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/mem_encrypt.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/pgtable.h>
  43 #include <asm/tlbflush.h>
  44 #include <asm/proto.h>
  45 #include <asm/efi.h>
  46 #include <asm/cacheflush.h>
  47 #include <asm/fixmap.h>
  48 #include <asm/realmode.h>
  49 #include <asm/time.h>
  50 #include <asm/pgalloc.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
  58 struct efi_scratch efi_scratch;
  59
  60 static void __init early_code_mapping_set_exec(int executable)
  61 {
  62         efi_memory_desc_t *md;
  63
  64         if (!(__supported_pte_mask & _PAGE_NX))
  65                 return;
  66
  67         /* Make EFI service code area executable */
  68         for_each_efi_memory_desc(md) {
  69                 if (md->type == EFI_RUNTIME_SERVICES_CODE ||
  70                     md->type == EFI_BOOT_SERVICES_CODE)
  71                         efi_set_executable(md, executable);
  72         }
  73 }
  74
  75 pgd_t * __init efi_call_phys_prolog(void)
  76 {
  77         unsigned long vaddr, addr_pgd, addr_p4d, addr_pud;
  78         pgd_t *save_pgd, *pgd_k, *pgd_efi;
  79         p4d_t *p4d, *p4d_k, *p4d_efi;
  80         pud_t *pud;
  81
  82         int pgd;
  83         int n_pgds, i, j;
  84
  85         if (!efi_enabled(EFI_OLD_MEMMAP)) {
  86                 efi_switch_mm(&efi_mm);
  87                 return efi_mm.pgd;
  88         }
  89
  90         early_code_mapping_set_exec(1);
  91
  92         n_pgds = DIV_ROUND_UP((max_pfn << PAGE_SHIFT), PGDIR_SIZE);
  93         save_pgd = kmalloc_array(n_pgds, sizeof(*save_pgd), GFP_KERNEL);
  94         if (!save_pgd)
  95                 return NULL;
  96
  97         /*
  98          * Build 1:1 identity mapping for efi=old_map usage. Note that
  99          * PAGE_OFFSET is PGDIR_SIZE aligned when KASLR is disabled, while
 100          * it is PUD_SIZE ALIGNED with KASLR enabled. So for a given physical
 101          * address X, the pud_index(X) != pud_index(__va(X)), we can only copy
 102          * PUD entry of __va(X) to fill in pud entry of X to build 1:1 mapping.
 103          * This means here we can only reuse the PMD tables of the direct mapping.
 104          */
 105         for (pgd = 0; pgd < n_pgds; pgd++) {
 106                 addr_pgd = (unsigned long)(pgd * PGDIR_SIZE);
 107                 vaddr = (unsigned long)__va(pgd * PGDIR_SIZE);
 108                 pgd_efi = pgd_offset_k(addr_pgd);
 109                 save_pgd[pgd] = *pgd_efi;
 110
 111                 p4d = p4d_alloc(&init_mm, pgd_efi, addr_pgd);
 112                 if (!p4d) {
 113                         pr_err("Failed to allocate p4d table!\n");
 114                         goto out;
 115                 }
 116
 117                 for (i = 0; i < PTRS_PER_P4D; i++) {
 118                         addr_p4d = addr_pgd + i * P4D_SIZE;
 119                         p4d_efi = p4d + p4d_index(addr_p4d);
 120
 121                         pud = pud_alloc(&init_mm, p4d_efi, addr_p4d);
 122                         if (!pud) {
 123                                 pr_err("Failed to allocate pud table!\n");
 124                                 goto out;
 125                         }
 126
 127                         for (j = 0; j < PTRS_PER_PUD; j++) {
 128                                 addr_pud = addr_p4d + j * PUD_SIZE;
 129
 130                                 if (addr_pud > (max_pfn << PAGE_SHIFT))
 131                                         break;
 132
 133                                 vaddr = (unsigned long)__va(addr_pud);
 134
 135                                 pgd_k = pgd_offset_k(vaddr);
 136                                 p4d_k = p4d_offset(pgd_k, vaddr);
 137                                 pud[j] = *pud_offset(p4d_k, vaddr);
 138                         }
 139                 }
 140                 pgd_offset_k(pgd * PGDIR_SIZE)->pgd &= ~_PAGE_NX;
 141         }
 142
 143         __flush_tlb_all();
 144         return save_pgd;
 145 out:
 146         efi_call_phys_epilog(save_pgd);
 147         return NULL;
 148 }
 149
 150 void __init efi_call_phys_epilog(pgd_t *save_pgd)
 151 {
 152         /*
 153          * After the lock is released, the original page table is restored.
