[deliverable/linux.git] / arch / x86 / mm / dump_pagetables.c

/*
 * Debug helper to dump the current kernel pagetables of the system
 * so that we can see what the various memory ranges are set to.
 *
 * (C) Copyright 2008 Intel Corporation
 *
 * Author: Arjan van de Ven <arjan@linux.intel.com>
 *
 * This program is free software; you can redistribute it and/or
 * modify it under the terms of the GNU General Public License
 * as published by the Free Software Foundation; version 2
 * of the License.
 */

#include <linux/debugfs.h>
#include <linux/mm.h>
#include <linux/module.h>
#include <linux/seq_file.h>

#include <asm/pgtable.h>

/*
 * The dumper groups pagetable entries of the same type into one, and for
 * that it needs to keep some state when walking, and flush this state
 * when a "break" in the continuity is found.
 */
struct pg_state {
	int level;
	pgprot_t current_prot;
	unsigned long start_address;
	unsigned long current_address;
	const struct addr_marker *marker;
	bool to_dmesg;
};

struct addr_marker {
	unsigned long start_address;
	const char *name;
};

/* indices for address_markers; keep sync'd w/ address_markers below */
enum address_markers_idx {
	USER_SPACE_NR = 0,
#ifdef CONFIG_X86_64
	KERNEL_SPACE_NR,
	LOW_KERNEL_NR,
	VMALLOC_START_NR,
	VMEMMAP_START_NR,
	HIGH_KERNEL_NR,
	MODULES_VADDR_NR,
	MODULES_END_NR,
#else
	KERNEL_SPACE_NR,
	VMALLOC_START_NR,
	VMALLOC_END_NR,
# ifdef CONFIG_HIGHMEM
	PKMAP_BASE_NR,
# endif
	FIXADDR_START_NR,
#endif
};

/* Address space markers hints */
static struct addr_marker address_markers[] = {
	{ 0, "User Space" },
#ifdef CONFIG_X86_64
	{ 0x8000000000000000UL, "Kernel Space" },
	{ PAGE_OFFSET,		"Low Kernel Mapping" },
	{ VMALLOC_START,        "vmalloc() Area" },
	{ VMEMMAP_START,        "Vmemmap" },
	{ __START_KERNEL_map,   "High Kernel Mapping" },
	{ MODULES_VADDR,        "Modules" },
	{ MODULES_END,          "End Modules" },
#else
	{ PAGE_OFFSET,          "Kernel Mapping" },
	{ 0/* VMALLOC_START */, "vmalloc() Area" },
	{ 0/*VMALLOC_END*/,     "vmalloc() End" },
# ifdef CONFIG_HIGHMEM
	{ 0/*PKMAP_BASE*/,      "Persisent kmap() Area" },
# endif
	{ 0/*FIXADDR_START*/,   "Fixmap Area" },
#endif
	{ -1, NULL }		/* End of list */
};

/* Multipliers for offsets within the PTEs */
#define PTE_LEVEL_MULT (PAGE_SIZE)
#define PMD_LEVEL_MULT (PTRS_PER_PTE * PTE_LEVEL_MULT)
#define PUD_LEVEL_MULT (PTRS_PER_PMD * PMD_LEVEL_MULT)
#define PGD_LEVEL_MULT (PTRS_PER_PUD * PUD_LEVEL_MULT)

#define pt_dump_seq_printf(m, to_dmesg, fmt, args...)		\
({								\
	if (to_dmesg)					\
		printk(KERN_INFO fmt, ##args);			\
	else							\
		if (m)						\
			seq_printf(m, fmt, ##args);		\
})

#define pt_dump_cont_printf(m, to_dmesg, fmt, args...)		\
({								\
	if (to_dmesg)					\
		printk(KERN_CONT fmt, ##args);			\
	else							\
		if (m)						\
			seq_printf(m, fmt, ##args);		\
})

