1/*
  2 * Copyright (c) 2014, The Linux Foundation. All rights reserved.
  3 * Debug helper to dump the current kernel pagetables of the system
  4 * so that we can see what the various memory ranges are set to.
  5 *
  6 * Derived from x86 and arm implementation:
  7 * (C) Copyright 2008 Intel Corporation
  8 *
  9 * Author: Arjan van de Ven <arjan@linux.intel.com>
 10 *
 11 * This program is free software; you can redistribute it and/or
 12 * modify it under the terms of the GNU General Public License
 13 * as published by the Free Software Foundation; version 2
 14 * of the License.
 15 */
 16#include <linux/debugfs.h>
 17#include <linux/errno.h>
 18#include <linux/fs.h>
 19#include <linux/io.h>
 20#include <linux/init.h>
 21#include <linux/mm.h>
 22#include <linux/sched.h>
 23#include <linux/seq_file.h>
 24
 25#include <asm/fixmap.h>
 26#include <asm/memory.h>
 27#include <asm/pgtable.h>
 28#include <asm/pgtable-hwdef.h>
 29
 30struct addr_marker {
 31	unsigned long start_address;
 32	const char *name;
 33};
 34
 35enum address_markers_idx {
 36	MODULES_START_NR = 0,
 37	MODULES_END_NR,
 38	VMALLOC_START_NR,
 39	VMALLOC_END_NR,
 40#ifdef CONFIG_SPARSEMEM_VMEMMAP
 41	VMEMMAP_START_NR,
 42	VMEMMAP_END_NR,
 43#endif
 44	FIXADDR_START_NR,
 45	FIXADDR_END_NR,
 46	PCI_START_NR,
 47	PCI_END_NR,
 48	KERNEL_SPACE_NR,
 49};
 50
 51static struct addr_marker address_markers[] = {
 52	{ MODULES_VADDR,	"Modules start" },
 53	{ MODULES_END,		"Modules end" },
 54	{ VMALLOC_START,	"vmalloc() Area" },
 55	{ VMALLOC_END,		"vmalloc() End" },
 56#ifdef CONFIG_SPARSEMEM_VMEMMAP
 57	{ 0,			"vmemmap start" },
 58	{ 0,			"vmemmap end" },
 59#endif
 60	{ FIXADDR_START,	"Fixmap start" },
 61	{ FIXADDR_TOP,		"Fixmap end" },
 62	{ PCI_IO_START,		"PCI I/O start" },
 63	{ PCI_IO_END,		"PCI I/O end" },
 64	{ PAGE_OFFSET,		"Linear Mapping" },
 65	{ -1,			NULL },
 66};
 67
 68/*
 69 * The page dumper groups page table entries of the same type into a single
 70 * description. It uses pg_state to track the range information while
 71 * iterating over the pte entries. When the continuity is broken it then
 72 * dumps out a description of the range.
 73 */
 74struct pg_state {
 75	struct seq_file *seq;
 76	const struct addr_marker *marker;
 77	unsigned long start_address;
 78	unsigned level;
 79	u64 current_prot;
 80};
 81
 82struct prot_bits {
 83	u64		mask;
 84	u64		val;
 85	const char	*set;
 86	const char	*clear;
 87};
 88
 89static const struct prot_bits pte_bits[] = {
 90	{
 91		.mask	= PTE_VALID,
 92		.val	= PTE_VALID,
 93		.set	= " ",
 94		.clear	= "F",
 95	}, {
 96		.mask	= PTE_USER,
 97		.val	= PTE_USER,
 98		.set	= "USR",
 99		.clear	= "   ",
100	}, {
101		.mask	= PTE_RDONLY,
102		.val	= PTE_RDONLY,
103		.set	= "ro",
104		.clear	= "RW",
105	}, {
106		.mask	= PTE_PXN,
107		.val	= PTE_PXN,
108		.set	= "NX",
109		.clear	= "x ",
110	}, {
111		.mask	= PTE_SHARED,
112		.val	= PTE_SHARED,
113		.set	= "SHD",
114		.clear	= "   ",
115	}, {
116		.mask	= PTE_AF,
117		.val	= PTE_AF,
