Review request: 8004697: SIGSEGV on Solaris sparc with -XX:+UseNUMA

Stefan Karlsson stefan.karlsson at oracle.com
Tue Mar 12 21:09:00 UTC 2013


Thanks, Jesper.

StefanK

On 2013-03-11 12:33, Jesper Wilhelmsson wrote:
> The change looks good.
> I especially like the assert :-)
>
> Ship it!
> /Jesper
>
> On 11/3/13 11:04 AM, Stefan Karlsson wrote:
>> http://cr.openjdk.java.net/~stefank/8004697/webrev.00/
>>
>> The JVM crashes because it accidentally frees memory in the survivor 
>> regions, at
>> one point where it is supposed to only free freeing memory in the eden.
>>
>> The code in os::scan_pages(start, end, ...) in solaris_os.cpp looks 
>> at pages
>> starting beyond 'end'. This causes 
>> MutableNUMASpace::LGRPSpace::scan_pages to
>> free memory outside eden, when the last page starts in, but ends 
>> outside of eden.
>>
>> Testing: The JVM used to crash within in a couple of minutes. With 
>> the patch,
>> the test has been running over the weekend without any crashes.
>>
>> thanks,
>> StefanK




More information about the hotspot-gc-dev mailing list