| 1 | // SPDX-License-Identifier: GPL-2.0 |
| 2 | /* |
| 3 | * x86_64 specific EFI support functions |
| 4 | * Based on Extensible Firmware Interface Specification version 1.0 |
| 5 | * |
| 6 | * Copyright (C) 2005-2008 Intel Co. |
| 7 | * Fenghua Yu <fenghua.yu@intel.com> |
| 8 | * Bibo Mao <bibo.mao@intel.com> |
| 9 | * Chandramouli Narayanan <mouli@linux.intel.com> |
| 10 | * Huang Ying <ying.huang@intel.com> |
| 11 | * |
| 12 | * Code to convert EFI to E820 map has been implemented in elilo bootloader |
| 13 | * based on a EFI patch by Edgar Hucek. Based on the E820 map, the page table |
| 14 | * is setup appropriately for EFI runtime code. |
| 15 | * - mouli 06/14/2007. |
| 16 | * |
| 17 | */ |
| 18 | |
| 19 | #define pr_fmt(fmt) "efi: " fmt |
| 20 | |
| 21 | #include <linux/kernel.h> |
| 22 | #include <linux/init.h> |
| 23 | #include <linux/mm.h> |
| 24 | #include <linux/types.h> |
| 25 | #include <linux/spinlock.h> |
| 26 | #include <linux/memblock.h> |
| 27 | #include <linux/ioport.h> |
| 28 | #include <linux/mc146818rtc.h> |
| 29 | #include <linux/efi.h> |
| 30 | #include <linux/export.h> |
| 31 | #include <linux/uaccess.h> |
| 32 | #include <linux/io.h> |
| 33 | #include <linux/reboot.h> |
| 34 | #include <linux/slab.h> |
| 35 | #include <linux/ucs2_string.h> |
| 36 | #include <linux/cc_platform.h> |
| 37 | #include <linux/sched/task.h> |
| 38 | |
| 39 | #include <asm/setup.h> |
| 40 | #include <asm/page.h> |
| 41 | #include <asm/e820/api.h> |
| 42 | #include <asm/tlbflush.h> |
| 43 | #include <asm/proto.h> |
| 44 | #include <asm/efi.h> |
| 45 | #include <asm/cacheflush.h> |
| 46 | #include <asm/fixmap.h> |
| 47 | #include <asm/realmode.h> |
| 48 | #include <asm/time.h> |
| 49 | #include <asm/pgalloc.h> |
| 50 | #include <asm/sev.h> |
| 51 | |
| 52 | /* |
| 53 | * We allocate runtime services regions top-down, starting from -4G, i.e. |
| 54 | * 0xffff_ffff_0000_0000 and limit EFI VA mapping space to 64G. |
| 55 | */ |
| 56 | static u64 efi_va = EFI_VA_START; |
| 57 | static struct mm_struct *efi_prev_mm; |
| 58 | |
| 59 | /* |
| 60 | * We need our own copy of the higher levels of the page tables |
| 61 | * because we want to avoid inserting EFI region mappings (EFI_VA_END |
| 62 | * to EFI_VA_START) into the standard kernel page tables. Everything |
| 63 | * else can be shared, see efi_sync_low_kernel_mappings(). |
| 64 | * |
| 65 | * We don't want the pgd on the pgd_list and cannot use pgd_alloc() for the |
| 66 | * allocation. |
| 67 | */ |
| 68 | int __init efi_alloc_page_tables(void) |
| 69 | { |
| 70 | pgd_t *pgd, *efi_pgd; |
| 71 | p4d_t *p4d; |
| 72 | pud_t *pud; |
| 73 | gfp_t gfp_mask; |
| 74 | |
| 75 | gfp_mask = GFP_KERNEL | __GFP_ZERO; |
| 76 | efi_pgd = (pgd_t *)__get_free_pages(gfp_mask, pgd_allocation_order()); |
| 77 | if (!efi_pgd) |
| 78 | goto fail; |
| 79 | |
| 80 | pgd = efi_pgd + pgd_index(EFI_VA_END); |
| 81 | p4d = p4d_alloc(mm: &init_mm, pgd, EFI_VA_END); |
| 82 | if (!p4d) |
| 83 | goto free_pgd; |
| 84 | |
| 85 | pud = pud_alloc(mm: &init_mm, p4d, EFI_VA_END); |
| 86 | if (!pud) |
| 87 | goto free_p4d; |
| 88 | |
| 89 | efi_mm.pgd = efi_pgd; |
| 90 | mm_init_cpumask(mm: &efi_mm); |
| 91 | init_new_context(NULL, mm: &efi_mm); |
| 92 | set_notrack_mm(&efi_mm); |
| 93 | |
| 94 | return 0; |
| 95 | |
| 96 | free_p4d: |
| 97 | if (pgtable_l5_enabled()) |
| 98 | free_page((unsigned long)pgd_page_vaddr(*pgd)); |
| 99 | free_pgd: |
| 100 | free_pages(addr: (unsigned long)efi_pgd, order: pgd_allocation_order()); |
| 101 | fail: |
| 102 | return -ENOMEM; |
| 103 | } |
| 104 | |
| 105 | /* |
| 106 | * Add low kernel mappings for passing arguments to EFI functions. |
| 107 | */ |
| 108 | void efi_sync_low_kernel_mappings(void) |
| 109 | { |
| 110 | unsigned num_entries; |
