arch/x86/platform/efi/efi.c

   1 /*
   2  * Common EFI (Extensible Firmware Interface) support functions
   3  * Based on Extensible Firmware Interface Specification version 1.0
   4  *
   5  * Copyright (C) 1999 VA Linux Systems
   6  * Copyright (C) 1999 Walt Drummond <drummond@valinux.com>
   7  * Copyright (C) 1999-2002 Hewlett-Packard Co.
   8  *      David Mosberger-Tang <davidm@hpl.hp.com>
   9  *      Stephane Eranian <eranian@hpl.hp.com>
  10  * Copyright (C) 2005-2008 Intel Co.
  11  *      Fenghua Yu <fenghua.yu@intel.com>
  12  *      Bibo Mao <bibo.mao@intel.com>
  13  *      Chandramouli Narayanan <mouli@linux.intel.com>
  14  *      Huang Ying <ying.huang@intel.com>
  15  * Copyright (C) 2013 SuSE Labs
  16  *      Borislav Petkov <bp@suse.de> - runtime services VA mapping
  17  *
  18  * Copied from efi_32.c to eliminate the duplicated code between EFI
  19  * 32/64 support code. --ying 2007-10-26
  20  *
  21  * All EFI Runtime Services are not implemented yet as EFI only
  22  * supports physical mode addressing on SoftSDV. This is to be fixed
  23  * in a future version.  --drummond 1999-07-20
  24  *
  25  * Implemented EFI runtime services and virtual mode calls.  --davidm
  26  *
  27  * Goutham Rao: <goutham.rao@intel.com>
  28  *      Skip non-WB memory and ignore empty memory ranges.
  29  */
  30
  31 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  32
  33 #include <linux/kernel.h>
  34 #include <linux/init.h>
  35 #include <linux/efi.h>
  36 #include <linux/efi-bgrt.h>
  37 #include <linux/export.h>
  38 #include <linux/bootmem.h>
  39 #include <linux/slab.h>
  40 #include <linux/memblock.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/time.h>
  51 #include <asm/cacheflush.h>
  52 #include <asm/tlbflush.h>
  53 #include <asm/x86_init.h>
  54 #include <asm/uv/uv.h>
  55
  56 static struct efi efi_phys __initdata;
  57 static efi_system_table_t efi_systab __initdata;
  58
  59 static efi_config_table_type_t arch_tables[] __initdata = {
  60 #ifdef CONFIG_X86_UV
  61         {UV_SYSTEM_TABLE_GUID, "UVsystab", &efi.uv_systab},
  62 #endif
  63         {NULL_GUID, NULL, NULL},
  64 };
  65
  66 u64 efi_setup;          /* efi setup_data physical address */
  67
  68 static int add_efi_memmap __initdata;
  69 static int __init setup_add_efi_memmap(char *arg)
  70 {
  71         add_efi_memmap = 1;
  72         return 0;
  73 }
  74 early_param("add_efi_memmap", setup_add_efi_memmap);
  75
  76 static efi_status_t __init phys_efi_set_virtual_address_map(
  77         unsigned long memory_map_size,
  78         unsigned long descriptor_size,
  79         u32 descriptor_version,
  80         efi_memory_desc_t *virtual_map)
  81 {
  82         efi_status_t status;
  83         unsigned long flags;
  84         pgd_t *save_pgd;
  85
  86         save_pgd = efi_call_phys_prolog();
  87
  88         /* Disable interrupts around EFI calls: */
  89         local_irq_save(flags);
  90         status = efi_call_phys(efi_phys.set_virtual_address_map,
  91                                memory_map_size, descriptor_size,
  92                                descriptor_version, virtual_map);
  93         local_irq_restore(flags);
  94
  95         efi_call_phys_epilog(save_pgd);
  96
  97         return status;
  98 }
  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         for_each_efi_memory_desc(md) {
 106                 unsigned long long start = md->phys_addr;
 107                 unsigned long long size = md->num_pages << EFI_PAGE_SHIFT;
 108
 109                 total_size += size;
 110                 if (md->attribute & EFI_MEMORY_MORE_RELIABLE) {
 111                         memblock_mark_mirror(start, size);
 112                         mirror_size += size;
 113                 }
 114         }
 115         if (mirror_size)
 116                 pr_info("Memory: %lldM/%lldM mirrored memory\n",
 117                         mirror_size>>20, total_size>>20);
 118 }
 119
 120 /*
 121  * Tell the kernel about the EFI memory map.  This might include
 122  * more than the max 128 entries that can fit in the e820 legacy
 123  * (zeropage) memory map.