 154          */
 155         int pgd_idx, i;
 156         int nr_pgds;
 157         pgd_t *pgd;
 158         p4d_t *p4d;
 159         pud_t *pud;
 160
 161         if (!efi_enabled(EFI_OLD_MEMMAP)) {
 162                 efi_switch_mm(efi_scratch.prev_mm);
 163                 return;
 164         }
 165
 166         nr_pgds = DIV_ROUND_UP((max_pfn << PAGE_SHIFT) , PGDIR_SIZE);
 167
 168         for (pgd_idx = 0; pgd_idx < nr_pgds; pgd_idx++) {
 169                 pgd = pgd_offset_k(pgd_idx * PGDIR_SIZE);
 170                 set_pgd(pgd_offset_k(pgd_idx * PGDIR_SIZE), save_pgd[pgd_idx]);
 171
 172                 if (!pgd_present(*pgd))
 173                         continue;
 174
 175                 for (i = 0; i < PTRS_PER_P4D; i++) {
 176                         p4d = p4d_offset(pgd,
 177                                          pgd_idx * PGDIR_SIZE + i * P4D_SIZE);
 178
 179                         if (!p4d_present(*p4d))
 180                                 continue;
 181
 182                         pud = (pud_t *)p4d_page_vaddr(*p4d);
 183                         pud_free(&init_mm, pud);
 184                 }
 185
 186                 p4d = (p4d_t *)pgd_page_vaddr(*pgd);
 187                 p4d_free(&init_mm, p4d);
 188         }
 189
 190         kfree(save_pgd);
 191
 192         __flush_tlb_all();
 193         early_code_mapping_set_exec(0);
 194 }
 195
 196 EXPORT_SYMBOL_GPL(efi_mm);
 197
 198 /*
 199  * We need our own copy of the higher levels of the page tables
 200  * because we want to avoid inserting EFI region mappings (EFI_VA_END
 201  * to EFI_VA_START) into the standard kernel page tables. Everything
 202  * else can be shared, see efi_sync_low_kernel_mappings().
 203  *
 204  * We don't want the pgd on the pgd_list and cannot use pgd_alloc() for the
 205  * allocation.
 206  */
 207 int __init efi_alloc_page_tables(void)
 208 {
 209         pgd_t *pgd, *efi_pgd;
 210         p4d_t *p4d;
 211         pud_t *pud;
 212         gfp_t gfp_mask;
 213
 214         if (efi_enabled(EFI_OLD_MEMMAP))
 215                 return 0;
 216
 217         gfp_mask = GFP_KERNEL | __GFP_ZERO;
 218         efi_pgd = (pgd_t *)__get_free_pages(gfp_mask, PGD_ALLOCATION_ORDER);
 219         if (!efi_pgd)
 220                 goto fail;
 221
 222         pgd = efi_pgd + pgd_index(EFI_VA_END);
 223         p4d = p4d_alloc(&init_mm, pgd, EFI_VA_END);
 224         if (!p4d)
 225                 goto free_pgd;
 226
 227         pud = pud_alloc(&init_mm, p4d, EFI_VA_END);
 228         if (!pud)
 229                 goto free_p4d;
 230
 231         efi_mm.pgd = efi_pgd;
 232         mm_init_cpumask(&efi_mm);
 233         init_new_context(NULL, &efi_mm);
 234
 235         return 0;
 236
 237 free_p4d:
 238         if (pgtable_l5_enabled())
 239                 free_page((unsigned long)pgd_page_vaddr(*pgd));
 240 free_pgd:
 241         free_pages((unsigned long)efi_pgd, PGD_ALLOCATION_ORDER);
 242 fail:
 243         return -ENOMEM;
 244 }
 245
 246 /*
 247  * Add low kernel mappings for passing arguments to EFI functions.
 248  */
 249 void efi_sync_low_kernel_mappings(void)
 250 {
 251         unsigned num_entries;
 252         pgd_t *pgd_k, *pgd_efi;
 253         p4d_t *p4d_k, *p4d_efi;
 254         pud_t *pud_k, *pud_efi;
 255         pgd_t *efi_pgd = efi_mm.pgd;
 256
 257         if (efi_enabled(EFI_OLD_MEMMAP))
 258                 return;
 259
 260         /*
 261          * We can share all PGD entries apart from the one entry that
 262          * covers the EFI runtime mapping space.
 263          *
 264          * Make sure the EFI runtime region mappings are guaranteed to
 265          * only span a single PGD entry and that the entry also maps
 266          * other important kernel regions.
 267          */
 268         MAYBE_BUILD_BUG_ON(pgd_index(EFI_VA_END) != pgd_index(MODULES_END));
 269         MAYBE_BUILD_BUG_ON((EFI_VA_START & PGDIR_MASK) !=
 270                         (EFI_VA_END & PGDIR_MASK));
 271
 272         pgd_efi = efi_pgd + pgd_index(PAGE_OFFSET);
 273         pgd_k = pgd_offset_k(PAGE_OFFSET);
 274
 275         num_entries = pgd_index(EFI_VA_END) - pgd_index(PAGE_OFFSET);
 276         memcpy(pgd_efi, pgd_k, sizeof(pgd_t) * num_entries);
 277
 278         /*
 279          * As with PGDs, we share all P4D entries apart from the one entry
 280          * that covers the EFI runtime mapping space.
