[9] RFR(S): 8065740: compiler/codecache/OverflowCodeCacheTest.java fails with SIGSEGV

[Tobias Hartmann]

Thanks, Igor.

Best,
Tobias

On 26.11.2014 07:49, Igor Ignatyev wrote:
> On 11/26/2014 09:48 AM, Tobias Hartmann wrote:
>> Albert, Igor, thanks for the review.
>>
>> On 25.11.2014 16:22, Igor Ignatyev wrote:
>>> Hi Tobias,
>>>
>>> isn't it another manifestation of JDK-8064669[1,2] ?
>>
>> Yes, I missed that. I'll close the bug as a duplicate.
>>
>> Since you didn't push yet, could you add the two code cleanups (typo in
>> whitebox.cpp and removal of comment in sweeper.cpp) to your change?
> sure.
>>
>> Thanks,
>> Tobias
>>
>>>
>>> [1] https://jbs.oracle.com/bugs/browse/JDK-8064669
>>> [2]
>>> http://mail.openjdk.java.net/pipermail/hotspot-compiler-dev/2014-November/016283.html
>>>
>>>
>>>
>>> Igor
>>>
>>> On 11/25/2014 05:58 PM, Tobias Hartmann wrote:
>>>> Hi,
>>>>
>>>> please review the following patch.
>>>>
>>>> Bug: https://bugs.openjdk.java.net/browse/JDK-8065740
>>>> Webrev: http://cr.openjdk.java.net/~thartmann/8065740/webrev.00/
>>>>
>>>> Problem:
>>>> The test allocates CodeBlobs in the code cache by using the Whitebox API method
>>>> 'WhiteBox::allocate_code_blob' introduced by JDK-8059624. The VM crashes while
>>>> sweeping because it encounters an allocated but uninitialized CodeBlob in the
>>>> code cache (crash happens in NMethodIterator::next_method because we call
>>>> '_code_blob->is_nmethod()' on an uninitialised CodeBlob).
>>>> The problem is that 'WhiteBox::allocate_code_blob' gives up the CodeCache_lock
>>>> before initializing the CodeBlob. This allows the sweeper thread to encounter a
>>>> yet uninitialized CodeBlob while iterating over the code cache.
>>>>
>>>> Solution:
>>>> I moved the initialization of the CodeBlob into the scope of the
>>>> CodeCache_lock.
>>>> I also fixed some typos and removed a comment that became meaningless with
>>>> JDK-8046809.
>>>>
>>>> Testing:
>>>> - Failing testcase
>>>> - JPRT
>>>>
>>>> Thanks,
>>>> Tobias
>>>>