一、cache_t 内部结构分析
1.1 在iOS类的结构分析中,我们已经分析过类(Class
)的本质是一个结构体 ,结构体内部结构如下 :
typedef struct objc_class *Class;
typedef struct objc_object *id;
struct objc_class : objc_object {
// Class ISA;
Class superclass;
cache_t cache; // formerly cache pointer and vtable
class_data_bits_t bits; // class_rw_t * plus custom rr/alloc flags
class_rw_t *data() const {
return bits.data();
}
...
}
-
Class ISA
:指向关联类 , 继承自objc_object
。 参考 isa底层结构分析 -
Class superclass
:父类指针 , 同样参考上述文章中有详细指向探索。 -
cache_t cache
, 方法缓存存储数据结构。 -
class_data_bits_t bits
,bit
中存储了属性,方法等类的源数据。
1.2 在iOS类的结构分析中,我们已经分析过 cache_t
结构体,分为以下四个部分:
struct cache_t {
struct bucket_t * _buckets; // 缓存数组,即哈希桶
mask_t _mask; // 缓存数组的容量临界值,实际上是为了 capacity 服务
uint16_t _flags; // 位置标记
uint16_t _occupied; // 缓存数组中已缓存方法数量
...省略
}
- _buckets:是
bucket_t
结构体的数组,bucket_t
是用来存放方法编号SEL
和函数指针IMP
的。
struct bucket_t {
explicit_atomic<uintptr_t> _imp;
explicit_atomic<SEL> _sel;
}
- _mask:
mask_t m = capacity - 1
;(capacity = MAX_CACHE_SIZE;)
,用作掩码。因为这里缓存Cache
的容量Size
一直是2倍扩容的,所以MAX_CACHE_SIZE
是2的整数次幂,所以mask
的二进制位000011, 000111, 001111
)刚好可以用作Hash
取余数的掩码。刚好保证相与后不超过缓存大小。
capacity = capacity ? capacity * 2 : INIT_CACHE_SIZE; // 扩容至两倍
- _flags: 位置标记
- _occupied是当前已缓存的方法数量。即数组中已使用了多少位置。
二、方法缓存原理探索
源码如下:
@interface LGPerson : NSObject
- (void)sayHello;
- (void)sayCode;
- (void)sayMaster;
- (void)sayNB;
+ (void)sayHappy;
@end
#import "LGPerson.h"
@implementation LGPerson
- (void)sayHello{
NSLog(@"LGPerson say : %s",__func__);
}
- (void)sayCode{
NSLog(@"LGPerson say : %s",__func__);
}
- (void)sayMaster{
NSLog(@"LGPerson say : %s",__func__);
}
- (void)sayNB{
NSLog(@"LGPerson say : %s",__func__);
}
+ (void)sayHappy{
NSLog(@"LGPerson say : %s",__func__);
}
@end
#import <Foundation/Foundation.h>
#import "LGPerson.h"
#import <objc/runtime.h>
// cache_t
int main(int argc, const char * argv[]) {
@autoreleasepool {
// insert code here...
