Jdk1.6 JUC源码解析(1)-atomic-AtomicXXX
作者：大飞
功能简介：
- 原子量和普通变量相比，主要体现在读写的线程安全上。对原子量的是原子的(比如多线程下的共享变量i++就不是原子的)，由CAS操作保证原子性。对原子量的读可以读到最新值，由volatile关键字来保证可见性。
- 原子量多用于数据统计(如接口调用次数)、一些序列生成(多线程环境下)以及一些同步数据结构中。
源码分析：
- 首先，原子量的一些较底层的操作都是来自sun.misc.Unsafe类，所以原子量内部有一个Unsafe的静态引用。
private static final Unsafe unsafe = Unsafe.getUnsafe();
在openJdk代码中可以找到这个类，目录openJdk的jdk/share/classes/sun/misc/。
这个类里面大多数方法都是native的，方法实现可以在openJdk的hotspot/share/vm/prims/unsafe.cpp里面找到。
- 接下来，先看下AtomicInteger的源码。
在AtomicInteger源码中，由内部的一个int域来保存值：
private volatile int value;注意到这个int域由volatile关键字修饰，可以保证可见性。
细节：volatile怎么保证可见性呢？对于被volatile修饰的域来说，对域进行的写入操作，在指令层面会在必要的时候(多核CPU)加入内存屏障(如:lock addl $0x0)，这个内存屏障的作用是令本次写操作刷回主存，同时使其他CPU的cacheline中相应数据失效。所以当其他CPU需要访问相应数据的时候，会到主存中访问，从而保证了多线程环境下相应域的可见性。
接下来看一下CAS操作，AtomicInteger中的CAS操作体现在方法compareAndSet。它的实现在unsafe.cpp里面：
/*
* Implementation of class sun.misc.Unsafe
*/
...
UNSAFE_ENTRY(jboolean, Unsafe_CompareAndSwapInt(JNIEnv *env, jobject unsafe, jobject obj, jlong offset, jint e, jint x))
UnsafeWrapper("Unsafe_CompareAndSwapInt");
oop p = JNIHandles::resolve(obj);
jint* addr = (jint *) index_oop_from_field_offset_long(p, offset);
return (jint)(Atomic::cmpxchg(x, addr, e)) == e;
UNSAFE_END 这里调用了Atomic的cmpxchg方法，继续找一下。这个方法定义在hotspot/share/vm/runtime/atomic.hpp中，实现在hotspot/share/vm/runtime/atomic.cpp中，最终实现取决于底层OS，比如linux x86，实现内联在hotspot部分代码os_cpu/linux_x86/vm/atomic_linux_x86.inline.hpp：
// Adding a lock prefix to an instruction on MP machine
#define LOCK_IF_MP(mp) "cmp $0, " #mp "; je 1f; lock; 1: "
...
inline jint Atomic::cmpxchg (jint exchange_value, volatile jint* dest, jint compare_value) {
int mp = os::is_MP();
__asm__ volatile (LOCK_IF_MP(%4) "cmpxchgl %1,(%3)"
: "=a" (exchange_value)
: "r" (exchange_value), "a" (compare_value), "r" (dest), "r" (mp)
: "cc", "memory");
return exchange_value;
} 从上面的代码中可以看到，如果是CPU是多核(multi processors)的话，会添加一个lock;前缀，这个lock;前缀也是内存屏障，它的作用是在执行后面指令的过程中锁总线(或者是锁cacheline)，保证一致性。后面的指令cmpxchgl就是x86的比较并交换指令了。
接下来有一个和compareAndSet类似的方法，weakCompareAndSet。
从注释看，这个方法会发生fail spuriously(伪失败)，而且不保证(指令)顺序，只能在一些特性场景(一些计数和统计)下替换compareAndSet。但从方法实现上看和compareAndSet没什么区别：
/**
* Atomically sets the value to the given updated value
* if the current value {@code ==} the expected value.
*
* <p>May <a href="package-summary.html#Spurious">fail spuriously</a>
* and does not provide ordering guarantees, so is only rarely an
* appropriate alternative to {@code compareAndSet}.
*
* @param expect the expected value
* @param update the new value
* @return true if successful.
