xref: /openbmc/linux/arch/x86/mm/dump_pagetables.c (revision b34e08d5)
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 	bool to_dmesg;
34 };
35 
36 struct addr_marker {
37 	unsigned long start_address;
38 	const char *name;
39 };
40 
41 /* indices for address_markers; keep sync'd w/ address_markers below */
42 enum address_markers_idx {
43 	USER_SPACE_NR = 0,
44 #ifdef CONFIG_X86_64
45 	KERNEL_SPACE_NR,
46 	LOW_KERNEL_NR,
47 	VMALLOC_START_NR,
48 	VMEMMAP_START_NR,
49 	HIGH_KERNEL_NR,
50 	MODULES_VADDR_NR,
51 	MODULES_END_NR,
52 #else
53 	KERNEL_SPACE_NR,
54 	VMALLOC_START_NR,
55 	VMALLOC_END_NR,
56 # ifdef CONFIG_HIGHMEM
57 	PKMAP_BASE_NR,
58 # endif
59 	FIXADDR_START_NR,
60 #endif
61 };
62 
63 /* Address space markers hints */
64 static struct addr_marker address_markers[] = {
65 	{ 0, "User Space" },
66 #ifdef CONFIG_X86_64
67 	{ 0x8000000000000000UL, "Kernel Space" },
68 	{ PAGE_OFFSET,		"Low Kernel Mapping" },
69 	{ VMALLOC_START,        "vmalloc() Area" },
70 	{ VMEMMAP_START,        "Vmemmap" },
71 	{ __START_KERNEL_map,   "High Kernel Mapping" },
72 	{ MODULES_VADDR,        "Modules" },
73 	{ MODULES_END,          "End Modules" },
74 #else
75 	{ PAGE_OFFSET,          "Kernel Mapping" },
76 	{ 0/* VMALLOC_START */, "vmalloc() Area" },
77 	{ 0/*VMALLOC_END*/,     "vmalloc() End" },
78 # ifdef CONFIG_HIGHMEM
79 	{ 0/*PKMAP_BASE*/,      "Persisent kmap() Area" },
80 # endif
81 	{ 0/*FIXADDR_START*/,   "Fixmap Area" },
82 #endif
83 	{ -1, NULL }		/* End of list */
84 };
85 
86 /* Multipliers for offsets within the PTEs */
87 #define PTE_LEVEL_MULT (PAGE_SIZE)
88 #define PMD_LEVEL_MULT (PTRS_PER_PTE * PTE_LEVEL_MULT)
89 #define PUD_LEVEL_MULT (PTRS_PER_PMD * PMD_LEVEL_MULT)
90 #define PGD_LEVEL_MULT (PTRS_PER_PUD * PUD_LEVEL_MULT)
91 
92 #define pt_dump_seq_printf(m, to_dmesg, fmt, args...)		\
93 ({								\
94 	if (to_dmesg)					\
95 		printk(KERN_INFO fmt, ##args);			\
96 	else							\
97 		if (m)						\
98 			seq_printf(m, fmt, ##args);		\
99 })
100 
101 #define pt_dump_cont_printf(m, to_dmesg, fmt, args...)		\
102 ({								\
103 	if (to_dmesg)					\
104 		printk(KERN_CONT fmt, ##args);			\
105 	else							\
106 		if (m)						\
107 			seq_printf(m, fmt, ##args);		\
108 })
109 
110 /*
111  * Print a readable form of a pgprot_t to the seq_file
112  */
113 static void printk_prot(struct seq_file *m, pgprot_t prot, int level, bool dmsg)
114 {
115 	pgprotval_t pr = pgprot_val(prot);
116 	static const char * const level_name[] =
117 		{ "cr3", "pgd", "pud", "pmd", "pte" };
118 
119 	if (!pgprot_val(prot)) {
120 		/* Not present */
121 		pt_dump_cont_printf(m, dmsg, "                          ");
122 	} else {
123 		if (pr & _PAGE_USER)
124 			pt_dump_cont_printf(m, dmsg, "USR ");
125 		else
126 			pt_dump_cont_printf(m, dmsg, "    ");
127 		if (pr & _PAGE_RW)
128 			pt_dump_cont_printf(m, dmsg, "RW ");
129 		else
130 			pt_dump_cont_printf(m, dmsg, "ro ");
131 		if (pr & _PAGE_PWT)
132 			pt_dump_cont_printf(m, dmsg, "PWT ");
133 		else
134 			pt_dump_cont_printf(m, dmsg, "    ");
135 		if (pr & _PAGE_PCD)
136 			pt_dump_cont_printf(m, dmsg, "PCD ");
137 		else
138 			pt_dump_cont_printf(m, dmsg, "    ");
139 
140 		/* Bit 9 has a different meaning on level 3 vs 4 */
141 		if (level <= 3) {
142 			if (pr & _PAGE_PSE)
143 				pt_dump_cont_printf(m, dmsg, "PSE ");
144 			else
145 				pt_dump_cont_printf(m, dmsg, "    ");
146 		} else {
147 			if (pr & _PAGE_PAT)
148 				pt_dump_cont_printf(m, dmsg, "pat ");
149 			else
150 				pt_dump_cont_printf(m, dmsg, "    ");
151 		}
152 		if (pr & _PAGE_GLOBAL)
153 			pt_dump_cont_printf(m, dmsg, "GLB ");
154 		else
155 			pt_dump_cont_printf(m, dmsg, "    ");
156 		if (pr & _PAGE_NX)
157 			pt_dump_cont_printf(m, dmsg, "NX ");
158 		else
159 			pt_dump_cont_printf(m, dmsg, "x  ");
160 	}
161 	pt_dump_cont_printf(m, dmsg, "%s\n", level_name[level]);
162 }
163 
164 /*
165  * On 64 bits, sign-extend the 48 bit address to 64 bit
166  */
167 static unsigned long normalize_addr(unsigned long u)
168 {
169 #ifdef CONFIG_X86_64
170 	return (signed long)(u << 16) >> 16;
171 #else
172 	return u;
173 #endif
174 }
175 
176 /*
177  * This function gets called on a break in a continuous series
178  * of PTE entries; the next one is different so we need to
179  * print what we collected so far.
