Fix UB (data race) in Q_GLOBAL_STATIC

The store to guard in the inner function's critical
section was not synchronized-with the load at the
start of the function:

    T1                    T2
    guard.load()
    mutex.lock()
    guard.load()
    d = new Type
    guard.store()
                          guard.load()
                          // use d
    mutex.unlock()

The use of d in T2 does not synchronize with
the write to d in T1 -> data race -> UB.

Fix by storing with release memory ordering,
so that the guard.load() in T2 synchronizes
with the guard.store() in T1.

Change-Id: I5c1cd1fa097c6397cb0b48b0d8e8012f95978558
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
Reviewed-by: Olivier Goffart (Woboq GmbH) <ogoffart@woboq.com>

1 files changed, 1 insertions, 1 deletions

diff --git a/src/corelib/global/qglobalstatic.h b/src/corelib/global/qglobalstatic.h
index 41fc151652..22194d8be7 100644
--- a/src/corelib/global/qglobalstatic.h
+++ b/src/corelib/global/qglobalstatic.h
@@ -110,7 +110,7 @@ QT_BEGIN_NAMESPACE
                         guard.store(QtGlobalStatic::Destroyed);         \
                     }                                                   \
                 } cleanup;                                              \
-                guard.store(QtGlobalStatic::Initialized);               \
+                guard.storeRelease(QtGlobalStatic::Initialized);        \
             }                                                           \
         }                                                               \
         return d;                                                       \