RFR: 8027230: Overflow in java.lang.instrument.Instrumentation.getObjectSize() method

[David Holmes]

On 21/05/2014 11:01 PM, Peter Allwin wrote:
> D’oh! Great catch, I’ll update before the push if that’s OK.

Sure - I assume it is the comment that will be updated :)

Thanks,
David

>
> Thanks,
>
> /peter
>
> On 21 May 2014, at 13:32, David Holmes <david.holmes at oracle.com> wrote:
>
>> Hi Peter,
>>
>> One inconsistency
>>
>> 27  * Test to verify GetObjectSize does not overflow on a 600K element int[]
>>
>> but
>>
>>   34         int[] a = new int[600_000_000];
>>
>> looks more like 600M
>>
>> David
>>
>> On 21/05/2014 9:10 PM, Peter Allwin wrote:
>>> Thanks Leonid, Serguei and David for your reviews!
>>>
>>> Updated webrev is here: http://cr.openjdk.java.net/~allwin/8027230/webrev.01
>>>
>>> Changes:
>>>
>>> 	- Agent process is now started trough ProcessBuilder
>>> 	- Non 64bit platforms are immediately skipped
>>> 	- Spacing before if/catch
>>> 	- Test TEST.groups updated
>>> 		Need compact3:
>>> 			serviceability/jvmti/GetObjectSizeOverflow.java (uses java.lang.Instrument)
>>> 			serviceability/jvmti/TestRedefineWithUnresolvedClass.java (uses java.lang.Instrument)
>>> 		Need JDK
>>> 			serviceability/jvmti/8036666/GetObjectLockCount.java (used com.sun.jdi)
>>>
>>> Thanks!
>>> /peter
>>>
>>>
>>> On 21 May 2014, at 07:54, David Holmes <david.holmes at oracle.com> wrote:
>>>
>>>> On 21/05/2014 1:19 AM, Leonid Mesnik wrote:
>>>>> Peter
>>>>>
>>>>>    35  * @run main/othervm -Xmx4000m -javaagent:agent.jar
>>>>> GetObjectSizeOverflowAgent
>>>>>
>>>>> I think that "-Xmx4000m" cause test failure for 32-bit VM. It would be
>>>>> better to use another 1 process builder.
>>>>
>>>> If you need a 4GB heap to test this you will have to limit it to 64-bit platforms.
>>>>
>>>>> Also I think you need to add your test into need_jdk because you use
>>>>> jar. Could you please check this with embedded team.
>>>>
>>>> JDKToolFinder should get jar from the compile JDK rather than the test JDK so that should be okay.
>>>>
>>>> However the use of the agent/instrumentation is limited to compact3 profile (if java.lang.instrument is used) and not the minimal VM, so changes are needed in TEST.groups. It looks like we have a number of missing updates to the groups file for the  test/serviceability/jvmti tests.
>>>>
>>>> David
>>>> -----
>>>>
>>>>
>>>>> Leonid
>>>>>
>>>>> On 20.05.2014 19:02, Peter Allwin wrote:
>>>>>> Hello!
>>>>>>
>>>>>> Please review this simple fix for an integer overflow in JVMTI
>>>>>> GetObjectSize().
>>>>>>
>>>>>> webrev: http://cr.openjdk.java.net/~allwin/8027230/webrev.00/
>>>>>> cr: https://bugs.openjdk.java.net/browse/JDK-8027230
>>>>>>
>>>>>>
>>>>>> Testing:
>>>>>>     New regression test
>>>>>>     nsk.quick-jvmti.testlist
>>>>>>
>>>>>> Thanks!
>>>>>> /peter
>>>>>
>>>
>