1 // SPDX-License-Identifier: GPL-2.0-only
3 * Copyright 2016, Rashmica Gupta, IBM Corp.
5 * This traverses the kernel pagetables and dumps the
6 * information about the used sections of memory to
7 * /sys/kernel/debug/kernel_pagetables.
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.
13 #include <linux/debugfs.h>
15 #include <linux/hugetlb.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>
24 #include <asm/hugetlb.h>
26 #include <mm/mmu_decl.h>
31 * To visualise what is happening,
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
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}
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
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.
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.
58 const struct addr_marker *marker;
59 unsigned long start_address;
60 unsigned long start_pa;
64 unsigned long wx_pages;
68 unsigned long start_address;
72 static struct addr_marker address_markers[] = {
73 { 0, "Start of kernel VM" },
75 { 0, "modules start" },
78 { 0, "vmalloc() Area" },
79 { 0, "vmalloc() End" },
81 { 0, "isa I/O start" },
83 { 0, "phb I/O start" },
85 { 0, "I/O remap start" },
86 { 0, "I/O remap end" },
87 { 0, "vmemmap start" },
89 { 0, "Early I/O remap start" },
90 { 0, "Early I/O remap end" },
92 { 0, "Highmem PTEs start" },
93 { 0, "Highmem PTEs end" },
95 { 0, "Fixmap start" },
99 { 0, "kasan shadow mem start" },
100 { 0, "kasan shadow mem end" },
105 #define pt_dump_seq_printf(m, fmt, args...) \
108 seq_printf(m, fmt, ##args); \
111 #define pt_dump_seq_putc(m, c) \
117 void pt_dump_size(struct seq_file *m, unsigned long size)
119 static const char units[] = "KMGTPE";
120 const char *unit = units;
122 /* Work out what appropriate unit to use */
123 while (!(size & 1023) && unit[1]) {
127 pt_dump_seq_printf(m, "%9lu%c ", size, *unit);
130 static void dump_flag_info(struct pg_state *st, const struct flag_info
131 *flag, u64 pte, int num)
135 for (i = 0; i < num; i++, flag++) {
136 const char *s = NULL;
139 /* flag not defined so don't check it */
142 /* Some 'flags' are actually values */
144 val = pte & flag->val;
146 val = val >> flag->shift;
147 pt_dump_seq_printf(st->seq, " %s:%llx", flag->set, val);
149 if ((pte & flag->mask) == flag->val)
154 pt_dump_seq_printf(st->seq, " %s", s);
156 st->current_flags &= ~flag->mask;
158 if (st->current_flags != 0)
159 pt_dump_seq_printf(st->seq, " unknown flags:%llx", st->current_flags);
162 static void dump_addr(struct pg_state *st, unsigned long addr)
165 #define REG "0x%016lx"
167 #define REG "0x%08lx"
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);
175 static void note_prot_wx(struct pg_state *st, unsigned long addr)
177 pte_t pte = __pte(st->current_flags);
179 if (!IS_ENABLED(CONFIG_PPC_DEBUG_WX) || !st->check_wx)
182 if (!pte_write(pte) || !pte_exec(pte))
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);
188 st->wx_pages += (addr - st->start_address) / PAGE_SIZE;
191 static void note_page_update_state(struct pg_state *st, unsigned long addr,
192 unsigned int level, u64 val, unsigned long page_size)
194 u64 flag = val & pg_level[level].mask;
195 u64 pa = val & PTE_RPN_MASK;
198 st->current_flags = flag;
199 st->start_address = addr;
202 while (addr >= st->marker[1].start_address) {
204 pt_dump_seq_printf(st->seq, "---[ %s ]---\n", st->marker->name);
208 static void note_page(struct pg_state *st, unsigned long addr,
209 unsigned int level, u64 val, unsigned long page_size)
211 u64 flag = val & pg_level[level].mask;
213 /* At first no level is set */
215 pt_dump_seq_printf(st->seq, "---[ %s ]---\n", st->marker->name);
216 note_page_update_state(st, addr, level, val, page_size);
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.
