linux/arch/x86/platform/efi/efi_64.c
<<
>>
Prefs
   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/bootmem.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 */
  56static u64 efi_va = EFI_VA_START;
  57
  58struct efi_scratch efi_scratch;
  59
  60static 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
  75pgd_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 NULL;
  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
  95        /*
  96         * Build 1:1 identity mapping for efi=old_map usage. Note that
  97         * PAGE_OFFSET is PGDIR_SIZE aligned when KASLR is disabled, while
  98         * it is PUD_SIZE ALIGNED with KASLR enabled. So for a given physical
  99         * address X, the pud_index(X) != pud_index(__va(X)), we can only copy
 100         * PUD entry of __va(X) to fill in pud entry of X to build 1:1 mapping.
 101         * This means here we can only reuse the PMD tables of the direct mapping.
 102         */
 103        for (pgd = 0; pgd < n_pgds; pgd++) {
 104                addr_pgd = (unsigned long)(pgd * PGDIR_SIZE);
 105                vaddr = (unsigned long)__va(pgd * PGDIR_SIZE);
 106                pgd_efi = pgd_offset_k(addr_pgd);
 107                save_pgd[pgd] = *pgd_efi;
 108
 109                p4d = p4d_alloc(&init_mm, pgd_efi, addr_pgd);
 110                if (!p4d) {
 111                        pr_err("Failed to allocate p4d table!\n");
 112                        goto out;
 113                }
 114
 115                for (i = 0; i < PTRS_PER_P4D; i++) {
 116                        addr_p4d = addr_pgd + i * P4D_SIZE;
 117                        p4d_efi = p4d + p4d_index(addr_p4d);
 118
 119                        pud = pud_alloc(&init_mm, p4d_efi, addr_p4d);
 120                        if (!pud) {
 121                                pr_err("Failed to allocate pud table!\n");
 122                                goto out;
 123                        }
 124
 125                        for (j = 0; j < PTRS_PER_PUD; j++) {
 126                                addr_pud = addr_p4d + j * PUD_SIZE;
 127
 128                                if (addr_pud > (max_pfn << PAGE_SHIFT))
 129                                        break;
 130
 131                                vaddr = (unsigned long)__va(addr_pud);
 132
 133                                pgd_k = pgd_offset_k(vaddr);
 134                                p4d_k = p4d_offset(pgd_k, vaddr);
 135                                pud[j] = *pud_offset(p4d_k, vaddr);
 136                        }
 137                }
 138                pgd_offset_k(pgd * PGDIR_SIZE)->pgd &= ~_PAGE_NX;
 139        }
 140
 141out:
 142        __flush_tlb_all();
 143
 144        return save_pgd;
 145}
 146
 147void __init efi_call_phys_epilog(pgd_t *save_pgd)
 148{
 149        /*
 150         * After the lock is released, the original page table is restored.
 151         */
 152        int pgd_idx, i;
 153        int nr_pgds;
 154        pgd_t *pgd;
 155        p4d_t *p4d;
 156        pud_t *pud;
 157
 158        if (!efi_enabled(EFI_OLD_MEMMAP)) {
 159                efi_switch_mm(efi_scratch.prev_mm);
 160                return;
 161        }
 162
 163        nr_pgds = DIV_ROUND_UP((max_pfn << PAGE_SHIFT) , PGDIR_SIZE);
 164
 165        for (pgd_idx = 0; pgd_idx < nr_pgds; pgd_idx++) {
 166                pgd = pgd_offset_k(pgd_idx * PGDIR_SIZE);
 167                set_pgd(pgd_offset_k(pgd_idx * PGDIR_SIZE), save_pgd[pgd_idx]);
 168
 169                if (!(pgd_val(*pgd) & _PAGE_PRESENT))
 170                        continue;
 171
 172                for (i = 0; i < PTRS_PER_P4D; i++) {
 173                        p4d = p4d_offset(pgd,
 174                                         pgd_idx * PGDIR_SIZE + i * P4D_SIZE);
 175
 176                        if (!(p4d_val(*p4d) & _PAGE_PRESENT))
 177                                continue;
 178
 179                        pud = (pud_t *)p4d_page_vaddr(*p4d);
 180                        pud_free(&init_mm, pud);
 181                }
 182
 183                p4d = (p4d_t *)pgd_page_vaddr(*pgd);
 184                p4d_free(&init_mm, p4d);
 185        }
 186
 187        kfree(save_pgd);
 188
 189        __flush_tlb_all();
 190        early_code_mapping_set_exec(0);
 191}
 192
 193EXPORT_SYMBOL_GPL(efi_mm);
 194
 195/*
 196 * We need our own copy of the higher levels of the page tables
 197 * because we want to avoid inserting EFI region mappings (EFI_VA_END
 198 * to EFI_VA_START) into the standard kernel page tables. Everything
 199 * else can be shared, see efi_sync_low_kernel_mappings().
