xref: /openbmc/linux/arch/powerpc/mm/ptdump/ptdump.c (revision 088e88be)
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 #ifdef CONFIG_PPC32
30 #define KERN_VIRT_START	PAGE_OFFSET
31 #endif
32 
33 /*
34  * To visualise what is happening,
35  *
36  *  - PTRS_PER_P** = how many entries there are in the corresponding P**
37  *  - P**_SHIFT = how many bits of the address we use to index into the
38  * corresponding P**
39  *  - P**_SIZE is how much memory we can access through the table - not the
40  * size of the table itself.
41  * P**={PGD, PUD, PMD, PTE}
42  *
43  *
44  * Each entry of the PGD points to a PUD. Each entry of a PUD points to a
45  * PMD. Each entry of a PMD points to a PTE. And every PTE entry points to
46  * a page.
47  *
48  * In the case where there are only 3 levels, the PUD is folded into the
49  * PGD: every PUD has only one entry which points to the PMD.
50  *
51  * The page dumper groups page table entries of the same type into a single
52  * description. It uses pg_state to track the range information while
53  * iterating over the PTE entries. When the continuity is broken it then
54  * dumps out a description of the range - ie PTEs that are virtually contiguous
55  * with the same PTE flags are chunked together. This is to make it clear how
56  * different areas of the kernel virtual memory are used.
57  *
58  */
59 struct pg_state {
60 	struct seq_file *seq;
61 	const struct addr_marker *marker;
62 	unsigned long start_address;
63 	unsigned long start_pa;
64 	unsigned long last_pa;
65 	unsigned int level;
66 	u64 current_flags;
67 	bool check_wx;
68 	unsigned long wx_pages;
69 };
70 
71 struct addr_marker {
72 	unsigned long start_address;
73 	const char *name;
74 };
75 
76 static struct addr_marker address_markers[] = {
77 	{ 0,	"Start of kernel VM" },
78 	{ 0,	"vmalloc() Area" },
79 	{ 0,	"vmalloc() End" },
80 #ifdef CONFIG_PPC64
81 	{ 0,	"isa I/O start" },
82 	{ 0,	"isa I/O end" },
83 	{ 0,	"phb I/O start" },
84 	{ 0,	"phb I/O end" },
85 	{ 0,	"I/O remap start" },
86 	{ 0,	"I/O remap end" },
87 	{ 0,	"vmemmap start" },
88 #else
89 	{ 0,	"Early I/O remap start" },
90 	{ 0,	"Early I/O remap end" },
91 #ifdef CONFIG_NOT_COHERENT_CACHE
92 	{ 0,	"Consistent mem start" },
93 	{ 0,	"Consistent mem end" },
94 #endif
95 #ifdef CONFIG_HIGHMEM
96 	{ 0,	"Highmem PTEs start" },
97 	{ 0,	"Highmem PTEs end" },
98 #endif
99 	{ 0,	"Fixmap start" },
100 	{ 0,	"Fixmap end" },
101 #endif
102 #ifdef CONFIG_KASAN
103 	{ 0,	"kasan shadow mem start" },
104 	{ 0,	"kasan shadow mem end" },
105 #endif
106 	{ -1,	NULL },
107 };
108 
109 #define pt_dump_seq_printf(m, fmt, args...)	\
110 ({						\
111 	if (m)					\
112 		seq_printf(m, fmt, ##args);	\
113 })
114 
115 #define pt_dump_seq_putc(m, c)		\
116 ({					\
117 	if (m)				\
118 		seq_putc(m, c);		\
119 })
120 
121 static void dump_flag_info(struct pg_state *st, const struct flag_info
