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# ifdef CONFIG_EFI
80	{ EFI_VA_END,		"EFI Runtime Services" },
81# endif
82	{ __START_KERNEL_map,   "High Kernel Mapping" },
83	{ MODULES_VADDR,        "Modules" },
84	{ MODULES_END,          "End Modules" },
85#else
86	{ PAGE_OFFSET,          "Kernel Mapping" },
87	{ 0/* VMALLOC_START */, "vmalloc() Area" },
88	{ 0/*VMALLOC_END*/,     "vmalloc() End" },
89# ifdef CONFIG_HIGHMEM
90	{ 0/*PKMAP_BASE*/,      "Persisent kmap() Area" },
91# endif
92	{ 0/*FIXADDR_START*/,   "Fixmap Area" },
93#endif
94	{ -1, NULL }		/* End of list */
95};
96
97/* Multipliers for offsets within the PTEs */
98#define PTE_LEVEL_MULT (PAGE_SIZE)
99#define PMD_LEVEL_MULT (PTRS_PER_PTE * PTE_LEVEL_MULT)
100#define PUD_LEVEL_MULT (PTRS_PER_PMD * PMD_LEVEL_MULT)
101#define PGD_LEVEL_MULT (PTRS_PER_PUD * PUD_LEVEL_MULT)
102
103#define pt_dump_seq_printf(m, to_dmesg, fmt, args...)		\
104({								\
105	if (to_dmesg)					\
106		printk(KERN_INFO fmt, ##args);			\
107	else							\
108		if (m)						\
109			seq_printf(m, fmt, ##args);		\
110})
111
112#define pt_dump_cont_printf(m, to_dmesg, fmt, args...)		\
113({								\
114	if (to_dmesg)					\
115		printk(KERN_CONT fmt, ##args);			\
116	else							\
117		if (m)						\
118			seq_printf(m, fmt, ##args);		\
119})
120
121/*
122 * Print a readable form of a pgprot_t to the seq_file
123 */
124static void printk_prot(struct seq_file *m, pgprot_t prot, int level, bool dmsg)
125{
126	pgprotval_t pr = pgprot_val(prot);
127	static const char * const level_name[] =
128		{ "cr3", "pgd", "pud", "pmd", "pte" };
129
130	if (!pgprot_val(prot)) {
131		/* Not present */
132		pt_dump_cont_printf(m, dmsg, "                              ");
133	} else {
134		if (pr & _PAGE_USER)
135			pt_dump_cont_printf(m, dmsg, "USR ");
136		else
137			pt_dump_cont_printf(m, dmsg, "    ");
138		if (pr & _PAGE_RW)
139			pt_dump_cont_printf(m, dmsg, "RW ");
140		else
141			pt_dump_cont_printf(m, dmsg, "ro ");
142		if (pr & _PAGE_PWT)
143			pt_dump_cont_printf(m, dmsg, "PWT ");
144		else
145			pt_dump_cont_printf(m, dmsg, "    ");
146		if (pr & _PAGE_PCD)
147			pt_dump_cont_printf(m, dmsg, "PCD ");
148		else
149			pt_dump_cont_printf(m, dmsg, "    ");
150
151		/* Bit 7 has a different meaning on level 3 vs 4 */
152		if (level <= 3 && pr & _PAGE_PSE)
153			pt_dump_cont_printf(m, dmsg, "PSE ");
154		else
155			pt_dump_cont_printf(m, dmsg, "    ");
156		if ((level == 4 && pr & _PAGE_PAT) ||
157		    ((level == 3 || level == 2) && pr & _PAGE_PAT_LARGE))
158			pt_dump_cont_printf(m, dmsg, "pat ");
159		else
160			pt_dump_cont_printf(m, dmsg, "    ");
161		if (pr & _PAGE_GLOBAL)
162			pt_dump_cont_printf(m, dmsg, "GLB ");
163		else
164			pt_dump_cont_printf(m, dmsg, "    ");
165		if (pr & _PAGE_NX)
166			pt_dump_cont_printf(m, dmsg, "NX ");
167		else
168			pt_dump_cont_printf(m, dmsg, "x  ");
169	}
170	pt_dump_cont_printf(m, dmsg, "%s\n", level_name[level]);
171}
172
173/*
174 * On 64 bits, sign-extend the 48 bit address to 64 bit
175 */
176static unsigned long normalize_addr(unsigned long u)
177{
178#ifdef CONFIG_X86_64
179	return (signed long)(u << 16) >> 16;
180#else
181	return u;
182#endif
183}
184
185/*
186 * This function gets called on a break in a continuous series
187 * of PTE entries; the next one is different so we need to
188 * print what we collected so far.