| 111 | pgd_t *pgd_k, *pgd_efi; |
| 112 | p4d_t *p4d_k, *p4d_efi; |
| 113 | pud_t *pud_k, *pud_efi; |
| 114 | pgd_t *efi_pgd = efi_mm.pgd; |
| 115 | |
| 116 | pgd_efi = efi_pgd + pgd_index(PAGE_OFFSET); |
| 117 | pgd_k = pgd_offset_k(PAGE_OFFSET); |
| 118 | |
| 119 | num_entries = pgd_index(EFI_VA_END) - pgd_index(PAGE_OFFSET); |
| 120 | memcpy(to: pgd_efi, from: pgd_k, len: sizeof(pgd_t) * num_entries); |
| 121 | |
| 122 | pgd_efi = efi_pgd + pgd_index(EFI_VA_END); |
| 123 | pgd_k = pgd_offset_k(EFI_VA_END); |
| 124 | p4d_efi = p4d_offset(pgd: pgd_efi, address: 0); |
| 125 | p4d_k = p4d_offset(pgd: pgd_k, address: 0); |
| 126 | |
| 127 | num_entries = p4d_index(EFI_VA_END); |
| 128 | memcpy(to: p4d_efi, from: p4d_k, len: sizeof(p4d_t) * num_entries); |
| 129 | |
| 130 | /* |
| 131 | * We share all the PUD entries apart from those that map the |
| 132 | * EFI regions. Copy around them. |
| 133 | */ |
| 134 | BUILD_BUG_ON((EFI_VA_START & ~PUD_MASK) != 0); |
| 135 | BUILD_BUG_ON((EFI_VA_END & ~PUD_MASK) != 0); |
| 136 | |
| 137 | p4d_efi = p4d_offset(pgd: pgd_efi, EFI_VA_END); |
| 138 | p4d_k = p4d_offset(pgd: pgd_k, EFI_VA_END); |
| 139 | pud_efi = pud_offset(p4d: p4d_efi, address: 0); |
| 140 | pud_k = pud_offset(p4d: p4d_k, address: 0); |
| 141 | |
| 142 | num_entries = pud_index(EFI_VA_END); |
| 143 | memcpy(to: pud_efi, from: pud_k, len: sizeof(pud_t) * num_entries); |
| 144 | |
| 145 | pud_efi = pud_offset(p4d: p4d_efi, EFI_VA_START); |
| 146 | pud_k = pud_offset(p4d: p4d_k, EFI_VA_START); |
| 147 | |
| 148 | num_entries = PTRS_PER_PUD - pud_index(EFI_VA_START); |
| 149 | memcpy(to: pud_efi, from: pud_k, len: sizeof(pud_t) * num_entries); |
| 150 | } |
| 151 | |
| 152 | /* |
| 153 | * Wrapper for slow_virt_to_phys() that handles NULL addresses. |
| 154 | */ |
| 155 | static inline phys_addr_t |
| 156 | virt_to_phys_or_null_size(void *va, unsigned long size) |
| 157 | { |
| 158 | phys_addr_t pa; |
| 159 | |
| 160 | if (!va) |
| 161 | return 0; |
| 162 | |
| 163 | if (virt_addr_valid(va)) |
| 164 | return virt_to_phys(address: va); |
| 165 | |
| 166 | pa = slow_virt_to_phys(address: va); |
| 167 | |
| 168 | /* check if the object crosses a page boundary */ |
| 169 | if (WARN_ON((pa ^ (pa + size - 1)) & PAGE_MASK)) |
| 170 | return 0; |
| 171 | |
| 172 | return pa; |
| 173 | } |
| 174 | |
| 175 | #define virt_to_phys_or_null(addr) \ |
| 176 | virt_to_phys_or_null_size((addr), sizeof(*(addr))) |
| 177 | |
| 178 | int __init efi_setup_page_tables(unsigned long pa_memmap, unsigned num_pages) |
| 179 | { |
| 180 | extern const u8 __efi64_thunk_ret_tramp[]; |
| 181 | unsigned long pfn, text, pf, rodata, tramp; |
| 182 | struct page *page; |
| 183 | unsigned npages; |
| 184 | pgd_t *pgd = efi_mm.pgd; |
| 185 | |
| 186 | /* |
| 187 | * It can happen that the physical address of new_memmap lands in memory |
| 188 | * which is not mapped in the EFI page table. Therefore we need to go |
| 189 | * and ident-map those pages containing the map before calling |
| 190 | * phys_efi_set_virtual_address_map(). |
| 191 | */ |
| 192 | pfn = pa_memmap >> PAGE_SHIFT; |
| 193 | pf = _PAGE_NX | _PAGE_RW | _PAGE_ENC; |
| 194 | if (kernel_map_pages_in_pgd(pgd, pfn, address: pa_memmap, numpages: num_pages, page_flags: pf)) { |
| 195 | pr_err("Error ident-mapping new memmap (0x%lx)!\n" , pa_memmap); |
| 196 | return 1; |
| 197 | } |
| 198 | |
| 199 | /* |
| 200 | * Certain firmware versions are way too sentimental and still believe |
| 201 | * they are exclusive and unquestionable owners of the first physical page, |
| 202 | * even though they explicitly mark it as EFI_CONVENTIONAL_MEMORY |
| 203 | * (but then write-access it later during SetVirtualAddressMap()). |
| 204 | * |
