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
30 #define LOWEST_ADDR (UL(0xffffffffffffffff) << VA_BITS)
31
32 struct addr_marker {
33 unsigned long start_address;
34 const char *name;
35 };
36
37 enum address_markers_idx {
38 VMALLOC_START_NR = 0,
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 MODULES_START_NR,
49 MODUELS_END_NR,
50 KERNEL_SPACE_NR,
51 };
52
53 static struct addr_marker address_markers[] = {
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 { MODULES_VADDR, "Modules start" },
65 { MODULES_END, "Modules end" },
66 { PAGE_OFFSET, "Kernel Mapping" },
67 { -1, NULL },
68 };
69
70 struct pg_state {
71 struct seq_file *seq;
72 const struct addr_marker *marker;
73 unsigned long start_address;
74 unsigned level;
75 u64 current_prot;
76 };
77
78 struct prot_bits {
79 u64 mask;
80 u64 val;
81 const char *set;
82 const char *clear;
83 };
84
85 static const struct prot_bits pte_bits[] = {
86 {
87 .mask = PTE_USER,
88 .val = PTE_USER,
89 .set = "USR",
90 .clear = " ",
91 }, {
92 .mask = PTE_RDONLY,
93 .val = PTE_RDONLY,
94 .set = "ro",
95 .clear = "RW",
96 }, {
97 .mask = PTE_PXN,
98 .val = PTE_PXN,
99 .set = "NX",
100 .clear = "x ",
101 }, {
102 .mask = PTE_SHARED,
103 .val = PTE_SHARED,
104 .set = "SHD",
105 .clear = " ",
106 }, {
107 .mask = PTE_AF,
108 .val = PTE_AF,
109 .set = "AF",
110 .clear = " ",
111 }, {
112 .mask = PTE_NG,
113 .val = PTE_NG,
114 .set = "NG",
115 .clear = " ",
116 }, {
117 .mask = PTE_UXN,
118 .val = PTE_UXN,
119 .set = "UXN",
120 }, {
121 .mask = PTE_ATTRINDX_MASK,
122 .val = PTE_ATTRINDX(MT_DEVICE_nGnRnE),
123 .set = "DEVICE/nGnRnE",
124 }, {
125 .mask = PTE_ATTRINDX_MASK,
126 .val = PTE_ATTRINDX(MT_DEVICE_nGnRE),
127 .set = "DEVICE/nGnRE",
128 }, {
129 .mask = PTE_ATTRINDX_MASK,
130 .val = PTE_ATTRINDX(MT_DEVICE_GRE),
131 .set = "DEVICE/GRE",
132 }, {
133 .mask = PTE_ATTRINDX_MASK,
134 .val = PTE_ATTRINDX(MT_NORMAL_NC),
135 .set = "MEM/NORMAL-NC",
136 }, {
137 .mask = PTE_ATTRINDX_MASK,
138 .val = PTE_ATTRINDX(MT_NORMAL),
139 .set = "MEM/NORMAL",
140 }
141 };
142
143 struct pg_level {
144 const struct prot_bits *bits;
145 size_t num;
146 u64 mask;
147 };
148
149 static struct pg_level pg_level[] = {
150 {
151 }, { /* pgd */
152 .bits = pte_bits,
153 .num = ARRAY_SIZE(pte_bits),
154 }, { /* pud */
155 .bits = pte_bits,
156 .num = ARRAY_SIZE(pte_bits),
157 }, { /* pmd */
158 .bits = pte_bits,
159 .num = ARRAY_SIZE(pte_bits),
160 }, { /* pte */
161 .bits = pte_bits,
162 .num = ARRAY_SIZE(pte_bits),
163 },
164 };
165
dump_prot(struct pg_state * st,const struct prot_bits * bits,size_t num)166 static void dump_prot(struct pg_state *st, const struct prot_bits *bits,
167 size_t num)
168 {
169 unsigned i;
170
171 for (i = 0; i < num; i++, bits++) {
172 const char *s;
173
174 if ((st->current_prot & bits->mask) == bits->val)
175 s = bits->set;
176 else
177 s = bits->clear;
178
179 if (s)
180 seq_printf(st->seq, " %s", s);
181 }
182 }
183
note_page(struct pg_state * st,unsigned long addr,unsigned level,u64 val)184 static void note_page(struct pg_state *st, unsigned long addr, unsigned level,
185 u64 val)
186 {
187 static const char units[] = "KMGTPE";
188 u64 prot = val & pg_level[level].mask;
189
190 if (!st->level) {
191 st->level = level;
192 st->current_prot = prot;
193 st->start_address = addr;
194 seq_printf(st->seq, "---[ %s ]---\n", st->marker->name);
195 } else if (prot != st->current_prot || level != st->level ||
196 addr >= st->marker[1].start_address) {
197 const char *unit = units;
198 unsigned long delta;
199
200 if (st->current_prot) {
201 seq_printf(st->seq, "0x%16lx-0x%16lx ",
202 st->start_address, addr);
203
204 delta = (addr - st->start_address) >> 10;
205 while (!(delta & 1023) && unit[1]) {
206 delta >>= 10;
