diff --git a/ReadMe.md b/ReadMe.md
index a7b0d2f..da1666c 100644
--- a/ReadMe.md
+++ b/ReadMe.md
@@ -57,7 +57,8 @@ YY-Thunks（鸭船），存在的目的就是抹平不同系统的差异，编
 2. 【链接器】-【输入】-【附加依赖项】，添加 
    `objs\$(PlatformShortName)\YY_Thunks_for_WinXP.obj`。
 3. 【链接器】-【系统】-【所需的最低版本】，根据实际情况填写 `5.01`（WinXP 32位） 或者 `5.02`（WinXP x64、2003）。
-4. 重新编译代码。
+4. 如果是编译DLL，那么【链接器】-【高级】-【自定义入口点】更改为`_DllMainCRTStartupForYY_Thunks`（如果不更改那么XP下使用TLS时可能崩溃！）
+5. 重新编译代码。
 
 > 温馨提示：如果需要兼容 Vista，【所需的最低版本】无需修改，但是【附加依赖项】请选择 
   `objs\$(PlatformShortName)\YY_Thunks_for_Vista.obj`。
diff --git a/src/Thunks/DllMainCRTStartup.hpp b/src/Thunks/DllMainCRTStartup.hpp
new file mode 100644
index 0000000..0c93a5d
--- /dev/null
+++ b/src/Thunks/DllMainCRTStartup.hpp
@@ -0,0 +1,292 @@
+#if defined(YY_Thunks_Implemented) && YY_Thunks_Support_Version < NTDDI_WIN6
+
+namespace YY::Thunks::internal
+{
+    extern "C" BOOL WINAPI _DllMainCRTStartup(
+        HINSTANCE const instance,
+        DWORD     const reason,
+        LPVOID    const reserved
+        );
+
+    extern "C" ULONG _tls_index;
+    static ULONG _tls_index_old;
+#ifdef _WIN64
+    extern "C" const IMAGE_TLS_DIRECTORY64 _tls_used;
+#else
+    extern "C" const IMAGE_TLS_DIRECTORY _tls_used;
+#endif
+
+    struct TlsRawItem
+    {
+        TlsRawItem* pNext;
+        TlsRawItem* pPrev;
+        BYTE* pBase;
+    };
+    static thread_local TlsRawItem s_CurrentNode;
+
+    static volatile LONG uStatus = 0;
+    static TlsRawItem* volatile pRoot = nullptr;
+
+    static ULONG __fastcall GetTlsIndexBufferCount(TEB* _pTeb)
+    {
+        auto _ppThreadLocalStoragePointer = (void**)_pTeb->ThreadLocalStoragePointer;
+        if (!_ppThreadLocalStoragePointer)
+            return 0;
+
+        return HeapSize(_pTeb->ProcessEnvironmentBlock->ProcessHeap, 0, _ppThreadLocalStoragePointer) / sizeof(void*);
+    }
+    static PVOID NTAPI RtlImageDirectoryEntryToData(
+        __in PVOID pBaseAddress,
+        __in ULONG dwDirectory,
+        __out PULONG pSize
+        )
+    {
+        auto _pDosHeader = (PIMAGE_DOS_HEADER)pBaseAddress;
+        auto _pNtHerder = reinterpret_cast<PIMAGE_NT_HEADERS>(PBYTE(pBaseAddress) + _pDosHeader->e_lfanew);
+        auto& _DataDirectory = _pNtHerder->OptionalHeader.DataDirectory[dwDirectory];
+
+        *pSize = _DataDirectory.Size;
+        if (_DataDirectory.Size == 0 || _DataDirectory.VirtualAddress == 0)
+            return nullptr;
+
+        return PBYTE(pBaseAddress) + _DataDirectory.VirtualAddress;
+    }
+    
+    static ULONG __fastcall GetMaxTlsIndex() noexcept
+    {
+        ULONG uMaxTlsIndex = 0;
+
+        auto _pLdr = (_PEB_LDR_DATA_XP*)(TEB*)NtCurrentTeb()->ProcessEnvironmentBlock->Ldr;
+        auto _pHerder = &_pLdr->InLoadOrderModuleList;
+        for (auto _pItem = _pHerder->Flink; _pItem != _pHerder; )
+        {
+            auto _pEntry = CONTAINING_RECORD(_pItem, _LDR_DATA_TABLE_ENTRY_XP, InLoadOrderModuleList);
+            _pItem = _pItem->Flink;
+
+            /*if (!_pEntry->TlsIndex)
+                continue;*/
+
+            ULONG TlsSize;
+            auto pTlsImage = (PIMAGE_TLS_DIRECTORY)RtlImageDirectoryEntryToData(
+                _pEntry->BaseAddress,
+                IMAGE_DIRECTORY_ENTRY_TLS,
+                &TlsSize);
+
+            if (pTlsImage == nullptr || pTlsImage->AddressOfIndex == 0)
+            {
+                continue;
+            }
+
+
+            __try
+            {
+
+                const auto _TlsIndex = *(ULONG*)pTlsImage->AddressOfIndex;
+                if (uMaxTlsIndex < _TlsIndex)
+                    uMaxTlsIndex = _TlsIndex;
+            }
+            __except (EXCEPTION_EXECUTE_HANDLER)
+            {
+            }
+        }
+
+        return uMaxTlsIndex;
