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
15 #include <linux/debugfs.h>
16 #include <linux/mm.h>
17 #include <linux/module.h>
18 #include <linux/seq_file.h>
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;
32 const struct addr_marker *marker;
33 unsigned long lines;
34 };
35
36 struct addr_marker {
37 unsigned long start_address;
38 const char *name;
39 unsigned long max_lines;
40 };
41
42 /* Address space markers hints */
43 static struct addr_marker address_markers[] = {
44 { 0, "User Space" },
45 #ifdef CONFIG_X86_64
46 { 0x8000000000000000UL, "Kernel Space" },
47 { PAGE_OFFSET, "Low Kernel Mapping" },
48 { VMALLOC_START, "vmalloc() Area" },
49 { VMEMMAP_START, "Vmemmap" },
50 { ESPFIX_BASE_ADDR, "ESPfix Area", 16 },
51 { __START_KERNEL_map, "High Kernel Mapping" },
52 { MODULES_VADDR, "Modules" },
53 { MODULES_END, "End Modules" },
54 #else
55 { PAGE_OFFSET, "Kernel Mapping" },
56 { 0/* VMALLOC_START */, "vmalloc() Area" },
57 { 0/*VMALLOC_END*/, "vmalloc() End" },
58 # ifdef CONFIG_HIGHMEM
59 { 0/*PKMAP_BASE*/, "Persisent kmap() Area" },
60 # endif
61 { 0/*FIXADDR_START*/, "Fixmap Area" },
62 #endif
63 { -1, NULL } /* End of list */
64 };
65
66 /* Multipliers for offsets within the PTEs */
67 #define PTE_LEVEL_MULT (PAGE_SIZE)
68 #define PMD_LEVEL_MULT (PTRS_PER_PTE * PTE_LEVEL_MULT)
69 #define PUD_LEVEL_MULT (PTRS_PER_PMD * PMD_LEVEL_MULT)
70 #define PGD_LEVEL_MULT (PTRS_PER_PUD * PUD_LEVEL_MULT)
71
72 /*
73 * Print a readable form of a pgprot_t to the seq_file
74 */
75 static void printk_prot(struct seq_file *m, pgprot_t prot, int level)
76 {
77 pgprotval_t pr = pgprot_val(prot);
78 static const char * const level_name[] =
79 { "cr3", "pgd", "pud", "pmd", "pte" };
80
81 if (!pgprot_val(prot)) {
82 /* Not present */
83 seq_printf(m, " ");
84 } else {
85 if (pr & _PAGE_USER)
86 seq_printf(m, "USR ");
87 else
88 seq_printf(m, " ");
89 if (pr & _PAGE_RW)
90 seq_printf(m, "RW ");
91 else
92 seq_printf(m, "ro ");
93 if (pr & _PAGE_PWT)
94 seq_printf(m, "PWT ");
95 else
96 seq_printf(m, " ");
97 if (pr & _PAGE_PCD)
98 seq_printf(m, "PCD ");
99 else
100 seq_printf(m, " ");
101
102 /* Bit 9 has a different meaning on level 3 vs 4 */
103 if (level <= 3) {
104 if (pr & _PAGE_PSE)
105 seq_printf(m, "PSE ");
106 else
107 seq_printf(m, " ");
108 } else {
109 if (pr & _PAGE_PAT)
110 seq_printf(m, "pat ");
111 else
112 seq_printf(m, " ");
113 }
114 if (pr & _PAGE_GLOBAL)
115 seq_printf(m, "GLB ");
116 else
117 seq_printf(m, " ");
118 if (pr & _PAGE_NX)
119 seq_printf(m, "NX ");
120 else
121 seq_printf(m, "x ");
122 }
123 seq_printf(m, "%s\n", level_name[level]);
124 }
125
126 /*
127 * On 64 bits, sign-extend the 48 bit address to 64 bit
128 */
129 static unsigned long normalize_addr(unsigned long u)
130 {
131 #ifdef CONFIG_X86_64
132 return (signed long)(u << 16) >> 16;
133 #else
134 return u;
135 #endif
136 }
137
138 /*
139 * This function gets called on a break in a continuous series
140 * of PTE entries; the next one is different so we need to
141 * print what we collected so far.
