membarrier: introduce qemu/sys_membarrier.h

This new header file provides heavy-weight "global" memory barriers that
enforce memory ordering on each running thread belonging to the current
process.  For now, use a dummy implementation that issues memory barriers
on both sides (matching what QEMU has been doing so far).

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
diff --git a/util/rcu.c b/util/rcu.c
index 7366dc5..5676c22 100644
--- a/util/rcu.c
+++ b/util/rcu.c
@@ -93,10 +93,10 @@
         }
 
         /* Here, order the stores to index->waiting before the loads of
-         * index->ctr.  Pairs with smp_mb() in rcu_read_unlock(),
+         * index->ctr.  Pairs with smp_mb_placeholder() in rcu_read_unlock(),
          * ensuring that the loads of index->ctr are sequentially consistent.
          */
-        smp_mb();
+        smp_mb_global();
 
         QLIST_FOREACH_SAFE(index, &registry, node, tmp) {
             if (!rcu_gp_ongoing(&index->ctr)) {
@@ -145,9 +145,9 @@
     qemu_mutex_lock(&rcu_sync_lock);
 
     /* Write RCU-protected pointers before reading p_rcu_reader->ctr.
-     * Pairs with smp_mb() in rcu_read_lock().
+     * Pairs with smp_mb_placeholder() in rcu_read_lock().
      */
-    smp_mb();
+    smp_mb_global();
 
     qemu_mutex_lock(&rcu_registry_lock);
     if (!QLIST_EMPTY(&registry)) {
@@ -376,6 +376,7 @@
 
 static void __attribute__((__constructor__)) rcu_init(void)
 {
+    smp_mb_global_init();
 #ifdef CONFIG_POSIX
     pthread_atfork(rcu_init_lock, rcu_init_unlock, rcu_init_child);
 #endif