Linux Audio

Check our new training course

Linux debugging, profiling, tracing and performance analysis training

Mar 24-27, 2025, special US time zones
Register
Loading...
v6.2
  1// SPDX-License-Identifier: GPL-2.0-only
  2/*
  3 * Hibernation support for x86-64
  4 *
 
 
  5 * Copyright (c) 2007 Rafael J. Wysocki <rjw@sisk.pl>
  6 * Copyright (c) 2002 Pavel Machek <pavel@ucw.cz>
  7 * Copyright (c) 2001 Patrick Mochel <mochel@osdl.org>
  8 */
  9
 10#include <linux/gfp.h>
 11#include <linux/smp.h>
 12#include <linux/suspend.h>
 13#include <linux/scatterlist.h>
 14#include <linux/kdebug.h>
 15#include <linux/pgtable.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/mtrr.h>
 24#include <asm/sections.h>
 25#include <asm/suspend.h>
 26#include <asm/tlbflush.h>
 27
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 28static int set_up_temporary_text_mapping(pgd_t *pgd)
 29{
 30	pmd_t *pmd;
 31	pud_t *pud;
 32	p4d_t *p4d = NULL;
 33	pgprot_t pgtable_prot = __pgprot(_KERNPG_TABLE);
 34	pgprot_t pmd_text_prot = __pgprot(__PAGE_KERNEL_LARGE_EXEC);
 35
 36	/* Filter out unsupported __PAGE_KERNEL* bits: */
 37	pgprot_val(pmd_text_prot) &= __default_kernel_pte_mask;
 38	pgprot_val(pgtable_prot)  &= __default_kernel_pte_mask;
 39
 40	/*
 41	 * The new mapping only has to cover the page containing the image
 42	 * kernel's entry point (jump_address_phys), because the switch over to
 43	 * it is carried out by relocated code running from a page allocated
 44	 * specifically for this purpose and covered by the identity mapping, so
 45	 * the temporary kernel text mapping is only needed for the final jump.
 46	 * Moreover, in that mapping the virtual address of the image kernel's
 47	 * entry point must be the same as its virtual address in the image
 48	 * kernel (restore_jump_address), so the image kernel's
 49	 * restore_registers() code doesn't find itself in a different area of
 50	 * the virtual address space after switching over to the original page
 51	 * tables used by the image kernel.
 52	 */
 53
 54	if (pgtable_l5_enabled()) {
 55		p4d = (p4d_t *)get_safe_page(GFP_ATOMIC);
 56		if (!p4d)
 57			return -ENOMEM;
 58	}
 59
 60	pud = (pud_t *)get_safe_page(GFP_ATOMIC);
 61	if (!pud)
 62		return -ENOMEM;
 63
 64	pmd = (pmd_t *)get_safe_page(GFP_ATOMIC);
 65	if (!pmd)
 66		return -ENOMEM;
 67
 68	set_pmd(pmd + pmd_index(restore_jump_address),
 69		__pmd((jump_address_phys & PMD_MASK) | pgprot_val(pmd_text_prot)));
 70	set_pud(pud + pud_index(restore_jump_address),
 71		__pud(__pa(pmd) | pgprot_val(pgtable_prot)));
 72	if (p4d) {
 73		p4d_t new_p4d = __p4d(__pa(pud) | pgprot_val(pgtable_prot));
 74		pgd_t new_pgd = __pgd(__pa(p4d) | pgprot_val(pgtable_prot));
 75
 76		set_p4d(p4d + p4d_index(restore_jump_address), new_p4d);
 77		set_pgd(pgd + pgd_index(restore_jump_address), new_pgd);
 78	} else {
 79		/* No p4d for 4-level paging: point the pgd to the pud page table */
 80		pgd_t new_pgd = __pgd(__pa(pud) | pgprot_val(pgtable_prot));
 81		set_pgd(pgd + pgd_index(restore_jump_address), new_pgd);
 82	}
 83
 84	return 0;
 85}
 86
 87static void *alloc_pgt_page(void *context)
 88{
 89	return (void *)get_safe_page(GFP_ATOMIC);
 90}
 91
 92static int set_up_temporary_mappings(void)
 93{
 94	struct x86_mapping_info info = {
 95		.alloc_pgt_page	= alloc_pgt_page,
 96		.page_flag	= __PAGE_KERNEL_LARGE_EXEC,
 97		.offset		= __PAGE_OFFSET,
 98	};
 99	unsigned long mstart, mend;
100	pgd_t *pgd;
101	int result;
102	int i;
103
