[NTOSKRNL/DDK]

- add definition for KeMemoryBarrierWithoutFence
- add x64 version of KeMemoryBarrier and some related defines
- Use KeMemoryBarrierWithoutFence in spinlocks as suggested by Alex

svn path=/trunk/; revision=52079

reactos/include/ddk/wdm.h

@@ -170,7 +170,7 @@ typedef struct _DMA_ADAPTER *PADAPTER_OBJECT;
 #elif defined(_WDM_INCLUDED_)
 typedef struct _DMA_ADAPTER *PADAPTER_OBJECT;
 #else
-typedef struct _ADAPTER_OBJECT *PADAPTER_OBJECT; 
+typedef struct _ADAPTER_OBJECT *PADAPTER_OBJECT;
 #endif
 
 #ifndef DEFINE_GUIDEX
@@ -191,7 +191,7 @@ typedef struct _ADAPTER_OBJECT *PADAPTER_OBJECT;
 #ifdef __cplusplus
 inline int IsEqualGUIDAligned(REFGUID guid1, REFGUID guid2)
 {
-    return ( (*(PLONGLONG)(&guid1) == *(PLONGLONG)(&guid2)) && 
+    return ( (*(PLONGLONG)(&guid1) == *(PLONGLONG)(&guid2)) &&
              (*((PLONGLONG)(&guid1) + 1) == *((PLONGLONG)(&guid2) + 1)) );
 }
 #else
@@ -7718,6 +7718,8 @@ KeMemoryBarrier(VOID)
 #endif
 }
 
+#define KeMemoryBarrierWithoutFence() _ReadWriteBarrier()
+
 NTHALAPI
 KIRQL
 NTAPI
@@ -7875,6 +7877,21 @@ typedef XSAVE_FORMAT XMM_SAVE_AREA32, *PXMM_SAVE_AREA32;
 #define KeGetDcacheFillSize() 1L
 
 #define YieldProcessor _mm_pause
+#define FastFence __faststorefence
+#define LoadFence _mm_lfence
+#define MemoryFence _mm_mfence
+#define StoreFence _mm_sfence
+#define LFENCE_ACQUIRE() LoadFence()
+
+FORCEINLINE
+VOID
+KeMemoryBarrier(VOID)
+{
+  FastFence();
+  LFENCE_ACQUIRE();
+}
+
+#define KeMemoryBarrierWithoutFence() _ReadWriteBarrier()
 
 FORCEINLINE
 KIRQL
@@ -13290,14 +13307,14 @@ ExInterlockedFlushSList(
 #if defined(_WIN2K_COMPAT_SLIST_USAGE) && defined(_X86_)
 
 NTKERNELAPI
-PSINGLE_LIST_ENTRY 
+PSINGLE_LIST_ENTRY
 FASTCALL
 ExInterlockedPopEntrySList(
   IN PSLIST_HEADER ListHead,
   IN PKSPIN_LOCK Lock);
 
 NTKERNELAPI
-PSINGLE_LIST_ENTRY 
+PSINGLE_LIST_ENTRY
 FASTCALL
 ExInterlockedPushEntrySList(
   IN PSLIST_HEADER ListHead,

reactos/include/xdk/amd64/ke.h

@@ -45,6 +45,22 @@ typedef XSAVE_FORMAT XMM_SAVE_AREA32, *PXMM_SAVE_AREA32;
 #define KeGetDcacheFillSize() 1L
 
 #define YieldProcessor _mm_pause
+#define FastFence __faststorefence
+#define LoadFence _mm_lfence
+#define MemoryFence _mm_mfence
+#define StoreFence _mm_sfence
+#define LFENCE_ACQUIRE() LoadFence()
+
+FORCEINLINE
+VOID
+KeMemoryBarrier(VOID)
+{
+  // FIXME: Do we really need lfence after the __faststorefence ?
+  FastFence();
+  LFENCE_ACQUIRE();
+}
+
+#define KeMemoryBarrierWithoutFence() _ReadWriteBarrier()
 
 FORCEINLINE
 KIRQL

reactos/include/xdk/x86/ke.h

@@ -59,6 +59,8 @@ KeMemoryBarrier(VOID)
 #endif
 }
 
+#define KeMemoryBarrierWithoutFence() _ReadWriteBarrier()
+
 NTHALAPI
 KIRQL
 NTAPI

reactos/ntoskrnl/include/internal/spinlock.h

@@ -24,7 +24,7 @@ KxAcquireSpinLock(IN PKSPIN_LOCK SpinLock)
 
     /* Add an explicit memory barrier to prevent the compiler from reordering
        memory accesses across the borders of spinlocks */
-    _ReadWriteBarrier();
+    KeMemoryBarrierWithoutFence();
 }
 
 //
@@ -39,7 +39,7 @@ KxReleaseSpinLock(IN PKSPIN_LOCK SpinLock)
 
     /* Add an explicit memory barrier to prevent the compiler from reordering
        memory accesses across the borders of spinlocks */
-    _ReadWriteBarrier();
+    KeMemoryBarrierWithoutFence();
 }
 
 #else

reactos/ntoskrnl/ke/amd64/spinlock.c

@@ -173,7 +173,7 @@ KeTryToAcquireQueuedSpinLockRaiseToSynch(IN KSPIN_LOCK_QUEUE_NUMBER LockNumber,
 
     /* Add an explicit memory barrier to prevent the compiler from reordering
        memory accesses across the borders of spinlocks */
-    _ReadWriteBarrier();
+    KeMemoryBarrierWithoutFence();
 
     /* Always return true on UP Machines */
     return TRUE;
@@ -196,7 +196,7 @@ KeTryToAcquireQueuedSpinLock(IN KSPIN_LOCK_QUEUE_NUMBER LockNumber,
 
     /* Add an explicit memory barrier to prevent the compiler from reordering
        memory accesses across the borders of spinlocks */
-    _ReadWriteBarrier();
+    KeMemoryBarrierWithoutFence();
 
     /* Always return true on UP Machines */
     return TRUE;