xref: /openbmc/linux/arch/s390/mm/dump_pagetables.c (revision 9d749629)
1 #include <linux/seq_file.h>
2 #include <linux/debugfs.h>
3 #include <linux/module.h>
4 #include <linux/mm.h>
5 #include <asm/sections.h>
6 #include <asm/pgtable.h>
7 
8 static unsigned long max_addr;
9 
10 struct addr_marker {
11 	unsigned long start_address;
12 	const char *name;
13 };
14 
15 enum address_markers_idx {
16 	IDENTITY_NR = 0,
17 	KERNEL_START_NR,
18 	KERNEL_END_NR,
19 	VMEMMAP_NR,
20 	VMALLOC_NR,
21 #ifdef CONFIG_64BIT
22 	MODULES_NR,
23 #endif
24 };
25 
26 static struct addr_marker address_markers[] = {
27 	[IDENTITY_NR]	  = {0, "Identity Mapping"},
28 	[KERNEL_START_NR] = {(unsigned long)&_stext, "Kernel Image Start"},
29 	[KERNEL_END_NR]	  = {(unsigned long)&_end, "Kernel Image End"},
30 	[VMEMMAP_NR]	  = {0, "vmemmap Area"},
31 	[VMALLOC_NR]	  = {0, "vmalloc Area"},
32 #ifdef CONFIG_64BIT
33 	[MODULES_NR]	  = {0, "Modules Area"},
34 #endif
35 	{ -1, NULL }
36 };
37 
38 struct pg_state {
39 	int level;
40 	unsigned int current_prot;
41 	unsigned long start_address;
42 	unsigned long current_address;
43 	const struct addr_marker *marker;
44 };
45 
46 static void print_prot(struct seq_file *m, unsigned int pr, int level)
47 {
48 	static const char * const level_name[] =
49 		{ "ASCE", "PGD", "PUD", "PMD", "PTE" };
50 
51 	seq_printf(m, "%s ", level_name[level]);
52 	if (pr & _PAGE_INVALID)
53 		seq_printf(m, "I\n");
54 	else
55 		seq_printf(m, "%s\n", pr & _PAGE_RO ? "RO" : "RW");
56 }
57 
58 static void note_page(struct seq_file *m, struct pg_state *st,
59 		     unsigned int new_prot, int level)
60 {
61 	static const char units[] = "KMGTPE";
62 	int width = sizeof(unsigned long) * 2;
63 	const char *unit = units;
64 	unsigned int prot, cur;
65 	unsigned long delta;
66 
67 	/*
68 	 * If we have a "break" in the series, we need to flush the state
69 	 * that we have now. "break" is either changing perms, levels or
70 	 * address space marker.
71 	 */
72 	prot = new_prot;
73 	cur = st->current_prot;
74 
75 	if (!st->level) {
76 		/* First entry */
77 		st->current_prot = new_prot;
78 		st->level = level;
79 		st->marker = address_markers;
80 		seq_printf(m, "---[ %s ]---\n", st->marker->name);
81 	} else if (prot != cur || level != st->level ||
82 		   st->current_address >= st->marker[1].start_address) {
83 		/* Print the actual finished series */
84 		seq_printf(m, "0x%0*lx-0x%0*lx",
85 			   width, st->start_address,
86 			   width, st->current_address);
87 		delta = (st->current_address - st->start_address) >> 10;
88 		while (!(delta & 0x3ff) && unit[1]) {
89 			delta >>= 10;
90 			unit++;
91 		}
92 		seq_printf(m, "%9lu%c ", delta, *unit);
93 		print_prot(m, st->current_prot, st->level);
94 		if (st->current_address >= st->marker[1].start_address) {
95 			st->marker++;
96 			seq_printf(m, "---[ %s ]---\n", st->marker->name);
97 		}
98 		st->start_address = st->current_address;
99 		st->current_prot = new_prot;
100 		st->level = level;
101 	}
102 }
103 
104 /*
105  * The actual page table walker functions. In order to keep the implementation
106  * of print_prot() short, we only check and pass _PAGE_INVALID and _PAGE_RO
107  * flags to note_page() if a region, segment or page table entry is invalid or
108  * read-only.
109  * After all it's just a hint that the current level being walked contains an
110  * invalid or read-only entry.
