X-Git-Url: https://git.ucc.asn.au/?a=blobdiff_plain;f=Kernel%2Farch%2Fx86_64%2Fmm_virt.c;h=c04fcf3a359216ea57d8fa76f48a5eac165fe79e;hb=463074dd4899f1cc851bff9c74356f5c6def1bb2;hp=d86ca1a92b6e8c14d602594738cfc13c1cc01f66;hpb=ccd85ddbb5551d2849363c45b41e1ba495b05905;p=tpg%2Facess2.git

diff --git a/Kernel/arch/x86_64/mm_virt.c b/Kernel/arch/x86_64/mm_virt.c
index d86ca1a9..c04fcf3a 100644
--- a/Kernel/arch/x86_64/mm_virt.c
+++ b/Kernel/arch/x86_64/mm_virt.c
@@ -3,7 +3,7 @@
  * 
  * Virtual Memory Manager
  */
-#define DEBUG	1
+#define DEBUG	0
 #include <acess.h>
 #include <mm_virt.h>
 #include <threads_int.h>
@@ -11,6 +11,7 @@
 
 // === CONSTANTS ===
 #define PHYS_BITS	52	// TODO: Move out
+#define VIRT_BITS	48
 
 #define PML4_SHIFT	39
 #define PDP_SHIFT	30
@@ -18,27 +19,28 @@
 #define PTAB_SHIFT	12
 
 #define	PADDR_MASK	0x7FFFFFFF##FFFFF000
-#define PAGE_MASK	(((Uint)1 << 36)-1)
-#define TABLE_MASK	(((Uint)1 << 27)-1)
-#define PDP_MASK	(((Uint)1 << 18)-1)
-#define PML4_MASK	(((Uint)1 << 9)-1)
+#define PAGE_MASK	((1LL << 36)-1)
+#define TABLE_MASK	((1LL << 27)-1)
+#define PDP_MASK	((1LL << 18)-1)
+#define PML4_MASK	((1LL << 9)-1)
 
 #define	PF_PRESENT	0x001
 #define	PF_WRITE	0x002
 #define	PF_USER		0x004
-#define	PF_LARGE	0x000
+#define	PF_LARGE	0x080
+#define	PF_GLOBAL	0x100
 #define	PF_COW		0x200
 #define	PF_PAGED	0x400
 #define	PF_NX		0x80000000##00000000
 
 // === MACROS ===
-#define PAGETABLE(idx)	(*((tPAddr*)MM_FRACTAL_BASE+((idx)&PAGE_MASK)))
+#define PAGETABLE(idx)	(*((Uint64*)MM_FRACTAL_BASE+((idx)&PAGE_MASK)))
 #define PAGEDIR(idx)	PAGETABLE((MM_FRACTAL_BASE>>12)+((idx)&TABLE_MASK))
 #define PAGEDIRPTR(idx)	PAGEDIR((MM_FRACTAL_BASE>>21)+((idx)&PDP_MASK))
 #define PAGEMAPLVL4(idx)	PAGEDIRPTR((MM_FRACTAL_BASE>>30)+((idx)&PML4_MASK))
 
 #define TMPCR3()	PAGEMAPLVL4(MM_TMPFRAC_BASE>>39)
-#define TMPTABLE(idx)	(*((tPAddr*)MM_TMPFRAC_BASE+((idx)&PAGE_MASK)))
+#define TMPTABLE(idx)	(*((Uint64*)MM_TMPFRAC_BASE+((idx)&PAGE_MASK)))
 #define TMPDIR(idx)	PAGETABLE((MM_TMPFRAC_BASE>>12)+((idx)&TABLE_MASK))
 #define TMPDIRPTR(idx)	PAGEDIR((MM_TMPFRAC_BASE>>21)+((idx)&PDP_MASK))
 #define TMPMAPLVL4(idx)	PAGEDIRPTR((MM_TMPFRAC_BASE>>30)+((idx)&PML4_MASK))
@@ -53,26 +55,33 @@
 // === IMPORTS ===
 extern void	Error_Backtrace(Uint IP, Uint BP);
 extern tPAddr	gInitialPML4[512];
+extern void	Threads_SegFault(tVAddr Addr);
+extern char	_UsertextBase[];
 
 // === PROTOTYPES ===
 void	MM_InitVirt(void);
 //void	MM_FinishVirtualInit(void);
-void	MM_PageFault(tVAddr Addr, Uint ErrorCode, tRegs *Regs);
+void	MM_int_ClonePageEnt( Uint64 *Ent, void *NextLevel, tVAddr Addr, int bTable );
+ int	MM_PageFault(tVAddr Addr, Uint ErrorCode, tRegs *Regs);
+void	MM_int_DumpTablesEnt(tVAddr RangeStart, size_t Length, tPAddr Expected);
 void	MM_DumpTables(tVAddr Start, tVAddr End);
  int	MM_GetPageEntryPtr(tVAddr Addr, BOOL bTemp, BOOL bAllocate, BOOL bLargePage, tPAddr **Pointer);
  int	MM_MapEx(tVAddr VAddr, tPAddr PAddr, BOOL bTemp, BOOL bLarge);
 // int	MM_Map(tVAddr VAddr, tPAddr PAddr);
 void	MM_Unmap(tVAddr VAddr);
+void	MM_int_ClearTableLevel(tVAddr VAddr, int LevelBits, int MaxEnts);
 void	MM_ClearUser(void);
  int	MM_GetPageEntry(tVAddr Addr, tPAddr *Phys, Uint *Flags);
 
