Temporarily restore support for Win7 / Server 2008 R2 (microsoft#4857)

stl/CMakeLists.txt

@@ -556,7 +556,7 @@ function(add_stl_dlls D_SUFFIX REL_OR_DBG)
     generate_satellite_def("atomic_wait" "${D_SUFFIX}")
 
     add_library(msvcp${D_SUFFIX}_atomic_wait SHARED "${CMAKE_BINARY_DIR}/msvcp_atomic_wait${D_SUFFIX}.def")
-    target_link_libraries(msvcp${D_SUFFIX}_atomic_wait PRIVATE msvcp${D_SUFFIX}_atomic_wait_objects msvcp${D_SUFFIX}_satellite_objects msvcp${D_SUFFIX}_implib_objects "msvcp${D_SUFFIX}" "${TOOLSET_LIB}/vcruntime${D_SUFFIX}.lib" "${TOOLSET_LIB}/msvcrt${D_SUFFIX}.lib" "ucrt${D_SUFFIX}.lib" "advapi32.lib" "synchronization.lib")
+    target_link_libraries(msvcp${D_SUFFIX}_atomic_wait PRIVATE msvcp${D_SUFFIX}_atomic_wait_objects msvcp${D_SUFFIX}_satellite_objects msvcp${D_SUFFIX}_implib_objects "msvcp${D_SUFFIX}" "${TOOLSET_LIB}/vcruntime${D_SUFFIX}.lib" "${TOOLSET_LIB}/msvcrt${D_SUFFIX}.lib" "ucrt${D_SUFFIX}.lib" "advapi32.lib")
     set_target_properties(msvcp${D_SUFFIX}_atomic_wait PROPERTIES ARCHIVE_OUTPUT_NAME "msvcp140_atomic_wait${D_SUFFIX}${VCLIBS_SUFFIX}")
     set_target_properties(msvcp${D_SUFFIX}_atomic_wait PROPERTIES ARCHIVE_OUTPUT_DIRECTORY "${CMAKE_CURRENT_BINARY_DIR}")
     set_target_properties(msvcp${D_SUFFIX}_atomic_wait PROPERTIES OUTPUT_NAME "msvcp140${D_SUFFIX}_atomic_wait${VCLIBS_SUFFIX}")

stl/inc/xatomic_wait.h

@@ -21,9 +21,23 @@ _STL_DISABLE_CLANG_WARNINGS
 extern "C" {
 inline constexpr unsigned long __std_atomic_wait_no_timeout = 0xFFFF'FFFF; // Pass as partial timeout
 
+enum class __std_atomic_api_level : unsigned long {
+    __not_set,
+    __detecting,
+    __has_srwlock,
+    __has_wait_on_address,
+};
+
+// This function allows testing the atomic wait support while always using the APIs for a platform with fewer
+// capabilities; it attempts to lock the APIs used to the level `_Requested_api_level`, and returns the actual API level
+// in use. Once the API level has been set by calling this function (or detected by a call to one of the atomic wait
+// functions), it can no longer be changed.
+__std_atomic_api_level __stdcall __std_atomic_set_api_level(__std_atomic_api_level _Requested_api_level) noexcept;
+
 // Support for atomic waits.
 // The "direct" functions are used when the underlying infrastructure can use WaitOnAddress directly; that is, _Size is
-// 1, 2, 4, or 8. The contract is the same as the WaitOnAddress function from the Windows SDK.
+// 1, 2, 4, or 8. The contract is the same as the WaitOnAddress function from the Windows SDK. If WaitOnAddress is not
+// available on the current platform, falls back to a similar solution based on SRWLOCK and CONDITION_VARIABLE.
 int __stdcall __std_atomic_wait_direct(
     const void* _Storage, void* _Comparand, size_t _Size, unsigned long _Remaining_timeout) noexcept;
 void __stdcall __std_atomic_notify_one_direct(const void* _Storage) noexcept;

stl/inc/yvals_core.h

@@ -1939,6 +1939,7 @@ compiler option, or define _ALLOW_RTCc_IN_STL to suppress this error.
 #endif // defined(MRTDLL) && !defined(_M_CEE_PURE)
 
