arch/x86/platform/efi/efi_64.c

   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(&init_mm, pgd, EFI_VA_END);
  82         if (!p4d)
  83                 goto free_pgd;
  84
  85         pud = pud_alloc(&init_mm, p4d, EFI_VA_END);
  86         if (!pud)
  87                 goto free_p4d;
  88
  89         efi_mm.pgd = efi_pgd;
  90         mm_init_cpumask(&efi_mm);
  91         init_new_context(NULL, &efi_mm);
  92
  93         return 0;
  94
  95 free_p4d:
  96         if (pgtable_l5_enabled())
  97                 free_page((unsigned long)pgd_page_vaddr(*pgd));
  98 free_pgd:
  99         free_pages((unsigned long)efi_pgd, PGD_ALLOCATION_ORDER);
 100 fail:
 101         return -ENOMEM;
 102 }
 103
 104 /*
 105  * Add low kernel mappings for passing arguments to EFI functions.
 106  */
 107 void efi_sync_low_kernel_mappings(void)
 108 {
 109         unsigned num_entries;
 110         pgd_t *pgd_k, *pgd_efi;
 111         p4d_t *p4d_k, *p4d_efi;
 112         pud_t *pud_k, *pud_efi;
 113         pgd_t *efi_pgd = efi_mm.pgd;
 114
 115         pgd_efi = efi_pgd + pgd_index(PAGE_OFFSET);
 116         pgd_k = pgd_offset_k(PAGE_OFFSET);
 117
 118         num_entries = pgd_index(EFI_VA_END) - pgd_index(PAGE_OFFSET);
 119         memcpy(pgd_efi, pgd_k, sizeof(pgd_t) * num_entries);
 120
 121         pgd_efi = efi_pgd + pgd_index(EFI_VA_END);
 122         pgd_k = pgd_offset_k(EFI_VA_END);
 123         p4d_efi = p4d_offset(pgd_efi, 0);
 124         p4d_k = p4d_offset(pgd_k, 0);
 125
 126         num_entries = p4d_index(EFI_VA_END);
 127         memcpy(p4d_efi, p4d_k, sizeof(p4d_t) * num_entries);
 128
 129         /*
 130          * We share all the PUD entries apart from those that map the
 131          * EFI regions. Copy around them.
 132          */
 133         BUILD_BUG_ON((EFI_VA_START & ~PUD_MASK) != 0);
 134         BUILD_BUG_ON((EFI_VA_END & ~PUD_MASK) != 0);
 135
 136         p4d_efi = p4d_offset(pgd_efi, EFI_VA_END);
 137         p4d_k = p4d_offset(pgd_k, EFI_VA_END);
 138         pud_efi = pud_offset(p4d_efi, 0);
 139         pud_k = pud_offset(p4d_k, 0);
 140
 141         num_entries = pud_index(EFI_VA_END);
 142         memcpy(pud_efi, pud_k, sizeof(pud_t) * num_entries);
 143
 144         pud_efi = pud_offset(p4d_efi, EFI_VA_START);
 145         pud_k = pud_offset(p4d_k, EFI_VA_START);
 146
 147         num_entries = PTRS_PER_PUD - pud_index(EFI_VA_START);
 148         memcpy(pud_efi, pud_k, sizeof(pud_t) * num_entries);
 149 }
 150
 151 /*
 152  * Wrapper for slow_virt_to_phys() that handles NULL addresses.