+        
+    }
+
+    static bool __fastcall AllocTlsData() noexcept
+    {
+        if (_tls_index == 0)
+            return false;
+
+        auto _pTeb= (TEB*)NtCurrentTeb();
+        auto _pTlsIndex = (void**)_pTeb->ThreadLocalStoragePointer;
+        auto _cTlsIndexLength = GetTlsIndexBufferCount(_pTeb);
+        if (_cTlsIndexLength <= _tls_index)
+        {
+            // Index不足，扩充……
+            auto _cNewTlsIndexLength = _tls_index + 128;
+            auto _pNewTlsIndex = (void**)Alloc(_cNewTlsIndexLength * sizeof(void*), HEAP_ZERO_MEMORY);
+            if (!_pNewTlsIndex)
+                return false;
+
+            memcpy(_pNewTlsIndex, _pTlsIndex, _cTlsIndexLength * sizeof(void*));
+            if ((void*)InterlockedCompareExchange((uintptr_t*)&_pTeb->ThreadLocalStoragePointer, (uintptr_t)_pNewTlsIndex, (uintptr_t)_pTlsIndex) != _pTlsIndex)
+            {
+                Free(_pNewTlsIndex);
+                return false;
+            }
+            Free(_pTlsIndex);
+            _pTlsIndex = _pNewTlsIndex;
+        }
+        const size_t _cbTlsRaw = _tls_used.EndAddressOfRawData - _tls_used.StartAddressOfRawData;
+        auto _pRawTlsData = (BYTE*)Alloc(_cbTlsRaw, HEAP_ZERO_MEMORY);
+        if (!_pRawTlsData)
+            return false;
+
+        memcpy(_pRawTlsData, (void*)_tls_used.StartAddressOfRawData, _cbTlsRaw);
+        InterlockedExchange((uintptr_t*)&_pTlsIndex[_tls_index], (uintptr_t)_pRawTlsData);
+
+        s_CurrentNode.pBase = _pRawTlsData;
+        for (;;)
+        {
+            if (!_interlockedbittestandset(&uStatus, 0))
+            {
+                s_CurrentNode.pNext = pRoot;
+                pRoot = &s_CurrentNode;
+                _interlockedbittestandreset(&uStatus, 0);
+                break;
+            }
+        }
+        return true;
+    }
+    
+    static void __fastcall FreeTlsData()
+    {
+        if (_tls_index == 0)
+            return;
+        auto _pTeb = (TEB*)NtCurrentTeb();
+        if (_tls_index >= GetTlsIndexBufferCount(_pTeb))
+            return;
+
+        auto _ppTlsIndex = (void**)_pTeb->ThreadLocalStoragePointer;
+        if (_ppTlsIndex[_tls_index] == nullptr)
+            return;
+
+        if (s_CurrentNode.pBase != _pTeb->ThreadLocalStoragePointer)
+            return;
+
+        s_CurrentNode.pBase = nullptr;
+
+        for (;;)
+        {
+            if (!_interlockedbittestandset(&uStatus, 0))
+            {
+                if (&s_CurrentNode == pRoot)
+                {
+                    s_CurrentNode.pNext->pPrev = nullptr;
+                    pRoot = s_CurrentNode.pNext;
+                }
+                else
+                {
+                    auto pPrev = s_CurrentNode.pPrev;
+                    auto pNext = s_CurrentNode.pNext;
+
+                    if (pPrev)
+                        pPrev->pNext = pNext;
+
+                    if (pNext)
+                        pNext->pPrev = pPrev;
+                }
+                _interlockedbittestandreset(&uStatus, 0);
+
+                auto _pTlsRawData = (void*)InterlockedExchange((uintptr_t*)&_ppTlsIndex[_tls_index], 0);
+                Free(_pTlsRawData);
+                break;
+            }
+        }
+    }
+
+    static ULONG __fastcall CreateTlsIndex() noexcept
+    {
+        return GetMaxTlsIndex() + 1;
+    }
+
+    static void __fastcall FreeTlsIndex() noexcept
+    {
+        if (_tls_index == 0)
+            return;
+
+        // 故意不加锁……
+        for (auto _pItem = pRoot; _pItem;)
+        {
+            auto _pNext = _pItem->pNext;
+            Free(_pItem->pBase);
+            _pItem = _pNext;
+        }
+    }
+
+    static void __fastcall CallTlsCallback(
+        HINSTANCE const _hInstance,
