arch/x86/platform/efi/efi_64.c

   1 /*
   2  * x86_64 specific EFI support functions
   3  * Based on Extensible Firmware Interface Specification version 1.0
   4  *
   5  * Copyright (C) 2005-2008 Intel Co.
   6  *      Fenghua Yu <fenghua.yu@intel.com>
   7  *      Bibo Mao <bibo.mao@intel.com>
   8  *      Chandramouli Narayanan <mouli@linux.intel.com>
   9  *      Huang Ying <ying.huang@intel.com>
  10  *
  11  * Code to convert EFI to E820 map has been implemented in elilo bootloader
  12  * based on a EFI patch by Edgar Hucek. Based on the E820 map, the page table
  13  * is setup appropriately for EFI runtime code.
  14  * - mouli 06/14/2007.
  15  *
  16  */
  17
  18 #define pr_fmt(fmt) "efi: " fmt
  19
  20 #include <linux/kernel.h>
  21 #include <linux/init.h>
  22 #include <linux/mm.h>
  23 #include <linux/types.h>
  24 #include <linux/spinlock.h>
  25 #include <linux/bootmem.h>
  26 #include <linux/ioport.h>
  27 #include <linux/init.h>
  28 #include <linux/mc146818rtc.h>
  29 #include <linux/efi.h>
  30 #include <linux/uaccess.h>
  31 #include <linux/io.h>
  32 #include <linux/reboot.h>
  33 #include <linux/slab.h>
  34 #include <linux/ucs2_string.h>
  35
  36 #include <asm/setup.h>
  37 #include <asm/page.h>
  38 #include <asm/e820/api.h>
  39 #include <asm/pgtable.h>
  40 #include <asm/tlbflush.h>
  41 #include <asm/proto.h>
  42 #include <asm/efi.h>
  43 #include <asm/cacheflush.h>
  44 #include <asm/fixmap.h>
  45 #include <asm/realmode.h>
  46 #include <asm/time.h>
  47 #include <asm/pgalloc.h>
  48
  49 /*
  50  * We allocate runtime services regions top-down, starting from -4G, i.e.
  51  * 0xffff_ffff_0000_0000 and limit EFI VA mapping space to 64G.
  52  */
  53 static u64 efi_va = EFI_VA_START;
  54
  55 struct efi_scratch efi_scratch;
  56
  57 static void __init early_code_mapping_set_exec(int executable)
  58 {
  59         efi_memory_desc_t *md;
  60
  61         if (!(__supported_pte_mask & _PAGE_NX))
  62                 return;
  63
  64         /* Make EFI service code area executable */
  65         for_each_efi_memory_desc(md) {
  66                 if (md->type == EFI_RUNTIME_SERVICES_CODE ||
  67                     md->type == EFI_BOOT_SERVICES_CODE)
  68                         efi_set_executable(md, executable);
  69         }
  70 }
  71
  72 pgd_t * __init efi_call_phys_prolog(void)
  73 {
  74         unsigned long vaddr, addr_pgd, addr_p4d, addr_pud;
  75         pgd_t *save_pgd, *pgd_k, *pgd_efi;
  76         p4d_t *p4d, *p4d_k, *p4d_efi;
  77         pud_t *pud;
  78
  79         int pgd;
  80         int n_pgds, i, j;
  81
  82         if (!efi_enabled(EFI_OLD_MEMMAP)) {
  83                 save_pgd = (pgd_t *)__read_cr3();
  84                 write_cr3((unsigned long)efi_scratch.efi_pgt);
  85                 goto out;
  86         }
  87
  88         early_code_mapping_set_exec(1);
  89
  90         n_pgds = DIV_ROUND_UP((max_pfn << PAGE_SHIFT), PGDIR_SIZE);
  91         save_pgd = kmalloc_array(n_pgds, sizeof(*save_pgd), GFP_KERNEL);
  92
  93         /*
  94          * Build 1:1 identity mapping for efi=old_map usage. Note that
  95          * PAGE_OFFSET is PGDIR_SIZE aligned when KASLR is disabled, while
  96          * it is PUD_SIZE ALIGNED with KASLR enabled. So for a given physical
  97          * address X, the pud_index(X) != pud_index(__va(X)), we can only copy
  98          * PUD entry of __va(X) to fill in pud entry of X to build 1:1 mapping.
  99          * This means here we can only reuse the PMD tables of the direct mapping.