189 */
190static void note_page(struct seq_file *m, struct pg_state *st,
191		      pgprot_t new_prot, int level)
192{
193	pgprotval_t prot, cur;
194	static const char units[] = "BKMGTPE";
195
196	/*
197	 * If we have a "break" in the series, we need to flush the state that
198	 * we have now. "break" is either changing perms, levels or
199	 * address space marker.
200	 */
201	prot = pgprot_val(new_prot) & PTE_FLAGS_MASK;
202	cur = pgprot_val(st->current_prot) & PTE_FLAGS_MASK;
203
204	if (!st->level) {
205		/* First entry */
206		st->current_prot = new_prot;
207		st->level = level;
208		st->marker = address_markers;
209		st->lines = 0;
210		pt_dump_seq_printf(m, st->to_dmesg, "---[ %s ]---\n",
211				   st->marker->name);
212	} else if (prot != cur || level != st->level ||
213		   st->current_address >= st->marker[1].start_address) {
214		const char *unit = units;
215		unsigned long delta;
216		int width = sizeof(unsigned long) * 2;
217
218		/*
219		 * Now print the actual finished series
220		 */
221		if (!st->marker->max_lines ||
222		    st->lines < st->marker->max_lines) {
223			pt_dump_seq_printf(m, st->to_dmesg,
224					   "0x%0*lx-0x%0*lx   ",
225					   width, st->start_address,
226					   width, st->current_address);
227
228			delta = st->current_address - st->start_address;
229			while (!(delta & 1023) && unit[1]) {
230				delta >>= 10;
231				unit++;
232			}
233			pt_dump_cont_printf(m, st->to_dmesg, "%9lu%c ",
234					    delta, *unit);
235			printk_prot(m, st->current_prot, st->level,
236				    st->to_dmesg);
237		}
238		st->lines++;
239
240		/*
241		 * We print markers for special areas of address space,
242		 * such as the start of vmalloc space etc.
243		 * This helps in the interpretation.
244		 */
245		if (st->current_address >= st->marker[1].start_address) {
246			if (st->marker->max_lines &&
247			    st->lines > st->marker->max_lines) {
248				unsigned long nskip =
249					st->lines - st->marker->max_lines;
250				pt_dump_seq_printf(m, st->to_dmesg,
251						   "... %lu entr%s skipped ... \n",
252						   nskip,
253						   nskip == 1 ? "y" : "ies");
254			}
255			st->marker++;
256			st->lines = 0;
257			pt_dump_seq_printf(m, st->to_dmesg, "---[ %s ]---\n",
258					   st->marker->name);
259		}
260
261		st->start_address = st->current_address;
262		st->current_prot = new_prot;
263		st->level = level;
264	}
265}
266
267static void walk_pte_level(struct seq_file *m, struct pg_state *st, pmd_t addr,
268							unsigned long P)
269{
270	int i;
271	pte_t *start;
272
273	start = (pte_t *) pmd_page_vaddr(addr);
274	for (i = 0; i < PTRS_PER_PTE; i++) {
275		pgprot_t prot = pte_pgprot(*start);
276
277		st->current_address = normalize_addr(P + i * PTE_LEVEL_MULT);
278		note_page(m, st, prot, 4);
279		start++;
280	}
281}
282
283#if PTRS_PER_PMD > 1
284
285static void walk_pmd_level(struct seq_file *m, struct pg_state *st, pud_t addr,
286							unsigned long P)
287{
288	int i;
289	pmd_t *start;
290
291	start = (pmd_t *) pud_page_vaddr(addr);
292	for (i = 0; i < PTRS_PER_PMD; i++) {
293		st->current_address = normalize_addr(P + i * PMD_LEVEL_MULT);
294		if (!pmd_none(*start)) {
295			pgprotval_t prot = pmd_val(*start) & PTE_FLAGS_MASK;
296
297			if (pmd_large(*start) || !pmd_present(*start))
298				note_page(m, st, __pgprot(prot), 3);
299			else
300				walk_pte_level(m, st, *start,
