最近在抽时间阅读 JDK 的源码,主要是 GC 还有 Safepoint 相关的源码,发现很多我在之前拜读网上各种 JVM 原理大作时候由于没有看源码导致我对于底层原理的误解。果然,一百个人读水浒传,就有一百种水浒传。还是需要更加深入的了解下源码,才能更好地理解 JVM,进行调优。这个系列,将在讲述 Java GC 各种原理的基础上,结合对应的源码分析,并附上源码地址。因为JVM的源码更新还是很快的,尤其是 GC 这一块,但是基本原理,应该大体不会变,附上源码地址,旨在让各位读者掌握这些原理最新实现情况。本文在撰写的时候,会紧跟最新版本的 JDK。
GC(Garbage Collection) 是目前很多编程语言自带的特性,例如Java,Python;GC是一个很好的特性,能让使用这个语言编程的程序员不去关心内存回收,并且降低内存泄漏和内存溢出发生的概率。
了解Java GC,需要先知道 Java 最基础的对象在内存中究竟是如何存储的。我们专注于 HotSpot 虚拟机实现,来详细阐述对象存储结构。首先我们来了解一个概念,对象指针(OOPs,Ordinary Object Pointers), 也就是对象头的主要部分。
1. 对象指针(OOPs,Ordinary Object Pointers)
对象指针的实现,可以参考oop.hpp:
class oopDesc {
private:
volatile markWord _mark;
union _metadata {
Klass* _klass;
narrowKlass _compressed_klass;
} _metadata;
}
从源代码可以简单看出,对象指针包括:
1.标记字(Mark Word): 一组标记,描述了对象的状态,包括对象默认哈希值(如果没有覆盖默认的 hashcode() 方法,则哈希值在 hashcode() 方法被调用之后,会被记录到标记字之中)、对象的形状(是否是数组)、锁状态(偏向锁等锁信息,值得一提的是偏向锁在 Java 15 中废弃:Disable and Deprecate Biased Locking)、数组长度(如果标记显示这个对象是数组,描述了数组的长度)。标记字的实现仅仅包含一个uintptr_t类型,所以,在 32 位和 64 位虚拟机上面,大小分别是 4 字节和 8 字节。可以参考源码: markWord.hpp。我们这里只讨论 64 位的 JVM,也就是标记字为 8 字节的情况。
class markWord {
private:
uintptr_t _value;
}
2.类型字(Class Word): 类型字是指向对象实现的 Class 的指针。Java 7 之前指向的区域位于持久带(Permanent Generation),Java 8 之后,持久带废弃,引入了元数据区的概念(Metaspace),所以Java 8 之后指向的是这个元数据区。
这个指针可能是被压缩的,就是压缩指针(Compressed OOPs)。当开启对象压缩时占用4字节(JVM默认开启),关闭时占用8字节
1.1. 标记字的具体结构
对于 64 位的虚拟机环境,标记字大小是 8 字节。先给出标记字结构:
(上图来自于:https://www.cnblogs.com/helloworldcode/p/11914053.html)
我们先来通过 jol (Java Object Layout) 工具,以及接下来的几个实例,来逐步看一下每种状态下的标记字结构。加入依赖:
<!-- https://mvnrepository.com/artifact/org.openjdk.jol/jol-core -->
<dependency>
<groupId>org.openjdk.jol</groupId>
<artifactId>jol-core</artifactId>
<version>0.13</version>
</dependency>
1.1.1. 哈希值计算几次?去哪里找哈希值?
如果类没有覆盖hashcode()
,那么实现hashcode()
的是一个native方法。
@HotSpotIntrinsicCandidate
public native int hashCode();
具体的实现参考源码synchronizer.cpp
:
intptr_t ObjectSynchronizer::FastHashCode(Thread* self, oop obj) {
//如果启用了偏向锁特性
if (UseBiasedLocking) {
//如果正处于偏向锁
if (obj->mark().has_bias_pattern()) {
//取消偏向锁,不会再处于偏向锁状态,并且在下次获取锁的时候,直接从轻量锁开始
Handle hobj(self, obj);
if (SafepointSynchronize::is_at_safepoint()) {
BiasedLocking::revoke_at_safepoint(hobj);
} else {
BiasedLocking::revoke(hobj, self);
}
obj = hobj();
assert(!obj->mark().has_bias_pattern(), "biases should be revoked by now");
}
}
while (true) {
ObjectMonitor* monitor = NULL;
markWord temp, test;
intptr_t hash;
markWord mark = read_stable_mark(obj);
// 不能处于偏向锁状态
assert(!mark.has_bias_pattern(), "invariant");
// 未处于任何锁状态的一般状态
if (mark.is_neutral()) {
hash = mark.hash();
if (hash != 0) { // 如果已经计算过哈希值,直接返回
return hash;
}
hash = get_next_hash(self, obj); // 否则,计算一个新的哈希值,这个方法的详细分析在另一篇文章
temp = mark.copy_set_hash(hash); // 设置哈希值
test = obj->cas_set_mark(temp, mark); //更新为设置成功
if (test == mark) { // 如果更新成功,则返回
return hash;
}
//如果设置失败,可能发生了锁膨胀,或者是多线程同时调用了hashcode方法,循环重新尝试
} else if (mark.has_monitor()) {
//如果处于重量级锁状态
//获取对应的monitor对象
monitor = mark.monitor();
//获取其header
temp = monitor->header();
assert(temp.is_neutral(), "invariant: header=" INTPTR_FORMAT, temp.value());
//查看header中的hashcode
hash = temp.hash();
//若monitor中有哈希值,并且没有发生异步monitor降级(Async Monitor Deflation),就使用这个哈希值
//异步monitor降级(Async Monitor Deflation),请参考:https://wiki.openjdk.java.net/display/HotSpot/Async+Monitor+Deflation,这是在 Java 15 之后开始引入的新特性
if (hash != 0) {
// It has a hash.