 100          */
 101         for (pgd = 0; pgd < n_pgds; pgd++) {
 102                 addr_pgd = (unsigned long)(pgd * PGDIR_SIZE);
 103                 vaddr = (unsigned long)__va(pgd * PGDIR_SIZE);
 104                 pgd_efi = pgd_offset_k(addr_pgd);
 105                 save_pgd[pgd] = *pgd_efi;
 106
 107                 p4d = p4d_alloc(&init_mm, pgd_efi, addr_pgd);
 108                 if (!p4d) {
 109                         pr_err("Failed to allocate p4d table!\n");
 110                         goto out;
 111                 }
 112
 113                 for (i = 0; i < PTRS_PER_P4D; i++) {
 114                         addr_p4d = addr_pgd + i * P4D_SIZE;
 115                         p4d_efi = p4d + p4d_index(addr_p4d);
 116
 117                         pud = pud_alloc(&init_mm, p4d_efi, addr_p4d);
 118                         if (!pud) {
 119                                 pr_err("Failed to allocate pud table!\n");
 120                                 goto out;
 121                         }
 122
 123                         for (j = 0; j < PTRS_PER_PUD; j++) {
 124                                 addr_pud = addr_p4d + j * PUD_SIZE;
 125
 126                                 if (addr_pud > (max_pfn << PAGE_SHIFT))
 127                                         break;
 128
 129                                 vaddr = (unsigned long)__va(addr_pud);
 130
 131                                 pgd_k = pgd_offset_k(vaddr);
 132                                 p4d_k = p4d_offset(pgd_k, vaddr);
 133                                 pud[j] = *pud_offset(p4d_k, vaddr);
 134                         }
 135                 }
 136         }
 137 out:
 138         __flush_tlb_all();
 139
 140         return save_pgd;
 141 }
 142
 143 void __init efi_call_phys_epilog(pgd_t *save_pgd)
 144 {
 145         /*
 146          * After the lock is released, the original page table is restored.
 147          */
 148         int pgd_idx, i;
 149         int nr_pgds;
 150         pgd_t *pgd;
 151         p4d_t *p4d;
 152         pud_t *pud;
 153
 154         if (!efi_enabled(EFI_OLD_MEMMAP)) {
 155                 write_cr3((unsigned long)save_pgd);
 156                 __flush_tlb_all();
 157                 return;
 158         }
 159
 160         nr_pgds = DIV_ROUND_UP((max_pfn << PAGE_SHIFT) , PGDIR_SIZE);
 161
 162         for (pgd_idx = 0; pgd_idx < nr_pgds; pgd_idx++) {
 163                 pgd = pgd_offset_k(pgd_idx * PGDIR_SIZE);
 164                 set_pgd(pgd_offset_k(pgd_idx * PGDIR_SIZE), save_pgd[pgd_idx]);
 165
 166                 if (!(pgd_val(*pgd) & _PAGE_PRESENT))
 167                         continue;
 168
 169                 for (i = 0; i < PTRS_PER_P4D; i++) {
 170                         p4d = p4d_offset(pgd,
 171                                          pgd_idx * PGDIR_SIZE + i * P4D_SIZE);
 172
 173                         if (!(p4d_val(*p4d) & _PAGE_PRESENT))
 174                                 continue;
 175
 176                         pud = (pud_t *)p4d_page_vaddr(*p4d);
 177                         pud_free(&init_mm, pud);
 178                 }
 179
 180                 p4d = (p4d_t *)pgd_page_vaddr(*pgd);
 181                 p4d_free(&init_mm, p4d);
 182         }
 183
 184         kfree(save_pgd);
 185
 186         __flush_tlb_all();
 187         early_code_mapping_set_exec(0);
 188 }
 189
 190 static pgd_t *efi_pgd;
 191
 192 /*
 193  * We need our own copy of the higher levels of the page tables
 194  * because we want to avoid inserting EFI region mappings (EFI_VA_END
 195  * to EFI_VA_START) into the standard kernel page tables. Everything
 196  * else can be shared, see efi_sync_low_kernel_mappings().
