yeskery

Java Object.hashCode()返回的是对象内存地址？

Category：软件开发 Tag：Java 2021-03-22 Views：22

基于OpenJDK 8

一直以为Java Object.hashCode()的结果就是通过对象的内存地址做相关运算得到的，但是无意在网上看到有相应的意见争论，故抽时间从源码层面验证了剖析了hashCode的默认计算方法。

先说结论：OpenJDK8 默认hashCode的计算方法是通过和当前线程有关的一个随机数+三个确定值，运用Marsaglia’s xorshift scheme随机数算法得到的一个随机数。和对象内存地址无关。

下面通过查找和分析OpenJDK8源码实现来一步步分析。

查找java.lang.Object.hashCode()源码

  1. public native int hashCode();

导出Object的JNI头文件

切换到Object.class文件所在目录，执行 javah -jni java.lang.Object，得到java_lang_Object.h文件，文件内容如下：

  1. /* DO NOT EDIT THIS FILE - it is machine generated */
  2. #include <jni.h>
  3. /* Header for class java_lang_Object */
  5. #ifndef _Included_java_lang_Object
  6. #define _Included_java_lang_Object
  7. #ifdef __cplusplus
  8. extern "C" {
  9. #endif
  10. /*
  11.  * Class:     java_lang_Object
  12.  * Method:    registerNatives
  13.  * Signature: ()V
  14.  */
  15. JNIEXPORT void JNICALL Java_java_lang_Object_registerNatives
  16.   (JNIEnv *, jclass);
  18. /*
  19.  * Class:     java_lang_Object
  20.  * Method:    getClass
  21.  * Signature: ()Ljava/lang/Class;
  22.  */
  23. JNIEXPORT jclass JNICALL Java_java_lang_Object_getClass
  24.   (JNIEnv *, jobject);
  26. /*
  27.  * Class:     java_lang_Object
  28.  * Method:    hashCode
  29.  * Signature: ()I
  30.  */
  31. JNIEXPORT jint JNICALL Java_java_lang_Object_hashCode
  32.   (JNIEnv *, jobject);
  34. /*
  35.  * Class:     java_lang_Object
  36.  * Method:    clone
  37.  * Signature: ()Ljava/lang/Object;
  38.  */
  39. JNIEXPORT jobject JNICALL Java_java_lang_Object_clone
  40.   (JNIEnv *, jobject);
  42. /*
  43.  * Class:     java_lang_Object
  44.  * Method:    notify
  45.  * Signature: ()V
  46.  */
  47. JNIEXPORT void JNICALL Java_java_lang_Object_notify
  48.   (JNIEnv *, jobject);
  50. /*
  51.  * Class:     java_lang_Object
  52.  * Method:    notifyAll
  53.  * Signature: ()V
  54.  */
  55. JNIEXPORT void JNICALL Java_java_lang_Object_notifyAll
  56.   (JNIEnv *, jobject);
  58. /*
  59.  * Class:     java_lang_Object
  60.  * Method:    wait
  61.  * Signature: (J)V
  62.  */
  63. JNIEXPORT void JNICALL Java_java_lang_Object_wait
  64.   (JNIEnv *, jobject, jlong);
  66. #ifdef __cplusplus
  67. }
  68. #endif
  69. #endif

查看Object的native方法实现

OpenJDK源码链接：http://hg.openjdk.java.net/jdk8u/jdk8u/jdk/file/3462d04401ba/src/share/native/java/lang/Object.c ，查看Object.c文件，可以看到hashCode()的方法被注册成由JVM_IHashCode方法指针来处理。

  1. static JNINativeMethod methods[] = {  
  2.     {"hashCode",    "()I",                    (void *)&JVM_IHashCode},//hashcode的方法指针JVM_IHashCode  
  3.     {"wait",        "(J)V",                   (void *)&JVM_MonitorWait},  
  4.     {"notify",      "()V",                    (void *)&JVM_MonitorNotify},  
  5.     {"notifyAll",   "()V",                    (void *)&JVM_MonitorNotifyAll},  
  6.     {"clone",       "()Ljava/lang/Object;",   (void *)&JVM_Clone},  
  7. };

