Loading...
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/ptdump.h>
10#include <asm/kasan.h>
11#include <asm/abs_lowcore.h>
12#include <asm/nospec-branch.h>
13#include <asm/sections.h>
14#include <asm/maccess.h>
15
16static unsigned long max_addr;
17
18struct addr_marker {
19 unsigned long start_address;
20 const char *name;
21};
22
23enum address_markers_idx {
24 IDENTITY_BEFORE_NR = 0,
25 IDENTITY_BEFORE_END_NR,
26 AMODE31_START_NR,
27 AMODE31_END_NR,
28 KERNEL_START_NR,
29 KERNEL_END_NR,
30#ifdef CONFIG_KFENCE
31 KFENCE_START_NR,
32 KFENCE_END_NR,
33#endif
34 IDENTITY_AFTER_NR,
35 IDENTITY_AFTER_END_NR,
36#ifdef CONFIG_KASAN
37 KASAN_SHADOW_START_NR,
38 KASAN_SHADOW_END_NR,
39#endif
40 VMEMMAP_NR,
41 VMEMMAP_END_NR,
42 VMALLOC_NR,
43 VMALLOC_END_NR,
44 MODULES_NR,
45 MODULES_END_NR,
46 ABS_LOWCORE_NR,
47 ABS_LOWCORE_END_NR,
48 MEMCPY_REAL_NR,
49 MEMCPY_REAL_END_NR,
50};
51
52static struct addr_marker address_markers[] = {
53 [IDENTITY_BEFORE_NR] = {0, "Identity Mapping Start"},
54 [IDENTITY_BEFORE_END_NR] = {(unsigned long)_stext, "Identity Mapping End"},
55 [AMODE31_START_NR] = {0, "Amode31 Area Start"},
56 [AMODE31_END_NR] = {0, "Amode31 Area End"},
57 [KERNEL_START_NR] = {(unsigned long)_stext, "Kernel Image Start"},
58 [KERNEL_END_NR] = {(unsigned long)_end, "Kernel Image End"},
59#ifdef CONFIG_KFENCE
60 [KFENCE_START_NR] = {0, "KFence Pool Start"},
61 [KFENCE_END_NR] = {0, "KFence Pool End"},
62#endif
63 [IDENTITY_AFTER_NR] = {(unsigned long)_end, "Identity Mapping Start"},
64 [IDENTITY_AFTER_END_NR] = {0, "Identity Mapping End"},
65#ifdef CONFIG_KASAN
66 [KASAN_SHADOW_START_NR] = {KASAN_SHADOW_START, "Kasan Shadow Start"},
67 [KASAN_SHADOW_END_NR] = {KASAN_SHADOW_END, "Kasan Shadow End"},
68#endif
69 [VMEMMAP_NR] = {0, "vmemmap Area Start"},
70 [VMEMMAP_END_NR] = {0, "vmemmap Area End"},
71 [VMALLOC_NR] = {0, "vmalloc Area Start"},
72 [VMALLOC_END_NR] = {0, "vmalloc Area End"},
73 [MODULES_NR] = {0, "Modules Area Start"},
74 [MODULES_END_NR] = {0, "Modules Area End"},
75 [ABS_LOWCORE_NR] = {0, "Lowcore Area Start"},
76 [ABS_LOWCORE_END_NR] = {0, "Lowcore Area End"},
77 [MEMCPY_REAL_NR] = {0, "Real Memory Copy Area Start"},
78 [MEMCPY_REAL_END_NR] = {0, "Real Memory Copy Area End"},
79 { -1, NULL }
80};
81
82struct pg_state {
83 struct ptdump_state ptdump;
84 struct seq_file *seq;
85 int level;
86 unsigned int current_prot;
87 bool check_wx;
88 unsigned long wx_pages;
89 unsigned long start_address;
90 const struct addr_marker *marker;
91};
92
93#define pt_dump_seq_printf(m, fmt, args...) \
94({ \
95 struct seq_file *__m = (m); \
96 \
97 if (__m) \
98 seq_printf(__m, fmt, ##args); \
99})
100
101#define pt_dump_seq_puts(m, fmt) \
102({ \