 197  */
 198 int __init efi_alloc_page_tables(void)
 199 {
 200         pgd_t *pgd;
 201         p4d_t *p4d;
 202         pud_t *pud;
 203         gfp_t gfp_mask;
 204
 205         if (efi_enabled(EFI_OLD_MEMMAP))
 206                 return 0;
 207
 208         gfp_mask = GFP_KERNEL | __GFP_NOTRACK | __GFP_ZERO;
 209         efi_pgd = (pgd_t *)__get_free_page(gfp_mask);
 210         if (!efi_pgd)
 211                 return -ENOMEM;
 212
 213         pgd = efi_pgd + pgd_index(EFI_VA_END);
 214         p4d = p4d_alloc(&init_mm, pgd, EFI_VA_END);
 215         if (!p4d) {
 216                 free_page((unsigned long)efi_pgd);
 217                 return -ENOMEM;
 218         }
 219
 220         pud = pud_alloc(&init_mm, p4d, EFI_VA_END);
 221         if (!pud) {
 222                 if (CONFIG_PGTABLE_LEVELS > 4)
 223                         free_page((unsigned long) pgd_page_vaddr(*pgd));
 224                 free_page((unsigned long)efi_pgd);
 225                 return -ENOMEM;
 226         }
 227
 228         return 0;
 229 }
 230
 231 /*
 232  * Add low kernel mappings for passing arguments to EFI functions.
 233  */
 234 void efi_sync_low_kernel_mappings(void)
 235 {
 236         unsigned num_entries;
 237         pgd_t *pgd_k, *pgd_efi;
 238         p4d_t *p4d_k, *p4d_efi;
 239         pud_t *pud_k, *pud_efi;
 240
 241         if (efi_enabled(EFI_OLD_MEMMAP))
 242                 return;
 243
 244         /*
 245          * We can share all PGD entries apart from the one entry that
 246          * covers the EFI runtime mapping space.
 247          *
 248          * Make sure the EFI runtime region mappings are guaranteed to
 249          * only span a single PGD entry and that the entry also maps
 250          * other important kernel regions.
 251          */
 252         BUILD_BUG_ON(pgd_index(EFI_VA_END) != pgd_index(MODULES_END));
 253         BUILD_BUG_ON((EFI_VA_START & PGDIR_MASK) !=
 254                         (EFI_VA_END & PGDIR_MASK));
 255
 256         pgd_efi = efi_pgd + pgd_index(PAGE_OFFSET);
 257         pgd_k = pgd_offset_k(PAGE_OFFSET);
 258
 259         num_entries = pgd_index(EFI_VA_END) - pgd_index(PAGE_OFFSET);
 260         memcpy(pgd_efi, pgd_k, sizeof(pgd_t) * num_entries);
 261
 262         /*
 263          * As with PGDs, we share all P4D entries apart from the one entry
 264          * that covers the EFI runtime mapping space.
 265          */
 266         BUILD_BUG_ON(p4d_index(EFI_VA_END) != p4d_index(MODULES_END));
 267         BUILD_BUG_ON((EFI_VA_START & P4D_MASK) != (EFI_VA_END & P4D_MASK));
 268
 269         pgd_efi = efi_pgd + pgd_index(EFI_VA_END);
 270         pgd_k = pgd_offset_k(EFI_VA_END);
 271         p4d_efi = p4d_offset(pgd_efi, 0);
 272         p4d_k = p4d_offset(pgd_k, 0);
 273
 274         num_entries = p4d_index(EFI_VA_END);
 275         memcpy(p4d_efi, p4d_k, sizeof(p4d_t) * num_entries);
 276
 277         /*
 278          * We share all the PUD entries apart from those that map the
 279          * EFI regions. Copy around them.
 280          */
 281         BUILD_BUG_ON((EFI_VA_START & ~PUD_MASK) != 0);
 282         BUILD_BUG_ON((EFI_VA_END & ~PUD_MASK) != 0);
 283
 284         p4d_efi = p4d_offset(pgd_efi, EFI_VA_END);
 285         p4d_k = p4d_offset(pgd_k, EFI_VA_END);
 286         pud_efi = pud_offset(p4d_efi, 0);
 287         pud_k = pud_offset(p4d_k, 0);
 288
 289         num_entries = pud_index(EFI_VA_END);
 290         memcpy(pud_efi, pud_k, sizeof(pud_t) * num_entries);
 291
 292         pud_efi = pud_offset(p4d_efi, EFI_VA_START);
 293         pud_k = pud_offset(p4d_k, EFI_VA_START);
 294
 295         num_entries = PTRS_PER_PUD - pud_index(EFI_VA_START);
 296         memcpy(pud_efi, pud_k, sizeof(pud_t) * num_entries);
 297 }
 298
 299 /*
 300  * Wrapper for slow_virt_to_phys() that handles NULL addresses.
 301  */
 302 static inline phys_addr_t
 303 virt_to_phys_or_null_size(void *va, unsigned long size)
 304 {
 305         bool bad_size;
 306
 307         if (!va)
 308                 return 0;
 309
 310         if (virt_addr_valid(va))
 311                 return virt_to_phys(va);
 312
 313         /*
 314          * A fully aligned variable on the stack is guaranteed not to
 315          * cross a page bounary. Try to catch strings on the stack by
 316          * checking that 'size' is a power of two.