 200 *
 201 * We don't want the pgd on the pgd_list and cannot use pgd_alloc() for the
 202 * allocation.
 203 */
 204int __init efi_alloc_page_tables(void)
 205{
 206        pgd_t *pgd, *efi_pgd;
 207        p4d_t *p4d;
 208        pud_t *pud;
 209        gfp_t gfp_mask;
 210
 211        if (efi_enabled(EFI_OLD_MEMMAP))
 212                return 0;
 213
 214        gfp_mask = GFP_KERNEL | __GFP_ZERO;
 215        efi_pgd = (pgd_t *)__get_free_pages(gfp_mask, PGD_ALLOCATION_ORDER);
 216        if (!efi_pgd)
 217                return -ENOMEM;
 218
 219        pgd = efi_pgd + pgd_index(EFI_VA_END);
 220        p4d = p4d_alloc(&init_mm, pgd, EFI_VA_END);
 221        if (!p4d) {
 222                free_page((unsigned long)efi_pgd);
 223                return -ENOMEM;
 224        }
 225
 226        pud = pud_alloc(&init_mm, p4d, EFI_VA_END);
 227        if (!pud) {
 228                if (pgtable_l5_enabled)
 229                        free_page((unsigned long) pgd_page_vaddr(*pgd));
 230                free_pages((unsigned long)efi_pgd, PGD_ALLOCATION_ORDER);
 231                return -ENOMEM;
 232        }
 233
 234        efi_mm.pgd = efi_pgd;
 235        mm_init_cpumask(&efi_mm);
 236        init_new_context(NULL, &efi_mm);
 237
 238        return 0;
 239}
 240
 241/*
 242 * Add low kernel mappings for passing arguments to EFI functions.
 243 */
 244void efi_sync_low_kernel_mappings(void)
 245{
 246        unsigned num_entries;
 247        pgd_t *pgd_k, *pgd_efi;
 248        p4d_t *p4d_k, *p4d_efi;
 249        pud_t *pud_k, *pud_efi;
 250        pgd_t *efi_pgd = efi_mm.pgd;
 251
 252        if (efi_enabled(EFI_OLD_MEMMAP))
 253                return;
 254
 255        /*
 256         * We can share all PGD entries apart from the one entry that
 257         * covers the EFI runtime mapping space.
 258         *
 259         * Make sure the EFI runtime region mappings are guaranteed to
 260         * only span a single PGD entry and that the entry also maps
 261         * other important kernel regions.
 262         */
 263        MAYBE_BUILD_BUG_ON(pgd_index(EFI_VA_END) != pgd_index(MODULES_END));
 264        MAYBE_BUILD_BUG_ON((EFI_VA_START & PGDIR_MASK) !=
 265                        (EFI_VA_END & PGDIR_MASK));
 266
 267        pgd_efi = efi_pgd + pgd_index(PAGE_OFFSET);
 268        pgd_k = pgd_offset_k(PAGE_OFFSET);
 269
 270        num_entries = pgd_index(EFI_VA_END) - pgd_index(PAGE_OFFSET);
 271        memcpy(pgd_efi, pgd_k, sizeof(pgd_t) * num_entries);
 272
 273        /*
 274         * As with PGDs, we share all P4D entries apart from the one entry
 275         * that covers the EFI runtime mapping space.
