1/*
2 * Debug helper to dump the current kernel pagetables of the system
3 * so that we can see what the various memory ranges are set to.
4 *
5 * (C) Copyright 2008 Intel Corporation
6 *
7 * Author: Arjan van de Ven <arjan@linux.intel.com>
8 *
9 * This program is free software; you can redistribute it and/or
10 * modify it under the terms of the GNU General Public License
11 * as published by the Free Software Foundation; version 2
12 * of the License.
13 */
14
15#include <linux/debugfs.h>
16#include <linux/mm.h>
17#include <linux/module.h>
18#include <linux/seq_file.h>
19
20#include <asm/pgtable.h>
21
22/*
23 * The dumper groups pagetable entries of the same type into one, and for
24 * that it needs to keep some state when walking, and flush this state
25 * when a "break" in the continuity is found.
26 */
27struct pg_state {
28	int level;
29	pgprot_t current_prot;
30	unsigned long start_address;
31	unsigned long current_address;
32	const struct addr_marker *marker;
33	unsigned long lines;
34	bool to_dmesg;
35};
36
37struct addr_marker {
38	unsigned long start_address;
39	const char *name;
40	unsigned long max_lines;
41};
42
43/* indices for address_markers; keep sync'd w/ address_markers below */
44enum address_markers_idx {
45	USER_SPACE_NR = 0,
46#ifdef CONFIG_X86_64
47	KERNEL_SPACE_NR,
48	LOW_KERNEL_NR,
49	VMALLOC_START_NR,
50	VMEMMAP_START_NR,
51# ifdef CONFIG_X86_ESPFIX64
52	ESPFIX_START_NR,
53# endif
54	HIGH_KERNEL_NR,
55	MODULES_VADDR_NR,
56	MODULES_END_NR,
57#else
58	KERNEL_SPACE_NR,
59	VMALLOC_START_NR,
60	VMALLOC_END_NR,
61# ifdef CONFIG_HIGHMEM
62	PKMAP_BASE_NR,
63# endif
64	FIXADDR_START_NR,
65#endif
66};
67
68/* Address space markers hints */
69static struct addr_marker address_markers[] = {
70	{ 0, "User Space" },
71#ifdef CONFIG_X86_64
72	{ 0x8000000000000000UL, "Kernel Space" },
73	{ PAGE_OFFSET,		"Low Kernel Mapping" },
74	{ VMALLOC_START,        "vmalloc() Area" },
75	{ VMEMMAP_START,        "Vmemmap" },
76# ifdef CONFIG_X86_ESPFIX64
77	{ ESPFIX_BASE_ADDR,	"ESPfix Area", 16 },
78# endif
79	{ __START_KERNEL_map,   "High Kernel Mapping" },
80	{ MODULES_VADDR,        "Modules" },
81	{ MODULES_END,          "End Modules" },
82#else
83	{ PAGE_OFFSET,          "Kernel Mapping" },
84	{ 0/* VMALLOC_START */, "vmalloc() Area" },
85	{ 0/*VMALLOC_END*/,     "vmalloc() End" },
86# ifdef CONFIG_HIGHMEM
87	{ 0/*PKMAP_BASE*/,      "Persisent kmap() Area" },
88# endif
89	{ 0/*FIXADDR_START*/,   "Fixmap Area" },
90#endif
91	{ -1, NULL }		/* End of list */
92};
93
94/* Multipliers for offsets within the PTEs */
95#define PTE_LEVEL_MULT (PAGE_SIZE)
96#define PMD_LEVEL_MULT (PTRS_PER_PTE * PTE_LEVEL_MULT)
97#define PUD_LEVEL_MULT (PTRS_PER_PMD * PMD_LEVEL_MULT)
98#define PGD_LEVEL_MULT (PTRS_PER_PUD * PUD_LEVEL_MULT)
99
