improved malloc zone handling on macOSX (not working yet)

include/mimalloc-internal.h

@@ -298,7 +298,7 @@ mi_heap_t*  _mi_heap_main_get(void);    // statically allocated main backing hea
 #endif
 
 #if defined(MI_TLS_SLOT)
-static inline void* mi_tls_slot(size_t slot);   // forward declaration
+static inline void* mi_tls_slot(size_t slot) mi_attr_noexcept;   // forward declaration
 #elif defined(MI_TLS_PTHREAD_SLOT_OFS)
 #include <pthread.h>
 static inline mi_heap_t** mi_tls_pthread_heap_slot(void) {

src/alloc-override-osx.c

@@ -14,6 +14,7 @@ terms of the MIT license. A copy of the license can be found in the file
 #error "this file should only be included on macOS"
 #endif
 
+#warning "malloc zones do not seem to work for now; use MI_INTERPOSE instead"
 /* ------------------------------------------------------
    Override system malloc on macOS
    This is done through the malloc zone interface.
@@ -35,34 +36,42 @@ extern malloc_zone_t* malloc_default_purgeable_zone(void) __attribute__((weak_im
 ------------------------------------------------------ */
 
 static size_t zone_size(malloc_zone_t* zone, const void* p) {
+  UNUSED(zone); UNUSED(p);
   return 0; // as we cannot guarantee that `p` comes from us, just return 0
 }
 
 static void* zone_malloc(malloc_zone_t* zone, size_t size) {
+  UNUSED(zone);
   return mi_malloc(size);
 }
 
 static void* zone_calloc(malloc_zone_t* zone, size_t count, size_t size) {
+  UNUSED(zone);
   return mi_calloc(count, size);
 }
 
 static void* zone_valloc(malloc_zone_t* zone, size_t size) {
+  UNUSED(zone);
   return mi_malloc_aligned(size, _mi_os_page_size());
 }
 
 static void zone_free(malloc_zone_t* zone, void* p) {
+  UNUSED(zone);
   return mi_free(p);
 }
 
 static void* zone_realloc(malloc_zone_t* zone, void* p, size_t newsize) {
+  UNUSED(zone);
   return mi_realloc(p, newsize);
 }
 
 static void* zone_memalign(malloc_zone_t* zone, size_t alignment, size_t size) {
+  UNUSED(zone);
   return mi_malloc_aligned(size,alignment);
 }
 
 static void zone_destroy(malloc_zone_t* zone) {
+  UNUSED(zone);
   // todo: ignore for now?
 }
 
@@ -83,11 +92,13 @@ static void zone_batch_free(malloc_zone_t* zone, void** ps, unsigned count) {
 }
 
 static size_t zone_pressure_relief(malloc_zone_t* zone, size_t size) {
+  UNUSED(zone); UNUSED(size);
   mi_collect(false);
   return 0;
 }
 
 static void zone_free_definite_size(malloc_zone_t* zone, void* p, size_t size) {
+  UNUSED(size);
   zone_free(zone,p);
 }
 
@@ -102,34 +113,43 @@ static kern_return_t intro_enumerator(task_t task, void* p,
                             vm_range_recorder_t recorder)
 {
   // todo: enumerate all memory
+  UNUSED(task); UNUSED(p); UNUSED(type_mask); UNUSED(zone_address);
+  UNUSED(reader); UNUSED(recorder);
   return KERN_SUCCESS;
 }
 
 static size_t intro_good_size(malloc_zone_t* zone, size_t size) {
+  UNUSED(zone);
   return mi_good_size(size);
 }
 
 static boolean_t intro_check(malloc_zone_t* zone) {
+  UNUSED(zone);
   return true;
 }
 
 static void intro_print(malloc_zone_t* zone, boolean_t verbose) {
+  UNUSED(zone); UNUSED(verbose);
   mi_stats_print(NULL);
 }
 
 static void intro_log(malloc_zone_t* zone, void* p) {
+  UNUSED(zone); UNUSED(p);
   // todo?
 }
 
 static void intro_force_lock(malloc_zone_t* zone) {
+  UNUSED(zone);
   // todo?
 }
 
 static void intro_force_unlock(malloc_zone_t* zone) {
+  UNUSED(zone);
   // todo?
 }
 
 static void intro_statistics(malloc_zone_t* zone, malloc_statistics_t* stats) {
+  UNUSED(zone);
   // todo...
   stats->blocks_in_use = 0;
   stats->size_in_use = 0;
@@ -138,6 +158,7 @@ static void intro_statistics(malloc_zone_t* zone, malloc_statistics_t* stats) {
 }
 
 static boolean_t intro_zone_locked(malloc_zone_t* zone) {
+  UNUSED(zone);
   return false;
 }
 