 281          */
 282         BUILD_BUG_ON(p4d_index(EFI_VA_END) != p4d_index(MODULES_END));
 283         BUILD_BUG_ON((EFI_VA_START & P4D_MASK) != (EFI_VA_END & P4D_MASK));
 284
 285         pgd_efi = efi_pgd + pgd_index(EFI_VA_END);
 286         pgd_k = pgd_offset_k(EFI_VA_END);
 287         p4d_efi = p4d_offset(pgd_efi, 0);
 288         p4d_k = p4d_offset(pgd_k, 0);
 289
 290         num_entries = p4d_index(EFI_VA_END);
 291         memcpy(p4d_efi, p4d_k, sizeof(p4d_t) * num_entries);
 292
 293         /*
 294          * We share all the PUD entries apart from those that map the
 295          * EFI regions. Copy around them.
 296          */
 297         BUILD_BUG_ON((EFI_VA_START & ~PUD_MASK) != 0);
 298         BUILD_BUG_ON((EFI_VA_END & ~PUD_MASK) != 0);
 299
 300         p4d_efi = p4d_offset(pgd_efi, EFI_VA_END);
 301         p4d_k = p4d_offset(pgd_k, EFI_VA_END);
 302         pud_efi = pud_offset(p4d_efi, 0);
 303         pud_k = pud_offset(p4d_k, 0);
 304
 305         num_entries = pud_index(EFI_VA_END);
 306         memcpy(pud_efi, pud_k, sizeof(pud_t) * num_entries);
 307
 308         pud_efi = pud_offset(p4d_efi, EFI_VA_START);
 309         pud_k = pud_offset(p4d_k, EFI_VA_START);
 310
 311         num_entries = PTRS_PER_PUD - pud_index(EFI_VA_START);
 312         memcpy(pud_efi, pud_k, sizeof(pud_t) * num_entries);
 313 }
 314
 315 /*
 316  * Wrapper for slow_virt_to_phys() that handles NULL addresses.
 317  */
 318 static inline phys_addr_t
 319 virt_to_phys_or_null_size(void *va, unsigned long size)
 320 {
 321         phys_addr_t pa;
 322
 323         if (!va)
 324                 return 0;
 325
 326         if (virt_addr_valid(va))
 327                 return virt_to_phys(va);
 328
 329         pa = slow_virt_to_phys(va);
 330
 331         /* check if the object crosses a page boundary */
 332         if (WARN_ON((pa ^ (pa + size - 1)) & PAGE_MASK))
 333                 return 0;
 334
 335         return pa;
 336 }
 337
 338 #define virt_to_phys_or_null(addr)                              \
 339         virt_to_phys_or_null_size((addr), sizeof(*(addr)))
 340
 341 int __init efi_setup_page_tables(unsigned long pa_memmap, unsigned num_pages)
 342 {
 343         unsigned long pfn, text, pf;
 344         struct page *page;
 345         unsigned npages;
 346         pgd_t *pgd = efi_mm.pgd;
 347
 348         if (efi_enabled(EFI_OLD_MEMMAP))
 349                 return 0;
 350
 351         /*
 352          * It can happen that the physical address of new_memmap lands in memory
 353          * which is not mapped in the EFI page table. Therefore we need to go
 354          * and ident-map those pages containing the map before calling
 355          * phys_efi_set_virtual_address_map().
 356          */
 357         pfn = pa_memmap >> PAGE_SHIFT;
 358         pf = _PAGE_NX | _PAGE_RW | _PAGE_ENC;
 359         if (kernel_map_pages_in_pgd(pgd, pfn, pa_memmap, num_pages, pf)) {
 360                 pr_err("Error ident-mapping new memmap (0x%lx)!\n", pa_memmap);
 361                 return 1;
 362         }
 363
 364         /*
 365          * Certain firmware versions are way too sentimential and still believe
 366          * they are exclusive and unquestionable owners of the first physical page,
 367          * even though they explicitly mark it as EFI_CONVENTIONAL_MEMORY
 368          * (but then write-access it later during SetVirtualAddressMap()).
 369          *
 370          * Create a 1:1 mapping for this page, to avoid triple faults during early
 371          * boot with such firmware. We are free to hand this page to the BIOS,
 372          * as trim_bios_range() will reserve the first page and isolate it away
 373          * from memory allocators anyway.
 374          */
 375         pf = _PAGE_RW;
 376         if (sev_active())
 377                 pf |= _PAGE_ENC;
 378
 379         if (kernel_map_pages_in_pgd(pgd, 0x0, 0x0, 1, pf)) {
 380                 pr_err("Failed to create 1:1 mapping for the first page!\n");
 381                 return 1;
 382         }
 383
 384         /*
 385          * When making calls to the firmware everything needs to be 1:1
 386          * mapped and addressable with 32-bit pointers. Map the kernel
 387          * text and allocate a new stack because we can't rely on the
 388          * stack pointer being < 4GB.
