Loading...
--- libmalloc/libmalloc-116/src/thresholds.h
+++ libmalloc/libmalloc-317.140.5/src/thresholds.h
@@ -25,83 +25,122 @@
#define __THRESHOLDS_H
/*
- * Tiny region size definitions; these are split into quanta of 16 bytes,
- * 64520 blocks is the magical value of how many quanta we can fit in a 1mb
+ * The actual threshold boundaries between allocators. These boundaries are
+ * where the next allocator will take over and they are *inclusive* of the
+ * limit value. That is, a TINY limit of X implies that an X-byte
+ * allocation will come from TINY.
+ *
+ * The LARGE allocator cuts in at whatever the last boundary limit is. So, when
+ * the medium allocator is compiled out, or not engaged, then large starts at
+ * the limit of SMALL.
+ */
+#if MALLOC_TARGET_64BIT
+#define TINY_LIMIT_THRESHOLD (1008)
+#else // MALLOC_TARGET_64BIT
+#define TINY_LIMIT_THRESHOLD (496)
+#endif // MALLOC_TARGET_64BIT
+
+#if MALLOC_TARGET_IOS
+#define SMALL_LIMIT_THRESHOLD (15 * 1024)
+#else // MALLOC_TARGET_IOS
+#define SMALL_LIMIT_THRESHOLD (32 * 1024)
+#endif // MALLOC_TARGET_IOS
+#define MEDIUM_LIMIT_THRESHOLD (8 * 1024 * 1024)
+
+/*
+ * Tiny region size definitions; these are split into quanta of 16 bytes,
+ * 64504 blocks is the magical value of how many quanta we can fit in a 1mb
* region including the region trailer and metadata.
*/
-#define SHIFT_TINY_QUANTUM 4
+#define SHIFT_TINY_QUANTUM 4ull
#define SHIFT_TINY_CEIL_BLOCKS 16 // ceil(log2(NUM_TINY_BLOCKS))
#define TINY_QUANTUM (1 << SHIFT_TINY_QUANTUM)
-#define NUM_TINY_BLOCKS 64520
+#define NUM_TINY_BLOCKS 64504
#define NUM_TINY_CEIL_BLOCKS (1 << SHIFT_TINY_CEIL_BLOCKS)
-
-/*
- * Small region size defintions.
+#define NUM_TINY_SLOTS (TINY_LIMIT_THRESHOLD >> SHIFT_TINY_QUANTUM)
+
+#if MALLOC_TARGET_64BIT
+#define TINY_BITMAP_RANGE_LIMIT 63
+#else
+#define TINY_BITMAP_RANGE_LIMIT 31
+#endif
+
+/*
+ * Small region size definitions.
*
* We can only represent up to 1<<15 for msize; but we choose to stay
* even below that to avoid the convention msize=0 => msize = (1<<15)
*/
#define SHIFT_SMALL_QUANTUM (SHIFT_TINY_QUANTUM + 5) // 9
-#define SMALL_QUANTUM (1 << SHIFT_SMALL_QUANTUM) // 512 bytes
#define SHIFT_SMALL_CEIL_BLOCKS 14 // ceil(log2(NUM_SMALL_BLOCKs))
+#define SMALL_QUANTUM (1 << SHIFT_SMALL_QUANTUM) // 512 bytes
+#define SMALL_BLOCKS_ALIGN (SHIFT_SMALL_CEIL_BLOCKS + SHIFT_SMALL_QUANTUM) // 23
#define NUM_SMALL_BLOCKS 16319
#define NUM_SMALL_CEIL_BLOCKS (1 << SHIFT_SMALL_CEIL_BLOCKS)
-#define SMALL_BLOCKS_ALIGN (SHIFT_SMALL_CEIL_BLOCKS + SHIFT_SMALL_QUANTUM) // 23
-
-/*
- * The number of slots in the free-list for small blocks. To avoid going to
- * vm system as often on large memory machines, increase the number of free list
- * spots above some amount of RAM installed in the system.
