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/tlbflush.h>
  43#include <asm/proto.h>
  44#include <asm/efi.h>
  45#include <asm/cacheflush.h>
  46#include <asm/fixmap.h>
  47#include <asm/realmode.h>
  48#include <asm/time.h>
  49#include <asm/pgalloc.h>
  50#include <asm/sev.h>
  51
  52/*
  53 * We allocate runtime services regions top-down, starting from -4G, i.e.
  54 * 0xffff_ffff_0000_0000 and limit EFI VA mapping space to 64G.
  55 */
  56static u64 efi_va = EFI_VA_START;
  57static struct mm_struct *efi_prev_mm;
  58
  59/*
  60 * We need our own copy of the higher levels of the page tables
  61 * because we want to avoid inserting EFI region mappings (EFI_VA_END
  62 * to EFI_VA_START) into the standard kernel page tables. Everything
  63 * else can be shared, see efi_sync_low_kernel_mappings().
  64 *
  65 * We don't want the pgd on the pgd_list and cannot use pgd_alloc() for the
  66 * allocation.
  67 */
  68int __init efi_alloc_page_tables(void)
  69{
  70        pgd_t *pgd, *efi_pgd;
  71        p4d_t *p4d;
  72        pud_t *pud;
  73        gfp_t gfp_mask;
  74
  75        gfp_mask = GFP_KERNEL | __GFP_ZERO;
  76        efi_pgd = (pgd_t *)__get_free_pages(gfp_mask, PGD_ALLOCATION_ORDER);
  77        if (!efi_pgd)
  78                goto fail;
  79
  80        pgd = efi_pgd + pgd_index(EFI_VA_END);
  81        p4d = p4d_alloc(&init_mm, pgd, EFI_VA_END);
  82        if (!p4d)
  83                goto free_pgd;
  84
  85        pud = pud_alloc(&init_mm, p4d, EFI_VA_END);
  86        if (!pud)
  87                goto free_p4d;
  88
  89        efi_mm.pgd = efi_pgd;
  90        mm_init_cpumask(&efi_mm);
  91        init_new_context(NULL, &efi_mm);
  92
  93        return 0;
  94
  95free_p4d:
  96        if (pgtable_l5_enabled())
  97                free_page((unsigned long)pgd_page_vaddr(*pgd));
  98free_pgd:
  99        free_pages((unsigned long)efi_pgd, PGD_ALLOCATION_ORDER);
 100fail:
 101        return -ENOMEM;
 102}
 103
 104/*
 105 * Add low kernel mappings for passing arguments to EFI functions.
 106 */
 107void efi_sync_low_kernel_mappings(void)
 108{
 109        unsigned num_entries;
 110        pgd_t *pgd_k, *pgd_efi;
 111        p4d_t *p4d_k, *p4d_efi;
 112        pud_t *pud_k, *pud_efi;
 113        pgd_t *efi_pgd = efi_mm.pgd;
 114
 115        pgd_efi = efi_pgd + pgd_index(PAGE_OFFSET);
 116        pgd_k = pgd_offset_k(PAGE_OFFSET);
 117
 118        num_entries = pgd_index(EFI_VA_END) - pgd_index(PAGE_OFFSET);
 119        memcpy(pgd_efi, pgd_k, sizeof(pgd_t) * num_entries);
 120
 121        pgd_efi = efi_pgd + pgd_index(EFI_VA_END);
 122        pgd_k = pgd_offset_k(EFI_VA_END);
 123        p4d_efi = p4d_offset(pgd_efi, 0);
 124        p4d_k = p4d_offset(pgd_k, 0);
 125
 126        num_entries = p4d_index(EFI_VA_END);
 127        memcpy(p4d_efi, p4d_k, sizeof(p4d_t) * num_entries);
 128
 129        /*
 130         * We share all the PUD entries apart from those that map the
 131         * EFI regions. Copy around them.