301					       P + i * PMD_LEVEL_MULT);
302		} else
303			note_page(m, st, __pgprot(0), 3);
304		start++;
305	}
306}
307
308#else
309#define walk_pmd_level(m,s,a,p) walk_pte_level(m,s,__pmd(pud_val(a)),p)
310#define pud_large(a) pmd_large(__pmd(pud_val(a)))
311#define pud_none(a)  pmd_none(__pmd(pud_val(a)))
312#endif
313
314#if PTRS_PER_PUD > 1
315
316static void walk_pud_level(struct seq_file *m, struct pg_state *st, pgd_t addr,
317							unsigned long P)
318{
319	int i;
320	pud_t *start;
321
322	start = (pud_t *) pgd_page_vaddr(addr);
323
324	for (i = 0; i < PTRS_PER_PUD; i++) {
325		st->current_address = normalize_addr(P + i * PUD_LEVEL_MULT);
326		if (!pud_none(*start)) {
327			pgprotval_t prot = pud_val(*start) & PTE_FLAGS_MASK;
328
329			if (pud_large(*start) || !pud_present(*start))
330				note_page(m, st, __pgprot(prot), 2);
331			else
332				walk_pmd_level(m, st, *start,
333					       P + i * PUD_LEVEL_MULT);
334		} else
335			note_page(m, st, __pgprot(0), 2);
336
337		start++;
338	}
339}
340
341#else
342#define walk_pud_level(m,s,a,p) walk_pmd_level(m,s,__pud(pgd_val(a)),p)
343#define pgd_large(a) pud_large(__pud(pgd_val(a)))
344#define pgd_none(a)  pud_none(__pud(pgd_val(a)))
345#endif
346
347void ptdump_walk_pgd_level(struct seq_file *m, pgd_t *pgd)
348{
349#ifdef CONFIG_X86_64
350	pgd_t *start = (pgd_t *) &init_level4_pgt;
351#else
352	pgd_t *start = swapper_pg_dir;
353#endif
354	int i;
355	struct pg_state st = {};
356
357	if (pgd) {
358		start = pgd;
359		st.to_dmesg = true;
360	}
361
362	for (i = 0; i < PTRS_PER_PGD; i++) {
363		st.current_address = normalize_addr(i * PGD_LEVEL_MULT);
364		if (!pgd_none(*start)) {
365			pgprotval_t prot = pgd_val(*start) & PTE_FLAGS_MASK;
366
367			if (pgd_large(*start) || !pgd_present(*start))
368				note_page(m, &st, __pgprot(prot), 1);
369			else
370				walk_pud_level(m, &st, *start,
371					       i * PGD_LEVEL_MULT);
372		} else
373			note_page(m, &st, __pgprot(0), 1);
374
375		start++;
376	}
377
378	/* Flush out the last page */
379	st.current_address = normalize_addr(PTRS_PER_PGD*PGD_LEVEL_MULT);
380	note_page(m, &st, __pgprot(0), 0);
381}
382
383static int ptdump_show(struct seq_file *m, void *v)
384{
385	ptdump_walk_pgd_level(m, NULL);
386	return 0;
387}
388
389static int ptdump_open(struct inode *inode, struct file *filp)
390{
391	return single_open(filp, ptdump_show, NULL);
392}
393
394static const struct file_operations ptdump_fops = {
395	.open		= ptdump_open,
396	.read		= seq_read,
397	.llseek		= seq_lseek,
398	.release	= single_release,
399};
400
401static int pt_dump_init(void)
402{
403	struct dentry *pe;
404
405#ifdef CONFIG_X86_32
406	/* Not a compile-time constant on x86-32 */
407	address_markers[VMALLOC_START_NR].start_address = VMALLOC_START;
408	address_markers[VMALLOC_END_NR].start_address = VMALLOC_END;
409# ifdef CONFIG_HIGHMEM
410	address_markers[PKMAP_BASE_NR].start_address = PKMAP_BASE;
411# endif
412	address_markers[FIXADDR_START_NR].start_address = FIXADDR_START;
413#endif
414
415	pe = debugfs_create_file("kernel_page_tables", 0600, NULL, NULL,
416				 &ptdump_fops);
417	if (!pe)
418		return -ENOMEM;
419
420	return 0;
421}
422
423__initcall(pt_dump_init);
424MODULE_LICENSE("GPL");
425MODULE_AUTHOR("Arjan van de Ven <arjan@linux.intel.com>");
426MODULE_DESCRIPTION("Kernel debugging helper that dumps pagetables");
427