- */
-#define NUM_SMALL_SLOTS 32
-#define NUM_SMALL_SLOTS_LARGEMEM 256
-#define SMALL_BITMAP_WORDS 8
-
-#if MALLOC_TARGET_64BIT
-#define NUM_TINY_SLOTS 64 // number of slots for free-lists
-#else // MALLOC_TARGET_64BIT
-#define NUM_TINY_SLOTS 32 // number of slots for free-lists
-#endif // MALLOC_TARGET_64BIT
-
-/*
- * The threshold above which we start allocating from the small
- * magazines. Computed from the largest allocation we can make
- * in the tiny region (currently 1008 bytes on 64-bit, and
- * 496 bytes on 32-bit).
- */
-#define SMALL_THRESHOLD ((NUM_TINY_SLOTS - 1) * TINY_QUANTUM)
-
-/*
- * The threshold above which we start allocating from the large
- * "region" (ie. direct vm_allocates). The LARGEMEM size is used
- * on systems that have more than 1GB RAM.
- */
-#define LARGE_THRESHOLD (15 * 1024)
-#define LARGE_THRESHOLD_LARGEMEM (127 * 1024)
-
-/*
- * When all memory is touched after a copy, vm_copy() is always a lose
- * But if the memory is only read, vm_copy() wins over memmove() at 3 or 4 pages
- * (on a G3/300MHz)
- *
- * This must be larger than LARGE_THRESHOLD
- */
-#define VM_COPY_THRESHOLD (40 * 1024)
-#define VM_COPY_THRESHOLD_LARGEMEM (128 * 1024)
+#define NUM_SMALL_SLOTS (SMALL_LIMIT_THRESHOLD >> SHIFT_SMALL_QUANTUM)
+
+/*
+ * Medium region size definitions.
+ *
+ * We can only represent up to 1<<15 for msize; but we choose to stay
+ * even below that to avoid the convention msize=0 => msize = (1<<15)
+ */
+#define SHIFT_MEDIUM_QUANTUM (SHIFT_SMALL_QUANTUM + 6) // 15
+#define SHIFT_MEDIUM_CEIL_BLOCKS 12ull // ceil(log2(NUM_MEDIUM_BLOCKS))
+#define MEDIUM_QUANTUM ((uint64_t)(1 << SHIFT_MEDIUM_QUANTUM)) // 32kbytes
+#define MEDIUM_BLOCKS_ALIGN (SHIFT_MEDIUM_CEIL_BLOCKS + SHIFT_MEDIUM_QUANTUM) // 27
+#define NUM_MEDIUM_BLOCKS 4095
+#define NUM_MEDIUM_CEIL_BLOCKS (1ull << SHIFT_MEDIUM_CEIL_BLOCKS)
+#define NUM_MEDIUM_SLOTS (MEDIUM_LIMIT_THRESHOLD >> SHIFT_MEDIUM_QUANTUM)
+#define MEDIUM_ACTIVATION_THRESHOLD (32ull * 1024 * 1024 * 1024)
+#define MEDIUM_CONDITIONAL_MADVISE_LIMIT (2 * 1024 * 1024)
+#define MEDIUM_MADVISE_SHIFT 4
+#define MEDIUM_MADVISE_MIN ((3 * 1024 * 1024) / 2) // 1.5 megabytes
+
+/*
+ * When performing a realloc() that must fallback to creating a new allocation
+ * and copying the previous contents to the new allocation, vm_copy is used if
+ * the allocation is greater than a given size.
+ *
+ * This threshold must be set such that all eligible allocations would have
+ * come from a page-sized, page-aligned allocator (so, medium or large).
+ *
+ * Note: iOS disables this threshold because the VM forces non-sharing from
+ * malloc-tagged allocations.
+ */
+#define VM_COPY_THRESHOLD (2 * 1024 * 1024)
+
+/*
+ * <rdar://problem/6881926&27190324> Extremely old versions of Microsoft Word
+ * (and, subsequently, versions of Adobe apps) required the Leopard behaviour
+ * where LARGE allocations were zero-filled prior to returning them to the
+ * caller.
