arch/arm64/kernel/hibernate.c

   1 /*:
   2  * Hibernate support specific for ARM64
   3  *
   4  * Derived from work on ARM hibernation support by:
   5  *
   6  * Ubuntu project, hibernation support for mach-dove
   7  * Copyright (C) 2010 Nokia Corporation (Hiroshi Doyu)
   8  * Copyright (C) 2010 Texas Instruments, Inc. (Teerth Reddy et al.)
   9  *  https://lkml.org/lkml/2010/6/18/4
  10  *  https://lists.linux-foundation.org/pipermail/linux-pm/2010-June/027422.html
  11  *  https://patchwork.kernel.org/patch/96442/
  12  *
  13  * Copyright (C) 2006 Rafael J. Wysocki <rjw@sisk.pl>
  14  *
  15  * License terms: GNU General Public License (GPL) version 2
  16  */
  17 #define pr_fmt(x) "hibernate: " x
  18 #include <linux/kvm_host.h>
  19 #include <linux/mm.h>
  20 #include <linux/notifier.h>
  21 #include <linux/pm.h>
  22 #include <linux/sched.h>
  23 #include <linux/suspend.h>
  24 #include <linux/utsname.h>
  25 #include <linux/version.h>
  26
  27 #include <asm/barrier.h>
  28 #include <asm/cacheflush.h>
  29 #include <asm/irqflags.h>
  30 #include <asm/memory.h>
  31 #include <asm/mmu_context.h>
  32 #include <asm/pgalloc.h>
  33 #include <asm/pgtable.h>
  34 #include <asm/pgtable-hwdef.h>
  35 #include <asm/sections.h>
  36 #include <asm/smp.h>
  37 #include <asm/suspend.h>
  38 #include <asm/virt.h>
  39
  40 /*
  41  * Hibernate core relies on this value being 0 on resume, and marks it
  42  * __nosavedata assuming it will keep the resume kernel's '0' value. This
  43  * doesn't happen with either KASLR.
  44  *
  45  * defined as "__visible int in_suspend __nosavedata" in
  46  * kernel/power/hibernate.c
  47  */
  48 extern int in_suspend;
  49
  50 /* Find a symbols alias in the linear map */
  51 #define LMADDR(x)       phys_to_virt(virt_to_phys(x))
  52
  53 /* Do we need to reset el2? */
  54 #define el2_reset_needed() (is_hyp_mode_available() && !is_kernel_in_hyp_mode())
  55
  56 /*
  57  * Start/end of the hibernate exit code, this must be copied to a 'safe'
  58  * location in memory, and executed from there.
  59  */
  60 extern char __hibernate_exit_text_start[], __hibernate_exit_text_end[];
  61
  62 /* temporary el2 vectors in the __hibernate_exit_text section. */
  63 extern char hibernate_el2_vectors[];
  64
  65 /* hyp-stub vectors, used to restore el2 during resume from hibernate. */
  66 extern char __hyp_stub_vectors[];
  67
  68 /*
  69  * Values that may not change over hibernate/resume. We put the build number
  70  * and date in here so that we guarantee not to resume with a different
  71  * kernel.
  72  */
  73 struct arch_hibernate_hdr_invariants {
  74         char            uts_version[__NEW_UTS_LEN + 1];
  75 };
  76
  77 /* These values need to be know across a hibernate/restore. */
  78 static struct arch_hibernate_hdr {
  79         struct arch_hibernate_hdr_invariants invariants;
  80
  81         /* These are needed to find the relocated kernel if built with kaslr */
  82         phys_addr_t     ttbr1_el1;
  83         void            (*reenter_kernel)(void);
  84
  85         /*
  86          * We need to know where the __hyp_stub_vectors are after restore to
  87          * re-configure el2.
