xref: /openbmc/linux/arch/x86/power/hibernate_64.c (revision 58fc7f73)
1 /*
2  * Hibernation support for x86-64
3  *
4  * Distribute under GPLv2
5  *
6  * Copyright (c) 2007 Rafael J. Wysocki <rjw@sisk.pl>
7  * Copyright (c) 2002 Pavel Machek <pavel@ucw.cz>
8  * Copyright (c) 2001 Patrick Mochel <mochel@osdl.org>
9  */
10 
11 #include <linux/gfp.h>
12 #include <linux/smp.h>
13 #include <linux/suspend.h>
14 #include <linux/scatterlist.h>
15 #include <linux/kdebug.h>
16 
17 #include <crypto/hash.h>
18 
19 #include <asm/e820/api.h>
20 #include <asm/init.h>
21 #include <asm/proto.h>
22 #include <asm/page.h>
23 #include <asm/pgtable.h>
24 #include <asm/mtrr.h>
25 #include <asm/sections.h>
26 #include <asm/suspend.h>
27 #include <asm/tlbflush.h>
28 
29 /* Defined in hibernate_asm_64.S */
30 extern asmlinkage __visible int restore_image(void);
31 
32 /*
33  * Address to jump to in the last phase of restore in order to get to the image
34  * kernel's text (this value is passed in the image header).
35  */
36 unsigned long restore_jump_address __visible;
37 unsigned long jump_address_phys;
38 
39 /*
40  * Value of the cr3 register from before the hibernation (this value is passed
41  * in the image header).
42  */
43 unsigned long restore_cr3 __visible;
44 
45 unsigned long temp_level4_pgt __visible;
46 
47 unsigned long relocated_restore_code __visible;
48 
49 static int set_up_temporary_text_mapping(pgd_t *pgd)
50 {
51 	pmd_t *pmd;
52 	pud_t *pud;
53 	p4d_t *p4d;
54 
55 	/*
56 	 * The new mapping only has to cover the page containing the image
57 	 * kernel's entry point (jump_address_phys), because the switch over to
58 	 * it is carried out by relocated code running from a page allocated
59 	 * specifically for this purpose and covered by the identity mapping, so
60 	 * the temporary kernel text mapping is only needed for the final jump.
61 	 * Moreover, in that mapping the virtual address of the image kernel's
62 	 * entry point must be the same as its virtual address in the image
63 	 * kernel (restore_jump_address), so the image kernel's
64 	 * restore_registers() code doesn't find itself in a different area of
65 	 * the virtual address space after switching over to the original page
66 	 * tables used by the image kernel.
67 	 */
68 
69 	if (IS_ENABLED(CONFIG_X86_5LEVEL)) {
70 		p4d = (p4d_t *)get_safe_page(GFP_ATOMIC);
71 		if (!p4d)
72 			return -ENOMEM;
73 	}
74 
75 	pud = (pud_t *)get_safe_page(GFP_ATOMIC);
76 	if (!pud)
77 		return -ENOMEM;
78 
79 	pmd = (pmd_t *)get_safe_page(GFP_ATOMIC);
80 	if (!pmd)
81 		return -ENOMEM;
82 
83 	set_pmd(pmd + pmd_index(restore_jump_address),
84 		__pmd((jump_address_phys & PMD_MASK) | __PAGE_KERNEL_LARGE_EXEC));
85 	set_pud(pud + pud_index(restore_jump_address),
86 		__pud(__pa(pmd) | _KERNPG_TABLE));
87 	if (IS_ENABLED(CONFIG_X86_5LEVEL)) {
88 		set_p4d(p4d + p4d_index(restore_jump_address), __p4d(__pa(pud) | _KERNPG_TABLE));
89 		set_pgd(pgd + pgd_index(restore_jump_address), __pgd(__pa(p4d) | _KERNPG_TABLE));
90 	} else {
91 		/* No p4d for 4-level paging: point the pgd to the pud page table */
92 		set_pgd(pgd + pgd_index(restore_jump_address), __pgd(__pa(pud) | _KERNPG_TABLE));
93 	}
94 
95 	return 0;
96 }
97 
98 static void *alloc_pgt_page(void *context)
99 {
100 	return (void *)get_safe_page(GFP_ATOMIC);
101 }
102 
103 static int set_up_temporary_mappings(void)
104 {
105 	struct x86_mapping_info info = {
