diff --git a/CMakeLists.txt b/CMakeLists.txt
index 7c0f67af..a6c95dc4 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -19,6 +19,7 @@ option(MI_OSX_INTERPOSE     "Use interpose to override standard malloc on macOS"
 option(MI_OSX_ZONE          "Use malloc zone to override standard malloc on macOS" ON)
 option(MI_WIN_REDIRECT      "Use redirection module ('mimalloc-redirect') on Windows if compiling mimalloc as a DLL" ON)
 option(MI_LOCAL_DYNAMIC_TLS "Use slightly slower, dlopen-compatible TLS mechanism (Unix)" OFF)
+option(MI_LIBC_MUSL         "Set this when linking with musl libc" OFF)
 option(MI_BUILD_SHARED      "Build shared library" ON)
 option(MI_BUILD_STATIC      "Build static library" ON)
 option(MI_BUILD_OBJECT      "Build object library" ON)
@@ -286,6 +287,12 @@ if(CMAKE_SYSTEM_NAME MATCHES "Linux|Android")
   endif()
 endif()
 
+if(MI_LIBC_MUSL)
+  message(STATUS "Assume using musl libc (MI_LIBC_MUSL=ON) (this implies MI_LOCAL_DYNAMIC_TLS=ON)")
+  set(MI_LOCAL_DYNAMIC_TLS "ON")
+  list(APPEND mi_defines MI_LIBC_MUSL=1)
+endif()
+
 # On Haiku use `-DCMAKE_INSTALL_PREFIX` instead, issue #788
 # if(CMAKE_SYSTEM_NAME MATCHES "Haiku")
 #   SET(CMAKE_INSTALL_LIBDIR ~/config/non-packaged/lib)
diff --git a/include/mimalloc/prim.h b/include/mimalloc/prim.h
index d14b885b..ebb31df2 100644
--- a/include/mimalloc/prim.h
+++ b/include/mimalloc/prim.h
@@ -203,12 +203,23 @@ static inline void mi_prim_tls_slot_set(size_t slot, void* value) mi_attr_noexce
 
 #endif
 
+// Do we have __builtin_thread_pointer? (do not make this a compound test as it fails on older gcc's, see issue #851)
+#if defined(__has_builtin)
+#if __has_builtin(__builtin_thread_pointer)
+#define MI_HAS_BUILTIN_THREAD_POINTER  1
+#endif
+#elif defined(__GNUC__) && (__GNUC__ >= 7) && defined(__aarch64__)  // special case aarch64 for older gcc versions (issue #851)
+#define MI_HAS_BUILTIN_THREAD_POINTER  1
+#endif
+
+
 // defined in `init.c`; do not use these directly
 extern mi_decl_thread mi_heap_t* _mi_heap_default;  // default heap to allocate from
 extern bool _mi_process_is_initialized;             // has mi_process_init been called?
 
 static inline mi_threadid_t _mi_prim_thread_id(void) mi_attr_noexcept;
 
+// Get a unique id for the current thread.
 #if defined(_WIN32)
 
 #define WIN32_LEAN_AND_MEAN
@@ -218,7 +229,7 @@ static inline mi_threadid_t _mi_prim_thread_id(void) mi_attr_noexcept {
   return (uintptr_t)NtCurrentTeb();
 }
 
-#elif defined(__has_builtin) && __has_builtin(__builtin_thread_pointer) && \
+#elif MI_HAS_BUILTIN_THREAD_POINTER && \
       (!defined(__APPLE__)) && /* on apple (M1) the wrong register is read (tpidr_el0 instead of tpidrro_el0) so fall back to TLS slot assembly (<https://github.com/microsoft/mimalloc/issues/343#issuecomment-763272369>)*/ \
       (!defined(__clang_major__) || __clang_major__ >= 14)  // older clang versions emit bad code; fall back to using the TLS slot (<https://lore.kernel.org/linux-arm-kernel/202110280952.352F66D8@keescook/T/>)
 
diff --git a/include/mimalloc/types.h b/include/mimalloc/types.h
index ab9ee160..97438569 100644
--- a/include/mimalloc/types.h
+++ b/include/mimalloc/types.h
@@ -161,17 +161,24 @@ typedef int32_t  mi_ssize_t;
 
 // Main tuning parameters for segment and page sizes
 // Sizes for 64-bit (usually divide by two for 32-bit)
+#ifndef MI_SEGMENT_SLICE_SHIFT
 #define MI_SEGMENT_SLICE_SHIFT            (13 + MI_INTPTR_SHIFT)         // 64KiB  (32KiB on 32-bit)
+#endif
 
+#ifndef MI_SEGMENT_SHIFT
 #if MI_INTPTR_SIZE > 4
 #define MI_SEGMENT_SHIFT                  ( 9 + MI_SEGMENT_SLICE_SHIFT)  // 32MiB
 #else
 #define MI_SEGMENT_SHIFT                  ( 7 + MI_SEGMENT_SLICE_SHIFT)  // 4MiB on 32-bit
 #endif
+#endif
 
+#ifndef MI_SMALL_PAGE_SHIFT
 #define MI_SMALL_PAGE_SHIFT               (MI_SEGMENT_SLICE_SHIFT)       // 64KiB
+#endif
+#ifndef MI_MEDIUM_PAGE_SHIFT
 #define MI_MEDIUM_PAGE_SHIFT              ( 3 + MI_SMALL_PAGE_SHIFT)     // 512KiB
-
+#endif
 
 // Derived constants
 #define MI_SEGMENT_SIZE                   (MI_ZU(1)<<MI_SEGMENT_SHIFT)
diff --git a/src/prim/unix/prim.c b/src/prim/unix/prim.c
index c4f1f58a..8d405363 100644
--- a/src/prim/unix/prim.c
+++ b/src/prim/unix/prim.c
@@ -148,13 +148,20 @@ void _mi_prim_mem_init( mi_os_mem_config_t* config )
   config->has_virtual_reserve = true; // todo: check if this true for NetBSD?  (for anonymous mmap with PROT_NONE)
 
   // disable transparent huge pages for this process?
-  #if defined(MI_NO_THP) && (defined(__linux__) || defined(__ANDROID__))
-  int val = 0;
-  if (prctl(PR_GET_THP_DISABLE, &val, 0, 0, 0) != 0) {
-    // Most likely since distros often come with always/madvise settings.
-    val = 1;
-    // Disabling only for mimalloc process rather than touching system wide settings
-    (void)prctl(PR_SET_THP_DISABLE, &val, 0, 0, 0);
+  #if (defined(__linux__) || defined(__ANDROID__)) && defined(PR_GET_THP_DISABLE)
+  #if defined(MI_NO_THP)
+  if (true)
+  #else
+  if (!mi_option_is_enabled(mi_option_allow_large_os_pages)) // disable THP also if large OS pages are not allowed in the options
+  #endif
+  {
+    int val = 0;
+    if (prctl(PR_GET_THP_DISABLE, &val, 0, 0, 0) != 0) {
+      // Most likely since distros often come with always/madvise settings.
+      val = 1;
+      // Disabling only for mimalloc process rather than touching system wide settings
+      (void)prctl(PR_SET_THP_DISABLE, &val, 0, 0, 0);
+    }
   }
   #endif
 }