// Separate load of dmw/header above from the loads in
// is_being_async_deflated().
if (support_IRIW_for_not_multiple_copy_atomic_cpu) {
// A non-multiple copy atomic (nMCA) machine needs a bigger
// hammer to separate the load above and the loads below.
OrderAccess::fence();
} else {
OrderAccess::loadload();
}
if (monitor->is_being_async_deflated()) {
// But we can't safely use the hash if we detect that async
// deflation has occurred. So we attempt to restore the
// header/dmw to the object's header so that we only retry
// once if the deflater thread happens to be slow.
monitor->install_displaced_markword_in_object(obj);
continue;
}
return hash;
}
} else if (self->is_lock_owned((address)mark.locker())) {
//如果处于轻量锁状态,读取指向锁记录的指针,获取其中的哈希值
temp = mark.displaced_mark_helper();
assert(temp.is_neutral(), "invariant: header=" INTPTR_FORMAT, temp.value());
hash = temp.hash();
if (hash != 0) { // if it has a hash, just return it
return hash;
}
}
//不处于以上状态或者处于以上锁状态之前,没有计算过哈希值,则直接膨胀锁到重量级锁
monitor = inflate(self, obj, inflate_cause_hash_code);
//查看是否已经有其他线程计算过哈希值
mark = monitor->header();
assert(mark.is_neutral(), "invariant: header=" INTPTR_FORMAT, mark.value());
hash = mark.hash();
//如果没有,则在这里计算,并设置
if (hash == 0) { // if it does not have a hash
hash = get_next_hash(self, obj); // get a new hash
temp = mark.copy_set_hash(hash); // merge the hash into header
assert(temp.is_neutral(), "invariant: header=" INTPTR_FORMAT, temp.value());
uintptr_t v = Atomic::cmpxchg((volatile uintptr_t*)monitor->header_addr(), mark.value(), temp.value());
test = markWord(v);
//设置失败,则有其他线程已经设置,尝试读取
if (test != mark) {
// The attempt to update the ObjectMonitor's header/dmw field
// did not work. This can happen if another thread managed to
// merge in the hash just before our cmpxchg().
// If we add any new usages of the header/dmw field, this code
// will need to be updated.
hash = test.hash();
assert(test.is_neutral(), "invariant: header=" INTPTR_FORMAT, test.value());
assert(hash != 0, "should only have lost the race to a thread that set a non-zero hash");
}
//如果有发生异步monitor降级(Async Monitor Deflation),跳过,重新判断
//异步monitor降级(Async Monitor Deflation),请参考:https://wiki.openjdk.java.net/display/HotSpot/Async+Monitor+Deflation,这是在 Java 15 之后开始引入的新特性
if (monitor->is_being_async_deflated()) {
// If we detect that async deflation has occurred, then we
// attempt to restore the header/dmw to the object's header
// so that we only retry once if the deflater thread happens
// to be slow.
monitor->install_displaced_markword_in_object(obj);
continue;
}
}
// We finally get the hash.