 317          */
 318         bad_size = size > PAGE_SIZE || !is_power_of_2(size);
 319
 320         WARN_ON(!IS_ALIGNED((unsigned long)va, size) || bad_size);
 321
 322         return slow_virt_to_phys(va);
 323 }
 324
 325 #define virt_to_phys_or_null(addr)                              \
 326         virt_to_phys_or_null_size((addr), sizeof(*(addr)))
 327
 328 int __init efi_setup_page_tables(unsigned long pa_memmap, unsigned num_pages)
 329 {
 330         unsigned long pfn, text;
 331         struct page *page;
 332         unsigned npages;
 333         pgd_t *pgd;
 334
 335         if (efi_enabled(EFI_OLD_MEMMAP))
 336                 return 0;
 337
 338         efi_scratch.efi_pgt = (pgd_t *)__pa(efi_pgd);
 339         pgd = efi_pgd;
 340
 341         /*
 342          * It can happen that the physical address of new_memmap lands in memory
 343          * which is not mapped in the EFI page table. Therefore we need to go
 344          * and ident-map those pages containing the map before calling
 345          * phys_efi_set_virtual_address_map().
 346          */
 347         pfn = pa_memmap >> PAGE_SHIFT;
 348         if (kernel_map_pages_in_pgd(pgd, pfn, pa_memmap, num_pages, _PAGE_NX | _PAGE_RW)) {
 349                 pr_err("Error ident-mapping new memmap (0x%lx)!\n", pa_memmap);
 350                 return 1;
 351         }
 352
 353         efi_scratch.use_pgd = true;
 354
 355         /*
 356          * Certain firmware versions are way too sentimential and still believe
 357          * they are exclusive and unquestionable owners of the first physical page,
 358          * even though they explicitly mark it as EFI_CONVENTIONAL_MEMORY
 359          * (but then write-access it later during SetVirtualAddressMap()).
 360          *
 361          * Create a 1:1 mapping for this page, to avoid triple faults during early
 362          * boot with such firmware. We are free to hand this page to the BIOS,
 363          * as trim_bios_range() will reserve the first page and isolate it away
 364          * from memory allocators anyway.
 365          */
 366         if (kernel_map_pages_in_pgd(pgd, 0x0, 0x0, 1, _PAGE_RW)) {
 367                 pr_err("Failed to create 1:1 mapping for the first page!\n");
 368                 return 1;
 369         }
 370
 371         /*
 372          * When making calls to the firmware everything needs to be 1:1
 373          * mapped and addressable with 32-bit pointers. Map the kernel
 374          * text and allocate a new stack because we can't rely on the
 375          * stack pointer being < 4GB.
 376          */
 377         if (!IS_ENABLED(CONFIG_EFI_MIXED) || efi_is_native())
 378                 return 0;
 379
 380         page = alloc_page(GFP_KERNEL|__GFP_DMA32);
 381         if (!page)
 382                 panic("Unable to allocate EFI runtime stack < 4GB\n");
 383
 384         efi_scratch.phys_stack = virt_to_phys(page_address(page));
 385         efi_scratch.phys_stack += PAGE_SIZE; /* stack grows down */
 386
 387         npages = (_etext - _text) >> PAGE_SHIFT;
 388         text = __pa(_text);
 389         pfn = text >> PAGE_SHIFT;
 390
 391         if (kernel_map_pages_in_pgd(pgd, pfn, text, npages, _PAGE_RW)) {
 392                 pr_err("Failed to map kernel text 1:1\n");
 393                 return 1;
 394         }
 395
 396         return 0;
 397 }
 398
 399 static void __init __map_region(efi_memory_desc_t *md, u64 va)
 400 {
 401         unsigned long flags = _PAGE_RW;
 402         unsigned long pfn;
 403         pgd_t *pgd = efi_pgd;
 404
 405         if (!(md->attribute & EFI_MEMORY_WB))
 406                 flags |= _PAGE_PCD;
 407
 408         pfn = md->phys_addr >> PAGE_SHIFT;
 409         if (kernel_map_pages_in_pgd(pgd, pfn, va, md->num_pages, flags))
 410                 pr_warn("Error mapping PA 0x%llx -> VA 0x%llx!\n",
 411                            md->phys_addr, va);
 412 }
 413
 414 void __init efi_map_region(efi_memory_desc_t *md)
 415 {
 416         unsigned long size = md->num_pages << PAGE_SHIFT;
 417         u64 pa = md->phys_addr;
 418
 419         if (efi_enabled(EFI_OLD_MEMMAP))
 420                 return old_map_region(md);
 421
 422         /*
 423          * Make sure the 1:1 mappings are present as a catch-all for b0rked
 424          * firmware which doesn't update all internal pointers after switching
 425          * to virtual mode and would otherwise crap on us.
