diff --git a/ide/vs2022/mimalloc.vcxproj b/ide/vs2022/mimalloc.vcxproj
index 0a45006c..69434c21 100644
--- a/ide/vs2022/mimalloc.vcxproj
+++ b/ide/vs2022/mimalloc.vcxproj
@@ -116,7 +116,7 @@
       <SDLCheck>true</SDLCheck>
       <ConformanceMode>true</ConformanceMode>
       <AdditionalIncludeDirectories>../../include</AdditionalIncludeDirectories>
-      <PreprocessorDefinitions>MI_DEBUG=3;%(PreprocessorDefinitions);</PreprocessorDefinitions>
+      <PreprocessorDefinitions>MI_DEBUG_TRACE=1;MI_DEBUG=3;%(PreprocessorDefinitions);</PreprocessorDefinitions>
       <CompileAs>CompileAsCpp</CompileAs>
       <SupportJustMyCode>false</SupportJustMyCode>
       <LanguageStandard>stdcpp20</LanguageStandard>
diff --git a/include/mimalloc-internal.h b/include/mimalloc-internal.h
index b94bf2a3..53932b2d 100644
--- a/include/mimalloc-internal.h
+++ b/include/mimalloc-internal.h
@@ -256,6 +256,12 @@ static inline uintptr_t _mi_align_down(uintptr_t sz, size_t alignment) {
     return ((sz / alignment) * alignment);
   }
 }
+  
+// Is aligned?
+static inline bool _mi_is_aligned(void* p, size_t alignment) {
+  mi_assert_internal(alignment != 0);
+  return (((uintptr_t)p % alignment) == 0);
+}
 
 // Divide upwards: `s <= _mi_divide_up(s,d)*d < s+d`.
 static inline uintptr_t _mi_divide_up(uintptr_t size, size_t divider) {
diff --git a/src/alloc.c b/src/alloc.c
index 4bf5bd53..730f8870 100644
--- a/src/alloc.c
+++ b/src/alloc.c
@@ -124,9 +124,9 @@ mi_decl_nodiscard extern inline mi_decl_restrict void* mi_malloc_small(size_t si
 }
 
 // The main allocation function
-extern inline mi_decl_restrict void* mi_heap_malloc(mi_heap_t* heap, size_t size) mi_attr_noexcept {
-  if mi_likely(size + MI_PADDING_SIZE <= MI_SMALL_SIZE_MAX + MI_PADDING_MINSIZE) {
-    return mi_heap_malloc_small(heap, size);
+extern inline void* _mi_heap_malloc_zero(mi_heap_t* heap, size_t size, bool zero) mi_attr_noexcept {
+  if mi_likely(size + MI_PADDING_SIZE <= MI_SMALL_SIZE_MAX + MI_PADDING_MINSIZE) {  
+    return mi_heap_malloc_small_zero(heap, size, zero);
   }
   else 
   {
@@ -236,7 +236,7 @@ static bool mi_verify_padding(const mi_page_t* page, const mi_block_t* block, si
 static void mi_check_padding(const mi_page_t* page, const mi_block_t* block) {
   size_t size;
   size_t wrong;
-  if (mi_unlikely(!mi_verify_padding(page,block,&size,&wrong))) {
+  if mi_unlikely(!mi_verify_padding(page,block,&size,&wrong)) {
     _mi_show_block_trace_with_predecessor(page, block, NULL);
     _mi_error_message(EFAULT, "buffer overflow in heap block %p of size %zu: write after %zu bytes\n", block, size, wrong );
   }
@@ -716,7 +716,9 @@ void* _mi_heap_realloc_zero(mi_heap_t* heap, void* p, size_t newsize, bool zero)
       memset((uint8_t*)newp + start, 0, newsize - start);
     }
     if mi_likely(p != NULL) {
-      _mi_memcpy_aligned(newp, p, (newsize > size ? size : newsize));
+      if mi_likely(_mi_is_aligned(p, sizeof(uintptr_t))) {
+        _mi_memcpy_aligned(newp, p, (newsize > size ? size : newsize));
+      }
       mi_free(p); // only free the original pointer if successful
     }
   }
diff --git a/src/init.c b/src/init.c
index 09152562..60cdd3fe 100644
--- a/src/init.c
+++ b/src/init.c
@@ -554,6 +554,25 @@ static void mi_detect_cpu_features(void) {
 }
 #endif
 
