- Used the already detected cache line size for RtlPrefetchMemoryNonTemporal.  
- Initialized RtlPrefetchMemoryNonTemporal earlier.  
- Change the memory protection during the initialisation of RtlPrefetchMemoryNonTemporal.

Modified: trunk/reactos/lib/rtl/i386/mem_asm.S
Modified: trunk/reactos/ntoskrnl/ke/i386/kernel.c

--- trunk/reactos/lib/rtl/i386/mem_asm.S	2005-11-30 00:17:14 UTC (rev 19763)
+++ trunk/reactos/lib/rtl/i386/mem_asm.S	2005-11-30 00:25:48 UTC (rev 19764)
@@ -151,13 +151,12 @@

 @RtlPrefetchMemoryNonTemporal@8:
 	ret         /* Overwritten by ntoskrnl/ke/i386/kernel.c if SSE is supported (see Ki386SetProcessorFeatures() ) */
 

-	mov eax, [_Ke386CacheGranularity]    // Get cache line size

+	mov eax, [_Ke386CacheAlignment]    // Get cache line size

 
 	// This is fastcall, so ecx = address, edx = size

-	fetch_next_line:

+fetch_next_line:

 	prefetchnta byte ptr [ecx]  // prefechnta(address)

-	sub edx, eax                // count = count - cache_line_size

 	add ecx, eax                // address = address + cache_line_size

-	cmp edx, 0                  // if(count) > 0

+	sub edx, eax                // count = count - cache_line_size

 	ja fetch_next_line          //     goto fetch_next_line
 	ret

--- trunk/reactos/ntoskrnl/ke/i386/kernel.c	2005-11-30 00:17:14 UTC (rev 19763)
+++ trunk/reactos/ntoskrnl/ke/i386/kernel.c	2005-11-30 00:25:48 UTC (rev 19764)
@@ -22,7 +22,6 @@

 ULONG Ke386CacheAlignment;
 CHAR Ke386CpuidModel[49] = {0,};
 ULONG Ke386L1CacheSize;

-ULONG Ke386CacheGranularity = 0x40;      /* FIXME: Default to 64 bytes for RtlPrefetchMemoryNonTemporal(), need real size */

 BOOLEAN Ke386NoExecute = FALSE;
 BOOLEAN Ke386Pae = FALSE;
 BOOLEAN Ke386GlobalPagesEnabled = FALSE;
@@ -418,6 +417,18 @@

       /* Target EIP. */
       Ke386Wrmsr(0x176, (ULONG_PTR)KiFastCallEntry, 0);
    }

+
+   /* Does the CPU Support 'prefetchnta' (SSE)  */
+   if(KPCR->PrcbData.FeatureBits & X86_FEATURE_SSE)
+   {
+       ULONG Protect;
+
+       Protect = MmGetPageProtect(NULL, (PVOID)RtlPrefetchMemoryNonTemporal);
+       MmSetPageProtect(NULL, (PVOID)RtlPrefetchMemoryNonTemporal, Protect | PAGE_IS_WRITABLE);
+       /* Replace the ret by a nop */
+       *(PCHAR)RtlPrefetchMemoryNonTemporal = 0x90;
+       MmSetPageProtect(NULL, (PVOID)RtlPrefetchMemoryNonTemporal, Protect);
+   }

 }
 
 VOID
@@ -523,13 +534,6 @@

    SharedUserData->ProcessorFeatures[PF_XMMI64_INSTRUCTIONS_AVAILABLE] =
       (Pcr->PrcbData.FeatureBits & X86_FEATURE_SSE2);
 

-   /* Does the CPU Support 'prefetchnta' (SSE)  */
-   if(Pcr->PrcbData.FeatureBits & X86_FEATURE_SSE)
-   {
-       /* Replace the ret by a nop */
-       /* FIXME - *(PCHAR)RtlPrefetchMemoryNonTemporal = 0x90; */
-   }
-

    /* Does the CPU Support Fast System Call? */   
    if (Pcr->PrcbData.FeatureBits & X86_FEATURE_SYSCALL) {