Merge phase #2 (PAT updates) of git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip

* 'x86-v28-for-linus-phase2-B' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip: (27 commits)
  x86, cpa: make the kernel physical mapping initialization a two pass sequence, fix
  x86, pat: cleanups
  x86: fix pagetable init 64-bit breakage
  x86: track memtype for RAM in page struct
  x86, cpa: srlz cpa(), global flush tlb after splitting big page and before doing cpa
  x86, cpa: remove cpa pool code
  x86, cpa: no need to check alias for __set_pages_p/__set_pages_np
  x86, cpa: dont use large pages for kernel identity mapping with DEBUG_PAGEALLOC
  x86, cpa: make the kernel physical mapping initialization a two pass sequence
  x86, cpa: remove USER permission from the very early identity mapping attribute
  x86, cpa: rename PTE attribute macros for kernel direct mapping in early boot
  x86: make sure the CPA test code's use of _PAGE_UNUSED1 is obvious
  linux-next: fix x86 tree build failure
  x86: have set_memory_array_{uc,wb} coalesce memtypes, fix
  agp: enable optimized agp_alloc_pages methods
  x86: have set_memory_array_{uc,wb} coalesce memtypes.
  x86: {reverve,free}_memtype() take a physical address
  x86: fix pageattr-test
  agp: add agp_generic_destroy_pages()
  agp: generic_alloc_pages()
  ...

28 files changed, 730 insertions, 307 deletions

diff --git a/arch/x86/kernel/head_32.S b/arch/x86/kernel/head_32.S
index a7010c3a377..e835b4eea70 100644
--- a/arch/x86/kernel/head_32.S
+++ b/arch/x86/kernel/head_32.S
@@ -172,10 +172,6 @@ num_subarch_entries = (. - subarch_entries) / 4
  *
  * Note that the stack is not yet set up!
  */
-#define PTE_ATTR	0x007		/* PRESENT+RW+USER */
-#define PDE_ATTR	0x067		/* PRESENT+RW+USER+DIRTY+ACCESSED */
-#define PGD_ATTR	0x001		/* PRESENT (no other attributes) */
-
 default_entry:
 #ifdef CONFIG_X86_PAE
 
@@ -196,9 +192,9 @@ default_entry:
 	movl $pa(pg0), %edi
 	movl %edi, pa(init_pg_tables_start)
 	movl $pa(swapper_pg_pmd), %edx
-	movl $PTE_ATTR, %eax
+	movl $PTE_IDENT_ATTR, %eax
 10:
-	leal PDE_ATTR(%edi),%ecx		/* Create PMD entry */
+	leal PDE_IDENT_ATTR(%edi),%ecx		/* Create PMD entry */
 	movl %ecx,(%edx)			/* Store PMD entry */
 						/* Upper half already zero */
 	addl $8,%edx
@@ -215,7 +211,7 @@ default_entry:
 	 * End condition: we must map up to and including INIT_MAP_BEYOND_END
 	 * bytes beyond the end of our own page tables.
 	 */
-	leal (INIT_MAP_BEYOND_END+PTE_ATTR)(%edi),%ebp
+	leal (INIT_MAP_BEYOND_END+PTE_IDENT_ATTR)(%edi),%ebp
 	cmpl %ebp,%eax
 	jb 10b
 1:
@@ -224,7 +220,7 @@ default_entry:
 	movl %eax, pa(max_pfn_mapped)
 
 	/* Do early initialization of the fixmap area */
-	movl $pa(swapper_pg_fixmap)+PDE_ATTR,%eax
+	movl $pa(swapper_pg_fixmap)+PDE_IDENT_ATTR,%eax
 	movl %eax,pa(swapper_pg_pmd+0x1000*KPMDS-8)
 #else	/* Not PAE */
 
@@ -233,9 +229,9 @@ page_pde_offset = (__PAGE_OFFSET >> 20);
 	movl $pa(pg0), %edi
 	movl %edi, pa(init_pg_tables_start)
 	movl $pa(swapper_pg_dir), %edx
-	movl $PTE_ATTR, %eax
+	movl $PTE_IDENT_ATTR, %eax
 10:
-	leal PDE_ATTR(%edi),%ecx		/* Create PDE entry */
+	leal PDE_IDENT_ATTR(%edi),%ecx		/* Create PDE entry */
 	movl %ecx,(%edx)			/* Store identity PDE entry */
 	movl %ecx,page_pde_offset(%edx)		/* Store kernel PDE entry */
 	addl $4,%edx
@@ -249,7 +245,7 @@ page_pde_offset = (__PAGE_OFFSET >> 20);
 	 * bytes beyond the end of our own page tables; the +0x007 is
 	 * the attribute bits
 	 */
-	leal (INIT_MAP_BEYOND_END+PTE_ATTR)(%edi),%ebp
+	leal (INIT_MAP_BEYOND_END+PTE_IDENT_ATTR)(%edi),%ebp
 	cmpl %ebp,%eax
 	jb 10b
 	movl %edi,pa(init_pg_tables_end)
@@ -257,7 +253,7 @@ page_pde_offset = (__PAGE_OFFSET >> 20);
 	movl %eax, pa(max_pfn_mapped)
 
