arm64/kernel/hibernate.c

0001 // SPDX-License-Identifier: GPL-2.0-only
0002 /*:
0003  * Hibernate support specific for ARM64
0004  *
0005  * Derived from work on ARM hibernation support by:
0006  *
0007  * Ubuntu project, hibernation support for mach-dove
0008  * Copyright (C) 2010 Nokia Corporation (Hiroshi Doyu)
0009  * Copyright (C) 2010 Texas Instruments, Inc. (Teerth Reddy et al.)
0010  * Copyright (C) 2006 Rafael J. Wysocki <rjw@sisk.pl>
0011  */
0012 #define pr_fmt(x) "hibernate: " x
0013 #include <linux/cpu.h>
0014 #include <linux/kvm_host.h>
0015 #include <linux/pm.h>
0016 #include <linux/sched.h>
0017 #include <linux/suspend.h>
0018 #include <linux/utsname.h>
0019
0020 #include <asm/barrier.h>
0021 #include <asm/cacheflush.h>
0022 #include <asm/cputype.h>
0023 #include <asm/daifflags.h>
0024 #include <asm/irqflags.h>
0025 #include <asm/kexec.h>
0026 #include <asm/memory.h>
0027 #include <asm/mmu_context.h>
0028 #include <asm/mte.h>
0029 #include <asm/sections.h>
0030 #include <asm/smp.h>
0031 #include <asm/smp_plat.h>
0032 #include <asm/suspend.h>
0033 #include <asm/sysreg.h>
0034 #include <asm/trans_pgd.h>
0035 #include <asm/virt.h>
0036
0037 /*
0038  * Hibernate core relies on this value being 0 on resume, and marks it
0039  * __nosavedata assuming it will keep the resume kernel's '0' value. This
0040  * doesn't happen with either KASLR.
0041  *
0042  * defined as "__visible int in_suspend __nosavedata" in
0043  * kernel/power/hibernate.c
0044  */
0045 extern int in_suspend;
0046
0047 /* Do we need to reset el2? */
0048 #define el2_reset_needed() (is_hyp_nvhe())
0049
0050 /* hyp-stub vectors, used to restore el2 during resume from hibernate. */
0051 extern char __hyp_stub_vectors[];
0052
0053 /*
0054  * The logical cpu number we should resume on, initialised to a non-cpu
0055  * number.
0056  */
0057 static int sleep_cpu = -EINVAL;
0058
0059 /*
0060  * Values that may not change over hibernate/resume. We put the build number
0061  * and date in here so that we guarantee not to resume with a different
0062  * kernel.
0063  */
0064 struct arch_hibernate_hdr_invariants {
0065     char        uts_version[__NEW_UTS_LEN + 1];
0066 };
0067
0068 /* These values need to be know across a hibernate/restore. */
0069 static struct arch_hibernate_hdr {
0070     struct arch_hibernate_hdr_invariants invariants;
0071
0072     /* These are needed to find the relocated kernel if built with kaslr */
0073     phys_addr_t ttbr1_el1;
0074     void        (*reenter_kernel)(void);
0075
0076     /*
0077      * We need to know where the __hyp_stub_vectors are after restore to
0078      * re-configure el2.
