上一篇分析了 HotSpot类模型之InstanceKlass ,这次主要分析表示java数组类型的C++类。
1、ArrayKlass类
ArrayKlass继承自Klass,是所有数组类的抽象基类,类及重要属性的定义如下:
class ArrayKlass: public Klass { ... private: int _dimension; // This is n'th-dimensional array. Klass* volatile _higher_dimension; // Refers the (n+1)'th-dimensional array (if present). Klass* volatile _lower_dimension; // Refers the (n-1)'th-dimensional array (if present). int _vtable_len; // size of vtable for this klass oop _component_mirror; // component type, as a java/lang/Class ... }
在Klass的基础上增加的属性如下表所示。
字段 | 作用 |
_dimension | int类型,表示数组的维度,记为n |
_higher_dimension | Klass指针,表示对n+1维数组Klass的引用 |
_lower_dimension | Klass指针,表示对n-1维数组Klass的引用 |
_vtable_len | int类型, 虚函数表的长度 |
_component_mirror | oop, 数组元素对应的java.lang.Class对象的Oop |
_vtable_len的值为5,因为数组是引用类型,父类为Object类,而Object类中有5个虚方法可被用来继承和重写,如下:
void finalize() boolean equals(Object) String toString() int hashCode() Object clone()
_dimension、_higher_dimension与_lower_dimension对于一维及多维数组的描述非常重要,属性值的设置相对简单,这里不在介绍。
2、ArrayKlass类的子类
(1)TypeArrayKlass类
TypeArrayKlass是ArrayKlass的子类,用于表示数组元素是基本类型的数组
class TypeArrayKlass : public ArrayKlass { ... private: jint _max_length; // maximum number of elements allowed in an array ... }
_max_length表示该数组允许的最大长度。
数组类和普通类不同,数组类没有对应的Class文件,所以数组类是直接被虚拟机创建的。HotSpot在初始化时就会创建好8个基本类型的一维数组对象TypeArrayKlass。之前在讲解HotSpot启动时讲到过,调用initializeJVM()方法初始化HotSpot,这个方法会最终调用到Universe::genesis()方法,在这个方法中初始化基本类型的一维数组对象TypeArrayKlass。例如初始化boolean类型的一维数组,调用语句如下:
_boolArrayKlassObj = TypeArrayKlass::create_klass(T_BOOLEAN, sizeof(jboolean), CHECK);
其中_boolArrayKlassObj是声明在universe.cpp文件中的全局变量,如下:
Klass* Universe::_boolArrayKlassObj = NULL;
调用TypeArrayKlass::create_klass()方法创建TypeArrayKlass对象,具体就是调用TypeArrayKlass::create_klass()方法来完成,方法的实现如下:
TypeArrayKlass* TypeArrayKlass::allocate(ClassLoaderData* loader_data, BasicType type, Symbol* name, TRAPS) { int x = TypeArrayKlass::header_size(); int size = ArrayKlass::static_size(x); // 调用的构造函数在下面 return new (loader_data, size, THREAD) TypeArrayKlass(type, name); }
非常类似于InstanceKlass等对象的创建,首先获取需要内存的大小size,然后通过重载new运算符完成对象内存分配后,调用TypeArrayKlass的构造函数初始化一些属性。
TypeArrayKlass的header_size()及static_size()函数的实现如下:
static int header_size(){ int k = sizeof(TypeArrayKlass); return k/HeapWordSize; } int ArrayKlass::static_size(int header_size) { // size of an array klass object assert(header_size <= InstanceKlass::header_size(), "bad header size"); // If this assert fails, see comments in base_create_array_klass. header_size = InstanceKlass::header_size(); // 为什么是InstanceKlass的大小??看ArrayKlass::start_of_vtable()函数有说明 int vtable_len = Universe::base_vtable_size(); // 值为5 int size = header_size + align_object_offset(vtable_len); // 对vtable_len进行对齐操作 return align_object_size(size); } static int header_size(){ return align_object_offset(sizeof(InstanceKlass)/HeapWordSize); }
注意header_size属性的值应该是TypeArrayKlass这个类自身占用的内存大小,但是现在却取的是InstanceKlass这个类自身占用内存的大小。这是因为InstanceKlass占用内存大小比TypeArrayKlass大,有足够内存存放相关数据,更重要的是为了统一从固定的偏移位置取出vtable_len属性的值。这样在实际操作过程中,无需关心是数组还是类,都直接偏移固定位置后取vtable_len属性值即可。
