xref: /linux/arch/powerpc/mm/ptdump/ptdump.c (revision 7a309195d11cde854eb75559fbd6b48f9e518f25)
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  */
56 struct 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 
69 struct addr_marker {
70 	unsigned long start_address;
71 	const char *name;
72 };
73 
74 static struct addr_marker address_markers[] = {
75 	{ 0,	"Start of kernel VM" },
76 	{ 0,	"vmalloc() Area" },
77 	{ 0,	"vmalloc() End" },
78 #ifdef CONFIG_PPC64
79 	{ 0,	"isa I/O start" },
80 	{ 0,	"isa I/O end" },
81 	{ 0,	"phb I/O start" },
82 	{ 0,	"phb I/O end" },
83 	{ 0,	"I/O remap start" },
84 	{ 0,	"I/O remap end" },
85 	{ 0,	"vmemmap start" },
86 #else
87 	{ 0,	"Early I/O remap start" },
88 	{ 0,	"Early I/O remap end" },
89 #ifdef CONFIG_HIGHMEM
90 	{ 0,	"Highmem PTEs start" },
91 	{ 0,	"Highmem PTEs end" },
92 #endif
93 	{ 0,	"Fixmap start" },
94 	{ 0,	"Fixmap end" },
95 #endif
96 #ifdef CONFIG_KASAN
97 	{ 0,	"kasan shadow mem start" },
98 	{ 0,	"kasan shadow mem end" },
99 #endif
100 	{ -1,	NULL },
101 };
102 
103 #define pt_dump_seq_printf(m, fmt, args...)	\
104 ({						\
105 	if (m)					\
106 		seq_printf(m, fmt, ##args);	\
107 })
108 
109 #define pt_dump_seq_putc(m, c)		\
110 ({					\
111 	if (m)				\
112 		seq_putc(m, c);		\
113 })
114 
115 void pt_dump_size(struct seq_file *m, unsigned long size)
116 {
117 	static const char units[] = "KMGTPE";
118 	const char *unit = units;
119 
120 	/* Work out what appropriate unit to use */
121 	while (!(size & 1023) && unit[1]) {
122 		size >>= 10;
123 		unit++;
124 	}
125 	pt_dump_seq_printf(m, "%9lu%c ", size, *unit);
126 }
127 
128 static void dump_flag_info(struct pg_state *st, const struct flag_info
129 		*flag, u64 pte, int num)
130 {
131 	unsigned int i;
132 
133 	for (i = 0; i < num; i++, flag++) {
134 		const char *s = NULL;
135 		u64 val;
136 
137 		/* flag not defined so don't check it */
138 		if (flag->mask == 0)
139 			continue;
140 		/* Some 'flags' are actually values */
141 		if (flag->is_val) {
142 			val = pte & flag->val;
143 			if (flag->shift)
144 				val = val >> flag->shift;
145 			pt_dump_seq_printf(st->seq, "  %s:%llx", flag->set, val);
146 		} else {
147 			if ((pte & flag->mask) == flag->val)
148 				s = flag->set;
149 			else
150 				s = flag->clear;
151 			if (s)
152 				pt_dump_seq_printf(st->seq, "  %s", s);
153 		}
154 		st->current_flags &= ~flag->mask;
155 	}
156 	if (st->current_flags != 0)
157 		pt_dump_seq_printf(st->seq, "  unknown flags:%llx", st->current_flags);
158 }
159 
160 static void dump_addr(struct pg_state *st, unsigned long addr)
161 {
162 	unsigned long delta;
163 
164 #ifdef CONFIG_PPC64
165 #define REG		"0x%016lx"
166 #else
167 #define REG		"0x%08lx"
168 #endif
169 
170 	pt_dump_seq_printf(st->seq, REG "-" REG " ", st->start_address, addr - 1);
171 	if (st->start_pa == st->last_pa && st->start_address + st->page_size != addr) {
172 		pt_dump_seq_printf(st->seq, "[" REG "]", st->start_pa);
173 		delta = st->page_size >> 10;
174 	} else {
175 		pt_dump_seq_printf(st->seq, " " REG " ", st->start_pa);
176 		delta = (addr - st->start_address) >> 10;
177 	}
178 	pt_dump_size(st->seq, delta);
179 }
180 
181 static void note_prot_wx(struct pg_state *st, unsigned long addr)
182 {
183 	pte_t pte = __pte(st->current_flags);
184 
185 	if (!IS_ENABLED(CONFIG_PPC_DEBUG_WX) || !st->check_wx)
186 		return;
187 
188 	if (!pte_write(pte) || !pte_exec(pte))
189 		return;
190 
191 	WARN_ONCE(1, "powerpc/mm: Found insecure W+X mapping at address %p/%pS\n",
192 		  (void *)st->start_address, (void *)st->start_address);
193 
194 	st->wx_pages += (addr - st->start_address) / PAGE_SIZE;
195 }
196 
197 static void note_page(struct pg_state *st, unsigned long addr,
198 	       unsigned int level, u64 val, unsigned long page_size)
199 {
200 	u64 flag = val & pg_level[level].mask;
201 	u64 pa = val & PTE_RPN_MASK;
202 
203 	/* At first no level is set */
204 	if (!st->level) {
205 		st->level = level;
206 		st->current_flags = flag;
207 		st->start_address = addr;
208 		st->start_pa = pa;
209 		st->last_pa = pa;
210 		st->page_size = page_size;
211 		pt_dump_seq_printf(st->seq, "---[ %s ]---\n", st->marker->name);
212 	/*
213 	 * Dump the section of virtual memory when:
214 	 *   - the PTE flags from one entry to the next differs.