*/
public final boolean weakCompareAndSet(int expect, int update) {
return unsafe.compareAndSwapInt(this, valueOffset, expect, update);
} 但还是应该按照API的说明来使用这两个方法，以防未来方法内部(实现或者底层内部机制)发生变化。
其余的大多数方法都是基于compareAndSet方法来实现的，来看其中一个，incrementAndGet方法：
/**
* Atomically increments by one the current value.
*
* @return the updated value
*/
public final int incrementAndGet() {
for (;;) {
int current = get();
int next = current + 1;
if (compareAndSet(current, next))
return next;
}
}
这个方法也体现了CAS一般是使用风格-CAS Loop，不断重试，直到成功。
当然，这也不是绝对的，JVM底层完全可以用更好的方式也替换这些方法，比如使用内联的lock;xadd就会比cmpxchg指令更好一些。
/**
* Eventually sets to the given value.
*
* @param newValue the new value
* @since 1.6
*/
public final void lazySet(int newValue) {
unsafe.putOrderedInt(this, valueOffset, newValue);
}
看下unsafe.cpp中putOrderedInt方法的实现：
{CC"putOrderedInt", CC"("OBJ"JI)V", FN_PTR(Unsafe_SetOrderedInt)},
...
// The non-intrinsified versions of setOrdered just use setVolatile
UNSAFE_ENTRY(void, Unsafe_SetOrderedInt(JNIEnv *env, jobject unsafe, jobject obj, jlong offset, jint x))
UnsafeWrapper("Unsafe_SetOrderedInt");
SET_FIELD_VOLATILE(obj, offset, jint, x);
UNSAFE_END 注：上面有句注释，说明这只是一个非内联的setOrdered方法的实现，使用了setVolatile(和setVolatile一样的效果)。
其中，SET_FIELD_VOLATILE的定义如下：
#define SET_FIELD_VOLATILE(obj, offset, type_name, x) \ oop p = JNIHandles::resolve(obj); \ OrderAccess::release_store_fence((volatile type_name*)index_oop_from_field_offset_long(p, offset), x);在hotspot/src/os_cpu/linux_x86/vm/orderAccess_linux_x86.inline.hpp找到了这个方法的内联实现。(hotspot/src/share/vm/runtime/orderAccess.hpp这个头文件里面的注释值得留意一下)：
inline void OrderAccess::release_store_fence(volatile jint* p, jint v) {
__asm__ volatile ( "xchgl (%2),%0"
: "=r" (v)
: "0" (v), "r" (p)
: "memory");
} 可见，这里是通过xchgl这个指令来实现的setOrdered。
但是，上面只是非内联的实现，我们看下内联的实现是什么样的。
在hotspot/src/share/vm/classfile/vmSymbols.hpp中有如下代码：
do_intrinsic(_putOrderedInt, sun_misc_Unsafe, putOrderedInt_name, putOrderedInt_signature, F_RN)然后找到hotspot/src/share/vm/opto/library_call.cpp中找到相应实现：
case vmIntrinsics::_putOrderedInt:
return inline_unsafe_ordered_store(T_INT);
...
bool LibraryCallKit::inline_unsafe_ordered_store(BasicType type) {
// This is another variant of inline_unsafe_access, differing in
// that it always issues store-store ("release") barrier and ensures
// store-atomicity (which only matters for "long").
if (callee()->is_static()) return false; // caller must have the capability!
...//省略不重要的部分
insert_mem_bar(Op_MemBarRelease);
insert_mem_bar(Op_MemBarCPUOrder);
// Ensure that the store is atomic for longs:
bool require_atomic_access = true;
Node* store;
if (type == T_OBJECT) // reference stores need a store barrier.
store = store_oop_to_unknown(control(), base, adr, adr_type, val, type);
else {
store = store_to_memory(control(), adr, val, type, adr_type, require_atomic_access);
}
insert_mem_bar(Op_MemBarCPUOrder);
return true;
} 我们看到这个方法里在保存动作的前后，有3个地方插入了内存屏障。
再看下hotspot/src/cpu/x86/vm/x86_64.ad：
instruct membar_release() %{
match(MemBarRelease);
ins_cost(400);
size(0);
format %{ "MEMBAR-release ! (empty encoding)" %}
ins_encode( );
ins_pipe(empty);
%}
...