 132         */
 133        BUILD_BUG_ON((EFI_VA_START & ~PUD_MASK) != 0);
 134        BUILD_BUG_ON((EFI_VA_END & ~PUD_MASK) != 0);
 135
 136        p4d_efi = p4d_offset(pgd_efi, EFI_VA_END);
 137        p4d_k = p4d_offset(pgd_k, EFI_VA_END);
 138        pud_efi = pud_offset(p4d_efi, 0);
 139        pud_k = pud_offset(p4d_k, 0);
 140
 141        num_entries = pud_index(EFI_VA_END);
 142        memcpy(pud_efi, pud_k, sizeof(pud_t) * num_entries);
 143
 144        pud_efi = pud_offset(p4d_efi, EFI_VA_START);
 145        pud_k = pud_offset(p4d_k, EFI_VA_START);
 146
 147        num_entries = PTRS_PER_PUD - pud_index(EFI_VA_START);
 148        memcpy(pud_efi, pud_k, sizeof(pud_t) * num_entries);
 149}
 150
 151/*
 152 * Wrapper for slow_virt_to_phys() that handles NULL addresses.
 153 */
 154static inline phys_addr_t
 155virt_to_phys_or_null_size(void *va, unsigned long size)
 156{
 157        phys_addr_t pa;
 158
 159        if (!va)
 160                return 0;
 161
 162        if (virt_addr_valid(va))
 163                return virt_to_phys(va);
 164
 165        pa = slow_virt_to_phys(va);
 166
 167        /* check if the object crosses a page boundary */
 168        if (WARN_ON((pa ^ (pa + size - 1)) & PAGE_MASK))
 169                return 0;
 170
 171        return pa;
 172}
 173
 174#define virt_to_phys_or_null(addr)                              \
 175        virt_to_phys_or_null_size((addr), sizeof(*(addr)))
 176
 177int __init efi_setup_page_tables(unsigned long pa_memmap, unsigned num_pages)
 178{
 179        unsigned long pfn, text, pf, rodata;
 180        struct page *page;
 181        unsigned npages;
 182        pgd_t *pgd = efi_mm.pgd;
 183
 184        /*
 185         * It can happen that the physical address of new_memmap lands in memory
 186         * which is not mapped in the EFI page table. Therefore we need to go
 187         * and ident-map those pages containing the map before calling
 188         * phys_efi_set_virtual_address_map().
 189         */
 190        pfn = pa_memmap >> PAGE_SHIFT;
 191        pf = _PAGE_NX | _PAGE_RW | _PAGE_ENC;
 192        if (kernel_map_pages_in_pgd(pgd, pfn, pa_memmap, num_pages, pf)) {
 193                pr_err("Error ident-mapping new memmap (0x%lx)!\n", pa_memmap);
 194                return 1;
 195        }
 196
 197        /*
 198         * Certain firmware versions are way too sentimental and still believe
 199         * they are exclusive and unquestionable owners of the first physical page,
 200         * even though they explicitly mark it as EFI_CONVENTIONAL_MEMORY
 201         * (but then write-access it later during SetVirtualAddressMap()).
 202         *
 203         * Create a 1:1 mapping for this page, to avoid triple faults during early
 204         * boot with such firmware. We are free to hand this page to the BIOS,
 205         * as trim_bios_range() will reserve the first page and isolate it away
 206         * from memory allocators anyway.
 207         */
 208        if (kernel_map_pages_in_pgd(pgd, 0x0, 0x0, 1, pf)) {
 209                pr_err("Failed to create 1:1 mapping for the first page!\n");
 210                return 1;
 211        }
 212
 213        /*
 214         * When SEV-ES is active, the GHCB as set by the kernel will be used
 215         * by firmware. Create a 1:1 unencrypted mapping for each GHCB.
 216         */
 217        if (sev_es_efi_map_ghcbs(pgd)) {
 218                pr_err("Failed to create 1:1 mapping for the GHCBs!\n");
 219                return 1;
 220        }
 221
 222        /*
 223         * When making calls to the firmware everything needs to be 1:1
 224         * mapped and addressable with 32-bit pointers. Map the kernel
 225         * text and allocate a new stack because we can't rely on the
 226         * stack pointer being < 4GB.
