x86: construct 32-bit boot time page tables in native format.

Specifically the boot time page tables in a CONFIG_X86_PAE=y enabled
kernel are in PAE format.

early_ioremap is updated to use the standard page table accessors.

Clear any mappings beyond max_low_pfn from the boot page tables in
native_pagetable_setup_start because the initial mappings can extend
beyond the range of physical memory and into the vmalloc area.

Derived from patches by Eric Biederman and H. Peter Anvin.

[ jeremy@goop.org: PAE swapper_pg_dir needs to be page-sized fix ]
Signed-off-by: Ian Campbell <ijc@hellion.org.uk>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Eric W. Biederman <ebiederm@xmission.com>
Cc: Andi Kleen <andi@firstfloor.org>
Cc: Mika Penttilä <mika.penttila@kolumbus.fi>
Cc: Jeremy Fitzhardinge <jeremy@goop.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

arch/x86/kernel/head_32.S

@@ -19,6 +19,10 @@
 #include <asm/thread_info.h>
 #include <asm/asm-offsets.h>
 #include <asm/setup.h>
+#include <asm/processor-flags.h>
+
+/* Physical address */
+#define pa(X) ((X) - __PAGE_OFFSET)
 
 /*
  * References to members of the new_cpu_data structure.
@@ -80,10 +84,6 @@ INIT_MAP_BEYOND_END = BOOTBITMAP_SIZE + (PAGE_TABLE_SIZE + ALLOCATOR_SLOP)*PAGE_
  */
 .section .text.head,"ax",@progbits
 ENTRY(startup_32)
-	/* check to see if KEEP_SEGMENTS flag is meaningful */
-	cmpw $0x207, BP_version(%esi)
-	jb 1f
-
 	/* test KEEP_SEGMENTS flag to see if the bootloader is asking
 		us to not reload segments */
 	testb $(1<<6), BP_loadflags(%esi)
@@ -92,7 +92,7 @@ ENTRY(startup_32)
 /*
  * Set segments to known values.
  */
-1:	lgdt boot_gdt_descr - __PAGE_OFFSET
+	lgdt pa(boot_gdt_descr)
 	movl $(__BOOT_DS),%eax
 	movl %eax,%ds
 	movl %eax,%es
@@ -105,8 +105,8 @@ ENTRY(startup_32)
  */
 	cld
 	xorl %eax,%eax
-	movl $__bss_start - __PAGE_OFFSET,%edi
-	movl $__bss_stop - __PAGE_OFFSET,%ecx
+	movl $pa(__bss_start),%edi
+	movl $pa(__bss_stop),%ecx
 	subl %edi,%ecx
 	shrl $2,%ecx
 	rep ; stosl
@@ -118,31 +118,32 @@ ENTRY(startup_32)
  * (kexec on panic case). Hence copy out the parameters before initializing
  * page tables.
  */
-	movl $(boot_params - __PAGE_OFFSET),%edi
+	movl $pa(boot_params),%edi
 	movl $(PARAM_SIZE/4),%ecx
 	cld
 	rep
 	movsl
-	movl boot_params - __PAGE_OFFSET + NEW_CL_POINTER,%esi
+	movl pa(boot_params) + NEW_CL_POINTER,%esi
 	andl %esi,%esi
 	jz 1f			# No comand line
-	movl $(boot_command_line - __PAGE_OFFSET),%edi
+	movl $pa(boot_command_line),%edi
 	movl $(COMMAND_LINE_SIZE/4),%ecx
 	rep
 	movsl
 1:
 
 #ifdef CONFIG_PARAVIRT
-	cmpw $0x207, (boot_params + BP_version - __PAGE_OFFSET)
+	/* This is can only trip for a broken bootloader... */
+	cmpw $0x207, pa(boot_params + BP_version)
 	jb default_entry
 
 	/* Paravirt-compatible boot parameters.  Look to see what architecture
 		we're booting under. */
-	movl (boot_params + BP_hardware_subarch - __PAGE_OFFSET), %eax
+	movl pa(boot_params + BP_hardware_subarch), %eax
 	cmpl $num_subarch_entries, %eax
 	jae bad_subarch
 
-	movl subarch_entries - __PAGE_OFFSET(,%eax,4), %eax
+	movl pa(subarch_entries)(,%eax,4), %eax
 	subl $__PAGE_OFFSET, %eax
 	jmp *%eax
 
