一、汇编缓存找不到

在《OC底层探究之从cache到objc_msgSend》文章中，我们把查找缓存的汇编流程过了一遍，汇编流程就是快速查找流程，但是还没有进入MissLabelDynamic流程，这就是慢速查找流程！

而MissLabelDynamic就是__objc_msgSend_uncached：

	STATIC_ENTRY __objc_msgSend_uncached
	UNWIND __objc_msgSend_uncached, FrameWithNoSaves

	// THIS IS NOT A CALLABLE C FUNCTION
	// Out-of-band p15 is the class to search
	
	MethodTableLookup
	TailCallFunctionPointer x17
复制代码

发现就执行了2个函数，先看MethodTableLookup：

.macro MethodTableLookup
	
	SAVE_REGS MSGSEND

	// lookUpImpOrForward(obj, sel, cls, LOOKUP_INITIALIZE | LOOKUP_RESOLVER)
	// receiver and selector already in x0 and x1
	mov	x2, x16
	mov	x3, #3
	bl	_lookUpImpOrForward

	// IMP in x0
	mov	x17, x0

	RESTORE_REGS MSGSEND

.endmacro
复制代码

这里就是存储了一些值以及调用了_lookUpImpOrForward。

在看TailCallFunctionPointer：

#if __has_feature(ptrauth_calls)
.macro TailCallFunctionPointer
	// $0 = function pointer value
	braaz	$0
.endmacro
#else
.macro TailCallMethodListImp
	// $0 = method list imp, $1 = address of method list imp
	br	$0
.endmacro
#endif
复制代码

发现这里直接就返回了$0，而$0就是传入的x17。

那么x17是什么呢？

在MethodTableLookup里有：

// IMP in x0
	mov	x17, x0
复制代码

即x17就x0，x0就是传入的第一个参数或者是返回的结果。

很明显这里的x0就是_lookUpImpOrForward返回的结果！

二、慢速查找流程的准备

1、总览lookUpImpOrForward

NEVER_INLINE
IMP lookUpImpOrForward(id inst, SEL sel, Class cls, int behavior)
{
    const IMP forward_imp = (IMP)_objc_msgForward_impcache;
    IMP imp = nil;
    Class curClass;

    runtimeLock.assertUnlocked();

    if (slowpath(!cls->isInitialized())) {
        // The first message sent to a class is often +new or +alloc, or +self
        // which goes through objc_opt_* or various optimized entry points.
        //
        // However, the class isn't realized/initialized yet at this point,
        // and the optimized entry points fall down through objc_msgSend,
        // which ends up here.
        //
        // We really want to avoid caching these, as it can cause IMP caches
        // to be made with a single entry forever.
        //
        // Note that this check is racy as several threads might try to
        // message a given class for the first time at the same time,
        // in which case we might cache anyway.
        behavior |= LOOKUP_NOCACHE;
    }

    // runtimeLock is held during isRealized and isInitialized checking
    // to prevent races against concurrent realization.

    // runtimeLock is held during method search to make
    // method-lookup + cache-fill atomic with respect to method addition.
    // Otherwise, a category could be added but ignored indefinitely because
    // the cache was re-filled with the old value after the cache flush on
    // behalf of the category.

    runtimeLock.lock();

    // We don't want people to be able to craft a binary blob that looks like
    // a class but really isn't one and do a CFI attack.
    //
    // To make these harder we want to make sure this is a class that was
    // either built into the binary or legitimately registered through
    // objc_duplicateClass, objc_initializeClassPair or objc_allocateClassPair.
    checkIsKnownClass(cls);

    cls = realizeAndInitializeIfNeeded_locked(inst, cls, behavior & LOOKUP_INITIALIZE);
    // runtimeLock may have been dropped but is now locked again
    runtimeLock.assertLocked();
    curClass = cls;

