xref: /openbmc/linux/arch/x86/mm/dump_pagetables.c (revision e3b9f1e8)
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/kasan.h>
17 #include <linux/mm.h>
18 #include <linux/init.h>
19 #include <linux/sched.h>
20 #include <linux/seq_file.h>
21 
22 #include <asm/pgtable.h>
23 
24 /*
25  * The dumper groups pagetable entries of the same type into one, and for
26  * that it needs to keep some state when walking, and flush this state
27  * when a "break" in the continuity is found.
28  */
29 struct pg_state {
30 	int level;
31 	pgprot_t current_prot;
32 	unsigned long start_address;
33 	unsigned long current_address;
34 	const struct addr_marker *marker;
35 	unsigned long lines;
36 	bool to_dmesg;
37 	bool check_wx;
38 	unsigned long wx_pages;
39 };
40 
41 struct addr_marker {
42 	unsigned long start_address;
43 	const char *name;
44 	unsigned long max_lines;
45 };
46 
47 /* Address space markers hints */
48 
49 #ifdef CONFIG_X86_64
50 
51 enum address_markers_idx {
52 	USER_SPACE_NR = 0,
53 	KERNEL_SPACE_NR,
54 	LOW_KERNEL_NR,
55 #if defined(CONFIG_MODIFY_LDT_SYSCALL) && defined(CONFIG_X86_5LEVEL)
56 	LDT_NR,
57 #endif
58 	VMALLOC_START_NR,
59 	VMEMMAP_START_NR,
60 #ifdef CONFIG_KASAN
61 	KASAN_SHADOW_START_NR,
62 	KASAN_SHADOW_END_NR,
63 #endif
64 	CPU_ENTRY_AREA_NR,
65 #if defined(CONFIG_MODIFY_LDT_SYSCALL) && !defined(CONFIG_X86_5LEVEL)
66 	LDT_NR,
67 #endif
68 #ifdef CONFIG_X86_ESPFIX64
69 	ESPFIX_START_NR,
70 #endif
71 #ifdef CONFIG_EFI
72 	EFI_END_NR,
73 #endif
74 	HIGH_KERNEL_NR,
75 	MODULES_VADDR_NR,
76 	MODULES_END_NR,
77 	FIXADDR_START_NR,
78 	END_OF_SPACE_NR,
79 };
80 
81 static struct addr_marker address_markers[] = {
82 	[USER_SPACE_NR]		= { 0,			"User Space" },
83 	[KERNEL_SPACE_NR]	= { (1UL << 63),	"Kernel Space" },
84 	[LOW_KERNEL_NR]		= { 0UL,		"Low Kernel Mapping" },
85 	[VMALLOC_START_NR]	= { 0UL,		"vmalloc() Area" },
86 	[VMEMMAP_START_NR]	= { 0UL,		"Vmemmap" },
87 #ifdef CONFIG_KASAN
88 	[KASAN_SHADOW_START_NR]	= { KASAN_SHADOW_START,	"KASAN shadow" },
89 	[KASAN_SHADOW_END_NR]	= { KASAN_SHADOW_END,	"KASAN shadow end" },
90 #endif
91 #ifdef CONFIG_MODIFY_LDT_SYSCALL
92 	[LDT_NR]		= { LDT_BASE_ADDR,	"LDT remap" },
93 #endif
94 	[CPU_ENTRY_AREA_NR]	= { CPU_ENTRY_AREA_BASE,"CPU entry Area" },
95 #ifdef CONFIG_X86_ESPFIX64
96 	[ESPFIX_START_NR]	= { ESPFIX_BASE_ADDR,	"ESPfix Area", 16 },
97 #endif
98 #ifdef CONFIG_EFI
99 	[EFI_END_NR]		= { EFI_VA_END,		"EFI Runtime Services" },
100 #endif
101 	[HIGH_KERNEL_NR]	= { __START_KERNEL_map,	"High Kernel Mapping" },
102 	[MODULES_VADDR_NR]	= { MODULES_VADDR,	"Modules" },
103 	[MODULES_END_NR]	= { MODULES_END,	"End Modules" },
104 	[FIXADDR_START_NR]	= { FIXADDR_START,	"Fixmap Area" },
105 	[END_OF_SPACE_NR]	= { -1,			NULL }
106 };
107 
108 #else /* CONFIG_X86_64 */
109 
110 enum address_markers_idx {
111 	USER_SPACE_NR = 0,
112 	KERNEL_SPACE_NR,
113 	VMALLOC_START_NR,
114 	VMALLOC_END_NR,
