xref: /openbmc/linux/arch/x86/mm/dump_pagetables.c (revision 6c33a6f4)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Debug helper to dump the current kernel pagetables of the system
4  * so that we can see what the various memory ranges are set to.
5  *
6  * (C) Copyright 2008 Intel Corporation
7  *
8  * Author: Arjan van de Ven <arjan@linux.intel.com>
9  */
10 
11 #include <linux/debugfs.h>
12 #include <linux/kasan.h>
13 #include <linux/mm.h>
14 #include <linux/init.h>
15 #include <linux/sched.h>
16 #include <linux/seq_file.h>
17 #include <linux/highmem.h>
18 #include <linux/pci.h>
19 #include <linux/ptdump.h>
20 
21 #include <asm/e820/types.h>
22 #include <asm/pgtable.h>
23 
24 /*
25  * The dumper groups pagetable entries of the same type into one, and for
26  * that it needs to keep some state when walking, and flush this state
27  * when a "break" in the continuity is found.
28  */
29 struct pg_state {
30 	struct ptdump_state ptdump;
31 	int level;
32 	pgprotval_t current_prot;
33 	pgprotval_t effective_prot;
34 	pgprotval_t prot_levels[5];
35 	unsigned long start_address;
36 	const struct addr_marker *marker;
37 	unsigned long lines;
38 	bool to_dmesg;
39 	bool check_wx;
40 	unsigned long wx_pages;
41 	struct seq_file *seq;
42 };
43 
44 struct addr_marker {
45 	unsigned long start_address;
46 	const char *name;
47 	unsigned long max_lines;
48 };
49 
50 /* Address space markers hints */
51 
52 #ifdef CONFIG_X86_64
53 
54 enum address_markers_idx {
55 	USER_SPACE_NR = 0,
56 	KERNEL_SPACE_NR,
57 #ifdef CONFIG_MODIFY_LDT_SYSCALL
58 	LDT_NR,
59 #endif
60 	LOW_KERNEL_NR,
61 	VMALLOC_START_NR,
62 	VMEMMAP_START_NR,
63 #ifdef CONFIG_KASAN
64 	KASAN_SHADOW_START_NR,
65 	KASAN_SHADOW_END_NR,
66 #endif
67 	CPU_ENTRY_AREA_NR,
68 #ifdef CONFIG_X86_ESPFIX64
69 	ESPFIX_START_NR,
70 #endif
71 #ifdef CONFIG_EFI
72 	EFI_END_NR,
73 #endif
74 	HIGH_KERNEL_NR,
75 	MODULES_VADDR_NR,
76 	MODULES_END_NR,
77 	FIXADDR_START_NR,
78 	END_OF_SPACE_NR,
79 };
80 
81 static struct addr_marker address_markers[] = {
82 	[USER_SPACE_NR]		= { 0,			"User Space" },
83 	[KERNEL_SPACE_NR]	= { (1UL << 63),	"Kernel Space" },
84 	[LOW_KERNEL_NR]		= { 0UL,		"Low Kernel Mapping" },
85 	[VMALLOC_START_NR]	= { 0UL,		"vmalloc() Area" },
86 	[VMEMMAP_START_NR]	= { 0UL,		"Vmemmap" },
87 #ifdef CONFIG_KASAN
88 	/*
89 	 * These fields get initialized with the (dynamic)
90 	 * KASAN_SHADOW_{START,END} values in pt_dump_init().