 // === GLOBALS ===
 tMutex	glMM_TempFractalLock;
+tPAddr	gMM_ZeroPage;
 
 // === CODE ===
 void MM_InitVirt(void)
 {
-	MM_DumpTables(0, -1L);
+//	Log_Debug("MMVirt", "&PAGEMAPLVL4(0) = %p", &PAGEMAPLVL4(0));
+//	MM_DumpTables(0, -1L);
 }
 
 void MM_FinishVirtualInit(void)
@@ -81,49 +90,121 @@ void MM_FinishVirtualInit(void)
 }
 
 /**
+ * \brief Clone a page from an entry
+ * \param Ent	Pointer to the entry in the PML4/PDP/PD/PT
+ * \param NextLevel	Pointer to contents of the entry
+ * \param Addr	Dest address
+ * \note Used in COW
+ */
+void MM_int_ClonePageEnt( Uint64 *Ent, void *NextLevel, tVAddr Addr, int bTable )
+{
+	tPAddr	curpage = *Ent & PADDR_MASK; 
+	if( MM_GetRefCount( curpage ) <= 0 ) {
+		Log_KernelPanic("MMVirt", "Page %P still marked COW, but unreferenced", curpage);
+	}
+	if( MM_GetRefCount( curpage ) == 1 )
+	{
+		*Ent &= ~PF_COW;
+		*Ent |= PF_PRESENT|PF_WRITE;
+//		Log_Debug("MMVirt", "COW ent at %p (%p) only %P", Ent, NextLevel, curpage);
+	}
+	else
+	{
+		void	*tmp;
+		tPAddr	paddr;
+		
+		if( !(paddr = MM_AllocPhys()) ) {
+			Threads_SegFault(Addr);
+			return ;
+		}
+
+		ASSERT(paddr != curpage);
+			
+		tmp = (void*)MM_MapTemp(paddr);
+		memcpy( tmp, NextLevel, 0x1000 );
+		MM_FreeTemp( (tVAddr)tmp );
+		
+//		Log_Debug("MMVirt", "COW ent at %p (%p) from %P to %P", Ent, NextLevel, curpage, paddr);
+
+		MM_DerefPhys( curpage );
+		*Ent &= PF_USER;
+		*Ent |= paddr|PF_PRESENT|PF_WRITE;
+	}
+	INVLPG( (tVAddr)NextLevel );
+	
+	// Mark COW on pages
+	if(bTable) 
+	{
+		Uint64	*dp = NextLevel;
+		 int	i;
+		for( i = 0; i < 512; i ++ )
+		{
+			if( !(dp[i] & PF_PRESENT) )	continue;
+			MM_RefPhys( dp[i] & PADDR_MASK );
+			if( dp[i] & PF_WRITE ) {
+				dp[i] &= ~PF_WRITE;
+				dp[i] |= PF_COW;
+			}
+		}
+	}
+}
+
+/*
  * \brief Called on a page fault
  */
-void MM_PageFault(tVAddr Addr, Uint ErrorCode, tRegs *Regs)
+int MM_PageFault(tVAddr Addr, Uint ErrorCode, tRegs *Regs)
 {
 	// TODO: Implement Copy-on-Write
-	#if 0
-	if( gaPageDir  [Addr>>22] & PF_PRESENT
-	 && gaPageTable[Addr>>12] & PF_PRESENT
-	 && gaPageTable[Addr>>12] & PF_COW )
+	#if 1
+	if( PAGEMAPLVL4(Addr>>39) & PF_PRESENT
+	 && PAGEDIRPTR (Addr>>30) & PF_PRESENT
+	 && PAGEDIR    (Addr>>21) & PF_PRESENT
+	 && PAGETABLE  (Addr>>12) & PF_PRESENT )
 	{
-		tPAddr	paddr;
-		if(MM_GetRefCount( gaPageTable[Addr>>12] & PADDR_MASK ) == 1)
+		// PML4 Entry
+		if( PAGEMAPLVL4(Addr>>39) & PF_COW )
 		{
-			gaPageTable[Addr>>12] &= ~PF_COW;
-			gaPageTable[Addr>>12] |= PF_PRESENT|PF_WRITE;
+			tPAddr	*dp = &PAGEDIRPTR((Addr>>39)*512);
+			MM_int_ClonePageEnt( &PAGEMAPLVL4(Addr>>39), dp, Addr, 1 );
+//			MM_DumpTables(Addr>>39 << 39, (((Addr>>39) + 1) << 39) - 1);
 		}
-		else
+		// PDP Entry
+		if( PAGEDIRPTR(Addr>>30) & PF_COW )
 		{
-			//Log("MM_PageFault: COW - MM_DuplicatePage(0x%x)", Addr);
-			paddr = MM_DuplicatePage( Addr );
-			MM_DerefPhys( gaPageTable[Addr>>12] & PADDR_MASK );
-			gaPageTable[Addr>>12] &= PF_USER;
-			gaPageTable[Addr>>12] |= paddr|PF_PRESENT|PF_WRITE;
+			tPAddr	*dp = &PAGEDIR( (Addr>>30)*512 );
+			MM_int_ClonePageEnt( &PAGEDIRPTR(Addr>>30), dp, Addr, 1 );
+//			MM_DumpTables(Addr>>30 << 30, (((Addr>>30) + 1) << 30) - 1);
+		}
+		// PD Entry
+		if( PAGEDIR(Addr>>21) & PF_COW )
+		{
+			tPAddr	*dp = &PAGETABLE( (Addr>>21)*512 );
+			MM_int_ClonePageEnt( &PAGEDIR(Addr>>21), dp, Addr, 1 );
+//			MM_DumpTables(Addr>>21 << 21, (((Addr>>21) + 1) << 21) - 1);
+		}
+		// PT Entry
+		if( PAGETABLE(Addr>>12) & PF_COW )
+		{
+			MM_int_ClonePageEnt( &PAGETABLE(Addr>>12), (void*)(Addr & ~0xFFF), Addr, 0 );
+			INVLPG( Addr & ~0xFFF );
+			return 0;
 		}
-		
-		INVLPG( Addr & ~0xFFF );
-		return;
 	}
 	#endif
 	
