xref: /openbmc/linux/arch/x86/mm/dump_pagetables.c (revision 2cf1c348)
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 
23 /*
24  * The dumper groups pagetable entries of the same type into one, and for
25  * that it needs to keep some state when walking, and flush this state
26  * when a "break" in the continuity is found.
27  */
28 struct pg_state {
29 	struct ptdump_state ptdump;
30 	int level;
31 	pgprotval_t current_prot;
32 	pgprotval_t effective_prot;
33 	pgprotval_t prot_levels[5];
34 	unsigned long start_address;
35 	const struct addr_marker *marker;
36 	unsigned long lines;
37 	bool to_dmesg;
38 	bool check_wx;
39 	unsigned long wx_pages;
40 	struct seq_file *seq;
41 };
42 
43 struct addr_marker {
44 	unsigned long start_address;
45 	const char *name;
46 	unsigned long max_lines;
47 };
48 
49 /* Address space markers hints */
50 
51 #ifdef CONFIG_X86_64
52 
53 enum address_markers_idx {
54 	USER_SPACE_NR = 0,
55 	KERNEL_SPACE_NR,
56 #ifdef CONFIG_MODIFY_LDT_SYSCALL
57 	LDT_NR,
58 #endif
59 	LOW_KERNEL_NR,
60 	VMALLOC_START_NR,
61 	VMEMMAP_START_NR,
62 #ifdef CONFIG_KASAN
63 	KASAN_SHADOW_START_NR,
64 	KASAN_SHADOW_END_NR,
65 #endif
66 	CPU_ENTRY_AREA_NR,
67 #ifdef CONFIG_X86_ESPFIX64
68 	ESPFIX_START_NR,
69 #endif
70 #ifdef CONFIG_EFI
71 	EFI_END_NR,
72 #endif
73 	HIGH_KERNEL_NR,
74 	MODULES_VADDR_NR,
75 	MODULES_END_NR,
76 	FIXADDR_START_NR,
77 	END_OF_SPACE_NR,
78 };
79 
80 static struct addr_marker address_markers[] = {
81 	[USER_SPACE_NR]		= { 0,			"User Space" },
82 	[KERNEL_SPACE_NR]	= { (1UL << 63),	"Kernel Space" },
83 	[LOW_KERNEL_NR]		= { 0UL,		"Low Kernel Mapping" },
84 	[VMALLOC_START_NR]	= { 0UL,		"vmalloc() Area" },
85 	[VMEMMAP_START_NR]	= { 0UL,		"Vmemmap" },
86 #ifdef CONFIG_KASAN
87 	/*
88 	 * These fields get initialized with the (dynamic)
89 	 * KASAN_SHADOW_{START,END} values in pt_dump_init().
90 	 */
91 	[KASAN_SHADOW_START_NR]	= { 0UL,		"KASAN shadow" },
92 	[KASAN_SHADOW_END_NR]	= { 0UL,		"KASAN shadow end" },
93 #endif
94 #ifdef CONFIG_MODIFY_LDT_SYSCALL
95 	[LDT_NR]		= { 0UL,		"LDT remap" },
96 #endif
97 	[CPU_ENTRY_AREA_NR]	= { CPU_ENTRY_AREA_BASE,"CPU entry Area" },
98 #ifdef CONFIG_X86_ESPFIX64
99 	[ESPFIX_START_NR]	= { ESPFIX_BASE_ADDR,	"ESPfix Area", 16 },
100 #endif
101 #ifdef CONFIG_EFI
102 	[EFI_END_NR]		= { EFI_VA_END,		"EFI Runtime Services" },
103 #endif
104 	[HIGH_KERNEL_NR]	= { __START_KERNEL_map,	"High Kernel Mapping" },
105 	[MODULES_VADDR_NR]	= { MODULES_VADDR,	"Modules" },
106 	[MODULES_END_NR]	= { MODULES_END,	"End Modules" },