TypeArrayKlass的构造函数如下:
TypeArrayKlass::TypeArrayKlass(BasicType type, Symbol* name) : ArrayKlass(name) { int lh = array_layout_helper(type); set_layout_helper(lh); assert(oop_is_array(), "sanity"); assert(oop_is_typeArray(), "sanity"); set_max_length(arrayOopDesc::max_array_length(type)); // 设置数组的最大长度 ... }
下面详细介绍一下对_layout_helper属性的设置。调用Klass::array_layout_helper()方法获取_layout_helper属性的值
jint Klass::array_layout_helper(BasicType etype) { assert(etype >= T_BOOLEAN && etype <= T_OBJECT, "valid etype"); // Note that T_ARRAY is not allowed here. int hsize = arrayOopDesc::base_offset_in_bytes(etype); // hsize表示数组元素的对象头部大小 int esize = type2aelembytes(etype); // 对应类型存储所需要的字节数 bool isobj = (etype == T_OBJECT); int tag = isobj ? _lh_array_tag_obj_value : _lh_array_tag_type_value; int esz = exact_log2(esize); int lh = array_layout_helper(tag, hsize, etype, esz); return lh; }
关于_layout_helper在之前已经介绍过,由于T_BOOLEAN为基本类型,所以tag的值取0xC0;hsize调用arrayOopDesc::base_offset_in_bytes()方法获取,值为16,后面在讲解arrayOopDesc时会介绍,数组对象其实是由对象头、对象字段数据和对齐填充组成,而这里获取的就是对象头的大小;esize表示对应类型存储所需要的字节数,对于T_BOOLEAN来说,只需要1个字节即可,所以esz为0。最后调用array_layout_helper()方法按照约定组合成一个int类型的数字即可。array_layout_helper()方法的实现如下:
static jint array_layout_helper(jint tag, int hsize, BasicType etype, int log2_esize) { return (tag << _lh_array_tag_shift) // 左移30位 | (hsize << _lh_header_size_shift) // 左移16位 | ((int)etype << _lh_element_type_shift) // 左移1位 | (log2_esize << _lh_log2_element_size_shift); // 左移0位 }
另外还有对_component_mirror属性的设置。对于一维基本类型的数组来说,这个值是java.lang.Class对象。Class对象使用oop对象来表示,调用java_lang_Class::create_basic_type_mirror()方法获取_component_mirror属性的值,通过java_lang_Class::create_mirror()方法完成属性的设置。例如获取boolean类型的属性值,调用语句如下:
void Universe::initialize_basic_type_mirrors(TRAPS) { ... _bool_mirror = java_lang_Class::create_basic_type_mirror("boolean",T_BOOLEAN, CHECK); ... }
方法create_basic_type_mirror()的实现如下:
oop java_lang_Class::create_basic_type_mirror(const char* basic_type_name, BasicType type, TRAPS) { // This should be improved by adding a field at the Java level or by // introducing a new VM klass (see comment in ClassFileParser) oop java_class = InstanceMirrorKlass::cast(SystemDictionary::Class_klass())->allocate_instance(NULL, CHECK_0); if (type != T_VOID) { Klass* aklass = Universe::typeArrayKlassObj(type); assert(aklass != NULL, "correct bootstrap"); set_array_klass(java_class, aklass); // 设置表示基本类型数组的TypeArrayKlass的 } return java_class; }
通过InstanceMirrorKlass对象(表示java.lang.Class类)来创建oop(表示java.lang.Class对象),_component_mirror最终设置的就是这个oop。引用类型组成的一维或多维数组的基本元素可以使用Klass对象来表示,如对于下面即将要介绍的Object[]来说,元素类型为Object,所以可以使用InstanceKlass来表示;基本类型组成的一维或多维数组的基本元素没有对应的Klass对象,所以只能使用Class对象来描述boolean、int等类型,这样就会与表示Class对象的oop对象产生关系,相关属性最终的值如下所示。
TypeArrayKlass._component_mirror=oop oop._array_klass_offset=TypeArrayKlass
oop表示java.lang.Class对象,用来描述Java类(包括数组类),而TypeArrayKlass也用来描述Java类(包括数组类),那么2者之间必须会的联系。可以通过_component_mirror属性(和_array_klass_offset属性找到对方,属性的设置过程在在java_lang_Class::create_mirror()函数中进行。