 	// If it was a user, tell the thread handler
 	if(ErrorCode & 4) {
-		Warning("%s %s %s memory%s",
-			(ErrorCode&4?"User":"Kernel"),
+		Warning("User %s %s memory%s",
 			(ErrorCode&2?"write to":"read from"),
 			(ErrorCode&1?"bad/locked":"non-present"),
 			(ErrorCode&16?" (Instruction Fetch)":"")
 			);
-		Warning("User Pagefault: Instruction at %04x:%08x accessed %p",
+		Warning("User Pagefault: Instruction at %04x:%p accessed %p",
 			Regs->CS, Regs->RIP, Addr);
 		__asm__ __volatile__ ("sti");	// Restart IRQs
-//		Threads_SegFault(Addr);
-		return ;
+		Error_Backtrace(Regs->RIP, Regs->RBP);
+		Threads_SegFault(Addr);
+		return 0;
 	}
 	
 	// Kernel #PF
@@ -133,23 +214,44 @@ void MM_PageFault(tVAddr Addr, Uint ErrorCode, tRegs *Regs)
 		Warning("Reserved Bits Trashed!");
 	else
 	{
-		Warning("%s %s %s memory%s",
-			(ErrorCode&4?"User":"Kernel"),
+		Warning("Kernel %s %s memory%s",
 			(ErrorCode&2?"write to":"read from"),
 			(ErrorCode&1?"bad/locked":"non-present"),
 			(ErrorCode&16?" (Instruction Fetch)":"")
 			);
 	}
 	
-	Log("Code at %p accessed %p", Regs->RIP, Addr);
+	Log("Thread %i - Code at %p accessed %p", Threads_GetTID(), Regs->RIP, Addr);
 	// Print Stack Backtrace
 	Error_Backtrace(Regs->RIP, Regs->RBP);
 	
 	MM_DumpTables(0, -1);
-	
-	__asm__ __volatile__ ("cli");
-	for( ;; )
-		HALT();
+
+	return 1;	
+}
+
+void MM_int_DumpTablesEnt(tVAddr RangeStart, size_t Length, tPAddr Expected)
+{
+	#define CANOICAL(addr)	((addr)&0x800000000000?(addr)|0xFFFF000000000000:(addr))
+	LogF("%016llx => ", CANOICAL(RangeStart));
+//	LogF("%6llx %6llx %6llx %016llx => ",
+//		MM_GetPhysAddr( (tVAddr)&PAGEDIRPTR(RangeStart>>30) ),
+//		MM_GetPhysAddr( (tVAddr)&PAGEDIR(RangeStart>>21) ),
+//		MM_GetPhysAddr( (tVAddr)&PAGETABLE(RangeStart>>12) ),
+//		CANOICAL(RangeStart)
+//		);
+	if( gMM_ZeroPage && (PAGETABLE(RangeStart>>12) & PADDR_MASK) == gMM_ZeroPage )
+		LogF("%13s", "zero" );
+	else
+		LogF("%13llx", PAGETABLE(RangeStart>>12) & PADDR_MASK );
+	LogF(" : 0x%6llx (%c%c%c%c)\r\n",
+		Length,
+		(Expected & PF_PAGED ? 'p' : '-'),
+		(Expected & PF_COW ? 'C' : '-'),
+		(Expected & PF_USER ? 'U' : '-'),
+		(Expected & PF_WRITE ? 'W' : '-')
+		);
+	#undef CANOICAL
 }
 
 /**
@@ -157,8 +259,7 @@ void MM_PageFault(tVAddr Addr, Uint ErrorCode, tRegs *Regs)
  */
 void MM_DumpTables(tVAddr Start, tVAddr End)
 {
-	#define CANOICAL(addr)	((addr)&0x800000000000?(addr)|0xFFFF000000000000:(addr))
-	const tPAddr	CHANGEABLE_BITS = 0xFF8;
+	const tPAddr	CHANGEABLE_BITS = ~(PF_PRESENT|PF_WRITE|PF_USER|PF_COW|PF_PAGED) & 0xFFF;
 	const tPAddr	MASK = ~CHANGEABLE_BITS;	// Physical address and access bits
 	tVAddr	rangeStart = 0;
 	tPAddr	expected = CHANGEABLE_BITS;	// CHANGEABLE_BITS is used because it's not a vaild value
@@ -175,50 +276,40 @@ void MM_DumpTables(tVAddr Start, tVAddr End)
 		page < End;
 		curPos += 0x1000, page++)
 	{
-		if( curPos == 0x800000000000L )
-			curPos = 0xFFFF800000000000L;
-		
 		//Debug("&PAGEMAPLVL4(%i page>>27) = %p", page>>27, &PAGEMAPLVL4(page>>27));
 		//Debug("&PAGEDIRPTR(%i page>>18) = %p", page>>18, &PAGEDIRPTR(page>>18));
 		//Debug("&PAGEDIR(%i page>>9) = %p", page>>9, &PAGEDIR(page>>9));
 		//Debug("&PAGETABLE(%i page) = %p", page, &PAGETABLE(page));
 		
