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 15 static unsigned long max_addr; 16 17 struct addr_marker { 18 unsigned long start_address; 19 const char *name; 20 }; 21 22 enum 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 #ifdef CONFIG_KMSAN 40 KMSAN_VMALLOC_SHADOW_START_NR, 41 KMSAN_VMALLOC_SHADOW_END_NR, 42 KMSAN_VMALLOC_ORIGIN_START_NR, 43 KMSAN_VMALLOC_ORIGIN_END_NR, 44 KMSAN_MODULES_SHADOW_START_NR, 45 KMSAN_MODULES_SHADOW_END_NR, 46 KMSAN_MODULES_ORIGIN_START_NR, 47 KMSAN_MODULES_ORIGIN_END_NR, 48 #endif 49 MODULES_NR, 50 MODULES_END_NR, 51 ABS_LOWCORE_NR, 52 ABS_LOWCORE_END_NR, 53 MEMCPY_REAL_NR, 54 MEMCPY_REAL_END_NR, 55 #ifdef CONFIG_KASAN 56 KASAN_SHADOW_START_NR, 57 KASAN_SHADOW_END_NR, 58 #endif 59 }; 60 61 static struct addr_marker address_markers[] = { 62 [IDENTITY_BEFORE_NR] = {0, "Identity Mapping Start"}, 63 [IDENTITY_BEFORE_END_NR] = {(unsigned long)_stext, "Identity Mapping End"}, 64 [AMODE31_START_NR] = {0, "Amode31 Area Start"}, 65 [AMODE31_END_NR] = {0, "Amode31 Area End"}, 66 [KERNEL_START_NR] = {(unsigned long)_stext, "Kernel Image Start"}, 67 [KERNEL_END_NR] = {(unsigned long)_end, "Kernel Image End"}, 68 #ifdef CONFIG_KFENCE 69 [KFENCE_START_NR] = {0, "KFence Pool Start"}, 70 [KFENCE_END_NR] = {0, "KFence Pool End"}, 71 #endif 72 [IDENTITY_AFTER_NR] = {(unsigned long)_end, "Identity Mapping Start"}, 73 [IDENTITY_AFTER_END_NR] = {0, "Identity Mapping End"}, 74 [VMEMMAP_NR] = {0, "vmemmap Area Start"}, 75 [VMEMMAP_END_NR] = {0, "vmemmap Area End"}, 76 [VMALLOC_NR] = {0, "vmalloc Area Start"}, 77 [VMALLOC_END_NR] = {0, "vmalloc Area End"}, 78 #ifdef CONFIG_KMSAN 79 [KMSAN_VMALLOC_SHADOW_START_NR] = {0, "Kmsan vmalloc Shadow Start"}, 80 [KMSAN_VMALLOC_SHADOW_END_NR] = {0, "Kmsan vmalloc Shadow End"}, 81 [KMSAN_VMALLOC_ORIGIN_START_NR] = {0, "Kmsan vmalloc Origins Start"}, 82 [KMSAN_VMALLOC_ORIGIN_END_NR] = {0, "Kmsan vmalloc Origins End"}, 83 [KMSAN_MODULES_SHADOW_START_NR] = {0, "Kmsan Modules Shadow Start"}, 84 [KMSAN_MODULES_SHADOW_END_NR] = {0, "Kmsan Modules Shadow End"}, 85 [KMSAN_MODULES_ORIGIN_START_NR] = {0, "Kmsan Modules Origins Start"}, 86 [KMSAN_MODULES_ORIGIN_END_NR] = {0, "Kmsan Modules Origins End"}, 87 #endif 88 [MODULES_NR] = {0, "Modules Area Start"}, 89 [MODULES_END_NR] = {0, "Modules Area End"}, 90 [ABS_LOWCORE_NR] = {0, "Lowcore Area Start"}, 91 [ABS_LOWCORE_END_NR] = {0, "Lowcore Area End"}, 92 [MEMCPY_REAL_NR] = {0, "Real Memory Copy Area Start"}, 93 [MEMCPY_REAL_END_NR] = {0, "Real Memory Copy Area End"}, 94 #ifdef CONFIG_KASAN 95 [KASAN_SHADOW_START_NR] = {KASAN_SHADOW_START, "Kasan Shadow Start"}, 96 [KASAN_SHADOW_END_NR] = {KASAN_SHADOW_END, "Kasan Shadow End"}, 97 #endif 98 { -1, NULL } 99 }; 100 101 struct pg_state { 102 struct ptdump_state ptdump; 103 struct seq_file *seq; 104 int level; 105 unsigned int current_prot; 106 bool check_wx; 107 unsigned long wx_pages; 108 unsigned long start_address; 109 const struct addr_marker *marker; 110 }; 111 112 #define pt_dump_seq_printf(m, fmt, args...) \ 113 ({ \ 114 struct seq_file *__m = (m); \ 115 \ 116 if (__m) \ 117 seq_printf(__m, fmt, ##args); \ 118 }) 119 120 #define pt_dump_seq_puts(m, fmt) \ 121 ({ \ 122 struct seq_file *__m = (m); \ 123 \ 124 if (__m) \ 125 seq_printf(__m, fmt); \ 126 }) 127 128 static void print_prot(struct seq_file *m, unsigned int pr, int level) 129 { 130 static const char * const level_name[] = 131 { "ASCE", "PGD", "PUD", "PMD", "PTE" }; 132 133 pt_dump_seq_printf(m, "%s ", level_name[level]); 134 if (pr & _PAGE_INVALID) { 135 pt_dump_seq_printf(m, "I\n"); 136 return; 137 } 138 pt_dump_seq_puts(m, (pr & _PAGE_PROTECT) ? "RO " : "RW "); 139 