blob: e04e67753238e65c75cab2c8b6484af15ba8883b [file] [log] [blame]
Arjan van de Ven926e5392008-04-17 17:40:45 +02001/*
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
H. Peter Anvinfe770bf02008-04-17 17:40:45 +020015#include <linux/debugfs.h>
16#include <linux/mm.h>
Arjan van de Ven926e5392008-04-17 17:40:45 +020017#include <linux/module.h>
18#include <linux/seq_file.h>
Arjan van de Ven926e5392008-04-17 17:40:45 +020019
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 */
27struct pg_state {
28 int level;
29 pgprot_t current_prot;
30 unsigned long start_address;
31 unsigned long current_address;
H. Peter Anvinfe770bf02008-04-17 17:40:45 +020032 const struct addr_marker *marker;
H. Peter Anvina7b854c2014-04-29 16:46:09 -070033 unsigned long lines;
H. Peter Anvinfe770bf02008-04-17 17:40:45 +020034};
35
36struct addr_marker {
37 unsigned long start_address;
38 const char *name;
H. Peter Anvina7b854c2014-04-29 16:46:09 -070039 unsigned long max_lines;
H. Peter Anvinfe770bf02008-04-17 17:40:45 +020040};
41
Andres Salomon92851e22010-07-20 15:19:46 -070042/* indices for address_markers; keep sync'd w/ address_markers below */
43enum address_markers_idx {
44 USER_SPACE_NR = 0,
45#ifdef CONFIG_X86_64
46 KERNEL_SPACE_NR,
47 LOW_KERNEL_NR,
48 VMALLOC_START_NR,
49 VMEMMAP_START_NR,
H. Peter Anvina7b854c2014-04-29 16:46:09 -070050 ESPFIX_START_NR,
Andres Salomon92851e22010-07-20 15:19:46 -070051 HIGH_KERNEL_NR,
52 MODULES_VADDR_NR,
53 MODULES_END_NR,
54#else
55 KERNEL_SPACE_NR,
56 VMALLOC_START_NR,
57 VMALLOC_END_NR,
58# ifdef CONFIG_HIGHMEM
59 PKMAP_BASE_NR,
60# endif
61 FIXADDR_START_NR,
62#endif
63};
64
H. Peter Anvinfe770bf02008-04-17 17:40:45 +020065/* Address space markers hints */
66static struct addr_marker address_markers[] = {
67 { 0, "User Space" },
68#ifdef CONFIG_X86_64
69 { 0x8000000000000000UL, "Kernel Space" },
Jiri Slaby684eb012008-05-12 15:43:37 +020070 { PAGE_OFFSET, "Low Kernel Mapping" },
H. Peter Anvinfe770bf02008-04-17 17:40:45 +020071 { VMALLOC_START, "vmalloc() Area" },
H. Peter Anvinfe770bf02008-04-17 17:40:45 +020072 { VMEMMAP_START, "Vmemmap" },
H. Peter Anvina7b854c2014-04-29 16:46:09 -070073 { ESPFIX_BASE_ADDR, "ESPfix Area", 16 },
H. Peter Anvinfe770bf02008-04-17 17:40:45 +020074 { __START_KERNEL_map, "High Kernel Mapping" },
Yinghai Lu9a79cf92008-03-07 19:17:55 -080075 { MODULES_VADDR, "Modules" },
76 { MODULES_END, "End Modules" },
H. Peter Anvinfe770bf02008-04-17 17:40:45 +020077#else
78 { PAGE_OFFSET, "Kernel Mapping" },
79 { 0/* VMALLOC_START */, "vmalloc() Area" },
80 { 0/*VMALLOC_END*/, "vmalloc() End" },
81# ifdef CONFIG_HIGHMEM
82 { 0/*PKMAP_BASE*/, "Persisent kmap() Area" },
83# endif
84 { 0/*FIXADDR_START*/, "Fixmap Area" },
85#endif
86 { -1, NULL } /* End of list */
Arjan van de Ven926e5392008-04-17 17:40:45 +020087};
88
89/* Multipliers for offsets within the PTEs */
H. Peter Anvinfe770bf02008-04-17 17:40:45 +020090#define PTE_LEVEL_MULT (PAGE_SIZE)
91#define PMD_LEVEL_MULT (PTRS_PER_PTE * PTE_LEVEL_MULT)
92#define PUD_LEVEL_MULT (PTRS_PER_PMD * PMD_LEVEL_MULT)
93#define PGD_LEVEL_MULT (PTRS_PER_PUD * PUD_LEVEL_MULT)