91 	 */
92 	[KASAN_SHADOW_START_NR]	= { 0UL,		"KASAN shadow" },
93 	[KASAN_SHADOW_END_NR]	= { 0UL,		"KASAN shadow end" },
94 #endif
95 #ifdef CONFIG_MODIFY_LDT_SYSCALL
96 	[LDT_NR]		= { 0UL,		"LDT remap" },
97 #endif
98 	[CPU_ENTRY_AREA_NR]	= { CPU_ENTRY_AREA_BASE,"CPU entry Area" },
99 #ifdef CONFIG_X86_ESPFIX64
100 	[ESPFIX_START_NR]	= { ESPFIX_BASE_ADDR,	"ESPfix Area", 16 },
101 #endif
102 #ifdef CONFIG_EFI
103 	[EFI_END_NR]		= { EFI_VA_END,		"EFI Runtime Services" },
104 #endif
105 	[HIGH_KERNEL_NR]	= { __START_KERNEL_map,	"High Kernel Mapping" },
106 	[MODULES_VADDR_NR]	= { MODULES_VADDR,	"Modules" },
107 	[MODULES_END_NR]	= { MODULES_END,	"End Modules" },
108 	[FIXADDR_START_NR]	= { FIXADDR_START,	"Fixmap Area" },
109 	[END_OF_SPACE_NR]	= { -1,			NULL }
110 };
111 
112 #define INIT_PGD	((pgd_t *) &init_top_pgt)
113 
114 #else /* CONFIG_X86_64 */
115 
116 enum address_markers_idx {
117 	USER_SPACE_NR = 0,
118 	KERNEL_SPACE_NR,
119 	VMALLOC_START_NR,
120 	VMALLOC_END_NR,
121 #ifdef CONFIG_HIGHMEM
122 	PKMAP_BASE_NR,
123 #endif
124 #ifdef CONFIG_MODIFY_LDT_SYSCALL
125 	LDT_NR,
126 #endif
127 	CPU_ENTRY_AREA_NR,
128 	FIXADDR_START_NR,
129 	END_OF_SPACE_NR,
130 };
131 
132 static struct addr_marker address_markers[] = {
133 	[USER_SPACE_NR]		= { 0,			"User Space" },
134 	[KERNEL_SPACE_NR]	= { PAGE_OFFSET,	"Kernel Mapping" },
135 	[VMALLOC_START_NR]	= { 0UL,		"vmalloc() Area" },
136 	[VMALLOC_END_NR]	= { 0UL,		"vmalloc() End" },
137 #ifdef CONFIG_HIGHMEM
138 	[PKMAP_BASE_NR]		= { 0UL,		"Persistent kmap() Area" },
139 #endif
140 #ifdef CONFIG_MODIFY_LDT_SYSCALL
141 	[LDT_NR]		= { 0UL,		"LDT remap" },
142 #endif
143 	[CPU_ENTRY_AREA_NR]	= { 0UL,		"CPU entry area" },
144 	[FIXADDR_START_NR]	= { 0UL,		"Fixmap area" },
145 	[END_OF_SPACE_NR]	= { -1,			NULL }
146 };
147 
148 #define INIT_PGD	(swapper_pg_dir)
149 
150 #endif /* !CONFIG_X86_64 */
151 
152 /* Multipliers for offsets within the PTEs */
153 #define PTE_LEVEL_MULT (PAGE_SIZE)
154 #define PMD_LEVEL_MULT (PTRS_PER_PTE * PTE_LEVEL_MULT)
155 #define PUD_LEVEL_MULT (PTRS_PER_PMD * PMD_LEVEL_MULT)
156 #define P4D_LEVEL_MULT (PTRS_PER_PUD * PUD_LEVEL_MULT)
157 #define PGD_LEVEL_MULT (PTRS_PER_P4D * P4D_LEVEL_MULT)
158 
159 #define pt_dump_seq_printf(m, to_dmesg, fmt, args...)		\
160 ({								\
161 	if (to_dmesg)					\
162 		printk(KERN_INFO fmt, ##args);			\
163 	else							\
164 		if (m)						\
165 			seq_printf(m, fmt, ##args);		\
166 })
167 
168 #define pt_dump_cont_printf(m, to_dmesg, fmt, args...)		\
169 ({								\
170 	if (to_dmesg)					\
171 		printk(KERN_CONT fmt, ##args);			\
172 	else							\
173 		if (m)						\
174 			seq_printf(m, fmt, ##args);		\
175 })
176 
177 /*
178  * Print a readable form of a pgprot_t to the seq_file
