[NativeAOT] Initialize COM before initializing FLS slot.

src/coreclr/nativeaot/Runtime.Base/src/System/Runtime/__Finalizer.cs

@@ -20,10 +20,6 @@ internal static class __Finalizer
         [UnmanagedCallersOnly(EntryPoint = "ProcessFinalizers")]
         public static void ProcessFinalizers()
         {
-#if INPLACE_RUNTIME
-            System.Runtime.FinalizerInitRunner.DoInitialize();
-#endif
-
             while (true)
             {
                 // Wait until there's some work to be done. If true is returned we should finalize objects,

src/coreclr/nativeaot/Runtime/FinalizerHelpers.cpp

@@ -29,14 +29,30 @@ CLREventStatic g_FinalizerDoneEvent;
 
 static HANDLE g_lowMemoryNotification = NULL;
 
+#ifdef TARGET_WINDOWS
+static bool g_ComAndFlsInitSucceeded = false;
+#endif
+
 EXTERN_C void QCALLTYPE ProcessFinalizers();
 
-// Unmanaged front-end to the finalizer thread. We require this because at the point the GC creates the
-// finalizer thread we can't run managed code. Instead this method waits
+// Unmanaged front-end to the finalizer thread. We require this because at the point when this thread is
+// created we can't run managed code. Instead this method waits
 // for the first finalization request (by which time everything must be up and running) and kicks off the
 // managed portion of the thread at that point
 uint32_t WINAPI FinalizerStart(void* pContext)
 {
+#ifdef TARGET_WINDOWS
+    g_ComAndFlsInitSucceeded = PalInitComAndFlsSlot();
+    // handshake with EE initialization, as now we can attach Thread objects to native threads.
+    UInt32_BOOL res = PalSetEvent(g_FinalizerDoneEvent.GetOSEvent());
+    ASSERT(res);
+
+    // if FLS initialization failed do not attach the current thread and just exit instead.
+    // we are going to fail the runtime initialization.
+    if (!g_ComAndFlsInitSucceeded)
+        return 0;
+#endif // DEBUG
+
     HANDLE hFinalizerEvent = (HANDLE)pContext;
 
     PalSetCurrentThreadName(".NET Finalizer");
@@ -86,12 +102,16 @@ bool RhInitializeFinalization()
     return true;
 }
 
-void RhEnableFinalization()
+#ifdef TARGET_WINDOWS
+bool RhWaitForFinalizerThreadStart()
 {
-    g_FinalizerEvent.Set();
+    g_FinalizerDoneEvent.Wait(INFINITE,FALSE);
+    g_FinalizerDoneEvent.Reset();
+    return g_ComAndFlsInitSucceeded;
 }
+#endif
 
-EXTERN_C void QCALLTYPE RhInitializeFinalizerThread()
+void RhEnableFinalization()
 {
     g_FinalizerEvent.Set();
 }

src/coreclr/nativeaot/Runtime/GCHelpers.cpp

@@ -43,11 +43,18 @@ GPTR_DECL(MethodTable, g_pFreeObjectEEType);
 GPTR_IMPL(Thread, g_pFinalizerThread);
 
 bool RhInitializeFinalization();
+#ifdef TARGET_WINDOWS
+bool RhWaitForFinalizerThreadStart();
+#endif
 
 // Perform any runtime-startup initialization needed by the GC, HandleTable or environmental code in gcenv.ee.
 // Returns true on success or false if a subsystem failed to initialize.
 bool InitializeGC()
 {
+    // Give some headstart to the finalizer thread by launching it early.
+    if (!RhInitializeFinalization())
+        return false;
+
     // Initialize the special MethodTable used to mark free list entries in the GC heap.
     g_FreeObjectEEType.InitializeAsGcFreeType();
     g_pFreeObjectEEType = &g_FreeObjectEEType;
@@ -78,13 +85,17 @@ bool InitializeGC()
     if (FAILED(hr))
         return false;
 
-    if (!RhInitializeFinalization())
-        return false;
-
     // Initialize HandleTable.
     if (!GCHandleUtilities::GetGCHandleManager()->Initialize())
         return false;
 