115 #ifdef CONFIG_HIGHMEM
116 	PKMAP_BASE_NR,
117 #endif
118 	CPU_ENTRY_AREA_NR,
119 	FIXADDR_START_NR,
120 	END_OF_SPACE_NR,
121 };
122 
123 static struct addr_marker address_markers[] = {
124 	[USER_SPACE_NR]		= { 0,			"User Space" },
125 	[KERNEL_SPACE_NR]	= { PAGE_OFFSET,	"Kernel Mapping" },
126 	[VMALLOC_START_NR]	= { 0UL,		"vmalloc() Area" },
127 	[VMALLOC_END_NR]	= { 0UL,		"vmalloc() End" },
128 #ifdef CONFIG_HIGHMEM
129 	[PKMAP_BASE_NR]		= { 0UL,		"Persistent kmap() Area" },
130 #endif
131 	[CPU_ENTRY_AREA_NR]	= { 0UL,		"CPU entry area" },
132 	[FIXADDR_START_NR]	= { 0UL,		"Fixmap area" },
133 	[END_OF_SPACE_NR]	= { -1,			NULL }
134 };
135 
136 #endif /* !CONFIG_X86_64 */
137 
138 /* Multipliers for offsets within the PTEs */
139 #define PTE_LEVEL_MULT (PAGE_SIZE)
140 #define PMD_LEVEL_MULT (PTRS_PER_PTE * PTE_LEVEL_MULT)
141 #define PUD_LEVEL_MULT (PTRS_PER_PMD * PMD_LEVEL_MULT)
142 #define P4D_LEVEL_MULT (PTRS_PER_PUD * PUD_LEVEL_MULT)
143 #define PGD_LEVEL_MULT (PTRS_PER_P4D * P4D_LEVEL_MULT)
144 
145 #define pt_dump_seq_printf(m, to_dmesg, fmt, args...)		\
146 ({								\
147 	if (to_dmesg)					\
148 		printk(KERN_INFO fmt, ##args);			\
149 	else							\
150 		if (m)						\
151 			seq_printf(m, fmt, ##args);		\
152 })
153 
154 #define pt_dump_cont_printf(m, to_dmesg, fmt, args...)		\
155 ({								\
156 	if (to_dmesg)					\
157 		printk(KERN_CONT fmt, ##args);			\
158 	else							\
159 		if (m)						\
160 			seq_printf(m, fmt, ##args);		\
161 })
162 
163 /*
164  * Print a readable form of a pgprot_t to the seq_file
165  */
166 static void printk_prot(struct seq_file *m, pgprot_t prot, int level, bool dmsg)
167 {
168 	pgprotval_t pr = pgprot_val(prot);
169 	static const char * const level_name[] =
170 		{ "cr3", "pgd", "p4d", "pud", "pmd", "pte" };
171 
172 	if (!(pr & _PAGE_PRESENT)) {
173 		/* Not present */
174 		pt_dump_cont_printf(m, dmsg, "                              ");
175 	} else {
176 		if (pr & _PAGE_USER)
177 			pt_dump_cont_printf(m, dmsg, "USR ");
178 		else
179 			pt_dump_cont_printf(m, dmsg, "    ");
180 		if (pr & _PAGE_RW)
181 			pt_dump_cont_printf(m, dmsg, "RW ");
182 		else
183 			pt_dump_cont_printf(m, dmsg, "ro ");
184 		if (pr & _PAGE_PWT)
185 			pt_dump_cont_printf(m, dmsg, "PWT ");
186 		else
187 			pt_dump_cont_printf(m, dmsg, "    ");
188 		if (pr & _PAGE_PCD)
189 			pt_dump_cont_printf(m, dmsg, "PCD ");
190 		else
191 			pt_dump_cont_printf(m, dmsg, "    ");
192 
193 		/* Bit 7 has a different meaning on level 3 vs 4 */
194 		if (level <= 4 && pr & _PAGE_PSE)
195 			pt_dump_cont_printf(m, dmsg, "PSE ");
196 		else
197 			pt_dump_cont_printf(m, dmsg, "    ");
198 		if ((level == 5 && pr & _PAGE_PAT) ||
199 		    ((level == 4 || level == 3) && pr & _PAGE_PAT_LARGE))
200 			pt_dump_cont_printf(m, dmsg, "PAT ");
201 		else
202 			pt_dump_cont_printf(m, dmsg, "    ");
203 		if (pr & _PAGE_GLOBAL)