 276         */
 277        BUILD_BUG_ON(p4d_index(EFI_VA_END) != p4d_index(MODULES_END));
 278        BUILD_BUG_ON((EFI_VA_START & P4D_MASK) != (EFI_VA_END & P4D_MASK));
 279
 280        pgd_efi = efi_pgd + pgd_index(EFI_VA_END);
 281        pgd_k = pgd_offset_k(EFI_VA_END);
 282        p4d_efi = p4d_offset(pgd_efi, 0);
 283        p4d_k = p4d_offset(pgd_k, 0);
 284
 285        num_entries = p4d_index(EFI_VA_END);
 286        memcpy(p4d_efi, p4d_k, sizeof(p4d_t) * num_entries);
 287
 288        /*
 289         * We share all the PUD entries apart from those that map the
 290         * EFI regions. Copy around them.
 291         */
 292        BUILD_BUG_ON((EFI_VA_START & ~PUD_MASK) != 0);
 293        BUILD_BUG_ON((EFI_VA_END & ~PUD_MASK) != 0);
 294
 295        p4d_efi = p4d_offset(pgd_efi, EFI_VA_END);
 296        p4d_k = p4d_offset(pgd_k, EFI_VA_END);
 297        pud_efi = pud_offset(p4d_efi, 0);
 298        pud_k = pud_offset(p4d_k, 0);
 299
 300        num_entries = pud_index(EFI_VA_END);
 301        memcpy(pud_efi, pud_k, sizeof(pud_t) * num_entries);
 302
 303        pud_efi = pud_offset(p4d_efi, EFI_VA_START);
 304        pud_k = pud_offset(p4d_k, EFI_VA_START);
 305
 306        num_entries = PTRS_PER_PUD - pud_index(EFI_VA_START);
 307        memcpy(pud_efi, pud_k, sizeof(pud_t) * num_entries);
 308}
 309
 310/*
 311 * Wrapper for slow_virt_to_phys() that handles NULL addresses.
 312 */
 313static inline phys_addr_t
 314virt_to_phys_or_null_size(void *va, unsigned long size)
 315{
 316        bool bad_size;
 317
 318        if (!va)
 319                return 0;
 320
 321        if (virt_addr_valid(va))
 322                return virt_to_phys(va);
 323
 324        /*
 325         * A fully aligned variable on the stack is guaranteed not to
 326         * cross a page bounary. Try to catch strings on the stack by
 327         * checking that 'size' is a power of two.
 328         */
 329        bad_size = size > PAGE_SIZE || !is_power_of_2(size);
 330
 331        WARN_ON(!IS_ALIGNED((unsigned long)va, size) || bad_size);
 332
 333        return slow_virt_to_phys(va);
 334}
 335
 336#define virt_to_phys_or_null(addr)                              \
 337        virt_to_phys_or_null_size((addr), sizeof(*(addr)))
 338
 339int __init efi_setup_page_tables(unsigned long pa_memmap, unsigned num_pages)
 340{
 341        unsigned long pfn, text, pf;
 342        struct page *page;
 343        unsigned npages;
 344        pgd_t *pgd = efi_mm.pgd;
 345
 346        if (efi_enabled(EFI_OLD_MEMMAP))
 347                return 0;
 348
 349        /*
 350         * It can happen that the physical address of new_memmap lands in memory
 351         * which is not mapped in the EFI page table. Therefore we need to go
 352         * and ident-map those pages containing the map before calling
 353         * phys_efi_set_virtual_address_map().
 354         */
 355        pfn = pa_memmap >> PAGE_SHIFT;
 356        pf = _PAGE_NX | _PAGE_RW | _PAGE_ENC;
 357        if (kernel_map_pages_in_pgd(pgd, pfn, pa_memmap, num_pages, pf)) {
 358                pr_err("Error ident-mapping new memmap (0x%lx)!\n", pa_memmap);
 359                return 1;
 360        }
 361
 362        /*
 363         * Certain firmware versions are way too sentimential and still believe
 364         * they are exclusive and unquestionable owners of the first physical page,
 365         * even though they explicitly mark it as EFI_CONVENTIONAL_MEMORY
 366         * (but then write-access it later during SetVirtualAddressMap()).