return hash;
}
}
可以看出,hashcode:
- 尽量只计算一次,计算出后,对于无锁对象,保存在对象标记字 Markword 中。
- 处于各种锁状态的话(除了偏向锁),都会修改并占用 Markword 导致需要其他地方缓存计算好的 hashcode,对于重量锁是对应的 Monitor 中保存,对于轻量锁是所指向的锁记录的指针中保存。
- 对于偏向锁,由于没法哈希值,所以只要计算过哈希值,就不会再进入偏向锁的状态,而是直接从轻量锁开始。
- 对于 JDK 15 之后引入的异步 Monitor 降级(Async Monitor Deflation),需要在这个过程完成或者未开始的时候读取 monitor对象的 hashcode 缓存。对于这个特性的详细说明,可以参考:Async Monitor Deflation
哈希值的具体计算,根据全局变量hashcode
的值决定计算方式(通过-XX:hashCode=设定)。默认的哈希值计算,是hashcode=5
的情况,不同情况的计算方式的详细说明,如果感兴趣,可以参考:JDK核心JAVA源码解析(9) - hashcode 方法
我们来通过一段代码以及输出来看看这些特性
public static void main(String[] args) throws Exception {
A a = new A();
B b = new B();
System.out.println("------After Initialization------\n" + ClassLayout.parseInstance(a).toPrintable() + "\n" + ClassLayout.parseInstance(b).toPrintable());
System.out.println("a.hashcode: " + a.hashCode());
System.out.println("b.hashcode: " + b.hashCode());
System.out.println("------After call hashcode------\n" + ClassLayout.parseInstance(a).toPrintable() + "\n" + ClassLayout.parseInstance(b).toPrintable());
}
//A没有覆盖默认的 hashcode 方法
public static class A {
long d;
}
//B覆盖了 hashcode 方法
public static class B {
long d;
@Override
public int hashCode() {
return (int) 5555;
}
}
输出:
------After Initialization------
com.hashjang.jdk.TestObjectAlign$A object internals:
OFFSET SIZE TYPE DESCRIPTION VALUE
0 4 (object header) 05 00 00 00 (00000101 00000000 00000000 00000000) (5)
4 4 (object header) 00 00 00 00 (00000000 00000000 00000000 00000000) (0)
8 4 (object header) 49 ce 00 20 (01001001 11001110 00000000 00100000) (536923721)
12 4 (alignment/padding gap)
16 8 long A.d 0
Instance size: 24 bytes
Space losses: 4 bytes internal + 0 bytes external = 4 bytes total
com.hashjang.jdk.TestObjectAlign$B object internals:
OFFSET SIZE TYPE DESCRIPTION VALUE
0 4 (object header) 05 00 00 00 (00000101 00000000 00000000 00000000) (5)
4 4 (object header) 00 00 00 00 (00000000 00000000 00000000 00000000) (0)
8 4 (object header) 8a ce 00 20 (10001010 11001110 00000000 00100000) (536923786)
12 4 (alignment/padding gap)
16 8 long B.d 0
Instance size: 24 bytes
Space losses: 4 bytes internal + 0 bytes external = 4 bytes total
a.hashcode: 2124046270
b.hashcode: 5555
------After call hashcode------
com.hashjang.jdk.TestObjectAlign$A object internals:
OFFSET SIZE TYPE DESCRIPTION VALUE
0 4 (object header) 01 be 5f 9a (00000001 10111110 01011111 10011010) (-1705001471)
4 4 (object header) 7e 00 00 00 (01111110 00000000 00000000 00000000) (126)
8 4 (object header) 49 ce 00 20 (01001001 11001110 00000000 00100000) (536923721)
12 4 (alignment/padding gap)
16 8 long A.d 0
Instance size: 24 bytes
Space losses: 4 bytes internal + 0 bytes external = 4 bytes total
com.hashjang.jdk.TestObjectAlign$B object internals:
OFFSET SIZE TYPE DESCRIPTION VALUE
0 4 (object header) 05 00 00 00 (00000101 00000000 00000000 00000000) (5)
4 4 (object header) 00 00 00 00 (00000000 00000000 00000000 00000000) (0)
8 4 (object header) 8a ce 00 20 (10001010 11001110 00000000 00100000) (536923786)
12 4 (alignment/padding gap)
16 8 long B.d 0
Instance size: 24 bytes
Space losses: 4 bytes internal + 0 bytes external = 4 bytes total
可以看出,对于未覆盖hashcode()
方法的 A 类,调用hashcode()
之后,哈希值存在了对象头。哈希值为 2124046270,转换为二进制为:1111110 10011010 01011111 10111110。由于 Windows 系统中为小端存储,所以打印出来的 Header 是反过来的,可以看到这个哈希值就存储在 Header 中:
00000001 10111110 01011111 10011010 01111110 00000000 00000000 00000000
其他特性,将在接下来的 1.1.3. 偏向锁,轻量锁,重量锁 这一小节详细说明。
1.1.2. 分代年龄
对象头中的分代年龄,用于分代 GC。分代 GC 我们在后面的章节会详细讲述,这里只是看一些特性。
记录分代年龄一共 4 bit,所以最大为 2^4 - 1 = 15。所以配置最大分代年龄-XX:MaxTenuringThreshold=n这个n不能大于15,当然也不能小于 0.等于 0 的话,就直接入老年代。等于 16 的话,就是从不进入老年代,这样不符合JVM规范,所以不能大于15(感谢 CSDN @JonsonJiao 指正)。默认是 15。