103 struct seq_file *__m = (m); \
104 \
105 if (__m) \
106 seq_printf(__m, fmt); \
107})
108
109static void print_prot(struct seq_file *m, unsigned int pr, int level)
110{
111 static const char * const level_name[] =
112 { "ASCE", "PGD", "PUD", "PMD", "PTE" };
113
114 pt_dump_seq_printf(m, "%s ", level_name[level]);
115 if (pr & _PAGE_INVALID) {
116 pt_dump_seq_printf(m, "I\n");
117 return;
118 }
119 pt_dump_seq_puts(m, (pr & _PAGE_PROTECT) ? "RO " : "RW ");
120 pt_dump_seq_puts(m, (pr & _PAGE_NOEXEC) ? "NX\n" : "X\n");
121}
122
123static void note_prot_wx(struct pg_state *st, unsigned long addr)
124{
125#ifdef CONFIG_DEBUG_WX
126 if (!st->check_wx)
127 return;
128 if (st->current_prot & _PAGE_INVALID)
129 return;
130 if (st->current_prot & _PAGE_PROTECT)
131 return;
132 if (st->current_prot & _PAGE_NOEXEC)
133 return;
134 /*
135 * The first lowcore page is W+X if spectre mitigations are using
136 * trampolines or the BEAR enhancements facility is not installed,
137 * in which case we have two lpswe instructions in lowcore that need
138 * to be executable.
139 */
140 if (addr == PAGE_SIZE && (nospec_uses_trampoline() || !static_key_enabled(&cpu_has_bear)))
141 return;
142 WARN_ONCE(1, "s390/mm: Found insecure W+X mapping at address %pS\n",
143 (void *)st->start_address);
144 st->wx_pages += (addr - st->start_address) / PAGE_SIZE;
145#endif /* CONFIG_DEBUG_WX */
146}
147
148static void note_page(struct ptdump_state *pt_st, unsigned long addr, int level, u64 val)
149{
150 int width = sizeof(unsigned long) * 2;
151 static const char units[] = "KMGTPE";
152 const char *unit = units;
153 unsigned long delta;
154 struct pg_state *st;
155 struct seq_file *m;
156 unsigned int prot;
157
158 st = container_of(pt_st, struct pg_state, ptdump);
159 m = st->seq;
160 prot = val & (_PAGE_PROTECT | _PAGE_NOEXEC);
161 if (level == 4 && (val & _PAGE_INVALID))
162 prot = _PAGE_INVALID;
163 /* For pmd_none() & friends val gets passed as zero. */
164 if (level != 4 && !val)
165 prot = _PAGE_INVALID;
166 /* Final flush from generic code. */
167 if (level == -1)
168 addr = max_addr;
169 if (st->level == -1) {
170 pt_dump_seq_printf(m, "---[ %s ]---\n", st->marker->name);
171 st->start_address = addr;
172 st->current_prot = prot;
173 st->level = level;
174 } else if (prot != st->current_prot || level != st->level ||
175 addr >= st->marker[1].start_address) {
176 note_prot_wx(st, addr);
177 pt_dump_seq_printf(m, "0x%0*lx-0x%0*lx ",
178 width, st->start_address,
179 width, addr);
180 delta = (addr - st->start_address) >> 10;
181 while (!(delta & 0x3ff) && unit[1]) {
182 delta >>= 10;
183 unit++;
184 }
185 pt_dump_seq_printf(m, "%9lu%c ", delta, *unit);
186 print_prot(m, st->current_prot, st->level);
