| 1 | // SPDX-License-Identifier: GPL-2.0-only |
| 2 | /* |
| 3 | * Extensible Firmware Interface |
| 4 | * |
| 5 | * Based on Extensible Firmware Interface Specification version 2.4 |
| 6 | * |
| 7 | * Copyright (C) 2013, 2014 Linaro Ltd. |
| 8 | */ |
| 9 | |
| 10 | #include <linux/efi.h> |
| 11 | #include <linux/init.h> |
| 12 | #include <linux/kmemleak.h> |
| 13 | #include <linux/kthread.h> |
| 14 | #include <linux/screen_info.h> |
| 15 | #include <linux/vmalloc.h> |
| 16 | |
| 17 | #include <asm/efi.h> |
| 18 | #include <asm/stacktrace.h> |
| 19 | #include <asm/vmap_stack.h> |
| 20 | |
| 21 | static bool region_is_misaligned(const efi_memory_desc_t *md) |
| 22 | { |
| 23 | if (PAGE_SIZE == EFI_PAGE_SIZE) |
| 24 | return false; |
| 25 | return !PAGE_ALIGNED(md->phys_addr) || |
| 26 | !PAGE_ALIGNED(md->num_pages << EFI_PAGE_SHIFT); |
| 27 | } |
| 28 | |
| 29 | /* |
| 30 | * Only regions of type EFI_RUNTIME_SERVICES_CODE need to be |
| 31 | * executable, everything else can be mapped with the XN bits |
| 32 | * set. Also take the new (optional) RO/XP bits into account. |
| 33 | */ |
| 34 | static __init ptdesc_t create_mapping_protection(efi_memory_desc_t *md) |
| 35 | { |
| 36 | u64 attr = md->attribute; |
| 37 | u32 type = md->type; |
| 38 | |
| 39 | if (type == EFI_MEMORY_MAPPED_IO) { |
| 40 | pgprot_t prot = __pgprot(PROT_DEVICE_nGnRE); |
| 41 | |
| 42 | if (arm64_is_protected_mmio(md->phys_addr, |
| 43 | md->num_pages << EFI_PAGE_SHIFT)) |
| 44 | prot = pgprot_encrypted(prot); |
| 45 | else |
| 46 | prot = pgprot_decrypted(prot); |
| 47 | return pgprot_val(prot); |
| 48 | } |
| 49 | |
| 50 | if (region_is_misaligned(md)) { |
| 51 | static bool __initdata code_is_misaligned; |
| 52 | |
| 53 | /* |
| 54 | * Regions that are not aligned to the OS page size cannot be |
| 55 | * mapped with strict permissions, as those might interfere |
| 56 | * with the permissions that are needed by the adjacent |
| 57 | * region's mapping. However, if we haven't encountered any |
| 58 | * misaligned runtime code regions so far, we can safely use |
| 59 | * non-executable permissions for non-code regions. |
| 60 | */ |
| 61 | code_is_misaligned |= (type == EFI_RUNTIME_SERVICES_CODE); |
| 62 | |
| 63 | return code_is_misaligned ? pgprot_val(PAGE_KERNEL_EXEC) |
| 64 | : pgprot_val(PAGE_KERNEL); |
| 65 | } |
| 66 | |
| 67 | /* R-- */ |
| 68 | if ((attr & (EFI_MEMORY_XP | EFI_MEMORY_RO)) == |
| 69 | (EFI_MEMORY_XP | EFI_MEMORY_RO)) |
| 70 | return pgprot_val(PAGE_KERNEL_RO); |
| 71 | |
| 72 | /* R-X */ |
| 73 | if (attr & EFI_MEMORY_RO) |
| 74 | return pgprot_val(PAGE_KERNEL_ROX); |
| 75 | |
| 76 | /* RW- */ |
| 77 | if (((attr & (EFI_MEMORY_RP | EFI_MEMORY_WP | EFI_MEMORY_XP)) == |
| 78 | EFI_MEMORY_XP) || |
| 79 | type != EFI_RUNTIME_SERVICES_CODE) |
| 80 | return pgprot_val(PAGE_KERNEL); |
| 81 | |
| 82 | /* RWX */ |
| 83 | return pgprot_val(PAGE_KERNEL_EXEC); |
| 84 | } |
| 85 | |
| 86 | int __init efi_create_mapping(struct mm_struct *mm, efi_memory_desc_t *md) |
| 87 | { |
| 88 | ptdesc_t prot_val = create_mapping_protection(md); |
| 89 | bool page_mappings_only = (md->type == EFI_RUNTIME_SERVICES_CODE || |
| 90 | md->type == EFI_RUNTIME_SERVICES_DATA); |
