Fix sem_post race (bug 14532)

[Joseph S. Myers]

On Fri, 31 Aug 2012, Chris Metcalf wrote:

> On 8/30/2012 5:09 PM, Joseph S. Myers wrote:
> > Bug 14532 is a race condition in sem_post (generic C version), where
> > it wrongly uses an acquire barrier rather than a release barrier, so
> > wrongly allowing memory operations from before the unlock to be moved
> > after it.
> >
> > This patch fixes this in the obvious way and adds a testcase.  Tested
> > mips64.
> 
> This bug is also present on tilegx, and fixed by your patch.

Do you think the patch is OK, then?  If so, I'll commit it.

-- 
Joseph S. Myers
joseph@codesourcery.com