0079      */
0080     phys_addr_t __hyp_stub_vectors;
0081
0082     u64     sleep_cpu_mpidr;
0083 } resume_hdr;
0084
0085 static inline void arch_hdr_invariants(struct arch_hibernate_hdr_invariants *i)
0086 {
0087     memset(i, 0, sizeof(*i));
0088     memcpy(i->uts_version, init_utsname()->version, sizeof(i->uts_version));
0089 }
0090
0091 int pfn_is_nosave(unsigned long pfn)
0092 {
0093     unsigned long nosave_begin_pfn = sym_to_pfn(&__nosave_begin);
0094     unsigned long nosave_end_pfn = sym_to_pfn(&__nosave_end - 1);
0095
0096     return ((pfn >= nosave_begin_pfn) && (pfn <= nosave_end_pfn)) ||
0097         crash_is_nosave(pfn);
0098 }
0099
0100 void notrace save_processor_state(void)
0101 {
0102     WARN_ON(num_online_cpus() != 1);
0103 }
0104
0105 void notrace restore_processor_state(void)
0106 {
0107 }
0108
0109 int arch_hibernation_header_save(void *addr, unsigned int max_size)
0110 {
0111     struct arch_hibernate_hdr *hdr = addr;
0112
0113     if (max_size < sizeof(*hdr))
0114         return -EOVERFLOW;
0115
0116     arch_hdr_invariants(&hdr->invariants);
0117     hdr->ttbr1_el1      = __pa_symbol(swapper_pg_dir);
0118     hdr->reenter_kernel = _cpu_resume;
0119
0120     /* We can't use __hyp_get_vectors() because kvm may still be loaded */
0121     if (el2_reset_needed())
0122         hdr->__hyp_stub_vectors = __pa_symbol(__hyp_stub_vectors);
0123     else
0124         hdr->__hyp_stub_vectors = 0;
0125
0126     /* Save the mpidr of the cpu we called cpu_suspend() on... */
0127     if (sleep_cpu < 0) {
0128         pr_err("Failing to hibernate on an unknown CPU.\n");
0129         return -ENODEV;
0130     }
0131     hdr->sleep_cpu_mpidr = cpu_logical_map(sleep_cpu);
0132     pr_info("Hibernating on CPU %d [mpidr:0x%llx]\n", sleep_cpu,
0133         hdr->sleep_cpu_mpidr);
0134
0135     return 0;
0136 }
0137 EXPORT_SYMBOL(arch_hibernation_header_save);
0138
0139 int arch_hibernation_header_restore(void *addr)
0140 {
0141     int ret;
0142     struct arch_hibernate_hdr_invariants invariants;
0143     struct arch_hibernate_hdr *hdr = addr;
0144
0145     arch_hdr_invariants(&invariants);
0146     if (memcmp(&hdr->invariants, &invariants, sizeof(invariants))) {
0147         pr_crit("Hibernate image not generated by this kernel!\n");
0148         return -EINVAL;
0149     }
0150
0151     sleep_cpu = get_logical_index(hdr->sleep_cpu_mpidr);
0152     pr_info("Hibernated on CPU %d [mpidr:0x%llx]\n", sleep_cpu,
0153         hdr->sleep_cpu_mpidr);
0154     if (sleep_cpu < 0) {
0155         pr_crit("Hibernated on a CPU not known to this kernel!\n");
0156         sleep_cpu = -EINVAL;
0157         return -EINVAL;
0158     }
0159
0160     ret = bringup_hibernate_cpu(sleep_cpu);
0161     if (ret) {
0162         sleep_cpu = -EINVAL;
0163         return ret;
0164     }
0165
0166     resume_hdr = *hdr;
0167
0168     return 0;
0169 }
0170 EXPORT_SYMBOL(arch_hibernation_header_restore);
0171
0172 static void *hibernate_page_alloc(void *arg)
0173 {
0174     return (void *)get_safe_page((__force gfp_t)(unsigned long)arg);
0175 }
0176
0177 /*
0178  * Copies length bytes, starting at src_start into an new page,
0179  * perform cache maintenance, then maps it at the specified address low
0180  * address as executable.
0181  *
0182  * This is used by hibernate to copy the code it needs to execute when
0183  * overwriting the kernel text. This function generates a new set of page
0184  * tables, which it loads into ttbr0.
0185  *
0186  * Length is provided as we probably only want 4K of data, even on a 64K
0187  * page system.
