RFR: JDK-8131645 [ARM64] crash on Cavium when using G1

[Tom Benson]

Looks good to me.  One comment, though, which you can choose to ignore:  
My personal preference would be for _release to be used only when 
matching _acquire, and in cases like this, storestore to be used 
instead.  I can't say that Hotspot code in general is consistent on 
that, tho.
Tom

On 9/21/2015 9:39 AM, Eric Caspole wrote:
> Thanks Thomas. I need one more Hotspot reviewer, anybody?
> Eric
>
>
> On 09/19/2015 11:10 AM, Thomas Schatzl wrote:
>> Hi,
>>
>> On Thu, 2015-09-10 at 09:15 -0400, Eric Caspole wrote:
>>> Hi everybody,
>>> Could you review this fix?
>>>
>>> https://bugs.openjdk.java.net/browse/JDK-8131645
>>>
>>> Webrev:
>>> http://cr.openjdk.java.net/~ecaspole/JDK-8131645/00/webrev/
>>>
>>> The problem was that without a store membar on the CodeRootSetTable
>>> creation, readers could see invalid memory leading to an assert or
>>> crash. This is more noticeable on ARM64 with relaxed memory semantics.
>>> Thanks Stefan K for noticing the exact spot for this fix.
>>
>>> This ran all night with the failing test with no crashes and passes 
>>> JPRT.
>>
>>    just in case another review is required, looks good.
>>
>> Thomas
>>