179  */
180 static void printk_prot(struct seq_file *m, pgprotval_t pr, int level, bool dmsg)
181 {
182 	static const char * const level_name[] =
183 		{ "pgd", "p4d", "pud", "pmd", "pte" };
184 
185 	if (!(pr & _PAGE_PRESENT)) {
186 		/* Not present */
187 		pt_dump_cont_printf(m, dmsg, "                              ");
188 	} else {
189 		if (pr & _PAGE_USER)
190 			pt_dump_cont_printf(m, dmsg, "USR ");
191 		else
192 			pt_dump_cont_printf(m, dmsg, "    ");
193 		if (pr & _PAGE_RW)
194 			pt_dump_cont_printf(m, dmsg, "RW ");
195 		else
196 			pt_dump_cont_printf(m, dmsg, "ro ");
197 		if (pr & _PAGE_PWT)
198 			pt_dump_cont_printf(m, dmsg, "PWT ");
199 		else
200 			pt_dump_cont_printf(m, dmsg, "    ");
201 		if (pr & _PAGE_PCD)
202 			pt_dump_cont_printf(m, dmsg, "PCD ");
203 		else
204 			pt_dump_cont_printf(m, dmsg, "    ");
205 
206 		/* Bit 7 has a different meaning on level 3 vs 4 */
207 		if (level <= 3 && pr & _PAGE_PSE)
208 			pt_dump_cont_printf(m, dmsg, "PSE ");
209 		else
210 			pt_dump_cont_printf(m, dmsg, "    ");
211 		if ((level == 4 && pr & _PAGE_PAT) ||
212 		    ((level == 3 || level == 2) && pr & _PAGE_PAT_LARGE))
213 			pt_dump_cont_printf(m, dmsg, "PAT ");
214 		else
215 			pt_dump_cont_printf(m, dmsg, "    ");
216 		if (pr & _PAGE_GLOBAL)
217 			pt_dump_cont_printf(m, dmsg, "GLB ");
218 		else
219 			pt_dump_cont_printf(m, dmsg, "    ");
220 		if (pr & _PAGE_NX)
221 			pt_dump_cont_printf(m, dmsg, "NX ");
222 		else
223 			pt_dump_cont_printf(m, dmsg, "x  ");
224 	}
225 	pt_dump_cont_printf(m, dmsg, "%s\n", level_name[level]);
226 }
227 
228 static void note_wx(struct pg_state *st, unsigned long addr)
229 {
230 	unsigned long npages;
231 
232 	npages = (addr - st->start_address) / PAGE_SIZE;
233 
234 #ifdef CONFIG_PCI_BIOS
235 	/*
236 	 * If PCI BIOS is enabled, the PCI BIOS area is forced to WX.
237 	 * Inform about it, but avoid the warning.
238 	 */
239 	if (pcibios_enabled && st->start_address >= PAGE_OFFSET + BIOS_BEGIN &&
240 	    addr <= PAGE_OFFSET + BIOS_END) {
241 		pr_warn_once("x86/mm: PCI BIOS W+X mapping %lu pages\n", npages);
242 		return;
243 	}
244 #endif
245 	/* Account the WX pages */
246 	st->wx_pages += npages;
247 	WARN_ONCE(__supported_pte_mask & _PAGE_NX,
248 		  "x86/mm: Found insecure W+X mapping at address %pS\n",
249 		  (void *)st->start_address);
250 }
251 
252 static inline pgprotval_t effective_prot(pgprotval_t prot1, pgprotval_t prot2)
253 {
254 	return (prot1 & prot2 & (_PAGE_USER | _PAGE_RW)) |
255 	       ((prot1 | prot2) & _PAGE_NX);
256 }
257 
258 /*
259  * This function gets called on a break in a continuous series
260  * of PTE entries; the next one is different so we need to
261  * print what we collected so far.