 #define _STL_WIN32_WINNT_VISTA 0x0600 // _WIN32_WINNT_VISTA from sdkddkver.h
+#define _STL_WIN32_WINNT_WIN7  0x0601 // _WIN32_WINNT_WIN7 from sdkddkver.h
 #define _STL_WIN32_WINNT_WIN8  0x0602 // _WIN32_WINNT_WIN8 from sdkddkver.h
 #define _STL_WIN32_WINNT_WIN10 0x0A00 // _WIN32_WINNT_WIN10 from sdkddkver.h
 
@@ -1947,10 +1948,13 @@ compiler option, or define _ALLOW_RTCc_IN_STL to suppress this error.
 #if defined(_M_ARM64)
 // The first ARM64 Windows was Windows 10
 #define _STL_WIN32_WINNT _STL_WIN32_WINNT_WIN10
-#else // ^^^ defined(_M_ARM64) / !defined(_M_ARM64) vvv
-// The earliest Windows supported by this implementation is Windows 8
+#elif defined(_M_ARM) || defined(_ONECORE) || defined(_CRT_APP)
+// The first ARM or OneCore or App Windows was Windows 8
 #define _STL_WIN32_WINNT _STL_WIN32_WINNT_WIN8
-#endif // ^^^ !defined(_M_ARM64) ^^^
+#else // ^^^ default to Win8 / default to Win7 vvv
+// The earliest Windows supported by this implementation is Windows 7
+#define _STL_WIN32_WINNT _STL_WIN32_WINNT_WIN7
+#endif // ^^^ !defined(_M_ARM) && !defined(_M_ARM64) && !defined(_ONECORE) && !defined(_CRT_APP) ^^^
 #endif // !defined(_STL_WIN32_WINNT)
 
 #ifdef __cpp_noexcept_function_type

stl/msbuild/stl_atomic_wait/msvcp_atomic_wait.settings.targets

@@ -70,7 +70,6 @@ SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
         <TargetLib Include="$(CrtLibPath)\vcruntime$(BuildSuffix)$(ClrLibSuffix).lib"/>
         <TargetLib Include="$(UniversalCRTLib)"/>
         <TargetLib Condition="'$(MsvcpFlavor)' == 'kernel32' or '$(MsvcpFlavor)' == 'netfx'" Include="$(SdkLibPath)\advapi32.lib"/>
-        <TargetLib Condition="'$(MsvcpFlavor)' == 'kernel32' or '$(MsvcpFlavor)' == 'netfx'" Include="$(SdkLibPath)\synchronization.lib"/>
     </ItemGroup>
 
     <!-- Copy the output dll and pdb to various destinations -->

stl/src/atomic_wait.cpp

@@ -11,8 +11,6 @@
 
 #include <Windows.h>
 
-#pragma comment(lib, "synchronization")
-
 namespace {
     constexpr unsigned long long _Atomic_wait_no_deadline = 0xFFFF'FFFF'FFFF'FFFF;
 