| 205 | * Create a 1:1 mapping for this page, to avoid triple faults during early |
| 206 | * boot with such firmware. We are free to hand this page to the BIOS, |
| 207 | * as trim_bios_range() will reserve the first page and isolate it away |
| 208 | * from memory allocators anyway. |
| 209 | */ |
| 210 | if (kernel_map_pages_in_pgd(pgd, pfn: 0x0, address: 0x0, numpages: 1, page_flags: pf)) { |
| 211 | pr_err("Failed to create 1:1 mapping for the first page!\n" ); |
| 212 | return 1; |
| 213 | } |
| 214 | |
| 215 | /* |
| 216 | * When SEV-ES is active, the GHCB as set by the kernel will be used |
| 217 | * by firmware. Create a 1:1 unencrypted mapping for each GHCB. |
| 218 | */ |
| 219 | if (sev_es_efi_map_ghcbs_cas(pgd)) { |
| 220 | pr_err("Failed to create 1:1 mapping for the GHCBs and CAs!\n" ); |
| 221 | return 1; |
| 222 | } |
| 223 | |
| 224 | /* |
| 225 | * When making calls to the firmware everything needs to be 1:1 |
| 226 | * mapped and addressable with 32-bit pointers. Map the kernel |
| 227 | * text and allocate a new stack because we can't rely on the |
| 228 | * stack pointer being < 4GB. |
| 229 | */ |
| 230 | if (!efi_is_mixed()) |
| 231 | return 0; |
| 232 | |
| 233 | page = alloc_page(GFP_KERNEL|__GFP_DMA32); |
| 234 | if (!page) { |
| 235 | pr_err("Unable to allocate EFI runtime stack < 4GB\n" ); |
| 236 | return 1; |
| 237 | } |
| 238 | |
| 239 | efi_mixed_mode_stack_pa = page_to_phys(page + 1); /* stack grows down */ |
| 240 | |
| 241 | npages = (_etext - _text) >> PAGE_SHIFT; |
| 242 | text = __pa(_text); |
| 243 | |
| 244 | if (kernel_unmap_pages_in_pgd(pgd, address: text, numpages: npages)) { |
| 245 | pr_err("Failed to unmap kernel text 1:1 mapping\n" ); |
| 246 | return 1; |
| 247 | } |
| 248 | |
| 249 | npages = (__end_rodata - __start_rodata) >> PAGE_SHIFT; |
| 250 | rodata = __pa(__start_rodata); |
| 251 | pfn = rodata >> PAGE_SHIFT; |
| 252 | |
| 253 | pf = _PAGE_NX | _PAGE_ENC; |
| 254 | if (kernel_map_pages_in_pgd(pgd, pfn, address: rodata, numpages: npages, page_flags: pf)) { |
| 255 | pr_err("Failed to map kernel rodata 1:1\n" ); |
| 256 | return 1; |
| 257 | } |
| 258 | |
| 259 | tramp = __pa(__efi64_thunk_ret_tramp); |
| 260 | pfn = tramp >> PAGE_SHIFT; |
| 261 | |
| 262 | pf = _PAGE_ENC; |
| 263 | if (kernel_map_pages_in_pgd(pgd, pfn, address: tramp, numpages: 1, page_flags: pf)) { |
| 264 | pr_err("Failed to map mixed mode return trampoline\n" ); |
| 265 | return 1; |
| 266 | } |
| 267 | |
| 268 | return 0; |
| 269 | } |
| 270 | |
| 271 | static void __init __map_region(efi_memory_desc_t *md, u64 va) |
| 272 | { |
| 273 | unsigned long flags = _PAGE_RW; |
| 274 | unsigned long pfn; |
| 275 | pgd_t *pgd = efi_mm.pgd; |
| 276 | |
| 277 | /* |
| 278 | * EFI_RUNTIME_SERVICES_CODE regions typically cover PE/COFF |
| 279 | * executable images in memory that consist of both R-X and |
| 280 | * RW- sections, so we cannot apply read-only or non-exec |
| 281 | * permissions just yet. However, modern EFI systems provide |
| 282 | * a memory attributes table that describes those sections |
| 283 | * with the appropriate restricted permissions, which are |
| 284 | * applied in efi_runtime_update_mappings() below. All other |
| 285 | * regions can be mapped non-executable at this point, with |
| 286 | * the exception of boot services code regions, but those will |
| 287 | * be unmapped again entirely in efi_free_boot_services(). |
| 288 | */ |
| 289 | if (md->type != EFI_BOOT_SERVICES_CODE && |
| 290 | md->type != EFI_RUNTIME_SERVICES_CODE) |
| 291 | flags |= _PAGE_NX; |
| 292 | |
| 293 | if (!(md->attribute & EFI_MEMORY_WB)) |
| 294 | flags |= _PAGE_PCD; |
| 295 | |
| 296 | if (cc_platform_has(attr: CC_ATTR_GUEST_MEM_ENCRYPT) && |
| 297 | md->type != EFI_MEMORY_MAPPED_IO) |
| 298 | flags |= _PAGE_ENC; |