 124  */
 125
 126 static void __init do_add_efi_memmap(void)
 127 {
 128         efi_memory_desc_t *md;
 129
 130         for_each_efi_memory_desc(md) {
 131                 unsigned long long start = md->phys_addr;
 132                 unsigned long long size = md->num_pages << EFI_PAGE_SHIFT;
 133                 int e820_type;
 134
 135                 switch (md->type) {
 136                 case EFI_LOADER_CODE:
 137                 case EFI_LOADER_DATA:
 138                 case EFI_BOOT_SERVICES_CODE:
 139                 case EFI_BOOT_SERVICES_DATA:
 140                 case EFI_CONVENTIONAL_MEMORY:
 141                         if (md->attribute & EFI_MEMORY_WB)
 142                                 e820_type = E820_RAM;
 143                         else
 144                                 e820_type = E820_RESERVED;
 145                         break;
 146                 case EFI_ACPI_RECLAIM_MEMORY:
 147                         e820_type = E820_ACPI;
 148                         break;
 149                 case EFI_ACPI_MEMORY_NVS:
 150                         e820_type = E820_NVS;
 151                         break;
 152                 case EFI_UNUSABLE_MEMORY:
 153                         e820_type = E820_UNUSABLE;
 154                         break;
 155                 case EFI_PERSISTENT_MEMORY:
 156                         e820_type = E820_PMEM;
 157                         break;
 158                 default:
 159                         /*
 160                          * EFI_RESERVED_TYPE EFI_RUNTIME_SERVICES_CODE
 161                          * EFI_RUNTIME_SERVICES_DATA EFI_MEMORY_MAPPED_IO
 162                          * EFI_MEMORY_MAPPED_IO_PORT_SPACE EFI_PAL_CODE
 163                          */
 164                         e820_type = E820_RESERVED;
 165                         break;
 166                 }
 167                 e820_add_region(start, size, e820_type);
 168         }
 169         sanitize_e820_map(e820->map, ARRAY_SIZE(e820->map), &e820->nr_map);
 170 }
 171
 172 int __init efi_memblock_x86_reserve_range(void)
 173 {
 174         struct efi_info *e = &boot_params.efi_info;
 175         struct efi_memory_map_data data;
 176         phys_addr_t pmap;
 177         int rv;
 178
 179         if (efi_enabled(EFI_PARAVIRT))
 180                 return 0;
 181
 182 #ifdef CONFIG_X86_32
 183         /* Can't handle data above 4GB at this time */
 184         if (e->efi_memmap_hi) {
 185                 pr_err("Memory map is above 4GB, disabling EFI.\n");
 186                 return -EINVAL;
 187         }
 188         pmap =  e->efi_memmap;
 189 #else
 190         pmap = (e->efi_memmap | ((__u64)e->efi_memmap_hi << 32));
 191 #endif
 192         data.phys_map           = pmap;
 193         data.size               = e->efi_memmap_size;
 194         data.desc_size          = e->efi_memdesc_size;
 195         data.desc_version       = e->efi_memdesc_version;
 196
 197         rv = efi_memmap_init_early(&data);
 198         if (rv)
 199                 return rv;
 200
 201         if (add_efi_memmap)
 202                 do_add_efi_memmap();
 203
 204         WARN(efi.memmap.desc_version != 1,
 205              "Unexpected EFI_MEMORY_DESCRIPTOR version %ld",
 206              efi.memmap.desc_version);
 207
 208         memblock_reserve(pmap, efi.memmap.nr_map * efi.memmap.desc_size);
 209
 210         return 0;
 211 }
 212
 213 void __init efi_print_memmap(void)
 214 {
 215         efi_memory_desc_t *md;
 216         int i = 0;
 217
 218         for_each_efi_memory_desc(md) {
 219                 char buf[64];
 220
 221                 pr_info("mem%02u: %s range=[0x%016llx-0x%016llx] (%lluMB)\n",
 222                         i++, efi_md_typeattr_format(buf, sizeof(buf), md),
 223                         md->phys_addr,
 224                         md->phys_addr + (md->num_pages << EFI_PAGE_SHIFT) - 1,
 225                         (md->num_pages >> (20 - EFI_PAGE_SHIFT)));
 226         }
 227 }
 228
 229 static int __init efi_systab_init(void *phys)
 230 {
 231         if (efi_enabled(EFI_64BIT)) {
 232                 efi_system_table_64_t *systab64;
 233                 struct efi_setup_data *data = NULL;
 234                 u64 tmp = 0;
 235
 236                 if (efi_setup) {
 237                         data = early_memremap(efi_setup, sizeof(*data));
 238                         if (!data)
 239                                 return -ENOMEM;
 240                 }
 241                 systab64 = early_memremap((unsigned long)phys,
 242                                          sizeof(*systab64));
 243                 if (systab64 == NULL) {
 244                         pr_err("Couldn't map the system table!\n");
 245                         if (data)
 246                                 early_memunmap(data, sizeof(*data));
 247                         return -ENOMEM;
 248                 }
 249
 250                 efi_systab.hdr = systab64->hdr;
 251                 efi_systab.fw_vendor = data ? (unsigned long)data->fw_vendor :
 252                                               systab64->fw_vendor;
 253                 tmp |= data ? data->fw_vendor : systab64->fw_vendor;
 254                 efi_systab.fw_revision = systab64->fw_revision;
 255                 efi_systab.con_in_handle = systab64->con_in_handle;
 256                 tmp |= systab64->con_in_handle;
 257                 efi_systab.con_in = systab64->con_in;
 258                 tmp |= systab64->con_in;
 259                 efi_systab.con_out_handle = systab64->con_out_handle;
 260                 tmp |= systab64->con_out_handle;
 261                 efi_systab.con_out = systab64->con_out;
 262                 tmp |= systab64->con_out;
 263                 efi_systab.stderr_handle = systab64->stderr_handle;
 264                 tmp |= systab64->stderr_handle;
 265                 efi_systab.stderr = systab64->stderr;
 266                 tmp |= systab64->stderr;
 267                 efi_systab.runtime = data ?