  88          */
  89         phys_addr_t     __hyp_stub_vectors;
  90 } resume_hdr;
  91
  92 static inline void arch_hdr_invariants(struct arch_hibernate_hdr_invariants *i)
  93 {
  94         memset(i, 0, sizeof(*i));
  95         memcpy(i->uts_version, init_utsname()->version, sizeof(i->uts_version));
  96 }
  97
  98 int pfn_is_nosave(unsigned long pfn)
  99 {
 100         unsigned long nosave_begin_pfn = virt_to_pfn(&__nosave_begin);
 101         unsigned long nosave_end_pfn = virt_to_pfn(&__nosave_end - 1);
 102
 103         return (pfn >= nosave_begin_pfn) && (pfn <= nosave_end_pfn);
 104 }
 105
 106 void notrace save_processor_state(void)
 107 {
 108         WARN_ON(num_online_cpus() != 1);
 109 }
 110
 111 void notrace restore_processor_state(void)
 112 {
 113 }
 114
 115 int arch_hibernation_header_save(void *addr, unsigned int max_size)
 116 {
 117         struct arch_hibernate_hdr *hdr = addr;
 118
 119         if (max_size < sizeof(*hdr))
 120                 return -EOVERFLOW;
 121
 122         arch_hdr_invariants(&hdr->invariants);
 123         hdr->ttbr1_el1          = virt_to_phys(swapper_pg_dir);
 124         hdr->reenter_kernel     = _cpu_resume;
 125
 126         /* We can't use __hyp_get_vectors() because kvm may still be loaded */
 127         if (el2_reset_needed())
 128                 hdr->__hyp_stub_vectors = virt_to_phys(__hyp_stub_vectors);
 129         else
 130                 hdr->__hyp_stub_vectors = 0;
 131
 132         return 0;
 133 }
 134 EXPORT_SYMBOL(arch_hibernation_header_save);
 135
 136 int arch_hibernation_header_restore(void *addr)
 137 {
 138         struct arch_hibernate_hdr_invariants invariants;
 139         struct arch_hibernate_hdr *hdr = addr;
 140
 141         arch_hdr_invariants(&invariants);
 142         if (memcmp(&hdr->invariants, &invariants, sizeof(invariants))) {
 143                 pr_crit("Hibernate image not generated by this kernel!\n");
 144                 return -EINVAL;
 145         }
 146
 147         resume_hdr = *hdr;
 148
 149         return 0;
 150 }
 151 EXPORT_SYMBOL(arch_hibernation_header_restore);
 152
 153 /*
 154  * Copies length bytes, starting at src_start into an new page,
 155  * perform cache maintentance, then maps it at the specified address low
 156  * address as executable.
 157  *
 158  * This is used by hibernate to copy the code it needs to execute when
 159  * overwriting the kernel text. This function generates a new set of page
 160  * tables, which it loads into ttbr0.
 161  *
 162  * Length is provided as we probably only want 4K of data, even on a 64K
 163  * page system.