/*
 * Print a readable form of a pgprot_t to the seq_file
 */
static void printk_prot(struct seq_file *m, pgprot_t prot, int level, bool dmsg)
{
	pgprotval_t pr = pgprot_val(prot);
	static const char * const level_name[] =
		{ "cr3", "pgd", "pud", "pmd", "pte" };

	if (!pgprot_val(prot)) {
		/* Not present */
		pt_dump_cont_printf(m, dmsg, "                          ");
	} else {
		if (pr & _PAGE_USER)
			pt_dump_cont_printf(m, dmsg, "USR ");
		else
			pt_dump_cont_printf(m, dmsg, "    ");
		if (pr & _PAGE_RW)
			pt_dump_cont_printf(m, dmsg, "RW ");
		else
			pt_dump_cont_printf(m, dmsg, "ro ");
		if (pr & _PAGE_PWT)
			pt_dump_cont_printf(m, dmsg, "PWT ");
		else
			pt_dump_cont_printf(m, dmsg, "    ");
		if (pr & _PAGE_PCD)
			pt_dump_cont_printf(m, dmsg, "PCD ");
		else
			pt_dump_cont_printf(m, dmsg, "    ");

		/* Bit 9 has a different meaning on level 3 vs 4 */
		if (level <= 3) {
			if (pr & _PAGE_PSE)
				pt_dump_cont_printf(m, dmsg, "PSE ");
			else
				pt_dump_cont_printf(m, dmsg, "    ");
		} else {
			if (pr & _PAGE_PAT)
				pt_dump_cont_printf(m, dmsg, "pat ");
			else
				pt_dump_cont_printf(m, dmsg, "    ");
		}
		if (pr & _PAGE_GLOBAL)
			pt_dump_cont_printf(m, dmsg, "GLB ");
		else
			pt_dump_cont_printf(m, dmsg, "    ");
		if (pr & _PAGE_NX)
			pt_dump_cont_printf(m, dmsg, "NX ");
		else
			pt_dump_cont_printf(m, dmsg, "x  ");
	}
	pt_dump_cont_printf(m, dmsg, "%s\n", level_name[level]);
}

/*
 * On 64 bits, sign-extend the 48 bit address to 64 bit
 */
static unsigned long normalize_addr(unsigned long u)
{
#ifdef CONFIG_X86_64
	return (signed long)(u << 16) >> 16;
#else
	return u;
#endif
}

/*
 * This function gets called on a break in a continuous series
 * of PTE entries; the next one is different so we need to
 * print what we collected so far.
 */
static void note_page(struct seq_file *m, struct pg_state *st,
		      pgprot_t new_prot, int level)
{
	pgprotval_t prot, cur;
	static const char units[] = "KMGTPE";

	/*
	 * If we have a "break" in the series, we need to flush the state that
	 * we have now. "break" is either changing perms, levels or
	 * address space marker.
	 */
	prot = pgprot_val(new_prot) & PTE_FLAGS_MASK;
	cur = pgprot_val(st->current_prot) & PTE_FLAGS_MASK;

	if (!st->level) {
		/* First entry */
		st->current_prot = new_prot;
		st->level = level;
		st->marker = address_markers;
		pt_dump_seq_printf(m, st->to_dmesg, "---[ %s ]---\n",
				   st->marker->name);
	} else if (prot != cur || level != st->level ||
		   st->current_address >= st->marker[1].start_address) {
		const char *unit = units;
		unsigned long delta;
		int width = sizeof(unsigned long) * 2;

		/*
		 * Now print the actual finished series
		 */
		pt_dump_seq_printf(m, st->to_dmesg,  "0x%0*lx-0x%0*lx   ",
				   width, st->start_address,
				   width, st->current_address);

		delta = (st->current_address - st->start_address) >> 10;
		while (!(delta & 1023) && unit[1]) {
			delta >>= 10;
			unit++;
		}
		pt_dump_cont_printf(m, st->to_dmesg, "%9lu%c ", delta, *unit);
		printk_prot(m, st->current_prot, st->level, st->to_dmesg);

