arch/x86/mm/dump_pagetables.c

   1 /*
   2  * Debug helper to dump the current kernel pagetables of the system
   3  * so that we can see what the various memory ranges are set to.
   4  *
   5  * (C) Copyright 2008 Intel Corporation
   6  *
   7  * Author: Arjan van de Ven <arjan@linux.intel.com>
   8  *
   9  * This program is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU General Public License
  11  * as published by the Free Software Foundation; version 2
  12  * of the License.
  13  */
  14
  15 #include <linux/debugfs.h>
  16 #include <linux/mm.h>
  17 #include <linux/module.h>
  18 #include <linux/seq_file.h>
  19
  20 #include <asm/pgtable.h>
  21
  22 /*
  23  * The dumper groups pagetable entries of the same type into one, and for
  24  * that it needs to keep some state when walking, and flush this state
  25  * when a "break" in the continuity is found.
  26  */
  27 struct pg_state {
  28         int level;
  29         pgprot_t current_prot;
  30         unsigned long start_address;
  31         unsigned long current_address;
  32         const struct addr_marker *marker;
  33         unsigned long lines;
  34 };
  35
  36 struct addr_marker {
  37         unsigned long start_address;
  38         const char *name;
  39         unsigned long max_lines;
  40 };
  41
  42 /* indices for address_markers; keep sync'd w/ address_markers below */
  43 enum address_markers_idx {
  44         USER_SPACE_NR = 0,
  45 #ifdef CONFIG_X86_64
  46         KERNEL_SPACE_NR,
  47         LOW_KERNEL_NR,
  48         VMALLOC_START_NR,
  49         VMEMMAP_START_NR,
  50         ESPFIX_START_NR,
  51         HIGH_KERNEL_NR,
  52         MODULES_VADDR_NR,
  53         MODULES_END_NR,
  54 #else
  55         KERNEL_SPACE_NR,
  56         VMALLOC_START_NR,
  57         VMALLOC_END_NR,
  58 # ifdef CONFIG_HIGHMEM
  59         PKMAP_BASE_NR,
  60 # endif
  61         FIXADDR_START_NR,
  62 #endif
  63 };
  64
  65 /* Address space markers hints */
  66 static struct addr_marker address_markers[] = {
  67         { 0, "User Space" },
  68 #ifdef CONFIG_X86_64
  69         { 0x8000000000000000UL, "Kernel Space" },
  70         { PAGE_OFFSET,          "Low Kernel Mapping" },
  71         { VMALLOC_START,        "vmalloc() Area" },
  72         { VMEMMAP_START,        "Vmemmap" },
  73         { ESPFIX_BASE_ADDR,     "ESPfix Area", 16 },
  74         { __START_KERNEL_map,   "High Kernel Mapping" },
  75         { MODULES_VADDR,        "Modules" },
  76         { MODULES_END,          "End Modules" },
  77 #else
  78         { PAGE_OFFSET,          "Kernel Mapping" },
  79         { 0/* VMALLOC_START */, "vmalloc() Area" },
  80         { 0/*VMALLOC_END*/,     "vmalloc() End" },
  81 # ifdef CONFIG_HIGHMEM
  82         { 0/*PKMAP_BASE*/,      "Persisent kmap() Area" },
  83 # endif
  84         { 0/*FIXADDR_START*/,   "Fixmap Area" },
  85 #endif
  86         { -1, NULL }            /* End of list */
  87 };
  88
  89 /* Multipliers for offsets within the PTEs */
  90 #define PTE_LEVEL_MULT (PAGE_SIZE)
  91 #define PMD_LEVEL_MULT (PTRS_PER_PTE * PTE_LEVEL_MULT)
  92 #define PUD_LEVEL_MULT (PTRS_PER_PMD * PMD_LEVEL_MULT)
  93 #define PGD_LEVEL_MULT (PTRS_PER_PUD * PUD_LEVEL_MULT)
  94
  95 /*
  96  * Print a readable form of a pgprot_t to the seq_file
  97  */
  98 static void printk_prot(struct seq_file *m, pgprot_t prot, int level)
  99 {
 100         pgprotval_t pr = pgprot_val(prot);
 101         static const char * const level_name[] =
 102                 { "cr3", "pgd", "pud", "pmd", "pte" };
 103
 104         if (!pgprot_val(prot)) {
 105                 /* Not present */
 106                 seq_printf(m, "                          ");
 107         } else {
 108                 if (pr & _PAGE_USER)
 109                         seq_printf(m, "USR ");
 110                 else
 111                         seq_printf(m, "    ");
 112                 if (pr & _PAGE_RW)
 113                         seq_printf(m, "RW ");
 114                 else
 115                         seq_printf(m, "ro ");
 116                 if (pr & _PAGE_PWT)
 117                         seq_printf(m, "PWT ");
 118                 else
 119                         seq_printf(m, "    ");
 120                 if (pr & _PAGE_PCD)
 121                         seq_printf(m, "PCD ");
 122                 else
 123                         seq_printf(m, "    ");
 124
 125                 /* Bit 9 has a different meaning on level 3 vs 4 */
 126                 if (level <= 3) {
 127                         if (pr & _PAGE_PSE)
 128                                 seq_printf(m, "PSE ");
 129                         else
 130                                 seq_printf(m, "    ");
 131                 } else {
 132                         if (pr & _PAGE_PAT)
 133                                 seq_printf(m, "pat ");
 134                         else
 135                                 seq_printf(m, "    ");
 136                 }
 137                 if (pr & _PAGE_GLOBAL)
 138                         seq_printf(m, "GLB ");
 139                 else
 140                         seq_printf(m, "    ");
 141                 if (pr & _PAGE_NX)
 142                         seq_printf(m, "NX ");
 143                 else
 144                         seq_printf(m, "x  ");
 145         }
 146         seq_printf(m, "%s\n", level_name[level]);
 147 }
 148
 149 /*
 150  * On 64 bits, sign-extend the 48 bit address to 64 bit
 151  */
 152 static unsigned long normalize_addr(unsigned long u)
 153 {
 154 #ifdef CONFIG_X86_64
 155         return (signed long)(u << 16) >> 16;
 156 #else
 157         return u;
 158 #endif
 159 }
 160
 161 /*
 162  * This function gets called on a break in a continuous series
 163  * of PTE entries; the next one is different so we need to
 164  * print what we collected so far.
