arch/powerpc/mm/ptdump/ptdump.c

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

DEFINITIONS