 389          */
 390         if (!IS_ENABLED(CONFIG_EFI_MIXED) || efi_is_native())
 391                 return 0;
 392
 393         page = alloc_page(GFP_KERNEL|__GFP_DMA32);
 394         if (!page) {
 395                 pr_err("Unable to allocate EFI runtime stack < 4GB\n");
 396                 return 1;
 397         }
 398
 399         efi_scratch.phys_stack = page_to_phys(page + 1); /* stack grows down */
 400
 401         npages = (_etext - _text) >> PAGE_SHIFT;
 402         text = __pa(_text);
 403         pfn = text >> PAGE_SHIFT;
 404
 405         pf = _PAGE_RW | _PAGE_ENC;
 406         if (kernel_map_pages_in_pgd(pgd, pfn, text, npages, pf)) {
 407                 pr_err("Failed to map kernel text 1:1\n");
 408                 return 1;
 409         }
 410
 411         return 0;
 412 }
 413
 414 static void __init __map_region(efi_memory_desc_t *md, u64 va)
 415 {
 416         unsigned long flags = _PAGE_RW;
 417         unsigned long pfn;
 418         pgd_t *pgd = efi_mm.pgd;
 419
 420         if (!(md->attribute & EFI_MEMORY_WB))
 421                 flags |= _PAGE_PCD;
 422
 423         if (sev_active() && md->type != EFI_MEMORY_MAPPED_IO)
 424                 flags |= _PAGE_ENC;
 425
 426         pfn = md->phys_addr >> PAGE_SHIFT;
 427         if (kernel_map_pages_in_pgd(pgd, pfn, va, md->num_pages, flags))
 428                 pr_warn("Error mapping PA 0x%llx -> VA 0x%llx!\n",
 429                            md->phys_addr, va);
 430 }
 431
 432 void __init efi_map_region(efi_memory_desc_t *md)
 433 {
 434         unsigned long size = md->num_pages << PAGE_SHIFT;
 435         u64 pa = md->phys_addr;
 436
 437         if (efi_enabled(EFI_OLD_MEMMAP))
 438                 return old_map_region(md);
 439
 440         /*
 441          * Make sure the 1:1 mappings are present as a catch-all for b0rked
 442          * firmware which doesn't update all internal pointers after switching
 443          * to virtual mode and would otherwise crap on us.
 444          */
 445         __map_region(md, md->phys_addr);
 446
 447         /*
 448          * Enforce the 1:1 mapping as the default virtual address when
 449          * booting in EFI mixed mode, because even though we may be
 450          * running a 64-bit kernel, the firmware may only be 32-bit.
 451          */
 452         if (!efi_is_native () && IS_ENABLED(CONFIG_EFI_MIXED)) {
 453                 md->virt_addr = md->phys_addr;
 454                 return;
 455         }
 456
 457         efi_va -= size;
 458
 459         /* Is PA 2M-aligned? */
 460         if (!(pa & (PMD_SIZE - 1))) {
 461                 efi_va &= PMD_MASK;
 462         } else {
 463                 u64 pa_offset = pa & (PMD_SIZE - 1);
 464                 u64 prev_va = efi_va;
 465
 466                 /* get us the same offset within this 2M page */
 467                 efi_va = (efi_va & PMD_MASK) + pa_offset;
 468
 469                 if (efi_va > prev_va)
 470                         efi_va -= PMD_SIZE;
 471         }
 472
 473         if (efi_va < EFI_VA_END) {
 474                 pr_warn(FW_WARN "VA address range overflow!\n");
 475                 return;
 476         }
 477
 478         /* Do the VA map */
 479         __map_region(md, efi_va);
 480         md->virt_addr = efi_va;
 481 }
 482
 483 /*
 484  * kexec kernel will use efi_map_region_fixed to map efi runtime memory ranges.
 485  * md->virt_addr is the original virtual address which had been mapped in kexec
 486  * 1st kernel.
 487  */
 488 void __init efi_map_region_fixed(efi_memory_desc_t *md)
 489 {
 490         __map_region(md, md->phys_addr);
 491         __map_region(md, md->virt_addr);
 492 }
 493
 494 void __iomem *__init efi_ioremap(unsigned long phys_addr, unsigned long size,
 495                                  u32 type, u64 attribute)
 496 {
 497         unsigned long last_map_pfn;
 498
 499         if (type == EFI_MEMORY_MAPPED_IO)
 500                 return ioremap(phys_addr, size);
 501
 502         last_map_pfn = init_memory_mapping(phys_addr, phys_addr + size);
 503         if ((last_map_pfn << PAGE_SHIFT) < phys_addr + size) {
 504                 unsigned long top = last_map_pfn << PAGE_SHIFT;
 505                 efi_ioremap(top, size - (top - phys_addr), type, attribute);
 506         }
 507
 508         if (!(attribute & EFI_MEMORY_WB))
 509                 efi_memory_uc((u64)(unsigned long)__va(phys_addr), size);
 510
 511         return (void __iomem *)__va(phys_addr);
 512 }
 513
 514 void __init parse_efi_setup(u64 phys_addr, u32 data_len)
 515 {
 516         efi_setup = phys_addr + sizeof(struct setup_data);
 517 }
 518
 519 static int __init efi_update_mappings(efi_memory_desc_t *md, unsigned long pf)
 520 {
 521         unsigned long pfn;
 522         pgd_t *pgd = efi_mm.pgd;
 523         int err1, err2;
 524
 525         /* Update the 1:1 mapping */
 526         pfn = md->phys_addr >> PAGE_SHIFT;
 527         err1 = kernel_map_pages_in_pgd(pgd, pfn, md->phys_addr, md->num_pages, pf);
