diff --git a/ide/vs2022/mimalloc-test-stress.vcxproj b/ide/vs2022/mimalloc-test-stress.vcxproj
index d6af71ce..128a4ff6 100644
--- a/ide/vs2022/mimalloc-test-stress.vcxproj
+++ b/ide/vs2022/mimalloc-test-stress.vcxproj
@@ -282,8 +282,8 @@
     </ClCompile>
   </ItemGroup>
   <ItemGroup>
-    <ProjectReference Include="mimalloc-lib.vcxproj">
-      <Project>{abb5eae7-b3e6-432e-b636-333449892ea6}</Project>
+    <ProjectReference Include="mimalloc-override-dll.vcxproj">
+      <Project>{abb5eae7-b3e6-432e-b636-333449892ea7}</Project>
     </ProjectReference>
   </ItemGroup>
   <Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
diff --git a/include/mimalloc/prim.h b/include/mimalloc/prim.h
index 03ebe7dd..011a330b 100644
--- a/include/mimalloc/prim.h
+++ b/include/mimalloc/prim.h
@@ -120,7 +120,6 @@ void _mi_prim_thread_associate_default_heap(mi_heap_t* heap);
 // Is this thread part of a thread pool?
 bool _mi_prim_thread_is_in_threadpool(void);
 
-#define MI_WIN_USE_FIXED_TLS  1
 
 //-------------------------------------------------------------------
 // Access to TLS (thread local storage) slots.
@@ -212,7 +211,7 @@ static inline void mi_prim_tls_slot_set(size_t slot, void* value) mi_attr_noexce
 // thread-local initialization checks in the fast path.
 // We allocate a user TLS slot at process initialization (see `windows/prim.c`)
 // and store the offset `_mi_win_tls_offset`.
-#define MI_HAS_TLS_SLOT      2              // 2 = we can reliably initialize the slot (saving a test on each malloc)
+#define MI_HAS_TLS_SLOT  1              // 2 = we can reliably initialize the slot (saving a test on each malloc)
 
 extern mi_decl_hidden size_t _mi_win_tls_offset;
 
diff --git a/src/init.c b/src/init.c
index 77630a50..3b143a37 100644
--- a/src/init.c
+++ b/src/init.c
@@ -372,7 +372,6 @@ mi_tld_t* _mi_thread_tld(void) mi_attr_noexcept {
   }
 }
 
-
 /* -----------------------------------------------------------
   Sub process
 ----------------------------------------------------------- */
diff --git a/src/prim/windows/prim.c b/src/prim/windows/prim.c
index 773d80c4..3c3e1175 100644
--- a/src/prim/windows/prim.c
+++ b/src/prim/windows/prim.c
@@ -650,18 +650,23 @@ bool _mi_prim_random_buf(void* buf, size_t buf_len) {
 mi_decl_cache_align size_t _mi_win_tls_offset = 0;  
 #endif
 
+//static void mi_debug_out(const char* s) {
+//  HANDLE h = GetStdHandle(STD_ERROR_HANDLE);
+//  WriteConsole(h, s, (DWORD)_mi_strlen(s), NULL, NULL);
+//}
+
 static void mi_win_tls_init(DWORD reason) {
-  #if MI_HAS_TLS_SLOT >= 2  // we must initialize the TLS slot before any allocation
-  #if MI_WIN_USE_FIXED_TLS==1
-  if (reason==DLL_PROCESS_ATTACH && _mi_win_tls_offset == 0) {
-    const DWORD tls_slot = TlsAlloc();  // usually returns slot 1
-    if (tls_slot == TLS_OUT_OF_INDEXES) {
-      _mi_error_message(EFAULT, "unable to allocate the a TLS slot (rebuild without MI_WIN_USE_FIXED_TLS?)\n");
-    }
-    _mi_win_tls_offset = (size_t)tls_slot * sizeof(void*);
-  }
-  #endif
   if (reason==DLL_PROCESS_ATTACH || reason==DLL_THREAD_ATTACH) {
+    #if MI_WIN_USE_FIXED_TLS==1  // we must allocate a TLS slot dynamically
+    if (_mi_win_tls_offset == 0 && reason=DLL_PROCESS_ATTACH) {      
+      const DWORD tls_slot = TlsAlloc();  // usually returns slot 1
+      if (tls_slot == TLS_OUT_OF_INDEXES) {
+        _mi_error_message(EFAULT, "unable to allocate the a TLS slot (rebuild without MI_WIN_USE_FIXED_TLS?)\n");
+      }
+      _mi_win_tls_offset = (size_t)tls_slot * sizeof(void*);
+    }
+    #endif
+    #if MI_HAS_TLS_SLOT >= 2  // we must initialize the TLS slot before any allocation
     if (mi_prim_get_default_heap() == NULL) {
       _mi_heap_set_default_direct((mi_heap_t*)&_mi_heap_empty);  
       #if MI_DEBUG && MI_WIN_USE_FIXED_TLS==1
@@ -669,10 +674,8 @@ static void mi_win_tls_init(DWORD reason) {
       mi_assert_internal(p == (void*)&_mi_heap_empty);
       #endif  
     }
-  }
-  #else
-  MI_UNUSED(reason);
-  #endif
+    #endif  
+  }  
 }
 
 static void NTAPI mi_win_main(PVOID module, DWORD reason, LPVOID reserved) {
@@ -695,7 +698,7 @@ static void NTAPI mi_win_main(PVOID module, DWORD reason, LPVOID reserved) {
   #define MI_PRIM_HAS_PROCESS_ATTACH  1
 
   // Windows DLL: easy to hook into process_init and thread_done
-  BOOL WINAPI DllMain(HINSTANCE inst, DWORD reason, LPVOID reserved) {
+  BOOL WINAPI DllMain(HINSTANCE inst, DWORD reason, LPVOID reserved) {    
     mi_win_main((PVOID)inst,reason,reserved);
     return TRUE;
   }