Loading...
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
8static unsigned long max_addr;
9
10struct addr_marker {
11 unsigned long start_address;
12 const char *name;
13};
14
15enum 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
24static 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
34struct 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
42static 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
56static 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 */
110static 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
126static 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
148static 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
170static 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
192static int ptdump_show(struct seq_file *m, void *v)
193{
194 walk_pgd_level(m);
195 return 0;
196}
197
198static int ptdump_open(struct inode *inode, struct file *filp)
199{
200 return single_open(filp, ptdump_show, NULL);
201}
202
203static const struct file_operations ptdump_fops = {
204 .open = ptdump_open,
205 .read = seq_read,
206 .llseek = seq_lseek,
207 .release = single_release,
208};
209
210static 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}
225device_initcall(pt_dump_init);
1// SPDX-License-Identifier: GPL-2.0
2#include <linux/set_memory.h>
3#include <linux/ptdump.h>
4#include <linux/seq_file.h>
5#include <linux/debugfs.h>
6#include <linux/mm.h>
7#include <linux/kfence.h>
8#include <linux/kasan.h>
9#include <asm/kasan.h>
10#include <asm/abs_lowcore.h>
11#include <asm/nospec-branch.h>
12#include <asm/sections.h>
13#include <asm/maccess.h>
14
15static unsigned long max_addr;
16
17struct addr_marker {
18 unsigned long start_address;
19 const char *name;
20};
21
22enum address_markers_idx {
23 IDENTITY_BEFORE_NR = 0,
24 IDENTITY_BEFORE_END_NR,
25 AMODE31_START_NR,
26 AMODE31_END_NR,
27 KERNEL_START_NR,
28 KERNEL_END_NR,
29#ifdef CONFIG_KFENCE
30 KFENCE_START_NR,
31 KFENCE_END_NR,
32#endif
33 IDENTITY_AFTER_NR,
34 IDENTITY_AFTER_END_NR,
35 VMEMMAP_NR,
36 VMEMMAP_END_NR,
37 VMALLOC_NR,
38 VMALLOC_END_NR,
39 MODULES_NR,
40 MODULES_END_NR,
41 ABS_LOWCORE_NR,
42 ABS_LOWCORE_END_NR,
43 MEMCPY_REAL_NR,
44 MEMCPY_REAL_END_NR,
45#ifdef CONFIG_KASAN
46 KASAN_SHADOW_START_NR,
47 KASAN_SHADOW_END_NR,
48#endif
49};
50
51static struct addr_marker address_markers[] = {
52 [IDENTITY_BEFORE_NR] = {0, "Identity Mapping Start"},
53 [IDENTITY_BEFORE_END_NR] = {(unsigned long)_stext, "Identity Mapping End"},
54 [AMODE31_START_NR] = {0, "Amode31 Area Start"},
55 [AMODE31_END_NR] = {0, "Amode31 Area End"},
56 [KERNEL_START_NR] = {(unsigned long)_stext, "Kernel Image Start"},
57 [KERNEL_END_NR] = {(unsigned long)_end, "Kernel Image End"},
58#ifdef CONFIG_KFENCE
59 [KFENCE_START_NR] = {0, "KFence Pool Start"},
60 [KFENCE_END_NR] = {0, "KFence Pool End"},
61#endif
62 [IDENTITY_AFTER_NR] = {(unsigned long)_end, "Identity Mapping Start"},
63 [IDENTITY_AFTER_END_NR] = {0, "Identity Mapping End"},
64 [VMEMMAP_NR] = {0, "vmemmap Area Start"},
65 [VMEMMAP_END_NR] = {0, "vmemmap Area End"},
66 [VMALLOC_NR] = {0, "vmalloc Area Start"},
67 [VMALLOC_END_NR] = {0, "vmalloc Area End"},
68 [MODULES_NR] = {0, "Modules Area Start"},
69 [MODULES_END_NR] = {0, "Modules Area End"},
70 [ABS_LOWCORE_NR] = {0, "Lowcore Area Start"},
71 [ABS_LOWCORE_END_NR] = {0, "Lowcore Area End"},
72 [MEMCPY_REAL_NR] = {0, "Real Memory Copy Area Start"},
73 [MEMCPY_REAL_END_NR] = {0, "Real Memory Copy Area End"},
74#ifdef CONFIG_KASAN
75 [KASAN_SHADOW_START_NR] = {KASAN_SHADOW_START, "Kasan Shadow Start"},
76 [KASAN_SHADOW_END_NR] = {KASAN_SHADOW_END, "Kasan Shadow End"},
77#endif
78 { -1, NULL }
79};
80
81struct pg_state {
82 struct ptdump_state ptdump;
83 struct seq_file *seq;
84 int level;
85 unsigned int current_prot;
86 bool check_wx;
87 unsigned long wx_pages;
88 unsigned long start_address;
89 const struct addr_marker *marker;
90};
91
92#define pt_dump_seq_printf(m, fmt, args...) \