 528         if (err1) {
 529                 pr_err("Error while updating 1:1 mapping PA 0x%llx -> VA 0x%llx!\n",
 530                            md->phys_addr, md->virt_addr);
 531         }
 532
 533         err2 = kernel_map_pages_in_pgd(pgd, pfn, md->virt_addr, md->num_pages, pf);
 534         if (err2) {
 535                 pr_err("Error while updating VA mapping PA 0x%llx -> VA 0x%llx!\n",
 536                            md->phys_addr, md->virt_addr);
 537         }
 538
 539         return err1 || err2;
 540 }
 541
 542 static int __init efi_update_mem_attr(struct mm_struct *mm, efi_memory_desc_t *md)
 543 {
 544         unsigned long pf = 0;
 545
 546         if (md->attribute & EFI_MEMORY_XP)
 547                 pf |= _PAGE_NX;
 548
 549         if (!(md->attribute & EFI_MEMORY_RO))
 550                 pf |= _PAGE_RW;
 551
 552         if (sev_active())
 553                 pf |= _PAGE_ENC;
 554
 555         return efi_update_mappings(md, pf);
 556 }
 557
 558 void __init efi_runtime_update_mappings(void)
 559 {
 560         efi_memory_desc_t *md;
 561
 562         if (efi_enabled(EFI_OLD_MEMMAP)) {
 563                 if (__supported_pte_mask & _PAGE_NX)
 564                         runtime_code_page_mkexec();
 565                 return;
 566         }
 567
 568         /*
 569          * Use the EFI Memory Attribute Table for mapping permissions if it
 570          * exists, since it is intended to supersede EFI_PROPERTIES_TABLE.
 571          */
 572         if (efi_enabled(EFI_MEM_ATTR)) {
 573                 efi_memattr_apply_permissions(NULL, efi_update_mem_attr);
 574                 return;
 575         }
 576
 577         /*
 578          * EFI_MEMORY_ATTRIBUTES_TABLE is intended to replace
 579          * EFI_PROPERTIES_TABLE. So, use EFI_PROPERTIES_TABLE to update
 580          * permissions only if EFI_MEMORY_ATTRIBUTES_TABLE is not
 581          * published by the firmware. Even if we find a buggy implementation of
 582          * EFI_MEMORY_ATTRIBUTES_TABLE, don't fall back to
 583          * EFI_PROPERTIES_TABLE, because of the same reason.
 584          */
 585
 586         if (!efi_enabled(EFI_NX_PE_DATA))
 587                 return;
 588
 589         for_each_efi_memory_desc(md) {
 590                 unsigned long pf = 0;
 591
 592                 if (!(md->attribute & EFI_MEMORY_RUNTIME))
 593                         continue;
 594
 595                 if (!(md->attribute & EFI_MEMORY_WB))
 596                         pf |= _PAGE_PCD;
 597
 598                 if ((md->attribute & EFI_MEMORY_XP) ||
 599                         (md->type == EFI_RUNTIME_SERVICES_DATA))
 600                         pf |= _PAGE_NX;
 601
 602                 if (!(md->attribute & EFI_MEMORY_RO) &&
 603                         (md->type != EFI_RUNTIME_SERVICES_CODE))
 604                         pf |= _PAGE_RW;
 605
 606                 if (sev_active())
 607                         pf |= _PAGE_ENC;
 608
 609                 efi_update_mappings(md, pf);
 610         }
 611 }
 612
 613 void __init efi_dump_pagetable(void)
 614 {
 615 #ifdef CONFIG_EFI_PGT_DUMP
 616         if (efi_enabled(EFI_OLD_MEMMAP))
 617                 ptdump_walk_pgd_level(NULL, swapper_pg_dir);
 618         else
 619                 ptdump_walk_pgd_level(NULL, efi_mm.pgd);
 620 #endif
 621 }
 622
 623 /*
 624  * Makes the calling thread switch to/from efi_mm context. Can be used
 625  * in a kernel thread and user context. Preemption needs to remain disabled
 626  * while the EFI-mm is borrowed. mmgrab()/mmdrop() is not used because the mm
 627  * can not change under us.
 628  * It should be ensured that there are no concurent calls to this function.
 629  */
 630 void efi_switch_mm(struct mm_struct *mm)
 631 {
 632         efi_scratch.prev_mm = current->active_mm;
 633         current->active_mm = mm;
 634         switch_mm(efi_scratch.prev_mm, mm, NULL);
 635 }
 636
 637 #ifdef CONFIG_EFI_MIXED
 638 extern efi_status_t efi64_thunk(u32, ...);
 639
 640 static DEFINE_SPINLOCK(efi_runtime_lock);
 641
 642 #define runtime_service32(func)                                          \
 643 ({                                                                       \
 644         u32 table = (u32)(unsigned long)efi.systab;                      \
 645         u32 *rt, *___f;                                                  \
 646                                                                          \
 647         rt = (u32 *)(table + offsetof(efi_system_table_32_t, runtime));  \
 648         ___f = (u32 *)(*rt + offsetof(efi_runtime_services_32_t, func)); \
 649         *___f;                                                           \
 650 })
 651
 652 /*
 653  * Switch to the EFI page tables early so that we can access the 1:1
 654  * runtime services mappings which are not mapped in any other page
 655  * tables. This function must be called before runtime_service32().