其它的属性设置很简单,这里不在介绍。
(2)ObjArrayKlass类
ObjArrayKlass是ArrayKlass的子类,用于表示数组元素是类或者数组
class ObjArrayKlass : public ArrayKlass { ... private: Klass* _element_klass; // The klass of the elements of this array type Klass* _bottom_klass; // The one-dimensional type (InstanceKlass or TypeArrayKlass) ... }
该类新增了2个属性,如下:
- _element_klass:数组元素对应的Klass对象,如果是多维数组,对应数组元素的ObjArrayKlass对象
- _bottom_klass:一维数组的类型,可以是InstanceKlass或者TypeArrayKlass。一维基本类型数组为TypeArrayKlass,而二维基本类型数组就会使用ObjArrayKlass来表示,所以其_bottom_klass会是TypeArrayKlass。
HotSpot在Universe::genesis()方法中创建Object数组,如下:
InstanceKlass* ik = InstanceKlass::cast(SystemDictionary::Object_klass()); _objectArrayKlassObj = ik->array_klass(1, CHECK); // 调用表示Object类的InstanceKlass类的array_klass()方法
调用array_klass()方法时传递的参数1表示创建一维数组。调用表示Object类的InstanceKlass对象的方法创建的,所以Object数组的创建要依赖于InstanceKlass对象(表示Object类)进行创建。
传递的参数1表示创建Object的一维数组类型,array_klass()函数及调用的相关函数的实现如下:
// array class with specific rank Klass* array_klass(int rank, TRAPS) { return array_klass_impl(false, rank, THREAD); } Klass* InstanceKlass::array_klass_impl(bool or_null, int n, TRAPS) { instanceKlassHandle this_oop(THREAD, this); return array_klass_impl(this_oop, or_null, n, THREAD); } Klass* InstanceKlass::array_klass_impl(instanceKlassHandle this_oop, bool or_null, int n, TRAPS) { if (this_oop->array_klasses() == NULL) { // 获取_array_klasses属性的值 if (or_null){ return NULL; } ResourceMark rm; JavaThread *jt = (JavaThread *)THREAD; { // Atomic creation of array_klasses MutexLocker mc(Compile_lock, THREAD); // for vtables MutexLocker ma(MultiArray_lock, THREAD); // Check if update has already taken place if (this_oop->array_klasses() == NULL) { ClassLoaderData* CLD = this_oop->class_loader_data(); Klass* k = ObjArrayKlass::allocate_objArray_klass(CLD, 1, this_oop, CHECK_NULL); this_oop->set_array_klasses(k); // 设置InstanceKlass::_array_klasses属性的值 } } } // _this will always be set at this point ObjArrayKlass* oak = (ObjArrayKlass*)this_oop->array_klasses(); // 获取InstanceKlass::_array_klasses属性的值 if (or_null) { return oak->array_klass_or_null(n); } return oak->array_klass(n, CHECK_NULL); // 在创建出一维的引用类型数组后,接着创建n维的引用类型数组 }
首次创建ObjTypeKlass时,InstanceKlass::_array_klasses属性的值为NULL,这样就会调用objArrayKlass::allocate_objArray_klass()函数,创建出一维的引用类型数组并保存到了InstanceKlass::_array_klasses属性中。有了一维的引用类型数组后就可以接着调用array_klass()方法创建n维的引用类型数组了。
(1)创建一维引用类型数组ObjArrayKlass::allocate_objArray_klass()
Klass* ObjArrayKlass::allocate_objArray_klass( ClassLoaderData* loader_data, int n, KlassHandle element_klass, TRAPS ) { // Eagerly allocate the direct array supertype. KlassHandle super_klass = KlassHandle(); if (!Universe::is_bootstrapping() || SystemDictionary::Object_klass_loaded()) { KlassHandle element_super (THREAD, element_klass->super()); if (element_super.not_null()) { // element_super是Object,Object的父类是null // The element type has a direct super. E.g., String[] has direct super of Object[]. super_klass = KlassHandle(THREAD, element_super->array_klass_or_null()); bool supers_exist = super_klass.not_null(); // Also, see if the element has secondary supertypes. // We need an array type for each. Array<Klass*>* element_supers = element_klass->secondary_supers(); for( int i = element_supers->length()-1; i >= 0; i-- ) { Klass* elem_super = element_supers->at(i); if (elem_super->array_klass_or_null() == NULL) { supers_exist = false; break; } } if (!supers_exist) { // Oops. Not allocated yet. Back out, allocate it, and retry. KlassHandle ek; { MutexUnlocker mu(MultiArray_lock); MutexUnlocker mc(Compile_lock); // for vtables Klass* sk = element_super->array_klass(CHECK_0); super_klass = KlassHandle(THREAD, sk); for( int i = element_supers->length()-1; i >= 0; i-- ) { KlassHandle elem_super(THREAD, element_supers->at(i)); elem_super->array_klass(CHECK_0); } // Now retry from the beginning Klass* klass_oop = element_klass->array_klass(n, CHECK_0); // Create a handle because the enclosing brace, when locking // can cause a gc. Better to have this function return a Handle. ek = KlassHandle(THREAD, klass_oop); } // re-lock return ek(); } } else { // element_super不是Object // The element type is already Object. Object[] has direct super of Object. super_klass = KlassHandle(THREAD, SystemDictionary::Object_klass()); } } // Create type name for klass. Symbol* name = NULL; if ( !element_klass->oop_is_instance() || (name = InstanceKlass::cast(element_klass())->array_name()) == NULL ){ ResourceMark rm(THREAD); char *name_str = element_klass->name()->as_C_string(); int len = element_klass->name()->utf8_length(); char *new_str = NEW_RESOURCE_ARRAY(char, len + 4); int idx = 0; new_str[idx++] = '['; if (element_klass->oop_is_instance()) { // it could be an array or simple type new_str[idx++] = 'L'; } memcpy(&new_str[idx], name_str, len * sizeof(char)); idx += len; if (element_klass->oop_is_instance()) { new_str[idx++] = ';'; } new_str[idx++] = ' '; name = SymbolTable::new_permanent_symbol(new_str, CHECK_0); if (element_klass->oop_is_instance()) { InstanceKlass* ik = InstanceKlass::cast(element_klass()); ik->set_array_name(name);// 设置InstanceKlass::_array_name的属性 } } // Initialize instance variables ObjArrayKlass* oak = ObjArrayKlass::allocate(loader_data, n, element_klass, name, CHECK_0); // Add all classes to our internal class loader list here, // including classes in the bootstrap (NULL) class loader. // GC walks these as strong roots. loader_data->add_class(oak); // Call complete_create_array_klass after all instance variables has been initialized. ArrayKlass::complete_create_array_klass(oak, super_klass, CHECK_0); return oak; }
调用的 ObjArrayKlass::allocate()函数的实现如下:
ObjArrayKlass* ObjArrayKlass::allocate(ClassLoaderData* loader_data, int n, KlassHandle klass_handle, Symbol* name, TRAPS) { assert(ObjArrayKlass::header_size() <= InstanceKlass::header_size(),"array klasses must be same size as InstanceKlass"); int x = ObjArrayKlass::header_size(); int size = ArrayKlass::static_size(x); return new (loader_data, size, THREAD) ObjArrayKlass(n, klass_handle, name); } int ArrayKlass::static_size(int header_size) { // size of an array klass object assert(header_size <= InstanceKlass::header_size(), "bad header size"); // If this assert fails, see comments in base_create_array_klass. header_size = InstanceKlass::header_size(); // 为什么是InstanceKlass的大小??看ArrayKlass::start_of_vtable()函数有说明 int vtable_len = Universe::base_vtable_size(); // 值为5 int size = header_size + align_object_offset(vtable_len); // 对vtable_len进行对齐操作 return align_object_size(size); }
ArrayKlass::complete_create_array_klass()函数的实现如下:
// Initialization of vtables and mirror object is done separatly from base_create_array_klass, // since a GC can happen. At this point all instance variables of the ArrayKlass must be setup. void ArrayKlass::complete_create_array_klass(ArrayKlass* k, KlassHandle super_klass, TRAPS) { ResourceMark rm(THREAD); Klass* curr_superklass = super_klass(); // super_klass是个参数,类型为KlassHandle k->initialize_supers(curr_superklass, CHECK); klassVtable* kv = k->vtable(); kv->initialize_vtable(false, CHECK); // 会初始化当前Klass的vtable(含有_length个vtableEntry) java_lang_Class::create_mirror(k, Handle(NULL), CHECK); }
调用initialize_vtalbe()完成虚函数表的初始化,调用java_lang_Class::create_mirror()函数完成当前ObjTypeArray对象对应的java.lang.Class对象的创建并设置了相关属性。
(2)创建n维引用类型数组ObjArrayKlass::array_klass()
Klass* ObjArrayKlass::array_klass_impl(bool or_null, int n, TRAPS) { assert(dimension() <= n, "check order of chain"); int dim = dimension(); if (dim == n) return this; if (higher_dimension() == NULL) { if (or_null) return NULL; ResourceMark rm; JavaThread *jt = (JavaThread *)THREAD; { MutexLocker mc(Compile_lock, THREAD); // for vtables // Ensure atomic creation of higher dimensions MutexLocker mu(MultiArray_lock, THREAD); // Check if another thread beat us if (higher_dimension() == NULL) { // Create multi-dim klass object and link them together Klass* k = ObjArrayKlass::allocate_objArray_klass(class_loader_data(), dim + 1, this, CHECK_NULL); ObjArrayKlass* ak = ObjArrayKlass::cast(k); ak->set_lower_dimension(this); OrderAccess::storestore(); set_higher_dimension(ak); assert(ak->oop_is_objArray(), "incorrect initialization of ObjArrayKlass"); } } } else { CHECK_UNHANDLED_OOPS_ONLY(Thread::current()->clear_unhandled_oops()); } ObjArrayKlass *ak = ObjArrayKlass::cast(higher_dimension()); if (or_null) { return ak->array_klass_or_null(n); } return ak->array_klass(n, CHECK_NULL); } Klass* ObjArrayKlass::array_klass_impl(bool or_null, TRAPS) { return array_klass_impl(or_null, dimension() + 1, CHECK_NULL); }
最终表示Object类的InstanceKlass与表示一维数组Object[]的ObjArrayKlass之间的相关属性如下:
ObjArrayKlass._element_klass=InstanceKlass ObjArrayKlass._bottom_klass=InstanceKlass InstanceKlass._array_name="[Ljava/lang/Object;" InstanceKlass._array_klasses=ObjArrayKlass
ObjArrayKlass中其它的属性设置也并不复杂,这里不在介绍。
其它参考文章:
1、在Ubuntu 16.04上编译OpenJDK8的源代码(配视频)
搭建过程中如果有问题可直接评论留言或加作者微信mazhimazh。
作者持续维护的个人博客 classloading.com。
B站上有HotSpot源码分析相关视频 https://space.bilibili.com/27533329
关注公众号,有HotSpot源码剖析系列文章!