xref: /linux/arch/arm64/mm/trans_pgd.c (revision 111b29599c189247919a514b6c3a93e828db27c2)
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 
trans_alloc(struct trans_pgd_info * info)29 static void *trans_alloc(struct trans_pgd_info *info)
30 {
31 	return info->trans_alloc_page(info->trans_alloc_arg);
32 }
33 
_copy_pte(pte_t * dst_ptep,pte_t * src_ptep,unsigned long addr)34 static 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 (!pte_none(pte)) {
46 		/*
47 		 * debug_pagealloc will removed the PTE_VALID bit if
48 		 * the page isn't in use by the resume kernel. It may have
49 		 * been in use by the original kernel, in which case we need
50 		 * to put it back in our copy to do the restore.
51 		 *
52 		 * Other cases include kfence / vmalloc / memfd_secret which
53 		 * may call `set_direct_map_invalid_noflush()`.
54 		 *
55 		 * Before marking this entry valid, check the pfn should
56 		 * be mapped.
57 		 */
58 		BUG_ON(!pfn_valid(pte_pfn(pte)));
59 
60 		__set_pte(dst_ptep, pte_mkvalid(pte_mkwrite_novma(pte)));
61 	}
62 }
63 
copy_pte(struct trans_pgd_info * info,pmd_t * dst_pmdp,pmd_t * src_pmdp,unsigned long start,unsigned long end)64 static int copy_pte(struct trans_pgd_info *info, pmd_t *dst_pmdp,
65 		    pmd_t *src_pmdp, unsigned long start, unsigned long end)
66 {
67 	pte_t *src_ptep;
68 	pte_t *dst_ptep;
69 	unsigned long addr = start;
70 
71 	dst_ptep = trans_alloc(info);
72 	if (!dst_ptep)
73 		return -ENOMEM;
74 	pmd_populate_kernel(NULL, dst_pmdp, dst_ptep);
75 	dst_ptep = pte_offset_kernel(dst_pmdp, start);
76 
77 	src_ptep = pte_offset_kernel(src_pmdp, start);
78 	do {
79 		_copy_pte(dst_ptep, src_ptep, addr);
80 	} while (dst_ptep++, src_ptep++, addr += PAGE_SIZE, addr != end);
81 
82 	return 0;
83 }
84 
copy_pmd(struct trans_pgd_info * info,pud_t * dst_pudp,pud_t * src_pudp,unsigned long start,unsigned long end)85 static int copy_pmd(struct trans_pgd_info *info, pud_t *dst_pudp,
86 		    pud_t *src_pudp, unsigned long start, unsigned long end)
87 {
88 	pmd_t *src_pmdp;
89 	pmd_t *dst_pmdp;
90 	unsigned long next;
91 	unsigned long addr = start;
92 
93 	if (pud_none(READ_ONCE(*dst_pudp))) {
94 		dst_pmdp = trans_alloc(info);
95 		if (!dst_pmdp)
96 			return -ENOMEM;
97 		pud_populate(NULL, dst_pudp, dst_pmdp);
98 	}
99 	dst_pmdp = pmd_offset(dst_pudp, start);
100 
101 	src_pmdp = pmd_offset(src_pudp, start);
102 	do {
103 		pmd_t pmd = READ_ONCE(*src_pmdp);
104 
105 		next = pmd_addr_end(addr, end);
106 		if (pmd_none(pmd))
107 			continue;
108 		if (pmd_table(pmd)) {
109 			if (copy_pte(info, dst_pmdp, src_pmdp, addr, next))
110 				return -ENOMEM;
111 		} else {
112 			set_pmd(dst_pmdp,
113 				__pmd(pmd_val(pmd) & ~PMD_SECT_RDONLY));
114 		}
115 	} while (dst_pmdp++, src_pmdp++, addr = next, addr != end);
116 
117 	return 0;
118 }
119 
copy_pud(struct trans_pgd_info * info,p4d_t * dst_p4dp,p4d_t * src_p4dp,unsigned long start,unsigned long end)120 static int copy_pud(struct trans_pgd_info *info, p4d_t *dst_p4dp,
