[PATCH 1/4] Pte drop ptep_get_and_clear paravirt op.patch

[PATCH 1/4] Pte drop ptep_get_and_clear paravirt op.patch

[Zachary Amsden]

In shadow mode hypervisors, ptep_get_and_clear achieves the desired
purpose of keeping the shadows in sync by issuing a native_get_and_clear,
followed by a call to pte_update, which indicates the PTE has been
modified.

Direct mode hypervisors (Xen) have no need for this anyway, and will trap
the update using writable pagetables.

This means no hypervisor makes use of ptep_get_and_clear; there is no
reason to have it in the paravirt-ops structure.  Change confusing
terminology about raw vs. native functions into consistent use of
native_pte_xxx for operations which do not invoke paravirt-ops.

Signed-off-by: Zachary Amsden <zach@vmware.com>

diff -r c02c6f5e882c arch/i386/kernel/paravirt.c
--- a/arch/i386/kernel/paravirt.c	Wed Apr 11 16:25:09 2007 -0700
+++ b/arch/i386/kernel/paravirt.c	Wed Apr 11 17:09:55 2007 -0700
@@ -315,8 +315,6 @@ struct paravirt_ops paravirt_ops = {
 	.pte_update = paravirt_nop,
 	.pte_update_defer = paravirt_nop,
 
-	.ptep_get_and_clear = native_ptep_get_and_clear,
-
 #ifdef CONFIG_HIGHPTE
 	.kmap_atomic_pte = kmap_atomic,
 #endif
diff -r c02c6f5e882c include/asm-i386/paravirt.h
--- a/include/asm-i386/paravirt.h	Wed Apr 11 16:25:09 2007 -0700
+++ b/include/asm-i386/paravirt.h	Wed Apr 11 17:12:03 2007 -0700
@@ -187,8 +187,6 @@ struct paravirt_ops
 	void (*pte_update)(struct mm_struct *mm, unsigned long addr, pte_t *ptep);
 	void (*pte_update_defer)(struct mm_struct *mm,
 				 unsigned long addr, pte_t *ptep);
-
- 	pte_t (*ptep_get_and_clear)(pte_t *ptep);
 
 #ifdef CONFIG_HIGHPTE
 	void *(*kmap_atomic_pte)(struct page *page, enum km_type type);
@@ -859,12 +857,8 @@ static inline void pmd_clear(pmd_t *pmdp
 	PVOP_VCALL1(pmd_clear, pmdp);
 }
 
-static inline pte_t raw_ptep_get_and_clear(pte_t *p)
-{
-	unsigned long long val = PVOP_CALL1(unsigned long long, ptep_get_and_clear, p);
-	return (pte_t) { val, val >> 32 };
-}
 #else  /* !CONFIG_X86_PAE */
+
 static inline pte_t __pte(unsigned long val)
 {
 	return (pte_t) { PVOP_CALL1(unsigned long, make_pte, val) };
@@ -899,11 +893,6 @@ static inline void set_pmd(pmd_t *pmdp, 
 static inline void set_pmd(pmd_t *pmdp, pmd_t pmdval)
 {
 	PVOP_VCALL2(set_pmd, pmdp, pmdval.pud.pgd.pgd);
-}
-
-static inline pte_t raw_ptep_get_and_clear(pte_t *p)
-{
-	return (pte_t) { PVOP_CALL1(unsigned long, ptep_get_and_clear, p) };
 }
 #endif	/* CONFIG_X86_PAE */
 
diff -r c02c6f5e882c include/asm-i386/pgtable.h
--- a/include/asm-i386/pgtable.h	Wed Apr 11 16:25:09 2007 -0700
+++ b/include/asm-i386/pgtable.h	Wed Apr 11 17:11:22 2007 -0700
@@ -265,8 +265,6 @@ static inline pte_t pte_mkhuge(pte_t pte
  */
 #define pte_update(mm, addr, ptep)		do { } while (0)
 #define pte_update_defer(mm, addr, ptep)	do { } while (0)
-
-#define raw_ptep_get_and_clear(xp)     native_ptep_get_and_clear(xp)
 #endif
 
 /*
@@ -340,7 +338,7 @@ do {									\
 #define __HAVE_ARCH_PTEP_GET_AND_CLEAR
 static inline pte_t ptep_get_and_clear(struct mm_struct *mm, unsigned long addr, pte_t *ptep)
 {
-	pte_t pte = raw_ptep_get_and_clear(ptep);
+	pte_t pte = native_ptep_get_and_clear(ptep);
 	pte_update(mm, addr, ptep);
 	return pte;
 }

Re: [PATCH 1/4] Pte drop ptep_get_and_clear paravirt op.patch

[Jeremy Fitzhardinge]

Zachary Amsden wrote:
> In shadow mode hypervisors, ptep_get_and_clear achieves the desired
> purpose of keeping the shadows in sync by issuing a native_get_and_clear,
> followed by a call to pte_update, which indicates the PTE has been
> modified.
>
> Direct mode hypervisors (Xen) have no need for this anyway, and will trap
> the update using writable pagetables.
>
> This means no hypervisor makes use of ptep_get_and_clear; there is no
> reason to have it in the paravirt-ops structure.  Change confusing
> terminology about raw vs. native functions into consistent use of
> native_pte_xxx for operations which do not invoke paravirt-ops.
>   

Looks OK to me.  Are you going to put this into the pvops queue?

