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 */
56struct 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 long last_pa;
62	unsigned long page_size;
63	unsigned int level;
64	u64 current_flags;
65	bool check_wx;
66	unsigned long wx_pages;
67};
68
69struct addr_marker {
70	unsigned long start_address;
71	const char *name;
72};
73
74static struct addr_marker address_markers[] = {
75	{ 0,	"Start of kernel VM" },
76#ifdef MODULES_VADDR
77	{ 0,	"modules start" },
78	{ 0,	"modules end" },
79#endif
80	{ 0,	"vmalloc() Area" },
81	{ 0,	"vmalloc() End" },
82#ifdef CONFIG_PPC64
83	{ 0,	"isa I/O start" },
84	{ 0,	"isa I/O end" },
85	{ 0,	"phb I/O start" },
86	{ 0,	"phb I/O end" },
87	{ 0,	"I/O remap start" },
88	{ 0,	"I/O remap end" },
89	{ 0,	"vmemmap start" },
90#else
91	{ 0,	"Early I/O remap start" },
92	{ 0,	"Early I/O remap end" },
93#ifdef CONFIG_HIGHMEM
94	{ 0,	"Highmem PTEs start" },
95	{ 0,	"Highmem PTEs end" },
96#endif
97	{ 0,	"Fixmap start" },
98	{ 0,	"Fixmap end" },
99#endif
100#ifdef CONFIG_KASAN
101	{ 0,	"kasan shadow mem start" },
102	{ 0,	"kasan shadow mem end" },
103#endif
104	{ -1,	NULL },
105};
106
107#define pt_dump_seq_printf(m, fmt, args...)	\
108({						\
109	if (m)					\
110		seq_printf(m, fmt, ##args);	\
111})
112
113#define pt_dump_seq_putc(m, c)		\
114({					\
115	if (m)				\
116		seq_putc(m, c);		\
117})
118
119void pt_dump_size(struct seq_file *m, unsigned long size)
120{
121	static const char units[] = "KMGTPE";
122	const char *unit = units;
123
124	/* Work out what appropriate unit to use */
125	while (!(size & 1023) && unit[1]) {
126		size >>= 10;
127		unit++;
128	}
129	pt_dump_seq_printf(m, "%9lu%c ", size, *unit);
130}
131
132static void dump_flag_info(struct pg_state *st, const struct flag_info
133		*flag, u64 pte, int num)
134{
135	unsigned int i;
136
137	for (i = 0; i < num; i++, flag++) {
138		const char *s = NULL;
139		u64 val;
140
141		/* flag not defined so don't check it */
142		if (flag->mask == 0)
143			continue;
144		/* Some 'flags' are actually values */
145		if (flag->is_val) {
146			val = pte & flag->val;
147			if (flag->shift)
148				val = val >> flag->shift;
149			pt_dump_seq_printf(st->seq, "  %s:%llx", flag->set, val);
150		} else {
151			if ((pte & flag->mask) == flag->val)
152				s = flag->set;
153			else
154				s = flag->clear;
155			if (s)
156				pt_dump_seq_printf(st->seq, "  %s", s);
157		}
158		st->current_flags &= ~flag->mask;
159	}
160	if (st->current_flags != 0)
161		pt_dump_seq_printf(st->seq, "  unknown flags:%llx", st->current_flags);
162}
163
164static void dump_addr(struct pg_state *st, unsigned long addr)
165{
166	unsigned long delta;
167
168#ifdef CONFIG_PPC64
169#define REG		"0x%016lx"
170#else
171#define REG		"0x%08lx"
172#endif
173
174	pt_dump_seq_printf(st->seq, REG "-" REG " ", st->start_address, addr - 1);
175	if (st->start_pa == st->last_pa && st->start_address + st->page_size != addr) {
176		pt_dump_seq_printf(st->seq, "[" REG "]", st->start_pa);
177		delta = st->page_size >> 10;
178	} else {
179		pt_dump_seq_printf(st->seq, " " REG " ", st->start_pa);
180		delta = (addr - st->start_address) >> 10;
181	}
182	pt_dump_size(st->seq, delta);
183}
184
185static void note_prot_wx(struct pg_state *st, unsigned long addr)
186{
187	pte_t pte = __pte(st->current_flags);
188
189	if (!IS_ENABLED(CONFIG_PPC_DEBUG_WX) || !st->check_wx)
190		return;
191
192	if (!pte_write(pte) || !pte_exec(pte))
193		return;
194
195	WARN_ONCE(1, "powerpc/mm: Found insecure W+X mapping at address %p/%pS\n",
196		  (void *)st->start_address, (void *)st->start_address);
197
198	st->wx_pages += (addr - st->start_address) / PAGE_SIZE;
199}
200
201static void note_page_update_state(struct pg_state *st, unsigned long addr,
202				   unsigned int level, u64 val, unsigned long page_size)
203{
204	u64 flag = val & pg_level[level].mask;
205	u64 pa = val & PTE_RPN_MASK;
206
207	st->level = level;
208	st->current_flags = flag;
209	st->start_address = addr;
210	st->start_pa = pa;
211	st->page_size = page_size;
212
213	while (addr >= st->marker[1].start_address) {
214		st->marker++;
215		pt_dump_seq_printf(st->seq, "---[ %s ]---\n", st->marker->name);
216	}
217}
218
219static void note_page(struct pg_state *st, unsigned long addr,
220	       unsigned int level, u64 val, unsigned long page_size)
221{
222	u64 flag = val & pg_level[level].mask;
223	u64 pa = val & PTE_RPN_MASK;
224
225	/* At first no level is set */
226	if (!st->level) {
227		pt_dump_seq_printf(st->seq, "---[ %s ]---\n", st->marker->name);
228		note_page_update_state(st, addr, level, val, page_size);
229	/*
230	 * Dump the section of virtual memory when:
231	 *   - the PTE flags from one entry to the next differs.