121 		    p4d_t *src_p4dp, unsigned long start,
122 		    unsigned long end)
123 {
124 	pud_t *dst_pudp;
125 	pud_t *src_pudp;
126 	unsigned long next;
127 	unsigned long addr = start;
128 
129 	if (p4d_none(READ_ONCE(*dst_p4dp))) {
130 		dst_pudp = trans_alloc(info);
131 		if (!dst_pudp)
132 			return -ENOMEM;
133 		p4d_populate(NULL, dst_p4dp, dst_pudp);
134 	}
135 	dst_pudp = pud_offset(dst_p4dp, start);
136 
137 	src_pudp = pud_offset(src_p4dp, start);
138 	do {
139 		pud_t pud = READ_ONCE(*src_pudp);
140 
141 		next = pud_addr_end(addr, end);
142 		if (pud_none(pud))
143 			continue;
144 		if (pud_table(pud)) {
145 			if (copy_pmd(info, dst_pudp, src_pudp, addr, next))
146 				return -ENOMEM;
147 		} else {
148 			set_pud(dst_pudp,
149 				__pud(pud_val(pud) & ~PUD_SECT_RDONLY));
150 		}
151 	} while (dst_pudp++, src_pudp++, addr = next, addr != end);
152 
153 	return 0;
154 }
155 
copy_p4d(struct trans_pgd_info * info,pgd_t * dst_pgdp,pgd_t * src_pgdp,unsigned long start,unsigned long end)156 static int copy_p4d(struct trans_pgd_info *info, pgd_t *dst_pgdp,
157 		    pgd_t *src_pgdp, unsigned long start,
158 		    unsigned long end)
159 {
160 	p4d_t *dst_p4dp;
161 	p4d_t *src_p4dp;
162 	unsigned long next;
163 	unsigned long addr = start;
164 
165 	if (pgd_none(READ_ONCE(*dst_pgdp))) {
166 		dst_p4dp = trans_alloc(info);
167 		if (!dst_p4dp)
168 			return -ENOMEM;
169 		pgd_populate(NULL, dst_pgdp, dst_p4dp);
170 	}
171 
172 	dst_p4dp = p4d_offset(dst_pgdp, start);
173 	src_p4dp = p4d_offset(src_pgdp, start);
174 	do {
175 		next = p4d_addr_end(addr, end);
176 		if (p4d_none(READ_ONCE(*src_p4dp)))
177 			continue;
178 		if (copy_pud(info, dst_p4dp, src_p4dp, addr, next))
179 			return -ENOMEM;
180 	} while (dst_p4dp++, src_p4dp++, addr = next, addr != end);
181 
182 	return 0;
183 }
184 
copy_page_tables(struct trans_pgd_info * info,pgd_t * dst_pgdp,unsigned long start,unsigned long end)185 static int copy_page_tables(struct trans_pgd_info *info, pgd_t *dst_pgdp,
186 			    unsigned long start, unsigned long end)
187 {
188 	unsigned long next;
189 	unsigned long addr = start;
190 	pgd_t *src_pgdp = pgd_offset_k(start);
191 
192 	dst_pgdp = pgd_offset_pgd(dst_pgdp, start);
193 	do {
194 		next = pgd_addr_end(addr, end);
195 		if (pgd_none(READ_ONCE(*src_pgdp)))
196 			continue;
197 		if (copy_p4d(info, dst_pgdp, src_pgdp, addr, next))
198 			return -ENOMEM;
199 	} while (dst_pgdp++, src_pgdp++, addr = next, addr != end);
200 
201 	return 0;
202 }
203 
204 /*
205  * Create trans_pgd and copy linear map.
206  * info:	contains allocator and its argument
207  * dst_pgdp:	new page table that is created, and to which map is copied.
208  * start:	Start of the interval (inclusive).
209  * end:		End of the interval (exclusive).
210  *
211  * Returns 0 on success, and -ENOMEM on failure.