 	/* Do early initialization of the fixmap area */
-	movl $pa(swapper_pg_fixmap)+PDE_ATTR,%eax
+	movl $pa(swapper_pg_fixmap)+PDE_IDENT_ATTR,%eax
 	movl %eax,pa(swapper_pg_dir+0xffc)
 #endif
 	jmp 3f
@@ -634,19 +630,19 @@ ENTRY(empty_zero_page)
 	/* Page-aligned for the benefit of paravirt? */
 	.align PAGE_SIZE_asm
 ENTRY(swapper_pg_dir)
-	.long	pa(swapper_pg_pmd+PGD_ATTR),0		/* low identity map */
+	.long	pa(swapper_pg_pmd+PGD_IDENT_ATTR),0	/* low identity map */
 # if KPMDS == 3
-	.long	pa(swapper_pg_pmd+PGD_ATTR),0
-	.long	pa(swapper_pg_pmd+PGD_ATTR+0x1000),0
-	.long	pa(swapper_pg_pmd+PGD_ATTR+0x2000),0
+	.long	pa(swapper_pg_pmd+PGD_IDENT_ATTR),0
+	.long	pa(swapper_pg_pmd+PGD_IDENT_ATTR+0x1000),0
+	.long	pa(swapper_pg_pmd+PGD_IDENT_ATTR+0x2000),0
 # elif KPMDS == 2
 	.long	0,0
-	.long	pa(swapper_pg_pmd+PGD_ATTR),0
-	.long	pa(swapper_pg_pmd+PGD_ATTR+0x1000),0
+	.long	pa(swapper_pg_pmd+PGD_IDENT_ATTR),0
+	.long	pa(swapper_pg_pmd+PGD_IDENT_ATTR+0x1000),0
 # elif KPMDS == 1
 	.long	0,0
 	.long	0,0
-	.long	pa(swapper_pg_pmd+PGD_ATTR),0
+	.long	pa(swapper_pg_pmd+PGD_IDENT_ATTR),0
 # else
 #  error "Kernel PMDs should be 1, 2 or 3"
 # endif
diff --git a/arch/x86/kernel/head_64.S b/arch/x86/kernel/head_64.S
index db3280afe88..26cfdc1d7c7 100644
--- a/arch/x86/kernel/head_64.S
+++ b/arch/x86/kernel/head_64.S
@@ -110,7 +110,7 @@ startup_64:
 	movq	%rdi, %rax
 	shrq	$PMD_SHIFT, %rax
 	andq	$(PTRS_PER_PMD - 1), %rax
-	leaq	__PAGE_KERNEL_LARGE_EXEC(%rdi), %rdx
+	leaq	__PAGE_KERNEL_IDENT_LARGE_EXEC(%rdi), %rdx
 	leaq	level2_spare_pgt(%rip), %rbx
 	movq	%rdx, 0(%rbx, %rax, 8)
 ident_complete:
@@ -374,7 +374,7 @@ NEXT_PAGE(level2_ident_pgt)
 	/* Since I easily can, map the first 1G.
 	 * Don't set NX because code runs from these pages.
 	 */
-	PMDS(0, __PAGE_KERNEL_LARGE_EXEC, PTRS_PER_PMD)
+	PMDS(0, __PAGE_KERNEL_IDENT_LARGE_EXEC, PTRS_PER_PMD)
 
 NEXT_PAGE(level2_kernel_pgt)
 	/*
diff --git a/arch/x86/mm/init_32.c b/arch/x86/mm/init_32.c
index 6b9a9358b33..c3789bb1930 100644
--- a/arch/x86/mm/init_32.c
+++ b/arch/x86/mm/init_32.c
@@ -195,11 +195,30 @@ static void __init kernel_physical_mapping_init(pgd_t *pgd_base,
 	pgd_t *pgd;
 	pmd_t *pmd;
 	pte_t *pte;
-	unsigned pages_2m = 0, pages_4k = 0;
+	unsigned pages_2m, pages_4k;
+	int mapping_iter;
+
+	/*
+	 * First iteration will setup identity mapping using large/small pages
+	 * based on use_pse, with other attributes same as set by
+	 * the early code in head_32.S
+	 *
+	 * Second iteration will setup the appropriate attributes (NX, GLOBAL..)
+	 * as desired for the kernel identity mapping.
+	 *
+	 * This two pass mechanism conforms to the TLB app note which says:
+	 *
+	 *     "Software should not write to a paging-structure entry in a way
+	 *      that would change, for any linear address, both the page size
+	 *      and either the page frame or attributes."
+	 */
+	mapping_iter = 1;
 
 	if (!cpu_has_pse)
 		use_pse = 0;
 