| 299 | |
| 300 | pfn = md->phys_addr >> PAGE_SHIFT; |
| 301 | if (kernel_map_pages_in_pgd(pgd, pfn, address: va, numpages: md->num_pages, page_flags: flags)) |
| 302 | pr_warn("Error mapping PA 0x%llx -> VA 0x%llx!\n" , |
| 303 | md->phys_addr, va); |
| 304 | } |
| 305 | |
| 306 | void __init efi_map_region(efi_memory_desc_t *md) |
| 307 | { |
| 308 | unsigned long size = md->num_pages << PAGE_SHIFT; |
| 309 | u64 pa = md->phys_addr; |
| 310 | |
| 311 | /* |
| 312 | * Make sure the 1:1 mappings are present as a catch-all for b0rked |
| 313 | * firmware which doesn't update all internal pointers after switching |
| 314 | * to virtual mode and would otherwise crap on us. |
| 315 | */ |
| 316 | __map_region(md, va: md->phys_addr); |
| 317 | |
| 318 | /* |
| 319 | * Enforce the 1:1 mapping as the default virtual address when |
| 320 | * booting in EFI mixed mode, because even though we may be |
| 321 | * running a 64-bit kernel, the firmware may only be 32-bit. |
| 322 | */ |
| 323 | if (efi_is_mixed()) { |
| 324 | md->virt_addr = md->phys_addr; |
| 325 | return; |
| 326 | } |
| 327 | |
| 328 | efi_va -= size; |
| 329 | |
| 330 | /* Is PA 2M-aligned? */ |
| 331 | if (!(pa & (PMD_SIZE - 1))) { |
| 332 | efi_va &= PMD_MASK; |
| 333 | } else { |
| 334 | u64 pa_offset = pa & (PMD_SIZE - 1); |
| 335 | u64 prev_va = efi_va; |
| 336 | |
| 337 | /* get us the same offset within this 2M page */ |
| 338 | efi_va = (efi_va & PMD_MASK) + pa_offset; |
| 339 | |
| 340 | if (efi_va > prev_va) |
| 341 | efi_va -= PMD_SIZE; |
| 342 | } |
| 343 | |
| 344 | if (efi_va < EFI_VA_END) { |
| 345 | pr_warn(FW_WARN "VA address range overflow!\n" ); |
| 346 | return; |
| 347 | } |
| 348 | |
| 349 | /* Do the VA map */ |
| 350 | __map_region(md, va: efi_va); |
| 351 | md->virt_addr = efi_va; |
| 352 | } |
| 353 | |
| 354 | /* |
| 355 | * kexec kernel will use efi_map_region_fixed to map efi runtime memory ranges. |
| 356 | * md->virt_addr is the original virtual address which had been mapped in kexec |
| 357 | * 1st kernel. |
| 358 | */ |
| 359 | void __init efi_map_region_fixed(efi_memory_desc_t *md) |
| 360 | { |
| 361 | __map_region(md, va: md->phys_addr); |
| 362 | __map_region(md, va: md->virt_addr); |
| 363 | } |
| 364 | |
| 365 | void __init parse_efi_setup(u64 phys_addr, u32 data_len) |
| 366 | { |
| 367 | efi_setup = phys_addr + sizeof(struct setup_data); |
| 368 | } |
| 369 | |
| 370 | static int __init efi_update_mappings(efi_memory_desc_t *md, unsigned long pf) |
| 371 | { |
| 372 | unsigned long pfn; |
| 373 | pgd_t *pgd = efi_mm.pgd; |
| 374 | int err1, err2; |
| 375 | |
| 376 | /* Update the 1:1 mapping */ |
| 377 | pfn = md->phys_addr >> PAGE_SHIFT; |
| 378 | err1 = kernel_map_pages_in_pgd(pgd, pfn, address: md->phys_addr, numpages: md->num_pages, page_flags: pf); |
| 379 | if (err1) { |
| 380 | pr_err("Error while updating 1:1 mapping PA 0x%llx -> VA 0x%llx!\n" , |
| 381 | md->phys_addr, md->virt_addr); |
| 382 | } |
| 383 | |
| 384 | err2 = kernel_map_pages_in_pgd(pgd, pfn, address: md->virt_addr, numpages: md->num_pages, page_flags: pf); |
| 385 | if (err2) { |
| 386 | pr_err("Error while updating VA mapping PA 0x%llx -> VA 0x%llx!\n" , |
| 387 | md->phys_addr, md->virt_addr); |
| 388 | } |
| 389 | |
| 390 | return err1 || err2; |
| 391 | } |
| 392 | |
| 393 | bool efi_disable_ibt_for_runtime __ro_after_init = true; |
| 394 | |
| 395 | static int __init efi_update_mem_attr(struct mm_struct *mm, efi_memory_desc_t *md, |
| 396 | bool has_ibt) |
| 397 | { |
| 398 | unsigned long pf = 0; |
| 399 | |
| 400 | efi_disable_ibt_for_runtime |= !has_ibt; |
| 401 | |
| 402 | if (md->attribute & EFI_MEMORY_XP) |
| 403 | pf |= _PAGE_NX; |
| 404 | |
| 405 | if (!(md->attribute & EFI_MEMORY_RO)) |
| 406 | pf |= _PAGE_RW; |