142 */
143 static void note_page(struct seq_file *m, struct pg_state *st,
144 pgprot_t new_prot, int level)
145 {
146 pgprotval_t prot, cur;
147 static const char units[] = "BKMGTPE";
148
149 /*
150 * If we have a "break" in the series, we need to flush the state that
151 * we have now. "break" is either changing perms, levels or
152 * address space marker.
153 */
154 prot = pgprot_val(new_prot) & PTE_FLAGS_MASK;
155 cur = pgprot_val(st->current_prot) & PTE_FLAGS_MASK;
156
157 if (!st->level) {
158 /* First entry */
159 st->current_prot = new_prot;
160 st->level = level;
161 st->marker = address_markers;
162 st->lines = 0;
163 seq_printf(m, "---[ %s ]---\n", st->marker->name);
164 } else if (prot != cur || level != st->level ||
165 st->current_address >= st->marker[1].start_address) {
166 const char *unit = units;
167 unsigned long delta;
168 int width = sizeof(unsigned long) * 2;
169
170 /*
171 * Now print the actual finished series
172 */
173 if (!st->marker->max_lines ||
174 st->lines < st->marker->max_lines) {
175 seq_printf(m, "0x%0*lx-0x%0*lx ",
176 width, st->start_address,
177 width, st->current_address);
178
179 delta = (st->current_address - st->start_address);
180 while (!(delta & 1023) && unit[1]) {
181 delta >>= 10;
182 unit++;
183 }
184 seq_printf(m, "%9lu%c ", delta, *unit);
185 printk_prot(m, st->current_prot, st->level);
186 }
187 st->lines++;
188
189 /*
190 * We print markers for special areas of address space,
191 * such as the start of vmalloc space etc.
192 * This helps in the interpretation.
193 */
194 if (st->current_address >= st->marker[1].start_address) {
195 if (st->marker->max_lines &&
196 st->lines > st->marker->max_lines) {
197 unsigned long nskip =
198 st->lines - st->marker->max_lines;
199 seq_printf(m, "... %lu entr%s skipped ... \n",
200 nskip, nskip == 1 ? "y" : "ies");
201 }
202 st->marker++;
203 st->lines = 0;
204 seq_printf(m, "---[ %s ]---\n", st->marker->name);
205 }
206
207 st->start_address = st->current_address;
208 st->current_prot = new_prot;
209 st->level = level;
210 }
211 }
212
213 static void walk_pte_level(struct seq_file *m, struct pg_state *st, pmd_t addr,
214 unsigned long P)
215 {
216 int i;
217 pte_t *start;
218
219 start = (pte_t *) pmd_page_vaddr(addr);
220 for (i = 0; i < PTRS_PER_PTE; i++) {
221 pgprot_t prot = pte_pgprot(*start);
222
223 st->current_address = normalize_addr(P + i * PTE_LEVEL_MULT);
224 note_page(m, st, prot, 4);
225 start++;
226 }
227 }
228
229 #if PTRS_PER_PMD > 1
230
231 static void walk_pmd_level(struct seq_file *m, struct pg_state *st, pud_t addr,
232 unsigned long P)
233 {
234 int i;
235 pmd_t *start;
236
237 start = (pmd_t *) pud_page_vaddr(addr);
238 for (i = 0; i < PTRS_PER_PMD; i++) {
239 st->current_address = normalize_addr(P + i * PMD_LEVEL_MULT);
240 if (!pmd_none(*start)) {
241 pgprotval_t prot = pmd_val(*start) & PTE_FLAGS_MASK;
242
243 if (pmd_large(*start) || !pmd_present(*start))
244 note_page(m, st, __pgprot(prot), 3);
245 else
246 walk_pte_level(m, st, *start,
247 P + i * PMD_LEVEL_MULT);
248 } else
249 note_page(m, st, __pgprot(0), 3);
250 start++;
251 }
252 }
253
254 #else
