arch/x86/power/hibernate_64.c

   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 #include <linux/cpu.h>
  17
  18 #include <crypto/hash.h>
  19
  20 #include <asm/e820/api.h>
  21 #include <asm/init.h>
  22 #include <asm/proto.h>
  23 #include <asm/page.h>
  24 #include <asm/pgtable.h>
  25 #include <asm/mtrr.h>
  26 #include <asm/sections.h>
  27 #include <asm/suspend.h>
  28 #include <asm/tlbflush.h>
  29
  30 /* Defined in hibernate_asm_64.S */
  31 extern asmlinkage __visible int restore_image(void);
  32
  33 /*
  34  * Address to jump to in the last phase of restore in order to get to the image
  35  * kernel's text (this value is passed in the image header).
  36  */
  37 unsigned long restore_jump_address __visible;
  38 unsigned long jump_address_phys;
  39
  40 /*
  41  * Value of the cr3 register from before the hibernation (this value is passed
  42  * in the image header).
  43  */
  44 unsigned long restore_cr3 __visible;
  45
  46 unsigned long temp_level4_pgt __visible;
  47
  48 unsigned long relocated_restore_code __visible;
  49
  50 static int set_up_temporary_text_mapping(pgd_t *pgd)
  51 {
  52         pmd_t *pmd;
  53         pud_t *pud;
  54         p4d_t *p4d = NULL;
  55         pgprot_t pgtable_prot = __pgprot(_KERNPG_TABLE);
  56         pgprot_t pmd_text_prot = __pgprot(__PAGE_KERNEL_LARGE_EXEC);
  57
  58         /* Filter out unsupported __PAGE_KERNEL* bits: */
  59         pgprot_val(pmd_text_prot) &= __default_kernel_pte_mask;
  60         pgprot_val(pgtable_prot)  &= __default_kernel_pte_mask;
  61
  62         /*
  63          * The new mapping only has to cover the page containing the image
  64          * kernel's entry point (jump_address_phys), because the switch over to
  65          * it is carried out by relocated code running from a page allocated
  66          * specifically for this purpose and covered by the identity mapping, so
  67          * the temporary kernel text mapping is only needed for the final jump.
  68          * Moreover, in that mapping the virtual address of the image kernel's
  69          * entry point must be the same as its virtual address in the image
  70          * kernel (restore_jump_address), so the image kernel's
  71          * restore_registers() code doesn't find itself in a different area of
  72          * the virtual address space after switching over to the original page
  73          * tables used by the image kernel.
  74          */
  75
  76         if (pgtable_l5_enabled()) {
  77                 p4d = (p4d_t *)get_safe_page(GFP_ATOMIC);
  78                 if (!p4d)
  79                         return -ENOMEM;
  80         }
  81
  82         pud = (pud_t *)get_safe_page(GFP_ATOMIC);
  83         if (!pud)
  84                 return -ENOMEM;
  85
  86         pmd = (pmd_t *)get_safe_page(GFP_ATOMIC);
  87         if (!pmd)
  88                 return -ENOMEM;
  89
  90         set_pmd(pmd + pmd_index(restore_jump_address),
  91                 __pmd((jump_address_phys & PMD_MASK) | pgprot_val(pmd_text_prot)));
  92         set_pud(pud + pud_index(restore_jump_address),
  93                 __pud(__pa(pmd) | pgprot_val(pgtable_prot)));
  94         if (p4d) {
  95                 p4d_t new_p4d = __p4d(__pa(pud) | pgprot_val(pgtable_prot));
  96                 pgd_t new_pgd = __pgd(__pa(p4d) | pgprot_val(pgtable_prot));
  97
  98                 set_p4d(p4d + p4d_index(restore_jump_address), new_p4d);
  99                 set_pgd(pgd + pgd_index(restore_jump_address), new_pgd);
 100         } else {
 101                 /* No p4d for 4-level paging: point the pgd to the pud page table */
 102                 pgd_t new_pgd = __pgd(__pa(pud) | pgprot_val(pgtable_prot));
 103                 set_pgd(pgd + pgd_index(restore_jump_address), new_pgd);
 104         }
 105
 106         return 0;
 107 }
 108
 109 static void *alloc_pgt_page(void *context)
 110 {
 111         return (void *)get_safe_page(GFP_ATOMIC);
 112 }
 113
 114 static int set_up_temporary_mappings(void)
 115 {
 116         struct x86_mapping_info info = {
 117                 .alloc_pgt_page = alloc_pgt_page,
 118                 .page_flag      = __PAGE_KERNEL_LARGE_EXEC,
 119                 .offset         = __PAGE_OFFSET,
 120         };
 121         unsigned long mstart, mend;
 122         pgd_t *pgd;
 123         int result;
 124         int i;
 125
 126         pgd = (pgd_t *)get_safe_page(GFP_ATOMIC);
 127         if (!pgd)
 128                 return -ENOMEM;
 129
 130         /* Prepare a temporary mapping for the kernel text */