| 407 | |
| 408 | if (cc_platform_has(attr: CC_ATTR_GUEST_MEM_ENCRYPT)) |
| 409 | pf |= _PAGE_ENC; |
| 410 | |
| 411 | return efi_update_mappings(md, pf); |
| 412 | } |
| 413 | |
| 414 | void __init efi_runtime_update_mappings(void) |
| 415 | { |
| 416 | if (efi_enabled(EFI_MEM_ATTR)) { |
| 417 | efi_disable_ibt_for_runtime = false; |
| 418 | efi_memattr_apply_permissions(NULL, fn: efi_update_mem_attr); |
| 419 | } |
| 420 | } |
| 421 | |
| 422 | void __init efi_dump_pagetable(void) |
| 423 | { |
| 424 | #ifdef CONFIG_EFI_PGT_DUMP |
| 425 | ptdump_walk_pgd_level(NULL, mm: &efi_mm); |
| 426 | #endif |
| 427 | } |
| 428 | |
| 429 | /* |
| 430 | * Makes the calling thread switch to/from efi_mm context. Can be used |
| 431 | * in a kernel thread and user context. Preemption needs to remain disabled |
| 432 | * while the EFI-mm is borrowed. mmgrab()/mmdrop() is not used because the mm |
| 433 | * can not change under us. |
| 434 | * It should be ensured that there are no concurrent calls to this function. |
| 435 | */ |
| 436 | static void efi_enter_mm(void) |
| 437 | { |
| 438 | efi_prev_mm = use_temporary_mm(temp_mm: &efi_mm); |
| 439 | } |
| 440 | |
| 441 | static void efi_leave_mm(void) |
| 442 | { |
| 443 | unuse_temporary_mm(prev_mm: efi_prev_mm); |
| 444 | } |
| 445 | |
| 446 | void arch_efi_call_virt_setup(void) |
| 447 | { |
| 448 | efi_sync_low_kernel_mappings(); |
| 449 | efi_fpu_begin(); |
| 450 | firmware_restrict_branch_speculation_start(); |
| 451 | efi_enter_mm(); |
| 452 | } |
| 453 | |
| 454 | void arch_efi_call_virt_teardown(void) |
| 455 | { |
| 456 | efi_leave_mm(); |
| 457 | firmware_restrict_branch_speculation_end(); |
| 458 | efi_fpu_end(); |
| 459 | } |
| 460 | |
| 461 | static DEFINE_SPINLOCK(efi_runtime_lock); |
| 462 | |
| 463 | /* |
| 464 | * DS and ES contain user values. We need to save them. |
| 465 | * The 32-bit EFI code needs a valid DS, ES, and SS. There's no |
| 466 | * need to save the old SS: __KERNEL_DS is always acceptable. |
| 467 | */ |
| 468 | #define __efi_thunk(func, ...) \ |
| 469 | ({ \ |
| 470 | unsigned short __ds, __es; \ |
| 471 | efi_status_t ____s; \ |
| 472 | \ |
| 473 | savesegment(ds, __ds); \ |
| 474 | savesegment(es, __es); \ |
| 475 | \ |
| 476 | loadsegment(ss, __KERNEL_DS); \ |
| 477 | loadsegment(ds, __KERNEL_DS); \ |
| 478 | loadsegment(es, __KERNEL_DS); \ |
| 479 | \ |
| 480 | ____s = efi64_thunk(efi.runtime->mixed_mode.func, __VA_ARGS__); \ |
| 481 | \ |
| 482 | loadsegment(ds, __ds); \ |
| 483 | loadsegment(es, __es); \ |
| 484 | \ |
| 485 | ____s ^= (____s & BIT(31)) | (____s & BIT_ULL(31)) << 32; \ |
| 486 | ____s; \ |
| 487 | }) |
| 488 | |
| 489 | /* |
| 490 | * Switch to the EFI page tables early so that we can access the 1:1 |
| 491 | * runtime services mappings which are not mapped in any other page |
| 492 | * tables. |
| 493 | * |
| 494 | * Also, disable interrupts because the IDT points to 64-bit handlers, |
| 495 | * which aren't going to function correctly when we switch to 32-bit. |
| 496 | */ |
| 497 | #define efi_thunk(func...) \ |
| 498 | ({ \ |
| 499 | efi_status_t __s; \ |
| 500 | \ |
| 501 | arch_efi_call_virt_setup(); \ |
| 502 | \ |
| 503 | __s = __efi_thunk(func); \ |
| 504 | \ |
| 505 | arch_efi_call_virt_teardown(); \ |
| 506 | \ |
| 507 | __s; \ |
| 508 | }) |
| 509 | |
| 510 | static efi_status_t __init __no_sanitize_address |
| 511 | efi_thunk_set_virtual_address_map(unsigned long memory_map_size, |
| 512 | unsigned long descriptor_size, |
| 513 | u32 descriptor_version, |
| 514 | efi_memory_desc_t *virtual_map) |
| 515 | { |
| 516 | efi_status_t status; |
| 517 | unsigned long flags; |
| 518 | |
| 519 | efi_sync_low_kernel_mappings(); |
| 520 | local_irq_save(flags); |
| 521 | |
| 522 | efi_enter_mm(); |