instruct membar_volatile(eFlagsReg cr) %{
match(MemBarVolatile);
effect(KILL cr);
ins_cost(400);
format %{
$$template
if (os::is_MP()) {
$$emit$$"LOCK ADDL [ESP + #0], 0\t! membar_volatile"
} else {
$$emit$$"MEMBAR-volatile ! (empty encoding)"
}
%}
ins_encode %{
__ membar(Assembler::StoreLoad);
%}
ins_pipe(pipe_slow);
%} 可见，除了membar_volatile中会添加LOCK ADDL这些指令，其他的貌似没什么卵用。 所以上面那3个insert_mem_bar也相当于没有加任何内存屏障。在这种情况下，lazySet就相当于对一个普通域的写操作喽。
- 再看下AtomicBoolean的源码。
AtomicBoolean内部是用一个int域来表示布尔状态，1表示true;0表示false：
private volatile int value;
/**
* Creates a new {@code AtomicBoolean} with the given initial value.
*
* @param initialValue the initial value
*/
public AtomicBoolean(boolean initialValue) {
value = initialValue ? 1 : 0;
} CAS、lazySet方法也都分别调用unsafe的compareAndSwapInt和putOrderedInt，上面分析过了。
- 继续看下AtomicLong的源码。
AtomicLong内部是用一个long域来保存值：
/**
* Records whether the underlying JVM supports lockless
* compareAndSwap for longs. While the Unsafe.compareAndSwapLong
* method works in either case, some constructions should be
* handled at Java level to avoid locking user-visible locks.
*/
static final boolean VM_SUPPORTS_LONG_CAS = VMSupportsCS8();
/**
* Returns whether underlying JVM supports lockless CompareAndSet
* for longs. Called only once and cached in VM_SUPPORTS_LONG_CAS.
*/
private static native boolean VMSupportsCS8();
static {
try {
valueOffset = unsafe.objectFieldOffset
(AtomicLong.class.getDeclaredField("value"));
} catch (Exception ex) { throw new Error(ex); }
}
private volatile long value; 这里注意到，AtomicLong中还提供了一个包内可见的静态域VM_SUPPORTS_LONG_CAS来表示底层是否支持Long类型(8字节)的lockless CAS操作。
AtomicLong内部整体结构和AtomicInteger类似，主要来看下内部使用的unsafe的方法有什么不同，首先CAS操作使用了unsafe的compareAndSwapLong方法：
/**
* Atomically sets the value to the given updated value
* if the current value {@code ==} the expected value.
*
* @param expect the expected value
* @param update the new value
* @return true if successful. False return indicates that
* the actual value was not equal to the expected value.
*/
public final boolean compareAndSet(long expect, long update) {
return unsafe.compareAndSwapLong(this, valueOffset, expect, update);
} 在unsafe.cpp中找到实现：
{CC"compareAndSwapLong", CC"("OBJ"J""J""J"")Z", FN_PTR(Unsafe_CompareAndSwapLong)},
...
UNSAFE_ENTRY(jboolean, Unsafe_CompareAndSwapLong(JNIEnv *env, jobject unsafe, jobject obj, jlong offset, jlong e, jlong x))
UnsafeWrapper("Unsafe_CompareAndSwapLong");
Handle p (THREAD, JNIHandles::resolve(obj));
jlong* addr = (jlong*)(index_oop_from_field_offset_long(p(), offset));
if (VM_Version::supports_cx8())
return (jlong)(Atomic::cmpxchg(x, addr, e)) == e;
else {
jboolean success = false;
ObjectLocker ol(p, THREAD);
if (*addr == e) { *addr = x; success = true; }
return success;
}
UNSAFE_END 从实现中可以看到，如果平台不支持8字节的CAS操作，就会加锁然后进行设置操作；如果支持，就会调用Atomic::cmpxchg方法，方法实现可以参考具体平台内联代码hotspot/os_cpu/linux_x86/vm/atomic_linux_x86.inline.hpp：
// Adding a lock prefix to an instruction on MP machine
#define LOCK_IF_MP(mp) "cmp $0, " #mp "; je 1f; lock; 1: "
inline jlong Atomic::cmpxchg (jlong exchange_value, volatile jlong* dest, jlong compare_value) {
bool mp = os::is_MP();
__asm__ __volatile__ (LOCK_IF_MP(%4) "cmpxchgq %1,(%3)"
: "=a" (exchange_value)
: "r" (exchange_value), "a" (compare_value), "r" (dest), "r" (mp)
: "cc", "memory");
return exchange_value;
} 其实就是(多核情况下带lock前缀的)cmpxchgq指令。
然后看一下lazySet中使用到的unsafe的putOrderedLong方法。
/**
* Eventually sets to the given value.