@@ -91,13 +89,134 @@ namespace {
         }
 #endif // defined(_DEBUG)
     }
+
+#ifndef _ATOMIC_WAIT_ON_ADDRESS_STATICALLY_AVAILABLE
+#if _STL_WIN32_WINNT >= _STL_WIN32_WINNT_WIN8
+#define _ATOMIC_WAIT_ON_ADDRESS_STATICALLY_AVAILABLE 1
+#else // ^^^ _STL_WIN32_WINNT >= _STL_WIN32_WINNT_WIN8 / _STL_WIN32_WINNT < _STL_WIN32_WINNT_WIN8 vvv
+#define _ATOMIC_WAIT_ON_ADDRESS_STATICALLY_AVAILABLE 0
+#endif // ^^^ _STL_WIN32_WINNT < _STL_WIN32_WINNT_WIN8 ^^^
+#endif // !defined(_ATOMIC_WAIT_ON_ADDRESS_STATICALLY_AVAILABLE)
+
+#if _ATOMIC_WAIT_ON_ADDRESS_STATICALLY_AVAILABLE
+
+#pragma comment(lib, "synchronization")
+
+#define __crtWaitOnAddress       WaitOnAddress
+#define __crtWakeByAddressSingle WakeByAddressSingle
+#define __crtWakeByAddressAll    WakeByAddressAll
+
+#else // ^^^ _ATOMIC_WAIT_ON_ADDRESS_STATICALLY_AVAILABLE / !_ATOMIC_WAIT_ON_ADDRESS_STATICALLY_AVAILABLE vvv
+
+    struct _Wait_functions_table {
+        _STD atomic<decltype(&::WaitOnAddress)> _Pfn_WaitOnAddress{nullptr};
+        _STD atomic<decltype(&::WakeByAddressSingle)> _Pfn_WakeByAddressSingle{nullptr};
+        _STD atomic<decltype(&::WakeByAddressAll)> _Pfn_WakeByAddressAll{nullptr};
+        _STD atomic<__std_atomic_api_level> _Api_level{__std_atomic_api_level::__not_set};
+    };
+
+    _Wait_functions_table _Wait_functions;
+
+    void _Force_wait_functions_srwlock_only() noexcept {
+        auto _Local = _Wait_functions._Api_level.load(_STD memory_order_acquire);
+        if (_Local <= __std_atomic_api_level::__detecting) {
+            while (!_Wait_functions._Api_level.compare_exchange_weak(
+                _Local, __std_atomic_api_level::__has_srwlock, _STD memory_order_acq_rel)) {
+                if (_Local > __std_atomic_api_level::__detecting) {
+                    return;
+                }
+            }
+        }
+    }
+
+    [[nodiscard]] __std_atomic_api_level _Init_wait_functions(__std_atomic_api_level _Level) {
+        while (!_Wait_functions._Api_level.compare_exchange_weak(
+            _Level, __std_atomic_api_level::__detecting, _STD memory_order_acq_rel)) {
+            if (_Level > __std_atomic_api_level::__detecting) {
+                return _Level;
+            }
+        }
+
+        _Level = __std_atomic_api_level::__has_srwlock;
+
+        const HMODULE _Sync_module = GetModuleHandleW(L"api-ms-win-core-synch-l1-2-0.dll");
+        if (_Sync_module != nullptr) {
+            const auto _Wait_on_address =
+                reinterpret_cast<decltype(&::WaitOnAddress)>(GetProcAddress(_Sync_module, "WaitOnAddress"));
+            const auto _Wake_by_address_single =
+                reinterpret_cast<decltype(&::WakeByAddressSingle)>(GetProcAddress(_Sync_module, "WakeByAddressSingle"));
+            const auto _Wake_by_address_all =
+                reinterpret_cast<decltype(&::WakeByAddressAll)>(GetProcAddress(_Sync_module, "WakeByAddressAll"));
+
+            if (_Wait_on_address != nullptr && _Wake_by_address_single != nullptr && _Wake_by_address_all != nullptr) {
+                _Wait_functions._Pfn_WaitOnAddress.store(_Wait_on_address, _STD memory_order_relaxed);
+                _Wait_functions._Pfn_WakeByAddressSingle.store(_Wake_by_address_single, _STD memory_order_relaxed);
+                _Wait_functions._Pfn_WakeByAddressAll.store(_Wake_by_address_all, _STD memory_order_relaxed);
+                _Level = __std_atomic_api_level::__has_wait_on_address;
+            }
+        }
+
+        // for __has_srwlock, relaxed would have been enough, not distinguishing for consistency
+        _Wait_functions._Api_level.store(_Level, _STD memory_order_release);
+        return _Level;
+    }
+
+    [[nodiscard]] __std_atomic_api_level _Acquire_wait_functions() noexcept {
+        auto _Level = _Wait_functions._Api_level.load(_STD memory_order_acquire);
+        if (_Level <= __std_atomic_api_level::__detecting) {
+            _Level = _Init_wait_functions(_Level);
+        }
+
+        return _Level;
+    }
+
+    [[nodiscard]] BOOL __crtWaitOnAddress(
+        volatile VOID* Address, PVOID CompareAddress, SIZE_T AddressSize, DWORD dwMilliseconds) {
+        const auto _Wait_on_address = _Wait_functions._Pfn_WaitOnAddress.load(_STD memory_order_relaxed);
+        return _Wait_on_address(Address, CompareAddress, AddressSize, dwMilliseconds);
+    }
+
+    VOID __crtWakeByAddressSingle(PVOID Address) {
+        const auto _Wake_by_address_single = _Wait_functions._Pfn_WakeByAddressSingle.load(_STD memory_order_relaxed);
+        _Wake_by_address_single(Address);
+    }
+
+    VOID __crtWakeByAddressAll(PVOID Address) {
+        const auto _Wake_by_address_all = _Wait_functions._Pfn_WakeByAddressAll.load(_STD memory_order_relaxed);
+        _Wake_by_address_all(Address);
+    }
+
+    bool __stdcall _Atomic_wait_are_equal_direct_fallback(
+        const void* _Storage, void* _Comparand, size_t _Size, void*) noexcept {
+        switch (_Size) {
+        case 1:
+            return __iso_volatile_load8(static_cast<const char*>(_Storage)) == *static_cast<const char*>(_Comparand);
+        case 2:
+            return __iso_volatile_load16(static_cast<const short*>(_Storage)) == *static_cast<const short*>(_Comparand);
+        case 4:
+            return __iso_volatile_load32(static_cast<const int*>(_Storage)) == *static_cast<const int*>(_Comparand);
+        case 8:
+            return __iso_volatile_load64(static_cast<const long long*>(_Storage))
+                == *static_cast<const long long*>(_Comparand);
+        default:
+            _CSTD abort();
+        }
+    }
+#endif // _ATOMIC_WAIT_ON_ADDRESS_STATICALLY_AVAILABLE
 } // unnamed namespace
 