		/*
		 * We print markers for special areas of address space,
		 * such as the start of vmalloc space etc.
		 * This helps in the interpretation.
		 */
		if (st->current_address >= st->marker[1].start_address) {
			st->marker++;
			pt_dump_seq_printf(m, st->to_dmesg, "---[ %s ]---\n",
					   st->marker->name);
		}

		st->start_address = st->current_address;
		st->current_prot = new_prot;
		st->level = level;
	}
}

static void walk_pte_level(struct seq_file *m, struct pg_state *st, pmd_t addr,
							unsigned long P)
{
	int i;
	pte_t *start;

	start = (pte_t *) pmd_page_vaddr(addr);
	for (i = 0; i < PTRS_PER_PTE; i++) {
		pgprot_t prot = pte_pgprot(*start);

		st->current_address = normalize_addr(P + i * PTE_LEVEL_MULT);
		note_page(m, st, prot, 4);
		start++;
	}
}

#if PTRS_PER_PMD > 1

static void walk_pmd_level(struct seq_file *m, struct pg_state *st, pud_t addr,
							unsigned long P)
{
	int i;
	pmd_t *start;

	start = (pmd_t *) pud_page_vaddr(addr);
	for (i = 0; i < PTRS_PER_PMD; i++) {
		st->current_address = normalize_addr(P + i * PMD_LEVEL_MULT);
		if (!pmd_none(*start)) {
			pgprotval_t prot = pmd_val(*start) & PTE_FLAGS_MASK;

			if (pmd_large(*start) || !pmd_present(*start))
				note_page(m, st, __pgprot(prot), 3);
			else
				walk_pte_level(m, st, *start,
					       P + i * PMD_LEVEL_MULT);
		} else
			note_page(m, st, __pgprot(0), 3);
		start++;
	}
}

#else
#define walk_pmd_level(m,s,a,p) walk_pte_level(m,s,__pmd(pud_val(a)),p)
#define pud_large(a) pmd_large(__pmd(pud_val(a)))
#define pud_none(a)  pmd_none(__pmd(pud_val(a)))
#endif

#if PTRS_PER_PUD > 1

static void walk_pud_level(struct seq_file *m, struct pg_state *st, pgd_t addr,
							unsigned long P)
{
	int i;
	pud_t *start;

	start = (pud_t *) pgd_page_vaddr(addr);

	for (i = 0; i < PTRS_PER_PUD; i++) {
		st->current_address = normalize_addr(P + i * PUD_LEVEL_MULT);
		if (!pud_none(*start)) {
			pgprotval_t prot = pud_val(*start) & PTE_FLAGS_MASK;

			if (pud_large(*start) || !pud_present(*start))
				note_page(m, st, __pgprot(prot), 2);
			else
				walk_pmd_level(m, st, *start,
					       P + i * PUD_LEVEL_MULT);
		} else
			note_page(m, st, __pgprot(0), 2);

		start++;
	}
}

#else
#define walk_pud_level(m,s,a,p) walk_pmd_level(m,s,__pud(pgd_val(a)),p)
#define pgd_large(a) pud_large(__pud(pgd_val(a)))
#define pgd_none(a)  pud_none(__pud(pgd_val(a)))
#endif

void ptdump_walk_pgd_level(struct seq_file *m, pgd_t *pgd)
{
#ifdef CONFIG_X86_64
	pgd_t *start = (pgd_t *) &init_level4_pgt;
#else
	pgd_t *start = swapper_pg_dir;
#endif
	int i;
	struct pg_state st = {};

	if (pgd) {
		start = pgd;
		st.to_dmesg = true;
	}

	for (i = 0; i < PTRS_PER_PGD; i++) {
		st.current_address = normalize_addr(i * PGD_LEVEL_MULT);
		if (!pgd_none(*start)) {
			pgprotval_t prot = pgd_val(*start) & PTE_FLAGS_MASK;

			if (pgd_large(*start) || !pgd_present(*start))
				note_page(m, &st, __pgprot(prot), 1);
			else
				walk_pud_level(m, &st, *start,
					       i * PGD_LEVEL_MULT);
		} else
			note_page(m, &st, __pgprot(0), 1);

		start++;
	}

	/* Flush out the last page */
	st.current_address = normalize_addr(PTRS_PER_PGD*PGD_LEVEL_MULT);
	note_page(m, &st, __pgprot(0), 0);
}

static int ptdump_show(struct seq_file *m, void *v)
{
	ptdump_walk_pgd_level(m, NULL);
	return 0;
}

static int ptdump_open(struct inode *inode, struct file *filp)
{
	return single_open(filp, ptdump_show, NULL);
}

static const struct file_operations ptdump_fops = {
	.open		= ptdump_open,
	.read		= seq_read,
	.llseek		= seq_lseek,
	.release	= single_release,
};

static int pt_dump_init(void)
{
	struct dentry *pe;