Arjan van de Ven926e5392008-04-17 17:40:45 +020094
95/*
96 * Print a readable form of a pgprot_t to the seq_file
97 */
98static void printk_prot(struct seq_file *m, pgprot_t prot, int level)
99{
H. Peter Anvinfe770bf02008-04-17 17:40:45 +0200100 pgprotval_t pr = pgprot_val(prot);
101 static const char * const level_name[] =
102 { "cr3", "pgd", "pud", "pmd", "pte" };
Arjan van de Ven926e5392008-04-17 17:40:45 +0200103
H. Peter Anvinfe770bf02008-04-17 17:40:45 +0200104 if (!pgprot_val(prot)) {
105 /* Not present */
106 seq_printf(m, " ");
Arjan van de Ven926e5392008-04-17 17:40:45 +0200107 } else {
H. Peter Anvinfe770bf02008-04-17 17:40:45 +0200108 if (pr & _PAGE_USER)
109 seq_printf(m, "USR ");
Arjan van de Ven926e5392008-04-17 17:40:45 +0200110 else
111 seq_printf(m, " ");
H. Peter Anvinfe770bf02008-04-17 17:40:45 +0200112 if (pr & _PAGE_RW)
113 seq_printf(m, "RW ");
114 else
115 seq_printf(m, "ro ");
116 if (pr & _PAGE_PWT)
117 seq_printf(m, "PWT ");
118 else
119 seq_printf(m, " ");
120 if (pr & _PAGE_PCD)
121 seq_printf(m, "PCD ");
122 else
123 seq_printf(m, " ");
124
125 /* Bit 9 has a different meaning on level 3 vs 4 */
126 if (level <= 3) {
127 if (pr & _PAGE_PSE)
128 seq_printf(m, "PSE ");
129 else
130 seq_printf(m, " ");
131 } else {
132 if (pr & _PAGE_PAT)
133 seq_printf(m, "pat ");
134 else
135 seq_printf(m, " ");
136 }
137 if (pr & _PAGE_GLOBAL)
138 seq_printf(m, "GLB ");
139 else
140 seq_printf(m, " ");
141 if (pr & _PAGE_NX)
142 seq_printf(m, "NX ");
143 else
144 seq_printf(m, "x ");
Arjan van de Ven926e5392008-04-17 17:40:45 +0200145 }
H. Peter Anvinfe770bf02008-04-17 17:40:45 +0200146 seq_printf(m, "%s\n", level_name[level]);
Arjan van de Ven926e5392008-04-17 17:40:45 +0200147}
148
149/*
H. Peter Anvinfe770bf02008-04-17 17:40:45 +0200150 * On 64 bits, sign-extend the 48 bit address to 64 bit
Arjan van de Ven926e5392008-04-17 17:40:45 +0200151 */
H. Peter Anvinfe770bf02008-04-17 17:40:45 +0200152static unsigned long normalize_addr(unsigned long u)
Arjan van de Ven926e5392008-04-17 17:40:45 +0200153{
H. Peter Anvinfe770bf02008-04-17 17:40:45 +0200154#ifdef CONFIG_X86_64
155 return (signed long)(u << 16) >> 16;
156#else
Arjan van de Ven926e5392008-04-17 17:40:45 +0200157 return u;
H. Peter Anvinfe770bf02008-04-17 17:40:45 +0200158#endif
Arjan van de Ven926e5392008-04-17 17:40:45 +0200159}
160
161/*
162 * This function gets called on a break in a continuous series
163 * of PTE entries; the next one is different so we need to
164 * print what we collected so far.
165 */
166static void note_page(struct seq_file *m, struct pg_state *st,
H. Peter Anvinfe770bf02008-04-17 17:40:45 +0200167 pgprot_t new_prot, int level)
Arjan van de Ven926e5392008-04-17 17:40:45 +0200168{
H. Peter Anvinfe770bf02008-04-17 17:40:45 +0200169 pgprotval_t prot, cur;
H. Peter Anvina7b854c2014-04-29 16:46:09 -0700170 static const char units[] = "BKMGTPE";
Arjan van de Ven926e5392008-04-17 17:40:45 +0200171
172 /*
173 * If we have a "break" in the series, we need to flush the state that
H. Peter Anvinfe770bf02008-04-17 17:40:45 +0200174 * we have now. "break" is either changing perms, levels or
175 * address space marker.
