Bug 1858250 - malloc_initialized should not be SequentiallyConsistent. r=glandium

I think ReleaseAcquire semantics are sound here.

It might be that relaxed is fine too, but that might be technically
unsound because it could make the writes in malloc_init_hard not visible
to the other threads in time.

Depends on D190623

Differential Revision: https://phabricator.services.mozilla.com/D190624

memory/build/mozjemalloc.cpp

@@ -639,7 +639,7 @@ static void* base_alloc(size_t aSize);
 // threads are created.
 static bool malloc_initialized;
 #else
-static Atomic<bool, SequentiallyConsistent> malloc_initialized;
+static Atomic<bool, MemoryOrdering::ReleaseAcquire> malloc_initialized;
 #endif
 
 static StaticMutex gInitLock MOZ_UNANNOTATED = {STATIC_MUTEX_INIT};
@@ -1518,10 +1518,9 @@ FORK_HOOK void _malloc_postfork_child(void);
 // initialization.
 // Returns whether the allocator was successfully initialized.
 static inline bool malloc_init() {
-  if (malloc_initialized == false) {
+  if (!malloc_initialized) {
     return malloc_init_hard();
   }
-
   return true;
 }
 
@@ -3492,7 +3491,7 @@ class AllocInfo {
   template <bool Validate = false>
   static inline AllocInfo Get(const void* aPtr) {
     // If the allocator is not initialized, the pointer can't belong to it.
-    if (Validate && malloc_initialized == false) {
+    if (Validate && !malloc_initialized) {
       return AllocInfo();
     }