    // The code used to lookup the class's cache again right after
    // we take the lock but for the vast majority of the cases
    // evidence shows this is a miss most of the time, hence a time loss.
    //
    // The only codepath calling into this without having performed some
    // kind of cache lookup is class_getInstanceMethod().

    for (unsigned attempts = unreasonableClassCount();;) {
        if (curClass->cache.isConstantOptimizedCache(/* strict */true)) {
#if CONFIG_USE_PREOPT_CACHES
            imp = cache_getImp(curClass, sel);
            if (imp) goto done_unlock;
            curClass = curClass->cache.preoptFallbackClass();
#endif
        } else {
            // curClass method list.
            Method meth = getMethodNoSuper_nolock(curClass, sel);
            if (meth) {
                imp = meth->imp(false);
                goto done;
            }

            if (slowpath((curClass = curClass->getSuperclass()) == nil)) {
                // No implementation found, and method resolver didn't help.
                // Use forwarding.
                imp = forward_imp;
                break;
            }
        }

        // Halt if there is a cycle in the superclass chain.
        if (slowpath(--attempts == 0)) {
            _objc_fatal("Memory corruption in class list.");
        }

        // Superclass cache.
        imp = cache_getImp(curClass, sel);
        if (slowpath(imp == forward_imp)) {
            // Found a forward:: entry in a superclass.
            // Stop searching, but don't cache yet; call method
            // resolver for this class first.
            break;
        }
        if (fastpath(imp)) {
            // Found the method in a superclass. Cache it in this class.
            goto done;
        }
    }

    // No implementation found. Try method resolver once.

    if (slowpath(behavior & LOOKUP_RESOLVER)) {
        behavior ^= LOOKUP_RESOLVER;
        return resolveMethod_locked(inst, sel, cls, behavior);
    }

 done:
    if (fastpath((behavior & LOOKUP_NOCACHE) == 0)) {
#if CONFIG_USE_PREOPT_CACHES
        while (cls->cache.isConstantOptimizedCache(/* strict */true)) {
            cls = cls->cache.preoptFallbackClass();
        }
#endif
        log_and_fill_cache(cls, imp, sel, inst, curClass);
    }
 done_unlock:
    runtimeLock.unlock();
    if (slowpath((behavior & LOOKUP_NIL) && imp == forward_imp)) {
        return nil;
    }
    return imp;
}
复制代码

代码很多，但是我们到这个方法的主要目的是找imp，看return也可以知道是imp。

imp大多都是for (unsigned attempts = unreasonableClassCount();;)这个循环内，在因此可得知这个循环是重点！

2、判断初始化

从头开始看，一进来先判断了是否初始化：

if (slowpath(!cls->isInitialized())) {
        behavior |= LOOKUP_NOCACHE;
    }
复制代码

3、检测类

接着调用checkIsKnownClass，检测当前的类是否已经注册到当前的缓存表里面：

/***********************************************************************
* checkIsKnownClass
* Checks the given class against the list of all known classes. Dies
* with a fatal error if the class is not known.
* Locking: runtimeLock must be held by the caller.
**********************************************************************/
ALWAYS_INLINE
static void
checkIsKnownClass(Class cls)
{
    if (slowpath(!isKnownClass(cls))) {
        _objc_fatal("Attempt to use unknown class %p.", cls);
    }
}
复制代码

通过中间方法，进入isKnownClass：

/***********************************************************************
* isKnownClass
* Return true if the class is known to the runtime (located within the
* shared cache, within the data segment of a loaded image, or has been
* allocated with obj_allocateClassPair).
*
* The result of this operation is cached on the class in a "witness"
* value that is cheaply checked in the fastpath.
**********************************************************************/
ALWAYS_INLINE
static bool
isKnownClass(Class cls)
{
    if (fastpath(objc::dataSegmentsRanges.contains(cls->data()->witness, (uintptr_t)cls))) {
        return true;
    }
    auto &set = objc::allocatedClasses.get();
    return set.find(cls) != set.end() || dataSegmentsContain(cls);
}
复制代码

