1 // SPDX-License-Identifier: GPL-2.0-only 2 /* 3 * Copyright 2016, Rashmica Gupta, IBM Corp. 4 * 5 * This traverses the kernel pagetables and dumps the 6 * information about the used sections of memory to 7 * /sys/kernel/debug/kernel_pagetables. 8 * 9 * Derived from the arm64 implementation: 10 * Copyright (c) 2014, The Linux Foundation, Laura Abbott. 11 * (C) Copyright 2008 Intel Corporation, Arjan van de Ven. 12 */ 13 #include <linux/debugfs.h> 14 #include <linux/fs.h> 15 #include <linux/hugetlb.h> 16 #include <linux/io.h> 17 #include <linux/mm.h> 18 #include <linux/highmem.h> 19 #include <linux/sched.h> 20 #include <linux/seq_file.h> 21 #include <asm/fixmap.h> 22 #include <linux/const.h> 23 #include <asm/page.h> 24 #include <asm/hugetlb.h> 25 26 #include <mm/mmu_decl.h> 27 28 #include "ptdump.h" 29 30 /* 31 * To visualise what is happening, 32 * 33 * - PTRS_PER_P** = how many entries there are in the corresponding P** 34 * - P**_SHIFT = how many bits of the address we use to index into the 35 * corresponding P** 36 * - P**_SIZE is how much memory we can access through the table - not the 37 * size of the table itself. 38 * P**={PGD, PUD, PMD, PTE} 39 * 40 * 41 * Each entry of the PGD points to a PUD. Each entry of a PUD points to a 42 * PMD. Each entry of a PMD points to a PTE. And every PTE entry points to 43 * a page. 44 * 45 * In the case where there are only 3 levels, the PUD is folded into the 46 * PGD: every PUD has only one entry which points to the PMD. 47 * 48 * The page dumper groups page table entries of the same type into a single 49 * description. It uses pg_state to track the range information while 50 * iterating over the PTE entries. When the continuity is broken it then 51 * dumps out a description of the range - ie PTEs that are virtually contiguous 52 * with the same PTE flags are chunked together. This is to make it clear how 53 * different areas of the kernel virtual memory are used. 54 * 55 */ 56 struct pg_state { 57 struct seq_file *seq; 58 const struct addr_marker *marker; 59 unsigned long start_address; 60 unsigned long start_pa; 61 unsigned int level; 62 u64 current_flags; 63 bool check_wx; 64 unsigned long wx_pages; 65 }; 66 67 struct addr_marker { 68 unsigned long start_address; 69 const char *name; 70 }; 71 72 static struct addr_marker address_markers[] = { 73 { 0, "Start of kernel VM" }, 74 #ifdef MODULES_VADDR 75 { 0, "modules start" }, 76 { 0, "modules end" }, 77 #endif 78 { 0, "vmalloc() Area" }, 79 { 0, "vmalloc() End" }, 80 #ifdef CONFIG_PPC64 81 { 0, "isa I/O start" }, 82 { 0, "isa I/O end" }, 83 { 0, "phb I/O start" }, 84 { 0, "phb I/O end" }, 85 { 0, "I/O remap start" }, 86 { 0, "I/O remap end" }, 87 { 0, "vmemmap start" }, 88 #else 89 { 0, "Early I/O remap start" }, 90 { 0, "Early I/O remap end" }, 91 #ifdef CONFIG_HIGHMEM 92 { 0, "Highmem PTEs start" }, 93 { 0, "Highmem PTEs end" }, 94 #endif 95 { 0, "Fixmap start" }, 96 { 0, "Fixmap end" }, 97 #endif 98 #ifdef CONFIG_KASAN 99 { 0, "kasan shadow mem start" }, 100 { 0, "kasan shadow mem end" }, 101 #endif 102 { -1, NULL }, 103 }; 104 105 #define pt_dump_seq_printf(m, fmt, args...) \ 106 ({ \ 107 if (m) \ 108 seq_printf(m, fmt, ##args); \ 109 }) 110 111 #define pt_dump_seq_putc(m, c) \ 112 ({ \ 113 if (m) \ 114 seq_putc(m, c); \ 115 }) 116 117 void pt_dump_size(struct seq_file *m, unsigned long size) 118 { 119 static const char units[] = "KMGTPE"; 120 const char *unit = units; 121 122 /* Work out what appropriate unit to use */ 123 while (!