104	pgd = (pgd_t *)get_safe_page(GFP_ATOMIC);
105	if (!pgd)
106		return -ENOMEM;
107
108	/* Prepare a temporary mapping for the kernel text */
109	result = set_up_temporary_text_mapping(pgd);
110	if (result)
111		return result;
112
113	/* Set up the direct mapping from scratch */
114	for (i = 0; i < nr_pfn_mapped; i++) {
115		mstart = pfn_mapped[i].start << PAGE_SHIFT;
116		mend   = pfn_mapped[i].end << PAGE_SHIFT;
117
118		result = kernel_ident_mapping_init(&info, pgd, mstart, mend);
119		if (result)
120			return result;
121	}
122
123	temp_pgt = __pa(pgd);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
124	return 0;
125}
126
127asmlinkage int swsusp_arch_resume(void)
128{
129	int error;
130
131	/* We have got enough memory and from now on we cannot recover */
132	error = set_up_temporary_mappings();
133	if (error)
134		return error;
135
136	error = relocate_restore_code();
137	if (error)
138		return error;
139
140	restore_image();
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
141	return 0;
142}
v4.17
 
  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 */
 30extern 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 */
 36unsigned long restore_jump_address __visible;
 37unsigned 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 */
 43unsigned long restore_cr3 __visible;
 44
 45unsigned long temp_level4_pgt __visible;
 46
 47unsigned long relocated_restore_code __visible;
 48
 49static int set_up_temporary_text_mapping(pgd_t *pgd)
 50{
 51	pmd_t *pmd;
 52	pud_t *pud;
 53	p4d_t *p4d = NULL;
 54	pgprot_t pgtable_prot = __pgprot(_KERNPG_TABLE);
 55	pgprot_t pmd_text_prot = __pgprot(__PAGE_KERNEL_LARGE_EXEC);
 56
 57	/* Filter out unsupported __PAGE_KERNEL* bits: */
 58	pgprot_val(pmd_text_prot) &= __default_kernel_pte_mask;
 59	pgprot_val(pgtable_prot)  &= __default_kernel_pte_mask;
 60
 61	/*
 62	 * The new mapping only has to cover the page containing the image
 63	 * kernel's entry point (jump_address_phys), because the switch over to
 64	 * it is carried out by relocated code running from a page allocated
 65	 * specifically for this purpose and covered by the identity mapping, so
 66	 * the temporary kernel text mapping is only needed for the final jump.
 67	 * Moreover, in that mapping the virtual address of the image kernel's
 68	 * entry point must be the same as its virtual address in the image
 69	 * kernel (restore_jump_address), so the image kernel's
 70	 * restore_registers() code doesn't find itself in a different area of
 71	 * the virtual address space after switching over to the original page
 72	 * tables used by the image kernel.
 73	 */
 74
 75	if (pgtable_l5_enabled) {
 76		p4d = (p4d_t *)get_safe_page(GFP_ATOMIC);
 77		if (!p4d)
 78			return -ENOMEM;
 79	}
 80
 81	pud = (pud_t *)get_safe_page(GFP_ATOMIC);
 82	if (!pud)
 83		return -ENOMEM;
 84
 85	pmd = (pmd_t *)get_safe_page(GFP_ATOMIC);
 86	if (!pmd)
 87		return -ENOMEM;
 88
 89	set_pmd(pmd + pmd_index(restore_jump_address),
 90		__pmd((jump_address_phys & PMD_MASK) | pgprot_val(pmd_text_prot)));
 91	set_pud(pud + pud_index(restore_jump_address),
 92		__pud(__pa(pmd) | pgprot_val(pgtable_prot)));
 93	if (p4d) {
 94		p4d_t new_p4d = __p4d(__pa(pud) | pgprot_val(pgtable_prot));
 95		pgd_t new_pgd = __pgd(__pa(p4d) | pgprot_val(pgtable_prot));
 96
 97		set_p4d(p4d + p4d_index(restore_jump_address), new_p4d);
 98		set_pgd(pgd + pgd_index(restore_jump_address), new_pgd);
 99	} else {
100		/* No p4d for 4-level paging: point the pgd to the pud page table */
