internal/refcount.h: overhaul fencing and add _MSC_VER section.

Relax memory_order on counter decrement itself, because mutable
members of the reference-counted structure should be visible on all
processors independently on counter. [Even re-format and minimize
dependency on other headers.]

Reviewed-by: Kurt Roeckx <kurt@roeckx.be>
(Merged from https://github.com/openssl/openssl/pull/6900)

1 files changed, 62 insertions, 11 deletions

diff --git a/include/internal/refcount.h b/include/internal/refcount.h
index c1663a0f83..75d70a6418 100644
--- a/include/internal/refcount.h
+++ b/include/internal/refcount.h
@@ -1,5 +1,5 @@
 /*
- * Copyright 2016 The OpenSSL Project Authors. All Rights Reserved.
+ * Copyright 2016-2018 The OpenSSL Project Authors. All Rights Reserved.
  *
  * Licensed under the OpenSSL license (the "License").  You may not use
  * this file except in compliance with the License.  You can obtain a copy
@@ -18,26 +18,36 @@
 
 # if defined(__STDC_VERSION__) && __STDC_VERSION__ >= 201112L \
      && !defined(__STDC_NO_ATOMICS__)
-# include <stdatomic.h>
-# define HAVE_C11_ATOMICS
+#  include <stdatomic.h>
+#  define HAVE_C11_ATOMICS
 # endif
 
 # if defined(HAVE_C11_ATOMICS) && defined(ATOMIC_INT_LOCK_FREE) \
      && ATOMIC_INT_LOCK_FREE > 0
 
-# define HAVE_ATOMICS 1
+#  define HAVE_ATOMICS 1
 
 typedef _Atomic int CRYPTO_REF_COUNT;
 
-static ossl_inline int CRYPTO_UP_REF(_Atomic int *val, int *ret, void *lock)
+static inline int CRYPTO_UP_REF(_Atomic int *val, int *ret, void *lock)
 {
     *ret = atomic_fetch_add_explicit(val, 1, memory_order_relaxed) + 1;
     return 1;
 }
 
-static ossl_inline int CRYPTO_DOWN_REF(_Atomic int *val, int *ret, void *lock)
+/*
+ * Changes to shared structure other than reference counter have to be
+ * serialized. And any kind of serialization implies a release fence. This
+ * means that by the time reference counter is decremented all other
+ * changes are visible on all processors. Hence decrement itself can be
+ * relaxed. In case it hits zero, object will be destructed. Since it's
+ * last use of the object, destructor programmer might reason that access
+ * to mutable members doesn't have to be serialized anymore, which would
+ * otherwise imply an acquire fence. Hence conditional acquire fence...
+ */
+static inline int CRYPTO_DOWN_REF(_Atomic int *val, int *ret, void *lock)
 {
-    *ret = atomic_fetch_sub_explicit(val, 1, memory_order_release) - 1;
+    *ret = atomic_fetch_sub_explicit(val, 1, memory_order_relaxed) - 1;
     if (*ret == 0)
         atomic_thread_fence(memory_order_acquire);
     return 1;
@@ -45,24 +55,65 @@ static ossl_inline int CRYPTO_DOWN_REF(_Atomic int *val, int *ret, void *lock)
 
 # elif defined(__GNUC__) && defined(__ATOMIC_RELAXED) && __GCC_ATOMIC_INT_LOCK_FREE > 0
 
-# define HAVE_ATOMICS 1
+#  define HAVE_ATOMICS 1
 
 typedef int CRYPTO_REF_COUNT;
 
-static ossl_inline int CRYPTO_UP_REF(int *val, int *ret, void *lock)
+static __inline__ int CRYPTO_UP_REF(int *val, int *ret, void *lock)
 {
     *ret = __atomic_fetch_add(val, 1, __ATOMIC_RELAXED) + 1;
     return 1;
 }
 
-static ossl_inline int CRYPTO_DOWN_REF(int *val, int *ret, void *lock)
+static __inline__ int CRYPTO_DOWN_REF(int *val, int *ret, void *lock)
 {
-    *ret = __atomic_fetch_sub(val, 1, __ATOMIC_RELEASE) - 1;
+    *ret = __atomic_fetch_sub(val, 1, __ATOMIC_RELAXED) - 1;
     if (*ret == 0)
         __atomic_thread_fence(__ATOMIC_ACQUIRE);
     return 1;
 }
 
+# elif defined(_MSC_VER) && _MSC_VER>=1200
+
+#  define HAVE_ATOMICS 1
+
+typedef volatile int CRYPTO_REF_COUNT;
+
+#  if (defined(_M_ARM) && _M_ARM>=7) || defined(_M_ARM64)
+#   include <intrin.h>
+#   if defined(_M_ARM64) && !defined(_ARM_BARRIER_ISH)
+#    define _ARM_BARRIER_ISH _ARM64_BARRIER_ISH
+#   endif
+
+static __inline int CRYPTO_UP_REF(volatile int *val, int *ret, void *lock)
+{
+    *ret = _InterlockedExchangeAdd_nf(val, 1) + 1;
+    return 1;
+}
+
+static __inline int CRYPTO_DOWN_REF(volatile int *val, int *ret, void *lock)
+{
+    *ret = _InterlockedExchangeAdd_nf(val, -1) - 1;
+    if (*ret == 0)
+        __dmb(_ARM_BARRIER_ISH);
+    return 1;
+}
+#  else
+#   pragma intrinsic(_InterlockedExchangeAdd)
+
+static __inline int CRYPTO_UP_REF(volatile int *val, int *ret, void *lock)
+{
+    *ret = _InterlockedExchangeAdd(val, 1) + 1;
+    return 1;
+}
+
+static __inline int CRYPTO_DOWN_REF(volatile int *val, int *ret, void *lock)
+{
+    *ret = _InterlockedExchangeAdd(val, -1) - 1;
+    return 1;
+}
+#  endif
+
 # else
 
 typedef int CRYPTO_REF_COUNT;