在发生 Young GC,更准确地说是在 Survivor 区复制的时候,存活的对象的分代年龄会加1。我们编写程序测试下,由于 编译器会优化代码,同时调用System.gc()
并不是立刻触发 GC,并且是 Full GC,可能会使对象直接进入老年代,分代年龄不再增长,所以我们可以使用 volatile 属性辅助我们真正创建对象,避免编译器优化:
static volatile Object consumer;
public static void main(String[] args) throws Exception {
//这是我们要观察的对象
Object instance = new Object();
long lastAddr = VM.current().addressOf(instance);
for (int i = 0; i < 10000; i++) {
//查看地址是否发生了变化,代表是否发生了 Survivor 复制,或者是移动到老年代
long currentAddr = VM.current().addressOf(instance);
if (currentAddr != lastAddr) {
//地址发生变化的时候,打印对象结构
ClassLayout layout = ClassLayout.parseInstance(instance);
System.out.println(layout.toPrintable());
lastAddr = currentAddr;
}
for (int j = 0; j < 10000; j++) {
//一直创建新对象
//因为是volatile的属性更新,不会被编译器优化
consumer = new Object();
}
}
}
可以配合 GC 日志一起观察,关于 JVM 日志配置可以参考这篇文章:OpenJDK 11 JVM日志相关参数解析与使用
首先我们用这个参数运行程序-Xmx128m -Xlog:gc=info
,输出:
[0.016s][info][gc] Using G1
# WARNING: Unable to get Instrumentation. Dynamic Attach failed. You may add this JAR as -javaagent manually, or supply -Djdk.attach.allowAttachSelf
[2.540s][info][gc] GC(0) Pause Young (Normal) (G1 Evacuation Pause) 24M->1M(128M) 2.600ms
java.lang.Object object internals:
OFFSET SIZE TYPE DESCRIPTION VALUE
0 4 (object header) 0d 00 00 00 (00001101 00000000 00000000 00000000) (13)
4 4 (object header) 00 00 00 00 (00000000 00000000 00000000 00000000) (0)
8 4 (object header) 00 02 00 20 (00000000 00000010 00000000 00100000) (536871424)
12 4 (loss due to the next object alignment)
Instance size: 16 bytes
Space losses: 0 bytes internal + 4 bytes external = 4 bytes total
[2.627s][info][gc] GC(1) Pause Young (Normal) (G1 Evacuation Pause) 75M->1M(128M) 2.273ms
java.lang.Object object internals:
OFFSET SIZE TYPE DESCRIPTION VALUE
0 4 (object header) 15 00 00 00 (00010101 00000000 00000000 00000000) (21)
4 4 (object header) 00 00 00 00 (00000000 00000000 00000000 00000000) (0)
8 4 (object header) 00 02 00 20 (00000000 00000010 00000000 00100000) (536871424)
12 4 (loss due to the next object alignment)
Instance size: 16 bytes
Space losses: 0 bytes internal + 4 bytes external = 4 bytes total
[2.675s][info][gc] GC(2) Pause Young (Normal) (G1 Evacuation Pause) 75M->1M(128M) 2.063ms
java.lang.Object object internals:
OFFSET SIZE TYPE DESCRIPTION VALUE
0 4 (object header) 1d 00 00 00 (00011101 00000000 00000000 00000000) (29)
4 4 (object header) 00 00 00 00 (00000000 00000000 00000000 00000000) (0)
8 4 (object header) 00 02 00 20 (00000000 00000010 00000000 00100000) (536871424)
12 4 (loss due to the next object alignment)
Instance size: 16 bytes
Space losses: 0 bytes internal + 4 bytes external = 4 bytes total
[2.724s][info][gc] GC(3) Pause Young (Normal) (G1 Evacuation Pause) 75M->1M(128M) 2.068ms
java.lang.Object object internals:
OFFSET SIZE TYPE DESCRIPTION VALUE
0 4 (object header) 25 00 00 00 (00100101 00000000 00000000 00000000) (37)
4 4 (object header) 00 00 00 00 (00000000 00000000 00000000 00000000) (0)
8 4 (object header) 00 02 00 20 (00000000 00000010 00000000 00100000) (536871424)
12 4 (loss due to the next object alignment)
Instance size: 16 bytes
Space losses: 0 bytes internal + 4 bytes external = 4 bytes total
[2.772s][info][gc] GC(4) Pause Young (Normal) (G1 Evacuation Pause) 75M->1M(128M) 2.212ms
java.lang.Object object internals:
OFFSET SIZE TYPE DESCRIPTION VALUE
0 4 (object header) 2d 00 00 00 (00101101 00000000 00000000 00000000) (45)
4 4 (object header) 00 00 00 00 (00000000 00000000 00000000 00000000) (0)