187 while (addr >= st->marker[1].start_address) {
188 st->marker++;
189 pt_dump_seq_printf(m, "---[ %s ]---\n", st->marker->name);
190 }
191 st->start_address = addr;
192 st->current_prot = prot;
193 st->level = level;
194 }
195}
196
197#ifdef CONFIG_DEBUG_WX
198void ptdump_check_wx(void)
199{
200 struct pg_state st = {
201 .ptdump = {
202 .note_page = note_page,
203 .range = (struct ptdump_range[]) {
204 {.start = 0, .end = max_addr},
205 {.start = 0, .end = 0},
206 }
207 },
208 .seq = NULL,
209 .level = -1,
210 .current_prot = 0,
211 .check_wx = true,
212 .wx_pages = 0,
213 .start_address = 0,
214 .marker = (struct addr_marker[]) {
215 { .start_address = 0, .name = NULL},
216 { .start_address = -1, .name = NULL},
217 },
218 };
219
220 if (!MACHINE_HAS_NX)
221 return;
222 ptdump_walk_pgd(&st.ptdump, &init_mm, NULL);
223 if (st.wx_pages)
224 pr_warn("Checked W+X mappings: FAILED, %lu W+X pages found\n", st.wx_pages);
225 else
226 pr_info("Checked W+X mappings: passed, no %sW+X pages found\n",
227 (nospec_uses_trampoline() || !static_key_enabled(&cpu_has_bear)) ?
228 "unexpected " : "");
229}
230#endif /* CONFIG_DEBUG_WX */
231
232#ifdef CONFIG_PTDUMP_DEBUGFS
233static int ptdump_show(struct seq_file *m, void *v)
234{
235 struct pg_state st = {
236 .ptdump = {
237 .note_page = note_page,
238 .range = (struct ptdump_range[]) {
239 {.start = 0, .end = max_addr},
240 {.start = 0, .end = 0},
241 }
242 },
243 .seq = m,
244 .level = -1,
245 .current_prot = 0,
246 .check_wx = false,
247 .wx_pages = 0,
248 .start_address = 0,
249 .marker = address_markers,
250 };
251
252 get_online_mems();
253 mutex_lock(&cpa_mutex);
254 ptdump_walk_pgd(&st.ptdump, &init_mm, NULL);
255 mutex_unlock(&cpa_mutex);
256 put_online_mems();
257 return 0;
258}
259DEFINE_SHOW_ATTRIBUTE(ptdump);
260#endif /* CONFIG_PTDUMP_DEBUGFS */
261
262/*
263 * Heapsort from lib/sort.c is not a stable sorting algorithm, do a simple
264 * insertion sort to preserve the original order of markers with the same
265 * start address.
266 */
267static void sort_address_markers(void)
268{
269 struct addr_marker tmp;
270 int i, j;
271
272 for (i = 1; i < ARRAY_SIZE(address_markers) - 1; i++) {
273 tmp = address_markers[i];
274 for (j = i - 1; j >= 0 && address_markers[j].start_address > tmp.start_address; j--)
275 address_markers[j + 1] = address_markers[j];
276 address_markers[j + 1] = tmp;
277 }
278}
279
280static int pt_dump_init(void)
281{
282#ifdef CONFIG_KFENCE
283 unsigned long kfence_start = (unsigned long)__kfence_pool;
284#endif
285 /*
286 * Figure out the maximum virtual address being accessible with the
287 * kernel ASCE. We need this to keep the page table walker functions
288 * from accessing non-existent entries.
