[people/pmueller/ipfire-2.x.git] / src / patches / suse-2.6.27.25 / patches.xen / xen-x86_64-pgd-alloc-order

From: jbeulich@novell.com
Subject: don't require order-1 allocations for pgd-s
Patch-mainline: obsolete

At the same time remove the useless user mode pair of init_level4_pgt.

--- sle11-2009-04-09.orig/arch/x86/kernel/cpu/common_64-xen.c	2008-11-17 14:07:10.000000000 +0100
+++ sle11-2009-04-09/arch/x86/kernel/cpu/common_64-xen.c	2009-03-16 16:40:52.000000000 +0100
@@ -530,8 +530,7 @@ static void __init_refok switch_pt(int c
 #ifdef CONFIG_XEN
 	if (cpu == 0)
 		xen_init_pt();
-	xen_pt_switch(__pa_symbol(init_level4_pgt));
-	xen_new_user_pt(__pa_symbol(__user_pgd(init_level4_pgt)));
+	xen_pt_switch(init_level4_pgt);
 #endif
 }
 
--- sle11-2009-04-09.orig/arch/x86/kernel/head_64-xen.S	2009-03-16 16:38:16.000000000 +0100
+++ sle11-2009-04-09/arch/x86/kernel/head_64-xen.S	2009-03-16 16:40:52.000000000 +0100
@@ -44,14 +44,6 @@ ENTRY(name)
 
 NEXT_PAGE(init_level4_pgt)
 	.fill	512,8,0
-        /*
-         * We update two pgd entries to make kernel and user pgd consistent
-         * at pgd_populate(). It can be used for kernel modules. So we place 
-         * this page here for those cases to avoid memory corruption.
-         * We also use this page to establish the initial mapping for the
-         * vsyscall area.
-         */
-	.fill	512,8,0
 
 NEXT_PAGE(level3_kernel_pgt)
 	.fill	512,8,0
--- sle11-2009-04-09.orig/arch/x86/mm/hypervisor.c	2009-03-16 16:40:50.000000000 +0100
+++ sle11-2009-04-09/arch/x86/mm/hypervisor.c	2009-03-30 12:19:20.000000000 +0200
@@ -423,7 +423,7 @@ void xen_l3_entry_update(pud_t *ptr, pud
 #endif
 
 #ifdef CONFIG_X86_64
-void xen_l4_entry_update(pgd_t *ptr, int user, pgd_t val)
+void xen_l4_entry_update(pgd_t *ptr, pgd_t val)
 {
 	mmu_update_t u[2];
 	struct page *page = NULL;
@@ -436,8 +436,10 @@ void xen_l4_entry_update(pgd_t *ptr, int
 	}
 	u[0].ptr = virt_to_machine(ptr);
 	u[0].val = __pgd_val(val);
-	if (user) {
-		u[1].ptr = virt_to_machine(__user_pgd(ptr));
+	if (((unsigned long)ptr & ~PAGE_MASK)
+	    < pgd_index(__HYPERVISOR_VIRT_START) * sizeof(*ptr)
+	    && (ptr = __user_pgd(ptr)) != NULL) {
+		u[1].ptr = virt_to_machine(ptr);
 		u[1].val = __pgd_val(val);
 		do_lN_entry_update(u, 2, page);
 	} else
@@ -445,21 +447,25 @@ void xen_l4_entry_update(pgd_t *ptr, int
 }
 #endif /* CONFIG_X86_64 */
 
-void xen_pt_switch(unsigned long ptr)
+#ifdef CONFIG_X86_64
+void xen_pt_switch(pgd_t *pgd)
 {
 	struct mmuext_op op;
 	op.cmd = MMUEXT_NEW_BASEPTR;
-	op.arg1.mfn = pfn_to_mfn(ptr >> PAGE_SHIFT);
+	op.arg1.mfn = pfn_to_mfn(__pa(pgd) >> PAGE_SHIFT);
 	BUG_ON(HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF) < 0);
 }
 
-void xen_new_user_pt(unsigned long ptr)
+void xen_new_user_pt(pgd_t *pgd)
 {
 	struct mmuext_op op;
+
+	pgd = __user_pgd(pgd);
 	op.cmd = MMUEXT_NEW_USER_BASEPTR;
-	op.arg1.mfn = pfn_to_mfn(ptr >> PAGE_SHIFT);
+	op.arg1.mfn = pgd ? pfn_to_mfn(__pa(pgd) >> PAGE_SHIFT) : 0;
 	BUG_ON(HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF) < 0);
 }
+#endif
 
 void xen_tlb_flush(void)
 {
@@ -529,28 +535,38 @@ void xen_invlpg_mask(cpumask_t *mask, un
 void xen_pgd_pin(pgd_t *pgd)
 {
 	struct mmuext_op op[NR_PGD_PIN_OPS];
+	unsigned int nr = NR_PGD_PIN_OPS;
 