Acked-by: Jeremy Fitzhardinge <jeremy@xensource.com>

    J

Re: [PATCH 1/4] Pte drop ptep_get_and_clear paravirt op.patch

[Zachary Amsden]

Jeremy Fitzhardinge wrote:
> Zachary Amsden wrote:
>   
>> In shadow mode hypervisors, ptep_get_and_clear achieves the desired
>> purpose of keeping the shadows in sync by issuing a native_get_and_clear,
>> followed by a call to pte_update, which indicates the PTE has been
>> modified.
>>
>> Direct mode hypervisors (Xen) have no need for this anyway, and will trap
>> the update using writable pagetables.
>>
>> This means no hypervisor makes use of ptep_get_and_clear; there is no
>> reason to have it in the paravirt-ops structure.  Change confusing
>> terminology about raw vs. native functions into consistent use of
>> native_pte_xxx for operations which do not invoke paravirt-ops.
>>   
>>     
>
> Looks OK to me.  Are you going to put this into the pvops queue?
>
> Acked-by: Jeremy Fitzhardinge <jeremy@xensource.com>

Yes, the rest are still undergoing feedback.  I'll kick this in and 
fixup the damage it causes.

Zach

Re: [PATCH 1/4] Pte drop ptep_get_and_clear paravirt op.patch

[Jeremy Fitzhardinge]

Zachary Amsden wrote:
> In shadow mode hypervisors, ptep_get_and_clear achieves the desired
> purpose of keeping the shadows in sync by issuing a native_get_and_clear,
> followed by a call to pte_update, which indicates the PTE has been
> modified.
>
> Direct mode hypervisors (Xen) have no need for this anyway, and will trap
> the update using writable pagetables.
>   

Jan Beulich just posted a patch for Xen/linux which specialzes
ptep_get_and_clear, which apparently improves kernel-compile performance
by 25-30% on some configurations.  I think we'll want to keep this.

    J

Re: [PATCH 1/4] Pte drop ptep_get_and_clear paravirt op.patch

[Zachary Amsden]

Jeremy Fitzhardinge wrote:
> Zachary Amsden wrote:
>   
>> In shadow mode hypervisors, ptep_get_and_clear achieves the desired
>> purpose of keeping the shadows in sync by issuing a native_get_and_clear,
>> followed by a call to pte_update, which indicates the PTE has been
>> modified.
>>
>> Direct mode hypervisors (Xen) have no need for this anyway, and will trap
>> the update using writable pagetables.
>>   
>>     
>
> Jan Beulich just posted a patch for Xen/linux which specialzes
> ptep_get_and_clear, which apparently improves kernel-compile performance
> by 25-30% on some configurations.  I think we'll want to keep this.
>   

Are you sure that it still wins even with these patches?  I can't see 
ptep_get_and_clear getting much faster than a pure non-emulated, and the 
stress case which wins 25-30% is fork/exit being able to drop the pte 
updates in zap_pte_range from trapping and / or going into a pte update 
queue.  That is what my patches addressed for shadow paging performance 
- the call to pte_update can be dropped - but you should already be 
getting this benefit for Xen, since there pte_update is a nop.  Are you 
sure you are unpinning the page tables and mapping them back as writable 
pages prior to address space destruction?

Or is there another case which exercises zap_pte_range via munmap?  In 
which case, you might be able to do things faster than trap and emulate 
with a hypercall, but a 25-30% speedup sounds rather para-normal for 
this type of change.  I would like to see a patch and some basic data 
showing the speedup if possible.

I certainly have no objection to adding back ptep_get_and_clear hook, 
but we need to fix the naming to be consistent (raw vs. native), which 
is one thing the patch (dropping the hook) is trying to do.  We can 
always re-add specialization for ptep_get_and_clear; the question is 
what level is appropriate.  Do you allow ptep_get_and_clear itself to be 
redefined or allow redefine of native_ptep_get_and_clear?