100#define pt_dump_seq_printf(m, to_dmesg, fmt, args...)		\
101({								\
102	if (to_dmesg)					\
103		printk(KERN_INFO fmt, ##args);			\
104	else							\
105		if (m)						\
106			seq_printf(m, fmt, ##args);		\
107})
108
109#define pt_dump_cont_printf(m, to_dmesg, fmt, args...)		\
110({								\
111	if (to_dmesg)					\
112		printk(KERN_CONT fmt, ##args);			\
113	else							\
114		if (m)						\
115			seq_printf(m, fmt, ##args);		\
116})
117
118/*
119 * Print a readable form of a pgprot_t to the seq_file
120 */
121static void printk_prot(struct seq_file *m, pgprot_t prot, int level, bool dmsg)
122{
123	pgprotval_t pr = pgprot_val(prot);
124	static const char * const level_name[] =
125		{ "cr3", "pgd", "pud", "pmd", "pte" };
126
127	if (!pgprot_val(prot)) {
128		/* Not present */
129		pt_dump_cont_printf(m, dmsg, "                          ");
130	} else {
131		if (pr & _PAGE_USER)
132			pt_dump_cont_printf(m, dmsg, "USR ");
133		else
134			pt_dump_cont_printf(m, dmsg, "    ");
135		if (pr & _PAGE_RW)
136			pt_dump_cont_printf(m, dmsg, "RW ");
137		else
138			pt_dump_cont_printf(m, dmsg, "ro ");
139		if (pr & _PAGE_PWT)
140			pt_dump_cont_printf(m, dmsg, "PWT ");
141		else
142			pt_dump_cont_printf(m, dmsg, "    ");
143		if (pr & _PAGE_PCD)
144			pt_dump_cont_printf(m, dmsg, "PCD ");
145		else
146			pt_dump_cont_printf(m, dmsg, "    ");
147
148		/* Bit 9 has a different meaning on level 3 vs 4 */
149		if (level <= 3) {
150			if (pr & _PAGE_PSE)
151				pt_dump_cont_printf(m, dmsg, "PSE ");
152			else
153				pt_dump_cont_printf(m, dmsg, "    ");
154		} else {
155			if (pr & _PAGE_PAT)
156				pt_dump_cont_printf(m, dmsg, "pat ");
157			else
158				pt_dump_cont_printf(m, dmsg, "    ");
159		}
160		if (pr & _PAGE_GLOBAL)
161			pt_dump_cont_printf(m, dmsg, "GLB ");
162		else
163			pt_dump_cont_printf(m, dmsg, "    ");
164		if (pr & _PAGE_NX)
165			pt_dump_cont_printf(m, dmsg, "NX ");
166		else
167			pt_dump_cont_printf(m, dmsg, "x  ");
168	}
169	pt_dump_cont_printf(m, dmsg, "%s\n", level_name[level]);
170}
171
172/*
173 * On 64 bits, sign-extend the 48 bit address to 64 bit
174 */
175static unsigned long normalize_addr(unsigned long u)
176{
177#ifdef CONFIG_X86_64
178	return (signed long)(u << 16) >> 16;
179#else
180	return u;
181#endif
182}
183
184/*
185 * This function gets called on a break in a continuous series
186 * of PTE entries; the next one is different so we need to
187 * print what we collected so far.
188 */
189static void note_page(struct seq_file *m, struct pg_state *st,
190		      pgprot_t new_prot, int level)
191{
192	pgprotval_t prot, cur;
193	static const char units[] = "BKMGTPE";
194
195	/*
196	 * If we have a "break" in the series, we need to flush the state that
197	 * we have now. "break" is either changing perms, levels or
198	 * address space marker.
