RFR(s): 8152160: SIGFPE in CompactibleFreeListSpaceLAB::compute_desired_plab_size
jon.masamitsu at oracle.com
Tue Mar 22 17:43:54 UTC 2016
On 3/22/2016 10:19 AM, sangheon wrote:
> Hi Per,
> Thank you for reviewing this issue.
> On 03/22/2016 08:59 AM, Per Liden wrote:
>> Hi Sangheon,
>> On 2016-03-22 03:12, sangheon wrote:
>>> Hi all,
>>> Could I have a couple of reviews for this change?
>>> With large value of CMSOldPLABNumRefills, SIGFPE would happen at
>>> CompactibleFreeListSpaceLAB::compute_desired_plab_size() because
>>> routine would be zero by an overflow.
>>> Type-casting to 'double' before the problematic division will resolve
>>> the problem.
>> Casting to double doesn't quite seem like the right thing to do here.
>> Can't this be rephrased to avoid the overflow completely, something
>> _global_num_blocks[i] / _global_num_workers[i] / CMSOldPLABNumRefills
> You are right!
> I like this much more.
> Webrev: http://cr.openjdk.java.net/~sangheki/8152160/webrev.01
>>> CR: https://bugs.openjdk.java.net/browse/JDK-8152160
>>> Webrev: http://cr.openjdk.java.net/~sangheki/8152160/webrev.00
>>> Testing: JPRT, TestOptionsWithRanges.java via RBT and manual test as CR
More information about the hotspot-gc-dev