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/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 */
  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 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;
 145out:
 146        efi_call_phys_epilog(save_pgd);
 147        return NULL;
 148}
 149
 150void __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
 196EXPORT_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 */
 207int __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                return -ENOMEM;
 221
 222        pgd = efi_pgd + pgd_index(EFI_VA_END);
 223        p4d = p4d_alloc(&init_mm, pgd, EFI_VA_END);
 224        if (!p4d) {
 225                free_page((unsigned long)efi_pgd);
 226                return -ENOMEM;
 227        }
 228
 229        pud = pud_alloc(&init_mm, p4d, EFI_VA_END);
 230        if (!pud) {
 231                if (pgtable_l5_enabled())
 232                        free_page((unsigned long) pgd_page_vaddr(*pgd));
 233                free_pages((unsigned long)efi_pgd, PGD_ALLOCATION_ORDER);
 234                return -ENOMEM;
 235        }
 236
 237        efi_mm.pgd = efi_pgd;
 238        mm_init_cpumask(&efi_mm);
 239        init_new_context(NULL, &efi_mm);
 240
 241        return 0;
 242}
 243
 244/*
 245 * Add low kernel mappings for passing arguments to EFI functions.
 246 */
 247void efi_sync_low_kernel_mappings(void)
 248{
 249        unsigned num_entries;
 250        pgd_t *pgd_k, *pgd_efi;
 251        p4d_t *p4d_k, *p4d_efi;
 252        pud_t *pud_k, *pud_efi;
 253        pgd_t *efi_pgd = efi_mm.pgd;
 254
 255        if (efi_enabled(EFI_OLD_MEMMAP))
 256                return;
 257
 258        /*
 259         * We can share all PGD entries apart from the one entry that
 260         * covers the EFI runtime mapping space.
 261         *
 262         * Make sure the EFI runtime region mappings are guaranteed to
 263         * only span a single PGD entry and that the entry also maps
 264         * other important kernel regions.
 265         */
 266        MAYBE_BUILD_BUG_ON(pgd_index(EFI_VA_END) != pgd_index(MODULES_END));
 267        MAYBE_BUILD_BUG_ON((EFI_VA_START & PGDIR_MASK) !=
 268                        (EFI_VA_END & PGDIR_MASK));
 269
 270        pgd_efi = efi_pgd + pgd_index(PAGE_OFFSET);
 271        pgd_k = pgd_offset_k(PAGE_OFFSET);
 272
 273        num_entries = pgd_index(EFI_VA_END) - pgd_index(PAGE_OFFSET);
 274        memcpy(pgd_efi, pgd_k, sizeof(pgd_t) * num_entries);
 275
 276        /*
 277         * As with PGDs, we share all P4D entries apart from the one entry
 278         * that covers the EFI runtime mapping space.
 279         */
 280        BUILD_BUG_ON(p4d_index(EFI_VA_END) != p4d_index(MODULES_END));
 281        BUILD_BUG_ON((EFI_VA_START & P4D_MASK) != (EFI_VA_END & P4D_MASK));
 282
 283        pgd_efi = efi_pgd + pgd_index(EFI_VA_END);
 284        pgd_k = pgd_offset_k(EFI_VA_END);
 285        p4d_efi = p4d_offset(pgd_efi, 0);
 286        p4d_k = p4d_offset(pgd_k, 0);
 287
 288        num_entries = p4d_index(EFI_VA_END);
 289        memcpy(p4d_efi, p4d_k, sizeof(p4d_t) * num_entries);
 290
 291        /*
 292         * We share all the PUD entries apart from those that map the
 293         * EFI regions. Copy around them.
 294         */
 295        BUILD_BUG_ON((EFI_VA_START & ~PUD_MASK) != 0);
 296        BUILD_BUG_ON((EFI_VA_END & ~PUD_MASK) != 0);
 297
 298        p4d_efi = p4d_offset(pgd_efi, EFI_VA_END);
 299        p4d_k = p4d_offset(pgd_k, EFI_VA_END);
 300        pud_efi = pud_offset(p4d_efi, 0);
 301        pud_k = pud_offset(p4d_k, 0);
 302
 303        num_entries = pud_index(EFI_VA_END);
 304        memcpy(pud_efi, pud_k, sizeof(pud_t) * num_entries);
 305
 306        pud_efi = pud_offset(p4d_efi, EFI_VA_START);
 307        pud_k = pud_offset(p4d_k, EFI_VA_START);
 308
 309        num_entries = PTRS_PER_PUD - pud_index(EFI_VA_START);
 310        memcpy(pud_efi, pud_k, sizeof(pud_t) * num_entries);
 311}
 312
 313/*
 314 * Wrapper for slow_virt_to_phys() that handles NULL addresses.