 153  */
 154 static inline phys_addr_t
 155 virt_to_phys_or_null_size(void *va, unsigned long size)
 156 {
 157         phys_addr_t pa;
 158
 159         if (!va)
 160                 return 0;
 161
 162         if (virt_addr_valid(va))
 163                 return virt_to_phys(va);
 164
 165         pa = slow_virt_to_phys(va);
 166
 167         /* check if the object crosses a page boundary */
 168         if (WARN_ON((pa ^ (pa + size - 1)) & PAGE_MASK))
 169                 return 0;
 170
 171         return pa;
 172 }
 173
 174 #define virt_to_phys_or_null(addr)                              \
 175         virt_to_phys_or_null_size((addr), sizeof(*(addr)))
 176
 177 int __init efi_setup_page_tables(unsigned long pa_memmap, unsigned num_pages)
 178 {
 179         extern const u8 __efi64_thunk_ret_tramp[];
 180         unsigned long pfn, text, pf, rodata, tramp;
 181         struct page *page;
 182         unsigned npages;
 183         pgd_t *pgd = efi_mm.pgd;
 184
 185         /*
 186          * It can happen that the physical address of new_memmap lands in memory
 187          * which is not mapped in the EFI page table. Therefore we need to go
 188          * and ident-map those pages containing the map before calling
 189          * phys_efi_set_virtual_address_map().
 190          */
 191         pfn = pa_memmap >> PAGE_SHIFT;
 192         pf = _PAGE_NX | _PAGE_RW | _PAGE_ENC;
 193         if (kernel_map_pages_in_pgd(pgd, pfn, pa_memmap, num_pages, pf)) {
 194                 pr_err("Error ident-mapping new memmap (0x%lx)!\n", pa_memmap);
 195                 return 1;
 196         }
 197
 198         /*
 199          * Certain firmware versions are way too sentimental and still believe
 200          * they are exclusive and unquestionable owners of the first physical page,
 201          * even though they explicitly mark it as EFI_CONVENTIONAL_MEMORY
 202          * (but then write-access it later during SetVirtualAddressMap()).
 203          *
 204          * Create a 1:1 mapping for this page, to avoid triple faults during early
 205          * boot with such firmware. We are free to hand this page to the BIOS,
 206          * as trim_bios_range() will reserve the first page and isolate it away
 207          * from memory allocators anyway.
 208          */
 209         if (kernel_map_pages_in_pgd(pgd, 0x0, 0x0, 1, pf)) {
 210                 pr_err("Failed to create 1:1 mapping for the first page!\n");
 211                 return 1;
 212         }
 213
 214         /*
 215          * When SEV-ES is active, the GHCB as set by the kernel will be used
 216          * by firmware. Create a 1:1 unencrypted mapping for each GHCB.
 217          */
 218         if (sev_es_efi_map_ghcbs(pgd)) {
 219                 pr_err("Failed to create 1:1 mapping for the GHCBs!\n");
 220                 return 1;
 221         }
 222
 223         /*
 224          * When making calls to the firmware everything needs to be 1:1
 225          * mapped and addressable with 32-bit pointers. Map the kernel
 226          * text and allocate a new stack because we can't rely on the
 227          * stack pointer being < 4GB.
 228          */
 229         if (!efi_is_mixed())
 230                 return 0;
 231
 232         page = alloc_page(GFP_KERNEL|__GFP_DMA32);
 233         if (!page) {
 234                 pr_err("Unable to allocate EFI runtime stack < 4GB\n");
 235                 return 1;
 236         }
 237
 238         efi_mixed_mode_stack_pa = page_to_phys(page + 1); /* stack grows down */
 239
 240         npages = (_etext - _text) >> PAGE_SHIFT;
 241         text = __pa(_text);
 242
 243         if (kernel_unmap_pages_in_pgd(pgd, text, npages)) {
 244                 pr_err("Failed to unmap kernel text 1:1 mapping\n");
 245                 return 1;
 246         }
 247
 248         npages = (__end_rodata - __start_rodata) >> PAGE_SHIFT;
 249         rodata = __pa(__start_rodata);
 250         pfn = rodata >> PAGE_SHIFT;
 251
 252         pf = _PAGE_NX | _PAGE_ENC;
 253         if (kernel_map_pages_in_pgd(pgd, pfn, rodata, npages, pf)) {
 254                 pr_err("Failed to map kernel rodata 1:1\n");
 255                 return 1;
 256         }
 257
 258         tramp = __pa(__efi64_thunk_ret_tramp);
 259         pfn = tramp >> PAGE_SHIFT;
 260
 261         pf = _PAGE_ENC;
 262         if (kernel_map_pages_in_pgd(pgd, pfn, tramp, 1, pf)) {
 263                 pr_err("Failed to map mixed mode return trampoline\n");
 264                 return 1;
 265         }
 266
 267         return 0;
 268 }
 269
 270 static void __init __map_region(efi_memory_desc_t *md, u64 va)
 271 {
 272         unsigned long flags = _PAGE_RW;
 273         unsigned long pfn;
 274         pgd_t *pgd = efi_mm.pgd;
 275
 276         /*
 277          * EFI_RUNTIME_SERVICES_CODE regions typically cover PE/COFF
 278          * executable images in memory that consist of both R-X and
 279          * RW- sections, so we cannot apply read-only or non-exec
 280          * permissions just yet. However, modern EFI systems provide
 281          * a memory attributes table that describes those sections
 282          * with the appropriate restricted permissions, which are
 283          * applied in efi_runtime_update_mappings() below. All other
 284          * regions can be mapped non-executable at this point, with
 285          * the exception of boot services code regions, but those will
 286          * be unmapped again entirely in efi_free_boot_services().
