherwise what
>> do we do if we have a pending exception but the universe is not
>> initialized?
>
> There won't be a pending exception without the universe initialized
> because (more secret code) Exceptions::_throw() will do a
> vm_exit_during_initialization() if that's the case - can't create an
> exception object yet.

In that case can you add a comment like:

// else: no pending exception possible - VM will already have aborted

Thanks,
David

>>
>> In threads.cpp you dropped the "if (EnableInvokeDynamic) " guard on
>> the jsr292 class initialization.
>>
> Oops, will add that back.
>> Otherwise I like the way you now use CHECK_JNIERR everywhere, even
>> though we still abort in CreateJavaVM :) One step at a time.
>>
>
> Yes, this has to be incremental.   thanks!
>
> Coleen
>> Thanks,
>> David
>> -----
>>
>>> open webrev at http://cr.openjdk.java.net/~coleenp/8028275/
>>> bug link https://bugs.openjdk.java.net/browse/JDK-8028275
>>>
>>> Tested with nsk.quick.testlist and with the hotspot jtreg tests, and
>>> some custom code to test exception paths.
>>>
>>> Thanks,
>>> Coleen
>>>
>