arch/x86/platform/efi/efi.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * Common EFI (Extensible Firmware Interface) support functions
   4  * Based on Extensible Firmware Interface Specification version 1.0
   5  *
   6  * Copyright (C) 1999 VA Linux Systems
   7  * Copyright (C) 1999 Walt Drummond <drummond@valinux.com>
   8  * Copyright (C) 1999-2002 Hewlett-Packard Co.
   9  *      David Mosberger-Tang <davidm@hpl.hp.com>
  10  *      Stephane Eranian <eranian@hpl.hp.com>
  11  * Copyright (C) 2005-2008 Intel Co.
  12  *      Fenghua Yu <fenghua.yu@intel.com>
  13  *      Bibo Mao <bibo.mao@intel.com>
  14  *      Chandramouli Narayanan <mouli@linux.intel.com>
  15  *      Huang Ying <ying.huang@intel.com>
  16  * Copyright (C) 2013 SuSE Labs
  17  *      Borislav Petkov <bp@suse.de> - runtime services VA mapping
  18  *
  19  * Copied from efi_32.c to eliminate the duplicated code between EFI
  20  * 32/64 support code. --ying 2007-10-26
  21  *
  22  * All EFI Runtime Services are not implemented yet as EFI only
  23  * supports physical mode addressing on SoftSDV. This is to be fixed
  24  * in a future version.  --drummond 1999-07-20
  25  *
  26  * Implemented EFI runtime services and virtual mode calls.  --davidm
  27  *
  28  * Goutham Rao: <goutham.rao@intel.com>
  29  *      Skip non-WB memory and ignore empty memory ranges.
  30  */
  31
  32 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  33
  34 #include <linux/kernel.h>
  35 #include <linux/init.h>
  36 #include <linux/efi.h>
  37 #include <linux/efi-bgrt.h>
  38 #include <linux/export.h>
  39 #include <linux/memblock.h>
  40 #include <linux/slab.h>
  41 #include <linux/spinlock.h>
  42 #include <linux/uaccess.h>
  43 #include <linux/time.h>
  44 #include <linux/io.h>
  45 #include <linux/reboot.h>
  46 #include <linux/bcd.h>
  47
  48 #include <asm/setup.h>
  49 #include <asm/efi.h>
  50 #include <asm/e820/api.h>
  51 #include <asm/time.h>
  52 #include <asm/set_memory.h>
  53 #include <asm/tlbflush.h>
  54 #include <asm/x86_init.h>
  55 #include <asm/uv/uv.h>
  56
  57 static efi_system_table_t efi_systab __initdata;
  58 static u64 efi_systab_phys __initdata;
  59
  60 static efi_config_table_type_t arch_tables[] __initdata = {
  61 #ifdef CONFIG_X86_UV
  62         {UV_SYSTEM_TABLE_GUID, "UVsystab", &uv_systab_phys},
  63 #endif
  64         {NULL_GUID, NULL, NULL},
  65 };
  66
  67 static const unsigned long * const efi_tables[] = {
  68         &efi.mps,
  69         &efi.acpi,
  70         &efi.acpi20,
  71         &efi.smbios,
  72         &efi.smbios3,
  73         &efi.boot_info,
  74         &efi.hcdp,
  75         &efi.uga,
  76 #ifdef CONFIG_X86_UV
  77         &uv_systab_phys,
  78 #endif
  79         &efi.fw_vendor,
  80         &efi.runtime,
  81         &efi.config_table,
  82         &efi.esrt,
  83         &efi.properties_table,
  84         &efi.mem_attr_table,
  85 #ifdef CONFIG_EFI_RCI2_TABLE
  86         &rci2_table_phys,
  87 #endif
  88 };
  89
  90 u64 efi_setup;          /* efi setup_data physical address */
  91
  92 static int add_efi_memmap __initdata;
  93 static int __init setup_add_efi_memmap(char *arg)
  94 {
  95         add_efi_memmap = 1;
  96         return 0;
  97 }
  98 early_param("add_efi_memmap", setup_add_efi_memmap);
  99
 100 void __init efi_find_mirror(void)
 101 {
 102         efi_memory_desc_t *md;
 103         u64 mirror_size = 0, total_size = 0;
 104
 105         if (!efi_enabled(EFI_MEMMAP))
 106                 return;
 107
 108         for_each_efi_memory_desc(md) {
 109                 unsigned long long start = md->phys_addr;
 110                 unsigned long long size = md->num_pages << EFI_PAGE_SHIFT;
 111
 112                 total_size += size;
 113                 if (md->attribute & EFI_MEMORY_MORE_RELIABLE) {
 114                         memblock_mark_mirror(start, size);
 115                         mirror_size += size;
 116                 }
 117         }
 118         if (mirror_size)
 119                 pr_info("Memory: %lldM/%lldM mirrored memory\n",
 120                         mirror_size>>20, total_size>>20);
 121 }
 122
 123 /*
 124  * Tell the kernel about the EFI memory map.  This might include
 125  * more than the max 128 entries that can fit in the passed in e820
 126  * legacy (zeropage) memory map, but the kernel's e820 table can hold
 127  * E820_MAX_ENTRIES.