LGPerson *p = [LGPerson alloc];
Class pClass = [LGPerson class];
[p sayHello];
[p sayCode];
[p sayMaster];
[p sayNB];
NSLog(@"%@",pClass);
}
return 0;
}
2.1 我们再sayHello
方法前设置断点,LLDB调试
其中的 cache_t
的数据
因为在类结构体中 cache_t
前面有 Class ISA指针
和 Class superclass 父类指针
,所以要偏移16位。
(lldb) p/x pClass
(Class) $0 = 0x00000001000022a0 LGPerson
(lldb) p (cache_t *)0x00000001000022b0
(cache_t *) $1 = 0x00000001000022b0
(lldb) p *$1
(cache_t) $2 = {
_buckets = {
std::__1::atomic<bucket_t *> = 0x000000010032e420 {
_sel = {
std::__1::atomic<objc_selector *> = (null)
}
_imp = {
std::__1::atomic<unsigned long> = 0
}
}
}
_mask = {
std::__1::atomic<unsigned int> = 0
}
_flags = 32804
_occupied = 0
}
2.2 然后执行一步 sayHello 方法
,再次进行 LLDB调试
,查看 cache_t
的数据
2020-09-17 22:37:33.187060+0800 KCObjc[34953:549295] LGPerson say : -[LGPerson sayHello]
(lldb) p *$1
(cache_t) $3 = {
_buckets = {
std::__1::atomic<bucket_t *> = 0x00000001006ad5f0 {
_sel = {
std::__1::atomic<objc_selector *> = ""
}
_imp = {
std::__1::atomic<unsigned long> = 11936
}
}
}
_mask = {
std::__1::atomic<unsigned int> = 3
}
_flags = 32804
_occupied = 1
}
2.3 走到这里,大家应该发现 _buckets
、_mask
、_occupied
的变化了。其中_occupied
从0变为1,也证明了执行完 sayHello 方法
之后,缓存方法数量 + 1 。接下来我们查看一下哈希桶 _buckets
的变化,哈希桶数据类型 struct bucket_t
我们点进去查看如下:
struct bucket_t {
public:
inline SEL sel() const { return _sel.load(memory_order::memory_order_relaxed); }
inline IMP imp(Class cls) const {
uintptr_t imp = _imp.load(memory_order::memory_order_relaxed);
if (!imp) return nil;
#if CACHE_IMP_ENCODING == CACHE_IMP_ENCODING_PTRAUTH
SEL sel = _sel.load(memory_order::memory_order_relaxed);
return (IMP)
ptrauth_auth_and_resign((const void *)imp,
ptrauth_key_process_dependent_code,
modifierForSEL(sel, cls),
ptrauth_key_function_pointer, 0);
#elif CACHE_IMP_ENCODING == CACHE_IMP_ENCODING_ISA_XOR
return (IMP)(imp ^ (uintptr_t)cls);
#elif CACHE_IMP_ENCODING == CACHE_IMP_ENCODING_NONE
return (IMP)imp;
#else
#error Unknown method cache IMP encoding.
#endif
}
}
我们就可以查看 _buckets
的 SEL
和 IMP
信息。
(lldb) p $3.buckets()
(bucket_t *) $4 = 0x00000001006ad5f0
(lldb) p *$4
(bucket_t) $5 = {
_sel = {
std::__1::atomic<objc_selector *> = ""
}
_imp = {
std::__1::atomic<unsigned long> = 11936
}
}
(lldb) p $5.sel()
(SEL) $6 = "sayHello"
(lldb) p $5.imp(pClass)
(IMP) $7 = 0x0000000100000c00 (KCObjc`-[LGPerson sayHello])
然后我们也可以打开 MachOView
查看一下 sayHello
方法的 IMP
指针
与我们
LLDB调试
结果不谋而合,完美~
2.4 接下来我们继续执行 sayMaster 方法
和 sayNB 方法
,进行 LLDB调试
,查看 cache_t
的数据
2020-09-17 23:12:37.095330+0800 KCObjc[34953:549295] LGPerson say : -[LGPerson sayCode]
(lldb) p *$1
(cache_t) $8 = {
_buckets = {
std::__1::atomic<bucket_t *> = 0x00000001006ad5f0 {
_sel = {
std::__1::atomic<objc_selector *> = ""
}
_imp = {
std::__1::atomic<unsigned long> = 11936
}
}
}
_mask = {
std::__1::atomic<unsigned int> = 3
}
_flags = 32804
_occupied = 2
}
2020-09-17 23:12:59.163825+0800 KCObjc[34953:549295] LGPerson say : -[LGPerson sayMaster]
(lldb) p *$1
(cache_t) $9 = {
_buckets = {
std::__1::atomic<bucket_t *> = 0x0000000103b4c7d0 {
_sel = {
std::__1::atomic<objc_selector *> = (null)
}
_imp = {
std::__1::atomic<unsigned long> = 0
}
}
}
_mask = {
std::__1::atomic<unsigned int> = 7
}
_flags = 32804
_occupied = 1
}
走到这里,我们发现:
问题①. _occupied
由 2 变为了 1 ,缓存方法数量 _occupied
为什么会减少呢?