在allocatedClasses方法获取表：

/***********************************************************************
* allocatedClasses
* A table of all classes (and metaclasses) which have been allocated
* with objc_allocateClassPair.
**********************************************************************/
namespace objc {
static ExplicitInitDenseSet<Class> allocatedClasses;
}
复制代码

4、类的实现

接着到类的实现方法：

cls = realizeAndInitializeIfNeeded_locked(inst, cls, behavior & LOOKUP_INITIALIZE);
复制代码

进过几次中间方法，最终进入到realizeClassWithoutSwift方法：

/***********************************************************************
* realizeClassWithoutSwift
* Performs first-time initialization on class cls, 
* including allocating its read-write data.
* Does not perform any Swift-side initialization.
* Returns the real class structure for the class. 
* Locking: runtimeLock must be write-locked by the caller
**********************************************************************/
static Class realizeClassWithoutSwift(Class cls, Class previously)
{
    runtimeLock.assertLocked();

    class_rw_t *rw;
    Class supercls;
    Class metacls;

    if (!cls) return nil;
    if (cls->isRealized()) {
        validateAlreadyRealizedClass(cls);
        return cls;
    }
    ASSERT(cls == remapClass(cls));

    // fixme verify class is not in an un-dlopened part of the shared cache?

    auto ro = (const class_ro_t *)cls->data();
    auto isMeta = ro->flags & RO_META;
    if (ro->flags & RO_FUTURE) {
        // This was a future class. rw data is already allocated.
        rw = cls->data();
        ro = cls->data()->ro();
        ASSERT(!isMeta);
        cls->changeInfo(RW_REALIZED|RW_REALIZING, RW_FUTURE);
    } else {
        // Normal class. Allocate writeable class data.
        rw = objc::zalloc<class_rw_t>();
        rw->set_ro(ro);
        rw->flags = RW_REALIZED|RW_REALIZING|isMeta;
        cls->setData(rw);
    }

    cls->cache.initializeToEmptyOrPreoptimizedInDisguise();

#if FAST_CACHE_META
    if (isMeta) cls->cache.setBit(FAST_CACHE_META);
#endif

    // Choose an index for this class.
    // Sets cls->instancesRequireRawIsa if indexes no more indexes are available
    cls->chooseClassArrayIndex();

    if (PrintConnecting) {
        _objc_inform("CLASS: realizing class '%s'%s %p %p #%u %s%s",
                     cls->nameForLogging(), isMeta ? " (meta)" : "", 
                     (void*)cls, ro, cls->classArrayIndex(),
                     cls->isSwiftStable() ? "(swift)" : "",
                     cls->isSwiftLegacy() ? "(pre-stable swift)" : "");
    }

    // Realize superclass and metaclass, if they aren't already.
    // This needs to be done after RW_REALIZED is set above, for root classes.
    // This needs to be done after class index is chosen, for root metaclasses.
    // This assumes that none of those classes have Swift contents,
    //   or that Swift's initializers have already been called.
    //   fixme that assumption will be wrong if we add support
    //   for ObjC subclasses of Swift classes.
    supercls = realizeClassWithoutSwift(remapClass(cls->getSuperclass()), nil);
    metacls = realizeClassWithoutSwift(remapClass(cls->ISA()), nil);

#if SUPPORT_NONPOINTER_ISA
    if (isMeta) {
        // Metaclasses do not need any features from non pointer ISA
        // This allows for a faspath for classes in objc_retain/objc_release.
        cls->setInstancesRequireRawIsa();
    } else {
        // Disable non-pointer isa for some classes and/or platforms.
        // Set instancesRequireRawIsa.
        bool instancesRequireRawIsa = cls->instancesRequireRawIsa();
        bool rawIsaIsInherited = false;
        static bool hackedDispatch = false;