 287          */
 288         if (md->type != EFI_BOOT_SERVICES_CODE &&
 289             md->type != EFI_RUNTIME_SERVICES_CODE)
 290                 flags |= _PAGE_NX;
 291
 292         if (!(md->attribute & EFI_MEMORY_WB))
 293                 flags |= _PAGE_PCD;
 294
 295         if (cc_platform_has(CC_ATTR_GUEST_MEM_ENCRYPT) &&
 296             md->type != EFI_MEMORY_MAPPED_IO)
 297                 flags |= _PAGE_ENC;
 298
 299         pfn = md->phys_addr >> PAGE_SHIFT;
 300         if (kernel_map_pages_in_pgd(pgd, pfn, va, md->num_pages, flags))
 301                 pr_warn("Error mapping PA 0x%llx -> VA 0x%llx!\n",
 302                            md->phys_addr, va);
 303 }
 304
 305 void __init efi_map_region(efi_memory_desc_t *md)
 306 {
 307         unsigned long size = md->num_pages << PAGE_SHIFT;
 308         u64 pa = md->phys_addr;
 309
 310         /*
 311          * Make sure the 1:1 mappings are present as a catch-all for b0rked
 312          * firmware which doesn't update all internal pointers after switching
 313          * to virtual mode and would otherwise crap on us.
 314          */
 315         __map_region(md, md->phys_addr);
 316
 317         /*
 318          * Enforce the 1:1 mapping as the default virtual address when
 319          * booting in EFI mixed mode, because even though we may be
 320          * running a 64-bit kernel, the firmware may only be 32-bit.
 321          */
 322         if (efi_is_mixed()) {
 323                 md->virt_addr = md->phys_addr;
 324                 return;
 325         }
 326
 327         efi_va -= size;
 328
 329         /* Is PA 2M-aligned? */
 330         if (!(pa & (PMD_SIZE - 1))) {
 331                 efi_va &= PMD_MASK;
 332         } else {
 333                 u64 pa_offset = pa & (PMD_SIZE - 1);
 334                 u64 prev_va = efi_va;
 335
 336                 /* get us the same offset within this 2M page */
 337                 efi_va = (efi_va & PMD_MASK) + pa_offset;
 338
 339                 if (efi_va > prev_va)
 340                         efi_va -= PMD_SIZE;
 341         }
 342
 343         if (efi_va < EFI_VA_END) {
 344                 pr_warn(FW_WARN "VA address range overflow!\n");
 345                 return;
 346         }
 347
 348         /* Do the VA map */
 349         __map_region(md, efi_va);
 350         md->virt_addr = efi_va;
 351 }
 352
 353 /*
 354  * kexec kernel will use efi_map_region_fixed to map efi runtime memory ranges.
 355  * md->virt_addr is the original virtual address which had been mapped in kexec
 356  * 1st kernel.