 656  *
 657  * Also, disable interrupts because the IDT points to 64-bit handlers,
 658  * which aren't going to function correctly when we switch to 32-bit.
 659  */
 660 #define efi_thunk(f, ...)                                               \
 661 ({                                                                      \
 662         efi_status_t __s;                                               \
 663         u32 __func;                                                     \
 664                                                                         \
 665         arch_efi_call_virt_setup();                                     \
 666                                                                         \
 667         __func = runtime_service32(f);                                  \
 668         __s = efi64_thunk(__func, __VA_ARGS__);                         \
 669                                                                         \
 670         arch_efi_call_virt_teardown();                                  \
 671                                                                         \
 672         __s;                                                            \
 673 })
 674
 675 efi_status_t efi_thunk_set_virtual_address_map(
 676         void *phys_set_virtual_address_map,
 677         unsigned long memory_map_size,
 678         unsigned long descriptor_size,
 679         u32 descriptor_version,
 680         efi_memory_desc_t *virtual_map)
 681 {
 682         efi_status_t status;
 683         unsigned long flags;
 684         u32 func;
 685
 686         efi_sync_low_kernel_mappings();
 687         local_irq_save(flags);
 688
 689         efi_switch_mm(&efi_mm);
 690
 691         func = (u32)(unsigned long)phys_set_virtual_address_map;
 692         status = efi64_thunk(func, memory_map_size, descriptor_size,
 693                              descriptor_version, virtual_map);
 694
 695         efi_switch_mm(efi_scratch.prev_mm);
 696         local_irq_restore(flags);
 697
 698         return status;
 699 }
 700
 701 static efi_status_t efi_thunk_get_time(efi_time_t *tm, efi_time_cap_t *tc)
 702 {
 703         efi_status_t status;
 704         u32 phys_tm, phys_tc;
 705         unsigned long flags;
 706
 707         spin_lock(&rtc_lock);
 708         spin_lock_irqsave(&efi_runtime_lock, flags);
 709
 710         phys_tm = virt_to_phys_or_null(tm);
 711         phys_tc = virt_to_phys_or_null(tc);
 712
 713         status = efi_thunk(get_time, phys_tm, phys_tc);
 714
 715         spin_unlock_irqrestore(&efi_runtime_lock, flags);
 716         spin_unlock(&rtc_lock);
 717
 718         return status;
 719 }
 720
 721 static efi_status_t efi_thunk_set_time(efi_time_t *tm)
 722 {
 723         efi_status_t status;
 724         u32 phys_tm;
 725         unsigned long flags;
 726
 727         spin_lock(&rtc_lock);
 728         spin_lock_irqsave(&efi_runtime_lock, flags);
 729
 730         phys_tm = virt_to_phys_or_null(tm);
 731
 732         status = efi_thunk(set_time, phys_tm);
 733
 734         spin_unlock_irqrestore(&efi_runtime_lock, flags);
 735         spin_unlock(&rtc_lock);
 736
 737         return status;
 738 }
 739
 740 static efi_status_t
 741 efi_thunk_get_wakeup_time(efi_bool_t *enabled, efi_bool_t *pending,
 742                           efi_time_t *tm)
 743 {
 744         efi_status_t status;
 745         u32 phys_enabled, phys_pending, phys_tm;
 746         unsigned long flags;
 747
 748         spin_lock(&rtc_lock);
 749         spin_lock_irqsave(&efi_runtime_lock, flags);
 750
 751         phys_enabled = virt_to_phys_or_null(enabled);
 752         phys_pending = virt_to_phys_or_null(pending);
 753         phys_tm = virt_to_phys_or_null(tm);
 754
 755         status = efi_thunk(get_wakeup_time, phys_enabled,
 756                              phys_pending, phys_tm);
 757
 758         spin_unlock_irqrestore(&efi_runtime_lock, flags);
 759         spin_unlock(&rtc_lock);
 760
 761         return status;
 762 }
 763
 764 static efi_status_t
 765 efi_thunk_set_wakeup_time(efi_bool_t enabled, efi_time_t *tm)
 766 {
 767         efi_status_t status;
 768         u32 phys_tm;
 769         unsigned long flags;
 770
 771         spin_lock(&rtc_lock);
 772         spin_lock_irqsave(&efi_runtime_lock, flags);
 773
 774         phys_tm = virt_to_phys_or_null(tm);
 775
 776         status = efi_thunk(set_wakeup_time, enabled, phys_tm);
 777
 778         spin_unlock_irqrestore(&efi_runtime_lock, flags);
 779         spin_unlock(&rtc_lock);
 780
 781         return status;
 782 }
 783
 784 static unsigned long efi_name_size(efi_char16_t *name)
 785 {
 786         return ucs2_strsize(name, EFI_VAR_NAME_LEN) + 1;
 787 }
 788
 789 static efi_status_t
 790 efi_thunk_get_variable(efi_char16_t *name, efi_guid_t *vendor,
 791                        u32 *attr, unsigned long *data_size, void *data)
 792 {
 793         u8 buf[24] __aligned(8);
 794         efi_guid_t *vnd = PTR_ALIGN((efi_guid_t *)buf, sizeof(*vnd));
 795         efi_status_t status;
 796         u32 phys_name, phys_vendor, phys_attr;
 797         u32 phys_data_size, phys_data;
 798         unsigned long flags;
 799
 800         spin_lock_irqsave(&efi_runtime_lock, flags);
 801
 802         *vnd = *vendor;
 803
 804         phys_data_size = virt_to_phys_or_null(data_size);
 805         phys_vendor = virt_to_phys_or_null(vnd);