@@ -170,17 +171,68 @@ num_subarch_entries = (. - subarch_entries) / 4
  * Mappings are created both at virtual address 0 (identity mapping)
  * and PAGE_OFFSET for up to _end+sizeof(page tables)+INIT_MAP_BEYOND_END.
  *
- * Warning: don't use %esi or the stack in this code.  However, %esp
- * can be used as a GPR if you really need it...
+ * Note that the stack is not yet set up!
  */
-page_pde_offset = (__PAGE_OFFSET >> 20);
+#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:
-	movl $(pg0 - __PAGE_OFFSET), %edi
-	movl $(swapper_pg_dir - __PAGE_OFFSET), %edx
-	movl $0x007, %eax			/* 0x007 = PRESENT+RW+USER */
+#ifdef CONFIG_X86_PAE
+
+	/*
+	 * In PAE mode swapper_pg_dir is statically defined to contain enough
+	 * entries to cover the VMSPLIT option (that is the top 1, 2 or 3
+	 * entries). The identity mapping is handled by pointing two PGD
+	 * entries to the first kernel PMD.
+	 *
+	 * Note the upper half of each PMD or PTE are always zero at
+	 * this stage.
+	 */
+
+#define KPMDS ((0x100000000-__PAGE_OFFSET) >> 30) /* Number of kernel PMDs */
+
+	xorl %ebx,%ebx				/* %ebx is kept at zero */
+
+	movl $pa(pg0), %edi
+	movl $pa(swapper_pg_pmd), %edx
+	movl $PTE_ATTR, %eax
+10:
+	leal PDE_ATTR(%edi),%ecx		/* Create PMD entry */
+	movl %ecx,(%edx)			/* Store PMD entry */
+						/* Upper half already zero */
+	addl $8,%edx
+	movl $512,%ecx
+11:
+	stosl
+	xchgl %eax,%ebx
+	stosl
+	xchgl %eax,%ebx
+	addl $0x1000,%eax
+	loop 11b
+
+	/*
+	 * 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
+	cmpl %ebp,%eax
+	jb 10b
+1:
+	movl %edi,pa(init_pg_tables_end)
+
+	/* Do early initialization of the fixmap area */
+	movl $pa(swapper_pg_fixmap)+PDE_ATTR,%eax
+	movl %eax,pa(swapper_pg_pmd+0x1000*KPMDS-8)
+#else	/* Not PAE */
+
+page_pde_offset = (__PAGE_OFFSET >> 20);
+
+	movl $pa(pg0), %edi
+	movl $pa(swapper_pg_dir), %edx
+	movl $PTE_ATTR, %eax
 10:
-	leal 0x007(%edi),%ecx			/* Create PDE entry */
+	leal PDE_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
@@ -189,19 +241,20 @@ default_entry:
 	stosl
 	addl $0x1000,%eax
 	loop 11b
-	/* End condition: we must map up to and including INIT_MAP_BEYOND_END */
-	/* bytes beyond the end of our own page tables; the +0x007 is the attribute bits */
-	leal (INIT_MAP_BEYOND_END+0x007)(%edi),%ebp
+	/*
+	 * End condition: we must map up to and including INIT_MAP_BEYOND_END
+	 * bytes beyond the end of our own page tables; the +0x007 is
+	 * the attribute bits
+	 */
+	leal (INIT_MAP_BEYOND_END+PTE_ATTR)(%edi),%ebp
 	cmpl %ebp,%eax
 	jb 10b
-	movl %edi,(init_pg_tables_end - __PAGE_OFFSET)
-
-	/* Do an early initialization of the fixmap area */
-	movl $(swapper_pg_dir - __PAGE_OFFSET), %edx
-	movl $(swapper_pg_pmd - __PAGE_OFFSET), %eax
-	addl $0x67, %eax			/* 0x67 == _PAGE_TABLE */
-	movl %eax, 4092(%edx)
+	movl %edi,pa(init_pg_tables_end)
 