@@ -161,7 +182,6 @@ static malloc_zone_t* mi_get_default_zone()
   }
 }
 
-
 static void __attribute__((constructor)) _mi_macos_override_malloc()
 {
   static malloc_introspection_t intro;
@@ -201,6 +221,7 @@ static void __attribute__((constructor)) _mi_macos_override_malloc()
   zone.free_definite_size = &zone_free_definite_size;
   zone.pressure_relief = &zone_pressure_relief;
   intro.zone_locked = &intro_zone_locked;
+  intro.statistics = &intro_statistics;
 
   // force the purgeable zone to exist to avoid strange bugs
   if (malloc_default_purgeable_zone) {
@@ -225,6 +246,7 @@ static void __attribute__((constructor)) _mi_macos_override_malloc()
     malloc_zone_unregister(purgeable_zone);
     malloc_zone_register(purgeable_zone);
   }
+
 }
 
 #endif // MI_MALLOC_OVERRIDE

src/alloc-override.c

@@ -13,7 +13,7 @@ terms of the MIT license. A copy of the license can be found in the file
 #error "It is only possible to override "malloc" on Windows when building as a DLL (and linking the C runtime as a DLL)"
 #endif
 
-#if defined(MI_MALLOC_OVERRIDE) && !defined(_WIN32)
+#if defined(MI_MALLOC_OVERRIDE) && !(defined(_WIN32) || (defined(__MACH__) && !defined(MI_INTERPOSE)))
 
 // ------------------------------------------------------
 // Override system malloc
@@ -68,10 +68,10 @@ terms of the MIT license. A copy of the license can be found in the file
   // we just override new/delete which does work in a static library.
 #else
   // On all other systems forward to our API
-  void* malloc(size_t size)              mi_attr_noexcept  MI_FORWARD1(mi_malloc, size);
-  void* calloc(size_t size, size_t n)    mi_attr_noexcept  MI_FORWARD2(mi_calloc, size, n);
-  void* realloc(void* p, size_t newsize) mi_attr_noexcept  MI_FORWARD2(mi_realloc, p, newsize);
-  void  free(void* p)                    mi_attr_noexcept  MI_FORWARD0(mi_free, p);
+  void* malloc(size_t size)              MI_FORWARD1(mi_malloc, size);
+  void* calloc(size_t size, size_t n)    MI_FORWARD2(mi_calloc, size, n);
+  void* realloc(void* p, size_t newsize) MI_FORWARD2(mi_realloc, p, newsize);
+  void  free(void* p)                    MI_FORWARD0(mi_free, p);
 #endif
 
 #if (defined(__GNUC__) || defined(__clang__)) && !defined(__MACH__)
@@ -99,8 +99,8 @@ terms of the MIT license. A copy of the license can be found in the file
   void* operator new[](std::size_t n, const std::nothrow_t& tag) noexcept { UNUSED(tag); return mi_new_nothrow(n); }
 
   #if (__cplusplus >= 201402L || _MSC_VER >= 1916)
-  void operator delete  (void* p, std::size_t n) MI_FORWARD02(mi_free_size,p,n);
-  void operator delete[](void* p, std::size_t n) MI_FORWARD02(mi_free_size,p,n);
+  void operator delete  (void* p, std::size_t n) noexcept MI_FORWARD02(mi_free_size,p,n);
+  void operator delete[](void* p, std::size_t n) noexcept MI_FORWARD02(mi_free_size,p,n);
   #endif
 
   #if (__cplusplus > 201402L || defined(__cpp_aligned_new)) && (!defined(__GNUC__) || (__GNUC__ > 5))

src/alloc.c

@@ -212,7 +212,7 @@ static size_t mi_page_usable_size_of(const mi_page_t* page, const mi_block_t* bl
   size_t delta;
   bool ok = mi_page_decode_padding(page, block, &delta, &bsize);
   mi_assert_internal(ok); mi_assert_internal(delta <= bsize);
-  return (ok ? bsize - delta : 0); 
+  return (ok ? bsize - delta : 0);
 }
 