 164  */
 165 static int create_safe_exec_page(void *src_start, size_t length,
 166                                  unsigned long dst_addr,
 167                                  phys_addr_t *phys_dst_addr,
 168                                  void *(*allocator)(gfp_t mask),
 169                                  gfp_t mask)
 170 {
 171         int rc = 0;
 172         pgd_t *pgd;
 173         pud_t *pud;
 174         pmd_t *pmd;
 175         pte_t *pte;
 176         unsigned long dst = (unsigned long)allocator(mask);
 177
 178         if (!dst) {
 179                 rc = -ENOMEM;
 180                 goto out;
 181         }
 182
 183         memcpy((void *)dst, src_start, length);
 184         flush_icache_range(dst, dst + length);
 185
 186         pgd = pgd_offset_raw(allocator(mask), dst_addr);
 187         if (pgd_none(*pgd)) {
 188                 pud = allocator(mask);
 189                 if (!pud) {
 190                         rc = -ENOMEM;
 191                         goto out;
 192                 }
 193                 pgd_populate(&init_mm, pgd, pud);
 194         }
 195
 196         pud = pud_offset(pgd, dst_addr);
 197         if (pud_none(*pud)) {
 198                 pmd = allocator(mask);
 199                 if (!pmd) {
 200                         rc = -ENOMEM;
 201                         goto out;
 202                 }
 203                 pud_populate(&init_mm, pud, pmd);
 204         }
 205
 206         pmd = pmd_offset(pud, dst_addr);
 207         if (pmd_none(*pmd)) {
 208                 pte = allocator(mask);
 209                 if (!pte) {
 210                         rc = -ENOMEM;
 211                         goto out;
 212                 }
 213                 pmd_populate_kernel(&init_mm, pmd, pte);
 214         }
 215
 216         pte = pte_offset_kernel(pmd, dst_addr);
 217         set_pte(pte, __pte(virt_to_phys((void *)dst) |
 218                          pgprot_val(PAGE_KERNEL_EXEC)));
 219
 220         /* Load our new page tables */
 221         asm volatile("msr       ttbr0_el1, %0;"
 222                      "isb;"
 223                      "tlbi      vmalle1is;"
 224                      "dsb       ish;"
 225                      "isb" : : "r"(virt_to_phys(pgd)));
 226
 227         *phys_dst_addr = virt_to_phys((void *)dst);
 228
 229 out:
 230         return rc;
 231 }
 232
 233
 234 int swsusp_arch_suspend(void)
 235 {
 236         int ret = 0;
 237         unsigned long flags;
 238         struct sleep_stack_data state;
 239
 240         if (cpus_are_stuck_in_kernel()) {
 241                 pr_err("Can't hibernate: no mechanism to offline secondary CPUs.\n");
 242                 return -EBUSY;
 243         }
 244
 245         local_dbg_save(flags);
 246
 247         if (__cpu_suspend_enter(&state)) {
 248                 ret = swsusp_save();
 249         } else {
 250                 /* Clean kernel to PoC for secondary core startup */
 251                 __flush_dcache_area(LMADDR(KERNEL_START), KERNEL_END - KERNEL_START);
 252
 253                 /*
 254                  * Tell the hibernation core that we've just restored
 255                  * the memory
 256                  */
 257                 in_suspend = 0;
 258
 259                 __cpu_suspend_exit();
 260         }
 261
 262         local_dbg_restore(flags);
 263
 264         return ret;
 265 }
 266
 267 static int copy_pte(pmd_t *dst_pmd, pmd_t *src_pmd, unsigned long start,
 268                     unsigned long end)
 269 {
 270         pte_t *src_pte;
 271         pte_t *dst_pte;
 272         unsigned long addr = start;
 273
 274         dst_pte = (pte_t *)get_safe_page(GFP_ATOMIC);
 275         if (!dst_pte)
 276                 return -ENOMEM;
 277         pmd_populate_kernel(&init_mm, dst_pmd, dst_pte);
 278         dst_pte = pte_offset_kernel(dst_pmd, start);
 279
 280         src_pte = pte_offset_kernel(src_pmd, start);
 281         do {
 282                 if (!pte_none(*src_pte))
 283                         /*
 284                          * Resume will overwrite areas that may be marked
 285                          * read only (code, rodata). Clear the RDONLY bit from
 286                          * the temporary mappings we use during restore.
