diff --git a/include/mimalloc/atomic.h b/include/mimalloc/atomic.h
index 971b374a..fe79fbca 100644
--- a/include/mimalloc/atomic.h
+++ b/include/mimalloc/atomic.h
@@ -275,6 +275,15 @@ static inline intptr_t mi_atomic_subi(_Atomic(intptr_t)*p, intptr_t sub) {
   return (intptr_t)mi_atomic_addi(p, -sub);
 }
 
+typedef _Atomic(uintptr_t) mi_atomic_once_t;
+
+// Returns true only on the first invocation
+static inline bool mi_atomic_once( mi_atomic_once_t* once ) {
+  if (mi_atomic_load_relaxed(once) != 0) return false;     // quick test 
+  uintptr_t expected = 0;
+  return mi_atomic_cas_strong_acq_rel(once, &expected, 1); // try to set to 1
+}
+
 // Yield
 #if defined(__cplusplus)
 #include <thread>
diff --git a/src/init.c b/src/init.c
index f82aa534..b402a3df 100644
--- a/src/init.c
+++ b/src/init.c
@@ -540,7 +540,8 @@ static void mi_detect_cpu_features(void) {
 // Initialize the process; called by thread_init or the process loader
 void mi_process_init(void) mi_attr_noexcept {
   // ensure we are called once
-  if (_mi_process_is_initialized) return;
+  static mi_atomic_once_t process_init;
+  if (!mi_atomic_once(&process_init)) return;
   _mi_verbose_message("process init: 0x%zx\n", _mi_thread_id());
   _mi_process_is_initialized = true;
   mi_process_setup_auto_thread_done();
diff --git a/src/prim/osx/alloc-override-zone.c b/src/prim/osx/alloc-override-zone.c
index a517ddea..80bcfa93 100644
--- a/src/prim/osx/alloc-override-zone.c
+++ b/src/prim/osx/alloc-override-zone.c
@@ -420,7 +420,7 @@ __attribute__((constructor(0)))
 #else
 __attribute__((constructor))      // seems not supported by g++-11 on the M1
 #endif
-static void _mi_macos_override_malloc() {
+static void _mi_macos_override_malloc(void) {
   malloc_zone_t* purgeable_zone = NULL;
 
   #if defined(MAC_OS_X_VERSION_10_6) && (MAC_OS_X_VERSION_MAX_ALLOWED >= MAC_OS_X_VERSION_10_6)