xref: /openbmc/linux/arch/x86/mm/dump_pagetables.c (revision 9cfc5c90)
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 	bool to_dmesg;
35 	bool check_wx;
36 	unsigned long wx_pages;
37 };
38 
39 struct addr_marker {
40 	unsigned long start_address;
41 	const char *name;
42 	unsigned long max_lines;
43 };
44 
45 /* indices for address_markers; keep sync'd w/ address_markers below */
46 enum address_markers_idx {
47 	USER_SPACE_NR = 0,
48 #ifdef CONFIG_X86_64
49 	KERNEL_SPACE_NR,
50 	LOW_KERNEL_NR,
51 	VMALLOC_START_NR,
52 	VMEMMAP_START_NR,
53 # ifdef CONFIG_X86_ESPFIX64
54 	ESPFIX_START_NR,
55 # endif
56 	HIGH_KERNEL_NR,
57 	MODULES_VADDR_NR,
58 	MODULES_END_NR,
59 #else
60 	KERNEL_SPACE_NR,
61 	VMALLOC_START_NR,
62 	VMALLOC_END_NR,
63 # ifdef CONFIG_HIGHMEM
64 	PKMAP_BASE_NR,
65 # endif
66 	FIXADDR_START_NR,
67 #endif
68 };
69 
70 /* Address space markers hints */
71 static struct addr_marker address_markers[] = {
72 	{ 0, "User Space" },
73 #ifdef CONFIG_X86_64
74 	{ 0x8000000000000000UL, "Kernel Space" },
75 	{ PAGE_OFFSET,		"Low Kernel Mapping" },
76 	{ VMALLOC_START,        "vmalloc() Area" },
77 	{ VMEMMAP_START,        "Vmemmap" },
78 # ifdef CONFIG_X86_ESPFIX64
79 	{ ESPFIX_BASE_ADDR,	"ESPfix Area", 16 },
80 # endif
81 # ifdef CONFIG_EFI
82 	{ EFI_VA_END,		"EFI Runtime Services" },
83 # endif
84 	{ __START_KERNEL_map,   "High Kernel Mapping" },
85 	{ MODULES_VADDR,        "Modules" },
86 	{ MODULES_END,          "End Modules" },
87 #else
88 	{ PAGE_OFFSET,          "Kernel Mapping" },
89 	{ 0/* VMALLOC_START */, "vmalloc() Area" },
90 	{ 0/*VMALLOC_END*/,     "vmalloc() End" },
91 # ifdef CONFIG_HIGHMEM
92 	{ 0/*PKMAP_BASE*/,      "Persisent kmap() Area" },
93 # endif
94 	{ 0/*FIXADDR_START*/,   "Fixmap Area" },
95 #endif
96 	{ -1, NULL }		/* End of list */
97 };
98 
99 /* Multipliers for offsets within the PTEs */
100 #define PTE_LEVEL_MULT (PAGE_SIZE)
101 #define PMD_LEVEL_MULT (PTRS_PER_PTE * PTE_LEVEL_MULT)
102 #define PUD_LEVEL_MULT (PTRS_PER_PMD * PMD_LEVEL_MULT)
103 #define PGD_LEVEL_MULT (PTRS_PER_PUD * PUD_LEVEL_MULT)
104 
105 #define pt_dump_seq_printf(m, to_dmesg, fmt, args...)		\
106 ({								\
107 	if (to_dmesg)					\
108 		printk(KERN_INFO fmt, ##args);			\
109 	else							\
110 		if (m)						\
111 			seq_printf(m, fmt, ##args);		\
112 })
113 
114 #define pt_dump_cont_printf(m, to_dmesg, fmt, args...)		\
115 ({								\
116 	if (to_dmesg)					\
117 		printk(KERN_CONT fmt, ##args);			\
118 	else							\
119 		if (m)						\
120 			seq_printf(m, fmt, ##args);		\
121 })
122 
123 /*
124  * Print a readable form of a pgprot_t to the seq_file
125  */
126 static void printk_prot(struct seq_file *m, pgprot_t prot, int level, bool dmsg)
127 {
128 	pgprotval_t pr = pgprot_val(prot);
129 	static const char * const level_name[] =
130 		{ "cr3", "pgd", "pud", "pmd", "pte" };
131 
132 	if (!pgprot_val(prot)) {
133 		/* Not present */
134 		pt_dump_cont_printf(m, dmsg, "                              ");
135 	} else {
136 		if (pr & _PAGE_USER)
137 			pt_dump_cont_printf(m, dmsg, "USR ");
138 		else
139 			pt_dump_cont_printf(m, dmsg, "    ");
140 		if (pr & _PAGE_RW)
141 			pt_dump_cont_printf(m, dmsg, "RW ");
142 		else
143 			pt_dump_cont_printf(m, dmsg, "ro ");
144 		if (pr & _PAGE_PWT)
145 			pt_dump_cont_printf(m, dmsg, "PWT ");
146 		else
147 			pt_dump_cont_printf(m, dmsg, "    ");
148 		if (pr & _PAGE_PCD)
