ntdll: Use log-linear bucketing for free lists.

Currently, the free list consists of a "small list" for sizes below 256,
which are linearly spaced, and a "large list" which is manually split
into a few chunks.

This patch replaces it with a single log-linear policy, while expanding
the range the large list covers.

The old implementation had issues when a lot of large allocations
happened. In this case, all the allocations went in the last catch-all
bucket in the "large list", and what happens is:
1. The linked list grew in size over time, causing searching cost to
   skyrocket.
2. With the first-fit allocation policy, fragmentation was also making
   the problem worse.

The new bucketing covers the entire range up until we start allocating
large blocks, which will not enter the free list. It also makes the
allocation policy closer to best-fit (although not exactly), reducing
fragmentation.

The increase in number of free lists does incur some cost when it needs
to be skipped over, but the improvement in allocation performance
outweighs it.

For future work, these ideas (mostly from glibc) might or might not
benefit performance:
- Use an exact best-fit allocation policy.
- Add a bitmap for freelist, allowing empty lists to be skipped with a
  single bit scan.

Signed-off-by: Tatsuyuki Ishi <[email protected]>
(cherry picked from commit a612ab6)

Author: ishitatsuyuki

dlls/kernel32/tests/heap.c

@@ -452,9 +452,7 @@ static void test_HeapCreate(void)
 
     SetLastError( 0xdeadbeef );
     ptr1 = HeapAlloc( heap, 0, alloc_size - (0x200 + 0x80 * sizeof(void *)) );
-    todo_wine
     ok( !ptr1, "HeapAlloc succeeded\n" );
-    todo_wine
     ok( GetLastError() == ERROR_NOT_ENOUGH_MEMORY, "got error %lu\n", GetLastError() );
     ret = HeapFree( heap, 0, ptr1 );
     ok( ret, "HeapFree failed, error %lu\n", GetLastError() );

dlls/ntdll/heap.c

@@ -146,7 +146,8 @@ C_ASSERT( sizeof(ARENA_LARGE) == 4 * BLOCK_ALIGN );
 
 #define ROUND_ADDR(addr, mask) ((void *)((UINT_PTR)(addr) & ~(UINT_PTR)(mask)))
 #define ROUND_SIZE(size, mask) ((((SIZE_T)(size) + (mask)) & ~(SIZE_T)(mask)))
-#define FIELD_MAX(type, field) (((SIZE_T)1 << (sizeof(((type *)0)->field) * 8)) - 1)
+#define FIELD_BITS(type, field) (sizeof(((type *)0)->field) * 8)
+#define FIELD_MAX(type, field) (((SIZE_T)1 << FIELD_BITS(type, field)) - 1)
 
 #define HEAP_MIN_BLOCK_SIZE   ROUND_SIZE(sizeof(struct entry) + BLOCK_ALIGN, BLOCK_ALIGN - 1)
 
@@ -168,17 +169,11 @@ C_ASSERT( HEAP_MAX_FREE_BLOCK_SIZE >= HEAP_MAX_BLOCK_REGION_SIZE );
 /* minimum size to start allocating large blocks */
 #define HEAP_MIN_LARGE_BLOCK_SIZE  (HEAP_MAX_USED_BLOCK_SIZE - 0x1000)
 
-/* There will be a free list bucket for every arena size up to and including this value */
-#define HEAP_MAX_SMALL_FREE_LIST 0x100
-C_ASSERT( HEAP_MAX_SMALL_FREE_LIST % BLOCK_ALIGN == 0 );
-#define HEAP_NB_SMALL_FREE_LISTS (((HEAP_MAX_SMALL_FREE_LIST - HEAP_MIN_BLOCK_SIZE) / BLOCK_ALIGN) + 1)
-
-/* Max size of the blocks on the free lists above HEAP_MAX_SMALL_FREE_LIST */
-static const SIZE_T free_list_sizes[] =
-{
-    0x200, 0x400, 0x1000, ~(SIZE_T)0
-};
-#define HEAP_NB_FREE_LISTS (ARRAY_SIZE(free_list_sizes) + HEAP_NB_SMALL_FREE_LISTS)
+#define FREE_LIST_LINEAR_BITS 2
+#define FREE_LIST_LINEAR_MASK ((1 << FREE_LIST_LINEAR_BITS) - 1)
+#define FREE_LIST_COUNT ((FIELD_BITS( struct block, block_size ) - FREE_LIST_LINEAR_BITS + 1) * (1 << FREE_LIST_LINEAR_BITS) + 1)
+/* for reference, update this when changing parameters */
+C_ASSERT( FREE_LIST_COUNT == 0x3d );
 
 typedef struct DECLSPEC_ALIGN(BLOCK_ALIGN) tagSUBHEAP
 {
@@ -304,7 +299,7 @@ struct heap
     DWORD            pending_pos;   /* Position in pending free requests ring */
     struct block   **pending_free;  /* Ring buffer for pending free requests */
     RTL_CRITICAL_SECTION cs;
-    struct entry     free_lists[HEAP_NB_FREE_LISTS];
+    struct entry     free_lists[FREE_LIST_COUNT];
     struct bin      *bins;
     SUBHEAP          subheap;
 };
@@ -569,23 +564,6 @@ static void valgrind_notify_free_all( SUBHEAP *subheap, const struct heap *heap
 #endif
 }
 
