xref: /openbmc/linux/arch/s390/mm/dump_pagetables.c (revision 7af6fbdd)
1 // SPDX-License-Identifier: GPL-2.0
2 #include <linux/seq_file.h>
3 #include <linux/debugfs.h>
4 #include <linux/sched.h>
5 #include <linux/mm.h>
6 #include <linux/kasan.h>
7 #include <asm/kasan.h>
8 #include <asm/sections.h>
9 
10 static unsigned long max_addr;
11 
12 struct addr_marker {
13 	unsigned long start_address;
14 	const char *name;
15 };
16 
17 enum address_markers_idx {
18 	IDENTITY_NR = 0,
19 	KERNEL_START_NR,
20 	KERNEL_END_NR,
21 #ifdef CONFIG_KASAN
22 	KASAN_SHADOW_START_NR,
23 	KASAN_SHADOW_END_NR,
24 #endif
25 	VMEMMAP_NR,
26 	VMALLOC_NR,
27 	MODULES_NR,
28 };
29 
30 static struct addr_marker address_markers[] = {
31 	[IDENTITY_NR]		= {0, "Identity Mapping"},
32 	[KERNEL_START_NR]	= {(unsigned long)_stext, "Kernel Image Start"},
33 	[KERNEL_END_NR]		= {(unsigned long)_end, "Kernel Image End"},
34 #ifdef CONFIG_KASAN
35 	[KASAN_SHADOW_START_NR]	= {KASAN_SHADOW_START, "Kasan Shadow Start"},
36 	[KASAN_SHADOW_END_NR]	= {KASAN_SHADOW_END, "Kasan Shadow End"},
37 #endif
38 	[VMEMMAP_NR]		= {0, "vmemmap Area"},
39 	[VMALLOC_NR]		= {0, "vmalloc Area"},
40 	[MODULES_NR]		= {0, "Modules Area"},
41 	{ -1, NULL }
42 };
43 
44 struct pg_state {
45 	int level;
46 	unsigned int current_prot;
47 	unsigned long start_address;
48 	unsigned long current_address;
49 	const struct addr_marker *marker;
50 };
51 
52 static void print_prot(struct seq_file *m, unsigned int pr, int level)
53 {
54 	static const char * const level_name[] =
55 		{ "ASCE", "PGD", "PUD", "PMD", "PTE" };
56 
57 	seq_printf(m, "%s ", level_name[level]);
58 	if (pr & _PAGE_INVALID) {
59 		seq_printf(m, "I\n");
60 		return;
61 	}
62 	seq_puts(m, (pr & _PAGE_PROTECT) ? "RO " : "RW ");
63 	seq_puts(m, (pr & _PAGE_NOEXEC) ? "NX\n" : "X\n");
64 }
65 
66 static void note_page(struct seq_file *m, struct pg_state *st,
67 		     unsigned int new_prot, int level)
68 {
69 	static const char units[] = "KMGTPE";
70 	int width = sizeof(unsigned long) * 2;
71 	const char *unit = units;
72 	unsigned int prot, cur;
73 	unsigned long delta;
74 
75 	/*
76 	 * If we have a "break" in the series, we need to flush the state
77 	 * that we have now. "break" is either changing perms, levels or
78 	 * address space marker.
79 	 */
80 	prot = new_prot;
81 	cur = st->current_prot;
82 
83 	if (!st->level) {
84 		/* First entry */
85 		st->current_prot = new_prot;
86 		st->level = level;
87 		st->marker = address_markers;
88 		seq_printf(m, "---[ %s ]---\n", st->marker->name);
89 	} else if (prot != cur || level != st->level ||
90 		   st->current_address >= st->marker[1].start_address) {
91 		/* Print the actual finished series */
92 		seq_printf(m, "0x%0*lx-0x%0*lx ",
93 			   width, st->start_address,
94 			   width, st->current_address);
95 		delta = (st->current_address - st->start_address) >> 10;
96 		while (!(delta & 0x3ff) && unit[1]) {
97 			delta >>= 10;
98 			unit++;
99 		}
100 		seq_printf(m, "%9lu%c ", delta, *unit);
101 		print_prot(m, st->current_prot, st->level);
102 		while (st->current_address >= st->marker[1].start_address) {
103 			st->marker++;
104 			seq_printf(m, "---[ %s ]---\n", st->marker->name);
105 		}
106 		st->start_address = st->current_address;
107 		st->current_prot = new_prot;
108 		st->level = level;
109 	}
110 }
111 
112 #ifdef CONFIG_KASAN
113 static void note_kasan_early_shadow_page(struct seq_file *m,
114 						struct pg_state *st)
115 {
116 	unsigned int prot;
117 
118 	prot = pte_val(*kasan_early_shadow_pte) &
119 		(_PAGE_PROTECT | _PAGE_INVALID | _PAGE_NOEXEC);
120 	note_page(m, st, prot, 4);
121 }
122 #endif
123 
124 /*
125  * The actual page table walker functions. In order to keep the
126  * implementation of print_prot() short, we only check and pass
127  * _PAGE_INVALID and _PAGE_PROTECT flags to note_page() if a region,
128  * segment or page table entry is invalid or read-only.
129  * After all it's just a hint that the current level being walked
130  * contains an invalid or read-only entry.