 268                                      (void *)(unsigned long)data->runtime :
 269                                      (void *)(unsigned long)systab64->runtime;
 270                 tmp |= data ? data->runtime : systab64->runtime;
 271                 efi_systab.boottime = (void *)(unsigned long)systab64->boottime;
 272                 tmp |= systab64->boottime;
 273                 efi_systab.nr_tables = systab64->nr_tables;
 274                 efi_systab.tables = data ? (unsigned long)data->tables :
 275                                            systab64->tables;
 276                 tmp |= data ? data->tables : systab64->tables;
 277
 278                 early_memunmap(systab64, sizeof(*systab64));
 279                 if (data)
 280                         early_memunmap(data, sizeof(*data));
 281 #ifdef CONFIG_X86_32
 282                 if (tmp >> 32) {
 283                         pr_err("EFI data located above 4GB, disabling EFI.\n");
 284                         return -EINVAL;
 285                 }
 286 #endif
 287         } else {
 288                 efi_system_table_32_t *systab32;
 289
 290                 systab32 = early_memremap((unsigned long)phys,
 291                                          sizeof(*systab32));
 292                 if (systab32 == NULL) {
 293                         pr_err("Couldn't map the system table!\n");
 294                         return -ENOMEM;
 295                 }
 296
 297                 efi_systab.hdr = systab32->hdr;
 298                 efi_systab.fw_vendor = systab32->fw_vendor;
 299                 efi_systab.fw_revision = systab32->fw_revision;
 300                 efi_systab.con_in_handle = systab32->con_in_handle;
 301                 efi_systab.con_in = systab32->con_in;
 302                 efi_systab.con_out_handle = systab32->con_out_handle;
 303                 efi_systab.con_out = systab32->con_out;
 304                 efi_systab.stderr_handle = systab32->stderr_handle;
 305                 efi_systab.stderr = systab32->stderr;
 306                 efi_systab.runtime = (void *)(unsigned long)systab32->runtime;
 307                 efi_systab.boottime = (void *)(unsigned long)systab32->boottime;
 308                 efi_systab.nr_tables = systab32->nr_tables;
 309                 efi_systab.tables = systab32->tables;
 310
 311                 early_memunmap(systab32, sizeof(*systab32));
 312         }
 313
 314         efi.systab = &efi_systab;
 315
 316         /*
 317          * Verify the EFI Table
 318          */
 319         if (efi.systab->hdr.signature != EFI_SYSTEM_TABLE_SIGNATURE) {
 320                 pr_err("System table signature incorrect!\n");
 321                 return -EINVAL;
 322         }
 323         if ((efi.systab->hdr.revision >> 16) == 0)
 324                 pr_err("Warning: System table version %d.%02d, expected 1.00 or greater!\n",
 325                        efi.systab->hdr.revision >> 16,
 326                        efi.systab->hdr.revision & 0xffff);
 327
 328         return 0;
 329 }
 330
 331 static int __init efi_runtime_init32(void)
 332 {
 333         efi_runtime_services_32_t *runtime;
 334
 335         runtime = early_memremap((unsigned long)efi.systab->runtime,
 336                         sizeof(efi_runtime_services_32_t));
 337         if (!runtime) {
 338                 pr_err("Could not map the runtime service table!\n");
 339                 return -ENOMEM;
 340         }
 341
 342         /*
 343          * We will only need *early* access to the SetVirtualAddressMap
 344          * EFI runtime service. All other runtime services will be called
 345          * via the virtual mapping.