(size & 1023) && unit[1]) { 124 size >>= 10; 125 unit++; 126 } 127 pt_dump_seq_printf(m, "%9lu%c ", size, *unit); 128 } 129 130 static void dump_flag_info(struct pg_state *st, const struct flag_info 131 *flag, u64 pte, int num) 132 { 133 unsigned int i; 134 135 for (i = 0; i < num; i++, flag++) { 136 const char *s = NULL; 137 u64 val; 138 139 /* flag not defined so don't check it */ 140 if (flag->mask == 0) 141 continue; 142 /* Some 'flags' are actually values */ 143 if (flag->is_val) { 144 val = pte & flag->val; 145 if (flag->shift) 146 val = val >> flag->shift; 147 pt_dump_seq_printf(st->seq, " %s:%llx", flag->set, val); 148 } else { 149 if ((pte & flag->mask) == flag->val) 150 s = flag->set; 151 else 152 s = flag->clear; 153 if (s) 154 pt_dump_seq_printf(st->seq, " %s", s); 155 } 156 st->current_flags &= ~flag->mask; 157 } 158 if (st->current_flags != 0) 159 pt_dump_seq_printf(st->seq, " unknown flags:%llx", st->current_flags); 160 } 161 162 static void dump_addr(struct pg_state *st, unsigned long addr) 163 { 164 #ifdef CONFIG_PPC64 165 #define REG "0x%016lx" 166 #else 167 #define REG "0x%08lx" 168 #endif 169 170 pt_dump_seq_printf(st->seq, REG "-" REG " ", st->start_address, addr - 1); 171 pt_dump_seq_printf(st->seq, " " REG " ", st->start_pa); 172 pt_dump_size(st->seq, (addr - st->start_address) >> 10); 173 } 174 175 static void note_prot_wx(struct pg_state *st, unsigned long addr) 176 { 177 pte_t pte = __pte(st->current_flags); 178 179 if (!IS_ENABLED(CONFIG_PPC_DEBUG_WX) || !st->check_wx) 180 return; 181 182 if (!pte_write(pte) || !pte_exec(pte)) 183 return; 184 185 WARN_ONCE(1, "powerpc/mm: Found insecure W+X mapping at address %p/%pS\n", 186 (void *)st->start_address, (void *)st->start_address); 187 188 st->wx_pages += (addr - st->start_address) / PAGE_SIZE; 189 } 190 191 static void note_page_update_state(struct pg_state *st, unsigned long addr, 192 unsigned int level, u64 val, unsigned long page_size) 193 { 194 u64 flag = val & pg_level[level].mask; 195 u64 pa = val & PTE_RPN_MASK; 196 197 st->level = level; 198 st->current_flags = flag; 199 st->start_address = addr; 200 st->start_pa = pa; 201 202 while (addr >= st->marker[1].start_address) { 203 st->marker++; 204 pt_dump_seq_printf(st->seq, "---[ %s ]---\n", st->marker->name); 205 } 206 } 207 208 static void note_page(struct pg_state *st, unsigned long addr, 209 unsigned int level, u64 val, unsigned long page_size) 210 { 211 u64 flag = val & pg_level[level].mask; 212 213 /* At first no level is set */ 214 if (!st->level) { 215 pt_dump_seq_printf(st->seq, "---[ %s ]---\n", st->marker->name); 216 note_page_update_state(st, addr, level, val, page_size); 217 /* 218 * Dump the section of virtual memory when: 219 * - the PTE flags from one entry to the next differs. 220 * - we change levels in the tree. 221 * - the address is in a different section of memory and is thus 222 * used for a different purpose, regardless of the flags. 