xref: /linux/arch/powerpc/mm/ptdump/ptdump.c (revision 1e525507)
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/ptdump.h>
20 #include <linux/sched.h>
21 #include <linux/seq_file.h>
22 #include <asm/fixmap.h>
23 #include <linux/const.h>
24 #include <linux/kasan.h>
25 #include <asm/page.h>
26 #include <asm/hugetlb.h>
27 
28 #include <mm/mmu_decl.h>
29 
30 #include "ptdump.h"
31 
32 /*
33  * To visualise what is happening,
34  *
35  *  - PTRS_PER_P** = how many entries there are in the corresponding P**
36  *  - P**_SHIFT = how many bits of the address we use to index into the
37  * corresponding P**
38  *  - P**_SIZE is how much memory we can access through the table - not the
39  * size of the table itself.
40  * P**={PGD, PUD, PMD, PTE}
41  *
42  *
43  * Each entry of the PGD points to a PUD. Each entry of a PUD points to a
44  * PMD. Each entry of a PMD points to a PTE. And every PTE entry points to
45  * a page.
46  *
47  * In the case where there are only 3 levels, the PUD is folded into the
48  * PGD: every PUD has only one entry which points to the PMD.
49  *
50  * The page dumper groups page table entries of the same type into a single
51  * description. It uses pg_state to track the range information while
52  * iterating over the PTE entries. When the continuity is broken it then
53  * dumps out a description of the range - ie PTEs that are virtually contiguous
54  * with the same PTE flags are chunked together. This is to make it clear how
55  * different areas of the kernel virtual memory are used.
56  *
57  */
58 struct pg_state {
59 	struct ptdump_state ptdump;
60 	struct seq_file *seq;
61 	const struct addr_marker *marker;
62 	unsigned long start_address;
63 	unsigned long start_pa;
64 	int level;
65 	u64 current_flags;
66 	bool check_wx;
67 	unsigned long wx_pages;
68 };
69 
70 struct addr_marker {
71 	unsigned long start_address;
72 	const char *name;
73 };
74 
75 static struct addr_marker address_markers[] = {
76 	{ 0,	"Start of kernel VM" },
77 #ifdef MODULES_VADDR
78 	{ 0,	"modules start" },
79 	{ 0,	"modules end" },
80 #endif
81 	{ 0,	"vmalloc() Area" },
82 	{ 0,	"vmalloc() End" },
83 #ifdef CONFIG_PPC64
84 	{ 0,	"isa I/O start" },
85 	{ 0,	"isa I/O end" },
86 	{ 0,	"phb I/O start" },
87 	{ 0,	"phb I/O end" },
88 	{ 0,	"I/O remap start" },
89 	{ 0,	"I/O remap end" },
90 	{ 0,	"vmemmap start" },
91 #else
92 	{ 0,	"Early I/O remap start" },
93 	{ 0,	"Early I/O remap end" },
94 #ifdef CONFIG_HIGHMEM
95 	{ 0,	"Highmem PTEs start" },
96 	{ 0,	"Highmem PTEs end" },
97 #endif
98 	{ 0,	"Fixmap start" },
99 	{ 0,	"Fixmap end" },
100 #endif
101 #ifdef CONFIG_KASAN
102 	{ 0,	"kasan shadow mem start" },
103 	{ 0,	"kasan shadow mem end" },
104 #endif
105 	{ -1,	NULL },
106 };
107 
108 static struct ptdump_range ptdump_range[] __ro_after_init = {
109 	{TASK_SIZE_MAX, ~0UL},
110 	{0, 0}
111 };
112 
113 #define pt_dump_seq_printf(m, fmt, args...)	\
114 ({						\
115 	if (m)					\
116 		seq_printf(m, fmt, ##args);	\
117 })
118 
119 #define pt_dump_seq_putc(m, c)		\
120 ({					\
121 	if (m)				\
122 		seq_putc(m, c);		\
123 })
124 
125 void pt_dump_size(struct seq_file *m, unsigned long size)
126 {
127 	static const char units[] = " KMGTPE";