 	op[0].cmd = MMUEXT_PIN_L3_TABLE;
 	op[0].arg1.mfn = pfn_to_mfn(__pa(pgd) >> PAGE_SHIFT);
 #ifdef CONFIG_X86_64
 	op[1].cmd = op[0].cmd = MMUEXT_PIN_L4_TABLE;
-	op[1].arg1.mfn = pfn_to_mfn(__pa(__user_pgd(pgd)) >> PAGE_SHIFT);
+	pgd = __user_pgd(pgd);
+	if (pgd)
+		op[1].arg1.mfn = pfn_to_mfn(__pa(pgd) >> PAGE_SHIFT);
+	else
+		nr = 1;
 #endif
-	if (HYPERVISOR_mmuext_op(op, NR_PGD_PIN_OPS, NULL, DOMID_SELF) < 0)
+	if (HYPERVISOR_mmuext_op(op, nr, NULL, DOMID_SELF) < 0)
 		BUG();
 }
 
 void xen_pgd_unpin(pgd_t *pgd)
 {
 	struct mmuext_op op[NR_PGD_PIN_OPS];
+	unsigned int nr = NR_PGD_PIN_OPS;
 
 	op[0].cmd = MMUEXT_UNPIN_TABLE;
 	op[0].arg1.mfn = pfn_to_mfn(__pa(pgd) >> PAGE_SHIFT);
 #ifdef CONFIG_X86_64
-	op[1].cmd = MMUEXT_UNPIN_TABLE;
-	op[1].arg1.mfn = pfn_to_mfn(__pa(__user_pgd(pgd)) >> PAGE_SHIFT);
+	pgd = __user_pgd(pgd);
+	if (pgd) {
+		op[1].cmd = MMUEXT_UNPIN_TABLE;
+		op[1].arg1.mfn = pfn_to_mfn(__pa(pgd) >> PAGE_SHIFT);
+	} else
+		nr = 1;
 #endif
-	if (HYPERVISOR_mmuext_op(op, NR_PGD_PIN_OPS, NULL, DOMID_SELF) < 0)
+	if (HYPERVISOR_mmuext_op(op, nr, NULL, DOMID_SELF) < 0)
 		BUG();
 }
 
--- sle11-2009-04-09.orig/arch/x86/mm/init_64-xen.c	2009-03-16 16:40:50.000000000 +0100
+++ sle11-2009-04-09/arch/x86/mm/init_64-xen.c	2009-03-16 16:40:52.000000000 +0100
@@ -616,9 +616,6 @@ void __init xen_init_pt(void)
 		__pud(__pa_symbol(level2_kernel_pgt) | _PAGE_TABLE);
 	memcpy(level2_kernel_pgt, page, PAGE_SIZE);
 
-	__user_pgd(init_level4_pgt)[pgd_index(VSYSCALL_START)] =
-		__pgd(__pa_symbol(level3_user_pgt) | _PAGE_TABLE);
-
 	/* Do an early initialization of the fixmap area. */
 	addr = __fix_to_virt(FIX_EARLYCON_MEM_BASE);
 	level3_kernel_pgt[pud_index(addr)] =
@@ -628,8 +625,6 @@ void __init xen_init_pt(void)
 
 	early_make_page_readonly(init_level4_pgt,
 				 XENFEAT_writable_page_tables);
-	early_make_page_readonly(__user_pgd(init_level4_pgt),
-				 XENFEAT_writable_page_tables);
 	early_make_page_readonly(level3_kernel_pgt,
 				 XENFEAT_writable_page_tables);
 	early_make_page_readonly(level3_user_pgt,
--- sle11-2009-04-09.orig/arch/x86/mm/pgtable-xen.c	2009-03-16 16:40:50.000000000 +0100
+++ sle11-2009-04-09/arch/x86/mm/pgtable-xen.c	2009-04-09 14:54:18.000000000 +0200
@@ -270,9 +270,11 @@ static void pgd_walk(pgd_t *pgd_base, pg
 			BUG();
 		seq = 0;
 	}
+	pgd = __user_pgd(pgd_base);
+	BUG_ON(!pgd);
 	MULTI_update_va_mapping(mcl + seq,
-	       (unsigned long)__user_pgd(pgd_base),
-	       pfn_pte(virt_to_phys(__user_pgd(pgd_base))>>PAGE_SHIFT, flags),
+	       (unsigned long)pgd,
+	       pfn_pte(virt_to_phys(pgd)>>PAGE_SHIFT, flags),
 	       0);
 	MULTI_update_va_mapping(mcl + seq + 1,
 	       (unsigned long)pgd_base,
@@ -662,12 +664,29 @@ static void pgd_prepopulate_pmd(struct m
 	}
 }
 