 346          */
 347         efi_phys.set_virtual_address_map =
 348                         (efi_set_virtual_address_map_t *)
 349                         (unsigned long)runtime->set_virtual_address_map;
 350         early_memunmap(runtime, sizeof(efi_runtime_services_32_t));
 351
 352         return 0;
 353 }
 354
 355 static int __init efi_runtime_init64(void)
 356 {
 357         efi_runtime_services_64_t *runtime;
 358
 359         runtime = early_memremap((unsigned long)efi.systab->runtime,
 360                         sizeof(efi_runtime_services_64_t));
 361         if (!runtime) {
 362                 pr_err("Could not map the runtime service table!\n");
 363                 return -ENOMEM;
 364         }
 365
 366         /*
 367          * We will only need *early* access to the SetVirtualAddressMap
 368          * EFI runtime service. All other runtime services will be called
 369          * via the virtual mapping.
 370          */
 371         efi_phys.set_virtual_address_map =
 372                         (efi_set_virtual_address_map_t *)
 373                         (unsigned long)runtime->set_virtual_address_map;
 374         early_memunmap(runtime, sizeof(efi_runtime_services_64_t));
 375
 376         return 0;
 377 }
 378
 379 static int __init efi_runtime_init(void)
 380 {
 381         int rv;
 382
 383         /*
 384          * Check out the runtime services table. We need to map
 385          * the runtime services table so that we can grab the physical
 386          * address of several of the EFI runtime functions, needed to
 387          * set the firmware into virtual mode.
 388          *
 389          * When EFI_PARAVIRT is in force then we could not map runtime
 390          * service memory region because we do not have direct access to it.
 391          * However, runtime services are available through proxy functions
 392          * (e.g. in case of Xen dom0 EFI implementation they call special
 393          * hypercall which executes relevant EFI functions) and that is why
 394          * they are always enabled.
 395          */
 396
 397         if (!efi_enabled(EFI_PARAVIRT)) {
 398                 if (efi_enabled(EFI_64BIT))
 399                         rv = efi_runtime_init64();
 400                 else
 401                         rv = efi_runtime_init32();
 402
 403                 if (rv)
 404                         return rv;
 405         }
 406
 407         set_bit(EFI_RUNTIME_SERVICES, &efi.flags);
 408
 409         return 0;
 410 }
 411
 412 void __init efi_init(void)
 413 {
 414         efi_char16_t *c16;
 415         char vendor[100] = "unknown";
 416         int i = 0;
 417         void *tmp;
 418
 419 #ifdef CONFIG_X86_32
 420         if (boot_params.efi_info.efi_systab_hi ||
 421             boot_params.efi_info.efi_memmap_hi) {
 422                 pr_info("Table located above 4GB, disabling EFI.\n");
 423                 return;
 424         }
 425         efi_phys.systab = (efi_system_table_t *)boot_params.efi_info.efi_systab;
 426 #else
 427         efi_phys.systab = (efi_system_table_t *)
 428                           (boot_params.efi_info.efi_systab |
 429                           ((__u64)boot_params.efi_info.efi_systab_hi<<32));
 430 #endif
 431
 432         if (efi_systab_init(efi_phys.systab))
 433                 return;
 434
 435         efi.config_table = (unsigned long)efi.systab->tables;
 436         efi.fw_vendor    = (unsigned long)efi.systab->fw_vendor;
 437         efi.runtime      = (unsigned long)efi.systab->runtime;
 438
 439         /*
 440          * Show what we know for posterity
 441          */
 442         c16 = tmp = early_memremap(efi.systab->fw_vendor, 2);
 443         if (c16) {
 444                 for (i = 0; i < sizeof(vendor) - 1 && *c16; ++i)
 445                         vendor[i] = *c16++;
 446                 vendor[i] = '\0';
 447         } else
 448                 pr_err("Could not map the firmware vendor!\n");
 449         early_memunmap(tmp, 2);
 450
 451         pr_info("EFI v%u.%.02u by %s\n",
 452                 efi.systab->hdr.revision >> 16,
 453                 efi.systab->hdr.revision & 0xffff, vendor);
 454
 455         if (efi_reuse_config(efi.systab->tables, efi.systab->nr_tables))
 456                 return;
 457
 458         if (efi_config_init(arch_tables))
 459                 return;
 460
 461         /*
 462          * Note: We currently don't support runtime services on an EFI
 463          * that doesn't match the kernel 32/64-bit mode.