107 	[FIXADDR_START_NR]	= { FIXADDR_START,	"Fixmap Area" },
108 	[END_OF_SPACE_NR]	= { -1,			NULL }
109 };
110 
111 #define INIT_PGD	((pgd_t *) &init_top_pgt)
112 
113 #else /* CONFIG_X86_64 */
114 
115 enum address_markers_idx {
116 	USER_SPACE_NR = 0,
117 	KERNEL_SPACE_NR,
118 	VMALLOC_START_NR,
119 	VMALLOC_END_NR,
120 #ifdef CONFIG_HIGHMEM
121 	PKMAP_BASE_NR,
122 #endif
123 #ifdef CONFIG_MODIFY_LDT_SYSCALL
124 	LDT_NR,
125 #endif
126 	CPU_ENTRY_AREA_NR,
127 	FIXADDR_START_NR,
128 	END_OF_SPACE_NR,
129 };
130 
131 static struct addr_marker address_markers[] = {
132 	[USER_SPACE_NR]		= { 0,			"User Space" },
133 	[KERNEL_SPACE_NR]	= { PAGE_OFFSET,	"Kernel Mapping" },
134 	[VMALLOC_START_NR]	= { 0UL,		"vmalloc() Area" },
135 	[VMALLOC_END_NR]	= { 0UL,		"vmalloc() End" },
136 #ifdef CONFIG_HIGHMEM
137 	[PKMAP_BASE_NR]		= { 0UL,		"Persistent kmap() Area" },
138 #endif
139 #ifdef CONFIG_MODIFY_LDT_SYSCALL
140 	[LDT_NR]		= { 0UL,		"LDT remap" },
141 #endif
142 	[CPU_ENTRY_AREA_NR]	= { 0UL,		"CPU entry area" },
143 	[FIXADDR_START_NR]	= { 0UL,		"Fixmap area" },
144 	[END_OF_SPACE_NR]	= { -1,			NULL }
145 };
146 
147 #define INIT_PGD	(swapper_pg_dir)
148 
149 #endif /* !CONFIG_X86_64 */
150 
151 /* Multipliers for offsets within the PTEs */
152 #define PTE_LEVEL_MULT (PAGE_SIZE)
153 #define PMD_LEVEL_MULT (PTRS_PER_PTE * PTE_LEVEL_MULT)
154 #define PUD_LEVEL_MULT (PTRS_PER_PMD * PMD_LEVEL_MULT)
155 #define P4D_LEVEL_MULT (PTRS_PER_PUD * PUD_LEVEL_MULT)
156 #define PGD_LEVEL_MULT (PTRS_PER_P4D * P4D_LEVEL_MULT)
157 
158 #define pt_dump_seq_printf(m, to_dmesg, fmt, args...)		\
159 ({								\
160 	if (to_dmesg)					\
161 		printk(KERN_INFO fmt, ##args);			\
162 	else							\
163 		if (m)						\
164 			seq_printf(m, fmt, ##args);		\
165 })
166 
167 #define pt_dump_cont_printf(m, to_dmesg, fmt, args...)		\
168 ({								\
169 	if (to_dmesg)					\
170 		printk(KERN_CONT fmt, ##args);			\
171 	else							\
172 		if (m)						\
173 			seq_printf(m, fmt, ##args);		\
174 })
175 
176 /*
177  * Print a readable form of a pgprot_t to the seq_file
178  */
179 static void printk_prot(struct seq_file *m, pgprotval_t pr, int level, bool dmsg)
180 {
181 	static const char * const level_name[] =
182 		{ "pgd", "p4d", "pud", "pmd", "pte" };
183 
184 	if (!(pr & _PAGE_PRESENT)) {
185 		/* Not present */
186 		pt_dump_cont_printf(m, dmsg, "                              ");
187 	} else {
188 		if (pr & _PAGE_USER)
189 			pt_dump_cont_printf(m, dmsg, "USR ");
190 		else
191 			pt_dump_cont_printf(m, dmsg, "    ");
192 		if (pr & _PAGE_RW)
193 			pt_dump_cont_printf(m, dmsg, "RW ");