+ *
+ * We've always used LARGE_THRESHOLD to denote this boundary but as we keep
+ * moving it around it's better to fix it at the point it was originally.
+ */
+#define LEGACY_ZEROING_THRESHOLD (127 * 1024)
/*
* Large entry cache (death row) sizes. The large cache is bounded with
* an overall top limit size, each entry is allowed a given slice of
* that limit.
*/
+
+#define LARGE_CACHE_EXPANDED_THRESHOLD (32ull * 1024 * 1024 * 1024)
+
#if MALLOC_TARGET_64BIT
-#define LARGE_ENTRY_CACHE_SIZE 16
-#define LARGE_CACHE_SIZE_LIMIT ((vm_size_t)0x80000000) /* 2Gb */
-#define LARGE_CACHE_SIZE_ENTRY_LIMIT (LARGE_CACHE_SIZE_LIMIT / LARGE_ENTRY_CACHE_SIZE)
+#define LARGE_ENTRY_CACHE_SIZE_HIGH 64
+#define LARGE_ENTRY_SIZE_ENTRY_LIMIT_HIGH (512 * 1024 * 1024)
+// lowmem config
+#define LARGE_ENTRY_CACHE_SIZE_LOW 16
+#define LARGE_ENTRY_SIZE_ENTRY_LIMIT_LOW (128 * 1024 * 1024)
#else // MALLOC_TARGET_64BIT
-#define LARGE_ENTRY_CACHE_SIZE 8
-#define LARGE_CACHE_SIZE_LIMIT ((vm_size_t)0x02000000) /* 32Mb */
-#define LARGE_CACHE_SIZE_ENTRY_LIMIT (LARGE_CACHE_SIZE_LIMIT / LARGE_ENTRY_CACHE_SIZE)
+#define LARGE_ENTRY_CACHE_SIZE_HIGH 8
+#define LARGE_ENTRY_SIZE_ENTRY_LIMIT_HIGH (32 * 1024 * 1024)
+// lowmem config same as "highmem"
+#define LARGE_ENTRY_CACHE_SIZE_LOW LARGE_ENTRY_CACHE_SIZE_HIGH
+#define LARGE_ENTRY_SIZE_ENTRY_LIMIT_LOW LARGE_ENTRY_SIZE_ENTRY_LIMIT_HIGH
#endif // MALLOC_TARGET_64BIT
/*
@@ -114,20 +153,76 @@
#define SZONE_FLOTSAM_THRESHOLD_HIGH (1024 * 1024)
/*
+ * The magazine freelist array must be large enough to accomodate the allocation
+ * granularity of the tiny, small and medium allocators. In addition, the last
+ * slot in the list is special and reserved for coalesced regions bigger than
+ * the overall max allocation size of the allocator.
+ */
+#define MAGAZINE_FREELIST_SLOTS (NUM_MEDIUM_SLOTS + 1)
+#define MAGAZINE_FREELIST_BITMAP_WORDS ((MAGAZINE_FREELIST_SLOTS + 31) >> 5)
+
+/*
* Density threshold used in determining the level of emptiness before
* moving regions to the recirc depot.
*/
#define DENSITY_THRESHOLD(a) \
((a) - ((a) >> 2)) // "Emptiness" f = 0.25, so "Density" is (1 - f)*a. Generally: ((a) - ((a) >> -log2(f)))
+/*
+ * Minimum number of regions to retain in a recirc depot.
+ */
+#define DEFAULT_RECIRC_RETAINED_REGIONS 2
+
/* Sanity checks. */
-#if (LARGE_THRESHOLD > NUM_SMALL_SLOTS * SMALL_QUANTUM)
-#error LARGE_THRESHOLD should always be less than NUM_SMALL_SLOTS * SMALL_QUANTUM
-#endif
-
-#if (LARGE_THRESHOLD_LARGEMEM > NUM_SMALL_SLOTS_LARGEMEM * SMALL_QUANTUM)
-#error LARGE_THRESHOLD_LARGEMEM should always be less than NUM_SMALL_SLOTS * SMALL_QUANTUM
-#endif
+// Tiny performs an ffsl of a uint64_t in order to determine how big an
+// allocation is. Therefore, the total allocation size of small cannot exceed
+// 63-bits worth of 16-byte quanta (64-bits but minus one for the start of the
+// allocation itself).