131  */
132 static void walk_pte_level(struct seq_file *m, struct pg_state *st,
133 			   pmd_t *pmd, unsigned long addr)
134 {
135 	unsigned int prot;
136 	pte_t *pte;
137 	int i;
138 
139 	for (i = 0; i < PTRS_PER_PTE && addr < max_addr; i++) {
140 		st->current_address = addr;
141 		pte = pte_offset_kernel(pmd, addr);
142 		prot = pte_val(*pte) &
143 			(_PAGE_PROTECT | _PAGE_INVALID | _PAGE_NOEXEC);
144 		note_page(m, st, prot, 4);
145 		addr += PAGE_SIZE;
146 	}
147 }
148 
149 static void walk_pmd_level(struct seq_file *m, struct pg_state *st,
150 			   pud_t *pud, unsigned long addr)
151 {
152 	unsigned int prot;
153 	pmd_t *pmd;
154 	int i;
155 
156 #ifdef CONFIG_KASAN
157 	if ((pud_val(*pud) & PAGE_MASK) == __pa(kasan_early_shadow_pmd)) {
158 		note_kasan_early_shadow_page(m, st);
159 		return;
160 	}
161 #endif
162 
163 	pmd = pmd_offset(pud, addr);
164 	for (i = 0; i < PTRS_PER_PMD && addr < max_addr; i++, pmd++) {
165 		st->current_address = addr;
166 		if (!pmd_none(*pmd)) {
167 			if (pmd_large(*pmd)) {
168 				prot = pmd_val(*pmd) &
169 					(_SEGMENT_ENTRY_PROTECT |
170 					 _SEGMENT_ENTRY_NOEXEC);
171 				note_page(m, st, prot, 3);
172 			} else
173 				walk_pte_level(m, st, pmd, addr);
174 		} else
175 			note_page(m, st, _PAGE_INVALID, 3);
176 		addr += PMD_SIZE;
177 	}
178 }
179 
180 static void walk_pud_level(struct seq_file *m, struct pg_state *st,
181 			   p4d_t *p4d, unsigned long addr)
182 {
183 	unsigned int prot;
184 	pud_t *pud;
185 	int i;
186 
187 #ifdef CONFIG_KASAN
188 	if ((p4d_val(*p4d) & PAGE_MASK) == __pa(kasan_early_shadow_pud)) {
189 		note_kasan_early_shadow_page(m, st);
190 		return;
191 	}
192 #endif
193 
194 	pud = pud_offset(p4d, addr);
195 	for (i = 0; i < PTRS_PER_PUD && addr < max_addr; i++, pud++) {
196 		st->current_address = addr;
197 		if (!pud_none(*pud))
198 			if (pud_large(*pud)) {
199 				prot = pud_val(*pud) &
200 					(_REGION_ENTRY_PROTECT |
201 					 _REGION_ENTRY_NOEXEC);
202 				note_page(m, st, prot, 2);
203 			} else
204 				walk_pmd_level(m, st, pud, addr);
205 		else
206 			note_page(m, st, _PAGE_INVALID, 2);
207 		addr += PUD_SIZE;
208 	}
209 }
210 
211 static void walk_p4d_level(struct seq_file *m, struct pg_state *st,
212 			   pgd_t *pgd, unsigned long addr)
213 {
214 	p4d_t *p4d;
215 	int i;
216 
217 #ifdef CONFIG_KASAN
218 	if ((pgd_val(*pgd) & PAGE_MASK) == __pa(kasan_early_shadow_p4d)) {
219 		note_kasan_early_shadow_page(m, st);
220 		return;
221 	}
222 #endif
223 
224 	p4d = p4d_offset(pgd, addr);
225 	for (i = 0; i < PTRS_PER_P4D && addr < max_addr; i++, p4d++) {
226 		st->current_address = addr;
227 		if (!p4d_none(*p4d))
228 			walk_pud_level(m, st, p4d, addr);
229 		else
230 			note_page(m, st, _PAGE_INVALID, 2);
231 		addr += P4D_SIZE;
232 	}
233 }
234 
235 static void walk_pgd_level(struct seq_file *m)
236 {
237 	unsigned long addr = 0;
238 	struct pg_state st;
239 	pgd_t *pgd;
240 	int i;
241 
242 	memset(&st, 0, sizeof(st));
243 	for (i = 0; i < PTRS_PER_PGD && addr < max_addr; i++) {
244 		st.current_address = addr;
245 		pgd = pgd_offset_k(addr);
246 		if (!pgd_none(*pgd))
247 			walk_p4d_level(m, &st, pgd, addr);
248 		else
249 			note_page(m, &st, _PAGE_INVALID, 1);
250 		addr += PGDIR_SIZE;
251 		cond_resched();
252 	}
253 	/* Flush out the last page */
254 	st.current_address = max_addr;
255 	note_page(m, &st, 0, 0);
256 }
257 
258 static int ptdump_show(struct seq_file *m, void *v)
259 {
260 	walk_pgd_level(m);
261 	return 0;
262 }
263 
264 static int ptdump_open(struct inode *inode, struct file *filp)
265 {
266 	return single_open(filp, ptdump_show, NULL);
267 }
268 
269 static const struct file_operations ptdump_fops = {
270 	.open		= ptdump_open,
271 	.read		= seq_read,
272 	.llseek		= seq_lseek,
273 	.release	= single_release,
274 };
275 
276 static int pt_dump_init(void)
277 {
278 	/*
279 	 * Figure out the maximum virtual address being accessible with the
280 	 * kernel ASCE. We need this to keep the page table walker functions
281 	 * from accessing non-existent entries.
282 	 */
283 	max_addr = (S390_lowcore.kernel_asce & _REGION_ENTRY_TYPE_MASK) >> 2;
284 	max_addr = 1UL << (max_addr * 11 + 31);
285 	address_markers[MODULES_NR].start_address = MODULES_VADDR;
286 	address_markers[VMEMMAP_NR].start_address = (unsigned long) vmemmap;
287 	address_markers[VMALLOC_NR].start_address = VMALLOC_START;
288 	debugfs_create_file("kernel_page_tables", 0400, NULL, NULL, &ptdump_fops);
289 	return 0;
290 }
291 device_initcall(pt_dump_init);
292