255 #define walk_pmd_level(m,s,a,p) walk_pte_level(m,s,__pmd(pud_val(a)),p)
256 #define pud_large(a) pmd_large(__pmd(pud_val(a)))
257 #define pud_none(a) pmd_none(__pmd(pud_val(a)))
258 #endif
259
260 #if PTRS_PER_PUD > 1
261
262 static void walk_pud_level(struct seq_file *m, struct pg_state *st, pgd_t addr,
263 unsigned long P)
264 {
265 int i;
266 pud_t *start;
267
268 start = (pud_t *) pgd_page_vaddr(addr);
269
270 for (i = 0; i < PTRS_PER_PUD; i++) {
271 st->current_address = normalize_addr(P + i * PUD_LEVEL_MULT);
272 if (!pud_none(*start)) {
273 pgprotval_t prot = pud_val(*start) & PTE_FLAGS_MASK;
274
275 if (pud_large(*start) || !pud_present(*start))
276 note_page(m, st, __pgprot(prot), 2);
277 else
278 walk_pmd_level(m, st, *start,
279 P + i * PUD_LEVEL_MULT);
280 } else
281 note_page(m, st, __pgprot(0), 2);
282
283 start++;
284 }
285 }
286
287 #else
288 #define walk_pud_level(m,s,a,p) walk_pmd_level(m,s,__pud(pgd_val(a)),p)
289 #define pgd_large(a) pud_large(__pud(pgd_val(a)))
290 #define pgd_none(a) pud_none(__pud(pgd_val(a)))
291 #endif
292
293 static void walk_pgd_level(struct seq_file *m)
294 {
295 #ifdef CONFIG_X86_64
296 pgd_t *start = (pgd_t *) &init_level4_pgt;
297 #else
298 pgd_t *start = swapper_pg_dir;
299 #endif
300 int i;
301 struct pg_state st;
302
303 memset(&st, 0, sizeof(st));
304
305 for (i = 0; i < PTRS_PER_PGD; i++) {
306 st.current_address = normalize_addr(i * PGD_LEVEL_MULT);
307 if (!pgd_none(*start)) {
308 pgprotval_t prot = pgd_val(*start) & PTE_FLAGS_MASK;
309
310 if (pgd_large(*start) || !pgd_present(*start))
311 note_page(m, &st, __pgprot(prot), 1);
312 else
313 walk_pud_level(m, &st, *start,
314 i * PGD_LEVEL_MULT);
315 } else
316 note_page(m, &st, __pgprot(0), 1);
317
318 start++;
319 }
320
321 /* Flush out the last page */
322 st.current_address = normalize_addr(PTRS_PER_PGD*PGD_LEVEL_MULT);
323 note_page(m, &st, __pgprot(0), 0);
324 }
325
326 static int ptdump_show(struct seq_file *m, void *v)
327 {
328 walk_pgd_level(m);
329 return 0;
330 }
331
332 static int ptdump_open(struct inode *inode, struct file *filp)
333 {
334 return single_open(filp, ptdump_show, NULL);
335 }
336
337 static const struct file_operations ptdump_fops = {
338 .open = ptdump_open,
339 .read = seq_read,
340 .llseek = seq_lseek,
341 .release = single_release,
342 };
343
344 static int pt_dump_init(void)
345 {
346 struct dentry *pe;
347
348 #ifdef CONFIG_X86_32
349 /* Not a compile-time constant on x86-32 */
350 address_markers[2].start_address = VMALLOC_START;
351 address_markers[3].start_address = VMALLOC_END;
352 # ifdef CONFIG_HIGHMEM
353 address_markers[4].start_address = PKMAP_BASE;
354 address_markers[5].start_address = FIXADDR_START;
355 # else
356 address_markers[4].start_address = FIXADDR_START;
357 # endif
358 #endif
359
360 pe = debugfs_create_file("kernel_page_tables", 0600, NULL, NULL,
361 &ptdump_fops);
362 if (!pe)
363 return -ENOMEM;
364
365 return 0;
366 }
367
368 __initcall(pt_dump_init);
369 MODULE_LICENSE("GPL");
370 MODULE_AUTHOR("Arjan van de Ven <arjan@linux.intel.com>");
371 MODULE_DESCRIPTION("Kernel debugging helper that dumps pagetables");
372
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.