+static inline pgd_t *user_pgd_alloc(pgd_t *pgd)
+{
 #ifdef CONFIG_X86_64
-/* We allocate two contiguous pages for kernel and user. */
-#define PGD_ORDER 1
-#else
-#define PGD_ORDER 0
+	if (pgd) {
+		pgd_t *upgd = (void *)__get_free_page(GFP_KERNEL|__GFP_ZERO);
+
+		if (upgd)
+			virt_to_page(pgd)->index = (long)upgd;
+		else {
+			free_page((unsigned long)pgd);
+			pgd = NULL;
+		}
+	}
+#endif
+	return pgd;
+}
+
+static inline void user_pgd_free(pgd_t *pgd)
+{
+#ifdef CONFIG_X86_64
+	free_page(virt_to_page(pgd)->index);
 #endif
+}
 
 pgd_t *pgd_alloc(struct mm_struct *mm)
 {
@@ -675,7 +694,7 @@ pgd_t *pgd_alloc(struct mm_struct *mm)
 	pmd_t *pmds[PREALLOCATED_PMDS];
 	unsigned long flags;
 
-	pgd = (pgd_t *)__get_free_pages(GFP_KERNEL | __GFP_ZERO, PGD_ORDER);
+	pgd = user_pgd_alloc((void *)__get_free_page(GFP_KERNEL|__GFP_ZERO));
 
 	if (pgd == NULL)
 		goto out;
@@ -714,7 +733,8 @@ pgd_t *pgd_alloc(struct mm_struct *mm)
 out_free_pmds:
 	free_pmds(pmds, mm, !xen_feature(XENFEAT_pae_pgdir_above_4gb));
 out_free_pgd:
-	free_pages((unsigned long)pgd, PGD_ORDER);
+	user_pgd_free(pgd);
+	free_page((unsigned long)pgd);
 out:
 	return NULL;
 }
@@ -733,7 +753,8 @@ void pgd_free(struct mm_struct *mm, pgd_
 
 	pgd_mop_up_pmds(mm, pgd);
 	paravirt_pgd_free(mm, pgd);
-	free_pages((unsigned long)pgd, PGD_ORDER);
+	user_pgd_free(pgd);
+	free_page((unsigned long)pgd);
 }
 
 /* blktap and gntdev need this, as otherwise they would implicitly (and
--- sle11-2009-04-09.orig/drivers/xen/core/machine_reboot.c	2009-02-17 12:25:29.000000000 +0100
+++ sle11-2009-04-09/drivers/xen/core/machine_reboot.c	2009-03-16 16:40:52.000000000 +0100
@@ -191,8 +191,7 @@ static int take_machine_down(void *_susp
 		 * in fast-suspend mode as that implies a new enough Xen.
 		 */
 		if (!suspend->fast_suspend)
-			xen_new_user_pt(__pa(__user_pgd(
-				current->active_mm->pgd)));
+			xen_new_user_pt(current->active_mm->pgd);
 #endif
 	}
 
--- sle11-2009-04-09.orig/include/asm-x86/mach-xen/asm/hypervisor.h	2009-03-16 16:40:50.000000000 +0100
+++ sle11-2009-04-09/include/asm-x86/mach-xen/asm/hypervisor.h	2009-03-16 16:40:52.000000000 +0100
@@ -85,8 +85,8 @@ void do_hypervisor_callback(struct pt_re
  * be MACHINE addresses.
  */
 
-void xen_pt_switch(unsigned long ptr);
-void xen_new_user_pt(unsigned long ptr); /* x86_64 only */
+void xen_pt_switch(pgd_t *);
+void xen_new_user_pt(pgd_t *); /* x86_64 only */
 void xen_load_gs(unsigned int selector); /* x86_64 only */
 void xen_tlb_flush(void);
 void xen_invlpg(unsigned long ptr);
@@ -94,7 +94,7 @@ void xen_invlpg(unsigned long ptr);
 void xen_l1_entry_update(pte_t *ptr, pte_t val);
 void xen_l2_entry_update(pmd_t *ptr, pmd_t val);
 void xen_l3_entry_update(pud_t *ptr, pud_t val); /* x86_64/PAE */
-void xen_l4_entry_update(pgd_t *ptr, int user, pgd_t val); /* x86_64 only */
+void xen_l4_entry_update(pgd_t *ptr, pgd_t val); /* x86_64 only */
 void xen_pgd_pin(pgd_t *);
 void xen_pgd_unpin(pgd_t *);
 
--- sle11-2009-04-09.orig/include/asm-x86/mach-xen/asm/mmu_context_64.h	2009-03-16 16:38:16.000000000 +0100
+++ sle11-2009-04-09/include/asm-x86/mach-xen/asm/mmu_context_64.h	2009-03-16 16:40:52.000000000 +0100
@@ -46,6 +46,7 @@ static inline void switch_mm(struct mm_s
 {
 	unsigned cpu = smp_processor_id();
 	struct mmuext_op _op[3], *op = _op;
+	pgd_t *upgd;
 
 	if (likely(prev != next)) {
 		BUG_ON(!xen_feature(XENFEAT_writable_page_tables) &&
@@ -64,9 +65,11 @@ static inline void switch_mm(struct mm_s
 		op->arg1.mfn = pfn_to_mfn(__pa(next->pgd) >> PAGE_SHIFT);
 		op++;
 
-		/* xen_new_user_pt(__pa(__user_pgd(next->pgd))) */
+		/* xen_new_user_pt(next->pgd) */
 		op->cmd = MMUEXT_NEW_USER_BASEPTR;
-		op->arg1.mfn = pfn_to_mfn(__pa(__user_pgd(next->pgd)) >> PAGE_SHIFT);
+		upgd = __user_pgd(next->pgd);
+		op->arg1.mfn = likely(upgd)
+			       ? pfn_to_mfn(__pa(upgd) >> PAGE_SHIFT) : 0;
 		op++;
 		
 		if (unlikely(next->context.ldt != prev->context.ldt)) {
@@ -90,7 +93,7 @@ static inline void switch_mm(struct mm_s
 			 * to make sure to use no freed page tables.
 			 */
                         load_cr3(next->pgd);
-                        xen_new_user_pt(__pa(__user_pgd(next->pgd)));		
+			xen_new_user_pt(next->pgd);
 			load_LDT_nolock(&next->context);
 		}
 	}
--- sle11-2009-04-09.orig/include/asm-x86/mach-xen/asm/pgalloc.h	2009-03-16 16:40:37.000000000 +0100
+++ sle11-2009-04-09/include/asm-x86/mach-xen/asm/pgalloc.h	2009-03-16 16:40:52.000000000 +0100
@@ -106,15 +106,13 @@ static inline void pud_populate(struct m
 #endif	/* CONFIG_X86_PAE */
 
 #if PAGETABLE_LEVELS > 3
-#define __user_pgd(pgd) ((pgd) + PTRS_PER_PGD)
-
 static inline void pgd_populate(struct mm_struct *mm, pgd_t *pgd, pud_t *pud)
 {
 	pgd_t ent = __pgd(_PAGE_TABLE | __pa(pud));
 
 	paravirt_alloc_pud(mm, __pa(pud) >> PAGE_SHIFT);
 	if (unlikely(PagePinned(virt_to_page(pgd))))
-		xen_l4_entry_update(pgd, 1, ent);
+		xen_l4_entry_update(pgd, ent);
 	else
 		*__user_pgd(pgd) = *pgd = ent;
 }
--- sle11-2009-04-09.orig/include/asm-x86/mach-xen/asm/pgtable_64.h	2009-03-16 16:40:37.000000000 +0100
+++ sle11-2009-04-09/include/asm-x86/mach-xen/asm/pgtable_64.h	2009-03-16 16:40:52.000000000 +0100
@@ -131,18 +131,25 @@ static inline void xen_set_pud(pud_t *pu
 	: (void)(*__pudp = xen_make_pud(0));	\
 })
 
-#define __user_pgd(pgd) ((pgd) + PTRS_PER_PGD)
+static inline pgd_t *__user_pgd(pgd_t *pgd)
+{
+	if (unlikely(((unsigned long)pgd & PAGE_MASK)
+		     == (unsigned long)init_level4_pgt))
+		return NULL;
+	return (pgd_t *)(virt_to_page(pgd)->index
+			 + ((unsigned long)pgd & ~PAGE_MASK));
+}
 
 static inline void xen_set_pgd(pgd_t *pgdp, pgd_t pgd)
 {
-	xen_l4_entry_update(pgdp, 0, pgd);
+	xen_l4_entry_update(pgdp, pgd);
 }
 
 #define xen_pgd_clear(pgd)			\
 ({						\
 	pgd_t *__pgdp = (pgd);			\
 	PagePinned(virt_to_page(__pgdp))	\
-	? xen_l4_entry_update(__pgdp, 1, xen_make_pgd(0)) \
+	? xen_l4_entry_update(__pgdp, xen_make_pgd(0)) \
 	: (void)(*__user_pgd(__pgdp) = *__pgdp = xen_make_pgd(0)); \
 })
Commit	Line	Data
cc90b958 BS	1	From: jbeulich@novell.com
	2	Subject: don't require order-1 allocations for pgd-s
	3	Patch-mainline: obsolete
	4
	5	At the same time remove the useless user mode pair of init_level4_pgt.
	6
00e5a55c BS	7	--- sle11-2009-04-09.orig/arch/x86/kernel/cpu/common_64-xen.c 2008-11-17 14:07:10.000000000 +0100
00e5a55c BS	8	+++ sle11-2009-04-09/arch/x86/kernel/cpu/common_64-xen.c 2009-03-16 16:40:52.000000000 +0100
cc90b958 BS	9	@@ -530,8 +530,7 @@ static void __init_refok switch_pt(int c
	10	#ifdef CONFIG_XEN
	11	if (cpu == 0)
	12	xen_init_pt();
	13	- xen_pt_switch(__pa_symbol(init_level4_pgt));
	14	- xen_new_user_pt(__pa_symbol(__user_pgd(init_level4_pgt)));
	15	+ xen_pt_switch(init_level4_pgt);
	16	#endif
	17	}
	18
00e5a55c BS	19	--- sle11-2009-04-09.orig/arch/x86/kernel/head_64-xen.S 2009-03-16 16:38:16.000000000 +0100
00e5a55c BS	20	+++ sle11-2009-04-09/arch/x86/kernel/head_64-xen.S 2009-03-16 16:40:52.000000000 +0100
cc90b958 BS	21	@@ -44,14 +44,6 @@ ENTRY(name)
	22
	23	NEXT_PAGE(init_level4_pgt)
	24	.fill 512,8,0
	25	- /*
	26	- * We update two pgd entries to make kernel and user pgd consistent
	27	- * at pgd_populate(). It can be used for kernel modules. So we place
	28	- * this page here for those cases to avoid memory corruption.
	29	- * We also use this page to establish the initial mapping for the
	30	- * vsyscall area.
	31	- */
	32	- .fill 512,8,0
	33
	34	NEXT_PAGE(level3_kernel_pgt)
	35	.fill 512,8,0
00e5a55c BS	36	--- sle11-2009-04-09.orig/arch/x86/mm/hypervisor.c 2009-03-16 16:40:50.000000000 +0100
	37	+++ sle11-2009-04-09/arch/x86/mm/hypervisor.c 2009-03-30 12:19:20.000000000 +0200
	38	@@ -423,7 +423,7 @@ void xen_l3_entry_update(pud_t *ptr, pud
cc90b958 BS	39	#endif
	40
	41	#ifdef CONFIG_X86_64
	42	-void xen_l4_entry_update(pgd_t *ptr, int user, pgd_t val)
	43	+void xen_l4_entry_update(pgd_t *ptr, pgd_t val)
	44	{
	45	mmu_update_t u[2];
	46	struct page *page = NULL;
00e5a55c	47	@@ -436,8 +436,10 @@ void xen_l4_entry_update(pgd_t *ptr, int
cc90b958 BS	48	}
	49	u[0].ptr = virt_to_machine(ptr);
	50	u[0].val = __pgd_val(val);
	51	- if (user) {
	52	- u[1].ptr = virt_to_machine(__user_pgd(ptr));
	53	+ if (((unsigned long)ptr & ~PAGE_MASK)
	54	+ < pgd_index(__HYPERVISOR_VIRT_START) * sizeof(*ptr)
	55	+ && (ptr = __user_pgd(ptr)) != NULL) {
	56	+ u[1].ptr = virt_to_machine(ptr);
	57	u[1].val = __pgd_val(val);
	58	do_lN_entry_update(u, 2, page);
	59	} else
00e5a55c	60	@@ -445,21 +447,25 @@ void xen_l4_entry_update(pgd_t *ptr, int
cc90b958 BS	61	}
	62	#endif /* CONFIG_X86_64 */
	63
	64	-void xen_pt_switch(unsigned long ptr)
	65	+#ifdef CONFIG_X86_64
	66	+void xen_pt_switch(pgd_t *pgd)
	67	{
	68	struct mmuext_op op;
	69	op.cmd = MMUEXT_NEW_BASEPTR;
	70	- op.arg1.mfn = pfn_to_mfn(ptr >> PAGE_SHIFT);
	71	+ op.arg1.mfn = pfn_to_mfn(__pa(pgd) >> PAGE_SHIFT);
	72	BUG_ON(HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF) < 0);
	73	}
	74
	75	-void xen_new_user_pt(unsigned long ptr)
	76	+void xen_new_user_pt(pgd_t *pgd)
	77	{
	78	struct mmuext_op op;
	79	+
	80	+ pgd = __user_pgd(pgd);
	81	op.cmd = MMUEXT_NEW_USER_BASEPTR;
	82	- op.arg1.mfn = pfn_to_mfn(ptr >> PAGE_SHIFT);
	83	+ op.arg1.mfn = pgd ? pfn_to_mfn(__pa(pgd) >> PAGE_SHIFT) : 0;
	84	BUG_ON(HYPERVISOR_mmuext_op(&op, 1, NULL, DOMID_SELF) < 0);
	85	}
	86	+#endif
	87
	88	void xen_tlb_flush(void)
	89	{
00e5a55c	90	@@ -529,28 +535,38 @@ void xen_invlpg_mask(cpumask_t *mask, un
cc90b958 BS	91	void xen_pgd_pin(pgd_t *pgd)
	92	{
	93	struct mmuext_op op[NR_PGD_PIN_OPS];
	94	+ unsigned int nr = NR_PGD_PIN_OPS;
	95
	96	op[0].cmd = MMUEXT_PIN_L3_TABLE;
	97	op[0].arg1.mfn = pfn_to_mfn(__pa(pgd) >> PAGE_SHIFT);
	98	#ifdef CONFIG_X86_64
	99	op[1].cmd = op[0].cmd = MMUEXT_PIN_L4_TABLE;
	100	- op[1].arg1.mfn = pfn_to_mfn(__pa(__user_pgd(pgd)) >> PAGE_SHIFT);
	101	+ pgd = __user_pgd(pgd);
	102	+ if (pgd)
	103	+ op[1].arg1.mfn = pfn_to_mfn(__pa(pgd) >> PAGE_SHIFT);
	104	+ else
	105	+ nr = 1;
	106	#endif
	107	- if (HYPERVISOR_mmuext_op(op, NR_PGD_PIN_OPS, NULL, DOMID_SELF) < 0)
	108	+ if (HYPERVISOR_mmuext_op(op, nr, NULL, DOMID_SELF) < 0)
	109	BUG();
	110	}
	111
	112	void xen_pgd_unpin(pgd_t *pgd)
	113	{
	114	struct mmuext_op op[NR_PGD_PIN_OPS];
	115	+ unsigned int nr = NR_PGD_PIN_OPS;
	116
	117	op[0].cmd = MMUEXT_UNPIN_TABLE;
	118	op[0].arg1.mfn = pfn_to_mfn(__pa(pgd) >> PAGE_SHIFT);
	119	#ifdef CONFIG_X86_64
	120	- op[1].cmd = MMUEXT_UNPIN_TABLE;
	121	- op[1].arg1.mfn = pfn_to_mfn(__pa(__user_pgd(pgd)) >> PAGE_SHIFT);
	122	+ pgd = __user_pgd(pgd);
	123	+ if (pgd) {
	124	+ op[1].cmd = MMUEXT_UNPIN_TABLE;
	125	+ op[1].arg1.mfn = pfn_to_mfn(__pa(pgd) >> PAGE_SHIFT);
	126	+ } else
	127	+ nr = 1;
	128	#endif
	129	- if (HYPERVISOR_mmuext_op(op, NR_PGD_PIN_OPS, NULL, DOMID_SELF) < 0)
	130	+ if (HYPERVISOR_mmuext_op(op, nr, NULL, DOMID_SELF) < 0)
	131	BUG();
	132	}
	133
00e5a55c BS	134	--- sle11-2009-04-09.orig/arch/x86/mm/init_64-xen.c 2009-03-16 16:40:50.000000000 +0100
	135	+++ sle11-2009-04-09/arch/x86/mm/init_64-xen.c 2009-03-16 16:40:52.000000000 +0100
	136	@@ -616,9 +616,6 @@ void __init xen_init_pt(void)
cc90b958 BS	137	__pud(__pa_symbol(level2_kernel_pgt) \| _PAGE_TABLE);
	138	memcpy(level2_kernel_pgt, page, PAGE_SIZE);
	139
	140	- __user_pgd(init_level4_pgt)[pgd_index(VSYSCALL_START)] =
	141	- __pgd(__pa_symbol(level3_user_pgt) \| _PAGE_TABLE);
	142	-
	143	/* Do an early initialization of the fixmap area. */
	144	addr = __fix_to_virt(FIX_EARLYCON_MEM_BASE);
	145	level3_kernel_pgt[pud_index(addr)] =
00e5a55c	146	@@ -628,8 +625,6 @@ void __init xen_init_pt(void)
cc90b958 BS	147
	148	early_make_page_readonly(init_level4_pgt,
	149	XENFEAT_writable_page_tables);
	150	- early_make_page_readonly(__user_pgd(init_level4_pgt),
	151	- XENFEAT_writable_page_tables);
	152	early_make_page_readonly(level3_kernel_pgt,
	153	XENFEAT_writable_page_tables);
	154	early_make_page_readonly(level3_user_pgt,
00e5a55c BS	155	--- sle11-2009-04-09.orig/arch/x86/mm/pgtable-xen.c 2009-03-16 16:40:50.000000000 +0100
00e5a55c BS	156	+++ sle11-2009-04-09/arch/x86/mm/pgtable-xen.c 2009-04-09 14:54:18.000000000 +0200
cc90b958 BS	157	@@ -270,9 +270,11 @@ static void pgd_walk(pgd_t *pgd_base, pg
	158	BUG();
	159	seq = 0;
	160	}
	161	+ pgd = __user_pgd(pgd_base);
	162	+ BUG_ON(!pgd);
	163	MULTI_update_va_mapping(mcl + seq,
	164	- (unsigned long)__user_pgd(pgd_base),
	165	- pfn_pte(virt_to_phys(__user_pgd(pgd_base))>>PAGE_SHIFT, flags),
	166	+ (unsigned long)pgd,
	167	+ pfn_pte(virt_to_phys(pgd)>>PAGE_SHIFT, flags),
	168	0);
	169	MULTI_update_va_mapping(mcl + seq + 1,
	170	(unsigned long)pgd_base,