One line of reasoning - for consistency, it seems that 
native_ptep_get_and_clear should be just that, a non-paravirtualized pte 
udpate.  Then it follows we should not allow native_ptep_get_and_clear 
to be overridden.  In this case, for cleanliness is it better to add a 
midlevel indirection; use raw_ptep_get_and_clear as the paravirt-hook, 
which is called from ptep_get_and_clear, and allow access to the 
non-paravirtualized native functions with native_ptep_get_and_clear.  If 
that sounds agreeable, then this patch should be backed out.  I think 
backing it out will damage the immediately following patches due to 
sensitivity from code proximity.

Maybe the best strategy is to just re-add ptep_get_and_clear as a 
paravirt-op after the whole set of patches and before Jan's speedup patch?

Zach

Re: [PATCH 1/4] Pte drop ptep_get_and_clear paravirt op.patch

[Jeremy Fitzhardinge]

[-- Attachment #1: Type: text/plain, Size: 1371 bytes --]

Zachary Amsden wrote:
> Are you sure that it still wins even with these patches?  I can't see
> ptep_get_and_clear getting much faster than a pure non-emulated, and
> the stress case which wins 25-30% is fork/exit being able to drop the
> pte updates in zap_pte_range from trapping and / or going into a pte
> update queue.  That is what my patches addressed for shadow paging
> performance - the call to pte_update can be dropped - but you should
> already be getting this benefit for Xen, since there pte_update is a
> nop.  Are you sure you are unpinning the page tables and mapping them
> back as writable pages prior to address space destruction?

I'll ask Jan for more details later today, but he did this work on
xen-unstable.  The specific change is that it only bothers to update the
pte if it wasn't already not-present.  I guess that's a generally
applicable optimisation, though it wouldn't do much for native.

> Maybe the best strategy is to just re-add ptep_get_and_clear as a
> paravirt-op after the whole set of patches and before Jan's speedup
> patch? 

Yes, I'm happy with native_ptep_get_and_clear to be a direct pagetable
write for pulldown, and then use ptep_get_and_clear as a hook for
manipulating a live pagetable.  I don't see any particular need to put
Jan's patch into the pvops Xen right now, but it would be nice to have
the option.

    J

[-- Attachment #2: jan.patch --]
[-- Type: text/x-patch, Size: 18524 bytes --]

changeset:   14731:14c25e48a557
user:        kfraser@localhost.localdomain
date:        Thu Apr 05 09:10:33 2007 +0100
files:       linux-2.6-xen-sparse/include/asm-i386/mach-xen/asm/pgtable-2level.h linux-2.6-xen-sparse/include/asm-i386/mach-xen/asm/pgtable-3level.h linux-2.6-xen-sparse/include/asm-i386/mach-xen/asm/pgtable.h linux-2.6-xen-sparse/include/asm-x86_64/mach-xen/asm/pgtable.h
description:
linux: improve x86 page table handling performance

Where possible,
- use hypercalls instead of writing to read-only pages
- fold TLB flushes into page table update hypercalls
- on PAE, use single-access updates instead of two-access ones

The single change to PAE pte_clear() yields a 25-30% boost for kernel
builds on a 4x2x2 CPUs, 8Gb box; the other changes together yield
improvements of 2-5%.

Also, adjust backward compatibility handling in a few more places.

Signed-off-by: Jan Beulich <jbeulich@novell.com>


diff -r 07d3208c0ca3 -r 14c25e48a557 linux-2.6-xen-sparse/include/asm-i386/mach-xen/asm/pgtable-2level.h
--- a/linux-2.6-xen-sparse/include/asm-i386/mach-xen/asm/pgtable-2level.h	Thu Apr 05 08:59:12 2007 +0100
+++ b/linux-2.6-xen-sparse/include/asm-i386/mach-xen/asm/pgtable-2level.h	Thu Apr 05 09:10:33 2007 +0100
@@ -36,8 +36,37 @@
 #define pte_clear(mm,addr,xp)	do { set_pte_at(mm, addr, xp, __pte(0)); } while (0)
 #define pmd_clear(xp)	do { set_pmd(xp, __pmd(0)); } while (0)
 