8 4 (object header) 00 02 00 20 (00000000 00000010 00000000 00100000) (536871424)
12 4 (loss due to the next object alignment)
Instance size: 16 bytes
Space losses: 0 bytes internal + 4 bytes external = 4 bytes total
[2.821s][info][gc] GC(5) Pause Young (Normal) (G1 Evacuation Pause) 75M->1M(128M) 2.202ms
java.lang.Object object internals:
OFFSET SIZE TYPE DESCRIPTION VALUE
0 4 (object header) 35 00 00 00 (00110101 00000000 00000000 00000000) (53)
4 4 (object header) 00 00 00 00 (00000000 00000000 00000000 00000000) (0)
8 4 (object header) 00 02 00 20 (00000000 00000010 00000000 00100000) (536871424)
12 4 (loss due to the next object alignment)
Instance size: 16 bytes
Space losses: 0 bytes internal + 4 bytes external = 4 bytes total
[2.869s][info][gc] GC(6) Pause Young (Normal) (G1 Evacuation Pause) 75M->1M(128M) 2.143ms
java.lang.Object object internals:
OFFSET SIZE TYPE DESCRIPTION VALUE
0 4 (object header) 3d 00 00 00 (00111101 00000000 00000000 00000000) (61)
4 4 (object header) 00 00 00 00 (00000000 00000000 00000000 00000000) (0)
8 4 (object header) 00 02 00 20 (00000000 00000010 00000000 00100000) (536871424)
12 4 (loss due to the next object alignment)
Instance size: 16 bytes
Space losses: 0 bytes internal + 4 bytes external = 4 bytes total
[2.917s][info][gc] GC(7) Pause Young (Normal) (G1 Evacuation Pause) 75M->1M(128M) 2.313ms
java.lang.Object object internals:
OFFSET SIZE TYPE DESCRIPTION VALUE
0 4 (object header) 45 00 00 00 (01000101 00000000 00000000 00000000) (69)
4 4 (object header) 00 00 00 00 (00000000 00000000 00000000 00000000) (0)
8 4 (object header) 00 02 00 20 (00000000 00000010 00000000 00100000) (536871424)
12 4 (loss due to the next object alignment)
Instance size: 16 bytes
Space losses: 0 bytes internal + 4 bytes external = 4 bytes total
[2.969s][info][gc] GC(8) Pause Young (Normal) (G1 Evacuation Pause) 75M->1M(128M) 2.473ms
java.lang.Object object internals:
OFFSET SIZE TYPE DESCRIPTION VALUE
0 4 (object header) 4d 00 00 00 (01001101 00000000 00000000 00000000) (77)
4 4 (object header) 00 00 00 00 (00000000 00000000 00000000 00000000) (0)
8 4 (object header) 00 02 00 20 (00000000 00000010 00000000 00100000) (536871424)
12 4 (loss due to the next object alignment)
Instance size: 16 bytes
Space losses: 0 bytes internal + 4 bytes external = 4 bytes total
[3.021s][info][gc] GC(9) Pause Young (Normal) (G1 Evacuation Pause) 75M->1M(128M) 2.283ms
java.lang.Object object internals:
OFFSET SIZE TYPE DESCRIPTION VALUE
0 4 (object header) 55 00 00 00 (01010101 00000000 00000000 00000000) (85)
4 4 (object header) 00 00 00 00 (00000000 00000000 00000000 00000000) (0)
8 4 (object header) 00 02 00 20 (00000000 00000010 00000000 00100000) (536871424)
12 4 (loss due to the next object alignment)
Instance size: 16 bytes
Space losses: 0 bytes internal + 4 bytes external = 4 bytes total
[3.072s][info][gc] GC(10) Pause Young (Normal) (G1 Evacuation Pause) 75M->1M(128M) 2.648ms
java.lang.Object object internals:
OFFSET SIZE TYPE DESCRIPTION VALUE
0 4 (object header) 5d 00 00 00 (01011101 00000000 00000000 00000000) (93)
4 4 (object header) 00 00 00 00 (00000000 00000000 00000000 00000000) (0)
8 4 (object header) 00 02 00 20 (00000000 00000010 00000000 00100000) (536871424)
12 4 (loss due to the next object alignment)
Instance size: 16 bytes
Space losses: 0 bytes internal + 4 bytes external = 4 bytes total
[3.122s][info][gc] GC(11) Pause Young (Normal) (G1 Evacuation Pause) 75M->1M(128M) 2.585ms
java.lang.Object object internals:
OFFSET SIZE TYPE DESCRIPTION VALUE
0 4 (object header) 65 00 00 00 (01100101 00000000 00000000 00000000) (101)