101		pgd_t new_pgd = __pgd(__pa(pud) | pgprot_val(pgtable_prot));
102		set_pgd(pgd + pgd_index(restore_jump_address), new_pgd);
103	}
104
105	return 0;
106}
107
108static void *alloc_pgt_page(void *context)
109{
110	return (void *)get_safe_page(GFP_ATOMIC);
111}
112
113static int set_up_temporary_mappings(void)
114{
115	struct x86_mapping_info info = {
116		.alloc_pgt_page	= alloc_pgt_page,
117		.page_flag	= __PAGE_KERNEL_LARGE_EXEC,
118		.offset		= __PAGE_OFFSET,
119	};
120	unsigned long mstart, mend;
121	pgd_t *pgd;
122	int result;
123	int i;
124
125	pgd = (pgd_t *)get_safe_page(GFP_ATOMIC);
126	if (!pgd)
127		return -ENOMEM;
128
129	/* Prepare a temporary mapping for the kernel text */
130	result = set_up_temporary_text_mapping(pgd);
131	if (result)
132		return result;
133
134	/* Set up the direct mapping from scratch */
135	for (i = 0; i < nr_pfn_mapped; i++) {
136		mstart = pfn_mapped[i].start << PAGE_SHIFT;
137		mend   = pfn_mapped[i].end << PAGE_SHIFT;
138
139		result = kernel_ident_mapping_init(&info, pgd, mstart, mend);
140		if (result)
141			return result;
142	}
143
144	temp_level4_pgt = __pa(pgd);
145	return 0;
146}
147
148static int relocate_restore_code(void)
149{
150	pgd_t *pgd;
151	p4d_t *p4d;
152	pud_t *pud;
153	pmd_t *pmd;
154	pte_t *pte;
155
156	relocated_restore_code = get_safe_page(GFP_ATOMIC);
157	if (!relocated_restore_code)
158		return -ENOMEM;
159
160	memcpy((void *)relocated_restore_code, core_restore_code, PAGE_SIZE);
161
162	/* Make the page containing the relocated code executable */
163	pgd = (pgd_t *)__va(read_cr3_pa()) +
164		pgd_index(relocated_restore_code);
165	p4d = p4d_offset(pgd, relocated_restore_code);
166	if (p4d_large(*p4d)) {
167		set_p4d(p4d, __p4d(p4d_val(*p4d) & ~_PAGE_NX));
168		goto out;
169	}
170	pud = pud_offset(p4d, relocated_restore_code);
171	if (pud_large(*pud)) {
172		set_pud(pud, __pud(pud_val(*pud) & ~_PAGE_NX));
173		goto out;
174	}
175	pmd = pmd_offset(pud, relocated_restore_code);
176	if (pmd_large(*pmd)) {
177		set_pmd(pmd, __pmd(pmd_val(*pmd) & ~_PAGE_NX));
178		goto out;
179	}
180	pte = pte_offset_kernel(pmd, relocated_restore_code);
181	set_pte(pte, __pte(pte_val(*pte) & ~_PAGE_NX));
182out:
183	__flush_tlb_all();
184	return 0;
185}
186
187asmlinkage int swsusp_arch_resume(void)
188{
189	int error;
190
191	/* We have got enough memory and from now on we cannot recover */
192	error = set_up_temporary_mappings();
193	if (error)
194		return error;
195
196	error = relocate_restore_code();
197	if (error)
198		return error;
199
200	restore_image();
201	return 0;
202}
203
204/*
205 *	pfn_is_nosave - check if given pfn is in the 'nosave' section
206 */
207
208int pfn_is_nosave(unsigned long pfn)
209{
210	unsigned long nosave_begin_pfn = __pa_symbol(&__nosave_begin) >> PAGE_SHIFT;
211	unsigned long nosave_end_pfn = PAGE_ALIGN(__pa_symbol(&__nosave_end)) >> PAGE_SHIFT;
212	return (pfn >= nosave_begin_pfn) && (pfn < nosave_end_pfn);
213}
214
215#define MD5_DIGEST_SIZE 16
216
217struct restore_data_record {
218	unsigned long jump_address;
219	unsigned long jump_address_phys;
220	unsigned long cr3;
221	unsigned long magic;
222	u8 e820_digest[MD5_DIGEST_SIZE];
223};
224
225#define RESTORE_MAGIC	0x23456789ABCDEF01UL
226
227#if IS_BUILTIN(CONFIG_CRYPTO_MD5)
228/**
229 * get_e820_md5 - calculate md5 according to given e820 table
230 *
231 * @table: the e820 table to be calculated
232 * @buf: the md5 result to be stored to
233 */
234static int get_e820_md5(struct e820_table *table, void *buf)