262  */
263 static void note_page(struct ptdump_state *pt_st, unsigned long addr, int level,
264 		      unsigned long val)
265 {
266 	struct pg_state *st = container_of(pt_st, struct pg_state, ptdump);
267 	pgprotval_t new_prot, new_eff;
268 	pgprotval_t cur, eff;
269 	static const char units[] = "BKMGTPE";
270 	struct seq_file *m = st->seq;
271 
272 	new_prot = val & PTE_FLAGS_MASK;
273 
274 	if (level > 0) {
275 		new_eff = effective_prot(st->prot_levels[level - 1],
276 					 new_prot);
277 	} else {
278 		new_eff = new_prot;
279 	}
280 
281 	if (level >= 0)
282 		st->prot_levels[level] = new_eff;
283 
284 	/*
285 	 * If we have a "break" in the series, we need to flush the state that
286 	 * we have now. "break" is either changing perms, levels or
287 	 * address space marker.
288 	 */
289 	cur = st->current_prot;
290 	eff = st->effective_prot;
291 
292 	if (st->level == -1) {
293 		/* First entry */
294 		st->current_prot = new_prot;
295 		st->effective_prot = new_eff;
296 		st->level = level;
297 		st->marker = address_markers;
298 		st->lines = 0;
299 		pt_dump_seq_printf(m, st->to_dmesg, "---[ %s ]---\n",
300 				   st->marker->name);
301 	} else if (new_prot != cur || new_eff != eff || level != st->level ||
302 		   addr >= st->marker[1].start_address) {
303 		const char *unit = units;
304 		unsigned long delta;
305 		int width = sizeof(unsigned long) * 2;
306 
307 		if (st->check_wx && (eff & _PAGE_RW) && !(eff & _PAGE_NX))
308 			note_wx(st, addr);
309 
310 		/*
311 		 * Now print the actual finished series
312 		 */
313 		if (!st->marker->max_lines ||
314 		    st->lines < st->marker->max_lines) {
315 			pt_dump_seq_printf(m, st->to_dmesg,
316 					   "0x%0*lx-0x%0*lx   ",
317 					   width, st->start_address,
318 					   width, addr);
319 
320 			delta = addr - st->start_address;
321 			while (!(delta & 1023) && unit[1]) {
322 				delta >>= 10;
323 				unit++;
324 			}
325 			pt_dump_cont_printf(m, st->to_dmesg, "%9lu%c ",
326 					    delta, *unit);
327 			printk_prot(m, st->current_prot, st->level,
328 				    st->to_dmesg);
329 		}
330 		st->lines++;
331 
332 		/*
333 		 * We print markers for special areas of address space,
334 		 * such as the start of vmalloc space etc.
335 		 * This helps in the interpretation.
336 		 */
337 		if (addr >= st->marker[1].start_address) {
338 			if (st->marker->max_lines &&
339 			    st->lines > st->marker->max_lines) {
340 				unsigned long nskip =
341 					st->lines - st->marker->max_lines;
342 				pt_dump_seq_printf(m, st->to_dmesg,
343 						   "... %lu entr%s skipped ... \n",
344 						   nskip,
345 						   nskip == 1 ? "y" : "ies");
346 			}
347 			st->marker++;
348 			st->lines = 0;
349 			pt_dump_seq_printf(m, st->to_dmesg, "---[ %s ]---\n",
350 					   st->marker->name);
351 		}
352 
353 		st->start_address = addr;
354 		st->current_prot = new_prot;
355 		st->effective_prot = new_eff;
356 		st->level = level;
357 	}
358 }
359 
360 static void ptdump_walk_pgd_level_core(struct seq_file *m,
361 				       struct mm_struct *mm, pgd_t *pgd,
362 				       bool checkwx, bool dmesg)
363 {