+#ifdef TARGET_WINDOWS
+    // By now finalizer thread should have initialized FLS slot for thread cleanup notifications.
+    // And ensured that COM is initialized (must happen before allocating FLS slot).
+    // Make sure that this was done.
+    if (!RhWaitForFinalizerThreadStart())
+        return false;
+#endif
     return true;
 }
 

src/coreclr/nativeaot/Runtime/PalRedhawk.h

@@ -295,6 +295,7 @@ typedef uint32_t (__stdcall *BackgroundCallback)(_In_opt_ void* pCallbackContext
 REDHAWK_PALIMPORT bool REDHAWK_PALAPI PalSetCurrentThreadName(const char* name);
 #ifdef TARGET_WINDOWS
 REDHAWK_PALIMPORT bool REDHAWK_PALAPI PalSetCurrentThreadNameW(const WCHAR* name);
+REDHAWK_PALIMPORT bool REDHAWK_PALAPI PalInitComAndFlsSlot();
 #endif
 REDHAWK_PALIMPORT bool REDHAWK_PALAPI PalStartBackgroundGCThread(_In_ BackgroundCallback callback, _In_opt_ void* pCallbackContext);
 REDHAWK_PALIMPORT bool REDHAWK_PALAPI PalStartFinalizerThread(_In_ BackgroundCallback callback, _In_opt_ void* pCallbackContext);
@@ -321,7 +322,6 @@ REDHAWK_PALIMPORT uint32_t REDHAWK_PALAPI PalCompatibleWaitAny(UInt32_BOOL alert
 REDHAWK_PALIMPORT HANDLE PalCreateLowMemoryResourceNotification();
 
 REDHAWK_PALIMPORT void REDHAWK_PALAPI PalAttachThread(void* thread);
-REDHAWK_PALIMPORT bool REDHAWK_PALAPI PalDetachThread(void* thread);
 
 REDHAWK_PALIMPORT uint64_t PalGetCurrentOSThreadId();
 

src/coreclr/nativeaot/Runtime/amd64/AsmMacros.inc

@@ -329,7 +329,6 @@ TAILJMP_RAX TEXTEQU <DB 048h, 0FFh, 0E0h>
 ;;
 ;; CONSTANTS -- INTEGER
 ;;
-TSF_Attached                    equ 01h
 TSF_SuppressGcStress            equ 08h
 TSF_DoNotTriggerGc              equ 10h
 

src/coreclr/nativeaot/Runtime/arm64/AsmMacros.h

@@ -9,13 +9,11 @@
 ;;
 ;; CONSTANTS -- INTEGER
 ;;
-TSF_Attached                    equ 0x01
 TSF_SuppressGcStress            equ 0x08
 TSF_DoNotTriggerGc              equ 0x10
 TSF_SuppressGcStress__OR__TSF_DoNotTriggerGC equ 0x18
 
 ;; Bit position for the flags above, to be used with tbz/tbnz instructions
-TSF_Attached_Bit                equ 0
 TSF_SuppressGcStress_Bit        equ 3
 TSF_DoNotTriggerGc_Bit          equ 4
 

src/coreclr/nativeaot/Runtime/i386/AsmMacros.inc

@@ -110,7 +110,6 @@ endm
 ;;
 ;; CONSTANTS -- INTEGER
 ;;
-TSF_Attached                    equ 01h
 TSF_SuppressGcStress            equ 08h
 TSF_DoNotTriggerGc              equ 10h
 

src/coreclr/nativeaot/Runtime/thread.cpp

@@ -1096,7 +1096,10 @@ bool Thread::IsDetached()
 void Thread::SetDetached()
 {
     ASSERT(!IsStateSet(TSF_Detached));
+    ASSERT(IsStateSet(TSF_Attached));
+
     SetState(TSF_Detached);
+    ClearState(TSF_Attached);
 }
 
 bool Thread::IsActivationPending()

src/coreclr/nativeaot/Runtime/thread.h

@@ -189,7 +189,9 @@ class Thread : private RuntimeThreadLocals
     {
         TSF_Unknown             = 0x00000000,       // Threads are created in this state
         TSF_Attached            = 0x00000001,       // Thread was inited by first U->M transition on this thread
-        TSF_Detached            = 0x00000002,       // Thread was detached by DllMain
+                                                    // ...Prior to setting this bit the state is TSF_Unknown.
+        TSF_Detached            = 0x00000002,       // Thread was detached and no longer can run managed code.
+                                                    // ...TSF_Attached is cleared when TSF_Detached is set.
         TSF_SuppressGcStress    = 0x00000008,       // Do not allow gc stress on this thread, used in DllMain
                                                     // ...and on the Finalizer thread
         TSF_DoNotTriggerGc      = 0x00000010,       // Do not allow hijacking of this thread, also intended to

src/coreclr/nativeaot/Runtime/threadstore.cpp

@@ -110,9 +110,16 @@ void ThreadStore::AttachCurrentThread(bool fAcquireThreadStoreLock)
     // we want to avoid at construction time because the loader lock is held then.
     Thread * pAttachingThread = RawGetCurrentThread();
 