 357  */
 358 void __init efi_map_region_fixed(efi_memory_desc_t *md)
 359 {
 360         __map_region(md, md->phys_addr);
 361         __map_region(md, md->virt_addr);
 362 }
 363
 364 void __init parse_efi_setup(u64 phys_addr, u32 data_len)
 365 {
 366         efi_setup = phys_addr + sizeof(struct setup_data);
 367 }
 368
 369 static int __init efi_update_mappings(efi_memory_desc_t *md, unsigned long pf)
 370 {
 371         unsigned long pfn;
 372         pgd_t *pgd = efi_mm.pgd;
 373         int err1, err2;
 374
 375         /* Update the 1:1 mapping */
 376         pfn = md->phys_addr >> PAGE_SHIFT;
 377         err1 = kernel_map_pages_in_pgd(pgd, pfn, md->phys_addr, md->num_pages, pf);
 378         if (err1) {
 379                 pr_err("Error while updating 1:1 mapping PA 0x%llx -> VA 0x%llx!\n",
 380                            md->phys_addr, md->virt_addr);
 381         }
 382
 383         err2 = kernel_map_pages_in_pgd(pgd, pfn, md->virt_addr, md->num_pages, pf);
 384         if (err2) {
 385                 pr_err("Error while updating VA mapping PA 0x%llx -> VA 0x%llx!\n",
 386                            md->phys_addr, md->virt_addr);
 387         }
 388
 389         return err1 || err2;
 390 }
 391
 392 static int __init efi_update_mem_attr(struct mm_struct *mm, efi_memory_desc_t *md,
 393                                       bool has_ibt)
 394 {
 395         unsigned long pf = 0;
 396
 397         if (md->attribute & EFI_MEMORY_XP)
 398                 pf |= _PAGE_NX;
 399
 400         if (!(md->attribute & EFI_MEMORY_RO))
 401                 pf |= _PAGE_RW;
 402
 403         if (cc_platform_has(CC_ATTR_GUEST_MEM_ENCRYPT))
 404                 pf |= _PAGE_ENC;
 405
 406         return efi_update_mappings(md, pf);
 407 }
 408
 409 void __init efi_runtime_update_mappings(void)
 410 {
 411         efi_memory_desc_t *md;
 412
 413         /*
 414          * Use the EFI Memory Attribute Table for mapping permissions if it
 415          * exists, since it is intended to supersede EFI_PROPERTIES_TABLE.
 416          */
 417         if (efi_enabled(EFI_MEM_ATTR)) {
 418                 efi_memattr_apply_permissions(NULL, efi_update_mem_attr);
 419                 return;
 420         }
 421
 422         /*
 423          * EFI_MEMORY_ATTRIBUTES_TABLE is intended to replace
 424          * EFI_PROPERTIES_TABLE. So, use EFI_PROPERTIES_TABLE to update
 425          * permissions only if EFI_MEMORY_ATTRIBUTES_TABLE is not
 426          * published by the firmware. Even if we find a buggy implementation of
 427          * EFI_MEMORY_ATTRIBUTES_TABLE, don't fall back to
 428          * EFI_PROPERTIES_TABLE, because of the same reason.
 429          */
 430
 431         if (!efi_enabled(EFI_NX_PE_DATA))
 432                 return;
 433
 434         for_each_efi_memory_desc(md) {
 435                 unsigned long pf = 0;
 436
 437                 if (!(md->attribute & EFI_MEMORY_RUNTIME))
 438                         continue;
 439
 440                 if (!(md->attribute & EFI_MEMORY_WB))
 441                         pf |= _PAGE_PCD;
 442
 443                 if ((md->attribute & EFI_MEMORY_XP) ||
 444                         (md->type == EFI_RUNTIME_SERVICES_DATA))
 445                         pf |= _PAGE_NX;
 446
 447                 if (!(md->attribute & EFI_MEMORY_RO) &&
 448                         (md->type != EFI_RUNTIME_SERVICES_CODE))
 449                         pf |= _PAGE_RW;
 450
 451                 if (cc_platform_has(CC_ATTR_GUEST_MEM_ENCRYPT))
 452                         pf |= _PAGE_ENC;
 453
 454                 efi_update_mappings(md, pf);
 455         }
 456 }
 457
 458 void __init efi_dump_pagetable(void)
 459 {
 460 #ifdef CONFIG_EFI_PGT_DUMP
 461         ptdump_walk_pgd_level(NULL, &efi_mm);
 462 #endif
 463 }
 464
 465 /*
 466  * Makes the calling thread switch to/from efi_mm context. Can be used
 467  * in a kernel thread and user context. Preemption needs to remain disabled
 468  * while the EFI-mm is borrowed. mmgrab()/mmdrop() is not used because the mm
 469  * can not change under us.