122 		*flag, u64 pte, int num)
123 {
124 	unsigned int i;
125 
126 	for (i = 0; i < num; i++, flag++) {
127 		const char *s = NULL;
128 		u64 val;
129 
130 		/* flag not defined so don't check it */
131 		if (flag->mask == 0)
132 			continue;
133 		/* Some 'flags' are actually values */
134 		if (flag->is_val) {
135 			val = pte & flag->val;
136 			if (flag->shift)
137 				val = val >> flag->shift;
138 			pt_dump_seq_printf(st->seq, "  %s:%llx", flag->set, val);
139 		} else {
140 			if ((pte & flag->mask) == flag->val)
141 				s = flag->set;
142 			else
143 				s = flag->clear;
144 			if (s)
145 				pt_dump_seq_printf(st->seq, "  %s", s);
146 		}
147 		st->current_flags &= ~flag->mask;
148 	}
149 	if (st->current_flags != 0)
150 		pt_dump_seq_printf(st->seq, "  unknown flags:%llx", st->current_flags);
151 }
152 
153 static void dump_addr(struct pg_state *st, unsigned long addr)
154 {
155 	static const char units[] = "KMGTPE";
156 	const char *unit = units;
157 	unsigned long delta;
158 
159 #ifdef CONFIG_PPC64
160 #define REG		"0x%016lx"
161 #else
162 #define REG		"0x%08lx"
163 #endif
164 
165 	pt_dump_seq_printf(st->seq, REG "-" REG " ", st->start_address, addr - 1);
166 	if (st->start_pa == st->last_pa && st->start_address + PAGE_SIZE != addr) {
167 		pt_dump_seq_printf(st->seq, "[" REG "]", st->start_pa);
168 		delta = PAGE_SIZE >> 10;
169 	} else {
170 		pt_dump_seq_printf(st->seq, " " REG " ", st->start_pa);
171 		delta = (addr - st->start_address) >> 10;
172 	}
173 	/* Work out what appropriate unit to use */
174 	while (!(delta & 1023) && unit[1]) {
175 		delta >>= 10;
176 		unit++;
177 	}
178 	pt_dump_seq_printf(st->seq, "%9lu%c", delta, *unit);
179 
180 }
181 
182 static void note_prot_wx(struct pg_state *st, unsigned long addr)
183 {
184 	if (!st->check_wx)
185 		return;
186 
187 	if (!((st->current_flags & pgprot_val(PAGE_KERNEL_X)) == pgprot_val(PAGE_KERNEL_X)))
188 		return;
189 
190 	WARN_ONCE(1, "powerpc/mm: Found insecure W+X mapping at address %p/%pS\n",
191 		  (void *)st->start_address, (void *)st->start_address);
192 
193 	st->wx_pages += (addr - st->start_address) / PAGE_SIZE;
194 }
195 
196 static void note_page(struct pg_state *st, unsigned long addr,
197 	       unsigned int level, u64 val)
198 {
199 	u64 flag = val & pg_level[level].mask;
200 	u64 pa = val & PTE_RPN_MASK;
201 
202 	/* At first no level is set */
203 	if (!st->level) {
204 		st->level = level;
205 		st->current_flags = flag;
206 		st->start_address = addr;
207 		st->start_pa = pa;
208 		st->last_pa = pa;
209 		pt_dump_seq_printf(st->seq, "---[ %s ]---\n", st->marker->name);
210 	/*
211 	 * Dump the section of virtual memory when:
212 	 *   - the PTE flags from one entry to the next differs.
213 	 *   - we change levels in the tree.
214 	 *   - the address is in a different section of memory and is thus
215 	 *   used for a different purpose, regardless of the flags.