289 */
290 max_addr = (S390_lowcore.kernel_asce & _REGION_ENTRY_TYPE_MASK) >> 2;
291 max_addr = 1UL << (max_addr * 11 + 31);
292 address_markers[IDENTITY_AFTER_END_NR].start_address = ident_map_size;
293 address_markers[AMODE31_START_NR].start_address = __samode31;
294 address_markers[AMODE31_END_NR].start_address = __eamode31;
295 address_markers[MODULES_NR].start_address = MODULES_VADDR;
296 address_markers[MODULES_END_NR].start_address = MODULES_END;
297 address_markers[ABS_LOWCORE_NR].start_address = __abs_lowcore;
298 address_markers[ABS_LOWCORE_END_NR].start_address = __abs_lowcore + ABS_LOWCORE_MAP_SIZE;
299 address_markers[MEMCPY_REAL_NR].start_address = __memcpy_real_area;
300 address_markers[MEMCPY_REAL_END_NR].start_address = __memcpy_real_area + PAGE_SIZE;
301 address_markers[VMEMMAP_NR].start_address = (unsigned long) vmemmap;
302 address_markers[VMEMMAP_END_NR].start_address = (unsigned long)vmemmap + vmemmap_size;
303 address_markers[VMALLOC_NR].start_address = VMALLOC_START;
304 address_markers[VMALLOC_END_NR].start_address = VMALLOC_END;
305#ifdef CONFIG_KFENCE
306 address_markers[KFENCE_START_NR].start_address = kfence_start;
307 address_markers[KFENCE_END_NR].start_address = kfence_start + KFENCE_POOL_SIZE;
308#endif
309 sort_address_markers();
310#ifdef CONFIG_PTDUMP_DEBUGFS
311 debugfs_create_file("kernel_page_tables", 0400, NULL, NULL, &ptdump_fops);
312#endif /* CONFIG_PTDUMP_DEBUGFS */
313 return 0;
314}
315device_initcall(pt_dump_init);
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#include <asm/pgtable.h>
10
11static unsigned long max_addr;
12
13struct addr_marker {
14 unsigned long start_address;
15 const char *name;
16};
17
18enum address_markers_idx {
19 IDENTITY_NR = 0,
20 KERNEL_START_NR,
21 KERNEL_END_NR,
22#ifdef CONFIG_KASAN
23 KASAN_SHADOW_START_NR,
24 KASAN_SHADOW_END_NR,
25#endif
26 VMEMMAP_NR,
27 VMALLOC_NR,
28 MODULES_NR,
29};
30
31static struct addr_marker address_markers[] = {
32 [IDENTITY_NR] = {0, "Identity Mapping"},
33 [KERNEL_START_NR] = {(unsigned long)_stext, "Kernel Image Start"},
34 [KERNEL_END_NR] = {(unsigned long)_end, "Kernel Image End"},
35#ifdef CONFIG_KASAN
36 [KASAN_SHADOW_START_NR] = {KASAN_SHADOW_START, "Kasan Shadow Start"},
37 [KASAN_SHADOW_END_NR] = {KASAN_SHADOW_END, "Kasan Shadow End"},
38#endif
39 [VMEMMAP_NR] = {0, "vmemmap Area"},
40 [VMALLOC_NR] = {0, "vmalloc Area"},
41 [MODULES_NR] = {0, "Modules Area"},
42 { -1, NULL }
43};
44
45struct pg_state {
46 int level;
47 unsigned int current_prot;
48 unsigned long start_address;
49 unsigned long current_address;
50 const struct addr_marker *marker;
51};
52
53static void print_prot(struct seq_file *m, unsigned int pr, int level)
54{
55 static const char * const level_name[] =
56 { "ASCE", "PGD", "PUD", "PMD", "PTE" };
57
58 seq_printf(m, "%s ", level_name[level]);
59 if (pr & _PAGE_INVALID) {
60 seq_printf(m, "I\n");
61 return;
62 }
63 seq_puts(m, (pr & _PAGE_PROTECT) ? "RO " : "RW ");
64 seq_puts(m, (pr & _PAGE_NOEXEC) ? "NX\n" : "X\n");
65}
66
67static void note_page(struct seq_file *m, struct pg_state *st,
68 unsigned int new_prot, int level)
69{
70 static const char units[] = "KMGTPE";
71 int width = sizeof(unsigned long) * 2;
72 const char *unit = units;
73 unsigned int prot, cur;
74 unsigned long delta;
75
76 /*
77 * If we have a "break" in the series, we need to flush the state
78 * that we have now. "break" is either changing perms, levels or
79 * address space marker.