204 			pt_dump_cont_printf(m, dmsg, "GLB ");
205 		else
206 			pt_dump_cont_printf(m, dmsg, "    ");
207 		if (pr & _PAGE_NX)
208 			pt_dump_cont_printf(m, dmsg, "NX ");
209 		else
210 			pt_dump_cont_printf(m, dmsg, "x  ");
211 	}
212 	pt_dump_cont_printf(m, dmsg, "%s\n", level_name[level]);
213 }
214 
215 /*
216  * On 64 bits, sign-extend the 48 bit address to 64 bit
217  */
218 static unsigned long normalize_addr(unsigned long u)
219 {
220 	int shift;
221 	if (!IS_ENABLED(CONFIG_X86_64))
222 		return u;
223 
224 	shift = 64 - (__VIRTUAL_MASK_SHIFT + 1);
225 	return (signed long)(u << shift) >> shift;
226 }
227 
228 /*
229  * This function gets called on a break in a continuous series
230  * of PTE entries; the next one is different so we need to
231  * print what we collected so far.
232  */
233 static void note_page(struct seq_file *m, struct pg_state *st,
234 		      pgprot_t new_prot, int level)
235 {
236 	pgprotval_t prot, cur;
237 	static const char units[] = "BKMGTPE";
238 
239 	/*
240 	 * If we have a "break" in the series, we need to flush the state that
241 	 * we have now. "break" is either changing perms, levels or
242 	 * address space marker.
243 	 */
244 	prot = pgprot_val(new_prot);
245 	cur = pgprot_val(st->current_prot);
246 
247 	if (!st->level) {
248 		/* First entry */
249 		st->current_prot = new_prot;
250 		st->level = level;
251 		st->marker = address_markers;
252 		st->lines = 0;
253 		pt_dump_seq_printf(m, st->to_dmesg, "---[ %s ]---\n",
254 				   st->marker->name);
255 	} else if (prot != cur || level != st->level ||
256 		   st->current_address >= st->marker[1].start_address) {
257 		const char *unit = units;
258 		unsigned long delta;
259 		int width = sizeof(unsigned long) * 2;
260 		pgprotval_t pr = pgprot_val(st->current_prot);
261 
262 		if (st->check_wx && (pr & _PAGE_RW) && !(pr & _PAGE_NX)) {
263 			WARN_ONCE(1,
264 				  "x86/mm: Found insecure W+X mapping at address %p/%pS\n",
265 				  (void *)st->start_address,
266 				  (void *)st->start_address);
267 			st->wx_pages += (st->current_address -
268 					 st->start_address) / PAGE_SIZE;
269 		}
270 
271 		/*
272 		 * Now print the actual finished series
273 		 */
274 		if (!st->marker->max_lines ||
275 		    st->lines < st->marker->max_lines) {
276 			pt_dump_seq_printf(m, st->to_dmesg,
277 					   "0x%0*lx-0x%0*lx   ",
278 					   width, st->start_address,
279 					   width, st->current_address);
280 
281 			delta = st->current_address - st->start_address;
282 			while (!(delta & 1023) && unit[1]) {
283 				delta >>= 10;
284 				unit++;
285 			}
286 			pt_dump_cont_printf(m, st->to_dmesg, "%9lu%c ",
287 					    delta, *unit);
288 			printk_prot(m, st->current_prot, st->level,
289 				    st->to_dmesg);
290 		}
291 		st->lines++;
292 
293 		/*
294 		 * We print markers for special areas of address space,
295 		 * such as the start of vmalloc space etc.