 426          */
 427         __map_region(md, md->phys_addr);
 428
 429         /*
 430          * Enforce the 1:1 mapping as the default virtual address when
 431          * booting in EFI mixed mode, because even though we may be
 432          * running a 64-bit kernel, the firmware may only be 32-bit.
 433          */
 434         if (!efi_is_native () && IS_ENABLED(CONFIG_EFI_MIXED)) {
 435                 md->virt_addr = md->phys_addr;
 436                 return;
 437         }
 438
 439         efi_va -= size;
 440
 441         /* Is PA 2M-aligned? */
 442         if (!(pa & (PMD_SIZE - 1))) {
 443                 efi_va &= PMD_MASK;
 444         } else {
 445                 u64 pa_offset = pa & (PMD_SIZE - 1);
 446                 u64 prev_va = efi_va;
 447
 448                 /* get us the same offset within this 2M page */
 449                 efi_va = (efi_va & PMD_MASK) + pa_offset;
 450
 451                 if (efi_va > prev_va)
 452                         efi_va -= PMD_SIZE;
 453         }
 454
 455         if (efi_va < EFI_VA_END) {
 456                 pr_warn(FW_WARN "VA address range overflow!\n");
 457                 return;
 458         }
 459
 460         /* Do the VA map */
 461         __map_region(md, efi_va);
 462         md->virt_addr = efi_va;
 463 }
 464
 465 /*
 466  * kexec kernel will use efi_map_region_fixed to map efi runtime memory ranges.
 467  * md->virt_addr is the original virtual address which had been mapped in kexec
 468  * 1st kernel.
 469  */
 470 void __init efi_map_region_fixed(efi_memory_desc_t *md)
 471 {
 472         __map_region(md, md->phys_addr);
 473         __map_region(md, md->virt_addr);
 474 }
 475
 476 void __iomem *__init efi_ioremap(unsigned long phys_addr, unsigned long size,
 477                                  u32 type, u64 attribute)
 478 {
 479         unsigned long last_map_pfn;
 480
 481         if (type == EFI_MEMORY_MAPPED_IO)
 482                 return ioremap(phys_addr, size);
 483
 484         last_map_pfn = init_memory_mapping(phys_addr, phys_addr + size);
 485         if ((last_map_pfn << PAGE_SHIFT) < phys_addr + size) {
 486                 unsigned long top = last_map_pfn << PAGE_SHIFT;
 487                 efi_ioremap(top, size - (top - phys_addr), type, attribute);
 488         }
 489
 490         if (!(attribute & EFI_MEMORY_WB))
 491                 efi_memory_uc((u64)(unsigned long)__va(phys_addr), size);
 492
 493         return (void __iomem *)__va(phys_addr);
 494 }
 495
 496 void __init parse_efi_setup(u64 phys_addr, u32 data_len)
 497 {
 498         efi_setup = phys_addr + sizeof(struct setup_data);
 499 }
 500
 501 static int __init efi_update_mappings(efi_memory_desc_t *md, unsigned long pf)
 502 {
 503         unsigned long pfn;
 504         pgd_t *pgd = efi_pgd;
 505         int err1, err2;
 506
 507         /* Update the 1:1 mapping */
 508         pfn = md->phys_addr >> PAGE_SHIFT;
 509         err1 = kernel_map_pages_in_pgd(pgd, pfn, md->phys_addr, md->num_pages, pf);
 510         if (err1) {
 511                 pr_err("Error while updating 1:1 mapping PA 0x%llx -> VA 0x%llx!\n",
 512                            md->phys_addr, md->virt_addr);
 513         }
 514
 515         err2 = kernel_map_pages_in_pgd(pgd, pfn, md->virt_addr, md->num_pages, pf);
 516         if (err2) {
 517                 pr_err("Error while updating VA mapping PA 0x%llx -> VA 0x%llx!\n",
 518                            md->phys_addr, md->virt_addr);
 519         }
 520
 521         return err1 || err2;
 522 }
 523
 524 static int __init efi_update_mem_attr(struct mm_struct *mm, efi_memory_desc_t *md)
 525 {
 526         unsigned long pf = 0;
 527
 528         if (md->attribute & EFI_MEMORY_XP)
 529                 pf |= _PAGE_NX;
 530
 531         if (!(md->attribute & EFI_MEMORY_RO))
 532                 pf |= _PAGE_RW;
 533
 534         return efi_update_mappings(md, pf);
 535 }
 536
 537 void __init efi_runtime_update_mappings(void)
 538 {
 539         efi_memory_desc_t *md;
 540
 541         if (efi_enabled(EFI_OLD_MEMMAP)) {
 542                 if (__supported_pte_mask & _PAGE_NX)
 543                         runtime_code_page_mkexec();
 544                 return;
 545         }
 546
 547         /*
 548          * Use the EFI Memory Attribute Table for mapping permissions if it
 549          * exists, since it is intended to supersede EFI_PROPERTIES_TABLE.