106 		.alloc_pgt_page	= alloc_pgt_page,
107 		.page_flag	= __PAGE_KERNEL_LARGE_EXEC,
108 		.offset		= __PAGE_OFFSET,
109 	};
110 	unsigned long mstart, mend;
111 	pgd_t *pgd;
112 	int result;
113 	int i;
114 
115 	pgd = (pgd_t *)get_safe_page(GFP_ATOMIC);
116 	if (!pgd)
117 		return -ENOMEM;
118 
119 	/* Prepare a temporary mapping for the kernel text */
120 	result = set_up_temporary_text_mapping(pgd);
121 	if (result)
122 		return result;
123 
124 	/* Set up the direct mapping from scratch */
125 	for (i = 0; i < nr_pfn_mapped; i++) {
126 		mstart = pfn_mapped[i].start << PAGE_SHIFT;
127 		mend   = pfn_mapped[i].end << PAGE_SHIFT;
128 
129 		result = kernel_ident_mapping_init(&info, pgd, mstart, mend);
130 		if (result)
131 			return result;
132 	}
133 
134 	temp_level4_pgt = __pa(pgd);
135 	return 0;
136 }
137 
138 static int relocate_restore_code(void)
139 {
140 	pgd_t *pgd;
141 	p4d_t *p4d;
142 	pud_t *pud;
143 	pmd_t *pmd;
144 	pte_t *pte;
145 
146 	relocated_restore_code = get_safe_page(GFP_ATOMIC);
147 	if (!relocated_restore_code)
148 		return -ENOMEM;
149 
150 	memcpy((void *)relocated_restore_code, core_restore_code, PAGE_SIZE);
151 
152 	/* Make the page containing the relocated code executable */
153 	pgd = (pgd_t *)__va(read_cr3_pa()) +
154 		pgd_index(relocated_restore_code);
155 	p4d = p4d_offset(pgd, relocated_restore_code);
156 	if (p4d_large(*p4d)) {
157 		set_p4d(p4d, __p4d(p4d_val(*p4d) & ~_PAGE_NX));
158 		goto out;
159 	}
160 	pud = pud_offset(p4d, relocated_restore_code);
161 	if (pud_large(*pud)) {
162 		set_pud(pud, __pud(pud_val(*pud) & ~_PAGE_NX));
163 		goto out;
164 	}
165 	pmd = pmd_offset(pud, relocated_restore_code);
166 	if (pmd_large(*pmd)) {
167 		set_pmd(pmd, __pmd(pmd_val(*pmd) & ~_PAGE_NX));
168 		goto out;
169 	}
170 	pte = pte_offset_kernel(pmd, relocated_restore_code);
171 	set_pte(pte, __pte(pte_val(*pte) & ~_PAGE_NX));
172 out:
173 	__flush_tlb_all();
174 	return 0;
175 }
176 
177 int swsusp_arch_resume(void)
178 {
179 	int error;
180 
181 	/* We have got enough memory and from now on we cannot recover */
182 	error = set_up_temporary_mappings();
183 	if (error)
184 		return error;
185 
186 	error = relocate_restore_code();
187 	if (error)
188 		return error;
189 
190 	restore_image();
191 	return 0;
192 }
193 
194 /*
195  *	pfn_is_nosave - check if given pfn is in the 'nosave' section
196  */
197 
198 int pfn_is_nosave(unsigned long pfn)
199 {
200 	unsigned long nosave_begin_pfn = __pa_symbol(&__nosave_begin) >> PAGE_SHIFT;
201 	unsigned long nosave_end_pfn = PAGE_ALIGN(__pa_symbol(&__nosave_end)) >> PAGE_SHIFT;
202 	return (pfn >= nosave_begin_pfn) && (pfn < nosave_end_pfn);
203 }
204 
205 #define MD5_DIGEST_SIZE 16
206 
207 struct restore_data_record {
208 	unsigned long jump_address;
209 	unsigned long jump_address_phys;
210 	unsigned long cr3;
211 	unsigned long magic;
212 	u8 e820_digest[MD5_DIGEST_SIZE];
213 };
214 
215 #define RESTORE_MAGIC	0x23456789ABCDEF01UL
216 
217 #if IS_BUILTIN(CONFIG_CRYPTO_MD5)
218 /**
219  * get_e820_md5 - calculate md5 according to given e820 table
220  *
221  * @table: the e820 table to be calculated
222  * @buf: the md5 result to be stored to
223  */
224 static int get_e820_md5(struct e820_table *table, void *buf)
225 {
226 	struct scatterlist sg;
227 	struct crypto_ahash *tfm;
228 	int size;
229 	int ret = 0;
230 
231 	tfm = crypto_alloc_ahash("md5", 0, CRYPTO_ALG_ASYNC);