Arjan van de Ven926e5392008-04-17 17:40:45 +0200176 */
Jeremy Fitzhardinge27990ea2008-08-19 13:10:07 -0700177 prot = pgprot_val(new_prot) & PTE_FLAGS_MASK;
178 cur = pgprot_val(st->current_prot) & PTE_FLAGS_MASK;
Arjan van de Ven926e5392008-04-17 17:40:45 +0200179
H. Peter Anvinfe770bf02008-04-17 17:40:45 +0200180 if (!st->level) {
181 /* First entry */
182 st->current_prot = new_prot;
183 st->level = level;
184 st->marker = address_markers;
H. Peter Anvina7b854c2014-04-29 16:46:09 -0700185 st->lines = 0;
H. Peter Anvinfe770bf02008-04-17 17:40:45 +0200186 seq_printf(m, "---[ %s ]---\n", st->marker->name);
187 } else if (prot != cur || level != st->level ||
188 st->current_address >= st->marker[1].start_address) {
189 const char *unit = units;
Arjan van de Ven926e5392008-04-17 17:40:45 +0200190 unsigned long delta;
Yinghai Lu6424fb32009-04-13 23:51:46 -0700191 int width = sizeof(unsigned long) * 2;
Arjan van de Ven926e5392008-04-17 17:40:45 +0200192
193 /*
H. Peter Anvinfe770bf02008-04-17 17:40:45 +0200194 * Now print the actual finished series
195 */
H. Peter Anvina7b854c2014-04-29 16:46:09 -0700196 if (!st->marker->max_lines ||
197 st->lines < st->marker->max_lines) {
198 seq_printf(m, "0x%0*lx-0x%0*lx ",
199 width, st->start_address,
200 width, st->current_address);
H. Peter Anvinfe770bf02008-04-17 17:40:45 +0200201
H. Peter Anvina7b854c2014-04-29 16:46:09 -0700202 delta = (st->current_address - st->start_address);
203 while (!(delta & 1023) && unit[1]) {
204 delta >>= 10;
205 unit++;
206 }
207 seq_printf(m, "%9lu%c ", delta, *unit);
208 printk_prot(m, st->current_prot, st->level);
H. Peter Anvinfe770bf02008-04-17 17:40:45 +0200209 }
H. Peter Anvina7b854c2014-04-29 16:46:09 -0700210 st->lines++;
H. Peter Anvinfe770bf02008-04-17 17:40:45 +0200211
212 /*
Arjan van de Ven926e5392008-04-17 17:40:45 +0200213 * We print markers for special areas of address space,
214 * such as the start of vmalloc space etc.
215 * This helps in the interpretation.
216 */
H. Peter Anvinfe770bf02008-04-17 17:40:45 +0200217 if (st->current_address >= st->marker[1].start_address) {
H. Peter Anvina7b854c2014-04-29 16:46:09 -0700218 if (st->marker->max_lines &&
219 st->lines > st->marker->max_lines) {
220 unsigned long nskip =
221 st->lines - st->marker->max_lines;
222 seq_printf(m, "... %lu entr%s skipped ... \n",
223 nskip, nskip == 1 ? "y" : "ies");
224 }
H. Peter Anvinfe770bf02008-04-17 17:40:45 +0200225 st->marker++;
H. Peter Anvina7b854c2014-04-29 16:46:09 -0700226 st->lines = 0;
H. Peter Anvinfe770bf02008-04-17 17:40:45 +0200227 seq_printf(m, "---[ %s ]---\n", st->marker->name);
Arjan van de Ven926e5392008-04-17 17:40:45 +0200228 }
229
Arjan van de Ven926e5392008-04-17 17:40:45 +0200230 st->start_address = st->current_address;
231 st->current_prot = new_prot;
232 st->level = level;
H. Peter Anvinfe770bf02008-04-17 17:40:45 +0200233 }
Arjan van de Ven926e5392008-04-17 17:40:45 +0200234}
235
H. Peter Anvinfe770bf02008-04-17 17:40:45 +0200236static void walk_pte_level(struct seq_file *m, struct pg_state *st, pmd_t addr,
Arjan van de Ven926e5392008-04-17 17:40:45 +0200237 unsigned long P)
238{
239 int i;
240 pte_t *start;
241
242 start = (pte_t *) pmd_page_vaddr(addr);
243 for (i = 0; i < PTRS_PER_PTE; i++) {