128 	const char *unit = units;
129 
130 	/* Work out what appropriate unit to use */
131 	while (!(size & 1023) && unit[1]) {
132 		size >>= 10;
133 		unit++;
134 	}
135 	pt_dump_seq_printf(m, "%9lu%c ", size, *unit);
136 }
137 
138 static void dump_flag_info(struct pg_state *st, const struct flag_info
139 		*flag, u64 pte, int num)
140 {
141 	unsigned int i;
142 
143 	for (i = 0; i < num; i++, flag++) {
144 		const char *s = NULL;
145 		u64 val;
146 
147 		/* flag not defined so don't check it */
148 		if (flag->mask == 0)
149 			continue;
150 		/* Some 'flags' are actually values */
151 		if (flag->is_val) {
152 			val = pte & flag->val;
153 			if (flag->shift)
154 				val = val >> flag->shift;
155 			pt_dump_seq_printf(st->seq, "  %s:%llx", flag->set, val);
156 		} else {
157 			if ((pte & flag->mask) == flag->val)
158 				s = flag->set;
159 			else
160 				s = flag->clear;
161 			if (s)
162 				pt_dump_seq_printf(st->seq, "  %s", s);
163 		}
164 		st->current_flags &= ~flag->mask;
165 	}
166 	if (st->current_flags != 0)
167 		pt_dump_seq_printf(st->seq, "  unknown flags:%llx", st->current_flags);
168 }
169 
170 static void dump_addr(struct pg_state *st, unsigned long addr)
171 {
172 #ifdef CONFIG_PPC64
173 #define REG		"0x%016lx"
174 #else
175 #define REG		"0x%08lx"
176 #endif
177 
178 	pt_dump_seq_printf(st->seq, REG "-" REG " ", st->start_address, addr - 1);
179 	pt_dump_seq_printf(st->seq, " " REG " ", st->start_pa);
180 	pt_dump_size(st->seq, addr - st->start_address);
181 }
182 
183 static void note_prot_wx(struct pg_state *st, unsigned long addr)
184 {
185 	pte_t pte = __pte(st->current_flags);
186 
187 	if (!st->check_wx)
188 		return;
189 
190 	if (!pte_write(pte) || !pte_exec(pte))
191 		return;
192 
193 	WARN_ONCE(IS_ENABLED(CONFIG_DEBUG_WX),
194 		  "powerpc/mm: Found insecure W+X mapping at address %p/%pS\n",
195 		  (void *)st->start_address, (void *)st->start_address);
196 
197 	st->wx_pages += (addr - st->start_address) / PAGE_SIZE;
198 }
199 
200 static void note_page_update_state(struct pg_state *st, unsigned long addr, int level, u64 val)
201 {
202 	u64 flag = level >= 0 ? val & pg_level[level].mask : 0;
203 	u64 pa = val & PTE_RPN_MASK;
204 
205 	st->level = level;
206 	st->current_flags = flag;
207 	st->start_address = addr;
208 	st->start_pa = pa;
209 
210 	while (addr >= st->marker[1].start_address) {
211 		st->marker++;
212 		pt_dump_seq_printf(st->seq, "---[ %s ]---\n", st->marker->name);
213 	}
214 }
215 
216 static void note_page(struct ptdump_state *pt_st, unsigned long addr, int level, u64 val)
217 {
218 	u64 flag = level >= 0 ? val & pg_level[level].mask : 0;
219 	struct pg_state *st = container_of(pt_st, struct pg_state, ptdump);
220 
221 	/* At first no level is set */
222 	if (st->level == -1) {
223 		pt_dump_seq_printf(st->seq, "---[ %s ]---\n", st->marker->name);
224 		note_page_update_state(st, addr, level, val);
225 	/*
226 	 * Dump the section of virtual memory when:
227 	 *   - the PTE flags from one entry to the next differs.
228 	 *   - we change levels in the tree.
229 	 *   - the address is in a different section of memory and is thus
230 	 *   used for a different purpose, regardless of the flags.