pt_dump_seq_puts(m, (pr & _PAGE_NOEXEC) ? "NX\n" : "X\n"); 140 } 141 142 static void note_prot_wx(struct pg_state *st, unsigned long addr) 143 { 144 if (!st->check_wx) 145 return; 146 if (st->current_prot & _PAGE_INVALID) 147 return; 148 if (st->current_prot & _PAGE_PROTECT) 149 return; 150 if (st->current_prot & _PAGE_NOEXEC) 151 return; 152 /* 153 * The first lowcore page is W+X if spectre mitigations are using 154 * trampolines or the BEAR enhancements facility is not installed, 155 * in which case we have two lpswe instructions in lowcore that need 156 * to be executable. 157 */ 158 if (addr == PAGE_SIZE && (nospec_uses_trampoline() || !static_key_enabled(&cpu_has_bear))) 159 return; 160 WARN_ONCE(IS_ENABLED(CONFIG_DEBUG_WX), 161 "s390/mm: Found insecure W+X mapping at address %pS\n", 162 (void *)st->start_address); 163 st->wx_pages += (addr - st->start_address) / PAGE_SIZE; 164 } 165 166 static void note_page(struct ptdump_state *pt_st, unsigned long addr, int level, u64 val) 167 { 168 int width = sizeof(unsigned long) * 2; 169 static const char units[] = "KMGTPE"; 170 const char *unit = units; 171 unsigned long delta; 172 struct pg_state *st; 173 struct seq_file *m; 174 unsigned int prot; 175 176 st = container_of(pt_st, struct pg_state, ptdump); 177 m = st->seq; 178 prot = val & (_PAGE_PROTECT | _PAGE_NOEXEC); 179 if (level == 4 && (val & _PAGE_INVALID)) 180 prot = _PAGE_INVALID; 181 /* For pmd_none() & friends val gets passed as zero. */ 182 if (level != 4 && !val) 183 prot = _PAGE_INVALID; 184 /* Final flush from generic code. */ 185 if (level == -1) 186 addr = max_addr; 187 if (st->level == -1) { 188 pt_dump_seq_printf(m, "---[ %s ]---\n", st->marker->name); 189 st->start_address = addr; 190 st->current_prot = prot; 191 st->level = level; 192 } else if (prot != st->current_prot || level != st->level || 193 addr >= st->marker[1].start_address) { 194 note_prot_wx(st, addr); 195 pt_dump_seq_printf(m, "0x%0*lx-0x%0*lx ", 196 width, st->start_address, 197 width, addr); 198 delta = (addr - st->start_address) >> 10; 199 while (!(delta & 0x3ff) && unit[1]) { 200 delta >>= 10; 201 unit++; 202 } 203 pt_dump_seq_printf(m, "%9lu%c ", delta, *unit); 204 print_prot(m, st->current_prot, st->level); 205 while (addr >= st->marker[1].start_address) { 206 st->marker++; 207 pt_dump_seq_printf(m, "---[ %s ]---\n", st->marker->name); 208 } 209 st->start_address = addr; 210 st->current_prot = prot; 211 st->level = level; 212 } 213 } 214 215 bool ptdump_check_wx(void) 216 { 217 struct pg_state st = { 218 .ptdump = { 219 .note_page = note_page, 220 .range = (struct ptdump_range[]) { 221 {.start = 0, .end = max_addr}, 222 {.start = 0, .end = 0}, 223 } 224 }, 225 .seq = NULL, 226 .level = -1, 227 .current_prot = 0, 228 .check_wx = true, 229 .wx_pages = 0, 230 .start_address = 0, 231 .marker = (struct addr_marker[]) { 232 { .start_address = 0, .name = NULL}, 233 { .start_address = -1, .name = NULL}, 234 }, 235 }; 236 237 if (!MACHINE_HAS_NX) 238 return true; 239 ptdump_walk_pgd(&st.ptdump, &init_mm, NULL); 240 if (st.wx_pages) { 241 pr_warn("Checked W+X mappings: FAILED, %lu W+X pages found\n", st.wx_pages); 242 243 return false; 244 } else { 245 pr_info("Checked W+X mappings: passed, no %sW+X pages found\n", 246 (nospec_uses_trampoline() || !static_key_enabled(&cpu_has_bear)) ? 