#ifdef CONFIG_X86_32
	/* Not a compile-time constant on x86-32 */
	address_markers[VMALLOC_START_NR].start_address = VMALLOC_START;
	address_markers[VMALLOC_END_NR].start_address = VMALLOC_END;
# ifdef CONFIG_HIGHMEM
	address_markers[PKMAP_BASE_NR].start_address = PKMAP_BASE;
# endif
	address_markers[FIXADDR_START_NR].start_address = FIXADDR_START;
#endif

	pe = debugfs_create_file("kernel_page_tables", 0600, NULL, NULL,
				 &ptdump_fops);
	if (!pe)
		return -ENOMEM;

	return 0;
}

__initcall(pt_dump_init);
MODULE_LICENSE("GPL");
MODULE_AUTHOR("Arjan van de Ven <arjan@linux.intel.com>");
MODULE_DESCRIPTION("Kernel debugging helper that dumps pagetables");
Commit	Line	Data
926e5392 AV	1	/*
	2	* Debug helper to dump the current kernel pagetables of the system
	3	* so that we can see what the various memory ranges are set to.
	4	*
	5	* (C) Copyright 2008 Intel Corporation
	6	*
	7	* Author: Arjan van de Ven <arjan@linux.intel.com>
	8	*
	9	* This program is free software; you can redistribute it and/or
	10	* modify it under the terms of the GNU General Public License
	11	* as published by the Free Software Foundation; version 2
	12	* of the License.
	13	*/
	14
fe770bf0 PA	15	#include <linux/debugfs.h>
fe770bf0 PA	16	#include <linux/mm.h>
926e5392 AV	17	#include <linux/module.h>
926e5392 AV	18	#include <linux/seq_file.h>
926e5392 AV	19
	20	#include <asm/pgtable.h>
	21
	22	/*
	23	* The dumper groups pagetable entries of the same type into one, and for
	24	* that it needs to keep some state when walking, and flush this state
	25	* when a "break" in the continuity is found.
	26	*/
	27	struct pg_state {
	28	int level;
	29	pgprot_t current_prot;
	30	unsigned long start_address;
	31	unsigned long current_address;
fe770bf0	32	const struct addr_marker *marker;
ef6bea6d	33	bool to_dmesg;
926e5392 AV	34	};
926e5392 AV	35
fe770bf0 PA	36	struct addr_marker {
	37	unsigned long start_address;
	38	const char *name;
	39	};
	40
92851e2f AS	41	/* indices for address_markers; keep sync'd w/ address_markers below */
	42	enum address_markers_idx {
	43	USER_SPACE_NR = 0,
	44	#ifdef CONFIG_X86_64
	45	KERNEL_SPACE_NR,
	46	LOW_KERNEL_NR,
	47	VMALLOC_START_NR,
	48	VMEMMAP_START_NR,
	49	HIGH_KERNEL_NR,
	50	MODULES_VADDR_NR,
	51	MODULES_END_NR,
	52	#else
	53	KERNEL_SPACE_NR,
	54	VMALLOC_START_NR,
	55	VMALLOC_END_NR,
	56	# ifdef CONFIG_HIGHMEM
	57	PKMAP_BASE_NR,
	58	# endif
	59	FIXADDR_START_NR,
	60	#endif
	61	};
	62
fe770bf0 PA	63	/* Address space markers hints */
	64	static struct addr_marker address_markers[] = {
	65	{ 0, "User Space" },
	66	#ifdef CONFIG_X86_64
	67	{ 0x8000000000000000UL, "Kernel Space" },
684eb016	68	{ PAGE_OFFSET, "Low Kernel Mapping" },
fe770bf0	69	{ VMALLOC_START, "vmalloc() Area" },