+repeat:
+	pages_2m = pages_4k = 0;
 	pfn = start_pfn;
 	pgd_idx = pgd_index((pfn<<PAGE_SHIFT) + PAGE_OFFSET);
 	pgd = pgd_base + pgd_idx;
@@ -225,6 +244,13 @@ static void __init kernel_physical_mapping_init(pgd_t *pgd_base,
 			if (use_pse) {
 				unsigned int addr2;
 				pgprot_t prot = PAGE_KERNEL_LARGE;
+				/*
+				 * first pass will use the same initial
+				 * identity mapping attribute + _PAGE_PSE.
+				 */
+				pgprot_t init_prot =
+					__pgprot(PTE_IDENT_ATTR |
+						 _PAGE_PSE);
 
 				addr2 = (pfn + PTRS_PER_PTE-1) * PAGE_SIZE +
 					PAGE_OFFSET + PAGE_SIZE-1;
@@ -234,7 +260,10 @@ static void __init kernel_physical_mapping_init(pgd_t *pgd_base,
 					prot = PAGE_KERNEL_LARGE_EXEC;
 
 				pages_2m++;
-				set_pmd(pmd, pfn_pmd(pfn, prot));
+				if (mapping_iter == 1)
+					set_pmd(pmd, pfn_pmd(pfn, init_prot));
+				else
+					set_pmd(pmd, pfn_pmd(pfn, prot));
 
 				pfn += PTRS_PER_PTE;
 				continue;
@@ -246,17 +275,43 @@ static void __init kernel_physical_mapping_init(pgd_t *pgd_base,
 			for (; pte_ofs < PTRS_PER_PTE && pfn < end_pfn;
 			     pte++, pfn++, pte_ofs++, addr += PAGE_SIZE) {
 				pgprot_t prot = PAGE_KERNEL;
+				/*
+				 * first pass will use the same initial
+				 * identity mapping attribute.
+				 */
+				pgprot_t init_prot = __pgprot(PTE_IDENT_ATTR);
 
 				if (is_kernel_text(addr))
 					prot = PAGE_KERNEL_EXEC;
 
 				pages_4k++;
-				set_pte(pte, pfn_pte(pfn, prot));
+				if (mapping_iter == 1)
+					set_pte(pte, pfn_pte(pfn, init_prot));
+				else
+					set_pte(pte, pfn_pte(pfn, prot));
 			}
 		}
 	}
-	update_page_count(PG_LEVEL_2M, pages_2m);
-	update_page_count(PG_LEVEL_4K, pages_4k);
+	if (mapping_iter == 1) {
+		/*
+		 * update direct mapping page count only in the first
+		 * iteration.
+		 */
+		update_page_count(PG_LEVEL_2M, pages_2m);
+		update_page_count(PG_LEVEL_4K, pages_4k);
+
+		/*
+		 * local global flush tlb, which will flush the previous
+		 * mappings present in both small and large page TLB's.
+		 */
+		__flush_tlb_all();
+
+		/*
+		 * Second iteration will set the actual desired PTE attributes.
+		 */
+		mapping_iter = 2;
+		goto repeat;
+	}
 }
 
 /*
@@ -719,7 +774,7 @@ void __init setup_bootmem_allocator(void)
 	after_init_bootmem = 1;
 }
 
-static void __init find_early_table_space(unsigned long end)
+static void __init find_early_table_space(unsigned long end, int use_pse)
 {
 	unsigned long puds, pmds, ptes, tables, start;
 
@@ -729,7 +784,7 @@ static void __init find_early_table_space(unsigned long end)
 	pmds = (end + PMD_SIZE - 1) >> PMD_SHIFT;
 	tables += PAGE_ALIGN(pmds * sizeof(pmd_t));
 
-	if (cpu_has_pse) {
+	if (use_pse) {
 		unsigned long extra;
 
 		extra = end - ((end>>PMD_SHIFT) << PMD_SHIFT);
@@ -769,12 +824,22 @@ unsigned long __init_refok init_memory_mapping(unsigned long start,
 	pgd_t *pgd_base = swapper_pg_dir;
 	unsigned long start_pfn, end_pfn;
 	unsigned long big_page_start;
+#ifdef CONFIG_DEBUG_PAGEALLOC
+	/*
+	 * For CONFIG_DEBUG_PAGEALLOC, identity mapping will use small pages.
+	 * This will simplify cpa(), which otherwise needs to support splitting
+	 * large pages into small in interrupt context, etc.
+	 */
+	int use_pse = 0;
+#else
+	int use_pse = cpu_has_pse;
+#endif
 