216 	 *   - the pa of this page is not adjacent to the last inspected page
217 	 */
218 	} else if (flag != st->current_flags || level != st->level ||
219 		   addr >= st->marker[1].start_address ||
220 		   (pa != st->last_pa + PAGE_SIZE &&
221 		    (pa != st->start_pa || st->start_pa != st->last_pa))) {
222 
223 		/* Check the PTE flags */
224 		if (st->current_flags) {
225 			note_prot_wx(st, addr);
226 			dump_addr(st, addr);
227 
228 			/* Dump all the flags */
229 			if (pg_level[st->level].flag)
230 				dump_flag_info(st, pg_level[st->level].flag,
231 					  st->current_flags,
232 					  pg_level[st->level].num);
233 
234 			pt_dump_seq_putc(st->seq, '\n');
235 		}
236 
237 		/*
238 		 * Address indicates we have passed the end of the
239 		 * current section of virtual memory
240 		 */
241 		while (addr >= st->marker[1].start_address) {
242 			st->marker++;
243 			pt_dump_seq_printf(st->seq, "---[ %s ]---\n", st->marker->name);
244 		}
245 		st->start_address = addr;
246 		st->start_pa = pa;
247 		st->last_pa = pa;
248 		st->current_flags = flag;
249 		st->level = level;
250 	} else {
251 		st->last_pa = pa;
252 	}
253 }
254 
255 static void walk_pte(struct pg_state *st, pmd_t *pmd, unsigned long start)
256 {
257 	pte_t *pte = pte_offset_kernel(pmd, 0);
258 	unsigned long addr;
259 	unsigned int i;
260 
261 	for (i = 0; i < PTRS_PER_PTE; i++, pte++) {
262 		addr = start + i * PAGE_SIZE;
263 		note_page(st, addr, 4, pte_val(*pte));
264 
265 	}
266 }
267 
268 static void walk_pmd(struct pg_state *st, pud_t *pud, unsigned long start)
269 {
270 	pmd_t *pmd = pmd_offset(pud, 0);
271 	unsigned long addr;
272 	unsigned int i;
273 
274 	for (i = 0; i < PTRS_PER_PMD; i++, pmd++) {
275 		addr = start + i * PMD_SIZE;
276 		if (!pmd_none(*pmd) && !pmd_is_leaf(*pmd))
277 			/* pmd exists */
278 			walk_pte(st, pmd, addr);
279 		else
280 			note_page(st, addr, 3, pmd_val(*pmd));
281 	}
282 }
283 
284 static void walk_pud(struct pg_state *st, pgd_t *pgd, unsigned long start)
285 {
286 	pud_t *pud = pud_offset(pgd, 0);
287 	unsigned long addr;
288 	unsigned int i;
289 
290 	for (i = 0; i < PTRS_PER_PUD; i++, pud++) {
291 		addr = start + i * PUD_SIZE;
292 		if (!pud_none(*pud) && !pud_is_leaf(*pud))
293 			/* pud exists */
294 			walk_pmd(st, pud, addr);
295 		else
296 			note_page(st, addr, 2, pud_val(*pud));
297 	}
298 }
299 
300 static void walk_pagetables(struct pg_state *st)
301 {
302 	pgd_t *pgd = pgd_offset_k(0UL);
303 	unsigned int i;
304 	unsigned long addr;
305 
306 	addr = st->start_address;
307 
308 	/*
309 	 * Traverse the linux pagetable structure and dump pages that are in
310 	 * the hash pagetable.