fe770bf0 PA	70	{ VMEMMAP_START, "Vmemmap" },
fe770bf0 PA	71	{ __START_KERNEL_map, "High Kernel Mapping" },
9a79cf9c YL	72	{ MODULES_VADDR, "Modules" },
9a79cf9c YL	73	{ MODULES_END, "End Modules" },
fe770bf0 PA	74	#else
	75	{ PAGE_OFFSET, "Kernel Mapping" },
	76	{ 0/* VMALLOC_START */, "vmalloc() Area" },
	77	{ 0/VMALLOC_END/, "vmalloc() End" },
	78	# ifdef CONFIG_HIGHMEM
	79	{ 0/PKMAP_BASE/, "Persisent kmap() Area" },
	80	# endif
	81	{ 0/FIXADDR_START/, "Fixmap Area" },
	82	#endif
	83	{ -1, NULL } /* End of list */
	84	};
926e5392	85
fe770bf0 PA	86	/* Multipliers for offsets within the PTEs */
	87	#define PTE_LEVEL_MULT (PAGE_SIZE)
	88	#define PMD_LEVEL_MULT (PTRS_PER_PTE * PTE_LEVEL_MULT)
	89	#define PUD_LEVEL_MULT (PTRS_PER_PMD * PMD_LEVEL_MULT)
	90	#define PGD_LEVEL_MULT (PTRS_PER_PUD * PUD_LEVEL_MULT)
926e5392	91
ef6bea6d BP	92	#define pt_dump_seq_printf(m, to_dmesg, fmt, args...) \
	93	({ \
	94	if (to_dmesg) \
	95	printk(KERN_INFO fmt, ##args); \
	96	else \
	97	if (m) \
	98	seq_printf(m, fmt, ##args); \
	99	})
	100
	101	#define pt_dump_cont_printf(m, to_dmesg, fmt, args...) \
	102	({ \
	103	if (to_dmesg) \
	104	printk(KERN_CONT fmt, ##args); \
	105	else \
	106	if (m) \
	107	seq_printf(m, fmt, ##args); \
	108	})
	109
926e5392 AV	110	/*
	111	* Print a readable form of a pgprot_t to the seq_file
	112	*/
ef6bea6d	113	static void printk_prot(struct seq_file *m, pgprot_t prot, int level, bool dmsg)
926e5392	114	{
fe770bf0 PA	115	pgprotval_t pr = pgprot_val(prot);
	116	static const char * const level_name[] =
	117	{ "cr3", "pgd", "pud", "pmd", "pte" };
	118
	119	if (!pgprot_val(prot)) {
	120	/* Not present */
ef6bea6d	121	pt_dump_cont_printf(m, dmsg, " ");
fe770bf0 PA	122	} else {
fe770bf0 PA	123	if (pr & _PAGE_USER)
ef6bea6d	124	pt_dump_cont_printf(m, dmsg, "USR ");
926e5392	125	else
ef6bea6d	126	pt_dump_cont_printf(m, dmsg, " ");
fe770bf0	127	if (pr & _PAGE_RW)
ef6bea6d	128	pt_dump_cont_printf(m, dmsg, "RW ");
fe770bf0	129	else
ef6bea6d	130	pt_dump_cont_printf(m, dmsg, "ro ");
fe770bf0	131	if (pr & _PAGE_PWT)
ef6bea6d	132	pt_dump_cont_printf(m, dmsg, "PWT ");
fe770bf0	133	else
ef6bea6d	134	pt_dump_cont_printf(m, dmsg, " ");
fe770bf0	135	if (pr & _PAGE_PCD)
ef6bea6d	136	pt_dump_cont_printf(m, dmsg, "PCD ");
926e5392	137	else
ef6bea6d	138	pt_dump_cont_printf(m, dmsg, " ");
fe770bf0 PA	139
	140	/* Bit 9 has a different meaning on level 3 vs 4 */
	141	if (level <= 3) {
	142	if (pr & _PAGE_PSE)
ef6bea6d	143	pt_dump_cont_printf(m, dmsg, "PSE ");
fe770bf0	144	else
ef6bea6d	145	pt_dump_cont_printf(m, dmsg, " ");
fe770bf0 PA	146	} else {
fe770bf0 PA	147	if (pr & _PAGE_PAT)
ef6bea6d	148	pt_dump_cont_printf(m, dmsg, "pat ");
fe770bf0	149	else
ef6bea6d	150	pt_dump_cont_printf(m, dmsg, " ");
fe770bf0 PA	151	}
