From 5c705166da112395dcff70685a4d19ea6e9f426b Mon Sep 17 00:00:00 2001
From: Man Cao <manc@openjdk.org>
Date: Thu, 15 Jun 2023 20:13:36 +0000
Subject: [PATCH] 8309688: Data race on
 java.io.ClassCache$CacheRef.strongReferent

Reviewed-by: rriggs, shade
---
 src/java.base/share/classes/java/io/ClassCache.java | 5 +++++
 1 file changed, 5 insertions(+)

diff --git a/src/java.base/share/classes/java/io/ClassCache.java b/src/java.base/share/classes/java/io/ClassCache.java
index 48e7a36c05a..4b41cf4671b 100644
--- a/src/java.base/share/classes/java/io/ClassCache.java
+++ b/src/java.base/share/classes/java/io/ClassCache.java
@@ -39,6 +39,11 @@ abstract class ClassCache<T> {
 
     private static class CacheRef<T> extends SoftReference<T> {
         private final Class<?> type;
+        // This field is deliberately accessed without sychronization. ClassValue
+        // provides synchronization when CacheRef is published. However, when
+        // a thread reads this field, while another thread is clearing the field, it
+        // would formally constitute a data race. But that data race is benign, and
+        // fixing it could introduce noticeable performance penalty, see JDK-8309688.
         private T strongReferent;
 
         CacheRef(T referent, ReferenceQueue<T> queue, Class<?> type) {