 		// End of a range
-		if(
-			!(PAGEMAPLVL4(page>>27) & PF_PRESENT)
-		||	!(PAGEDIRPTR(page>>18) & PF_PRESENT)
-		||	!(PAGEDIR(page>>9) & PF_PRESENT)
-		||  !(PAGETABLE(page) & PF_PRESENT)
-		||  (PAGETABLE(page) & MASK) != expected)
+		if(!(PAGEMAPLVL4(page>>27) & PF_PRESENT)
+		|| !(PAGEDIRPTR(page>>18) & PF_PRESENT)
+		|| !(PAGEDIR(page>>9) & PF_PRESENT)
+		|| !(PAGETABLE(page) & PF_PRESENT)
+		|| (PAGETABLE(page) & MASK) != expected)
 		{			
-			if(expected != CHANGEABLE_BITS) {
-				Log("%016llx => %013llx : 0x%6llx (%c%c%c%c)",
-					CANOICAL(rangeStart),
-					PAGETABLE(rangeStart>>12) & PADDR_MASK,
-					curPos - rangeStart,
-					(expected & PF_PAGED ? 'p' : '-'),
-					(expected & PF_COW ? 'C' : '-'),
-					(expected & PF_USER ? 'U' : '-'),
-					(expected & PF_WRITE ? 'W' : '-')
-					);
+			if(expected != CHANGEABLE_BITS)
+			{
+				MM_int_DumpTablesEnt( rangeStart, curPos - rangeStart, expected );
 				expected = CHANGEABLE_BITS;
 			}
+			
+			if( curPos == 0x800000000000L )
+				curPos = 0xFFFF800000000000L;
+		
 			if( !(PAGEMAPLVL4(page>>27) & PF_PRESENT) ) {
 				page += (1 << 27) - 1;
 				curPos += (1L << 39) - 0x1000;
-				//Debug("pml4 ent unset (page = 0x%x now)", page);
 				continue;
 			}
 			if( !(PAGEDIRPTR(page>>18) & PF_PRESENT) ) {
 				page += (1 << 18) - 1;
 				curPos += (1L << 30) - 0x1000;
-				//Debug("pdp ent unset (page = 0x%x now)", page);
 				continue;
 			}
 			if( !(PAGEDIR(page>>9) & PF_PRESENT) ) {
 				page += (1 << 9) - 1;
 				curPos += (1L << 21) - 0x1000;
-				//Debug("pd ent unset (page = 0x%x now)", page);
 				continue;
 			}
 			if( !(PAGETABLE(page) & PF_PRESENT) )	continue;
@@ -226,23 +317,16 @@ void MM_DumpTables(tVAddr Start, tVAddr End)
 			expected = (PAGETABLE(page) & MASK);
 			rangeStart = curPos;
 		}
-		if(expected != CHANGEABLE_BITS)
+		if(gMM_ZeroPage && (expected & PADDR_MASK) == gMM_ZeroPage )
+			expected = expected;
+		else if(expected != CHANGEABLE_BITS)
 			expected += 0x1000;
 	}
 	
 	if(expected != CHANGEABLE_BITS) {
-		Log("%016llx => %013llx : 0x%6llx (%c%c%c%c)",
-			CANOICAL(rangeStart),
-			PAGETABLE(rangeStart>>12) & PADDR_MASK,
-			curPos - rangeStart,
-			(expected & PF_PAGED ? 'p' : '-'),
-			(expected & PF_COW ? 'C' : '-'),
-			(expected & PF_USER ? 'U' : '-'),
-			(expected & PF_WRITE ? 'W' : '-')
-			);
+		MM_int_DumpTablesEnt( rangeStart, curPos - rangeStart, expected );
 		expected = 0;
 	}
-	#undef CANOICAL
 }
 
 /**
@@ -258,10 +342,10 @@ int MM_GetPageEntryPtr(tVAddr Addr, BOOL bTemp, BOOL bAllocate, BOOL bLargePage,
 {
 	tPAddr	*pmlevels[4];
 	tPAddr	tmp;
-	const int	ADDR_SIZES[] = {39, 30, 21, 12};
-	const int	nADDR_SIZES = sizeof(ADDR_SIZES)/sizeof(ADDR_SIZES[0]);
-	 int	i;
+	 int	i, size;
 	
+	#define BITMASK(bits)	( (1LL << (bits))-1 )
+
 	if( bTemp )
 	{
 		pmlevels[3] = &TMPTABLE(0);	// Page Table
@@ -272,47 +356,52 @@ int MM_GetPageEntryPtr(tVAddr Addr, BOOL bTemp, BOOL bAllocate, BOOL bLargePage,
 	else
 	{
 		pmlevels[3] = (void*)MM_FRACTAL_BASE;	// Page Table
-		pmlevels[2] = &pmlevels[3][(MM_FRACTAL_BASE>>12)&PAGE_MASK];	// PDIR
-		pmlevels[1] = &pmlevels[2][(MM_FRACTAL_BASE>>21)&TABLE_MASK];	// PDPT
-		pmlevels[0] = &pmlevels[1][(MM_FRACTAL_BASE>>30)&PDP_MASK];	// PML4
+		pmlevels[2] = &pmlevels[3][(MM_FRACTAL_BASE>>12)&BITMASK(VIRT_BITS-12)];	// PDIR
+		pmlevels[1] = &pmlevels[2][(MM_FRACTAL_BASE>>21)&BITMASK(VIRT_BITS-21)];	// PDPT
+		pmlevels[0] = &pmlevels[1][(MM_FRACTAL_BASE>>30)&BITMASK(VIRT_BITS-30)];	// PML4
 	}
 	
