diff --git a/CMakeLists.txt b/CMakeLists.txt
index 299aec1b..083abce8 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -190,14 +190,6 @@ if (MSVC AND MSVC_VERSION GREATER_EQUAL 1914)
   list(APPEND mi_cflags /Zc:__cplusplus)
 endif()
 
-# Architecture flags
-if(${CMAKE_HOST_SYSTEM_PROCESSOR} MATCHES "arm" AND NOT APPLE)
-    check_cxx_compiler_flag(-march=native CXX_SUPPORTS_MARCH_NATIVE)
-    if (CXX_SUPPORTS_MARCH_NATIVE)
-        list(APPEND mi_cflags -march=native)
-    endif()
-endif()
-
 # extra needed libraries
 if(WIN32)
   list(APPEND mi_libraries psapi shell32 user32 advapi32 bcrypt)
diff --git a/include/mimalloc-atomic.h b/include/mimalloc-atomic.h
index 2d725a25..db885319 100644
--- a/include/mimalloc-atomic.h
+++ b/include/mimalloc-atomic.h
@@ -118,7 +118,7 @@ static inline void mi_atomic_maxi64_relaxed(volatile int64_t* p, int64_t x) {
 
 // MSVC C compilation wrapper that uses Interlocked operations to model C11 atomics.
 #define WIN32_LEAN_AND_MEAN
-#include <Windows.h>
+#include <windows.h>
 #include <intrin.h>
 #ifdef _WIN64
 typedef LONG64   msc_intptr_t;
diff --git a/src/alloc.c b/src/alloc.c
index 1d973ca4..2de26b78 100644
--- a/src/alloc.c
+++ b/src/alloc.c
@@ -726,7 +726,7 @@ mi_decl_restrict char* mi_strndup(const char* s, size_t n) mi_attr_noexcept {
 #ifndef PATH_MAX
 #define PATH_MAX MAX_PATH
 #endif
-#include <Windows.h>
+#include <windows.h>
 mi_decl_restrict char* mi_heap_realpath(mi_heap_t* heap, const char* fname, char* resolved_name) mi_attr_noexcept {
   // todo: use GetFullPathNameW to allow longer file names
   char buf[PATH_MAX];
diff --git a/src/arena.c b/src/arena.c
index ea3a1abd..fae4ecaa 100644
--- a/src/arena.c
+++ b/src/arena.c
@@ -211,7 +211,10 @@ void* _mi_arena_alloc_aligned(size_t size, size_t alignment, bool* commit, bool*
   }
 
   // finally, fall back to the OS
-  if (mi_option_is_enabled(mi_option_limit_os_alloc)) return NULL;
+  if (mi_option_is_enabled(mi_option_limit_os_alloc)) {
+    errno = ENOMEM;
+    return NULL;
+  }
   *is_zero = true;
   *memid   = MI_MEMID_OS;  
   void* p = _mi_os_alloc_aligned(size, alignment, *commit, large, tld->stats);
diff --git a/src/bitmap.c b/src/bitmap.c
index ea55d33e..fa1b16c3 100644
--- a/src/bitmap.c
+++ b/src/bitmap.c
@@ -371,13 +371,13 @@ static bool mi_bitmap_is_claimedx_across(mi_bitmap_t bitmap, size_t bitmap_field
   if ((prev & pre_mask) != 0) any_ones = true;
   while (mid_count-- > 0) {
     prev = mi_atomic_load_relaxed(field++);
-    if ((prev & pre_mask) != pre_mask) all_ones = false;
-    if ((prev & pre_mask) != 0) any_ones = true;
+    if ((prev & mid_mask) != mid_mask) all_ones = false;
+    if ((prev & mid_mask) != 0) any_ones = true;
   }
   if (post_mask!=0) {
     prev = mi_atomic_load_relaxed(field);
-    if ((prev & pre_mask) != pre_mask) all_ones = false;
-    if ((prev & pre_mask) != 0) any_ones = true;
+    if ((prev & post_mask) != post_mask) all_ones = false;
+    if ((prev & post_mask) != 0) any_ones = true;
   }  
   if (pany_ones != NULL) *pany_ones = any_ones;
   return all_ones;
diff --git a/src/init.c b/src/init.c
index 2161fdd8..aee08f5a 100644
--- a/src/init.c
+++ b/src/init.c
@@ -309,7 +309,7 @@ static void _mi_thread_done(mi_heap_t* default_heap);
   // nothing to do as it is done in DllMain
 #elif defined(_WIN32) && !defined(MI_SHARED_LIB)
   // use thread local storage keys to detect thread ending
-  #include <Windows.h>
+  #include <windows.h>
   #include <fibersapi.h>
   #if (_WIN32_WINNT < 0x600)  // before Windows Vista 
   WINBASEAPI DWORD WINAPI FlsAlloc( _In_opt_ PFLS_CALLBACK_FUNCTION lpCallback );
diff --git a/src/options.c b/src/options.c
index 0560b937..baa387ad 100644
--- a/src/options.c
+++ b/src/options.c
@@ -425,7 +425,7 @@ static inline int mi_strnicmp(const char* s, const char* t, size_t n) {
 // reliably even when this is invoked before the C runtime is initialized.
 // i.e. when `_mi_preloading() == true`.
 // Note: on windows, environment names are not case sensitive.
-#include <Windows.h>
+#include <windows.h>
 static bool mi_getenv(const char* name, char* result, size_t result_size) {
   result[0] = 0;
   size_t len = GetEnvironmentVariableA(name, result, (DWORD)result_size);
diff --git a/src/os.c b/src/os.c
index 7f94dd5a..21c9d638 100644
--- a/src/os.c
+++ b/src/os.c
@@ -28,7 +28,7 @@ terms of the MIT license. A copy of the license can be found in the file
 
 
 #if defined(_WIN32)
-#include <Windows.h>
+#include <windows.h>
 #elif defined(__wasi__)
 // stdlib.h is all we need, and has already been included in mimalloc.h
 #else
diff --git a/src/random.c b/src/random.c
index b6067ecf..113ba0fd 100644
--- a/src/random.c
+++ b/src/random.c
@@ -244,7 +244,7 @@ static bool os_random_buf(void* buf, size_t buf_len) {
 #endif
 
 #if defined(_WIN32)
-#include <Windows.h>
+#include <windows.h>
 #elif defined(__APPLE__)
 #include <mach/mach_time.h>
 #else
diff --git a/src/stats.c b/src/stats.c
index 10be20b0..ab21ec79 100644
--- a/src/stats.c
+++ b/src/stats.c
@@ -393,7 +393,7 @@ void mi_thread_stats_print_out(mi_output_fun* out, void* arg) mi_attr_noexcept {
 // Basic timer for convenience; use milli-seconds to avoid doubles
 // ----------------------------------------------------------------
 #ifdef _WIN32
-#include <Windows.h>
+#include <windows.h>
 static mi_msecs_t mi_to_msecs(LARGE_INTEGER t) {
   static LARGE_INTEGER mfreq; // = 0
   if (mfreq.QuadPart == 0LL) {
@@ -448,7 +448,7 @@ mi_msecs_t _mi_clock_end(mi_msecs_t start) {
 // --------------------------------------------------------
 
 #if defined(_WIN32)
-#include <Windows.h>
+#include <windows.h>
 #include <psapi.h>
 #pragma comment(lib,"psapi.lib")