215 	 *   - we change levels in the tree.
216 	 *   - the address is in a different section of memory and is thus
217 	 *   used for a different purpose, regardless of the flags.
218 	 *   - the pa of this page is not adjacent to the last inspected page
219 	 */
220 	} else if (flag != st->current_flags || level != st->level ||
221 		   addr >= st->marker[1].start_address ||
222 		   (pa != st->last_pa + st->page_size &&
223 		    (pa != st->start_pa || st->start_pa != st->last_pa))) {
224 
225 		/* Check the PTE flags */
226 		if (st->current_flags) {
227 			note_prot_wx(st, addr);
228 			dump_addr(st, addr);
229 
230 			/* Dump all the flags */
231 			if (pg_level[st->level].flag)
232 				dump_flag_info(st, pg_level[st->level].flag,
233 					  st->current_flags,
234 					  pg_level[st->level].num);
235 
236 			pt_dump_seq_putc(st->seq, '\n');
237 		}
238 
239 		/*
240 		 * Address indicates we have passed the end of the
241 		 * current section of virtual memory
242 		 */
243 		while (addr >= st->marker[1].start_address) {
244 			st->marker++;
245 			pt_dump_seq_printf(st->seq, "---[ %s ]---\n", st->marker->name);
246 		}
247 		st->start_address = addr;
248 		st->start_pa = pa;
249 		st->last_pa = pa;
250 		st->page_size = page_size;
251 		st->current_flags = flag;
252 		st->level = level;
253 	} else {
254 		st->last_pa = pa;
255 	}
256 }
257 
258 static void walk_pte(struct pg_state *st, pmd_t *pmd, unsigned long start)
259 {
260 	pte_t *pte = pte_offset_kernel(pmd, 0);
261 	unsigned long addr;
262 	unsigned int i;
263 
264 	for (i = 0; i < PTRS_PER_PTE; i++, pte++) {
265 		addr = start + i * PAGE_SIZE;
266 		note_page(st, addr, 4, pte_val(*pte), PAGE_SIZE);
267 
268 	}
269 }
270 
271 static void walk_hugepd(struct pg_state *st, hugepd_t *phpd, unsigned long start,
272 			int pdshift, int level)
273 {
274 #ifdef CONFIG_ARCH_HAS_HUGEPD
275 	unsigned int i;
276 	int shift = hugepd_shift(*phpd);
277 	int ptrs_per_hpd = pdshift - shift > 0 ? 1 << (pdshift - shift) : 1;
278 
279 	if (start & ((1 << shift) - 1))
280 		return;
281 
282 	for (i = 0; i < ptrs_per_hpd; i++) {
283 		unsigned long addr = start + (i << shift);
284 		pte_t *pte = hugepte_offset(*phpd, addr, pdshift);
285 
286 		note_page(st, addr, level + 1, pte_val(*pte), 1 << shift);
287 	}
288 #endif
289 }
290 
291 static void walk_pmd(struct pg_state *st, pud_t *pud, unsigned long start)
292 {
293 	pmd_t *pmd = pmd_offset(pud, 0);
294 	unsigned long addr;
295 	unsigned int i;
296 
297 	for (i = 0; i < PTRS_PER_PMD; i++, pmd++) {
298 		addr = start + i * PMD_SIZE;
299 		if (!pmd_none(*pmd) && !pmd_is_leaf(*pmd))
300 			/* pmd exists */
301 			walk_pte(st, pmd, addr);
302 		else
303 			note_page(st, addr, 3, pmd_val(*pmd), PMD_SIZE);
304 	}
305 }
306 
307 static void walk_pud(struct pg_state *st, p4d_t *p4d, unsigned long start)
308 {
309 	pud_t *pud = pud_offset(p4d, 0);
310 	unsigned long addr;
311 	unsigned int i;
312 
313 	for (i = 0; i < PTRS_PER_PUD; i++, pud++) {
314 		addr = start + i * PUD_SIZE;
315 		if (!pud_none(*pud) && !pud_is_leaf(*pud))
316 			/* pud exists */
317 			walk_pmd(st, pud, addr);
318 		else
319 			note_page(st, addr, 2, pud_val(*pud), PUD_SIZE);
320 	}
321 }
322 
323 static void walk_pagetables(struct pg_state *st)
324 {
325 	unsigned int i;
326 	unsigned long addr = st->start_address & PGDIR_MASK;
327 	pgd_t *pgd = pgd_offset_k(addr);
328 
329 	/*
330 	 * Traverse the linux pagetable structure and dump pages that are in
331 	 * the hash pagetable.