*
* @param newValue the new value
* @since 1.6
*/
public final void lazySet(long newValue) {
unsafe.putOrderedLong(this, valueOffset, newValue);
} 同样在unsafe.cpp中可以找到该方法的实现：
{CC"putOrderedLong", CC"("OBJ"JJ)V", FN_PTR(Unsafe_SetOrderedLong)},
...
UNSAFE_ENTRY(void, Unsafe_SetOrderedLong(JNIEnv *env, jobject unsafe, jobject obj, jlong offset, jlong x))
UnsafeWrapper("Unsafe_SetOrderedLong");
#if defined(SPARC) || defined(X86)
// Sparc and X86 have atomic jlong (8 bytes) instructions
SET_FIELD_VOLATILE(obj, offset, jlong, x);
#else
// Keep old code for platforms which may not have atomic long (8 bytes) instructions
{
if (VM_Version::supports_cx8()) {
SET_FIELD_VOLATILE(obj, offset, jlong, x);
}
else {
Handle p (THREAD, JNIHandles::resolve(obj));
jlong* addr = (jlong*)(index_oop_from_field_offset_long(p(), offset));
ObjectLocker ol(p, THREAD);
*addr = x;
}
}
#endif
UNSAFE_END 从实现上看，如果平台是SPARC或者X86或者平台支持8字节CAS，就相当于执行了一个volatile write；否则，加锁写。
当然还要看一下内联方法，在hotspot/src/share/vm/classfile/vmSymbols.hpp中有如下代码：
do_intrinsic(_putOrderedLong, sun_misc_Unsafe, putOrderedLong_name, putOrderedLong_signature, F_RN) \然后找到hotspot/src/share/vm/opto/library_call.cpp中找到相应实现。
case vmIntrinsics::_putOrderedLong:
return inline_unsafe_ordered_store(T_LONG);
...
bool LibraryCallKit::inline_unsafe_ordered_store(BasicType type) {
// This is another variant of inline_unsafe_access, differing in
// that it always issues store-store ("release") barrier and ensures
// store-atomicity (which only matters for "long").
if (callee()->is_static()) return false; // caller must have the capability!
...//忽略不重要部分
// Ensure that the store is atomic for longs:
bool require_atomic_access = true;
Node* store;
if (type == T_OBJECT) // reference stores need a store barrier.
store = store_oop_to_unknown(control(), base, adr, adr_type, val, type);
else {
store = store_to_memory(control(), adr, val, type, adr_type, require_atomic_access);
}
insert_mem_bar(Op_MemBarCPUOrder);
return true;
} 从代码的注释上可以发现，require_atomic_access设置为true，为了保证long写操作的原子性。继续跟代码，找到hotspot/src/share/vm/opto/graphKit.cpp：
Node* GraphKit::store_to_memory(Node* ctl, Node* adr, Node *val, BasicType bt,
int adr_idx,
bool require_atomic_access) {
assert(adr_idx != Compile::AliasIdxTop, "use other store_to_memory factory" );
const TypePtr* adr_type = NULL;
debug_only(adr_type = C->get_adr_type(adr_idx));
Node *mem = memory(adr_idx);
Node* st;
if (require_atomic_access && bt == T_LONG) {
st = StoreLNode::make_atomic(C, ctl, mem, adr, adr_type, val);
} else {
st = StoreNode::make(_gvn, ctl, mem, adr, adr_type, val, bt);
}
st = _gvn.transform(st);
set_memory(st, adr_idx);
// Back-to-back stores can only remove intermediate store with DU info
// so push on worklist for optimizer.
if (mem->req() > MemNode::Address && adr == mem->in(MemNode::Address))
record_for_igvn(st);
return st;
} 可见，这里会针对long做原子写操作(这里的原子操作应该指的是将long的高4字节和低4字节的操作合并成一个原子操作，比如某些平台不支持非volatile的long/double域的原子操作)。
- 最后看下AtomicReference的源码。
AtomicReference内部构成和其他原子量基本一致，区别只是这个类内部保存一个对象引用。
public class AtomicReference<V> implements java.io.Serializable {
private static final long serialVersionUID = -1848883965231344442L;
private static final Unsafe unsafe = Unsafe.getUnsafe();
private static final long valueOffset;
static {
try {
valueOffset = unsafe.objectFieldOffset
(AtomicReference.class.getDeclaredField("value"));
} catch (Exception ex) { throw new Error(ex); }
}
private volatile V value; 重点看一下内部调用的unsafe的compareAndSwapObject和putOrderedObject方法，先看一下compareAndSwapObject：
/**
* Atomically sets the value to the given updated value
* if the current value {@code ==} the expected value.