-#define ptep_get_and_clear(mm,addr,xp)	__pte_ma(xchg(&(xp)->pte_low, 0))
+#define pte_none(x) (!(x).pte_low)
+
+static inline pte_t ptep_get_and_clear(struct mm_struct *mm, unsigned long addr, pte_t *ptep)
+{
+	pte_t pte = *ptep;
+	if (!pte_none(pte)) {
+		if (mm != &init_mm)
+			pte = __pte_ma(xchg(&ptep->pte_low, 0));
+		else
+			HYPERVISOR_update_va_mapping(addr, __pte(0), 0);
+	}
+	return pte;
+}
+
+#define ptep_clear_flush(vma, addr, ptep)			\
+({								\
+	pte_t *__ptep = (ptep);					\
+	pte_t __res = *__ptep;					\
+	if (!pte_none(__res) &&					\
+	    ((vma)->vm_mm != current->mm ||			\
+	     HYPERVISOR_update_va_mapping(addr, __pte(0),	\
+			(unsigned long)(vma)->vm_mm->cpu_vm_mask.bits| \
+				UVMF_INVLPG|UVMF_MULTI))) {	\
+		__ptep->pte_low = 0;				\
+		flush_tlb_page(vma, addr);			\
+	}							\
+	__res;							\
+})
+
 #define pte_same(a, b)		((a).pte_low == (b).pte_low)
+
 #define __pte_mfn(_pte) ((_pte).pte_low >> PAGE_SHIFT)
 #define pte_mfn(_pte) ((_pte).pte_low & _PAGE_PRESENT ? \
 	__pte_mfn(_pte) : pfn_to_mfn(__pte_mfn(_pte)))
@@ -46,7 +75,6 @@
 
 #define pte_page(_pte) pfn_to_page(pte_pfn(_pte))
 
-#define pte_none(x)		(!(x).pte_low)
 #define pfn_pte(pfn, prot)	__pte(((pfn) << PAGE_SHIFT) | pgprot_val(prot))
 #define pfn_pmd(pfn, prot)	__pmd(((pfn) << PAGE_SHIFT) | pgprot_val(prot))
 
diff -r 07d3208c0ca3 -r 14c25e48a557 linux-2.6-xen-sparse/include/asm-i386/mach-xen/asm/pgtable-3level.h
--- a/linux-2.6-xen-sparse/include/asm-i386/mach-xen/asm/pgtable-3level.h	Thu Apr 05 08:59:12 2007 +0100
+++ b/linux-2.6-xen-sparse/include/asm-i386/mach-xen/asm/pgtable-3level.h	Thu Apr 05 09:10:33 2007 +0100
@@ -99,6 +99,11 @@ static inline void pud_clear (pud_t * pu
 #define pmd_offset(pud, address) ((pmd_t *) pud_page(*(pud)) + \
 			pmd_index(address))
 
+static inline int pte_none(pte_t pte)
+{
+	return !(pte.pte_low | pte.pte_high);
+}
+
 /*
  * For PTEs and PDEs, we must clear the P-bit first when clearing a page table
  * entry, so clear the bottom half first and enforce ordering with a compiler
@@ -106,24 +111,50 @@ static inline void pud_clear (pud_t * pu
  */
 static inline void pte_clear(struct mm_struct *mm, unsigned long addr, pte_t *ptep)
 {
-	ptep->pte_low = 0;
-	smp_wmb();
-	ptep->pte_high = 0;
+	if ((mm != current->mm && mm != &init_mm)
+	    || HYPERVISOR_update_va_mapping(addr, __pte(0), 0)) {
+		ptep->pte_low = 0;
+		smp_wmb();
+		ptep->pte_high = 0;
+	}
 }
 
 #define pmd_clear(xp)	do { set_pmd(xp, __pmd(0)); } while (0)
 
 static inline pte_t ptep_get_and_clear(struct mm_struct *mm, unsigned long addr, pte_t *ptep)
 {
-	pte_t res;
-
-	/* xchg acts as a barrier before the setting of the high bits */
-	res.pte_low = xchg(&ptep->pte_low, 0);
-	res.pte_high = ptep->pte_high;
-	ptep->pte_high = 0;
-
-	return res;
-}
+	pte_t pte = *ptep;
+	if (!pte_none(pte)) {
+		if (mm != &init_mm) {
+			uint64_t val = pte_val_ma(pte);
+			if (__cmpxchg64(ptep, val, 0) != val) {
+				/* xchg acts as a barrier before the setting of the high bits */
+				pte.pte_low = xchg(&ptep->pte_low, 0);
+				pte.pte_high = ptep->pte_high;
+				ptep->pte_high = 0;
+			}
+		} else
+			HYPERVISOR_update_va_mapping(addr, __pte(0), 0);
+	}
+	return pte;
+}
+
+#define ptep_clear_flush(vma, addr, ptep)			\
+({								\
+	pte_t *__ptep = (ptep);					\
+	pte_t __res = *__ptep;					\
+	if (!pte_none(__res) &&					\
+	    ((vma)->vm_mm != current->mm ||			\
+	     HYPERVISOR_update_va_mapping(addr,	__pte(0),	\
+			(unsigned long)(vma)->vm_mm->cpu_vm_mask.bits| \
+				UVMF_INVLPG|UVMF_MULTI))) {	\
+		__ptep->pte_low = 0;				\
+		smp_wmb();					\
+		__ptep->pte_high = 0;				\
+		flush_tlb_page(vma, addr);			\
+	}							\
+	__res;							\
+})
 
 static inline int pte_same(pte_t a, pte_t b)
 {
@@ -131,11 +162,6 @@ static inline int pte_same(pte_t a, pte_
 }
 
