RFR: 8138966: Intermittent SEGV running ParallelGC

[Dean Long]

Oh, and thanks for finding this bug!

dl

On 10/29/2015 1:32 PM, Dean Long wrote:
> I'll let GC folks review for correctness, but I was just wondering if 
> you could use
> load_aquire() and release_store() instead of acquire() and release().
>
> dl
>
> On 10/29/2015 11:12 AM, Andrew Haley wrote:
>> We have observed intermittent crashes when running with
>> -XX:+UseParallelGC on a many-core AArch64 system.  These appear as
>> apparently random segfaults.
>>
>> I've tracked it down to a race between threads when updating the block
>> table in ParallelCompactData::calc_new_pointer() :
>>
>>     if (!region_ptr->blocks_filled()) {
>>      PSParallelCompact::fill_blocks(addr_to_region_idx(addr));
>>      region_ptr->set_blocks_filled();
>>    }
>>
>> Neither blocks_filled() nor set_blocks_filled() have any memory fences
>> so it is possible for a thread to observe a partially-filled block
>> table.
>>
>> http://cr.openjdk.java.net/~aph/8138966/
>>
>> This is shared code so I'll need someone to push this for me.
>>
>> Andrew.
>