merge from dev

include/mimalloc/internal.h

@@ -91,6 +91,7 @@ void       _mi_tld_init(mi_tld_t* tld, mi_heap_t* bheap);
 mi_threadid_t _mi_thread_id(void) mi_attr_noexcept;
 mi_heap_t*    _mi_heap_main_get(void);     // statically allocated main backing heap
 mi_subproc_t* _mi_subproc_from_id(mi_subproc_id_t subproc_id);
+void       _mi_heap_guarded_init(mi_heap_t* heap);
 
 // os.c
 void       _mi_os_init(void);                                            // called from process init
@@ -639,16 +640,40 @@ static inline void mi_page_set_has_aligned(mi_page_t* page, bool has_aligned) {
   page->flags.x.has_aligned = has_aligned;
 }
 
-#if MI_DEBUG_GUARDED
-static inline bool mi_page_has_guarded(const mi_page_t* page) {
-  return page->flags.x.has_guarded;
+/* -------------------------------------------------------------------
+  Guarded objects
+------------------------------------------------------------------- */
+#if MI_GUARDED
+static inline bool mi_block_ptr_is_guarded(const mi_block_t* block, const void* p) {
+  const ptrdiff_t offset = (uint8_t*)p - (uint8_t*)block;
+  return (offset >= (ptrdiff_t)(sizeof(mi_block_t)) && block->next == MI_BLOCK_TAG_GUARDED);
 }
 
-static inline void mi_page_set_has_guarded(mi_page_t* page, bool has_guarded) {
-  page->flags.x.has_guarded = has_guarded;
+static inline bool mi_heap_malloc_use_guarded(mi_heap_t* heap, size_t size) {
+  // this code is written to result in fast assembly as it is on the hot path for allocation
+  const size_t count = heap->guarded_sample_count - 1;  // if the rate was 0, this will underflow and count for a long time..
+  if mi_likely(count != 0) {
+    // no sample
+    heap->guarded_sample_count = count;
+    return false;
+  }
+  else if (size >= heap->guarded_size_min && size <= heap->guarded_size_max) {
+    // use guarded allocation
+    heap->guarded_sample_count = heap->guarded_sample_rate;  // reset
+    return (heap->guarded_sample_rate != 0);
+  }
+  else {
+    // failed size criteria, rewind count (but don't write to an empty heap)
+    if (heap->guarded_sample_rate != 0) { heap->guarded_sample_count = 1; } 
+    return false;
+  }  
 }
+
+mi_decl_restrict void* _mi_heap_malloc_guarded(mi_heap_t* heap, size_t size, bool zero) mi_attr_noexcept;
+
 #endif
 
+
 /* -------------------------------------------------------------------
 Encoding/Decoding the free list next pointers
 

include/mimalloc/prim.h

@@ -25,6 +25,8 @@ typedef struct mi_os_mem_config_s {
   size_t  page_size;            // default to 4KiB
   size_t  large_page_size;      // 0 if not supported, usually 2MiB (4MiB on Windows)
   size_t  alloc_granularity;    // smallest allocation size (usually 4KiB, on Windows 64KiB)
+  size_t  physical_memory;      // physical memory size
+  size_t  virtual_address_bits; // usually 48 or 56 bits on 64-bit systems. (used to determine secure randomization)
   bool    has_overcommit;       // can we reserve more memory than can be actually committed?
   bool    has_partial_free;     // can allocated blocks be freed partially? (true for mmap, false for VirtualAlloc)
   bool    has_virtual_reserve;  // supports virtual address space reservation? (if true we can reserve virtual address space without using commit or physical memory)
@@ -41,9 +43,10 @@ int _mi_prim_free(void* addr, size_t size );
 // If `commit` is false, the virtual memory range only needs to be reserved (with no access)
 // which will later be committed explicitly using `_mi_prim_commit`.
 // `is_zero` is set to true if the memory was zero initialized (as on most OS's)
+// The `hint_addr` address is either `NULL` or a preferred allocation address but can be ignored.
 // pre: !commit => !allow_large
 //      try_alignment >= _mi_os_page_size() and a power of 2
-int _mi_prim_alloc(size_t size, size_t try_alignment, bool commit, bool allow_large, bool* is_large, bool* is_zero, void** addr);
+int _mi_prim_alloc(void* hint_addr, size_t size, size_t try_alignment, bool commit, bool allow_large, bool* is_large, bool* is_zero, void** addr);
 