fe770bf0 PA	152	if (pr & _PAGE_GLOBAL)
ef6bea6d	153	pt_dump_cont_printf(m, dmsg, "GLB ");
fe770bf0	154	else
ef6bea6d	155	pt_dump_cont_printf(m, dmsg, " ");
fe770bf0	156	if (pr & _PAGE_NX)
ef6bea6d	157	pt_dump_cont_printf(m, dmsg, "NX ");
fe770bf0	158	else
ef6bea6d	159	pt_dump_cont_printf(m, dmsg, "x ");
926e5392	160	}
ef6bea6d	161	pt_dump_cont_printf(m, dmsg, "%s\n", level_name[level]);
926e5392 AV	162	}
	163
	164	/*
fe770bf0	165	* On 64 bits, sign-extend the 48 bit address to 64 bit
926e5392	166	*/
fe770bf0	167	static unsigned long normalize_addr(unsigned long u)
926e5392	168	{
fe770bf0 PA	169	#ifdef CONFIG_X86_64
	170	return (signed long)(u << 16) >> 16;
	171	#else
926e5392	172	return u;
fe770bf0	173	#endif
926e5392 AV	174	}
	175
	176	/*
	177	* This function gets called on a break in a continuous series
	178	* of PTE entries; the next one is different so we need to
	179	* print what we collected so far.
	180	*/
	181	static void note_page(struct seq_file m, struct pg_state st,
fe770bf0	182	pgprot_t new_prot, int level)
926e5392	183	{
fe770bf0 PA	184	pgprotval_t prot, cur;
fe770bf0 PA	185	static const char units[] = "KMGTPE";
926e5392 AV	186
	187	/*
	188	* If we have a "break" in the series, we need to flush the state that
fe770bf0 PA	189	* we have now. "break" is either changing perms, levels or
fe770bf0 PA	190	* address space marker.
926e5392	191	*/
27990eac JF	192	prot = pgprot_val(new_prot) & PTE_FLAGS_MASK;
27990eac JF	193	cur = pgprot_val(st->current_prot) & PTE_FLAGS_MASK;
926e5392	194
fe770bf0 PA	195	if (!st->level) {
	196	/* First entry */
	197	st->current_prot = new_prot;
	198	st->level = level;
	199	st->marker = address_markers;
ef6bea6d BP	200	pt_dump_seq_printf(m, st->to_dmesg, "---[ %s ]---\n",
ef6bea6d BP	201	st->marker->name);
fe770bf0 PA	202	} else if (prot != cur \|\| level != st->level \|\|
	203	st->current_address >= st->marker[1].start_address) {
	204	const char *unit = units;
926e5392	205	unsigned long delta;
6424fb38	206	int width = sizeof(unsigned long) * 2;
926e5392	207
926e5392 AV	208	/*
	209	* Now print the actual finished series
	210	*/
ef6bea6d BP	211	pt_dump_seq_printf(m, st->to_dmesg, "0x%0lx-0x%0lx ",
	212	width, st->start_address,
	213	width, st->current_address);
926e5392 AV	214
926e5392 AV	215	delta = (st->current_address - st->start_address) >> 10;
fe770bf0 PA	216	while (!(delta & 1023) && unit[1]) {
	217	delta >>= 10;
	218	unit++;
926e5392	219	}
ef6bea6d BP	220	pt_dump_cont_printf(m, st->to_dmesg, "%9lu%c ", delta, *unit);
ef6bea6d BP	221	printk_prot(m, st->current_prot, st->level, st->to_dmesg);
fe770bf0 PA	222
	223	/*
	224	* We print markers for special areas of address space,
	225	* such as the start of vmalloc space etc.