+	/* Do early initialization of the fixmap area */
+	movl $pa(swapper_pg_fixmap)+PDE_ATTR,%eax
+	movl %eax,pa(swapper_pg_dir+0xffc)
+#endif
 	jmp 3f
 /*
  * Non-boot CPU entry point; entered from trampoline.S
@@ -241,7 +294,7 @@ ENTRY(startup_32_smp)
  *	NOTE! We have to correct for the fact that we're
  *	not yet offset PAGE_OFFSET..
  */
-#define cr4_bits mmu_cr4_features-__PAGE_OFFSET
+#define cr4_bits pa(mmu_cr4_features)
 	movl cr4_bits,%edx
 	andl %edx,%edx
 	jz 6f
@@ -276,10 +329,10 @@ ENTRY(startup_32_smp)
 /*
  * Enable paging
  */
-	movl $swapper_pg_dir-__PAGE_OFFSET,%eax
+	movl $pa(swapper_pg_dir),%eax
 	movl %eax,%cr3		/* set the page table pointer.. */
 	movl %cr0,%eax
-	orl $0x80000000,%eax
+	orl  $X86_CR0_PG,%eax
 	movl %eax,%cr0		/* ..and set paging (PG) bit */
 	ljmp $__BOOT_CS,$1f	/* Clear prefetch and normalize %eip */
 1:
@@ -552,16 +605,44 @@ ENTRY(_stext)
  */
 .section ".bss.page_aligned","wa"
 	.align PAGE_SIZE_asm
+#ifdef CONFIG_X86_PAE
+ENTRY(swapper_pg_pmd)
+	.fill 1024*KPMDS,4,0
+#else
 ENTRY(swapper_pg_dir)
 	.fill 1024,4,0
-ENTRY(swapper_pg_pmd)
+#endif
+ENTRY(swapper_pg_fixmap)
 	.fill 1024,4,0
 ENTRY(empty_zero_page)
 	.fill 4096,1,0
-
 /*
  * This starts the data section.
  */
+#ifdef CONFIG_X86_PAE
+.section ".data.page_aligned","wa"
+	/* 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 */
+# 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
+# elif KPMDS == 2
+	.long	0,0
+	.long	pa(swapper_pg_pmd+PGD_ATTR),0
+	.long	pa(swapper_pg_pmd+PGD_ATTR+0x1000),0
+# elif KPMDS == 1
+	.long	0,0
+	.long	0,0
+	.long	pa(swapper_pg_pmd+PGD_ATTR),0
+# else
+#  error "Kernel PMDs should be 1, 2 or 3"
+# endif
+	.align PAGE_SIZE_asm		/* needs to be page-sized too */
+#endif
+
 .data
 ENTRY(stack_start)
 	.long init_thread_union+THREAD_SIZE

arch/x86/kernel/setup_32.c

@@ -154,7 +154,11 @@ struct cpuinfo_x86 new_cpu_data __cpuinitdata = { 0, 0, 0, 0, -1, 1, 0, 0, -1 };
 struct cpuinfo_x86 boot_cpu_data __read_mostly = { 0, 0, 0, 0, -1, 1, 0, 0, -1 };
 EXPORT_SYMBOL(boot_cpu_data);
 
+#ifndef CONFIG_X86_PAE
 unsigned long mmu_cr4_features;
+#else
+unsigned long mmu_cr4_features = X86_CR4_PAE;
+#endif
 
 /* for MCA, but anyone else can use it if they want */
 unsigned int machine_id;

arch/x86/mm/init_32.c

@@ -46,6 +46,7 @@
 #include <asm/pgalloc.h>
 #include <asm/sections.h>
 #include <asm/paravirt.h>
+#include <asm/setup.h>
 
 unsigned int __VMALLOC_RESERVE = 128 << 20;
 
@@ -328,44 +329,38 @@ pteval_t __PAGE_KERNEL_EXEC = _PAGE_KERNEL_EXEC;
 
 void __init native_pagetable_setup_start(pgd_t *base)
 {
-#ifdef CONFIG_X86_PAE
-	int i;
+	unsigned long pfn, va;
+	pgd_t *pgd;
+	pud_t *pud;
+	pmd_t *pmd;
+	pte_t *pte;
 
 	/*
-	 * Init entries of the first-level page table to the
-	 * zero page, if they haven't already been set up.
-	 *
-	 * In a normal native boot, we'll be running on a
-	 * pagetable rooted in swapper_pg_dir, but not in PAE
-	 * mode, so this will end up clobbering the mappings
-	 * for the lower 24Mbytes of the address space,
-	 * without affecting the kernel address space.
+	 * Remove any mappings which extend past the end of physical
+	 * memory from the boot time page table:
 	 */
-	for (i = 0; i < USER_PTRS_PER_PGD; i++)
-		set_pgd(&base[i],
-			__pgd(__pa(empty_zero_page) | _PAGE_PRESENT));
-
-	/* Make sure kernel address space is empty so that a pagetable
-	   will be allocated for it. */
-	memset(&base[USER_PTRS_PER_PGD], 0,
-	       KERNEL_PGD_PTRS * sizeof(pgd_t));
-#else
+	for (pfn = max_low_pfn + 1; pfn < 1<<(32-PAGE_SHIFT); pfn++) {
+		va = PAGE_OFFSET + (pfn<<PAGE_SHIFT);
+		pgd = base + pgd_index(va);
+		if (!pgd_present(*pgd))
+			break;
+
+		pud = pud_offset(pgd, va);
+		pmd = pmd_offset(pud, va);
+		if (!pmd_present(*pmd))
+			break;
+
+		pte = pte_offset_kernel(pmd, va);
+		if (!pte_present(*pte))
+			break;
+
+		pte_clear(NULL, va, pte);
+	}
 	paravirt_alloc_pd(&init_mm, __pa(base) >> PAGE_SHIFT);
-#endif
 }
 