而JVM_IHashCode方法指针在 openjdk\hotspot\src\share\vm\prims\jvm.cpp中定义为：

  1. JVM_ENTRY(jint, JVM_IHashCode(JNIEnv* env, jobject handle))  
  2.   JVMWrapper("JVM_IHashCode");  
  3.   // as implemented in the classic virtual machine; return 0 if object is NULL  
  4.   return handle == NULL ? 0 : ObjectSynchronizer::FastHashCode (THREAD, JNIHandles::resolve_non_null(handle)) ;  
  5. JVM_END

从而得知，真正计算获得hashCode的值是ObjectSynchronizer::FastHashCode

ObjectSynchronizer::fashHashCode方法的实现

openjdk\hotspot\src\share\vm\runtime\synchronizer.cpp 找到其实现方法。

  1. intptr_t ObjectSynchronizer::FastHashCode (Thread * Self, oop obj) {
  2.   if (UseBiasedLocking) {
  3.     // NOTE: many places throughout the JVM do not expect a safepoint
  4.     // to be taken here, in particular most operations on perm gen
  5.     // objects. However, we only ever bias Java instances and all of
  6.     // the call sites of identity_hash that might revoke biases have
  7.     // been checked to make sure they can handle a safepoint. The
  8.     // added check of the bias pattern is to avoid useless calls to
  9.     // thread-local storage.
  10.     if (obj->mark()->has_bias_pattern()) {
  11.       // Box and unbox the raw reference just in case we cause a STW safepoint.
  12.       Handle hobj (Self, obj) ;
  13.       // Relaxing assertion for bug 6320749.
  14.       assert (Universe::verify_in_progress() ||
  15.               !SafepointSynchronize::is_at_safepoint(),
  16.              "biases should not be seen by VM thread here");
  17.       BiasedLocking::revoke_and_rebias(hobj, false, JavaThread::current());
  18.       obj = hobj() ;
  19.       assert(!obj->mark()->has_bias_pattern(), "biases should be revoked by now");
  20.     }
  21.   }
  23.   // hashCode() is a heap mutator ...
  24.   // Relaxing assertion for bug 6320749.
  25.   assert (Universe::verify_in_progress() ||
  26.           !SafepointSynchronize::is_at_safepoint(), "invariant") ;
  27.   assert (Universe::verify_in_progress() ||
  28.           Self->is_Java_thread() , "invariant") ;
  29.   assert (Universe::verify_in_progress() ||
  30.          ((JavaThread *)Self)->thread_state() != _thread_blocked, "invariant") ;
  32.   ObjectMonitor* monitor = NULL;
  33.   markOop temp, test;
  34.   intptr_t hash;
  35.   markOop mark = ReadStableMark (obj);
  37.   // object should remain ineligible for biased locking
  38.   assert (!mark->has_bias_pattern(), "invariant") ;
  40.   if (mark->is_neutral()) {
  41.     hash = mark->hash();              // this is a normal header
  42.     if (hash) {                       // if it has hash, just return it
  43.       return hash;
  44.     }
  45.     hash = get_next_hash(Self, obj);  // allocate a new hash code
  46.     temp = mark->copy_set_hash(hash); // merge the hash code into header
  47.     // use (machine word version) atomic operation to install the hash
  48.     test = (markOop) Atomic::cmpxchg_ptr(temp, obj->mark_addr(), mark);
  49.     if (test == mark) {
  50.       return hash;
  51.     }
  52.     // If atomic operation failed, we must inflate the header
  53.     // into heavy weight monitor. We could add more code here
  54.     // for fast path, but it does not worth the complexity.
  55.   } else if (mark->has_monitor()) {
  56.     monitor = mark->monitor();
  57.     temp = monitor->header();
  58.     assert (temp->is_neutral(), "invariant") ;
  59.     hash = temp->hash();
  60.     if (hash) {
  61.       return hash;
  62.     }
  63.     // Skip to the following code to reduce code size
  64.   } else if (Self->is_lock_owned((address)mark->locker())) {
  65.     temp = mark->displaced_mark_helper(); // this is a lightweight monitor owned
  66.     assert (temp->is_neutral(), "invariant") ;
  67.     hash = temp->hash();              // by current thread, check if the displaced
  68.     if (hash) {                       // header contains hash code
  69.       return hash;
  70.     }
  71.     // WARNING:
  72.     //   The displaced header is strictly immutable.
  73.     // It can NOT be changed in ANY cases. So we have
  74.     // to inflate the header into heavyweight monitor
  75.     // even the current thread owns the lock. The reason
  76.     // is the BasicLock (stack slot) will be asynchronously
  77.     // read by other threads during the inflate() function.
  78.     // Any change to stack may not propagate to other threads
  79.     // correctly.
  80.   }
  82.   // Inflate the monitor to set hash code
  83.   monitor = ObjectSynchronizer::inflate(Self, obj);
  84.   // Load displaced header and check it has hash code
  85.   mark = monitor->header();
  86.   assert (mark->is_neutral(), "invariant") ;
  87.   hash = mark->hash();
  88.   if (hash == 0) {
  89.     hash = get_next_hash(Self, obj);
  90.     temp = mark->copy_set_hash(hash); // merge hash code into header
  91.     assert (temp->is_neutral(), "invariant") ;
  92.     test = (markOop) Atomic::cmpxchg_ptr(temp, monitor, mark);
  93.     if (test != mark) {
  94.       // The only update to the header in the monitor (outside GC)
  95.       // is install the hash code. If someone add new usage of
  96.       // displaced header, please update this code
  97.       hash = test->hash();
  98.       assert (test->is_neutral(), "invariant") ;
  99.       assert (hash != 0, "Trivial unexpected object/monitor header usage.");
  100.     }
  101.   }
  102.   // We finally get the hash
  103.   return hash;
  104. }