        if (DisableNonpointerIsa) {
            // Non-pointer isa disabled by environment or app SDK version
            instancesRequireRawIsa = true;
        }
        else if (!hackedDispatch  &&  0 == strcmp(ro->getName(), "OS_object"))
        {
            // hack for libdispatch et al - isa also acts as vtable pointer
            hackedDispatch = true;
            instancesRequireRawIsa = true;
        }
        else if (supercls  &&  supercls->getSuperclass()  &&
                 supercls->instancesRequireRawIsa())
        {
            // This is also propagated by addSubclass()
            // but nonpointer isa setup needs it earlier.
            // Special case: instancesRequireRawIsa does not propagate
            // from root class to root metaclass
            instancesRequireRawIsa = true;
            rawIsaIsInherited = true;
        }

        if (instancesRequireRawIsa) {
            cls->setInstancesRequireRawIsaRecursively(rawIsaIsInherited);
        }
    }
// SUPPORT_NONPOINTER_ISA
#endif

    // Update superclass and metaclass in case of remapping
    cls->setSuperclass(supercls);
    cls->initClassIsa(metacls);

    // Reconcile instance variable offsets / layout.
    // This may reallocate class_ro_t, updating our ro variable.
    if (supercls  &&  !isMeta) reconcileInstanceVariables(cls, supercls, ro);

    // Set fastInstanceSize if it wasn't set already.
    cls->setInstanceSize(ro->instanceSize);

    // Copy some flags from ro to rw
    if (ro->flags & RO_HAS_CXX_STRUCTORS) {
        cls->setHasCxxDtor();
        if (! (ro->flags & RO_HAS_CXX_DTOR_ONLY)) {
            cls->setHasCxxCtor();
        }
    }
    
    // Propagate the associated objects forbidden flag from ro or from
    // the superclass.
    if ((ro->flags & RO_FORBIDS_ASSOCIATED_OBJECTS) ||
        (supercls && supercls->forbidsAssociatedObjects()))
    {
        rw->flags |= RW_FORBIDS_ASSOCIATED_OBJECTS;
    }

    // Connect this class to its superclass's subclass lists
    if (supercls) {
        addSubclass(supercls, cls);
    } else {
        addRootClass(cls);
    }

    // Attach categories
    methodizeClass(cls, previously);

    return cls;
}
复制代码

先是rw、ro赋值，即属性列表、方法列表等等。

然后实现元类和父类：

supercls = realizeClassWithoutSwift(remapClass(cls->getSuperclass()), nil);
metacls = realizeClassWithoutSwift(remapClass(cls->ISA()), nil);
复制代码

这里是一个递归调用。

接着把元类和父类赋值给当前类：

 // Update superclass and metaclass in case of remapping
    cls->setSuperclass(supercls);
    cls->initClassIsa(metacls);
复制代码

就是在这样的递归调用中把isa走位图的所有相关类都实现了！

为什么要把所有的相关类都实现呢？

因为要找方法！

对象方法要是当前类找不到方法就需要去父类中寻找！

类方法要是当前类的isa找不到方法就需要去父类的isa中寻找！

以上就是为了for循环做的准备！

三、二分查找流程

按照我们一般的理解，查找对象方法：

1、在自己这里查找，即在methodlist里面查找sel和imp。

2、找不到则去父类找，再去NSObject里面去找，最后到nil报错。

接着我们就来看看是不是这样呢？

进入到for (unsigned attempts = unreasonableClassCount();;)循环，发现这是一个死循环！

即只有break或者goto等才能跳出循环！

进来的第一个判断：

if (curClass->cache.isConstantOptimizedCache(/* strict */true)) {
#if CONFIG_USE_PREOPT_CACHES
            imp = cache_getImp(curClass, sel);
            if (imp) goto done_unlock;
            curClass = curClass->cache.preoptFallbackClass();
#endif
        }
复制代码