 #define pte_page(x)	pfn_to_page(pte_pfn(x))
-
-static inline int pte_none(pte_t pte)
-{
-	return !pte.pte_low && !pte.pte_high;
-}
 
 #define __pte_mfn(_pte) (((_pte).pte_low >> PAGE_SHIFT) | \
 			 ((_pte).pte_high << (32-PAGE_SHIFT)))
diff -r 07d3208c0ca3 -r 14c25e48a557 linux-2.6-xen-sparse/include/asm-i386/mach-xen/asm/pgtable.h
--- a/linux-2.6-xen-sparse/include/asm-i386/mach-xen/asm/pgtable.h	Thu Apr 05 08:59:12 2007 +0100
+++ b/linux-2.6-xen-sparse/include/asm-i386/mach-xen/asm/pgtable.h	Thu Apr 05 09:10:33 2007 +0100
@@ -210,9 +210,13 @@ extern unsigned long pg0[];
 
 /* To avoid harmful races, pmd_none(x) should check only the lower when PAE */
 #define pmd_none(x)	(!(unsigned long)pmd_val(x))
+#ifdef CONFIG_XEN_COMPAT_030002
 /* pmd_present doesn't just test the _PAGE_PRESENT bit since wr.p.t.
    can temporarily clear it. */
 #define pmd_present(x)	(pmd_val(x))
+#else
+#define pmd_present(x)	(pmd_val(x) & _PAGE_PRESENT)
+#endif
 #define pmd_bad(x)	((pmd_val(x) & (~PAGE_MASK & ~_PAGE_USER & ~_PAGE_PRESENT)) != (_KERNPG_TABLE & ~_PAGE_PRESENT))
 
 
@@ -252,36 +256,47 @@ static inline pte_t pte_mkhuge(pte_t pte
 # include <asm/pgtable-2level.h>
 #endif
 
-static inline int ptep_test_and_clear_dirty(struct vm_area_struct *vma, unsigned long addr, pte_t *ptep)
-{
-	if (!pte_dirty(*ptep))
-		return 0;
-	return test_and_clear_bit(_PAGE_BIT_DIRTY, &ptep->pte_low);
-}
-
-static inline int ptep_test_and_clear_young(struct vm_area_struct *vma, unsigned long addr, pte_t *ptep)
-{
-	if (!pte_young(*ptep))
-		return 0;
-	return test_and_clear_bit(_PAGE_BIT_ACCESSED, &ptep->pte_low);
-}
-
-static inline pte_t ptep_get_and_clear_full(struct mm_struct *mm, unsigned long addr, pte_t *ptep, int full)
-{
-	pte_t pte;
-	if (full) {
-		pte = *ptep;
-		pte_clear(mm, addr, ptep);
-	} else {
-		pte = ptep_get_and_clear(mm, addr, ptep);
-	}
-	return pte;
-}
+#define ptep_test_and_clear_dirty(vma, addr, ptep)			\
+({									\
+	pte_t __pte = *(ptep);						\
+	int __ret = pte_dirty(__pte);					\
+	if (__ret) {							\
+		__pte = pte_mkclean(__pte);				\
+		if ((vma)->vm_mm != current->mm ||			\
+		    HYPERVISOR_update_va_mapping(addr, __pte, 0))	\
+			(ptep)->pte_low = __pte.pte_low;		\
+	}								\
+	__ret;								\
+})
+
+#define ptep_test_and_clear_young(vma, addr, ptep)			\
+({									\
+	pte_t __pte = *(ptep);						\
+	int __ret = pte_young(__pte);					\
+	if (__ret)							\
+		__pte = pte_mkold(__pte);				\
+		if ((vma)->vm_mm != current->mm ||			\
+		    HYPERVISOR_update_va_mapping(addr, __pte, 0))	\
+			(ptep)->pte_low = __pte.pte_low;		\
+	__ret;								\
+})
+
+#define ptep_get_and_clear_full(mm, addr, ptep, full)			\
+	((full) ? ({							\
+		pte_t __res = *(ptep);					\
+		if (test_bit(PG_pinned, &virt_to_page((mm)->pgd)->flags)) \
+			xen_l1_entry_update(ptep, __pte(0));		\
+		else							\
+			*(ptep) = __pte(0);				\
+		__res;							\
+	 }) :								\
+	 ptep_get_and_clear(mm, addr, ptep))
 
 static inline void ptep_set_wrprotect(struct mm_struct *mm, unsigned long addr, pte_t *ptep)
 {
-	if (pte_write(*ptep))
-		clear_bit(_PAGE_BIT_RW, &ptep->pte_low);
+	pte_t pte = *ptep;
+	if (pte_write(pte))
+		set_pte_at(mm, addr, ptep, pte_wrprotect(pte));
 }
 