* @param expect the expected value
* @param update the new value
* @return true if successful. False return indicates that
* the actual value was not equal to the expected value.
*/
public final boolean compareAndSet(V expect, V update) {
return unsafe.compareAndSwapObject(this, valueOffset, expect, update);
} 在unsafe.cpp中可以找到实现：
{CC"compareAndSwapObject", CC"("OBJ"J"OBJ""OBJ")Z", FN_PTR(Unsafe_CompareAndSwapObject)},
...
UNSAFE_ENTRY(jboolean, Unsafe_CompareAndSwapObject(JNIEnv *env, jobject unsafe, jobject obj, jlong offset, jobject e_h, jobject x_h))
UnsafeWrapper("Unsafe_CompareAndSwapObject");
oop x = JNIHandles::resolve(x_h);
oop e = JNIHandles::resolve(e_h);
oop p = JNIHandles::resolve(obj);
HeapWord* addr = (HeapWord *)index_oop_from_field_offset_long(p, offset);
if (UseCompressedOops) {
update_barrier_set_pre((narrowOop*)addr, e);
} else {
update_barrier_set_pre((oop*)addr, e);
}
oop res = oopDesc::atomic_compare_exchange_oop(x, addr, e);
jboolean success = (res == e);
if (success)
update_barrier_set((void*)addr, x);
return success;
UNSAFE_END 可以看到里面实际上是执行oopDesc::atomic_compare_exchange_oop这个方法。找到hotspot/src/share/vm/oops/oop.inline.hpp中该方法实现：
inline oop oopDesc::atomic_compare_exchange_oop(oop exchange_value,
volatile HeapWord *dest,
oop compare_value) {
if (UseCompressedOops) {
// encode exchange and compare value from oop to T
narrowOop val = encode_heap_oop(exchange_value);
narrowOop cmp = encode_heap_oop(compare_value);
narrowOop old = (narrowOop) Atomic::cmpxchg(val, (narrowOop*)dest, cmp);
// decode old from T to oop
return decode_heap_oop(old);
} else {
return (oop)Atomic::cmpxchg_ptr(exchange_value, (oop*)dest, compare_value);
}
} cmpxchg之前分析过，看看cmpxche_ptr，找到hotspot/os_cpu/linux_x86/vm/atomic_linux_x86.inline.hpp中实现：
inline intptr_t Atomic::cmpxchg_ptr(intptr_t exchange_value, volatile intptr_t* dest, intptr_t compare_value) {
return (intptr_t)cmpxchg((jlong)exchange_value, (volatile jlong*)dest, (jlong)compare_value);
}
inline jlong Atomic::cmpxchg (jlong exchange_value, volatile jlong* dest, jlong compare_value) {
bool mp = os::is_MP();
__asm__ __volatile__ (LOCK_IF_MP(%4) "cmpxchgq %1,(%3)"
: "=a" (exchange_value)
: "r" (exchange_value), "a" (compare_value), "r" (dest), "r" (mp)
: "cc", "memory");
return exchange_value;
} 就是(多核下带lock前缀的)cmpxchgq命令了。
putOrderedObject方法按之前几篇的查找方法，会发现内联之后，相当于一个普通写操作了。
OK，源码分析到此结束！
参考资料：
http://gee.cs.oswego.edu/dl/jmm/cookbook.html
http://x86.renejeschke.de/html/file_module_x86_id_41.html
http://stackoverflow.com/questions/2443239/how-can-weakcompareandset-fail-spuriously-if-it-is-implemented-exactly-like-comp
http://faydoc.tripod.com/cpu/lock.htm
http://ifeve.com/juc-atomic-class-lazyset-que/
http://bugs.java.com/bugdatabase/view_bug.do?bug_id=6275329
http://stackoverflow.com/questions/4232660/which-is-a-better-write-barrier-on-x86-lockaddl-or-xchgl
http://bugs.java.com/bugdatabase/view_bug.do?bug_id=7023898