| 523 | |
| 524 | status = __efi_thunk(set_virtual_address_map, memory_map_size, |
| 525 | descriptor_size, descriptor_version, virtual_map); |
| 526 | |
| 527 | efi_leave_mm(); |
| 528 | local_irq_restore(flags); |
| 529 | |
| 530 | return status; |
| 531 | } |
| 532 | |
| 533 | static efi_status_t efi_thunk_get_time(efi_time_t *tm, efi_time_cap_t *tc) |
| 534 | { |
| 535 | return EFI_UNSUPPORTED; |
| 536 | } |
| 537 | |
| 538 | static efi_status_t efi_thunk_set_time(efi_time_t *tm) |
| 539 | { |
| 540 | return EFI_UNSUPPORTED; |
| 541 | } |
| 542 | |
| 543 | static efi_status_t |
| 544 | efi_thunk_get_wakeup_time(efi_bool_t *enabled, efi_bool_t *pending, |
| 545 | efi_time_t *tm) |
| 546 | { |
| 547 | return EFI_UNSUPPORTED; |
| 548 | } |
| 549 | |
| 550 | static efi_status_t |
| 551 | efi_thunk_set_wakeup_time(efi_bool_t enabled, efi_time_t *tm) |
| 552 | { |
| 553 | return EFI_UNSUPPORTED; |
| 554 | } |
| 555 | |
| 556 | static unsigned long efi_name_size(efi_char16_t *name) |
| 557 | { |
| 558 | return ucs2_strsize(data: name, EFI_VAR_NAME_LEN) + 1; |
| 559 | } |
| 560 | |
| 561 | static efi_status_t |
| 562 | efi_thunk_get_variable(efi_char16_t *name, efi_guid_t *vendor, |
| 563 | u32 *attr, unsigned long *data_size, void *data) |
| 564 | { |
| 565 | u8 buf[24] __aligned(8); |
| 566 | efi_guid_t *vnd = PTR_ALIGN((efi_guid_t *)buf, sizeof(*vnd)); |
| 567 | efi_status_t status; |
| 568 | u32 phys_name, phys_vendor, phys_attr; |
| 569 | u32 phys_data_size, phys_data; |
| 570 | unsigned long flags; |
| 571 | |
| 572 | spin_lock_irqsave(&efi_runtime_lock, flags); |
| 573 | |
| 574 | *vnd = *vendor; |
| 575 | |
| 576 | phys_data_size = virt_to_phys_or_null(data_size); |
| 577 | phys_vendor = virt_to_phys_or_null(vnd); |
| 578 | phys_name = virt_to_phys_or_null_size(va: name, size: efi_name_size(name)); |
| 579 | phys_attr = virt_to_phys_or_null(attr); |
| 580 | phys_data = virt_to_phys_or_null_size(va: data, size: *data_size); |
| 581 | |
| 582 | if (!phys_name || (data && !phys_data)) |
| 583 | status = EFI_INVALID_PARAMETER; |
| 584 | else |
| 585 | status = efi_thunk(get_variable, phys_name, phys_vendor, |
| 586 | phys_attr, phys_data_size, phys_data); |
| 587 | |
| 588 | spin_unlock_irqrestore(lock: &efi_runtime_lock, flags); |
| 589 | |
| 590 | return status; |
| 591 | } |
| 592 | |
| 593 | static efi_status_t |
| 594 | efi_thunk_set_variable(efi_char16_t *name, efi_guid_t *vendor, |
| 595 | u32 attr, unsigned long data_size, void *data) |
| 596 | { |
| 597 | u8 buf[24] __aligned(8); |
| 598 | efi_guid_t *vnd = PTR_ALIGN((efi_guid_t *)buf, sizeof(*vnd)); |
| 599 | u32 phys_name, phys_vendor, phys_data; |
| 600 | efi_status_t status; |
| 601 | unsigned long flags; |
| 602 | |
| 603 | spin_lock_irqsave(&efi_runtime_lock, flags); |
| 604 | |
| 605 | *vnd = *vendor; |
| 606 | |
| 607 | phys_name = virt_to_phys_or_null_size(va: name, size: efi_name_size(name)); |
| 608 | phys_vendor = virt_to_phys_or_null(vnd); |
| 609 | phys_data = virt_to_phys_or_null_size(va: data, size: data_size); |
| 610 | |
| 611 | if (!phys_name || (data && !phys_data)) |
| 612 | status = EFI_INVALID_PARAMETER; |
| 613 | else |
| 614 | status = efi_thunk(set_variable, phys_name, phys_vendor, |
| 615 | attr, data_size, phys_data); |
| 616 | |
| 617 | spin_unlock_irqrestore(lock: &efi_runtime_lock, flags); |
| 618 | |
| 619 | return status; |
| 620 | } |
| 621 | |
| 622 | static efi_status_t |
| 623 | efi_thunk_set_variable_nonblocking(efi_char16_t *name, efi_guid_t *vendor, |
| 624 | u32 attr, unsigned long data_size, |
| 625 | void *data) |
| 626 | { |
| 627 | u8 buf[24] __aligned(8); |
| 628 | efi_guid_t *vnd = PTR_ALIGN((efi_guid_t *)buf, sizeof(*vnd)); |
| 629 | u32 phys_name, phys_vendor, phys_data; |