207 unit++;
208 }
209 seq_printf(st->seq, "%9lu%c", delta, *unit);
210 if (pg_level[st->level].bits)
211 dump_prot(st, pg_level[st->level].bits,
212 pg_level[st->level].num);
213 seq_puts(st->seq, "\n");
214 }
215
216 if (addr >= st->marker[1].start_address) {
217 st->marker++;
218 seq_printf(st->seq, "---[ %s ]---\n", st->marker->name);
219 }
220
221 st->start_address = addr;
222 st->current_prot = prot;
223 st->level = level;
224 }
225
226 if (addr >= st->marker[1].start_address) {
227 st->marker++;
228 seq_printf(st->seq, "---[ %s ]---\n", st->marker->name);
229 }
230
231 }
232
walk_pte(struct pg_state * st,pmd_t * pmd,unsigned long start)233 static void walk_pte(struct pg_state *st, pmd_t *pmd, unsigned long start)
234 {
235 pte_t *pte = pte_offset_kernel(pmd, 0);
236 unsigned long addr;
237 unsigned i;
238
239 for (i = 0; i < PTRS_PER_PTE; i++, pte++) {
240 addr = start + i * PAGE_SIZE;
241 note_page(st, addr, 4, pte_val(*pte));
242 }
243 }
244
walk_pmd(struct pg_state * st,pud_t * pud,unsigned long start)245 static void walk_pmd(struct pg_state *st, pud_t *pud, unsigned long start)
246 {
247 pmd_t *pmd = pmd_offset(pud, 0);
248 unsigned long addr;
249 unsigned i;
250
251 for (i = 0; i < PTRS_PER_PMD; i++, pmd++) {
252 addr = start + i * PMD_SIZE;
253 if (pmd_none(*pmd) || pmd_sect(*pmd)) {
254 note_page(st, addr, 3, pmd_val(*pmd));
255 } else {
256 BUG_ON(pmd_bad(*pmd));
257 walk_pte(st, pmd, addr);
258 }
259 }
260 }
261
walk_pud(struct pg_state * st,pgd_t * pgd,unsigned long start)262 static void walk_pud(struct pg_state *st, pgd_t *pgd, unsigned long start)
263 {
264 pud_t *pud = pud_offset(pgd, 0);
265 unsigned long addr;
266 unsigned i;
267
268 for (i = 0; i < PTRS_PER_PUD; i++, pud++) {
269 addr = start + i * PUD_SIZE;
270 if (pud_none(*pud) || pud_sect(*pud)) {
271 note_page(st, addr, 2, pud_val(*pud));
272 } else {
273 BUG_ON(pud_bad(*pud));
274 walk_pmd(st, pud, addr);
275 }
276 }
277 }
278
walk_pgd(struct pg_state * st,struct mm_struct * mm,unsigned long start)279 static void walk_pgd(struct pg_state *st, struct mm_struct *mm, unsigned long start)
280 {
281 pgd_t *pgd = pgd_offset(mm, 0UL);
282 unsigned i;
283 unsigned long addr;
284
285 for (i = 0; i < PTRS_PER_PGD; i++, pgd++) {
286 addr = start + i * PGDIR_SIZE;
287 if (pgd_none(*pgd)) {
288 note_page(st, addr, 1, pgd_val(*pgd));
289 } else {
290 BUG_ON(pgd_bad(*pgd));
291 walk_pud(st, pgd, addr);
292 }
293 }
294 }
295
ptdump_show(struct seq_file * m,void * v)296 static int ptdump_show(struct seq_file *m, void *v)
297 {
298 struct pg_state st = {
299 .seq = m,
300 .marker = address_markers,
301 };
302
303 walk_pgd(&st, &init_mm, LOWEST_ADDR);
304
305 note_page(&st, 0, 0, 0);
306 return 0;
307 }
308
ptdump_open(struct inode * inode,struct file * file)309 static int ptdump_open(struct inode *inode, struct file *file)
310 {
311 return single_open(file, ptdump_show, NULL);
312 }
313
314 static const struct file_operations ptdump_fops = {
315 .open = ptdump_open,
316 .read = seq_read,
317 .llseek = seq_lseek,
318 .release = single_release,
319 };
320
ptdump_init(void)321 static int ptdump_init(void)
322 {
323 struct dentry *pe;
324 unsigned i, j;
325
326 for (i = 0; i < ARRAY_SIZE(pg_level); i++)
327 if (pg_level[i].bits)
328 for (j = 0; j < pg_level[i].num; j++)
329 pg_level[i].mask |= pg_level[i].bits[j].mask;
330
331 #ifdef CONFIG_SPARSEMEM_VMEMMAP
332 address_markers[VMEMMAP_START_NR].start_address =
333 (unsigned long)virt_to_page(PAGE_OFFSET);
334 address_markers[VMEMMAP_END_NR].start_address =
335 (unsigned long)virt_to_page(high_memory);
336 #endif
337
338 pe = debugfs_create_file("kernel_page_tables", 0400, NULL, NULL,
339 &ptdump_fops);
340 return pe ? 0 : -ENOMEM;
341 }
342 device_initcall(ptdump_init);
343