 315 */
 316static inline phys_addr_t
 317virt_to_phys_or_null_size(void *va, unsigned long size)
 318{
 319        bool bad_size;
 320
 321        if (!va)
 322                return 0;
 323
 324        if (virt_addr_valid(va))
 325                return virt_to_phys(va);
 326
 327        /*
 328         * A fully aligned variable on the stack is guaranteed not to
 329         * cross a page bounary. Try to catch strings on the stack by
 330         * checking that 'size' is a power of two.
 331         */
 332        bad_size = size > PAGE_SIZE || !is_power_of_2(size);
 333
 334        WARN_ON(!IS_ALIGNED((unsigned long)va, size) || bad_size);
 335
 336        return slow_virt_to_phys(va);
 337}
 338
 339#define virt_to_phys_or_null(addr)                              \
 340        virt_to_phys_or_null_size((addr), sizeof(*(addr)))
 341
 342int __init efi_setup_page_tables(unsigned long pa_memmap, unsigned num_pages)
 343{
 344        unsigned long pfn, text, pf;
 345        struct page *page;
 346        unsigned npages;
 347        pgd_t *pgd = efi_mm.pgd;
 348
 349        if (efi_enabled(EFI_OLD_MEMMAP))
 350                return 0;
 351
 352        /*
 353         * It can happen that the physical address of new_memmap lands in memory
 354         * which is not mapped in the EFI page table. Therefore we need to go
 355         * and ident-map those pages containing the map before calling
 356         * phys_efi_set_virtual_address_map().
 357         */
 358        pfn = pa_memmap >> PAGE_SHIFT;
 359        pf = _PAGE_NX | _PAGE_RW | _PAGE_ENC;
 360        if (kernel_map_pages_in_pgd(pgd, pfn, pa_memmap, num_pages, pf)) {
 361                pr_err("Error ident-mapping new memmap (0x%lx)!\n", pa_memmap);
 362                return 1;
 363        }
 364
 365        /*
 366         * Certain firmware versions are way too sentimential and still believe
 367         * they are exclusive and unquestionable owners of the first physical page,
 368         * even though they explicitly mark it as EFI_CONVENTIONAL_MEMORY
 369         * (but then write-access it later during SetVirtualAddressMap()).
 370         *
 371         * Create a 1:1 mapping for this page, to avoid triple faults during early
 372         * boot with such firmware. We are free to hand this page to the BIOS,
 373         * as trim_bios_range() will reserve the first page and isolate it away
 374         * from memory allocators anyway.
 375         */
 376        pf = _PAGE_RW;
 377        if (sev_active())
 378                pf |= _PAGE_ENC;
 379
 380        if (kernel_map_pages_in_pgd(pgd, 0x0, 0x0, 1, pf)) {
 381                pr_err("Failed to create 1:1 mapping for the first page!\n");
 382                return 1;
 383        }
 384
 385        /*
 386         * When making calls to the firmware everything needs to be 1:1
 387         * mapped and addressable with 32-bit pointers. Map the kernel
 388         * text and allocate a new stack because we can't rely on the
 389         * stack pointer being < 4GB.