	226	* This helps in the interpretation.
	227	*/
	228	if (st->current_address >= st->marker[1].start_address) {
	229	st->marker++;
ef6bea6d BP	230	pt_dump_seq_printf(m, st->to_dmesg, "---[ %s ]---\n",
ef6bea6d BP	231	st->marker->name);
926e5392	232	}
fe770bf0	233
926e5392 AV	234	st->start_address = st->current_address;
	235	st->current_prot = new_prot;
	236	st->level = level;
fe770bf0	237	}
926e5392 AV	238	}
926e5392 AV	239
fe770bf0	240	static void walk_pte_level(struct seq_file m, struct pg_state st, pmd_t addr,
926e5392 AV	241	unsigned long P)
	242	{
	243	int i;
	244	pte_t *start;
	245
	246	start = (pte_t *) pmd_page_vaddr(addr);
	247	for (i = 0; i < PTRS_PER_PTE; i++) {
	248	pgprot_t prot = pte_pgprot(*start);
	249
fe770bf0	250	st->current_address = normalize_addr(P + i * PTE_LEVEL_MULT);
926e5392 AV	251	note_page(m, st, prot, 4);
	252	start++;
	253	}
	254	}
	255
fe770bf0	256	#if PTRS_PER_PMD > 1
926e5392	257
fe770bf0	258	static void walk_pmd_level(struct seq_file m, struct pg_state st, pud_t addr,
926e5392 AV	259	unsigned long P)
	260	{
	261	int i;
	262	pmd_t *start;
	263
	264	start = (pmd_t *) pud_page_vaddr(addr);
	265	for (i = 0; i < PTRS_PER_PMD; i++) {
fe770bf0	266	st->current_address = normalize_addr(P + i * PMD_LEVEL_MULT);
926e5392	267	if (!pmd_none(*start)) {
77be1fab	268	pgprotval_t prot = pmd_val(*start) & PTE_FLAGS_MASK;
926e5392	269
fe770bf0	270	if (pmd_large(start) \|\| !pmd_present(start))
926e5392 AV	271	note_page(m, st, __pgprot(prot), 3);
926e5392 AV	272	else
fe770bf0 PA	273	walk_pte_level(m, st, *start,
fe770bf0 PA	274	P + i * PMD_LEVEL_MULT);
926e5392 AV	275	} else
	276	note_page(m, st, __pgprot(0), 3);
	277	start++;
	278	}
	279	}
	280
fe770bf0 PA	281	#else
	282	#define walk_pmd_level(m,s,a,p) walk_pte_level(m,s,__pmd(pud_val(a)),p)
	283	#define pud_large(a) pmd_large(__pmd(pud_val(a)))
	284	#define pud_none(a) pmd_none(__pmd(pud_val(a)))
	285	#endif
926e5392	286
fe770bf0 PA	287	#if PTRS_PER_PUD > 1
	288
	289	static void walk_pud_level(struct seq_file m, struct pg_state st, pgd_t addr,
926e5392 AV	290	unsigned long P)
	291	{
	292	int i;
	293	pud_t *start;
	294
	295	start = (pud_t *) pgd_page_vaddr(addr);
	296
	297	for (i = 0; i < PTRS_PER_PUD; i++) {
fe770bf0	298	st->current_address = normalize_addr(P + i * PUD_LEVEL_MULT);
926e5392	299	if (!pud_none(*start)) {
77be1fab	300	pgprotval_t prot = pud_val(*start) & PTE_FLAGS_MASK;
926e5392	301
fe770bf0	302	if (pud_large(start) \|\| !pud_present(start))
926e5392 AV	303	note_page(m, st, __pgprot(prot), 2);
926e5392 AV	304	else
fe770bf0 PA	305	walk_pmd_level(m, st, *start,
fe770bf0 PA	306	P + i * PUD_LEVEL_MULT);
926e5392 AV	307	} else
	308	note_page(m, st, __pgprot(0), 2);
	309
	310	start++;
	311	}
	312	}
	313
fe770bf0 PA	314	#else