问题②. _mask
由 3 变为了 7 ,至于 _mask
的变化,大家可以能想到,前面我们讲过, _mask
是受缓存容量 CACHE SIZE
2 倍扩容的影响。缓存容量 CACHE SIZE
由 4 变为了 8 。
问题③. _buckets
里面的 SEL
和 IMP
消失了。
2.5 接下来,我们来一探究竟。在 void incrementOccupied();
方法中我们看到了 _occupied++;
。
void cache_t::incrementOccupied()
{
_occupied++;
}
2.6 然后我们在源码中找一下,什么地方执行了 incrementOccupied();
这个方法。惊喜来了,cache_t::insert()
方法中执行了 incrementOccupied();
这个方法。从名称我们就可以发现,这是向缓存插入的方法。
void cache_t::insert(Class cls, SEL sel, IMP imp, id receiver)
{
#if CONFIG_USE_CACHE_LOCK
cacheUpdateLock.assertLocked();
#else
runtimeLock.assertLocked();
#endif
ASSERT(sel != 0 && cls->isInitialized());
// Use the cache as-is if it is less than 3/4 full
mask_t newOccupied = occupied() + 1;
unsigned oldCapacity = capacity(), capacity = oldCapacity;
if (slowpath(isConstantEmptyCache())) {
// Cache is read-only. Replace it.
if (!capacity) capacity = INIT_CACHE_SIZE;
reallocate(oldCapacity, capacity, /* freeOld */false);
}
else if (fastpath(newOccupied + CACHE_END_MARKER <= capacity / 4 * 3)) { // 4 3 + 1 bucket cache_t
// Cache is less than 3/4 full. Use it as-is.
}
else {
capacity = capacity ? capacity * 2 : INIT_CACHE_SIZE; // 扩容两倍 4
if (capacity > MAX_CACHE_SIZE) {
capacity = MAX_CACHE_SIZE;
}
reallocate(oldCapacity, capacity, true); // 内存 库容完毕
}
bucket_t *b = buckets();
mask_t m = capacity - 1;
mask_t begin = cache_hash(sel, m);
mask_t i = begin;
// Scan for the first unused slot and insert there.
// There is guaranteed to be an empty slot because the
// minimum size is 4 and we resized at 3/4 full.
do {
if (fastpath(b[i].sel() == 0)) {
incrementOccupied();
b[i].set<Atomic, Encoded>(sel, imp, cls);
return;
}
if (b[i].sel() == sel) {
// The entry was added to the cache by some other thread
// before we grabbed the cacheUpdateLock.
return;
}
} while (fastpath((i = cache_next(i, m)) != begin));
cache_t::bad_cache(receiver, (SEL)sel, cls);
}
2.6.1 接下来我们分析一下这个小概率事件 -> 初始化方法:
如果缓存为空,则开辟缓存 INIT_CACHE_SIZE
:4。然后利用 reallocate() 方法
开辟空间。
enum {
INIT_CACHE_SIZE_LOG2 = 2,
INIT_CACHE_SIZE = (1 << INIT_CACHE_SIZE_LOG2),
MAX_CACHE_SIZE_LOG2 = 16,
MAX_CACHE_SIZE = (1 << MAX_CACHE_SIZE_LOG2),
};
if (slowpath(isConstantEmptyCache())) { // 小概率事件 -> 初始化方法
// Cache is read-only. Replace it.
if (!capacity) capacity = INIT_CACHE_SIZE; // 4 (枚举定义:1 左移 2 位)
reallocate(oldCapacity, capacity, /* freeOld */false);
}
reallocate() 方法
- 申请
newCapacity
大小的地址 - 调用
setBucketsAndMask() 方法
初始化bucket
void cache_t::reallocate(mask_t oldCapacity, mask_t newCapacity, bool freeOld)
{
bucket_t *oldBuckets = buckets();
bucket_t *newBuckets = allocateBuckets(newCapacity);
// Cache's old contents are not propagated.