 464          */
 465
 466         if (!efi_runtime_supported())
 467                 pr_info("No EFI runtime due to 32/64-bit mismatch with kernel\n");
 468         else {
 469                 if (efi_runtime_disabled() || efi_runtime_init()) {
 470                         efi_memmap_unmap();
 471                         return;
 472                 }
 473         }
 474
 475         if (efi_enabled(EFI_DBG))
 476                 efi_print_memmap();
 477 }
 478
 479 void __init efi_late_init(void)
 480 {
 481         efi_bgrt_init();
 482 }
 483
 484 void __init efi_set_executable(efi_memory_desc_t *md, bool executable)
 485 {
 486         u64 addr, npages;
 487
 488         addr = md->virt_addr;
 489         npages = md->num_pages;
 490
 491         memrange_efi_to_native(&addr, &npages);
 492
 493         if (executable)
 494                 set_memory_x(addr, npages);
 495         else
 496                 set_memory_nx(addr, npages);
 497 }
 498
 499 void __init runtime_code_page_mkexec(void)
 500 {
 501         efi_memory_desc_t *md;
 502
 503         /* Make EFI runtime service code area executable */
 504         for_each_efi_memory_desc(md) {
 505                 if (md->type != EFI_RUNTIME_SERVICES_CODE)
 506                         continue;
 507
 508                 efi_set_executable(md, true);
 509         }
 510 }
 511
 512 void __init efi_memory_uc(u64 addr, unsigned long size)
 513 {
 514         unsigned long page_shift = 1UL << EFI_PAGE_SHIFT;
 515         u64 npages;
 516
 517         npages = round_up(size, page_shift) / page_shift;
 518         memrange_efi_to_native(&addr, &npages);
 519         set_memory_uc(addr, npages);
 520 }
 521
 522 void __init old_map_region(efi_memory_desc_t *md)
 523 {
 524         u64 start_pfn, end_pfn, end;
 525         unsigned long size;
 526         void *va;
 527
 528         start_pfn = PFN_DOWN(md->phys_addr);
 529         size      = md->num_pages << PAGE_SHIFT;
 530         end       = md->phys_addr + size;
 531         end_pfn   = PFN_UP(end);
 532
 533         if (pfn_range_is_mapped(start_pfn, end_pfn)) {
 534                 va = __va(md->phys_addr);
 535
 536                 if (!(md->attribute & EFI_MEMORY_WB))
 537                         efi_memory_uc((u64)(unsigned long)va, size);
 538         } else
 539                 va = efi_ioremap(md->phys_addr, size,
 540                                  md->type, md->attribute);
 541
 542         md->virt_addr = (u64) (unsigned long) va;
 543         if (!va)
 544                 pr_err("ioremap of 0x%llX failed!\n",
 545                        (unsigned long long)md->phys_addr);
 546 }
 547
 548 /* Merge contiguous regions of the same type and attribute */
 549 static void __init efi_merge_regions(void)
 550 {
 551         efi_memory_desc_t *md, *prev_md = NULL;
 552
 553         for_each_efi_memory_desc(md) {
 554                 u64 prev_size;
 555
 556                 if (!prev_md) {
 557                         prev_md = md;
 558                         continue;
 559                 }
 560
 561                 if (prev_md->type != md->type ||
 562                     prev_md->attribute != md->attribute) {
 563                         prev_md = md;
 564                         continue;
 565                 }
 566
 567                 prev_size = prev_md->num_pages << EFI_PAGE_SHIFT;
 568
 569                 if (md->phys_addr == (prev_md->phys_addr + prev_size)) {
 570                         prev_md->num_pages += md->num_pages;
 571                         md->type = EFI_RESERVED_TYPE;
 572                         md->attribute = 0;
 573                         continue;
 574                 }
 575                 prev_md = md;
 576         }
 577 }
 578
 579 static void __init get_systab_virt_addr(efi_memory_desc_t *md)
 580 {
 581         unsigned long size;
 582         u64 end, systab;
 583
 584         size = md->num_pages << EFI_PAGE_SHIFT;
 585         end = md->phys_addr + size;
 586         systab = (u64)(unsigned long)efi_phys.systab;
 587         if (md->phys_addr <= systab && systab < end) {
 588                 systab += md->virt_addr - md->phys_addr;
 589                 efi.systab = (efi_system_table_t *)(unsigned long)systab;
 590         }
 591 }
 592
 593 static void *realloc_pages(void *old_memmap, int old_shift)
 594 {
 595         void *ret;
 596
 597         ret = (void *)__get_free_pages(GFP_KERNEL, old_shift + 1);
 598         if (!ret)
 599                 goto out;
 600
 601         /*
 602          * A first-time allocation doesn't have anything to copy.