-/* locate a free list entry of the appropriate size */
-/* size is the size of the whole block including the arena header */
-static inline struct entry *find_free_list( struct heap *heap, SIZE_T block_size, BOOL last )
-{
-    struct entry *list, *end = heap->free_lists + ARRAY_SIZE(heap->free_lists);
-    unsigned int i;
-
-    if (block_size <= HEAP_MAX_SMALL_FREE_LIST)
-        i = (block_size - HEAP_MIN_BLOCK_SIZE) / BLOCK_ALIGN;
-    else for (i = HEAP_NB_SMALL_FREE_LISTS; i < HEAP_NB_FREE_LISTS - 1; i++)
-        if (block_size <= free_list_sizes[i - HEAP_NB_SMALL_FREE_LISTS]) break;
-
-    list = heap->free_lists + i;
-    if (last && ++list == end) list = heap->free_lists;
-    return list;
-}
-
 /* get the memory protection type to use for a given heap */
 static inline ULONG get_protection_type( DWORD flags )
 {
@@ -624,10 +602,60 @@ static void heap_set_status( const struct heap *heap, ULONG flags, NTSTATUS stat
     if (status) RtlSetLastWin32ErrorAndNtStatusFromNtStatus( status );
 }
 
-static size_t get_free_list_block_size( unsigned int index )
+static SIZE_T get_free_list_block_size( unsigned int index )
+{
+    DWORD log = index >> FREE_LIST_LINEAR_BITS;
+    DWORD linear = index & FREE_LIST_LINEAR_MASK;
+
+    if (log == 0) return index * BLOCK_ALIGN;
+
+    return (((1 << FREE_LIST_LINEAR_BITS) + linear) << (log - 1)) * BLOCK_ALIGN;
+}
+
+/*
+ * Given a size, return its index in the block size list for freelists.
+ *
+ * With FREE_LIST_LINEAR_BITS=2, the list looks like this
+ * (with respect to size / BLOCK_ALIGN):
+ *   0,
+ *   1,   2,   3,   4,   5,   6,   7,   8,
+ *  10,  12,  14,  16,  20,  24,  28,  32,
+ *  40,  48,  56,  64,  80,  96, 112, 128,
+ * 160, 192, 224, 256, 320, 384, 448, 512,
+ * ...
+ */
+static unsigned int get_free_list_index( SIZE_T block_size )
+{
+    DWORD bit, log, linear;
+
+    if (block_size > get_free_list_block_size( FREE_LIST_COUNT - 1 ))
+        return FREE_LIST_COUNT - 1;
+
+    block_size /= BLOCK_ALIGN;
+    /* find the highest bit */
+    if (!BitScanReverse( &bit, block_size ) || bit < FREE_LIST_LINEAR_BITS)
+    {
+        /* for small values, the index is same as block_size. */
+        log = 0;
+        linear = block_size;
+    }
+    else
+    {
+        /* the highest bit is always set, ignore it and encode the next FREE_LIST_LINEAR_BITS bits
+         * as a linear scale, combined with the shift as a log scale, in the free list index. */
+        log = bit - FREE_LIST_LINEAR_BITS + 1;
+        linear = (block_size >> (bit - FREE_LIST_LINEAR_BITS)) & FREE_LIST_LINEAR_MASK;
+    }
+
+    return (log << FREE_LIST_LINEAR_BITS) + linear;
+}
+
+/* locate a free list entry of the appropriate size */
+static inline struct entry *find_free_list( struct heap *heap, SIZE_T block_size, BOOL last )
 {
-    if (index < HEAP_NB_SMALL_FREE_LISTS) return HEAP_MIN_BLOCK_SIZE + index * BLOCK_ALIGN;
-    return free_list_sizes[index - HEAP_NB_SMALL_FREE_LISTS];
+    unsigned int index = get_free_list_index( block_size );
+    if (last && ++index == FREE_LIST_COUNT) index = 0;
+    return &heap->free_lists[index];
 }
 
 static void heap_dump( const struct heap *heap )
@@ -651,7 +679,7 @@ static void heap_dump( const struct heap *heap )
     }
 
     TRACE( "  free_lists: %p\n", heap->free_lists );
-    for (i = 0; i < HEAP_NB_FREE_LISTS; i++)
+    for (i = 0; i < FREE_LIST_COUNT; i++)
         TRACE( "    %p: size %#8Ix, prev %p, next %p\n", heap->free_lists + i, get_free_list_block_size( i ),
                LIST_ENTRY( heap->free_lists[i].entry.prev, struct entry, entry ),
                LIST_ENTRY( heap->free_lists[i].entry.next, struct entry, entry ) );
@@ -1126,7 +1154,7 @@ static BOOL is_valid_free_block( const struct heap *heap, const struct block *bl
     unsigned int i;
 
     if ((subheap = find_subheap( heap, block, FALSE ))) return TRUE;
-    for (i = 0; i < HEAP_NB_FREE_LISTS; i++) if (block == &heap->free_lists[i].block) return TRUE;
+    for (i = 0; i < FREE_LIST_COUNT; i++) if (block == &heap->free_lists[i].block) return TRUE;
     return FALSE;
 }
 
@@ -1510,7 +1538,7 @@ HANDLE WINAPI RtlCreateHeap( ULONG flags, void *addr, SIZE_T total_size, SIZE_T
     list_init( &heap->large_list );
 
     list_init( &heap->free_lists[0].entry );
-    for (i = 0, entry = heap->free_lists; i < HEAP_NB_FREE_LISTS; i++, entry++)
+    for (i = 0, entry = heap->free_lists; i < FREE_LIST_COUNT; i++, entry++)
     {
         block_set_flags( &entry->block, ~0, BLOCK_FLAG_FREE_LINK );
         block_set_size( &entry->block, 0 );