 void __init native_pagetable_setup_done(pgd_t *base)
 {
-#ifdef CONFIG_X86_PAE
-	/*
-	 * Add low memory identity-mappings - SMP needs it when
-	 * starting up on an AP from real-mode. In the non-PAE
-	 * case we already have these mappings through head.S.
-	 * All user-space mappings are explicitly cleared after
-	 * SMP startup.
-	 */
-	set_pgd(&base[0], base[USER_PTRS_PER_PGD]);
-#endif
 }
 
 /*
@@ -374,9 +369,8 @@ void __init native_pagetable_setup_done(pgd_t *base)
  * the boot process.
  *
  * If we're booting on native hardware, this will be a pagetable
- * constructed in arch/i386/kernel/head.S, and not running in PAE mode
- * (even if we'll end up running in PAE).  The root of the pagetable
- * will be swapper_pg_dir.
+ * constructed in arch/x86/kernel/head_32.S.  The root of the
+ * pagetable will be swapper_pg_dir.
  *
  * If we're booting paravirtualized under a hypervisor, then there are
  * more options: we may already be running PAE, and the pagetable may
@@ -537,14 +531,6 @@ void __init paging_init(void)
 
 	load_cr3(swapper_pg_dir);
 
-#ifdef CONFIG_X86_PAE
-	/*
-	 * We will bail out later - printk doesn't work right now so
-	 * the user would just see a hanging kernel.
-	 */
-	if (cpu_has_pae)
-		set_in_cr4(X86_CR4_PAE);
-#endif
 	__flush_tlb_all();
 
 	kmap_init();
@@ -675,10 +661,6 @@ void __init mem_init(void)
 	BUG_ON((unsigned long)high_memory		> VMALLOC_START);
 #endif /* double-sanity-check paranoia */
 
-#ifdef CONFIG_X86_PAE
-	if (!cpu_has_pae)
-		panic("cannot execute a PAE-enabled kernel on a PAE-less CPU!");
-#endif
 	if (boot_cpu_data.wp_works_ok < 0)
 		test_wp_bit();
 

arch/x86/mm/ioremap.c

@@ -260,37 +260,42 @@ static int __init early_ioremap_debug_setup(char *str)
 early_param("early_ioremap_debug", early_ioremap_debug_setup);
 
 static __initdata int after_paging_init;
-static __initdata unsigned long bm_pte[1024]
+static __initdata pte_t bm_pte[PAGE_SIZE/sizeof(pte_t)]
 				__attribute__((aligned(PAGE_SIZE)));
 
-static inline unsigned long * __init early_ioremap_pgd(unsigned long addr)
+static inline pmd_t * __init early_ioremap_pmd(unsigned long addr)
 {
-	return (unsigned long *)swapper_pg_dir + ((addr >> 22) & 1023);
+	pgd_t *pgd = &swapper_pg_dir[pgd_index(addr)];
+	pud_t *pud = pud_offset(pgd, addr);
+	pmd_t *pmd = pmd_offset(pud, addr);
+
+	return pmd;
 }
 
-static inline unsigned long * __init early_ioremap_pte(unsigned long addr)
+static inline pte_t * __init early_ioremap_pte(unsigned long addr)
 {
-	return bm_pte + ((addr >> PAGE_SHIFT) & 1023);
+	return &bm_pte[pte_index(addr)];
 }
 
 void __init early_ioremap_init(void)
 {
-	unsigned long *pgd;
+	pmd_t *pmd;
 
 	if (early_ioremap_debug)
 		printk(KERN_INFO "early_ioremap_init()\n");
 
