Review request: 8043086 : Hotspot is expected to report OOM which is occurred String.intern(), but crashes in JDK8u5

[Daniel D. Daugherty]

On 5/28/14 4:45 AM, Poonam Bajaj wrote:
> Hi David,
>
> Here's the updated webrev:
> http://cr.openjdk.java.net/~poonam/8043086/webrev.01/

line 566: _chunk = k;                 // restore the value of _chunk
     Perhaps: // restore the previous value of _chunk

No need for me to re-review if you decide to take the tweak.

Dan


>
> Thanks,
> Poonam
>
> On 5/28/2014 10:54 AM, David Holmes wrote:
>> Hi Poonam,
>>
>> On 28/05/2014 12:20 PM, Poonam Bajaj wrote:
>>> Hi,
>>>
>>> Request for reviews for the following fix:
>>> Bug: 8043086 <https://bugs.openjdk.java.net/browse/JDK-8043086>: 
>>> Hotspot
>>> is expected to report OOM which is occurred String.intern(), but 
>>> crashes
>>> in JDK8u5
>>> Webrev: http://cr.openjdk.java.net/~poonam/8043086/webrev.00/
>>>
>>> Fix details: Bug  'JDK-8011773: Some tests on Interned String crashed
>>> JVM with OOM' was fixed in JDK 8.  But with this fix another crash
>>> happens while throwing java.lang.OutOfMemoryError exception when an out
>>> of memory error occurs in String.intern(). The reason being in case of
>>> an allocation failure, the _chunk of ResourceMark could be left null 
>>> and
>>> that can cause crash in the destructor of ResourceMark. These changes
>>> fix that.
>>
>> You already have a "tmp" copy of chunk_, so this should suffice as a 
>> fix:
>>
>>  562   Chunk *k = _chunk;            // Get filled-up chunk address
>>  563   _chunk = new (alloc_failmode, len) Chunk(len);
>>  564
>>  565   if (_chunk == NULL) {
>>
>> +        _chunk = k;  // restore
>>
>>  566     return NULL;
>>  567   }
>>
>> David
>> -----
>>
>>> Thanks,
>>> Poonam
>>>