 165  */
 166 static void note_page(struct seq_file *m, struct pg_state *st,
 167                       pgprot_t new_prot, int level)
 168 {
 169         pgprotval_t prot, cur;
 170         static const char units[] = "BKMGTPE";
 171
 172         /*
 173          * If we have a "break" in the series, we need to flush the state that
 174          * we have now. "break" is either changing perms, levels or
 175          * address space marker.
 176          */
 177         prot = pgprot_val(new_prot) & PTE_FLAGS_MASK;
 178         cur = pgprot_val(st->current_prot) & PTE_FLAGS_MASK;
 179
 180         if (!st->level) {
 181                 /* First entry */
 182                 st->current_prot = new_prot;
 183                 st->level = level;
 184                 st->marker = address_markers;
 185                 st->lines = 0;
 186                 seq_printf(m, "---[ %s ]---\n", st->marker->name);
 187         } else if (prot != cur || level != st->level ||
 188                    st->current_address >= st->marker[1].start_address) {
 189                 const char *unit = units;
 190                 unsigned long delta;
 191                 int width = sizeof(unsigned long) * 2;
 192
 193                 /*
 194                  * Now print the actual finished series
 195                  */
 196                 if (!st->marker->max_lines ||
 197                     st->lines < st->marker->max_lines) {
 198                         seq_printf(m, "0x%0*lx-0x%0*lx   ",
 199                                    width, st->start_address,
 200                                    width, st->current_address);
 201
 202                         delta = (st->current_address - st->start_address) >> 10;
 203                         while (!(delta & 1023) && unit[1]) {
 204                                 delta >>= 10;
 205                                 unit++;
 206                         }
 207                         seq_printf(m, "%9lu%c ", delta, *unit);
 208                         printk_prot(m, st->current_prot, st->level);
 209                 }
 210                 st->lines++;
 211
 212                 /*
 213                  * We print markers for special areas of address space,
 214                  * such as the start of vmalloc space etc.
 215                  * This helps in the interpretation.
 216                  */
 217                 if (st->current_address >= st->marker[1].start_address) {
 218                         st->marker++;
 219                         seq_printf(m, "---[ %s ]---\n", st->marker->name);
 220                 }
 221
 222                 st->start_address = st->current_address;
 223                 st->current_prot = new_prot;
 224                 st->level = level;
 225         }
 226 }
 227
 228 static void walk_pte_level(struct seq_file *m, struct pg_state *st, pmd_t addr,
 229                                                         unsigned long P)
 230 {
 231         int i;
 232         pte_t *start;
 233
 234         start = (pte_t *) pmd_page_vaddr(addr);
 235         for (i = 0; i < PTRS_PER_PTE; i++) {
 236                 pgprot_t prot = pte_pgprot(*start);
 237
 238                 st->current_address = normalize_addr(P + i * PTE_LEVEL_MULT);
 239                 note_page(m, st, prot, 4);
 240                 start++;
 241         }
 242 }
 243
 244 #if PTRS_PER_PMD > 1
 245
 246 static void walk_pmd_level(struct seq_file *m, struct pg_state *st, pud_t addr,
 247                                                         unsigned long P)
 248 {
 249         int i;
 250         pmd_t *start;
 251
 252         start = (pmd_t *) pud_page_vaddr(addr);
 253         for (i = 0; i < PTRS_PER_PMD; i++) {
 254                 st->current_address = normalize_addr(P + i * PMD_LEVEL_MULT);
 255                 if (!pmd_none(*start)) {
 256                         pgprotval_t prot = pmd_val(*start) & PTE_FLAGS_MASK;
 257
 258                         if (pmd_large(*start) || !pmd_present(*start))
 259                                 note_page(m, st, __pgprot(prot), 3);
 260                         else
 261                                 walk_pte_level(m, st, *start,
 262                                                P + i * PMD_LEVEL_MULT);
 263                 } else