 227         */
 228        if (!efi_is_mixed())
 229                return 0;
 230
 231        page = alloc_page(GFP_KERNEL|__GFP_DMA32);
 232        if (!page) {
 233                pr_err("Unable to allocate EFI runtime stack < 4GB\n");
 234                return 1;
 235        }
 236
 237        efi_mixed_mode_stack_pa = page_to_phys(page + 1); /* stack grows down */
 238
 239        npages = (_etext - _text) >> PAGE_SHIFT;
 240        text = __pa(_text);
 241        pfn = text >> PAGE_SHIFT;
 242
 243        pf = _PAGE_ENC;
 244        if (kernel_map_pages_in_pgd(pgd, pfn, text, npages, pf)) {
 245                pr_err("Failed to map kernel text 1:1\n");
 246                return 1;
 247        }
 248
 249        npages = (__end_rodata - __start_rodata) >> PAGE_SHIFT;
 250        rodata = __pa(__start_rodata);
 251        pfn = rodata >> PAGE_SHIFT;
 252
 253        pf = _PAGE_NX | _PAGE_ENC;
 254        if (kernel_map_pages_in_pgd(pgd, pfn, rodata, npages, pf)) {
 255                pr_err("Failed to map kernel rodata 1:1\n");
 256                return 1;
 257        }
 258
 259        return 0;
 260}
 261
 262static void __init __map_region(efi_memory_desc_t *md, u64 va)
 263{
 264        unsigned long flags = _PAGE_RW;
 265        unsigned long pfn;
 266        pgd_t *pgd = efi_mm.pgd;
 267
 268        /*
 269         * EFI_RUNTIME_SERVICES_CODE regions typically cover PE/COFF
 270         * executable images in memory that consist of both R-X and
 271         * RW- sections, so we cannot apply read-only or non-exec
 272         * permissions just yet. However, modern EFI systems provide
 273         * a memory attributes table that describes those sections
 274         * with the appropriate restricted permissions, which are
 275         * applied in efi_runtime_update_mappings() below. All other
 276         * regions can be mapped non-executable at this point, with
 277         * the exception of boot services code regions, but those will
 278         * be unmapped again entirely in efi_free_boot_services().
 279         */
 280        if (md->type != EFI_BOOT_SERVICES_CODE &&
 281            md->type != EFI_RUNTIME_SERVICES_CODE)
 282                flags |= _PAGE_NX;
 283
 284        if (!(md->attribute & EFI_MEMORY_WB))
 285                flags |= _PAGE_PCD;
 286
 287        if (sev_active() && md->type != EFI_MEMORY_MAPPED_IO)
 288                flags |= _PAGE_ENC;
 289
 290        pfn = md->phys_addr >> PAGE_SHIFT;
 291        if (kernel_map_pages_in_pgd(pgd, pfn, va, md->num_pages, flags))
 292                pr_warn("Error mapping PA 0x%llx -> VA 0x%llx!\n",
 293                           md->phys_addr, va);
 294}
 295
 296void __init efi_map_region(efi_memory_desc_t *md)
 297{
 298        unsigned long size = md->num_pages << PAGE_SHIFT;
 299        u64 pa = md->phys_addr;
 300
 301        /*
 302         * Make sure the 1:1 mappings are present as a catch-all for b0rked
 303         * firmware which doesn't update all internal pointers after switching
 304         * to virtual mode and would otherwise crap on us.
 305         */
 306        __map_region(md, md->phys_addr);
 307
 308        /*
 309         * Enforce the 1:1 mapping as the default virtual address when
 310         * booting in EFI mixed mode, because even though we may be
 311         * running a 64-bit kernel, the firmware may only be 32-bit.
 312         */
 313        if (efi_is_mixed()) {
 314                md->virt_addr = md->phys_addr;
 315                return;
 316        }
 317
 318        efi_va -= size;
 319
 320        /* Is PA 2M-aligned? */
 321        if (!(pa & (PMD_SIZE - 1))) {
 322                efi_va &= PMD_MASK;
 323        } else {
 324                u64 pa_offset = pa & (PMD_SIZE - 1);
 325                u64 prev_va = efi_va;
 326
 327                /* get us the same offset within this 2M page */
 328                efi_va = (efi_va & PMD_MASK) + pa_offset;
 329
 330                if (efi_va > prev_va)
 331                        efi_va -= PMD_SIZE;
 332        }
 333
 334        if (efi_va < EFI_VA_END) {
 335                pr_warn(FW_WARN "VA address range overflow!\n");
 336                return;
 337        }
 338
 339        /* Do the VA map */
 340        __map_region(md, efi_va);
 341        md->virt_addr = efi_va;
 342}
 343
 344/*
 345 * kexec kernel will use efi_map_region_fixed to map efi runtime memory ranges.