 	/*
 	 * Find space for the kernel direct mapping tables.
 	 */
 	if (!after_init_bootmem)
-		find_early_table_space(end);
+		find_early_table_space(end, use_pse);
 
 #ifdef CONFIG_X86_PAE
 	set_nx();
@@ -820,7 +885,7 @@ unsigned long __init_refok init_memory_mapping(unsigned long start,
 	end_pfn = (end>>PMD_SHIFT) << (PMD_SHIFT - PAGE_SHIFT);
 	if (start_pfn < end_pfn)
 		kernel_physical_mapping_init(pgd_base, start_pfn, end_pfn,
-						cpu_has_pse);
+					     use_pse);
 
 	/* tail is not big page alignment ? */
 	start_pfn = end_pfn;
@@ -983,7 +1048,6 @@ void __init mem_init(void)
 	if (boot_cpu_data.wp_works_ok < 0)
 		test_wp_bit();
 
-	cpa_init();
 	save_pg_dir();
 	zap_low_mappings();
 }
diff --git a/arch/x86/mm/init_64.c b/arch/x86/mm/init_64.c
index 770536ebf7e..fb30486c82f 100644
--- a/arch/x86/mm/init_64.c
+++ b/arch/x86/mm/init_64.c
@@ -271,7 +271,8 @@ static __ref void unmap_low_page(void *adr)
 }
 
 static unsigned long __meminit
-phys_pte_init(pte_t *pte_page, unsigned long addr, unsigned long end)
+phys_pte_init(pte_t *pte_page, unsigned long addr, unsigned long end,
+	      pgprot_t prot)
 {
 	unsigned pages = 0;
 	unsigned long last_map_addr = end;
@@ -289,36 +290,43 @@ phys_pte_init(pte_t *pte_page, unsigned long addr, unsigned long end)
 			break;
 		}
 
+		/*
+		 * We will re-use the existing mapping.
+		 * Xen for example has some special requirements, like mapping
+		 * pagetable pages as RO. So assume someone who pre-setup
+		 * these mappings are more intelligent.
+		 */
 		if (pte_val(*pte))
 			continue;
 
 		if (0)
 			printk("   pte=%p addr=%lx pte=%016lx\n",
 			       pte, addr, pfn_pte(addr >> PAGE_SHIFT, PAGE_KERNEL).pte);
-		set_pte(pte, pfn_pte(addr >> PAGE_SHIFT, PAGE_KERNEL));
-		last_map_addr = (addr & PAGE_MASK) + PAGE_SIZE;
 		pages++;
+		set_pte(pte, pfn_pte(addr >> PAGE_SHIFT, prot));
+		last_map_addr = (addr & PAGE_MASK) + PAGE_SIZE;
 	}
+
 	update_page_count(PG_LEVEL_4K, pages);
 
 	return last_map_addr;
 }
 
 static unsigned long __meminit
-phys_pte_update(pmd_t *pmd, unsigned long address, unsigned long end)
+phys_pte_update(pmd_t *pmd, unsigned long address, unsigned long end,
+		pgprot_t prot)
 {
 	pte_t *pte = (pte_t *)pmd_page_vaddr(*pmd);
 
-	return phys_pte_init(pte, address, end);
+	return phys_pte_init(pte, address, end, prot);
 }
 
 static unsigned long __meminit
 phys_pmd_init(pmd_t *pmd_page, unsigned long address, unsigned long end,
-			 unsigned long page_size_mask)
+	      unsigned long page_size_mask, pgprot_t prot)
 {
 	unsigned long pages = 0;
 	unsigned long last_map_addr = end;
-	unsigned long start = address;
 
 	int i = pmd_index(address);
 
@@ -326,6 +334,7 @@ phys_pmd_init(pmd_t *pmd_page, unsigned long address, unsigned long end,
 		unsigned long pte_phys;
 		pmd_t *pmd = pmd_page + pmd_index(address);
 		pte_t *pte;
+		pgprot_t new_prot = prot;
 
 		if (address >= end) {
 			if (!after_bootmem) {
@@ -339,27 +348,40 @@ phys_pmd_init(pmd_t *pmd_page, unsigned long address, unsigned long end,
 			if (!pmd_large(*pmd)) {
 				spin_lock(&init_mm.page_table_lock);
 				last_map_addr = phys_pte_update(pmd, address,
-								end);
+								end, prot);
 				spin_unlock(&init_mm.page_table_lock);
+				continue;
 			}
-			/* Count entries we're using from level2_ident_pgt */
-			if (start == 0)
-				pages++;
-			continue;
+			/*
+			 * If we are ok with PG_LEVEL_2M mapping, then we will
+			 * use the existing mapping,
+			 *
+			 * Otherwise, we will split the large page mapping but
+			 * use the same existing protection bits except for
+			 * large page, so that we don't violate Intel's TLB
+			 * Application note (317080) which says, while changing
+			 * the page sizes, new and old translations should
+			 * not differ with respect to page frame and
+			 * attributes.
+			 */
+			if (page_size_mask & (1 << PG_LEVEL_2M))
+				continue;
+			new_prot = pte_pgprot(pte_clrhuge(*(pte_t *)pmd));
 		}
 