199	 */
200	prot = pgprot_val(new_prot) & PTE_FLAGS_MASK;
201	cur = pgprot_val(st->current_prot) & PTE_FLAGS_MASK;
202
203	if (!st->level) {
204		/* First entry */
205		st->current_prot = new_prot;
206		st->level = level;
207		st->marker = address_markers;
208		st->lines = 0;
209		pt_dump_seq_printf(m, st->to_dmesg, "---[ %s ]---\n",
210				   st->marker->name);
211	} else if (prot != cur || level != st->level ||
212		   st->current_address >= st->marker[1].start_address) {
213		const char *unit = units;
214		unsigned long delta;
215		int width = sizeof(unsigned long) * 2;
216
217		/*
218		 * Now print the actual finished series
219		 */
220		if (!st->marker->max_lines ||
221		    st->lines < st->marker->max_lines) {
222			pt_dump_seq_printf(m, st->to_dmesg,
223					   "0x%0*lx-0x%0*lx   ",
224					   width, st->start_address,
225					   width, st->current_address);
226
227			delta = st->current_address - st->start_address;
228			while (!(delta & 1023) && unit[1]) {
229				delta >>= 10;
230				unit++;
231			}
232			pt_dump_cont_printf(m, st->to_dmesg, "%9lu%c ",
233					    delta, *unit);
234			printk_prot(m, st->current_prot, st->level,
235				    st->to_dmesg);
236		}
237		st->lines++;
238
239		/*
240		 * We print markers for special areas of address space,
241		 * such as the start of vmalloc space etc.
242		 * This helps in the interpretation.
243		 */
244		if (st->current_address >= st->marker[1].start_address) {
245			if (st->marker->max_lines &&
246			    st->lines > st->marker->max_lines) {
247				unsigned long nskip =
248					st->lines - st->marker->max_lines;
249				pt_dump_seq_printf(m, st->to_dmesg,
250						   "... %lu entr%s skipped ... \n",
251						   nskip,
252						   nskip == 1 ? "y" : "ies");
253			}
254			st->marker++;
255			st->lines = 0;
256			pt_dump_seq_printf(m, st->to_dmesg, "---[ %s ]---\n",
257					   st->marker->name);
258		}
259
260		st->start_address = st->current_address;
261		st->current_prot = new_prot;
262		st->level = level;
263	}
264}
265
266static void walk_pte_level(struct seq_file *m, struct pg_state *st, pmd_t addr,
267							unsigned long P)
268{
269	int i;
270	pte_t *start;
271
272	start = (pte_t *) pmd_page_vaddr(addr);
273	for (i = 0; i < PTRS_PER_PTE; i++) {
274		pgprot_t prot = pte_pgprot(*start);
275
276		st->current_address = normalize_addr(P + i * PTE_LEVEL_MULT);
277		note_page(m, st, prot, 4);
278		start++;
279	}
280}
281
282#if PTRS_PER_PMD > 1
283
284static void walk_pmd_level(struct seq_file *m, struct pg_state *st, pud_t addr,
285							unsigned long P)
286{
287	int i;
288	pmd_t *start;
289
290	start = (pmd_t *) pud_page_vaddr(addr);
291	for (i = 0; i < PTRS_PER_PMD; i++) {
292		st->current_address = normalize_addr(P + i * PMD_LEVEL_MULT);
293		if (!pmd_none(*start)) {
294			pgprotval_t prot = pmd_val(*start) & PTE_FLAGS_MASK;
295
296			if (pmd_large(*start) || !pmd_present(*start))
297				note_page(m, st, __pgprot(prot), 3);
298			else
299				walk_pte_level(m, st, *start,
300					       P + i * PMD_LEVEL_MULT);
301		} else
302			note_page(m, st, __pgprot(0), 3);
303		start++;
304	}
305}
306
307#else
308#define walk_pmd_level(m,s,a,p) walk_pte_level(m,s,__pmd(pud_val(a)),p)