 131         result = set_up_temporary_text_mapping(pgd);
 132         if (result)
 133                 return result;
 134
 135         /* Set up the direct mapping from scratch */
 136         for (i = 0; i < nr_pfn_mapped; i++) {
 137                 mstart = pfn_mapped[i].start << PAGE_SHIFT;
 138                 mend   = pfn_mapped[i].end << PAGE_SHIFT;
 139
 140                 result = kernel_ident_mapping_init(&info, pgd, mstart, mend);
 141                 if (result)
 142                         return result;
 143         }
 144
 145         temp_level4_pgt = __pa(pgd);
 146         return 0;
 147 }
 148
 149 static int relocate_restore_code(void)
 150 {
 151         pgd_t *pgd;
 152         p4d_t *p4d;
 153         pud_t *pud;
 154         pmd_t *pmd;
 155         pte_t *pte;
 156
 157         relocated_restore_code = get_safe_page(GFP_ATOMIC);
 158         if (!relocated_restore_code)
 159                 return -ENOMEM;
 160
 161         memcpy((void *)relocated_restore_code, core_restore_code, PAGE_SIZE);
 162
 163         /* Make the page containing the relocated code executable */
 164         pgd = (pgd_t *)__va(read_cr3_pa()) +
 165                 pgd_index(relocated_restore_code);
 166         p4d = p4d_offset(pgd, relocated_restore_code);
 167         if (p4d_large(*p4d)) {
 168                 set_p4d(p4d, __p4d(p4d_val(*p4d) & ~_PAGE_NX));
 169                 goto out;
 170         }
 171         pud = pud_offset(p4d, relocated_restore_code);
 172         if (pud_large(*pud)) {
 173                 set_pud(pud, __pud(pud_val(*pud) & ~_PAGE_NX));
 174                 goto out;
 175         }
 176         pmd = pmd_offset(pud, relocated_restore_code);
 177         if (pmd_large(*pmd)) {
 178                 set_pmd(pmd, __pmd(pmd_val(*pmd) & ~_PAGE_NX));
 179                 goto out;
 180         }
 181         pte = pte_offset_kernel(pmd, relocated_restore_code);
 182         set_pte(pte, __pte(pte_val(*pte) & ~_PAGE_NX));
 183 out:
 184         __flush_tlb_all();
 185         return 0;
 186 }
 187
 188 asmlinkage int swsusp_arch_resume(void)
 189 {
 190         int error;
 191
 192         /* We have got enough memory and from now on we cannot recover */
 193         error = set_up_temporary_mappings();
 194         if (error)
 195                 return error;
 196
 197         error = relocate_restore_code();
 198         if (error)
 199                 return error;
 200
 201         restore_image();
 202         return 0;
 203 }
 204
 205 /*
 206  *      pfn_is_nosave - check if given pfn is in the 'nosave' section
 207  */
 208
 209 int pfn_is_nosave(unsigned long pfn)
 210 {
 211         unsigned long nosave_begin_pfn = __pa_symbol(&__nosave_begin) >> PAGE_SHIFT;
 212         unsigned long nosave_end_pfn = PAGE_ALIGN(__pa_symbol(&__nosave_end)) >> PAGE_SHIFT;
 213         return (pfn >= nosave_begin_pfn) && (pfn < nosave_end_pfn);
 214 }
 215
 216 #define MD5_DIGEST_SIZE 16
 217
 218 struct restore_data_record {
 219         unsigned long jump_address;
 220         unsigned long jump_address_phys;
 221         unsigned long cr3;
 222         unsigned long magic;
 223         u8 e820_digest[MD5_DIGEST_SIZE];
 224 };
 225
 226 #define RESTORE_MAGIC   0x23456789ABCDEF01UL
 227
 228 #if IS_BUILTIN(CONFIG_CRYPTO_MD5)
 229 /**
 230  * get_e820_md5 - calculate md5 according to given e820 table
 231  *
 232  * @table: the e820 table to be calculated
 233  * @buf: the md5 result to be stored to
 234  */
 235 static int get_e820_md5(struct e820_table *table, void *buf)
 236 {
 237         struct crypto_shash *tfm;
 238         struct shash_desc *desc;
 239         int size;
 240         int ret = 0;
 241
 242         tfm = crypto_alloc_shash("md5", 0, 0);
 243         if (IS_ERR(tfm))
 244                 return -ENOMEM;
 245
 246         desc = kmalloc(sizeof(struct shash_desc) + crypto_shash_descsize(tfm),
 247                        GFP_KERNEL);
 248         if (!desc) {
 249                 ret = -ENOMEM;
 250                 goto free_tfm;
 251         }
 252
 253         desc->tfm = tfm;
 254         desc->flags = 0;
 255
 256         size = offsetof(struct e820_table, entries) +
 257                 sizeof(struct e820_entry) * table->nr_entries;
 258
 259         if (crypto_shash_digest(desc, (u8 *)table, size, buf))
 260                 ret = -EINVAL;
 261
 262         kzfree(desc);
 263
 264 free_tfm:
 265         crypto_free_shash(tfm);
 266         return ret;
 267 }
 268
 269 static void hibernation_e820_save(void *buf)
 270 {