 603          */
 604         if (!old_memmap)
 605                 return ret;
 606
 607         memcpy(ret, old_memmap, PAGE_SIZE << old_shift);
 608
 609 out:
 610         free_pages((unsigned long)old_memmap, old_shift);
 611         return ret;
 612 }
 613
 614 /*
 615  * Iterate the EFI memory map in reverse order because the regions
 616  * will be mapped top-down. The end result is the same as if we had
 617  * mapped things forward, but doesn't require us to change the
 618  * existing implementation of efi_map_region().
 619  */
 620 static inline void *efi_map_next_entry_reverse(void *entry)
 621 {
 622         /* Initial call */
 623         if (!entry)
 624                 return efi.memmap.map_end - efi.memmap.desc_size;
 625
 626         entry -= efi.memmap.desc_size;
 627         if (entry < efi.memmap.map)
 628                 return NULL;
 629
 630         return entry;
 631 }
 632
 633 /*
 634  * efi_map_next_entry - Return the next EFI memory map descriptor
 635  * @entry: Previous EFI memory map descriptor
 636  *
 637  * This is a helper function to iterate over the EFI memory map, which
 638  * we do in different orders depending on the current configuration.
 639  *
 640  * To begin traversing the memory map @entry must be %NULL.
 641  *
 642  * Returns %NULL when we reach the end of the memory map.
 643  */
 644 static void *efi_map_next_entry(void *entry)
 645 {
 646         if (!efi_enabled(EFI_OLD_MEMMAP) && efi_enabled(EFI_64BIT)) {
 647                 /*
 648                  * Starting in UEFI v2.5 the EFI_PROPERTIES_TABLE
 649                  * config table feature requires us to map all entries
 650                  * in the same order as they appear in the EFI memory
 651                  * map. That is to say, entry N must have a lower
 652                  * virtual address than entry N+1. This is because the
 653                  * firmware toolchain leaves relative references in
 654                  * the code/data sections, which are split and become
 655                  * separate EFI memory regions. Mapping things
 656                  * out-of-order leads to the firmware accessing
 657                  * unmapped addresses.
 658                  *
 659                  * Since we need to map things this way whether or not
 660                  * the kernel actually makes use of
 661                  * EFI_PROPERTIES_TABLE, let's just switch to this
 662                  * scheme by default for 64-bit.
 663                  */
 664                 return efi_map_next_entry_reverse(entry);
 665         }
 666
 667         /* Initial call */
 668         if (!entry)
 669                 return efi.memmap.map;
 670
 671         entry += efi.memmap.desc_size;
 672         if (entry >= efi.memmap.map_end)
 673                 return NULL;
 674
 675         return entry;
 676 }
 677
 678 static bool should_map_region(efi_memory_desc_t *md)
 679 {
 680         /*
 681          * Runtime regions always require runtime mappings (obviously).
 682          */
 683         if (md->attribute & EFI_MEMORY_RUNTIME)
 684                 return true;
 685
 686         /*
 687          * 32-bit EFI doesn't suffer from the bug that requires us to
 688          * reserve boot services regions, and mixed mode support
 689          * doesn't exist for 32-bit kernels.
 690          */
 691         if (IS_ENABLED(CONFIG_X86_32))
 692                 return false;
 693
 694         /*
 695          * Map all of RAM so that we can access arguments in the 1:1
 696          * mapping when making EFI runtime calls.
 697          */
 698         if (IS_ENABLED(CONFIG_EFI_MIXED) && !efi_is_native()) {
 699                 if (md->type == EFI_CONVENTIONAL_MEMORY ||
 700                     md->type == EFI_LOADER_DATA ||
 701                     md->type == EFI_LOADER_CODE)
 702                         return true;
 703         }
 704
 705         /*
 706          * Map boot services regions as a workaround for buggy
 707          * firmware that accesses them even when they shouldn't.
 708          *
 709          * See efi_{reserve,free}_boot_services().
 710          */
 711         if (md->type == EFI_BOOT_SERVICES_CODE ||
 712             md->type == EFI_BOOT_SERVICES_DATA)
 713                 return true;
 714
 715         return false;
 716 }
 717
 718 /*
 719  * Map the efi memory ranges of the runtime services and update new_mmap with
 720  * virtual addresses.