 264                         note_page(m, st, __pgprot(0), 3);
 265                 start++;
 266         }
 267 }
 268
 269 #else
 270 #define walk_pmd_level(m,s,a,p) walk_pte_level(m,s,__pmd(pud_val(a)),p)
 271 #define pud_large(a) pmd_large(__pmd(pud_val(a)))
 272 #define pud_none(a)  pmd_none(__pmd(pud_val(a)))
 273 #endif
 274
 275 #if PTRS_PER_PUD > 1
 276
 277 static void walk_pud_level(struct seq_file *m, struct pg_state *st, pgd_t addr,
 278                                                         unsigned long P)
 279 {
 280         int i;
 281         pud_t *start;
 282
 283         start = (pud_t *) pgd_page_vaddr(addr);
 284
 285         for (i = 0; i < PTRS_PER_PUD; i++) {
 286                 st->current_address = normalize_addr(P + i * PUD_LEVEL_MULT);
 287                 if (!pud_none(*start)) {
 288                         pgprotval_t prot = pud_val(*start) & PTE_FLAGS_MASK;
 289
 290                         if (pud_large(*start) || !pud_present(*start))
 291                                 note_page(m, st, __pgprot(prot), 2);
 292                         else
 293                                 walk_pmd_level(m, st, *start,
 294                                                P + i * PUD_LEVEL_MULT);
 295                 } else
 296                         note_page(m, st, __pgprot(0), 2);
 297
 298                 start++;
 299         }
 300 }
 301
 302 #else
 303 #define walk_pud_level(m,s,a,p) walk_pmd_level(m,s,__pud(pgd_val(a)),p)
 304 #define pgd_large(a) pud_large(__pud(pgd_val(a)))
 305 #define pgd_none(a)  pud_none(__pud(pgd_val(a)))
 306 #endif
 307
 308 static void walk_pgd_level(struct seq_file *m)
 309 {
 310 #ifdef CONFIG_X86_64
 311         pgd_t *start = (pgd_t *) &init_level4_pgt;
 312 #else
 313         pgd_t *start = swapper_pg_dir;
 314 #endif
 315         int i;
 316         struct pg_state st;
 317
 318         memset(&st, 0, sizeof(st));
 319
 320         for (i = 0; i < PTRS_PER_PGD; i++) {
 321                 st.current_address = normalize_addr(i * PGD_LEVEL_MULT);
 322                 if (!pgd_none(*start)) {
 323                         pgprotval_t prot = pgd_val(*start) & PTE_FLAGS_MASK;
 324
 325                         if (pgd_large(*start) || !pgd_present(*start))
 326                                 note_page(m, &st, __pgprot(prot), 1);
 327                         else
 328                                 walk_pud_level(m, &st, *start,
 329                                                i * PGD_LEVEL_MULT);
 330                 } else
 331                         note_page(m, &st, __pgprot(0), 1);
 332
 333                 start++;
 334         }
 335
 336         /* Flush out the last page */
 337         st.current_address = normalize_addr(PTRS_PER_PGD*PGD_LEVEL_MULT);
 338         note_page(m, &st, __pgprot(0), 0);
 339 }
 340
 341 static int ptdump_show(struct seq_file *m, void *v)
 342 {
 343         walk_pgd_level(m);
 344         return 0;
 345 }
 346
 347 static int ptdump_open(struct inode *inode, struct file *filp)
 348 {
 349         return single_open(filp, ptdump_show, NULL);
 350 }
 351
 352 static const struct file_operations ptdump_fops = {
 353         .open           = ptdump_open,
 354         .read           = seq_read,
 355         .llseek         = seq_lseek,
 356         .release        = single_release,
 357 };
 358
 359 static int pt_dump_init(void)
 360 {
 361         struct dentry *pe;
 362
 363 #ifdef CONFIG_X86_32
 364         /* Not a compile-time constant on x86-32 */
 365         address_markers[VMALLOC_START_NR].start_address = VMALLOC_START;
 366         address_markers[VMALLOC_END_NR].start_address = VMALLOC_END;
 367 # ifdef CONFIG_HIGHMEM
 368         address_markers[PKMAP_BASE_NR].start_address = PKMAP_BASE;
 369 # endif
 370         address_markers[FIXADDR_START_NR].start_address = FIXADDR_START;
 371 #endif
 372
 373         pe = debugfs_create_file("kernel_page_tables", 0600, NULL, NULL,
 374                                  &ptdump_fops);
 375         if (!pe)
 376                 return -ENOMEM;
 377
 378         return 0;
 379 }
 380
 381 __initcall(pt_dump_init);
 382 MODULE_LICENSE("GPL");
 383 MODULE_AUTHOR("Arjan van de Ven <arjan@linux.intel.com>");
 384 MODULE_DESCRIPTION("Kernel debugging helper that dumps pagetables");