 271         get_e820_md5(e820_table_firmware, buf);
 272 }
 273
 274 static bool hibernation_e820_mismatch(void *buf)
 275 {
 276         int ret;
 277         u8 result[MD5_DIGEST_SIZE];
 278
 279         memset(result, 0, MD5_DIGEST_SIZE);
 280         /* If there is no digest in suspend kernel, let it go. */
 281         if (!memcmp(result, buf, MD5_DIGEST_SIZE))
 282                 return false;
 283
 284         ret = get_e820_md5(e820_table_firmware, result);
 285         if (ret)
 286                 return true;
 287
 288         return memcmp(result, buf, MD5_DIGEST_SIZE) ? true : false;
 289 }
 290 #else
 291 static void hibernation_e820_save(void *buf)
 292 {
 293 }
 294
 295 static bool hibernation_e820_mismatch(void *buf)
 296 {
 297         /* If md5 is not builtin for restore kernel, let it go. */
 298         return false;
 299 }
 300 #endif
 301
 302 /**
 303  *      arch_hibernation_header_save - populate the architecture specific part
 304  *              of a hibernation image header
 305  *      @addr: address to save the data at
 306  */
 307 int arch_hibernation_header_save(void *addr, unsigned int max_size)
 308 {
 309         struct restore_data_record *rdr = addr;
 310
 311         if (max_size < sizeof(struct restore_data_record))
 312                 return -EOVERFLOW;
 313         rdr->jump_address = (unsigned long)restore_registers;
 314         rdr->jump_address_phys = __pa_symbol(restore_registers);
 315
 316         /*
 317          * The restore code fixes up CR3 and CR4 in the following sequence:
 318          *
 319          * [in hibernation asm]
 320          * 1. CR3 <= temporary page tables
 321          * 2. CR4 <= mmu_cr4_features (from the kernel that restores us)
 322          * 3. CR3 <= rdr->cr3
 323          * 4. CR4 <= mmu_cr4_features (from us, i.e. the image kernel)
 324          * [in restore_processor_state()]
 325          * 5. CR4 <= saved CR4
 326          * 6. CR3 <= saved CR3
 327          *
 328          * Our mmu_cr4_features has CR4.PCIDE=0, and toggling
 329          * CR4.PCIDE while CR3's PCID bits are nonzero is illegal, so
 330          * rdr->cr3 needs to point to valid page tables but must not
 331          * have any of the PCID bits set.
 332          */
 333         rdr->cr3 = restore_cr3 & ~CR3_PCID_MASK;
 334
 335         rdr->magic = RESTORE_MAGIC;
 336
 337         hibernation_e820_save(rdr->e820_digest);
 338
 339         return 0;
 340 }
 341
 342 /**
 343  *      arch_hibernation_header_restore - read the architecture specific data
 344  *              from the hibernation image header
 345  *      @addr: address to read the data from
 346  */
 347 int arch_hibernation_header_restore(void *addr)
 348 {
 349         struct restore_data_record *rdr = addr;
 350
 351         restore_jump_address = rdr->jump_address;
 352         jump_address_phys = rdr->jump_address_phys;
 353         restore_cr3 = rdr->cr3;
 354
 355         if (rdr->magic != RESTORE_MAGIC) {
 356                 pr_crit("Unrecognized hibernate image header format!\n");
 357                 return -EINVAL;
 358         }
 359
 360         if (hibernation_e820_mismatch(rdr->e820_digest)) {
 361                 pr_crit("Hibernate inconsistent memory map detected!\n");
 362                 return -ENODEV;
 363         }
 364
 365         return 0;
 366 }
 367
 368 int arch_resume_nosmt(void)
 369 {
 370         int ret = 0;
 371         /*
 372          * We reached this while coming out of hibernation. This means
 373          * that SMT siblings are sleeping in hlt, as mwait is not safe
 374          * against control transition during resume (see comment in
 375          * hibernate_resume_nonboot_cpu_disable()).
 376          *
 377          * If the resumed kernel has SMT disabled, we have to take all the
 378          * SMT siblings out of hlt, and offline them again so that they
 379          * end up in mwait proper.
 380          *
 381          * Called with hotplug disabled.
 382          */
 383         cpu_hotplug_enable();
 384         if (cpu_smt_control == CPU_SMT_DISABLED ||
 385                         cpu_smt_control == CPU_SMT_FORCE_DISABLED) {
 386                 enum cpuhp_smt_control old = cpu_smt_control;
 387
 388                 ret = cpuhp_smt_enable();
 389                 if (ret)
 390                         goto out;
 391                 ret = cpuhp_smt_disable(old);
 392                 if (ret)
 393                         goto out;
 394         }
 395 out:
 396         cpu_hotplug_disable();
 397         return ret;
 398 }