 367         *
 368         * Create a 1:1 mapping for this page, to avoid triple faults during early
 369         * boot with such firmware. We are free to hand this page to the BIOS,
 370         * as trim_bios_range() will reserve the first page and isolate it away
 371         * from memory allocators anyway.
 372         */
 373        pf = _PAGE_RW;
 374        if (sev_active())
 375                pf |= _PAGE_ENC;
 376
 377        if (kernel_map_pages_in_pgd(pgd, 0x0, 0x0, 1, pf)) {
 378                pr_err("Failed to create 1:1 mapping for the first page!\n");
 379                return 1;
 380        }
 381
 382        /*
 383         * When making calls to the firmware everything needs to be 1:1
 384         * mapped and addressable with 32-bit pointers. Map the kernel
 385         * text and allocate a new stack because we can't rely on the
 386         * stack pointer being < 4GB.
 387         */
 388        if (!IS_ENABLED(CONFIG_EFI_MIXED) || efi_is_native())
 389                return 0;
 390
 391        page = alloc_page(GFP_KERNEL|__GFP_DMA32);
 392        if (!page)
 393                panic("Unable to allocate EFI runtime stack < 4GB\n");
 394
 395        efi_scratch.phys_stack = virt_to_phys(page_address(page));
 396        efi_scratch.phys_stack += PAGE_SIZE; /* stack grows down */
 397
 398        npages = (_etext - _text) >> PAGE_SHIFT;
 399        text = __pa(_text);
 400        pfn = text >> PAGE_SHIFT;
 401
 402        pf = _PAGE_RW | _PAGE_ENC;
 403        if (kernel_map_pages_in_pgd(pgd, pfn, text, npages, pf)) {
 404                pr_err("Failed to map kernel text 1:1\n");
 405                return 1;
 406        }
 407
 408        return 0;
 409}
 410
 411static void __init __map_region(efi_memory_desc_t *md, u64 va)
 412{
 413        unsigned long flags = _PAGE_RW;
 414        unsigned long pfn;
 415        pgd_t *pgd = efi_mm.pgd;
 416
 417        if (!(md->attribute & EFI_MEMORY_WB))
 418                flags |= _PAGE_PCD;
 419
 420        if (sev_active())
 421                flags |= _PAGE_ENC;
 422
 423        pfn = md->phys_addr >> PAGE_SHIFT;
 424        if (kernel_map_pages_in_pgd(pgd, pfn, va, md->num_pages, flags))
 425                pr_warn("Error mapping PA 0x%llx -> VA 0x%llx!\n",
 426                           md->phys_addr, va);
 427}
 428
 429void __init efi_map_region(efi_memory_desc_t *md)
 430{
 431        unsigned long size = md->num_pages << PAGE_SHIFT;
 432        u64 pa = md->phys_addr;
 433
 434        if (efi_enabled(EFI_OLD_MEMMAP))
 435                return old_map_region(md);
 436
 437        /*
 438         * Make sure the 1:1 mappings are present as a catch-all for b0rked
 439         * firmware which doesn't update all internal pointers after switching
 440         * to virtual mode and would otherwise crap on us.
 441         */
 442        __map_region(md, md->phys_addr);
 443
 444        /*
 445         * Enforce the 1:1 mapping as the default virtual address when
 446         * booting in EFI mixed mode, because even though we may be
 447         * running a 64-bit kernel, the firmware may only be 32-bit.
