merge from dev

src/alloc-aligned.c

@@ -44,7 +44,8 @@ static mi_decl_noinline void* mi_heap_malloc_zero_aligned_at_fallback(mi_heap_t*
       return NULL;
     }
     oversize = (size <= MI_SMALL_SIZE_MAX ? MI_SMALL_SIZE_MAX + 1 /* ensure we use generic malloc path */ : size);
-    p = _mi_heap_malloc_zero_ex(heap, oversize, zero, alignment); // the page block size should be large enough to align in the single huge page block
+    p = _mi_heap_malloc_zero_ex(heap, oversize, false, alignment); // the page block size should be large enough to align in the single huge page block
+    // zero afterwards as only the area from the aligned_p may be committed!
     if (p == NULL) return NULL;    
   }
   else {
@@ -63,6 +64,13 @@ static mi_decl_noinline void* mi_heap_malloc_zero_aligned_at_fallback(mi_heap_t*
   mi_assert_internal(p == _mi_page_ptr_unalign(_mi_ptr_segment(aligned_p), _mi_ptr_page(aligned_p), aligned_p));
   mi_assert_internal(((uintptr_t)aligned_p + offset) % alignment == 0);
   mi_assert_internal(mi_page_usable_block_size(_mi_ptr_page(p)) >= adjust + size);
+
+  // now zero the block if needed
+  if (zero && alignment > MI_ALIGNMENT_MAX) {
+    const ptrdiff_t diff = (uint8_t*)aligned_p - (uint8_t*)p;
+    const ptrdiff_t zsize = mi_page_usable_block_size(_mi_ptr_page(p)) - diff - MI_PADDING_SIZE;
+    if (zsize > 0) { _mi_memzero(aligned_p, zsize); }
+  }
   
   #if MI_TRACK_ENABLED
   if (p != aligned_p) {

src/alloc.c

@@ -51,7 +51,9 @@ extern inline void* _mi_page_malloc(mi_heap_t* heap, mi_page_t* page, size_t siz
   }
 
 #if (MI_DEBUG>0) && !MI_TRACK_ENABLED
-  if (!page->is_zero && !zero) { memset(block, MI_DEBUG_UNINIT, mi_page_usable_block_size(page)); }
+  if (!page->is_zero && !zero && !mi_page_is_huge(page)) { 
+    memset(block, MI_DEBUG_UNINIT, mi_page_usable_block_size(page)); 
+  }
 #elif (MI_SECURE!=0)
   if (!zero) { block->next = 0; } // don't leak internal data
 #endif
@@ -77,9 +79,11 @@ extern inline void* _mi_page_malloc(mi_heap_t* heap, mi_page_t* page, size_t siz
   #endif
   padding->canary = (uint32_t)(mi_ptr_encode(page,block,page->keys));
   padding->delta  = (uint32_t)(delta);
-  uint8_t* fill = (uint8_t*)padding - delta;
-  const size_t maxpad = (delta > MI_MAX_ALIGN_SIZE ? MI_MAX_ALIGN_SIZE : delta); // set at most N initial padding bytes
-  for (size_t i = 0; i < maxpad; i++) { fill[i] = MI_DEBUG_PADDING; }
+  if (!mi_page_is_huge(page)) {
+    uint8_t* fill = (uint8_t*)padding - delta;
+    const size_t maxpad = (delta > MI_MAX_ALIGN_SIZE ? MI_MAX_ALIGN_SIZE : delta); // set at most N initial padding bytes
+    for (size_t i = 0; i < maxpad; i++) { fill[i] = MI_DEBUG_PADDING; }
+  }
 #endif
 
