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 MODULES_NR,
22 };
23
24 static struct addr_marker address_markers[] = {
25 [IDENTITY_NR] = {0, "Identity Mapping"},
26 [KERNEL_START_NR] = {(unsigned long)&_stext, "Kernel Image Start"},
27 [KERNEL_END_NR] = {(unsigned long)&_end, "Kernel Image End"},
28 [VMEMMAP_NR] = {0, "vmemmap Area"},
29 [VMALLOC_NR] = {0, "vmalloc Area"},
30 [MODULES_NR] = {0, "Modules Area"},
31 { -1, NULL }
32 };
33
34 struct pg_state {
35 int level;
36 unsigned int current_prot;
37 unsigned long start_address;
38 unsigned long current_address;
39 const struct addr_marker *marker;
40 };
41
print_prot(struct seq_file * m,unsigned int pr,int level)42 static void print_prot(struct seq_file *m, unsigned int pr, int level)
43 {
44 static const char * const level_name[] =
45 { "ASCE", "PGD", "PUD", "PMD", "PTE" };
46
47 seq_printf(m, "%s ", level_name[level]);
48 if (pr & _PAGE_INVALID) {
49 seq_printf(m, "I\n");
50 return;
51 }
52 seq_printf(m, "%s", pr & _PAGE_PROTECT ? "RO " : "RW ");
53 seq_putc(m, '\n');
54 }
55
note_page(struct seq_file * m,struct pg_state * st,unsigned int new_prot,int level)56 static void note_page(struct seq_file *m, struct pg_state *st,
57 unsigned int new_prot, int level)
58 {
59 static const char units[] = "KMGTPE";
60 int width = sizeof(unsigned long) * 2;
61 const char *unit = units;
62 unsigned int prot, cur;
63 unsigned long delta;
64
65 /*
66 * If we have a "break" in the series, we need to flush the state
67 * that we have now. "break" is either changing perms, levels or
68 * address space marker.
69 */
70 prot = new_prot;
71 cur = st->current_prot;
72
73 if (!st->level) {
74 /* First entry */
75 st->current_prot = new_prot;
76 st->level = level;
77 st->marker = address_markers;
78 seq_printf(m, "---[ %s ]---\n", st->marker->name);
79 } else if (prot != cur || level != st->level ||
80 st->current_address >= st->marker[1].start_address) {
81 /* Print the actual finished series */
82 seq_printf(m, "0x%0*lx-0x%0*lx",
83 width, st->start_address,
84 width, st->current_address);
85 delta = (st->current_address - st->start_address) >> 10;
86 while (!(delta & 0x3ff) && unit[1]) {
87 delta >>= 10;
88 unit++;
89 }
90 seq_printf(m, "%9lu%c ", delta, *unit);
91 print_prot(m, st->current_prot, st->level);
92 if (st->current_address >= st->marker[1].start_address) {
93 st->marker++;
94 seq_printf(m, "---[ %s ]---\n", st->marker->name);
95 }
96 st->start_address = st->current_address;
97 st->current_prot = new_prot;
98 st->level = level;
99 }
100 }
101
102 /*
103 * The actual page table walker functions. In order to keep the
104 * implementation of print_prot() short, we only check and pass
105 * _PAGE_INVALID and _PAGE_PROTECT flags to note_page() if a region,
106 * segment or page table entry is invalid or read-only.
107 * After all it's just a hint that the current level being walked
108 * contains an invalid or read-only entry.