 346 * md->virt_addr is the original virtual address which had been mapped in kexec
 347 * 1st kernel.
 348 */
 349void __init efi_map_region_fixed(efi_memory_desc_t *md)
 350{
 351        __map_region(md, md->phys_addr);
 352        __map_region(md, md->virt_addr);
 353}
 354
 355void __init parse_efi_setup(u64 phys_addr, u32 data_len)
 356{
 357        efi_setup = phys_addr + sizeof(struct setup_data);
 358}
 359
 360static int __init efi_update_mappings(efi_memory_desc_t *md, unsigned long pf)
 361{
 362        unsigned long pfn;
 363        pgd_t *pgd = efi_mm.pgd;
 364        int err1, err2;
 365
 366        /* Update the 1:1 mapping */
 367        pfn = md->phys_addr >> PAGE_SHIFT;
 368        err1 = kernel_map_pages_in_pgd(pgd, pfn, md->phys_addr, md->num_pages, pf);
 369        if (err1) {
 370                pr_err("Error while updating 1:1 mapping PA 0x%llx -> VA 0x%llx!\n",
 371                           md->phys_addr, md->virt_addr);
 372        }
 373
 374        err2 = kernel_map_pages_in_pgd(pgd, pfn, md->virt_addr, md->num_pages, pf);
 375        if (err2) {
 376                pr_err("Error while updating VA mapping PA 0x%llx -> VA 0x%llx!\n",
 377                           md->phys_addr, md->virt_addr);
 378        }
 379
 380        return err1 || err2;
 381}
 382
 383static int __init efi_update_mem_attr(struct mm_struct *mm, efi_memory_desc_t *md)
 384{
 385        unsigned long pf = 0;
 386
 387        if (md->attribute & EFI_MEMORY_XP)
 388                pf |= _PAGE_NX;
 389
 390        if (!(md->attribute & EFI_MEMORY_RO))
 391                pf |= _PAGE_RW;
 392
 393        if (sev_active())
 394                pf |= _PAGE_ENC;
 395
 396        return efi_update_mappings(md, pf);
 397}
 398
 399void __init efi_runtime_update_mappings(void)
 400{
 401        efi_memory_desc_t *md;
 402
 403        /*
 404         * Use the EFI Memory Attribute Table for mapping permissions if it
 405         * exists, since it is intended to supersede EFI_PROPERTIES_TABLE.
 406         */
 407        if (efi_enabled(EFI_MEM_ATTR)) {
 408                efi_memattr_apply_permissions(NULL, efi_update_mem_attr);
 409                return;
 410        }
 411
 412        /*
 413         * EFI_MEMORY_ATTRIBUTES_TABLE is intended to replace
 414         * EFI_PROPERTIES_TABLE. So, use EFI_PROPERTIES_TABLE to update
 415         * permissions only if EFI_MEMORY_ATTRIBUTES_TABLE is not
 416         * published by the firmware. Even if we find a buggy implementation of
 417         * EFI_MEMORY_ATTRIBUTES_TABLE, don't fall back to
 418         * EFI_PROPERTIES_TABLE, because of the same reason.