   return block;
@@ -250,17 +254,19 @@ static bool mi_verify_padding(const mi_page_t* page, const mi_block_t* block, si
   if (!ok) return false;
   mi_assert_internal(bsize >= delta);
   *size = bsize - delta;
-  uint8_t* fill = (uint8_t*)block + bsize - delta;
-  const size_t maxpad = (delta > MI_MAX_ALIGN_SIZE ? MI_MAX_ALIGN_SIZE : delta); // check at most the first N padding bytes
-  mi_track_mem_defined(fill,maxpad);
-  for (size_t i = 0; i < maxpad; i++) {
-    if (fill[i] != MI_DEBUG_PADDING) {
-      *wrong = bsize - delta + i;
-      ok = false;
-      break;
+  if (!mi_page_is_huge(page)) {
+    uint8_t* fill = (uint8_t*)block + bsize - delta;
+    const size_t maxpad = (delta > MI_MAX_ALIGN_SIZE ? MI_MAX_ALIGN_SIZE : delta); // check at most the first N padding bytes
+    mi_track_mem_defined(fill, maxpad);
+    for (size_t i = 0; i < maxpad; i++) {
+      if (fill[i] != MI_DEBUG_PADDING) {
+        *wrong = bsize - delta + i;
+        ok = false;
+        break;
+      }
     }
+    mi_track_mem_noaccess(fill, maxpad);
   }
-  mi_track_mem_noaccess(fill,maxpad);
   return ok;
 }
 
@@ -367,10 +373,7 @@ static mi_decl_noinline void _mi_free_block_mt(mi_page_t* page, mi_block_t* bloc
   // that is safe as these are constant and the page won't be freed (as the block is not freed yet).
   mi_check_padding(page, block);
   mi_padding_shrink(page, block, sizeof(mi_block_t));       // for small size, ensure we can fit the delayed thread pointers without triggering overflow detection
-  #if (MI_DEBUG!=0) && !MI_TRACK_ENABLED                    // note: when tracking, cannot use mi_usable_size with multi-threading
-  memset(block, MI_DEBUG_FREED, mi_usable_size(block));
-  #endif
-
+  
   // huge page segments are always abandoned and can be freed immediately
   mi_segment_t* segment = _mi_page_segment(page);
   if (segment->kind==MI_SEGMENT_HUGE) {
@@ -379,6 +382,10 @@ static mi_decl_noinline void _mi_free_block_mt(mi_page_t* page, mi_block_t* bloc
     return;
   }
 
+  #if (MI_DEBUG!=0) && !MI_TRACK_ENABLED                    // note: when tracking, cannot use mi_usable_size with multi-threading
+  memset(block, MI_DEBUG_FREED, mi_usable_size(block));
+  #endif
+
   // Try to put the block on either the page-local thread free list, or the heap delayed free list.
   mi_thread_free_t tfreex;
   bool use_delayed;

src/segment.c

@@ -1531,10 +1531,16 @@ static mi_page_t* mi_segment_huge_page_alloc(size_t size, size_t page_alignment,
   #if MI_DEBUG > 3
   if (page_alignment > 0) {
     size_t psize;
-    void* p = _mi_segment_page_start(segment, page, &psize);
-    void* aligned_p = (void*)_mi_align_up((uintptr_t)p, page_alignment);
-    mi_assert_internal(page_alignment == 0 || _mi_is_aligned(aligned_p, page_alignment));
-    mi_assert_internal(page_alignment == 0 || psize - ((uint8_t*)aligned_p - (uint8_t*)p) >= size);
+    uint8_t* p = _mi_segment_page_start(segment, page, &psize);
+    uint8_t* aligned_p = (uint8_t*)_mi_align_up((uintptr_t)p, page_alignment);
+    mi_assert_internal(_mi_is_aligned(aligned_p, page_alignment));
+    mi_assert_internal(psize - (aligned_p - p) >= size);
+    if (!segment->mem_is_pinned && page->is_committed) {
+       // decommit the part of the page that is unused; this can be quite large (close to MI_SEGMENT_SIZE)
+      uint8_t* decommit_start = p + sizeof(mi_block_t); // for the free list
+      ptrdiff_t decommit_size = aligned_p - decommit_start;
+      _mi_os_decommit(decommit_start, decommit_size, os_tld->stats);
+    }
   }
   #endif
   // for huge pages we initialize the xblock_size as we may