93({ \
94 struct seq_file *__m = (m); \
95 \
96 if (__m) \
97 seq_printf(__m, fmt, ##args); \
98})
99
100#define pt_dump_seq_puts(m, fmt) \
101({ \
102 struct seq_file *__m = (m); \
103 \
104 if (__m) \
105 seq_printf(__m, fmt); \
106})
107
108static void print_prot(struct seq_file *m, unsigned int pr, int level)
109{
110 static const char * const level_name[] =
111 { "ASCE", "PGD", "PUD", "PMD", "PTE" };
112
113 pt_dump_seq_printf(m, "%s ", level_name[level]);
114 if (pr & _PAGE_INVALID) {
115 pt_dump_seq_printf(m, "I\n");
116 return;
117 }
118 pt_dump_seq_puts(m, (pr & _PAGE_PROTECT) ? "RO " : "RW ");
119 pt_dump_seq_puts(m, (pr & _PAGE_NOEXEC) ? "NX\n" : "X\n");
120}
121
122static void note_prot_wx(struct pg_state *st, unsigned long addr)
123{
124 if (!st->check_wx)
125 return;
126 if (st->current_prot & _PAGE_INVALID)
127 return;
128 if (st->current_prot & _PAGE_PROTECT)
129 return;
130 if (st->current_prot & _PAGE_NOEXEC)
131 return;
132 /*
133 * The first lowcore page is W+X if spectre mitigations are using
134 * trampolines or the BEAR enhancements facility is not installed,
135 * in which case we have two lpswe instructions in lowcore that need
136 * to be executable.
137 */
138 if (addr == PAGE_SIZE && (nospec_uses_trampoline() || !static_key_enabled(&cpu_has_bear)))
139 return;
140 WARN_ONCE(IS_ENABLED(CONFIG_DEBUG_WX),
141 "s390/mm: Found insecure W+X mapping at address %pS\n",
142 (void *)st->start_address);
143 st->wx_pages += (addr - st->start_address) / PAGE_SIZE;
144}
145
146static void note_page(struct ptdump_state *pt_st, unsigned long addr, int level, u64 val)
147{
148 int width = sizeof(unsigned long) * 2;
149 static const char units[] = "KMGTPE";
150 const char *unit = units;
151 unsigned long delta;
152 struct pg_state *st;
153 struct seq_file *m;
154 unsigned int prot;
155
156 st = container_of(pt_st, struct pg_state, ptdump);
157 m = st->seq;
158 prot = val & (_PAGE_PROTECT | _PAGE_NOEXEC);
159 if (level == 4 && (val & _PAGE_INVALID))
160 prot = _PAGE_INVALID;
161 /* For pmd_none() & friends val gets passed as zero. */
162 if (level != 4 && !val)
163 prot = _PAGE_INVALID;
164 /* Final flush from generic code. */
165 if (level == -1)
166 addr = max_addr;
167 if (st->level == -1) {
168 pt_dump_seq_printf(m, "---[ %s ]---\n", st->marker->name);
169 st->start_address = addr;
170 st->current_prot = prot;
171 st->level = level;
172 } else if (prot != st->current_prot || level != st->level ||
173 addr >= st->marker[1].start_address) {
174 note_prot_wx(st, addr);
175 pt_dump_seq_printf(m, "0x%0*lx-0x%0*lx ",
176 width, st->start_address,
177 width, addr);
178 delta = (addr - st->start_address) >> 10;
179 while (!(delta & 0x3ff) && unit[1]) {
180 delta >>= 10;
181 unit++;
182 }
183 pt_dump_seq_printf(m, "%9lu%c ", delta, *unit);
184 print_prot(m, st->current_prot, st->level);
185 while (addr >= st->marker[1].start_address) {
186 st->marker++;
187 pt_dump_seq_printf(m, "---[ %s ]---\n", st->marker->name);
188 }
189 st->start_address = addr;
190 st->current_prot = prot;
191 st->level = level;
192 }
193}
194
195bool ptdump_check_wx(void)
196{
197 struct pg_state st = {
198 .ptdump = {
199 .note_page = note_page,
200 .range = (struct ptdump_range[]) {
201 {.start = 0, .end = max_addr},
202 {.start = 0, .end = 0},
203 }
204 },
205 .seq = NULL,
206 .level = -1,
207 .current_prot = 0,
208 .check_wx = true,
209 .wx_pages = 0,
210 .start_address = 0,
211 .marker = (struct addr_marker[]) {
212 { .start_address = 0, .name = NULL},
213 { .start_address = -1, .name = NULL},
214 },
215 };
216
217 if (!MACHINE_HAS_NX)
218 return true;
219 ptdump_walk_pgd(&st.ptdump, &init_mm, NULL);
220 if (st.wx_pages) {
221 pr_warn("Checked W+X mappings: FAILED, %lu W+X pages found\n", st.wx_pages);
222
223 return false;
224 } else {
225 pr_info("Checked W+X mappings: passed, no %sW+X pages found\n",
226 (nospec_uses_trampoline() || !static_key_enabled(&cpu_has_bear)) ?
