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 16 #include <crypto/hash.h> 17 18 #include <asm/e820/api.h> 19 #include <asm/init.h> 20 #include <asm/proto.h> 21 #include <asm/page.h> 22 #include <asm/pgtable.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 */ 32 unsigned long restore_jump_address __visible; 33 unsigned 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 */ 39 unsigned long restore_cr3 __visible; 40 unsigned long temp_pgt __visible; 41 unsigned long relocated_restore_code __visible; 42 43 /** 44 * pfn_is_nosave - check if given pfn is in the 'nosave' section 45 */ 46 int pfn_is_nosave(unsigned long pfn) 47 { 48 unsigned long nosave_begin_pfn; 49 unsigned long nosave_end_pfn; 50 51 nosave_begin_pfn = __pa_symbol(&__nosave_begin) >> PAGE_SHIFT; 52 nosave_end_pfn = PAGE_ALIGN(__pa_symbol(&__nosave_end)) >> PAGE_SHIFT; 53 54 return pfn >= nosave_begin_pfn && pfn < nosave_end_pfn; 55 } 56 57 58 #define MD5_DIGEST_SIZE 16 59 60 struct restore_data_record { 61 unsigned long jump_address; 62 unsigned long jump_address_phys; 63 unsigned long cr3; 64 unsigned long magic; 65 u8 e820_digest[MD5_DIGEST_SIZE]; 66 }; 67 68 #if IS_BUILTIN(CONFIG_CRYPTO_MD5) 69 /** 70 * get_e820_md5 - calculate md5 according to given e820 table 71 * 72 * @table: the e820 table to be calculated 73 * @buf: the md5 result to be stored to 74 */ 75 static int get_e820_md5(struct e820_table *table, void *buf) 76 { 77 struct crypto_shash *tfm; 78 struct shash_desc *desc; 79 int size; 80 int ret = 0; 81 82 tfm = crypto_alloc_shash("md5", 0, 0); 83 if (IS_ERR(tfm)) 84 return -ENOMEM; 85 86 desc = kmalloc(sizeof(struct shash_desc) + crypto_shash_descsize(tfm), 87 GFP_KERNEL); 88 if (!desc) { 89 ret = -ENOMEM; 90 goto free_tfm; 91 } 92 93 desc->tfm = tfm; 94 95 size = offsetof(struct e820_table, entries) + 96 sizeof(struct e820_entry) * table->nr_entries; 97 98 if (crypto_shash_digest(desc, (u8 *)table, size, buf)) 99 ret = -EINVAL; 100 101 kzfree(desc); 102 103 free_tfm: 104 crypto_free_shash(tfm); 105 return ret; 106 } 107 108 static int hibernation_e820_save(void *buf) 109 { 110 return get_e820_md5(e820_table_firmware, buf); 111 } 112 113 static bool hibernation_e820_mismatch(void *buf) 114 { 115 int ret; 116 u8 result[MD5_DIGEST_SIZE]; 117 118 memset(result, 0, MD5_DIGEST_SIZE); 119 /* If there is no digest in suspend kernel, let it go. */ 120 if (!memcmp(result, buf, MD5_DIGEST_SIZE)) 121 return false; 122 123 ret = get_e820_md5(e820_table_firmware, result); 124 if (ret) 125 return true; 126 127 return memcmp(result, buf, MD5_DIGEST_SIZE) ? true : false; 128 } 129 #else 130 static int hibernation_e820_save(void *buf) 131 { 132 return 0; 133 } 134 135 static bool hibernation_e820_mismatch(void *buf) 136 { 137 /* If md5 is not builtin for restore kernel, let it go. */ 138 return false; 139 } 140 #endif 141 142 #ifdef CONFIG_X86_64 143 #define RESTORE_MAGIC 0x23456789ABCDEF01UL 144 #else 145 #define RESTORE_MAGIC 0x12345678UL 146 #endif 147 148 /** 149 * arch_hibernation_header_save - populate the architecture specific part 150 * of a hibernation image header 151 * @addr: address to save the data at 152 */ 153 int arch_hibernation_header_save(void *addr, unsigned int max_size) 154 { 155 struct restore_data_record *rdr = addr; 156 157 if (max_size < sizeof(struct restore_data_record)) 158 return -EOVERFLOW; 159 rdr->magic = RESTORE_MAGIC; 160 rdr->jump_address = (unsigned long)restore_registers; 161 rdr->jump_address_phys = __pa_symbol(restore_registers); 162 163 /* 164 * The restore code fixes up CR3 and CR4 in the following sequence: 165 * 166 * [in hibernation asm] 167 * 1. CR3 <= temporary page tables 168 * 2. CR4 <= mmu_cr4_features (from the kernel that restores us) 169 * 3. CR3 <= rdr->cr3 170 * 4. CR4 <= mmu_cr4_features (from us, i.e. the image kernel) 171 * [in restore_processor_state()] 172 * 5. CR4 <= saved CR4 173 * 6. CR3 <= saved CR3 174 * 175 * Our mmu_cr4_features has CR4.PCIDE=0, and toggling 176 * CR4.PCIDE while CR3's PCID bits are nonzero is illegal, so 177 * rdr->cr3 needs to point to valid page tables but must not 178 * have any of the PCID bits set. 179 */ 180 rdr->cr3 = restore_cr3 & ~CR3_PCID_MASK; 181 182 return hibernation_e820_save(rdr->e820_digest); 183 } 184 185 /** 186 * arch_hibernation_header_restore - read the architecture specific data 187 * from the hibernation image header 188 * @addr: address to read the data from 189 */ 190 int arch_hibernation_header_restore(void *addr) 191 { 192 struct restore_data_record *rdr = addr; 193 194 if (rdr->magic != RESTORE_MAGIC) { 195 pr_crit("Unrecognized hibernate image header format!\n"); 196 return -EINVAL; 197 } 198 199 restore_jump_address = rdr->jump_address; 200 jump_address_phys = rdr->jump_address_phys; 201 restore_cr3 = rdr->cr3; 202 203 if (hibernation_e820_mismatch(rdr->e820_digest)) { 204 pr_crit("Hibernate inconsistent memory map detected!\n"); 205 return -ENODEV; 206 } 207 208 return 0; 209 } 210 211 int relocate_restore_code(void) 212 { 213 pgd_t *pgd; 214 p4d_t *p4d; 215 pud_t *pud; 216 pmd_t *pmd; 217 pte_t *pte; 218 219 relocated_restore_code = get_safe_page(GFP_ATOMIC); 220 if (!relocated_restore_code) 221 return -ENOMEM; 222 223 memcpy((void *)relocated_restore_code, core_restore_code, PAGE_SIZE); 224 225 /* Make the page containing the relocated code executable */ 226 pgd = (pgd_t *)__va(read_cr3_pa()) + 227 pgd_index(relocated_restore_code); 228 p4d = p4d_offset(pgd, relocated_restore_code); 229 if (p4d_large(*p4d)) { 230 set_p4d(p4d, __p4d(p4d_val(*p4d) & ~_PAGE_NX)); 231 goto out; 232 } 233 pud = pud_offset(p4d, relocated_restore_code); 234 if (pud_large(*pud)) { 235 set_pud(pud, __pud(pud_val(*pud) & ~_PAGE_NX)); 236 goto out; 237 } 238 pmd = pmd_offset(pud, relocated_restore_code); 239 if (pmd_large(*pmd)) { 240 set_pmd(pmd, __pmd(pmd_val(*pmd) & ~_PAGE_NX)); 241 goto out; 242 } 243 pte = pte_offset_kernel(pmd, relocated_restore_code); 244 set_pte(pte, __pte(pte_val(*pte) & ~_PAGE_NX)); 245 out: 246 __flush_tlb_all(); 247 return 0; 248 } 249 250 int arch_resume_nosmt(void) 251 { 252 int ret = 0; 253 /* 254 * We reached this while coming out of hibernation. This means 255 * that SMT siblings are sleeping in hlt, as mwait is not safe 256 * against control transition during resume (see comment in 257 * hibernate_resume_nonboot_cpu_disable()). 258 * 259 * If the resumed kernel has SMT disabled, we have to take all the 260 * SMT siblings out of hlt, and offline them again so that they 261 * end up in mwait proper. 262 * 263 * Called with hotplug disabled. 264 */ 265 cpu_hotplug_enable(); 266 if (cpu_smt_control == CPU_SMT_DISABLED || 267 cpu_smt_control == CPU_SMT_FORCE_DISABLED) { 268 enum cpuhp_smt_control old = cpu_smt_control; 269 270 ret = cpuhp_smt_enable(); 271 if (ret) 272 goto out; 273 ret = cpuhp_smt_disable(old); 274 if (ret) 275 goto out; 276 } 277 out: 278 cpu_hotplug_disable(); 279 return ret; 280 } 281