 extern "C" {
 int __stdcall __std_atomic_wait_direct(const void* const _Storage, void* const _Comparand, const size_t _Size,
     const unsigned long _Remaining_timeout) noexcept {
-    const auto _Result =
-        WaitOnAddress(const_cast<volatile void*>(_Storage), const_cast<void*>(_Comparand), _Size, _Remaining_timeout);
+#if _ATOMIC_WAIT_ON_ADDRESS_STATICALLY_AVAILABLE == 0
+    if (_Acquire_wait_functions() < __std_atomic_api_level::__has_wait_on_address) {
+        return __std_atomic_wait_indirect(
+            _Storage, _Comparand, _Size, nullptr, &_Atomic_wait_are_equal_direct_fallback, _Remaining_timeout);
+    }
+#endif // _ATOMIC_WAIT_ON_ADDRESS_STATICALLY_AVAILABLE == 0
+
+    const auto _Result = __crtWaitOnAddress(
+        const_cast<volatile void*>(_Storage), const_cast<void*>(_Comparand), _Size, _Remaining_timeout);
 
     if (!_Result) {
         _Assume_timeout();
@@ -106,11 +225,25 @@ int __stdcall __std_atomic_wait_direct(const void* const _Storage, void* const _
 }
 
 void __stdcall __std_atomic_notify_one_direct(const void* const _Storage) noexcept {
-    WakeByAddressSingle(const_cast<void*>(_Storage));
+#if _ATOMIC_WAIT_ON_ADDRESS_STATICALLY_AVAILABLE == 0
+    if (_Acquire_wait_functions() < __std_atomic_api_level::__has_wait_on_address) {
+        __std_atomic_notify_one_indirect(_Storage);
+        return;
+    }
+#endif // _ATOMIC_WAIT_ON_ADDRESS_STATICALLY_AVAILABLE = 0
+
+    __crtWakeByAddressSingle(const_cast<void*>(_Storage));
 }
 