118		.set	= "AF",
119		.clear	= "  ",
120	}, {
121		.mask	= PTE_NG,
122		.val	= PTE_NG,
123		.set	= "NG",
124		.clear	= "  ",
125	}, {
126		.mask	= PTE_CONT,
127		.val	= PTE_CONT,
128		.set	= "CON",
129		.clear	= "   ",
130	}, {
131		.mask	= PTE_TABLE_BIT,
132		.val	= PTE_TABLE_BIT,
133		.set	= "   ",
134		.clear	= "BLK",
135	}, {
136		.mask	= PTE_UXN,
137		.val	= PTE_UXN,
138		.set	= "UXN",
139	}, {
140		.mask	= PTE_ATTRINDX_MASK,
141		.val	= PTE_ATTRINDX(MT_DEVICE_nGnRnE),
142		.set	= "DEVICE/nGnRnE",
143	}, {
144		.mask	= PTE_ATTRINDX_MASK,
145		.val	= PTE_ATTRINDX(MT_DEVICE_nGnRE),
146		.set	= "DEVICE/nGnRE",
147	}, {
148		.mask	= PTE_ATTRINDX_MASK,
149		.val	= PTE_ATTRINDX(MT_DEVICE_GRE),
150		.set	= "DEVICE/GRE",
151	}, {
152		.mask	= PTE_ATTRINDX_MASK,
153		.val	= PTE_ATTRINDX(MT_NORMAL_NC),
154		.set	= "MEM/NORMAL-NC",
155	}, {
156		.mask	= PTE_ATTRINDX_MASK,
157		.val	= PTE_ATTRINDX(MT_NORMAL),
158		.set	= "MEM/NORMAL",
159	}
160};
161
162struct pg_level {
163	const struct prot_bits *bits;
164	size_t num;
165	u64 mask;
166};
167
168static struct pg_level pg_level[] = {
169	{
170	}, { /* pgd */
171		.bits	= pte_bits,
172		.num	= ARRAY_SIZE(pte_bits),
173	}, { /* pud */
174		.bits	= pte_bits,
175		.num	= ARRAY_SIZE(pte_bits),
176	}, { /* pmd */
177		.bits	= pte_bits,
178		.num	= ARRAY_SIZE(pte_bits),
179	}, { /* pte */
180		.bits	= pte_bits,
181		.num	= ARRAY_SIZE(pte_bits),
182	},
183};
184
185static void dump_prot(struct pg_state *st, const struct prot_bits *bits,
186			size_t num)
187{
188	unsigned i;
189
190	for (i = 0; i < num; i++, bits++) {
191		const char *s;
192
193		if ((st->current_prot & bits->mask) == bits->val)
194			s = bits->set;
195		else
196			s = bits->clear;
197
198		if (s)
199			seq_printf(st->seq, " %s", s);
200	}
201}
202
203static void note_page(struct pg_state *st, unsigned long addr, unsigned level,
204				u64 val)
205{
206	static const char units[] = "KMGTPE";
207	u64 prot = val & pg_level[level].mask;
208
209	if (!st->level) {
210		st->level = level;
211		st->current_prot = prot;
212		st->start_address = addr;
213		seq_printf(st->seq, "---[ %s ]---\n", st->marker->name);
214	} else if (prot != st->current_prot || level != st->level ||
215		   addr >= st->marker[1].start_address) {
216		const char *unit = units;
217		unsigned long delta;
218
219		if (st->current_prot) {
220			seq_printf(st->seq, "0x%016lx-0x%016lx   ",
221				   st->start_address, addr);
222
223			delta = (addr - st->start_address) >> 10;
224			while (!(delta & 1023) && unit[1]) {
225				delta >>= 10;
226				unit++;
227			}
228			seq_printf(st->seq, "%9lu%c", delta, *unit);
229			if (pg_level[st->level].bits)
230				dump_prot(st, pg_level[st->level].bits,
231					  pg_level[st->level].num);
232			seq_puts(st->seq, "\n");
233		}
234
235		if (addr >= st->marker[1].start_address) {
236			st->marker++;
237			seq_printf(st->seq, "---[ %s ]---\n", st->marker->name);
238		}
239
240		st->start_address = addr;
241		st->current_prot = prot;