 		if (page_size_mask & (1<<PG_LEVEL_2M)) {
 			pages++;
 			spin_lock(&init_mm.page_table_lock);
 			set_pte((pte_t *)pmd,
-				pfn_pte(address >> PAGE_SHIFT, PAGE_KERNEL_LARGE));
+				pfn_pte(address >> PAGE_SHIFT,
+					__pgprot(pgprot_val(prot) | _PAGE_PSE)));
 			spin_unlock(&init_mm.page_table_lock);
 			last_map_addr = (address & PMD_MASK) + PMD_SIZE;
 			continue;
 		}
 
 		pte = alloc_low_page(&pte_phys);
-		last_map_addr = phys_pte_init(pte, address, end);
+		last_map_addr = phys_pte_init(pte, address, end, new_prot);
 		unmap_low_page(pte);
 
 		spin_lock(&init_mm.page_table_lock);
@@ -372,12 +394,12 @@ phys_pmd_init(pmd_t *pmd_page, unsigned long address, unsigned long end,
 
 static unsigned long __meminit
 phys_pmd_update(pud_t *pud, unsigned long address, unsigned long end,
-			 unsigned long page_size_mask)
+		unsigned long page_size_mask, pgprot_t prot)
 {
 	pmd_t *pmd = pmd_offset(pud, 0);
 	unsigned long last_map_addr;
 
-	last_map_addr = phys_pmd_init(pmd, address, end, page_size_mask);
+	last_map_addr = phys_pmd_init(pmd, address, end, page_size_mask, prot);
 	__flush_tlb_all();
 	return last_map_addr;
 }
@@ -394,6 +416,7 @@ phys_pud_init(pud_t *pud_page, unsigned long addr, unsigned long end,
 		unsigned long pmd_phys;
 		pud_t *pud = pud_page + pud_index(addr);
 		pmd_t *pmd;
+		pgprot_t prot = PAGE_KERNEL;
 
 		if (addr >= end)
 			break;
@@ -405,10 +428,26 @@ phys_pud_init(pud_t *pud_page, unsigned long addr, unsigned long end,
 		}
 
 		if (pud_val(*pud)) {
-			if (!pud_large(*pud))
+			if (!pud_large(*pud)) {
 				last_map_addr = phys_pmd_update(pud, addr, end,
-							 page_size_mask);
-			continue;
+							 page_size_mask, prot);
+				continue;
+			}
+			/*
+			 * If we are ok with PG_LEVEL_1G mapping, then we will
+			 * use the existing mapping.
+			 *
+			 * Otherwise, we will split the gbpage mapping but use
+			 * the same existing protection  bits except for large
+			 * page, so that we don't violate Intel's TLB
+			 * Application note (317080) which says, while changing
+			 * the page sizes, new and old translations should
+			 * not differ with respect to page frame and
+			 * attributes.
+			 */
+			if (page_size_mask & (1 << PG_LEVEL_1G))
+				continue;
+			prot = pte_pgprot(pte_clrhuge(*(pte_t *)pud));
 		}
 
 		if (page_size_mask & (1<<PG_LEVEL_1G)) {
@@ -422,7 +461,8 @@ phys_pud_init(pud_t *pud_page, unsigned long addr, unsigned long end,
 		}
 
 		pmd = alloc_low_page(&pmd_phys);
-		last_map_addr = phys_pmd_init(pmd, addr, end, page_size_mask);
+		last_map_addr = phys_pmd_init(pmd, addr, end, page_size_mask,
+					      prot);
 		unmap_low_page(pmd);
 
 		spin_lock(&init_mm.page_table_lock);
@@ -430,6 +470,7 @@ phys_pud_init(pud_t *pud_page, unsigned long addr, unsigned long end,
 		spin_unlock(&init_mm.page_table_lock);
 	}
 	__flush_tlb_all();
+
 	update_page_count(PG_LEVEL_1G, pages);
 
 	return last_map_addr;
@@ -446,13 +487,14 @@ phys_pud_update(pgd_t *pgd, unsigned long addr, unsigned long end,
 	return phys_pud_init(pud, addr, end, page_size_mask);
 }
 