180  */
181 static void note_page(struct seq_file *m, struct pg_state *st,
182 		      pgprot_t new_prot, int level)
183 {
184 	pgprotval_t prot, cur;
185 	static const char units[] = "KMGTPE";
186 
187 	/*
188 	 * If we have a "break" in the series, we need to flush the state that
189 	 * we have now. "break" is either changing perms, levels or
190 	 * address space marker.
191 	 */
192 	prot = pgprot_val(new_prot) & PTE_FLAGS_MASK;
193 	cur = pgprot_val(st->current_prot) & PTE_FLAGS_MASK;
194 
195 	if (!st->level) {
196 		/* First entry */
197 		st->current_prot = new_prot;
198 		st->level = level;
199 		st->marker = address_markers;
200 		pt_dump_seq_printf(m, st->to_dmesg, "---[ %s ]---\n",
201 				   st->marker->name);
202 	} else if (prot != cur || level != st->level ||
203 		   st->current_address >= st->marker[1].start_address) {
204 		const char *unit = units;
205 		unsigned long delta;
206 		int width = sizeof(unsigned long) * 2;
207 
208 		/*
209 		 * Now print the actual finished series
210 		 */
211 		pt_dump_seq_printf(m, st->to_dmesg,  "0x%0*lx-0x%0*lx   ",
212 				   width, st->start_address,
213 				   width, st->current_address);
214 
215 		delta = (st->current_address - st->start_address) >> 10;
216 		while (!(delta & 1023) && unit[1]) {
217 			delta >>= 10;
218 			unit++;
219 		}
220 		pt_dump_cont_printf(m, st->to_dmesg, "%9lu%c ", delta, *unit);
221 		printk_prot(m, st->current_prot, st->level, st->to_dmesg);
222 
223 		/*
224 		 * We print markers for special areas of address space,
225 		 * such as the start of vmalloc space etc.
226 		 * This helps in the interpretation.
227 		 */
228 		if (st->current_address >= st->marker[1].start_address) {
229 			st->marker++;
230 			pt_dump_seq_printf(m, st->to_dmesg, "---[ %s ]---\n",
231 					   st->marker->name);
232 		}
233 
234 		st->start_address = st->current_address;
235 		st->current_prot = new_prot;
236 		st->level = level;
237 	}
238 }
239 
240 static void walk_pte_level(struct seq_file *m, struct pg_state *st, pmd_t addr,
241 							unsigned long P)
242 {
243 	int i;
244 	pte_t *start;
245 
246 	start = (pte_t *) pmd_page_vaddr(addr);
247 	for (i = 0; i < PTRS_PER_PTE; i++) {
248 		pgprot_t prot = pte_pgprot(*start);
249 
250 		st->current_address = normalize_addr(P + i * PTE_LEVEL_MULT);
251 		note_page(m, st, prot, 4);
252 		start++;
253 	}
254 }
255 
256 #if PTRS_PER_PMD > 1
257 
258 static void walk_pmd_level(struct seq_file *m, struct pg_state *st, pud_t addr,
259 							unsigned long P)
260 {
261 	int i;
262 	pmd_t *start;
263 
264 	start = (pmd_t *) pud_page_vaddr(addr);
265 	for (i = 0; i < PTRS_PER_PMD; i++) {
266 		st->current_address = normalize_addr(P + i * PMD_LEVEL_MULT);
267 		if (!pmd_none(*start)) {
268 			pgprotval_t prot = pmd_val(*start) & PTE_FLAGS_MASK;
269 
270 			if (pmd_large(*start) || !pmd_present(*start))
271 				note_page(m, st, __pgprot(prot), 3);
272 			else
273 				walk_pte_level(m, st, *start,
274 					       P + i * PMD_LEVEL_MULT);
275 		} else
276 			note_page(m, st, __pgprot(0), 3);
277 		start++;
278 	}
279 }
280 
281 #else
282 #define walk_pmd_level(m,s,a,p) walk_pte_level(m,s,__pmd(pud_val(a)),p)
283 #define pud_large(a) pmd_large(__pmd(pud_val(a)))