 448         */
 449        if (!efi_is_native () && IS_ENABLED(CONFIG_EFI_MIXED)) {
 450                md->virt_addr = md->phys_addr;
 451                return;
 452        }
 453
 454        efi_va -= size;
 455
 456        /* Is PA 2M-aligned? */
 457        if (!(pa & (PMD_SIZE - 1))) {
 458                efi_va &= PMD_MASK;
 459        } else {
 460                u64 pa_offset = pa & (PMD_SIZE - 1);
 461                u64 prev_va = efi_va;
 462
 463                /* get us the same offset within this 2M page */
 464                efi_va = (efi_va & PMD_MASK) + pa_offset;
 465
 466                if (efi_va > prev_va)
 467                        efi_va -= PMD_SIZE;
 468        }
 469
 470        if (efi_va < EFI_VA_END) {
 471                pr_warn(FW_WARN "VA address range overflow!\n");
 472                return;
 473        }
 474
 475        /* Do the VA map */
 476        __map_region(md, efi_va);
 477        md->virt_addr = efi_va;
 478}
 479
 480/*
 481 * kexec kernel will use efi_map_region_fixed to map efi runtime memory ranges.
 482 * md->virt_addr is the original virtual address which had been mapped in kexec
 483 * 1st kernel.
 484 */
 485void __init efi_map_region_fixed(efi_memory_desc_t *md)
 486{
 487        __map_region(md, md->phys_addr);
 488        __map_region(md, md->virt_addr);
 489}
 490
 491void __iomem *__init efi_ioremap(unsigned long phys_addr, unsigned long size,
 492                                 u32 type, u64 attribute)
 493{
 494        unsigned long last_map_pfn;
 495
 496        if (type == EFI_MEMORY_MAPPED_IO)
 497                return ioremap(phys_addr, size);
 498
 499        last_map_pfn = init_memory_mapping(phys_addr, phys_addr + size);
 500        if ((last_map_pfn << PAGE_SHIFT) < phys_addr + size) {
 501                unsigned long top = last_map_pfn << PAGE_SHIFT;
 502                efi_ioremap(top, size - (top - phys_addr), type, attribute);
 503        }
 504
 505        if (!(attribute & EFI_MEMORY_WB))
 506                efi_memory_uc((u64)(unsigned long)__va(phys_addr), size);
 507
 508        return (void __iomem *)__va(phys_addr);
 509}
 510
 511void __init parse_efi_setup(u64 phys_addr, u32 data_len)
 512{
 513        efi_setup = phys_addr + sizeof(struct setup_data);
 514}
 515
 516static int __init efi_update_mappings(efi_memory_desc_t *md, unsigned long pf)
 517{
 518        unsigned long pfn;
 519        pgd_t *pgd = efi_mm.pgd;
 520        int err1, err2;
 521
 522        /* Update the 1:1 mapping */
 523        pfn = md->phys_addr >> PAGE_SHIFT;
 524        err1 = kernel_map_pages_in_pgd(pgd, pfn, md->phys_addr, md->num_pages, pf);
 525        if (err1) {
 526                pr_err("Error while updating 1:1 mapping PA 0x%llx -> VA 0x%llx!\n",
 527                           md->phys_addr, md->virt_addr);
 528        }
 529
 530        err2 = kernel_map_pages_in_pgd(pgd, pfn, md->virt_addr, md->num_pages, pf);
 531        if (err2) {
 532                pr_err("Error while updating VA mapping PA 0x%llx -> VA 0x%llx!\n",
 533                           md->phys_addr, md->virt_addr);
 534        }
 535
 536        return err1 || err2;
 537}
 538
 539static int __init efi_update_mem_attr(struct mm_struct *mm, efi_memory_desc_t *md)
 540{
 541        unsigned long pf = 0;
 542
 543        if (md->attribute & EFI_MEMORY_XP)
 544                pf |= _PAGE_NX;
 545
 546        if (!(md->attribute & EFI_MEMORY_RO))
 547                pf |= _PAGE_RW;
 548
 549        if (sev_active())
 550                pf |= _PAGE_ENC;
 551
 552        return efi_update_mappings(md, pf);
 553}
 554
 555void __init efi_runtime_update_mappings(void)
 556{
 557        efi_memory_desc_t *md;
 558
 559        if (efi_enabled(EFI_OLD_MEMMAP)) {
 560                if (__supported_pte_mask & _PAGE_NX)
 561                        runtime_code_page_mkexec();
 562                return;
 563        }
 564
 565        /*
 566         * Use the EFI Memory Attribute Table for mapping permissions if it
 567         * exists, since it is intended to supersede EFI_PROPERTIES_TABLE.