 /*
@@ -418,6 +433,20 @@ extern void noexec_setup(const char *str
 #define pte_unmap_nested(pte) do { } while (0)
 #endif
 
+#define __HAVE_ARCH_PTEP_ESTABLISH
+#define ptep_establish(vma, address, ptep, pteval)			\
+	do {								\
+		if ( likely((vma)->vm_mm == current->mm) ) {		\
+			BUG_ON(HYPERVISOR_update_va_mapping(address,	\
+				pteval,					\
+				(unsigned long)(vma)->vm_mm->cpu_vm_mask.bits| \
+					UVMF_INVLPG|UVMF_MULTI));	\
+		} else {						\
+			xen_l1_entry_update(ptep, pteval);		\
+			flush_tlb_page(vma, address);			\
+		}							\
+	} while (0)
+
 /*
  * The i386 doesn't have any external MMU info: the kernel page
  * tables contain all the necessary information.
@@ -430,26 +459,11 @@ extern void noexec_setup(const char *str
  */
 #define update_mmu_cache(vma,address,pte) do { } while (0)
 #define  __HAVE_ARCH_PTEP_SET_ACCESS_FLAGS
-#define ptep_set_access_flags(__vma, __address, __ptep, __entry, __dirty) \
-	do {								  \
-		if (__dirty) {						  \
-			if ( likely((__vma)->vm_mm == current->mm) ) {	  \
-				BUG_ON(HYPERVISOR_update_va_mapping(__address, \
-					__entry,			  \
-					(unsigned long)(__vma)->vm_mm->cpu_vm_mask.bits| \
-					UVMF_INVLPG|UVMF_MULTI));	  \
-			} else {					  \
-				xen_l1_entry_update(__ptep, __entry);	  \
-				flush_tlb_page(__vma, __address);	  \
-			}						  \
-		}							  \
+#define ptep_set_access_flags(vma, address, ptep, entry, dirty)		\
+	do {								\
+		if (dirty)						\
+			ptep_establish(vma, address, ptep, entry);	\
 	} while (0)
-
-#define __HAVE_ARCH_PTEP_ESTABLISH
-#define ptep_establish(__vma, __address, __ptep, __entry)		\
-do {				  					\
-	ptep_set_access_flags(__vma, __address, __ptep, __entry, 1);	\
-} while (0)
 
 #include <xen/features.h>
 void make_lowmem_page_readonly(void *va, unsigned int feature);
@@ -508,6 +522,7 @@ direct_remap_pfn_range(vma,from,pfn,size
 #define __HAVE_ARCH_PTEP_TEST_AND_CLEAR_DIRTY
 #define __HAVE_ARCH_PTEP_GET_AND_CLEAR
 #define __HAVE_ARCH_PTEP_GET_AND_CLEAR_FULL
+#define __HAVE_ARCH_PTEP_CLEAR_FLUSH
 #define __HAVE_ARCH_PTEP_SET_WRPROTECT
 #define __HAVE_ARCH_PTE_SAME
 #include <asm-generic/pgtable.h>
diff -r 07d3208c0ca3 -r 14c25e48a557 linux-2.6-xen-sparse/include/asm-x86_64/mach-xen/asm/pgtable.h
--- a/linux-2.6-xen-sparse/include/asm-x86_64/mach-xen/asm/pgtable.h	Thu Apr 05 08:59:12 2007 +0100
+++ b/linux-2.6-xen-sparse/include/asm-x86_64/mach-xen/asm/pgtable.h	Thu Apr 05 09:10:33 2007 +0100
@@ -93,11 +93,6 @@ extern unsigned long empty_zero_page[PAG
 #define pgd_none(x)	(!pgd_val(x))
 #define pud_none(x)	(!pud_val(x))
 
-#define set_pte_batched(pteptr, pteval) \
-	queue_l1_entry_update(pteptr, (pteval))
-
-extern inline int pud_present(pud_t pud)	{ return !pud_none(pud); }
-
 static inline void set_pte(pte_t *dst, pte_t val)
 {
 	*dst = val;
@@ -122,41 +117,6 @@ static inline void pgd_clear (pgd_t * pg
 