232	 *   - we change levels in the tree.
233	 *   - the address is in a different section of memory and is thus
234	 *   used for a different purpose, regardless of the flags.
235	 *   - the pa of this page is not adjacent to the last inspected page
236	 */
237	} else if (flag != st->current_flags || level != st->level ||
238		   addr >= st->marker[1].start_address ||
239		   (pa != st->last_pa + st->page_size &&
240		    (pa != st->start_pa || st->start_pa != st->last_pa))) {
241
242		/* Check the PTE flags */
243		if (st->current_flags) {
244			note_prot_wx(st, addr);
245			dump_addr(st, addr);
246
247			/* Dump all the flags */
248			if (pg_level[st->level].flag)
249				dump_flag_info(st, pg_level[st->level].flag,
250					  st->current_flags,
251					  pg_level[st->level].num);
252
253			pt_dump_seq_putc(st->seq, '\n');
254		}
255
256		/*
257		 * Address indicates we have passed the end of the
258		 * current section of virtual memory
259		 */
260		note_page_update_state(st, addr, level, val, page_size);
261	}
262	st->last_pa = pa;
263}
264
265static void walk_pte(struct pg_state *st, pmd_t *pmd, unsigned long start)
266{
267	pte_t *pte = pte_offset_kernel(pmd, 0);
268	unsigned long addr;
269	unsigned int i;
270
271	for (i = 0; i < PTRS_PER_PTE; i++, pte++) {
272		addr = start + i * PAGE_SIZE;
273		note_page(st, addr, 4, pte_val(*pte), PAGE_SIZE);
274
275	}
276}
277
278static void walk_hugepd(struct pg_state *st, hugepd_t *phpd, unsigned long start,
279			int pdshift, int level)
280{
281#ifdef CONFIG_ARCH_HAS_HUGEPD
282	unsigned int i;
283	int shift = hugepd_shift(*phpd);
284	int ptrs_per_hpd = pdshift - shift > 0 ? 1 << (pdshift - shift) : 1;
285
286	if (start & ((1 << shift) - 1))
287		return;
288
289	for (i = 0; i < ptrs_per_hpd; i++) {
290		unsigned long addr = start + (i << shift);
291		pte_t *pte = hugepte_offset(*phpd, addr, pdshift);
292
293		note_page(st, addr, level + 1, pte_val(*pte), 1 << shift);
294	}
295#endif
296}
297
298static void walk_pmd(struct pg_state *st, pud_t *pud, unsigned long start)
299{
300	pmd_t *pmd = pmd_offset(pud, 0);
301	unsigned long addr;
302	unsigned int i;
303
304	for (i = 0; i < PTRS_PER_PMD; i++, pmd++) {
305		addr = start + i * PMD_SIZE;
306		if (!pmd_none(*pmd) && !pmd_is_leaf(*pmd))
307			/* pmd exists */
308			walk_pte(st, pmd, addr);
309		else
310			note_page(st, addr, 3, pmd_val(*pmd), PMD_SIZE);
311	}
312}
313
314static void walk_pud(struct pg_state *st, p4d_t *p4d, unsigned long start)
315{
316	pud_t *pud = pud_offset(p4d, 0);
317	unsigned long addr;
318	unsigned int i;
319
320	for (i = 0; i < PTRS_PER_PUD; i++, pud++) {
321		addr = start + i * PUD_SIZE;
322		if (!pud_none(*pud) && !pud_is_leaf(*pud))
323			/* pud exists */
324			walk_pmd(st, pud, addr);
325		else
326			note_page(st, addr, 2, pud_val(*pud), PUD_SIZE);
327	}
328}
329
330static void walk_pagetables(struct pg_state *st)
331{
332	unsigned int i;
333	unsigned long addr = st->start_address & PGDIR_MASK;
334	pgd_t *pgd = pgd_offset_k(addr);
335
336	/*
337	 * Traverse the linux pagetable structure and dump pages that are in
338	 * the hash pagetable.