 419         */
 420
 421        if (!efi_enabled(EFI_NX_PE_DATA))
 422                return;
 423
 424        for_each_efi_memory_desc(md) {
 425                unsigned long pf = 0;
 426
 427                if (!(md->attribute & EFI_MEMORY_RUNTIME))
 428                        continue;
 429
 430                if (!(md->attribute & EFI_MEMORY_WB))
 431                        pf |= _PAGE_PCD;
 432
 433                if ((md->attribute & EFI_MEMORY_XP) ||
 434                        (md->type == EFI_RUNTIME_SERVICES_DATA))
 435                        pf |= _PAGE_NX;
 436
 437                if (!(md->attribute & EFI_MEMORY_RO) &&
 438                        (md->type != EFI_RUNTIME_SERVICES_CODE))
 439                        pf |= _PAGE_RW;
 440
 441                if (sev_active())
 442                        pf |= _PAGE_ENC;
 443
 444                efi_update_mappings(md, pf);
 445        }
 446}
 447
 448void __init efi_dump_pagetable(void)
 449{
 450#ifdef CONFIG_EFI_PGT_DUMP
 451        ptdump_walk_pgd_level(NULL, &efi_mm);
 452#endif
 453}
 454
 455/*
 456 * Makes the calling thread switch to/from efi_mm context. Can be used
 457 * in a kernel thread and user context. Preemption needs to remain disabled
 458 * while the EFI-mm is borrowed. mmgrab()/mmdrop() is not used because the mm
 459 * can not change under us.
 460 * It should be ensured that there are no concurrent calls to this function.
 461 */
 462void efi_enter_mm(void)
 463{
 464        efi_prev_mm = current->active_mm;
 465        current->active_mm = &efi_mm;
 466        switch_mm(efi_prev_mm, &efi_mm, NULL);
 467}
 468
 469void efi_leave_mm(void)
 470{
 471        current->active_mm = efi_prev_mm;
 472        switch_mm(&efi_mm, efi_prev_mm, NULL);
 473}
 474
 475static DEFINE_SPINLOCK(efi_runtime_lock);
 476
 477/*
 478 * DS and ES contain user values.  We need to save them.
 479 * The 32-bit EFI code needs a valid DS, ES, and SS.  There's no
 480 * need to save the old SS: __KERNEL_DS is always acceptable.
 481 */
 482#define __efi_thunk(func, ...)                                          \
 483({                                                                      \
 484        unsigned short __ds, __es;                                      \
 485        efi_status_t ____s;                                             \
 486                                                                        \
 487        savesegment(ds, __ds);                                          \
 488        savesegment(es, __es);                                          \
 489                                                                        \
 490        loadsegment(ss, __KERNEL_DS);                                   \
 491        loadsegment(ds, __KERNEL_DS);                                   \
 492        loadsegment(es, __KERNEL_DS);                                   \
 493                                                                        \
 494        ____s = efi64_thunk(efi.runtime->mixed_mode.func, __VA_ARGS__); \
 495                                                                        \
 496        loadsegment(ds, __ds);                                          \
 497        loadsegment(es, __es);                                          \
 498                                                                        \
 499        ____s ^= (____s & BIT(31)) | (____s & BIT_ULL(31)) << 32;       \
 500        ____s;                                                          \
 501})
 502
 503/*
 504 * Switch to the EFI page tables early so that we can access the 1:1
 505 * runtime services mappings which are not mapped in any other page
 506 * tables.
 507 *
 508 * Also, disable interrupts because the IDT points to 64-bit handlers,
 509 * which aren't going to function correctly when we switch to 32-bit.