232 	if (IS_ERR(tfm))
233 		return -ENOMEM;
234 
235 	{
236 		AHASH_REQUEST_ON_STACK(req, tfm);
237 		size = offsetof(struct e820_table, entries) + sizeof(struct e820_entry) * table->nr_entries;
238 		ahash_request_set_tfm(req, tfm);
239 		sg_init_one(&sg, (u8 *)table, size);
240 		ahash_request_set_callback(req, 0, NULL, NULL);
241 		ahash_request_set_crypt(req, &sg, buf, size);
242 
243 		if (crypto_ahash_digest(req))
244 			ret = -EINVAL;
245 		ahash_request_zero(req);
246 	}
247 	crypto_free_ahash(tfm);
248 
249 	return ret;
250 }
251 
252 static void hibernation_e820_save(void *buf)
253 {
254 	get_e820_md5(e820_table_firmware, buf);
255 }
256 
257 static bool hibernation_e820_mismatch(void *buf)
258 {
259 	int ret;
260 	u8 result[MD5_DIGEST_SIZE];
261 
262 	memset(result, 0, MD5_DIGEST_SIZE);
263 	/* If there is no digest in suspend kernel, let it go. */
264 	if (!memcmp(result, buf, MD5_DIGEST_SIZE))
265 		return false;
266 
267 	ret = get_e820_md5(e820_table_firmware, result);
268 	if (ret)
269 		return true;
270 
271 	return memcmp(result, buf, MD5_DIGEST_SIZE) ? true : false;
272 }
273 #else
274 static void hibernation_e820_save(void *buf)
275 {
276 }
277 
278 static bool hibernation_e820_mismatch(void *buf)
279 {
280 	/* If md5 is not builtin for restore kernel, let it go. */
281 	return false;
282 }
283 #endif
284 
285 /**
286  *	arch_hibernation_header_save - populate the architecture specific part
287  *		of a hibernation image header
288  *	@addr: address to save the data at
289  */
290 int arch_hibernation_header_save(void *addr, unsigned int max_size)
291 {
292 	struct restore_data_record *rdr = addr;
293 
294 	if (max_size < sizeof(struct restore_data_record))
295 		return -EOVERFLOW;
296 	rdr->jump_address = (unsigned long)restore_registers;
297 	rdr->jump_address_phys = __pa_symbol(restore_registers);
298 
299 	/*
300 	 * The restore code fixes up CR3 and CR4 in the following sequence:
301 	 *
302 	 * [in hibernation asm]
303 	 * 1. CR3 <= temporary page tables
304 	 * 2. CR4 <= mmu_cr4_features (from the kernel that restores us)
305 	 * 3. CR3 <= rdr->cr3
306 	 * 4. CR4 <= mmu_cr4_features (from us, i.e. the image kernel)
307 	 * [in restore_processor_state()]
308 	 * 5. CR4 <= saved CR4
309 	 * 6. CR3 <= saved CR3
310 	 *
311 	 * Our mmu_cr4_features has CR4.PCIDE=0, and toggling
312 	 * CR4.PCIDE while CR3's PCID bits are nonzero is illegal, so
313 	 * rdr->cr3 needs to point to valid page tables but must not
314 	 * have any of the PCID bits set.
315 	 */
316 	rdr->cr3 = restore_cr3 & ~CR3_PCID_MASK;
317 
318 	rdr->magic = RESTORE_MAGIC;
319 
320 	hibernation_e820_save(rdr->e820_digest);
321 
322 	return 0;
323 }
324 
325 /**
326  *	arch_hibernation_header_restore - read the architecture specific data
327  *		from the hibernation image header
328  *	@addr: address to read the data from
329  */
330 int arch_hibernation_header_restore(void *addr)
331 {
332 	struct restore_data_record *rdr = addr;
333 
334 	restore_jump_address = rdr->jump_address;
335 	jump_address_phys = rdr->jump_address_phys;
336 	restore_cr3 = rdr->cr3;
337 
338 	if (rdr->magic != RESTORE_MAGIC) {
339 		pr_crit("Unrecognized hibernate image header format!\n");
340 		return -EINVAL;
341 	}
342 
343 	if (hibernation_e820_mismatch(rdr->e820_digest)) {
344 		pr_crit("Hibernate inconsistent memory map detected!\n");
345 		return -ENODEV;
346 	}
347 
348 	return 0;
349 }
350