 721  */
 722 static void * __init efi_map_regions(int *count, int *pg_shift)
 723 {
 724         void *p, *new_memmap = NULL;
 725         unsigned long left = 0;
 726         unsigned long desc_size;
 727         efi_memory_desc_t *md;
 728
 729         desc_size = efi.memmap.desc_size;
 730
 731         p = NULL;
 732         while ((p = efi_map_next_entry(p))) {
 733                 md = p;
 734
 735                 if (!should_map_region(md))
 736                         continue;
 737
 738                 efi_map_region(md);
 739                 get_systab_virt_addr(md);
 740
 741                 if (left < desc_size) {
 742                         new_memmap = realloc_pages(new_memmap, *pg_shift);
 743                         if (!new_memmap)
 744                                 return NULL;
 745
 746                         left += PAGE_SIZE << *pg_shift;
 747                         (*pg_shift)++;
 748                 }
 749
 750                 memcpy(new_memmap + (*count * desc_size), md, desc_size);
 751
 752                 left -= desc_size;
 753                 (*count)++;
 754         }
 755
 756         return new_memmap;
 757 }
 758
 759 static void __init kexec_enter_virtual_mode(void)
 760 {
 761 #ifdef CONFIG_KEXEC_CORE
 762         efi_memory_desc_t *md;
 763         unsigned int num_pages;
 764
 765         efi.systab = NULL;
 766
 767         /*
 768          * We don't do virtual mode, since we don't do runtime services, on
 769          * non-native EFI
 770          */
 771         if (!efi_is_native()) {
 772                 efi_memmap_unmap();
 773                 clear_bit(EFI_RUNTIME_SERVICES, &efi.flags);
 774                 return;
 775         }
 776
 777         if (efi_alloc_page_tables()) {
 778                 pr_err("Failed to allocate EFI page tables\n");
 779                 clear_bit(EFI_RUNTIME_SERVICES, &efi.flags);
 780                 return;
 781         }
 782
 783         /*
 784         * Map efi regions which were passed via setup_data. The virt_addr is a
 785         * fixed addr which was used in first kernel of a kexec boot.
 786         */
 787         for_each_efi_memory_desc(md) {
 788                 efi_map_region_fixed(md); /* FIXME: add error handling */
 789                 get_systab_virt_addr(md);
 790         }
 791
 792         /*
 793          * Unregister the early EFI memmap from efi_init() and install
 794          * the new EFI memory map.
 795          */
 796         efi_memmap_unmap();
 797
 798         if (efi_memmap_init_late(efi.memmap.phys_map,
 799                                  efi.memmap.desc_size * efi.memmap.nr_map)) {
 800                 pr_err("Failed to remap late EFI memory map\n");
 801                 clear_bit(EFI_RUNTIME_SERVICES, &efi.flags);
 802                 return;
 803         }
 804
 805         BUG_ON(!efi.systab);
 806
 807         num_pages = ALIGN(efi.memmap.nr_map * efi.memmap.desc_size, PAGE_SIZE);
 808         num_pages >>= PAGE_SHIFT;
 809
 810         if (efi_setup_page_tables(efi.memmap.phys_map, num_pages)) {
 811                 clear_bit(EFI_RUNTIME_SERVICES, &efi.flags);
 812                 return;
 813         }
 814
 815         efi_sync_low_kernel_mappings();
 816
 817         /*
 818          * Now that EFI is in virtual mode, update the function
 819          * pointers in the runtime service table to the new virtual addresses.
 820          *
 821          * Call EFI services through wrapper functions.
 822          */
 823         efi.runtime_version = efi_systab.hdr.revision;
 824
 825         efi_native_runtime_setup();
 826
 827         efi.set_virtual_address_map = NULL;
 828
 829         if (efi_enabled(EFI_OLD_MEMMAP) && (__supported_pte_mask & _PAGE_NX))
 830                 runtime_code_page_mkexec();
 831
 832         /* clean DUMMY object */
 833         efi_delete_dummy_variable();
 834 #endif
 835 }
 836
 837 /*
 838  * This function will switch the EFI runtime services to virtual mode.
 839  * Essentially, we look through the EFI memmap and map every region that
 840  * has the runtime attribute bit set in its memory descriptor into the
 841  * efi_pgd page table.
 842  *
 843  * The old method which used to update that memory descriptor with the
 844  * virtual address obtained from ioremap() is still supported when the
 845  * kernel is booted with efi=old_map on its command line. Same old
 846  * method enabled the runtime services to be called without having to
 847  * thunk back into physical mode for every invocation.
 848  *
 849  * The new method does a pagetable switch in a preemption-safe manner
 850  * so that we're in a different address space when calling a runtime
 851  * function. For function arguments passing we do copy the PUDs of the
 852  * kernel page table into efi_pgd prior to each call.
 853  *
 854  * Specially for kexec boot, efi runtime maps in previous kernel should
 855  * be passed in via setup_data. In that case runtime ranges will be mapped
 856  * to the same virtual addresses as the first kernel, see
 857  * kexec_enter_virtual_mode().