 568         */
 569        if (efi_enabled(EFI_MEM_ATTR)) {
 570                efi_memattr_apply_permissions(NULL, efi_update_mem_attr);
 571                return;
 572        }
 573
 574        /*
 575         * EFI_MEMORY_ATTRIBUTES_TABLE is intended to replace
 576         * EFI_PROPERTIES_TABLE. So, use EFI_PROPERTIES_TABLE to update
 577         * permissions only if EFI_MEMORY_ATTRIBUTES_TABLE is not
 578         * published by the firmware. Even if we find a buggy implementation of
 579         * EFI_MEMORY_ATTRIBUTES_TABLE, don't fall back to
 580         * EFI_PROPERTIES_TABLE, because of the same reason.
 581         */
 582
 583        if (!efi_enabled(EFI_NX_PE_DATA))
 584                return;
 585
 586        for_each_efi_memory_desc(md) {
 587                unsigned long pf = 0;
 588
 589                if (!(md->attribute & EFI_MEMORY_RUNTIME))
 590                        continue;
 591
 592                if (!(md->attribute & EFI_MEMORY_WB))
 593                        pf |= _PAGE_PCD;
 594
 595                if ((md->attribute & EFI_MEMORY_XP) ||
 596                        (md->type == EFI_RUNTIME_SERVICES_DATA))
 597                        pf |= _PAGE_NX;
 598
 599                if (!(md->attribute & EFI_MEMORY_RO) &&
 600                        (md->type != EFI_RUNTIME_SERVICES_CODE))
 601                        pf |= _PAGE_RW;
 602
 603                if (sev_active())
 604                        pf |= _PAGE_ENC;
 605
 606                efi_update_mappings(md, pf);
 607        }
 608}
 609
 610void __init efi_dump_pagetable(void)
 611{
 612#ifdef CONFIG_EFI_PGT_DUMP
 613        if (efi_enabled(EFI_OLD_MEMMAP))
 614                ptdump_walk_pgd_level(NULL, swapper_pg_dir);
 615        else
 616                ptdump_walk_pgd_level(NULL, efi_mm.pgd);
 617#endif
 618}
 619
 620/*
 621 * Makes the calling thread switch to/from efi_mm context. Can be used
 622 * for SetVirtualAddressMap() i.e. current->active_mm == init_mm as well
 623 * as during efi runtime calls i.e current->active_mm == current_mm.
 624 * We are not mm_dropping()/mm_grabbing() any mm, because we are not
 625 * losing/creating any references.
 626 */
 627void efi_switch_mm(struct mm_struct *mm)
 628{
 629        task_lock(current);
 630        efi_scratch.prev_mm = current->active_mm;
 631        current->active_mm = mm;
 632        switch_mm(efi_scratch.prev_mm, mm, NULL);
 633        task_unlock(current);
 634}
 635
 636#ifdef CONFIG_EFI_MIXED
 637extern efi_status_t efi64_thunk(u32, ...);
 638
 639#define runtime_service32(func)                                          \
 640({                                                                       \
 641        u32 table = (u32)(unsigned long)efi.systab;                      \
 642        u32 *rt, *___f;                                                  \
 643                                                                         \
 644        rt = (u32 *)(table + offsetof(efi_system_table_32_t, runtime));  \
 645        ___f = (u32 *)(*rt + offsetof(efi_runtime_services_32_t, func)); \
 646        *___f;                                                           \
 647})
 648
 649/*
 650 * Switch to the EFI page tables early so that we can access the 1:1
 651 * runtime services mappings which are not mapped in any other page
 652 * tables. This function must be called before runtime_service32().
 653 *
 654 * Also, disable interrupts because the IDT points to 64-bit handlers,
 655 * which aren't going to function correctly when we switch to 32-bit.
