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/sort.h> 7 #include <linux/mm.h> 8 #include <linux/kfence.h> 9 #include <linux/kasan.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 16 static unsigned long max_addr; 17 18 struct addr_marker { 19 int is_start; 20 unsigned long start_address; 21 unsigned long size; 22 const char *name; 23 }; 24 25 static struct addr_marker *markers; 26 static unsigned int markers_cnt; 27 28 struct pg_state { 29 struct ptdump_state ptdump; 30 struct seq_file *seq; 31 int level; 32 unsigned int current_prot; 33 bool check_wx; 34 unsigned long wx_pages; 35 unsigned long start_address; 36 const struct addr_marker *marker; 37 }; 38 39 #define pt_dump_seq_printf(m, fmt, args...) \ 40 ({ \ 41 struct seq_file *__m = (m); \ 42 \ 43 if (__m) \ 44 seq_printf(__m, fmt, ##args); \ 45 }) 46 47 #define pt_dump_seq_puts(m, fmt) \ 48 ({ \ 49 struct seq_file *__m = (m); \ 50 \ 51 if (__m) \ 52 seq_printf(__m, fmt); \ 53 }) 54 55 static void print_prot(struct seq_file *m, unsigned int pr, int level) 56 { 57 static const char * const level_name[] = 58 { "ASCE", "PGD", "PUD", "PMD", "PTE" }; 59 60 pt_dump_seq_printf(m, "%s ", level_name[level]); 61 if (pr & _PAGE_INVALID) { 62 pt_dump_seq_printf(m, "I\n"); 63 return; 64 } 65 pt_dump_seq_puts(m, (pr & _PAGE_PROTECT) ? "RO " : "RW "); 66 pt_dump_seq_puts(m, (pr & _PAGE_NOEXEC) ? "NX\n" : "X\n"); 67 } 68 69 static void note_prot_wx(struct pg_state *st, unsigned long addr) 70 { 71 if (!st->check_wx) 72 return; 73 if (st->current_prot & _PAGE_INVALID) 74 return; 75 if (st->current_prot & _PAGE_PROTECT) 76 return; 77 if (st->current_prot & _PAGE_NOEXEC) 78 return; 79 /* 80 * The first lowcore page is W+X if spectre mitigations are using 81 * trampolines or the BEAR enhancements facility is not installed, 82 * in which case we have two lpswe instructions in lowcore that need 83 * to be executable. 84 */ 85 if (addr == PAGE_SIZE && (nospec_uses_trampoline() || !static_key_enabled(&cpu_has_bear))) 86 return; 87 WARN_ONCE(IS_ENABLED(CONFIG_DEBUG_WX), 88 "s390/mm: Found insecure W+X mapping at address %pS\n", 89 (void *)st->start_address); 90 st->wx_pages += (addr - st->start_address) / PAGE_SIZE; 91 } 92 93 static void note_page_update_state(struct pg_state *st, unsigned long addr, unsigned int prot, int level) 94 { 95 struct seq_file *m = st->seq; 96 97 while (addr >= st->marker[1].start_address) { 98 st->marker++; 99 pt_dump_seq_printf(m, "---[ %s %s ]---\n", st->marker->name, 100 st->marker->is_start ? "Start" : "End"); 101 } 102 st->start_address = addr; 103 st->current_prot = prot; 104 st->level = level; 105 } 106 107 static void note_page(struct ptdump_state *pt_st, unsigned long addr, int level, u64 val) 108 { 109 int width = sizeof(unsigned long) * 2; 110 static const char units[] = "KMGTPE"; 111 const char *unit = units; 112 unsigned long delta; 113 struct pg_state *st; 114 struct seq_file *m; 115 unsigned int prot; 116 117 st = container_of(pt_st, struct pg_state, ptdump); 118 m = st->seq; 119 prot = val & (_PAGE_PROTECT | _PAGE_NOEXEC); 120 if (level == 4 && (val & _PAGE_INVALID)) 121 prot = _PAGE_INVALID; 122 /* For pmd_none() & friends val gets passed