364 	const struct ptdump_range ptdump_ranges[] = {
365 #ifdef CONFIG_X86_64
366 	{0, PTRS_PER_PGD * PGD_LEVEL_MULT / 2},
367 	{GUARD_HOLE_END_ADDR, ~0UL},
368 #else
369 	{0, ~0UL},
370 #endif
371 	{0, 0}
372 };
373 
374 	struct pg_state st = {
375 		.ptdump = {
376 			.note_page	= note_page,
377 			.range		= ptdump_ranges
378 		},
379 		.level = -1,
380 		.to_dmesg	= dmesg,
381 		.check_wx	= checkwx,
382 		.seq		= m
383 	};
384 
385 	ptdump_walk_pgd(&st.ptdump, mm, pgd);
386 
387 	if (!checkwx)
388 		return;
389 	if (st.wx_pages)
390 		pr_info("x86/mm: Checked W+X mappings: FAILED, %lu W+X pages found.\n",
391 			st.wx_pages);
392 	else
393 		pr_info("x86/mm: Checked W+X mappings: passed, no W+X pages found.\n");
394 }
395 
396 void ptdump_walk_pgd_level(struct seq_file *m, struct mm_struct *mm)
397 {
398 	ptdump_walk_pgd_level_core(m, mm, mm->pgd, false, true);
399 }
400 
401 void ptdump_walk_pgd_level_debugfs(struct seq_file *m, struct mm_struct *mm,
402 				   bool user)
403 {
404 	pgd_t *pgd = mm->pgd;
405 #ifdef CONFIG_PAGE_TABLE_ISOLATION
406 	if (user && boot_cpu_has(X86_FEATURE_PTI))
407 		pgd = kernel_to_user_pgdp(pgd);
408 #endif
409 	ptdump_walk_pgd_level_core(m, mm, pgd, false, false);
410 }
411 EXPORT_SYMBOL_GPL(ptdump_walk_pgd_level_debugfs);
412 
413 void ptdump_walk_user_pgd_level_checkwx(void)
414 {
415 #ifdef CONFIG_PAGE_TABLE_ISOLATION
416 	pgd_t *pgd = INIT_PGD;
417 
418 	if (!(__supported_pte_mask & _PAGE_NX) ||
419 	    !boot_cpu_has(X86_FEATURE_PTI))
420 		return;
421 
422 	pr_info("x86/mm: Checking user space page tables\n");
423 	pgd = kernel_to_user_pgdp(pgd);
424 	ptdump_walk_pgd_level_core(NULL, &init_mm, pgd, true, false);
425 #endif
426 }
427 
428 void ptdump_walk_pgd_level_checkwx(void)
429 {
430 	ptdump_walk_pgd_level_core(NULL, &init_mm, INIT_PGD, true, false);
431 }
432 
433 static int __init pt_dump_init(void)
434 {
435 	/*
436 	 * Various markers are not compile-time constants, so assign them
437 	 * here.
438 	 */
439 #ifdef CONFIG_X86_64
440 	address_markers[LOW_KERNEL_NR].start_address = PAGE_OFFSET;
441 	address_markers[VMALLOC_START_NR].start_address = VMALLOC_START;
442 	address_markers[VMEMMAP_START_NR].start_address = VMEMMAP_START;
443 #ifdef CONFIG_MODIFY_LDT_SYSCALL
444 	address_markers[LDT_NR].start_address = LDT_BASE_ADDR;
445 #endif
446 #ifdef CONFIG_KASAN
447 	address_markers[KASAN_SHADOW_START_NR].start_address = KASAN_SHADOW_START;
448 	address_markers[KASAN_SHADOW_END_NR].start_address = KASAN_SHADOW_END;
449 #endif
450 #endif
451 #ifdef CONFIG_X86_32
452 	address_markers[VMALLOC_START_NR].start_address = VMALLOC_START;
453 	address_markers[VMALLOC_END_NR].start_address = VMALLOC_END;
454 # ifdef CONFIG_HIGHMEM
455 	address_markers[PKMAP_BASE_NR].start_address = PKMAP_BASE;
456 # endif
457 	address_markers[FIXADDR_START_NR].start_address = FIXADDR_START;
458 	address_markers[CPU_ENTRY_AREA_NR].start_address = CPU_ENTRY_AREA_BASE;
459 # ifdef CONFIG_MODIFY_LDT_SYSCALL
460 	address_markers[LDT_NR].start_address = LDT_BASE_ADDR;
461 # endif
462 #endif
463 	return 0;
464 }
465 __initcall(pt_dump_init);
466