1// SPDX-License-Identifier: GPL-2.0
2
3/*
4 * Transitional page tables for kexec and hibernate
5 *
6 * This file derived from: arch/arm64/kernel/hibernate.c
7 *
8 * Copyright (c) 2021, Microsoft Corporation.
9 * Pasha Tatashin <pasha.tatashin@soleen.com>
10 *
11 */
12
13/*
14 * Transitional tables are used during system transferring from one world to
15 * another: such as during hibernate restore, and kexec reboots. During these
16 * phases one cannot rely on page table not being overwritten. This is because
17 * hibernate and kexec can overwrite the current page tables during transition.
18 */
19
20#include <asm/trans_pgd.h>
21#include <asm/pgalloc.h>
22#include <asm/pgtable.h>
23#include <linux/suspend.h>
24#include <linux/bug.h>
25#include <linux/mm.h>
26#include <linux/mmzone.h>
27#include <linux/kfence.h>
28
29static void *trans_alloc(struct trans_pgd_info *info)
30{
31	return info->trans_alloc_page(info->trans_alloc_arg);
32}
33
34static void _copy_pte(pte_t *dst_ptep, pte_t *src_ptep, unsigned long addr)
35{
36	pte_t pte = __ptep_get(src_ptep);
37
38	if (pte_valid(pte)) {
39		/*
40		 * Resume will overwrite areas that may be marked
41		 * read only (code, rodata). Clear the RDONLY bit from
42		 * the temporary mappings we use during restore.
43		 */
44		__set_pte(dst_ptep, pte_mkwrite_novma(pte));
45	} else if ((debug_pagealloc_enabled() ||
46		   is_kfence_address((void *)addr)) && !pte_none(pte)) {
47		/*
48		 * debug_pagealloc will removed the PTE_VALID bit if
49		 * the page isn't in use by the resume kernel. It may have
50		 * been in use by the original kernel, in which case we need
51		 * to put it back in our copy to do the restore.
52		 *
53		 * Before marking this entry valid, check the pfn should
54		 * be mapped.
55		 */
56		BUG_ON(!pfn_valid(pte_pfn(pte)));
57
58		__set_pte(dst_ptep, pte_mkpresent(pte_mkwrite_novma(pte)));
59	}
60}
61
62static int copy_pte(struct trans_pgd_info *info, pmd_t *dst_pmdp,
63		    pmd_t *src_pmdp, unsigned long start, unsigned long end)
64{
65	pte_t *src_ptep;
66	pte_t *dst_ptep;
67	unsigned long addr = start;
68
69	dst_ptep = trans_alloc(info);
70	if (!dst_ptep)
71		return -ENOMEM;
72	pmd_populate_kernel(NULL, dst_pmdp, dst_ptep);
73	dst_ptep = pte_offset_kernel(dst_pmdp, start);
74
75	src_ptep = pte_offset_kernel(src_pmdp, start);
76	do {
77		_copy_pte(dst_ptep, src_ptep, addr);
78	} while (dst_ptep++, src_ptep++, addr += PAGE_SIZE, addr != end);
79
80	return 0;
81}
82
83static int copy_pmd(struct trans_pgd_info *info, pud_t *dst_pudp,
84		    pud_t *src_pudp, unsigned long start, unsigned long end)
85{
86	pmd_t *src_pmdp;
87	pmd_t *dst_pmdp;
88	unsigned long next;
89	unsigned long addr = start;
90
91	if (pud_none(READ_ONCE(*dst_pudp))) {
92		dst_pmdp = trans_alloc(info);
93		if (!dst_pmdp)
94			return -ENOMEM;
95		pud_populate(NULL, dst_pudp, dst_pmdp);
96	}
97	dst_pmdp = pmd_offset(dst_pudp, start);
98
99	src_pmdp = pmd_offset(src_pudp, start);
100	do {
101		pmd_t pmd = READ_ONCE(*src_pmdp);
102
103		next = pmd_addr_end(addr, end);
104		if (pmd_none(pmd))
105			continue;
106		if (pmd_table(pmd)) {
107			if (copy_pte(info, dst_pmdp, src_pmdp, addr, next))
108				return -ENOMEM;
109		} else {
110			set_pmd(dst_pmdp,
111				__pmd(pmd_val(pmd) & ~PMD_SECT_RDONLY));
112		}
113	} while (dst_pmdp++, src_pmdp++, addr = next, addr != end);
114
115	return 0;
116}
117
118static int copy_pud(struct trans_pgd_info *info, p4d_t *dst_p4dp,
119		    p4d_t *src_p4dp, unsigned long start,
120		    unsigned long end)
121{
122	pud_t *dst_pudp;
123	pud_t *src_pudp;
124	unsigned long next;
125	unsigned long addr = start;
126
127	if (p4d_none(READ_ONCE(*dst_p4dp))) {
128		dst_pudp = trans_alloc(info);
129		if (!dst_pudp)
130			return -ENOMEM;
131		p4d_populate(NULL, dst_p4dp, dst_pudp);
132	}
133	dst_pudp = pud_offset(dst_p4dp, start);
134
135	src_pudp = pud_offset(src_p4dp, start);
136	do {
137		pud_t pud = READ_ONCE(*src_pudp);
138
139		next = pud_addr_end(addr, end);
140		if (pud_none(pud))
141			continue;
142		if (pud_table(pud)) {
143			if (copy_pmd(info, dst_pudp, src_pudp, addr, next))
144				return -ENOMEM;
145		} else {
146			set_pud(dst_pudp,
147				__pud(pud_val(pud) & ~PUD_SECT_RDONLY));
148		}
149	} while (dst_pudp++, src_pudp++, addr = next, addr != end);
150
151	return 0;
152}
153
154static int copy_p4d(struct trans_pgd_info *info, pgd_t *dst_pgdp,
155		    pgd_t *src_pgdp, unsigned long start,
156		    unsigned long end)
157{
158	p4d_t *dst_p4dp;
159	p4d_t *src_p4dp;
160	unsigned long next;
161	unsigned long addr = start;
162
163	dst_p4dp = p4d_offset(dst_pgdp, start);
164	src_p4dp = p4d_offset(src_pgdp, start);
165	do {
166		next = p4d_addr_end(addr, end);
167		if (p4d_none(READ_ONCE(*src_p4dp)))
168			continue;
169		if (copy_pud(info, dst_p4dp, src_p4dp, addr, next))
170			return -ENOMEM;
171	} while (dst_p4dp++, src_p4dp++, addr = next, addr != end);
172
173	return 0;
174}
175
176static int copy_page_tables(struct trans_pgd_info *info, pgd_t *dst_pgdp,
177			    unsigned long start, unsigned long end)
178{
179	unsigned long next;
180	unsigned long addr = start;
181	pgd_t *src_pgdp = pgd_offset_k(start);
182
183	dst_pgdp = pgd_offset_pgd(dst_pgdp, start);
184	do {
185		next = pgd_addr_end(addr, end);
186		if (pgd_none(READ_ONCE(*src_pgdp)))
187			continue;
188		if (copy_p4d(info, dst_pgdp, src_pgdp, addr, next))
189			return -ENOMEM;
190	} while (dst_pgdp++, src_pgdp++, addr = next, addr != end);
191
192	return 0;
193}
194
195/*
196 * Create trans_pgd and copy linear map.