231 	 */
232 	} else if (flag != st->current_flags || level != st->level ||
233 		   addr >= st->marker[1].start_address) {
234 
235 		/* Check the PTE flags */
236 		if (st->current_flags) {
237 			note_prot_wx(st, addr);
238 			dump_addr(st, addr);
239 
240 			/* Dump all the flags */
241 			if (pg_level[st->level].flag)
242 				dump_flag_info(st, pg_level[st->level].flag,
243 					  st->current_flags,
244 					  pg_level[st->level].num);
245 
246 			pt_dump_seq_putc(st->seq, '\n');
247 		}
248 
249 		/*
250 		 * Address indicates we have passed the end of the
251 		 * current section of virtual memory
252 		 */
253 		note_page_update_state(st, addr, level, val);
254 	}
255 }
256 
257 static void populate_markers(void)
258 {
259 	int i = 0;
260 
261 #ifdef CONFIG_PPC64
262 	address_markers[i++].start_address = PAGE_OFFSET;
263 #else
264 	address_markers[i++].start_address = TASK_SIZE;
265 #endif
266 #ifdef MODULES_VADDR
267 	address_markers[i++].start_address = MODULES_VADDR;
268 	address_markers[i++].start_address = MODULES_END;
269 #endif
270 	address_markers[i++].start_address = VMALLOC_START;
271 	address_markers[i++].start_address = VMALLOC_END;
272 #ifdef CONFIG_PPC64
273 	address_markers[i++].start_address = ISA_IO_BASE;
274 	address_markers[i++].start_address = ISA_IO_END;
275 	address_markers[i++].start_address = PHB_IO_BASE;
276 	address_markers[i++].start_address = PHB_IO_END;
277 	address_markers[i++].start_address = IOREMAP_BASE;
278 	address_markers[i++].start_address = IOREMAP_END;
279 	/* What is the ifdef about? */
280 #ifdef CONFIG_PPC_BOOK3S_64
281 	address_markers[i++].start_address =  H_VMEMMAP_START;
282 #else
283 	address_markers[i++].start_address =  VMEMMAP_BASE;
284 #endif
285 #else /* !CONFIG_PPC64 */
286 	address_markers[i++].start_address = ioremap_bot;
287 	address_markers[i++].start_address = IOREMAP_TOP;
288 #ifdef CONFIG_HIGHMEM
289 	address_markers[i++].start_address = PKMAP_BASE;
290 	address_markers[i++].start_address = PKMAP_ADDR(LAST_PKMAP);
291 #endif
292 	address_markers[i++].start_address = FIXADDR_START;
293 	address_markers[i++].start_address = FIXADDR_TOP;
294 #endif /* CONFIG_PPC64 */
295 #ifdef CONFIG_KASAN
296 	address_markers[i++].start_address = KASAN_SHADOW_START;
297 	address_markers[i++].start_address = KASAN_SHADOW_END;
298 #endif
299 }
300 
301 static int ptdump_show(struct seq_file *m, void *v)
302 {
303 	struct pg_state st = {
304 		.seq = m,
305 		.marker = address_markers,
306 		.level = -1,
307 		.ptdump = {
308 			.note_page = note_page,
309 			.range = ptdump_range,
310 		}
311 	};
312 
313 	/* Traverse kernel page tables */
314 	ptdump_walk_pgd(&st.ptdump, &init_mm, NULL);
315 	return 0;
316 }
317 
318 DEFINE_SHOW_ATTRIBUTE(ptdump);
319 
320 static void __init build_pgtable_complete_mask(void)
321 {
322 	unsigned int i, j;
323 
324 	for (i = 0; i < ARRAY_SIZE(pg_level); i++)
325 		if (pg_level[i].flag)
326 			for (j = 0; j < pg_level[i].num; j++)
327 				pg_level[i].mask |= pg_level[i].flag[j].mask;
328 }
329 
330 bool ptdump_check_wx(void)
331 {
332 	struct pg_state st = {
333 		.seq = NULL,
334 		.marker = (struct addr_marker[]) {
335 			{ 0, NULL},
336 			{ -1, NULL},
337 		},
338 		.level = -1,
339 		.check_wx = true,
340 		.ptdump = {
341 			.note_page = note_page,
342 			.range = ptdump_range,
343 		}
344 	};
345 
346 	if (IS_ENABLED(CONFIG_PPC_BOOK3S_64) && !mmu_has_feature(MMU_FTR_KERNEL_RO))
347 		return true;
348 
349 	ptdump_walk_pgd(&st.ptdump, &init_mm, NULL);
350 
351 	if (st.wx_pages) {
352 		pr_warn("Checked W+X mappings: FAILED, %lu W+X pages found\n",
353 			st.wx_pages);
354 
355 		return false;
356 	} else {
357 		pr_info("Checked W+X mappings: passed, no W+X pages found\n");
358 
359 		return true;
360 	}
361 }
362 
363 static int __init ptdump_init(void)
364 {
365 #ifdef CONFIG_PPC64
366 	if (!radix_enabled())
367 		ptdump_range[0].start = KERN_VIRT_START;
368 	else
369 		ptdump_range[0].start = PAGE_OFFSET;
370 
371 	ptdump_range[0].end = PAGE_OFFSET + (PGDIR_SIZE * PTRS_PER_PGD);
372 #endif
373 
374 	populate_markers();
375 	build_pgtable_complete_mask();
376 
377 	if (IS_ENABLED(CONFIG_PTDUMP_DEBUGFS))
378 		debugfs_create_file("kernel_page_tables", 0400, NULL, NULL, &ptdump_fops);
379 
380 	return 0;
381 }
382 device_initcall(ptdump_init);
383