 128  */
 129
 130 static void __init do_add_efi_memmap(void)
 131 {
 132         efi_memory_desc_t *md;
 133
 134         if (!efi_enabled(EFI_MEMMAP))
 135                 return;
 136
 137         for_each_efi_memory_desc(md) {
 138                 unsigned long long start = md->phys_addr;
 139                 unsigned long long size = md->num_pages << EFI_PAGE_SHIFT;
 140                 int e820_type;
 141
 142                 switch (md->type) {
 143                 case EFI_LOADER_CODE:
 144                 case EFI_LOADER_DATA:
 145                 case EFI_BOOT_SERVICES_CODE:
 146                 case EFI_BOOT_SERVICES_DATA:
 147                 case EFI_CONVENTIONAL_MEMORY:
 148                         if (efi_soft_reserve_enabled()
 149                             && (md->attribute & EFI_MEMORY_SP))
 150                                 e820_type = E820_TYPE_SOFT_RESERVED;
 151                         else if (md->attribute & EFI_MEMORY_WB)
 152                                 e820_type = E820_TYPE_RAM;
 153                         else
 154                                 e820_type = E820_TYPE_RESERVED;
 155                         break;
 156                 case EFI_ACPI_RECLAIM_MEMORY:
 157                         e820_type = E820_TYPE_ACPI;
 158                         break;
 159                 case EFI_ACPI_MEMORY_NVS:
 160                         e820_type = E820_TYPE_NVS;
 161                         break;
 162                 case EFI_UNUSABLE_MEMORY:
 163                         e820_type = E820_TYPE_UNUSABLE;
 164                         break;
 165                 case EFI_PERSISTENT_MEMORY:
 166                         e820_type = E820_TYPE_PMEM;
 167                         break;
 168                 default:
 169                         /*
 170                          * EFI_RESERVED_TYPE EFI_RUNTIME_SERVICES_CODE
 171                          * EFI_RUNTIME_SERVICES_DATA EFI_MEMORY_MAPPED_IO
 172                          * EFI_MEMORY_MAPPED_IO_PORT_SPACE EFI_PAL_CODE
 173                          */
 174                         e820_type = E820_TYPE_RESERVED;
 175                         break;
 176                 }
 177
 178                 e820__range_add(start, size, e820_type);
 179         }
 180         e820__update_table(e820_table);
 181 }
 182
 183 /*
 184  * Given add_efi_memmap defaults to 0 and there there is no alternative
 185  * e820 mechanism for soft-reserved memory, import the full EFI memory
 186  * map if soft reservations are present and enabled. Otherwise, the
 187  * mechanism to disable the kernel's consideration of EFI_MEMORY_SP is
 188  * the efi=nosoftreserve option.
 189  */
 190 static bool do_efi_soft_reserve(void)
 191 {
 192         efi_memory_desc_t *md;
 193
 194         if (!efi_enabled(EFI_MEMMAP))
 195                 return false;
 196
 197         if (!efi_soft_reserve_enabled())
 198                 return false;
 199
 200         for_each_efi_memory_desc(md)
 201                 if (md->type == EFI_CONVENTIONAL_MEMORY &&
 202                     (md->attribute & EFI_MEMORY_SP))
 203                         return true;
 204         return false;
 205 }
 206
 207 int __init efi_memblock_x86_reserve_range(void)
 208 {
 209         struct efi_info *e = &boot_params.efi_info;
 210         struct efi_memory_map_data data;
 211         phys_addr_t pmap;
 212         int rv;
 213
 214         if (efi_enabled(EFI_PARAVIRT))
 215                 return 0;
 216
 217 #ifdef CONFIG_X86_32
 218         /* Can't handle data above 4GB at this time */
 219         if (e->efi_memmap_hi) {
 220                 pr_err("Memory map is above 4GB, disabling EFI.\n");
 221                 return -EINVAL;
 222         }
 223         pmap =  e->efi_memmap;
 224 #else
 225         pmap = (e->efi_memmap | ((__u64)e->efi_memmap_hi << 32));
 226 #endif
 227         data.phys_map           = pmap;
 228         data.size               = e->efi_memmap_size;
 229         data.desc_size          = e->efi_memdesc_size;
 230         data.desc_version       = e->efi_memdesc_version;
 231
 232         rv = efi_memmap_init_early(&data);
 233         if (rv)
 234                 return rv;
 235
 236         if (add_efi_memmap || do_efi_soft_reserve())
 237                 do_add_efi_memmap();
 238
 239         efi_fake_memmap_early();
 240
 241         WARN(efi.memmap.desc_version != 1,
 242              "Unexpected EFI_MEMORY_DESCRIPTOR version %ld",
 243              efi.memmap.desc_version);
 244
 245         memblock_reserve(pmap, efi.memmap.nr_map * efi.memmap.desc_size);
 246
 247         return 0;
 248 }
 249
 250 #define OVERFLOW_ADDR_SHIFT     (64 - EFI_PAGE_SHIFT)
 251 #define OVERFLOW_ADDR_MASK      (U64_MAX << OVERFLOW_ADDR_SHIFT)
 252 #define U64_HIGH_BIT            (~(U64_MAX >> 1))
 253
 254 static bool __init efi_memmap_entry_valid(const efi_memory_desc_t *md, int i)
 255 {
 256         u64 end = (md->num_pages << EFI_PAGE_SHIFT) + md->phys_addr - 1;
 257         u64 end_hi = 0;
 258         char buf[64];