0188  */
0189 static int create_safe_exec_page(void *src_start, size_t length,
0190                  phys_addr_t *phys_dst_addr)
0191 {
0192     struct trans_pgd_info trans_info = {
0193         .trans_alloc_page   = hibernate_page_alloc,
0194         .trans_alloc_arg    = (__force void *)GFP_ATOMIC,
0195     };
0196
0197     void *page = (void *)get_safe_page(GFP_ATOMIC);
0198     phys_addr_t trans_ttbr0;
0199     unsigned long t0sz;
0200     int rc;
0201
0202     if (!page)
0203         return -ENOMEM;
0204
0205     memcpy(page, src_start, length);
0206     caches_clean_inval_pou((unsigned long)page, (unsigned long)page + length);
0207     rc = trans_pgd_idmap_page(&trans_info, &trans_ttbr0, &t0sz, page);
0208     if (rc)
0209         return rc;
0210
0211     cpu_install_ttbr0(trans_ttbr0, t0sz);
0212     *phys_dst_addr = virt_to_phys(page);
0213
0214     return 0;
0215 }
0216
0217 #ifdef CONFIG_ARM64_MTE
0218
0219 static DEFINE_XARRAY(mte_pages);
0220
0221 static int save_tags(struct page *page, unsigned long pfn)
0222 {
0223     void *tag_storage, *ret;
0224
0225     tag_storage = mte_allocate_tag_storage();
0226     if (!tag_storage)
0227         return -ENOMEM;
0228
0229     mte_save_page_tags(page_address(page), tag_storage);
0230
0231     ret = xa_store(&mte_pages, pfn, tag_storage, GFP_KERNEL);
0232     if (WARN(xa_is_err(ret), "Failed to store MTE tags")) {
0233         mte_free_tag_storage(tag_storage);
0234         return xa_err(ret);
0235     } else if (WARN(ret, "swsusp: %s: Duplicate entry", __func__)) {
0236         mte_free_tag_storage(ret);
0237     }
0238
0239     return 0;
0240 }
0241
0242 static void swsusp_mte_free_storage(void)
0243 {
0244     XA_STATE(xa_state, &mte_pages, 0);
0245     void *tags;
0246
0247     xa_lock(&mte_pages);
0248     xas_for_each(&xa_state, tags, ULONG_MAX) {
0249         mte_free_tag_storage(tags);
0250     }
0251     xa_unlock(&mte_pages);
0252
0253     xa_destroy(&mte_pages);
0254 }
0255
0256 static int swsusp_mte_save_tags(void)
0257 {
0258     struct zone *zone;
0259     unsigned long pfn, max_zone_pfn;
0260     int ret = 0;
0261     int n = 0;
0262
0263     if (!system_supports_mte())
0264         return 0;
0265
0266     for_each_populated_zone(zone) {
0267         max_zone_pfn = zone_end_pfn(zone);
0268         for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++) {
0269             struct page *page = pfn_to_online_page(pfn);
0270
0271             if (!page)
0272                 continue;
0273
0274             if (!test_bit(PG_mte_tagged, &page->flags))
0275                 continue;
0276
0277             ret = save_tags(page, pfn);
0278             if (ret) {
0279                 swsusp_mte_free_storage();
0280                 goto out;
0281             }
0282
0283             n++;
0284         }
0285     }
0286     pr_info("Saved %d MTE pages\n", n);
0287
0288 out:
0289     return ret;
0290 }
0291
0292 static void swsusp_mte_restore_tags(void)
0293 {
0294     XA_STATE(xa_state, &mte_pages, 0);
0295     int n = 0;
0296     void *tags;
0297
0298     xa_lock(&mte_pages);
0299     xas_for_each(&xa_state, tags, ULONG_MAX) {
0300         unsigned long pfn = xa_state.xa_index;
0301         struct page *page = pfn_to_online_page(pfn);
0302
0303         mte_restore_page_tags(page_address(page), tags);
0304
0305         mte_free_tag_storage(tags);
0306         n++;
0307     }
0308     xa_unlock(&mte_pages);
0309
0310     pr_info("Restored %d MTE pages\n", n);
0311
0312     xa_destroy(&mte_pages);
0313 }
0314
0315 #else   /* CONFIG_ARM64_MTE */
0316
0317 static int swsusp_mte_save_tags(void)
0318 {
0319     return 0;
0320 }
0321
0322 static void swsusp_mte_restore_tags(void)
0323 {
0324 }
0325
0326 #endif  /* CONFIG_ARM64_MTE */
0327
0328 int swsusp_arch_suspend(void)
0329 {
0330     int ret = 0;
0331     unsigned long flags;
0332     struct sleep_stack_data state;
0333
0334     if (cpus_are_stuck_in_kernel()) {
0335         pr_err("Can't hibernate: no mechanism to offline secondary CPUs.\n");
0336         return -EBUSY;
0337     }
0338
0339     flags = local_daif_save();
0340
0341     if (__cpu_suspend_enter(&state)) {
0342         /* make the crash dump kernel image visible/saveable */
0343         crash_prepare_suspend();
0344
0345         ret = swsusp_mte_save_tags();
0346         if (ret)
0347             return ret;
0348
0349         sleep_cpu = smp_processor_id();
0350         ret = swsusp_save();
0351     } else {
0352         /* Clean kernel core startup/idle code to PoC*/
0353         dcache_clean_inval_poc((unsigned long)__mmuoff_data_start,
0354                     (unsigned long)__mmuoff_data_end);
0355         dcache_clean_inval_poc((unsigned long)__idmap_text_start,
0356                     (unsigned long)__idmap_text_end);
0357
0358         /* Clean kvm setup code to PoC? */
0359         if (el2_reset_needed()) {
0360             dcache_clean_inval_poc(
0361                 (unsigned long)__hyp_idmap_text_start,
0362                 (unsigned long)__hyp_idmap_text_end);
0363             dcache_clean_inval_poc((unsigned long)__hyp_text_start,
0364                         (unsigned long)__hyp_text_end);
0365         }
0366
0367         swsusp_mte_restore_tags();
0368
0369         /* make the crash dump kernel image protected again */
0370         crash_post_resume();
0371
0372         /*
0373          * Tell the hibernation core that we've just restored
0374          * the memory
0375          */
0376         in_suspend = 0;
0377
0378         sleep_cpu = -EINVAL;
0379         __cpu_suspend_exit();
0380
0381         /*
0382          * Just in case the boot kernel did turn the SSBD
0383          * mitigation off behind our back, let's set the state
0384          * to what we expect it to be.