4 4 (object header) 00 00 00 00 (00000000 00000000 00000000 00000000) (0)
8 4 (object header) 00 02 00 20 (00000000 00000010 00000000 00100000) (536871424)
12 4 (loss due to the next object alignment)
Instance size: 16 bytes
Space losses: 0 bytes internal + 4 bytes external = 4 bytes total
[3.173s][info][gc] GC(12) Pause Young (Normal) (G1 Evacuation Pause) 75M->1M(128M) 2.130ms
java.lang.Object object internals:
OFFSET SIZE TYPE DESCRIPTION VALUE
0 4 (object header) 6d 00 00 00 (01101101 00000000 00000000 00000000) (109)
4 4 (object header) 00 00 00 00 (00000000 00000000 00000000 00000000) (0)
8 4 (object header) 00 02 00 20 (00000000 00000010 00000000 00100000) (536871424)
12 4 (loss due to the next object alignment)
Instance size: 16 bytes
Space losses: 0 bytes internal + 4 bytes external = 4 bytes total
[3.224s][info][gc] GC(13) Pause Young (Normal) (G1 Evacuation Pause) 75M->1M(128M) 2.078ms
java.lang.Object object internals:
OFFSET SIZE TYPE DESCRIPTION VALUE
0 4 (object header) 75 00 00 00 (01110101 00000000 00000000 00000000) (117)
4 4 (object header) 00 00 00 00 (00000000 00000000 00000000 00000000) (0)
8 4 (object header) 00 02 00 20 (00000000 00000010 00000000 00100000) (536871424)
12 4 (loss due to the next object alignment)
Instance size: 16 bytes
Space losses: 0 bytes internal + 4 bytes external = 4 bytes total
[3.273s][info][gc] GC(14) Pause Young (Normal) (G1 Evacuation Pause) 75M->1M(128M) 2.135ms
java.lang.Object object internals:
OFFSET SIZE TYPE DESCRIPTION VALUE
0 4 (object header) 7d 00 00 00 (01111101 00000000 00000000 00000000) (125)
4 4 (object header) 00 00 00 00 (00000000 00000000 00000000 00000000) (0)
8 4 (object header) 00 02 00 20 (00000000 00000010 00000000 00100000) (536871424)
12 4 (loss due to the next object alignment)
Instance size: 16 bytes
Space losses: 0 bytes internal + 4 bytes external = 4 bytes total
[3.322s][info][gc] GC(15) Pause Young (Normal) (G1 Evacuation Pause) 75M->1M(128M) 2.467ms
java.lang.Object object internals:
OFFSET SIZE TYPE DESCRIPTION VALUE
0 4 (object header) 7d 00 00 00 (01111101 00000000 00000000 00000000) (125)
4 4 (object header) 00 00 00 00 (00000000 00000000 00000000 00000000) (0)
8 4 (object header) 00 02 00 20 (00000000 00000010 00000000 00100000) (536871424)
12 4 (loss due to the next object alignment)
Instance size: 16 bytes
Space losses: 0 bytes internal + 4 bytes external = 4 bytes total
[3.404s][info][gc] GC(16) Pause Young (Normal) (G1 Evacuation Pause) 76M->1M(128M) 0.556ms
[3.485s][info][gc] GC(17) Pause Young (Normal) (G1 Evacuation Pause) 76M->1M(128M) 0.303ms
[3.566s][info][gc] GC(18) Pause Young (Normal) (G1 Evacuation Pause) 76M->1M(128M) 0.288ms
[3.647s][info][gc] GC(19) Pause Young (Normal) (G1 Evacuation Pause) 76M->1M(128M) 0.317ms
[3.727s][info][gc] GC(20) Pause Young (Normal) (G1 Evacuation Pause) 76M->1M(128M) 0.286ms
可以看到,在第 15 次 GC 的时候,对象进入了老年代,内存地址不再随着 Young GC 的进行而变化。
各位读者可以将-XX:MaxTenuringThreshold=
设置为 0 和 16 看看效果。
1.1.3. 偏向锁,轻量锁,重量锁
我们来编写测试代码:
A a = new A();
System.out.println("------After Initialization------\n" + ClassLayout.parseInstance(a).toPrintable());
//偏向锁
synchronized (a) {
System.out.println("------After Fetched Lock------\n" + ClassLayout.parseInstance(a).toPrintable());
}
System.out.println("------After Released Lock------\n" + ClassLayout.parseInstance(a).toPrintable());
System.out.println("a.hashcode: " + a.hashCode());
System.out.println("------After call hashcode------\n" + ClassLayout.parseInstance(a).toPrintable());
//由于调用了 hashcode,这里直接升级成为轻量锁
synchronized (a) {