242		st->level = level;
243	}
244
245	if (addr >= st->marker[1].start_address) {
246		st->marker++;
247		seq_printf(st->seq, "---[ %s ]---\n", st->marker->name);
248	}
249
250}
251
252static void walk_pte(struct pg_state *st, pmd_t *pmd, unsigned long start)
253{
254	pte_t *pte = pte_offset_kernel(pmd, 0);
255	unsigned long addr;
256	unsigned i;
257
258	for (i = 0; i < PTRS_PER_PTE; i++, pte++) {
259		addr = start + i * PAGE_SIZE;
260		note_page(st, addr, 4, pte_val(*pte));
261	}
262}
263
264static void walk_pmd(struct pg_state *st, pud_t *pud, unsigned long start)
265{
266	pmd_t *pmd = pmd_offset(pud, 0);
267	unsigned long addr;
268	unsigned i;
269
270	for (i = 0; i < PTRS_PER_PMD; i++, pmd++) {
271		addr = start + i * PMD_SIZE;
272		if (pmd_none(*pmd) || pmd_sect(*pmd)) {
273			note_page(st, addr, 3, pmd_val(*pmd));
274		} else {
275			BUG_ON(pmd_bad(*pmd));
276			walk_pte(st, pmd, addr);
277		}
278	}
279}
280
281static void walk_pud(struct pg_state *st, pgd_t *pgd, unsigned long start)
282{
283	pud_t *pud = pud_offset(pgd, 0);
284	unsigned long addr;
285	unsigned i;
286
287	for (i = 0; i < PTRS_PER_PUD; i++, pud++) {
288		addr = start + i * PUD_SIZE;
289		if (pud_none(*pud) || pud_sect(*pud)) {
290			note_page(st, addr, 2, pud_val(*pud));
291		} else {
292			BUG_ON(pud_bad(*pud));
293			walk_pmd(st, pud, addr);
294		}
295	}
296}
297
298static void walk_pgd(struct pg_state *st, struct mm_struct *mm, unsigned long start)
299{
300	pgd_t *pgd = pgd_offset(mm, 0UL);
301	unsigned i;
302	unsigned long addr;
303
304	for (i = 0; i < PTRS_PER_PGD; i++, pgd++) {
305		addr = start + i * PGDIR_SIZE;
306		if (pgd_none(*pgd)) {
307			note_page(st, addr, 1, pgd_val(*pgd));
308		} else {
309			BUG_ON(pgd_bad(*pgd));
310			walk_pud(st, pgd, addr);
311		}
312	}
313}
314
315static int ptdump_show(struct seq_file *m, void *v)
316{
317	struct pg_state st = {
318		.seq = m,
319		.marker = address_markers,
320	};
321
322	walk_pgd(&st, &init_mm, VA_START);
323
324	note_page(&st, 0, 0, 0);
325	return 0;
326}
327
328static int ptdump_open(struct inode *inode, struct file *file)
329{
330	return single_open(file, ptdump_show, NULL);
331}
332
333static const struct file_operations ptdump_fops = {
334	.open		= ptdump_open,
335	.read		= seq_read,
336	.llseek		= seq_lseek,
337	.release	= single_release,
338};
339
340static int ptdump_init(void)
341{
342	struct dentry *pe;
343	unsigned i, j;
344
345	for (i = 0; i < ARRAY_SIZE(pg_level); i++)
346		if (pg_level[i].bits)
347			for (j = 0; j < pg_level[i].num; j++)
348				pg_level[i].mask |= pg_level[i].bits[j].mask;
349
350#ifdef CONFIG_SPARSEMEM_VMEMMAP
351	address_markers[VMEMMAP_START_NR].start_address =
352				(unsigned long)virt_to_page(PAGE_OFFSET);
353	address_markers[VMEMMAP_END_NR].start_address =
354				(unsigned long)virt_to_page(high_memory);
355#endif
356
357	pe = debugfs_create_file("kernel_page_tables", 0400, NULL, NULL,
358				 &ptdump_fops);
359	return pe ? 0 : -ENOMEM;
360}
361device_initcall(ptdump_init);