-	pgd = early_ioremap_pgd(fix_to_virt(FIX_BTMAP_BEGIN));
-	*pgd = __pa(bm_pte) | _PAGE_TABLE;
+	pmd = early_ioremap_pmd(fix_to_virt(FIX_BTMAP_BEGIN));
 	memset(bm_pte, 0, sizeof(bm_pte));
+	set_pmd(pmd, __pmd(__pa(bm_pte) | _PAGE_TABLE));
+
 	/*
-	 * The boot-ioremap range spans multiple pgds, for which
+	 * The boot-ioremap range spans multiple pmds, for which
 	 * we are not prepared:
 	 */
-	if (pgd != early_ioremap_pgd(fix_to_virt(FIX_BTMAP_END))) {
+	if (pmd != early_ioremap_pmd(fix_to_virt(FIX_BTMAP_END))) {
 		WARN_ON(1);
-		printk(KERN_WARNING "pgd %p != %p\n",
-		       pgd, early_ioremap_pgd(fix_to_virt(FIX_BTMAP_END)));
+		printk(KERN_WARNING "pmd %p != %p\n",
+		       pmd, early_ioremap_pmd(fix_to_virt(FIX_BTMAP_END)));
 		printk(KERN_WARNING "fix_to_virt(FIX_BTMAP_BEGIN): %08lx\n",
 			fix_to_virt(FIX_BTMAP_BEGIN));
 		printk(KERN_WARNING "fix_to_virt(FIX_BTMAP_END):   %08lx\n",
@@ -304,28 +309,29 @@ void __init early_ioremap_init(void)
 
 void __init early_ioremap_clear(void)
 {
-	unsigned long *pgd;
+	pmd_t *pmd;
 
 	if (early_ioremap_debug)
 		printk(KERN_INFO "early_ioremap_clear()\n");
 
-	pgd = early_ioremap_pgd(fix_to_virt(FIX_BTMAP_BEGIN));
-	*pgd = 0;
-	paravirt_release_pt(__pa(pgd) >> PAGE_SHIFT);
+	pmd = early_ioremap_pmd(fix_to_virt(FIX_BTMAP_BEGIN));
+	pmd_clear(pmd);
+	paravirt_release_pt(__pa(pmd) >> PAGE_SHIFT);
 	__flush_tlb_all();
 }
 
 void __init early_ioremap_reset(void)
 {
 	enum fixed_addresses idx;
-	unsigned long *pte, phys, addr;
+	unsigned long addr, phys;
+	pte_t *pte;
 
 	after_paging_init = 1;
 	for (idx = FIX_BTMAP_BEGIN; idx >= FIX_BTMAP_END; idx--) {
 		addr = fix_to_virt(idx);
 		pte = early_ioremap_pte(addr);
-		if (*pte & _PAGE_PRESENT) {
-			phys = *pte & PAGE_MASK;
+		if (pte_present(*pte)) {
+			phys = pte_val(*pte) & PAGE_MASK;
 			set_fixmap(idx, phys);
 		}
 	}
@@ -334,7 +340,8 @@ void __init early_ioremap_reset(void)
 static void __init __early_set_fixmap(enum fixed_addresses idx,
 				   unsigned long phys, pgprot_t flags)
 {
-	unsigned long *pte, addr = __fix_to_virt(idx);
+	unsigned long addr = __fix_to_virt(idx);
+	pte_t *pte;
 
 	if (idx >= __end_of_fixed_addresses) {
 		BUG();
@@ -342,9 +349,9 @@ static void __init __early_set_fixmap(enum fixed_addresses idx,
 	}
 	pte = early_ioremap_pte(addr);
 	if (pgprot_val(flags))
-		*pte = (phys & PAGE_MASK) | pgprot_val(flags);
+		set_pte(pte, pfn_pte(phys >> PAGE_SHIFT, flags));
 	else
-		*pte = 0;
+		pte_clear(NULL, addr, pte);
 	__flush_tlb_one(addr);
 }
 

include/asm-x86/page_32.h

@@ -48,7 +48,6 @@ typedef unsigned long	pgprotval_t;
 typedef unsigned long	phys_addr_t;
 
 typedef union { pteval_t pte, pte_low; } pte_t;
-typedef pte_t boot_pte_t;
 
 #endif	/* __ASSEMBLY__ */
 #endif	/* CONFIG_X86_PAE */

include/asm-x86/pgtable_32.h

@@ -52,10 +52,6 @@ void paging_init(void);
 #define USER_PGD_PTRS (PAGE_OFFSET >> PGDIR_SHIFT)
 #define KERNEL_PGD_PTRS (PTRS_PER_PGD-USER_PGD_PTRS)
 
-#define TWOLEVEL_PGDIR_SHIFT	22
-#define BOOT_USER_PGD_PTRS (__PAGE_OFFSET >> TWOLEVEL_PGDIR_SHIFT)
-#define BOOT_KERNEL_PGD_PTRS (1024-BOOT_USER_PGD_PTRS)
-
 /* Just any arbitrary offset to the start of the vmalloc VM area: the
  * current 8MB value just means that there will be a 8MB "hole" after the
  * physical memory until the kernel virtual memory starts.  That means that