80 */
81 prot = new_prot;
82 cur = st->current_prot;
83
84 if (!st->level) {
85 /* First entry */
86 st->current_prot = new_prot;
87 st->level = level;
88 st->marker = address_markers;
89 seq_printf(m, "---[ %s ]---\n", st->marker->name);
90 } else if (prot != cur || level != st->level ||
91 st->current_address >= st->marker[1].start_address) {
92 /* Print the actual finished series */
93 seq_printf(m, "0x%0*lx-0x%0*lx ",
94 width, st->start_address,
95 width, st->current_address);
96 delta = (st->current_address - st->start_address) >> 10;
97 while (!(delta & 0x3ff) && unit[1]) {
98 delta >>= 10;
99 unit++;
100 }
101 seq_printf(m, "%9lu%c ", delta, *unit);
102 print_prot(m, st->current_prot, st->level);
103 while (st->current_address >= st->marker[1].start_address) {
104 st->marker++;
105 seq_printf(m, "---[ %s ]---\n", st->marker->name);
106 }
107 st->start_address = st->current_address;
108 st->current_prot = new_prot;
109 st->level = level;
110 }
111}
112
113#ifdef CONFIG_KASAN
114static void note_kasan_early_shadow_page(struct seq_file *m,
115 struct pg_state *st)
116{
117 unsigned int prot;
118
119 prot = pte_val(*kasan_early_shadow_pte) &
120 (_PAGE_PROTECT | _PAGE_INVALID | _PAGE_NOEXEC);
121 note_page(m, st, prot, 4);
122}
123#endif
124
125/*
126 * The actual page table walker functions. In order to keep the
127 * implementation of print_prot() short, we only check and pass
128 * _PAGE_INVALID and _PAGE_PROTECT flags to note_page() if a region,
129 * segment or page table entry is invalid or read-only.
130 * After all it's just a hint that the current level being walked
131 * contains an invalid or read-only entry.
132 */
133static void walk_pte_level(struct seq_file *m, struct pg_state *st,
134 pmd_t *pmd, unsigned long addr)
135{
136 unsigned int prot;
137 pte_t *pte;
138 int i;
139
140 for (i = 0; i < PTRS_PER_PTE && addr < max_addr; i++) {
141 st->current_address = addr;
142 pte = pte_offset_kernel(pmd, addr);
143 prot = pte_val(*pte) &
144 (_PAGE_PROTECT | _PAGE_INVALID | _PAGE_NOEXEC);
145 note_page(m, st, prot, 4);
146 addr += PAGE_SIZE;
147 }
148}
149
150static void walk_pmd_level(struct seq_file *m, struct pg_state *st,
151 pud_t *pud, unsigned long addr)
152{
153 unsigned int prot;
154 pmd_t *pmd;
155 int i;
156
157#ifdef CONFIG_KASAN
158 if ((pud_val(*pud) & PAGE_MASK) == __pa(kasan_early_shadow_pmd)) {
159 note_kasan_early_shadow_page(m, st);
160 return;
161 }
162#endif
163
164 pmd = pmd_offset(pud, addr);
165 for (i = 0; i < PTRS_PER_PMD && addr < max_addr; i++, pmd++) {
166 st->current_address = addr;
167 if (!pmd_none(*pmd)) {
168 if (pmd_large(*pmd)) {
169 prot = pmd_val(*pmd) &
170 (_SEGMENT_ENTRY_PROTECT |
171 _SEGMENT_ENTRY_NOEXEC);
172 note_page(m, st, prot, 3);
173 } else
174 walk_pte_level(m, st, pmd, addr);
175 } else
176 note_page(m, st, _PAGE_INVALID, 3);
177 addr += PMD_SIZE;
178 }
179}
180
181static void walk_pud_level(struct seq_file *m, struct pg_state *st,