| 91 | |
| 92 | /* |
| 93 | * If this region is not aligned to the page size used by the OS, the |
| 94 | * mapping will be rounded outwards, and may end up sharing a page |
| 95 | * frame with an adjacent runtime memory region. Given that the page |
| 96 | * table descriptor covering the shared page will be rewritten when the |
| 97 | * adjacent region gets mapped, we must avoid block mappings here so we |
| 98 | * don't have to worry about splitting them when that happens. |
| 99 | */ |
| 100 | if (region_is_misaligned(md)) |
| 101 | page_mappings_only = true; |
| 102 | |
| 103 | create_pgd_mapping(mm, md->phys_addr, md->virt_addr, |
| 104 | md->num_pages << EFI_PAGE_SHIFT, |
| 105 | __pgprot(prot_val | PTE_NG), page_mappings_only); |
| 106 | return 0; |
| 107 | } |
| 108 | |
| 109 | struct set_perm_data { |
| 110 | const efi_memory_desc_t *md; |
| 111 | bool has_bti; |
| 112 | }; |
| 113 | |
| 114 | static int __init set_permissions(pte_t *ptep, unsigned long addr, void *data) |
| 115 | { |
| 116 | struct set_perm_data *spd = data; |
| 117 | const efi_memory_desc_t *md = spd->md; |
| 118 | pte_t pte = __ptep_get(ptep); |
| 119 | |
| 120 | if (md->attribute & EFI_MEMORY_RO) |
| 121 | pte = set_pte_bit(pte, __pgprot(PTE_RDONLY)); |
| 122 | if (md->attribute & EFI_MEMORY_XP) |
| 123 | pte = set_pte_bit(pte, __pgprot(PTE_PXN)); |
| 124 | else if (system_supports_bti_kernel() && spd->has_bti) |
| 125 | pte = set_pte_bit(pte, __pgprot(PTE_GP)); |
| 126 | __set_pte(ptep, pte); |
| 127 | return 0; |
| 128 | } |
| 129 | |
| 130 | int __init efi_set_mapping_permissions(struct mm_struct *mm, |
| 131 | efi_memory_desc_t *md, |
| 132 | bool has_bti) |
| 133 | { |
| 134 | struct set_perm_data data = { md, has_bti }; |
| 135 | |
| 136 | BUG_ON(md->type != EFI_RUNTIME_SERVICES_CODE && |
| 137 | md->type != EFI_RUNTIME_SERVICES_DATA); |
| 138 | |
| 139 | if (region_is_misaligned(md)) |
| 140 | return 0; |
| 141 | |
| 142 | /* |
| 143 | * Calling apply_to_page_range() is only safe on regions that are |
| 144 | * guaranteed to be mapped down to pages. Since we are only called |
| 145 | * for regions that have been mapped using efi_create_mapping() above |
| 146 | * (and this is checked by the generic Memory Attributes table parsing |
| 147 | * routines), there is no need to check that again here. |
| 148 | */ |
| 149 | return apply_to_page_range(mm, address: md->virt_addr, |
| 150 | size: md->num_pages << EFI_PAGE_SHIFT, |
| 151 | fn: set_permissions, data: &data); |
| 152 | } |
| 153 | |
| 154 | /* |
| 155 | * UpdateCapsule() depends on the system being shutdown via |
| 156 | * ResetSystem(). |
| 157 | */ |
| 158 | bool efi_poweroff_required(void) |
| 159 | { |
| 160 | return efi_enabled(EFI_RUNTIME_SERVICES); |
| 161 | } |
| 162 | |
| 163 | asmlinkage efi_status_t efi_handle_corrupted_x18(efi_status_t s, const char *f) |
| 164 | { |
| 165 | pr_err_ratelimited(FW_BUG "register x18 corrupted by EFI %s\n" , f); |
| 166 | return s; |
| 167 | } |
| 168 | |
| 169 | void arch_efi_call_virt_setup(void) |
| 170 | { |
| 171 | efi_runtime_assert_lock_held(); |
| 172 | |
| 173 | if (preemptible() && (current->flags & PF_KTHREAD)) { |
| 174 | /* |