 510 */
 511#define efi_thunk(func...)                                              \
 512({                                                                      \
 513        efi_status_t __s;                                               \
 514                                                                        \
 515        arch_efi_call_virt_setup();                                     \
 516                                                                        \
 517        __s = __efi_thunk(func);                                        \
 518                                                                        \
 519        arch_efi_call_virt_teardown();                                  \
 520                                                                        \
 521        __s;                                                            \
 522})
 523
 524static efi_status_t __init __no_sanitize_address
 525efi_thunk_set_virtual_address_map(unsigned long memory_map_size,
 526                                  unsigned long descriptor_size,
 527                                  u32 descriptor_version,
 528                                  efi_memory_desc_t *virtual_map)
 529{
 530        efi_status_t status;
 531        unsigned long flags;
 532
 533        efi_sync_low_kernel_mappings();
 534        local_irq_save(flags);
 535
 536        efi_enter_mm();
 537
 538        status = __efi_thunk(set_virtual_address_map, memory_map_size,
 539                             descriptor_size, descriptor_version, virtual_map);
 540
 541        efi_leave_mm();
 542        local_irq_restore(flags);
 543
 544        return status;
 545}
 546
 547static efi_status_t efi_thunk_get_time(efi_time_t *tm, efi_time_cap_t *tc)
 548{
 549        return EFI_UNSUPPORTED;
 550}
 551
 552static efi_status_t efi_thunk_set_time(efi_time_t *tm)
 553{
 554        return EFI_UNSUPPORTED;
 555}
 556
 557static efi_status_t
 558efi_thunk_get_wakeup_time(efi_bool_t *enabled, efi_bool_t *pending,
 559                          efi_time_t *tm)
 560{
 561        return EFI_UNSUPPORTED;
 562}
 563
 564static efi_status_t
 565efi_thunk_set_wakeup_time(efi_bool_t enabled, efi_time_t *tm)
 566{
 567        return EFI_UNSUPPORTED;
 568}
 569
 570static unsigned long efi_name_size(efi_char16_t *name)
 571{
 572        return ucs2_strsize(name, EFI_VAR_NAME_LEN) + 1;
 573}
 574
 575static efi_status_t
 576efi_thunk_get_variable(efi_char16_t *name, efi_guid_t *vendor,
 577                       u32 *attr, unsigned long *data_size, void *data)
 578{
 579        u8 buf[24] __aligned(8);
 580        efi_guid_t *vnd = PTR_ALIGN((efi_guid_t *)buf, sizeof(*vnd));
 581        efi_status_t status;
 582        u32 phys_name, phys_vendor, phys_attr;
 583        u32 phys_data_size, phys_data;
 584        unsigned long flags;
 585
 586        spin_lock_irqsave(&efi_runtime_lock, flags);
 587
 588        *vnd = *vendor;
 589
 590        phys_data_size = virt_to_phys_or_null(data_size);
 591        phys_vendor = virt_to_phys_or_null(vnd);
 592        phys_name = virt_to_phys_or_null_size(name, efi_name_size(name));
 593        phys_attr = virt_to_phys_or_null(attr);
 594        phys_data = virt_to_phys_or_null_size(data, *data_size);
 595
 596        if (!phys_name || (data && !phys_data))
 597                status = EFI_INVALID_PARAMETER;
 598        else
 599                status = efi_thunk(get_variable, phys_name, phys_vendor,
 600                                   phys_attr, phys_data_size, phys_data);
 601
 602        spin_unlock_irqrestore(&efi_runtime_lock, flags);
 603
 604        return status;
 605}
 606
 607static efi_status_t
 608efi_thunk_set_variable(efi_char16_t *name, efi_guid_t *vendor,
 609                       u32 attr, unsigned long data_size, void *data)
 610{
 611        u8 buf[24] __aligned(8);
 612        efi_guid_t *vnd = PTR_ALIGN((efi_guid_t *)buf, sizeof(*vnd));
 613        u32 phys_name, phys_vendor, phys_data;
 614        efi_status_t status;