-static void __init find_early_table_space(unsigned long end)
+static void __init find_early_table_space(unsigned long end, int use_pse,
+					  int use_gbpages)
 {
 	unsigned long puds, pmds, ptes, tables, start;
 
 	puds = (end + PUD_SIZE - 1) >> PUD_SHIFT;
 	tables = roundup(puds * sizeof(pud_t), PAGE_SIZE);
-	if (direct_gbpages) {
+	if (use_gbpages) {
 		unsigned long extra;
 		extra = end - ((end>>PUD_SHIFT) << PUD_SHIFT);
 		pmds = (extra + PMD_SIZE - 1) >> PMD_SHIFT;
@@ -460,7 +502,7 @@ static void __init find_early_table_space(unsigned long end)
 		pmds = (end + PMD_SIZE - 1) >> PMD_SHIFT;
 	tables += roundup(pmds * sizeof(pmd_t), PAGE_SIZE);
 
-	if (cpu_has_pse) {
+	if (use_pse) {
 		unsigned long extra;
 		extra = end - ((end>>PMD_SHIFT) << PMD_SHIFT);
 		ptes = (extra + PAGE_SIZE - 1) >> PAGE_SHIFT;
@@ -528,6 +570,7 @@ static unsigned long __init kernel_physical_mapping_init(unsigned long start,
 		pgd_populate(&init_mm, pgd, __va(pud_phys));
 		spin_unlock(&init_mm.page_table_lock);
 	}
+	__flush_tlb_all();
 
 	return last_map_addr;
 }
@@ -571,6 +614,7 @@ unsigned long __init_refok init_memory_mapping(unsigned long start,
 
 	struct map_range mr[NR_RANGE_MR];
 	int nr_range, i;
+	int use_pse, use_gbpages;
 
 	printk(KERN_INFO "init_memory_mapping\n");
 
@@ -584,9 +628,21 @@ unsigned long __init_refok init_memory_mapping(unsigned long start,
 	if (!after_bootmem)
 		init_gbpages();
 
-	if (direct_gbpages)
+#ifdef CONFIG_DEBUG_PAGEALLOC
+	/*
+	 * For CONFIG_DEBUG_PAGEALLOC, identity mapping will use small pages.
+	 * This will simplify cpa(), which otherwise needs to support splitting
+	 * large pages into small in interrupt context, etc.
+	 */
+	use_pse = use_gbpages = 0;
+#else
+	use_pse = cpu_has_pse;
+	use_gbpages = direct_gbpages;
+#endif
+
+	if (use_gbpages)
 		page_size_mask |= 1 << PG_LEVEL_1G;
-	if (cpu_has_pse)
+	if (use_pse)
 		page_size_mask |= 1 << PG_LEVEL_2M;
 
 	memset(mr, 0, sizeof(mr));
@@ -647,7 +703,7 @@ unsigned long __init_refok init_memory_mapping(unsigned long start,
 			 (mr[i].page_size_mask & (1<<PG_LEVEL_2M))?"2M":"4k"));
 
 	if (!after_bootmem)
-		find_early_table_space(end);
+		find_early_table_space(end, use_pse, use_gbpages);
 
 	for (i = 0; i < nr_range; i++)
 		last_map_addr = kernel_physical_mapping_init(
@@ -806,8 +862,6 @@ void __init mem_init(void)
 		reservedpages << (PAGE_SHIFT-10),
 		datasize >> 10,
 		initsize >> 10);
-
-	cpa_init();
 }
 
 void free_init_pages(char *what, unsigned long begin, unsigned long end)
diff --git a/arch/x86/mm/ioremap.c b/arch/x86/mm/ioremap.c
index cac6da54203..6ab3196d12b 100644
--- a/arch/x86/mm/ioremap.c
+++ b/arch/x86/mm/ioremap.c
@@ -83,6 +83,25 @@ int page_is_ram(unsigned long pagenr)
 	return 0;
 }
 
+int pagerange_is_ram(unsigned long start, unsigned long end)
+{
+	int ram_page = 0, not_rampage = 0;
+	unsigned long page_nr;
+
+	for (page_nr = (start >> PAGE_SHIFT); page_nr < (end >> PAGE_SHIFT);
+	     ++page_nr) {
+		if (page_is_ram(page_nr))
+			ram_page = 1;
+		else
+			not_rampage = 1;
+
+		if (ram_page == not_rampage)
+			return -1;
+	}
+
+	return ram_page;
+}
+
 /*
  * Fix up the linear direct mapping of the kernel to avoid cache attribute
  * conflicts.
diff --git a/arch/x86/mm/pageattr-test.c b/arch/x86/mm/pageattr-test.c
index d4aa503caaa..e1d10690921 100644
--- a/arch/x86/mm/pageattr-test.c
+++ b/arch/x86/mm/pageattr-test.c
@@ -32,7 +32,7 @@ enum {
 	GPS			= (1<<30)
 };
 