	315	#define walk_pud_level(m,s,a,p) walk_pmd_level(m,s,__pud(pgd_val(a)),p)
	316	#define pgd_large(a) pud_large(__pud(pgd_val(a)))
	317	#define pgd_none(a) pud_none(__pud(pgd_val(a)))
	318	#endif
	319
ef6bea6d	320	void ptdump_walk_pgd_level(struct seq_file m, pgd_t pgd)
926e5392	321	{
fe770bf0	322	#ifdef CONFIG_X86_64
926e5392	323	pgd_t start = (pgd_t ) &init_level4_pgt;
fe770bf0 PA	324	#else
	325	pgd_t *start = swapper_pg_dir;
	326	#endif
926e5392	327	int i;
ef6bea6d	328	struct pg_state st = {};
926e5392	329
ef6bea6d BP	330	if (pgd) {
	331	start = pgd;
	332	st.to_dmesg = true;
	333	}
926e5392 AV	334
926e5392 AV	335	for (i = 0; i < PTRS_PER_PGD; i++) {
fe770bf0 PA	336	st.current_address = normalize_addr(i * PGD_LEVEL_MULT);
fe770bf0 PA	337	if (!pgd_none(*start)) {
77be1fab	338	pgprotval_t prot = pgd_val(*start) & PTE_FLAGS_MASK;
fe770bf0 PA	339
	340	if (pgd_large(start) \|\| !pgd_present(start))
	341	note_page(m, &st, __pgprot(prot), 1);
	342	else
	343	walk_pud_level(m, &st, *start,
	344	i * PGD_LEVEL_MULT);
	345	} else
926e5392	346	note_page(m, &st, __pgprot(0), 1);
fe770bf0	347
926e5392 AV	348	start++;
926e5392 AV	349	}
fe770bf0 PA	350
	351	/* Flush out the last page */
	352	st.current_address = normalize_addr(PTRS_PER_PGD*PGD_LEVEL_MULT);
	353	note_page(m, &st, __pgprot(0), 0);
926e5392 AV	354	}
	355
	356	static int ptdump_show(struct seq_file m, void v)
	357	{
ef6bea6d	358	ptdump_walk_pgd_level(m, NULL);
926e5392 AV	359	return 0;
	360	}
	361
	362	static int ptdump_open(struct inode inode, struct file filp)
	363	{
	364	return single_open(filp, ptdump_show, NULL);
	365	}
	366
	367	static const struct file_operations ptdump_fops = {
	368	.open = ptdump_open,
	369	.read = seq_read,
	370	.llseek = seq_lseek,
	371	.release = single_release,
	372	};
	373
a4928cff	374	static int pt_dump_init(void)
926e5392 AV	375	{
	376	struct dentry *pe;
	377
fe770bf0 PA	378	#ifdef CONFIG_X86_32
fe770bf0 PA	379	/* Not a compile-time constant on x86-32 */
92851e2f AS	380	address_markers[VMALLOC_START_NR].start_address = VMALLOC_START;
92851e2f AS	381	address_markers[VMALLOC_END_NR].start_address = VMALLOC_END;
fe770bf0	382	# ifdef CONFIG_HIGHMEM
92851e2f	383	address_markers[PKMAP_BASE_NR].start_address = PKMAP_BASE;
fe770bf0	384	# endif
92851e2f	385	address_markers[FIXADDR_START_NR].start_address = FIXADDR_START;
fe770bf0 PA	386	#endif
fe770bf0 PA	387
926e5392 AV	388	pe = debugfs_create_file("kernel_page_tables", 0600, NULL, NULL,
	389	&ptdump_fops);
	390	if (!pe)
	391	return -ENOMEM;
	392
	393	return 0;
	394	}
	395
	396	__initcall(pt_dump_init);
	397	MODULE_LICENSE("GPL");
	398	MODULE_AUTHOR("Arjan van de Ven <arjan@linux.intel.com>");
	399	MODULE_DESCRIPTION("Kernel debugging helper that dumps pagetables");