247 "unexpected " : ""); 248 249 return true; 250 } 251 } 252 253 #ifdef CONFIG_PTDUMP_DEBUGFS 254 static int ptdump_show(struct seq_file *m, void *v) 255 { 256 struct pg_state st = { 257 .ptdump = { 258 .note_page = note_page, 259 .range = (struct ptdump_range[]) { 260 {.start = 0, .end = max_addr}, 261 {.start = 0, .end = 0}, 262 } 263 }, 264 .seq = m, 265 .level = -1, 266 .current_prot = 0, 267 .check_wx = false, 268 .wx_pages = 0, 269 .start_address = 0, 270 .marker = address_markers, 271 }; 272 273 get_online_mems(); 274 mutex_lock(&cpa_mutex); 275 ptdump_walk_pgd(&st.ptdump, &init_mm, NULL); 276 mutex_unlock(&cpa_mutex); 277 put_online_mems(); 278 return 0; 279 } 280 DEFINE_SHOW_ATTRIBUTE(ptdump); 281 #endif /* CONFIG_PTDUMP_DEBUGFS */ 282 283 /* 284 * Heapsort from lib/sort.c is not a stable sorting algorithm, do a simple 285 * insertion sort to preserve the original order of markers with the same 286 * start address. 287 */ 288 static void sort_address_markers(void) 289 { 290 struct addr_marker tmp; 291 int i, j; 292 293 for (i = 1; i < ARRAY_SIZE(address_markers) - 1; i++) { 294 tmp = address_markers[i]; 295 for (j = i - 1; j >= 0 && address_markers[j].start_address > tmp.start_address; j--) 296 address_markers[j + 1] = address_markers[j]; 297 address_markers[j + 1] = tmp; 298 } 299 } 300 301 static int pt_dump_init(void) 302 { 303 #ifdef CONFIG_KFENCE 304 unsigned long kfence_start = (unsigned long)__kfence_pool; 305 #endif 306 /* 307 * Figure out the maximum virtual address being accessible with the 308 * kernel ASCE. We need this to keep the page table walker functions 309 * from accessing non-existent entries. 310 */ 311 max_addr = (get_lowcore()->kernel_asce.val & _REGION_ENTRY_TYPE_MASK) >> 2; 312 max_addr = 1UL << (max_addr * 11 + 31); 313 address_markers[IDENTITY_AFTER_END_NR].start_address = ident_map_size; 314 address_markers[AMODE31_START_NR].start_address = (unsigned long)__samode31; 315 address_markers[AMODE31_END_NR].start_address = (unsigned long)__eamode31; 316 address_markers[MODULES_NR].start_address = MODULES_VADDR; 317 address_markers[MODULES_END_NR].start_address = MODULES_END; 318 address_markers[ABS_LOWCORE_NR].start_address = __abs_lowcore; 319 address_markers[ABS_LOWCORE_END_NR].start_address = __abs_lowcore + ABS_LOWCORE_MAP_SIZE; 320 address_markers[MEMCPY_REAL_NR].start_address = __memcpy_real_area; 321 address_markers[MEMCPY_REAL_END_NR].start_address = __memcpy_real_area + MEMCPY_REAL_SIZE; 322 address_markers[VMEMMAP_NR].start_address = (unsigned long) vmemmap; 323 address_markers[VMEMMAP_END_NR].start_address = (unsigned long)vmemmap + vmemmap_size; 324 address_markers[VMALLOC_NR].start_address = VMALLOC_START; 325 address_markers[VMALLOC_END_NR].start_address = VMALLOC_END; 326 #ifdef CONFIG_KFENCE 327 address_markers[KFENCE_START_NR].start_address = kfence_start; 328 address_markers[KFENCE_END_NR].start_address = kfence_start + KFENCE_POOL_SIZE; 329 #endif 330 #ifdef CONFIG_KMSAN 331 address_markers[KMSAN_VMALLOC_SHADOW_START_NR].start_address = KMSAN_VMALLOC_SHADOW_START; 332 address_markers[KMSAN_VMALLOC_SHADOW_END_NR].start_address = KMSAN_VMALLOC_SHADOW_END; 333 address_markers[KMSAN_VMALLOC_ORIGIN_START_NR].start_address = KMSAN_VMALLOC_ORIGIN_START; 334 address_markers[KMSAN_VMALLOC_ORIGIN_END_NR].start_address = KMSAN_VMALLOC_ORIGIN_END; 335 address_markers[KMSAN_MODULES_SHADOW_START_NR].start_address = KMSAN_MODULES_SHADOW_START; 336 address_markers[KMSAN_MODULES_SHADOW_END_NR].start_address = KMSAN_MODULES_SHADOW_END; 337 address_markers[KMSAN_MODULES_ORIGIN_START_NR].start_address = KMSAN_MODULES_ORIGIN_START; 338 address_markers[KMSAN_MODULES_ORIGIN_END_NR].start_address = KMSAN_MODULES_ORIGIN_END; 339 #endif 340 sort_address_markers(); 341 #ifdef CONFIG_PTDUMP_DEBUGFS 342 debugfs_create_file("kernel_page_tables", 0400, NULL, NULL, &ptdump_fops); 343 #endif /* CONFIG_PTDUMP_DEBUGFS */ 344 return 0; 345 } 346 device_initcall(pt_dump_init); 347