 void __stdcall __std_atomic_notify_all_direct(const void* const _Storage) noexcept {
-    WakeByAddressAll(const_cast<void*>(_Storage));
+#if _ATOMIC_WAIT_ON_ADDRESS_STATICALLY_AVAILABLE == 0
+    if (_Acquire_wait_functions() < __std_atomic_api_level::__has_wait_on_address) {
+        __std_atomic_notify_all_indirect(_Storage);
+        return;
+    }
+#endif // _ATOMIC_WAIT_ON_ADDRESS_STATICALLY_AVAILABLE == 0
+
+    __crtWakeByAddressAll(const_cast<void*>(_Storage));
 }
 
 void __stdcall __std_atomic_notify_one_indirect(const void* const _Storage) noexcept {
@@ -206,10 +339,25 @@ unsigned long __stdcall __std_atomic_wait_get_remaining_timeout(unsigned long lo
     return static_cast<unsigned long>(_Remaining);
 }
 
-// TRANSITION, ABI: preserved for binary compatibility
-enum class __std_atomic_api_level : unsigned long { __not_set, __detecting, __has_srwlock, __has_wait_on_address };
-__std_atomic_api_level __stdcall __std_atomic_set_api_level(__std_atomic_api_level) noexcept {
+__std_atomic_api_level __stdcall __std_atomic_set_api_level(__std_atomic_api_level _Requested_api_level) noexcept {
+#if _ATOMIC_WAIT_ON_ADDRESS_STATICALLY_AVAILABLE
+    (void) _Requested_api_level;
     return __std_atomic_api_level::__has_wait_on_address;
+#else // ^^^ _ATOMIC_WAIT_ON_ADDRESS_STATICALLY_AVAILABLE / !_ATOMIC_WAIT_ON_ADDRESS_STATICALLY_AVAILABLE vvv
+    switch (_Requested_api_level) {
+    case __std_atomic_api_level::__not_set:
+    case __std_atomic_api_level::__detecting:
+        _CSTD abort();
+    case __std_atomic_api_level::__has_srwlock:
+        _Force_wait_functions_srwlock_only();
+        break;
+    case __std_atomic_api_level::__has_wait_on_address:
+    default: // future compat: new header using an old DLL will get the highest requested level supported
+        break;
+    }
+
+    return _Acquire_wait_functions();
+#endif // !_ATOMIC_WAIT_ON_ADDRESS_STATICALLY_AVAILABLE
 }
 
 #pragma warning(push)

stl/src/awint.hpp

@@ -14,6 +14,17 @@
 
 _CRT_BEGIN_C_HEADER
 
+#if _STL_WIN32_WINNT >= _WIN32_WINNT_WIN8
+
+#define __crtGetSystemTimePreciseAsFileTime(lpSystemTimeAsFileTime) \
+    GetSystemTimePreciseAsFileTime(lpSystemTimeAsFileTime)
+
+#else // ^^^ _STL_WIN32_WINNT >= _WIN32_WINNT_WIN8 / _STL_WIN32_WINNT < _WIN32_WINNT_WIN8 vvv
+
+_CRTIMP2 void __cdecl __crtGetSystemTimePreciseAsFileTime(_Out_ LPFILETIME lpSystemTimeAsFileTime) noexcept;
+
+#endif // ^^^ _STL_WIN32_WINNT < _WIN32_WINNT_WIN8 ^^^
+
 _CRTIMP2 int __cdecl __crtCompareStringA(_In_z_ LPCWSTR _LocaleName, _In_ DWORD _DwCmpFlags,
     _In_reads_(_CchCount1) LPCSTR _LpString1, _In_ int _CchCount1, _In_reads_(_CchCount2) LPCSTR _LpString2,
     _In_ int _CchCount2, _In_ int _CodePage) noexcept;

stl/src/ppltasks.cpp

@@ -26,6 +26,12 @@ namespace Concurrency {
 
     namespace details {
         [[noreturn]] _CRTIMP2 void __cdecl _ReportUnobservedException() {
+#if (defined(_M_IX86) || defined(_M_X64)) && !defined(_CRT_APP) && _STL_WIN32_WINNT < _WIN32_WINNT_WIN8
+            if (!IsProcessorFeaturePresent(PF_FASTFAIL_AVAILABLE)) {
+                std::abort();
+            }
+#endif // ^^^ __fastfail conditionally available ^^^
+
             __fastfail(FAST_FAIL_INVALID_ARG);
         }
 

stl/src/winapisupp.cpp

@@ -14,6 +14,10 @@ namespace {
 // Use this macro for defining the following function pointers
 #define DEFINEFUNCTIONPOINTER(fn_name) decltype(&fn_name) __KERNEL32Function_##fn_name = nullptr
 