as zero. */ 123 if (level != 4 && !val) 124 prot = _PAGE_INVALID; 125 /* Final flush from generic code. */ 126 if (level == -1) 127 addr = max_addr; 128 if (st->level == -1) { 129 pt_dump_seq_puts(m, "---[ Kernel Virtual Address Space ]---\n"); 130 note_page_update_state(st, addr, prot, level); 131 } else if (prot != st->current_prot || level != st->level || 132 addr >= st->marker[1].start_address) { 133 note_prot_wx(st, addr); 134 pt_dump_seq_printf(m, "0x%0*lx-0x%0*lx ", 135 width, st->start_address, 136 width, addr); 137 delta = (addr - st->start_address) >> 10; 138 while (!(delta & 0x3ff) && unit[1]) { 139 delta >>= 10; 140 unit++; 141 } 142 pt_dump_seq_printf(m, "%9lu%c ", delta, *unit); 143 print_prot(m, st->current_prot, st->level); 144 note_page_update_state(st, addr, prot, level); 145 } 146 } 147 148 bool ptdump_check_wx(void) 149 { 150 struct pg_state st = { 151 .ptdump = { 152 .note_page = note_page, 153 .range = (struct ptdump_range[]) { 154 {.start = 0, .end = max_addr}, 155 {.start = 0, .end = 0}, 156 } 157 }, 158 .seq = NULL, 159 .level = -1, 160 .current_prot = 0, 161 .check_wx = true, 162 .wx_pages = 0, 163 .start_address = 0, 164 .marker = (struct addr_marker[]) { 165 { .start_address = 0, .name = NULL}, 166 { .start_address = -1, .name = NULL}, 167 }, 168 }; 169 170 if (!MACHINE_HAS_NX) 171 return true; 172 ptdump_walk_pgd(&st.ptdump, &init_mm, NULL); 173 if (st.wx_pages) { 174 pr_warn("Checked W+X mappings: FAILED, %lu W+X pages found\n", st.wx_pages); 175 176 return false; 177 } else { 178 pr_info("Checked W+X mappings: passed, no %sW+X pages found\n", 179 (nospec_uses_trampoline() || !static_key_enabled(&cpu_has_bear)) ? 180 "unexpected " : ""); 181 182 return true; 183 } 184 } 185 186 #ifdef CONFIG_PTDUMP_DEBUGFS 187 static int ptdump_show(struct seq_file *m, void *v) 188 { 189 struct pg_state st = { 190 .ptdump = { 191 .note_page = note_page, 192 .range = (struct ptdump_range[]) { 193 {.start = 0, .end = max_addr}, 194 {.start = 0, .end = 0}, 195 } 196 }, 197 .seq = m, 198 .level = -1, 199 .current_prot = 0, 200 .check_wx = false, 201 .wx_pages = 0, 202 .start_address = 0, 203 .marker = markers, 204 }; 205 206 get_online_mems(); 207 mutex_lock(&cpa_mutex); 208 ptdump_walk_pgd(&st.ptdump, &init_mm, NULL); 209 mutex_unlock(&cpa_mutex); 210 put_online_mems(); 211 return 0; 212 } 213 DEFINE_SHOW_ATTRIBUTE(ptdump); 214 #endif /* CONFIG_PTDUMP_DEBUGFS */ 215 216 static int ptdump_cmp(const void *a, const void *b) 217 { 218 const struct addr_marker *ama = a; 219 const struct addr_marker *amb = b; 220 221 if (ama->start_address > amb->start_address) 222 return 1; 223 if (ama->start_address < amb->start_address) 224 return -1; 225 /* 226 * If the start addresses of two markers are identical sort markers in an 227 * order that considers areas contained within other areas correctly. 