 390         */
 391        if (!IS_ENABLED(CONFIG_EFI_MIXED) || efi_is_native())
 392                return 0;
 393
 394        page = alloc_page(GFP_KERNEL|__GFP_DMA32);
 395        if (!page)
 396                panic("Unable to allocate EFI runtime stack < 4GB\n");
 397
 398        efi_scratch.phys_stack = virt_to_phys(page_address(page));
 399        efi_scratch.phys_stack += PAGE_SIZE; /* 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
 414static 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
 432void __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 */
 488void __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
 494void __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
 514void __init parse_efi_setup(u64 phys_addr, u32 data_len)
 515{
 516        efi_setup = phys_addr + sizeof(struct setup_data);
 517}
 518
 519static 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
 542static 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
 558void __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
 613void __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 */
 630void 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
 638extern efi_status_t efi64_thunk(u32, ...);
 639
 640static 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
 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        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
 721static 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
 740static efi_status_t
 741efi_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
 764static efi_status_t
 765efi_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
 784static unsigned long efi_name_size(efi_char16_t *name)
 785{
 786        return ucs2_strsize(name, EFI_VAR_NAME_LEN) + 1;
 787}
 788
 789static efi_status_t
 790efi_thunk_get_variable(efi_char16_t *name, efi_guid_t *vendor,
 791                       u32 *attr, unsigned long *data_size, void *data)
 792{
 793        efi_status_t status;
 794        u32 phys_name, phys_vendor, phys_attr;
 795        u32 phys_data_size, phys_data;
 796        unsigned long flags;
 797
 798        spin_lock_irqsave(&efi_runtime_lock, flags);
 799
 800        phys_data_size = virt_to_phys_or_null(data_size);
 801        phys_vendor = virt_to_phys_or_null(vendor);
 802        phys_name = virt_to_phys_or_null_size(name, efi_name_size(name));
 803        phys_attr = virt_to_phys_or_null(attr);
 804        phys_data = virt_to_phys_or_null_size(data, *data_size);
 805
 806        status = efi_thunk(get_variable, phys_name, phys_vendor,
 807                           phys_attr, phys_data_size, phys_data);
 808
 809        spin_unlock_irqrestore(&efi_runtime_lock, flags);
 810
 811        return status;
 812}
 813
 814static efi_status_t
 815efi_thunk_set_variable(efi_char16_t *name, efi_guid_t *vendor,
 816                       u32 attr, unsigned long data_size, void *data)
 817{
 818        u32 phys_name, phys_vendor, phys_data;
 819        efi_status_t status;
 820        unsigned long flags;
 821
 822        spin_lock_irqsave(&efi_runtime_lock, flags);
 823
 824        phys_name = virt_to_phys_or_null_size(name, efi_name_size(name));
 825        phys_vendor = virt_to_phys_or_null(vendor);
 826        phys_data = virt_to_phys_or_null_size(data, data_size);
 827
 828        /* If data_size is > sizeof(u32) we've got problems */
 829        status = efi_thunk(set_variable, phys_name, phys_vendor,
 830                           attr, data_size, phys_data);
 831
 832        spin_unlock_irqrestore(&efi_runtime_lock, flags);
 833
 834        return status;
 835}
 836
 837static efi_status_t
 838efi_thunk_set_variable_nonblocking(efi_char16_t *name, efi_guid_t *vendor,
 839                                   u32 attr, unsigned long data_size,
 840                                   void *data)
 841{
 842        u32 phys_name, phys_vendor, phys_data;
 843        efi_status_t status;
 844        unsigned long flags;
 845
 846        if (!spin_trylock_irqsave(&efi_runtime_lock, flags))
 847                return EFI_NOT_READY;
 848
 849        phys_name = virt_to_phys_or_null_size(name, efi_name_size(name));
 850        phys_vendor = virt_to_phys_or_null(vendor);
 851        phys_data = virt_to_phys_or_null_size(data, data_size);
 852
 853        /* If data_size is > sizeof(u32) we've got problems */
 854        status = efi_thunk(set_variable, phys_name, phys_vendor,
 855                           attr, data_size, phys_data);
 856
 857        spin_unlock_irqrestore(&efi_runtime_lock, flags);
 858
 859        return status;
 860}
 861
 862static efi_status_t
 863efi_thunk_get_next_variable(unsigned long *name_size,
 864                            efi_char16_t *name,
 865                            efi_guid_t *vendor)
 866{
 867        efi_status_t status;
 868        u32 phys_name_size, phys_name, phys_vendor;
 869        unsigned long flags;
 870
 871        spin_lock_irqsave(&efi_runtime_lock, flags);
 872
 873        phys_name_size = virt_to_phys_or_null(name_size);
 874        phys_vendor = virt_to_phys_or_null(vendor);
 875        phys_name = virt_to_phys_or_null_size(name, *name_size);
 876
 877        status = efi_thunk(get_next_variable, phys_name_size,
 878                           phys_name, phys_vendor);
 879
 880        spin_unlock_irqrestore(&efi_runtime_lock, flags);
 881
 882        return status;
 883}
 884
 885static efi_status_t
 886efi_thunk_get_next_high_mono_count(u32 *count)
 887{
 888        efi_status_t status;
 889        u32 phys_count;
 890        unsigned long flags;
 891
 892        spin_lock_irqsave(&efi_runtime_lock, flags);
 893
 894        phys_count = virt_to_phys_or_null(count);
 895        status = efi_thunk(get_next_high_mono_count, phys_count);
 896
 897        spin_unlock_irqrestore(&efi_runtime_lock, flags);
 898
 899        return status;
 900}
 901
 902static void
 903efi_thunk_reset_system(int reset_type, efi_status_t status,
 904                       unsigned long data_size, efi_char16_t *data)
 905{
 906        u32 phys_data;
 907        unsigned long flags;
 908
 909        spin_lock_irqsave(&efi_runtime_lock, flags);
 910
 911        phys_data = virt_to_phys_or_null_size(data, data_size);
 912
 913        efi_thunk(reset_system, reset_type, status, data_size, phys_data);
 914
 915        spin_unlock_irqrestore(&efi_runtime_lock, flags);
 916}
 917
 918static efi_status_t
 919efi_thunk_update_capsule(efi_capsule_header_t **capsules,
 920                         unsigned long count, unsigned long sg_list)
 921{
 922        /*
 923         * To properly support this function we would need to repackage
 924         * 'capsules' because the firmware doesn't understand 64-bit
 925         * pointers.