 806         phys_name = virt_to_phys_or_null_size(name, efi_name_size(name));
 807         phys_attr = virt_to_phys_or_null(attr);
 808         phys_data = virt_to_phys_or_null_size(data, *data_size);
 809
 810         if (!phys_name || (data && !phys_data))
 811                 status = EFI_INVALID_PARAMETER;
 812         else
 813                 status = efi_thunk(get_variable, phys_name, phys_vendor,
 814                                    phys_attr, phys_data_size, phys_data);
 815
 816         spin_unlock_irqrestore(&efi_runtime_lock, flags);
 817
 818         return status;
 819 }
 820
 821 static efi_status_t
 822 efi_thunk_set_variable(efi_char16_t *name, efi_guid_t *vendor,
 823                        u32 attr, unsigned long data_size, void *data)
 824 {
 825         u8 buf[24] __aligned(8);
 826         efi_guid_t *vnd = PTR_ALIGN((efi_guid_t *)buf, sizeof(*vnd));
 827         u32 phys_name, phys_vendor, phys_data;
 828         efi_status_t status;
 829         unsigned long flags;
 830
 831         spin_lock_irqsave(&efi_runtime_lock, flags);
 832
 833         *vnd = *vendor;
 834
 835         phys_name = virt_to_phys_or_null_size(name, efi_name_size(name));
 836         phys_vendor = virt_to_phys_or_null(vnd);
 837         phys_data = virt_to_phys_or_null_size(data, data_size);
 838
 839         if (!phys_name || (data && !phys_data))
 840                 status = EFI_INVALID_PARAMETER;
 841         else
 842                 status = efi_thunk(set_variable, phys_name, phys_vendor,
 843                                    attr, data_size, phys_data);
 844
 845         spin_unlock_irqrestore(&efi_runtime_lock, flags);
 846
 847         return status;
 848 }
 849
 850 static efi_status_t
 851 efi_thunk_set_variable_nonblocking(efi_char16_t *name, efi_guid_t *vendor,
 852                                    u32 attr, unsigned long data_size,
 853                                    void *data)
 854 {
 855         u8 buf[24] __aligned(8);
 856         efi_guid_t *vnd = PTR_ALIGN((efi_guid_t *)buf, sizeof(*vnd));
 857         u32 phys_name, phys_vendor, phys_data;
 858         efi_status_t status;
 859         unsigned long flags;
 860
 861         if (!spin_trylock_irqsave(&efi_runtime_lock, flags))
 862                 return EFI_NOT_READY;
 863
 864         *vnd = *vendor;
 865
 866         phys_name = virt_to_phys_or_null_size(name, efi_name_size(name));
 867         phys_vendor = virt_to_phys_or_null(vnd);
 868         phys_data = virt_to_phys_or_null_size(data, data_size);
 869
 870         if (!phys_name || (data && !phys_data))
 871                 status = EFI_INVALID_PARAMETER;
 872         else
 873                 status = efi_thunk(set_variable, phys_name, phys_vendor,
 874                                    attr, data_size, phys_data);
 875
 876         spin_unlock_irqrestore(&efi_runtime_lock, flags);
 877
 878         return status;
 879 }
 880
 881 static efi_status_t
 882 efi_thunk_get_next_variable(unsigned long *name_size,
 883                             efi_char16_t *name,
 884                             efi_guid_t *vendor)
 885 {
 886         u8 buf[24] __aligned(8);
 887         efi_guid_t *vnd = PTR_ALIGN((efi_guid_t *)buf, sizeof(*vnd));
 888         efi_status_t status;
 889         u32 phys_name_size, phys_name, phys_vendor;
 890         unsigned long flags;
 891
 892         spin_lock_irqsave(&efi_runtime_lock, flags);
 893
 894         *vnd = *vendor;
 895
 896         phys_name_size = virt_to_phys_or_null(name_size);
 897         phys_vendor = virt_to_phys_or_null(vnd);
 898         phys_name = virt_to_phys_or_null_size(name, *name_size);
 899
 900         if (!phys_name)
 901                 status = EFI_INVALID_PARAMETER;
 902         else
 903                 status = efi_thunk(get_next_variable, phys_name_size,
 904                                    phys_name, phys_vendor);
 905
 906         spin_unlock_irqrestore(&efi_runtime_lock, flags);
 907
 908         *vendor = *vnd;
 909         return status;
 910 }
 911
 912 static efi_status_t
 913 efi_thunk_get_next_high_mono_count(u32 *count)
 914 {
 915         efi_status_t status;
 916         u32 phys_count;
 917         unsigned long flags;
 918
 919         spin_lock_irqsave(&efi_runtime_lock, flags);
 920
 921         phys_count = virt_to_phys_or_null(count);
 922         status = efi_thunk(get_next_high_mono_count, phys_count);
 923
 924         spin_unlock_irqrestore(&efi_runtime_lock, flags);
 925
 926         return status;
 927 }
 928
 929 static void
 930 efi_thunk_reset_system(int reset_type, efi_status_t status,
 931                        unsigned long data_size, efi_char16_t *data)
 932 {
 933         u32 phys_data;
 934         unsigned long flags;
 935
 936         spin_lock_irqsave(&efi_runtime_lock, flags);
 937
 938         phys_data = virt_to_phys_or_null_size(data, data_size);
 939
 940         efi_thunk(reset_system, reset_type, status, data_size, phys_data);
 941
 942         spin_unlock_irqrestore(&efi_runtime_lock, flags);
 943 }
 944
 945 static efi_status_t
 946 efi_thunk_update_capsule(efi_capsule_header_t **capsules,
 947                          unsigned long count, unsigned long sg_list)
 948 {
 949         /*
 950          * To properly support this function we would need to repackage
 951          * 'capsules' because the firmware doesn't understand 64-bit
 952          * pointers.