这里是去共享缓存里面进行查找！共享缓存一般是存储系统的方法，我们的方法一般是不在共享缓存里面的。

直接进入else：

else {
            // curClass method list.
            Method meth = getMethodNoSuper_nolock(curClass, sel);
            if (meth) {
                imp = meth->imp(false);
                goto done;
            }

            if (slowpath((curClass = curClass->getSuperclass()) == nil)) {
                // No implementation found, and method resolver didn't help.
                // Use forwarding.
                imp = forward_imp;
                break;
            }
        }
复制代码

首先调用getMethodNoSuper_nolock获取方法列表：

/***********************************************************************
 * getMethodNoSuper_nolock
 * fixme
 * Locking: runtimeLock must be read- or write-locked by the caller
 **********************************************************************/
static method_t *
getMethodNoSuper_nolock(Class cls, SEL sel)
{
    runtimeLock.assertLocked();

    ASSERT(cls->isRealized());
    // fixme nil cls? 
    // fixme nil sel?

    auto const methods = cls->data()->methods();
    for (auto mlists = methods.beginLists(),
              end = methods.endLists();
         mlists != end;
         ++mlists)
    {
        // <rdar://problem/46904873> getMethodNoSuper_nolock is the hottest
        // caller of search_method_list, inlining it turns
        // getMethodNoSuper_nolock into a frame-less function and eliminates
        // any store from this codepath.
        method_t *m = search_method_list_inline(*mlists, sel);
        if (m) return m;
    }

    return nil;
}
复制代码

先通过cls->data()->methods()获取方法列表，在《OC底层探究之类的底层原理结构》文章中进行过探索。

然后遍历方法列表调用search_method_list_inline获取方法method_t：

ALWAYS_INLINE static method_t *
search_method_list_inline(const method_list_t *mlist, SEL sel)
{
    int methodListIsFixedUp = mlist->isFixedUp();
    int methodListHasExpectedSize = mlist->isExpectedSize();
    
    if (fastpath(methodListIsFixedUp && methodListHasExpectedSize)) {
        return findMethodInSortedMethodList(sel, mlist);
    } else {
        // Linear search of unsorted method list
        if (auto *m = findMethodInUnsortedMethodList(sel, mlist))
            return m;
    }

#if DEBUG
    // sanity-check negative results
    if (mlist->isFixedUp()) {
        for (auto& meth : *mlist) {
            if (meth.name() == sel) {
                _objc_fatal("linear search worked when binary search did not");
            }
        }
    }
#endif

    return nil;
}
复制代码

然后进入findMethodInSortedMethodList通过排好序的方法列表进行查找：

ALWAYS_INLINE static method_t *
findMethodInSortedMethodList(SEL key, const method_list_t *list)
{
    if (list->isSmallList()) {
        if (CONFIG_SHARED_CACHE_RELATIVE_DIRECT_SELECTORS && objc::inSharedCache((uintptr_t)list)) {
            return findMethodInSortedMethodList(key, list, [](method_t &m) { return m.getSmallNameAsSEL(); });
        } else {
            return findMethodInSortedMethodList(key, list, [](method_t &m) { return m.getSmallNameAsSELRef(); });
        }
    } else {
        return findMethodInSortedMethodList(key, list, [](method_t &m) { return m.big().name; });
    }
}
复制代码

这里是判断是SmallList还是BigList。

BigList是传统的电脑，所以先看else，进入findMethodInSortedMethodList：

/***********************************************************************
 * search_method_list_inline
 **********************************************************************/
template<class getNameFunc>
ALWAYS_INLINE static method_t *
findMethodInSortedMethodList(SEL key, const method_list_t *list, const getNameFunc &getName)
{
    ASSERT(list);

    auto first = list->begin();
    auto base = first;
    decltype(first) probe;

    uintptr_t keyValue = (uintptr_t)key;
    uint32_t count;
    
    for (count = list->count; count != 0; count >>= 1) {
        probe = base + (count >> 1);
        
        uintptr_t probeValue = (uintptr_t)getName(probe);
        