该方法中

  1. // Load displaced header and check it has hash code
  2.   mark = monitor->header();
  3.   assert (mark->is_neutral(), "invariant") ;
  4.   hash = mark->hash();
  5.   if (hash == 0) {
  6.     hash = get_next_hash(Self, obj);
  7. ...
  8. }

对hash值真正进行了计算，查看get_next_hash方法源码http://hg.openjdk.java.net/jdk8u/jdk8u/hotspot/file/87ee5ee27509/src/share/vm/runtime/synchronizer.cpp#l555

  1. static inline intptr_t get_next_hash(Thread * Self, oop obj) {
  2.   intptr_t value = 0 ;
  3.   if (hashCode == 0) {
  4.      // This form uses an unguarded global Park-Miller RNG,
  5.      // so it's possible for two threads to race and generate the same RNG.
  6.      // On MP system we'll have lots of RW access to a global, so the
  7.      // mechanism induces lots of coherency traffic.
  8.      value = os::random() ;
  9.   } else
  10.   if (hashCode == 1) {
  11.      // This variation has the property of being stable (idempotent)
  12.      // between STW operations.  This can be useful in some of the 1-0
  13.      // synchronization schemes.
  14.      intptr_t addrBits = cast_from_oop<intptr_t>(obj) >> 3 ;
  15.      value = addrBits ^ (addrBits >> 5) ^ GVars.stwRandom ;
  16.   } else
  17.   if (hashCode == 2) {
  18.      value = 1 ;            // for sensitivity testing
  19.   } else
  20.   if (hashCode == 3) {
  21.      value = ++GVars.hcSequence ;
  22.   } else
  23.   if (hashCode == 4) {
  24.      value = cast_from_oop<intptr_t>(obj) ;
  25.   } else {
  26.      // Marsaglia's xor-shift scheme with thread-specific state
  27.      // This is probably the best overall implementation -- we'll
  28.      // likely make this the default in future releases.
  29.      unsigned t = Self->_hashStateX ;
  30.      t ^= (t << 11) ;
  31.      Self->_hashStateX = Self->_hashStateY ;
  32.      Self->_hashStateY = Self->_hashStateZ ;
  33.      Self->_hashStateZ = Self->_hashStateW ;
  34.      unsigned v = Self->_hashStateW ;
  35.      v = (v ^ (v >> 19)) ^ (t ^ (t >> 8)) ;
  36.      Self->_hashStateW = v ;
  37.      value = v ;
  38.   }
  40.   value &= markOopDesc::hash_mask;
  41.   if (value == 0) value = 0xBAD ;
  42.   assert (value != markOopDesc::no_hash, "invariant") ;
  43.   TEVENT (hashCode: GENERATE) ;
  44.   return value;
  45. }

对于OpenJDK8版本，其默认配置http://hg.openjdk.java.net/jdk8u/jdk8u/hotspot/file/87ee5ee27509/src/share/vm/runtime/globals.hpp#l1127 为：

  1.  product(intx, hashCode, 5,                                                \
  2.           "(Unstable) select hashCode generation algorithm")                \

