RFR(s): 8152182: Possible overflow in initialzation of _rescan_task_size and _marking_task_size

Jesper Wilhelmsson jesper.wilhelmsson at oracle.com
Tue Apr 5 21:24:39 UTC 2016


Looks good!
/Jesper

Den 5/4/16 kl. 23:19, skrev sangheon:
> Hi all,
>
> After some discussion with Jesper, I updated below:
> 1. Removed a flag value recommendation part. Maybe we need more tests on various
> machines and this way is mostly exception comparing to other flags' handling. So
> removing it.
> 2. Changed comments 'would' -> 'could'.
>
> Updated webrev:
> http://cr.openjdk.java.net/~sangheki/8152182/webrev.01_to_00
> http://cr.openjdk.java.net/~sangheki/8152182/webrev.01_to_00
>
> Thanks,
> Sangheon
>
>
> On 04/05/2016 10:24 AM, sangheon wrote:
>> Hi all,
>>
>> Please review this change for CMSRescanMultiple and CMSConcMarkMultiple flags.
>>
>> Both flags are set by "CardTableModRefBS::card_size_in_words * BitsPerWord *
>> flag" which potentially would make an overflow with their maximum value
>> without setting range. And these flags also would make an arithmetic overflow
>> when calculating with the size and the start address of reserved area. In
>> addition, CMSRescanMultiple needs an alignment check.
>>
>> CR: https://bugs.openjdk.java.net/browse/JDK-8152182
>> Webrev: http://cr.openjdk.java.net/~sangheki/8152182/webrev.00
>> Testing: JPRT, runtime/commandline JTREG for all platforms
>>
>> Thanks,
>> Sangheon
>


More information about the hotspot-gc-dev mailing list