        if (keyValue == probeValue) {
            // `probe` is a match.
            // Rewind looking for the *first* occurrence of this value.
            // This is required for correct category overrides.
            while (probe > first && keyValue == (uintptr_t)getName((probe - 1))) {
                probe--;
            }
            return &*probe;
        }
        
        if (keyValue > probeValue) {
            base = probe + 1;
            count--;
        }
    }
    
    return nil;
}
复制代码

先看for循环，每次循环完成后，右移1位，其实就除2！

然后看循环内：

先是probe赋值为base + (count >> 1)，base是开始位置，即为0，所以这里probe为count / 2。

然后取probe处的名字即sel，然后让需要查找的sel和当前sel对比，如果一样且probe大于起始位置，则让需要查找的sel继续比较下一个sel，如果还一样，则probe--，最后返回probe的地址！

为什么要比较下一个名字呢？因为可能存在对象方法和类方法名字一样的时候！

如果不一致，则对比大小，如果要查找的sel比当前的sel大，则说明要查找的sel在后面，否则就在前面！

在后面的话base赋值为probe + 1，同时count--，最后继续循环。

四、慢速查找递归流程

1、查到方法

当在二分法中找到method_t（方法）后，直接跳转到done：

 done:
    if (fastpath((behavior & LOOKUP_NOCACHE) == 0)) {
#if CONFIG_USE_PREOPT_CACHES
        while (cls->cache.isConstantOptimizedCache(/* strict */true)) {
            cls = cls->cache.preoptFallbackClass();
        }
#endif
        log_and_fill_cache(cls, imp, sel, inst, curClass);
    }
复制代码

执行填充缓存方法log_and_fill_cache：

/***********************************************************************
* log_and_fill_cache
* Log this method call. If the logger permits it, fill the method cache.
* cls is the method whose cache should be filled. 
* implementer is the class that owns the implementation in question.
**********************************************************************/
static void
log_and_fill_cache(Class cls, IMP imp, SEL sel, id receiver, Class implementer)
{
#if SUPPORT_MESSAGE_LOGGING
    if (slowpath(objcMsgLogEnabled && implementer)) {
        bool cacheIt = logMessageSend(implementer->isMetaClass(), 
                                      cls->nameForLogging(),
                                      implementer->nameForLogging(), 
                                      sel);
        if (!cacheIt) return;
    }
#endif
    cls->cache.insert(sel, imp, receiver);
}
复制代码

最后进入到我们熟悉的insert函数！

2、去父类中查找

如果在方法列表中没有找到方法，先是判断if (slowpath((curClass = curClass->getSuperclass()) == nil))，即当前的class赋值为superclass且为nil的时候，break出去！什么时候会是nil？NSObject的父类就是nil！说明到这里已经把父类全部查完了，都没有查到！然后就会进入消息转发流程！

如果还有父类，则开始去父类中查找：

// Superclass cache.
imp = cache_getImp(curClass, sel);
复制代码

进入到cache_getImp：

// returns:
// - the cached IMP when one is found
// - nil if there's no cached value and the cache is dynamic
// - `value_on_constant_cache_miss` if there's no cached value and the cache is preoptimized
extern "C" IMP cache_getImp(Class cls, SEL sel, IMP value_on_constant_cache_miss = nil);
复制代码

发现没有后续了！

全局搜索cache_getImp：

	STATIC_ENTRY _cache_getImp

	GetClassFromIsa_p16 p0, 0
	CacheLookup GETIMP, _cache_getImp, LGetImpMissDynamic, LGetImpMissConstant
复制代码

发现这就是我们之前的汇编查找缓存的流程，即快速查找流程！

说明当前类找不到方法后，就去父类中进行快速查找流程，查找缓存中的方法！

如果在父类中找到了方法，进入done，进行缓存。

如果在父类中没有找到方法，继续循环，进行慢速查找流程！

3、方法决议

如果跳出循环，则进入到方法决议resolveMethod_locked！

五、总结