 615        unsigned long flags;
 616
 617        spin_lock_irqsave(&efi_runtime_lock, flags);
 618
 619        *vnd = *vendor;
 620
 621        phys_name = virt_to_phys_or_null_size(name, efi_name_size(name));
 622        phys_vendor = virt_to_phys_or_null(vnd);
 623        phys_data = virt_to_phys_or_null_size(data, data_size);
 624
 625        if (!phys_name || (data && !phys_data))
 626                status = EFI_INVALID_PARAMETER;
 627        else
 628                status = efi_thunk(set_variable, phys_name, phys_vendor,
 629                                   attr, data_size, phys_data);
 630
 631        spin_unlock_irqrestore(&efi_runtime_lock, flags);
 632
 633        return status;
 634}
 635
 636static efi_status_t
 637efi_thunk_set_variable_nonblocking(efi_char16_t *name, efi_guid_t *vendor,
 638                                   u32 attr, unsigned long data_size,
 639                                   void *data)
 640{
 641        u8 buf[24] __aligned(8);
 642        efi_guid_t *vnd = PTR_ALIGN((efi_guid_t *)buf, sizeof(*vnd));
 643        u32 phys_name, phys_vendor, phys_data;
 644        efi_status_t status;
 645        unsigned long flags;
 646
 647        if (!spin_trylock_irqsave(&efi_runtime_lock, flags))
 648                return EFI_NOT_READY;
 649
 650        *vnd = *vendor;
 651
 652        phys_name = virt_to_phys_or_null_size(name, efi_name_size(name));
 653        phys_vendor = virt_to_phys_or_null(vnd);
 654        phys_data = virt_to_phys_or_null_size(data, data_size);
 655
 656        if (!phys_name || (data && !phys_data))
 657                status = EFI_INVALID_PARAMETER;
 658        else
 659                status = efi_thunk(set_variable, phys_name, phys_vendor,
 660                                   attr, data_size, phys_data);
 661
 662        spin_unlock_irqrestore(&efi_runtime_lock, flags);
 663
 664        return status;
 665}
 666
 667static efi_status_t
 668efi_thunk_get_next_variable(unsigned long *name_size,
 669                            efi_char16_t *name,
 670                            efi_guid_t *vendor)
 671{
 672        u8 buf[24] __aligned(8);
 673        efi_guid_t *vnd = PTR_ALIGN((efi_guid_t *)buf, sizeof(*vnd));
 674        efi_status_t status;
 675        u32 phys_name_size, phys_name, phys_vendor;
 676        unsigned long flags;
 677
 678        spin_lock_irqsave(&efi_runtime_lock, flags);
 679
 680        *vnd = *vendor;
 681
 682        phys_name_size = virt_to_phys_or_null(name_size);
 683        phys_vendor = virt_to_phys_or_null(vnd);
 684        phys_name = virt_to_phys_or_null_size(name, *name_size);
 685
 686        if (!phys_name)
 687                status = EFI_INVALID_PARAMETER;
 688        else
 689                status = efi_thunk(get_next_variable, phys_name_size,
 690                                   phys_name, phys_vendor);
 691
 692        spin_unlock_irqrestore(&efi_runtime_lock, flags);
 693
 694        *vendor = *vnd;
 695        return status;
 696}
 697
 698static efi_status_t
 699efi_thunk_get_next_high_mono_count(u32 *count)
 700{
 701        return EFI_UNSUPPORTED;
 702}
 703
 704static void
 705efi_thunk_reset_system(int reset_type, efi_status_t status,
 706                       unsigned long data_size, efi_char16_t *data)
 707{
 708        u32 phys_data;
 709        unsigned long flags;
 710
 711        spin_lock_irqsave(&efi_runtime_lock, flags);
 712
 713        phys_data = virt_to_phys_or_null_size(data, data_size);
 714
 715        efi_thunk(reset_system, reset_type, status, data_size, phys_data);
 716
 717        spin_unlock_irqrestore(&efi_runtime_lock, flags);
 718}
 719
 720static efi_status_t
 721efi_thunk_update_capsule(efi_capsule_header_t **capsules,
 722                         unsigned long count, unsigned long sg_list)
 723{
 724        /*
 725         * To properly support this function we would need to repackage
 726         * 'capsules' because the firmware doesn't understand 64-bit
 727         * pointers.