 953          */
 954         return EFI_UNSUPPORTED;
 955 }
 956
 957 static efi_status_t
 958 efi_thunk_query_variable_info(u32 attr, u64 *storage_space,
 959                               u64 *remaining_space,
 960                               u64 *max_variable_size)
 961 {
 962         efi_status_t status;
 963         u32 phys_storage, phys_remaining, phys_max;
 964         unsigned long flags;
 965
 966         if (efi.runtime_version < EFI_2_00_SYSTEM_TABLE_REVISION)
 967                 return EFI_UNSUPPORTED;
 968
 969         spin_lock_irqsave(&efi_runtime_lock, flags);
 970
 971         phys_storage = virt_to_phys_or_null(storage_space);
 972         phys_remaining = virt_to_phys_or_null(remaining_space);
 973         phys_max = virt_to_phys_or_null(max_variable_size);
 974
 975         status = efi_thunk(query_variable_info, attr, phys_storage,
 976                            phys_remaining, phys_max);
 977
 978         spin_unlock_irqrestore(&efi_runtime_lock, flags);
 979
 980         return status;
 981 }
 982
 983 static efi_status_t
 984 efi_thunk_query_variable_info_nonblocking(u32 attr, u64 *storage_space,
 985                                           u64 *remaining_space,
 986                                           u64 *max_variable_size)
 987 {
 988         efi_status_t status;
 989         u32 phys_storage, phys_remaining, phys_max;
 990         unsigned long flags;
 991
 992         if (efi.runtime_version < EFI_2_00_SYSTEM_TABLE_REVISION)
 993                 return EFI_UNSUPPORTED;
 994
 995         if (!spin_trylock_irqsave(&efi_runtime_lock, flags))
 996                 return EFI_NOT_READY;
 997
 998         phys_storage = virt_to_phys_or_null(storage_space);
 999         phys_remaining = virt_to_phys_or_null(remaining_space);
1000         phys_max = virt_to_phys_or_null(max_variable_size);
1001
1002         status = efi_thunk(query_variable_info, attr, phys_storage,
1003                            phys_remaining, phys_max);
1004
1005         spin_unlock_irqrestore(&efi_runtime_lock, flags);
1006
1007         return status;
1008 }
1009
1010 static efi_status_t
1011 efi_thunk_query_capsule_caps(efi_capsule_header_t **capsules,
1012                              unsigned long count, u64 *max_size,
1013                              int *reset_type)
1014 {
1015         /*
1016          * To properly support this function we would need to repackage
1017          * 'capsules' because the firmware doesn't understand 64-bit
1018          * pointers.
1019          */
1020         return EFI_UNSUPPORTED;
1021 }
1022
1023 void efi_thunk_runtime_setup(void)
1024 {
1025         efi.get_time = efi_thunk_get_time;
1026         efi.set_time = efi_thunk_set_time;
1027         efi.get_wakeup_time = efi_thunk_get_wakeup_time;
1028         efi.set_wakeup_time = efi_thunk_set_wakeup_time;
1029         efi.get_variable = efi_thunk_get_variable;
1030         efi.get_next_variable = efi_thunk_get_next_variable;
1031         efi.set_variable = efi_thunk_set_variable;
1032         efi.set_variable_nonblocking = efi_thunk_set_variable_nonblocking;
1033         efi.get_next_high_mono_count = efi_thunk_get_next_high_mono_count;
1034         efi.reset_system = efi_thunk_reset_system;
1035         efi.query_variable_info = efi_thunk_query_variable_info;
1036         efi.query_variable_info_nonblocking = efi_thunk_query_variable_info_nonblocking;
1037         efi.update_capsule = efi_thunk_update_capsule;
1038         efi.query_capsule_caps = efi_thunk_query_capsule_caps;
1039 }
1040 #endif /* CONFIG_EFI_MIXED */