 858  */
 859 static void __init __efi_enter_virtual_mode(void)
 860 {
 861         int count = 0, pg_shift = 0;
 862         void *new_memmap = NULL;
 863         efi_status_t status;
 864         phys_addr_t pa;
 865
 866         efi.systab = NULL;
 867
 868         if (efi_alloc_page_tables()) {
 869                 pr_err("Failed to allocate EFI page tables\n");
 870                 clear_bit(EFI_RUNTIME_SERVICES, &efi.flags);
 871                 return;
 872         }
 873
 874         efi_merge_regions();
 875         new_memmap = efi_map_regions(&count, &pg_shift);
 876         if (!new_memmap) {
 877                 pr_err("Error reallocating memory, EFI runtime non-functional!\n");
 878                 clear_bit(EFI_RUNTIME_SERVICES, &efi.flags);
 879                 return;
 880         }
 881
 882         pa = __pa(new_memmap);
 883
 884         /*
 885          * Unregister the early EFI memmap from efi_init() and install
 886          * the new EFI memory map that we are about to pass to the
 887          * firmware via SetVirtualAddressMap().
 888          */
 889         efi_memmap_unmap();
 890
 891         if (efi_memmap_init_late(pa, efi.memmap.desc_size * count)) {
 892                 pr_err("Failed to remap late EFI memory map\n");
 893                 clear_bit(EFI_RUNTIME_SERVICES, &efi.flags);
 894                 return;
 895         }
 896
 897         BUG_ON(!efi.systab);
 898
 899         if (efi_setup_page_tables(pa, 1 << pg_shift)) {
 900                 clear_bit(EFI_RUNTIME_SERVICES, &efi.flags);
 901                 return;
 902         }
 903
 904         efi_sync_low_kernel_mappings();
 905
 906         if (efi_is_native()) {
 907                 status = phys_efi_set_virtual_address_map(
 908                                 efi.memmap.desc_size * count,
 909                                 efi.memmap.desc_size,
 910                                 efi.memmap.desc_version,
 911                                 (efi_memory_desc_t *)pa);
 912         } else {
 913                 status = efi_thunk_set_virtual_address_map(
 914                                 efi_phys.set_virtual_address_map,
 915                                 efi.memmap.desc_size * count,
 916                                 efi.memmap.desc_size,
 917                                 efi.memmap.desc_version,
 918                                 (efi_memory_desc_t *)pa);
 919         }
 920
 921         if (status != EFI_SUCCESS) {
 922                 pr_alert("Unable to switch EFI into virtual mode (status=%lx)!\n",
 923                          status);
 924                 panic("EFI call to SetVirtualAddressMap() failed!");
 925         }
 926
 927         /*
 928          * Now that EFI is in virtual mode, update the function
 929          * pointers in the runtime service table to the new virtual addresses.
 930          *
 931          * Call EFI services through wrapper functions.
 932          */
 933         efi.runtime_version = efi_systab.hdr.revision;
 934
 935         if (efi_is_native())
 936                 efi_native_runtime_setup();
 937         else
 938                 efi_thunk_runtime_setup();
 939
 940         efi.set_virtual_address_map = NULL;
 941
 942         /*
 943          * Apply more restrictive page table mapping attributes now that
 944          * SVAM() has been called and the firmware has performed all
 945          * necessary relocation fixups for the new virtual addresses.
 946          */
 947         efi_runtime_update_mappings();
 948         efi_dump_pagetable();
 949
 950         /* clean DUMMY object */
 951         efi_delete_dummy_variable();
 952 }
 953
 954 void __init efi_enter_virtual_mode(void)
 955 {
 956         if (efi_enabled(EFI_PARAVIRT))
 957                 return;
 958
 959         if (efi_setup)
 960                 kexec_enter_virtual_mode();
 961         else
 962                 __efi_enter_virtual_mode();
 963 }
 964
 965 /*
 966  * Convenience functions to obtain memory types and attributes
 967  */
 968 u32 efi_mem_type(unsigned long phys_addr)
 969 {
 970         efi_memory_desc_t *md;
 971
 972         if (!efi_enabled(EFI_MEMMAP))
 973                 return 0;
 974
 975         for_each_efi_memory_desc(md) {
 976                 if ((md->phys_addr <= phys_addr) &&
 977                     (phys_addr < (md->phys_addr +
 978                                   (md->num_pages << EFI_PAGE_SHIFT))))
 979                         return md->type;
 980         }
 981         return 0;
 982 }
 983
 984 static int __init arch_parse_efi_cmdline(char *str)
 985 {
 986         if (!str) {
 987                 pr_warn("need at least one option\n");
 988                 return -EINVAL;
 989         }
 990
 991         if (parse_option_str(str, "old_map"))
 992                 set_bit(EFI_OLD_MEMMAP, &efi.flags);
 993
 994         return 0;
 995 }
 996 early_param("efi", arch_parse_efi_cmdline);