// This is thought to save cache memory at the cost of extra cache fills.
// fixme re-measure this
ASSERT(newCapacity > 0);
ASSERT((uintptr_t)(mask_t)(newCapacity-1) == newCapacity-1);
setBucketsAndMask(newBuckets, newCapacity - 1);
if (freeOld) {
cache_collect_free(oldBuckets, oldCapacity);
}
}
setBucketsAndMask() 方法
- 将
旧bucket
存入新bucket
- 将
_occupied = 0
,这里我们留意到了reallocate() 方法
会将_occupied = 0
。
void cache_t::setBucketsAndMask(struct bucket_t *newBuckets, mask_t newMask)
{
// objc_msgSend uses mask and buckets with no locks.
// It is safe for objc_msgSend to see new buckets but old mask.
// (It will get a cache miss but not overrun the buckets' bounds).
// It is unsafe for objc_msgSend to see old buckets and new mask.
// Therefore we write new buckets, wait a lot, then write new mask.
// objc_msgSend reads mask first, then buckets.
#ifdef __arm__
// ensure other threads see buckets contents before buckets pointer
mega_barrier();
_buckets.store(newBuckets, memory_order::memory_order_relaxed);
// ensure other threads see new buckets before new mask
mega_barrier();
_mask.store(newMask, memory_order::memory_order_relaxed);
_occupied = 0;
#elif __x86_64__ || i386
// ensure other threads see buckets contents before buckets pointer
_buckets.store(newBuckets, memory_order::memory_order_release);
// ensure other threads see new buckets before new mask
_mask.store(newMask, memory_order::memory_order_release);
_occupied = 0;
#else
#error Don't know how to do setBucketsAndMask on this architecture.
#endif
}
2.6.2 接下来就是大概率事件方法
如果缓存 newOccupied
+ CACHE_END_MARKER(1)
< capacity / 4 * 3,则什么都不需要做。
#define CACHE_END_MARKER 1
else if (fastpath(newOccupied + CACHE_END_MARKER <= capacity / 4 * 3)) { // 4 3 + 1 bucket cache_t
// Cache is less than 3/4 full. Use it as-is.
}
2.6.3 接下来就是扩容方法
- 如果大于总容量的 3 / 4 的时候,就需要扩容了(扩容至2倍)。
- 扩容之后仍然需要利用
reallocate() 方法
开辟空间,在 2.6.1 的
setBucketsAndMask() 方法 中我们讲过,reallocate() 方法
会将_occupied = 0
。到这,我们终于理解了2.4 当中的 问题② ,为什么_occupied
会减少,因为扩容之后_occupied
会初始化至 0,重新计算。
else {
capacity = capacity ? capacity * 2 : INIT_CACHE_SIZE; // 扩容至两倍 4
if (capacity > MAX_CACHE_SIZE) {
capacity = MAX_CACHE_SIZE;
}
reallocate(oldCapacity, capacity, true); // 内存 扩容完毕
}
2.6.4 reallocate() 方法
- 调用
setBucketsAndMask() 方法
初始化bucket
,因为bucket
受扩容影响重新初始化了,所以2.4 当中的 问题③ 的原因就在这里。
void cache_t::reallocate(mask_t oldCapacity, mask_t newCapacity, bool freeOld)
{
bucket_t *oldBuckets = buckets();
bucket_t *newBuckets = allocateBuckets(newCapacity);
// Cache's old contents are not propagated.
// This is thought to save cache memory at the cost of extra cache fills.
// fixme re-measure this
ASSERT(newCapacity > 0);
ASSERT((uintptr_t)(mask_t)(newCapacity-1) == newCapacity-1);
setBucketsAndMask(newBuckets, newCapacity - 1);
if (freeOld) {
cache_collect_free(oldBuckets, oldCapacity);
}
}
2.6.5 接下来就是 _mask
变化的方法,在2.6.3 中我们知道容量扩容到 2 倍,那么 mask
的值就是 2 的 n次幂
- 1 , 所以 2.4 当中的 问题① 便迎刃而解了。
mask_t m = capacity - 1;