 656 */
 657#define efi_thunk(f, ...)                                               \
 658({                                                                      \
 659        efi_status_t __s;                                               \
 660        unsigned long __flags;                                          \
 661        u32 __func;                                                     \
 662                                                                        \
 663        local_irq_save(__flags);                                        \
 664        arch_efi_call_virt_setup();                                     \
 665                                                                        \
 666        __func = runtime_service32(f);                                  \
 667        __s = efi64_thunk(__func, __VA_ARGS__);                         \
 668                                                                        \
 669        arch_efi_call_virt_teardown();                                  \
 670        local_irq_restore(__flags);                                     \
 671                                                                        \
 672        __s;                                                            \
 673})
 674
 675efi_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
 701static 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
 706        spin_lock(&rtc_lock);
 707
 708        phys_tm = virt_to_phys_or_null(tm);
 709        phys_tc = virt_to_phys_or_null(tc);
 710
 711        status = efi_thunk(get_time, phys_tm, phys_tc);
 712
 713        spin_unlock(&rtc_lock);
 714
 715        return status;
 716}
 717
 718static efi_status_t efi_thunk_set_time(efi_time_t *tm)
 719{
 720        efi_status_t status;
 721        u32 phys_tm;
 722
 723        spin_lock(&rtc_lock);
 724
 725        phys_tm = virt_to_phys_or_null(tm);
 726
 727        status = efi_thunk(set_time, phys_tm);
 728
 729        spin_unlock(&rtc_lock);
 730
 731        return status;
 732}
 733
 734static efi_status_t
 735efi_thunk_get_wakeup_time(efi_bool_t *enabled, efi_bool_t *pending,
 736                          efi_time_t *tm)
 737{
 738        efi_status_t status;
 739        u32 phys_enabled, phys_pending, phys_tm;
 740
 741        spin_lock(&rtc_lock);
 742
 743        phys_enabled = virt_to_phys_or_null(enabled);
 744        phys_pending = virt_to_phys_or_null(pending);
 745        phys_tm = virt_to_phys_or_null(tm);
 746
 747        status = efi_thunk(get_wakeup_time, phys_enabled,
 748                             phys_pending, phys_tm);
 749
 750        spin_unlock(&rtc_lock);
 751
 752        return status;
 753}
 754
 755static efi_status_t
 756efi_thunk_set_wakeup_time(efi_bool_t enabled, efi_time_t *tm)
 757{
 758        efi_status_t status;
 759        u32 phys_tm;
 760
 761        spin_lock(&rtc_lock);
 762
 763        phys_tm = virt_to_phys_or_null(tm);
 764
 765        status = efi_thunk(set_wakeup_time, enabled, phys_tm);
 766
 767        spin_unlock(&rtc_lock);
 768
 769        return status;
 770}
 771
 772static unsigned long efi_name_size(efi_char16_t *name)
 773{
 774        return ucs2_strsize(name, EFI_VAR_NAME_LEN) + 1;
 775}
 776
 777static efi_status_t
 778efi_thunk_get_variable(efi_char16_t *name, efi_guid_t *vendor,
 779                       u32 *attr, unsigned long *data_size, void *data)
 780{
 781        efi_status_t status;
 782        u32 phys_name, phys_vendor, phys_attr;
 783        u32 phys_data_size, phys_data;
 784
 785        phys_data_size = virt_to_phys_or_null(data_size);
 786        phys_vendor = virt_to_phys_or_null(vendor);
 787        phys_name = virt_to_phys_or_null_size(name, efi_name_size(name));
 788        phys_attr = virt_to_phys_or_null(attr);
 789        phys_data = virt_to_phys_or_null_size(data, *data_size);
 790
 791        status = efi_thunk(get_variable, phys_name, phys_vendor,
 792                           phys_attr, phys_data_size, phys_data);
 793
 794        return status;
 795}
 796
 797static efi_status_t
 798efi_thunk_set_variable(efi_char16_t *name, efi_guid_t *vendor,
 799                       u32 attr, unsigned long data_size, void *data)