 259
 260         if (md->num_pages == 0) {
 261                 end = 0;
 262         } else if (md->num_pages > EFI_PAGES_MAX ||
 263                    EFI_PAGES_MAX - md->num_pages <
 264                    (md->phys_addr >> EFI_PAGE_SHIFT)) {
 265                 end_hi = (md->num_pages & OVERFLOW_ADDR_MASK)
 266                         >> OVERFLOW_ADDR_SHIFT;
 267
 268                 if ((md->phys_addr & U64_HIGH_BIT) && !(end & U64_HIGH_BIT))
 269                         end_hi += 1;
 270         } else {
 271                 return true;
 272         }
 273
 274         pr_warn_once(FW_BUG "Invalid EFI memory map entries:\n");
 275
 276         if (end_hi) {
 277                 pr_warn("mem%02u: %s range=[0x%016llx-0x%llx%016llx] (invalid)\n",
 278                         i, efi_md_typeattr_format(buf, sizeof(buf), md),
 279                         md->phys_addr, end_hi, end);
 280         } else {
 281                 pr_warn("mem%02u: %s range=[0x%016llx-0x%016llx] (invalid)\n",
 282                         i, efi_md_typeattr_format(buf, sizeof(buf), md),
 283                         md->phys_addr, end);
 284         }
 285         return false;
 286 }
 287
 288 static void __init efi_clean_memmap(void)
 289 {
 290         efi_memory_desc_t *out = efi.memmap.map;
 291         const efi_memory_desc_t *in = out;
 292         const efi_memory_desc_t *end = efi.memmap.map_end;
 293         int i, n_removal;
 294
 295         for (i = n_removal = 0; in < end; i++) {
 296                 if (efi_memmap_entry_valid(in, i)) {
 297                         if (out != in)
 298                                 memcpy(out, in, efi.memmap.desc_size);
 299                         out = (void *)out + efi.memmap.desc_size;
 300                 } else {
 301                         n_removal++;
 302                 }
 303                 in = (void *)in + efi.memmap.desc_size;
 304         }
 305
 306         if (n_removal > 0) {
 307                 struct efi_memory_map_data data = {
 308                         .phys_map = efi.memmap.phys_map,
 309                         .desc_version = efi.memmap.desc_version,
 310                         .desc_size = efi.memmap.desc_size,
 311                         .size = efi.memmap.desc_size * (efi.memmap.nr_map - n_removal),
 312                         .flags = 0,
 313                 };
 314
 315                 pr_warn("Removing %d invalid memory map entries.\n", n_removal);
 316                 efi_memmap_install(&data);
 317         }
 318 }
 319
 320 void __init efi_print_memmap(void)
 321 {
 322         efi_memory_desc_t *md;
 323         int i = 0;
 324
 325         for_each_efi_memory_desc(md) {
 326                 char buf[64];
 327
 328                 pr_info("mem%02u: %s range=[0x%016llx-0x%016llx] (%lluMB)\n",
 329                         i++, efi_md_typeattr_format(buf, sizeof(buf), md),
 330                         md->phys_addr,
 331                         md->phys_addr + (md->num_pages << EFI_PAGE_SHIFT) - 1,
 332                         (md->num_pages >> (20 - EFI_PAGE_SHIFT)));
 333         }
 334 }
 335
 336 static int __init efi_systab_init(u64 phys)
 337 {
 338         int size = efi_enabled(EFI_64BIT) ? sizeof(efi_system_table_64_t)
 339                                           : sizeof(efi_system_table_32_t);
 340         bool over4g = false;
 341         void *p;
 342
 343         p = early_memremap_ro(phys, size);
 344         if (p == NULL) {
 345                 pr_err("Couldn't map the system table!\n");
 346                 return -ENOMEM;
 347         }
 348
 349         if (efi_enabled(EFI_64BIT)) {
 350                 const efi_system_table_64_t *systab64 = p;
 351
 352                 efi_systab.hdr                  = systab64->hdr;
 353                 efi_systab.fw_vendor            = systab64->fw_vendor;
 354                 efi_systab.fw_revision          = systab64->fw_revision;
 355                 efi_systab.con_in_handle        = systab64->con_in_handle;
 356                 efi_systab.con_in               = systab64->con_in;
 357                 efi_systab.con_out_handle       = systab64->con_out_handle;
 358                 efi_systab.con_out              = (void *)(unsigned long)systab64->con_out;
 359                 efi_systab.stderr_handle        = systab64->stderr_handle;
 360                 efi_systab.stderr               = systab64->stderr;
 361                 efi_systab.runtime              = (void *)(unsigned long)systab64->runtime;
 362                 efi_systab.boottime             = (void *)(unsigned long)systab64->boottime;
 363                 efi_systab.nr_tables            = systab64->nr_tables;
 364                 efi_systab.tables               = systab64->tables;
 365
 366                 over4g = systab64->con_in_handle        > U32_MAX ||
 367                          systab64->con_in               > U32_MAX ||
 368                          systab64->con_out_handle       > U32_MAX ||
 369                          systab64->con_out              > U32_MAX ||