194 		else
195 			pt_dump_cont_printf(m, dmsg, "ro ");
196 		if (pr & _PAGE_PWT)
197 			pt_dump_cont_printf(m, dmsg, "PWT ");
198 		else
199 			pt_dump_cont_printf(m, dmsg, "    ");
200 		if (pr & _PAGE_PCD)
201 			pt_dump_cont_printf(m, dmsg, "PCD ");
202 		else
203 			pt_dump_cont_printf(m, dmsg, "    ");
204 
205 		/* Bit 7 has a different meaning on level 3 vs 4 */
206 		if (level <= 3 && pr & _PAGE_PSE)
207 			pt_dump_cont_printf(m, dmsg, "PSE ");
208 		else
209 			pt_dump_cont_printf(m, dmsg, "    ");
210 		if ((level == 4 && pr & _PAGE_PAT) ||
211 		    ((level == 3 || level == 2) && pr & _PAGE_PAT_LARGE))
212 			pt_dump_cont_printf(m, dmsg, "PAT ");
213 		else
214 			pt_dump_cont_printf(m, dmsg, "    ");
215 		if (pr & _PAGE_GLOBAL)
216 			pt_dump_cont_printf(m, dmsg, "GLB ");
217 		else
218 			pt_dump_cont_printf(m, dmsg, "    ");
219 		if (pr & _PAGE_NX)
220 			pt_dump_cont_printf(m, dmsg, "NX ");
221 		else
222 			pt_dump_cont_printf(m, dmsg, "x  ");
223 	}
224 	pt_dump_cont_printf(m, dmsg, "%s\n", level_name[level]);
225 }
226 
227 static void note_wx(struct pg_state *st, unsigned long addr)
228 {
229 	unsigned long npages;
230 
231 	npages = (addr - st->start_address) / PAGE_SIZE;
232 
233 #ifdef CONFIG_PCI_BIOS
234 	/*
235 	 * If PCI BIOS is enabled, the PCI BIOS area is forced to WX.
236 	 * Inform about it, but avoid the warning.
237 	 */
238 	if (pcibios_enabled && st->start_address >= PAGE_OFFSET + BIOS_BEGIN &&
239 	    addr <= PAGE_OFFSET + BIOS_END) {
240 		pr_warn_once("x86/mm: PCI BIOS W+X mapping %lu pages\n", npages);
241 		return;
242 	}
243 #endif
244 	/* Account the WX pages */
245 	st->wx_pages += npages;
246 	WARN_ONCE(__supported_pte_mask & _PAGE_NX,
247 		  "x86/mm: Found insecure W+X mapping at address %pS\n",
248 		  (void *)st->start_address);
249 }
250 
251 static void effective_prot(struct ptdump_state *pt_st, int level, u64 val)
252 {
253 	struct pg_state *st = container_of(pt_st, struct pg_state, ptdump);
254 	pgprotval_t prot = val & PTE_FLAGS_MASK;
255 	pgprotval_t effective;
256 
257 	if (level > 0) {
258 		pgprotval_t higher_prot = st->prot_levels[level - 1];
259 
260 		effective = (higher_prot & prot & (_PAGE_USER | _PAGE_RW)) |
261 			    ((higher_prot | prot) & _PAGE_NX);
262 	} else {
263 		effective = prot;
264 	}
265 
266 	st->prot_levels[level] = effective;
267 }
268 
269 /*
270  * This function gets called on a break in a continuous series
271  * of PTE entries; the next one is different so we need to
272  * print what we collected so far.