0385          */
0386         spectre_v4_enable_mitigation(NULL);
0387     }
0388
0389     local_daif_restore(flags);
0390
0391     return ret;
0392 }
0393
0394 /*
0395  * Setup then Resume from the hibernate image using swsusp_arch_suspend_exit().
0396  *
0397  * Memory allocated by get_safe_page() will be dealt with by the hibernate code,
0398  * we don't need to free it here.
0399  */
0400 int swsusp_arch_resume(void)
0401 {
0402     int rc;
0403     void *zero_page;
0404     size_t exit_size;
0405     pgd_t *tmp_pg_dir;
0406     phys_addr_t el2_vectors;
0407     void __noreturn (*hibernate_exit)(phys_addr_t, phys_addr_t, void *,
0408                       void *, phys_addr_t, phys_addr_t);
0409     struct trans_pgd_info trans_info = {
0410         .trans_alloc_page   = hibernate_page_alloc,
0411         .trans_alloc_arg    = (void *)GFP_ATOMIC,
0412     };
0413
0414     /*
0415      * Restoring the memory image will overwrite the ttbr1 page tables.
0416      * Create a second copy of just the linear map, and use this when
0417      * restoring.
0418      */
0419     rc = trans_pgd_create_copy(&trans_info, &tmp_pg_dir, PAGE_OFFSET,
0420                    PAGE_END);
0421     if (rc)
0422         return rc;
0423
0424     /*
0425      * We need a zero page that is zero before & after resume in order
0426      * to break before make on the ttbr1 page tables.
0427      */
0428     zero_page = (void *)get_safe_page(GFP_ATOMIC);
0429     if (!zero_page) {
0430         pr_err("Failed to allocate zero page.\n");
0431         return -ENOMEM;
0432     }
0433
0434     if (el2_reset_needed()) {
0435         rc = trans_pgd_copy_el2_vectors(&trans_info, &el2_vectors);
0436         if (rc) {
0437             pr_err("Failed to setup el2 vectors\n");
0438             return rc;
0439         }
0440     }
0441
0442     exit_size = __hibernate_exit_text_end - __hibernate_exit_text_start;
0443     /*
0444      * Copy swsusp_arch_suspend_exit() to a safe page. This will generate
0445      * a new set of ttbr0 page tables and load them.
0446      */
0447     rc = create_safe_exec_page(__hibernate_exit_text_start, exit_size,
0448                    (phys_addr_t *)&hibernate_exit);
0449     if (rc) {
0450         pr_err("Failed to create safe executable page for hibernate_exit code.\n");
0451         return rc;
0452     }
0453
0454     /*
0455      * KASLR will cause the el2 vectors to be in a different location in
0456      * the resumed kernel. Load hibernate's temporary copy into el2.
0457      *
0458      * We can skip this step if we booted at EL1, or are running with VHE.
0459      */
0460     if (el2_reset_needed())
0461         __hyp_set_vectors(el2_vectors);
0462
0463     hibernate_exit(virt_to_phys(tmp_pg_dir), resume_hdr.ttbr1_el1,
0464                resume_hdr.reenter_kernel, restore_pblist,
0465                resume_hdr.__hyp_stub_vectors, virt_to_phys(zero_page));
0466
0467     return 0;
0468 }
0469
0470 int hibernate_resume_nonboot_cpu_disable(void)
0471 {
0472     if (sleep_cpu < 0) {
0473         pr_err("Failing to resume from hibernate on an unknown CPU.\n");
0474         return -ENODEV;
0475     }
0476
0477     return freeze_secondary_cpus(sleep_cpu);
0478 }