-#define PAGE_TESTBIT	__pgprot(_PAGE_UNUSED1)
+#define PAGE_CPA_TEST	__pgprot(_PAGE_CPA_TEST)
 
 static int pte_testbit(pte_t pte)
 {
@@ -118,6 +118,7 @@ static int pageattr_test(void)
 	unsigned int level;
 	int i, k;
 	int err;
+	unsigned long test_addr;
 
 	if (print)
 		printk(KERN_INFO "CPA self-test:\n");
@@ -172,7 +173,8 @@ static int pageattr_test(void)
 			continue;
 		}
 
-		err = change_page_attr_set(addr[i], len[i], PAGE_TESTBIT);
+		test_addr = addr[i];
+		err = change_page_attr_set(&test_addr, len[i], PAGE_CPA_TEST, 0);
 		if (err < 0) {
 			printk(KERN_ERR "CPA %d failed %d\n", i, err);
 			failed++;
@@ -204,7 +206,8 @@ static int pageattr_test(void)
 			failed++;
 			continue;
 		}
-		err = change_page_attr_clear(addr[i], len[i], PAGE_TESTBIT);
+		test_addr = addr[i];
+		err = change_page_attr_clear(&test_addr, len[i], PAGE_CPA_TEST, 0);
 		if (err < 0) {
 			printk(KERN_ERR "CPA reverting failed: %d\n", err);
 			failed++;
diff --git a/arch/x86/mm/pageattr.c b/arch/x86/mm/pageattr.c
index 898fad617ab..a9ec89c3fbc 100644
--- a/arch/x86/mm/pageattr.c
+++ b/arch/x86/mm/pageattr.c
@@ -25,15 +25,27 @@
  * The current flushing context - we pass it instead of 5 arguments:
  */
 struct cpa_data {
-	unsigned long	vaddr;
+	unsigned long	*vaddr;
 	pgprot_t	mask_set;
 	pgprot_t	mask_clr;
 	int		numpages;
-	int		flushtlb;
+	int		flags;
 	unsigned long	pfn;
 	unsigned	force_split : 1;
+	int		curpage;
 };
 
+/*
+ * Serialize cpa() (for !DEBUG_PAGEALLOC which uses large identity mappings)
+ * using cpa_lock. So that we don't allow any other cpu, with stale large tlb
+ * entries change the page attribute in parallel to some other cpu
+ * splitting a large page entry along with changing the attribute.
+ */
+static DEFINE_SPINLOCK(cpa_lock);
+
+#define CPA_FLUSHTLB 1
+#define CPA_ARRAY 2
+
 #ifdef CONFIG_PROC_FS
 static unsigned long direct_pages_count[PG_LEVEL_NUM];
 
@@ -190,6 +202,41 @@ static void cpa_flush_range(unsigned long start, int numpages, int cache)
 	}
 }
 
+static void cpa_flush_array(unsigned long *start, int numpages, int cache)
+{
+	unsigned int i, level;
+	unsigned long *addr;
+
+	BUG_ON(irqs_disabled());
+
+	on_each_cpu(__cpa_flush_range, NULL, 1);
+
+	if (!cache)
+		return;
+
+	/* 4M threshold */
+	if (numpages >= 1024) {
+		if (boot_cpu_data.x86_model >= 4)
+			wbinvd();
+		return;
+	}
+	/*
+	 * We only need to flush on one CPU,
+	 * clflush is a MESI-coherent instruction that
+	 * will cause all other CPUs to flush the same
+	 * cachelines:
+	 */
+	for (i = 0, addr = start; i < numpages; i++, addr++) {
+		pte_t *pte = lookup_address(*addr, &level);
+
+		/*
+		 * Only flush present addresses:
+		 */
+		if (pte && (pte_val(*pte) & _PAGE_PRESENT))
+			clflush_cache_range((void *) *addr, PAGE_SIZE);
+	}
+}
+
 /*
  * Certain areas of memory on x86 require very specific protection flags,
  * for example the BIOS area or kernel text. Callers don't always get this
@@ -398,7 +445,7 @@ try_preserve_large_page(pte_t *kpte, unsigned long address,
 		 */
 		new_pte = pfn_pte(pte_pfn(old_pte), canon_pgprot(new_prot));
 		__set_pmd_pte(kpte, address, new_pte);
-		cpa->flushtlb = 1;
+		cpa->flags |= CPA_FLUSHTLB;
 		do_split = 0;
 	}
 