| 175 | * Disable migration to ensure that a preempted EFI runtime |
| 176 | * service call will be resumed on the same CPU. This avoids |
| 177 | * potential issues with EFI runtime calls that are preempted |
| 178 | * while polling for an asynchronous completion of a secure |
| 179 | * firmware call, which may not permit the CPU to change. |
| 180 | */ |
| 181 | migrate_disable(); |
| 182 | kthread_use_mm(mm: &efi_mm); |
| 183 | } else { |
| 184 | efi_virtmap_load(); |
| 185 | } |
| 186 | |
| 187 | /* |
| 188 | * Enable access to the valid TTBR0_EL1 and invoke the errata |
| 189 | * workaround directly since there is no return from exception when |
| 190 | * invoking the EFI run-time services. |
| 191 | */ |
| 192 | uaccess_ttbr0_enable(); |
| 193 | post_ttbr_update_workaround(); |
| 194 | |
| 195 | __efi_fpsimd_begin(); |
| 196 | } |
| 197 | |
| 198 | void arch_efi_call_virt_teardown(void) |
| 199 | { |
| 200 | __efi_fpsimd_end(); |
| 201 | |
| 202 | /* |
| 203 | * Defer the switch to the current thread's TTBR0_EL1 until |
| 204 | * uaccess_enable(). Do so before efi_virtmap_unload() updates the |
| 205 | * saved TTBR0 value, so the userland page tables are not activated |
| 206 | * inadvertently over the back of an exception. |
| 207 | */ |
| 208 | uaccess_ttbr0_disable(); |
| 209 | |
| 210 | if (preemptible() && (current->flags & PF_KTHREAD)) { |
| 211 | kthread_unuse_mm(mm: &efi_mm); |
| 212 | migrate_enable(); |
| 213 | } else { |
| 214 | efi_virtmap_unload(); |
| 215 | } |
| 216 | } |
| 217 | |
| 218 | asmlinkage u64 *efi_rt_stack_top __ro_after_init; |
| 219 | |
| 220 | asmlinkage efi_status_t __efi_rt_asm_recover(void); |
| 221 | |
| 222 | bool efi_runtime_fixup_exception(struct pt_regs *regs, const char *msg) |
| 223 | { |
| 224 | /* Check whether the exception occurred while running the firmware */ |
| 225 | if (!current_in_efi() || regs->pc >= TASK_SIZE_64) |
| 226 | return false; |
| 227 | |
| 228 | pr_err(FW_BUG "Unable to handle %s in EFI runtime service\n" , msg); |
| 229 | add_taint(TAINT_FIRMWARE_WORKAROUND, LOCKDEP_STILL_OK); |
| 230 | clear_bit(EFI_RUNTIME_SERVICES, addr: &efi.flags); |
| 231 | |
| 232 | regs->regs[0] = EFI_ABORTED; |
| 233 | regs->regs[30] = efi_rt_stack_top[-1]; |
| 234 | regs->pc = (u64)__efi_rt_asm_recover; |
| 235 | |
| 236 | if (IS_ENABLED(CONFIG_SHADOW_CALL_STACK)) |
| 237 | regs->regs[18] = efi_rt_stack_top[-2]; |
| 238 | |
| 239 | return true; |
| 240 | } |
| 241 | |
| 242 | /* EFI requires 8 KiB of stack space for runtime services */ |
| 243 | static_assert(THREAD_SIZE >= SZ_8K); |
| 244 | |
| 245 | static int __init arm64_efi_rt_init(void) |
| 246 | { |
| 247 | void *p; |
| 248 | |
| 249 | if (!efi_enabled(EFI_RUNTIME_SERVICES)) |
| 250 | return 0; |
| 251 | |
| 252 | p = arch_alloc_vmap_stack(THREAD_SIZE, NUMA_NO_NODE); |
| 253 | if (!p) { |
| 254 | pr_warn("Failed to allocate EFI runtime stack\n" ); |
| 255 | clear_bit(EFI_RUNTIME_SERVICES, addr: &efi.flags); |
| 256 | return -ENOMEM; |
| 257 | } |
| 258 | |
| 259 | kmemleak_not_leak(ptr: p); |
| 260 | efi_rt_stack_top = p + THREAD_SIZE; |
| 261 | return 0; |
| 262 | } |
| 263 | core_initcall(arm64_efi_rt_init); |
| 264 | |