 470  * It should be ensured that there are no concurrent calls to this function.
 471  */
 472 void efi_enter_mm(void)
 473 {
 474         efi_prev_mm = current->active_mm;
 475         current->active_mm = &efi_mm;
 476         switch_mm(efi_prev_mm, &efi_mm, NULL);
 477 }
 478
 479 void efi_leave_mm(void)
 480 {
 481         current->active_mm = efi_prev_mm;
 482         switch_mm(&efi_mm, efi_prev_mm, NULL);
 483 }
 484
 485 static DEFINE_SPINLOCK(efi_runtime_lock);
 486
 487 /*
 488  * DS and ES contain user values.  We need to save them.
 489  * The 32-bit EFI code needs a valid DS, ES, and SS.  There's no
 490  * need to save the old SS: __KERNEL_DS is always acceptable.
 491  */
 492 #define __efi_thunk(func, ...)                                          \
 493 ({                                                                      \
 494         unsigned short __ds, __es;                                      \
 495         efi_status_t ____s;                                             \
 496                                                                         \
 497         savesegment(ds, __ds);                                          \
 498         savesegment(es, __es);                                          \
 499                                                                         \
 500         loadsegment(ss, __KERNEL_DS);                                   \
 501         loadsegment(ds, __KERNEL_DS);                                   \
 502         loadsegment(es, __KERNEL_DS);                                   \
 503                                                                         \
 504         ____s = efi64_thunk(efi.runtime->mixed_mode.func, __VA_ARGS__); \
 505                                                                         \
 506         loadsegment(ds, __ds);                                          \
 507         loadsegment(es, __es);                                          \
 508                                                                         \
 509         ____s ^= (____s & BIT(31)) | (____s & BIT_ULL(31)) << 32;       \
 510         ____s;                                                          \
 511 })
 512
 513 /*
 514  * Switch to the EFI page tables early so that we can access the 1:1
 515  * runtime services mappings which are not mapped in any other page
 516  * tables.
 517  *
 518  * Also, disable interrupts because the IDT points to 64-bit handlers,
 519  * which aren't going to function correctly when we switch to 32-bit.
 520  */
 521 #define efi_thunk(func...)                                              \
 522 ({                                                                      \
 523         efi_status_t __s;                                               \
 524                                                                         \
 525         arch_efi_call_virt_setup();                                     \
 526                                                                         \
 527         __s = __efi_thunk(func);                                        \
 528                                                                         \
 529         arch_efi_call_virt_teardown();                                  \
 530                                                                         \
 531         __s;                                                            \
 532 })
 533
 534 static efi_status_t __init __no_sanitize_address
 535 efi_thunk_set_virtual_address_map(unsigned long memory_map_size,