@@ -408,84 +455,6 @@ out_unlock:
 	return do_split;
 }
 
-static LIST_HEAD(page_pool);
-static unsigned long pool_size, pool_pages, pool_low;
-static unsigned long pool_used, pool_failed;
-
-static void cpa_fill_pool(struct page **ret)
-{
-	gfp_t gfp = GFP_KERNEL;
-	unsigned long flags;
-	struct page *p;
-
-	/*
-	 * Avoid recursion (on debug-pagealloc) and also signal
-	 * our priority to get to these pagetables:
-	 */
-	if (current->flags & PF_MEMALLOC)
-		return;
-	current->flags |= PF_MEMALLOC;
-
-	/*
-	 * Allocate atomically from atomic contexts:
-	 */
-	if (in_atomic() || irqs_disabled() || debug_pagealloc)
-		gfp =  GFP_ATOMIC | __GFP_NORETRY | __GFP_NOWARN;
-
-	while (pool_pages < pool_size || (ret && !*ret)) {
-		p = alloc_pages(gfp, 0);
-		if (!p) {
-			pool_failed++;
-			break;
-		}
-		/*
-		 * If the call site needs a page right now, provide it:
-		 */
-		if (ret && !*ret) {
-			*ret = p;
-			continue;
-		}
-		spin_lock_irqsave(&pgd_lock, flags);
-		list_add(&p->lru, &page_pool);
-		pool_pages++;
-		spin_unlock_irqrestore(&pgd_lock, flags);
-	}
-
-	current->flags &= ~PF_MEMALLOC;
-}
-
-#define SHIFT_MB		(20 - PAGE_SHIFT)
-#define ROUND_MB_GB		((1 << 10) - 1)
-#define SHIFT_MB_GB		10
-#define POOL_PAGES_PER_GB	16
-
-void __init cpa_init(void)
-{
-	struct sysinfo si;
-	unsigned long gb;
-
-	si_meminfo(&si);
-	/*
-	 * Calculate the number of pool pages:
-	 *
-	 * Convert totalram (nr of pages) to MiB and round to the next
-	 * GiB. Shift MiB to Gib and multiply the result by
-	 * POOL_PAGES_PER_GB:
-	 */
-	if (debug_pagealloc) {
-		gb = ((si.totalram >> SHIFT_MB) + ROUND_MB_GB) >> SHIFT_MB_GB;
-		pool_size = POOL_PAGES_PER_GB * gb;
-	} else {
-		pool_size = 1;
-	}
-	pool_low = pool_size;
-
-	cpa_fill_pool(NULL);
-	printk(KERN_DEBUG
-	       "CPA: page pool initialized %lu of %lu pages preallocated\n",
-	       pool_pages, pool_size);
-}
-
 static int split_large_page(pte_t *kpte, unsigned long address)
 {
 	unsigned long flags, pfn, pfninc = 1;
@@ -494,28 +463,15 @@ static int split_large_page(pte_t *kpte, unsigned long address)
 	pgprot_t ref_prot;
 	struct page *base;
 
-	/*
-	 * Get a page from the pool. The pool list is protected by the
-	 * pgd_lock, which we have to take anyway for the split
-	 * operation:
-	 */
-	spin_lock_irqsave(&pgd_lock, flags);
-	if (list_empty(&page_pool)) {
-		spin_unlock_irqrestore(&pgd_lock, flags);
-		base = NULL;
-		cpa_fill_pool(&base);
-		if (!base)
-			return -ENOMEM;
-		spin_lock_irqsave(&pgd_lock, flags);
-	} else {
-		base = list_first_entry(&page_pool, struct page, lru);
-		list_del(&base->lru);
-		pool_pages--;
-
-		if (pool_pages < pool_low)
-			pool_low = pool_pages;
-	}
+	if (!debug_pagealloc)
+		spin_unlock(&cpa_lock);
+	base = alloc_pages(GFP_KERNEL, 0);
+	if (!debug_pagealloc)
+		spin_lock(&cpa_lock);
+	if (!base)
+		return -ENOMEM;
 
+	spin_lock_irqsave(&pgd_lock, flags);
 	/*
 	 * Check for races, another CPU might have split this page
 	 * up for us already:
@@ -572,11 +528,8 @@ out_unlock:
 	 * If we dropped out via the lookup_address check under
 	 * pgd_lock then stick the page back into the pool:
 	 */
-	if (base) {
-		list_add(&base->lru, &page_pool);
-		pool_pages++;
-	} else
-		pool_used++;
+	if (base)
+		__free_page(base);
 	spin_unlock_irqrestore(&pgd_lock, flags);
 
 	return 0;
@@ -584,11 +537,16 @@ out_unlock:
 
 static int __change_page_attr(struct cpa_data *cpa, int primary)
 {
-	unsigned long address = cpa->vaddr;
+	unsigned long address;
 	int do_split, err;
 	unsigned int level;
 	pte_t *kpte, old_pte;
 
+	if (cpa->flags & CPA_ARRAY)
+		address = cpa->vaddr[cpa->curpage];
+	else
+		address = *cpa->vaddr;
+
 repeat: