1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
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 implementation:
7 * (C) Copyright 2008 Intel Corporation
8 *
9 * Author: Arjan van de Ven <arjan@linux.intel.com>
10 */
11 #include <linux/debugfs.h>
12 #include <linux/fs.h>
13 #include <linux/mm.h>
14 #include <linux/seq_file.h>
15
16 #include <asm/domain.h>
17 #include <asm/fixmap.h>
18 #include <asm/page.h>
19 #include <asm/ptdump.h>
20
21 static struct addr_marker address_markers[] = {
22 #ifdef CONFIG_KASAN
23 { KASAN_SHADOW_START, "Kasan shadow start"},
24 { KASAN_SHADOW_END, "Kasan shadow end"},
25 #endif
26 { MODULES_VADDR, "Modules" },
27 { PAGE_OFFSET, "Kernel Mapping" },
28 { 0, "vmalloc() Area" },
29 { FDT_FIXED_BASE, "FDT Area" },
30 { FIXADDR_START, "Fixmap Area" },
31 { VECTORS_BASE, "Vectors" },
32 { VECTORS_BASE + PAGE_SIZE * 2, "Vectors End" },
33 { -1, NULL },
34 };
35
36 #define pt_dump_seq_printf(m, fmt, args...) \
37 ({ \
38 if (m) \
39 seq_printf(m, fmt, ##args); \
40 })
41
42 #define pt_dump_seq_puts(m, fmt) \
43 ({ \
44 if (m) \
45 seq_printf(m, fmt); \
46 })
47
48 struct pg_state {
49 struct seq_file *seq;
50 const struct addr_marker *marker;
51 unsigned long start_address;
52 unsigned level;
53 u64 current_prot;
54 bool check_wx;
55 unsigned long wx_pages;
56 const char *current_domain;
57 };
58
59 struct prot_bits {
60 u64 mask;
61 u64 val;
62 const char *set;
63 const char *clear;
64 bool ro_bit;
65 bool nx_bit;
66 };
67
68 static const struct prot_bits pte_bits[] = {
69 {
70 .mask = L_PTE_USER,
71 .val = L_PTE_USER,
72 .set = "USR",
73 .clear = " ",
74 }, {
75 .mask = L_PTE_RDONLY,
76 .val = L_PTE_RDONLY,
77 .set = "ro",
78 .clear = "RW",
79 .ro_bit = true,
80 }, {
81 .mask = L_PTE_XN,
82 .val = L_PTE_XN,
83 .set = "NX",
84 .clear = "x ",
85 .nx_bit = true,
86 }, {
87 .mask = L_PTE_SHARED,
88 .val = L_PTE_SHARED,
89 .set = "SHD",
90 .clear = " ",
91 }, {
92 .mask = L_PTE_MT_MASK,
93 .val = L_PTE_MT_UNCACHED,
94 .set = "SO/UNCACHED",
95 }, {
96 .mask = L_PTE_MT_MASK,
97 .val = L_PTE_MT_BUFFERABLE,
98 .set = "MEM/BUFFERABLE/WC",
99 }, {
100 .mask = L_PTE_MT_MASK,
101 .val = L_PTE_MT_WRITETHROUGH,
102 .set = "MEM/CACHED/WT",
103 }, {
104 .mask = L_PTE_MT_MASK,
105 .val = L_PTE_MT_WRITEBACK,
106 .set = "MEM/CACHED/WBRA",
107 #ifndef CONFIG_ARM_LPAE
108 }, {
109 .mask = L_PTE_MT_MASK,
110 .val = L_PTE_MT_MINICACHE,
111 .set = "MEM/MINICACHE",
112 #endif
113 }, {
114 .mask = L_PTE_MT_MASK,
115 .val = L_PTE_MT_WRITEALLOC,
116 .set = "MEM/CACHED/WBWA",
117 }, {
118 .mask = L_PTE_MT_MASK,
119 .val = L_PTE_MT_DEV_SHARED,
120 .set = "DEV/SHARED",
121 #ifndef CONFIG_ARM_LPAE
122 }, {
123 .mask = L_PTE_MT_MASK,
124 .val = L_PTE_MT_DEV_NONSHARED,
125 .set = "DEV/NONSHARED",
126 #endif
127 }, {
128 .mask = L_PTE_MT_MASK,
129 .val = L_PTE_MT_DEV_WC,
130 .set = "DEV/WC",
131 }, {
132 .mask = L_PTE_MT_MASK,
133 .val = L_PTE_MT_DEV_CACHED,
134 .set = "DEV/CACHED",
135 },
136 };
137
138 static const struct prot_bits section_bits[] = {
139 #ifdef CONFIG_ARM_LPAE
140 {
141 .mask = PMD_SECT_USER,
142 .val = PMD_SECT_USER,
143 .set = "USR",