311 	 */
312 	for (i = 0; i < PTRS_PER_PGD; i++, pgd++, addr += PGDIR_SIZE) {
313 		if (!pgd_none(*pgd) && !pgd_is_leaf(*pgd))
314 			/* pgd exists */
315 			walk_pud(st, pgd, addr);
316 		else
317 			note_page(st, addr, 1, pgd_val(*pgd));
318 	}
319 }
320 
321 static void populate_markers(void)
322 {
323 	int i = 0;
324 
325 	address_markers[i++].start_address = PAGE_OFFSET;
326 	address_markers[i++].start_address = VMALLOC_START;
327 	address_markers[i++].start_address = VMALLOC_END;
328 #ifdef CONFIG_PPC64
329 	address_markers[i++].start_address = ISA_IO_BASE;
330 	address_markers[i++].start_address = ISA_IO_END;
331 	address_markers[i++].start_address = PHB_IO_BASE;
332 	address_markers[i++].start_address = PHB_IO_END;
333 	address_markers[i++].start_address = IOREMAP_BASE;
334 	address_markers[i++].start_address = IOREMAP_END;
335 	/* What is the ifdef about? */
336 #ifdef CONFIG_PPC_BOOK3S_64
337 	address_markers[i++].start_address =  H_VMEMMAP_START;
338 #else
339 	address_markers[i++].start_address =  VMEMMAP_BASE;
340 #endif
341 #else /* !CONFIG_PPC64 */
342 	address_markers[i++].start_address = ioremap_bot;
343 	address_markers[i++].start_address = IOREMAP_TOP;
344 #ifdef CONFIG_NOT_COHERENT_CACHE
345 	address_markers[i++].start_address = IOREMAP_TOP;
346 	address_markers[i++].start_address = IOREMAP_TOP +
347 					     CONFIG_CONSISTENT_SIZE;
348 #endif
349 #ifdef CONFIG_HIGHMEM
350 	address_markers[i++].start_address = PKMAP_BASE;
351 	address_markers[i++].start_address = PKMAP_ADDR(LAST_PKMAP);
352 #endif
353 	address_markers[i++].start_address = FIXADDR_START;
354 	address_markers[i++].start_address = FIXADDR_TOP;
355 #ifdef CONFIG_KASAN
356 	address_markers[i++].start_address = KASAN_SHADOW_START;
357 	address_markers[i++].start_address = KASAN_SHADOW_END;
358 #endif
359 #endif /* CONFIG_PPC64 */
360 }
361 
362 static int ptdump_show(struct seq_file *m, void *v)
363 {
364 	struct pg_state st = {
365 		.seq = m,
366 		.marker = address_markers,
367 	};
368 
369 	if (radix_enabled())
370 		st.start_address = PAGE_OFFSET;
371 	else
372 		st.start_address = KERN_VIRT_START;
373 
374 	/* Traverse kernel page tables */
375 	walk_pagetables(&st);
376 	note_page(&st, 0, 0, 0);
377 	return 0;
378 }
379 
380 
381 static int ptdump_open(struct inode *inode, struct file *file)
382 {
383 	return single_open(file, ptdump_show, NULL);
384 }
385 
386 static const struct file_operations ptdump_fops = {
387 	.open		= ptdump_open,
388 	.read		= seq_read,
389 	.llseek		= seq_lseek,
390 	.release	= single_release,
391 };
392 
393 static void build_pgtable_complete_mask(void)
394 {
395 	unsigned int i, j;
396 
397 	for (i = 0; i < ARRAY_SIZE(pg_level); i++)
398 		if (pg_level[i].flag)
399 			for (j = 0; j < pg_level[i].num; j++)
400 				pg_level[i].mask |= pg_level[i].flag[j].mask;
401 }
402 
403 #ifdef CONFIG_PPC_DEBUG_WX
404 void ptdump_check_wx(void)
405 {
406 	struct pg_state st = {
407 		.seq = NULL,
408 		.marker = address_markers,
409 		.check_wx = true,
410 	};
411 
412 	if (radix_enabled())
413 		st.start_address = PAGE_OFFSET;
414 	else
415 		st.start_address = KERN_VIRT_START;
416 
417 	walk_pagetables(&st);
418 
419 	if (st.wx_pages)
420 		pr_warn("Checked W+X mappings: FAILED, %lu W+X pages found\n",
421 			st.wx_pages);
422 	else
423 		pr_info("Checked W+X mappings: passed, no W+X pages found\n");
424 }
425 #endif
426 
427 static int ptdump_init(void)
428 {
429 	struct dentry *debugfs_file;
430 
431 	populate_markers();
432 	build_pgtable_complete_mask();
433 	debugfs_file = debugfs_create_file("kernel_page_tables", 0400, NULL,
434 			NULL, &ptdump_fops);
435 	return debugfs_file ? 0 : -ENOMEM;
436 }
437 device_initcall(ptdump_init);
438