+#if defined(_WIN32) && (MI_DEBUG_TRACE > 0)
+#include <dbghelp.h>
+static void mi_debug_init(void) {
+  if (SymInitialize(GetCurrentProcess(), NULL, TRUE) != TRUE) {  // initialize here as it is single threaded.
+    _mi_warning_message("unable to initialize debug symbol information (error 0x%x)", GetLastError());
+  }
+}
+static void mi_debug_done(void) {
+  SymCleanup(GetCurrentProcess());
+}
+#else
+static void mi_debug_init(void) {
+  // nothing
+}
+static void mi_debug_done(void) {
+  // nothing
+}
+#endif
+
 // Initialize the process; called by thread_init or the process loader
 void mi_process_init(void) mi_attr_noexcept {
   // ensure we are called once
@@ -570,6 +589,7 @@ void mi_process_init(void) mi_attr_noexcept {
   _mi_verbose_message("debug level : %d\n", MI_DEBUG);
   #endif
   _mi_verbose_message("secure level: %d\n", MI_SECURE);
+  mi_debug_init();
   mi_thread_init();
 
   #if defined(_WIN32) && !defined(MI_SHARED_LIB)
@@ -578,7 +598,7 @@ void mi_process_init(void) mi_attr_noexcept {
   // will not call _mi_thread_done on the (still executing) main thread. See issue #508.
   FlsSetValue(mi_fls_key, NULL);
   #endif
-
+ 
   mi_stats_reset();  // only call stat reset *after* thread init (or the heap tld == NULL)
 
   if (mi_option_is_enabled(mi_option_reserve_huge_os_pages)) {
@@ -624,6 +644,7 @@ static void mi_process_done(void) {
     mi_stats_print(NULL);
   }
   mi_allocator_done();  
+  mi_debug_done();
   _mi_verbose_message("process done: 0x%zx\n", _mi_heap_main.thread_id);
   os_preloading = true; // don't call the C runtime anymore
 }
diff --git a/src/options.c b/src/options.c
index 63d629eb..4cd45724 100644
--- a/src/options.c
+++ b/src/options.c
@@ -425,21 +425,14 @@ void _mi_stack_trace_print(const char* msg, void** strace, size_t len, const mi_
     info->MaxNameLen = 255;
     info->SizeOfStruct = sizeof(SYMBOL_INFO);
     HANDLE current_process = GetCurrentProcess();
-    bool initialized = (SymInitialize(current_process, NULL, TRUE) == TRUE);
     for (size_t i = 0; i < len && strace[i] != NULL; i++) {
-      if (!initialized) {
-        _mi_fprintf(NULL, NULL, "  %2zu: %8p: <unknown address>\n", i, strace[i]);
-      }
-      else if (SymFromAddr(current_process, (DWORD64)(strace[i]), 0, info)) {
+      if (SymFromAddr(current_process, (DWORD64)(strace[i]), 0, info)) {
         _mi_fprintf(NULL, NULL, "  %2zu: %8p: %s\n", i, strace[i], info->Name);
       }
       else {
         _mi_fprintf(NULL, NULL, "  %2zu: %8p: <unknown address: error: 0x%04x>\n", i, strace[i], GetLastError());
       }
-    }  
-    if (initialized) {
-      SymCleanup(current_process);
-    }
+    }      
   }
 }
 #elif (MI_DEBUG_TRACE > 0) && (defined(__linux__) || defined(__FreeBSD__) || defined(__APPLE__))