 536                                   unsigned long descriptor_size,
 537                                   u32 descriptor_version,
 538                                   efi_memory_desc_t *virtual_map)
 539 {
 540         efi_status_t status;
 541         unsigned long flags;
 542
 543         efi_sync_low_kernel_mappings();
 544         local_irq_save(flags);
 545
 546         efi_enter_mm();
 547
 548         status = __efi_thunk(set_virtual_address_map, memory_map_size,
 549                              descriptor_size, descriptor_version, virtual_map);
 550
 551         efi_leave_mm();
 552         local_irq_restore(flags);
 553
 554         return status;
 555 }
 556
 557 static efi_status_t efi_thunk_get_time(efi_time_t *tm, efi_time_cap_t *tc)
 558 {
 559         return EFI_UNSUPPORTED;
 560 }
 561
 562 static efi_status_t efi_thunk_set_time(efi_time_t *tm)
 563 {
 564         return EFI_UNSUPPORTED;
 565 }
 566
 567 static efi_status_t
 568 efi_thunk_get_wakeup_time(efi_bool_t *enabled, efi_bool_t *pending,
 569                           efi_time_t *tm)
 570 {
 571         return EFI_UNSUPPORTED;
 572 }
 573
 574 static efi_status_t
 575 efi_thunk_set_wakeup_time(efi_bool_t enabled, efi_time_t *tm)
 576 {
 577         return EFI_UNSUPPORTED;
 578 }
 579
 580 static unsigned long efi_name_size(efi_char16_t *name)
 581 {
 582         return ucs2_strsize(name, EFI_VAR_NAME_LEN) + 1;
 583 }
 584
 585 static efi_status_t
 586 efi_thunk_get_variable(efi_char16_t *name, efi_guid_t *vendor,
 587                        u32 *attr, unsigned long *data_size, void *data)
 588 {
 589         u8 buf[24] __aligned(8);
 590         efi_guid_t *vnd = PTR_ALIGN((efi_guid_t *)buf, sizeof(*vnd));
 591         efi_status_t status;
 592         u32 phys_name, phys_vendor, phys_attr;
 593         u32 phys_data_size, phys_data;
 594         unsigned long flags;
 595
 596         spin_lock_irqsave(&efi_runtime_lock, flags);
 597
 598         *vnd = *vendor;
 599
 600         phys_data_size = virt_to_phys_or_null(data_size);
 601         phys_vendor = virt_to_phys_or_null(vnd);
 602         phys_name = virt_to_phys_or_null_size(name, efi_name_size(name));
 603         phys_attr = virt_to_phys_or_null(attr);
 604         phys_data = virt_to_phys_or_null_size(data, *data_size);
 605
 606         if (!phys_name || (data && !phys_data))
 607                 status = EFI_INVALID_PARAMETER;
 608         else
 609                 status = efi_thunk(get_variable, phys_name, phys_vendor,
 610                                    phys_attr, phys_data_size, phys_data);
 611
 612         spin_unlock_irqrestore(&efi_runtime_lock, flags);
 613
 614         return status;
 615 }
 616
 617 static efi_status_t
 618 efi_thunk_set_variable(efi_char16_t *name, efi_guid_t *vendor,
 619                        u32 attr, unsigned long data_size, void *data)
 620 {
 621         u8 buf[24] __aligned(8);
 622         efi_guid_t *vnd = PTR_ALIGN((efi_guid_t *)buf, sizeof(*vnd));
 623         u32 phys_name, phys_vendor, phys_data;
 624         efi_status_t status;
 625         unsigned long flags;
 626
 627         spin_lock_irqsave(&efi_runtime_lock, flags);
 628
 629         *vnd = *vendor;
 630
 631         phys_name = virt_to_phys_or_null_size(name, efi_name_size(name));
 632         phys_vendor = virt_to_phys_or_null(vnd);
 633         phys_data = virt_to_phys_or_null_size(data, data_size);
 634
 635         if (!phys_name || (data && !phys_data))
 636                 status = EFI_INVALID_PARAMETER;
 637         else
 638                 status = efi_thunk(set_variable, phys_name, phys_vendor,