223 */ 224 } else if (flag != st->current_flags || level != st->level || 225 addr >= st->marker[1].start_address) { 226 227 /* Check the PTE flags */ 228 if (st->current_flags) { 229 note_prot_wx(st, addr); 230 dump_addr(st, addr); 231 232 /* Dump all the flags */ 233 if (pg_level[st->level].flag) 234 dump_flag_info(st, pg_level[st->level].flag, 235 st->current_flags, 236 pg_level[st->level].num); 237 238 pt_dump_seq_putc(st->seq, '\n'); 239 } 240 241 /* 242 * Address indicates we have passed the end of the 243 * current section of virtual memory 244 */ 245 note_page_update_state(st, addr, level, val, page_size); 246 } 247 } 248 249 static void walk_pte(struct pg_state *st, pmd_t *pmd, unsigned long start) 250 { 251 pte_t *pte = pte_offset_kernel(pmd, 0); 252 unsigned long addr; 253 unsigned int i; 254 255 for (i = 0; i < PTRS_PER_PTE; i++, pte++) { 256 addr = start + i * PAGE_SIZE; 257 note_page(st, addr, 4, pte_val(*pte), PAGE_SIZE); 258 259 } 260 } 261 262 static void walk_hugepd(struct pg_state *st, hugepd_t *phpd, unsigned long start, 263 int pdshift, int level) 264 { 265 #ifdef CONFIG_ARCH_HAS_HUGEPD 266 unsigned int i; 267 int shift = hugepd_shift(*phpd); 268 int ptrs_per_hpd = pdshift - shift > 0 ? 1 << (pdshift - shift) : 1; 269 270 if (start & ((1 << shift) - 1)) 271 return; 272 273 for (i = 0; i < ptrs_per_hpd; i++) { 274 unsigned long addr = start + (i << shift); 275 pte_t *pte = hugepte_offset(*phpd, addr, pdshift); 276 277 note_page(st, addr, level + 1, pte_val(*pte), 1 << shift); 278 } 279 #endif 280 } 281 282 static void walk_pmd(struct pg_state *st, pud_t *pud, unsigned long start) 283 { 284 pmd_t *pmd = pmd_offset(pud, 0); 285 unsigned long addr; 286 unsigned int i; 287 288 for (i = 0; i < PTRS_PER_PMD; i++, pmd++) { 289 addr = start + i * PMD_SIZE; 290 if (!pmd_none(*pmd) && !pmd_is_leaf(*pmd)) 291 /* pmd exists */ 292 walk_pte(st, pmd, addr); 293 else 294 note_page(st, addr, 3, pmd_val(*pmd), PMD_SIZE); 295 } 296 } 297 298 static void walk_pud(struct pg_state *st, p4d_t *p4d, unsigned long start) 299 { 300 pud_t *pud = pud_offset(p4d, 0); 301 unsigned long addr; 302 unsigned int i; 303 304 for (i = 0; i < PTRS_PER_PUD; i++, pud++) { 305 addr = start + i * PUD_SIZE; 306 if (!pud_none(*pud) && !pud_is_leaf(*pud)) 307 /* pud exists */ 308 walk_pmd(st, pud, addr); 309 else 310 note_page(st, addr, 2, pud_val(*pud), PUD_SIZE); 311 } 312 } 313 314 static void walk_pagetables(struct pg_state *st) 315 { 316 unsigned int i; 317 unsigned long addr = st->start_address & PGDIR_MASK; 318 pgd_t *pgd = pgd_offset_k(addr); 319 320 /* 321 * Traverse the linux pagetable structure and dump pages that are in 322 * the hash pagetable. 323 */ 324 for (i = pgd_index(addr); i < PTRS_PER_PGD; i++, pgd++, addr += PGDIR_SIZE) { 325 p4d_t *p4d = p4d_offset(pgd, 0); 326 327 if (p4d_none(*p4d) || p4d_is_leaf(*p4d)) 328 note_page(st, addr, 1, p4d_val(*p4d), PGDIR_SIZE); 329 else if (is_hugepd(__hugepd(p4d_val(*p4d)))) 330 walk_hugepd(st, (hugepd_t *)p4d, addr, PGDIR_SHIFT, 1); 331 else 332 /* p4d exists */ 333 walk_pud(st, p4d, addr); 334 } 335 } 336 337 static void populate_markers(void) 338 { 339 int i = 0; 340 341 #ifdef CONFIG_PPC64 342 address_markers[i++].start_address = PAGE_OFFSET; 343 #else 344 address_markers[i++].start_address = TASK_SIZE; 