00e5a55c	171	@@ -662,12 +664,29 @@ static void pgd_prepopulate_pmd(struct m
cc90b958 BS	172	}
	173	}
	174
	175	+static inline pgd_t user_pgd_alloc(pgd_t pgd)
	176	+{
	177	#ifdef CONFIG_X86_64
	178	-/* We allocate two contiguous pages for kernel and user. */
	179	-#define PGD_ORDER 1
	180	-#else
	181	-#define PGD_ORDER 0
	182	+ if (pgd) {
	183	+ pgd_t upgd = (void )__get_free_page(GFP_KERNEL\|__GFP_ZERO);
	184	+
	185	+ if (upgd)
	186	+ virt_to_page(pgd)->index = (long)upgd;
	187	+ else {
	188	+ free_page((unsigned long)pgd);
	189	+ pgd = NULL;
	190	+ }
	191	+ }
	192	+#endif
	193	+ return pgd;
	194	+}
	195	+
	196	+static inline void user_pgd_free(pgd_t *pgd)
	197	+{
	198	+#ifdef CONFIG_X86_64
	199	+ free_page(virt_to_page(pgd)->index);
	200	#endif
	201	+}
	202
	203	pgd_t pgd_alloc(struct mm_struct mm)
	204	{
00e5a55c	205	@@ -675,7 +694,7 @@ pgd_t pgd_alloc(struct mm_struct mm)
cc90b958 BS	206	pmd_t *pmds[PREALLOCATED_PMDS];
	207	unsigned long flags;
	208
	209	- pgd = (pgd_t *)__get_free_pages(GFP_KERNEL \| __GFP_ZERO, PGD_ORDER);
	210	+ pgd = user_pgd_alloc((void *)__get_free_page(GFP_KERNEL\|__GFP_ZERO));
	211
	212	if (pgd == NULL)
	213	goto out;
00e5a55c	214	@@ -714,7 +733,8 @@ pgd_t pgd_alloc(struct mm_struct mm)
cc90b958 BS	215	out_free_pmds:
	216	free_pmds(pmds, mm, !xen_feature(XENFEAT_pae_pgdir_above_4gb));
	217	out_free_pgd:
	218	- free_pages((unsigned long)pgd, PGD_ORDER);
	219	+ user_pgd_free(pgd);
	220	+ free_page((unsigned long)pgd);
	221	out:
	222	return NULL;
	223	}
00e5a55c	224	@@ -733,7 +753,8 @@ void pgd_free(struct mm_struct *mm, pgd_
cc90b958 BS	225
	226	pgd_mop_up_pmds(mm, pgd);
	227	paravirt_pgd_free(mm, pgd);
	228	- free_pages((unsigned long)pgd, PGD_ORDER);
	229	+ user_pgd_free(pgd);
	230	+ free_page((unsigned long)pgd);
	231	}
	232
	233	/* blktap and gntdev need this, as otherwise they would implicitly (and
00e5a55c BS	234	--- sle11-2009-04-09.orig/drivers/xen/core/machine_reboot.c 2009-02-17 12:25:29.000000000 +0100
	235	+++ sle11-2009-04-09/drivers/xen/core/machine_reboot.c 2009-03-16 16:40:52.000000000 +0100
	236	@@ -191,8 +191,7 @@ static int take_machine_down(void *_susp
cc90b958 BS	237	* in fast-suspend mode as that implies a new enough Xen.
	238	*/
	239	if (!suspend->fast_suspend)
	240	- xen_new_user_pt(__pa(__user_pgd(
	241	- current->active_mm->pgd)));
	242	+ xen_new_user_pt(current->active_mm->pgd);
	243	#endif
	244	}
	245
00e5a55c BS	246	--- sle11-2009-04-09.orig/include/asm-x86/mach-xen/asm/hypervisor.h 2009-03-16 16:40:50.000000000 +0100
00e5a55c BS	247	+++ sle11-2009-04-09/include/asm-x86/mach-xen/asm/hypervisor.h 2009-03-16 16:40:52.000000000 +0100
cc90b958 BS	248	@@ -85,8 +85,8 @@ void do_hypervisor_callback(struct pt_re
	249	* be MACHINE addresses.
	250	*/
	251
	252	-void xen_pt_switch(unsigned long ptr);
	253	-void xen_new_user_pt(unsigned long ptr); /* x86_64 only */
	254	+void xen_pt_switch(pgd_t *);
	255	+void xen_new_user_pt(pgd_t ); / x86_64 only */
	256	void xen_load_gs(unsigned int selector); /* x86_64 only */
	257	void xen_tlb_flush(void);
	258	void xen_invlpg(unsigned long ptr);
	259	@@ -94,7 +94,7 @@ void xen_invlpg(unsigned long ptr);
	260	void xen_l1_entry_update(pte_t *ptr, pte_t val);
	261	void xen_l2_entry_update(pmd_t *ptr, pmd_t val);
	262	void xen_l3_entry_update(pud_t ptr, pud_t val); / x86_64/PAE */
	263	-void xen_l4_entry_update(pgd_t ptr, int user, pgd_t val); / x86_64 only */
	264	+void xen_l4_entry_update(pgd_t ptr, pgd_t val); / x86_64 only */