 800{
 801        u32 phys_name, phys_vendor, phys_data;
 802        efi_status_t status;
 803
 804        phys_name = virt_to_phys_or_null_size(name, efi_name_size(name));
 805        phys_vendor = virt_to_phys_or_null(vendor);
 806        phys_data = virt_to_phys_or_null_size(data, data_size);
 807
 808        /* If data_size is > sizeof(u32) we've got problems */
 809        status = efi_thunk(set_variable, phys_name, phys_vendor,
 810                           attr, data_size, phys_data);
 811
 812        return status;
 813}
 814
 815static efi_status_t
 816efi_thunk_get_next_variable(unsigned long *name_size,
 817                            efi_char16_t *name,
 818                            efi_guid_t *vendor)
 819{
 820        efi_status_t status;
 821        u32 phys_name_size, phys_name, phys_vendor;
 822
 823        phys_name_size = virt_to_phys_or_null(name_size);
 824        phys_vendor = virt_to_phys_or_null(vendor);
 825        phys_name = virt_to_phys_or_null_size(name, *name_size);
 826
 827        status = efi_thunk(get_next_variable, phys_name_size,
 828                           phys_name, phys_vendor);
 829
 830        return status;
 831}
 832
 833static efi_status_t
 834efi_thunk_get_next_high_mono_count(u32 *count)
 835{
 836        efi_status_t status;
 837        u32 phys_count;
 838
 839        phys_count = virt_to_phys_or_null(count);
 840        status = efi_thunk(get_next_high_mono_count, phys_count);
 841
 842        return status;
 843}
 844
 845static void
 846efi_thunk_reset_system(int reset_type, efi_status_t status,
 847                       unsigned long data_size, efi_char16_t *data)
 848{
 849        u32 phys_data;
 850
 851        phys_data = virt_to_phys_or_null_size(data, data_size);
 852
 853        efi_thunk(reset_system, reset_type, status, data_size, phys_data);
 854}
 855
 856static efi_status_t
 857efi_thunk_update_capsule(efi_capsule_header_t **capsules,
 858                         unsigned long count, unsigned long sg_list)
 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
 868static efi_status_t
 869efi_thunk_query_variable_info(u32 attr, u64 *storage_space,
 870                              u64 *remaining_space,
 871                              u64 *max_variable_size)
 872{
 873        efi_status_t status;
 874        u32 phys_storage, phys_remaining, phys_max;
 875
 876        if (efi.runtime_version < EFI_2_00_SYSTEM_TABLE_REVISION)
 877                return EFI_UNSUPPORTED;
 878
 879        phys_storage = virt_to_phys_or_null(storage_space);
 880        phys_remaining = virt_to_phys_or_null(remaining_space);
 881        phys_max = virt_to_phys_or_null(max_variable_size);
 882
 883        status = efi_thunk(query_variable_info, attr, phys_storage,
 884                           phys_remaining, phys_max);
 885
 886        return status;
 887}
 888
 889static efi_status_t
 890efi_thunk_query_capsule_caps(efi_capsule_header_t **capsules,
 891                             unsigned long count, u64 *max_size,
 892                             int *reset_type)
 893{
 894        /*
 895         * To properly support this function we would need to repackage
 896         * 'capsules' because the firmware doesn't understand 64-bit
 897         * pointers.
 898         */
 899        return EFI_UNSUPPORTED;
 900}
 901
 902void efi_thunk_runtime_setup(void)
 903{
 904        efi.get_time = efi_thunk_get_time;
 905        efi.set_time = efi_thunk_set_time;
 906        efi.get_wakeup_time = efi_thunk_get_wakeup_time;
 907        efi.set_wakeup_time = efi_thunk_set_wakeup_time;
 908        efi.get_variable = efi_thunk_get_variable;
 909        efi.get_next_variable = efi_thunk_get_next_variable;
 910        efi.set_variable = efi_thunk_set_variable;
 911        efi.get_next_high_mono_count = efi_thunk_get_next_high_mono_count;
 912        efi.reset_system = efi_thunk_reset_system;
 913        efi.query_variable_info = efi_thunk_query_variable_info;
 914        efi.update_capsule = efi_thunk_update_capsule;
 915        efi.query_capsule_caps = efi_thunk_query_capsule_caps;
 916}
 917#endif /* CONFIG_EFI_MIXED */
 918