-    // The thread was already initialized, so it is already attached
+    if (pAttachingThread->IsDetached())
+    {
+        ASSERT_UNCONDITIONALLY("Attempt to execute managed code after the .NET runtime thread state has been destroyed.");
+        RhFailFast();
+    }
+
+    // The thread was already initialized, so it is already attached.
     if (pAttachingThread->IsInitialized())
     {
+        ASSERT((pAttachingThread->m_ThreadStateFlags & Thread::TSF_Attached) != 0);
         return;
     }
 
@@ -156,12 +163,8 @@ void ThreadStore::DetachCurrentThread()
         return;
     }
 
-    // Unregister from OS notifications
-    // This can return false if a thread did not register for OS notification.
-    if (!PalDetachThread(pDetachingThread))
-    {
-        return;
-    }
+    // detach callback should not call us twice
+    ASSERT(!pDetachingThread->IsDetached());
 
     // Run pre-mortem callbacks while we still can run managed code and not holding locks.
     // NOTE: background GC threads are attached/suspendable threads, but should not run ordinary

src/coreclr/nativeaot/Runtime/unix/PalRedhawkUnix.cpp

@@ -517,17 +517,6 @@ extern "C" void PalAttachThread(void* thread)
     UnmaskActivationSignal();
 }
 
-// Detach thread from OS notifications.
-// Parameters:
-//  thread        - thread to detach
-// Return:
-//  true if the thread was detached, false if there was no attached thread
-extern "C" bool PalDetachThread(void* thread)
-{
-    UNREFERENCED_PARAMETER(thread);
-    return true;
-}
-
 #if !defined(USE_PORTABLE_HELPERS) && !defined(FEATURE_RX_THUNKS)
 
 REDHAWK_PALEXPORT UInt32_BOOL REDHAWK_PALAPI PalAllocateThunksFromTemplate(HANDLE hTemplateModule, uint32_t templateRva, size_t templateSize, void** newThunksOut)

src/coreclr/nativeaot/Runtime/windows/PalRedhawkMinWin.cpp

@@ -53,13 +53,47 @@ static uint32_t g_flsIndex = FLS_OUT_OF_INDEXES;
 void __stdcall FiberDetachCallback(void* lpFlsData)
 {
     ASSERT(g_flsIndex != FLS_OUT_OF_INDEXES);
+    ASSERT(g_flsIndex != NULL);
     ASSERT(lpFlsData == FlsGetValue(g_flsIndex));
 
-    if (lpFlsData != NULL)
+    // The current fiber is the home fiber of a thread, so the thread is shutting down
+    RuntimeThreadShutdown(lpFlsData);
+}
+
+REDHAWK_PALEXPORT bool REDHAWK_PALAPI PalInitComAndFlsSlot()
+{
+    ASSERT(g_flsIndex == FLS_OUT_OF_INDEXES);
+
+    // Making finalizer thread MTA early ensures that COM is initialized before we initialize our thread
+    // termination callback.
+    HRESULT hr = CoInitializeEx(NULL, COINIT_MULTITHREADED);
+    if (FAILED(hr))
+        return false;
+
+    // We use fiber detach callbacks to run our thread shutdown code because the fiber detach
+    // callback is made without the OS loader lock
+    g_flsIndex = FlsAlloc(FiberDetachCallback);
+    return g_flsIndex != FLS_OUT_OF_INDEXES;
+}
+
+// Register the thread with OS to be notified when thread is about to be destroyed
+// It fails fast if a different thread was already registered with the current fiber.
+// Parameters:
+//  thread        - thread to attach
+REDHAWK_PALEXPORT void REDHAWK_PALAPI PalAttachThread(void* thread)
+{
+    void* threadFromCurrentFiber = FlsGetValue(g_flsIndex);
+
+    if (threadFromCurrentFiber != NULL)
     {
-        // The current fiber is the home fiber of a thread, so the thread is shutting down
-        RuntimeThreadShutdown(lpFlsData);
+        ASSERT_UNCONDITIONALLY("Multiple threads encountered from a single fiber");
+        RhFailFast();
     }
+
+    // Associate the current fiber with the current thread.  This makes the current fiber the thread's "home"
+    // fiber.  This fiber is the only fiber allowed to execute managed code on this thread.  When this fiber
+    // is destroyed, we consider the thread to be destroyed.
+    FlsSetValue(g_flsIndex, thread);
 }
 
 static HMODULE LoadKernel32dll()
@@ -159,14 +193,6 @@ void InitializeCurrentProcessCpuCount()
 // initialization and false on failure.
 REDHAWK_PALEXPORT bool REDHAWK_PALAPI PalInit()
 {
-    // We use fiber detach callbacks to run our thread shutdown code because the fiber detach
-    // callback is made without the OS loader lock
-    g_flsIndex = FlsAlloc(FiberDetachCallback);
-    if (g_flsIndex == FLS_OUT_OF_INDEXES)
-    {
-        return false;
-    }
-
     GCConfig::Initialize();
 
     if (!GCToOSInterface::Initialize())
@@ -179,54 +205,6 @@ REDHAWK_PALEXPORT bool REDHAWK_PALAPI PalInit()
     return true;
 }
 
