RFR(XS) 8152846: Creation of ModuleEntryTable Investigate Need For OrderAccess::storestore()

[David Holmes]

Hi Harold,

On 6/04/2016 3:52 AM, harold seigel wrote:
> Hi,
>
> Please review this small change to fix bug 8152846.
>
> This change removes an unneeded OrderAccess::storestore() call.
>
> Webrev: http://javaweb.us.oracle.com/~hseigel/webrev/bug_8152846/

Please ensure you provide an accessible webrev link please.

I have a concern with this code. I think the storeStore as written was 
in the wrong place. The main concern with lock-free read of a field like 
_packages is that if you see a non-null value then you also see all the 
stores that were made as part of the construction of the object 
_packages refers to. In that regard I think the original code:

     // Ensure _packages is stable, since it is examined without a lock
      OrderAccess::storestore();
     _packages = new 
PackageEntryTable(PackageEntryTable::_packagetable_entry_size);
    }
   return _packages;

Should really have been:

  temp = new PackageEntryTable(PackageEntryTable::_packagetable_entry_size);
  OrderAccess::storeStore();
  _packages = tmp;
  ...

though to be sure we need to examine the lock-free usage of _packages.

And to be really correct the store to _packages should be a storeRelease 
(no storestore barrier) and the lock-free read should be a readAcquire.

David
-----

> JBS Bug: https://bugs.openjdk.java.net/browse/JDK-8152846
>
> The fix was tested with JCK API, Lang, and VM tests, the UTE
> non-colocated quick tests, and the hotspot, JDK vm, java/io, java/lang,
> java/util, and java/security tests.  Testing with UTE colocated tests is
> in progress.
>
> Thanks, Harold