Thanks for bringing this in Paul. The change looks ok to me, and suitable for backport.
Trivially, the test does not need the Classpath exception in its license header. -Chris. On 23 Apr 2015, at 14:00, Paul Sandoz <paul.san...@oracle.com> wrote: > Hi. > > Please review this patch from Doug that fixes an issue where a submitted task > to a f/j pool never gets executed. > > > http://cr.openjdk.java.net/~psandoz/jdk9/JDK-8078490-fj-missed-submissions/webrev/ > > I already counted myself and Alexey as reviewers. > > The issue manifests itself under certain conditions where by a task is > submitted to a ForkJoinPool of a certain size and that task is not joined. In > ForkJoin.externalPush a release write is too weak and needs to be > straightened to a volatile write, which stamps in a memory fence before > signalling that there is work to be performed. > > I anticipate Doug will provide a clearer fix at some point in the future. > This fix is sufficient for now and for backporting to 8u60. > > Paul.