296 		 * This helps in the interpretation.
297 		 */
298 		if (st->current_address >= st->marker[1].start_address) {
299 			if (st->marker->max_lines &&
300 			    st->lines > st->marker->max_lines) {
301 				unsigned long nskip =
302 					st->lines - st->marker->max_lines;
303 				pt_dump_seq_printf(m, st->to_dmesg,
304 						   "... %lu entr%s skipped ... \n",
305 						   nskip,
306 						   nskip == 1 ? "y" : "ies");
307 			}
308 			st->marker++;
309 			st->lines = 0;
310 			pt_dump_seq_printf(m, st->to_dmesg, "---[ %s ]---\n",
311 					   st->marker->name);
312 		}
313 
314 		st->start_address = st->current_address;
315 		st->current_prot = new_prot;
316 		st->level = level;
317 	}
318 }
319 
320 static void walk_pte_level(struct seq_file *m, struct pg_state *st, pmd_t addr, unsigned long P)
321 {
322 	int i;
323 	pte_t *start;
324 	pgprotval_t prot;
325 
326 	start = (pte_t *)pmd_page_vaddr(addr);
327 	for (i = 0; i < PTRS_PER_PTE; i++) {
328 		prot = pte_flags(*start);
329 		st->current_address = normalize_addr(P + i * PTE_LEVEL_MULT);
330 		note_page(m, st, __pgprot(prot), 5);
331 		start++;
332 	}
333 }
334 #ifdef CONFIG_KASAN
335 
336 /*
337  * This is an optimization for KASAN=y case. Since all kasan page tables
338  * eventually point to the kasan_zero_page we could call note_page()
339  * right away without walking through lower level page tables. This saves
340  * us dozens of seconds (minutes for 5-level config) while checking for
341  * W+X mapping or reading kernel_page_tables debugfs file.
342  */
343 static inline bool kasan_page_table(struct seq_file *m, struct pg_state *st,
344 				void *pt)
345 {
346 	if (__pa(pt) == __pa(kasan_zero_pmd) ||
347 #ifdef CONFIG_X86_5LEVEL
348 	    __pa(pt) == __pa(kasan_zero_p4d) ||
349 #endif
350 	    __pa(pt) == __pa(kasan_zero_pud)) {
351 		pgprotval_t prot = pte_flags(kasan_zero_pte[0]);
352 		note_page(m, st, __pgprot(prot), 5);
353 		return true;
354 	}
355 	return false;
356 }
357 #else
358 static inline bool kasan_page_table(struct seq_file *m, struct pg_state *st,
359 				void *pt)
360 {
361 	return false;
362 }
363 #endif
364 
365 #if PTRS_PER_PMD > 1
366 
367 static void walk_pmd_level(struct seq_file *m, struct pg_state *st, pud_t addr, unsigned long P)
368 {
369 	int i;
370 	pmd_t *start, *pmd_start;
371 	pgprotval_t prot;
372 
373 	pmd_start = start = (pmd_t *)pud_page_vaddr(addr);
374 	for (i = 0; i < PTRS_PER_PMD; i++) {
375 		st->current_address = normalize_addr(P + i * PMD_LEVEL_MULT);
376 		if (!pmd_none(*start)) {
377 			if (pmd_large(*start) || !pmd_present(*start)) {
378 				prot = pmd_flags(*start);
379 				note_page(m, st, __pgprot(prot), 4);