 926         */
 927        return EFI_UNSUPPORTED;
 928}
 929
 930static efi_status_t
 931efi_thunk_query_variable_info(u32 attr, u64 *storage_space,
 932                              u64 *remaining_space,
 933                              u64 *max_variable_size)
 934{
 935        efi_status_t status;
 936        u32 phys_storage, phys_remaining, phys_max;
 937        unsigned long flags;
 938
 939        if (efi.runtime_version < EFI_2_00_SYSTEM_TABLE_REVISION)
 940                return EFI_UNSUPPORTED;
 941
 942        spin_lock_irqsave(&efi_runtime_lock, flags);
 943
 944        phys_storage = virt_to_phys_or_null(storage_space);
 945        phys_remaining = virt_to_phys_or_null(remaining_space);
 946        phys_max = virt_to_phys_or_null(max_variable_size);
 947
 948        status = efi_thunk(query_variable_info, attr, phys_storage,
 949                           phys_remaining, phys_max);
 950
 951        spin_unlock_irqrestore(&efi_runtime_lock, flags);
 952
 953        return status;
 954}
 955
 956static efi_status_t
 957efi_thunk_query_variable_info_nonblocking(u32 attr, u64 *storage_space,
 958                                          u64 *remaining_space,
 959                                          u64 *max_variable_size)
 960{
 961        efi_status_t status;
 962        u32 phys_storage, phys_remaining, phys_max;
 963        unsigned long flags;
 964
 965        if (efi.runtime_version < EFI_2_00_SYSTEM_TABLE_REVISION)
 966                return EFI_UNSUPPORTED;
 967
 968        if (!spin_trylock_irqsave(&efi_runtime_lock, flags))
 969                return EFI_NOT_READY;
 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
 983static efi_status_t
 984efi_thunk_query_capsule_caps(efi_capsule_header_t **capsules,
 985                             unsigned long count, u64 *max_size,
 986                             int *reset_type)
 987{
 988        /*
 989         * To properly support this function we would need to repackage
 990         * 'capsules' because the firmware doesn't understand 64-bit
 991         * pointers.
 992         */
 993        return EFI_UNSUPPORTED;
 994}
 995
 996void efi_thunk_runtime_setup(void)
 997{
 998        efi.get_time = efi_thunk_get_time;
 999        efi.set_time = efi_thunk_set_time;
1000        efi.get_wakeup_time = efi_thunk_get_wakeup_time;
1001        efi.set_wakeup_time = efi_thunk_set_wakeup_time;
1002        efi.get_variable = efi_thunk_get_variable;
1003        efi.get_next_variable = efi_thunk_get_next_variable;
1004        efi.set_variable = efi_thunk_set_variable;
1005        efi.set_variable_nonblocking = efi_thunk_set_variable_nonblocking;
1006        efi.get_next_high_mono_count = efi_thunk_get_next_high_mono_count;
1007        efi.reset_system = efi_thunk_reset_system;
1008        efi.query_variable_info = efi_thunk_query_variable_info;
1009        efi.query_variable_info_nonblocking = efi_thunk_query_variable_info_nonblocking;
1010        efi.update_capsule = efi_thunk_update_capsule;
1011        efi.query_capsule_caps = efi_thunk_query_capsule_caps;
1012}
1013#endif /* CONFIG_EFI_MIXED */
1014