 287                          */
 288                         set_pte(dst_pte, __pte(pte_val(*src_pte) & ~PTE_RDONLY));
 289         } while (dst_pte++, src_pte++, addr += PAGE_SIZE, addr != end);
 290
 291         return 0;
 292 }
 293
 294 static int copy_pmd(pud_t *dst_pud, pud_t *src_pud, unsigned long start,
 295                     unsigned long end)
 296 {
 297         pmd_t *src_pmd;
 298         pmd_t *dst_pmd;
 299         unsigned long next;
 300         unsigned long addr = start;
 301
 302         if (pud_none(*dst_pud)) {
 303                 dst_pmd = (pmd_t *)get_safe_page(GFP_ATOMIC);
 304                 if (!dst_pmd)
 305                         return -ENOMEM;
 306                 pud_populate(&init_mm, dst_pud, dst_pmd);
 307         }
 308         dst_pmd = pmd_offset(dst_pud, start);
 309
 310         src_pmd = pmd_offset(src_pud, start);
 311         do {
 312                 next = pmd_addr_end(addr, end);
 313                 if (pmd_none(*src_pmd))
 314                         continue;
 315                 if (pmd_table(*src_pmd)) {
 316                         if (copy_pte(dst_pmd, src_pmd, addr, next))
 317                                 return -ENOMEM;
 318                 } else {
 319                         set_pmd(dst_pmd,
 320                                 __pmd(pmd_val(*src_pmd) & ~PMD_SECT_RDONLY));
 321                 }
 322         } while (dst_pmd++, src_pmd++, addr = next, addr != end);
 323
 324         return 0;
 325 }
 326
 327 static int copy_pud(pgd_t *dst_pgd, pgd_t *src_pgd, unsigned long start,
 328                     unsigned long end)
 329 {
 330         pud_t *dst_pud;
 331         pud_t *src_pud;
 332         unsigned long next;
 333         unsigned long addr = start;
 334
 335         if (pgd_none(*dst_pgd)) {
 336                 dst_pud = (pud_t *)get_safe_page(GFP_ATOMIC);
 337                 if (!dst_pud)
 338                         return -ENOMEM;
 339                 pgd_populate(&init_mm, dst_pgd, dst_pud);
 340         }
 341         dst_pud = pud_offset(dst_pgd, start);
 342
 343         src_pud = pud_offset(src_pgd, start);
 344         do {
 345                 next = pud_addr_end(addr, end);
 346                 if (pud_none(*src_pud))
 347                         continue;
 348                 if (pud_table(*(src_pud))) {
 349                         if (copy_pmd(dst_pud, src_pud, addr, next))
 350                                 return -ENOMEM;
 351                 } else {
 352                         set_pud(dst_pud,
 353                                 __pud(pud_val(*src_pud) & ~PMD_SECT_RDONLY));
 354                 }
 355         } while (dst_pud++, src_pud++, addr = next, addr != end);
 356
 357         return 0;
 358 }
 359
 360 static int copy_page_tables(pgd_t *dst_pgd, unsigned long start,
 361                             unsigned long end)
 362 {
 363         unsigned long next;
 364         unsigned long addr = start;
 365         pgd_t *src_pgd = pgd_offset_k(start);
 366
 367         dst_pgd = pgd_offset_raw(dst_pgd, start);
 368         do {
 369                 next = pgd_addr_end(addr, end);
 370                 if (pgd_none(*src_pgd))
 371                         continue;
 372                 if (copy_pud(dst_pgd, src_pgd, addr, next))
 373                         return -ENOMEM;
 374         } while (dst_pgd++, src_pgd++, addr = next, addr != end);
 375
 376         return 0;
 377 }
 378
 379 /*
 380  * Setup then Resume from the hibernate image using swsusp_arch_suspend_exit().
 381  *
 382  * Memory allocated by get_safe_page() will be dealt with by the hibernate code,
 383  * we don't need to free it here.
 384  */
 385 int swsusp_arch_resume(void)
 386 {
 387         int rc = 0;
 388         void *zero_page;
 389         size_t exit_size;
 390         pgd_t *tmp_pg_dir;
 391         void *lm_restore_pblist;
 392         phys_addr_t phys_hibernate_exit;
 393         void __noreturn (*hibernate_exit)(phys_addr_t, phys_addr_t, void *,
 394                                           void *, phys_addr_t, phys_addr_t);
 395
 396         /*
 397          * Locate the exit code in the bottom-but-one page, so that *NULL
 398          * still has disastrous affects.