309#define pud_large(a) pmd_large(__pmd(pud_val(a)))
310#define pud_none(a)  pmd_none(__pmd(pud_val(a)))
311#endif
312
313#if PTRS_PER_PUD > 1
314
315static void walk_pud_level(struct seq_file *m, struct pg_state *st, pgd_t addr,
316							unsigned long P)
317{
318	int i;
319	pud_t *start;
320
321	start = (pud_t *) pgd_page_vaddr(addr);
322
323	for (i = 0; i < PTRS_PER_PUD; i++) {
324		st->current_address = normalize_addr(P + i * PUD_LEVEL_MULT);
325		if (!pud_none(*start)) {
326			pgprotval_t prot = pud_val(*start) & PTE_FLAGS_MASK;
327
328			if (pud_large(*start) || !pud_present(*start))
329				note_page(m, st, __pgprot(prot), 2);
330			else
331				walk_pmd_level(m, st, *start,
332					       P + i * PUD_LEVEL_MULT);
333		} else
334			note_page(m, st, __pgprot(0), 2);
335
336		start++;
337	}
338}
339
340#else
341#define walk_pud_level(m,s,a,p) walk_pmd_level(m,s,__pud(pgd_val(a)),p)
342#define pgd_large(a) pud_large(__pud(pgd_val(a)))
343#define pgd_none(a)  pud_none(__pud(pgd_val(a)))
344#endif
345
346void ptdump_walk_pgd_level(struct seq_file *m, pgd_t *pgd)
347{
348#ifdef CONFIG_X86_64
349	pgd_t *start = (pgd_t *) &init_level4_pgt;
350#else
351	pgd_t *start = swapper_pg_dir;
352#endif
353	int i;
354	struct pg_state st = {};
355
356	if (pgd) {
357		start = pgd;
358		st.to_dmesg = true;
359	}
360
361	for (i = 0; i < PTRS_PER_PGD; i++) {
362		st.current_address = normalize_addr(i * PGD_LEVEL_MULT);
363		if (!pgd_none(*start)) {
364			pgprotval_t prot = pgd_val(*start) & PTE_FLAGS_MASK;
365
366			if (pgd_large(*start) || !pgd_present(*start))
367				note_page(m, &st, __pgprot(prot), 1);
368			else
369				walk_pud_level(m, &st, *start,
370					       i * PGD_LEVEL_MULT);
371		} else
372			note_page(m, &st, __pgprot(0), 1);
373
374		start++;
375	}
376
377	/* Flush out the last page */
378	st.current_address = normalize_addr(PTRS_PER_PGD*PGD_LEVEL_MULT);
379	note_page(m, &st, __pgprot(0), 0);
380}
381
382static int ptdump_show(struct seq_file *m, void *v)
383{
384	ptdump_walk_pgd_level(m, NULL);
385	return 0;
386}
387
388static int ptdump_open(struct inode *inode, struct file *filp)
389{
390	return single_open(filp, ptdump_show, NULL);
391}
392
393static const struct file_operations ptdump_fops = {
394	.open		= ptdump_open,
395	.read		= seq_read,
396	.llseek		= seq_lseek,
397	.release	= single_release,
398};
399
400static int pt_dump_init(void)
401{
402	struct dentry *pe;
403
404#ifdef CONFIG_X86_32
405	/* Not a compile-time constant on x86-32 */
406	address_markers[VMALLOC_START_NR].start_address = VMALLOC_START;
407	address_markers[VMALLOC_END_NR].start_address = VMALLOC_END;
408# ifdef CONFIG_HIGHMEM
409	address_markers[PKMAP_BASE_NR].start_address = PKMAP_BASE;
410# endif
411	address_markers[FIXADDR_START_NR].start_address = FIXADDR_START;
412#endif
413
414	pe = debugfs_create_file("kernel_page_tables", 0600, NULL, NULL,
415				 &ptdump_fops);
416	if (!pe)
417		return -ENOMEM;
418
419	return 0;
420}
421
422__initcall(pt_dump_init);
423MODULE_LICENSE("GPL");
424MODULE_AUTHOR("Arjan van de Ven <arjan@linux.intel.com>");
425MODULE_DESCRIPTION("Kernel debugging helper that dumps pagetables");
426