228 */ 229 if (ama->is_start && amb->is_start) { 230 if (ama->size > amb->size) 231 return -1; 232 if (ama->size < amb->size) 233 return 1; 234 return 0; 235 } 236 if (!ama->is_start && !amb->is_start) { 237 if (ama->size > amb->size) 238 return 1; 239 if (ama->size < amb->size) 240 return -1; 241 return 0; 242 } 243 if (ama->is_start) 244 return 1; 245 if (amb->is_start) 246 return -1; 247 return 0; 248 } 249 250 static int add_marker(unsigned long start, unsigned long end, const char *name) 251 { 252 size_t oldsize, newsize; 253 254 oldsize = markers_cnt * sizeof(*markers); 255 newsize = oldsize + 2 * sizeof(*markers); 256 if (!oldsize) 257 markers = kvmalloc(newsize, GFP_KERNEL); 258 else 259 markers = kvrealloc(markers, newsize, GFP_KERNEL); 260 if (!markers) 261 goto error; 262 markers[markers_cnt].is_start = 1; 263 markers[markers_cnt].start_address = start; 264 markers[markers_cnt].size = end - start; 265 markers[markers_cnt].name = name; 266 markers_cnt++; 267 markers[markers_cnt].is_start = 0; 268 markers[markers_cnt].start_address = end; 269 markers[markers_cnt].size = end - start; 270 markers[markers_cnt].name = name; 271 markers_cnt++; 272 return 0; 273 error: 274 markers_cnt = 0; 275 return -ENOMEM; 276 } 277 278 static int pt_dump_init(void) 279 { 280 #ifdef CONFIG_KFENCE 281 unsigned long kfence_start = (unsigned long)__kfence_pool; 282 #endif 283 unsigned long lowcore = (unsigned long)get_lowcore(); 284 int rc; 285 286 /* 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 = (get_lowcore()->kernel_asce.val & _REGION_ENTRY_TYPE_MASK) >> 2; 292 max_addr = 1UL << (max_addr * 11 + 31); 293 /* start + end markers - must be added first */ 294 rc = add_marker(0, -1UL, NULL); 295 rc |= add_marker((unsigned long)_stext, (unsigned long)_end, "Kernel Image"); 296 rc |= add_marker(lowcore, lowcore + sizeof(struct lowcore), "Lowcore"); 297 rc |= add_marker(__identity_base, __identity_base + ident_map_size, "Identity Mapping"); 298 rc |= add_marker((unsigned long)__samode31, (unsigned long)__eamode31, "Amode31 Area"); 299 rc |= add_marker(MODULES_VADDR, MODULES_END, "Modules Area"); 300 rc |= add_marker(__abs_lowcore, __abs_lowcore + ABS_LOWCORE_MAP_SIZE, "Lowcore Area"); 301 rc |= add_marker(__memcpy_real_area, __memcpy_real_area + MEMCPY_REAL_SIZE, "Real Memory Copy Area"); 302 rc |= add_marker((unsigned long)vmemmap, (unsigned long)vmemmap + vmemmap_size, "vmemmap Area"); 303 rc |= add_marker(VMALLOC_START, VMALLOC_END, "vmalloc Area"); 304 #ifdef CONFIG_KFENCE 305 rc |= add_marker(kfence_start, kfence_start + KFENCE_POOL_SIZE, "KFence Pool"); 306 #endif 307 #ifdef CONFIG_KMSAN 308 rc |= add_marker(KMSAN_VMALLOC_SHADOW_START, KMSAN_VMALLOC_SHADOW_END, "Kmsan vmalloc Shadow"); 309 rc |= add_marker(KMSAN_VMALLOC_ORIGIN_START, KMSAN_VMALLOC_ORIGIN_END, "Kmsan vmalloc Origins"); 310 rc |= add_marker(KMSAN_MODULES_SHADOW_START, KMSAN_MODULES_SHADOW_END, "Kmsan Modules Shadow"); 311 rc |= add_marker(KMSAN_MODULES_ORIGIN_START, KMSAN_MODULES_ORIGIN_END, "Kmsan Modules Origins"); 312 #endif 313 #ifdef CONFIG_KASAN 314 rc |= add_marker(KASAN_SHADOW_START, KASAN_SHADOW_END, "Kasan Shadow"); 315 #endif 316 if (rc) 317 goto error; 318 sort(&markers[1], markers_cnt - 1, sizeof(*markers), ptdump_cmp, NULL); 319 #ifdef CONFIG_PTDUMP_DEBUGFS 320 debugfs_create_file("kernel_page_tables", 0400, NULL, NULL, &ptdump_fops); 321 #endif /* CONFIG_PTDUMP_DEBUGFS */ 322 return 0; 323 error: 324 kvfree(markers); 325 return -ENOMEM; 326 } 327 device_initcall(pt_dump_init); 328