109 */
walk_pte_level(struct seq_file * m,struct pg_state * st,pmd_t * pmd,unsigned long addr)110 static void walk_pte_level(struct seq_file *m, struct pg_state *st,
111 pmd_t *pmd, unsigned long addr)
112 {
113 unsigned int prot;
114 pte_t *pte;
115 int i;
116
117 for (i = 0; i < PTRS_PER_PTE && addr < max_addr; i++) {
118 st->current_address = addr;
119 pte = pte_offset_kernel(pmd, addr);
120 prot = pte_val(*pte) & (_PAGE_PROTECT | _PAGE_INVALID);
121 note_page(m, st, prot, 4);
122 addr += PAGE_SIZE;
123 }
124 }
125
walk_pmd_level(struct seq_file * m,struct pg_state * st,pud_t * pud,unsigned long addr)126 static void walk_pmd_level(struct seq_file *m, struct pg_state *st,
127 pud_t *pud, unsigned long addr)
128 {
129 unsigned int prot;
130 pmd_t *pmd;
131 int i;
132
133 for (i = 0; i < PTRS_PER_PMD && addr < max_addr; i++) {
134 st->current_address = addr;
135 pmd = pmd_offset(pud, addr);
136 if (!pmd_none(*pmd)) {
137 if (pmd_large(*pmd)) {
138 prot = pmd_val(*pmd) & _SEGMENT_ENTRY_PROTECT;
139 note_page(m, st, prot, 3);
140 } else
141 walk_pte_level(m, st, pmd, addr);
142 } else
143 note_page(m, st, _PAGE_INVALID, 3);
144 addr += PMD_SIZE;
145 }
146 }
147
walk_pud_level(struct seq_file * m,struct pg_state * st,pgd_t * pgd,unsigned long addr)148 static void walk_pud_level(struct seq_file *m, struct pg_state *st,
149 pgd_t *pgd, unsigned long addr)
150 {
151 unsigned int prot;
152 pud_t *pud;
153 int i;
154
155 for (i = 0; i < PTRS_PER_PUD && addr < max_addr; i++) {
156 st->current_address = addr;
157 pud = pud_offset(pgd, addr);
158 if (!pud_none(*pud))
159 if (pud_large(*pud)) {
160 prot = pud_val(*pud) & _REGION3_ENTRY_RO;
161 note_page(m, st, prot, 2);
162 } else
163 walk_pmd_level(m, st, pud, addr);
164 else
165 note_page(m, st, _PAGE_INVALID, 2);
166 addr += PUD_SIZE;
167 }
168 }
169
walk_pgd_level(struct seq_file * m)170 static void walk_pgd_level(struct seq_file *m)
171 {
172 unsigned long addr = 0;
173 struct pg_state st;
174 pgd_t *pgd;
175 int i;
176
177 memset(&st, 0, sizeof(st));
178 for (i = 0; i < PTRS_PER_PGD && addr < max_addr; i++) {
179 st.current_address = addr;
180 pgd = pgd_offset_k(addr);
181 if (!pgd_none(*pgd))
182 walk_pud_level(m, &st, pgd, addr);
183 else
184 note_page(m, &st, _PAGE_INVALID, 1);
185 addr += PGDIR_SIZE;
186 }
187 /* Flush out the last page */
188 st.current_address = max_addr;
189 note_page(m, &st, 0, 0);
190 }
191
ptdump_show(struct seq_file * m,void * v)192 static int ptdump_show(struct seq_file *m, void *v)
193 {
194 walk_pgd_level(m);
195 return 0;
196 }
197
ptdump_open(struct inode * inode,struct file * filp)198 static int ptdump_open(struct inode *inode, struct file *filp)
199 {
200 return single_open(filp, ptdump_show, NULL);
201 }
202
203 static const struct file_operations ptdump_fops = {
204 .open = ptdump_open,
205 .read = seq_read,
206 .llseek = seq_lseek,
207 .release = single_release,
208 };
209
pt_dump_init(void)210 static int pt_dump_init(void)
211 {
212 /*
213 * Figure out the maximum virtual address being accessible with the
214 * kernel ASCE. We need this to keep the page table walker functions
215 * from accessing non-existent entries.
216 */
217 max_addr = (S390_lowcore.kernel_asce & _REGION_ENTRY_TYPE_MASK) >> 2;
218 max_addr = 1UL << (max_addr * 11 + 31);
219 address_markers[MODULES_NR].start_address = MODULES_VADDR;
220 address_markers[VMEMMAP_NR].start_address = (unsigned long) vmemmap;
221 address_markers[VMALLOC_NR].start_address = VMALLOC_START;
222 debugfs_create_file("kernel_page_tables", 0400, NULL, NULL, &ptdump_fops);
223 return 0;
224 }
225 device_initcall(pt_dump_init);
226