+        DWORD     const _uReason)
+    {
+        if (_tls_index == 0)
+            return;
+
+        auto _pTeb = (TEB*)NtCurrentTeb();
+        auto _ppThreadLocalStoragePointer = (void**)_pTeb->ThreadLocalStoragePointer;
+        if (!_ppThreadLocalStoragePointer)
+            return;
+
+        if (_tls_index >= GetTlsIndexBufferCount(_pTeb))
+            return;
+
+        if (!_ppThreadLocalStoragePointer[_tls_index])
+            return;
+        
+        __try
+        {
+            auto _pfnTlsCallbacks = reinterpret_cast<PIMAGE_TLS_CALLBACK*>(_tls_used.AddressOfCallBacks);
+            if (_pfnTlsCallbacks)
+            {
+                for (; *_pfnTlsCallbacks; ++_pfnTlsCallbacks)
+                {
+                    (*_pfnTlsCallbacks)(_hInstance, _uReason, nullptr);
+                }
+            }
+        }
+        __except (EXCEPTION_EXECUTE_HANDLER)
+        {
+
+        }
+    }
+
+    extern "C" BOOL WINAPI DllMainCRTStartupForYY_Thunks(
+        HINSTANCE const _hInstance,
+        DWORD     const _uReason,
+        LPVOID    const _pReserved
+        )
+    {
+        if (internal::GetSystemVersion() < internal::MakeVersion(6, 0))
+        {
+            // Vista开始已经可以动态的处理TLS问题了，所以这里只针对老系统处理。
+            switch (_uReason)
+            {
+            case DLL_PROCESS_ATTACH:
+                _tls_index_old = _tls_index;
+                if (_tls_index == 0)
+                {
+                    _tls_index = CreateTlsIndex();
+                    AllocTlsData();
+                }
+                break;
+            case DLL_THREAD_ATTACH:
+                if (_tls_index_old == 0)
+                {
+                    AllocTlsData();
+                    CallTlsCallback(_hInstance, _uReason);
+                }
+                break;
+            case DLL_THREAD_DETACH:
+                if (_tls_index_old == 0)
+                {
+                    CallTlsCallback(_hInstance, _uReason);
+                    FreeTlsData();
+                }
+                break;
+            case DLL_PROCESS_DETACH:
+#if (YY_Thunks_Support_Version < NTDDI_WINXP)
+                __YY_Thunks_Process_Terminating = _pReserved != nullptr;
+#endif
+                if (_pReserved == nullptr && _tls_index_old == 0)
+                {
+                    FreeTlsIndex();
+                }
+                break;
+            }
+        }
+
+        return _DllMainCRTStartup(_hInstance, _uReason, _pReserved);
+    }
+}
+#endif
diff --git a/src/YY-Thunks.UnitTest/YY-Thunks.UnitTest.vcxproj b/src/YY-Thunks.UnitTest/YY-Thunks.UnitTest.vcxproj
index 5f2fd8f..071ef63 100644
--- a/src/YY-Thunks.UnitTest/YY-Thunks.UnitTest.vcxproj
+++ b/src/YY-Thunks.UnitTest/YY-Thunks.UnitTest.vcxproj
@@ -222,6 +222,7 @@
     <ClInclude Include="..\Thunks\d3d11.hpp" />
     <ClInclude Include="..\Thunks\d3d9.hpp" />
     <ClInclude Include="..\Thunks\dbghelp.hpp" />
+    <ClInclude Include="..\Thunks\DllMainCRTStartup.hpp" />
     <ClInclude Include="..\Thunks\dwmapi.hpp" />
     <ClInclude Include="..\Thunks\dwrite.hpp" />
     <ClInclude Include="..\Thunks\dxgi.hpp" />
diff --git a/src/YY-Thunks.UnitTest/YY-Thunks.UnitTest.vcxproj.filters b/src/YY-Thunks.UnitTest/YY-Thunks.UnitTest.vcxproj.filters
index b34511e..3cbda50 100644
--- a/src/YY-Thunks.UnitTest/YY-Thunks.UnitTest.vcxproj.filters
+++ b/src/YY-Thunks.UnitTest/YY-Thunks.UnitTest.vcxproj.filters
@@ -308,6 +308,9 @@
     <ClInclude Include="..\Thunks\dbghelp.hpp">
       <Filter>Thunks</Filter>
     </ClInclude>
+    <ClInclude Include="..\Thunks\DllMainCRTStartup.hpp">
+      <Filter>Thunks</Filter>
+    </ClInclude>
   </ItemGroup>
   <ItemGroup>
     <None Include="..\def\x86\PSAPI2Kernel32.def">