 #define pud_page(pud) \
     ((unsigned long) __va(pud_val(pud) & PHYSICAL_PAGE_MASK))
-
-/*
- * A note on implementation of this atomic 'get-and-clear' operation.
- * This is actually very simple because Xen Linux can only run on a single
- * processor. Therefore, we cannot race other processors setting the 'accessed'
- * or 'dirty' bits on a page-table entry.
- * Even if pages are shared between domains, that is not a problem because
- * each domain will have separate page tables, with their own versions of
- * accessed & dirty state.
- */
-#define ptep_get_and_clear(mm,addr,xp)	__pte_ma(xchg(&(xp)->pte, 0))
-
-#if 0
-static inline pte_t ptep_get_and_clear(struct mm_struct *mm, unsigned long addr, pte_t *xp)
-{
-        pte_t pte = *xp;
-        if (pte.pte)
-                set_pte(xp, __pte_ma(0));
-        return pte;
-}
-#endif
-
-struct mm_struct;
-
-static inline pte_t ptep_get_and_clear_full(struct mm_struct *mm, unsigned long addr, pte_t *ptep, int full)
-{
-	pte_t pte;
-	if (full) {
-		pte = *ptep;
-		*ptep = __pte(0);
-	} else {
-		pte = ptep_get_and_clear(mm, addr, ptep);
-	}
-	return pte;
-}
 
 #define pte_same(a, b)		((a).pte == (b).pte)
 
@@ -318,6 +278,46 @@ static inline pte_t pfn_pte(unsigned lon
 	return __pte(pte);
 }
 
+static inline pte_t ptep_get_and_clear(struct mm_struct *mm, unsigned long addr, pte_t *ptep)
+{
+	pte_t pte = *ptep;
+	if (!pte_none(pte)) {
+		if (mm != &init_mm)
+			pte = __pte_ma(xchg(&ptep->pte, 0));
+		else
+			HYPERVISOR_update_va_mapping(addr, __pte(0), 0);
+	}
+	return pte;
+}
+
+static inline pte_t ptep_get_and_clear_full(struct mm_struct *mm, unsigned long addr, pte_t *ptep, int full)
+{
+	if (full) {
+		pte_t pte = *ptep;
+		if (mm->context.pinned)
+			xen_l1_entry_update(ptep, __pte(0));
+		else
+			*ptep = __pte(0);
+		return pte;
+	}
+	return ptep_get_and_clear(mm, addr, ptep);
+}
+
+#define ptep_clear_flush(vma, addr, ptep)			\
+({								\
+	pte_t *__ptep = (ptep);					\
+	pte_t __res = *__ptep;					\
+	if (!pte_none(__res) &&					\
+	    ((vma)->vm_mm != current->mm ||			\
+	     HYPERVISOR_update_va_mapping(addr,	__pte(0), 	\
+			(unsigned long)(vma)->vm_mm->cpu_vm_mask.bits| \
+				UVMF_INVLPG|UVMF_MULTI))) {	\
+		__ptep->pte = 0;				\
+		flush_tlb_page(vma, addr);			\
+	}							\
+	__res;							\
+})
+
 /*
  * The following only work if pte_present() is true.
  * Undefined behaviour if not..
@@ -346,31 +346,29 @@ static inline pte_t pte_mkwrite(pte_t pt
 static inline pte_t pte_mkwrite(pte_t pte)	{ __pte_val(pte) |= _PAGE_RW; return pte; }
 static inline pte_t pte_mkhuge(pte_t pte)	{ __pte_val(pte) |= _PAGE_PSE; return pte; }
 
-struct vm_area_struct;
-
-static inline int ptep_test_and_clear_dirty(struct vm_area_struct *vma, unsigned long addr, pte_t *ptep)
-{
-	pte_t pte = *ptep;
-	int ret = pte_dirty(pte);
-	if (ret)
-		set_pte(ptep, pte_mkclean(pte));
-	return ret;
-}
-
-static inline int ptep_test_and_clear_young(struct vm_area_struct *vma, unsigned long addr, pte_t *ptep)
-{
-	pte_t pte = *ptep;
-	int ret = pte_young(pte);
-	if (ret)
-		set_pte(ptep, pte_mkold(pte));
-	return ret;
-}
+#define ptep_test_and_clear_dirty(vma, addr, ptep)			\
+({									\
+	pte_t __pte = *(ptep);						\
+	int __ret = pte_dirty(__pte);					\
+	if (__ret)							\
+		set_pte_at((vma)->vm_mm, addr, ptep, pte_mkclean(__pte)); \
+	__ret;								\
+})
+
+#define ptep_test_and_clear_young(vma, addr, ptep)			\
+({									\
+	pte_t __pte = *(ptep);						\
+	int __ret = pte_young(__pte);					\
+	if (__ret)							\
+		set_pte_at((vma)->vm_mm, addr, ptep, pte_mkold(__pte)); \
+	__ret;								\
+})
 
 static inline void ptep_set_wrprotect(struct mm_struct *mm, unsigned long addr, pte_t *ptep)
 {
 	pte_t pte = *ptep;
 	if (pte_write(pte))
-		set_pte(ptep, pte_wrprotect(pte));
+		set_pte_at(mm, addr, ptep, pte_wrprotect(pte));
 }
 