244 pgprot_t prot = pte_pgprot(*start);
245
H. Peter Anvinfe770bf02008-04-17 17:40:45 +0200246 st->current_address = normalize_addr(P + i * PTE_LEVEL_MULT);
Arjan van de Ven926e5392008-04-17 17:40:45 +0200247 note_page(m, st, prot, 4);
248 start++;
249 }
250}
251
H. Peter Anvinfe770bf02008-04-17 17:40:45 +0200252#if PTRS_PER_PMD > 1
Arjan van de Ven926e5392008-04-17 17:40:45 +0200253
H. Peter Anvinfe770bf02008-04-17 17:40:45 +0200254static void walk_pmd_level(struct seq_file *m, struct pg_state *st, pud_t addr,
Arjan van de Ven926e5392008-04-17 17:40:45 +0200255 unsigned long P)
256{
257 int i;
258 pmd_t *start;
259
260 start = (pmd_t *) pud_page_vaddr(addr);
261 for (i = 0; i < PTRS_PER_PMD; i++) {
H. Peter Anvinfe770bf02008-04-17 17:40:45 +0200262 st->current_address = normalize_addr(P + i * PMD_LEVEL_MULT);
Arjan van de Ven926e5392008-04-17 17:40:45 +0200263 if (!pmd_none(*start)) {
Jeremy Fitzhardinge77be1fa2008-07-21 22:59:56 -0700264 pgprotval_t prot = pmd_val(*start) & PTE_FLAGS_MASK;
Arjan van de Ven926e5392008-04-17 17:40:45 +0200265
H. Peter Anvinfe770bf02008-04-17 17:40:45 +0200266 if (pmd_large(*start) || !pmd_present(*start))
Arjan van de Ven926e5392008-04-17 17:40:45 +0200267 note_page(m, st, __pgprot(prot), 3);
268 else
H. Peter Anvinfe770bf02008-04-17 17:40:45 +0200269 walk_pte_level(m, st, *start,
270 P + i * PMD_LEVEL_MULT);
Arjan van de Ven926e5392008-04-17 17:40:45 +0200271 } else
272 note_page(m, st, __pgprot(0), 3);
273 start++;
274 }
275}
276
H. Peter Anvinfe770bf02008-04-17 17:40:45 +0200277#else
278#define walk_pmd_level(m,s,a,p) walk_pte_level(m,s,__pmd(pud_val(a)),p)
279#define pud_large(a) pmd_large(__pmd(pud_val(a)))
280#define pud_none(a) pmd_none(__pmd(pud_val(a)))
281#endif
Arjan van de Ven926e5392008-04-17 17:40:45 +0200282
H. Peter Anvinfe770bf02008-04-17 17:40:45 +0200283#if PTRS_PER_PUD > 1
284
285static void walk_pud_level(struct seq_file *m, struct pg_state *st, pgd_t addr,
Arjan van de Ven926e5392008-04-17 17:40:45 +0200286 unsigned long P)
287{
288 int i;
289 pud_t *start;
290
291 start = (pud_t *) pgd_page_vaddr(addr);
292
293 for (i = 0; i < PTRS_PER_PUD; i++) {
H. Peter Anvinfe770bf02008-04-17 17:40:45 +0200294 st->current_address = normalize_addr(P + i * PUD_LEVEL_MULT);
Arjan van de Ven926e5392008-04-17 17:40:45 +0200295 if (!pud_none(*start)) {
Jeremy Fitzhardinge77be1fa2008-07-21 22:59:56 -0700296 pgprotval_t prot = pud_val(*start) & PTE_FLAGS_MASK;
Arjan van de Ven926e5392008-04-17 17:40:45 +0200297
H. Peter Anvinfe770bf02008-04-17 17:40:45 +0200298 if (pud_large(*start) || !pud_present(*start))
Arjan van de Ven926e5392008-04-17 17:40:45 +0200299 note_page(m, st, __pgprot(prot), 2);
300 else
H. Peter Anvinfe770bf02008-04-17 17:40:45 +0200301 walk_pmd_level(m, st, *start,
302 P + i * PUD_LEVEL_MULT);
Arjan van de Ven926e5392008-04-17 17:40:45 +0200303 } else
304 note_page(m, st, __pgprot(0), 2);
305
306 start++;
307 }
308}
309
H. Peter Anvinfe770bf02008-04-17 17:40:45 +0200310#else
311#define walk_pud_level(m,s,a,p) walk_pmd_level(m,s,__pud(pgd_val(a)),p)
312#define pgd_large(a) pud_large(__pud(pgd_val(a)))
313#define pgd_none(a) pud_none(__pud(pgd_val(a)))
314#endif
315
316static void walk_pgd_level(struct seq_file *m)