273  */
274 static void note_page(struct ptdump_state *pt_st, unsigned long addr, int level,
275 		      u64 val)
276 {
277 	struct pg_state *st = container_of(pt_st, struct pg_state, ptdump);
278 	pgprotval_t new_prot, new_eff;
279 	pgprotval_t cur, eff;
280 	static const char units[] = "BKMGTPE";
281 	struct seq_file *m = st->seq;
282 
283 	new_prot = val & PTE_FLAGS_MASK;
284 	if (!val)
285 		new_eff = 0;
286 	else
287 		new_eff = st->prot_levels[level];
288 
289 	/*
290 	 * If we have a "break" in the series, we need to flush the state that
291 	 * we have now. "break" is either changing perms, levels or
292 	 * address space marker.
293 	 */
294 	cur = st->current_prot;
295 	eff = st->effective_prot;
296 
297 	if (st->level == -1) {
298 		/* First entry */
299 		st->current_prot = new_prot;
300 		st->effective_prot = new_eff;
301 		st->level = level;
302 		st->marker = address_markers;
303 		st->lines = 0;
304 		pt_dump_seq_printf(m, st->to_dmesg, "---[ %s ]---\n",
305 				   st->marker->name);
306 	} else if (new_prot != cur || new_eff != eff || level != st->level ||
307 		   addr >= st->marker[1].start_address) {
308 		const char *unit = units;
309 		unsigned long delta;
310 		int width = sizeof(unsigned long) * 2;
311 
312 		if (st->check_wx && (eff & _PAGE_RW) && !(eff & _PAGE_NX))
313 			note_wx(st, addr);
314 
315 		/*
316 		 * Now print the actual finished series
317 		 */
318 		if (!st->marker->max_lines ||
319 		    st->lines < st->marker->max_lines) {
320 			pt_dump_seq_printf(m, st->to_dmesg,
321 					   "0x%0*lx-0x%0*lx   ",
322 					   width, st->start_address,
323 					   width, addr);
324 
325 			delta = addr - st->start_address;
326 			while (!(delta & 1023) && unit[1]) {
327 				delta >>= 10;
328 				unit++;
329 			}
330 			pt_dump_cont_printf(m, st->to_dmesg, "%9lu%c ",
331 					    delta, *unit);
332 			printk_prot(m, st->current_prot, st->level,
333 				    st->to_dmesg);
334 		}
335 		st->lines++;
336 
337 		/*
338 		 * We print markers for special areas of address space,
339 		 * such as the start of vmalloc space etc.
340 		 * This helps in the interpretation.
341 		 */
342 		if (addr >= st->marker[1].start_address) {
343 			if (st->marker->max_lines &&
344 			    st->lines > st->marker->max_lines) {
345 				unsigned long nskip =
346 					st->lines - st->marker->max_lines;
347 				pt_dump_seq_printf(m, st->to_dmesg,
348 						   "... %lu entr%s skipped ... \n",
349 						   nskip,
350 						   nskip == 1 ? "y" : "ies");
351 			}
352 			st->marker++;
353 			st->lines = 0;
354 			pt_dump_seq_printf(m, st->to_dmesg, "---[ %s ]---\n",
355 					   st->marker->name);
356 		}
357 
358 		st->start_address = addr;
359 		st->current_prot = new_prot;
360 		st->effective_prot = new_eff;
361 		st->level = level;
362 	}
363 }
364 
365 static void ptdump_walk_pgd_level_core(struct seq_file *m,
366 				       struct mm_struct *mm, pgd_t *pgd,
367 				       bool checkwx, bool dmesg)
368 {
369 	const struct ptdump_range ptdump_ranges[] = {
370 #ifdef CONFIG_X86_64
371 	{0, PTRS_PER_PGD * PGD_LEVEL_MULT / 2},