其对应的hashCode计算方案为：

  1.     // Marsaglia's xor-shift scheme with thread-specific state
  2.      // This is probably the best overall implementation -- we'll
  3.      // likely make this the default in future releases.
  4.      unsigned t = Self->_hashStateX ;
  5.      t ^= (t << 11) ;
  6.      Self->_hashStateX = Self->_hashStateY ;
  7.      Self->_hashStateY = Self->_hashStateZ ;
  8.      Self->_hashStateZ = Self->_hashStateW ;
  9.      unsigned v = Self->_hashStateW ;
  10.      v = (v ^ (v >> 19)) ^ (t ^ (t >> 8)) ;
  11.      Self->_hashStateW = v ;
  12.      value = v ;

其中Thread->_hashStateX, Thread->_hashStateY, Thread->_hashStateZ, Thread->_hashStateW在http://hg.openjdk.java.net/jdk8u/jdk8u/hotspot/file/87ee5ee27509/src/share/vm/runtime/thread.cpp#I263 有定义：

  1.    // thread-specific hashCode stream generator state - Marsaglia shift-xor form
  2.   _hashStateX = os::random() ;
  3.   _hashStateY = 842502087 ;
  4.   _hashStateZ = 0x8767 ;    // (int)(3579807591LL & 0xffff) ;
  5.   _hashStateW = 273326509 ;

所以，JDK8 的默认hashCode的计算方法是通过和当前线程有关的一个随机数+三个确定值，运用Marsaglia’s xorshift scheme随机数算法得到的一个随机数。对xorshift算法有兴趣可以参考原论文：https://www.jstatsoft.org/article/view/v008i14/xorshift.pdf 。
xorshift是由George Marsaglia发现的一类伪随机数生成器，其通过移位和与或计算，能够在计算机上以极快的速度生成伪随机数序列。其算法的基本实现如下：

  1. unsigned long xor128(){
  2. static unsigned long x=123456789,y=362436069,z=521288629,w=88675123;
  3. unsigned long t;
  4. t=(xˆ(x<<11));x=y;y=z;z=w; return( w=(wˆ(w>>19))ˆ(tˆ(t>>8)) );

这就和上面计算hashCode的OpenJDK代码对应了起来。

其他几类hashCode计算方案：

  1. if (hashCode == 0) {
  2.      // This form uses an unguarded global Park-Miller RNG,
  3.      // so it's possible for two threads to race and generate the same RNG.
  4.      // On MP system we'll have lots of RW access to a global, so the
  5.      // mechanism induces lots of coherency traffic.
  6.      value = os::random() ;
  7.   }
  • hashCode == 1
    此类方案将对象的内存地址，做移位运算后与一个随机数进行异或得到结果
  1. if (hashCode == 1) {
  2.      // This variation has the property of being stable (idempotent)
  3.      // between STW operations.  This can be useful in some of the 1-0
  4.      // synchronization schemes.
  5.      intptr_t addrBits = cast_from_oop<intptr_t>(obj) >> 3 ;
  6.      value = addrBits ^ (addrBits >> 5) ^ GVars.stwRandom ;
  7.   }
  • hashCode == 2
    此类方案返回固定的1
  1. if (hashCode == 2) {
  2.      value = 1 ;            // for sensitivity testing
  3.   }
  • hashCode == 3
    此类方案返回一个自增序列的当前值
  1. if (hashCode == 3) {
  2.      value = ++GVars.hcSequence ;
  3.   }
  • hashCode == 4
    此类方案返回当前对象的内存地址
  1. if (hashCode == 4) {
  2.      value = cast_from_oop<intptr_t>(obj) ;
  3.   }

可以通过在JVM启动参数中添加-XX:hashCode=4，改变默认的hashCode计算方式。

参考资料：

  1. https://srvaroa.github.io/jvm/java/openjdk/biased-locking/2017/01/30/hashCode.html
  2. https://en.wikipedia.org/wiki/Xorshift
  3. http://www.cnblogs.com/mengyou0304/p/4763220.html
  4. http://stackoverflow.com/questions/2427631/how-is-hashcode-calculated-in-java
  5. http://hllvm.group.iteye.com/group/topic/39183

本文转载自：https://www.jianshu.com/p/be943b4958f4

评论

Post comment Click Refresh verification code

Tip

The feature is not yet available

Close