 550          */
 551         if (efi_enabled(EFI_MEM_ATTR)) {
 552                 efi_memattr_apply_permissions(NULL, efi_update_mem_attr);
 553                 return;
 554         }
 555
 556         /*
 557          * EFI_MEMORY_ATTRIBUTES_TABLE is intended to replace
 558          * EFI_PROPERTIES_TABLE. So, use EFI_PROPERTIES_TABLE to update
 559          * permissions only if EFI_MEMORY_ATTRIBUTES_TABLE is not
 560          * published by the firmware. Even if we find a buggy implementation of
 561          * EFI_MEMORY_ATTRIBUTES_TABLE, don't fall back to
 562          * EFI_PROPERTIES_TABLE, because of the same reason.
 563          */
 564
 565         if (!efi_enabled(EFI_NX_PE_DATA))
 566                 return;
 567
 568         for_each_efi_memory_desc(md) {
 569                 unsigned long pf = 0;
 570
 571                 if (!(md->attribute & EFI_MEMORY_RUNTIME))
 572                         continue;
 573
 574                 if (!(md->attribute & EFI_MEMORY_WB))
 575                         pf |= _PAGE_PCD;
 576
 577                 if ((md->attribute & EFI_MEMORY_XP) ||
 578                         (md->type == EFI_RUNTIME_SERVICES_DATA))
 579                         pf |= _PAGE_NX;
 580
 581                 if (!(md->attribute & EFI_MEMORY_RO) &&
 582                         (md->type != EFI_RUNTIME_SERVICES_CODE))
 583                         pf |= _PAGE_RW;
 584
 585                 efi_update_mappings(md, pf);
 586         }
 587 }
 588
 589 void __init efi_dump_pagetable(void)
 590 {
 591 #ifdef CONFIG_EFI_PGT_DUMP
 592         if (efi_enabled(EFI_OLD_MEMMAP))
 593                 ptdump_walk_pgd_level(NULL, swapper_pg_dir);
 594         else
 595                 ptdump_walk_pgd_level(NULL, efi_pgd);
 596 #endif
 597 }
 598
 599 #ifdef CONFIG_EFI_MIXED
 600 extern efi_status_t efi64_thunk(u32, ...);
 601
 602 #define runtime_service32(func)                                          \
 603 ({                                                                       \
 604         u32 table = (u32)(unsigned long)efi.systab;                      \
 605         u32 *rt, *___f;                                                  \
 606                                                                          \
 607         rt = (u32 *)(table + offsetof(efi_system_table_32_t, runtime));  \
 608         ___f = (u32 *)(*rt + offsetof(efi_runtime_services_32_t, func)); \
 609         *___f;                                                           \
 610 })
 611
 612 /*
 613  * Switch to the EFI page tables early so that we can access the 1:1
 614  * runtime services mappings which are not mapped in any other page
 615  * tables. This function must be called before runtime_service32().
 616  *
 617  * Also, disable interrupts because the IDT points to 64-bit handlers,
 618  * which aren't going to function correctly when we switch to 32-bit.