149 			pt_dump_cont_printf(m, dmsg, "PCD ");
150 		else
151 			pt_dump_cont_printf(m, dmsg, "    ");
152 
153 		/* Bit 7 has a different meaning on level 3 vs 4 */
154 		if (level <= 3 && pr & _PAGE_PSE)
155 			pt_dump_cont_printf(m, dmsg, "PSE ");
156 		else
157 			pt_dump_cont_printf(m, dmsg, "    ");
158 		if ((level == 4 && pr & _PAGE_PAT) ||
159 		    ((level == 3 || level == 2) && pr & _PAGE_PAT_LARGE))
160 			pt_dump_cont_printf(m, dmsg, "PAT ");
161 		else
162 			pt_dump_cont_printf(m, dmsg, "    ");
163 		if (pr & _PAGE_GLOBAL)
164 			pt_dump_cont_printf(m, dmsg, "GLB ");
165 		else
166 			pt_dump_cont_printf(m, dmsg, "    ");
167 		if (pr & _PAGE_NX)
168 			pt_dump_cont_printf(m, dmsg, "NX ");
169 		else
170 			pt_dump_cont_printf(m, dmsg, "x  ");
171 	}
172 	pt_dump_cont_printf(m, dmsg, "%s\n", level_name[level]);
173 }
174 
175 /*
176  * On 64 bits, sign-extend the 48 bit address to 64 bit
177  */
178 static unsigned long normalize_addr(unsigned long u)
179 {
180 #ifdef CONFIG_X86_64
181 	return (signed long)(u << 16) >> 16;
182 #else
183 	return u;
184 #endif
185 }
186 
187 /*
188  * This function gets called on a break in a continuous series
189  * of PTE entries; the next one is different so we need to
190  * print what we collected so far.
191  */
192 static void note_page(struct seq_file *m, struct pg_state *st,
193 		      pgprot_t new_prot, int level)
194 {
195 	pgprotval_t prot, cur;
196 	static const char units[] = "BKMGTPE";
197 
198 	/*
199 	 * If we have a "break" in the series, we need to flush the state that
200 	 * we have now. "break" is either changing perms, levels or
201 	 * address space marker.
202 	 */
203 	prot = pgprot_val(new_prot);
204 	cur = pgprot_val(st->current_prot);
205 
206 	if (!st->level) {
207 		/* First entry */
208 		st->current_prot = new_prot;
209 		st->level = level;
210 		st->marker = address_markers;
211 		st->lines = 0;
212 		pt_dump_seq_printf(m, st->to_dmesg, "---[ %s ]---\n",
213 				   st->marker->name);
214 	} else if (prot != cur || level != st->level ||
215 		   st->current_address >= st->marker[1].start_address) {
216 		const char *unit = units;
217 		unsigned long delta;
218 		int width = sizeof(unsigned long) * 2;
219 		pgprotval_t pr = pgprot_val(st->current_prot);
220 
221 		if (st->check_wx && (pr & _PAGE_RW) && !(pr & _PAGE_NX)) {
222 			WARN_ONCE(1,
223 				  "x86/mm: Found insecure W+X mapping at address %p/%pS\n",
224 				  (void *)st->start_address,
225 				  (void *)st->start_address);
226 			st->wx_pages += (st->current_address -
227 					 st->start_address) / PAGE_SIZE;
228 		}
229 
230 		/*
231 		 * Now print the actual finished series
232 		 */
233 		if (!st->marker->max_lines ||
234 		    st->lines < st->marker->max_lines) {
235 			pt_dump_seq_printf(m, st->to_dmesg,
236 					   "0x%0*lx-0x%0*lx   ",
237 					   width, st->start_address,
238 					   width, st->current_address);
239 
240 			delta = st->current_address - st->start_address;
241 			while (!(delta & 1023) && unit[1]) {
242 				delta >>= 10;
243 				unit++;
244 			}
245 			pt_dump_cont_printf(m, st->to_dmesg, "%9lu%c ",
246 					    delta, *unit);
247 			printk_prot(m, st->current_prot, st->level,
248 				    st->to_dmesg);
249 		}
250 		st->lines++;
251 
252 		/*
253 		 * We print markers for special areas of address space,
254 		 * such as the start of vmalloc space etc.