 399          */
 400         hibernate_exit = (void *)PAGE_SIZE;
 401         exit_size = __hibernate_exit_text_end - __hibernate_exit_text_start;
 402         /*
 403          * Copy swsusp_arch_suspend_exit() to a safe page. This will generate
 404          * a new set of ttbr0 page tables and load them.
 405          */
 406         rc = create_safe_exec_page(__hibernate_exit_text_start, exit_size,
 407                                    (unsigned long)hibernate_exit,
 408                                    &phys_hibernate_exit,
 409                                    (void *)get_safe_page, GFP_ATOMIC);
 410         if (rc) {
 411                 pr_err("Failed to create safe executable page for hibernate_exit code.");
 412                 goto out;
 413         }
 414
 415         /*
 416          * The hibernate exit text contains a set of el2 vectors, that will
 417          * be executed at el2 with the mmu off in order to reload hyp-stub.
 418          */
 419         __flush_dcache_area(hibernate_exit, exit_size);
 420
 421         /*
 422          * Restoring the memory image will overwrite the ttbr1 page tables.
 423          * Create a second copy of just the linear map, and use this when
 424          * restoring.
 425          */
 426         tmp_pg_dir = (pgd_t *)get_safe_page(GFP_ATOMIC);
 427         if (!tmp_pg_dir) {
 428                 pr_err("Failed to allocate memory for temporary page tables.");
 429                 rc = -ENOMEM;
 430                 goto out;
 431         }
 432         rc = copy_page_tables(tmp_pg_dir, PAGE_OFFSET, 0);
 433         if (rc)
 434                 goto out;
 435
 436         /*
 437          * Since we only copied the linear map, we need to find restore_pblist's
 438          * linear map address.
 439          */
 440         lm_restore_pblist = LMADDR(restore_pblist);
 441
 442         /*
 443          * KASLR will cause the el2 vectors to be in a different location in
 444          * the resumed kernel. Load hibernate's temporary copy into el2.
 445          *
 446          * We can skip this step if we booted at EL1, or are running with VHE.
 447          */
 448         if (el2_reset_needed()) {
 449                 phys_addr_t el2_vectors = phys_hibernate_exit;  /* base */
 450                 el2_vectors += hibernate_el2_vectors -
 451                                __hibernate_exit_text_start;     /* offset */
 452
 453                 __hyp_set_vectors(el2_vectors);
 454         }
 455
 456         /*
 457          * We need a zero page that is zero before & after resume in order to
 458          * to break before make on the ttbr1 page tables.
 459          */
 460         zero_page = (void *)get_safe_page(GFP_ATOMIC);
 461
 462         hibernate_exit(virt_to_phys(tmp_pg_dir), resume_hdr.ttbr1_el1,
 463                        resume_hdr.reenter_kernel, lm_restore_pblist,
 464                        resume_hdr.__hyp_stub_vectors, virt_to_phys(zero_page));
 465
 466 out:
 467         return rc;
 468 }
 469
 470 static int check_boot_cpu_online_pm_callback(struct notifier_block *nb,
 471                                              unsigned long action, void *ptr)
 472 {
 473         if (action == PM_HIBERNATION_PREPARE &&
 474              cpumask_first(cpu_online_mask) != 0) {
 475                 pr_warn("CPU0 is offline.\n");
 476                 return notifier_from_errno(-ENODEV);
 477         }
 478
 479         return NOTIFY_OK;
 480 }
 481
 482 static int __init check_boot_cpu_online_init(void)
 483 {
 484         /*
 485          * Set this pm_notifier callback with a lower priority than
 486          * cpu_hotplug_pm_callback, so that cpu_hotplug_pm_callback will be
 487          * called earlier to disable cpu hotplug before the cpu online check.
 488          */
 489         pm_notifier(check_boot_cpu_online_pm_callback, -INT_MAX);
 490
 491         return 0;
 492 }
 493 core_initcall(check_boot_cpu_online_init);