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>

include/qemu/rcu.h

@@ -27,6 +27,7 @@
 #include "qemu/thread.h"
 #include "qemu/queue.h"
 #include "qemu/atomic.h"
+#include "qemu/sys_membarrier.h"
 
 #ifdef __cplusplus
 extern "C" {
@@ -82,7 +83,7 @@ static inline void rcu_read_lock(void)
     atomic_set(&p_rcu_reader->ctr, ctr);
 
     /* Write p_rcu_reader->ctr before reading RCU-protected pointers.  */
-    smp_mb();
+    smp_mb_placeholder();
 }
 
 static inline void rcu_read_unlock(void)
@@ -96,13 +97,13 @@ static inline void rcu_read_unlock(void)
 
     /* Ensure that the critical section is seen to precede the
      * store to p_rcu_reader->ctr.  Together with the following
-     * smp_mb(), this ensures writes to p_rcu_reader->ctr
+     * smp_mb_placeholder(), this ensures writes to p_rcu_reader->ctr
      * are sequentially consistent.
      */
     atomic_store_release(&p_rcu_reader->ctr, 0);
 
     /* Write p_rcu_reader->ctr before reading p_rcu_reader->waiting.  */
-    smp_mb();
+    smp_mb_placeholder();
     if (unlikely(atomic_read(&p_rcu_reader->waiting))) {
         atomic_set(&p_rcu_reader->waiting, false);
         qemu_event_set(&rcu_gp_event);

include/qemu/sys_membarrier.h

@@ -0,0 +1,17 @@
+/*
+ * Process-global memory barriers
+ *
+ * Copyright (c) 2018 Red Hat, Inc.
+ *
+ * Author: Paolo Bonzini <pbonzini@redhat.com>
+ */
+
+#ifndef QEMU_SYS_MEMBARRIER_H
+#define QEMU_SYS_MEMBARRIER_H 1
+
+/* Keep it simple, execute a real memory barrier on both sides.  */
+static inline void smp_mb_global_init(void) {}
+#define smp_mb_global()            smp_mb()
+#define smp_mb_placeholder()       smp_mb()
+
+#endif

util/rcu.c

@@ -93,10 +93,10 @@ static void wait_for_readers(void)
         }
 
         /* 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 @@ void synchronize_rcu(void)
     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 rcu_init_child(void)
 
 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