224 } else if (flag != st->current_flags || level != st->level ||
225 addr >= st->marker[1].start_address) {
227 /* Check the PTE flags */
228 if (st->current_flags) {
229 note_prot_wx(st, addr);
232 /* Dump all the flags */
233 if (pg_level[st->level].flag)
234 dump_flag_info(st, pg_level[st->level].flag,
236 pg_level[st->level].num);
238 pt_dump_seq_putc(st->seq, '\n');
242 * Address indicates we have passed the end of the
243 * current section of virtual memory
245 note_page_update_state(st, addr, level, val, page_size);
249 static void walk_pte(struct pg_state *st, pmd_t *pmd, unsigned long start)
251 pte_t *pte = pte_offset_kernel(pmd, 0);
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);
262 static void walk_hugepd(struct pg_state *st, hugepd_t *phpd, unsigned long start,
263 int pdshift, int level)
265 #ifdef CONFIG_ARCH_HAS_HUGEPD
267 int shift = hugepd_shift(*phpd);
268 int ptrs_per_hpd = pdshift - shift > 0 ? 1 << (pdshift - shift) : 1;
270 if (start & ((1 << shift) - 1))
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);
277 note_page(st, addr, level + 1, pte_val(*pte), 1 << shift);
282 static void walk_pmd(struct pg_state *st, pud_t *pud, unsigned long start)
284 pmd_t *pmd = pmd_offset(pud, 0);
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))
292 walk_pte(st, pmd, addr);
294 note_page(st, addr, 3, pmd_val(*pmd), PMD_SIZE);
298 static void walk_pud(struct pg_state *st, p4d_t *p4d, unsigned long start)
300 pud_t *pud = pud_offset(p4d, 0);
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))
308 walk_pmd(st, pud, addr);
310 note_page(st, addr, 2, pud_val(*pud), PUD_SIZE);
314 static void walk_pagetables(struct pg_state *st)
317 unsigned long addr = st->start_address & PGDIR_MASK;
318 pgd_t *pgd = pgd_offset_k(addr);
321 * Traverse the linux pagetable structure and dump pages that are in
322 * the hash pagetable.
324 for (i = pgd_index(addr); i < PTRS_PER_PGD; i++, pgd++, addr += PGDIR_SIZE) {
325 p4d_t *p4d = p4d_offset(pgd, 0);
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);
333 walk_pud(st, p4d, addr);
337 static void populate_markers(void)
342 address_markers[i++].start_address = PAGE_OFFSET;
344 address_markers[i++].start_address = TASK_SIZE;
347 address_markers[i++].start_address = MODULES_VADDR;
348 address_markers[i++].start_address = MODULES_END;
350 address_markers[i++].start_address = VMALLOC_START;
351 address_markers[i++].start_address = VMALLOC_END;
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;
363 address_markers[i++].start_address = VMEMMAP_BASE;
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);
372 address_markers[i++].start_address = FIXADDR_START;
373 address_markers[i++].start_address = FIXADDR_TOP;
375 address_markers[i++].start_address = KASAN_SHADOW_START;
376 address_markers[i++].start_address = KASAN_SHADOW_END;
378 #endif /* CONFIG_PPC64 */
381 static int ptdump_show(struct seq_file *m, void *v)
383 struct pg_state st = {
385 .marker = address_markers,
386 .start_address = IS_ENABLED(CONFIG_PPC64) ? PAGE_OFFSET : TASK_SIZE,
390 if (!radix_enabled())
391 st.start_address = KERN_VIRT_START;
394 /* Traverse kernel page tables */
395 walk_pagetables(&st);
396 note_page(&st, 0, 0, 0, 0);
401 static int ptdump_open(struct inode *inode, struct file *file)
403 return single_open(file, ptdump_show, NULL);
406 static const struct file_operations ptdump_fops = {
410 .release = single_release,
413 static void build_pgtable_complete_mask(void)
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;
423 #ifdef CONFIG_PPC_DEBUG_WX
424 void ptdump_check_wx(void)
426 struct pg_state st = {
428 .marker = address_markers,
430 .start_address = IS_ENABLED(CONFIG_PPC64) ? PAGE_OFFSET : TASK_SIZE,
434 if (!radix_enabled())
435 st.start_address = KERN_VIRT_START;
438 walk_pagetables(&st);
441 pr_warn("Checked W+X mappings: FAILED, %lu W+X pages found\n",
444 pr_info("Checked W+X mappings: passed, no W+X pages found\n");
448 static int ptdump_init(void)
451 build_pgtable_complete_mask();
452 debugfs_create_file("kernel_page_tables", 0400, NULL, NULL,
456 device_initcall(ptdump_init);