235{
236	struct scatterlist sg;
237	struct crypto_ahash *tfm;
238	int size;
239	int ret = 0;
240
241	tfm = crypto_alloc_ahash("md5", 0, CRYPTO_ALG_ASYNC);
242	if (IS_ERR(tfm))
243		return -ENOMEM;
244
245	{
246		AHASH_REQUEST_ON_STACK(req, tfm);
247		size = offsetof(struct e820_table, entries) + sizeof(struct e820_entry) * table->nr_entries;
248		ahash_request_set_tfm(req, tfm);
249		sg_init_one(&sg, (u8 *)table, size);
250		ahash_request_set_callback(req, 0, NULL, NULL);
251		ahash_request_set_crypt(req, &sg, buf, size);
252
253		if (crypto_ahash_digest(req))
254			ret = -EINVAL;
255		ahash_request_zero(req);
256	}
257	crypto_free_ahash(tfm);
258
259	return ret;
260}
261
262static void hibernation_e820_save(void *buf)
263{
264	get_e820_md5(e820_table_firmware, buf);
265}
266
267static bool hibernation_e820_mismatch(void *buf)
268{
269	int ret;
270	u8 result[MD5_DIGEST_SIZE];
271
272	memset(result, 0, MD5_DIGEST_SIZE);
273	/* If there is no digest in suspend kernel, let it go. */
274	if (!memcmp(result, buf, MD5_DIGEST_SIZE))
275		return false;
276
277	ret = get_e820_md5(e820_table_firmware, result);
278	if (ret)
279		return true;
280
281	return memcmp(result, buf, MD5_DIGEST_SIZE) ? true : false;
282}
283#else
284static void hibernation_e820_save(void *buf)
285{
286}
287
288static bool hibernation_e820_mismatch(void *buf)
289{
290	/* If md5 is not builtin for restore kernel, let it go. */
291	return false;
292}
293#endif
294
295/**
296 *	arch_hibernation_header_save - populate the architecture specific part
297 *		of a hibernation image header
298 *	@addr: address to save the data at
299 */
300int arch_hibernation_header_save(void *addr, unsigned int max_size)
301{
302	struct restore_data_record *rdr = addr;
303
304	if (max_size < sizeof(struct restore_data_record))
305		return -EOVERFLOW;
306	rdr->jump_address = (unsigned long)restore_registers;
307	rdr->jump_address_phys = __pa_symbol(restore_registers);
308
309	/*
310	 * The restore code fixes up CR3 and CR4 in the following sequence:
311	 *
312	 * [in hibernation asm]
313	 * 1. CR3 <= temporary page tables
314	 * 2. CR4 <= mmu_cr4_features (from the kernel that restores us)
315	 * 3. CR3 <= rdr->cr3
316	 * 4. CR4 <= mmu_cr4_features (from us, i.e. the image kernel)
317	 * [in restore_processor_state()]
318	 * 5. CR4 <= saved CR4
319	 * 6. CR3 <= saved CR3
320	 *
321	 * Our mmu_cr4_features has CR4.PCIDE=0, and toggling
322	 * CR4.PCIDE while CR3's PCID bits are nonzero is illegal, so
323	 * rdr->cr3 needs to point to valid page tables but must not
324	 * have any of the PCID bits set.
325	 */
326	rdr->cr3 = restore_cr3 & ~CR3_PCID_MASK;
327
328	rdr->magic = RESTORE_MAGIC;
329
330	hibernation_e820_save(rdr->e820_digest);
331
332	return 0;
333}
334
335/**
336 *	arch_hibernation_header_restore - read the architecture specific data
337 *		from the hibernation image header
338 *	@addr: address to read the data from
339 */
340int arch_hibernation_header_restore(void *addr)
341{
342	struct restore_data_record *rdr = addr;
343
344	restore_jump_address = rdr->jump_address;
345	jump_address_phys = rdr->jump_address_phys;
346	restore_cr3 = rdr->cr3;
347
348	if (rdr->magic != RESTORE_MAGIC) {
349		pr_crit("Unrecognized hibernate image header format!\n");
350		return -EINVAL;
351	}
352
353	if (hibernation_e820_mismatch(rdr->e820_digest)) {
354		pr_crit("Hibernate inconsistent memory map detected!\n");
355		return -ENODEV;
356	}
357
358	return 0;
359}