182 p4d_t *p4d, unsigned long addr)
183{
184 unsigned int prot;
185 pud_t *pud;
186 int i;
187
188#ifdef CONFIG_KASAN
189 if ((p4d_val(*p4d) & PAGE_MASK) == __pa(kasan_early_shadow_pud)) {
190 note_kasan_early_shadow_page(m, st);
191 return;
192 }
193#endif
194
195 pud = pud_offset(p4d, addr);
196 for (i = 0; i < PTRS_PER_PUD && addr < max_addr; i++, pud++) {
197 st->current_address = addr;
198 if (!pud_none(*pud))
199 if (pud_large(*pud)) {
200 prot = pud_val(*pud) &
201 (_REGION_ENTRY_PROTECT |
202 _REGION_ENTRY_NOEXEC);
203 note_page(m, st, prot, 2);
204 } else
205 walk_pmd_level(m, st, pud, addr);
206 else
207 note_page(m, st, _PAGE_INVALID, 2);
208 addr += PUD_SIZE;
209 }
210}
211
212static void walk_p4d_level(struct seq_file *m, struct pg_state *st,
213 pgd_t *pgd, unsigned long addr)
214{
215 p4d_t *p4d;
216 int i;
217
218#ifdef CONFIG_KASAN
219 if ((pgd_val(*pgd) & PAGE_MASK) == __pa(kasan_early_shadow_p4d)) {
220 note_kasan_early_shadow_page(m, st);
221 return;
222 }
223#endif
224
225 p4d = p4d_offset(pgd, addr);
226 for (i = 0; i < PTRS_PER_P4D && addr < max_addr; i++, p4d++) {
227 st->current_address = addr;
228 if (!p4d_none(*p4d))
229 walk_pud_level(m, st, p4d, addr);
230 else
231 note_page(m, st, _PAGE_INVALID, 2);
232 addr += P4D_SIZE;
233 }
234}
235
236static void walk_pgd_level(struct seq_file *m)
237{
238 unsigned long addr = 0;
239 struct pg_state st;
240 pgd_t *pgd;
241 int i;
242
243 memset(&st, 0, sizeof(st));
244 for (i = 0; i < PTRS_PER_PGD && addr < max_addr; i++) {
245 st.current_address = addr;
246 pgd = pgd_offset_k(addr);
247 if (!pgd_none(*pgd))
248 walk_p4d_level(m, &st, pgd, addr);
249 else
250 note_page(m, &st, _PAGE_INVALID, 1);
251 addr += PGDIR_SIZE;
252 cond_resched();
253 }
254 /* Flush out the last page */
255 st.current_address = max_addr;
256 note_page(m, &st, 0, 0);
257}
258
259static int ptdump_show(struct seq_file *m, void *v)
260{
261 walk_pgd_level(m);
262 return 0;
263}
264
265static int ptdump_open(struct inode *inode, struct file *filp)
266{
267 return single_open(filp, ptdump_show, NULL);
268}
269
270static const struct file_operations ptdump_fops = {
271 .open = ptdump_open,
272 .read = seq_read,
273 .llseek = seq_lseek,
274 .release = single_release,
275};
276
277static int pt_dump_init(void)
278{
279 /*
280 * Figure out the maximum virtual address being accessible with the
281 * kernel ASCE. We need this to keep the page table walker functions
282 * from accessing non-existent entries.
283 */
284 max_addr = (S390_lowcore.kernel_asce & _REGION_ENTRY_TYPE_MASK) >> 2;
285 max_addr = 1UL << (max_addr * 11 + 31);
286 address_markers[MODULES_NR].start_address = MODULES_VADDR;
287 address_markers[VMEMMAP_NR].start_address = (unsigned long) vmemmap;
288 address_markers[VMALLOC_NR].start_address = VMALLOC_START;
289 debugfs_create_file("kernel_page_tables", 0400, NULL, NULL, &ptdump_fops);
290 return 0;
291}
292device_initcall(pt_dump_init);