 619  */
 620 #define efi_thunk(f, ...)                                               \
 621 ({                                                                      \
 622         efi_status_t __s;                                               \
 623         unsigned long __flags;                                          \
 624         u32 __func;                                                     \
 625                                                                         \
 626         local_irq_save(__flags);                                        \
 627         arch_efi_call_virt_setup();                                     \
 628                                                                         \
 629         __func = runtime_service32(f);                                  \
 630         __s = efi64_thunk(__func, __VA_ARGS__);                         \
 631                                                                         \
 632         arch_efi_call_virt_teardown();                                  \
 633         local_irq_restore(__flags);                                     \
 634                                                                         \
 635         __s;                                                            \
 636 })
 637
 638 efi_status_t efi_thunk_set_virtual_address_map(
 639         void *phys_set_virtual_address_map,
 640         unsigned long memory_map_size,
 641         unsigned long descriptor_size,
 642         u32 descriptor_version,
 643         efi_memory_desc_t *virtual_map)
 644 {
 645         efi_status_t status;
 646         unsigned long flags;
 647         u32 func;
 648
 649         efi_sync_low_kernel_mappings();
 650         local_irq_save(flags);
 651
 652         efi_scratch.prev_cr3 = __read_cr3();
 653         write_cr3((unsigned long)efi_scratch.efi_pgt);
 654         __flush_tlb_all();
 655
 656         func = (u32)(unsigned long)phys_set_virtual_address_map;
 657         status = efi64_thunk(func, memory_map_size, descriptor_size,
 658                              descriptor_version, virtual_map);
 659
 660         write_cr3(efi_scratch.prev_cr3);
 661         __flush_tlb_all();
 662         local_irq_restore(flags);
 663
 664         return status;
 665 }
 666
 667 static efi_status_t efi_thunk_get_time(efi_time_t *tm, efi_time_cap_t *tc)
 668 {
 669         efi_status_t status;
 670         u32 phys_tm, phys_tc;
 671
 672         spin_lock(&rtc_lock);
 673
 674         phys_tm = virt_to_phys_or_null(tm);
 675         phys_tc = virt_to_phys_or_null(tc);
 676
 677         status = efi_thunk(get_time, phys_tm, phys_tc);
 678
 679         spin_unlock(&rtc_lock);
 680
 681         return status;
 682 }
 683
 684 static efi_status_t efi_thunk_set_time(efi_time_t *tm)
 685 {
 686         efi_status_t status;
 687         u32 phys_tm;
 688
 689         spin_lock(&rtc_lock);
 690
 691         phys_tm = virt_to_phys_or_null(tm);
 692
 693         status = efi_thunk(set_time, phys_tm);
 694
 695         spin_unlock(&rtc_lock);
 696
 697         return status;
 698 }
 699
 700 static efi_status_t
 701 efi_thunk_get_wakeup_time(efi_bool_t *enabled, efi_bool_t *pending,
 702                           efi_time_t *tm)
 703 {
 704         efi_status_t status;
 705         u32 phys_enabled, phys_pending, phys_tm;
 706
 707         spin_lock(&rtc_lock);
 708
 709         phys_enabled = virt_to_phys_or_null(enabled);
 710         phys_pending = virt_to_phys_or_null(pending);
 711         phys_tm = virt_to_phys_or_null(tm);
 712
 713         status = efi_thunk(get_wakeup_time, phys_enabled,
 714                              phys_pending, phys_tm);
 715
 716         spin_unlock(&rtc_lock);
 717
 718         return status;
 719 }
 720
 721 static efi_status_t
 722 efi_thunk_set_wakeup_time(efi_bool_t enabled, efi_time_t *tm)
 723 {
 724         efi_status_t status;
 725         u32 phys_tm;
 726
 727         spin_lock(&rtc_lock);
 728
 729         phys_tm = virt_to_phys_or_null(tm);
 730
 731         status = efi_thunk(set_wakeup_time, enabled, phys_tm);
 732
 733         spin_unlock(&rtc_lock);
 734
 735         return status;
 736 }
 737
 738 static unsigned long efi_name_size(efi_char16_t *name)
 739 {
 740         return ucs2_strsize(name, EFI_VAR_NAME_LEN) + 1;
 741 }
 742
 743 static efi_status_t
 744 efi_thunk_get_variable(efi_char16_t *name, efi_guid_t *vendor,
 745                        u32 *attr, unsigned long *data_size, void *data)
 746 {
 747         efi_status_t status;
 748         u32 phys_name, phys_vendor, phys_attr;
 749         u32 phys_data_size, phys_data;
 750
 751         phys_data_size = virt_to_phys_or_null(data_size);
 752         phys_vendor = virt_to_phys_or_null(vendor);
 753         phys_name = virt_to_phys_or_null_size(name, efi_name_size(name));
 754         phys_attr = virt_to_phys_or_null(attr);