	265	void xen_pgd_pin(pgd_t *);
	266	void xen_pgd_unpin(pgd_t *);
	267
00e5a55c BS	268	--- sle11-2009-04-09.orig/include/asm-x86/mach-xen/asm/mmu_context_64.h 2009-03-16 16:38:16.000000000 +0100
00e5a55c BS	269	+++ sle11-2009-04-09/include/asm-x86/mach-xen/asm/mmu_context_64.h 2009-03-16 16:40:52.000000000 +0100
cc90b958 BS	270	@@ -46,6 +46,7 @@ static inline void switch_mm(struct mm_s
	271	{
	272	unsigned cpu = smp_processor_id();
	273	struct mmuext_op _op[3], *op = _op;
	274	+ pgd_t *upgd;
	275
	276	if (likely(prev != next)) {
	277	BUG_ON(!xen_feature(XENFEAT_writable_page_tables) &&
	278	@@ -64,9 +65,11 @@ static inline void switch_mm(struct mm_s
	279	op->arg1.mfn = pfn_to_mfn(__pa(next->pgd) >> PAGE_SHIFT);
	280	op++;
	281
	282	- /* xen_new_user_pt(__pa(__user_pgd(next->pgd))) */
	283	+ /* xen_new_user_pt(next->pgd) */
	284	op->cmd = MMUEXT_NEW_USER_BASEPTR;
	285	- op->arg1.mfn = pfn_to_mfn(__pa(__user_pgd(next->pgd)) >> PAGE_SHIFT);
	286	+ upgd = __user_pgd(next->pgd);
	287	+ op->arg1.mfn = likely(upgd)
	288	+ ? pfn_to_mfn(__pa(upgd) >> PAGE_SHIFT) : 0;
	289	op++;
	290
	291	if (unlikely(next->context.ldt != prev->context.ldt)) {
	292	@@ -90,7 +93,7 @@ static inline void switch_mm(struct mm_s
	293	* to make sure to use no freed page tables.
	294	*/
	295	load_cr3(next->pgd);
	296	- xen_new_user_pt(__pa(__user_pgd(next->pgd)));
	297	+ xen_new_user_pt(next->pgd);
	298	load_LDT_nolock(&next->context);
	299	}
	300	}
00e5a55c BS	301	--- sle11-2009-04-09.orig/include/asm-x86/mach-xen/asm/pgalloc.h 2009-03-16 16:40:37.000000000 +0100
00e5a55c BS	302	+++ sle11-2009-04-09/include/asm-x86/mach-xen/asm/pgalloc.h 2009-03-16 16:40:52.000000000 +0100
cc90b958 BS	303	@@ -106,15 +106,13 @@ static inline void pud_populate(struct m
	304	#endif /* CONFIG_X86_PAE */
	305
	306	#if PAGETABLE_LEVELS > 3
	307	-#define __user_pgd(pgd) ((pgd) + PTRS_PER_PGD)
	308	-
	309	static inline void pgd_populate(struct mm_struct mm, pgd_t pgd, pud_t *pud)
	310	{
	311	pgd_t ent = __pgd(_PAGE_TABLE \| __pa(pud));
	312
	313	paravirt_alloc_pud(mm, __pa(pud) >> PAGE_SHIFT);
	314	if (unlikely(PagePinned(virt_to_page(pgd))))
	315	- xen_l4_entry_update(pgd, 1, ent);
	316	+ xen_l4_entry_update(pgd, ent);
	317	else
	318	__user_pgd(pgd) = pgd = ent;
	319	}
00e5a55c BS	320	--- sle11-2009-04-09.orig/include/asm-x86/mach-xen/asm/pgtable_64.h 2009-03-16 16:40:37.000000000 +0100
00e5a55c BS	321	+++ sle11-2009-04-09/include/asm-x86/mach-xen/asm/pgtable_64.h 2009-03-16 16:40:52.000000000 +0100
cc90b958 BS	322	@@ -131,18 +131,25 @@ static inline void xen_set_pud(pud_t *pu
	323	: (void)(*__pudp = xen_make_pud(0)); \
	324	})
	325
	326	-#define __user_pgd(pgd) ((pgd) + PTRS_PER_PGD)
	327	+static inline pgd_t __user_pgd(pgd_t pgd)
	328	+{
	329	+ if (unlikely(((unsigned long)pgd & PAGE_MASK)
	330	+ == (unsigned long)init_level4_pgt))
	331	+ return NULL;
	332	+ return (pgd_t *)(virt_to_page(pgd)->index
	333	+ + ((unsigned long)pgd & ~PAGE_MASK));
	334	+}
	335
	336	static inline void xen_set_pgd(pgd_t *pgdp, pgd_t pgd)
	337	{
	338	- xen_l4_entry_update(pgdp, 0, pgd);
	339	+ xen_l4_entry_update(pgdp, pgd);
	340	}
	341
	342	#define xen_pgd_clear(pgd) \
	343	({ \
	344	pgd_t *__pgdp = (pgd); \
	345	PagePinned(virt_to_page(__pgdp)) \
	346	- ? xen_l4_entry_update(__pgdp, 1, xen_make_pgd(0)) \
	347	+ ? xen_l4_entry_update(__pgdp, xen_make_pgd(0)) \
	348	: (void)(__user_pgd(__pgdp) = __pgdp = xen_make_pgd(0)); \
	349	})
	350