 /*
@@ -403,6 +401,7 @@ static inline int pmd_large(pmd_t pte) {
 /* to find an entry in a page-table-directory. */
 #define pud_index(address) (((address) >> PUD_SHIFT) & (PTRS_PER_PUD-1))
 #define pud_offset(pgd, address) ((pud_t *) pgd_page(*(pgd)) + pud_index(address))
+#define pud_present(pud) (pud_val(pud) & _PAGE_PRESENT)
 
 /* PMD  - Level 2 access */
 #define pmd_page_kernel(pmd) ((unsigned long) __va(pmd_val(pmd) & PTE_MASK))
@@ -412,9 +411,13 @@ static inline int pmd_large(pmd_t pte) {
 #define pmd_offset(dir, address) ((pmd_t *) pud_page(*(dir)) + \
                                   pmd_index(address))
 #define pmd_none(x)	(!pmd_val(x))
+#ifdef CONFIG_XEN_COMPAT_030002
 /* pmd_present doesn't just test the _PAGE_PRESENT bit since wr.p.t.
    can temporarily clear it. */
 #define pmd_present(x)	(pmd_val(x))
+#else
+#define pmd_present(x)	(pmd_val(x) & _PAGE_PRESENT)
+#endif
 #define pmd_clear(xp)	do { set_pmd(xp, __pmd(0)); } while (0)
 #define pmd_bad(x) ((pmd_val(x) & ~(PTE_MASK | _PAGE_USER | _PAGE_PRESENT)) \
 		    != (_KERNPG_TABLE & ~(_PAGE_USER | _PAGE_PRESENT)))
@@ -468,25 +471,34 @@ static inline pte_t pte_modify(pte_t pte
 
 #define update_mmu_cache(vma,address,pte) do { } while (0)
 
+/*
+ * Rules for using ptep_establish: the pte MUST be a user pte, and
+ * must be a present->present transition.
+ */
+#define __HAVE_ARCH_PTEP_ESTABLISH
+#define ptep_establish(vma, address, ptep, pteval)			\
+	do {								\
+		if ( likely((vma)->vm_mm == current->mm) ) {		\
+			BUG_ON(HYPERVISOR_update_va_mapping(address,	\
+				pteval,					\
+				(unsigned long)(vma)->vm_mm->cpu_vm_mask.bits| \
+					UVMF_INVLPG|UVMF_MULTI));	\
+		} else {						\
+			xen_l1_entry_update(ptep, pteval);		\
+			flush_tlb_page(vma, address);			\
+		}							\
+	} while (0)
+
 /* We only update the dirty/accessed state if we set
  * the dirty bit by hand in the kernel, since the hardware
  * will do the accessed bit for us, and we don't want to
  * race with other CPU's that might be updating the dirty
  * bit at the same time. */
 #define  __HAVE_ARCH_PTEP_SET_ACCESS_FLAGS
-#define ptep_set_access_flags(__vma, __address, __ptep, __entry, __dirty) \
-	do {								  \
-		if (__dirty) {						  \
-			if ( likely((__vma)->vm_mm == current->mm) ) {	  \
-				BUG_ON(HYPERVISOR_update_va_mapping(__address, \
-					__entry,			  \
-					(unsigned long)(__vma)->vm_mm->cpu_vm_mask.bits| \
-						UVMF_INVLPG|UVMF_MULTI)); \
-			} else {					  \
-				xen_l1_entry_update(__ptep, __entry);	  \
-				flush_tlb_page(__vma, __address);	  \
-			}						  \
-		}							  \
+#define ptep_set_access_flags(vma, address, ptep, entry, dirty)		\
+	do {								\
+		if (dirty)						\
+			ptep_establish(vma, address, ptep, entry);	\
 	} while (0)
 
 /* Encode and de-code a swap entry */
@@ -506,6 +518,8 @@ extern int kern_addr_valid(unsigned long
 
 #define DOMID_LOCAL (0xFFFFU)
 
+struct vm_area_struct;
+
 int direct_remap_pfn_range(struct vm_area_struct *vma,
                             unsigned long address,
                             unsigned long mfn,
@@ -551,6 +565,7 @@ int touch_pte_range(struct mm_struct *mm
 #define __HAVE_ARCH_PTEP_TEST_AND_CLEAR_DIRTY
 #define __HAVE_ARCH_PTEP_GET_AND_CLEAR
 #define __HAVE_ARCH_PTEP_GET_AND_CLEAR_FULL
+#define __HAVE_ARCH_PTEP_CLEAR_FLUSH
 #define __HAVE_ARCH_PTEP_SET_WRPROTECT
 #define __HAVE_ARCH_PTE_SAME
 #include <asm-generic/pgtable.h>