144 }, {
145 .mask = L_PMD_SECT_RDONLY | PMD_SECT_AP2,
146 .val = L_PMD_SECT_RDONLY | PMD_SECT_AP2,
147 .set = "ro",
148 .clear = "RW",
149 .ro_bit = true,
150 #elif __LINUX_ARM_ARCH__ >= 6
151 {
152 .mask = PMD_SECT_APX | PMD_SECT_AP_READ | PMD_SECT_AP_WRITE,
153 .val = PMD_SECT_APX | PMD_SECT_AP_WRITE,
154 .set = " ro",
155 .ro_bit = true,
156 }, {
157 .mask = PMD_SECT_APX | PMD_SECT_AP_READ | PMD_SECT_AP_WRITE,
158 .val = PMD_SECT_AP_WRITE,
159 .set = " RW",
160 }, {
161 .mask = PMD_SECT_APX | PMD_SECT_AP_READ | PMD_SECT_AP_WRITE,
162 .val = PMD_SECT_AP_READ,
163 .set = "USR ro",
164 }, {
165 .mask = PMD_SECT_APX | PMD_SECT_AP_READ | PMD_SECT_AP_WRITE,
166 .val = PMD_SECT_AP_READ | PMD_SECT_AP_WRITE,
167 .set = "USR RW",
168 #else /* ARMv4/ARMv5 */
169 /* These are approximate */
170 {
171 .mask = PMD_SECT_AP_READ | PMD_SECT_AP_WRITE,
172 .val = 0,
173 .set = " ro",
174 .ro_bit = true,
175 }, {
176 .mask = PMD_SECT_AP_READ | PMD_SECT_AP_WRITE,
177 .val = PMD_SECT_AP_WRITE,
178 .set = " RW",
179 }, {
180 .mask = PMD_SECT_AP_READ | PMD_SECT_AP_WRITE,
181 .val = PMD_SECT_AP_READ,
182 .set = "USR ro",
183 }, {
184 .mask = PMD_SECT_AP_READ | PMD_SECT_AP_WRITE,
185 .val = PMD_SECT_AP_READ | PMD_SECT_AP_WRITE,
186 .set = "USR RW",
187 #endif
188 }, {
189 .mask = PMD_SECT_XN,
190 .val = PMD_SECT_XN,
191 .set = "NX",
192 .clear = "x ",
193 .nx_bit = true,
194 }, {
195 .mask = PMD_SECT_S,
196 .val = PMD_SECT_S,
197 .set = "SHD",
198 .clear = " ",
199 },
200 };
201
202 struct pg_level {
203 const char *name;
204 const struct prot_bits *bits;
205 size_t num;
206 u64 mask;
207 const struct prot_bits *ro_bit;
208 const struct prot_bits *nx_bit;
209 };
210
211 static struct pg_level pg_level[] = {
212 {
213 }, { /* pgd */
214 }, { /* p4d */
215 }, { /* pud */
216 }, { /* pmd */
217 .name = (CONFIG_PGTABLE_LEVELS > 2) ? "PMD" : "PGD",
218 .bits = section_bits,
219 .num = ARRAY_SIZE(section_bits),
220 }, { /* pte */
221 .name = "PTE",
222 .bits = pte_bits,
223 .num = ARRAY_SIZE(pte_bits),
224 },
225 };
226
dump_prot(struct pg_state * st,const struct prot_bits * bits,size_t num)227 static void dump_prot(struct pg_state *st, const struct prot_bits *bits, size_t num)
228 {
229 unsigned i;
230
231 for (i = 0; i < num; i++, bits++) {
232 const char *s;
233
234 if ((st->current_prot & bits->mask) == bits->val)
235 s = bits->set;
236 else
237 s = bits->clear;
238
239 if (s)
240 pt_dump_seq_printf(st->seq, " %s", s);
241 }
242 }
243
note_prot_wx(struct pg_state * st,unsigned long addr)244 static void note_prot_wx(struct pg_state *st, unsigned long addr)
245 {
246 if (!st->check_wx)
247 return;
248 if ((st->current_prot & pg_level[st->level].ro_bit->mask) ==
249 pg_level[st->level].ro_bit->val)
250 return;
251 if ((st->current_prot & pg_level[st->level].nx_bit->mask) ==
252 pg_level[st->level].nx_bit->val)
253 return;
254
255 WARN_ONCE(1, "arm/mm: Found insecure W+X mapping at address %pS\n",
256 (void *)st->start_address);
257
258 st->wx_pages += (addr - st->start_address) / PAGE_SIZE;
259 }
260
note_page(struct pg_state * st,unsigned long addr,unsigned int level,u64 val,const char * domain)261 static void note_page(struct pg_state *st, unsigned long addr,