+#if _STL_WIN32_WINNT < _WIN32_WINNT_WIN8
+    DEFINEFUNCTIONPOINTER(GetSystemTimePreciseAsFileTime);
+#endif // _STL_WIN32_WINNT < _WIN32_WINNT_WIN8
+
     DEFINEFUNCTIONPOINTER(GetTempPath2W);
 
 // Use this macro for caching a function pointer from a DLL
@@ -154,9 +158,15 @@ extern "C" _CRTIMP2 BOOL __cdecl __crtSetFileInformationByHandle(_In_ HANDLE con
 
 #if _STL_WIN32_WINNT < _WIN32_WINNT_WIN8
 
-// TRANSITION, ABI: preserved for binary compatibility
 extern "C" _CRTIMP2 void __cdecl __crtGetSystemTimePreciseAsFileTime(_Out_ LPFILETIME lpSystemTimeAsFileTime) noexcept {
-    GetSystemTimePreciseAsFileTime(lpSystemTimeAsFileTime);
+    // use GetSystemTimePreciseAsFileTime if it is available (only on Windows 8+)...
+    IFDYNAMICGETCACHEDFUNCTION(GetSystemTimePreciseAsFileTime) {
+        pfGetSystemTimePreciseAsFileTime(lpSystemTimeAsFileTime);
+        return;
+    }
+
+    // ...otherwise use GetSystemTimeAsFileTime.
+    GetSystemTimeAsFileTime(lpSystemTimeAsFileTime);
 }
 
 #endif // _STL_WIN32_WINNT < _WIN32_WINNT_WIN8
@@ -186,6 +196,10 @@ static int __cdecl initialize_pointers() noexcept {
     HINSTANCE hKernel32 = GetModuleHandleW(L"kernel32.dll");
     _Analysis_assume_(hKernel32);
 
+#if _STL_WIN32_WINNT < _WIN32_WINNT_WIN8
+    STOREFUNCTIONPOINTER(hKernel32, GetSystemTimePreciseAsFileTime);
+#endif // _STL_WIN32_WINNT < _WIN32_WINNT_WIN8
+
     // Note that GetTempPath2W is defined as of Windows 10 Build 20348 (a server release) or Windows 11,
     // but there is no "_WIN32_WINNT_WIN11" constant, so we will always dynamically load it
     STOREFUNCTIONPOINTER(hKernel32, GetTempPath2W);

stl/src/xtime.cpp

@@ -51,7 +51,7 @@ _CRTIMP2_PURE long long __cdecl _Xtime_get_ticks() noexcept {
     constexpr long long _Epoch = 0x19DB1DED53E8000LL;
 
     FILETIME ft;
-    GetSystemTimePreciseAsFileTime(&ft);
+    __crtGetSystemTimePreciseAsFileTime(&ft);
     return ((static_cast<long long>(ft.dwHighDateTime)) << 32) + static_cast<long long>(ft.dwLowDateTime) - _Epoch;
 }
 

tests/std/test.lst

@@ -523,6 +523,7 @@ tests\P1032R1_miscellaneous_constexpr
 tests\P1132R7_out_ptr
 tests\P1135R6_atomic_flag_test
 tests\P1135R6_atomic_wait
+tests\P1135R6_atomic_wait_win7
 tests\P1135R6_barrier
 tests\P1135R6_latch
 tests\P1135R6_semaphore

tests/std/tests/P1135R6_atomic_wait_win7/env.lst

@@ -0,0 +1,4 @@
+# Copyright (c) Microsoft Corporation.
+# SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+
+RUNALL_INCLUDE ..\usual_20_matrix.lst