339	 */
340	for (i = pgd_index(addr); i < PTRS_PER_PGD; i++, pgd++, addr += PGDIR_SIZE) {
341		p4d_t *p4d = p4d_offset(pgd, 0);
342
343		if (p4d_none(*p4d) || p4d_is_leaf(*p4d))
344			note_page(st, addr, 1, p4d_val(*p4d), PGDIR_SIZE);
345		else if (is_hugepd(__hugepd(p4d_val(*p4d))))
346			walk_hugepd(st, (hugepd_t *)p4d, addr, PGDIR_SHIFT, 1);
347		else
348			/* p4d exists */
349			walk_pud(st, p4d, addr);
350	}
351}
352
353static void populate_markers(void)
354{
355	int i = 0;
356
357#ifdef CONFIG_PPC64
358	address_markers[i++].start_address = PAGE_OFFSET;
359#else
360	address_markers[i++].start_address = TASK_SIZE;
361#endif
362#ifdef MODULES_VADDR
363	address_markers[i++].start_address = MODULES_VADDR;
364	address_markers[i++].start_address = MODULES_END;
365#endif
366	address_markers[i++].start_address = VMALLOC_START;
367	address_markers[i++].start_address = VMALLOC_END;
368#ifdef CONFIG_PPC64
369	address_markers[i++].start_address = ISA_IO_BASE;
370	address_markers[i++].start_address = ISA_IO_END;
371	address_markers[i++].start_address = PHB_IO_BASE;
372	address_markers[i++].start_address = PHB_IO_END;
373	address_markers[i++].start_address = IOREMAP_BASE;
374	address_markers[i++].start_address = IOREMAP_END;
375	/* What is the ifdef about? */
376#ifdef CONFIG_PPC_BOOK3S_64
377	address_markers[i++].start_address =  H_VMEMMAP_START;
378#else
379	address_markers[i++].start_address =  VMEMMAP_BASE;
380#endif
381#else /* !CONFIG_PPC64 */
382	address_markers[i++].start_address = ioremap_bot;
383	address_markers[i++].start_address = IOREMAP_TOP;
384#ifdef CONFIG_HIGHMEM
385	address_markers[i++].start_address = PKMAP_BASE;
386	address_markers[i++].start_address = PKMAP_ADDR(LAST_PKMAP);
387#endif
388	address_markers[i++].start_address = FIXADDR_START;
389	address_markers[i++].start_address = FIXADDR_TOP;
390#ifdef CONFIG_KASAN
391	address_markers[i++].start_address = KASAN_SHADOW_START;
392	address_markers[i++].start_address = KASAN_SHADOW_END;
393#endif
394#endif /* CONFIG_PPC64 */
395}
396
397static int ptdump_show(struct seq_file *m, void *v)
398{
399	struct pg_state st = {
400		.seq = m,
401		.marker = address_markers,
402		.start_address = IS_ENABLED(CONFIG_PPC64) ? PAGE_OFFSET : TASK_SIZE,
403	};
404
405#ifdef CONFIG_PPC64
406	if (!radix_enabled())
407		st.start_address = KERN_VIRT_START;
408#endif
409
410	/* Traverse kernel page tables */
411	walk_pagetables(&st);
412	note_page(&st, 0, 0, 0, 0);
413	return 0;
414}
415
416
417static int ptdump_open(struct inode *inode, struct file *file)
418{
419	return single_open(file, ptdump_show, NULL);
420}
421
422static const struct file_operations ptdump_fops = {
423	.open		= ptdump_open,
424	.read		= seq_read,
425	.llseek		= seq_lseek,
426	.release	= single_release,
427};
428
429static void build_pgtable_complete_mask(void)
430{
431	unsigned int i, j;
432
433	for (i = 0; i < ARRAY_SIZE(pg_level); i++)
434		if (pg_level[i].flag)
435			for (j = 0; j < pg_level[i].num; j++)
436				pg_level[i].mask |= pg_level[i].flag[j].mask;
437}
438
439#ifdef CONFIG_PPC_DEBUG_WX
440void ptdump_check_wx(void)
441{
442	struct pg_state st = {
443		.seq = NULL,
444		.marker = address_markers,
445		.check_wx = true,
446		.start_address = IS_ENABLED(CONFIG_PPC64) ? PAGE_OFFSET : TASK_SIZE,
447	};
448
449#ifdef CONFIG_PPC64
450	if (!radix_enabled())
451		st.start_address = KERN_VIRT_START;
452#endif
453
454	walk_pagetables(&st);
455
456	if (st.wx_pages)
457		pr_warn("Checked W+X mappings: FAILED, %lu W+X pages found\n",
458			st.wx_pages);
459	else
460		pr_info("Checked W+X mappings: passed, no W+X pages found\n");
461}
462#endif
463
464static int ptdump_init(void)
465{
466	populate_markers();
467	build_pgtable_complete_mask();
468	debugfs_create_file("kernel_page_tables", 0400, NULL, NULL,
469			    &ptdump_fops);
470	return 0;
471}
472device_initcall(ptdump_init);
473