| 630 | efi_status_t status; |
| 631 | unsigned long flags; |
| 632 | |
| 633 | if (!spin_trylock_irqsave(&efi_runtime_lock, flags)) |
| 634 | return EFI_NOT_READY; |
| 635 | |
| 636 | *vnd = *vendor; |
| 637 | |
| 638 | phys_name = virt_to_phys_or_null_size(va: name, size: efi_name_size(name)); |
| 639 | phys_vendor = virt_to_phys_or_null(vnd); |
| 640 | phys_data = virt_to_phys_or_null_size(va: data, size: data_size); |
| 641 | |
| 642 | if (!phys_name || (data && !phys_data)) |
| 643 | status = EFI_INVALID_PARAMETER; |
| 644 | else |
| 645 | status = efi_thunk(set_variable, phys_name, phys_vendor, |
| 646 | attr, data_size, phys_data); |
| 647 | |
| 648 | spin_unlock_irqrestore(lock: &efi_runtime_lock, flags); |
| 649 | |
| 650 | return status; |
| 651 | } |
| 652 | |
| 653 | static efi_status_t |
| 654 | efi_thunk_get_next_variable(unsigned long *name_size, |
| 655 | efi_char16_t *name, |
| 656 | efi_guid_t *vendor) |
| 657 | { |
| 658 | u8 buf[24] __aligned(8); |
| 659 | efi_guid_t *vnd = PTR_ALIGN((efi_guid_t *)buf, sizeof(*vnd)); |
| 660 | efi_status_t status; |
| 661 | u32 phys_name_size, phys_name, phys_vendor; |
| 662 | unsigned long flags; |
| 663 | |
| 664 | spin_lock_irqsave(&efi_runtime_lock, flags); |
| 665 | |
| 666 | *vnd = *vendor; |
| 667 | |
| 668 | phys_name_size = virt_to_phys_or_null(name_size); |
| 669 | phys_vendor = virt_to_phys_or_null(vnd); |
| 670 | phys_name = virt_to_phys_or_null_size(va: name, size: *name_size); |
| 671 | |
| 672 | if (!phys_name) |
| 673 | status = EFI_INVALID_PARAMETER; |
| 674 | else |
| 675 | status = efi_thunk(get_next_variable, phys_name_size, |
| 676 | phys_name, phys_vendor); |
| 677 | |
| 678 | spin_unlock_irqrestore(lock: &efi_runtime_lock, flags); |
| 679 | |
| 680 | *vendor = *vnd; |
| 681 | return status; |
| 682 | } |
| 683 | |
| 684 | static efi_status_t |
| 685 | efi_thunk_get_next_high_mono_count(u32 *count) |
| 686 | { |
| 687 | return EFI_UNSUPPORTED; |
| 688 | } |
| 689 | |
| 690 | static void |
| 691 | efi_thunk_reset_system(int reset_type, efi_status_t status, |
| 692 | unsigned long data_size, efi_char16_t *data) |
| 693 | { |
| 694 | u32 phys_data; |
| 695 | unsigned long flags; |
| 696 | |
| 697 | spin_lock_irqsave(&efi_runtime_lock, flags); |
| 698 | |
| 699 | phys_data = virt_to_phys_or_null_size(va: data, size: data_size); |
| 700 | |
| 701 | efi_thunk(reset_system, reset_type, status, data_size, phys_data); |
| 702 | |
| 703 | spin_unlock_irqrestore(lock: &efi_runtime_lock, flags); |
| 704 | } |
| 705 | |
| 706 | static efi_status_t |
| 707 | efi_thunk_update_capsule(efi_capsule_header_t **capsules, |
| 708 | unsigned long count, unsigned long sg_list) |
| 709 | { |
| 710 | /* |
| 711 | * To properly support this function we would need to repackage |
| 712 | * 'capsules' because the firmware doesn't understand 64-bit |
| 713 | * pointers. |
| 714 | */ |
| 715 | return EFI_UNSUPPORTED; |
| 716 | } |
| 717 | |
| 718 | static efi_status_t |
| 719 | efi_thunk_query_variable_info(u32 attr, u64 *storage_space, |
| 720 | u64 *remaining_space, |
| 721 | u64 *max_variable_size) |
| 722 | { |
| 723 | efi_status_t status; |
| 724 | u32 phys_storage, phys_remaining, phys_max; |
| 725 | unsigned long flags; |
| 726 | |
| 727 | if (efi.runtime_version < EFI_2_00_SYSTEM_TABLE_REVISION) |
| 728 | return EFI_UNSUPPORTED; |
| 729 | |
| 730 | spin_lock_irqsave(&efi_runtime_lock, flags); |
| 731 | |
| 732 | phys_storage = virt_to_phys_or_null(storage_space); |
| 733 | phys_remaining = virt_to_phys_or_null(remaining_space); |
| 734 | phys_max = virt_to_phys_or_null(max_variable_size); |
| 735 | |
| 736 | status = efi_thunk(query_variable_info, attr, phys_storage, |
| 737 | phys_remaining, phys_max); |
| 738 | |
| 739 | spin_unlock_irqrestore(lock: &efi_runtime_lock, flags); |
