RFR(XS): 8178497: Bug in MutableNUMASpace::ensure_parsability
sangheon.kim at oracle.com
Wed Nov 22 21:23:33 UTC 2017
Thank you for reviewing this.
On 11/21/2017 05:29 AM, Stefan Johansson wrote:
> Hi Sangheon,
> On 2017-11-20 20:05, sangheon.kim wrote:
>> Hi all,
>> Can I have some reviews for this tiny pointer arithmetic change?
>> Current code doesn't use pointer arithmetic, so the the resulting
>> values are wrong(too small). i.e. adding two values first and then
>> typecast to HeapWord* which is wrong here.
>> intptr_t cur_top = X;
>> size_t touched_words = XX;
>> align_down((HeapWord*)(cur_top + touched_words), XXX);
>> This should be 'align_down( (HeapWord*)cur_top + touched_words, XXXX);'.
>> As I don't see any good reason of using 'intptr_t', changed to use
>> 'HeapWord*' and changed related stuff.
>> I didn't add regression test or some further investigation as this is
>> clear that the calculation is wrong. And hard to provoke the problem
>> CR: https://bugs.openjdk.java.net/browse/JDK-8178497
>> Webrev: http://cr.openjdk.java.net/~sangheki/8178497/webrev.0
> Thanks for taking care of this. I think the changes is good, but I
> would feel even more certain if you ran some testing with NUMA and
> Parallel to make sure we haven't overlooked anything.
Okay, I added some explanation about what would happen.
And also ran hundred runs of GCOld with NUMA enabled on Solaris, of
course no problem.
Hope this is sufficient to go. :)
>> Testing: local building
More information about the hotspot-gc-dev