 370                          systab64->stderr_handle        > U32_MAX ||
 371                          systab64->stderr               > U32_MAX ||
 372                          systab64->boottime             > U32_MAX;
 373
 374                 if (efi_setup) {
 375                         struct efi_setup_data *data;
 376
 377                         data = early_memremap_ro(efi_setup, sizeof(*data));
 378                         if (!data) {
 379                                 early_memunmap(p, size);
 380                                 return -ENOMEM;
 381                         }
 382
 383                         efi_systab.fw_vendor    = (unsigned long)data->fw_vendor;
 384                         efi_systab.runtime      = (void *)(unsigned long)data->runtime;
 385                         efi_systab.tables       = (unsigned long)data->tables;
 386
 387                         over4g |= data->fw_vendor       > U32_MAX ||
 388                                   data->runtime         > U32_MAX ||
 389                                   data->tables          > U32_MAX;
 390
 391                         early_memunmap(data, sizeof(*data));
 392                 } else {
 393                         over4g |= systab64->fw_vendor   > U32_MAX ||
 394                                   systab64->runtime     > U32_MAX ||
 395                                   systab64->tables      > U32_MAX;
 396                 }
 397         } else {
 398                 const efi_system_table_32_t *systab32 = p;
 399
 400                 efi_systab.hdr                  = systab32->hdr;
 401                 efi_systab.fw_vendor            = systab32->fw_vendor;
 402                 efi_systab.fw_revision          = systab32->fw_revision;
 403                 efi_systab.con_in_handle        = systab32->con_in_handle;
 404                 efi_systab.con_in               = systab32->con_in;
 405                 efi_systab.con_out_handle       = systab32->con_out_handle;
 406                 efi_systab.con_out              = (void *)(unsigned long)systab32->con_out;
 407                 efi_systab.stderr_handle        = systab32->stderr_handle;
 408                 efi_systab.stderr               = systab32->stderr;
 409                 efi_systab.runtime              = (void *)(unsigned long)systab32->runtime;
 410                 efi_systab.boottime             = (void *)(unsigned long)systab32->boottime;
 411                 efi_systab.nr_tables            = systab32->nr_tables;
 412                 efi_systab.tables               = systab32->tables;
 413         }
 414
 415         early_memunmap(p, size);
 416
 417         if (IS_ENABLED(CONFIG_X86_32) && over4g) {
 418                 pr_err("EFI data located above 4GB, disabling EFI.\n");
 419                 return -EINVAL;
 420         }
 421
 422         efi.systab = &efi_systab;
 423
 424         /*
 425          * Verify the EFI Table
 426          */
 427         if (efi.systab->hdr.signature != EFI_SYSTEM_TABLE_SIGNATURE) {
 428                 pr_err("System table signature incorrect!\n");
 429                 return -EINVAL;
 430         }
 431         if ((efi.systab->hdr.revision >> 16) == 0)
 432                 pr_err("Warning: System table version %d.%02d, expected 1.00 or greater!\n",
 433                        efi.systab->hdr.revision >> 16,
 434                        efi.systab->hdr.revision & 0xffff);
 435
 436         return 0;
 437 }
 438
 439 void __init efi_init(void)
 440 {
 441         efi_char16_t *c16;
 442         char vendor[100] = "unknown";
 443         int i = 0;
 444
 445         if (IS_ENABLED(CONFIG_X86_32) &&
 446             (boot_params.efi_info.efi_systab_hi ||
 447              boot_params.efi_info.efi_memmap_hi)) {
 448                 pr_info("Table located above 4GB, disabling EFI.\n");
 449                 return;
 450         }
 451
 452         efi_systab_phys = boot_params.efi_info.efi_systab |
 453                           ((__u64)boot_params.efi_info.efi_systab_hi << 32);
 454
 455         if (efi_systab_init(efi_systab_phys))
 456                 return;
 457
 458         efi.config_table = (unsigned long)efi.systab->tables;
 459         efi.fw_vendor    = (unsigned long)efi.systab->fw_vendor;
 460         efi.runtime      = (unsigned long)efi.systab->runtime;
 461
 462         /*
 463          * Show what we know for posterity
 464          */
 465         c16 = early_memremap_ro(efi.systab->fw_vendor,
 466                                 sizeof(vendor) * sizeof(efi_char16_t));
 467         if (c16) {
 468                 for (i = 0; i < sizeof(vendor) - 1 && c16[i]; ++i)
 469                         vendor[i] = c16[i];
 470                 vendor[i] = '\0';
 471                 early_memunmap(c16, sizeof(vendor) * sizeof(efi_char16_t));
 472         } else {
 473                 pr_err("Could not map the firmware vendor!\n");
 474         }
 475
 476         pr_info("EFI v%u.%.02u by %s\n",
 477                 efi.systab->hdr.revision >> 16,