System.out.println("------After Fetched Lock------\n" + ClassLayout.parseInstance(a).toPrintable());
}
System.out.println("------After Released Lock------\n" + ClassLayout.parseInstance(a).toPrintable());
//测试重量级锁
Runnable r = () -> {
synchronized (a) {
System.out.println("------After " + Thread.currentThread() + " lock is fetched------\n" + ClassLayout.parseInstance(a).toPrintable());
try {
TimeUnit.SECONDS.sleep(1);
} catch (InterruptedException e) {
e.printStackTrace();
}
}
};
Thread [] threads = new Thread[2];
for (int i = 0; i < threads.length; i++) {
threads[i] = new Thread(r);
threads[i].start();
}
for (int i = 0; i < threads.length; i++) {
threads[i].join();
}
System.out.println("------After Released Lock------\n" + ClassLayout.parseInstance(a).toPrintable());
输出为(我们这里省略掉我们不关心的输出):
------After Initialization------
OFFSET SIZE TYPE DESCRIPTION VALUE
0 4 (object header) 05 00 00 00 (00000101 00000000 00000000 00000000) (5)
4 4 (object header) 00 00 00 00 (00000000 00000000 00000000 00000000) (0)
------After Fetched Lock------
OFFSET SIZE TYPE DESCRIPTION VALUE
0 4 (object header) 05 90 8a 5e (00000101 10010000 10001010 01011110) (1586139141)
4 4 (object header) c7 02 00 00 (11000111 00000010 00000000 00000000) (711)
------After Released Lock------
OFFSET SIZE TYPE DESCRIPTION VALUE
0 4 (object header) 05 90 8a 5e (00000101 10010000 10001010 01011110) (1586139141)
4 4 (object header) c7 02 00 00 (11000111 00000010 00000000 00000000) (711)
a.hashcode: 929776179
------After call hashcode------
OFFSET SIZE TYPE DESCRIPTION VALUE
0 4 (object header) 01 33 42 6b (00000001 00110011 01000010 01101011) (1799500545)
4 4 (object header) 37 00 00 00 (00110111 00000000 00000000 00000000) (55)
------After Fetched Lock------
OFFSET SIZE TYPE DESCRIPTION VALUE
0 4 (object header) 78 f2 0f 53 (01111000 11110010 00001111 01010011) (1393554040)
4 4 (object header) ee 00 00 00 (11101110 00000000 00000000 00000000) (238)
------After Released Lock------
OFFSET SIZE TYPE DESCRIPTION VALUE
0 4 (object header) 01 33 42 6b (00000001 00110011 01000010 01101011) (1799500545)
4 4 (object header) 37 00 00 00 (00110111 00000000 00000000 00000000) (55)
------After Thread[Thread-0,5,main] lock is fetched------
OFFSET SIZE TYPE DESCRIPTION VALUE
0 4 (object header) 02 3e e0 b0 (00000010 00111110 11100000 10110000) (-1327481342)
4 4 (object header) 69 02 00 00 (01101001 00000010 00000000 00000000) (617)
------After Thread[Thread-1,5,main] lock is fetched------
OFFSET SIZE TYPE DESCRIPTION VALUE
0 4 (object header) 02 3e e0 b0 (00000010 00111110 11100000 10110000) (-1327481342)
4 4 (object header) 69 02 00 00 (01101001 00000010 00000000 00000000) (617)
------After Released Lock------
OFFSET SIZE TYPE DESCRIPTION VALUE
0 4 (object header) 02 3e e0 b0 (00000010 00111110 11100000 10110000) (-1327481342)
4 4 (object header) 69 02 00 00 (01101001 00000010 00000000 00000000) (617)
我们这里通过第一字节 8 位的末尾两位判断锁状态:
- 创建对象后,初始为无锁状态:第一字节为00000101,01 代表处于无锁状态。同时,偏向锁是开启状态,因为00000101,倒数第三位为1,这个根据前面的结构图可以知道是偏向锁标记。
- 第一次 main 线程获取锁,由于没有争抢,同时启动参数中也没有关闭偏向锁,采用偏向锁:第一字节为00000101,代表处于偏向锁状态,后面保存的是指向线程的指针。
- 第一次 main 线程释放锁,由于没有其他争抢,保持这个偏向锁状态(Hotspot 取消偏向锁需要全局 safepoint,关于这个相关的分析,可以参考:JVM相关 - SafePoint 与 Stop The World 全解(基于OpenJDK 11版本))。
- 调用 hashcode,根据之前的源码分析,需要取消偏向锁,同时将 hashcode 写入 header:第一字节为00000001,代表处于无锁状态,偏向锁关闭,倒数第三位为 0。
- 调用 hashcode 之后 main 线程获取锁,由于偏向锁关闭,直接从轻量锁开始:第一字节为01111000,00 代表轻量锁。后面保存了指向锁记录的指针。
- 调用 hashcode 之后 main 线程释放锁,释放轻量锁,锁记录会被回收,所以 hashcode 回到 header 保存。
- 多线程导致对象升级为重量级锁之后:第一字节为 00000010,10 代表重量级锁,由于 monitor 一旦生成一直存在,所以这个对象头会一直保留 monitor 的指针,hashcode 也会保存在 monitor 上。