262 unsigned int level, u64 val, const char *domain)
263 {
264 static const char units[] = "KMGTPE";
265 u64 prot = val & pg_level[level].mask;
266
267 if (!st->level) {
268 st->level = level;
269 st->current_prot = prot;
270 st->current_domain = domain;
271 pt_dump_seq_printf(st->seq, "---[ %s ]---\n", st->marker->name);
272 } else if (prot != st->current_prot || level != st->level ||
273 domain != st->current_domain ||
274 addr >= st->marker[1].start_address) {
275 const char *unit = units;
276 unsigned long delta;
277
278 if (st->current_prot) {
279 note_prot_wx(st, addr);
280 pt_dump_seq_printf(st->seq, "0x%08lx-0x%08lx ",
281 st->start_address, addr);
282
283 delta = (addr - st->start_address) >> 10;
284 while (!(delta & 1023) && unit[1]) {
285 delta >>= 10;
286 unit++;
287 }
288 pt_dump_seq_printf(st->seq, "%9lu%c %s", delta, *unit,
289 pg_level[st->level].name);
290 if (st->current_domain)
291 pt_dump_seq_printf(st->seq, " %s",
292 st->current_domain);
293 if (pg_level[st->level].bits)
294 dump_prot(st, pg_level[st->level].bits, pg_level[st->level].num);
295 pt_dump_seq_printf(st->seq, "\n");
296 }
297
298 if (addr >= st->marker[1].start_address) {
299 st->marker++;
300 pt_dump_seq_printf(st->seq, "---[ %s ]---\n",
301 st->marker->name);
302 }
303 st->start_address = addr;
304 st->current_prot = prot;
305 st->current_domain = domain;
306 st->level = level;
307 }
308 }
309
walk_pte(struct pg_state * st,pmd_t * pmd,unsigned long start,const char * domain)310 static void walk_pte(struct pg_state *st, pmd_t *pmd, unsigned long start,
311 const char *domain)
312 {
313 pte_t *pte = pte_offset_kernel(pmd, 0);
314 unsigned long addr;
315 unsigned i;
316
317 for (i = 0; i < PTRS_PER_PTE; i++, pte++) {
318 addr = start + i * PAGE_SIZE;
319 note_page(st, addr, 5, pte_val(*pte), domain);
320 }
321 }
322
get_domain_name(pmd_t * pmd)323 static const char *get_domain_name(pmd_t *pmd)
324 {
325 #ifndef CONFIG_ARM_LPAE
326 switch (pmd_val(*pmd) & PMD_DOMAIN_MASK) {
327 case PMD_DOMAIN(DOMAIN_KERNEL):
328 return "KERNEL ";
329 case PMD_DOMAIN(DOMAIN_USER):
330 return "USER ";
331 case PMD_DOMAIN(DOMAIN_IO):
332 return "IO ";
333 case PMD_DOMAIN(DOMAIN_VECTORS):
334 return "VECTORS";
335 default:
336 return "unknown";
337 }
338 #endif
339 return NULL;
340 }
341
walk_pmd(struct pg_state * st,pud_t * pud,unsigned long start)342 static void walk_pmd(struct pg_state *st, pud_t *pud, unsigned long start)
343 {
344 pmd_t *pmd = pmd_offset(pud, 0);
345 unsigned long addr;
346 unsigned i;
347 const char *domain;
348
349 for (i = 0; i < PTRS_PER_PMD; i++, pmd++) {
350 addr = start + i * PMD_SIZE;
351 domain = get_domain_name(pmd);
352 if (pmd_none(*pmd) || pmd_leaf(*pmd) || !pmd_present(*pmd))
353 note_page(st, addr, 4, pmd_val(*pmd), domain);
354 else
355 walk_pte(st, pmd, addr, domain);
356
357 if (SECTION_SIZE < PMD_SIZE && pmd_leaf(pmd[1])) {
358 addr += SECTION_SIZE;
359 pmd++;
360 domain = get_domain_name(pmd);
361 note_page(st, addr, 4, pmd_val(*pmd), domain);
362 }
363 }
364 }
365