 478                 efi.systab->hdr.revision & 0xffff, vendor);
 479
 480         if (efi_reuse_config(efi.systab->tables, efi.systab->nr_tables))
 481                 return;
 482
 483         if (efi_config_init(arch_tables))
 484                 return;
 485
 486         /*
 487          * Note: We currently don't support runtime services on an EFI
 488          * that doesn't match the kernel 32/64-bit mode.
 489          */
 490
 491         if (!efi_runtime_supported())
 492                 pr_info("No EFI runtime due to 32/64-bit mismatch with kernel\n");
 493
 494         if (!efi_runtime_supported() || efi_runtime_disabled()) {
 495                 efi_memmap_unmap();
 496                 return;
 497         }
 498
 499         set_bit(EFI_RUNTIME_SERVICES, &efi.flags);
 500         efi_clean_memmap();
 501
 502         if (efi_enabled(EFI_DBG))
 503                 efi_print_memmap();
 504 }
 505
 506 #if defined(CONFIG_X86_32) || defined(CONFIG_X86_UV)
 507
 508 void __init efi_set_executable(efi_memory_desc_t *md, bool executable)
 509 {
 510         u64 addr, npages;
 511
 512         addr = md->virt_addr;
 513         npages = md->num_pages;
 514
 515         memrange_efi_to_native(&addr, &npages);
 516
 517         if (executable)
 518                 set_memory_x(addr, npages);
 519         else
 520                 set_memory_nx(addr, npages);
 521 }
 522
 523 void __init runtime_code_page_mkexec(void)
 524 {
 525         efi_memory_desc_t *md;
 526
 527         /* Make EFI runtime service code area executable */
 528         for_each_efi_memory_desc(md) {
 529                 if (md->type != EFI_RUNTIME_SERVICES_CODE)
 530                         continue;
 531
 532                 efi_set_executable(md, true);
 533         }
 534 }
 535
 536 void __init efi_memory_uc(u64 addr, unsigned long size)
 537 {
 538         unsigned long page_shift = 1UL << EFI_PAGE_SHIFT;
 539         u64 npages;
 540
 541         npages = round_up(size, page_shift) / page_shift;
 542         memrange_efi_to_native(&addr, &npages);
 543         set_memory_uc(addr, npages);
 544 }
 545
 546 void __init old_map_region(efi_memory_desc_t *md)
 547 {
 548         u64 start_pfn, end_pfn, end;
 549         unsigned long size;
 550         void *va;
 551
 552         start_pfn = PFN_DOWN(md->phys_addr);
 553         size      = md->num_pages << PAGE_SHIFT;
 554         end       = md->phys_addr + size;
 555         end_pfn   = PFN_UP(end);
 556
 557         if (pfn_range_is_mapped(start_pfn, end_pfn)) {
 558                 va = __va(md->phys_addr);
 559
 560                 if (!(md->attribute & EFI_MEMORY_WB))
 561                         efi_memory_uc((u64)(unsigned long)va, size);
 562         } else
 563                 va = efi_ioremap(md->phys_addr, size,
 564                                  md->type, md->attribute);
 565
 566         md->virt_addr = (u64) (unsigned long) va;
 567         if (!va)
 568                 pr_err("ioremap of 0x%llX failed!\n",
 569                        (unsigned long long)md->phys_addr);
 570 }
 571
 572 #endif
 573
 574 /* Merge contiguous regions of the same type and attribute */
 575 static void __init efi_merge_regions(void)
 576 {
 577         efi_memory_desc_t *md, *prev_md = NULL;
 578
 579         for_each_efi_memory_desc(md) {
 580                 u64 prev_size;
 581
 582                 if (!prev_md) {
 583                         prev_md = md;
 584                         continue;
 585                 }
 586
 587                 if (prev_md->type != md->type ||
 588                     prev_md->attribute != md->attribute) {
 589                         prev_md = md;
 590                         continue;
 591                 }
 592
 593                 prev_size = prev_md->num_pages << EFI_PAGE_SHIFT;
 594
 595                 if (md->phys_addr == (prev_md->phys_addr + prev_size)) {
 596                         prev_md->num_pages += md->num_pages;
 597                         md->type = EFI_RESERVED_TYPE;
 598                         md->attribute = 0;
 599                         continue;
 600                 }
 601                 prev_md = md;
 602         }
 603 }
 604
 605 static void __init get_systab_virt_addr(efi_memory_desc_t *md)
 606 {
 607         unsigned long size;
 608         u64 end, systab;
 609
 610         size = md->num_pages << EFI_PAGE_SHIFT;
 611         end = md->phys_addr + size;
 612         systab = efi_systab_phys;
 613         if (md->phys_addr <= systab && systab < end) {
 614                 systab += md->virt_addr - md->phys_addr;
 615                 efi.systab = (efi_system_table_t *)(unsigned long)systab;
 616         }
 617 }
 618
 619 static void *realloc_pages(void *old_memmap, int old_shift)
 620 {
 621         void *ret;
 622
 623         ret = (void *)__get_free_pages(GFP_KERNEL, old_shift + 1);
 624         if (!ret)
 625                 goto out;
 626
 627         /*
 628          * A first-time allocation doesn't have anything to copy.