 	// Mask address
 	Addr &= (1ULL << 48)-1;
 	
-	for( i = 0; i < nADDR_SIZES-1; i ++ )
+	for( size = 39, i = 0; size > 12; size -= 9, i ++ )
 	{
+		Uint64	*ent = &pmlevels[i][Addr >> size];
 //		INVLPG( &pmlevels[i][ (Addr >> ADDR_SIZES[i]) & 
 		
-		// Check for a large page
-		if( (Addr & ((1ULL << ADDR_SIZES[i])-1)) == 0 && bLargePage )
+		// Check for a free large page slot
+		// TODO: Better support with selectable levels
+		if( (Addr & ((1ULL << size)-1)) == 0 && bLargePage )
 		{
-			if(Pointer)	*Pointer = &pmlevels[i][Addr >> ADDR_SIZES[i]];
-			return ADDR_SIZES[i];
+			if(Pointer)	*Pointer = ent;
+			return size;
 		}
 		// Allocate an entry if required
-		if( !(pmlevels[i][Addr >> ADDR_SIZES[i]] & 1) )
+		if( !(*ent & PF_PRESENT) )
 		{
 			if( !bAllocate )	return -4;	// If allocation is not requested, error
-			tmp = MM_AllocPhys();
-			if(!tmp)	return -2;
-			pmlevels[i][Addr >> ADDR_SIZES[i]] = tmp | 3;
-			INVLPG( &pmlevels[i+1][ (Addr>>ADDR_SIZES[i])*512 ] );
-			memset( &pmlevels[i+1][ (Addr>>ADDR_SIZES[i])*512 ], 0, 0x1000 );
+			if( !(tmp = MM_AllocPhys()) )	return -2;
+			*ent = tmp | 3;
+			if( Addr < 0x800000000000 )
+				*ent |= PF_USER;
+			INVLPG( &pmlevels[i+1][ (Addr>>size)*512 ] );
+			memset( &pmlevels[i+1][ (Addr>>size)*512 ], 0, 0x1000 );
+			LOG("Init PML%i ent 0x%x %p with %P", 4 - i,
+				Addr>>size, (Addr>>size) << size, tmp);
 		}
 		// Catch large pages
-		else if( pmlevels[i][Addr >> ADDR_SIZES[i]] & PF_LARGE )
+		else if( *ent & PF_LARGE )
 		{
 			// Alignment
-			if( (Addr & ((1ULL << ADDR_SIZES[i])-1)) != 0 )	return -3;
-			if(Pointer)	*Pointer = &pmlevels[i][Addr >> ADDR_SIZES[i]];
-			return ADDR_SIZES[i];	// Large page warning
+			if( (Addr & ((1ULL << size)-1)) != 0 )	return -3;
+			if(Pointer)	*Pointer = ent;
+			return size;	// Large page warning
 		}
 	}
 	
 	// And, set the page table entry
-	if(Pointer)	*Pointer = &pmlevels[i][Addr >> ADDR_SIZES[i]];
-	return ADDR_SIZES[i];
+	if(Pointer)	*Pointer = &pmlevels[i][Addr >> size];
+	return size;
 }
 
 /**
@@ -327,7 +416,7 @@ int MM_MapEx(tVAddr VAddr, tPAddr PAddr, BOOL bTemp, BOOL bLarge)
 	tPAddr	*ent;
 	 int	rv;
 	
-	ENTER("xVAddr xPAddr", VAddr, PAddr);
+	ENTER("pVAddr PPAddr", VAddr, PAddr);
 	
 	// Get page pointer (Allow allocating)
 	rv = MM_GetPageEntryPtr(VAddr, bTemp, 1, bLarge, &ent);
@@ -336,7 +425,10 @@ int MM_MapEx(tVAddr VAddr, tPAddr PAddr, BOOL bTemp, BOOL bLarge)
 	if( *ent & 1 )	LEAVE_RET('i', 0);
 	
 	*ent = PAddr | 3;
-	
+
+	if( VAddr < 0x800000000000 )
+		*ent |= PF_USER;
+
 	INVLPG( VAddr );
 
 	LEAVE('i', 1);	
@@ -364,7 +456,7 @@ void MM_Unmap(tVAddr VAddr)
 	if( !(PAGEDIRPTR(VAddr >> 30) & 1) )	return ;
 	// Check Page Dir
 	if( !(PAGEDIR(VAddr >> 21) & 1) )	return ;
-	
+
 	PAGETABLE(VAddr >> PTAB_SHIFT) = 0;
 	INVLPG( VAddr );
 }
@@ -398,6 +490,26 @@ tPAddr MM_Allocate(tVAddr VAddr)
 	return ret;
 }
 