 639                                    attr, data_size, phys_data);
 640
 641         spin_unlock_irqrestore(&efi_runtime_lock, flags);
 642
 643         return status;
 644 }
 645
 646 static efi_status_t
 647 efi_thunk_set_variable_nonblocking(efi_char16_t *name, efi_guid_t *vendor,
 648                                    u32 attr, unsigned long data_size,
 649                                    void *data)
 650 {
 651         u8 buf[24] __aligned(8);
 652         efi_guid_t *vnd = PTR_ALIGN((efi_guid_t *)buf, sizeof(*vnd));
 653         u32 phys_name, phys_vendor, phys_data;
 654         efi_status_t status;
 655         unsigned long flags;
 656
 657         if (!spin_trylock_irqsave(&efi_runtime_lock, flags))
 658                 return EFI_NOT_READY;
 659
 660         *vnd = *vendor;
 661
 662         phys_name = virt_to_phys_or_null_size(name, efi_name_size(name));
 663         phys_vendor = virt_to_phys_or_null(vnd);
 664         phys_data = virt_to_phys_or_null_size(data, data_size);
 665
 666         if (!phys_name || (data && !phys_data))
 667                 status = EFI_INVALID_PARAMETER;
 668         else
 669                 status = efi_thunk(set_variable, phys_name, phys_vendor,
 670                                    attr, data_size, phys_data);
 671
 672         spin_unlock_irqrestore(&efi_runtime_lock, flags);
 673
 674         return status;
 675 }
 676
 677 static efi_status_t
 678 efi_thunk_get_next_variable(unsigned long *name_size,
 679                             efi_char16_t *name,
 680                             efi_guid_t *vendor)
 681 {
 682         u8 buf[24] __aligned(8);
 683         efi_guid_t *vnd = PTR_ALIGN((efi_guid_t *)buf, sizeof(*vnd));
 684         efi_status_t status;
 685         u32 phys_name_size, phys_name, phys_vendor;
 686         unsigned long flags;
 687
 688         spin_lock_irqsave(&efi_runtime_lock, flags);
 689
 690         *vnd = *vendor;
 691
 692         phys_name_size = virt_to_phys_or_null(name_size);
 693         phys_vendor = virt_to_phys_or_null(vnd);
 694         phys_name = virt_to_phys_or_null_size(name, *name_size);
 695
 696         if (!phys_name)
 697                 status = EFI_INVALID_PARAMETER;
 698         else
 699                 status = efi_thunk(get_next_variable, phys_name_size,
 700                                    phys_name, phys_vendor);
 701
 702         spin_unlock_irqrestore(&efi_runtime_lock, flags);
 703
 704         *vendor = *vnd;
 705         return status;
 706 }
 707
 708 static efi_status_t
 709 efi_thunk_get_next_high_mono_count(u32 *count)
 710 {
 711         return EFI_UNSUPPORTED;
 712 }
 713
 714 static void
 715 efi_thunk_reset_system(int reset_type, efi_status_t status,
 716                        unsigned long data_size, efi_char16_t *data)
 717 {
 718         u32 phys_data;
 719         unsigned long flags;
 720
 721         spin_lock_irqsave(&efi_runtime_lock, flags);
 722
 723         phys_data = virt_to_phys_or_null_size(data, data_size);
 724
 725         efi_thunk(reset_system, reset_type, status, data_size, phys_data);
 726
 727         spin_unlock_irqrestore(&efi_runtime_lock, flags);
 728 }
 729
 730 static efi_status_t
 731 efi_thunk_update_capsule(efi_capsule_header_t **capsules,
 732                          unsigned long count, unsigned long sg_list)
 733 {
 734         /*
 735          * To properly support this function we would need to repackage
 736          * 'capsules' because the firmware doesn't understand 64-bit
 737          * pointers.