 629          */
 630         if (!old_memmap)
 631                 return ret;
 632
 633         memcpy(ret, old_memmap, PAGE_SIZE << old_shift);
 634
 635 out:
 636         free_pages((unsigned long)old_memmap, old_shift);
 637         return ret;
 638 }
 639
 640 /*
 641  * Iterate the EFI memory map in reverse order because the regions
 642  * will be mapped top-down. The end result is the same as if we had
 643  * mapped things forward, but doesn't require us to change the
 644  * existing implementation of efi_map_region().
 645  */
 646 static inline void *efi_map_next_entry_reverse(void *entry)
 647 {
 648         /* Initial call */
 649         if (!entry)
 650                 return efi.memmap.map_end - efi.memmap.desc_size;
 651
 652         entry -= efi.memmap.desc_size;
 653         if (entry < efi.memmap.map)
 654                 return NULL;
 655
 656         return entry;
 657 }
 658
 659 /*
 660  * efi_map_next_entry - Return the next EFI memory map descriptor
 661  * @entry: Previous EFI memory map descriptor
 662  *
 663  * This is a helper function to iterate over the EFI memory map, which
 664  * we do in different orders depending on the current configuration.
 665  *
 666  * To begin traversing the memory map @entry must be %NULL.
 667  *
 668  * Returns %NULL when we reach the end of the memory map.
 669  */
 670 static void *efi_map_next_entry(void *entry)
 671 {
 672         if (!efi_have_uv1_memmap() && efi_enabled(EFI_64BIT)) {
 673                 /*
 674                  * Starting in UEFI v2.5 the EFI_PROPERTIES_TABLE
 675                  * config table feature requires us to map all entries
 676                  * in the same order as they appear in the EFI memory
 677                  * map. That is to say, entry N must have a lower
 678                  * virtual address than entry N+1. This is because the
 679                  * firmware toolchain leaves relative references in
 680                  * the code/data sections, which are split and become
 681                  * separate EFI memory regions. Mapping things
 682                  * out-of-order leads to the firmware accessing
 683                  * unmapped addresses.
 684                  *
 685                  * Since we need to map things this way whether or not
 686                  * the kernel actually makes use of
 687                  * EFI_PROPERTIES_TABLE, let's just switch to this
 688                  * scheme by default for 64-bit.
 689                  */
 690                 return efi_map_next_entry_reverse(entry);
 691         }
 692
 693         /* Initial call */
 694         if (!entry)
 695                 return efi.memmap.map;
 696
 697         entry += efi.memmap.desc_size;
 698         if (entry >= efi.memmap.map_end)
 699                 return NULL;
 700
 701         return entry;
 702 }
 703
 704 static bool should_map_region(efi_memory_desc_t *md)
 705 {
 706         /*
 707          * Runtime regions always require runtime mappings (obviously).
 708          */
 709         if (md->attribute & EFI_MEMORY_RUNTIME)
 710                 return true;
 711
 712         /*
 713          * 32-bit EFI doesn't suffer from the bug that requires us to
 714          * reserve boot services regions, and mixed mode support
 715          * doesn't exist for 32-bit kernels.
 716          */
 717         if (IS_ENABLED(CONFIG_X86_32))
 718                 return false;
 719
 720         /*
 721          * EFI specific purpose memory may be reserved by default
 722          * depending on kernel config and boot options.