380 			} else if (!kasan_page_table(m, st, pmd_start)) {
381 				walk_pte_level(m, st, *start,
382 					       P + i * PMD_LEVEL_MULT);
383 			}
384 		} else
385 			note_page(m, st, __pgprot(0), 4);
386 		start++;
387 	}
388 }
389 
390 #else
391 #define walk_pmd_level(m,s,a,p) walk_pte_level(m,s,__pmd(pud_val(a)),p)
392 #define pud_large(a) pmd_large(__pmd(pud_val(a)))
393 #define pud_none(a)  pmd_none(__pmd(pud_val(a)))
394 #endif
395 
396 #if PTRS_PER_PUD > 1
397 
398 static void walk_pud_level(struct seq_file *m, struct pg_state *st, p4d_t addr, unsigned long P)
399 {
400 	int i;
401 	pud_t *start, *pud_start;
402 	pgprotval_t prot;
403 	pud_t *prev_pud = NULL;
404 
405 	pud_start = start = (pud_t *)p4d_page_vaddr(addr);
406 
407 	for (i = 0; i < PTRS_PER_PUD; i++) {
408 		st->current_address = normalize_addr(P + i * PUD_LEVEL_MULT);
409 		if (!pud_none(*start)) {
410 			if (pud_large(*start) || !pud_present(*start)) {
411 				prot = pud_flags(*start);
412 				note_page(m, st, __pgprot(prot), 3);
413 			} else if (!kasan_page_table(m, st, pud_start)) {
414 				walk_pmd_level(m, st, *start,
415 					       P + i * PUD_LEVEL_MULT);
416 			}
417 		} else
418 			note_page(m, st, __pgprot(0), 3);
419 
420 		prev_pud = start;
421 		start++;
422 	}
423 }
424 
425 #else
426 #define walk_pud_level(m,s,a,p) walk_pmd_level(m,s,__pud(p4d_val(a)),p)
427 #define p4d_large(a) pud_large(__pud(p4d_val(a)))
428 #define p4d_none(a)  pud_none(__pud(p4d_val(a)))
429 #endif
430 
431 #if PTRS_PER_P4D > 1
432 
433 static void walk_p4d_level(struct seq_file *m, struct pg_state *st, pgd_t addr, unsigned long P)
434 {
435 	int i;
436 	p4d_t *start, *p4d_start;
437 	pgprotval_t prot;
438 
439 	p4d_start = start = (p4d_t *)pgd_page_vaddr(addr);
440 
441 	for (i = 0; i < PTRS_PER_P4D; i++) {
442 		st->current_address = normalize_addr(P + i * P4D_LEVEL_MULT);
443 		if (!p4d_none(*start)) {
444 			if (p4d_large(*start) || !p4d_present(*start)) {
445 				prot = p4d_flags(*start);
446 				note_page(m, st, __pgprot(prot), 2);
447 			} else if (!kasan_page_table(m, st, p4d_start)) {
448 				walk_pud_level(m, st, *start,
449 					       P + i * P4D_LEVEL_MULT);
450 			}
451 		} else
452 			note_page(m, st, __pgprot(0), 2);
453 
454 		start++;
455 	}
456 }
457 
458 #else
459 #define walk_p4d_level(m,s,a,p) walk_pud_level(m,s,__p4d(pgd_val(a)),p)
460 #define pgd_large(a) p4d_large(__p4d(pgd_val(a)))
461 #define pgd_none(a)  p4d_none(__p4d(pgd_val(a)))
462 #endif
463 
464 static inline bool is_hypervisor_range(int idx)
465 {
466 #ifdef CONFIG_X86_64
467 	/*
468 	 * ffff800000000000 - ffff87ffffffffff is reserved for
469 	 * the hypervisor.