212  */
trans_pgd_create_copy(struct trans_pgd_info * info,pgd_t ** dst_pgdp,unsigned long start,unsigned long end)213 int trans_pgd_create_copy(struct trans_pgd_info *info, pgd_t **dst_pgdp,
214 			  unsigned long start, unsigned long end)
215 {
216 	int rc;
217 	pgd_t *trans_pgd = trans_alloc(info);
218 
219 	if (!trans_pgd) {
220 		pr_err("Failed to allocate memory for temporary page tables.\n");
221 		return -ENOMEM;
222 	}
223 
224 	rc = copy_page_tables(info, trans_pgd, start, end);
225 	if (!rc)
226 		*dst_pgdp = trans_pgd;
227 
228 	return rc;
229 }
230 
231 /*
232  * The page we want to idmap may be outside the range covered by VA_BITS that
233  * can be built using the kernel's p?d_populate() helpers. As a one off, for a
234  * single page, we build these page tables bottom up and just assume that will
235  * need the maximum T0SZ.
236  *
237  * Returns 0 on success, and -ENOMEM on failure.
238  * On success trans_ttbr0 contains page table with idmapped page, t0sz is set to
239  * maximum T0SZ for this page.
240  */
trans_pgd_idmap_page(struct trans_pgd_info * info,phys_addr_t * trans_ttbr0,unsigned long * t0sz,void * page)241 int trans_pgd_idmap_page(struct trans_pgd_info *info, phys_addr_t *trans_ttbr0,
242 			 unsigned long *t0sz, void *page)
243 {
244 	phys_addr_t dst_addr = virt_to_phys(page);
245 	unsigned long pfn = __phys_to_pfn(dst_addr);
246 	int max_msb = (dst_addr & GENMASK(52, 48)) ? 51 : 47;
247 	int bits_mapped = PAGE_SHIFT - 4;
248 	unsigned long level_mask, prev_level_entry, *levels[4];
249 	int this_level, index, level_lsb, level_msb;
250 
251 	dst_addr &= PAGE_MASK;
252 	prev_level_entry = pte_val(pfn_pte(pfn, PAGE_KERNEL_ROX));
253 
254 	for (this_level = 3; this_level >= 0; this_level--) {
255 		levels[this_level] = trans_alloc(info);
256 		if (!levels[this_level])
257 			return -ENOMEM;
258 
259 		level_lsb = ARM64_HW_PGTABLE_LEVEL_SHIFT(this_level);
260 		level_msb = min(level_lsb + bits_mapped, max_msb);
261 		level_mask = GENMASK_ULL(level_msb, level_lsb);
262 
263 		index = (dst_addr & level_mask) >> level_lsb;
264 		*(levels[this_level] + index) = prev_level_entry;
265 
266 		pfn = virt_to_pfn(levels[this_level]);
267 		prev_level_entry = pte_val(pfn_pte(pfn,
268 						   __pgprot(PMD_TYPE_TABLE)));
269 
270 		if (level_msb == max_msb)
271 			break;
272 	}
273 
274 	*trans_ttbr0 = phys_to_ttbr(__pfn_to_phys(pfn));
275 	*t0sz = TCR_T0SZ(max_msb + 1);
276 
277 	return 0;
278 }
279 
280 /*
281  * Create a copy of the vector table so we can call HVC_SET_VECTORS or
282  * HVC_SOFT_RESTART from contexts where the table may be overwritten.
283  */
trans_pgd_copy_el2_vectors(struct trans_pgd_info * info,phys_addr_t * el2_vectors)284 int trans_pgd_copy_el2_vectors(struct trans_pgd_info *info,
285 			       phys_addr_t *el2_vectors)
286 {
287 	void *hyp_stub = trans_alloc(info);
288 
289 	if (!hyp_stub)
290 		return -ENOMEM;
291 	*el2_vectors = virt_to_phys(hyp_stub);
292 	memcpy(hyp_stub, &trans_pgd_stub_vectors, ARM64_VECTOR_TABLE_LEN);
293 	caches_clean_inval_pou((unsigned long)hyp_stub,
294 			       (unsigned long)hyp_stub +
295 			       ARM64_VECTOR_TABLE_LEN);
296 	dcache_clean_inval_poc((unsigned long)hyp_stub,
297 			       (unsigned long)hyp_stub +
298 			       ARM64_VECTOR_TABLE_LEN);
299 
300 	return 0;
301 }
302