372 	{GUARD_HOLE_END_ADDR, ~0UL},
373 #else
374 	{0, ~0UL},
375 #endif
376 	{0, 0}
377 };
378 
379 	struct pg_state st = {
380 		.ptdump = {
381 			.note_page	= note_page,
382 			.effective_prot = effective_prot,
383 			.range		= ptdump_ranges
384 		},
385 		.level = -1,
386 		.to_dmesg	= dmesg,
387 		.check_wx	= checkwx,
388 		.seq		= m
389 	};
390 
391 	ptdump_walk_pgd(&st.ptdump, mm, pgd);
392 
393 	if (!checkwx)
394 		return;
395 	if (st.wx_pages)
396 		pr_info("x86/mm: Checked W+X mappings: FAILED, %lu W+X pages found.\n",
397 			st.wx_pages);
398 	else
399 		pr_info("x86/mm: Checked W+X mappings: passed, no W+X pages found.\n");
400 }
401 
402 void ptdump_walk_pgd_level(struct seq_file *m, struct mm_struct *mm)
403 {
404 	ptdump_walk_pgd_level_core(m, mm, mm->pgd, false, true);
405 }
406 
407 void ptdump_walk_pgd_level_debugfs(struct seq_file *m, struct mm_struct *mm,
408 				   bool user)
409 {
410 	pgd_t *pgd = mm->pgd;
411 #ifdef CONFIG_PAGE_TABLE_ISOLATION
412 	if (user && boot_cpu_has(X86_FEATURE_PTI))
413 		pgd = kernel_to_user_pgdp(pgd);
414 #endif
415 	ptdump_walk_pgd_level_core(m, mm, pgd, false, false);
416 }
417 EXPORT_SYMBOL_GPL(ptdump_walk_pgd_level_debugfs);
418 
419 void ptdump_walk_user_pgd_level_checkwx(void)
420 {
421 #ifdef CONFIG_PAGE_TABLE_ISOLATION
422 	pgd_t *pgd = INIT_PGD;
423 
424 	if (!(__supported_pte_mask & _PAGE_NX) ||
425 	    !boot_cpu_has(X86_FEATURE_PTI))
426 		return;
427 
428 	pr_info("x86/mm: Checking user space page tables\n");
429 	pgd = kernel_to_user_pgdp(pgd);
430 	ptdump_walk_pgd_level_core(NULL, &init_mm, pgd, true, false);
431 #endif
432 }
433 
434 void ptdump_walk_pgd_level_checkwx(void)
435 {
436 	ptdump_walk_pgd_level_core(NULL, &init_mm, INIT_PGD, true, false);
437 }
438 
439 static int __init pt_dump_init(void)
440 {
441 	/*
442 	 * Various markers are not compile-time constants, so assign them
443 	 * here.
444 	 */
445 #ifdef CONFIG_X86_64
446 	address_markers[LOW_KERNEL_NR].start_address = PAGE_OFFSET;
447 	address_markers[VMALLOC_START_NR].start_address = VMALLOC_START;
448 	address_markers[VMEMMAP_START_NR].start_address = VMEMMAP_START;
449 #ifdef CONFIG_MODIFY_LDT_SYSCALL
450 	address_markers[LDT_NR].start_address = LDT_BASE_ADDR;
451 #endif
452 #ifdef CONFIG_KASAN
453 	address_markers[KASAN_SHADOW_START_NR].start_address = KASAN_SHADOW_START;
454 	address_markers[KASAN_SHADOW_END_NR].start_address = KASAN_SHADOW_END;
455 #endif
456 #endif
457 #ifdef CONFIG_X86_32
458 	address_markers[VMALLOC_START_NR].start_address = VMALLOC_START;
459 	address_markers[VMALLOC_END_NR].start_address = VMALLOC_END;
460 # ifdef CONFIG_HIGHMEM
461 	address_markers[PKMAP_BASE_NR].start_address = PKMAP_BASE;
462 # endif
463 	address_markers[FIXADDR_START_NR].start_address = FIXADDR_START;
464 	address_markers[CPU_ENTRY_AREA_NR].start_address = CPU_ENTRY_AREA_BASE;
465 # ifdef CONFIG_MODIFY_LDT_SYSCALL
466 	address_markers[LDT_NR].start_address = LDT_BASE_ADDR;
467 # endif
468 #endif
469 	return 0;
470 }
471 __initcall(pt_dump_init);
472