255 		 * This helps in the interpretation.
256 		 */
257 		if (st->current_address >= st->marker[1].start_address) {
258 			if (st->marker->max_lines &&
259 			    st->lines > st->marker->max_lines) {
260 				unsigned long nskip =
261 					st->lines - st->marker->max_lines;
262 				pt_dump_seq_printf(m, st->to_dmesg,
263 						   "... %lu entr%s skipped ... \n",
264 						   nskip,
265 						   nskip == 1 ? "y" : "ies");
266 			}
267 			st->marker++;
268 			st->lines = 0;
269 			pt_dump_seq_printf(m, st->to_dmesg, "---[ %s ]---\n",
270 					   st->marker->name);
271 		}
272 
273 		st->start_address = st->current_address;
274 		st->current_prot = new_prot;
275 		st->level = level;
276 	}
277 }
278 
279 static void walk_pte_level(struct seq_file *m, struct pg_state *st, pmd_t addr,
280 							unsigned long P)
281 {
282 	int i;
283 	pte_t *start;
284 	pgprotval_t prot;
285 
286 	start = (pte_t *) pmd_page_vaddr(addr);
287 	for (i = 0; i < PTRS_PER_PTE; i++) {
288 		prot = pte_flags(*start);
289 		st->current_address = normalize_addr(P + i * PTE_LEVEL_MULT);
290 		note_page(m, st, __pgprot(prot), 4);
291 		start++;
292 	}
293 }
294 
295 #if PTRS_PER_PMD > 1
296 
297 static void walk_pmd_level(struct seq_file *m, struct pg_state *st, pud_t addr,
298 							unsigned long P)
299 {
300 	int i;
301 	pmd_t *start;
302 	pgprotval_t prot;
303 
304 	start = (pmd_t *) pud_page_vaddr(addr);
305 	for (i = 0; i < PTRS_PER_PMD; i++) {
306 		st->current_address = normalize_addr(P + i * PMD_LEVEL_MULT);
307 		if (!pmd_none(*start)) {
308 			if (pmd_large(*start) || !pmd_present(*start)) {
309 				prot = pmd_flags(*start);
310 				note_page(m, st, __pgprot(prot), 3);
311 			} else {
312 				walk_pte_level(m, st, *start,
313 					       P + i * PMD_LEVEL_MULT);
314 			}
315 		} else
316 			note_page(m, st, __pgprot(0), 3);
317 		start++;
318 	}
319 }
320 
321 #else
322 #define walk_pmd_level(m,s,a,p) walk_pte_level(m,s,__pmd(pud_val(a)),p)
323 #define pud_large(a) pmd_large(__pmd(pud_val(a)))
324 #define pud_none(a)  pmd_none(__pmd(pud_val(a)))
325 #endif
326 
327 #if PTRS_PER_PUD > 1
328 
329 static void walk_pud_level(struct seq_file *m, struct pg_state *st, pgd_t addr,
330 							unsigned long P)
331 {
332 	int i;
333 	pud_t *start;
334 	pgprotval_t prot;
335 
336 	start = (pud_t *) pgd_page_vaddr(addr);
337 
338 	for (i = 0; i < PTRS_PER_PUD; i++) {
339 		st->current_address = normalize_addr(P + i * PUD_LEVEL_MULT);
340 		if (!pud_none(*start)) {
341 			if (pud_large(*start) || !pud_present(*start)) {
342 				prot = pud_flags(*start);
343 				note_page(m, st, __pgprot(prot), 2);
344 			} else {
345 				walk_pmd_level(m, st, *start,
346 					       P + i * PUD_LEVEL_MULT);
347 			}
348 		} else
349 			note_page(m, st, __pgprot(0), 2);
350 
351 		start++;
352 	}
353 }
354 
355 #else
356 #define walk_pud_level(m,s,a,p) walk_pmd_level(m,s,__pud(pgd_val(a)),p)
357 #define pgd_large(a) pud_large(__pud(pgd_val(a)))
358 #define pgd_none(a)  pud_none(__pud(pgd_val(a)))
359 #endif
360 
361 #ifdef CONFIG_X86_64
362 static inline bool is_hypervisor_range(int idx)
363 {
364 	/*
365 	 * ffff800000000000 - ffff87ffffffffff is reserved for
366 	 * the hypervisor.