+tPAddr MM_AllocateZero(tVAddr VAddr)
+{
+	tPAddr	ret = gMM_ZeroPage;
+	
+	MM_GetPageEntryPtr(VAddr, 0, 1, 0, NULL);
+
+	if(!gMM_ZeroPage) {
+		ret = gMM_ZeroPage = MM_AllocPhys();
+		MM_RefPhys(ret);	// Don't free this please
+		MM_Map(VAddr, ret);
+		memset((void*)VAddr, 0, 0x1000);
+	}
+	else {
+		MM_Map(VAddr, ret);
+	}
+	MM_RefPhys(ret);	// Refernce for this map
+	MM_SetFlags(VAddr, MM_PFLAG_COW, MM_PFLAG_COW);
+	return ret;
+}
+
 /**
  * \brief Deallocate a page at a virtual address
  */
@@ -446,6 +558,8 @@ tPAddr MM_GetPhysAddr(tVAddr Addr)
 	ret = MM_GetPageEntryPtr(Addr, 0, 0, 0, &ptr);
 	if( ret < 0 )	return 0;
 	
+	if( !(*ptr & 1) )	return 0;
+	
 	return (*ptr & PADDR_MASK) | (Addr & 0xFFF);
 }
 
@@ -554,6 +668,8 @@ tVAddr MM_MapHWPages(tPAddr PAddr, Uint Number)
 		}
 		if( num >= 0 )	continue;
 		
+//		Log_Debug("MMVirt", "Mapping %i pages to %p (base %P)", Number, ret-Number*0x1000, PAddr);
+
 		PAddr += 0x1000 * Number;
 		
 		while( Number -- )
@@ -561,12 +677,13 @@ tVAddr MM_MapHWPages(tPAddr PAddr, Uint Number)
 			ret -= 0x1000;
 			PAddr -= 0x1000;
 			MM_Map(ret, PAddr);
+			MM_RefPhys(PAddr);
 		}
 		
 		return ret;
 	}
 	
-	Log_KernelPanic("MM", "TODO: Implement MM_MapHWPages");
+	Log_Error("MM", "MM_MapHWPages - No space for %i pages", Number);
 	return 0;
 }
 
@@ -578,6 +695,7 @@ void MM_UnmapHWPages(tVAddr VAddr, Uint Number)
 //	Log_KernelPanic("MM", "TODO: Implement MM_UnmapHWPages");
 	while( Number -- )
 	{
+		MM_DerefPhys( MM_GetPhysAddr(VAddr) );
 		MM_Unmap(VAddr);
 		VAddr += 0x1000;
 	}
@@ -606,10 +724,7 @@ tVAddr MM_AllocDMA(int Pages, int MaxBits, tPAddr *PhysAddr)
 		phys = MM_AllocPhys();
 		*PhysAddr = phys;
 		ret = MM_MapHWPages(phys, 1);
-		if(ret == 0) {
-			MM_DerefPhys(phys);
-			return 0;
-		}
+		MM_DerefPhys(phys);
 		return ret;
 	}
 	
@@ -620,10 +735,11 @@ tVAddr MM_AllocDMA(int Pages, int MaxBits, tPAddr *PhysAddr)
 	
 	// Allocated successfully, now map
 	ret = MM_MapHWPages(phys, Pages);
+	// MapHWPages references the pages, so deref them back down to 1
+	for(;Pages--;phys+=0x1000)
+		MM_DerefPhys(phys);
 	if( ret == 0 ) {
 		// If it didn't map, free then return 0
-		for(;Pages--;phys+=0x1000)
-			MM_DerefPhys(phys);
 		return 0;
 	}
 	
@@ -634,13 +750,31 @@ tVAddr MM_AllocDMA(int Pages, int MaxBits, tPAddr *PhysAddr)
 // --- Tempory Mappings ---
 tVAddr MM_MapTemp(tPAddr PAddr)
 {
-	Log_KernelPanic("MM", "TODO: Implement MM_MapTemp");
+	const int max_slots = (MM_TMPMAP_END - MM_TMPMAP_BASE) / PAGE_SIZE;
+	tVAddr	ret = MM_TMPMAP_BASE;
+	 int	i;
+	
+	for( i = 0; i < max_slots; i ++, ret += PAGE_SIZE )
+	{
+		tPAddr	*ent;
+		if( MM_GetPageEntryPtr( ret, 0, 1, 0, &ent) < 0 ) {
+			continue ;
+		}
+
+		if( *ent & 1 )
+			continue ;
+
+		*ent = PAddr | 3;
+		MM_RefPhys(PAddr);
+		INVLPG(ret);
+		return ret;
+	}
 	return 0;
 }
 
 void MM_FreeTemp(tVAddr VAddr)
 {
-	Log_KernelPanic("MM", "TODO: Implement MM_FreeTemp");
+	MM_Deallocate(VAddr);
 	return ;
 }
 
@@ -650,8 +784,8 @@ tPAddr MM_Clone(void)
 {
 	tPAddr	ret;
 	 int	i;
-	tVAddr	kstackbase = Proc_GetCurThread()->KernelStack - KERNEL_STACK_SIZE + 0x1000;
-	
+	tVAddr	kstackbase;
+
 	// #1 Create a copy of the PML4
 	ret = MM_AllocPhys();
 	if(!ret)	return 0;
@@ -661,19 +795,19 @@ tPAddr MM_Clone(void)
 	TMPCR3() = ret | 3;
 	INVLPG_ALL();
 	