470 	 */
471 	return	(idx >= pgd_index(__PAGE_OFFSET) - 16) &&
472 		(idx <  pgd_index(__PAGE_OFFSET));
473 #else
474 	return false;
475 #endif
476 }
477 
478 static void ptdump_walk_pgd_level_core(struct seq_file *m, pgd_t *pgd,
479 				       bool checkwx, bool dmesg)
480 {
481 #ifdef CONFIG_X86_64
482 	pgd_t *start = (pgd_t *) &init_top_pgt;
483 #else
484 	pgd_t *start = swapper_pg_dir;
485 #endif
486 	pgprotval_t prot;
487 	int i;
488 	struct pg_state st = {};
489 
490 	if (pgd) {
491 		start = pgd;
492 		st.to_dmesg = dmesg;
493 	}
494 
495 	st.check_wx = checkwx;
496 	if (checkwx)
497 		st.wx_pages = 0;
498 
499 	for (i = 0; i < PTRS_PER_PGD; i++) {
500 		st.current_address = normalize_addr(i * PGD_LEVEL_MULT);
501 		if (!pgd_none(*start) && !is_hypervisor_range(i)) {
502 			if (pgd_large(*start) || !pgd_present(*start)) {
503 				prot = pgd_flags(*start);
504 				note_page(m, &st, __pgprot(prot), 1);
505 			} else {
506 				walk_p4d_level(m, &st, *start,
507 					       i * PGD_LEVEL_MULT);
508 			}
509 		} else
510 			note_page(m, &st, __pgprot(0), 1);
511 
512 		cond_resched();
513 		start++;
514 	}
515 
516 	/* Flush out the last page */
517 	st.current_address = normalize_addr(PTRS_PER_PGD*PGD_LEVEL_MULT);
518 	note_page(m, &st, __pgprot(0), 0);
519 	if (!checkwx)
520 		return;
521 	if (st.wx_pages)
522 		pr_info("x86/mm: Checked W+X mappings: FAILED, %lu W+X pages found.\n",
523 			st.wx_pages);
524 	else
525 		pr_info("x86/mm: Checked W+X mappings: passed, no W+X pages found.\n");
526 }
527 
528 void ptdump_walk_pgd_level(struct seq_file *m, pgd_t *pgd)
529 {
530 	ptdump_walk_pgd_level_core(m, pgd, false, true);
531 }
532 
533 void ptdump_walk_pgd_level_debugfs(struct seq_file *m, pgd_t *pgd, bool user)
534 {
535 #ifdef CONFIG_PAGE_TABLE_ISOLATION
536 	if (user && static_cpu_has(X86_FEATURE_PTI))
537 		pgd = kernel_to_user_pgdp(pgd);
538 #endif
539 	ptdump_walk_pgd_level_core(m, pgd, false, false);
540 }
541 EXPORT_SYMBOL_GPL(ptdump_walk_pgd_level_debugfs);
542 
543 static void ptdump_walk_user_pgd_level_checkwx(void)
544 {
545 #ifdef CONFIG_PAGE_TABLE_ISOLATION
546 	pgd_t *pgd = (pgd_t *) &init_top_pgt;
547 
548 	if (!static_cpu_has(X86_FEATURE_PTI))
549 		return;
550 
551 	pr_info("x86/mm: Checking user space page tables\n");
552 	pgd = kernel_to_user_pgdp(pgd);
553 	ptdump_walk_pgd_level_core(NULL, pgd, true, false);
554 #endif
555 }
556 
557 void ptdump_walk_pgd_level_checkwx(void)
558 {
559 	ptdump_walk_pgd_level_core(NULL, NULL, true, false);
560 	ptdump_walk_user_pgd_level_checkwx();
561 }
562 
563 static int __init pt_dump_init(void)
564 {
565 	/*
566 	 * Various markers are not compile-time constants, so assign them
567 	 * here.
568 	 */
569 #ifdef CONFIG_X86_64
570 	address_markers[LOW_KERNEL_NR].start_address = PAGE_OFFSET;
571 	address_markers[VMALLOC_START_NR].start_address = VMALLOC_START;
572 	address_markers[VMEMMAP_START_NR].start_address = VMEMMAP_START;
573 #endif
574 #ifdef CONFIG_X86_32
575 	address_markers[VMALLOC_START_NR].start_address = VMALLOC_START;
576 	address_markers[VMALLOC_END_NR].start_address = VMALLOC_END;
577 # ifdef CONFIG_HIGHMEM
578 	address_markers[PKMAP_BASE_NR].start_address = PKMAP_BASE;
579 # endif
580 	address_markers[FIXADDR_START_NR].start_address = FIXADDR_START;
581 	address_markers[CPU_ENTRY_AREA_NR].start_address = CPU_ENTRY_AREA_BASE;
582 #endif
583 	return 0;
584 }
585 __initcall(pt_dump_init);
586