332 	 */
333 	for (i = pgd_index(addr); i < PTRS_PER_PGD; i++, pgd++, addr += PGDIR_SIZE) {
334 		p4d_t *p4d = p4d_offset(pgd, 0);
335 
336 		if (p4d_none(*p4d) || p4d_is_leaf(*p4d))
337 			note_page(st, addr, 1, p4d_val(*p4d), PGDIR_SIZE);
338 		else if (is_hugepd(__hugepd(p4d_val(*p4d))))
339 			walk_hugepd(st, (hugepd_t *)p4d, addr, PGDIR_SHIFT, 1);
340 		else
341 			/* p4d exists */
342 			walk_pud(st, p4d, addr);
343 	}
344 }
345 
346 static void populate_markers(void)
347 {
348 	int i = 0;
349 
350 	address_markers[i++].start_address = PAGE_OFFSET;
351 	address_markers[i++].start_address = VMALLOC_START;
352 	address_markers[i++].start_address = VMALLOC_END;
353 #ifdef CONFIG_PPC64
354 	address_markers[i++].start_address = ISA_IO_BASE;
355 	address_markers[i++].start_address = ISA_IO_END;
356 	address_markers[i++].start_address = PHB_IO_BASE;
357 	address_markers[i++].start_address = PHB_IO_END;
358 	address_markers[i++].start_address = IOREMAP_BASE;
359 	address_markers[i++].start_address = IOREMAP_END;
360 	/* What is the ifdef about? */
361 #ifdef CONFIG_PPC_BOOK3S_64
362 	address_markers[i++].start_address =  H_VMEMMAP_START;
363 #else
364 	address_markers[i++].start_address =  VMEMMAP_BASE;
365 #endif
366 #else /* !CONFIG_PPC64 */
367 	address_markers[i++].start_address = ioremap_bot;
368 	address_markers[i++].start_address = IOREMAP_TOP;
369 #ifdef CONFIG_HIGHMEM
370 	address_markers[i++].start_address = PKMAP_BASE;
371 	address_markers[i++].start_address = PKMAP_ADDR(LAST_PKMAP);
372 #endif
373 	address_markers[i++].start_address = FIXADDR_START;
374 	address_markers[i++].start_address = FIXADDR_TOP;
375 #ifdef CONFIG_KASAN
376 	address_markers[i++].start_address = KASAN_SHADOW_START;
377 	address_markers[i++].start_address = KASAN_SHADOW_END;
378 #endif
379 #endif /* CONFIG_PPC64 */
380 }
381 
382 static int ptdump_show(struct seq_file *m, void *v)
383 {
384 	struct pg_state st = {
385 		.seq = m,
386 		.marker = address_markers,
387 		.start_address = PAGE_OFFSET,
388 	};
389 
390 #ifdef CONFIG_PPC64
391 	if (!radix_enabled())
392 		st.start_address = KERN_VIRT_START;
393 #endif
394 
395 	/* Traverse kernel page tables */
396 	walk_pagetables(&st);
397 	note_page(&st, 0, 0, 0, 0);
398 	return 0;
399 }
400 
401 
402 static int ptdump_open(struct inode *inode, struct file *file)
403 {
404 	return single_open(file, ptdump_show, NULL);
405 }
406 
407 static const struct file_operations ptdump_fops = {
408 	.open		= ptdump_open,
409 	.read		= seq_read,
410 	.llseek		= seq_lseek,
411 	.release	= single_release,
412 };
413 
414 static void build_pgtable_complete_mask(void)
415 {
416 	unsigned int i, j;
417 
418 	for (i = 0; i < ARRAY_SIZE(pg_level); i++)
419 		if (pg_level[i].flag)
420 			for (j = 0; j < pg_level[i].num; j++)
421 				pg_level[i].mask |= pg_level[i].flag[j].mask;
422 }
423 
424 #ifdef CONFIG_PPC_DEBUG_WX
425 void ptdump_check_wx(void)
426 {
427 	struct pg_state st = {
428 		.seq = NULL,
429 		.marker = address_markers,
430 		.check_wx = true,
431 		.start_address = PAGE_OFFSET,
432 	};
433 
434 #ifdef CONFIG_PPC64
435 	if (!radix_enabled())
436 		st.start_address = KERN_VIRT_START;
437 #endif
438 
439 	walk_pagetables(&st);
440 
441 	if (st.wx_pages)
442 		pr_warn("Checked W+X mappings: FAILED, %lu W+X pages found\n",
443 			st.wx_pages);
444 	else
445 		pr_info("Checked W+X mappings: passed, no W+X pages found\n");
446 }
447 #endif
448 
449 static int ptdump_init(void)
450 {
451 	populate_markers();
452 	build_pgtable_complete_mask();
453 	debugfs_create_file("kernel_page_tables", 0400, NULL, NULL,
454 			    &ptdump_fops);
455 	return 0;
456 }
457 device_initcall(ptdump_init);
458