Upgrade to 4.4.50-rt62

[kvmfornfv.git] / kernel / kernel / locking / mcs_spinlock.h

diff --git a/kernel/kernel/locking/mcs_spinlock.h b/kernel/kernel/locking/mcs_spinlock.h
index 5b9102a..c835270 100644 (file)
--- a/kernel/kernel/locking/mcs_spinlock.h
+++ b/kernel/kernel/locking/mcs_spinlock.h
@@ -67,7 +67,13 @@ void mcs_spin_lock(struct mcs_spinlock **lock, struct mcs_spinlock *node)
        node->locked = 0;
        node->next   = NULL;
 
-       prev = xchg_acquire(lock, node);
+       /*
+        * We rely on the full barrier with global transitivity implied by the
+        * below xchg() to order the initialization stores above against any
+        * observation of @node. And to provide the ACQUIRE ordering associated
+        * with a LOCK primitive.
+        */
+       prev = xchg(lock, node);
        if (likely(prev == NULL)) {
                /*
                 * Lock acquired, don't need to set node->locked to 1. Threads