111  */
112 static void walk_pte_level(struct seq_file *m, struct pg_state *st,
113 			   pmd_t *pmd, unsigned long addr)
114 {
115 	unsigned int prot;
116 	pte_t *pte;
117 	int i;
118 
119 	for (i = 0; i < PTRS_PER_PTE && addr < max_addr; i++) {
120 		st->current_address = addr;
121 		pte = pte_offset_kernel(pmd, addr);
122 		prot = pte_val(*pte) & (_PAGE_RO | _PAGE_INVALID);
123 		note_page(m, st, prot, 4);
124 		addr += PAGE_SIZE;
125 	}
126 }
127 
128 static void walk_pmd_level(struct seq_file *m, struct pg_state *st,
129 			   pud_t *pud, unsigned long addr)
130 {
131 	unsigned int prot;
132 	pmd_t *pmd;
133 	int i;
134 
135 	for (i = 0; i < PTRS_PER_PMD && addr < max_addr; i++) {
136 		st->current_address = addr;
137 		pmd = pmd_offset(pud, addr);
138 		if (!pmd_none(*pmd)) {
139 			if (pmd_large(*pmd)) {
140 				prot = pmd_val(*pmd) & _SEGMENT_ENTRY_RO;
141 				note_page(m, st, prot, 3);
142 			} else
143 				walk_pte_level(m, st, pmd, addr);
144 		} else
145 			note_page(m, st, _PAGE_INVALID, 3);
146 		addr += PMD_SIZE;
147 	}
148 }
149 
150 static void walk_pud_level(struct seq_file *m, struct pg_state *st,
151 			   pgd_t *pgd, unsigned long addr)
152 {
153 	unsigned int prot;
154 	pud_t *pud;
155 	int i;
156 
157 	for (i = 0; i < PTRS_PER_PUD && addr < max_addr; i++) {
158 		st->current_address = addr;
159 		pud = pud_offset(pgd, addr);
160 		if (!pud_none(*pud))
161 			if (pud_large(*pud)) {
162 				prot = pud_val(*pud) & _PAGE_RO;
163 				note_page(m, st, prot, 2);
164 			} else
165 				walk_pmd_level(m, st, pud, addr);
166 		else
167 			note_page(m, st, _PAGE_INVALID, 2);
168 		addr += PUD_SIZE;
169 	}
170 }
171 
172 static void walk_pgd_level(struct seq_file *m)
173 {
174 	unsigned long addr = 0;
175 	struct pg_state st;
176 	pgd_t *pgd;
177 	int i;
178 
179 	memset(&st, 0, sizeof(st));
180 	for (i = 0; i < PTRS_PER_PGD && addr < max_addr; i++) {
181 		st.current_address = addr;
182 		pgd = pgd_offset_k(addr);
183 		if (!pgd_none(*pgd))
184 			walk_pud_level(m, &st, pgd, addr);
185 		else
186 			note_page(m, &st, _PAGE_INVALID, 1);
187 		addr += PGDIR_SIZE;
188 	}
189 	/* Flush out the last page */
190 	st.current_address = max_addr;
191 	note_page(m, &st, 0, 0);
192 }
193 
194 static int ptdump_show(struct seq_file *m, void *v)
195 {
196 	walk_pgd_level(m);
197 	return 0;
198 }
199 
200 static int ptdump_open(struct inode *inode, struct file *filp)
201 {
202 	return single_open(filp, ptdump_show, NULL);
203 }
204 
205 static const struct file_operations ptdump_fops = {
206 	.open		= ptdump_open,
207 	.read		= seq_read,
208 	.llseek		= seq_lseek,
209 	.release	= single_release,
210 };
211 
212 static int pt_dump_init(void)
213 {
214 	/*
215 	 * Figure out the maximum virtual address being accessible with the
216 	 * kernel ASCE. We need this to keep the page table walker functions
217 	 * from accessing non-existent entries.
218 	 */
219 #ifdef CONFIG_32BIT
220 	max_addr = 1UL << 31;
221 #else
222 	max_addr = (S390_lowcore.kernel_asce & _REGION_ENTRY_TYPE_MASK) >> 2;
223 	max_addr = 1UL << (max_addr * 11 + 31);
224 	address_markers[MODULES_NR].start_address = MODULES_VADDR;
225 #endif
226 	address_markers[VMEMMAP_NR].start_address = (unsigned long) vmemmap;
227 	address_markers[VMALLOC_NR].start_address = VMALLOC_START;
228 	debugfs_create_file("kernel_page_tables", 0400, NULL, NULL, &ptdump_fops);
229 	return 0;
230 }
231 device_initcall(pt_dump_init);
232