+MALLOC_STATIC_ASSERT((TINY_LIMIT_THRESHOLD / TINY_QUANTUM) <= TINY_BITMAP_RANGE_LIMIT,
+ "TINY_LIMIT_THRESHOLD cannot exceed TINY_BITMAP_RANGE_LIMIT-bits worth of metadata");
+
+// Check that the given threshold limits are a round multiple of their
+// allocator's quantum size.
+MALLOC_STATIC_ASSERT((TINY_LIMIT_THRESHOLD % TINY_QUANTUM) == 0,
+ "TINY_LIMIT_THRESHOLD must be a multiple of TINY_QUANTUM");
+
+MALLOC_STATIC_ASSERT((SMALL_LIMIT_THRESHOLD % SMALL_QUANTUM) == 0,
+ "SMALL_LIMIT_THRESHOLD must be a multiple of SMALL_QUANTUM");
+
+MALLOC_STATIC_ASSERT((MEDIUM_LIMIT_THRESHOLD % MEDIUM_QUANTUM) == 0,
+ "MEDIUM_LIMIT_THRESHOLD must be a multiple of MEDIUM_QUANTUM");
+
+// All the "slot" counts are calculated as a shift of thresholds now but
+// in-case someone decides to try hand-crafted values, make sure they adhere to
+// the basic expectation that slot count must account for all valid sizes of
+// allocations.
+MALLOC_STATIC_ASSERT(NUM_TINY_SLOTS >= TINY_LIMIT_THRESHOLD >> SHIFT_TINY_QUANTUM,
+ "NUM_TINY_SLOTS must allow a free list for every valid TINY allocation");
+
+MALLOC_STATIC_ASSERT(NUM_SMALL_SLOTS >= SMALL_LIMIT_THRESHOLD >> SHIFT_SMALL_QUANTUM,
+ "NUM_SMALL_SLOTS must allow a free list for every valid SMALL allocation");
+
+MALLOC_STATIC_ASSERT(NUM_MEDIUM_SLOTS >= MEDIUM_LIMIT_THRESHOLD >> SHIFT_MEDIUM_QUANTUM,
+ "NUM_MEDIUM_SLOTS must allow a free list for every valid MEDIUM allocation");
+
+// MAGAZINE_FREELIST_SLOTS cannot be dynamically selected by the MAX() of all
+// three allocators, so it must match (at least) the maxmium slot count of the
+// allocator with the largest range.
+//
+// Additionally, each allocator assumes that there is one additional free-list
+// slot above their maximum allocation size. This allows each allocator to
+// store an unordered list of maximally-sized free list entries.
+MALLOC_STATIC_ASSERT(NUM_TINY_SLOTS < MAGAZINE_FREELIST_SLOTS,
+ "NUM_TINY_SLOTS must be less than MAGAZINE_FREELIST_SLOTS");
+
+MALLOC_STATIC_ASSERT(NUM_SMALL_SLOTS < MAGAZINE_FREELIST_SLOTS,
+ "NUM_SMALL_SLOTS must be less than MAGAZINE_FREELIST_SLOTS");
+
+MALLOC_STATIC_ASSERT(NUM_MEDIUM_SLOTS < MAGAZINE_FREELIST_SLOTS,
+ "NUM_MEDIUM_SLOTS must be less than MAGAZINE_FREELIST_SLOTS");
+
+MALLOC_STATIC_ASSERT(VM_COPY_THRESHOLD >= SMALL_LIMIT_THRESHOLD,
+ "VM_COPY_THRESHOLD must be larger than SMALL_LIMIT_THRESHOLD");
#endif // __THRESHOLDS_H