367 	 */
368 	return paravirt_enabled() &&
369 		(idx >= pgd_index(__PAGE_OFFSET) - 16) &&
370 		(idx < pgd_index(__PAGE_OFFSET));
371 }
372 #else
373 static inline bool is_hypervisor_range(int idx) { return false; }
374 #endif
375 
376 static void ptdump_walk_pgd_level_core(struct seq_file *m, pgd_t *pgd,
377 				       bool checkwx)
378 {
379 #ifdef CONFIG_X86_64
380 	pgd_t *start = (pgd_t *) &init_level4_pgt;
381 #else
382 	pgd_t *start = swapper_pg_dir;
383 #endif
384 	pgprotval_t prot;
385 	int i;
386 	struct pg_state st = {};
387 
388 	if (pgd) {
389 		start = pgd;
390 		st.to_dmesg = true;
391 	}
392 
393 	st.check_wx = checkwx;
394 	if (checkwx)
395 		st.wx_pages = 0;
396 
397 	for (i = 0; i < PTRS_PER_PGD; i++) {
398 		st.current_address = normalize_addr(i * PGD_LEVEL_MULT);
399 		if (!pgd_none(*start) && !is_hypervisor_range(i)) {
400 			if (pgd_large(*start) || !pgd_present(*start)) {
401 				prot = pgd_flags(*start);
402 				note_page(m, &st, __pgprot(prot), 1);
403 			} else {
404 				walk_pud_level(m, &st, *start,
405 					       i * PGD_LEVEL_MULT);
406 			}
407 		} else
408 			note_page(m, &st, __pgprot(0), 1);
409 
410 		start++;
411 	}
412 
413 	/* Flush out the last page */
414 	st.current_address = normalize_addr(PTRS_PER_PGD*PGD_LEVEL_MULT);
415 	note_page(m, &st, __pgprot(0), 0);
416 	if (!checkwx)
417 		return;
418 	if (st.wx_pages)
419 		pr_info("x86/mm: Checked W+X mappings: FAILED, %lu W+X pages found.\n",
420 			st.wx_pages);
421 	else
422 		pr_info("x86/mm: Checked W+X mappings: passed, no W+X pages found.\n");
423 }
424 
425 void ptdump_walk_pgd_level(struct seq_file *m, pgd_t *pgd)
426 {
427 	ptdump_walk_pgd_level_core(m, pgd, false);
428 }
429 
430 void ptdump_walk_pgd_level_checkwx(void)
431 {
432 	ptdump_walk_pgd_level_core(NULL, NULL, true);
433 }
434 
435 #ifdef CONFIG_X86_PTDUMP
436 static int ptdump_show(struct seq_file *m, void *v)
437 {
438 	ptdump_walk_pgd_level(m, NULL);
439 	return 0;
440 }
441 
442 static int ptdump_open(struct inode *inode, struct file *filp)
443 {
444 	return single_open(filp, ptdump_show, NULL);
445 }
446 
447 static const struct file_operations ptdump_fops = {
448 	.open		= ptdump_open,
449 	.read		= seq_read,
450 	.llseek		= seq_lseek,
451 	.release	= single_release,
452 };
453 #endif
454 
455 static int pt_dump_init(void)
456 {
457 #ifdef CONFIG_X86_PTDUMP
458 	struct dentry *pe;
459 #endif
460 
461 #ifdef CONFIG_X86_32
462 	/* Not a compile-time constant on x86-32 */
463 	address_markers[VMALLOC_START_NR].start_address = VMALLOC_START;
464 	address_markers[VMALLOC_END_NR].start_address = VMALLOC_END;
465 # ifdef CONFIG_HIGHMEM
466 	address_markers[PKMAP_BASE_NR].start_address = PKMAP_BASE;
467 # endif
468 	address_markers[FIXADDR_START_NR].start_address = FIXADDR_START;
469 #endif
470 
471 #ifdef CONFIG_X86_PTDUMP
472 	pe = debugfs_create_file("kernel_page_tables", 0600, NULL, NULL,
473 				 &ptdump_fops);
474 	if (!pe)
475 		return -ENOMEM;
476 #endif
477 
478 	return 0;
479 }
480 
481 __initcall(pt_dump_init);
482 MODULE_LICENSE("GPL");
483 MODULE_AUTHOR("Arjan van de Ven <arjan@linux.intel.com>");
484 MODULE_DESCRIPTION("Kernel debugging helper that dumps pagetables");
485