walk_pud(struct pg_state * st,p4d_t * p4d,unsigned long start)366 static void walk_pud(struct pg_state *st, p4d_t *p4d, unsigned long start)
367 {
368 pud_t *pud = pud_offset(p4d, 0);
369 unsigned long addr;
370 unsigned i;
371
372 for (i = 0; i < PTRS_PER_PUD; i++, pud++) {
373 addr = start + i * PUD_SIZE;
374 if (!pud_none(*pud)) {
375 walk_pmd(st, pud, addr);
376 } else {
377 note_page(st, addr, 3, pud_val(*pud), NULL);
378 }
379 }
380 }
381
walk_p4d(struct pg_state * st,pgd_t * pgd,unsigned long start)382 static void walk_p4d(struct pg_state *st, pgd_t *pgd, unsigned long start)
383 {
384 p4d_t *p4d = p4d_offset(pgd, 0);
385 unsigned long addr;
386 unsigned i;
387
388 for (i = 0; i < PTRS_PER_P4D; i++, p4d++) {
389 addr = start + i * P4D_SIZE;
390 if (!p4d_none(*p4d)) {
391 walk_pud(st, p4d, addr);
392 } else {
393 note_page(st, addr, 2, p4d_val(*p4d), NULL);
394 }
395 }
396 }
397
walk_pgd(struct pg_state * st,struct mm_struct * mm,unsigned long start)398 static void walk_pgd(struct pg_state *st, struct mm_struct *mm,
399 unsigned long start)
400 {
401 pgd_t *pgd = pgd_offset(mm, 0UL);
402 unsigned i;
403 unsigned long addr;
404
405 for (i = 0; i < PTRS_PER_PGD; i++, pgd++) {
406 addr = start + i * PGDIR_SIZE;
407 if (!pgd_none(*pgd)) {
408 walk_p4d(st, pgd, addr);
409 } else {
410 note_page(st, addr, 1, pgd_val(*pgd), NULL);
411 }
412 }
413 }
414
ptdump_walk_pgd(struct seq_file * m,struct ptdump_info * info)415 void ptdump_walk_pgd(struct seq_file *m, struct ptdump_info *info)
416 {
417 struct pg_state st = {
418 .seq = m,
419 .marker = info->markers,
420 .check_wx = false,
421 };
422
423 walk_pgd(&st, info->mm, info->base_addr);
424 note_page(&st, 0, 0, 0, NULL);
425 }
426
ptdump_initialize(void)427 static void __init ptdump_initialize(void)
428 {
429 unsigned i, j;
430
431 for (i = 0; i < ARRAY_SIZE(pg_level); i++)
432 if (pg_level[i].bits)
433 for (j = 0; j < pg_level[i].num; j++) {
434 pg_level[i].mask |= pg_level[i].bits[j].mask;
435 if (pg_level[i].bits[j].ro_bit)
436 pg_level[i].ro_bit = &pg_level[i].bits[j];
437 if (pg_level[i].bits[j].nx_bit)
438 pg_level[i].nx_bit = &pg_level[i].bits[j];
439 }
440 #ifdef CONFIG_KASAN
441 address_markers[4].start_address = VMALLOC_START;
442 #else
443 address_markers[2].start_address = VMALLOC_START;
444 #endif
445 }
446
447 static struct ptdump_info kernel_ptdump_info = {
448 .mm = &init_mm,
449 .markers = address_markers,
450 .base_addr = 0,
451 };
452
ptdump_check_wx(void)453 void ptdump_check_wx(void)
454 {
455 struct pg_state st = {
456 .seq = NULL,
457 .marker = (struct addr_marker[]) {
458 { 0, NULL},
459 { -1, NULL},
460 },
461 .check_wx = true,
462 };
463
464 walk_pgd(&st, &init_mm, 0);
465 note_page(&st, 0, 0, 0, NULL);
466 if (st.wx_pages)
467 pr_warn("Checked W+X mappings: FAILED, %lu W+X pages found\n",
468 st.wx_pages);
469 else
470 pr_info("Checked W+X mappings: passed, no W+X pages found\n");
471 }
472
ptdump_init(void)473 static int __init ptdump_init(void)
474 {
475 ptdump_initialize();
476 ptdump_debugfs_register(&kernel_ptdump_info, "kernel_page_tables");
477 return 0;
478 }
479 __initcall(ptdump_init);
480