- 最后锁释放后,header 没有改变,也是上面说的原因。
1.2. 类型字压缩指针与 JVM 最大内存
压缩指针这个属性默认是打开的,可以通过-XX:-UseCompressedOops
关闭。
首先说一下为何需要压缩指针呢?32 位的存储,可以描述多大的内存呢?假设每一个1代表1字节,那么可以描述 0~2^32-1 这 2^32 字节也就是 4 GB 的内存。
但是呢,Java 默认是 8 字节对齐的内存,也就是一个对象占用的空间,必须是 8 字节的整数倍,不足的话会填充到 8 字节的整数倍。也就是其实描述内存的时候,不用从 0 开始描述到 8(就是根本不需要定位到之间的1,2,3,4,5,6,7)因为对象起止肯定都是 8 的整数倍。所以,2^32 字节如果一个1代表8字节的话,那么最多可以描述 2^32 * 8 字节也就是 32 GB 的内存。
这就是压缩指针的原理。如果配置最大堆内存超过 32 GB(当 JVM 是 8 字节对齐),那么压缩指针会失效。 但是,这个 32 GB 是和字节对齐大小相关的,**也就是-XX:ObjectAlignmentInBytes
配置的大小(默认为8字节,也就是 Java 默认是 8 字节对齐)**。-XX:ObjectAlignmentInBytes
可以设置为 8 的整数倍,最大 128。也就是如果配置-XX:ObjectAlignmentInBytes
为 24,那么配置最大堆内存超过 96 GB 压缩指针才会失效。
编写程序测试下:
A a = new A();
System.out.println("------After Initialization------\n" + ClassLayout.parseInstance(a).toPrintable());
首先,以启动参数:-XX:ObjectAlignmentInBytes=8 -Xmx16g
执行:
------After Initialization------
com.hashjang.jdk.TestObjectAlign$A object internals:
OFFSET SIZE TYPE DESCRIPTION VALUE
0 4 (object header) 05 00 00 00 (00000101 00000000 00000000 00000000) (5)
4 4 (object header) 00 00 00 00 (00000000 00000000 00000000 00000000) (0)
8 4 (object header) 48 72 06 00 (01001000 01110010 00000110 00000000) (422472)
12 4 (alignment/padding gap)
16 8 long A.d 0
Instance size: 24 bytes
Space losses: 4 bytes internal + 0 bytes external = 4 bytes total
可以看到类型字大小为 4 字节48 72 06 00 (01001000 01110010 00000110 00000000) (422472)
,压缩指针生效。
首先,以启动参数:-XX:ObjectAlignmentInBytes=8 -Xmx32g
执行:
------After Initialization------
com.hashjang.jdk.TestObjectAlign$A object internals:
OFFSET SIZE TYPE DESCRIPTION VALUE
0 4 (object header) 05 00 00 00 (00000101 00000000 00000000 00000000) (5)
4 4 (object header) 00 00 00 00 (00000000 00000000 00000000 00000000) (0)
8 4 (object header) a0 5b c6 00 (10100000 01011011 11000110 00000000) (12999584)
12 4 (object header) b4 02 00 00 (10110100 00000010 00000000 00000000) (692)
16 8 long A.d 0
Instance size: 24 bytes
Space losses: 0 bytes internal + 0 bytes external = 0 bytes total
可以看到类型字大小为 8 字节,压缩指针失效:
a0 5b c6 00 (10100000 01011011 11000110 00000000) (12999584)
b4 02 00 00 (10110100 00000010 00000000 00000000) (692)
修改对齐大小为 16 字节,也就是以-XX:ObjectAlignmentInBytes=16 -Xmx32g
执行:
------After Initialization------
com.hashjang.jdk.TestObjectAlign$A object internals:
OFFSET SIZE TYPE DESCRIPTION VALUE
0 4 (object header) 05 00 00 00 (00000101 00000000 00000000 00000000) (5)
4 4 (object header) 00 00 00 00 (00000000 00000000 00000000 00000000) (0)
8 4 (object header) 48 72 06 00 (01001000 01110010 00000110 00000000) (422472)
12 4 (alignment/padding gap)
16 8 long A.d 0
24 8 (loss due to the next object alignment)
Instance size: 32 bytes
Space losses: 4 bytes internal + 8 bytes external = 12 bytes total
可以看到类型字大小为 4 字节48 72 06 00 (01001000 01110010 00000110 00000000) (422472)
,压缩指针生效。
总结
- 对象指针包括标记字与类型字。
- 标记字中保存了:对象默认哈希值(如果没有覆盖默认的 hashcode() 方法,则哈希值在 hashcode() 方法被调用之后,会被记录到标记字之中)、对象的形状(是否是数组)、锁状态(偏向锁等锁信息,值得一提的是偏向锁在 Java 15 中废弃:Disable and Deprecate Biased Locking)、数组长度(如果标记显示这个对象是数组,描述了数组的长度)。标记字的实现仅仅包含一个uintptr_t类型,所以,在 32 位和 64 位虚拟机上面,大小分别是 4 字节和 8 字节。可以参考源码: markWord.hpp
- 类型字中保存了:指向对象实现的 Class 的指针。类型字默认是被压缩的,压缩指针指的就是这里。
- 默认哈希值计算,对于偏向锁是否生效,是有影响的。就是默认哈希值与偏向锁不能共存。
- 默认哈希值有缓存:无锁缓存在标记字;轻量锁缓存在锁记录,标记字中有指针指向锁记录,轻量锁释放后,锁记录中的哈希值复制到标记字中;重量锁缓存在 monitor 对象,标记字中有指针指向 monitor 对象,释放后,哈希值依然缓存在 monitor 对象中。
- 默认哈希值计算,需要考虑异步 monitor 降级的情况,这是 Java 15 中的新特性:Async Monitor Deflation
- 分代年龄在每次 Young GC 复制之后 +1,最大是
-XX:MaxTenuringThreshold=n
配置的值,大于这个值就进入老年代了。 - 压缩指针是否启用和 Java 对齐字节大小(
-XX:ObjectAlignmentInBytes
,默认是 8,也就是 8 字节对齐)还有最大堆栈大小相关。