 738          */
 739         return EFI_UNSUPPORTED;
 740 }
 741
 742 static efi_status_t
 743 efi_thunk_query_variable_info(u32 attr, u64 *storage_space,
 744                               u64 *remaining_space,
 745                               u64 *max_variable_size)
 746 {
 747         efi_status_t status;
 748         u32 phys_storage, phys_remaining, phys_max;
 749         unsigned long flags;
 750
 751         if (efi.runtime_version < EFI_2_00_SYSTEM_TABLE_REVISION)
 752                 return EFI_UNSUPPORTED;
 753
 754         spin_lock_irqsave(&efi_runtime_lock, flags);
 755
 756         phys_storage = virt_to_phys_or_null(storage_space);
 757         phys_remaining = virt_to_phys_or_null(remaining_space);
 758         phys_max = virt_to_phys_or_null(max_variable_size);
 759
 760         status = efi_thunk(query_variable_info, attr, phys_storage,
 761                            phys_remaining, phys_max);
 762
 763         spin_unlock_irqrestore(&efi_runtime_lock, flags);
 764
 765         return status;
 766 }
 767
 768 static efi_status_t
 769 efi_thunk_query_variable_info_nonblocking(u32 attr, u64 *storage_space,
 770                                           u64 *remaining_space,
 771                                           u64 *max_variable_size)
 772 {
 773         efi_status_t status;
 774         u32 phys_storage, phys_remaining, phys_max;
 775         unsigned long flags;
 776
 777         if (efi.runtime_version < EFI_2_00_SYSTEM_TABLE_REVISION)
 778                 return EFI_UNSUPPORTED;
 779
 780         if (!spin_trylock_irqsave(&efi_runtime_lock, flags))
 781                 return EFI_NOT_READY;
 782
 783         phys_storage = virt_to_phys_or_null(storage_space);
 784         phys_remaining = virt_to_phys_or_null(remaining_space);
 785         phys_max = virt_to_phys_or_null(max_variable_size);
 786
 787         status = efi_thunk(query_variable_info, attr, phys_storage,
 788                            phys_remaining, phys_max);
 789
 790         spin_unlock_irqrestore(&efi_runtime_lock, flags);
 791
 792         return status;
 793 }
 794
 795 static efi_status_t
 796 efi_thunk_query_capsule_caps(efi_capsule_header_t **capsules,
 797                              unsigned long count, u64 *max_size,
 798                              int *reset_type)
 799 {
 800         /*
 801          * To properly support this function we would need to repackage
 802          * 'capsules' because the firmware doesn't understand 64-bit
 803          * pointers.
 804          */
 805         return EFI_UNSUPPORTED;
 806 }
 807
 808 void __init efi_thunk_runtime_setup(void)
 809 {
 810         if (!IS_ENABLED(CONFIG_EFI_MIXED))
 811                 return;
 812
 813         efi.get_time = efi_thunk_get_time;
 814         efi.set_time = efi_thunk_set_time;
 815         efi.get_wakeup_time = efi_thunk_get_wakeup_time;
 816         efi.set_wakeup_time = efi_thunk_set_wakeup_time;
 817         efi.get_variable = efi_thunk_get_variable;
 818         efi.get_next_variable = efi_thunk_get_next_variable;
 819         efi.set_variable = efi_thunk_set_variable;
 820         efi.set_variable_nonblocking = efi_thunk_set_variable_nonblocking;
 821         efi.get_next_high_mono_count = efi_thunk_get_next_high_mono_count;
 822         efi.reset_system = efi_thunk_reset_system;
 823         efi.query_variable_info = efi_thunk_query_variable_info;
 824         efi.query_variable_info_nonblocking = efi_thunk_query_variable_info_nonblocking;
 825         efi.update_capsule = efi_thunk_update_capsule;
 826         efi.query_capsule_caps = efi_thunk_query_capsule_caps;
 827 }
 828
 829 efi_status_t __init __no_sanitize_address
 830 efi_set_virtual_address_map(unsigned long memory_map_size,
 831                             unsigned long descriptor_size,
 832                             u32 descriptor_version,
 833                             efi_memory_desc_t *virtual_map,
 834                             unsigned long systab_phys)
 835 {
 836         const efi_system_table_t *systab = (efi_system_table_t *)systab_phys;
 837         efi_status_t status;
 838         unsigned long flags;
 839
 840         if (efi_is_mixed())
 841                 return efi_thunk_set_virtual_address_map(memory_map_size,
 842                                                          descriptor_size,
 843                                                          descriptor_version,
 844                                                          virtual_map);
 845         efi_enter_mm();
 846
 847         efi_fpu_begin();
 848
 849         /* Disable interrupts around EFI calls: */
 850         local_irq_save(flags);
 851         status = efi_call(efi.runtime->set_virtual_address_map,
 852                           memory_map_size, descriptor_size,
 853                           descriptor_version, virtual_map);
 854         local_irq_restore(flags);
 855
 856         efi_fpu_end();
 857
 858         /* grab the virtually remapped EFI runtime services table pointer */
 859         efi.runtime = READ_ONCE(systab->runtime);
 860
 861         efi_leave_mm();
 862
 863         return status;
 864 }