1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Hibernation support for x86
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#include <linux/gfp.h>
10#include <linux/smp.h>
11#include <linux/suspend.h>
12#include <linux/scatterlist.h>
13#include <linux/kdebug.h>
14#include <linux/cpu.h>
15#include <linux/pgtable.h>
16#include <linux/types.h>
17#include <linux/crc32.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
28/*
29 * Address to jump to in the last phase of restore in order to get to the image
30 * kernel's text (this value is passed in the image header).
31 */
32unsigned long restore_jump_address __visible;
33unsigned long jump_address_phys;
34
35/*
36 * Value of the cr3 register from before the hibernation (this value is passed
37 * in the image header).
38 */
39unsigned long restore_cr3 __visible;
40unsigned long temp_pgt __visible;
41unsigned long relocated_restore_code __visible;
42
43/**
44 * pfn_is_nosave - check if given pfn is in the 'nosave' section
45 * @pfn: the page frame number to check.
46 */
47int pfn_is_nosave(unsigned long pfn)
48{
49 unsigned long nosave_begin_pfn;
50 unsigned long nosave_end_pfn;
51
52 nosave_begin_pfn = __pa_symbol(&__nosave_begin) >> PAGE_SHIFT;
53 nosave_end_pfn = PAGE_ALIGN(__pa_symbol(&__nosave_end)) >> PAGE_SHIFT;
54
55 return pfn >= nosave_begin_pfn && pfn < nosave_end_pfn;
56}
57
58struct restore_data_record {
59 unsigned long jump_address;
60 unsigned long jump_address_phys;
61 unsigned long cr3;
62 unsigned long magic;
63 unsigned long e820_checksum;
64};
65
66/**
67 * compute_e820_crc32 - calculate crc32 of a given e820 table
68 *
69 * @table: the e820 table to be calculated
70 *
71 * Return: the resulting checksum
72 */
73static inline u32 compute_e820_crc32(struct e820_table *table)
74{
75 int size = offsetof(struct e820_table, entries) +
76 sizeof(struct e820_entry) * table->nr_entries;
77
78 return ~crc32_le(crc: ~0, p: (unsigned char const *)table, len: size);
79}
80
81#ifdef CONFIG_X86_64
82#define RESTORE_MAGIC 0x23456789ABCDEF02UL
83#else
84#define RESTORE_MAGIC 0x12345679UL
85#endif
86
87/**
88 * arch_hibernation_header_save - populate the architecture specific part
89 * of a hibernation image header
90 * @addr: address where architecture specific header data will be saved.
91 * @max_size: maximum size of architecture specific data in hibernation header.
92 *
93 * Return: 0 on success, -EOVERFLOW if max_size is insufficient.
94 */
95int arch_hibernation_header_save(void *addr, unsigned int max_size)
96{
97 struct restore_data_record *rdr = addr;
98
99 if (max_size < sizeof(struct restore_data_record))
100 return -EOVERFLOW;
101 rdr->magic = RESTORE_MAGIC;
102 rdr->jump_address = (unsigned long)restore_registers;
103 rdr->jump_address_phys = __pa_symbol(restore_registers);
104
105 /*
106 * The restore code fixes up CR3 and CR4 in the following sequence:
107 *
108 * [in hibernation asm]
109 * 1. CR3 <= temporary page tables
110 * 2. CR4 <= mmu_cr4_features (from the kernel that restores us)
111 * 3. CR3 <= rdr->cr3
112 * 4. CR4 <= mmu_cr4_features (from us, i.e. the image kernel)
113 * [in restore_processor_state()]
114 * 5. CR4 <= saved CR4
115 * 6. CR3 <= saved CR3
116 *
117 * Our mmu_cr4_features has CR4.PCIDE=0, and toggling
118 * CR4.PCIDE while CR3's PCID bits are nonzero is illegal, so
119 * rdr->cr3 needs to point to valid page tables but must not
120 * have any of the PCID bits set.
121 */
122 rdr->cr3 = restore_cr3 & ~CR3_PCID_MASK;
123
124 rdr->e820_checksum = compute_e820_crc32(table: e820_table_firmware);
125 return 0;
126}
127
128/**
129 * arch_hibernation_header_restore - read the architecture specific data
130 * from the hibernation image header
131 * @addr: address to read the data from
132 */
133int arch_hibernation_header_restore(void *addr)
134{
135 struct restore_data_record *rdr = addr;
136
137 if (rdr->magic != RESTORE_MAGIC) {
138 pr_crit("Unrecognized hibernate image header format!\n");
139 return -EINVAL;
140 }
141
142 restore_jump_address = rdr->jump_address;
143 jump_address_phys = rdr->jump_address_phys;
144 restore_cr3 = rdr->cr3;
145
146 if (rdr->e820_checksum != compute_e820_crc32(table: e820_table_firmware)) {
147 pr_crit("Hibernate inconsistent memory map detected!\n");
148 return -ENODEV;
149 }
150
151 return 0;
152}
153
154int relocate_restore_code(void)
155{
156 pgd_t *pgd;
157 p4d_t *p4d;
158 pud_t *pud;
159 pmd_t *pmd;
160 pte_t *pte;
161
162 relocated_restore_code = get_safe_page(GFP_ATOMIC);
163 if (!relocated_restore_code)
164 return -ENOMEM;
165
166 __memcpy(to: (void *)relocated_restore_code, from: core_restore_code, PAGE_SIZE);
167
168 /* Make the page containing the relocated code executable */
169 pgd = (pgd_t *)__va(read_cr3_pa()) +
170 pgd_index(relocated_restore_code);
171 p4d = p4d_offset(pgd, address: relocated_restore_code);
172 if (p4d_leaf(*p4d)) {
173 set_p4d(p4d, __p4d(p4d_val(*p4d) & ~_PAGE_NX));
174 goto out;
175 }
176 pud = pud_offset(p4d, address: relocated_restore_code);
177 if (pud_leaf(pud: *pud)) {
178 set_pud(pud, __pud(pud_val(*pud) & ~_PAGE_NX));
179 goto out;
180 }
181 pmd = pmd_offset(pud, address: relocated_restore_code);
182 if (pmd_leaf(pte: *pmd)) {
183 set_pmd(pmd, __pmd(pmd_val(*pmd) & ~_PAGE_NX));
184 goto out;
185 }
186 pte = pte_offset_kernel(pmd, address: relocated_restore_code);
187 set_pte(pte, __pte(pte_val(*pte) & ~_PAGE_NX));
188out:
189 __flush_tlb_all();
190 return 0;
191}
192
193int arch_resume_nosmt(void)
194{
195 int ret;
196
197 /*
198 * We reached this while coming out of hibernation. This means
199 * that SMT siblings are sleeping in hlt, as mwait is not safe
200 * against control transition during resume (see comment in
201 * hibernate_resume_nonboot_cpu_disable()).
202 *
203 * If the resumed kernel has SMT disabled, we have to take all the
204 * SMT siblings out of hlt, and offline them again so that they
205 * end up in mwait proper.
206 *
207 * Called with hotplug disabled.
208 */
209 cpu_hotplug_enable();
210
211 ret = arch_cpu_rescan_dead_smt_siblings();
212
213 cpu_hotplug_disable();
214
215 return ret;
216}
217