345 #endif 346 #ifdef MODULES_VADDR 347 address_markers[i++].start_address = MODULES_VADDR; 348 address_markers[i++].start_address = MODULES_END; 349 #endif 350 address_markers[i++].start_address = VMALLOC_START; 351 address_markers[i++].start_address = VMALLOC_END; 352 #ifdef CONFIG_PPC64 353 address_markers[i++].start_address = ISA_IO_BASE; 354 address_markers[i++].start_address = ISA_IO_END; 355 address_markers[i++].start_address = PHB_IO_BASE; 356 address_markers[i++].start_address = PHB_IO_END; 357 address_markers[i++].start_address = IOREMAP_BASE; 358 address_markers[i++].start_address = IOREMAP_END; 359 /* What is the ifdef about? */ 360 #ifdef CONFIG_PPC_BOOK3S_64 361 address_markers[i++].start_address = H_VMEMMAP_START; 362 #else 363 address_markers[i++].start_address = VMEMMAP_BASE; 364 #endif 365 #else /* !CONFIG_PPC64 */ 366 address_markers[i++].start_address = ioremap_bot; 367 address_markers[i++].start_address = IOREMAP_TOP; 368 #ifdef CONFIG_HIGHMEM 369 address_markers[i++].start_address = PKMAP_BASE; 370 address_markers[i++].start_address = PKMAP_ADDR(LAST_PKMAP); 371 #endif 372 address_markers[i++].start_address = FIXADDR_START; 373 address_markers[i++].start_address = FIXADDR_TOP; 374 #ifdef CONFIG_KASAN 375 address_markers[i++].start_address = KASAN_SHADOW_START; 376 address_markers[i++].start_address = KASAN_SHADOW_END; 377 #endif 378 #endif /* CONFIG_PPC64 */ 379 } 380 381 static int ptdump_show(struct seq_file *m, void *v) 382 { 383 struct pg_state st = { 384 .seq = m, 385 .marker = address_markers, 386 .start_address = IS_ENABLED(CONFIG_PPC64) ? PAGE_OFFSET : TASK_SIZE, 387 }; 388 389 #ifdef CONFIG_PPC64 390 if (!radix_enabled()) 391 st.start_address = KERN_VIRT_START; 392 #endif 393 394 /* Traverse kernel page tables */ 395 walk_pagetables(&st); 396 note_page(&st, 0, 0, 0, 0); 397 return 0; 398 } 399 400 401 static int ptdump_open(struct inode *inode, struct file *file) 402 { 403 return single_open(file, ptdump_show, NULL); 404 } 405 406 static const struct file_operations ptdump_fops = { 407 .open = ptdump_open, 408 .read = seq_read, 409 .llseek = seq_lseek, 410 .release = single_release, 411 }; 412 413 static void build_pgtable_complete_mask(void) 414 { 415 unsigned int i, j; 416 417 for (i = 0; i < ARRAY_SIZE(pg_level); i++) 418 if (pg_level[i].flag) 419 for (j = 0; j < pg_level[i].num; j++) 420 pg_level[i].mask |= pg_level[i].flag[j].mask; 421 } 422 423 #ifdef CONFIG_PPC_DEBUG_WX 424 void ptdump_check_wx(void) 425 { 426 struct pg_state st = { 427 .seq = NULL, 428 .marker = address_markers, 429 .check_wx = true, 430 .start_address = IS_ENABLED(CONFIG_PPC64) ? PAGE_OFFSET : TASK_SIZE, 431 }; 432 433 #ifdef CONFIG_PPC64 434 if (!radix_enabled()) 435 st.start_address = KERN_VIRT_START; 436 #endif 437 438 walk_pagetables(&st); 439 440 if (st.wx_pages) 441 pr_warn("Checked W+X mappings: FAILED, %lu W+X pages found\n", 442 st.wx_pages); 443 else 444 pr_info("Checked W+X mappings: passed, no W+X pages found\n"); 445 } 446 #endif 447 448 static int ptdump_init(void) 449 { 450 populate_markers(); 451 build_pgtable_complete_mask(); 452 debugfs_create_file("kernel_page_tables", 0400, NULL, NULL, 453 &ptdump_fops); 454 return 0; 455 } 456 device_initcall(ptdump_init); 457