Arjan van de Ven926e5392008-04-17 17:40:45 +0200317{
H. Peter Anvinfe770bf02008-04-17 17:40:45 +0200318#ifdef CONFIG_X86_64
Arjan van de Ven926e5392008-04-17 17:40:45 +0200319 pgd_t *start = (pgd_t *) &init_level4_pgt;
H. Peter Anvinfe770bf02008-04-17 17:40:45 +0200320#else
321 pgd_t *start = swapper_pg_dir;
322#endif
Arjan van de Ven926e5392008-04-17 17:40:45 +0200323 int i;
324 struct pg_state st;
325
326 memset(&st, 0, sizeof(st));
Arjan van de Ven926e5392008-04-17 17:40:45 +0200327
328 for (i = 0; i < PTRS_PER_PGD; i++) {
H. Peter Anvinfe770bf02008-04-17 17:40:45 +0200329 st.current_address = normalize_addr(i * PGD_LEVEL_MULT);
330 if (!pgd_none(*start)) {
Jeremy Fitzhardinge77be1fa2008-07-21 22:59:56 -0700331 pgprotval_t prot = pgd_val(*start) & PTE_FLAGS_MASK;
H. Peter Anvinfe770bf02008-04-17 17:40:45 +0200332
333 if (pgd_large(*start) || !pgd_present(*start))
334 note_page(m, &st, __pgprot(prot), 1);
335 else
336 walk_pud_level(m, &st, *start,
337 i * PGD_LEVEL_MULT);
338 } else
Arjan van de Ven926e5392008-04-17 17:40:45 +0200339 note_page(m, &st, __pgprot(0), 1);
H. Peter Anvinfe770bf02008-04-17 17:40:45 +0200340
Arjan van de Ven926e5392008-04-17 17:40:45 +0200341 start++;
342 }
H. Peter Anvinfe770bf02008-04-17 17:40:45 +0200343
344 /* Flush out the last page */
345 st.current_address = normalize_addr(PTRS_PER_PGD*PGD_LEVEL_MULT);
346 note_page(m, &st, __pgprot(0), 0);
Arjan van de Ven926e5392008-04-17 17:40:45 +0200347}
348
349static int ptdump_show(struct seq_file *m, void *v)
350{
H. Peter Anvinfe770bf02008-04-17 17:40:45 +0200351 walk_pgd_level(m);
Arjan van de Ven926e5392008-04-17 17:40:45 +0200352 return 0;
353}
354
355static int ptdump_open(struct inode *inode, struct file *filp)
356{
357 return single_open(filp, ptdump_show, NULL);
358}
359
360static const struct file_operations ptdump_fops = {
361 .open = ptdump_open,
362 .read = seq_read,
363 .llseek = seq_lseek,
364 .release = single_release,
365};
366
Ingo Molnara4928cf2008-04-23 13:20:56 +0200367static int pt_dump_init(void)
Arjan van de Ven926e5392008-04-17 17:40:45 +0200368{
369 struct dentry *pe;
370
H. Peter Anvinfe770bf02008-04-17 17:40:45 +0200371#ifdef CONFIG_X86_32
372 /* Not a compile-time constant on x86-32 */
Andres Salomon92851e22010-07-20 15:19:46 -0700373 address_markers[VMALLOC_START_NR].start_address = VMALLOC_START;
374 address_markers[VMALLOC_END_NR].start_address = VMALLOC_END;
H. Peter Anvinfe770bf02008-04-17 17:40:45 +0200375# ifdef CONFIG_HIGHMEM
Andres Salomon92851e22010-07-20 15:19:46 -0700376 address_markers[PKMAP_BASE_NR].start_address = PKMAP_BASE;
H. Peter Anvinfe770bf02008-04-17 17:40:45 +0200377# endif
Andres Salomon92851e22010-07-20 15:19:46 -0700378 address_markers[FIXADDR_START_NR].start_address = FIXADDR_START;
H. Peter Anvinfe770bf02008-04-17 17:40:45 +0200379#endif
380
Arjan van de Ven926e5392008-04-17 17:40:45 +0200381 pe = debugfs_create_file("kernel_page_tables", 0600, NULL, NULL,
382 &ptdump_fops);
383 if (!pe)
384 return -ENOMEM;
385
386 return 0;
387}
388
389__initcall(pt_dump_init);
390MODULE_LICENSE("GPL");
391MODULE_AUTHOR("Arjan van de Ven <arjan@linux.intel.com>");
392MODULE_DESCRIPTION("Kernel debugging helper that dumps pagetables");