 755         phys_data = virt_to_phys_or_null_size(data, *data_size);
 756
 757         status = efi_thunk(get_variable, phys_name, phys_vendor,
 758                            phys_attr, phys_data_size, phys_data);
 759
 760         return status;
 761 }
 762
 763 static efi_status_t
 764 efi_thunk_set_variable(efi_char16_t *name, efi_guid_t *vendor,
 765                        u32 attr, unsigned long data_size, void *data)
 766 {
 767         u32 phys_name, phys_vendor, phys_data;
 768         efi_status_t status;
 769
 770         phys_name = virt_to_phys_or_null_size(name, efi_name_size(name));
 771         phys_vendor = virt_to_phys_or_null(vendor);
 772         phys_data = virt_to_phys_or_null_size(data, data_size);
 773
 774         /* If data_size is > sizeof(u32) we've got problems */
 775         status = efi_thunk(set_variable, phys_name, phys_vendor,
 776                            attr, data_size, phys_data);
 777
 778         return status;
 779 }
 780
 781 static efi_status_t
 782 efi_thunk_get_next_variable(unsigned long *name_size,
 783                             efi_char16_t *name,
 784                             efi_guid_t *vendor)
 785 {
 786         efi_status_t status;
 787         u32 phys_name_size, phys_name, phys_vendor;
 788
 789         phys_name_size = virt_to_phys_or_null(name_size);
 790         phys_vendor = virt_to_phys_or_null(vendor);
 791         phys_name = virt_to_phys_or_null_size(name, *name_size);
 792
 793         status = efi_thunk(get_next_variable, phys_name_size,
 794                            phys_name, phys_vendor);
 795
 796         return status;
 797 }
 798
 799 static efi_status_t
 800 efi_thunk_get_next_high_mono_count(u32 *count)
 801 {
 802         efi_status_t status;
 803         u32 phys_count;
 804
 805         phys_count = virt_to_phys_or_null(count);
 806         status = efi_thunk(get_next_high_mono_count, phys_count);
 807
 808         return status;
 809 }
 810
 811 static void
 812 efi_thunk_reset_system(int reset_type, efi_status_t status,
 813                        unsigned long data_size, efi_char16_t *data)
 814 {
 815         u32 phys_data;
 816
 817         phys_data = virt_to_phys_or_null_size(data, data_size);
 818
 819         efi_thunk(reset_system, reset_type, status, data_size, phys_data);
 820 }
 821
 822 static efi_status_t
 823 efi_thunk_update_capsule(efi_capsule_header_t **capsules,
 824                          unsigned long count, unsigned long sg_list)
 825 {
 826         /*
 827          * To properly support this function we would need to repackage
 828          * 'capsules' because the firmware doesn't understand 64-bit
 829          * pointers.
 830          */
 831         return EFI_UNSUPPORTED;
 832 }
 833
 834 static efi_status_t
 835 efi_thunk_query_variable_info(u32 attr, u64 *storage_space,
 836                               u64 *remaining_space,
 837                               u64 *max_variable_size)
 838 {
 839         efi_status_t status;
 840         u32 phys_storage, phys_remaining, phys_max;
 841
 842         if (efi.runtime_version < EFI_2_00_SYSTEM_TABLE_REVISION)
 843                 return EFI_UNSUPPORTED;
 844
 845         phys_storage = virt_to_phys_or_null(storage_space);
 846         phys_remaining = virt_to_phys_or_null(remaining_space);
 847         phys_max = virt_to_phys_or_null(max_variable_size);
 848
 849         status = efi_thunk(query_variable_info, attr, phys_storage,
 850                            phys_remaining, phys_max);
 851
 852         return status;
 853 }
 854
 855 static efi_status_t
 856 efi_thunk_query_capsule_caps(efi_capsule_header_t **capsules,
 857                              unsigned long count, u64 *max_size,
 858                              int *reset_type)
 859 {
 860         /*
 861          * To properly support this function we would need to repackage
 862          * 'capsules' because the firmware doesn't understand 64-bit
 863          * pointers.
 864          */
 865         return EFI_UNSUPPORTED;
 866 }
 867
 868 void efi_thunk_runtime_setup(void)
 869 {
 870         efi.get_time = efi_thunk_get_time;
 871         efi.set_time = efi_thunk_set_time;
 872         efi.get_wakeup_time = efi_thunk_get_wakeup_time;
 873         efi.set_wakeup_time = efi_thunk_set_wakeup_time;
 874         efi.get_variable = efi_thunk_get_variable;
 875         efi.get_next_variable = efi_thunk_get_next_variable;
 876         efi.set_variable = efi_thunk_set_variable;
 877         efi.get_next_high_mono_count = efi_thunk_get_next_high_mono_count;
 878         efi.reset_system = efi_thunk_reset_system;
 879         efi.query_variable_info = efi_thunk_query_variable_info;
 880         efi.update_capsule = efi_thunk_update_capsule;
 881         efi.query_capsule_caps = efi_thunk_query_capsule_caps;
 882 }
 883 #endif /* CONFIG_EFI_MIXED */