227 "unexpected " : "");
228
229 return true;
230 }
231}
232
233#ifdef CONFIG_PTDUMP_DEBUGFS
234static int ptdump_show(struct seq_file *m, void *v)
235{
236 struct pg_state st = {
237 .ptdump = {
238 .note_page = note_page,
239 .range = (struct ptdump_range[]) {
240 {.start = 0, .end = max_addr},
241 {.start = 0, .end = 0},
242 }
243 },
244 .seq = m,
245 .level = -1,
246 .current_prot = 0,
247 .check_wx = false,
248 .wx_pages = 0,
249 .start_address = 0,
250 .marker = address_markers,
251 };
252
253 get_online_mems();
254 mutex_lock(&cpa_mutex);
255 ptdump_walk_pgd(&st.ptdump, &init_mm, NULL);
256 mutex_unlock(&cpa_mutex);
257 put_online_mems();
258 return 0;
259}
260DEFINE_SHOW_ATTRIBUTE(ptdump);
261#endif /* CONFIG_PTDUMP_DEBUGFS */
262
263/*
264 * Heapsort from lib/sort.c is not a stable sorting algorithm, do a simple
265 * insertion sort to preserve the original order of markers with the same
266 * start address.
267 */
268static void sort_address_markers(void)
269{
270 struct addr_marker tmp;
271 int i, j;
272
273 for (i = 1; i < ARRAY_SIZE(address_markers) - 1; i++) {
274 tmp = address_markers[i];
275 for (j = i - 1; j >= 0 && address_markers[j].start_address > tmp.start_address; j--)
276 address_markers[j + 1] = address_markers[j];
277 address_markers[j + 1] = tmp;
278 }
279}
280
281static int pt_dump_init(void)
282{
283#ifdef CONFIG_KFENCE
284 unsigned long kfence_start = (unsigned long)__kfence_pool;
285#endif
286 /*
287 * Figure out the maximum virtual address being accessible with the
288 * kernel ASCE. We need this to keep the page table walker functions
289 * from accessing non-existent entries.
290 */
291 max_addr = (S390_lowcore.kernel_asce.val & _REGION_ENTRY_TYPE_MASK) >> 2;
292 max_addr = 1UL << (max_addr * 11 + 31);
293 address_markers[IDENTITY_AFTER_END_NR].start_address = ident_map_size;
294 address_markers[AMODE31_START_NR].start_address = (unsigned long)__samode31;
295 address_markers[AMODE31_END_NR].start_address = (unsigned long)__eamode31;
296 address_markers[MODULES_NR].start_address = MODULES_VADDR;
297 address_markers[MODULES_END_NR].start_address = MODULES_END;
298 address_markers[ABS_LOWCORE_NR].start_address = __abs_lowcore;
299 address_markers[ABS_LOWCORE_END_NR].start_address = __abs_lowcore + ABS_LOWCORE_MAP_SIZE;
300 address_markers[MEMCPY_REAL_NR].start_address = __memcpy_real_area;
301 address_markers[MEMCPY_REAL_END_NR].start_address = __memcpy_real_area + MEMCPY_REAL_SIZE;
302 address_markers[VMEMMAP_NR].start_address = (unsigned long) vmemmap;
303 address_markers[VMEMMAP_END_NR].start_address = (unsigned long)vmemmap + vmemmap_size;
304 address_markers[VMALLOC_NR].start_address = VMALLOC_START;
305 address_markers[VMALLOC_END_NR].start_address = VMALLOC_END;
306#ifdef CONFIG_KFENCE
307 address_markers[KFENCE_START_NR].start_address = kfence_start;
308 address_markers[KFENCE_END_NR].start_address = kfence_start + KFENCE_POOL_SIZE;
309#endif
310 sort_address_markers();
311#ifdef CONFIG_PTDUMP_DEBUGFS
312 debugfs_create_file("kernel_page_tables", 0400, NULL, NULL, &ptdump_fops);
313#endif /* CONFIG_PTDUMP_DEBUGFS */
314 return 0;
315}
316device_initcall(pt_dump_init);