-// Register the thread with OS to be notified when thread is about to be destroyed
-// It fails fast if a different thread was already registered with the current fiber.
-// Parameters:
-//  thread        - thread to attach
-REDHAWK_PALEXPORT void REDHAWK_PALAPI PalAttachThread(void* thread)
-{
-    void* threadFromCurrentFiber = FlsGetValue(g_flsIndex);
-
-    if (threadFromCurrentFiber != NULL)
-    {
-        ASSERT_UNCONDITIONALLY("Multiple threads encountered from a single fiber");
-        RhFailFast();
-    }
-
-    // Associate the current fiber with the current thread.  This makes the current fiber the thread's "home"
-    // fiber.  This fiber is the only fiber allowed to execute managed code on this thread.  When this fiber
-    // is destroyed, we consider the thread to be destroyed.
-    FlsSetValue(g_flsIndex, thread);
-}
-
-// Detach thread from OS notifications.
-// It fails fast if some other thread value was attached to the current fiber.
-// Parameters:
-//  thread        - thread to detach
-// Return:
-//  true if the thread was detached, false if there was no attached thread
-REDHAWK_PALEXPORT bool REDHAWK_PALAPI PalDetachThread(void* thread)
-{
-    ASSERT(g_flsIndex != FLS_OUT_OF_INDEXES);
-    void* threadFromCurrentFiber = FlsGetValue(g_flsIndex);
-
-    if (threadFromCurrentFiber == NULL)
-    {
-        // we've seen this thread, but not this fiber.  It must be a "foreign" fiber that was
-        // borrowing this thread.
-        return false;
-    }
-
-    if (threadFromCurrentFiber != thread)
-    {
-        ASSERT_UNCONDITIONALLY("Detaching a thread from the wrong fiber");
-        RhFailFast();
-    }
-
-    FlsSetValue(g_flsIndex, NULL);
-    return true;
-}
-
 extern "C" uint64_t PalQueryPerformanceCounter()
 {
     return GCToOSInterface::QueryPerformanceCounter();

src/coreclr/nativeaot/System.Private.CoreLib/src/System.Private.CoreLib.csproj

@@ -201,7 +201,6 @@
     <Compile Include="System\Runtime\GCSettings.NativeAot.cs" />
     <Compile Include="System\Runtime\TypeLoaderExports.cs" />
     <Compile Include="System\Runtime\ThunkPool.cs" />
-    <Compile Include="System\Runtime\InitializeFinalizerThread.cs" />
     <Compile Include="System\Runtime\InteropServices\ComEventsHelper.NativeAot.cs" Condition="'$(FeatureCominterop)' == 'true'" />
     <Compile Include="System\Runtime\InteropServices\ComWrappers.NativeAot.cs" Condition="'$(FeatureComWrappers)' == 'true'" />
     <Compile Include="System\Runtime\InteropServices\GCHandle.NativeAot.cs" />

src/coreclr/nativeaot/System.Private.CoreLib/src/System/Runtime/RuntimeImports.cs

@@ -94,9 +94,6 @@ internal static void RhWaitForPendingFinalizers(bool allowReentrantWait)
             RhWaitForPendingFinalizers(allowReentrantWait ? 1 : 0);
         }
 
-        [LibraryImport(RuntimeLibrary)]
-        internal static partial void RhInitializeFinalizerThread();
-
         // Get maximum GC generation number.
         [MethodImplAttribute(MethodImplOptions.InternalCall)]
         [RuntimeImport(RuntimeLibrary, "RhGetMaxGcGeneration")]

src/coreclr/nativeaot/System.Private.CoreLib/src/System/Threading/Thread.NativeAot.Unix.cs

@@ -137,10 +137,6 @@ private static bool SetApartmentStateUnchecked(ApartmentState state, bool throwO
 
         partial void InitializeComOnNewThread();
 
-        internal static void InitializeComForFinalizerThread()
-        {
-        }
-
         public void DisableComObjectEagerCleanup() { }
 
         public void Interrupt() => WaitSubsystem.Interrupt(this);