| 740 | |
| 741 | return status; |
| 742 | } |
| 743 | |
| 744 | static efi_status_t |
| 745 | efi_thunk_query_variable_info_nonblocking(u32 attr, u64 *storage_space, |
| 746 | u64 *remaining_space, |
| 747 | u64 *max_variable_size) |
| 748 | { |
| 749 | efi_status_t status; |
| 750 | u32 phys_storage, phys_remaining, phys_max; |
| 751 | unsigned long flags; |
| 752 | |
| 753 | if (efi.runtime_version < EFI_2_00_SYSTEM_TABLE_REVISION) |
| 754 | return EFI_UNSUPPORTED; |
| 755 | |
| 756 | if (!spin_trylock_irqsave(&efi_runtime_lock, flags)) |
| 757 | return EFI_NOT_READY; |
| 758 | |
| 759 | phys_storage = virt_to_phys_or_null(storage_space); |
| 760 | phys_remaining = virt_to_phys_or_null(remaining_space); |
| 761 | phys_max = virt_to_phys_or_null(max_variable_size); |
| 762 | |
| 763 | status = efi_thunk(query_variable_info, attr, phys_storage, |
| 764 | phys_remaining, phys_max); |
| 765 | |
| 766 | spin_unlock_irqrestore(lock: &efi_runtime_lock, flags); |
| 767 | |
| 768 | return status; |
| 769 | } |
| 770 | |
| 771 | static efi_status_t |
| 772 | efi_thunk_query_capsule_caps(efi_capsule_header_t **capsules, |
| 773 | unsigned long count, u64 *max_size, |
| 774 | int *reset_type) |
| 775 | { |
| 776 | /* |
| 777 | * To properly support this function we would need to repackage |
| 778 | * 'capsules' because the firmware doesn't understand 64-bit |
| 779 | * pointers. |
| 780 | */ |
| 781 | return EFI_UNSUPPORTED; |
| 782 | } |
| 783 | |
| 784 | void __init efi_thunk_runtime_setup(void) |
| 785 | { |
| 786 | if (!IS_ENABLED(CONFIG_EFI_MIXED)) |
| 787 | return; |
| 788 | |
| 789 | efi.get_time = efi_thunk_get_time; |
| 790 | efi.set_time = efi_thunk_set_time; |
| 791 | efi.get_wakeup_time = efi_thunk_get_wakeup_time; |
| 792 | efi.set_wakeup_time = efi_thunk_set_wakeup_time; |
| 793 | efi.get_variable = efi_thunk_get_variable; |
| 794 | efi.get_next_variable = efi_thunk_get_next_variable; |
| 795 | efi.set_variable = efi_thunk_set_variable; |
| 796 | efi.set_variable_nonblocking = efi_thunk_set_variable_nonblocking; |
| 797 | efi.get_next_high_mono_count = efi_thunk_get_next_high_mono_count; |
| 798 | efi.reset_system = efi_thunk_reset_system; |
| 799 | efi.query_variable_info = efi_thunk_query_variable_info; |
| 800 | efi.query_variable_info_nonblocking = efi_thunk_query_variable_info_nonblocking; |
| 801 | efi.update_capsule = efi_thunk_update_capsule; |
| 802 | efi.query_capsule_caps = efi_thunk_query_capsule_caps; |
| 803 | } |
| 804 | |
| 805 | efi_status_t __init __no_sanitize_address |
| 806 | efi_set_virtual_address_map(unsigned long memory_map_size, |
| 807 | unsigned long descriptor_size, |
| 808 | u32 descriptor_version, |
| 809 | efi_memory_desc_t *virtual_map, |
| 810 | unsigned long systab_phys) |
| 811 | { |
| 812 | const efi_system_table_t *systab = (efi_system_table_t *)systab_phys; |
| 813 | efi_status_t status; |
| 814 | unsigned long flags; |
| 815 | |
| 816 | if (efi_is_mixed()) |
| 817 | return efi_thunk_set_virtual_address_map(memory_map_size, |
| 818 | descriptor_size, |
| 819 | descriptor_version, |
| 820 | virtual_map); |
| 821 | efi_enter_mm(); |
| 822 | |
| 823 | efi_fpu_begin(); |
| 824 | |
| 825 | /* Disable interrupts around EFI calls: */ |
| 826 | local_irq_save(flags); |
| 827 | status = arch_efi_call_virt(efi.runtime, set_virtual_address_map, |
| 828 | memory_map_size, descriptor_size, |
| 829 | descriptor_version, virtual_map); |
| 830 | local_irq_restore(flags); |
| 831 | |
| 832 | efi_fpu_end(); |
| 833 | |
| 834 | /* grab the virtually remapped EFI runtime services table pointer */ |
| 835 | efi.runtime = READ_ONCE(systab->runtime); |
| 836 | |
| 837 | efi_leave_mm(); |
| 838 | |
| 839 | return status; |
| 840 | } |
| 841 | |