 723          */
 724         if (md->type == EFI_CONVENTIONAL_MEMORY &&
 725             efi_soft_reserve_enabled() &&
 726             (md->attribute & EFI_MEMORY_SP))
 727                 return false;
 728
 729         /*
 730          * Map all of RAM so that we can access arguments in the 1:1
 731          * mapping when making EFI runtime calls.
 732          */
 733         if (efi_is_mixed()) {
 734                 if (md->type == EFI_CONVENTIONAL_MEMORY ||
 735                     md->type == EFI_LOADER_DATA ||
 736                     md->type == EFI_LOADER_CODE)
 737                         return true;
 738         }
 739
 740         /*
 741          * Map boot services regions as a workaround for buggy
 742          * firmware that accesses them even when they shouldn't.
 743          *
 744          * See efi_{reserve,free}_boot_services().
 745          */
 746         if (md->type == EFI_BOOT_SERVICES_CODE ||
 747             md->type == EFI_BOOT_SERVICES_DATA)
 748                 return true;
 749
 750         return false;
 751 }
 752
 753 /*
 754  * Map the efi memory ranges of the runtime services and update new_mmap with
 755  * virtual addresses.
 756  */
 757 static void * __init efi_map_regions(int *count, int *pg_shift)
 758 {
 759         void *p, *new_memmap = NULL;
 760         unsigned long left = 0;
 761         unsigned long desc_size;
 762         efi_memory_desc_t *md;
 763
 764         desc_size = efi.memmap.desc_size;
 765
 766         p = NULL;
 767         while ((p = efi_map_next_entry(p))) {
 768                 md = p;
 769
 770                 if (!should_map_region(md))
 771                         continue;
 772
 773                 efi_map_region(md);
 774                 get_systab_virt_addr(md);
 775
 776                 if (left < desc_size) {
 777                         new_memmap = realloc_pages(new_memmap, *pg_shift);
 778                         if (!new_memmap)
 779                                 return NULL;
 780
 781                         left += PAGE_SIZE << *pg_shift;
 782                         (*pg_shift)++;
 783                 }
 784
 785                 memcpy(new_memmap + (*count * desc_size), md, desc_size);
 786
 787                 left -= desc_size;
 788                 (*count)++;
 789         }
 790
 791         return new_memmap;
 792 }
 793
 794 static void __init kexec_enter_virtual_mode(void)
 795 {
 796 #ifdef CONFIG_KEXEC_CORE
 797         efi_memory_desc_t *md;
 798         unsigned int num_pages;
 799
 800         efi.systab = NULL;
 801
 802         /*
 803          * We don't do virtual mode, since we don't do runtime services, on
 804          * non-native EFI. With the UV1 memmap, we don't do runtime services in
 805          * kexec kernel because in the initial boot something else might
 806          * have been mapped at these virtual addresses.
 807          */
 808         if (efi_is_mixed() || efi_have_uv1_memmap()) {
 809                 efi_memmap_unmap();
 810                 clear_bit(EFI_RUNTIME_SERVICES, &efi.flags);
 811                 return;
 812         }
 813
 814         if (efi_alloc_page_tables()) {
 815                 pr_err("Failed to allocate EFI page tables\n");
 816                 clear_bit(EFI_RUNTIME_SERVICES, &efi.flags);
 817                 return;
 818         }
 819
 820         /*
 821         * Map efi regions which were passed via setup_data. The virt_addr is a
 822         * fixed addr which was used in first kernel of a kexec boot.
 823         */
 824         for_each_efi_memory_desc(md) {
 825                 efi_map_region_fixed(md); /* FIXME: add error handling */
 826                 get_systab_virt_addr(md);
 827         }
 828
 829         /*
 830          * Unregister the early EFI memmap from efi_init() and install
 831          * the new EFI memory map.
 832          */
 833         efi_memmap_unmap();
 834
 835         if (efi_memmap_init_late(efi.memmap.phys_map,
 836                                  efi.memmap.desc_size * efi.memmap.nr_map)) {
 837                 pr_err("Failed to remap late EFI memory map\n");
 838                 clear_bit(EFI_RUNTIME_SERVICES, &efi.flags);
 839                 return;
 840         }
 841
 842         BUG_ON(!efi.systab);
 843
 844         num_pages = ALIGN(efi.memmap.nr_map * efi.memmap.desc_size, PAGE_SIZE);
 845         num_pages >>= PAGE_SHIFT;
 846
 847         if (efi_setup_page_tables(efi.memmap.phys_map, num_pages)) {
 848                 clear_bit(EFI_RUNTIME_SERVICES, &efi.flags);
 849                 return;
 850         }
 851
 852         efi_sync_low_kernel_mappings();
 853
 854         /*
 855          * Now that EFI is in virtual mode, update the function
 856          * pointers in the runtime service table to the new virtual addresses.
 857          *
 858          * Call EFI services through wrapper functions.