 static bool mi_verify_padding(const mi_page_t* page, const mi_block_t* block, size_t* size, size_t* wrong) {
@@ -259,7 +259,7 @@ static void mi_padding_shrink(const mi_page_t* page, const mi_block_t* block, co
   mi_padding_t* padding = (mi_padding_t*)((uint8_t*)block + bsize);
   padding->delta = (uint32_t)new_delta;
 }
-#else 
+#else
 static void mi_check_padding(const mi_page_t* page, const mi_block_t* block) {
   UNUSED(page);
   UNUSED(block);
@@ -359,7 +359,7 @@ static inline void _mi_free_block(mi_page_t* page, bool local, mi_block_t* block
     }
     else if (mi_unlikely(mi_page_is_in_full(page))) {
       _mi_page_unfull(page);
-    }    
+    }
   }
   else {
     _mi_free_block_mt(page,block);
@@ -401,7 +401,7 @@ void mi_free(void* p) mi_attr_noexcept
       "(this may still be a valid very large allocation (over 64MiB))\n", p);
     if (mi_likely(_mi_ptr_cookie(segment) == segment->cookie)) {
       _mi_warning_message("(yes, the previous pointer %p was valid after all)\n", p);
-    }
+    } 
   }
 #endif
 #if (MI_DEBUG!=0 || MI_SECURE>=4)
@@ -421,11 +421,11 @@ void mi_free(void* p) mi_attr_noexcept
   mi_heap_stat_decrease(heap, malloc, bsize);
   if (bsize <= MI_LARGE_OBJ_SIZE_MAX) { // huge page stats are accounted for in `_mi_page_retire`
     mi_heap_stat_decrease(heap, normal[_mi_bin(bsize)], 1);
-  }  
+  }
 #endif
 
   if (mi_likely(tid == segment->thread_id && page->flags.full_aligned == 0)) {  // the thread id matches and it is not a full page, nor has aligned blocks
-    // local, and not full or aligned    
+    // local, and not full or aligned
     if (mi_unlikely(mi_check_is_double_free(page,block))) return;
     mi_check_padding(page, block);
     #if (MI_DEBUG!=0)
@@ -436,7 +436,7 @@ void mi_free(void* p) mi_attr_noexcept
     page->used--;
     if (mi_unlikely(mi_page_all_free(page))) {
       _mi_page_retire(page);
-    }    
+    }
   }
   else {
     // non-local, aligned blocks, or a full page; use the more generic path
@@ -473,7 +473,7 @@ size_t mi_usable_size(const void* p) mi_attr_noexcept {
   const mi_segment_t* const segment = _mi_ptr_segment(p);
   const mi_page_t* const page = _mi_segment_page_of(segment, p);
   const mi_block_t* const block = (const mi_block_t*)p;
-  const size_t size = mi_page_usable_size_of(page, block);  
+  const size_t size = mi_page_usable_size_of(page, block);
   if (mi_unlikely(mi_page_has_aligned(page))) {
     ptrdiff_t const adjust = (uint8_t*)p - (uint8_t*)_mi_page_ptr_unalign(segment,page,p);
     mi_assert_internal(adjust >= 0 && (size_t)adjust <= size);

src/init.c

@@ -34,7 +34,7 @@ const mi_page_t _mi_page_empty = {
 
 #if defined(MI_PADDING) && (MI_INTPTR_SIZE >= 8)
 #define MI_SMALL_PAGES_EMPTY  { MI_INIT128(MI_PAGE_EMPTY), MI_PAGE_EMPTY(), MI_PAGE_EMPTY() }
-#elif defined(MI_PADDING) 
+#elif defined(MI_PADDING)
 #define MI_SMALL_PAGES_EMPTY  { MI_INIT128(MI_PAGE_EMPTY), MI_PAGE_EMPTY(), MI_PAGE_EMPTY(), MI_PAGE_EMPTY() }
 #else
 #define MI_SMALL_PAGES_EMPTY  { MI_INIT128(MI_PAGE_EMPTY), MI_PAGE_EMPTY() }
@@ -190,7 +190,7 @@ static bool _mi_heap_init(void) {
     heap->cookie  = _mi_heap_random_next(heap) | 1;
     heap->keys[0] = _mi_heap_random_next(heap);
     heap->keys[1] = _mi_heap_random_next(heap);
-    heap->tld = tld;    
+    heap->tld = tld;
     tld->heap_backing = heap;
     tld->segments.stats = &tld->stats;
     tld->segments.os = &tld->os;
@@ -421,9 +421,9 @@ static void mi_process_load(void) {
   volatile mi_heap_t* dummy = _mi_heap_default; // access TLS to allocate it before setting tls_initialized to true;
   UNUSED(dummy);
   #endif
-  os_preloading = false;  
+  os_preloading = false;
   atexit(&mi_process_done);
-  _mi_options_init();  
+  _mi_options_init();
   mi_process_init();
   //mi_stats_reset();-
   if (mi_redirected) _mi_verbose_message("malloc is redirected.\n");

test/test-stress.c

@@ -38,7 +38,7 @@ static bool   allow_large_objects = true;    // allow very large objects?
 static size_t use_one_size = 0;              // use single object size of `N * sizeof(uintptr_t)`?
 
 
-#ifdef USE_STD_MALLOC
+#ifndef USE_STD_MALLOC
 #define custom_calloc(n,s)    calloc(n,s)
 #define custom_realloc(p,s)   realloc(p,s)
 #define custom_free(p)        free(p)