 728         */
 729        return EFI_UNSUPPORTED;
 730}
 731
 732static efi_status_t
 733efi_thunk_query_variable_info(u32 attr, u64 *storage_space,
 734                              u64 *remaining_space,
 735                              u64 *max_variable_size)
 736{
 737        efi_status_t status;
 738        u32 phys_storage, phys_remaining, phys_max;
 739        unsigned long flags;
 740
 741        if (efi.runtime_version < EFI_2_00_SYSTEM_TABLE_REVISION)
 742                return EFI_UNSUPPORTED;
 743
 744        spin_lock_irqsave(&efi_runtime_lock, flags);
 745
 746        phys_storage = virt_to_phys_or_null(storage_space);
 747        phys_remaining = virt_to_phys_or_null(remaining_space);
 748        phys_max = virt_to_phys_or_null(max_variable_size);
 749
 750        status = efi_thunk(query_variable_info, attr, phys_storage,
 751                           phys_remaining, phys_max);
 752
 753        spin_unlock_irqrestore(&efi_runtime_lock, flags);
 754
 755        return status;
 756}
 757
 758static efi_status_t
 759efi_thunk_query_variable_info_nonblocking(u32 attr, u64 *storage_space,
 760                                          u64 *remaining_space,
 761                                          u64 *max_variable_size)
 762{
 763        efi_status_t status;
 764        u32 phys_storage, phys_remaining, phys_max;
 765        unsigned long flags;
 766
 767        if (efi.runtime_version < EFI_2_00_SYSTEM_TABLE_REVISION)
 768                return EFI_UNSUPPORTED;
 769
 770        if (!spin_trylock_irqsave(&efi_runtime_lock, flags))
 771                return EFI_NOT_READY;
 772
 773        phys_storage = virt_to_phys_or_null(storage_space);
 774        phys_remaining = virt_to_phys_or_null(remaining_space);
 775        phys_max = virt_to_phys_or_null(max_variable_size);
 776
 777        status = efi_thunk(query_variable_info, attr, phys_storage,
 778                           phys_remaining, phys_max);
 779
 780        spin_unlock_irqrestore(&efi_runtime_lock, flags);
 781
 782        return status;
 783}
 784
 785static efi_status_t
 786efi_thunk_query_capsule_caps(efi_capsule_header_t **capsules,
 787                             unsigned long count, u64 *max_size,
 788                             int *reset_type)
 789{
 790        /*
 791         * To properly support this function we would need to repackage
 792         * 'capsules' because the firmware doesn't understand 64-bit
 793         * pointers.
 794         */
 795        return EFI_UNSUPPORTED;
 796}
 797
 798void __init efi_thunk_runtime_setup(void)
 799{
 800        if (!IS_ENABLED(CONFIG_EFI_MIXED))
 801                return;
 802
 803        efi.get_time = efi_thunk_get_time;
 804        efi.set_time = efi_thunk_set_time;
 805        efi.get_wakeup_time = efi_thunk_get_wakeup_time;
 806        efi.set_wakeup_time = efi_thunk_set_wakeup_time;
 807        efi.get_variable = efi_thunk_get_variable;
 808        efi.get_next_variable = efi_thunk_get_next_variable;
 809        efi.set_variable = efi_thunk_set_variable;
 810        efi.set_variable_nonblocking = efi_thunk_set_variable_nonblocking;
 811        efi.get_next_high_mono_count = efi_thunk_get_next_high_mono_count;
 812        efi.reset_system = efi_thunk_reset_system;
 813        efi.query_variable_info = efi_thunk_query_variable_info;
 814        efi.query_variable_info_nonblocking = efi_thunk_query_variable_info_nonblocking;
 815        efi.update_capsule = efi_thunk_update_capsule;
 816        efi.query_capsule_caps = efi_thunk_query_capsule_caps;
 817}
 818
 819efi_status_t __init __no_sanitize_address
 820efi_set_virtual_address_map(unsigned long memory_map_size,
 821                            unsigned long descriptor_size,
 822                            u32 descriptor_version,
 823                            efi_memory_desc_t *virtual_map,
 824                            unsigned long systab_phys)
 825{
 826        const efi_system_table_t *systab = (efi_system_table_t *)systab_phys;
 827        efi_status_t status;
 828        unsigned long flags;
 829
 830        if (efi_is_mixed())
 831                return efi_thunk_set_virtual_address_map(memory_map_size,
 832                                                         descriptor_size,
 833                                                         descriptor_version,
 834                                                         virtual_map);
 835        efi_enter_mm();
 836
 837        efi_fpu_begin();
 838
 839        /* Disable interrupts around EFI calls: */
 840        local_irq_save(flags);
 841        status = efi_call(efi.runtime->set_virtual_address_map,
 842                          memory_map_size, descriptor_size,
 843                          descriptor_version, virtual_map);
 844        local_irq_restore(flags);
 845
 846        efi_fpu_end();
 847
 848        /* grab the virtually remapped EFI runtime services table pointer */
 849        efi.runtime = READ_ONCE(systab->runtime);
 850
 851        efi_leave_mm();
 852
 853        return status;
 854}
 855