-//	Log_KernelPanic("MM", "TODO: Implement MM_Clone");
-	
 	// #3 Set Copy-On-Write to all user pages
 	for( i = 0; i < 256; i ++)
 	{
+		if( PAGEMAPLVL4(i) & PF_WRITE ) {
+			PAGEMAPLVL4(i) |= PF_COW;
+			PAGEMAPLVL4(i) &= ~PF_WRITE;
+		}
+
 		TMPMAPLVL4(i) = PAGEMAPLVL4(i);
 //		Log_Debug("MM", "TMPMAPLVL4(%i) = 0x%016llx", i, TMPMAPLVL4(i));
-		if( TMPMAPLVL4(i) & 1 )
-		{
-			MM_RefPhys( TMPMAPLVL4(i) & PADDR_MASK );
-			TMPMAPLVL4(i) |= PF_COW;
-			TMPMAPLVL4(i) &= ~PF_WRITE;
-		}
+		if( !(TMPMAPLVL4(i) & PF_PRESENT) )	continue ;
+		
+		MM_RefPhys( TMPMAPLVL4(i) & PADDR_MASK );
 	}
 	
 	// #4 Map in kernel pages
@@ -683,115 +817,89 @@ tPAddr MM_Clone(void)
 		// 320 0xFFFFA....	- Kernel Stacks
 		if( i == 320 )	continue;
 		// 509 0xFFFFFE0..	- Fractal mapping
-		if( i == 509 )	continue;
+		if( i == 508 )	continue;
 		// 510 0xFFFFFE8..	- Temp fractal mapping
-		if( i == 510 )	continue;
+		if( i == 509 )	continue;
+		
+		TMPMAPLVL4(i) = PAGEMAPLVL4(i);
+		if( TMPMAPLVL4(i) & 1 )
+			MM_RefPhys( TMPMAPLVL4(i) & PADDR_MASK );
 	}
+
+	// Mark Per-Process data as COW
+	TMPMAPLVL4(MM_PPD_BASE>>39) |= PF_COW;
+	TMPMAPLVL4(MM_PPD_BASE>>39) &= ~PF_WRITE;
 	
 	// #5 Set fractal mapping
-	TMPMAPLVL4(509) = ret | 3;
-	TMPMAPLVL4(510) = 0;	// Temp
+	TMPMAPLVL4(MM_FRACTAL_BASE>>39) = ret | 3;	// Main
+	TMPMAPLVL4(MM_TMPFRAC_BASE>>39) = 0;	// Temp
 	
 	// #6 Create kernel stack
-	TMPMAPLVL4(320) = 0;
-	for( i = 0; i < KERNEL_STACK_SIZE/0x1000-1; i ++ )
+	//  tThread->KernelStack is the top
+	//  There is 1 guard page below the stack
+	kstackbase = Proc_GetCurThread()->KernelStack - KERNEL_STACK_SIZE;
+
+	// Clone stack
+	TMPMAPLVL4(MM_KSTACK_BASE >> PML4_SHIFT) = 0;
+	for( i = 1; i < KERNEL_STACK_SIZE/0x1000; i ++ )
 	{
 		tPAddr	phys = MM_AllocPhys();
 		tVAddr	tmpmapping;
 		MM_MapEx(kstackbase+i*0x1000, phys, 1, 0);
 		
 		tmpmapping = MM_MapTemp(phys);
-		memcpy((void*)tmpmapping, (void*)(kstackbase+i*0x1000), 0x1000);
+		if( MM_GetPhysAddr( kstackbase+i*0x1000 ) )
+			memcpy((void*)tmpmapping, (void*)(kstackbase+i*0x1000), 0x1000);
+		else
+			memset((void*)tmpmapping, 0, 0x1000);
+//		if( i == 0xF )
+//			Debug_HexDump("MM_Clone: *tmpmapping = ", (void*)tmpmapping, 0x1000);
 		MM_FreeTemp(tmpmapping);
 	}
 	
+//	MAGIC_BREAK();
+
 	// #7 Return
 	TMPCR3() = 0;
 	INVLPG_ALL();
 	Mutex_Release(&glMM_TempFractalLock);
+//	Log("MM_Clone: RETURN %P", ret);
 	return ret;
 }
 