197 * info:	contains allocator and its argument
198 * dst_pgdp:	new page table that is created, and to which map is copied.
199 * start:	Start of the interval (inclusive).
200 * end:		End of the interval (exclusive).
201 *
202 * Returns 0 on success, and -ENOMEM on failure.
203 */
204int trans_pgd_create_copy(struct trans_pgd_info *info, pgd_t **dst_pgdp,
205			  unsigned long start, unsigned long end)
206{
207	int rc;
208	pgd_t *trans_pgd = trans_alloc(info);
209
210	if (!trans_pgd) {
211		pr_err("Failed to allocate memory for temporary page tables.\n");
212		return -ENOMEM;
213	}
214
215	rc = copy_page_tables(info, trans_pgd, start, end);
216	if (!rc)
217		*dst_pgdp = trans_pgd;
218
219	return rc;
220}
221
222/*
223 * The page we want to idmap may be outside the range covered by VA_BITS that
224 * can be built using the kernel's p?d_populate() helpers. As a one off, for a
225 * single page, we build these page tables bottom up and just assume that will
226 * need the maximum T0SZ.
227 *
228 * Returns 0 on success, and -ENOMEM on failure.
229 * On success trans_ttbr0 contains page table with idmapped page, t0sz is set to
230 * maximum T0SZ for this page.
231 */
232int trans_pgd_idmap_page(struct trans_pgd_info *info, phys_addr_t *trans_ttbr0,
233			 unsigned long *t0sz, void *page)
234{
235	phys_addr_t dst_addr = virt_to_phys(page);
236	unsigned long pfn = __phys_to_pfn(dst_addr);
237	int max_msb = (dst_addr & GENMASK(52, 48)) ? 51 : 47;
238	int bits_mapped = PAGE_SHIFT - 4;
239	unsigned long level_mask, prev_level_entry, *levels[4];
240	int this_level, index, level_lsb, level_msb;
241
242	dst_addr &= PAGE_MASK;
243	prev_level_entry = pte_val(pfn_pte(pfn, PAGE_KERNEL_ROX));
244
245	for (this_level = 3; this_level >= 0; this_level--) {
246		levels[this_level] = trans_alloc(info);
247		if (!levels[this_level])
248			return -ENOMEM;
249
250		level_lsb = ARM64_HW_PGTABLE_LEVEL_SHIFT(this_level);
251		level_msb = min(level_lsb + bits_mapped, max_msb);
252		level_mask = GENMASK_ULL(level_msb, level_lsb);
253
254		index = (dst_addr & level_mask) >> level_lsb;
255		*(levels[this_level] + index) = prev_level_entry;
256
257		pfn = virt_to_pfn(levels[this_level]);
258		prev_level_entry = pte_val(pfn_pte(pfn,
259						   __pgprot(PMD_TYPE_TABLE)));
260
261		if (level_msb == max_msb)
262			break;
263	}
264
265	*trans_ttbr0 = phys_to_ttbr(__pfn_to_phys(pfn));
266	*t0sz = TCR_T0SZ(max_msb + 1);
267
268	return 0;
269}
270
271/*
272 * Create a copy of the vector table so we can call HVC_SET_VECTORS or
273 * HVC_SOFT_RESTART from contexts where the table may be overwritten.
274 */
275int trans_pgd_copy_el2_vectors(struct trans_pgd_info *info,
276			       phys_addr_t *el2_vectors)
277{
278	void *hyp_stub = trans_alloc(info);
279
280	if (!hyp_stub)
281		return -ENOMEM;
282	*el2_vectors = virt_to_phys(hyp_stub);
283	memcpy(hyp_stub, &trans_pgd_stub_vectors, ARM64_VECTOR_TABLE_LEN);
284	caches_clean_inval_pou((unsigned long)hyp_stub,
285			       (unsigned long)hyp_stub +
286			       ARM64_VECTOR_TABLE_LEN);
287	dcache_clean_inval_poc((unsigned long)hyp_stub,
288			       (unsigned long)hyp_stub +
289			       ARM64_VECTOR_TABLE_LEN);
290
291	return 0;
292}
293