 859          */
 860         efi.runtime_version = efi_systab.hdr.revision;
 861
 862         efi_native_runtime_setup();
 863 #endif
 864 }
 865
 866 /*
 867  * This function will switch the EFI runtime services to virtual mode.
 868  * Essentially, we look through the EFI memmap and map every region that
 869  * has the runtime attribute bit set in its memory descriptor into the
 870  * efi_pgd page table.
 871  *
 872  * The old method which used to update that memory descriptor with the
 873  * virtual address obtained from ioremap() is still supported when the
 874  * kernel is booted on SG1 UV1 hardware. Same old method enabled the
 875  * runtime services to be called without having to thunk back into
 876  * physical mode for every invocation.
 877  *
 878  * The new method does a pagetable switch in a preemption-safe manner
 879  * so that we're in a different address space when calling a runtime
 880  * function. For function arguments passing we do copy the PUDs of the
 881  * kernel page table into efi_pgd prior to each call.
 882  *
 883  * Specially for kexec boot, efi runtime maps in previous kernel should
 884  * be passed in via setup_data. In that case runtime ranges will be mapped
 885  * to the same virtual addresses as the first kernel, see
 886  * kexec_enter_virtual_mode().
 887  */
 888 static void __init __efi_enter_virtual_mode(void)
 889 {
 890         int count = 0, pg_shift = 0;
 891         void *new_memmap = NULL;
 892         efi_status_t status;
 893         unsigned long pa;
 894
 895         efi.systab = NULL;
 896
 897         if (efi_alloc_page_tables()) {
 898                 pr_err("Failed to allocate EFI page tables\n");
 899                 goto err;
 900         }
 901
 902         efi_merge_regions();
 903         new_memmap = efi_map_regions(&count, &pg_shift);
 904         if (!new_memmap) {
 905                 pr_err("Error reallocating memory, EFI runtime non-functional!\n");
 906                 goto err;
 907         }
 908
 909         pa = __pa(new_memmap);
 910
 911         /*
 912          * Unregister the early EFI memmap from efi_init() and install
 913          * the new EFI memory map that we are about to pass to the
 914          * firmware via SetVirtualAddressMap().
 915          */
 916         efi_memmap_unmap();
 917
 918         if (efi_memmap_init_late(pa, efi.memmap.desc_size * count)) {
 919                 pr_err("Failed to remap late EFI memory map\n");
 920                 goto err;
 921         }
 922
 923         if (efi_enabled(EFI_DBG)) {
 924                 pr_info("EFI runtime memory map:\n");
 925                 efi_print_memmap();
 926         }
 927
 928         if (WARN_ON(!efi.systab))
 929                 goto err;
 930
 931         if (efi_setup_page_tables(pa, 1 << pg_shift))
 932                 goto err;
 933
 934         efi_sync_low_kernel_mappings();
 935
 936         status = efi_set_virtual_address_map(efi.memmap.desc_size * count,
 937                                              efi.memmap.desc_size,
 938                                              efi.memmap.desc_version,
 939                                              (efi_memory_desc_t *)pa);
 940         if (status != EFI_SUCCESS) {
 941                 pr_err("Unable to switch EFI into virtual mode (status=%lx)!\n",
 942                        status);
 943                 goto err;
 944         }
 945
 946         efi_free_boot_services();
 947
 948         /*
 949          * Now that EFI is in virtual mode, update the function
 950          * pointers in the runtime service table to the new virtual addresses.
 951          *
 952          * Call EFI services through wrapper functions.
 953          */
 954         efi.runtime_version = efi_systab.hdr.revision;
 955
 956         if (!efi_is_mixed())
 957                 efi_native_runtime_setup();
 958         else
 959                 efi_thunk_runtime_setup();
 960
 961         /*
 962          * Apply more restrictive page table mapping attributes now that
 963          * SVAM() has been called and the firmware has performed all
 964          * necessary relocation fixups for the new virtual addresses.
 965          */
 966         efi_runtime_update_mappings();
 967
 968         /* clean DUMMY object */
 969         efi_delete_dummy_variable();
 970         return;
 971
 972 err:
 973         clear_bit(EFI_RUNTIME_SERVICES, &efi.flags);
 974 }
 975
 976 void __init efi_enter_virtual_mode(void)
 977 {
 978         if (efi_enabled(EFI_PARAVIRT))
 979                 return;
 980
 981         if (efi_setup)
 982                 kexec_enter_virtual_mode();
 983         else
 984                 __efi_enter_virtual_mode();
 985
 986         efi_dump_pagetable();
 987 }
 988
 989 bool efi_is_table_address(unsigned long phys_addr)
 990 {
 991         unsigned int i;
 992
 993         if (phys_addr == EFI_INVALID_TABLE_ADDR)
 994                 return false;
 995
 996         for (i = 0; i < ARRAY_SIZE(efi_tables); i++)
 997                 if (*(efi_tables[i]) == phys_addr)
 998                         return true;
 999
1000         return false;
1001 }