284 #define pud_none(a)  pmd_none(__pmd(pud_val(a)))
285 #endif
286 
287 #if PTRS_PER_PUD > 1
288 
289 static void walk_pud_level(struct seq_file *m, struct pg_state *st, pgd_t addr,
290 							unsigned long P)
291 {
292 	int i;
293 	pud_t *start;
294 
295 	start = (pud_t *) pgd_page_vaddr(addr);
296 
297 	for (i = 0; i < PTRS_PER_PUD; i++) {
298 		st->current_address = normalize_addr(P + i * PUD_LEVEL_MULT);
299 		if (!pud_none(*start)) {
300 			pgprotval_t prot = pud_val(*start) & PTE_FLAGS_MASK;
301 
302 			if (pud_large(*start) || !pud_present(*start))
303 				note_page(m, st, __pgprot(prot), 2);
304 			else
305 				walk_pmd_level(m, st, *start,
306 					       P + i * PUD_LEVEL_MULT);
307 		} else
308 			note_page(m, st, __pgprot(0), 2);
309 
310 		start++;
311 	}
312 }
313 
314 #else
315 #define walk_pud_level(m,s,a,p) walk_pmd_level(m,s,__pud(pgd_val(a)),p)
316 #define pgd_large(a) pud_large(__pud(pgd_val(a)))
317 #define pgd_none(a)  pud_none(__pud(pgd_val(a)))
318 #endif
319 
320 void ptdump_walk_pgd_level(struct seq_file *m, pgd_t *pgd)
321 {
322 #ifdef CONFIG_X86_64
323 	pgd_t *start = (pgd_t *) &init_level4_pgt;
324 #else
325 	pgd_t *start = swapper_pg_dir;
326 #endif
327 	int i;
328 	struct pg_state st = {};
329 
330 	if (pgd) {
331 		start = pgd;
332 		st.to_dmesg = true;
333 	}
334 
335 	for (i = 0; i < PTRS_PER_PGD; i++) {
336 		st.current_address = normalize_addr(i * PGD_LEVEL_MULT);
337 		if (!pgd_none(*start)) {
338 			pgprotval_t prot = pgd_val(*start) & PTE_FLAGS_MASK;
339 
340 			if (pgd_large(*start) || !pgd_present(*start))
341 				note_page(m, &st, __pgprot(prot), 1);
342 			else
343 				walk_pud_level(m, &st, *start,
344 					       i * PGD_LEVEL_MULT);
345 		} else
346 			note_page(m, &st, __pgprot(0), 1);
347 
348 		start++;
349 	}
350 
351 	/* Flush out the last page */
352 	st.current_address = normalize_addr(PTRS_PER_PGD*PGD_LEVEL_MULT);
353 	note_page(m, &st, __pgprot(0), 0);
354 }
355 
356 static int ptdump_show(struct seq_file *m, void *v)
357 {
358 	ptdump_walk_pgd_level(m, NULL);
359 	return 0;
360 }
361 
362 static int ptdump_open(struct inode *inode, struct file *filp)
363 {
364 	return single_open(filp, ptdump_show, NULL);
365 }
366 
367 static const struct file_operations ptdump_fops = {
368 	.open		= ptdump_open,
369 	.read		= seq_read,
370 	.llseek		= seq_lseek,
371 	.release	= single_release,
372 };
373 
374 static int pt_dump_init(void)
375 {
376 	struct dentry *pe;
377 
378 #ifdef CONFIG_X86_32
379 	/* Not a compile-time constant on x86-32 */
380 	address_markers[VMALLOC_START_NR].start_address = VMALLOC_START;
381 	address_markers[VMALLOC_END_NR].start_address = VMALLOC_END;
382 # ifdef CONFIG_HIGHMEM
383 	address_markers[PKMAP_BASE_NR].start_address = PKMAP_BASE;
384 # endif
385 	address_markers[FIXADDR_START_NR].start_address = FIXADDR_START;
386 #endif
387 
388 	pe = debugfs_create_file("kernel_page_tables", 0600, NULL, NULL,
389 				 &ptdump_fops);
390 	if (!pe)
391 		return -ENOMEM;
392 
393 	return 0;
394 }
395 
396 __initcall(pt_dump_init);
397 MODULE_LICENSE("GPL");
398 MODULE_AUTHOR("Arjan van de Ven <arjan@linux.intel.com>");
399 MODULE_DESCRIPTION("Kernel debugging helper that dumps pagetables");
400