-void MM_ClearUser(void)
+void MM_int_ClearTableLevel(tVAddr VAddr, int LevelBits, int MaxEnts)
 {
-	tVAddr	addr = 0;
-	 int	pml4, pdpt, pd, pt;
-	
-	for( pml4 = 0; pml4 < 256; pml4 ++ )
+	Uint64	* const table_bases[] = {&PAGETABLE(0), &PAGEDIR(0), &PAGEDIRPTR(0), &PAGEMAPLVL4(0)};
+	Uint64	*table = table_bases[(LevelBits-12)/9] + (VAddr >> LevelBits);
+	 int	i;
+//	Log("MM_int_ClearTableLevel: (VAddr=%p, LevelBits=%i, MaxEnts=%i)", VAddr, LevelBits, MaxEnts);
+	for( i = 0; i < MaxEnts; i ++ )
 	{
-		// Catch an un-allocated PML4 entry
-		if( !(PAGEMAPLVL4(pml4) & 1) ) {
-			addr += 1ULL << PML4_SHIFT;
+		// Skip non-present tables
+		if( !(table[i] & PF_PRESENT) ) {
+			table[i] = 0;
 			continue ;
 		}
-		
-		// Catch a large COW
-		if( (PAGEMAPLVL4(pml4) & PF_COW) ) {
-			addr += 1ULL << PML4_SHIFT;
-		}
-		else
-		{
-			// TODO: Large pages
-			
-			// Child entries
-			for( pdpt = 0; pdpt < 512; pdpt ++ )
-			{
-				// Unallocated
-				if( !(PAGEDIRPTR(addr >> PDP_SHIFT) & 1) ) {
-					addr += 1ULL << PDP_SHIFT;
-					continue;
-				}
-			
-				// Catch a large COW
-				if( (PAGEDIRPTR(addr >> PDP_SHIFT) & PF_COW) ) {
-					addr += 1ULL << PDP_SHIFT;
-				}
-				else {
-					// Child entries
-					for( pd = 0; pd < 512; pd ++ )
-					{
-						// Unallocated PDir entry
-						if( !(PAGEDIR(addr >> PDIR_SHIFT) & 1) ) {
-							addr += 1ULL << PDIR_SHIFT;
-							continue;
-						}
-						
-						// COW Page Table
-						if( PAGEDIR(addr >> PDIR_SHIFT) & PF_COW ) {
-							addr += 1ULL << PDIR_SHIFT;
-						}
-						else
-						{
-							// TODO: Catch large pages
-							
-							// Child entries
-							for( pt = 0; pt < 512; pt ++ )
-							{
-								// Free page
-								if( PAGETABLE(addr >> PTAB_SHIFT) & 1 ) {
-									MM_DerefPhys( PAGETABLE(addr >> PTAB_SHIFT) & PADDR_MASK );
-									PAGETABLE(addr >> PTAB_SHIFT) = 0;
-								}
-								addr += 1ULL << 12;
-							}
-						}
-						// Free page table
-						MM_DerefPhys( PAGEDIR(addr >> PDIR_SHIFT) & PADDR_MASK );
-						PAGEDIR(addr >> PDIR_SHIFT) = 0;
-					}
-				}
-				// Free page directory
-				MM_DerefPhys( PAGEDIRPTR(addr >> PDP_SHIFT) & PADDR_MASK );
-				PAGEDIRPTR(addr >> PDP_SHIFT) = 0;
-			}
+	
+		if( (table[i] & PF_COW) && MM_GetRefCount(table[i] & PADDR_MASK) > 1 ) {
+			MM_DerefPhys(table[i] & PADDR_MASK);
+			table[i] = 0;
+			continue ;
 		}
-		// Free page directory pointer table (PML4 entry)
-		MM_DerefPhys( PAGEMAPLVL4(pml4) & PADDR_MASK );
-		PAGEMAPLVL4(pml4) = 0;
+		// Clear table contents (if it is a table)
+		if( LevelBits > 12 )
+			MM_int_ClearTableLevel(VAddr + ((tVAddr)i << LevelBits), LevelBits-9, 512);
+		MM_DerefPhys(table[i] & PADDR_MASK);
+		table[i] = 0;
 	}
 }
 
-tVAddr MM_NewWorkerStack(void)
+void MM_ClearUser(void)
+{
+	MM_int_ClearTableLevel(0, 39, 256);	
+}
+
+tVAddr MM_NewWorkerStack(void *StackData, size_t StackSize)
 {
 	tVAddr	ret;
 	 int	i;
@@ -803,7 +911,9 @@ tVAddr MM_NewWorkerStack(void)
 	// #2 Scan for a free stack addresss < 2^47
 	for(ret = 0x100000; ret < (1ULL << 47); ret += KERNEL_STACK_SIZE)
 	{
-		if( MM_GetPhysAddr(ret) == 0 )	break;
+		tPAddr	*ptr;
+		if( MM_GetPageEntryPtr(ret, 1, 0, 0, &ptr) <= 0 )	break;
+		if( !(*ptr & 1) )	break;
 	}
 	if( ret >= (1ULL << 47) ) {
 		Mutex_Release(&glMM_TempFractalLock);
@@ -822,6 +932,19 @@ tVAddr MM_NewWorkerStack(void)
 		}
 		MM_MapEx(ret + i*0x1000, phys, 1, 0);
 	}
+
+	if( StackSize > 0x1000 ) {
+		Log_Error("MM", "MM_NewWorkerStack: StackSize(0x%x) > 0x1000, cbf handling", StackSize);
+	}
+	else {
+		tPAddr	*ptr, paddr;
+		tVAddr	tmp_addr;
+		MM_GetPageEntryPtr(ret + i*0x1000, 1, 0, 0, &ptr);
+		paddr = *ptr & ~0xFFF;
+		tmp_addr = MM_MapTemp(paddr);
+		memcpy( (void*)(tmp_addr + (0x1000 - StackSize)), StackData, StackSize );
+		MM_FreeTemp(tmp_addr);
+	}
 	
 	Mutex_Release(&glMM_TempFractalLock);
 	
@@ -837,11 +960,11 @@ tVAddr MM_NewKStack(void)
 	Uint	i;
 	for( ; base < MM_KSTACK_TOP; base += KERNEL_STACK_SIZE )
 	{
-		if(MM_GetPhysAddr(base) != 0)
+		if(MM_GetPhysAddr(base+KERNEL_STACK_SIZE-0x1000) != 0)
 			continue;
 		
 		//Log("MM_NewKStack: Found one at %p", base + KERNEL_STACK_SIZE);
-		for( i = 0; i < KERNEL_STACK_SIZE; i += 0x1000)
+		for( i = 0x1000; i < KERNEL_STACK_SIZE; i += 0x1000)
 		{
 			if( !MM_Allocate(base+i) )
 			{