diff --git a/src/runtime/debug.go b/src/runtime/debug.go
index fd6cbd484e..82deefa200 100644
--- a/src/runtime/debug.go
+++ b/src/runtime/debug.go
@@ -45,7 +45,7 @@ func NumCPU() int {
 
 // NumCgoCall returns the number of cgo calls made by the current process.
 func NumCgoCall() int64 {
-	var n = int64(ncgocall)
+	var n = int64(atomic.Load64(&ncgocall))
 	for mp := (*m)(atomic.Loadp(unsafe.Pointer(&allm))); mp != nil; mp = mp.alllink {
 		n += int64(mp.ncgocall)
 	}
diff --git a/src/runtime/proc.go b/src/runtime/proc.go
index c4dc8717b0..4c92588a66 100644
--- a/src/runtime/proc.go
+++ b/src/runtime/proc.go
@@ -1507,7 +1507,6 @@ func mexit(osStack bool) {
 	}
 	throw("m not found in allm")
 found:
-	ncgocall += m.ncgocall
 	if !osStack {
 		// Delay reaping m until it's done with the stack.
 		//
@@ -1523,6 +1522,8 @@ found:
 	}
 	unlock(&sched.lock)
 
+	atomic.Xadd64(&ncgocall, int64(m.ncgocall))
+
 	// Release the P.
 	handoffp(releasep())
 	// After this point we must not have write barriers.