 // Commit memory. Returns error code or 0 on success.
 // For example, on Linux this would make the memory PROT_READ|PROT_WRITE.

include/mimalloc/types.h

@@ -75,8 +75,8 @@ terms of the MIT license. A copy of the license can be found in the file
 
 // Use guard pages behind objects of a certain size (set by the MIMALLOC_DEBUG_GUARDED_MIN/MAX options)
 // Padding should be disabled when using guard pages
-// #define MI_DEBUG_GUARDED 1
-#if defined(MI_DEBUG_GUARDED)
+// #define MI_GUARDED 1
+#if defined(MI_GUARDED)
 #define MI_PADDING  0
 #endif
 
@@ -244,6 +244,13 @@ typedef struct mi_block_s {
   mi_encoded_t next;
 } mi_block_t;
 
+#if MI_GUARDED
+// we always align guarded pointers in a block at an offset
+// the block `next` field is then used as a tag to distinguish regular offset aligned blocks from guarded ones
+#define MI_BLOCK_TAG_ALIGNED   ((mi_encoded_t)(0))
+#define MI_BLOCK_TAG_GUARDED   (~MI_BLOCK_TAG_ALIGNED)
+#endif
+
 
 // The delayed flags are used for efficient multi-threaded free-ing
 typedef enum mi_delayed_e {
@@ -262,7 +269,6 @@ typedef union mi_page_flags_s {
   struct {
     uint8_t in_full : 1;
     uint8_t has_aligned : 1;
-    uint8_t has_guarded : 1;  // only used with MI_DEBUG_GUARDED
   } x;
 } mi_page_flags_t;
 #else
@@ -272,7 +278,6 @@ typedef union mi_page_flags_s {
   struct {
     uint8_t in_full;
     uint8_t has_aligned;
-    uint8_t has_guarded; // only used with MI_DEBUG_GUARDED
   } x;
 } mi_page_flags_t;
 #endif
@@ -556,6 +561,13 @@ struct mi_heap_s {
   mi_heap_t*            next;                                // list of heaps per thread
   bool                  no_reclaim;                          // `true` if this heap should not reclaim abandoned pages
   uint8_t               tag;                                 // custom tag, can be used for separating heaps based on the object types
+  #if MI_GUARDED
+  size_t                guarded_size_min;                    // minimal size for guarded objects
+  size_t                guarded_size_max;                    // maximal size for guarded objects
+  size_t                guarded_sample_rate;                 // sample rate (set to 0 to disable guarded pages)
+  size_t                guarded_sample_seed;                 // starting sample count
+  size_t                guarded_sample_count;                // current sample count (counting down to 0)
+  #endif
   mi_page_t*            pages_free_direct[MI_PAGES_DIRECT];  // optimize: array where every entry points a page with possibly free blocks in the corresponding queue for that size.
   mi_page_queue_t       pages[MI_BIN_FULL + 1];              // queue of pages for each size class (or "bin")
 };
@@ -649,6 +661,7 @@ typedef struct mi_stats_s {
   mi_stat_counter_t arena_count;
   mi_stat_counter_t arena_crossover_count;
   mi_stat_counter_t arena_rollback_count;
+  mi_stat_counter_t guarded_alloc_count;
 #if MI_STAT>1
   mi_stat_count_t normal_bins[MI_BIN_HUGE+1];
 #endif